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