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