fs/nilfs2/super.c

   1 /*
   2  * super.c - NILFS module and super block management.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  */
  22 /*
  23  *  linux/fs/ext2/super.c
  24  *
  25  * Copyright (C) 1992, 1993, 1994, 1995
  26  * Remy Card (card@masi.ibp.fr)
  27  * Laboratoire MASI - Institut Blaise Pascal
  28  * Universite Pierre et Marie Curie (Paris VI)
  29  *
  30  *  from
  31  *
  32  *  linux/fs/minix/inode.c
  33  *
  34  *  Copyright (C) 1991, 1992  Linus Torvalds
  35  *
  36  *  Big-endian to little-endian byte-swapping/bitmaps by
  37  *        David S. Miller (davem@caip.rutgers.edu), 1995
  38  */
  39
  40 #include <linux/module.h>
  41 #include <linux/string.h>
  42 #include <linux/slab.h>
  43 #include <linux/init.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/parser.h>
  46 #include <linux/random.h>
  47 #include <linux/crc32.h>
  48 #include <linux/smp_lock.h>
  49 #include <linux/vfs.h>
  50 #include <linux/writeback.h>
  51 #include <linux/kobject.h>
  52 #include <linux/exportfs.h>
  53 #include <linux/seq_file.h>
  54 #include <linux/mount.h>
  55 #include "nilfs.h"
  56 #include "mdt.h"
  57 #include "alloc.h"
  58 #include "page.h"
  59 #include "cpfile.h"
  60 #include "ifile.h"
  61 #include "dat.h"
  62 #include "segment.h"
  63 #include "segbuf.h"
  64
  65 MODULE_AUTHOR("NTT Corp.");
  66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  67                    "(NILFS)");
  68 MODULE_LICENSE("GPL");
  69
  70 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  71
  72 /**
  73  * nilfs_error() - report failure condition on a filesystem
  74  *
  75  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
  76  * reporting an error message.  It should be called when NILFS detects
  77  * incoherences or defects of meta data on disk.  As for sustainable
  78  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
  79  * function should be used instead.
  80  *
  81  * The segment constructor must not call this function because it can
  82  * kill itself.
  83  */
  84 void nilfs_error(struct super_block *sb, const char *function,
  85                  const char *fmt, ...)
  86 {
  87         struct nilfs_sb_info *sbi = NILFS_SB(sb);
  88         va_list args;
  89
  90         va_start(args, fmt);
  91         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
  92         vprintk(fmt, args);
  93         printk("\n");
  94         va_end(args);
  95
  96         if (!(sb->s_flags & MS_RDONLY)) {
  97                 struct the_nilfs *nilfs = sbi->s_nilfs;
  98
  99                 if (!nilfs_test_opt(sbi, ERRORS_CONT))
 100                         nilfs_detach_segment_constructor(sbi);
 101
 102                 down_write(&nilfs->ns_sem);
 103                 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
 104                         nilfs->ns_mount_state |= NILFS_ERROR_FS;
 105                         nilfs->ns_sbp[0]->s_state |=
 106                                 cpu_to_le16(NILFS_ERROR_FS);
 107                         nilfs_commit_super(sbi, 1);
 108                 }
 109                 up_write(&nilfs->ns_sem);
 110
 111                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
 112                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 113                         sb->s_flags |= MS_RDONLY;
 114                 }
 115         }
 116
 117         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 118                 panic("NILFS (device %s): panic forced after error\n",
 119                       sb->s_id);
 120 }
 121
 122 void nilfs_warning(struct super_block *sb, const char *function,
 123                    const char *fmt, ...)
 124 {
 125         va_list args;
 126
 127         va_start(args, fmt);
 128         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
 129                sb->s_id, function);
 130         vprintk(fmt, args);
 131         printk("\n");
 132         va_end(args);
 133 }
 134
 135 static struct kmem_cache *nilfs_inode_cachep;
 136
 137 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
 138 {
 139         struct nilfs_inode_info *ii;
 140
 141         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 142         if (!ii)
 143                 return NULL;
 144         ii->i_bh = NULL;
 145         ii->i_state = 0;
 146         ii->vfs_inode.i_version = 1;
 147         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
 148         return &ii->vfs_inode;
 149 }
 150
 151 struct inode *nilfs_alloc_inode(struct super_block *sb)
 152 {
 153         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
 154 }
 155
 156 void nilfs_destroy_inode(struct inode *inode)
 157 {
 158         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 159 }
 160
 161 static void init_once(void *obj)
 162 {
 163         struct nilfs_inode_info *ii = obj;
 164
 165         INIT_LIST_HEAD(&ii->i_dirty);
 166 #ifdef CONFIG_NILFS_XATTR
 167         init_rwsem(&ii->xattr_sem);
 168 #endif
 169         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
 170         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
 171         inode_init_once(&ii->vfs_inode);
 172 }
 173
 174 static int nilfs_init_inode_cache(void)
 175 {
 176         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
 177                                                sizeof(struct nilfs_inode_info),
 178                                                0, SLAB_RECLAIM_ACCOUNT,
 179                                                init_once);
 180
 181         return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
 182 }
 183
 184 static inline void nilfs_destroy_inode_cache(void)
 185 {
 186         kmem_cache_destroy(nilfs_inode_cachep);
 187 }
 188
 189 static void nilfs_clear_inode(struct inode *inode)
 190 {
 191         struct nilfs_inode_info *ii = NILFS_I(inode);
 192
 193         /*
 194          * Free resources allocated in nilfs_read_inode(), here.
 195          */
 196         BUG_ON(!list_empty(&ii->i_dirty));
 197         brelse(ii->i_bh);
 198         ii->i_bh = NULL;
 199
 200         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 201                 nilfs_bmap_clear(ii->i_bmap);
 202
 203         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 204 }
 205
 206 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
 207 {
 208         struct the_nilfs *nilfs = sbi->s_nilfs;
 209         int err;
 210         int barrier_done = 0;
 211
 212         if (nilfs_test_opt(sbi, BARRIER)) {
 213                 set_buffer_ordered(nilfs->ns_sbh[0]);
 214                 barrier_done = 1;
 215         }
 216  retry:
 217         set_buffer_dirty(nilfs->ns_sbh[0]);
 218         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 219         if (err == -EOPNOTSUPP && barrier_done) {
 220                 nilfs_warning(sbi->s_super, __func__,
 221                               "barrier-based sync failed. "
 222                               "disabling barriers\n");
 223                 nilfs_clear_opt(sbi, BARRIER);
 224                 barrier_done = 0;
 225                 clear_buffer_ordered(nilfs->ns_sbh[0]);
 226                 goto retry;
 227         }
 228         if (unlikely(err)) {
 229                 printk(KERN_ERR
 230                        "NILFS: unable to write superblock (err=%d)\n", err);
 231                 if (err == -EIO && nilfs->ns_sbh[1]) {
 232                         nilfs_fall_back_super_block(nilfs);
 233                         goto retry;
 234                 }
 235         } else {
 236                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 237
 238                 /*
 239                  * The latest segment becomes trailable from the position
 240                  * written in superblock.
 241                  */
 242                 clear_nilfs_discontinued(nilfs);
 243
 244                 /* update GC protection for recent segments */
 245                 if (nilfs->ns_sbh[1]) {
 246                         sbp = NULL;
 247                         if (dupsb) {
 248                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 249                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
 250                                         sbp = nilfs->ns_sbp[1];
 251                         }
 252                 }
 253                 if (sbp) {
 254                         spin_lock(&nilfs->ns_last_segment_lock);
 255                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 256                         spin_unlock(&nilfs->ns_last_segment_lock);
 257                 }
 258         }
 259
 260         return err;
 261 }
 262
 263 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
 264 {
 265         struct the_nilfs *nilfs = sbi->s_nilfs;
 266         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 267         sector_t nfreeblocks;
 268         time_t t;
 269         int err;
 270
 271         /* nilfs->sem must be locked by the caller. */
 272         if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
 273                 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
 274                         nilfs_swap_super_block(nilfs);
 275                 else {
 276                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 277                                sbi->s_super->s_id);
 278                         return -EIO;
 279                 }
 280         }
 281         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 282         if (unlikely(err)) {
 283                 printk(KERN_ERR "NILFS: failed to count free blocks\n");
 284                 return err;
 285         }
 286         spin_lock(&nilfs->ns_last_segment_lock);
 287         sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 288         sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 289         sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 290         spin_unlock(&nilfs->ns_last_segment_lock);
 291
 292         t = get_seconds();
 293         nilfs->ns_sbwtime[0] = t;
 294         sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 295         sbp[0]->s_wtime = cpu_to_le64(t);
 296         sbp[0]->s_sum = 0;
 297         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 298                                              (unsigned char *)sbp[0],
 299                                              nilfs->ns_sbsize));
 300         if (dupsb && sbp[1]) {
 301                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 302                 nilfs->ns_sbwtime[1] = t;
 303         }
 304         sbi->s_super->s_dirt = 0;
 305         return nilfs_sync_super(sbi, dupsb);
 306 }
 307
 308 static void nilfs_put_super(struct super_block *sb)
 309 {
 310         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 311         struct the_nilfs *nilfs = sbi->s_nilfs;
 312
 313         lock_kernel();
 314
 315         nilfs_detach_segment_constructor(sbi);
 316
 317         if (!(sb->s_flags & MS_RDONLY)) {
 318                 down_write(&nilfs->ns_sem);
 319                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 320                 nilfs_commit_super(sbi, 1);
 321                 up_write(&nilfs->ns_sem);
 322         }
 323         down_write(&nilfs->ns_super_sem);
 324         if (nilfs->ns_current == sbi)
 325                 nilfs->ns_current = NULL;
 326         up_write(&nilfs->ns_super_sem);
 327
 328         nilfs_detach_checkpoint(sbi);
 329         put_nilfs(sbi->s_nilfs);
 330         sbi->s_super = NULL;
 331         sb->s_fs_info = NULL;
 332         nilfs_put_sbinfo(sbi);
 333
 334         unlock_kernel();
 335 }
 336
 337 static int nilfs_sync_fs(struct super_block *sb, int wait)
 338 {
 339         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 340         struct the_nilfs *nilfs = sbi->s_nilfs;
 341         int err = 0;
 342
 343         /* This function is called when super block should be written back */
 344         if (wait)
 345                 err = nilfs_construct_segment(sb);
 346
 347         down_write(&nilfs->ns_sem);
 348         if (sb->s_dirt)
 349                 nilfs_commit_super(sbi, 1);
 350         up_write(&nilfs->ns_sem);
 351
 352         return err;
 353 }
 354
 355 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
 356 {
 357         struct the_nilfs *nilfs = sbi->s_nilfs;
 358         struct nilfs_checkpoint *raw_cp;
 359         struct buffer_head *bh_cp;
 360         int err;
 361
 362         down_write(&nilfs->ns_super_sem);
 363         list_add(&sbi->s_list, &nilfs->ns_supers);
 364         up_write(&nilfs->ns_super_sem);
 365
 366         sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
 367         if (!sbi->s_ifile)
 368                 return -ENOMEM;
 369
 370         down_read(&nilfs->ns_segctor_sem);
 371         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 372                                           &bh_cp);
 373         up_read(&nilfs->ns_segctor_sem);
 374         if (unlikely(err)) {
 375                 if (err == -ENOENT || err == -EINVAL) {
 376                         printk(KERN_ERR
 377                                "NILFS: Invalid checkpoint "
 378                                "(checkpoint number=%llu)\n",
 379                                (unsigned long long)cno);
 380                         err = -EINVAL;
 381                 }
 382                 goto failed;
 383         }
 384         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
 385         if (unlikely(err))
 386                 goto failed_bh;
 387         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 388         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 389
 390         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 391         return 0;
 392
 393  failed_bh:
 394         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 395  failed:
 396         nilfs_mdt_destroy(sbi->s_ifile);
 397         sbi->s_ifile = NULL;
 398
 399         down_write(&nilfs->ns_super_sem);
 400         list_del_init(&sbi->s_list);
 401         up_write(&nilfs->ns_super_sem);
 402
 403         return err;
 404 }
 405
 406 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
 407 {
 408         struct the_nilfs *nilfs = sbi->s_nilfs;
 409
 410         nilfs_mdt_destroy(sbi->s_ifile);
 411         sbi->s_ifile = NULL;
 412         down_write(&nilfs->ns_super_sem);
 413         list_del_init(&sbi->s_list);
 414         up_write(&nilfs->ns_super_sem);
 415 }
 416
 417 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 418 {
 419         struct super_block *sb = dentry->d_sb;
 420         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 421         struct the_nilfs *nilfs = sbi->s_nilfs;
 422         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 423         unsigned long long blocks;
 424         unsigned long overhead;
 425         unsigned long nrsvblocks;
 426         sector_t nfreeblocks;
 427         int err;
 428
 429         /*
 430          * Compute all of the segment blocks
 431          *
 432          * The blocks before first segment and after last segment
 433          * are excluded.
 434          */
 435         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 436                 - nilfs->ns_first_data_block;
 437         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 438
 439         /*
 440          * Compute the overhead
 441          *
 442          * When distributing meta data blocks outside semgent structure,
 443          * We must count them as the overhead.
 444          */
 445         overhead = 0;
 446
 447         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 448         if (unlikely(err))
 449                 return err;
 450
 451         buf->f_type = NILFS_SUPER_MAGIC;
 452         buf->f_bsize = sb->s_blocksize;
 453         buf->f_blocks = blocks - overhead;
 454         buf->f_bfree = nfreeblocks;
 455         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 456                 (buf->f_bfree - nrsvblocks) : 0;
 457         buf->f_files = atomic_read(&sbi->s_inodes_count);
 458         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 459         buf->f_namelen = NILFS_NAME_LEN;
 460         buf->f_fsid.val[0] = (u32)id;
 461         buf->f_fsid.val[1] = (u32)(id >> 32);
 462
 463         return 0;
 464 }
 465
 466 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 467 {
 468         struct super_block *sb = vfs->mnt_sb;
 469         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 470
 471         if (!nilfs_test_opt(sbi, BARRIER))
 472                 seq_printf(seq, ",nobarrier");
 473         if (nilfs_test_opt(sbi, SNAPSHOT))
 474                 seq_printf(seq, ",cp=%llu",
 475                            (unsigned long long int)sbi->s_snapshot_cno);
 476         if (nilfs_test_opt(sbi, ERRORS_RO))
 477                 seq_printf(seq, ",errors=remount-ro");
 478         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 479                 seq_printf(seq, ",errors=panic");
 480         if (nilfs_test_opt(sbi, STRICT_ORDER))
 481                 seq_printf(seq, ",order=strict");
 482         if (nilfs_test_opt(sbi, NORECOVERY))
 483                 seq_printf(seq, ",norecovery");
 484         if (nilfs_test_opt(sbi, DISCARD))
 485                 seq_printf(seq, ",discard");
 486
 487         return 0;
 488 }
 489
 490 static const struct super_operations nilfs_sops = {
 491         .alloc_inode    = nilfs_alloc_inode,
 492         .destroy_inode  = nilfs_destroy_inode,
 493         .dirty_inode    = nilfs_dirty_inode,
 494         /* .write_inode    = nilfs_write_inode, */
 495         /* .put_inode      = nilfs_put_inode, */
 496         /* .drop_inode    = nilfs_drop_inode, */
 497         .delete_inode   = nilfs_delete_inode,
 498         .put_super      = nilfs_put_super,
 499         /* .write_super    = nilfs_write_super, */
 500         .sync_fs        = nilfs_sync_fs,
 501         /* .write_super_lockfs */
 502         /* .unlockfs */
 503         .statfs         = nilfs_statfs,
 504         .remount_fs     = nilfs_remount,
 505         .clear_inode    = nilfs_clear_inode,
 506         /* .umount_begin */
 507         .show_options = nilfs_show_options
 508 };
 509
 510 static struct inode *
 511 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
 512 {
 513         struct inode *inode;
 514
 515         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
 516             ino != NILFS_SKETCH_INO)
 517                 return ERR_PTR(-ESTALE);
 518
 519         inode = nilfs_iget(sb, ino);
 520         if (IS_ERR(inode))
 521                 return ERR_CAST(inode);
 522         if (generation && inode->i_generation != generation) {
 523                 iput(inode);
 524                 return ERR_PTR(-ESTALE);
 525         }
 526
 527         return inode;
 528 }
 529
 530 static struct dentry *
 531 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
 532                    int fh_type)
 533 {
 534         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 535                                     nilfs_nfs_get_inode);
 536 }
 537
 538 static struct dentry *
 539 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
 540                    int fh_type)
 541 {
 542         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 543                                     nilfs_nfs_get_inode);
 544 }
 545
 546 static const struct export_operations nilfs_export_ops = {
 547         .fh_to_dentry = nilfs_fh_to_dentry,
 548         .fh_to_parent = nilfs_fh_to_parent,
 549         .get_parent = nilfs_get_parent,
 550 };
 551
 552 enum {
 553         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 554         Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 555         Opt_discard, Opt_err,
 556 };
 557
 558 static match_table_t tokens = {
 559         {Opt_err_cont, "errors=continue"},
 560         {Opt_err_panic, "errors=panic"},
 561         {Opt_err_ro, "errors=remount-ro"},
 562         {Opt_nobarrier, "nobarrier"},
 563         {Opt_snapshot, "cp=%u"},
 564         {Opt_order, "order=%s"},
 565         {Opt_norecovery, "norecovery"},
 566         {Opt_discard, "discard"},
 567         {Opt_err, NULL}
 568 };
 569
 570 static int parse_options(char *options, struct super_block *sb)
 571 {
 572         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 573         char *p;
 574         substring_t args[MAX_OPT_ARGS];
 575         int option;
 576
 577         if (!options)
 578                 return 1;
 579
 580         while ((p = strsep(&options, ",")) != NULL) {
 581                 int token;
 582                 if (!*p)
 583                         continue;
 584
 585                 token = match_token(p, tokens, args);
 586                 switch (token) {
 587                 case Opt_nobarrier:
 588                         nilfs_clear_opt(sbi, BARRIER);
 589                         break;
 590                 case Opt_order:
 591                         if (strcmp(args[0].from, "relaxed") == 0)
 592                                 /* Ordered data semantics */
 593                                 nilfs_clear_opt(sbi, STRICT_ORDER);
 594                         else if (strcmp(args[0].from, "strict") == 0)
 595                                 /* Strict in-order semantics */
 596                                 nilfs_set_opt(sbi, STRICT_ORDER);
 597                         else
 598                                 return 0;
 599                         break;
 600                 case Opt_err_panic:
 601                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
 602                         break;
 603                 case Opt_err_ro:
 604                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
 605                         break;
 606                 case Opt_err_cont:
 607                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
 608                         break;
 609                 case Opt_snapshot:
 610                         if (match_int(&args[0], &option) || option <= 0)
 611                                 return 0;
 612                         if (!(sb->s_flags & MS_RDONLY))
 613                                 return 0;
 614                         sbi->s_snapshot_cno = option;
 615                         nilfs_set_opt(sbi, SNAPSHOT);
 616                         break;
 617                 case Opt_norecovery:
 618                         nilfs_set_opt(sbi, NORECOVERY);
 619                         break;
 620                 case Opt_discard:
 621                         nilfs_set_opt(sbi, DISCARD);
 622                         break;
 623                 default:
 624                         printk(KERN_ERR
 625                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 626                         return 0;
 627                 }
 628         }
 629         return 1;
 630 }
 631
 632 static inline void
 633 nilfs_set_default_options(struct nilfs_sb_info *sbi,
 634                           struct nilfs_super_block *sbp)
 635 {
 636         sbi->s_mount_opt =
 637                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
 638 }
 639
 640 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
 641 {
 642         struct the_nilfs *nilfs = sbi->s_nilfs;
 643         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 644         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
 645         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
 646
 647         /* nilfs->sem must be locked by the caller. */
 648         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 649                 printk(KERN_WARNING
 650                        "NILFS warning: mounting fs with errors\n");
 651 #if 0
 652         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 653                 printk(KERN_WARNING
 654                        "NILFS warning: maximal mount count reached\n");
 655 #endif
 656         }
 657         if (!max_mnt_count)
 658                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 659
 660         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
 661         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
 662         sbp->s_mtime = cpu_to_le64(get_seconds());
 663         return nilfs_commit_super(sbi, 1);
 664 }
 665
 666 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 667                                                  u64 pos, int blocksize,
 668                                                  struct buffer_head **pbh)
 669 {
 670         unsigned long long sb_index = pos;
 671         unsigned long offset;
 672
 673         offset = do_div(sb_index, blocksize);
 674         *pbh = sb_bread(sb, sb_index);
 675         if (!*pbh)
 676                 return NULL;
 677         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 678 }
 679
 680 int nilfs_store_magic_and_option(struct super_block *sb,
 681                                  struct nilfs_super_block *sbp,
 682                                  char *data)
 683 {
 684         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 685
 686         sb->s_magic = le16_to_cpu(sbp->s_magic);
 687
 688         /* FS independent flags */
 689 #ifdef NILFS_ATIME_DISABLE
 690         sb->s_flags |= MS_NOATIME;
 691 #endif
 692
 693         nilfs_set_default_options(sbi, sbp);
 694
 695         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
 696         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
 697         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
 698         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
 699
 700         return !parse_options(data, sb) ? -EINVAL : 0 ;
 701 }
 702
 703 /**
 704  * nilfs_fill_super() - initialize a super block instance
 705  * @sb: super_block
 706  * @data: mount options
 707  * @silent: silent mode flag
 708  * @nilfs: the_nilfs struct
 709  *
 710  * This function is called exclusively by nilfs->ns_mount_mutex.
 711  * So, the recovery process is protected from other simultaneous mounts.
 712  */
 713 static int
 714 nilfs_fill_super(struct super_block *sb, void *data, int silent,
 715                  struct the_nilfs *nilfs)
 716 {
 717         struct nilfs_sb_info *sbi;
 718         struct inode *root;
 719         __u64 cno;
 720         int err;
 721
 722         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 723         if (!sbi)
 724                 return -ENOMEM;
 725
 726         sb->s_fs_info = sbi;
 727
 728         get_nilfs(nilfs);
 729         sbi->s_nilfs = nilfs;
 730         sbi->s_super = sb;
 731         atomic_set(&sbi->s_count, 1);
 732
 733         err = init_nilfs(nilfs, sbi, (char *)data);
 734         if (err)
 735                 goto failed_sbi;
 736
 737         spin_lock_init(&sbi->s_inode_lock);
 738         INIT_LIST_HEAD(&sbi->s_dirty_files);
 739         INIT_LIST_HEAD(&sbi->s_list);
 740
 741         /*
 742          * Following initialization is overlapped because
 743          * nilfs_sb_info structure has been cleared at the beginning.
 744          * But we reserve them to keep our interest and make ready
 745          * for the future change.
 746          */
 747         get_random_bytes(&sbi->s_next_generation,
 748                          sizeof(sbi->s_next_generation));
 749         spin_lock_init(&sbi->s_next_gen_lock);
 750
 751         sb->s_op = &nilfs_sops;
 752         sb->s_export_op = &nilfs_export_ops;
 753         sb->s_root = NULL;
 754         sb->s_time_gran = 1;
 755
 756         err = load_nilfs(nilfs, sbi);
 757         if (err)
 758                 goto failed_sbi;
 759
 760         cno = nilfs_last_cno(nilfs);
 761
 762         if (sb->s_flags & MS_RDONLY) {
 763                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
 764                         down_read(&nilfs->ns_segctor_sem);
 765                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
 766                                                        sbi->s_snapshot_cno);
 767                         up_read(&nilfs->ns_segctor_sem);
 768                         if (err < 0) {
 769                                 if (err == -ENOENT)
 770                                         err = -EINVAL;
 771                                 goto failed_sbi;
 772                         }
 773                         if (!err) {
 774                                 printk(KERN_ERR
 775                                        "NILFS: The specified checkpoint is "
 776                                        "not a snapshot "
 777                                        "(checkpoint number=%llu).\n",
 778                                        (unsigned long long)sbi->s_snapshot_cno);
 779                                 err = -EINVAL;
 780                                 goto failed_sbi;
 781                         }
 782                         cno = sbi->s_snapshot_cno;
 783                 } else
 784                         /* Read-only mount */
 785                         sbi->s_snapshot_cno = cno;
 786         }
 787
 788         err = nilfs_attach_checkpoint(sbi, cno);
 789         if (err) {
 790                 printk(KERN_ERR "NILFS: error loading a checkpoint"
 791                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
 792                 goto failed_sbi;
 793         }
 794
 795         if (!(sb->s_flags & MS_RDONLY)) {
 796                 err = nilfs_attach_segment_constructor(sbi);
 797                 if (err)
 798                         goto failed_checkpoint;
 799         }
 800
 801         root = nilfs_iget(sb, NILFS_ROOT_INO);
 802         if (IS_ERR(root)) {
 803                 printk(KERN_ERR "NILFS: get root inode failed\n");
 804                 err = PTR_ERR(root);
 805                 goto failed_segctor;
 806         }
 807         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
 808                 iput(root);
 809                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 810                 err = -EINVAL;
 811                 goto failed_segctor;
 812         }
 813         sb->s_root = d_alloc_root(root);
 814         if (!sb->s_root) {
 815                 iput(root);
 816                 printk(KERN_ERR "NILFS: get root dentry failed\n");
 817                 err = -ENOMEM;
 818                 goto failed_segctor;
 819         }
 820
 821         if (!(sb->s_flags & MS_RDONLY)) {
 822                 down_write(&nilfs->ns_sem);
 823                 nilfs_setup_super(sbi);
 824                 up_write(&nilfs->ns_sem);
 825         }
 826
 827         down_write(&nilfs->ns_super_sem);
 828         if (!nilfs_test_opt(sbi, SNAPSHOT))
 829                 nilfs->ns_current = sbi;
 830         up_write(&nilfs->ns_super_sem);
 831
 832         return 0;
 833
 834  failed_segctor:
 835         nilfs_detach_segment_constructor(sbi);
 836
 837  failed_checkpoint:
 838         nilfs_detach_checkpoint(sbi);
 839
 840  failed_sbi:
 841         put_nilfs(nilfs);
 842         sb->s_fs_info = NULL;
 843         nilfs_put_sbinfo(sbi);
 844         return err;
 845 }
 846
 847 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
 848 {
 849         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 850         struct nilfs_super_block *sbp;
 851         struct the_nilfs *nilfs = sbi->s_nilfs;
 852         unsigned long old_sb_flags;
 853         struct nilfs_mount_options old_opts;
 854         int err;
 855
 856         lock_kernel();
 857
 858         down_write(&nilfs->ns_super_sem);
 859         old_sb_flags = sb->s_flags;
 860         old_opts.mount_opt = sbi->s_mount_opt;
 861         old_opts.snapshot_cno = sbi->s_snapshot_cno;
 862
 863         if (!parse_options(data, sb)) {
 864                 err = -EINVAL;
 865                 goto restore_opts;
 866         }
 867         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
 868
 869         if ((*flags & MS_RDONLY) &&
 870             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
 871                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 872                        "remount to a different snapshot. \n",
 873                        sb->s_id);
 874                 err = -EINVAL;
 875                 goto restore_opts;
 876         }
 877
 878         if (!nilfs_valid_fs(nilfs)) {
 879                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 880                        "remount because the filesystem is in an "
 881                        "incomplete recovery state.\n", sb->s_id);
 882                 err = -EINVAL;
 883                 goto restore_opts;
 884         }
 885
 886         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 887                 goto out;
 888         if (*flags & MS_RDONLY) {
 889                 /* Shutting down the segment constructor */
 890                 nilfs_detach_segment_constructor(sbi);
 891                 sb->s_flags |= MS_RDONLY;
 892
 893                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
 894                 /* nilfs_set_opt(sbi, SNAPSHOT); */
 895
 896                 /*
 897                  * Remounting a valid RW partition RDONLY, so set
 898                  * the RDONLY flag and then mark the partition as valid again.
 899                  */
 900                 down_write(&nilfs->ns_sem);
 901                 sbp = nilfs->ns_sbp[0];
 902                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
 903                     (nilfs->ns_mount_state & NILFS_VALID_FS))
 904                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
 905                 sbp->s_mtime = cpu_to_le64(get_seconds());
 906                 nilfs_commit_super(sbi, 1);
 907                 up_write(&nilfs->ns_sem);
 908         } else {
 909                 /*
 910                  * Mounting a RDONLY partition read-write, so reread and
 911                  * store the current valid flag.  (It may have been changed
 912                  * by fsck since we originally mounted the partition.)
 913                  */
 914                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
 915                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 916                                "remount because an RW-mount exists.\n",
 917                                sb->s_id);
 918                         err = -EBUSY;
 919                         goto restore_opts;
 920                 }
 921                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
 922                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 923                                "remount because the current RO-mount is not "
 924                                "the latest one.\n",
 925                                sb->s_id);
 926                         err = -EINVAL;
 927                         goto restore_opts;
 928                 }
 929                 sb->s_flags &= ~MS_RDONLY;
 930                 nilfs_clear_opt(sbi, SNAPSHOT);
 931                 sbi->s_snapshot_cno = 0;
 932
 933                 err = nilfs_attach_segment_constructor(sbi);
 934                 if (err)
 935                         goto restore_opts;
 936
 937                 down_write(&nilfs->ns_sem);
 938                 nilfs_setup_super(sbi);
 939                 up_write(&nilfs->ns_sem);
 940
 941                 nilfs->ns_current = sbi;
 942         }
 943  out:
 944         up_write(&nilfs->ns_super_sem);
 945         unlock_kernel();
 946         return 0;
 947
 948  restore_opts:
 949         sb->s_flags = old_sb_flags;
 950         sbi->s_mount_opt = old_opts.mount_opt;
 951         sbi->s_snapshot_cno = old_opts.snapshot_cno;
 952         up_write(&nilfs->ns_super_sem);
 953         unlock_kernel();
 954         return err;
 955 }
 956
 957 struct nilfs_super_data {
 958         struct block_device *bdev;
 959         struct nilfs_sb_info *sbi;
 960         __u64 cno;
 961         int flags;
 962 };
 963
 964 /**
 965  * nilfs_identify - pre-read mount options needed to identify mount instance
 966  * @data: mount options
 967  * @sd: nilfs_super_data
 968  */
 969 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
 970 {
 971         char *p, *options = data;
 972         substring_t args[MAX_OPT_ARGS];
 973         int option, token;
 974         int ret = 0;
 975
 976         do {
 977                 p = strsep(&options, ",");
 978                 if (p != NULL && *p) {
 979                         token = match_token(p, tokens, args);
 980                         if (token == Opt_snapshot) {
 981                                 if (!(sd->flags & MS_RDONLY))
 982                                         ret++;
 983                                 else {
 984                                         ret = match_int(&args[0], &option);
 985                                         if (!ret) {
 986                                                 if (option > 0)
 987                                                         sd->cno = option;
 988                                                 else
 989                                                         ret++;
 990                                         }
 991                                 }
 992                         }
 993                         if (ret)
 994                                 printk(KERN_ERR
 995                                        "NILFS: invalid mount option: %s\n", p);
 996                 }
 997                 if (!options)
 998                         break;
 999                 BUG_ON(options == data);
1000                 *(options - 1) = ',';
1001         } while (!ret);
1002         return ret;
1003 }
1004
1005 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1006 {
1007         struct nilfs_super_data *sd = data;
1008
1009         s->s_bdev = sd->bdev;
1010         s->s_dev = s->s_bdev->bd_dev;
1011         return 0;
1012 }
1013
1014 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1015 {
1016         struct nilfs_super_data *sd = data;
1017
1018         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1019 }
1020
1021 static int
1022 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1023              const char *dev_name, void *data, struct vfsmount *mnt)
1024 {
1025         struct nilfs_super_data sd;
1026         struct super_block *s;
1027         struct the_nilfs *nilfs;
1028         int err, need_to_close = 1;
1029
1030         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1031         if (IS_ERR(sd.bdev))
1032                 return PTR_ERR(sd.bdev);
1033
1034         /*
1035          * To get mount instance using sget() vfs-routine, NILFS needs
1036          * much more information than normal filesystems to identify mount
1037          * instance.  For snapshot mounts, not only a mount type (ro-mount
1038          * or rw-mount) but also a checkpoint number is required.
1039          */
1040         sd.cno = 0;
1041         sd.flags = flags;
1042         if (nilfs_identify((char *)data, &sd)) {
1043                 err = -EINVAL;
1044                 goto failed;
1045         }
1046
1047         nilfs = find_or_create_nilfs(sd.bdev);
1048         if (!nilfs) {
1049                 err = -ENOMEM;
1050                 goto failed;
1051         }
1052
1053         mutex_lock(&nilfs->ns_mount_mutex);
1054
1055         if (!sd.cno) {
1056                 /*
1057                  * Check if an exclusive mount exists or not.
1058                  * Snapshot mounts coexist with a current mount
1059                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1060                  * ro-mount are mutually exclusive.
1061                  */
1062                 down_read(&nilfs->ns_super_sem);
1063                 if (nilfs->ns_current &&
1064                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1065                      & MS_RDONLY)) {
1066                         up_read(&nilfs->ns_super_sem);
1067                         err = -EBUSY;
1068                         goto failed_unlock;
1069                 }
1070                 up_read(&nilfs->ns_super_sem);
1071         }
1072
1073         /*
1074          * Find existing nilfs_sb_info struct
1075          */
1076         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1077
1078         /*
1079          * Get super block instance holding the nilfs_sb_info struct.
1080          * A new instance is allocated if no existing mount is present or
1081          * existing instance has been unmounted.
1082          */
1083         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1084         if (sd.sbi)
1085                 nilfs_put_sbinfo(sd.sbi);
1086
1087         if (IS_ERR(s)) {
1088                 err = PTR_ERR(s);
1089                 goto failed_unlock;
1090         }
1091
1092         if (!s->s_root) {
1093                 char b[BDEVNAME_SIZE];
1094
1095                 /* New superblock instance created */
1096                 s->s_flags = flags;
1097                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1098                 sb_set_blocksize(s, block_size(sd.bdev));
1099
1100                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1101                 if (err)
1102                         goto cancel_new;
1103
1104                 s->s_flags |= MS_ACTIVE;
1105                 need_to_close = 0;
1106         }
1107
1108         mutex_unlock(&nilfs->ns_mount_mutex);
1109         put_nilfs(nilfs);
1110         if (need_to_close)
1111                 close_bdev_exclusive(sd.bdev, flags);
1112         simple_set_mnt(mnt, s);
1113         return 0;
1114
1115  failed_unlock:
1116         mutex_unlock(&nilfs->ns_mount_mutex);
1117         put_nilfs(nilfs);
1118  failed:
1119         close_bdev_exclusive(sd.bdev, flags);
1120
1121         return err;
1122
1123  cancel_new:
1124         /* Abandoning the newly allocated superblock */
1125         mutex_unlock(&nilfs->ns_mount_mutex);
1126         put_nilfs(nilfs);
1127         deactivate_locked_super(s);
1128         /*
1129          * deactivate_super() invokes close_bdev_exclusive().
1130          * We must finish all post-cleaning before this call;
1131          * put_nilfs() needs the block device.
1132          */
1133         return err;
1134 }
1135
1136 struct file_system_type nilfs_fs_type = {
1137         .owner    = THIS_MODULE,
1138         .name     = "nilfs2",
1139         .get_sb   = nilfs_get_sb,
1140         .kill_sb  = kill_block_super,
1141         .fs_flags = FS_REQUIRES_DEV,
1142 };
1143
1144 static int __init init_nilfs_fs(void)
1145 {
1146         int err;
1147
1148         err = nilfs_init_inode_cache();
1149         if (err)
1150                 goto failed;
1151
1152         err = nilfs_init_transaction_cache();
1153         if (err)
1154                 goto failed_inode_cache;
1155
1156         err = nilfs_init_segbuf_cache();
1157         if (err)
1158                 goto failed_transaction_cache;
1159
1160         err = nilfs_btree_path_cache_init();
1161         if (err)
1162                 goto failed_segbuf_cache;
1163
1164         err = register_filesystem(&nilfs_fs_type);
1165         if (err)
1166                 goto failed_btree_path_cache;
1167
1168         return 0;
1169
1170  failed_btree_path_cache:
1171         nilfs_btree_path_cache_destroy();
1172
1173  failed_segbuf_cache:
1174         nilfs_destroy_segbuf_cache();
1175
1176  failed_transaction_cache:
1177         nilfs_destroy_transaction_cache();
1178
1179  failed_inode_cache:
1180         nilfs_destroy_inode_cache();
1181
1182  failed:
1183         return err;
1184 }
1185
1186 static void __exit exit_nilfs_fs(void)
1187 {
1188         nilfs_destroy_segbuf_cache();
1189         nilfs_destroy_transaction_cache();
1190         nilfs_destroy_inode_cache();
1191         nilfs_btree_path_cache_destroy();
1192         unregister_filesystem(&nilfs_fs_type);
1193 }
1194
1195 module_init(init_nilfs_fs)
1196 module_exit(exit_nilfs_fs)