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