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 "nilfs.h"
  54 #include "mdt.h"
  55 #include "alloc.h"
  56 #include "page.h"
  57 #include "cpfile.h"
  58 #include "ifile.h"
  59 #include "dat.h"
  60 #include "segment.h"
  61 #include "segbuf.h"
  62
  63 MODULE_AUTHOR("NTT Corp.");
  64 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  65                    "(NILFS)");
  66 MODULE_LICENSE("GPL");
  67
  68 static void nilfs_write_super(struct super_block *sb);
  69 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  70
  71 /**
  72  * nilfs_error() - report failure condition on a filesystem
  73  *
  74  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
  75  * reporting an error message.  It should be called when NILFS detects
  76  * incoherences or defects of meta data on disk.  As for sustainable
  77  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
  78  * function should be used instead.
  79  *
  80  * The segment constructor must not call this function because it can
  81  * kill itself.
  82  */
  83 void nilfs_error(struct super_block *sb, const char *function,
  84                  const char *fmt, ...)
  85 {
  86         struct nilfs_sb_info *sbi = NILFS_SB(sb);
  87         va_list args;
  88
  89         va_start(args, fmt);
  90         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
  91         vprintk(fmt, args);
  92         printk("\n");
  93         va_end(args);
  94
  95         if (!(sb->s_flags & MS_RDONLY)) {
  96                 struct the_nilfs *nilfs = sbi->s_nilfs;
  97
  98                 if (!nilfs_test_opt(sbi, ERRORS_CONT))
  99                         nilfs_detach_segment_constructor(sbi);
 100
 101                 down_write(&nilfs->ns_sem);
 102                 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
 103                         nilfs->ns_mount_state |= NILFS_ERROR_FS;
 104                         nilfs->ns_sbp[0]->s_state |=
 105                                 cpu_to_le16(NILFS_ERROR_FS);
 106                         nilfs_commit_super(sbi, 1);
 107                 }
 108                 up_write(&nilfs->ns_sem);
 109
 110                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
 111                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 112                         sb->s_flags |= MS_RDONLY;
 113                 }
 114         }
 115
 116         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 117                 panic("NILFS (device %s): panic forced after error\n",
 118                       sb->s_id);
 119 }
 120
 121 void nilfs_warning(struct super_block *sb, const char *function,
 122                    const char *fmt, ...)
 123 {
 124         va_list args;
 125
 126         va_start(args, fmt);
 127         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
 128                sb->s_id, function);
 129         vprintk(fmt, args);
 130         printk("\n");
 131         va_end(args);
 132 }
 133
 134 static struct kmem_cache *nilfs_inode_cachep;
 135
 136 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
 137 {
 138         struct nilfs_inode_info *ii;
 139
 140         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 141         if (!ii)
 142                 return NULL;
 143         ii->i_bh = NULL;
 144         ii->i_state = 0;
 145         ii->vfs_inode.i_version = 1;
 146         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
 147         return &ii->vfs_inode;
 148 }
 149
 150 struct inode *nilfs_alloc_inode(struct super_block *sb)
 151 {
 152         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
 153 }
 154
 155 void nilfs_destroy_inode(struct inode *inode)
 156 {
 157         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 158 }
 159
 160 static void init_once(void *obj)
 161 {
 162         struct nilfs_inode_info *ii = obj;
 163
 164         INIT_LIST_HEAD(&ii->i_dirty);
 165 #ifdef CONFIG_NILFS_XATTR
 166         init_rwsem(&ii->xattr_sem);
 167 #endif
 168         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
 169         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
 170         inode_init_once(&ii->vfs_inode);
 171 }
 172
 173 static int nilfs_init_inode_cache(void)
 174 {
 175         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
 176                                                sizeof(struct nilfs_inode_info),
 177                                                0, SLAB_RECLAIM_ACCOUNT,
 178                                                init_once);
 179
 180         return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
 181 }
 182
 183 static inline void nilfs_destroy_inode_cache(void)
 184 {
 185         kmem_cache_destroy(nilfs_inode_cachep);
 186 }
 187
 188 static void nilfs_clear_inode(struct inode *inode)
 189 {
 190         struct nilfs_inode_info *ii = NILFS_I(inode);
 191
 192         /*
 193          * Free resources allocated in nilfs_read_inode(), here.
 194          */
 195         BUG_ON(!list_empty(&ii->i_dirty));
 196         brelse(ii->i_bh);
 197         ii->i_bh = NULL;
 198
 199         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 200                 nilfs_bmap_clear(ii->i_bmap);
 201
 202         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 203 }
 204
 205 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
 206 {
 207         struct the_nilfs *nilfs = sbi->s_nilfs;
 208         int err;
 209         int barrier_done = 0;
 210
 211         if (nilfs_test_opt(sbi, BARRIER)) {
 212                 set_buffer_ordered(nilfs->ns_sbh[0]);
 213                 barrier_done = 1;
 214         }
 215  retry:
 216         set_buffer_dirty(nilfs->ns_sbh[0]);
 217         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 218         if (err == -EOPNOTSUPP && barrier_done) {
 219                 nilfs_warning(sbi->s_super, __func__,
 220                               "barrier-based sync failed. "
 221                               "disabling barriers\n");
 222                 nilfs_clear_opt(sbi, BARRIER);
 223                 barrier_done = 0;
 224                 clear_buffer_ordered(nilfs->ns_sbh[0]);
 225                 goto retry;
 226         }
 227         if (unlikely(err)) {
 228                 printk(KERN_ERR
 229                        "NILFS: unable to write superblock (err=%d)\n", err);
 230                 if (err == -EIO && nilfs->ns_sbh[1]) {
 231                         nilfs_fall_back_super_block(nilfs);
 232                         goto retry;
 233                 }
 234         } else {
 235                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 236
 237                 /*
 238                  * The latest segment becomes trailable from the position
 239                  * written in superblock.
 240                  */
 241                 clear_nilfs_discontinued(nilfs);
 242
 243                 /* update GC protection for recent segments */
 244                 if (nilfs->ns_sbh[1]) {
 245                         sbp = NULL;
 246                         if (dupsb) {
 247                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 248                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
 249                                         sbp = nilfs->ns_sbp[1];
 250                         }
 251                 }
 252                 if (sbp) {
 253                         spin_lock(&nilfs->ns_last_segment_lock);
 254                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 255                         spin_unlock(&nilfs->ns_last_segment_lock);
 256                 }
 257         }
 258
 259         return err;
 260 }
 261
 262 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
 263 {
 264         struct the_nilfs *nilfs = sbi->s_nilfs;
 265         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 266         sector_t nfreeblocks;
 267         time_t t;
 268         int err;
 269
 270         /* nilfs->sem must be locked by the caller. */
 271         if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
 272                 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
 273                         nilfs_swap_super_block(nilfs);
 274                 else {
 275                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 276                                sbi->s_super->s_id);
 277                         return -EIO;
 278                 }
 279         }
 280         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 281         if (unlikely(err)) {
 282                 printk(KERN_ERR "NILFS: failed to count free blocks\n");
 283                 return err;
 284         }
 285         spin_lock(&nilfs->ns_last_segment_lock);
 286         sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 287         sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 288         sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 289         spin_unlock(&nilfs->ns_last_segment_lock);
 290
 291         t = get_seconds();
 292         nilfs->ns_sbwtime[0] = t;
 293         sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 294         sbp[0]->s_wtime = cpu_to_le64(t);
 295         sbp[0]->s_sum = 0;
 296         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 297                                              (unsigned char *)sbp[0],
 298                                              nilfs->ns_sbsize));
 299         if (dupsb && sbp[1]) {
 300                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 301                 nilfs->ns_sbwtime[1] = t;
 302         }
 303         sbi->s_super->s_dirt = 0;
 304         return nilfs_sync_super(sbi, dupsb);
 305 }
 306
 307 static void nilfs_put_super(struct super_block *sb)
 308 {
 309         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 310         struct the_nilfs *nilfs = sbi->s_nilfs;
 311
 312         lock_kernel();
 313
 314         if (sb->s_dirt)
 315                 nilfs_write_super(sb);
 316
 317         nilfs_detach_segment_constructor(sbi);
 318
 319         if (!(sb->s_flags & MS_RDONLY)) {
 320                 down_write(&nilfs->ns_sem);
 321                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 322                 nilfs_commit_super(sbi, 1);
 323                 up_write(&nilfs->ns_sem);
 324         }
 325         down_write(&nilfs->ns_super_sem);
 326         if (nilfs->ns_current == sbi)
 327                 nilfs->ns_current = NULL;
 328         up_write(&nilfs->ns_super_sem);
 329
 330         nilfs_detach_checkpoint(sbi);
 331         put_nilfs(sbi->s_nilfs);
 332         sbi->s_super = NULL;
 333         sb->s_fs_info = NULL;
 334         nilfs_put_sbinfo(sbi);
 335
 336         unlock_kernel();
 337 }
 338
 339 /**
 340  * nilfs_write_super - write super block(s) of NILFS
 341  * @sb: super_block
 342  *
 343  * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
 344  * clears s_dirt.  This function is called in the section protected by
 345  * lock_super().
 346  *
 347  * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
 348  * of the struct the_nilfs.  Lock order must be as follows:
 349  *
 350  *   1. lock_super()
 351  *   2.    down_write(&nilfs->ns_sem)
 352  *
 353  * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
 354  * of the super block (nilfs->ns_sbp[]).
 355  *
 356  * In most cases, VFS functions call lock_super() before calling these
 357  * methods.  So we must be careful not to bring on deadlocks when using
 358  * lock_super();  see generic_shutdown_super(), write_super(), and so on.
 359  *
 360  * Note that order of lock_kernel() and lock_super() depends on contexts
 361  * of VFS.  We should also note that lock_kernel() can be used in its
 362  * protective section and only the outermost one has an effect.
 363  */
 364 static void nilfs_write_super(struct super_block *sb)
 365 {
 366         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 367         struct the_nilfs *nilfs = sbi->s_nilfs;
 368
 369         down_write(&nilfs->ns_sem);
 370         if (!(sb->s_flags & MS_RDONLY)) {
 371                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
 372                 u64 t = get_seconds();
 373                 int dupsb;
 374
 375                 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
 376                     t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
 377                         up_write(&nilfs->ns_sem);
 378                         return;
 379                 }
 380                 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
 381                 nilfs_commit_super(sbi, dupsb);
 382         }
 383         sb->s_dirt = 0;
 384         up_write(&nilfs->ns_sem);
 385 }
 386
 387 static int nilfs_sync_fs(struct super_block *sb, int wait)
 388 {
 389         int err = 0;
 390
 391         nilfs_write_super(sb);
 392
 393         /* This function is called when super block should be written back */
 394         if (wait)
 395                 err = nilfs_construct_segment(sb);
 396         return err;
 397 }
 398
 399 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
 400 {
 401         struct the_nilfs *nilfs = sbi->s_nilfs;
 402         struct nilfs_checkpoint *raw_cp;
 403         struct buffer_head *bh_cp;
 404         int err;
 405
 406         down_write(&nilfs->ns_super_sem);
 407         list_add(&sbi->s_list, &nilfs->ns_supers);
 408         up_write(&nilfs->ns_super_sem);
 409
 410         sbi->s_ifile = nilfs_mdt_new(
 411                 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
 412         if (!sbi->s_ifile)
 413                 return -ENOMEM;
 414
 415         err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
 416         if (unlikely(err))
 417                 goto failed;
 418
 419         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 420                                           &bh_cp);
 421         if (unlikely(err)) {
 422                 if (err == -ENOENT || err == -EINVAL) {
 423                         printk(KERN_ERR
 424                                "NILFS: Invalid checkpoint "
 425                                "(checkpoint number=%llu)\n",
 426                                (unsigned long long)cno);
 427                         err = -EINVAL;
 428                 }
 429                 goto failed;
 430         }
 431         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
 432         if (unlikely(err))
 433                 goto failed_bh;
 434         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 435         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 436
 437         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 438         return 0;
 439
 440  failed_bh:
 441         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 442  failed:
 443         nilfs_mdt_destroy(sbi->s_ifile);
 444         sbi->s_ifile = NULL;
 445
 446         down_write(&nilfs->ns_super_sem);
 447         list_del_init(&sbi->s_list);
 448         up_write(&nilfs->ns_super_sem);
 449
 450         return err;
 451 }
 452
 453 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
 454 {
 455         struct the_nilfs *nilfs = sbi->s_nilfs;
 456
 457         nilfs_mdt_clear(sbi->s_ifile);
 458         nilfs_mdt_destroy(sbi->s_ifile);
 459         sbi->s_ifile = NULL;
 460         down_write(&nilfs->ns_super_sem);
 461         list_del_init(&sbi->s_list);
 462         up_write(&nilfs->ns_super_sem);
 463 }
 464
 465 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
 466 {
 467         struct the_nilfs *nilfs = sbi->s_nilfs;
 468         int err = 0;
 469
 470         down_write(&nilfs->ns_sem);
 471         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
 472                 nilfs->ns_mount_state |= NILFS_VALID_FS;
 473                 err = nilfs_commit_super(sbi, 1);
 474                 if (likely(!err))
 475                         printk(KERN_INFO "NILFS: recovery complete.\n");
 476         }
 477         up_write(&nilfs->ns_sem);
 478         return err;
 479 }
 480
 481 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 482 {
 483         struct super_block *sb = dentry->d_sb;
 484         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 485         struct the_nilfs *nilfs = sbi->s_nilfs;
 486         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 487         unsigned long long blocks;
 488         unsigned long overhead;
 489         unsigned long nrsvblocks;
 490         sector_t nfreeblocks;
 491         int err;
 492
 493         /*
 494          * Compute all of the segment blocks
 495          *
 496          * The blocks before first segment and after last segment
 497          * are excluded.
 498          */
 499         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 500                 - nilfs->ns_first_data_block;
 501         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 502
 503         /*
 504          * Compute the overhead
 505          *
 506          * When distributing meta data blocks outside semgent structure,
 507          * We must count them as the overhead.
 508          */
 509         overhead = 0;
 510
 511         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 512         if (unlikely(err))
 513                 return err;
 514
 515         buf->f_type = NILFS_SUPER_MAGIC;
 516         buf->f_bsize = sb->s_blocksize;
 517         buf->f_blocks = blocks - overhead;
 518         buf->f_bfree = nfreeblocks;
 519         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 520                 (buf->f_bfree - nrsvblocks) : 0;
 521         buf->f_files = atomic_read(&sbi->s_inodes_count);
 522         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 523         buf->f_namelen = NILFS_NAME_LEN;
 524         buf->f_fsid.val[0] = (u32)id;
 525         buf->f_fsid.val[1] = (u32)(id >> 32);
 526
 527         return 0;
 528 }
 529
 530 static struct super_operations nilfs_sops = {
 531         .alloc_inode    = nilfs_alloc_inode,
 532         .destroy_inode  = nilfs_destroy_inode,
 533         .dirty_inode    = nilfs_dirty_inode,
 534         /* .write_inode    = nilfs_write_inode, */
 535         /* .put_inode      = nilfs_put_inode, */
 536         /* .drop_inode    = nilfs_drop_inode, */
 537         .delete_inode   = nilfs_delete_inode,
 538         .put_super      = nilfs_put_super,
 539         .write_super    = nilfs_write_super,
 540         .sync_fs        = nilfs_sync_fs,
 541         /* .write_super_lockfs */
 542         /* .unlockfs */
 543         .statfs         = nilfs_statfs,
 544         .remount_fs     = nilfs_remount,
 545         .clear_inode    = nilfs_clear_inode,
 546         /* .umount_begin */
 547         /* .show_options */
 548 };
 549
 550 static struct inode *
 551 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
 552 {
 553         struct inode *inode;
 554
 555         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
 556             ino != NILFS_SKETCH_INO)
 557                 return ERR_PTR(-ESTALE);
 558
 559         inode = nilfs_iget(sb, ino);
 560         if (IS_ERR(inode))
 561                 return ERR_CAST(inode);
 562         if (generation && inode->i_generation != generation) {
 563                 iput(inode);
 564                 return ERR_PTR(-ESTALE);
 565         }
 566
 567         return inode;
 568 }
 569
 570 static struct dentry *
 571 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
 572                    int fh_type)
 573 {
 574         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 575                                     nilfs_nfs_get_inode);
 576 }
 577
 578 static struct dentry *
 579 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
 580                    int fh_type)
 581 {
 582         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 583                                     nilfs_nfs_get_inode);
 584 }
 585
 586 static struct export_operations nilfs_export_ops = {
 587         .fh_to_dentry = nilfs_fh_to_dentry,
 588         .fh_to_parent = nilfs_fh_to_parent,
 589         .get_parent = nilfs_get_parent,
 590 };
 591
 592 enum {
 593         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 594         Opt_barrier, Opt_snapshot, Opt_order,
 595         Opt_err,
 596 };
 597
 598 static match_table_t tokens = {
 599         {Opt_err_cont, "errors=continue"},
 600         {Opt_err_panic, "errors=panic"},
 601         {Opt_err_ro, "errors=remount-ro"},
 602         {Opt_barrier, "barrier=%s"},
 603         {Opt_snapshot, "cp=%u"},
 604         {Opt_order, "order=%s"},
 605         {Opt_err, NULL}
 606 };
 607
 608 static int match_bool(substring_t *s, int *result)
 609 {
 610         int len = s->to - s->from;
 611
 612         if (strncmp(s->from, "on", len) == 0)
 613                 *result = 1;
 614         else if (strncmp(s->from, "off", len) == 0)
 615                 *result = 0;
 616         else
 617                 return 1;
 618         return 0;
 619 }
 620
 621 static int parse_options(char *options, struct super_block *sb)
 622 {
 623         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 624         char *p;
 625         substring_t args[MAX_OPT_ARGS];
 626         int option;
 627
 628         if (!options)
 629                 return 1;
 630
 631         while ((p = strsep(&options, ",")) != NULL) {
 632                 int token;
 633                 if (!*p)
 634                         continue;
 635
 636                 token = match_token(p, tokens, args);
 637                 switch (token) {
 638                 case Opt_barrier:
 639                         if (match_bool(&args[0], &option))
 640                                 return 0;
 641                         if (option)
 642                                 nilfs_set_opt(sbi, BARRIER);
 643                         else
 644                                 nilfs_clear_opt(sbi, BARRIER);
 645                         break;
 646                 case Opt_order:
 647                         if (strcmp(args[0].from, "relaxed") == 0)
 648                                 /* Ordered data semantics */
 649                                 nilfs_clear_opt(sbi, STRICT_ORDER);
 650                         else if (strcmp(args[0].from, "strict") == 0)
 651                                 /* Strict in-order semantics */
 652                                 nilfs_set_opt(sbi, STRICT_ORDER);
 653                         else
 654                                 return 0;
 655                         break;
 656                 case Opt_err_panic:
 657                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
 658                         break;
 659                 case Opt_err_ro:
 660                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
 661                         break;
 662                 case Opt_err_cont:
 663                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
 664                         break;
 665                 case Opt_snapshot:
 666                         if (match_int(&args[0], &option) || option <= 0)
 667                                 return 0;
 668                         if (!(sb->s_flags & MS_RDONLY))
 669                                 return 0;
 670                         sbi->s_snapshot_cno = option;
 671                         nilfs_set_opt(sbi, SNAPSHOT);
 672                         break;
 673                 default:
 674                         printk(KERN_ERR
 675                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 676                         return 0;
 677                 }
 678         }
 679         return 1;
 680 }
 681
 682 static inline void
 683 nilfs_set_default_options(struct nilfs_sb_info *sbi,
 684                           struct nilfs_super_block *sbp)
 685 {
 686         sbi->s_mount_opt =
 687                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
 688 }
 689
 690 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
 691 {
 692         struct the_nilfs *nilfs = sbi->s_nilfs;
 693         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 694         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
 695         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
 696
 697         /* nilfs->sem must be locked by the caller. */
 698         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
 699                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 700         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 701                 printk(KERN_WARNING
 702                        "NILFS warning: mounting fs with errors\n");
 703 #if 0
 704         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 705                 printk(KERN_WARNING
 706                        "NILFS warning: maximal mount count reached\n");
 707 #endif
 708         }
 709         if (!max_mnt_count)
 710                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 711
 712         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
 713         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
 714         sbp->s_mtime = cpu_to_le64(get_seconds());
 715         return nilfs_commit_super(sbi, 1);
 716 }
 717
 718 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 719                                                  u64 pos, int blocksize,
 720                                                  struct buffer_head **pbh)
 721 {
 722         unsigned long long sb_index = pos;
 723         unsigned long offset;
 724
 725         offset = do_div(sb_index, blocksize);
 726         *pbh = sb_bread(sb, sb_index);
 727         if (!*pbh)
 728                 return NULL;
 729         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 730 }
 731
 732 int nilfs_store_magic_and_option(struct super_block *sb,
 733                                  struct nilfs_super_block *sbp,
 734                                  char *data)
 735 {
 736         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 737
 738         sb->s_magic = le16_to_cpu(sbp->s_magic);
 739
 740         /* FS independent flags */
 741 #ifdef NILFS_ATIME_DISABLE
 742         sb->s_flags |= MS_NOATIME;
 743 #endif
 744
 745         nilfs_set_default_options(sbi, sbp);
 746
 747         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
 748         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
 749         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
 750         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
 751
 752         return !parse_options(data, sb) ? -EINVAL : 0 ;
 753 }
 754
 755 /**
 756  * nilfs_fill_super() - initialize a super block instance
 757  * @sb: super_block
 758  * @data: mount options
 759  * @silent: silent mode flag
 760  * @nilfs: the_nilfs struct
 761  *
 762  * This function is called exclusively by nilfs->ns_mount_mutex.
 763  * So, the recovery process is protected from other simultaneous mounts.
 764  */
 765 static int
 766 nilfs_fill_super(struct super_block *sb, void *data, int silent,
 767                  struct the_nilfs *nilfs)
 768 {
 769         struct nilfs_sb_info *sbi;
 770         struct inode *root;
 771         __u64 cno;
 772         int err;
 773
 774         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 775         if (!sbi)
 776                 return -ENOMEM;
 777
 778         sb->s_fs_info = sbi;
 779
 780         get_nilfs(nilfs);
 781         sbi->s_nilfs = nilfs;
 782         sbi->s_super = sb;
 783         atomic_set(&sbi->s_count, 1);
 784
 785         err = init_nilfs(nilfs, sbi, (char *)data);
 786         if (err)
 787                 goto failed_sbi;
 788
 789         spin_lock_init(&sbi->s_inode_lock);
 790         INIT_LIST_HEAD(&sbi->s_dirty_files);
 791         INIT_LIST_HEAD(&sbi->s_list);
 792
 793         /*
 794          * Following initialization is overlapped because
 795          * nilfs_sb_info structure has been cleared at the beginning.
 796          * But we reserve them to keep our interest and make ready
 797          * for the future change.
 798          */
 799         get_random_bytes(&sbi->s_next_generation,
 800                          sizeof(sbi->s_next_generation));
 801         spin_lock_init(&sbi->s_next_gen_lock);
 802
 803         sb->s_op = &nilfs_sops;
 804         sb->s_export_op = &nilfs_export_ops;
 805         sb->s_root = NULL;
 806         sb->s_time_gran = 1;
 807
 808         if (!nilfs_loaded(nilfs)) {
 809                 err = load_nilfs(nilfs, sbi);
 810                 if (err)
 811                         goto failed_sbi;
 812         }
 813         cno = nilfs_last_cno(nilfs);
 814
 815         if (sb->s_flags & MS_RDONLY) {
 816                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
 817                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
 818                                                        sbi->s_snapshot_cno);
 819                         if (err < 0)
 820                                 goto failed_sbi;
 821                         if (!err) {
 822                                 printk(KERN_ERR
 823                                        "NILFS: The specified checkpoint is "
 824                                        "not a snapshot "
 825                                        "(checkpoint number=%llu).\n",
 826                                        (unsigned long long)sbi->s_snapshot_cno);
 827                                 err = -EINVAL;
 828                                 goto failed_sbi;
 829                         }
 830                         cno = sbi->s_snapshot_cno;
 831                 } else
 832                         /* Read-only mount */
 833                         sbi->s_snapshot_cno = cno;
 834         }
 835
 836         err = nilfs_attach_checkpoint(sbi, cno);
 837         if (err) {
 838                 printk(KERN_ERR "NILFS: error loading a checkpoint"
 839                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
 840                 goto failed_sbi;
 841         }
 842
 843         if (!(sb->s_flags & MS_RDONLY)) {
 844                 err = nilfs_attach_segment_constructor(sbi);
 845                 if (err)
 846                         goto failed_checkpoint;
 847         }
 848
 849         root = nilfs_iget(sb, NILFS_ROOT_INO);
 850         if (IS_ERR(root)) {
 851                 printk(KERN_ERR "NILFS: get root inode failed\n");
 852                 err = PTR_ERR(root);
 853                 goto failed_segctor;
 854         }
 855         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
 856                 iput(root);
 857                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 858                 err = -EINVAL;
 859                 goto failed_segctor;
 860         }
 861         sb->s_root = d_alloc_root(root);
 862         if (!sb->s_root) {
 863                 iput(root);
 864                 printk(KERN_ERR "NILFS: get root dentry failed\n");
 865                 err = -ENOMEM;
 866                 goto failed_segctor;
 867         }
 868
 869         if (!(sb->s_flags & MS_RDONLY)) {
 870                 down_write(&nilfs->ns_sem);
 871                 nilfs_setup_super(sbi);
 872                 up_write(&nilfs->ns_sem);
 873         }
 874
 875         err = nilfs_mark_recovery_complete(sbi);
 876         if (unlikely(err)) {
 877                 printk(KERN_ERR "NILFS: recovery failed.\n");
 878                 goto failed_root;
 879         }
 880
 881         down_write(&nilfs->ns_super_sem);
 882         if (!nilfs_test_opt(sbi, SNAPSHOT))
 883                 nilfs->ns_current = sbi;
 884         up_write(&nilfs->ns_super_sem);
 885
 886         return 0;
 887
 888  failed_root:
 889         dput(sb->s_root);
 890         sb->s_root = NULL;
 891
 892  failed_segctor:
 893         nilfs_detach_segment_constructor(sbi);
 894
 895  failed_checkpoint:
 896         nilfs_detach_checkpoint(sbi);
 897
 898  failed_sbi:
 899         put_nilfs(nilfs);
 900         sb->s_fs_info = NULL;
 901         nilfs_put_sbinfo(sbi);
 902         return err;
 903 }
 904
 905 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
 906 {
 907         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 908         struct nilfs_super_block *sbp;
 909         struct the_nilfs *nilfs = sbi->s_nilfs;
 910         unsigned long old_sb_flags;
 911         struct nilfs_mount_options old_opts;
 912         int err;
 913
 914         lock_kernel();
 915
 916         down_write(&nilfs->ns_super_sem);
 917         old_sb_flags = sb->s_flags;
 918         old_opts.mount_opt = sbi->s_mount_opt;
 919         old_opts.snapshot_cno = sbi->s_snapshot_cno;
 920
 921         if (!parse_options(data, sb)) {
 922                 err = -EINVAL;
 923                 goto restore_opts;
 924         }
 925         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
 926
 927         if ((*flags & MS_RDONLY) &&
 928             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
 929                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 930                        "remount to a different snapshot. \n",
 931                        sb->s_id);
 932                 err = -EINVAL;
 933                 goto restore_opts;
 934         }
 935
 936         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 937                 goto out;
 938         if (*flags & MS_RDONLY) {
 939                 /* Shutting down the segment constructor */
 940                 nilfs_detach_segment_constructor(sbi);
 941                 sb->s_flags |= MS_RDONLY;
 942
 943                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
 944                 /* nilfs_set_opt(sbi, SNAPSHOT); */
 945
 946                 /*
 947                  * Remounting a valid RW partition RDONLY, so set
 948                  * the RDONLY flag and then mark the partition as valid again.
 949                  */
 950                 down_write(&nilfs->ns_sem);
 951                 sbp = nilfs->ns_sbp[0];
 952                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
 953                     (nilfs->ns_mount_state & NILFS_VALID_FS))
 954                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
 955                 sbp->s_mtime = cpu_to_le64(get_seconds());
 956                 nilfs_commit_super(sbi, 1);
 957                 up_write(&nilfs->ns_sem);
 958         } else {
 959                 /*
 960                  * Mounting a RDONLY partition read-write, so reread and
 961                  * store the current valid flag.  (It may have been changed
 962                  * by fsck since we originally mounted the partition.)
 963                  */
 964                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
 965                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 966                                "remount because an RW-mount exists.\n",
 967                                sb->s_id);
 968                         err = -EBUSY;
 969                         goto restore_opts;
 970                 }
 971                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
 972                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 973                                "remount because the current RO-mount is not "
 974                                "the latest one.\n",
 975                                sb->s_id);
 976                         err = -EINVAL;
 977                         goto restore_opts;
 978                 }
 979                 sb->s_flags &= ~MS_RDONLY;
 980                 nilfs_clear_opt(sbi, SNAPSHOT);
 981                 sbi->s_snapshot_cno = 0;
 982
 983                 err = nilfs_attach_segment_constructor(sbi);
 984                 if (err)
 985                         goto restore_opts;
 986
 987                 down_write(&nilfs->ns_sem);
 988                 nilfs_setup_super(sbi);
 989                 up_write(&nilfs->ns_sem);
 990
 991                 nilfs->ns_current = sbi;
 992         }
 993  out:
 994         up_write(&nilfs->ns_super_sem);
 995         unlock_kernel();
 996         return 0;
 997
 998  restore_opts:
 999         sb->s_flags = old_sb_flags;
1000         sbi->s_mount_opt = old_opts.mount_opt;
1001         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1002         up_write(&nilfs->ns_super_sem);
1003         unlock_kernel();
1004         return err;
1005 }
1006
1007 struct nilfs_super_data {
1008         struct block_device *bdev;
1009         struct nilfs_sb_info *sbi;
1010         __u64 cno;
1011         int flags;
1012 };
1013
1014 /**
1015  * nilfs_identify - pre-read mount options needed to identify mount instance
1016  * @data: mount options
1017  * @sd: nilfs_super_data
1018  */
1019 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1020 {
1021         char *p, *options = data;
1022         substring_t args[MAX_OPT_ARGS];
1023         int option, token;
1024         int ret = 0;
1025
1026         do {
1027                 p = strsep(&options, ",");
1028                 if (p != NULL && *p) {
1029                         token = match_token(p, tokens, args);
1030                         if (token == Opt_snapshot) {
1031                                 if (!(sd->flags & MS_RDONLY))
1032                                         ret++;
1033                                 else {
1034                                         ret = match_int(&args[0], &option);
1035                                         if (!ret) {
1036                                                 if (option > 0)
1037                                                         sd->cno = option;
1038                                                 else
1039                                                         ret++;
1040                                         }
1041                                 }
1042                         }
1043                         if (ret)
1044                                 printk(KERN_ERR
1045                                        "NILFS: invalid mount option: %s\n", p);
1046                 }
1047                 if (!options)
1048                         break;
1049                 BUG_ON(options == data);
1050                 *(options - 1) = ',';
1051         } while (!ret);
1052         return ret;
1053 }
1054
1055 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1056 {
1057         struct nilfs_super_data *sd = data;
1058
1059         s->s_bdev = sd->bdev;
1060         s->s_dev = s->s_bdev->bd_dev;
1061         return 0;
1062 }
1063
1064 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1065 {
1066         struct nilfs_super_data *sd = data;
1067
1068         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1069 }
1070
1071 static int
1072 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1073              const char *dev_name, void *data, struct vfsmount *mnt)
1074 {
1075         struct nilfs_super_data sd;
1076         struct super_block *s;
1077         struct the_nilfs *nilfs;
1078         int err, need_to_close = 1;
1079
1080         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1081         if (IS_ERR(sd.bdev))
1082                 return PTR_ERR(sd.bdev);
1083
1084         /*
1085          * To get mount instance using sget() vfs-routine, NILFS needs
1086          * much more information than normal filesystems to identify mount
1087          * instance.  For snapshot mounts, not only a mount type (ro-mount
1088          * or rw-mount) but also a checkpoint number is required.
1089          */
1090         sd.cno = 0;
1091         sd.flags = flags;
1092         if (nilfs_identify((char *)data, &sd)) {
1093                 err = -EINVAL;
1094                 goto failed;
1095         }
1096
1097         nilfs = find_or_create_nilfs(sd.bdev);
1098         if (!nilfs) {
1099                 err = -ENOMEM;
1100                 goto failed;
1101         }
1102
1103         mutex_lock(&nilfs->ns_mount_mutex);
1104
1105         if (!sd.cno) {
1106                 /*
1107                  * Check if an exclusive mount exists or not.
1108                  * Snapshot mounts coexist with a current mount
1109                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1110                  * ro-mount are mutually exclusive.
1111                  */
1112                 down_read(&nilfs->ns_super_sem);
1113                 if (nilfs->ns_current &&
1114                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1115                      & MS_RDONLY)) {
1116                         up_read(&nilfs->ns_super_sem);
1117                         err = -EBUSY;
1118                         goto failed_unlock;
1119                 }
1120                 up_read(&nilfs->ns_super_sem);
1121         }
1122
1123         /*
1124          * Find existing nilfs_sb_info struct
1125          */
1126         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1127
1128         if (!sd.cno)
1129                 /* trying to get the latest checkpoint.  */
1130                 sd.cno = nilfs_last_cno(nilfs);
1131
1132         /*
1133          * Get super block instance holding the nilfs_sb_info struct.
1134          * A new instance is allocated if no existing mount is present or
1135          * existing instance has been unmounted.
1136          */
1137         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1138         if (sd.sbi)
1139                 nilfs_put_sbinfo(sd.sbi);
1140
1141         if (IS_ERR(s)) {
1142                 err = PTR_ERR(s);
1143                 goto failed_unlock;
1144         }
1145
1146         if (!s->s_root) {
1147                 char b[BDEVNAME_SIZE];
1148
1149                 /* New superblock instance created */
1150                 s->s_flags = flags;
1151                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1152                 sb_set_blocksize(s, block_size(sd.bdev));
1153
1154                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1155                 if (err)
1156                         goto cancel_new;
1157
1158                 s->s_flags |= MS_ACTIVE;
1159                 need_to_close = 0;
1160         }
1161
1162         mutex_unlock(&nilfs->ns_mount_mutex);
1163         put_nilfs(nilfs);
1164         if (need_to_close)
1165                 close_bdev_exclusive(sd.bdev, flags);
1166         simple_set_mnt(mnt, s);
1167         return 0;
1168
1169  failed_unlock:
1170         mutex_unlock(&nilfs->ns_mount_mutex);
1171         put_nilfs(nilfs);
1172  failed:
1173         close_bdev_exclusive(sd.bdev, flags);
1174
1175         return err;
1176
1177  cancel_new:
1178         /* Abandoning the newly allocated superblock */
1179         mutex_unlock(&nilfs->ns_mount_mutex);
1180         put_nilfs(nilfs);
1181         up_write(&s->s_umount);
1182         deactivate_super(s);
1183         /*
1184          * deactivate_super() invokes close_bdev_exclusive().
1185          * We must finish all post-cleaning before this call;
1186          * put_nilfs() needs the block device.
1187          */
1188         return err;
1189 }
1190
1191 struct file_system_type nilfs_fs_type = {
1192         .owner    = THIS_MODULE,
1193         .name     = "nilfs2",
1194         .get_sb   = nilfs_get_sb,
1195         .kill_sb  = kill_block_super,
1196         .fs_flags = FS_REQUIRES_DEV,
1197 };
1198
1199 static int __init init_nilfs_fs(void)
1200 {
1201         int err;
1202
1203         err = nilfs_init_inode_cache();
1204         if (err)
1205                 goto failed;
1206
1207         err = nilfs_init_transaction_cache();
1208         if (err)
1209                 goto failed_inode_cache;
1210
1211         err = nilfs_init_segbuf_cache();
1212         if (err)
1213                 goto failed_transaction_cache;
1214
1215         err = nilfs_btree_path_cache_init();
1216         if (err)
1217                 goto failed_segbuf_cache;
1218
1219         err = register_filesystem(&nilfs_fs_type);
1220         if (err)
1221                 goto failed_btree_path_cache;
1222
1223         return 0;
1224
1225  failed_btree_path_cache:
1226         nilfs_btree_path_cache_destroy();
1227
1228  failed_segbuf_cache:
1229         nilfs_destroy_segbuf_cache();
1230
1231  failed_transaction_cache:
1232         nilfs_destroy_transaction_cache();
1233
1234  failed_inode_cache:
1235         nilfs_destroy_inode_cache();
1236
1237  failed:
1238         return err;
1239 }
1240
1241 static void __exit exit_nilfs_fs(void)
1242 {
1243         nilfs_destroy_segbuf_cache();
1244         nilfs_destroy_transaction_cache();
1245         nilfs_destroy_inode_cache();
1246         nilfs_btree_path_cache_destroy();
1247         unregister_filesystem(&nilfs_fs_type);
1248 }
1249
1250 module_init(init_nilfs_fs)
1251 module_exit(exit_nilfs_fs)