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/crc32.h>
  47 #include <linux/vfs.h>
  48 #include <linux/writeback.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/mount.h>
  51 #include "nilfs.h"
  52 #include "export.h"
  53 #include "mdt.h"
  54 #include "alloc.h"
  55 #include "btree.h"
  56 #include "btnode.h"
  57 #include "page.h"
  58 #include "cpfile.h"
  59 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  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 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_setup_super(struct super_block *sb, int is_mount);
  76 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  77
  78 static void nilfs_set_error(struct super_block *sb)
  79 {
  80         struct the_nilfs *nilfs = sb->s_fs_info;
  81         struct nilfs_super_block **sbp;
  82
  83         down_write(&nilfs->ns_sem);
  84         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  85                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  86                 sbp = nilfs_prepare_super(sb, 0);
  87                 if (likely(sbp)) {
  88                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  89                         if (sbp[1])
  90                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  91                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
  92                 }
  93         }
  94         up_write(&nilfs->ns_sem);
  95 }
  96
  97 /**
  98  * nilfs_error() - report failure condition on a filesystem
  99  *
 100  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
 101  * reporting an error message.  It should be called when NILFS detects
 102  * incoherences or defects of meta data on disk.  As for sustainable
 103  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
 104  * function should be used instead.
 105  *
 106  * The segment constructor must not call this function because it can
 107  * kill itself.
 108  */
 109 void nilfs_error(struct super_block *sb, const char *function,
 110                  const char *fmt, ...)
 111 {
 112         struct the_nilfs *nilfs = sb->s_fs_info;
 113         struct va_format vaf;
 114         va_list args;
 115
 116         va_start(args, fmt);
 117
 118         vaf.fmt = fmt;
 119         vaf.va = &args;
 120
 121         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 122                sb->s_id, function, &vaf);
 123
 124         va_end(args);
 125
 126         if (!(sb->s_flags & MS_RDONLY)) {
 127                 nilfs_set_error(sb);
 128
 129                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 130                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 131                         sb->s_flags |= MS_RDONLY;
 132                 }
 133         }
 134
 135         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 136                 panic("NILFS (device %s): panic forced after error\n",
 137                       sb->s_id);
 138 }
 139
 140 void nilfs_warning(struct super_block *sb, const char *function,
 141                    const char *fmt, ...)
 142 {
 143         struct va_format vaf;
 144         va_list args;
 145
 146         va_start(args, fmt);
 147
 148         vaf.fmt = fmt;
 149         vaf.va = &args;
 150
 151         printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
 152                sb->s_id, function, &vaf);
 153
 154         va_end(args);
 155 }
 156
 157
 158 struct inode *nilfs_alloc_inode(struct super_block *sb)
 159 {
 160         struct nilfs_inode_info *ii;
 161
 162         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 163         if (!ii)
 164                 return NULL;
 165         ii->i_bh = NULL;
 166         ii->i_state = 0;
 167         ii->i_cno = 0;
 168         ii->vfs_inode.i_version = 1;
 169         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode, sb->s_bdi);
 170         return &ii->vfs_inode;
 171 }
 172
 173 static void nilfs_i_callback(struct rcu_head *head)
 174 {
 175         struct inode *inode = container_of(head, struct inode, i_rcu);
 176         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 177
 178         INIT_LIST_HEAD(&inode->i_dentry);
 179
 180         if (mdi) {
 181                 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
 182                 kfree(mdi);
 183         }
 184         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 185 }
 186
 187 void nilfs_destroy_inode(struct inode *inode)
 188 {
 189         call_rcu(&inode->i_rcu, nilfs_i_callback);
 190 }
 191
 192 static int nilfs_sync_super(struct super_block *sb, int flag)
 193 {
 194         struct the_nilfs *nilfs = sb->s_fs_info;
 195         int err;
 196
 197  retry:
 198         set_buffer_dirty(nilfs->ns_sbh[0]);
 199         if (nilfs_test_opt(nilfs, BARRIER)) {
 200                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 201                                           WRITE_SYNC | WRITE_FLUSH_FUA);
 202         } else {
 203                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 204         }
 205
 206         if (unlikely(err)) {
 207                 printk(KERN_ERR
 208                        "NILFS: unable to write superblock (err=%d)\n", err);
 209                 if (err == -EIO && nilfs->ns_sbh[1]) {
 210                         /*
 211                          * sbp[0] points to newer log than sbp[1],
 212                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 213                          */
 214                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 215                                nilfs->ns_sbsize);
 216                         nilfs_fall_back_super_block(nilfs);
 217                         goto retry;
 218                 }
 219         } else {
 220                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 221
 222                 nilfs->ns_sbwcount++;
 223
 224                 /*
 225                  * The latest segment becomes trailable from the position
 226                  * written in superblock.
 227                  */
 228                 clear_nilfs_discontinued(nilfs);
 229
 230                 /* update GC protection for recent segments */
 231                 if (nilfs->ns_sbh[1]) {
 232                         if (flag == NILFS_SB_COMMIT_ALL) {
 233                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 234                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 235                                         goto out;
 236                         }
 237                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 238                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 239                                 sbp = nilfs->ns_sbp[1];
 240                 }
 241
 242                 spin_lock(&nilfs->ns_last_segment_lock);
 243                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 244                 spin_unlock(&nilfs->ns_last_segment_lock);
 245         }
 246  out:
 247         return err;
 248 }
 249
 250 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 251                           struct the_nilfs *nilfs)
 252 {
 253         sector_t nfreeblocks;
 254
 255         /* nilfs->ns_sem must be locked by the caller. */
 256         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 257         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 258
 259         spin_lock(&nilfs->ns_last_segment_lock);
 260         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 261         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 262         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 263         spin_unlock(&nilfs->ns_last_segment_lock);
 264 }
 265
 266 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 267                                                int flip)
 268 {
 269         struct the_nilfs *nilfs = sb->s_fs_info;
 270         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 271
 272         /* nilfs->ns_sem must be locked by the caller. */
 273         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 274                 if (sbp[1] &&
 275                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 276                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 277                 } else {
 278                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 279                                sb->s_id);
 280                         return NULL;
 281                 }
 282         } else if (sbp[1] &&
 283                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 284                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 285         }
 286
 287         if (flip && sbp[1])
 288                 nilfs_swap_super_block(nilfs);
 289
 290         return sbp;
 291 }
 292
 293 int nilfs_commit_super(struct super_block *sb, int flag)
 294 {
 295         struct the_nilfs *nilfs = sb->s_fs_info;
 296         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 297         time_t t;
 298
 299         /* nilfs->ns_sem must be locked by the caller. */
 300         t = get_seconds();
 301         nilfs->ns_sbwtime = t;
 302         sbp[0]->s_wtime = cpu_to_le64(t);
 303         sbp[0]->s_sum = 0;
 304         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 305                                              (unsigned char *)sbp[0],
 306                                              nilfs->ns_sbsize));
 307         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 308                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 309                 sbp[1]->s_sum = 0;
 310                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 311                                             (unsigned char *)sbp[1],
 312                                             nilfs->ns_sbsize));
 313         }
 314         clear_nilfs_sb_dirty(nilfs);
 315         return nilfs_sync_super(sb, flag);
 316 }
 317
 318 /**
 319  * nilfs_cleanup_super() - write filesystem state for cleanup
 320  * @sb: super block instance to be unmounted or degraded to read-only
 321  *
 322  * This function restores state flags in the on-disk super block.
 323  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 324  * filesystem was not clean previously.
 325  */
 326 int nilfs_cleanup_super(struct super_block *sb)
 327 {
 328         struct the_nilfs *nilfs = sb->s_fs_info;
 329         struct nilfs_super_block **sbp;
 330         int flag = NILFS_SB_COMMIT;
 331         int ret = -EIO;
 332
 333         sbp = nilfs_prepare_super(sb, 0);
 334         if (sbp) {
 335                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 336                 nilfs_set_log_cursor(sbp[0], nilfs);
 337                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 338                         /*
 339                          * make the "clean" flag also to the opposite
 340                          * super block if both super blocks point to
 341                          * the same checkpoint.
 342                          */
 343                         sbp[1]->s_state = sbp[0]->s_state;
 344                         flag = NILFS_SB_COMMIT_ALL;
 345                 }
 346                 ret = nilfs_commit_super(sb, flag);
 347         }
 348         return ret;
 349 }
 350
 351 /**
 352  * nilfs_move_2nd_super - relocate secondary super block
 353  * @sb: super block instance
 354  * @sb2off: new offset of the secondary super block (in bytes)
 355  */
 356 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 357 {
 358         struct the_nilfs *nilfs = sb->s_fs_info;
 359         struct buffer_head *nsbh;
 360         struct nilfs_super_block *nsbp;
 361         sector_t blocknr, newblocknr;
 362         unsigned long offset;
 363         int sb2i = -1;  /* array index of the secondary superblock */
 364         int ret = 0;
 365
 366         /* nilfs->ns_sem must be locked by the caller. */
 367         if (nilfs->ns_sbh[1] &&
 368             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 369                 sb2i = 1;
 370                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 371         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 372                 sb2i = 0;
 373                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 374         }
 375         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 376                 goto out;  /* super block location is unchanged */
 377
 378         /* Get new super block buffer */
 379         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 380         offset = sb2off & (nilfs->ns_blocksize - 1);
 381         nsbh = sb_getblk(sb, newblocknr);
 382         if (!nsbh) {
 383                 printk(KERN_WARNING
 384                        "NILFS warning: unable to move secondary superblock "
 385                        "to block %llu\n", (unsigned long long)newblocknr);
 386                 ret = -EIO;
 387                 goto out;
 388         }
 389         nsbp = (void *)nsbh->b_data + offset;
 390         memset(nsbp, 0, nilfs->ns_blocksize);
 391
 392         if (sb2i >= 0) {
 393                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 394                 brelse(nilfs->ns_sbh[sb2i]);
 395                 nilfs->ns_sbh[sb2i] = nsbh;
 396                 nilfs->ns_sbp[sb2i] = nsbp;
 397         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 398                 /* secondary super block will be restored to index 1 */
 399                 nilfs->ns_sbh[1] = nsbh;
 400                 nilfs->ns_sbp[1] = nsbp;
 401         } else {
 402                 brelse(nsbh);
 403         }
 404 out:
 405         return ret;
 406 }
 407
 408 /**
 409  * nilfs_resize_fs - resize the filesystem
 410  * @sb: super block instance
 411  * @newsize: new size of the filesystem (in bytes)
 412  */
 413 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 414 {
 415         struct the_nilfs *nilfs = sb->s_fs_info;
 416         struct nilfs_super_block **sbp;
 417         __u64 devsize, newnsegs;
 418         loff_t sb2off;
 419         int ret;
 420
 421         ret = -ERANGE;
 422         devsize = i_size_read(sb->s_bdev->bd_inode);
 423         if (newsize > devsize)
 424                 goto out;
 425
 426         /*
 427          * Write lock is required to protect some functions depending
 428          * on the number of segments, the number of reserved segments,
 429          * and so forth.
 430          */
 431         down_write(&nilfs->ns_segctor_sem);
 432
 433         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 434         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 435         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 436
 437         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 438         up_write(&nilfs->ns_segctor_sem);
 439         if (ret < 0)
 440                 goto out;
 441
 442         ret = nilfs_construct_segment(sb);
 443         if (ret < 0)
 444                 goto out;
 445
 446         down_write(&nilfs->ns_sem);
 447         nilfs_move_2nd_super(sb, sb2off);
 448         ret = -EIO;
 449         sbp = nilfs_prepare_super(sb, 0);
 450         if (likely(sbp)) {
 451                 nilfs_set_log_cursor(sbp[0], nilfs);
 452                 /*
 453                  * Drop NILFS_RESIZE_FS flag for compatibility with
 454                  * mount-time resize which may be implemented in a
 455                  * future release.
 456                  */
 457                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 458                                               ~NILFS_RESIZE_FS);
 459                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 460                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 461                 if (sbp[1])
 462                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 463                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 464         }
 465         up_write(&nilfs->ns_sem);
 466
 467         /*
 468          * Reset the range of allocatable segments last.  This order
 469          * is important in the case of expansion because the secondary
 470          * superblock must be protected from log write until migration
 471          * completes.
 472          */
 473         if (!ret)
 474                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 475 out:
 476         return ret;
 477 }
 478
 479 static void nilfs_put_super(struct super_block *sb)
 480 {
 481         struct the_nilfs *nilfs = sb->s_fs_info;
 482
 483         nilfs_detach_log_writer(sb);
 484
 485         if (!(sb->s_flags & MS_RDONLY)) {
 486                 down_write(&nilfs->ns_sem);
 487                 nilfs_cleanup_super(sb);
 488                 up_write(&nilfs->ns_sem);
 489         }
 490
 491         iput(nilfs->ns_sufile);
 492         iput(nilfs->ns_cpfile);
 493         iput(nilfs->ns_dat);
 494
 495         destroy_nilfs(nilfs);
 496         sb->s_fs_info = NULL;
 497 }
 498
 499 static int nilfs_sync_fs(struct super_block *sb, int wait)
 500 {
 501         struct the_nilfs *nilfs = sb->s_fs_info;
 502         struct nilfs_super_block **sbp;
 503         int err = 0;
 504
 505         /* This function is called when super block should be written back */
 506         if (wait)
 507                 err = nilfs_construct_segment(sb);
 508
 509         down_write(&nilfs->ns_sem);
 510         if (nilfs_sb_dirty(nilfs)) {
 511                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 512                 if (likely(sbp)) {
 513                         nilfs_set_log_cursor(sbp[0], nilfs);
 514                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 515                 }
 516         }
 517         up_write(&nilfs->ns_sem);
 518
 519         return err;
 520 }
 521
 522 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 523                             struct nilfs_root **rootp)
 524 {
 525         struct the_nilfs *nilfs = sb->s_fs_info;
 526         struct nilfs_root *root;
 527         struct nilfs_checkpoint *raw_cp;
 528         struct buffer_head *bh_cp;
 529         int err = -ENOMEM;
 530
 531         root = nilfs_find_or_create_root(
 532                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 533         if (!root)
 534                 return err;
 535
 536         if (root->ifile)
 537                 goto reuse; /* already attached checkpoint */
 538
 539         down_read(&nilfs->ns_segctor_sem);
 540         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 541                                           &bh_cp);
 542         up_read(&nilfs->ns_segctor_sem);
 543         if (unlikely(err)) {
 544                 if (err == -ENOENT || err == -EINVAL) {
 545                         printk(KERN_ERR
 546                                "NILFS: Invalid checkpoint "
 547                                "(checkpoint number=%llu)\n",
 548                                (unsigned long long)cno);
 549                         err = -EINVAL;
 550                 }
 551                 goto failed;
 552         }
 553
 554         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 555                                &raw_cp->cp_ifile_inode, &root->ifile);
 556         if (err)
 557                 goto failed_bh;
 558
 559         atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 560         atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 561
 562         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 563
 564  reuse:
 565         *rootp = root;
 566         return 0;
 567
 568  failed_bh:
 569         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 570  failed:
 571         nilfs_put_root(root);
 572
 573         return err;
 574 }
 575
 576 static int nilfs_freeze(struct super_block *sb)
 577 {
 578         struct the_nilfs *nilfs = sb->s_fs_info;
 579         int err;
 580
 581         if (sb->s_flags & MS_RDONLY)
 582                 return 0;
 583
 584         /* Mark super block clean */
 585         down_write(&nilfs->ns_sem);
 586         err = nilfs_cleanup_super(sb);
 587         up_write(&nilfs->ns_sem);
 588         return err;
 589 }
 590
 591 static int nilfs_unfreeze(struct super_block *sb)
 592 {
 593         struct the_nilfs *nilfs = sb->s_fs_info;
 594
 595         if (sb->s_flags & MS_RDONLY)
 596                 return 0;
 597
 598         down_write(&nilfs->ns_sem);
 599         nilfs_setup_super(sb, false);
 600         up_write(&nilfs->ns_sem);
 601         return 0;
 602 }
 603
 604 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 605 {
 606         struct super_block *sb = dentry->d_sb;
 607         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
 608         struct the_nilfs *nilfs = root->nilfs;
 609         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 610         unsigned long long blocks;
 611         unsigned long overhead;
 612         unsigned long nrsvblocks;
 613         sector_t nfreeblocks;
 614         int err;
 615
 616         /*
 617          * Compute all of the segment blocks
 618          *
 619          * The blocks before first segment and after last segment
 620          * are excluded.
 621          */
 622         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 623                 - nilfs->ns_first_data_block;
 624         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 625
 626         /*
 627          * Compute the overhead
 628          *
 629          * When distributing meta data blocks outside segment structure,
 630          * We must count them as the overhead.
 631          */
 632         overhead = 0;
 633
 634         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 635         if (unlikely(err))
 636                 return err;
 637
 638         buf->f_type = NILFS_SUPER_MAGIC;
 639         buf->f_bsize = sb->s_blocksize;
 640         buf->f_blocks = blocks - overhead;
 641         buf->f_bfree = nfreeblocks;
 642         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 643                 (buf->f_bfree - nrsvblocks) : 0;
 644         buf->f_files = atomic_read(&root->inodes_count);
 645         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 646         buf->f_namelen = NILFS_NAME_LEN;
 647         buf->f_fsid.val[0] = (u32)id;
 648         buf->f_fsid.val[1] = (u32)(id >> 32);
 649
 650         return 0;
 651 }
 652
 653 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 654 {
 655         struct super_block *sb = vfs->mnt_sb;
 656         struct the_nilfs *nilfs = sb->s_fs_info;
 657         struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
 658
 659         if (!nilfs_test_opt(nilfs, BARRIER))
 660                 seq_puts(seq, ",nobarrier");
 661         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 662                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 663         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 664                 seq_puts(seq, ",errors=panic");
 665         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 666                 seq_puts(seq, ",errors=continue");
 667         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 668                 seq_puts(seq, ",order=strict");
 669         if (nilfs_test_opt(nilfs, NORECOVERY))
 670                 seq_puts(seq, ",norecovery");
 671         if (nilfs_test_opt(nilfs, DISCARD))
 672                 seq_puts(seq, ",discard");
 673
 674         return 0;
 675 }
 676
 677 static const struct super_operations nilfs_sops = {
 678         .alloc_inode    = nilfs_alloc_inode,
 679         .destroy_inode  = nilfs_destroy_inode,
 680         .dirty_inode    = nilfs_dirty_inode,
 681         /* .write_inode    = nilfs_write_inode, */
 682         /* .put_inode      = nilfs_put_inode, */
 683         /* .drop_inode    = nilfs_drop_inode, */
 684         .evict_inode    = nilfs_evict_inode,
 685         .put_super      = nilfs_put_super,
 686         /* .write_super    = nilfs_write_super, */
 687         .sync_fs        = nilfs_sync_fs,
 688         .freeze_fs      = nilfs_freeze,
 689         .unfreeze_fs    = nilfs_unfreeze,
 690         /* .write_super_lockfs */
 691         /* .unlockfs */
 692         .statfs         = nilfs_statfs,
 693         .remount_fs     = nilfs_remount,
 694         /* .umount_begin */
 695         .show_options = nilfs_show_options
 696 };
 697
 698 enum {
 699         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 700         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 701         Opt_discard, Opt_nodiscard, Opt_err,
 702 };
 703
 704 static match_table_t tokens = {
 705         {Opt_err_cont, "errors=continue"},
 706         {Opt_err_panic, "errors=panic"},
 707         {Opt_err_ro, "errors=remount-ro"},
 708         {Opt_barrier, "barrier"},
 709         {Opt_nobarrier, "nobarrier"},
 710         {Opt_snapshot, "cp=%u"},
 711         {Opt_order, "order=%s"},
 712         {Opt_norecovery, "norecovery"},
 713         {Opt_discard, "discard"},
 714         {Opt_nodiscard, "nodiscard"},
 715         {Opt_err, NULL}
 716 };
 717
 718 static int parse_options(char *options, struct super_block *sb, int is_remount)
 719 {
 720         struct the_nilfs *nilfs = sb->s_fs_info;
 721         char *p;
 722         substring_t args[MAX_OPT_ARGS];
 723
 724         if (!options)
 725                 return 1;
 726
 727         while ((p = strsep(&options, ",")) != NULL) {
 728                 int token;
 729                 if (!*p)
 730                         continue;
 731
 732                 token = match_token(p, tokens, args);
 733                 switch (token) {
 734                 case Opt_barrier:
 735                         nilfs_set_opt(nilfs, BARRIER);
 736                         break;
 737                 case Opt_nobarrier:
 738                         nilfs_clear_opt(nilfs, BARRIER);
 739                         break;
 740                 case Opt_order:
 741                         if (strcmp(args[0].from, "relaxed") == 0)
 742                                 /* Ordered data semantics */
 743                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 744                         else if (strcmp(args[0].from, "strict") == 0)
 745                                 /* Strict in-order semantics */
 746                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 747                         else
 748                                 return 0;
 749                         break;
 750                 case Opt_err_panic:
 751                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 752                         break;
 753                 case Opt_err_ro:
 754                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 755                         break;
 756                 case Opt_err_cont:
 757                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 758                         break;
 759                 case Opt_snapshot:
 760                         if (is_remount) {
 761                                 printk(KERN_ERR
 762                                        "NILFS: \"%s\" option is invalid "
 763                                        "for remount.\n", p);
 764                                 return 0;
 765                         }
 766                         break;
 767                 case Opt_norecovery:
 768                         nilfs_set_opt(nilfs, NORECOVERY);
 769                         break;
 770                 case Opt_discard:
 771                         nilfs_set_opt(nilfs, DISCARD);
 772                         break;
 773                 case Opt_nodiscard:
 774                         nilfs_clear_opt(nilfs, DISCARD);
 775                         break;
 776                 default:
 777                         printk(KERN_ERR
 778                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 779                         return 0;
 780                 }
 781         }
 782         return 1;
 783 }
 784
 785 static inline void
 786 nilfs_set_default_options(struct super_block *sb,
 787                           struct nilfs_super_block *sbp)
 788 {
 789         struct the_nilfs *nilfs = sb->s_fs_info;
 790
 791         nilfs->ns_mount_opt =
 792                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 793 }
 794
 795 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 796 {
 797         struct the_nilfs *nilfs = sb->s_fs_info;
 798         struct nilfs_super_block **sbp;
 799         int max_mnt_count;
 800         int mnt_count;
 801
 802         /* nilfs->ns_sem must be locked by the caller. */
 803         sbp = nilfs_prepare_super(sb, 0);
 804         if (!sbp)
 805                 return -EIO;
 806
 807         if (!is_mount)
 808                 goto skip_mount_setup;
 809
 810         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 811         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 812
 813         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 814                 printk(KERN_WARNING
 815                        "NILFS warning: mounting fs with errors\n");
 816 #if 0
 817         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 818                 printk(KERN_WARNING
 819                        "NILFS warning: maximal mount count reached\n");
 820 #endif
 821         }
 822         if (!max_mnt_count)
 823                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 824
 825         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 826         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 827
 828 skip_mount_setup:
 829         sbp[0]->s_state =
 830                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 831         /* synchronize sbp[1] with sbp[0] */
 832         if (sbp[1])
 833                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 834         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 835 }
 836
 837 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 838                                                  u64 pos, int blocksize,
 839                                                  struct buffer_head **pbh)
 840 {
 841         unsigned long long sb_index = pos;
 842         unsigned long offset;
 843
 844         offset = do_div(sb_index, blocksize);
 845         *pbh = sb_bread(sb, sb_index);
 846         if (!*pbh)
 847                 return NULL;
 848         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 849 }
 850
 851 int nilfs_store_magic_and_option(struct super_block *sb,
 852                                  struct nilfs_super_block *sbp,
 853                                  char *data)
 854 {
 855         struct the_nilfs *nilfs = sb->s_fs_info;
 856
 857         sb->s_magic = le16_to_cpu(sbp->s_magic);
 858
 859         /* FS independent flags */
 860 #ifdef NILFS_ATIME_DISABLE
 861         sb->s_flags |= MS_NOATIME;
 862 #endif
 863
 864         nilfs_set_default_options(sb, sbp);
 865
 866         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 867         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 868         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 869         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 870
 871         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
 872 }
 873
 874 int nilfs_check_feature_compatibility(struct super_block *sb,
 875                                       struct nilfs_super_block *sbp)
 876 {
 877         __u64 features;
 878
 879         features = le64_to_cpu(sbp->s_feature_incompat) &
 880                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 881         if (features) {
 882                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 883                        "optional features (%llx)\n",
 884                        (unsigned long long)features);
 885                 return -EINVAL;
 886         }
 887         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 888                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 889         if (!(sb->s_flags & MS_RDONLY) && features) {
 890                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
 891                        "unsupported optional features (%llx)\n",
 892                        (unsigned long long)features);
 893                 return -EINVAL;
 894         }
 895         return 0;
 896 }
 897
 898 static int nilfs_get_root_dentry(struct super_block *sb,
 899                                  struct nilfs_root *root,
 900                                  struct dentry **root_dentry)
 901 {
 902         struct inode *inode;
 903         struct dentry *dentry;
 904         int ret = 0;
 905
 906         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 907         if (IS_ERR(inode)) {
 908                 printk(KERN_ERR "NILFS: get root inode failed\n");
 909                 ret = PTR_ERR(inode);
 910                 goto out;
 911         }
 912         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 913                 iput(inode);
 914                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 915                 ret = -EINVAL;
 916                 goto out;
 917         }
 918
 919         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 920                 dentry = d_find_alias(inode);
 921                 if (!dentry) {
 922                         dentry = d_alloc_root(inode);
 923                         if (!dentry) {
 924                                 iput(inode);
 925                                 ret = -ENOMEM;
 926                                 goto failed_dentry;
 927                         }
 928                 } else {
 929                         iput(inode);
 930                 }
 931         } else {
 932                 dentry = d_obtain_alias(inode);
 933                 if (IS_ERR(dentry)) {
 934                         ret = PTR_ERR(dentry);
 935                         goto failed_dentry;
 936                 }
 937         }
 938         *root_dentry = dentry;
 939  out:
 940         return ret;
 941
 942  failed_dentry:
 943         printk(KERN_ERR "NILFS: get root dentry failed\n");
 944         goto out;
 945 }
 946
 947 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 948                                  struct dentry **root_dentry)
 949 {
 950         struct the_nilfs *nilfs = s->s_fs_info;
 951         struct nilfs_root *root;
 952         int ret;
 953
 954         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 955
 956         down_read(&nilfs->ns_segctor_sem);
 957         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 958         up_read(&nilfs->ns_segctor_sem);
 959         if (ret < 0) {
 960                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 961                 goto out;
 962         } else if (!ret) {
 963                 printk(KERN_ERR "NILFS: The specified checkpoint is "
 964                        "not a snapshot (checkpoint number=%llu).\n",
 965                        (unsigned long long)cno);
 966                 ret = -EINVAL;
 967                 goto out;
 968         }
 969
 970         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 971         if (ret) {
 972                 printk(KERN_ERR "NILFS: error loading snapshot "
 973                        "(checkpoint number=%llu).\n",
 974                (unsigned long long)cno);
 975                 goto out;
 976         }
 977         ret = nilfs_get_root_dentry(s, root, root_dentry);
 978         nilfs_put_root(root);
 979  out:
 980         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 981         return ret;
 982 }
 983
 984 static int nilfs_tree_was_touched(struct dentry *root_dentry)
 985 {
 986         return root_dentry->d_count > 1;
 987 }
 988
 989 /**
 990  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
 991  * @root_dentry: root dentry of the tree to be shrunk
 992  *
 993  * This function returns true if the tree was in-use.
 994  */
 995 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
 996 {
 997         if (have_submounts(root_dentry))
 998                 return true;
 999         shrink_dcache_parent(root_dentry);
1000         return nilfs_tree_was_touched(root_dentry);
1001 }
1002
1003 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1004 {
1005         struct the_nilfs *nilfs = sb->s_fs_info;
1006         struct nilfs_root *root;
1007         struct inode *inode;
1008         struct dentry *dentry;
1009         int ret;
1010
1011         if (cno < 0 || cno > nilfs->ns_cno)
1012                 return false;
1013
1014         if (cno >= nilfs_last_cno(nilfs))
1015                 return true;    /* protect recent checkpoints */
1016
1017         ret = false;
1018         root = nilfs_lookup_root(nilfs, cno);
1019         if (root) {
1020                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1021                 if (inode) {
1022                         dentry = d_find_alias(inode);
1023                         if (dentry) {
1024                                 if (nilfs_tree_was_touched(dentry))
1025                                         ret = nilfs_try_to_shrink_tree(dentry);
1026                                 dput(dentry);
1027                         }
1028                         iput(inode);
1029                 }
1030                 nilfs_put_root(root);
1031         }
1032         return ret;
1033 }
1034
1035 /**
1036  * nilfs_fill_super() - initialize a super block instance
1037  * @sb: super_block
1038  * @data: mount options
1039  * @silent: silent mode flag
1040  *
1041  * This function is called exclusively by nilfs->ns_mount_mutex.
1042  * So, the recovery process is protected from other simultaneous mounts.
1043  */
1044 static int
1045 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1046 {
1047         struct the_nilfs *nilfs;
1048         struct nilfs_root *fsroot;
1049         struct backing_dev_info *bdi;
1050         __u64 cno;
1051         int err;
1052
1053         nilfs = alloc_nilfs(sb->s_bdev);
1054         if (!nilfs)
1055                 return -ENOMEM;
1056
1057         sb->s_fs_info = nilfs;
1058
1059         err = init_nilfs(nilfs, sb, (char *)data);
1060         if (err)
1061                 goto failed_nilfs;
1062
1063         sb->s_op = &nilfs_sops;
1064         sb->s_export_op = &nilfs_export_ops;
1065         sb->s_root = NULL;
1066         sb->s_time_gran = 1;
1067
1068         bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1069         sb->s_bdi = bdi ? : &default_backing_dev_info;
1070
1071         err = load_nilfs(nilfs, sb);
1072         if (err)
1073                 goto failed_nilfs;
1074
1075         cno = nilfs_last_cno(nilfs);
1076         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1077         if (err) {
1078                 printk(KERN_ERR "NILFS: error loading last checkpoint "
1079                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
1080                 goto failed_unload;
1081         }
1082
1083         if (!(sb->s_flags & MS_RDONLY)) {
1084                 err = nilfs_attach_log_writer(sb, fsroot);
1085                 if (err)
1086                         goto failed_checkpoint;
1087         }
1088
1089         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1090         if (err)
1091                 goto failed_segctor;
1092
1093         nilfs_put_root(fsroot);
1094
1095         if (!(sb->s_flags & MS_RDONLY)) {
1096                 down_write(&nilfs->ns_sem);
1097                 nilfs_setup_super(sb, true);
1098                 up_write(&nilfs->ns_sem);
1099         }
1100
1101         return 0;
1102
1103  failed_segctor:
1104         nilfs_detach_log_writer(sb);
1105
1106  failed_checkpoint:
1107         nilfs_put_root(fsroot);
1108
1109  failed_unload:
1110         iput(nilfs->ns_sufile);
1111         iput(nilfs->ns_cpfile);
1112         iput(nilfs->ns_dat);
1113
1114  failed_nilfs:
1115         destroy_nilfs(nilfs);
1116         return err;
1117 }
1118
1119 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1120 {
1121         struct the_nilfs *nilfs = sb->s_fs_info;
1122         unsigned long old_sb_flags;
1123         unsigned long old_mount_opt;
1124         int err;
1125
1126         old_sb_flags = sb->s_flags;
1127         old_mount_opt = nilfs->ns_mount_opt;
1128
1129         if (!parse_options(data, sb, 1)) {
1130                 err = -EINVAL;
1131                 goto restore_opts;
1132         }
1133         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1134
1135         err = -EINVAL;
1136
1137         if (!nilfs_valid_fs(nilfs)) {
1138                 printk(KERN_WARNING "NILFS (device %s): couldn't "
1139                        "remount because the filesystem is in an "
1140                        "incomplete recovery state.\n", sb->s_id);
1141                 goto restore_opts;
1142         }
1143
1144         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1145                 goto out;
1146         if (*flags & MS_RDONLY) {
1147                 /* Shutting down log writer */
1148                 nilfs_detach_log_writer(sb);
1149                 sb->s_flags |= MS_RDONLY;
1150
1151                 /*
1152                  * Remounting a valid RW partition RDONLY, so set
1153                  * the RDONLY flag and then mark the partition as valid again.
1154                  */
1155                 down_write(&nilfs->ns_sem);
1156                 nilfs_cleanup_super(sb);
1157                 up_write(&nilfs->ns_sem);
1158         } else {
1159                 __u64 features;
1160                 struct nilfs_root *root;
1161
1162                 /*
1163                  * Mounting a RDONLY partition read-write, so reread and
1164                  * store the current valid flag.  (It may have been changed
1165                  * by fsck since we originally mounted the partition.)
1166                  */
1167                 down_read(&nilfs->ns_sem);
1168                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1169                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1170                 up_read(&nilfs->ns_sem);
1171                 if (features) {
1172                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1173                                "remount RDWR because of unsupported optional "
1174                                "features (%llx)\n",
1175                                sb->s_id, (unsigned long long)features);
1176                         err = -EROFS;
1177                         goto restore_opts;
1178                 }
1179
1180                 sb->s_flags &= ~MS_RDONLY;
1181
1182                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1183                 err = nilfs_attach_log_writer(sb, root);
1184                 if (err)
1185                         goto restore_opts;
1186
1187                 down_write(&nilfs->ns_sem);
1188                 nilfs_setup_super(sb, true);
1189                 up_write(&nilfs->ns_sem);
1190         }
1191  out:
1192         return 0;
1193
1194  restore_opts:
1195         sb->s_flags = old_sb_flags;
1196         nilfs->ns_mount_opt = old_mount_opt;
1197         return err;
1198 }
1199
1200 struct nilfs_super_data {
1201         struct block_device *bdev;
1202         __u64 cno;
1203         int flags;
1204 };
1205
1206 /**
1207  * nilfs_identify - pre-read mount options needed to identify mount instance
1208  * @data: mount options
1209  * @sd: nilfs_super_data
1210  */
1211 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1212 {
1213         char *p, *options = data;
1214         substring_t args[MAX_OPT_ARGS];
1215         int token;
1216         int ret = 0;
1217
1218         do {
1219                 p = strsep(&options, ",");
1220                 if (p != NULL && *p) {
1221                         token = match_token(p, tokens, args);
1222                         if (token == Opt_snapshot) {
1223                                 if (!(sd->flags & MS_RDONLY)) {
1224                                         ret++;
1225                                 } else {
1226                                         sd->cno = simple_strtoull(args[0].from,
1227                                                                   NULL, 0);
1228                                         /*
1229                                          * No need to see the end pointer;
1230                                          * match_token() has done syntax
1231                                          * checking.
1232                                          */
1233                                         if (sd->cno == 0)
1234                                                 ret++;
1235                                 }
1236                         }
1237                         if (ret)
1238                                 printk(KERN_ERR
1239                                        "NILFS: invalid mount option: %s\n", p);
1240                 }
1241                 if (!options)
1242                         break;
1243                 BUG_ON(options == data);
1244                 *(options - 1) = ',';
1245         } while (!ret);
1246         return ret;
1247 }
1248
1249 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1250 {
1251         s->s_bdev = data;
1252         s->s_dev = s->s_bdev->bd_dev;
1253         return 0;
1254 }
1255
1256 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1257 {
1258         return (void *)s->s_bdev == data;
1259 }
1260
1261 static struct dentry *
1262 nilfs_mount(struct file_system_type *fs_type, int flags,
1263              const char *dev_name, void *data)
1264 {
1265         struct nilfs_super_data sd;
1266         struct super_block *s;
1267         fmode_t mode = FMODE_READ | FMODE_EXCL;
1268         struct dentry *root_dentry;
1269         int err, s_new = false;
1270
1271         if (!(flags & MS_RDONLY))
1272                 mode |= FMODE_WRITE;
1273
1274         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1275         if (IS_ERR(sd.bdev))
1276                 return ERR_CAST(sd.bdev);
1277
1278         sd.cno = 0;
1279         sd.flags = flags;
1280         if (nilfs_identify((char *)data, &sd)) {
1281                 err = -EINVAL;
1282                 goto failed;
1283         }
1284
1285         /*
1286          * once the super is inserted into the list by sget, s_umount
1287          * will protect the lockfs code from trying to start a snapshot
1288          * while we are mounting
1289          */
1290         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1291         if (sd.bdev->bd_fsfreeze_count > 0) {
1292                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1293                 err = -EBUSY;
1294                 goto failed;
1295         }
1296         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1297         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1298         if (IS_ERR(s)) {
1299                 err = PTR_ERR(s);
1300                 goto failed;
1301         }
1302
1303         if (!s->s_root) {
1304                 char b[BDEVNAME_SIZE];
1305
1306                 s_new = true;
1307
1308                 /* New superblock instance created */
1309                 s->s_flags = flags;
1310                 s->s_mode = mode;
1311                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1312                 sb_set_blocksize(s, block_size(sd.bdev));
1313
1314                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1315                 if (err)
1316                         goto failed_super;
1317
1318                 s->s_flags |= MS_ACTIVE;
1319         } else if (!sd.cno) {
1320                 int busy = false;
1321
1322                 if (nilfs_tree_was_touched(s->s_root)) {
1323                         busy = nilfs_try_to_shrink_tree(s->s_root);
1324                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1325                                 printk(KERN_ERR "NILFS: the device already "
1326                                        "has a %s mount.\n",
1327                                        (s->s_flags & MS_RDONLY) ?
1328                                        "read-only" : "read/write");
1329                                 err = -EBUSY;
1330                                 goto failed_super;
1331                         }
1332                 }
1333                 if (!busy) {
1334                         /*
1335                          * Try remount to setup mount states if the current
1336                          * tree is not mounted and only snapshots use this sb.
1337                          */
1338                         err = nilfs_remount(s, &flags, data);
1339                         if (err)
1340                                 goto failed_super;
1341                 }
1342         }
1343
1344         if (sd.cno) {
1345                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1346                 if (err)
1347                         goto failed_super;
1348         } else {
1349                 root_dentry = dget(s->s_root);
1350         }
1351
1352         if (!s_new)
1353                 blkdev_put(sd.bdev, mode);
1354
1355         return root_dentry;
1356
1357  failed_super:
1358         deactivate_locked_super(s);
1359
1360  failed:
1361         if (!s_new)
1362                 blkdev_put(sd.bdev, mode);
1363         return ERR_PTR(err);
1364 }
1365
1366 struct file_system_type nilfs_fs_type = {
1367         .owner    = THIS_MODULE,
1368         .name     = "nilfs2",
1369         .mount    = nilfs_mount,
1370         .kill_sb  = kill_block_super,
1371         .fs_flags = FS_REQUIRES_DEV,
1372 };
1373
1374 static void nilfs_inode_init_once(void *obj)
1375 {
1376         struct nilfs_inode_info *ii = obj;
1377
1378         INIT_LIST_HEAD(&ii->i_dirty);
1379 #ifdef CONFIG_NILFS_XATTR
1380         init_rwsem(&ii->xattr_sem);
1381 #endif
1382         address_space_init_once(&ii->i_btnode_cache);
1383         ii->i_bmap = &ii->i_bmap_data;
1384         inode_init_once(&ii->vfs_inode);
1385 }
1386
1387 static void nilfs_segbuf_init_once(void *obj)
1388 {
1389         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1390 }
1391
1392 static void nilfs_destroy_cachep(void)
1393 {
1394         if (nilfs_inode_cachep)
1395                 kmem_cache_destroy(nilfs_inode_cachep);
1396         if (nilfs_transaction_cachep)
1397                 kmem_cache_destroy(nilfs_transaction_cachep);
1398         if (nilfs_segbuf_cachep)
1399                 kmem_cache_destroy(nilfs_segbuf_cachep);
1400         if (nilfs_btree_path_cache)
1401                 kmem_cache_destroy(nilfs_btree_path_cache);
1402 }
1403
1404 static int __init nilfs_init_cachep(void)
1405 {
1406         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1407                         sizeof(struct nilfs_inode_info), 0,
1408                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1409         if (!nilfs_inode_cachep)
1410                 goto fail;
1411
1412         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1413                         sizeof(struct nilfs_transaction_info), 0,
1414                         SLAB_RECLAIM_ACCOUNT, NULL);
1415         if (!nilfs_transaction_cachep)
1416                 goto fail;
1417
1418         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1419                         sizeof(struct nilfs_segment_buffer), 0,
1420                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1421         if (!nilfs_segbuf_cachep)
1422                 goto fail;
1423
1424         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1425                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1426                         0, 0, NULL);
1427         if (!nilfs_btree_path_cache)
1428                 goto fail;
1429
1430         return 0;
1431
1432 fail:
1433         nilfs_destroy_cachep();
1434         return -ENOMEM;
1435 }
1436
1437 static int __init init_nilfs_fs(void)
1438 {
1439         int err;
1440
1441         err = nilfs_init_cachep();
1442         if (err)
1443                 goto fail;
1444
1445         err = register_filesystem(&nilfs_fs_type);
1446         if (err)
1447                 goto free_cachep;
1448
1449         printk(KERN_INFO "NILFS version 2 loaded\n");
1450         return 0;
1451
1452 free_cachep:
1453         nilfs_destroy_cachep();
1454 fail:
1455         return err;
1456 }
1457
1458 static void __exit exit_nilfs_fs(void)
1459 {
1460         nilfs_destroy_cachep();
1461         unregister_filesystem(&nilfs_fs_type);
1462 }
1463
1464 module_init(init_nilfs_fs)
1465 module_exit(exit_nilfs_fs)