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