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         down_read(&nilfs->ns_segctor_sem);
 420         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 421                                           &bh_cp);
 422         up_read(&nilfs->ns_segctor_sem);
 423         if (unlikely(err)) {
 424                 if (err == -ENOENT || err == -EINVAL) {
 425                         printk(KERN_ERR
 426                                "NILFS: Invalid checkpoint "
 427                                "(checkpoint number=%llu)\n",
 428                                (unsigned long long)cno);
 429                         err = -EINVAL;
 430                 }
 431                 goto failed;
 432         }
 433         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
 434         if (unlikely(err))
 435                 goto failed_bh;
 436         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 437         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 438
 439         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 440         return 0;
 441
 442  failed_bh:
 443         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 444  failed:
 445         nilfs_mdt_destroy(sbi->s_ifile);
 446         sbi->s_ifile = NULL;
 447
 448         down_write(&nilfs->ns_super_sem);
 449         list_del_init(&sbi->s_list);
 450         up_write(&nilfs->ns_super_sem);
 451
 452         return err;
 453 }
 454
 455 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
 456 {
 457         struct the_nilfs *nilfs = sbi->s_nilfs;
 458
 459         nilfs_mdt_clear(sbi->s_ifile);
 460         nilfs_mdt_destroy(sbi->s_ifile);
 461         sbi->s_ifile = NULL;
 462         down_write(&nilfs->ns_super_sem);
 463         list_del_init(&sbi->s_list);
 464         up_write(&nilfs->ns_super_sem);
 465 }
 466
 467 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
 468 {
 469         struct the_nilfs *nilfs = sbi->s_nilfs;
 470         int err = 0;
 471
 472         down_write(&nilfs->ns_sem);
 473         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
 474                 nilfs->ns_mount_state |= NILFS_VALID_FS;
 475                 err = nilfs_commit_super(sbi, 1);
 476                 if (likely(!err))
 477                         printk(KERN_INFO "NILFS: recovery complete.\n");
 478         }
 479         up_write(&nilfs->ns_sem);
 480         return err;
 481 }
 482
 483 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 484 {
 485         struct super_block *sb = dentry->d_sb;
 486         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 487         struct the_nilfs *nilfs = sbi->s_nilfs;
 488         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 489         unsigned long long blocks;
 490         unsigned long overhead;
 491         unsigned long nrsvblocks;
 492         sector_t nfreeblocks;
 493         int err;
 494
 495         /*
 496          * Compute all of the segment blocks
 497          *
 498          * The blocks before first segment and after last segment
 499          * are excluded.
 500          */
 501         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 502                 - nilfs->ns_first_data_block;
 503         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 504
 505         /*
 506          * Compute the overhead
 507          *
 508          * When distributing meta data blocks outside semgent structure,
 509          * We must count them as the overhead.
 510          */
 511         overhead = 0;
 512
 513         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 514         if (unlikely(err))
 515                 return err;
 516
 517         buf->f_type = NILFS_SUPER_MAGIC;
 518         buf->f_bsize = sb->s_blocksize;
 519         buf->f_blocks = blocks - overhead;
 520         buf->f_bfree = nfreeblocks;
 521         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 522                 (buf->f_bfree - nrsvblocks) : 0;
 523         buf->f_files = atomic_read(&sbi->s_inodes_count);
 524         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 525         buf->f_namelen = NILFS_NAME_LEN;
 526         buf->f_fsid.val[0] = (u32)id;
 527         buf->f_fsid.val[1] = (u32)(id >> 32);
 528
 529         return 0;
 530 }
 531
 532 static struct super_operations nilfs_sops = {
 533         .alloc_inode    = nilfs_alloc_inode,
 534         .destroy_inode  = nilfs_destroy_inode,
 535         .dirty_inode    = nilfs_dirty_inode,
 536         /* .write_inode    = nilfs_write_inode, */
 537         /* .put_inode      = nilfs_put_inode, */
 538         /* .drop_inode    = nilfs_drop_inode, */
 539         .delete_inode   = nilfs_delete_inode,
 540         .put_super      = nilfs_put_super,
 541         .write_super    = nilfs_write_super,
 542         .sync_fs        = nilfs_sync_fs,
 543         /* .write_super_lockfs */
 544         /* .unlockfs */
 545         .statfs         = nilfs_statfs,
 546         .remount_fs     = nilfs_remount,
 547         .clear_inode    = nilfs_clear_inode,
 548         /* .umount_begin */
 549         /* .show_options */
 550 };
 551
 552 static struct inode *
 553 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
 554 {
 555         struct inode *inode;
 556
 557         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
 558             ino != NILFS_SKETCH_INO)
 559                 return ERR_PTR(-ESTALE);
 560
 561         inode = nilfs_iget(sb, ino);
 562         if (IS_ERR(inode))
 563                 return ERR_CAST(inode);
 564         if (generation && inode->i_generation != generation) {
 565                 iput(inode);
 566                 return ERR_PTR(-ESTALE);
 567         }
 568
 569         return inode;
 570 }
 571
 572 static struct dentry *
 573 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
 574                    int fh_type)
 575 {
 576         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 577                                     nilfs_nfs_get_inode);
 578 }
 579
 580 static struct dentry *
 581 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
 582                    int fh_type)
 583 {
 584         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 585                                     nilfs_nfs_get_inode);
 586 }
 587
 588 static struct export_operations nilfs_export_ops = {
 589         .fh_to_dentry = nilfs_fh_to_dentry,
 590         .fh_to_parent = nilfs_fh_to_parent,
 591         .get_parent = nilfs_get_parent,
 592 };
 593
 594 enum {
 595         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 596         Opt_barrier, Opt_snapshot, Opt_order,
 597         Opt_err,
 598 };
 599
 600 static match_table_t tokens = {
 601         {Opt_err_cont, "errors=continue"},
 602         {Opt_err_panic, "errors=panic"},
 603         {Opt_err_ro, "errors=remount-ro"},
 604         {Opt_barrier, "barrier=%s"},
 605         {Opt_snapshot, "cp=%u"},
 606         {Opt_order, "order=%s"},
 607         {Opt_err, NULL}
 608 };
 609
 610 static int match_bool(substring_t *s, int *result)
 611 {
 612         int len = s->to - s->from;
 613
 614         if (strncmp(s->from, "on", len) == 0)
 615                 *result = 1;
 616         else if (strncmp(s->from, "off", len) == 0)
 617                 *result = 0;
 618         else
 619                 return 1;
 620         return 0;
 621 }
 622
 623 static int parse_options(char *options, struct super_block *sb)
 624 {
 625         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 626         char *p;
 627         substring_t args[MAX_OPT_ARGS];
 628         int option;
 629
 630         if (!options)
 631                 return 1;
 632
 633         while ((p = strsep(&options, ",")) != NULL) {
 634                 int token;
 635                 if (!*p)
 636                         continue;
 637
 638                 token = match_token(p, tokens, args);
 639                 switch (token) {
 640                 case Opt_barrier:
 641                         if (match_bool(&args[0], &option))
 642                                 return 0;
 643                         if (option)
 644                                 nilfs_set_opt(sbi, BARRIER);
 645                         else
 646                                 nilfs_clear_opt(sbi, BARRIER);
 647                         break;
 648                 case Opt_order:
 649                         if (strcmp(args[0].from, "relaxed") == 0)
 650                                 /* Ordered data semantics */
 651                                 nilfs_clear_opt(sbi, STRICT_ORDER);
 652                         else if (strcmp(args[0].from, "strict") == 0)
 653                                 /* Strict in-order semantics */
 654                                 nilfs_set_opt(sbi, STRICT_ORDER);
 655                         else
 656                                 return 0;
 657                         break;
 658                 case Opt_err_panic:
 659                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
 660                         break;
 661                 case Opt_err_ro:
 662                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
 663                         break;
 664                 case Opt_err_cont:
 665                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
 666                         break;
 667                 case Opt_snapshot:
 668                         if (match_int(&args[0], &option) || option <= 0)
 669                                 return 0;
 670                         if (!(sb->s_flags & MS_RDONLY))
 671                                 return 0;
 672                         sbi->s_snapshot_cno = option;
 673                         nilfs_set_opt(sbi, SNAPSHOT);
 674                         break;
 675                 default:
 676                         printk(KERN_ERR
 677                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 678                         return 0;
 679                 }
 680         }
 681         return 1;
 682 }
 683
 684 static inline void
 685 nilfs_set_default_options(struct nilfs_sb_info *sbi,
 686                           struct nilfs_super_block *sbp)
 687 {
 688         sbi->s_mount_opt =
 689                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
 690 }
 691
 692 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
 693 {
 694         struct the_nilfs *nilfs = sbi->s_nilfs;
 695         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 696         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
 697         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
 698
 699         /* nilfs->sem must be locked by the caller. */
 700         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
 701                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 702         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 703                 printk(KERN_WARNING
 704                        "NILFS warning: mounting fs with errors\n");
 705 #if 0
 706         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 707                 printk(KERN_WARNING
 708                        "NILFS warning: maximal mount count reached\n");
 709 #endif
 710         }
 711         if (!max_mnt_count)
 712                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 713
 714         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
 715         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
 716         sbp->s_mtime = cpu_to_le64(get_seconds());
 717         return nilfs_commit_super(sbi, 1);
 718 }
 719
 720 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 721                                                  u64 pos, int blocksize,
 722                                                  struct buffer_head **pbh)
 723 {
 724         unsigned long long sb_index = pos;
 725         unsigned long offset;
 726
 727         offset = do_div(sb_index, blocksize);
 728         *pbh = sb_bread(sb, sb_index);
 729         if (!*pbh)
 730                 return NULL;
 731         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 732 }
 733
 734 int nilfs_store_magic_and_option(struct super_block *sb,
 735                                  struct nilfs_super_block *sbp,
 736                                  char *data)
 737 {
 738         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 739
 740         sb->s_magic = le16_to_cpu(sbp->s_magic);
 741
 742         /* FS independent flags */
 743 #ifdef NILFS_ATIME_DISABLE
 744         sb->s_flags |= MS_NOATIME;
 745 #endif
 746
 747         nilfs_set_default_options(sbi, sbp);
 748
 749         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
 750         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
 751         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
 752         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
 753
 754         return !parse_options(data, sb) ? -EINVAL : 0 ;
 755 }
 756
 757 /**
 758  * nilfs_fill_super() - initialize a super block instance
 759  * @sb: super_block
 760  * @data: mount options
 761  * @silent: silent mode flag
 762  * @nilfs: the_nilfs struct
 763  *
 764  * This function is called exclusively by nilfs->ns_mount_mutex.
 765  * So, the recovery process is protected from other simultaneous mounts.
 766  */
 767 static int
 768 nilfs_fill_super(struct super_block *sb, void *data, int silent,
 769                  struct the_nilfs *nilfs)
 770 {
 771         struct nilfs_sb_info *sbi;
 772         struct inode *root;
 773         __u64 cno;
 774         int err;
 775
 776         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 777         if (!sbi)
 778                 return -ENOMEM;
 779
 780         sb->s_fs_info = sbi;
 781
 782         get_nilfs(nilfs);
 783         sbi->s_nilfs = nilfs;
 784         sbi->s_super = sb;
 785         atomic_set(&sbi->s_count, 1);
 786
 787         err = init_nilfs(nilfs, sbi, (char *)data);
 788         if (err)
 789                 goto failed_sbi;
 790
 791         spin_lock_init(&sbi->s_inode_lock);
 792         INIT_LIST_HEAD(&sbi->s_dirty_files);
 793         INIT_LIST_HEAD(&sbi->s_list);
 794
 795         /*
 796          * Following initialization is overlapped because
 797          * nilfs_sb_info structure has been cleared at the beginning.
 798          * But we reserve them to keep our interest and make ready
 799          * for the future change.
 800          */
 801         get_random_bytes(&sbi->s_next_generation,
 802                          sizeof(sbi->s_next_generation));
 803         spin_lock_init(&sbi->s_next_gen_lock);
 804
 805         sb->s_op = &nilfs_sops;
 806         sb->s_export_op = &nilfs_export_ops;
 807         sb->s_root = NULL;
 808         sb->s_time_gran = 1;
 809
 810         if (!nilfs_loaded(nilfs)) {
 811                 err = load_nilfs(nilfs, sbi);
 812                 if (err)
 813                         goto failed_sbi;
 814         }
 815         cno = nilfs_last_cno(nilfs);
 816
 817         if (sb->s_flags & MS_RDONLY) {
 818                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
 819                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
 820                                                        sbi->s_snapshot_cno);
 821                         if (err < 0)
 822                                 goto failed_sbi;
 823                         if (!err) {
 824                                 printk(KERN_ERR
 825                                        "NILFS: The specified checkpoint is "
 826                                        "not a snapshot "
 827                                        "(checkpoint number=%llu).\n",
 828                                        (unsigned long long)sbi->s_snapshot_cno);
 829                                 err = -EINVAL;
 830                                 goto failed_sbi;
 831                         }
 832                         cno = sbi->s_snapshot_cno;
 833                 } else
 834                         /* Read-only mount */
 835                         sbi->s_snapshot_cno = cno;
 836         }
 837
 838         err = nilfs_attach_checkpoint(sbi, cno);
 839         if (err) {
 840                 printk(KERN_ERR "NILFS: error loading a checkpoint"
 841                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
 842                 goto failed_sbi;
 843         }
 844
 845         if (!(sb->s_flags & MS_RDONLY)) {
 846                 err = nilfs_attach_segment_constructor(sbi);
 847                 if (err)
 848                         goto failed_checkpoint;
 849         }
 850
 851         root = nilfs_iget(sb, NILFS_ROOT_INO);
 852         if (IS_ERR(root)) {
 853                 printk(KERN_ERR "NILFS: get root inode failed\n");
 854                 err = PTR_ERR(root);
 855                 goto failed_segctor;
 856         }
 857         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
 858                 iput(root);
 859                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 860                 err = -EINVAL;
 861                 goto failed_segctor;
 862         }
 863         sb->s_root = d_alloc_root(root);
 864         if (!sb->s_root) {
 865                 iput(root);
 866                 printk(KERN_ERR "NILFS: get root dentry failed\n");
 867                 err = -ENOMEM;
 868                 goto failed_segctor;
 869         }
 870
 871         if (!(sb->s_flags & MS_RDONLY)) {
 872                 down_write(&nilfs->ns_sem);
 873                 nilfs_setup_super(sbi);
 874                 up_write(&nilfs->ns_sem);
 875         }
 876
 877         err = nilfs_mark_recovery_complete(sbi);
 878         if (unlikely(err)) {
 879                 printk(KERN_ERR "NILFS: recovery failed.\n");
 880                 goto failed_root;
 881         }
 882
 883         down_write(&nilfs->ns_super_sem);
 884         if (!nilfs_test_opt(sbi, SNAPSHOT))
 885                 nilfs->ns_current = sbi;
 886         up_write(&nilfs->ns_super_sem);
 887
 888         return 0;
 889
 890  failed_root:
 891         dput(sb->s_root);
 892         sb->s_root = NULL;
 893
 894  failed_segctor:
 895         nilfs_detach_segment_constructor(sbi);
 896
 897  failed_checkpoint:
 898         nilfs_detach_checkpoint(sbi);
 899
 900  failed_sbi:
 901         put_nilfs(nilfs);
 902         sb->s_fs_info = NULL;
 903         nilfs_put_sbinfo(sbi);
 904         return err;
 905 }
 906
 907 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
 908 {
 909         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 910         struct nilfs_super_block *sbp;
 911         struct the_nilfs *nilfs = sbi->s_nilfs;
 912         unsigned long old_sb_flags;
 913         struct nilfs_mount_options old_opts;
 914         int err;
 915
 916         lock_kernel();
 917
 918         down_write(&nilfs->ns_super_sem);
 919         old_sb_flags = sb->s_flags;
 920         old_opts.mount_opt = sbi->s_mount_opt;
 921         old_opts.snapshot_cno = sbi->s_snapshot_cno;
 922
 923         if (!parse_options(data, sb)) {
 924                 err = -EINVAL;
 925                 goto restore_opts;
 926         }
 927         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
 928
 929         if ((*flags & MS_RDONLY) &&
 930             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
 931                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 932                        "remount to a different snapshot. \n",
 933                        sb->s_id);
 934                 err = -EINVAL;
 935                 goto restore_opts;
 936         }
 937
 938         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 939                 goto out;
 940         if (*flags & MS_RDONLY) {
 941                 /* Shutting down the segment constructor */
 942                 nilfs_detach_segment_constructor(sbi);
 943                 sb->s_flags |= MS_RDONLY;
 944
 945                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
 946                 /* nilfs_set_opt(sbi, SNAPSHOT); */
 947
 948                 /*
 949                  * Remounting a valid RW partition RDONLY, so set
 950                  * the RDONLY flag and then mark the partition as valid again.
 951                  */
 952                 down_write(&nilfs->ns_sem);
 953                 sbp = nilfs->ns_sbp[0];
 954                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
 955                     (nilfs->ns_mount_state & NILFS_VALID_FS))
 956                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
 957                 sbp->s_mtime = cpu_to_le64(get_seconds());
 958                 nilfs_commit_super(sbi, 1);
 959                 up_write(&nilfs->ns_sem);
 960         } else {
 961                 /*
 962                  * Mounting a RDONLY partition read-write, so reread and
 963                  * store the current valid flag.  (It may have been changed
 964                  * by fsck since we originally mounted the partition.)
 965                  */
 966                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
 967                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 968                                "remount because an RW-mount exists.\n",
 969                                sb->s_id);
 970                         err = -EBUSY;
 971                         goto restore_opts;
 972                 }
 973                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
 974                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 975                                "remount because the current RO-mount is not "
 976                                "the latest one.\n",
 977                                sb->s_id);
 978                         err = -EINVAL;
 979                         goto restore_opts;
 980                 }
 981                 sb->s_flags &= ~MS_RDONLY;
 982                 nilfs_clear_opt(sbi, SNAPSHOT);
 983                 sbi->s_snapshot_cno = 0;
 984
 985                 err = nilfs_attach_segment_constructor(sbi);
 986                 if (err)
 987                         goto restore_opts;
 988
 989                 down_write(&nilfs->ns_sem);
 990                 nilfs_setup_super(sbi);
 991                 up_write(&nilfs->ns_sem);
 992
 993                 nilfs->ns_current = sbi;
 994         }
 995  out:
 996         up_write(&nilfs->ns_super_sem);
 997         unlock_kernel();
 998         return 0;
 999
1000  restore_opts:
1001         sb->s_flags = old_sb_flags;
1002         sbi->s_mount_opt = old_opts.mount_opt;
1003         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1004         up_write(&nilfs->ns_super_sem);
1005         unlock_kernel();
1006         return err;
1007 }
1008
1009 struct nilfs_super_data {
1010         struct block_device *bdev;
1011         struct nilfs_sb_info *sbi;
1012         __u64 cno;
1013         int flags;
1014 };
1015
1016 /**
1017  * nilfs_identify - pre-read mount options needed to identify mount instance
1018  * @data: mount options
1019  * @sd: nilfs_super_data
1020  */
1021 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1022 {
1023         char *p, *options = data;
1024         substring_t args[MAX_OPT_ARGS];
1025         int option, token;
1026         int ret = 0;
1027
1028         do {
1029                 p = strsep(&options, ",");
1030                 if (p != NULL && *p) {
1031                         token = match_token(p, tokens, args);
1032                         if (token == Opt_snapshot) {
1033                                 if (!(sd->flags & MS_RDONLY))
1034                                         ret++;
1035                                 else {
1036                                         ret = match_int(&args[0], &option);
1037                                         if (!ret) {
1038                                                 if (option > 0)
1039                                                         sd->cno = option;
1040                                                 else
1041                                                         ret++;
1042                                         }
1043                                 }
1044                         }
1045                         if (ret)
1046                                 printk(KERN_ERR
1047                                        "NILFS: invalid mount option: %s\n", p);
1048                 }
1049                 if (!options)
1050                         break;
1051                 BUG_ON(options == data);
1052                 *(options - 1) = ',';
1053         } while (!ret);
1054         return ret;
1055 }
1056
1057 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1058 {
1059         struct nilfs_super_data *sd = data;
1060
1061         s->s_bdev = sd->bdev;
1062         s->s_dev = s->s_bdev->bd_dev;
1063         return 0;
1064 }
1065
1066 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1067 {
1068         struct nilfs_super_data *sd = data;
1069
1070         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1071 }
1072
1073 static int
1074 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1075              const char *dev_name, void *data, struct vfsmount *mnt)
1076 {
1077         struct nilfs_super_data sd;
1078         struct super_block *s;
1079         struct the_nilfs *nilfs;
1080         int err, need_to_close = 1;
1081
1082         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1083         if (IS_ERR(sd.bdev))
1084                 return PTR_ERR(sd.bdev);
1085
1086         /*
1087          * To get mount instance using sget() vfs-routine, NILFS needs
1088          * much more information than normal filesystems to identify mount
1089          * instance.  For snapshot mounts, not only a mount type (ro-mount
1090          * or rw-mount) but also a checkpoint number is required.
1091          */
1092         sd.cno = 0;
1093         sd.flags = flags;
1094         if (nilfs_identify((char *)data, &sd)) {
1095                 err = -EINVAL;
1096                 goto failed;
1097         }
1098
1099         nilfs = find_or_create_nilfs(sd.bdev);
1100         if (!nilfs) {
1101                 err = -ENOMEM;
1102                 goto failed;
1103         }
1104
1105         mutex_lock(&nilfs->ns_mount_mutex);
1106
1107         if (!sd.cno) {
1108                 /*
1109                  * Check if an exclusive mount exists or not.
1110                  * Snapshot mounts coexist with a current mount
1111                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1112                  * ro-mount are mutually exclusive.
1113                  */
1114                 down_read(&nilfs->ns_super_sem);
1115                 if (nilfs->ns_current &&
1116                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1117                      & MS_RDONLY)) {
1118                         up_read(&nilfs->ns_super_sem);
1119                         err = -EBUSY;
1120                         goto failed_unlock;
1121                 }
1122                 up_read(&nilfs->ns_super_sem);
1123         }
1124
1125         /*
1126          * Find existing nilfs_sb_info struct
1127          */
1128         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1129
1130         if (!sd.cno)
1131                 /* trying to get the latest checkpoint.  */
1132                 sd.cno = nilfs_last_cno(nilfs);
1133
1134         /*
1135          * Get super block instance holding the nilfs_sb_info struct.
1136          * A new instance is allocated if no existing mount is present or
1137          * existing instance has been unmounted.
1138          */
1139         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1140         if (sd.sbi)
1141                 nilfs_put_sbinfo(sd.sbi);
1142
1143         if (IS_ERR(s)) {
1144                 err = PTR_ERR(s);
1145                 goto failed_unlock;
1146         }
1147
1148         if (!s->s_root) {
1149                 char b[BDEVNAME_SIZE];
1150
1151                 /* New superblock instance created */
1152                 s->s_flags = flags;
1153                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1154                 sb_set_blocksize(s, block_size(sd.bdev));
1155
1156                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1157                 if (err)
1158                         goto cancel_new;
1159
1160                 s->s_flags |= MS_ACTIVE;
1161                 need_to_close = 0;
1162         }
1163
1164         mutex_unlock(&nilfs->ns_mount_mutex);
1165         put_nilfs(nilfs);
1166         if (need_to_close)
1167                 close_bdev_exclusive(sd.bdev, flags);
1168         simple_set_mnt(mnt, s);
1169         return 0;
1170
1171  failed_unlock:
1172         mutex_unlock(&nilfs->ns_mount_mutex);
1173         put_nilfs(nilfs);
1174  failed:
1175         close_bdev_exclusive(sd.bdev, flags);
1176
1177         return err;
1178
1179  cancel_new:
1180         /* Abandoning the newly allocated superblock */
1181         mutex_unlock(&nilfs->ns_mount_mutex);
1182         put_nilfs(nilfs);
1183         up_write(&s->s_umount);
1184         deactivate_super(s);
1185         /*
1186          * deactivate_super() invokes close_bdev_exclusive().
1187          * We must finish all post-cleaning before this call;
1188          * put_nilfs() needs the block device.
1189          */
1190         return err;
1191 }
1192
1193 struct file_system_type nilfs_fs_type = {
1194         .owner    = THIS_MODULE,
1195         .name     = "nilfs2",
1196         .get_sb   = nilfs_get_sb,
1197         .kill_sb  = kill_block_super,
1198         .fs_flags = FS_REQUIRES_DEV,
1199 };
1200
1201 static int __init init_nilfs_fs(void)
1202 {
1203         int err;
1204
1205         err = nilfs_init_inode_cache();
1206         if (err)
1207                 goto failed;
1208
1209         err = nilfs_init_transaction_cache();
1210         if (err)
1211                 goto failed_inode_cache;
1212
1213         err = nilfs_init_segbuf_cache();
1214         if (err)
1215                 goto failed_transaction_cache;
1216
1217         err = nilfs_btree_path_cache_init();
1218         if (err)
1219                 goto failed_segbuf_cache;
1220
1221         err = register_filesystem(&nilfs_fs_type);
1222         if (err)
1223                 goto failed_btree_path_cache;
1224
1225         return 0;
1226
1227  failed_btree_path_cache:
1228         nilfs_btree_path_cache_destroy();
1229
1230  failed_segbuf_cache:
1231         nilfs_destroy_segbuf_cache();
1232
1233  failed_transaction_cache:
1234         nilfs_destroy_transaction_cache();
1235
1236  failed_inode_cache:
1237         nilfs_destroy_inode_cache();
1238
1239  failed:
1240         return err;
1241 }
1242
1243 static void __exit exit_nilfs_fs(void)
1244 {
1245         nilfs_destroy_segbuf_cache();
1246         nilfs_destroy_transaction_cache();
1247         nilfs_destroy_inode_cache();
1248         nilfs_btree_path_cache_destroy();
1249         unregister_filesystem(&nilfs_fs_type);
1250 }
1251
1252 module_init(init_nilfs_fs)
1253 module_exit(exit_nilfs_fs)