SAFE public projects git trees. - safe/jmp/linux-2.6/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_mdt_new(nilfs, sbi->s_super, NILFS_IFILE_INO, 0);
 367         if (!sbi->s_ifile)
 368                 return -ENOMEM;
 369
 370         err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
 371         if (unlikely(err))
 372                 goto failed;
 373
 374         down_read(&nilfs->ns_segctor_sem);
 375         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 376                                           &bh_cp);
 377         up_read(&nilfs->ns_segctor_sem);
 378         if (unlikely(err)) {
 379                 if (err == -ENOENT || err == -EINVAL) {
 380                         printk(KERN_ERR
 381                                "NILFS: Invalid checkpoint "
 382                                "(checkpoint number=%llu)\n",
 383                                (unsigned long long)cno);
 384                         err = -EINVAL;
 385                 }
 386                 goto failed;
 387         }
 388         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
 389         if (unlikely(err))
 390                 goto failed_bh;
 391         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 392         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 393
 394         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 395         return 0;
 396
 397  failed_bh:
 398         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 399  failed:
 400         nilfs_mdt_destroy(sbi->s_ifile);
 401         sbi->s_ifile = NULL;
 402
 403         down_write(&nilfs->ns_super_sem);
 404         list_del_init(&sbi->s_list);
 405         up_write(&nilfs->ns_super_sem);
 406
 407         return err;
 408 }
 409
 410 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
 411 {
 412         struct the_nilfs *nilfs = sbi->s_nilfs;
 413
 414         nilfs_mdt_clear(sbi->s_ifile);
 415         nilfs_mdt_destroy(sbi->s_ifile);
 416         sbi->s_ifile = NULL;
 417         down_write(&nilfs->ns_super_sem);
 418         list_del_init(&sbi->s_list);
 419         up_write(&nilfs->ns_super_sem);
 420 }
 421
 422 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
 423 {
 424         struct the_nilfs *nilfs = sbi->s_nilfs;
 425         int err = 0;
 426
 427         down_write(&nilfs->ns_sem);
 428         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
 429                 nilfs->ns_mount_state |= NILFS_VALID_FS;
 430                 err = nilfs_commit_super(sbi, 1);
 431                 if (likely(!err))
 432                         printk(KERN_INFO "NILFS: recovery complete.\n");
 433         }
 434         up_write(&nilfs->ns_sem);
 435         return err;
 436 }
 437
 438 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 439 {
 440         struct super_block *sb = dentry->d_sb;
 441         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 442         struct the_nilfs *nilfs = sbi->s_nilfs;
 443         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 444         unsigned long long blocks;
 445         unsigned long overhead;
 446         unsigned long nrsvblocks;
 447         sector_t nfreeblocks;
 448         int err;
 449
 450         /*
 451          * Compute all of the segment blocks
 452          *
 453          * The blocks before first segment and after last segment
 454          * are excluded.
 455          */
 456         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 457                 - nilfs->ns_first_data_block;
 458         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 459
 460         /*
 461          * Compute the overhead
 462          *
 463          * When distributing meta data blocks outside semgent structure,
 464          * We must count them as the overhead.
 465          */
 466         overhead = 0;
 467
 468         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 469         if (unlikely(err))
 470                 return err;
 471
 472         buf->f_type = NILFS_SUPER_MAGIC;
 473         buf->f_bsize = sb->s_blocksize;
 474         buf->f_blocks = blocks - overhead;
 475         buf->f_bfree = nfreeblocks;
 476         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 477                 (buf->f_bfree - nrsvblocks) : 0;
 478         buf->f_files = atomic_read(&sbi->s_inodes_count);
 479         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 480         buf->f_namelen = NILFS_NAME_LEN;
 481         buf->f_fsid.val[0] = (u32)id;
 482         buf->f_fsid.val[1] = (u32)(id >> 32);
 483
 484         return 0;
 485 }
 486
 487 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 488 {
 489         struct super_block *sb = vfs->mnt_sb;
 490         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 491
 492         if (!nilfs_test_opt(sbi, BARRIER))
 493                 seq_printf(seq, ",nobarrier");
 494         if (nilfs_test_opt(sbi, SNAPSHOT))
 495                 seq_printf(seq, ",cp=%llu",
 496                            (unsigned long long int)sbi->s_snapshot_cno);
 497         if (nilfs_test_opt(sbi, ERRORS_RO))
 498                 seq_printf(seq, ",errors=remount-ro");
 499         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 500                 seq_printf(seq, ",errors=panic");
 501         if (nilfs_test_opt(sbi, STRICT_ORDER))
 502                 seq_printf(seq, ",order=strict");
 503
 504         return 0;
 505 }
 506
 507 static const struct super_operations nilfs_sops = {
 508         .alloc_inode    = nilfs_alloc_inode,
 509         .destroy_inode  = nilfs_destroy_inode,
 510         .dirty_inode    = nilfs_dirty_inode,
 511         /* .write_inode    = nilfs_write_inode, */
 512         /* .put_inode      = nilfs_put_inode, */
 513         /* .drop_inode    = nilfs_drop_inode, */
 514         .delete_inode   = nilfs_delete_inode,
 515         .put_super      = nilfs_put_super,
 516         /* .write_super    = nilfs_write_super, */
 517         .sync_fs        = nilfs_sync_fs,
 518         /* .write_super_lockfs */
 519         /* .unlockfs */
 520         .statfs         = nilfs_statfs,
 521         .remount_fs     = nilfs_remount,
 522         .clear_inode    = nilfs_clear_inode,
 523         /* .umount_begin */
 524         .show_options = nilfs_show_options
 525 };
 526
 527 static struct inode *
 528 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
 529 {
 530         struct inode *inode;
 531
 532         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
 533             ino != NILFS_SKETCH_INO)
 534                 return ERR_PTR(-ESTALE);
 535
 536         inode = nilfs_iget(sb, ino);
 537         if (IS_ERR(inode))
 538                 return ERR_CAST(inode);
 539         if (generation && inode->i_generation != generation) {
 540                 iput(inode);
 541                 return ERR_PTR(-ESTALE);
 542         }
 543
 544         return inode;
 545 }
 546
 547 static struct dentry *
 548 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
 549                    int fh_type)
 550 {
 551         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 552                                     nilfs_nfs_get_inode);
 553 }
 554
 555 static struct dentry *
 556 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
 557                    int fh_type)
 558 {
 559         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 560                                     nilfs_nfs_get_inode);
 561 }
 562
 563 static const struct export_operations nilfs_export_ops = {
 564         .fh_to_dentry = nilfs_fh_to_dentry,
 565         .fh_to_parent = nilfs_fh_to_parent,
 566         .get_parent = nilfs_get_parent,
 567 };
 568
 569 enum {
 570         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 571         Opt_nobarrier, Opt_snapshot, Opt_order,
 572         Opt_err,
 573 };
 574
 575 static match_table_t tokens = {
 576         {Opt_err_cont, "errors=continue"},
 577         {Opt_err_panic, "errors=panic"},
 578         {Opt_err_ro, "errors=remount-ro"},
 579         {Opt_nobarrier, "nobarrier"},
 580         {Opt_snapshot, "cp=%u"},
 581         {Opt_order, "order=%s"},
 582         {Opt_err, NULL}
 583 };
 584
 585 static int parse_options(char *options, struct super_block *sb)
 586 {
 587         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 588         char *p;
 589         substring_t args[MAX_OPT_ARGS];
 590         int option;
 591
 592         if (!options)
 593                 return 1;
 594
 595         while ((p = strsep(&options, ",")) != NULL) {
 596                 int token;
 597                 if (!*p)
 598                         continue;
 599
 600                 token = match_token(p, tokens, args);
 601                 switch (token) {
 602                 case Opt_nobarrier:
 603                         nilfs_clear_opt(sbi, BARRIER);
 604                         break;
 605                 case Opt_order:
 606                         if (strcmp(args[0].from, "relaxed") == 0)
 607                                 /* Ordered data semantics */
 608                                 nilfs_clear_opt(sbi, STRICT_ORDER);
 609                         else if (strcmp(args[0].from, "strict") == 0)
 610                                 /* Strict in-order semantics */
 611                                 nilfs_set_opt(sbi, STRICT_ORDER);
 612                         else
 613                                 return 0;
 614                         break;
 615                 case Opt_err_panic:
 616                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
 617                         break;
 618                 case Opt_err_ro:
 619                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
 620                         break;
 621                 case Opt_err_cont:
 622                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
 623                         break;
 624                 case Opt_snapshot:
 625                         if (match_int(&args[0], &option) || option <= 0)
 626                                 return 0;
 627                         if (!(sb->s_flags & MS_RDONLY))
 628                                 return 0;
 629                         sbi->s_snapshot_cno = option;
 630                         nilfs_set_opt(sbi, SNAPSHOT);
 631                         break;
 632                 default:
 633                         printk(KERN_ERR
 634                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 635                         return 0;
 636                 }
 637         }
 638         return 1;
 639 }
 640
 641 static inline void
 642 nilfs_set_default_options(struct nilfs_sb_info *sbi,
 643                           struct nilfs_super_block *sbp)
 644 {
 645         sbi->s_mount_opt =
 646                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
 647 }
 648
 649 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
 650 {
 651         struct the_nilfs *nilfs = sbi->s_nilfs;
 652         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 653         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
 654         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
 655
 656         /* nilfs->sem must be locked by the caller. */
 657         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
 658                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 659         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 660                 printk(KERN_WARNING
 661                        "NILFS warning: mounting fs with errors\n");
 662 #if 0
 663         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 664                 printk(KERN_WARNING
 665                        "NILFS warning: maximal mount count reached\n");
 666 #endif
 667         }
 668         if (!max_mnt_count)
 669                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 670
 671         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
 672         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
 673         sbp->s_mtime = cpu_to_le64(get_seconds());
 674         return nilfs_commit_super(sbi, 1);
 675 }
 676
 677 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 678                                                  u64 pos, int blocksize,
 679                                                  struct buffer_head **pbh)
 680 {
 681         unsigned long long sb_index = pos;
 682         unsigned long offset;
 683
 684         offset = do_div(sb_index, blocksize);
 685         *pbh = sb_bread(sb, sb_index);
 686         if (!*pbh)
 687                 return NULL;
 688         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 689 }
 690
 691 int nilfs_store_magic_and_option(struct super_block *sb,
 692                                  struct nilfs_super_block *sbp,
 693                                  char *data)
 694 {
 695         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 696
 697         sb->s_magic = le16_to_cpu(sbp->s_magic);
 698
 699         /* FS independent flags */
 700 #ifdef NILFS_ATIME_DISABLE
 701         sb->s_flags |= MS_NOATIME;
 702 #endif
 703
 704         nilfs_set_default_options(sbi, sbp);
 705
 706         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
 707         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
 708         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
 709         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
 710
 711         return !parse_options(data, sb) ? -EINVAL : 0 ;
 712 }
 713
 714 /**
 715  * nilfs_fill_super() - initialize a super block instance
 716  * @sb: super_block
 717  * @data: mount options
 718  * @silent: silent mode flag
 719  * @nilfs: the_nilfs struct
 720  *
 721  * This function is called exclusively by nilfs->ns_mount_mutex.
 722  * So, the recovery process is protected from other simultaneous mounts.
 723  */
 724 static int
 725 nilfs_fill_super(struct super_block *sb, void *data, int silent,
 726                  struct the_nilfs *nilfs)
 727 {
 728         struct nilfs_sb_info *sbi;
 729         struct inode *root;
 730         __u64 cno;
 731         int err;
 732
 733         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 734         if (!sbi)
 735                 return -ENOMEM;
 736
 737         sb->s_fs_info = sbi;
 738
 739         get_nilfs(nilfs);
 740         sbi->s_nilfs = nilfs;
 741         sbi->s_super = sb;
 742         atomic_set(&sbi->s_count, 1);
 743
 744         err = init_nilfs(nilfs, sbi, (char *)data);
 745         if (err)
 746                 goto failed_sbi;
 747
 748         spin_lock_init(&sbi->s_inode_lock);
 749         INIT_LIST_HEAD(&sbi->s_dirty_files);
 750         INIT_LIST_HEAD(&sbi->s_list);
 751
 752         /*
 753          * Following initialization is overlapped because
 754          * nilfs_sb_info structure has been cleared at the beginning.
 755          * But we reserve them to keep our interest and make ready
 756          * for the future change.
 757          */
 758         get_random_bytes(&sbi->s_next_generation,
 759                          sizeof(sbi->s_next_generation));
 760         spin_lock_init(&sbi->s_next_gen_lock);
 761
 762         sb->s_op = &nilfs_sops;
 763         sb->s_export_op = &nilfs_export_ops;
 764         sb->s_root = NULL;
 765         sb->s_time_gran = 1;
 766
 767         if (!nilfs_loaded(nilfs)) {
 768                 err = load_nilfs(nilfs, sbi);
 769                 if (err)
 770                         goto failed_sbi;
 771         }
 772         cno = nilfs_last_cno(nilfs);
 773
 774         if (sb->s_flags & MS_RDONLY) {
 775                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
 776                         down_read(&nilfs->ns_segctor_sem);
 777                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
 778                                                        sbi->s_snapshot_cno);
 779                         up_read(&nilfs->ns_segctor_sem);
 780                         if (err < 0) {
 781                                 if (err == -ENOENT)
 782                                         err = -EINVAL;
 783                                 goto failed_sbi;
 784                         }
 785                         if (!err) {
 786                                 printk(KERN_ERR
 787                                        "NILFS: The specified checkpoint is "
 788                                        "not a snapshot "
 789                                        "(checkpoint number=%llu).\n",
 790                                        (unsigned long long)sbi->s_snapshot_cno);
 791                                 err = -EINVAL;
 792                                 goto failed_sbi;
 793                         }
 794                         cno = sbi->s_snapshot_cno;
 795                 } else
 796                         /* Read-only mount */
 797                         sbi->s_snapshot_cno = cno;
 798         }
 799
 800         err = nilfs_attach_checkpoint(sbi, cno);
 801         if (err) {
 802                 printk(KERN_ERR "NILFS: error loading a checkpoint"
 803                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
 804                 goto failed_sbi;
 805         }
 806
 807         if (!(sb->s_flags & MS_RDONLY)) {
 808                 err = nilfs_attach_segment_constructor(sbi);
 809                 if (err)
 810                         goto failed_checkpoint;
 811         }
 812
 813         root = nilfs_iget(sb, NILFS_ROOT_INO);
 814         if (IS_ERR(root)) {
 815                 printk(KERN_ERR "NILFS: get root inode failed\n");
 816                 err = PTR_ERR(root);
 817                 goto failed_segctor;
 818         }
 819         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
 820                 iput(root);
 821                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 822                 err = -EINVAL;
 823                 goto failed_segctor;
 824         }
 825         sb->s_root = d_alloc_root(root);
 826         if (!sb->s_root) {
 827                 iput(root);
 828                 printk(KERN_ERR "NILFS: get root dentry failed\n");
 829                 err = -ENOMEM;
 830                 goto failed_segctor;
 831         }
 832
 833         if (!(sb->s_flags & MS_RDONLY)) {
 834                 down_write(&nilfs->ns_sem);
 835                 nilfs_setup_super(sbi);
 836                 up_write(&nilfs->ns_sem);
 837         }
 838
 839         err = nilfs_mark_recovery_complete(sbi);
 840         if (unlikely(err)) {
 841                 printk(KERN_ERR "NILFS: recovery failed.\n");
 842                 goto failed_root;
 843         }
 844
 845         down_write(&nilfs->ns_super_sem);
 846         if (!nilfs_test_opt(sbi, SNAPSHOT))
 847                 nilfs->ns_current = sbi;
 848         up_write(&nilfs->ns_super_sem);
 849
 850         return 0;
 851
 852  failed_root:
 853         dput(sb->s_root);
 854         sb->s_root = NULL;
 855
 856  failed_segctor:
 857         nilfs_detach_segment_constructor(sbi);
 858
 859  failed_checkpoint:
 860         nilfs_detach_checkpoint(sbi);
 861
 862  failed_sbi:
 863         put_nilfs(nilfs);
 864         sb->s_fs_info = NULL;
 865         nilfs_put_sbinfo(sbi);
 866         return err;
 867 }
 868
 869 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
 870 {
 871         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 872         struct nilfs_super_block *sbp;
 873         struct the_nilfs *nilfs = sbi->s_nilfs;
 874         unsigned long old_sb_flags;
 875         struct nilfs_mount_options old_opts;
 876         int err;
 877
 878         lock_kernel();
 879
 880         down_write(&nilfs->ns_super_sem);
 881         old_sb_flags = sb->s_flags;
 882         old_opts.mount_opt = sbi->s_mount_opt;
 883         old_opts.snapshot_cno = sbi->s_snapshot_cno;
 884
 885         if (!parse_options(data, sb)) {
 886                 err = -EINVAL;
 887                 goto restore_opts;
 888         }
 889         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
 890
 891         if ((*flags & MS_RDONLY) &&
 892             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
 893                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 894                        "remount to a different snapshot. \n",
 895                        sb->s_id);
 896                 err = -EINVAL;
 897                 goto restore_opts;
 898         }
 899
 900         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 901                 goto out;
 902         if (*flags & MS_RDONLY) {
 903                 /* Shutting down the segment constructor */
 904                 nilfs_detach_segment_constructor(sbi);
 905                 sb->s_flags |= MS_RDONLY;
 906
 907                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
 908                 /* nilfs_set_opt(sbi, SNAPSHOT); */
 909
 910                 /*
 911                  * Remounting a valid RW partition RDONLY, so set
 912                  * the RDONLY flag and then mark the partition as valid again.
 913                  */
 914                 down_write(&nilfs->ns_sem);
 915                 sbp = nilfs->ns_sbp[0];
 916                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
 917                     (nilfs->ns_mount_state & NILFS_VALID_FS))
 918                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
 919                 sbp->s_mtime = cpu_to_le64(get_seconds());
 920                 nilfs_commit_super(sbi, 1);
 921                 up_write(&nilfs->ns_sem);
 922         } else {
 923                 /*
 924                  * Mounting a RDONLY partition read-write, so reread and
 925                  * store the current valid flag.  (It may have been changed
 926                  * by fsck since we originally mounted the partition.)
 927                  */
 928                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
 929                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 930                                "remount because an RW-mount exists.\n",
 931                                sb->s_id);
 932                         err = -EBUSY;
 933                         goto restore_opts;
 934                 }
 935                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
 936                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 937                                "remount because the current RO-mount is not "
 938                                "the latest one.\n",
 939                                sb->s_id);
 940                         err = -EINVAL;
 941                         goto restore_opts;
 942                 }
 943                 sb->s_flags &= ~MS_RDONLY;
 944                 nilfs_clear_opt(sbi, SNAPSHOT);
 945                 sbi->s_snapshot_cno = 0;
 946
 947                 err = nilfs_attach_segment_constructor(sbi);
 948                 if (err)
 949                         goto restore_opts;
 950
 951                 down_write(&nilfs->ns_sem);
 952                 nilfs_setup_super(sbi);
 953                 up_write(&nilfs->ns_sem);
 954
 955                 nilfs->ns_current = sbi;
 956         }
 957  out:
 958         up_write(&nilfs->ns_super_sem);
 959         unlock_kernel();
 960         return 0;
 961
 962  restore_opts:
 963         sb->s_flags = old_sb_flags;
 964         sbi->s_mount_opt = old_opts.mount_opt;
 965         sbi->s_snapshot_cno = old_opts.snapshot_cno;
 966         up_write(&nilfs->ns_super_sem);
 967         unlock_kernel();
 968         return err;
 969 }
 970
 971 struct nilfs_super_data {
 972         struct block_device *bdev;
 973         struct nilfs_sb_info *sbi;
 974         __u64 cno;
 975         int flags;
 976 };
 977
 978 /**
 979  * nilfs_identify - pre-read mount options needed to identify mount instance
 980  * @data: mount options
 981  * @sd: nilfs_super_data
 982  */
 983 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
 984 {
 985         char *p, *options = data;
 986         substring_t args[MAX_OPT_ARGS];
 987         int option, token;
 988         int ret = 0;
 989
 990         do {
 991                 p = strsep(&options, ",");
 992                 if (p != NULL && *p) {
 993                         token = match_token(p, tokens, args);
 994                         if (token == Opt_snapshot) {
 995                                 if (!(sd->flags & MS_RDONLY))
 996                                         ret++;
 997                                 else {
 998                                         ret = match_int(&args[0], &option);
 999                                         if (!ret) {
1000                                                 if (option > 0)
1001                                                         sd->cno = option;
1002                                                 else
1003                                                         ret++;
1004                                         }
1005                                 }
1006                         }
1007                         if (ret)
1008                                 printk(KERN_ERR
1009                                        "NILFS: invalid mount option: %s\n", p);
1010                 }
1011                 if (!options)
1012                         break;
1013                 BUG_ON(options == data);
1014                 *(options - 1) = ',';
1015         } while (!ret);
1016         return ret;
1017 }
1018
1019 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1020 {
1021         struct nilfs_super_data *sd = data;
1022
1023         s->s_bdev = sd->bdev;
1024         s->s_dev = s->s_bdev->bd_dev;
1025         return 0;
1026 }
1027
1028 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1029 {
1030         struct nilfs_super_data *sd = data;
1031
1032         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1033 }
1034
1035 static int
1036 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1037              const char *dev_name, void *data, struct vfsmount *mnt)
1038 {
1039         struct nilfs_super_data sd;
1040         struct super_block *s;
1041         struct the_nilfs *nilfs;
1042         int err, need_to_close = 1;
1043
1044         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1045         if (IS_ERR(sd.bdev))
1046                 return PTR_ERR(sd.bdev);
1047
1048         /*
1049          * To get mount instance using sget() vfs-routine, NILFS needs
1050          * much more information than normal filesystems to identify mount
1051          * instance.  For snapshot mounts, not only a mount type (ro-mount
1052          * or rw-mount) but also a checkpoint number is required.
1053          */
1054         sd.cno = 0;
1055         sd.flags = flags;
1056         if (nilfs_identify((char *)data, &sd)) {
1057                 err = -EINVAL;
1058                 goto failed;
1059         }
1060
1061         nilfs = find_or_create_nilfs(sd.bdev);
1062         if (!nilfs) {
1063                 err = -ENOMEM;
1064                 goto failed;
1065         }
1066
1067         mutex_lock(&nilfs->ns_mount_mutex);
1068
1069         if (!sd.cno) {
1070                 /*
1071                  * Check if an exclusive mount exists or not.
1072                  * Snapshot mounts coexist with a current mount
1073                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1074                  * ro-mount are mutually exclusive.
1075                  */
1076                 down_read(&nilfs->ns_super_sem);
1077                 if (nilfs->ns_current &&
1078                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1079                      & MS_RDONLY)) {
1080                         up_read(&nilfs->ns_super_sem);
1081                         err = -EBUSY;
1082                         goto failed_unlock;
1083                 }
1084                 up_read(&nilfs->ns_super_sem);
1085         }
1086
1087         /*
1088          * Find existing nilfs_sb_info struct
1089          */
1090         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1091
1092         /*
1093          * Get super block instance holding the nilfs_sb_info struct.
1094          * A new instance is allocated if no existing mount is present or
1095          * existing instance has been unmounted.
1096          */
1097         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1098         if (sd.sbi)
1099                 nilfs_put_sbinfo(sd.sbi);
1100
1101         if (IS_ERR(s)) {
1102                 err = PTR_ERR(s);
1103                 goto failed_unlock;
1104         }
1105
1106         if (!s->s_root) {
1107                 char b[BDEVNAME_SIZE];
1108
1109                 /* New superblock instance created */
1110                 s->s_flags = flags;
1111                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1112                 sb_set_blocksize(s, block_size(sd.bdev));
1113
1114                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1115                 if (err)
1116                         goto cancel_new;
1117
1118                 s->s_flags |= MS_ACTIVE;
1119                 need_to_close = 0;
1120         }
1121
1122         mutex_unlock(&nilfs->ns_mount_mutex);
1123         put_nilfs(nilfs);
1124         if (need_to_close)
1125                 close_bdev_exclusive(sd.bdev, flags);
1126         simple_set_mnt(mnt, s);
1127         return 0;
1128
1129  failed_unlock:
1130         mutex_unlock(&nilfs->ns_mount_mutex);
1131         put_nilfs(nilfs);
1132  failed:
1133         close_bdev_exclusive(sd.bdev, flags);
1134
1135         return err;
1136
1137  cancel_new:
1138         /* Abandoning the newly allocated superblock */
1139         mutex_unlock(&nilfs->ns_mount_mutex);
1140         put_nilfs(nilfs);
1141         up_write(&s->s_umount);
1142         deactivate_super(s);
1143         /*
1144          * deactivate_super() invokes close_bdev_exclusive().
1145          * We must finish all post-cleaning before this call;
1146          * put_nilfs() needs the block device.
1147          */
1148         return err;
1149 }
1150
1151 struct file_system_type nilfs_fs_type = {
1152         .owner    = THIS_MODULE,
1153         .name     = "nilfs2",
1154         .get_sb   = nilfs_get_sb,
1155         .kill_sb  = kill_block_super,
1156         .fs_flags = FS_REQUIRES_DEV,
1157 };
1158
1159 static int __init init_nilfs_fs(void)
1160 {
1161         int err;
1162
1163         err = nilfs_init_inode_cache();
1164         if (err)
1165                 goto failed;
1166
1167         err = nilfs_init_transaction_cache();
1168         if (err)
1169                 goto failed_inode_cache;
1170
1171         err = nilfs_init_segbuf_cache();
1172         if (err)
1173                 goto failed_transaction_cache;
1174
1175         err = nilfs_btree_path_cache_init();
1176         if (err)
1177                 goto failed_segbuf_cache;
1178
1179         err = register_filesystem(&nilfs_fs_type);
1180         if (err)
1181                 goto failed_btree_path_cache;
1182
1183         return 0;
1184
1185  failed_btree_path_cache:
1186         nilfs_btree_path_cache_destroy();
1187
1188  failed_segbuf_cache:
1189         nilfs_destroy_segbuf_cache();
1190
1191  failed_transaction_cache:
1192         nilfs_destroy_transaction_cache();
1193
1194  failed_inode_cache:
1195         nilfs_destroy_inode_cache();
1196
1197  failed:
1198         return err;
1199 }
1200
1201 static void __exit exit_nilfs_fs(void)
1202 {
1203         nilfs_destroy_segbuf_cache();
1204         nilfs_destroy_transaction_cache();
1205         nilfs_destroy_inode_cache();
1206         nilfs_btree_path_cache_destroy();
1207         unregister_filesystem(&nilfs_fs_type);
1208 }
1209
1210 module_init(init_nilfs_fs)
1211 module_exit(exit_nilfs_fs)