Pileus Git - ~andy/linux/blob - fs/nilfs2/super.c

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