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