SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/nilfs2/super.c

   1 /*
   2  * super.c - NILFS module and super block management.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  */
  22 /*
  23  *  linux/fs/ext2/super.c
  24  *
  25  * Copyright (C) 1992, 1993, 1994, 1995
  26  * Remy Card (card@masi.ibp.fr)
  27  * Laboratoire MASI - Institut Blaise Pascal
  28  * Universite Pierre et Marie Curie (Paris VI)
  29  *
  30  *  from
  31  *
  32  *  linux/fs/minix/inode.c
  33  *
  34  *  Copyright (C) 1991, 1992  Linus Torvalds
  35  *
  36  *  Big-endian to little-endian byte-swapping/bitmaps by
  37  *        David S. Miller (davem@caip.rutgers.edu), 1995
  38  */
  39
  40 #include <linux/module.h>
  41 #include <linux/string.h>
  42 #include <linux/slab.h>
  43 #include <linux/init.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/parser.h>
  46 #include <linux/random.h>
  47 #include <linux/crc32.h>
  48 #include <linux/smp_lock.h>
  49 #include <linux/vfs.h>
  50 #include <linux/writeback.h>
  51 #include <linux/kobject.h>
  52 #include <linux/exportfs.h>
  53 #include <linux/seq_file.h>
  54 #include <linux/mount.h>
  55 #include "nilfs.h"
  56 #include "mdt.h"
  57 #include "alloc.h"
  58 #include "page.h"
  59 #include "cpfile.h"
  60 #include "ifile.h"
  61 #include "dat.h"
  62 #include "segment.h"
  63 #include "segbuf.h"
  64
  65 MODULE_AUTHOR("NTT Corp.");
  66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  67                    "(NILFS)");
  68 MODULE_LICENSE("GPL");
  69
  70 struct kmem_cache *nilfs_inode_cachep;
  71 struct kmem_cache *nilfs_transaction_cachep;
  72 struct kmem_cache *nilfs_segbuf_cachep;
  73 struct kmem_cache *nilfs_btree_path_cache;
  74
  75 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  76
  77 /**
  78  * nilfs_error() - report failure condition on a filesystem
  79  *
  80  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
  81  * reporting an error message.  It should be called when NILFS detects
  82  * incoherences or defects of meta data on disk.  As for sustainable
  83  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
  84  * function should be used instead.
  85  *
  86  * The segment constructor must not call this function because it can
  87  * kill itself.
  88  */
  89 void nilfs_error(struct super_block *sb, const char *function,
  90                  const char *fmt, ...)
  91 {
  92         struct nilfs_sb_info *sbi = NILFS_SB(sb);
  93         va_list args;
  94
  95         va_start(args, fmt);
  96         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
  97         vprintk(fmt, args);
  98         printk("\n");
  99         va_end(args);
 100
 101         if (!(sb->s_flags & MS_RDONLY)) {
 102                 struct the_nilfs *nilfs = sbi->s_nilfs;
 103
 104                 down_write(&nilfs->ns_sem);
 105                 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
 106                         nilfs->ns_mount_state |= NILFS_ERROR_FS;
 107                         nilfs->ns_sbp[0]->s_state |=
 108                                 cpu_to_le16(NILFS_ERROR_FS);
 109                         nilfs_commit_super(sbi, 1);
 110                 }
 111                 up_write(&nilfs->ns_sem);
 112
 113                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
 114                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 115                         sb->s_flags |= MS_RDONLY;
 116                 }
 117         }
 118
 119         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 120                 panic("NILFS (device %s): panic forced after error\n",
 121                       sb->s_id);
 122 }
 123
 124 void nilfs_warning(struct super_block *sb, const char *function,
 125                    const char *fmt, ...)
 126 {
 127         va_list args;
 128
 129         va_start(args, fmt);
 130         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
 131                sb->s_id, function);
 132         vprintk(fmt, args);
 133         printk("\n");
 134         va_end(args);
 135 }
 136
 137
 138 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
 139 {
 140         struct nilfs_inode_info *ii;
 141
 142         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 143         if (!ii)
 144                 return NULL;
 145         ii->i_bh = NULL;
 146         ii->i_state = 0;
 147         ii->vfs_inode.i_version = 1;
 148         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
 149         return &ii->vfs_inode;
 150 }
 151
 152 struct inode *nilfs_alloc_inode(struct super_block *sb)
 153 {
 154         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
 155 }
 156
 157 void nilfs_destroy_inode(struct inode *inode)
 158 {
 159         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 160 }
 161
 162 static void nilfs_clear_inode(struct inode *inode)
 163 {
 164         struct nilfs_inode_info *ii = NILFS_I(inode);
 165
 166         /*
 167          * Free resources allocated in nilfs_read_inode(), here.
 168          */
 169         BUG_ON(!list_empty(&ii->i_dirty));
 170         brelse(ii->i_bh);
 171         ii->i_bh = NULL;
 172
 173         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 174                 nilfs_bmap_clear(ii->i_bmap);
 175
 176         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 177 }
 178
 179 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
 180 {
 181         struct the_nilfs *nilfs = sbi->s_nilfs;
 182         int err;
 183         int barrier_done = 0;
 184
 185         if (nilfs_test_opt(sbi, BARRIER)) {
 186                 set_buffer_ordered(nilfs->ns_sbh[0]);
 187                 barrier_done = 1;
 188         }
 189  retry:
 190         set_buffer_dirty(nilfs->ns_sbh[0]);
 191         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 192         if (err == -EOPNOTSUPP && barrier_done) {
 193                 nilfs_warning(sbi->s_super, __func__,
 194                               "barrier-based sync failed. "
 195                               "disabling barriers\n");
 196                 nilfs_clear_opt(sbi, BARRIER);
 197                 barrier_done = 0;
 198                 clear_buffer_ordered(nilfs->ns_sbh[0]);
 199                 goto retry;
 200         }
 201         if (unlikely(err)) {
 202                 printk(KERN_ERR
 203                        "NILFS: unable to write superblock (err=%d)\n", err);
 204                 if (err == -EIO && nilfs->ns_sbh[1]) {
 205                         nilfs_fall_back_super_block(nilfs);
 206                         goto retry;
 207                 }
 208         } else {
 209                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 210
 211                 /*
 212                  * The latest segment becomes trailable from the position
 213                  * written in superblock.
 214                  */
 215                 clear_nilfs_discontinued(nilfs);
 216
 217                 /* update GC protection for recent segments */
 218                 if (nilfs->ns_sbh[1]) {
 219                         sbp = NULL;
 220                         if (dupsb) {
 221                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 222                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
 223                                         sbp = nilfs->ns_sbp[1];
 224                         }
 225                 }
 226                 if (sbp) {
 227                         spin_lock(&nilfs->ns_last_segment_lock);
 228                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 229                         spin_unlock(&nilfs->ns_last_segment_lock);
 230                 }
 231         }
 232
 233         return err;
 234 }
 235
 236 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
 237 {
 238         struct the_nilfs *nilfs = sbi->s_nilfs;
 239         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 240         sector_t nfreeblocks;
 241         time_t t;
 242         int err;
 243
 244         /* nilfs->sem must be locked by the caller. */
 245         if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
 246                 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
 247                         nilfs_swap_super_block(nilfs);
 248                 else {
 249                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 250                                sbi->s_super->s_id);
 251                         return -EIO;
 252                 }
 253         }
 254         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 255         if (unlikely(err)) {
 256                 printk(KERN_ERR "NILFS: failed to count free blocks\n");
 257                 return err;
 258         }
 259         spin_lock(&nilfs->ns_last_segment_lock);
 260         sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 261         sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 262         sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 263         spin_unlock(&nilfs->ns_last_segment_lock);
 264
 265         t = get_seconds();
 266         nilfs->ns_sbwtime[0] = t;
 267         sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 268         sbp[0]->s_wtime = cpu_to_le64(t);
 269         sbp[0]->s_sum = 0;
 270         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 271                                              (unsigned char *)sbp[0],
 272                                              nilfs->ns_sbsize));
 273         if (dupsb && sbp[1]) {
 274                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 275                 nilfs->ns_sbwtime[1] = t;
 276         }
 277         clear_nilfs_sb_dirty(nilfs);
 278         return nilfs_sync_super(sbi, dupsb);
 279 }
 280
 281 static void nilfs_put_super(struct super_block *sb)
 282 {
 283         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 284         struct the_nilfs *nilfs = sbi->s_nilfs;
 285
 286         lock_kernel();
 287
 288         nilfs_detach_segment_constructor(sbi);
 289
 290         if (!(sb->s_flags & MS_RDONLY)) {
 291                 down_write(&nilfs->ns_sem);
 292                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 293                 nilfs_commit_super(sbi, 1);
 294                 up_write(&nilfs->ns_sem);
 295         }
 296         down_write(&nilfs->ns_super_sem);
 297         if (nilfs->ns_current == sbi)
 298                 nilfs->ns_current = NULL;
 299         up_write(&nilfs->ns_super_sem);
 300
 301         nilfs_detach_checkpoint(sbi);
 302         put_nilfs(sbi->s_nilfs);
 303         sbi->s_super = NULL;
 304         sb->s_fs_info = NULL;
 305         nilfs_put_sbinfo(sbi);
 306
 307         unlock_kernel();
 308 }
 309
 310 static int nilfs_sync_fs(struct super_block *sb, int wait)
 311 {
 312         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 313         struct the_nilfs *nilfs = sbi->s_nilfs;
 314         int err = 0;
 315
 316         /* This function is called when super block should be written back */
 317         if (wait)
 318                 err = nilfs_construct_segment(sb);
 319
 320         down_write(&nilfs->ns_sem);
 321         if (nilfs_sb_dirty(nilfs))
 322                 nilfs_commit_super(sbi, 1);
 323         up_write(&nilfs->ns_sem);
 324
 325         return err;
 326 }
 327
 328 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
 329 {
 330         struct the_nilfs *nilfs = sbi->s_nilfs;
 331         struct nilfs_checkpoint *raw_cp;
 332         struct buffer_head *bh_cp;
 333         int err;
 334
 335         down_write(&nilfs->ns_super_sem);
 336         list_add(&sbi->s_list, &nilfs->ns_supers);
 337         up_write(&nilfs->ns_super_sem);
 338
 339         sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
 340         if (!sbi->s_ifile)
 341                 return -ENOMEM;
 342
 343         down_read(&nilfs->ns_segctor_sem);
 344         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 345                                           &bh_cp);
 346         up_read(&nilfs->ns_segctor_sem);
 347         if (unlikely(err)) {
 348                 if (err == -ENOENT || err == -EINVAL) {
 349                         printk(KERN_ERR
 350                                "NILFS: Invalid checkpoint "
 351                                "(checkpoint number=%llu)\n",
 352                                (unsigned long long)cno);
 353                         err = -EINVAL;
 354                 }
 355                 goto failed;
 356         }
 357         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
 358         if (unlikely(err))
 359                 goto failed_bh;
 360         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 361         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 362
 363         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 364         return 0;
 365
 366  failed_bh:
 367         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 368  failed:
 369         nilfs_mdt_destroy(sbi->s_ifile);
 370         sbi->s_ifile = NULL;
 371
 372         down_write(&nilfs->ns_super_sem);
 373         list_del_init(&sbi->s_list);
 374         up_write(&nilfs->ns_super_sem);
 375
 376         return err;
 377 }
 378
 379 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
 380 {
 381         struct the_nilfs *nilfs = sbi->s_nilfs;
 382
 383         nilfs_mdt_destroy(sbi->s_ifile);
 384         sbi->s_ifile = NULL;
 385         down_write(&nilfs->ns_super_sem);
 386         list_del_init(&sbi->s_list);
 387         up_write(&nilfs->ns_super_sem);
 388 }
 389
 390 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 391 {
 392         struct super_block *sb = dentry->d_sb;
 393         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 394         struct the_nilfs *nilfs = sbi->s_nilfs;
 395         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 396         unsigned long long blocks;
 397         unsigned long overhead;
 398         unsigned long nrsvblocks;
 399         sector_t nfreeblocks;
 400         int err;
 401
 402         /*
 403          * Compute all of the segment blocks
 404          *
 405          * The blocks before first segment and after last segment
 406          * are excluded.
 407          */
 408         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 409                 - nilfs->ns_first_data_block;
 410         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 411
 412         /*
 413          * Compute the overhead
 414          *
 415          * When distributing meta data blocks outside segment structure,
 416          * We must count them as the overhead.
 417          */
 418         overhead = 0;
 419
 420         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 421         if (unlikely(err))
 422                 return err;
 423
 424         buf->f_type = NILFS_SUPER_MAGIC;
 425         buf->f_bsize = sb->s_blocksize;
 426         buf->f_blocks = blocks - overhead;
 427         buf->f_bfree = nfreeblocks;
 428         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 429                 (buf->f_bfree - nrsvblocks) : 0;
 430         buf->f_files = atomic_read(&sbi->s_inodes_count);
 431         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 432         buf->f_namelen = NILFS_NAME_LEN;
 433         buf->f_fsid.val[0] = (u32)id;
 434         buf->f_fsid.val[1] = (u32)(id >> 32);
 435
 436         return 0;
 437 }
 438
 439 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 440 {
 441         struct super_block *sb = vfs->mnt_sb;
 442         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 443
 444         if (!nilfs_test_opt(sbi, BARRIER))
 445                 seq_printf(seq, ",nobarrier");
 446         if (nilfs_test_opt(sbi, SNAPSHOT))
 447                 seq_printf(seq, ",cp=%llu",
 448                            (unsigned long long int)sbi->s_snapshot_cno);
 449         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 450                 seq_printf(seq, ",errors=panic");
 451         if (nilfs_test_opt(sbi, ERRORS_CONT))
 452                 seq_printf(seq, ",errors=continue");
 453         if (nilfs_test_opt(sbi, STRICT_ORDER))
 454                 seq_printf(seq, ",order=strict");
 455         if (nilfs_test_opt(sbi, NORECOVERY))
 456                 seq_printf(seq, ",norecovery");
 457         if (nilfs_test_opt(sbi, DISCARD))
 458                 seq_printf(seq, ",discard");
 459
 460         return 0;
 461 }
 462
 463 static const struct super_operations nilfs_sops = {
 464         .alloc_inode    = nilfs_alloc_inode,
 465         .destroy_inode  = nilfs_destroy_inode,
 466         .dirty_inode    = nilfs_dirty_inode,
 467         /* .write_inode    = nilfs_write_inode, */
 468         /* .put_inode      = nilfs_put_inode, */
 469         /* .drop_inode    = nilfs_drop_inode, */
 470         .delete_inode   = nilfs_delete_inode,
 471         .put_super      = nilfs_put_super,
 472         /* .write_super    = nilfs_write_super, */
 473         .sync_fs        = nilfs_sync_fs,
 474         /* .write_super_lockfs */
 475         /* .unlockfs */
 476         .statfs         = nilfs_statfs,
 477         .remount_fs     = nilfs_remount,
 478         .clear_inode    = nilfs_clear_inode,
 479         /* .umount_begin */
 480         .show_options = nilfs_show_options
 481 };
 482
 483 static struct inode *
 484 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
 485 {
 486         struct inode *inode;
 487
 488         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
 489             ino != NILFS_SKETCH_INO)
 490                 return ERR_PTR(-ESTALE);
 491
 492         inode = nilfs_iget(sb, ino);
 493         if (IS_ERR(inode))
 494                 return ERR_CAST(inode);
 495         if (generation && inode->i_generation != generation) {
 496                 iput(inode);
 497                 return ERR_PTR(-ESTALE);
 498         }
 499
 500         return inode;
 501 }
 502
 503 static struct dentry *
 504 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
 505                    int fh_type)
 506 {
 507         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 508                                     nilfs_nfs_get_inode);
 509 }
 510
 511 static struct dentry *
 512 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
 513                    int fh_type)
 514 {
 515         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 516                                     nilfs_nfs_get_inode);
 517 }
 518
 519 static const struct export_operations nilfs_export_ops = {
 520         .fh_to_dentry = nilfs_fh_to_dentry,
 521         .fh_to_parent = nilfs_fh_to_parent,
 522         .get_parent = nilfs_get_parent,
 523 };
 524
 525 enum {
 526         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 527         Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 528         Opt_discard, Opt_err,
 529 };
 530
 531 static match_table_t tokens = {
 532         {Opt_err_cont, "errors=continue"},
 533         {Opt_err_panic, "errors=panic"},
 534         {Opt_err_ro, "errors=remount-ro"},
 535         {Opt_nobarrier, "nobarrier"},
 536         {Opt_snapshot, "cp=%u"},
 537         {Opt_order, "order=%s"},
 538         {Opt_norecovery, "norecovery"},
 539         {Opt_discard, "discard"},
 540         {Opt_err, NULL}
 541 };
 542
 543 static int parse_options(char *options, struct super_block *sb)
 544 {
 545         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 546         char *p;
 547         substring_t args[MAX_OPT_ARGS];
 548         int option;
 549
 550         if (!options)
 551                 return 1;
 552
 553         while ((p = strsep(&options, ",")) != NULL) {
 554                 int token;
 555                 if (!*p)
 556                         continue;
 557
 558                 token = match_token(p, tokens, args);
 559                 switch (token) {
 560                 case Opt_nobarrier:
 561                         nilfs_clear_opt(sbi, BARRIER);
 562                         break;
 563                 case Opt_order:
 564                         if (strcmp(args[0].from, "relaxed") == 0)
 565                                 /* Ordered data semantics */
 566                                 nilfs_clear_opt(sbi, STRICT_ORDER);
 567                         else if (strcmp(args[0].from, "strict") == 0)
 568                                 /* Strict in-order semantics */
 569                                 nilfs_set_opt(sbi, STRICT_ORDER);
 570                         else
 571                                 return 0;
 572                         break;
 573                 case Opt_err_panic:
 574                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
 575                         break;
 576                 case Opt_err_ro:
 577                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
 578                         break;
 579                 case Opt_err_cont:
 580                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
 581                         break;
 582                 case Opt_snapshot:
 583                         if (match_int(&args[0], &option) || option <= 0)
 584                                 return 0;
 585                         if (!(sb->s_flags & MS_RDONLY))
 586                                 return 0;
 587                         sbi->s_snapshot_cno = option;
 588                         nilfs_set_opt(sbi, SNAPSHOT);
 589                         break;
 590                 case Opt_norecovery:
 591                         nilfs_set_opt(sbi, NORECOVERY);
 592                         break;
 593                 case Opt_discard:
 594                         nilfs_set_opt(sbi, DISCARD);
 595                         break;
 596                 default:
 597                         printk(KERN_ERR
 598                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 599                         return 0;
 600                 }
 601         }
 602         return 1;
 603 }
 604
 605 static inline void
 606 nilfs_set_default_options(struct nilfs_sb_info *sbi,
 607                           struct nilfs_super_block *sbp)
 608 {
 609         sbi->s_mount_opt =
 610                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 611 }
 612
 613 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
 614 {
 615         struct the_nilfs *nilfs = sbi->s_nilfs;
 616         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 617         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
 618         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
 619
 620         /* nilfs->sem must be locked by the caller. */
 621         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 622                 printk(KERN_WARNING
 623                        "NILFS warning: mounting fs with errors\n");
 624 #if 0
 625         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 626                 printk(KERN_WARNING
 627                        "NILFS warning: maximal mount count reached\n");
 628 #endif
 629         }
 630         if (!max_mnt_count)
 631                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 632
 633         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
 634         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
 635         sbp->s_mtime = cpu_to_le64(get_seconds());
 636         return nilfs_commit_super(sbi, 1);
 637 }
 638
 639 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 640                                                  u64 pos, int blocksize,
 641                                                  struct buffer_head **pbh)
 642 {
 643         unsigned long long sb_index = pos;
 644         unsigned long offset;
 645
 646         offset = do_div(sb_index, blocksize);
 647         *pbh = sb_bread(sb, sb_index);
 648         if (!*pbh)
 649                 return NULL;
 650         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 651 }
 652
 653 int nilfs_store_magic_and_option(struct super_block *sb,
 654                                  struct nilfs_super_block *sbp,
 655                                  char *data)
 656 {
 657         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 658
 659         sb->s_magic = le16_to_cpu(sbp->s_magic);
 660
 661         /* FS independent flags */
 662 #ifdef NILFS_ATIME_DISABLE
 663         sb->s_flags |= MS_NOATIME;
 664 #endif
 665
 666         nilfs_set_default_options(sbi, sbp);
 667
 668         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
 669         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
 670         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
 671         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
 672
 673         return !parse_options(data, sb) ? -EINVAL : 0 ;
 674 }
 675
 676 /**
 677  * nilfs_fill_super() - initialize a super block instance
 678  * @sb: super_block
 679  * @data: mount options
 680  * @silent: silent mode flag
 681  * @nilfs: the_nilfs struct
 682  *
 683  * This function is called exclusively by nilfs->ns_mount_mutex.
 684  * So, the recovery process is protected from other simultaneous mounts.
 685  */
 686 static int
 687 nilfs_fill_super(struct super_block *sb, void *data, int silent,
 688                  struct the_nilfs *nilfs)
 689 {
 690         struct nilfs_sb_info *sbi;
 691         struct inode *root;
 692         __u64 cno;
 693         int err;
 694
 695         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 696         if (!sbi)
 697                 return -ENOMEM;
 698
 699         sb->s_fs_info = sbi;
 700
 701         get_nilfs(nilfs);
 702         sbi->s_nilfs = nilfs;
 703         sbi->s_super = sb;
 704         atomic_set(&sbi->s_count, 1);
 705
 706         err = init_nilfs(nilfs, sbi, (char *)data);
 707         if (err)
 708                 goto failed_sbi;
 709
 710         spin_lock_init(&sbi->s_inode_lock);
 711         INIT_LIST_HEAD(&sbi->s_dirty_files);
 712         INIT_LIST_HEAD(&sbi->s_list);
 713
 714         /*
 715          * Following initialization is overlapped because
 716          * nilfs_sb_info structure has been cleared at the beginning.
 717          * But we reserve them to keep our interest and make ready
 718          * for the future change.
 719          */
 720         get_random_bytes(&sbi->s_next_generation,
 721                          sizeof(sbi->s_next_generation));
 722         spin_lock_init(&sbi->s_next_gen_lock);
 723
 724         sb->s_op = &nilfs_sops;
 725         sb->s_export_op = &nilfs_export_ops;
 726         sb->s_root = NULL;
 727         sb->s_time_gran = 1;
 728         sb->s_bdi = nilfs->ns_bdi;
 729
 730         err = load_nilfs(nilfs, sbi);
 731         if (err)
 732                 goto failed_sbi;
 733
 734         cno = nilfs_last_cno(nilfs);
 735
 736         if (sb->s_flags & MS_RDONLY) {
 737                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
 738                         down_read(&nilfs->ns_segctor_sem);
 739                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
 740                                                        sbi->s_snapshot_cno);
 741                         up_read(&nilfs->ns_segctor_sem);
 742                         if (err < 0) {
 743                                 if (err == -ENOENT)
 744                                         err = -EINVAL;
 745                                 goto failed_sbi;
 746                         }
 747                         if (!err) {
 748                                 printk(KERN_ERR
 749                                        "NILFS: The specified checkpoint is "
 750                                        "not a snapshot "
 751                                        "(checkpoint number=%llu).\n",
 752                                        (unsigned long long)sbi->s_snapshot_cno);
 753                                 err = -EINVAL;
 754                                 goto failed_sbi;
 755                         }
 756                         cno = sbi->s_snapshot_cno;
 757                 } else
 758                         /* Read-only mount */
 759                         sbi->s_snapshot_cno = cno;
 760         }
 761
 762         err = nilfs_attach_checkpoint(sbi, cno);
 763         if (err) {
 764                 printk(KERN_ERR "NILFS: error loading a checkpoint"
 765                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
 766                 goto failed_sbi;
 767         }
 768
 769         if (!(sb->s_flags & MS_RDONLY)) {
 770                 err = nilfs_attach_segment_constructor(sbi);
 771                 if (err)
 772                         goto failed_checkpoint;
 773         }
 774
 775         root = nilfs_iget(sb, NILFS_ROOT_INO);
 776         if (IS_ERR(root)) {
 777                 printk(KERN_ERR "NILFS: get root inode failed\n");
 778                 err = PTR_ERR(root);
 779                 goto failed_segctor;
 780         }
 781         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
 782                 iput(root);
 783                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 784                 err = -EINVAL;
 785                 goto failed_segctor;
 786         }
 787         sb->s_root = d_alloc_root(root);
 788         if (!sb->s_root) {
 789                 iput(root);
 790                 printk(KERN_ERR "NILFS: get root dentry failed\n");
 791                 err = -ENOMEM;
 792                 goto failed_segctor;
 793         }
 794
 795         if (!(sb->s_flags & MS_RDONLY)) {
 796                 down_write(&nilfs->ns_sem);
 797                 nilfs_setup_super(sbi);
 798                 up_write(&nilfs->ns_sem);
 799         }
 800
 801         down_write(&nilfs->ns_super_sem);
 802         if (!nilfs_test_opt(sbi, SNAPSHOT))
 803                 nilfs->ns_current = sbi;
 804         up_write(&nilfs->ns_super_sem);
 805
 806         return 0;
 807
 808  failed_segctor:
 809         nilfs_detach_segment_constructor(sbi);
 810
 811  failed_checkpoint:
 812         nilfs_detach_checkpoint(sbi);
 813
 814  failed_sbi:
 815         put_nilfs(nilfs);
 816         sb->s_fs_info = NULL;
 817         nilfs_put_sbinfo(sbi);
 818         return err;
 819 }
 820
 821 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
 822 {
 823         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 824         struct nilfs_super_block *sbp;
 825         struct the_nilfs *nilfs = sbi->s_nilfs;
 826         unsigned long old_sb_flags;
 827         struct nilfs_mount_options old_opts;
 828         int err;
 829
 830         lock_kernel();
 831
 832         down_write(&nilfs->ns_super_sem);
 833         old_sb_flags = sb->s_flags;
 834         old_opts.mount_opt = sbi->s_mount_opt;
 835         old_opts.snapshot_cno = sbi->s_snapshot_cno;
 836
 837         if (!parse_options(data, sb)) {
 838                 err = -EINVAL;
 839                 goto restore_opts;
 840         }
 841         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
 842
 843         if ((*flags & MS_RDONLY) &&
 844             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
 845                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 846                        "remount to a different snapshot.\n",
 847                        sb->s_id);
 848                 err = -EINVAL;
 849                 goto restore_opts;
 850         }
 851
 852         if (!nilfs_valid_fs(nilfs)) {
 853                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 854                        "remount because the filesystem is in an "
 855                        "incomplete recovery state.\n", sb->s_id);
 856                 err = -EINVAL;
 857                 goto restore_opts;
 858         }
 859
 860         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 861                 goto out;
 862         if (*flags & MS_RDONLY) {
 863                 /* Shutting down the segment constructor */
 864                 nilfs_detach_segment_constructor(sbi);
 865                 sb->s_flags |= MS_RDONLY;
 866
 867                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
 868                 /* nilfs_set_opt(sbi, SNAPSHOT); */
 869
 870                 /*
 871                  * Remounting a valid RW partition RDONLY, so set
 872                  * the RDONLY flag and then mark the partition as valid again.
 873                  */
 874                 down_write(&nilfs->ns_sem);
 875                 sbp = nilfs->ns_sbp[0];
 876                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
 877                     (nilfs->ns_mount_state & NILFS_VALID_FS))
 878                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
 879                 sbp->s_mtime = cpu_to_le64(get_seconds());
 880                 nilfs_commit_super(sbi, 1);
 881                 up_write(&nilfs->ns_sem);
 882         } else {
 883                 /*
 884                  * Mounting a RDONLY partition read-write, so reread and
 885                  * store the current valid flag.  (It may have been changed
 886                  * by fsck since we originally mounted the partition.)
 887                  */
 888                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
 889                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 890                                "remount because an RW-mount exists.\n",
 891                                sb->s_id);
 892                         err = -EBUSY;
 893                         goto restore_opts;
 894                 }
 895                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
 896                         printk(KERN_WARNING "NILFS (device %s): couldn't "
 897                                "remount because the current RO-mount is not "
 898                                "the latest one.\n",
 899                                sb->s_id);
 900                         err = -EINVAL;
 901                         goto restore_opts;
 902                 }
 903                 sb->s_flags &= ~MS_RDONLY;
 904                 nilfs_clear_opt(sbi, SNAPSHOT);
 905                 sbi->s_snapshot_cno = 0;
 906
 907                 err = nilfs_attach_segment_constructor(sbi);
 908                 if (err)
 909                         goto restore_opts;
 910
 911                 down_write(&nilfs->ns_sem);
 912                 nilfs_setup_super(sbi);
 913                 up_write(&nilfs->ns_sem);
 914
 915                 nilfs->ns_current = sbi;
 916         }
 917  out:
 918         up_write(&nilfs->ns_super_sem);
 919         unlock_kernel();
 920         return 0;
 921
 922  restore_opts:
 923         sb->s_flags = old_sb_flags;
 924         sbi->s_mount_opt = old_opts.mount_opt;
 925         sbi->s_snapshot_cno = old_opts.snapshot_cno;
 926         up_write(&nilfs->ns_super_sem);
 927         unlock_kernel();
 928         return err;
 929 }
 930
 931 struct nilfs_super_data {
 932         struct block_device *bdev;
 933         struct nilfs_sb_info *sbi;
 934         __u64 cno;
 935         int flags;
 936 };
 937
 938 /**
 939  * nilfs_identify - pre-read mount options needed to identify mount instance
 940  * @data: mount options
 941  * @sd: nilfs_super_data
 942  */
 943 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
 944 {
 945         char *p, *options = data;
 946         substring_t args[MAX_OPT_ARGS];
 947         int option, token;
 948         int ret = 0;
 949
 950         do {
 951                 p = strsep(&options, ",");
 952                 if (p != NULL && *p) {
 953                         token = match_token(p, tokens, args);
 954                         if (token == Opt_snapshot) {
 955                                 if (!(sd->flags & MS_RDONLY))
 956                                         ret++;
 957                                 else {
 958                                         ret = match_int(&args[0], &option);
 959                                         if (!ret) {
 960                                                 if (option > 0)
 961                                                         sd->cno = option;
 962                                                 else
 963                                                         ret++;
 964                                         }
 965                                 }
 966                         }
 967                         if (ret)
 968                                 printk(KERN_ERR
 969                                        "NILFS: invalid mount option: %s\n", p);
 970                 }
 971                 if (!options)
 972                         break;
 973                 BUG_ON(options == data);
 974                 *(options - 1) = ',';
 975         } while (!ret);
 976         return ret;
 977 }
 978
 979 static int nilfs_set_bdev_super(struct super_block *s, void *data)
 980 {
 981         struct nilfs_super_data *sd = data;
 982
 983         s->s_bdev = sd->bdev;
 984         s->s_dev = s->s_bdev->bd_dev;
 985         return 0;
 986 }
 987
 988 static int nilfs_test_bdev_super(struct super_block *s, void *data)
 989 {
 990         struct nilfs_super_data *sd = data;
 991
 992         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
 993 }
 994
 995 static int
 996 nilfs_get_sb(struct file_system_type *fs_type, int flags,
 997              const char *dev_name, void *data, struct vfsmount *mnt)
 998 {
 999         struct nilfs_super_data sd;
1000         struct super_block *s;
1001         struct the_nilfs *nilfs;
1002         int err, need_to_close = 1;
1003
1004         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1005         if (IS_ERR(sd.bdev))
1006                 return PTR_ERR(sd.bdev);
1007
1008         /*
1009          * To get mount instance using sget() vfs-routine, NILFS needs
1010          * much more information than normal filesystems to identify mount
1011          * instance.  For snapshot mounts, not only a mount type (ro-mount
1012          * or rw-mount) but also a checkpoint number is required.
1013          */
1014         sd.cno = 0;
1015         sd.flags = flags;
1016         if (nilfs_identify((char *)data, &sd)) {
1017                 err = -EINVAL;
1018                 goto failed;
1019         }
1020
1021         nilfs = find_or_create_nilfs(sd.bdev);
1022         if (!nilfs) {
1023                 err = -ENOMEM;
1024                 goto failed;
1025         }
1026
1027         mutex_lock(&nilfs->ns_mount_mutex);
1028
1029         if (!sd.cno) {
1030                 /*
1031                  * Check if an exclusive mount exists or not.
1032                  * Snapshot mounts coexist with a current mount
1033                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1034                  * ro-mount are mutually exclusive.
1035                  */
1036                 down_read(&nilfs->ns_super_sem);
1037                 if (nilfs->ns_current &&
1038                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1039                      & MS_RDONLY)) {
1040                         up_read(&nilfs->ns_super_sem);
1041                         err = -EBUSY;
1042                         goto failed_unlock;
1043                 }
1044                 up_read(&nilfs->ns_super_sem);
1045         }
1046
1047         /*
1048          * Find existing nilfs_sb_info struct
1049          */
1050         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1051
1052         /*
1053          * Get super block instance holding the nilfs_sb_info struct.
1054          * A new instance is allocated if no existing mount is present or
1055          * existing instance has been unmounted.
1056          */
1057         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1058         if (sd.sbi)
1059                 nilfs_put_sbinfo(sd.sbi);
1060
1061         if (IS_ERR(s)) {
1062                 err = PTR_ERR(s);
1063                 goto failed_unlock;
1064         }
1065
1066         if (!s->s_root) {
1067                 char b[BDEVNAME_SIZE];
1068
1069                 /* New superblock instance created */
1070                 s->s_flags = flags;
1071                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1072                 sb_set_blocksize(s, block_size(sd.bdev));
1073
1074                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1075                 if (err)
1076                         goto cancel_new;
1077
1078                 s->s_flags |= MS_ACTIVE;
1079                 need_to_close = 0;
1080         }
1081
1082         mutex_unlock(&nilfs->ns_mount_mutex);
1083         put_nilfs(nilfs);
1084         if (need_to_close)
1085                 close_bdev_exclusive(sd.bdev, flags);
1086         simple_set_mnt(mnt, s);
1087         return 0;
1088
1089  failed_unlock:
1090         mutex_unlock(&nilfs->ns_mount_mutex);
1091         put_nilfs(nilfs);
1092  failed:
1093         close_bdev_exclusive(sd.bdev, flags);
1094
1095         return err;
1096
1097  cancel_new:
1098         /* Abandoning the newly allocated superblock */
1099         mutex_unlock(&nilfs->ns_mount_mutex);
1100         put_nilfs(nilfs);
1101         deactivate_locked_super(s);
1102         /*
1103          * deactivate_super() invokes close_bdev_exclusive().
1104          * We must finish all post-cleaning before this call;
1105          * put_nilfs() needs the block device.
1106          */
1107         return err;
1108 }
1109
1110 struct file_system_type nilfs_fs_type = {
1111         .owner    = THIS_MODULE,
1112         .name     = "nilfs2",
1113         .get_sb   = nilfs_get_sb,
1114         .kill_sb  = kill_block_super,
1115         .fs_flags = FS_REQUIRES_DEV,
1116 };
1117
1118 static void nilfs_inode_init_once(void *obj)
1119 {
1120         struct nilfs_inode_info *ii = obj;
1121
1122         INIT_LIST_HEAD(&ii->i_dirty);
1123 #ifdef CONFIG_NILFS_XATTR
1124         init_rwsem(&ii->xattr_sem);
1125 #endif
1126         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1127         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
1128         inode_init_once(&ii->vfs_inode);
1129 }
1130
1131 static void nilfs_segbuf_init_once(void *obj)
1132 {
1133         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1134 }
1135
1136 static void nilfs_destroy_cachep(void)
1137 {
1138          if (nilfs_inode_cachep)
1139                 kmem_cache_destroy(nilfs_inode_cachep);
1140          if (nilfs_transaction_cachep)
1141                 kmem_cache_destroy(nilfs_transaction_cachep);
1142          if (nilfs_segbuf_cachep)
1143                 kmem_cache_destroy(nilfs_segbuf_cachep);
1144          if (nilfs_btree_path_cache)
1145                 kmem_cache_destroy(nilfs_btree_path_cache);
1146 }
1147
1148 static int __init nilfs_init_cachep(void)
1149 {
1150         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1151                         sizeof(struct nilfs_inode_info), 0,
1152                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1153         if (!nilfs_inode_cachep)
1154                 goto fail;
1155
1156         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1157                         sizeof(struct nilfs_transaction_info), 0,
1158                         SLAB_RECLAIM_ACCOUNT, NULL);
1159         if (!nilfs_transaction_cachep)
1160                 goto fail;
1161
1162         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1163                         sizeof(struct nilfs_segment_buffer), 0,
1164                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1165         if (!nilfs_segbuf_cachep)
1166                 goto fail;
1167
1168         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1169                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1170                         0, 0, NULL);
1171         if (!nilfs_btree_path_cache)
1172                 goto fail;
1173
1174         return 0;
1175
1176 fail:
1177         nilfs_destroy_cachep();
1178         return -ENOMEM;
1179 }
1180
1181 static int __init init_nilfs_fs(void)
1182 {
1183         int err;
1184
1185         err = nilfs_init_cachep();
1186         if (err)
1187                 goto fail;
1188
1189         err = register_filesystem(&nilfs_fs_type);
1190         if (err)
1191                 goto free_cachep;
1192
1193         return 0;
1194
1195 free_cachep:
1196         nilfs_destroy_cachep();
1197 fail:
1198         return err;
1199 }
1200
1201 static void __exit exit_nilfs_fs(void)
1202 {
1203         nilfs_destroy_cachep();
1204         unregister_filesystem(&nilfs_fs_type);
1205 }
1206
1207 module_init(init_nilfs_fs)
1208 module_exit(exit_nilfs_fs)