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/seq_file.h>
  51 #include <linux/mount.h>
  52 #include "nilfs.h"
  53 #include "export.h"
  54 #include "mdt.h"
  55 #include "alloc.h"
  56 #include "btree.h"
  57 #include "btnode.h"
  58 #include "page.h"
  59 #include "cpfile.h"
  60 #include "ifile.h"
  61 #include "dat.h"
  62 #include "segment.h"
  63 #include "segbuf.h"
  64
  65 MODULE_AUTHOR("NTT Corp.");
  66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  67                    "(NILFS)");
  68 MODULE_LICENSE("GPL");
  69
  70 static struct kmem_cache *nilfs_inode_cachep;
  71 struct kmem_cache *nilfs_transaction_cachep;
  72 struct kmem_cache *nilfs_segbuf_cachep;
  73 struct kmem_cache *nilfs_btree_path_cache;
  74
  75 static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount);
  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 the_nilfs *nilfs = sbi->s_nilfs;
 113         struct va_format vaf;
 114         va_list args;
 115
 116         va_start(args, fmt);
 117
 118         vaf.fmt = fmt;
 119         vaf.va = &args;
 120
 121         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 122                sb->s_id, function, &vaf);
 123
 124         va_end(args);
 125
 126         if (!(sb->s_flags & MS_RDONLY)) {
 127                 nilfs_set_error(sbi);
 128
 129                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 130                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 131                         sb->s_flags |= MS_RDONLY;
 132                 }
 133         }
 134
 135         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 136                 panic("NILFS (device %s): panic forced after error\n",
 137                       sb->s_id);
 138 }
 139
 140 void nilfs_warning(struct super_block *sb, const char *function,
 141                    const char *fmt, ...)
 142 {
 143         struct va_format vaf;
 144         va_list args;
 145
 146         va_start(args, fmt);
 147
 148         vaf.fmt = fmt;
 149         vaf.va = &args;
 150
 151         printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
 152                sb->s_id, function, &vaf);
 153
 154         va_end(args);
 155 }
 156
 157
 158 struct inode *nilfs_alloc_inode(struct super_block *sb)
 159 {
 160         struct nilfs_inode_info *ii;
 161
 162         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 163         if (!ii)
 164                 return NULL;
 165         ii->i_bh = NULL;
 166         ii->i_state = 0;
 167         ii->i_cno = 0;
 168         ii->vfs_inode.i_version = 1;
 169         nilfs_btnode_cache_init(&ii->i_btnode_cache, sb->s_bdi);
 170         return &ii->vfs_inode;
 171 }
 172
 173 static void nilfs_i_callback(struct rcu_head *head)
 174 {
 175         struct inode *inode = container_of(head, struct inode, i_rcu);
 176         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 177
 178         INIT_LIST_HEAD(&inode->i_dentry);
 179
 180         if (mdi) {
 181                 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
 182                 kfree(mdi);
 183         }
 184         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 185 }
 186
 187 void nilfs_destroy_inode(struct inode *inode)
 188 {
 189         call_rcu(&inode->i_rcu, nilfs_i_callback);
 190 }
 191
 192 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
 193 {
 194         struct the_nilfs *nilfs = sbi->s_nilfs;
 195         int err;
 196
 197  retry:
 198         set_buffer_dirty(nilfs->ns_sbh[0]);
 199         if (nilfs_test_opt(nilfs, BARRIER)) {
 200                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 201                                           WRITE_SYNC | WRITE_FLUSH_FUA);
 202         } else {
 203                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 204         }
 205
 206         if (unlikely(err)) {
 207                 printk(KERN_ERR
 208                        "NILFS: unable to write superblock (err=%d)\n", err);
 209                 if (err == -EIO && nilfs->ns_sbh[1]) {
 210                         /*
 211                          * sbp[0] points to newer log than sbp[1],
 212                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 213                          */
 214                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 215                                nilfs->ns_sbsize);
 216                         nilfs_fall_back_super_block(nilfs);
 217                         goto retry;
 218                 }
 219         } else {
 220                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 221
 222                 nilfs->ns_sbwcount++;
 223
 224                 /*
 225                  * The latest segment becomes trailable from the position
 226                  * written in superblock.
 227                  */
 228                 clear_nilfs_discontinued(nilfs);
 229
 230                 /* update GC protection for recent segments */
 231                 if (nilfs->ns_sbh[1]) {
 232                         if (flag == NILFS_SB_COMMIT_ALL) {
 233                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 234                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 235                                         goto out;
 236                         }
 237                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 238                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 239                                 sbp = nilfs->ns_sbp[1];
 240                 }
 241
 242                 spin_lock(&nilfs->ns_last_segment_lock);
 243                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 244                 spin_unlock(&nilfs->ns_last_segment_lock);
 245         }
 246  out:
 247         return err;
 248 }
 249
 250 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 251                           struct the_nilfs *nilfs)
 252 {
 253         sector_t nfreeblocks;
 254
 255         /* nilfs->ns_sem must be locked by the caller. */
 256         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 257         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 258
 259         spin_lock(&nilfs->ns_last_segment_lock);
 260         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 261         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 262         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 263         spin_unlock(&nilfs->ns_last_segment_lock);
 264 }
 265
 266 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
 267                                                int flip)
 268 {
 269         struct the_nilfs *nilfs = sbi->s_nilfs;
 270         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 271
 272         /* nilfs->ns_sem must be locked by the caller. */
 273         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 274                 if (sbp[1] &&
 275                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 276                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 277                 } else {
 278                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 279                                sbi->s_super->s_id);
 280                         return NULL;
 281                 }
 282         } else if (sbp[1] &&
 283                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 284                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 285         }
 286
 287         if (flip && sbp[1])
 288                 nilfs_swap_super_block(nilfs);
 289
 290         return sbp;
 291 }
 292
 293 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
 294 {
 295         struct the_nilfs *nilfs = sbi->s_nilfs;
 296         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 297         time_t t;
 298
 299         /* nilfs->ns_sem must be locked by the caller. */
 300         t = get_seconds();
 301         nilfs->ns_sbwtime = t;
 302         sbp[0]->s_wtime = cpu_to_le64(t);
 303         sbp[0]->s_sum = 0;
 304         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 305                                              (unsigned char *)sbp[0],
 306                                              nilfs->ns_sbsize));
 307         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 308                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 309                 sbp[1]->s_sum = 0;
 310                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 311                                             (unsigned char *)sbp[1],
 312                                             nilfs->ns_sbsize));
 313         }
 314         clear_nilfs_sb_dirty(nilfs);
 315         return nilfs_sync_super(sbi, flag);
 316 }
 317
 318 /**
 319  * nilfs_cleanup_super() - write filesystem state for cleanup
 320  * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
 321  *
 322  * This function restores state flags in the on-disk super block.
 323  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 324  * filesystem was not clean previously.
 325  */
 326 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
 327 {
 328         struct nilfs_super_block **sbp;
 329         int flag = NILFS_SB_COMMIT;
 330         int ret = -EIO;
 331
 332         sbp = nilfs_prepare_super(sbi, 0);
 333         if (sbp) {
 334                 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
 335                 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
 336                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 337                         /*
 338                          * make the "clean" flag also to the opposite
 339                          * super block if both super blocks point to
 340                          * the same checkpoint.
 341                          */
 342                         sbp[1]->s_state = sbp[0]->s_state;
 343                         flag = NILFS_SB_COMMIT_ALL;
 344                 }
 345                 ret = nilfs_commit_super(sbi, flag);
 346         }
 347         return ret;
 348 }
 349
 350 static void nilfs_put_super(struct super_block *sb)
 351 {
 352         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 353         struct the_nilfs *nilfs = sbi->s_nilfs;
 354
 355         nilfs_detach_segment_constructor(sbi);
 356
 357         if (!(sb->s_flags & MS_RDONLY)) {
 358                 down_write(&nilfs->ns_sem);
 359                 nilfs_cleanup_super(sbi);
 360                 up_write(&nilfs->ns_sem);
 361         }
 362
 363         iput(nilfs->ns_sufile);
 364         iput(nilfs->ns_cpfile);
 365         iput(nilfs->ns_dat);
 366
 367         destroy_nilfs(nilfs);
 368         sbi->s_super = NULL;
 369         sb->s_fs_info = NULL;
 370         kfree(sbi);
 371 }
 372
 373 static int nilfs_sync_fs(struct super_block *sb, int wait)
 374 {
 375         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 376         struct the_nilfs *nilfs = sbi->s_nilfs;
 377         struct nilfs_super_block **sbp;
 378         int err = 0;
 379
 380         /* This function is called when super block should be written back */
 381         if (wait)
 382                 err = nilfs_construct_segment(sb);
 383
 384         down_write(&nilfs->ns_sem);
 385         if (nilfs_sb_dirty(nilfs)) {
 386                 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
 387                 if (likely(sbp)) {
 388                         nilfs_set_log_cursor(sbp[0], nilfs);
 389                         nilfs_commit_super(sbi, NILFS_SB_COMMIT);
 390                 }
 391         }
 392         up_write(&nilfs->ns_sem);
 393
 394         return err;
 395 }
 396
 397 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
 398                             struct nilfs_root **rootp)
 399 {
 400         struct the_nilfs *nilfs = sbi->s_nilfs;
 401         struct nilfs_root *root;
 402         struct nilfs_checkpoint *raw_cp;
 403         struct buffer_head *bh_cp;
 404         int err = -ENOMEM;
 405
 406         root = nilfs_find_or_create_root(
 407                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 408         if (!root)
 409                 return err;
 410
 411         if (root->ifile)
 412                 goto reuse; /* already attached checkpoint */
 413
 414         down_read(&nilfs->ns_segctor_sem);
 415         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 416                                           &bh_cp);
 417         up_read(&nilfs->ns_segctor_sem);
 418         if (unlikely(err)) {
 419                 if (err == -ENOENT || err == -EINVAL) {
 420                         printk(KERN_ERR
 421                                "NILFS: Invalid checkpoint "
 422                                "(checkpoint number=%llu)\n",
 423                                (unsigned long long)cno);
 424                         err = -EINVAL;
 425                 }
 426                 goto failed;
 427         }
 428
 429         err = nilfs_ifile_read(sbi->s_super, root, nilfs->ns_inode_size,
 430                                &raw_cp->cp_ifile_inode, &root->ifile);
 431         if (err)
 432                 goto failed_bh;
 433
 434         atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 435         atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 436
 437         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 438
 439  reuse:
 440         *rootp = root;
 441         return 0;
 442
 443  failed_bh:
 444         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 445  failed:
 446         nilfs_put_root(root);
 447
 448         return err;
 449 }
 450
 451 static int nilfs_freeze(struct super_block *sb)
 452 {
 453         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 454         struct the_nilfs *nilfs = sbi->s_nilfs;
 455         int err;
 456
 457         if (sb->s_flags & MS_RDONLY)
 458                 return 0;
 459
 460         /* Mark super block clean */
 461         down_write(&nilfs->ns_sem);
 462         err = nilfs_cleanup_super(sbi);
 463         up_write(&nilfs->ns_sem);
 464         return err;
 465 }
 466
 467 static int nilfs_unfreeze(struct super_block *sb)
 468 {
 469         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 470         struct the_nilfs *nilfs = sbi->s_nilfs;
 471
 472         if (sb->s_flags & MS_RDONLY)
 473                 return 0;
 474
 475         down_write(&nilfs->ns_sem);
 476         nilfs_setup_super(sbi, false);
 477         up_write(&nilfs->ns_sem);
 478         return 0;
 479 }
 480
 481 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 482 {
 483         struct super_block *sb = dentry->d_sb;
 484         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
 485         struct the_nilfs *nilfs = root->nilfs;
 486         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 487         unsigned long long blocks;
 488         unsigned long overhead;
 489         unsigned long nrsvblocks;
 490         sector_t nfreeblocks;
 491         int err;
 492
 493         /*
 494          * Compute all of the segment blocks
 495          *
 496          * The blocks before first segment and after last segment
 497          * are excluded.
 498          */
 499         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 500                 - nilfs->ns_first_data_block;
 501         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 502
 503         /*
 504          * Compute the overhead
 505          *
 506          * When distributing meta data blocks outside segment structure,
 507          * We must count them as the overhead.
 508          */
 509         overhead = 0;
 510
 511         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 512         if (unlikely(err))
 513                 return err;
 514
 515         buf->f_type = NILFS_SUPER_MAGIC;
 516         buf->f_bsize = sb->s_blocksize;
 517         buf->f_blocks = blocks - overhead;
 518         buf->f_bfree = nfreeblocks;
 519         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 520                 (buf->f_bfree - nrsvblocks) : 0;
 521         buf->f_files = atomic_read(&root->inodes_count);
 522         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 523         buf->f_namelen = NILFS_NAME_LEN;
 524         buf->f_fsid.val[0] = (u32)id;
 525         buf->f_fsid.val[1] = (u32)(id >> 32);
 526
 527         return 0;
 528 }
 529
 530 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 531 {
 532         struct super_block *sb = vfs->mnt_sb;
 533         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
 534         struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
 535
 536         if (!nilfs_test_opt(nilfs, BARRIER))
 537                 seq_puts(seq, ",nobarrier");
 538         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 539                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 540         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 541                 seq_puts(seq, ",errors=panic");
 542         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 543                 seq_puts(seq, ",errors=continue");
 544         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 545                 seq_puts(seq, ",order=strict");
 546         if (nilfs_test_opt(nilfs, NORECOVERY))
 547                 seq_puts(seq, ",norecovery");
 548         if (nilfs_test_opt(nilfs, DISCARD))
 549                 seq_puts(seq, ",discard");
 550
 551         return 0;
 552 }
 553
 554 static const struct super_operations nilfs_sops = {
 555         .alloc_inode    = nilfs_alloc_inode,
 556         .destroy_inode  = nilfs_destroy_inode,
 557         .dirty_inode    = nilfs_dirty_inode,
 558         /* .write_inode    = nilfs_write_inode, */
 559         /* .put_inode      = nilfs_put_inode, */
 560         /* .drop_inode    = nilfs_drop_inode, */
 561         .evict_inode    = nilfs_evict_inode,
 562         .put_super      = nilfs_put_super,
 563         /* .write_super    = nilfs_write_super, */
 564         .sync_fs        = nilfs_sync_fs,
 565         .freeze_fs      = nilfs_freeze,
 566         .unfreeze_fs    = nilfs_unfreeze,
 567         /* .write_super_lockfs */
 568         /* .unlockfs */
 569         .statfs         = nilfs_statfs,
 570         .remount_fs     = nilfs_remount,
 571         /* .umount_begin */
 572         .show_options = nilfs_show_options
 573 };
 574
 575 enum {
 576         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 577         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 578         Opt_discard, Opt_nodiscard, Opt_err,
 579 };
 580
 581 static match_table_t tokens = {
 582         {Opt_err_cont, "errors=continue"},
 583         {Opt_err_panic, "errors=panic"},
 584         {Opt_err_ro, "errors=remount-ro"},
 585         {Opt_barrier, "barrier"},
 586         {Opt_nobarrier, "nobarrier"},
 587         {Opt_snapshot, "cp=%u"},
 588         {Opt_order, "order=%s"},
 589         {Opt_norecovery, "norecovery"},
 590         {Opt_discard, "discard"},
 591         {Opt_nodiscard, "nodiscard"},
 592         {Opt_err, NULL}
 593 };
 594
 595 static int parse_options(char *options, struct super_block *sb, int is_remount)
 596 {
 597         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
 598         char *p;
 599         substring_t args[MAX_OPT_ARGS];
 600
 601         if (!options)
 602                 return 1;
 603
 604         while ((p = strsep(&options, ",")) != NULL) {
 605                 int token;
 606                 if (!*p)
 607                         continue;
 608
 609                 token = match_token(p, tokens, args);
 610                 switch (token) {
 611                 case Opt_barrier:
 612                         nilfs_set_opt(nilfs, BARRIER);
 613                         break;
 614                 case Opt_nobarrier:
 615                         nilfs_clear_opt(nilfs, BARRIER);
 616                         break;
 617                 case Opt_order:
 618                         if (strcmp(args[0].from, "relaxed") == 0)
 619                                 /* Ordered data semantics */
 620                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 621                         else if (strcmp(args[0].from, "strict") == 0)
 622                                 /* Strict in-order semantics */
 623                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 624                         else
 625                                 return 0;
 626                         break;
 627                 case Opt_err_panic:
 628                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 629                         break;
 630                 case Opt_err_ro:
 631                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 632                         break;
 633                 case Opt_err_cont:
 634                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 635                         break;
 636                 case Opt_snapshot:
 637                         if (is_remount) {
 638                                 printk(KERN_ERR
 639                                        "NILFS: \"%s\" option is invalid "
 640                                        "for remount.\n", p);
 641                                 return 0;
 642                         }
 643                         break;
 644                 case Opt_norecovery:
 645                         nilfs_set_opt(nilfs, NORECOVERY);
 646                         break;
 647                 case Opt_discard:
 648                         nilfs_set_opt(nilfs, DISCARD);
 649                         break;
 650                 case Opt_nodiscard:
 651                         nilfs_clear_opt(nilfs, DISCARD);
 652                         break;
 653                 default:
 654                         printk(KERN_ERR
 655                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 656                         return 0;
 657                 }
 658         }
 659         return 1;
 660 }
 661
 662 static inline void
 663 nilfs_set_default_options(struct super_block *sb,
 664                           struct nilfs_super_block *sbp)
 665 {
 666         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
 667
 668         nilfs->ns_mount_opt =
 669                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 670 }
 671
 672 static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount)
 673 {
 674         struct the_nilfs *nilfs = sbi->s_nilfs;
 675         struct nilfs_super_block **sbp;
 676         int max_mnt_count;
 677         int mnt_count;
 678
 679         /* nilfs->ns_sem must be locked by the caller. */
 680         sbp = nilfs_prepare_super(sbi, 0);
 681         if (!sbp)
 682                 return -EIO;
 683
 684         if (!is_mount)
 685                 goto skip_mount_setup;
 686
 687         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 688         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 689
 690         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 691                 printk(KERN_WARNING
 692                        "NILFS warning: mounting fs with errors\n");
 693 #if 0
 694         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 695                 printk(KERN_WARNING
 696                        "NILFS warning: maximal mount count reached\n");
 697 #endif
 698         }
 699         if (!max_mnt_count)
 700                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 701
 702         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 703         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 704
 705 skip_mount_setup:
 706         sbp[0]->s_state =
 707                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 708         /* synchronize sbp[1] with sbp[0] */
 709         if (sbp[1])
 710                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 711         return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
 712 }
 713
 714 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 715                                                  u64 pos, int blocksize,
 716                                                  struct buffer_head **pbh)
 717 {
 718         unsigned long long sb_index = pos;
 719         unsigned long offset;
 720
 721         offset = do_div(sb_index, blocksize);
 722         *pbh = sb_bread(sb, sb_index);
 723         if (!*pbh)
 724                 return NULL;
 725         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 726 }
 727
 728 int nilfs_store_magic_and_option(struct super_block *sb,
 729                                  struct nilfs_super_block *sbp,
 730                                  char *data)
 731 {
 732         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
 733
 734         sb->s_magic = le16_to_cpu(sbp->s_magic);
 735
 736         /* FS independent flags */
 737 #ifdef NILFS_ATIME_DISABLE
 738         sb->s_flags |= MS_NOATIME;
 739 #endif
 740
 741         nilfs_set_default_options(sb, sbp);
 742
 743         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 744         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 745         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 746         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 747
 748         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
 749 }
 750
 751 int nilfs_check_feature_compatibility(struct super_block *sb,
 752                                       struct nilfs_super_block *sbp)
 753 {
 754         __u64 features;
 755
 756         features = le64_to_cpu(sbp->s_feature_incompat) &
 757                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 758         if (features) {
 759                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 760                        "optional features (%llx)\n",
 761                        (unsigned long long)features);
 762                 return -EINVAL;
 763         }
 764         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 765                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 766         if (!(sb->s_flags & MS_RDONLY) && features) {
 767                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
 768                        "unsupported optional features (%llx)\n",
 769                        (unsigned long long)features);
 770                 return -EINVAL;
 771         }
 772         return 0;
 773 }
 774
 775 static int nilfs_get_root_dentry(struct super_block *sb,
 776                                  struct nilfs_root *root,
 777                                  struct dentry **root_dentry)
 778 {
 779         struct inode *inode;
 780         struct dentry *dentry;
 781         int ret = 0;
 782
 783         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 784         if (IS_ERR(inode)) {
 785                 printk(KERN_ERR "NILFS: get root inode failed\n");
 786                 ret = PTR_ERR(inode);
 787                 goto out;
 788         }
 789         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 790                 iput(inode);
 791                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 792                 ret = -EINVAL;
 793                 goto out;
 794         }
 795
 796         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 797                 dentry = d_find_alias(inode);
 798                 if (!dentry) {
 799                         dentry = d_alloc_root(inode);
 800                         if (!dentry) {
 801                                 iput(inode);
 802                                 ret = -ENOMEM;
 803                                 goto failed_dentry;
 804                         }
 805                 } else {
 806                         iput(inode);
 807                 }
 808         } else {
 809                 dentry = d_obtain_alias(inode);
 810                 if (IS_ERR(dentry)) {
 811                         ret = PTR_ERR(dentry);
 812                         goto failed_dentry;
 813                 }
 814         }
 815         *root_dentry = dentry;
 816  out:
 817         return ret;
 818
 819  failed_dentry:
 820         printk(KERN_ERR "NILFS: get root dentry failed\n");
 821         goto out;
 822 }
 823
 824 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 825                                  struct dentry **root_dentry)
 826 {
 827         struct the_nilfs *nilfs = NILFS_SB(s)->s_nilfs;
 828         struct nilfs_root *root;
 829         int ret;
 830
 831         down_read(&nilfs->ns_segctor_sem);
 832         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 833         up_read(&nilfs->ns_segctor_sem);
 834         if (ret < 0) {
 835                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 836                 goto out;
 837         } else if (!ret) {
 838                 printk(KERN_ERR "NILFS: The specified checkpoint is "
 839                        "not a snapshot (checkpoint number=%llu).\n",
 840                        (unsigned long long)cno);
 841                 ret = -EINVAL;
 842                 goto out;
 843         }
 844
 845         ret = nilfs_attach_checkpoint(NILFS_SB(s), cno, false, &root);
 846         if (ret) {
 847                 printk(KERN_ERR "NILFS: error loading snapshot "
 848                        "(checkpoint number=%llu).\n",
 849                (unsigned long long)cno);
 850                 goto out;
 851         }
 852         ret = nilfs_get_root_dentry(s, root, root_dentry);
 853         nilfs_put_root(root);
 854  out:
 855         return ret;
 856 }
 857
 858 static int nilfs_tree_was_touched(struct dentry *root_dentry)
 859 {
 860         return root_dentry->d_count > 1;
 861 }
 862
 863 /**
 864  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
 865  * @root_dentry: root dentry of the tree to be shrunk
 866  *
 867  * This function returns true if the tree was in-use.
 868  */
 869 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
 870 {
 871         if (have_submounts(root_dentry))
 872                 return true;
 873         shrink_dcache_parent(root_dentry);
 874         return nilfs_tree_was_touched(root_dentry);
 875 }
 876
 877 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
 878 {
 879         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
 880         struct nilfs_root *root;
 881         struct inode *inode;
 882         struct dentry *dentry;
 883         int ret;
 884
 885         if (cno < 0 || cno > nilfs->ns_cno)
 886                 return false;
 887
 888         if (cno >= nilfs_last_cno(nilfs))
 889                 return true;    /* protect recent checkpoints */
 890
 891         ret = false;
 892         root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
 893         if (root) {
 894                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
 895                 if (inode) {
 896                         dentry = d_find_alias(inode);
 897                         if (dentry) {
 898                                 if (nilfs_tree_was_touched(dentry))
 899                                         ret = nilfs_try_to_shrink_tree(dentry);
 900                                 dput(dentry);
 901                         }
 902                         iput(inode);
 903                 }
 904                 nilfs_put_root(root);
 905         }
 906         return ret;
 907 }
 908
 909 /**
 910  * nilfs_fill_super() - initialize a super block instance
 911  * @sb: super_block
 912  * @data: mount options
 913  * @silent: silent mode flag
 914  *
 915  * This function is called exclusively by nilfs->ns_mount_mutex.
 916  * So, the recovery process is protected from other simultaneous mounts.
 917  */
 918 static int
 919 nilfs_fill_super(struct super_block *sb, void *data, int silent)
 920 {
 921         struct the_nilfs *nilfs;
 922         struct nilfs_sb_info *sbi;
 923         struct nilfs_root *fsroot;
 924         struct backing_dev_info *bdi;
 925         __u64 cno;
 926         int err;
 927
 928         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 929         if (!sbi)
 930                 return -ENOMEM;
 931
 932         sb->s_fs_info = sbi;
 933         sbi->s_super = sb;
 934
 935         nilfs = alloc_nilfs(sb->s_bdev);
 936         if (!nilfs) {
 937                 err = -ENOMEM;
 938                 goto failed_sbi;
 939         }
 940         sbi->s_nilfs = nilfs;
 941
 942         err = init_nilfs(nilfs, sbi, (char *)data);
 943         if (err)
 944                 goto failed_nilfs;
 945
 946         /*
 947          * Following initialization is overlapped because
 948          * nilfs_sb_info structure has been cleared at the beginning.
 949          * But we reserve them to keep our interest and make ready
 950          * for the future change.
 951          */
 952         get_random_bytes(&sbi->s_next_generation,
 953                          sizeof(sbi->s_next_generation));
 954         spin_lock_init(&sbi->s_next_gen_lock);
 955
 956         sb->s_op = &nilfs_sops;
 957         sb->s_export_op = &nilfs_export_ops;
 958         sb->s_root = NULL;
 959         sb->s_time_gran = 1;
 960
 961         bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
 962         sb->s_bdi = bdi ? : &default_backing_dev_info;
 963
 964         err = load_nilfs(nilfs, sbi);
 965         if (err)
 966                 goto failed_nilfs;
 967
 968         cno = nilfs_last_cno(nilfs);
 969         err = nilfs_attach_checkpoint(sbi, cno, true, &fsroot);
 970         if (err) {
 971                 printk(KERN_ERR "NILFS: error loading last checkpoint "
 972                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
 973                 goto failed_unload;
 974         }
 975
 976         if (!(sb->s_flags & MS_RDONLY)) {
 977                 err = nilfs_attach_segment_constructor(sbi, fsroot);
 978                 if (err)
 979                         goto failed_checkpoint;
 980         }
 981
 982         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
 983         if (err)
 984                 goto failed_segctor;
 985
 986         nilfs_put_root(fsroot);
 987
 988         if (!(sb->s_flags & MS_RDONLY)) {
 989                 down_write(&nilfs->ns_sem);
 990                 nilfs_setup_super(sbi, true);
 991                 up_write(&nilfs->ns_sem);
 992         }
 993
 994         return 0;
 995
 996  failed_segctor:
 997         nilfs_detach_segment_constructor(sbi);
 998
 999  failed_checkpoint:
1000         nilfs_put_root(fsroot);
1001
1002  failed_unload:
1003         iput(nilfs->ns_sufile);
1004         iput(nilfs->ns_cpfile);
1005         iput(nilfs->ns_dat);
1006
1007  failed_nilfs:
1008         destroy_nilfs(nilfs);
1009
1010  failed_sbi:
1011         sb->s_fs_info = NULL;
1012         kfree(sbi);
1013         return err;
1014 }
1015
1016 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1017 {
1018         struct nilfs_sb_info *sbi = NILFS_SB(sb);
1019         struct the_nilfs *nilfs = sbi->s_nilfs;
1020         unsigned long old_sb_flags;
1021         unsigned long old_mount_opt;
1022         int err;
1023
1024         old_sb_flags = sb->s_flags;
1025         old_mount_opt = nilfs->ns_mount_opt;
1026
1027         if (!parse_options(data, sb, 1)) {
1028                 err = -EINVAL;
1029                 goto restore_opts;
1030         }
1031         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1032
1033         err = -EINVAL;
1034
1035         if (!nilfs_valid_fs(nilfs)) {
1036                 printk(KERN_WARNING "NILFS (device %s): couldn't "
1037                        "remount because the filesystem is in an "
1038                        "incomplete recovery state.\n", sb->s_id);
1039                 goto restore_opts;
1040         }
1041
1042         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1043                 goto out;
1044         if (*flags & MS_RDONLY) {
1045                 /* Shutting down the segment constructor */
1046                 nilfs_detach_segment_constructor(sbi);
1047                 sb->s_flags |= MS_RDONLY;
1048
1049                 /*
1050                  * Remounting a valid RW partition RDONLY, so set
1051                  * the RDONLY flag and then mark the partition as valid again.
1052                  */
1053                 down_write(&nilfs->ns_sem);
1054                 nilfs_cleanup_super(sbi);
1055                 up_write(&nilfs->ns_sem);
1056         } else {
1057                 __u64 features;
1058                 struct nilfs_root *root;
1059
1060                 /*
1061                  * Mounting a RDONLY partition read-write, so reread and
1062                  * store the current valid flag.  (It may have been changed
1063                  * by fsck since we originally mounted the partition.)
1064                  */
1065                 down_read(&nilfs->ns_sem);
1066                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1067                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1068                 up_read(&nilfs->ns_sem);
1069                 if (features) {
1070                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1071                                "remount RDWR because of unsupported optional "
1072                                "features (%llx)\n",
1073                                sb->s_id, (unsigned long long)features);
1074                         err = -EROFS;
1075                         goto restore_opts;
1076                 }
1077
1078                 sb->s_flags &= ~MS_RDONLY;
1079
1080                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1081                 err = nilfs_attach_segment_constructor(sbi, root);
1082                 if (err)
1083                         goto restore_opts;
1084
1085                 down_write(&nilfs->ns_sem);
1086                 nilfs_setup_super(sbi, true);
1087                 up_write(&nilfs->ns_sem);
1088         }
1089  out:
1090         return 0;
1091
1092  restore_opts:
1093         sb->s_flags = old_sb_flags;
1094         nilfs->ns_mount_opt = old_mount_opt;
1095         return err;
1096 }
1097
1098 struct nilfs_super_data {
1099         struct block_device *bdev;
1100         struct nilfs_sb_info *sbi;
1101         __u64 cno;
1102         int flags;
1103 };
1104
1105 /**
1106  * nilfs_identify - pre-read mount options needed to identify mount instance
1107  * @data: mount options
1108  * @sd: nilfs_super_data
1109  */
1110 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1111 {
1112         char *p, *options = data;
1113         substring_t args[MAX_OPT_ARGS];
1114         int token;
1115         int ret = 0;
1116
1117         do {
1118                 p = strsep(&options, ",");
1119                 if (p != NULL && *p) {
1120                         token = match_token(p, tokens, args);
1121                         if (token == Opt_snapshot) {
1122                                 if (!(sd->flags & MS_RDONLY)) {
1123                                         ret++;
1124                                 } else {
1125                                         sd->cno = simple_strtoull(args[0].from,
1126                                                                   NULL, 0);
1127                                         /*
1128                                          * No need to see the end pointer;
1129                                          * match_token() has done syntax
1130                                          * checking.
1131                                          */
1132                                         if (sd->cno == 0)
1133                                                 ret++;
1134                                 }
1135                         }
1136                         if (ret)
1137                                 printk(KERN_ERR
1138                                        "NILFS: invalid mount option: %s\n", p);
1139                 }
1140                 if (!options)
1141                         break;
1142                 BUG_ON(options == data);
1143                 *(options - 1) = ',';
1144         } while (!ret);
1145         return ret;
1146 }
1147
1148 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1149 {
1150         s->s_bdev = data;
1151         s->s_dev = s->s_bdev->bd_dev;
1152         return 0;
1153 }
1154
1155 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1156 {
1157         return (void *)s->s_bdev == data;
1158 }
1159
1160 static struct dentry *
1161 nilfs_mount(struct file_system_type *fs_type, int flags,
1162              const char *dev_name, void *data)
1163 {
1164         struct nilfs_super_data sd;
1165         struct super_block *s;
1166         fmode_t mode = FMODE_READ | FMODE_EXCL;
1167         struct dentry *root_dentry;
1168         int err, s_new = false;
1169
1170         if (!(flags & MS_RDONLY))
1171                 mode |= FMODE_WRITE;
1172
1173         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1174         if (IS_ERR(sd.bdev))
1175                 return ERR_CAST(sd.bdev);
1176
1177         sd.cno = 0;
1178         sd.flags = flags;
1179         if (nilfs_identify((char *)data, &sd)) {
1180                 err = -EINVAL;
1181                 goto failed;
1182         }
1183
1184         /*
1185          * once the super is inserted into the list by sget, s_umount
1186          * will protect the lockfs code from trying to start a snapshot
1187          * while we are mounting
1188          */
1189         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1190         if (sd.bdev->bd_fsfreeze_count > 0) {
1191                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1192                 err = -EBUSY;
1193                 goto failed;
1194         }
1195         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1196         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1197         if (IS_ERR(s)) {
1198                 err = PTR_ERR(s);
1199                 goto failed;
1200         }
1201
1202         if (!s->s_root) {
1203                 char b[BDEVNAME_SIZE];
1204
1205                 s_new = true;
1206
1207                 /* New superblock instance created */
1208                 s->s_flags = flags;
1209                 s->s_mode = mode;
1210                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1211                 sb_set_blocksize(s, block_size(sd.bdev));
1212
1213                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1214                 if (err)
1215                         goto failed_super;
1216
1217                 s->s_flags |= MS_ACTIVE;
1218         } else if (!sd.cno) {
1219                 int busy = false;
1220
1221                 if (nilfs_tree_was_touched(s->s_root)) {
1222                         busy = nilfs_try_to_shrink_tree(s->s_root);
1223                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1224                                 printk(KERN_ERR "NILFS: the device already "
1225                                        "has a %s mount.\n",
1226                                        (s->s_flags & MS_RDONLY) ?
1227                                        "read-only" : "read/write");
1228                                 err = -EBUSY;
1229                                 goto failed_super;
1230                         }
1231                 }
1232                 if (!busy) {
1233                         /*
1234                          * Try remount to setup mount states if the current
1235                          * tree is not mounted and only snapshots use this sb.
1236                          */
1237                         err = nilfs_remount(s, &flags, data);
1238                         if (err)
1239                                 goto failed_super;
1240                 }
1241         }
1242
1243         if (sd.cno) {
1244                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1245                 if (err)
1246                         goto failed_super;
1247         } else {
1248                 root_dentry = dget(s->s_root);
1249         }
1250
1251         if (!s_new)
1252                 blkdev_put(sd.bdev, mode);
1253
1254         return root_dentry;
1255
1256  failed_super:
1257         deactivate_locked_super(s);
1258
1259  failed:
1260         if (!s_new)
1261                 blkdev_put(sd.bdev, mode);
1262         return ERR_PTR(err);
1263 }
1264
1265 struct file_system_type nilfs_fs_type = {
1266         .owner    = THIS_MODULE,
1267         .name     = "nilfs2",
1268         .mount    = nilfs_mount,
1269         .kill_sb  = kill_block_super,
1270         .fs_flags = FS_REQUIRES_DEV,
1271 };
1272
1273 static void nilfs_inode_init_once(void *obj)
1274 {
1275         struct nilfs_inode_info *ii = obj;
1276
1277         INIT_LIST_HEAD(&ii->i_dirty);
1278 #ifdef CONFIG_NILFS_XATTR
1279         init_rwsem(&ii->xattr_sem);
1280 #endif
1281         address_space_init_once(&ii->i_btnode_cache);
1282         ii->i_bmap = &ii->i_bmap_data;
1283         inode_init_once(&ii->vfs_inode);
1284 }
1285
1286 static void nilfs_segbuf_init_once(void *obj)
1287 {
1288         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1289 }
1290
1291 static void nilfs_destroy_cachep(void)
1292 {
1293         if (nilfs_inode_cachep)
1294                 kmem_cache_destroy(nilfs_inode_cachep);
1295         if (nilfs_transaction_cachep)
1296                 kmem_cache_destroy(nilfs_transaction_cachep);
1297         if (nilfs_segbuf_cachep)
1298                 kmem_cache_destroy(nilfs_segbuf_cachep);
1299         if (nilfs_btree_path_cache)
1300                 kmem_cache_destroy(nilfs_btree_path_cache);
1301 }
1302
1303 static int __init nilfs_init_cachep(void)
1304 {
1305         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1306                         sizeof(struct nilfs_inode_info), 0,
1307                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1308         if (!nilfs_inode_cachep)
1309                 goto fail;
1310
1311         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1312                         sizeof(struct nilfs_transaction_info), 0,
1313                         SLAB_RECLAIM_ACCOUNT, NULL);
1314         if (!nilfs_transaction_cachep)
1315                 goto fail;
1316
1317         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1318                         sizeof(struct nilfs_segment_buffer), 0,
1319                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1320         if (!nilfs_segbuf_cachep)
1321                 goto fail;
1322
1323         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1324                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1325                         0, 0, NULL);
1326         if (!nilfs_btree_path_cache)
1327                 goto fail;
1328
1329         return 0;
1330
1331 fail:
1332         nilfs_destroy_cachep();
1333         return -ENOMEM;
1334 }
1335
1336 static int __init init_nilfs_fs(void)
1337 {
1338         int err;
1339
1340         err = nilfs_init_cachep();
1341         if (err)
1342                 goto fail;
1343
1344         err = register_filesystem(&nilfs_fs_type);
1345         if (err)
1346                 goto free_cachep;
1347
1348         printk(KERN_INFO "NILFS version 2 loaded\n");
1349         return 0;
1350
1351 free_cachep:
1352         nilfs_destroy_cachep();
1353 fail:
1354         return err;
1355 }
1356
1357 static void __exit exit_nilfs_fs(void)
1358 {
1359         nilfs_destroy_cachep();
1360         unregister_filesystem(&nilfs_fs_type);
1361 }
1362
1363 module_init(init_nilfs_fs)
1364 module_exit(exit_nilfs_fs)