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         /* .write_inode    = nilfs_write_inode, */
 680         /* .put_inode      = nilfs_put_inode, */
 681         /* .drop_inode    = nilfs_drop_inode, */
 682         .evict_inode    = nilfs_evict_inode,
 683         .put_super      = nilfs_put_super,
 684         /* .write_super    = nilfs_write_super, */
 685         .sync_fs        = nilfs_sync_fs,
 686         .freeze_fs      = nilfs_freeze,
 687         .unfreeze_fs    = nilfs_unfreeze,
 688         /* .write_super_lockfs */
 689         /* .unlockfs */
 690         .statfs         = nilfs_statfs,
 691         .remount_fs     = nilfs_remount,
 692         /* .umount_begin */
 693         .show_options = nilfs_show_options
 694 };
 695
 696 enum {
 697         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 698         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 699         Opt_discard, Opt_nodiscard, Opt_err,
 700 };
 701
 702 static match_table_t tokens = {
 703         {Opt_err_cont, "errors=continue"},
 704         {Opt_err_panic, "errors=panic"},
 705         {Opt_err_ro, "errors=remount-ro"},
 706         {Opt_barrier, "barrier"},
 707         {Opt_nobarrier, "nobarrier"},
 708         {Opt_snapshot, "cp=%u"},
 709         {Opt_order, "order=%s"},
 710         {Opt_norecovery, "norecovery"},
 711         {Opt_discard, "discard"},
 712         {Opt_nodiscard, "nodiscard"},
 713         {Opt_err, NULL}
 714 };
 715
 716 static int parse_options(char *options, struct super_block *sb, int is_remount)
 717 {
 718         struct the_nilfs *nilfs = sb->s_fs_info;
 719         char *p;
 720         substring_t args[MAX_OPT_ARGS];
 721
 722         if (!options)
 723                 return 1;
 724
 725         while ((p = strsep(&options, ",")) != NULL) {
 726                 int token;
 727                 if (!*p)
 728                         continue;
 729
 730                 token = match_token(p, tokens, args);
 731                 switch (token) {
 732                 case Opt_barrier:
 733                         nilfs_set_opt(nilfs, BARRIER);
 734                         break;
 735                 case Opt_nobarrier:
 736                         nilfs_clear_opt(nilfs, BARRIER);
 737                         break;
 738                 case Opt_order:
 739                         if (strcmp(args[0].from, "relaxed") == 0)
 740                                 /* Ordered data semantics */
 741                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 742                         else if (strcmp(args[0].from, "strict") == 0)
 743                                 /* Strict in-order semantics */
 744                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 745                         else
 746                                 return 0;
 747                         break;
 748                 case Opt_err_panic:
 749                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 750                         break;
 751                 case Opt_err_ro:
 752                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 753                         break;
 754                 case Opt_err_cont:
 755                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 756                         break;
 757                 case Opt_snapshot:
 758                         if (is_remount) {
 759                                 printk(KERN_ERR
 760                                        "NILFS: \"%s\" option is invalid "
 761                                        "for remount.\n", p);
 762                                 return 0;
 763                         }
 764                         break;
 765                 case Opt_norecovery:
 766                         nilfs_set_opt(nilfs, NORECOVERY);
 767                         break;
 768                 case Opt_discard:
 769                         nilfs_set_opt(nilfs, DISCARD);
 770                         break;
 771                 case Opt_nodiscard:
 772                         nilfs_clear_opt(nilfs, DISCARD);
 773                         break;
 774                 default:
 775                         printk(KERN_ERR
 776                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 777                         return 0;
 778                 }
 779         }
 780         return 1;
 781 }
 782
 783 static inline void
 784 nilfs_set_default_options(struct super_block *sb,
 785                           struct nilfs_super_block *sbp)
 786 {
 787         struct the_nilfs *nilfs = sb->s_fs_info;
 788
 789         nilfs->ns_mount_opt =
 790                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 791 }
 792
 793 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 794 {
 795         struct the_nilfs *nilfs = sb->s_fs_info;
 796         struct nilfs_super_block **sbp;
 797         int max_mnt_count;
 798         int mnt_count;
 799
 800         /* nilfs->ns_sem must be locked by the caller. */
 801         sbp = nilfs_prepare_super(sb, 0);
 802         if (!sbp)
 803                 return -EIO;
 804
 805         if (!is_mount)
 806                 goto skip_mount_setup;
 807
 808         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 809         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 810
 811         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 812                 printk(KERN_WARNING
 813                        "NILFS warning: mounting fs with errors\n");
 814 #if 0
 815         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 816                 printk(KERN_WARNING
 817                        "NILFS warning: maximal mount count reached\n");
 818 #endif
 819         }
 820         if (!max_mnt_count)
 821                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 822
 823         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 824         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 825
 826 skip_mount_setup:
 827         sbp[0]->s_state =
 828                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 829         /* synchronize sbp[1] with sbp[0] */
 830         if (sbp[1])
 831                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 832         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 833 }
 834
 835 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 836                                                  u64 pos, int blocksize,
 837                                                  struct buffer_head **pbh)
 838 {
 839         unsigned long long sb_index = pos;
 840         unsigned long offset;
 841
 842         offset = do_div(sb_index, blocksize);
 843         *pbh = sb_bread(sb, sb_index);
 844         if (!*pbh)
 845                 return NULL;
 846         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 847 }
 848
 849 int nilfs_store_magic_and_option(struct super_block *sb,
 850                                  struct nilfs_super_block *sbp,
 851                                  char *data)
 852 {
 853         struct the_nilfs *nilfs = sb->s_fs_info;
 854
 855         sb->s_magic = le16_to_cpu(sbp->s_magic);
 856
 857         /* FS independent flags */
 858 #ifdef NILFS_ATIME_DISABLE
 859         sb->s_flags |= MS_NOATIME;
 860 #endif
 861
 862         nilfs_set_default_options(sb, sbp);
 863
 864         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 865         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 866         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 867         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 868
 869         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
 870 }
 871
 872 int nilfs_check_feature_compatibility(struct super_block *sb,
 873                                       struct nilfs_super_block *sbp)
 874 {
 875         __u64 features;
 876
 877         features = le64_to_cpu(sbp->s_feature_incompat) &
 878                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 879         if (features) {
 880                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 881                        "optional features (%llx)\n",
 882                        (unsigned long long)features);
 883                 return -EINVAL;
 884         }
 885         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 886                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 887         if (!(sb->s_flags & MS_RDONLY) && features) {
 888                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
 889                        "unsupported optional features (%llx)\n",
 890                        (unsigned long long)features);
 891                 return -EINVAL;
 892         }
 893         return 0;
 894 }
 895
 896 static int nilfs_get_root_dentry(struct super_block *sb,
 897                                  struct nilfs_root *root,
 898                                  struct dentry **root_dentry)
 899 {
 900         struct inode *inode;
 901         struct dentry *dentry;
 902         int ret = 0;
 903
 904         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 905         if (IS_ERR(inode)) {
 906                 printk(KERN_ERR "NILFS: get root inode failed\n");
 907                 ret = PTR_ERR(inode);
 908                 goto out;
 909         }
 910         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 911                 iput(inode);
 912                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 913                 ret = -EINVAL;
 914                 goto out;
 915         }
 916
 917         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 918                 dentry = d_find_alias(inode);
 919                 if (!dentry) {
 920                         dentry = d_make_root(inode);
 921                         if (!dentry) {
 922                                 ret = -ENOMEM;
 923                                 goto failed_dentry;
 924                         }
 925                 } else {
 926                         iput(inode);
 927                 }
 928         } else {
 929                 dentry = d_obtain_alias(inode);
 930                 if (IS_ERR(dentry)) {
 931                         ret = PTR_ERR(dentry);
 932                         goto failed_dentry;
 933                 }
 934         }
 935         *root_dentry = dentry;
 936  out:
 937         return ret;
 938
 939  failed_dentry:
 940         printk(KERN_ERR "NILFS: get root dentry failed\n");
 941         goto out;
 942 }
 943
 944 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 945                                  struct dentry **root_dentry)
 946 {
 947         struct the_nilfs *nilfs = s->s_fs_info;
 948         struct nilfs_root *root;
 949         int ret;
 950
 951         down_read(&nilfs->ns_segctor_sem);
 952         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 953         up_read(&nilfs->ns_segctor_sem);
 954         if (ret < 0) {
 955                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 956                 goto out;
 957         } else if (!ret) {
 958                 printk(KERN_ERR "NILFS: The specified checkpoint is "
 959                        "not a snapshot (checkpoint number=%llu).\n",
 960                        (unsigned long long)cno);
 961                 ret = -EINVAL;
 962                 goto out;
 963         }
 964
 965         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 966         if (ret) {
 967                 printk(KERN_ERR "NILFS: error loading snapshot "
 968                        "(checkpoint number=%llu).\n",
 969                (unsigned long long)cno);
 970                 goto out;
 971         }
 972         ret = nilfs_get_root_dentry(s, root, root_dentry);
 973         nilfs_put_root(root);
 974  out:
 975         return ret;
 976 }
 977
 978 static int nilfs_tree_was_touched(struct dentry *root_dentry)
 979 {
 980         return root_dentry->d_count > 1;
 981 }
 982
 983 /**
 984  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
 985  * @root_dentry: root dentry of the tree to be shrunk
 986  *
 987  * This function returns true if the tree was in-use.
 988  */
 989 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
 990 {
 991         if (have_submounts(root_dentry))
 992                 return true;
 993         shrink_dcache_parent(root_dentry);
 994         return nilfs_tree_was_touched(root_dentry);
 995 }
 996
 997 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
 998 {
 999         struct the_nilfs *nilfs = sb->s_fs_info;
1000         struct nilfs_root *root;
1001         struct inode *inode;
1002         struct dentry *dentry;
1003         int ret;
1004
1005         if (cno < 0 || cno > nilfs->ns_cno)
1006                 return false;
1007
1008         if (cno >= nilfs_last_cno(nilfs))
1009                 return true;    /* protect recent checkpoints */
1010
1011         ret = false;
1012         root = nilfs_lookup_root(nilfs, cno);
1013         if (root) {
1014                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1015                 if (inode) {
1016                         dentry = d_find_alias(inode);
1017                         if (dentry) {
1018                                 if (nilfs_tree_was_touched(dentry))
1019                                         ret = nilfs_try_to_shrink_tree(dentry);
1020                                 dput(dentry);
1021                         }
1022                         iput(inode);
1023                 }
1024                 nilfs_put_root(root);
1025         }
1026         return ret;
1027 }
1028
1029 /**
1030  * nilfs_fill_super() - initialize a super block instance
1031  * @sb: super_block
1032  * @data: mount options
1033  * @silent: silent mode flag
1034  *
1035  * This function is called exclusively by nilfs->ns_mount_mutex.
1036  * So, the recovery process is protected from other simultaneous mounts.
1037  */
1038 static int
1039 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1040 {
1041         struct the_nilfs *nilfs;
1042         struct nilfs_root *fsroot;
1043         struct backing_dev_info *bdi;
1044         __u64 cno;
1045         int err;
1046
1047         nilfs = alloc_nilfs(sb->s_bdev);
1048         if (!nilfs)
1049                 return -ENOMEM;
1050
1051         sb->s_fs_info = nilfs;
1052
1053         err = init_nilfs(nilfs, sb, (char *)data);
1054         if (err)
1055                 goto failed_nilfs;
1056
1057         sb->s_op = &nilfs_sops;
1058         sb->s_export_op = &nilfs_export_ops;
1059         sb->s_root = NULL;
1060         sb->s_time_gran = 1;
1061         sb->s_max_links = NILFS_LINK_MAX;
1062
1063         bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1064         sb->s_bdi = bdi ? : &default_backing_dev_info;
1065
1066         err = load_nilfs(nilfs, sb);
1067         if (err)
1068                 goto failed_nilfs;
1069
1070         cno = nilfs_last_cno(nilfs);
1071         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1072         if (err) {
1073                 printk(KERN_ERR "NILFS: error loading last checkpoint "
1074                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
1075                 goto failed_unload;
1076         }
1077
1078         if (!(sb->s_flags & MS_RDONLY)) {
1079                 err = nilfs_attach_log_writer(sb, fsroot);
1080                 if (err)
1081                         goto failed_checkpoint;
1082         }
1083
1084         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1085         if (err)
1086                 goto failed_segctor;
1087
1088         nilfs_put_root(fsroot);
1089
1090         if (!(sb->s_flags & MS_RDONLY)) {
1091                 down_write(&nilfs->ns_sem);
1092                 nilfs_setup_super(sb, true);
1093                 up_write(&nilfs->ns_sem);
1094         }
1095
1096         return 0;
1097
1098  failed_segctor:
1099         nilfs_detach_log_writer(sb);
1100
1101  failed_checkpoint:
1102         nilfs_put_root(fsroot);
1103
1104  failed_unload:
1105         iput(nilfs->ns_sufile);
1106         iput(nilfs->ns_cpfile);
1107         iput(nilfs->ns_dat);
1108
1109  failed_nilfs:
1110         destroy_nilfs(nilfs);
1111         return err;
1112 }
1113
1114 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1115 {
1116         struct the_nilfs *nilfs = sb->s_fs_info;
1117         unsigned long old_sb_flags;
1118         unsigned long old_mount_opt;
1119         int err;
1120
1121         old_sb_flags = sb->s_flags;
1122         old_mount_opt = nilfs->ns_mount_opt;
1123
1124         if (!parse_options(data, sb, 1)) {
1125                 err = -EINVAL;
1126                 goto restore_opts;
1127         }
1128         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1129
1130         err = -EINVAL;
1131
1132         if (!nilfs_valid_fs(nilfs)) {
1133                 printk(KERN_WARNING "NILFS (device %s): couldn't "
1134                        "remount because the filesystem is in an "
1135                        "incomplete recovery state.\n", sb->s_id);
1136                 goto restore_opts;
1137         }
1138
1139         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1140                 goto out;
1141         if (*flags & MS_RDONLY) {
1142                 /* Shutting down log writer */
1143                 nilfs_detach_log_writer(sb);
1144                 sb->s_flags |= MS_RDONLY;
1145
1146                 /*
1147                  * Remounting a valid RW partition RDONLY, so set
1148                  * the RDONLY flag and then mark the partition as valid again.
1149                  */
1150                 down_write(&nilfs->ns_sem);
1151                 nilfs_cleanup_super(sb);
1152                 up_write(&nilfs->ns_sem);
1153         } else {
1154                 __u64 features;
1155                 struct nilfs_root *root;
1156
1157                 /*
1158                  * Mounting a RDONLY partition read-write, so reread and
1159                  * store the current valid flag.  (It may have been changed
1160                  * by fsck since we originally mounted the partition.)
1161                  */
1162                 down_read(&nilfs->ns_sem);
1163                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1164                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1165                 up_read(&nilfs->ns_sem);
1166                 if (features) {
1167                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1168                                "remount RDWR because of unsupported optional "
1169                                "features (%llx)\n",
1170                                sb->s_id, (unsigned long long)features);
1171                         err = -EROFS;
1172                         goto restore_opts;
1173                 }
1174
1175                 sb->s_flags &= ~MS_RDONLY;
1176
1177                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1178                 err = nilfs_attach_log_writer(sb, root);
1179                 if (err)
1180                         goto restore_opts;
1181
1182                 down_write(&nilfs->ns_sem);
1183                 nilfs_setup_super(sb, true);
1184                 up_write(&nilfs->ns_sem);
1185         }
1186  out:
1187         return 0;
1188
1189  restore_opts:
1190         sb->s_flags = old_sb_flags;
1191         nilfs->ns_mount_opt = old_mount_opt;
1192         return err;
1193 }
1194
1195 struct nilfs_super_data {
1196         struct block_device *bdev;
1197         __u64 cno;
1198         int flags;
1199 };
1200
1201 /**
1202  * nilfs_identify - pre-read mount options needed to identify mount instance
1203  * @data: mount options
1204  * @sd: nilfs_super_data
1205  */
1206 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1207 {
1208         char *p, *options = data;
1209         substring_t args[MAX_OPT_ARGS];
1210         int token;
1211         int ret = 0;
1212
1213         do {
1214                 p = strsep(&options, ",");
1215                 if (p != NULL && *p) {
1216                         token = match_token(p, tokens, args);
1217                         if (token == Opt_snapshot) {
1218                                 if (!(sd->flags & MS_RDONLY)) {
1219                                         ret++;
1220                                 } else {
1221                                         sd->cno = simple_strtoull(args[0].from,
1222                                                                   NULL, 0);
1223                                         /*
1224                                          * No need to see the end pointer;
1225                                          * match_token() has done syntax
1226                                          * checking.
1227                                          */
1228                                         if (sd->cno == 0)
1229                                                 ret++;
1230                                 }
1231                         }
1232                         if (ret)
1233                                 printk(KERN_ERR
1234                                        "NILFS: invalid mount option: %s\n", p);
1235                 }
1236                 if (!options)
1237                         break;
1238                 BUG_ON(options == data);
1239                 *(options - 1) = ',';
1240         } while (!ret);
1241         return ret;
1242 }
1243
1244 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1245 {
1246         s->s_bdev = data;
1247         s->s_dev = s->s_bdev->bd_dev;
1248         return 0;
1249 }
1250
1251 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1252 {
1253         return (void *)s->s_bdev == data;
1254 }
1255
1256 static struct dentry *
1257 nilfs_mount(struct file_system_type *fs_type, int flags,
1258              const char *dev_name, void *data)
1259 {
1260         struct nilfs_super_data sd;
1261         struct super_block *s;
1262         fmode_t mode = FMODE_READ | FMODE_EXCL;
1263         struct dentry *root_dentry;
1264         int err, s_new = false;
1265
1266         if (!(flags & MS_RDONLY))
1267                 mode |= FMODE_WRITE;
1268
1269         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1270         if (IS_ERR(sd.bdev))
1271                 return ERR_CAST(sd.bdev);
1272
1273         sd.cno = 0;
1274         sd.flags = flags;
1275         if (nilfs_identify((char *)data, &sd)) {
1276                 err = -EINVAL;
1277                 goto failed;
1278         }
1279
1280         /*
1281          * once the super is inserted into the list by sget, s_umount
1282          * will protect the lockfs code from trying to start a snapshot
1283          * while we are mounting
1284          */
1285         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1286         if (sd.bdev->bd_fsfreeze_count > 0) {
1287                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1288                 err = -EBUSY;
1289                 goto failed;
1290         }
1291         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1292         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1293         if (IS_ERR(s)) {
1294                 err = PTR_ERR(s);
1295                 goto failed;
1296         }
1297
1298         if (!s->s_root) {
1299                 char b[BDEVNAME_SIZE];
1300
1301                 s_new = true;
1302
1303                 /* New superblock instance created */
1304                 s->s_flags = flags;
1305                 s->s_mode = mode;
1306                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1307                 sb_set_blocksize(s, block_size(sd.bdev));
1308
1309                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1310                 if (err)
1311                         goto failed_super;
1312
1313                 s->s_flags |= MS_ACTIVE;
1314         } else if (!sd.cno) {
1315                 int busy = false;
1316
1317                 if (nilfs_tree_was_touched(s->s_root)) {
1318                         busy = nilfs_try_to_shrink_tree(s->s_root);
1319                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1320                                 printk(KERN_ERR "NILFS: the device already "
1321                                        "has a %s mount.\n",
1322                                        (s->s_flags & MS_RDONLY) ?
1323                                        "read-only" : "read/write");
1324                                 err = -EBUSY;
1325                                 goto failed_super;
1326                         }
1327                 }
1328                 if (!busy) {
1329                         /*
1330                          * Try remount to setup mount states if the current
1331                          * tree is not mounted and only snapshots use this sb.
1332                          */
1333                         err = nilfs_remount(s, &flags, data);
1334                         if (err)
1335                                 goto failed_super;
1336                 }
1337         }
1338
1339         if (sd.cno) {
1340                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1341                 if (err)
1342                         goto failed_super;
1343         } else {
1344                 root_dentry = dget(s->s_root);
1345         }
1346
1347         if (!s_new)
1348                 blkdev_put(sd.bdev, mode);
1349
1350         return root_dentry;
1351
1352  failed_super:
1353         deactivate_locked_super(s);
1354
1355  failed:
1356         if (!s_new)
1357                 blkdev_put(sd.bdev, mode);
1358         return ERR_PTR(err);
1359 }
1360
1361 struct file_system_type nilfs_fs_type = {
1362         .owner    = THIS_MODULE,
1363         .name     = "nilfs2",
1364         .mount    = nilfs_mount,
1365         .kill_sb  = kill_block_super,
1366         .fs_flags = FS_REQUIRES_DEV,
1367 };
1368
1369 static void nilfs_inode_init_once(void *obj)
1370 {
1371         struct nilfs_inode_info *ii = obj;
1372
1373         INIT_LIST_HEAD(&ii->i_dirty);
1374 #ifdef CONFIG_NILFS_XATTR
1375         init_rwsem(&ii->xattr_sem);
1376 #endif
1377         address_space_init_once(&ii->i_btnode_cache);
1378         ii->i_bmap = &ii->i_bmap_data;
1379         inode_init_once(&ii->vfs_inode);
1380 }
1381
1382 static void nilfs_segbuf_init_once(void *obj)
1383 {
1384         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1385 }
1386
1387 static void nilfs_destroy_cachep(void)
1388 {
1389         if (nilfs_inode_cachep)
1390                 kmem_cache_destroy(nilfs_inode_cachep);
1391         if (nilfs_transaction_cachep)
1392                 kmem_cache_destroy(nilfs_transaction_cachep);
1393         if (nilfs_segbuf_cachep)
1394                 kmem_cache_destroy(nilfs_segbuf_cachep);
1395         if (nilfs_btree_path_cache)
1396                 kmem_cache_destroy(nilfs_btree_path_cache);
1397 }
1398
1399 static int __init nilfs_init_cachep(void)
1400 {
1401         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1402                         sizeof(struct nilfs_inode_info), 0,
1403                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1404         if (!nilfs_inode_cachep)
1405                 goto fail;
1406
1407         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1408                         sizeof(struct nilfs_transaction_info), 0,
1409                         SLAB_RECLAIM_ACCOUNT, NULL);
1410         if (!nilfs_transaction_cachep)
1411                 goto fail;
1412
1413         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1414                         sizeof(struct nilfs_segment_buffer), 0,
1415                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1416         if (!nilfs_segbuf_cachep)
1417                 goto fail;
1418
1419         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1420                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1421                         0, 0, NULL);
1422         if (!nilfs_btree_path_cache)
1423                 goto fail;
1424
1425         return 0;
1426
1427 fail:
1428         nilfs_destroy_cachep();
1429         return -ENOMEM;
1430 }
1431
1432 static int __init init_nilfs_fs(void)
1433 {
1434         int err;
1435
1436         err = nilfs_init_cachep();
1437         if (err)
1438                 goto fail;
1439
1440         err = register_filesystem(&nilfs_fs_type);
1441         if (err)
1442                 goto free_cachep;
1443
1444         printk(KERN_INFO "NILFS version 2 loaded\n");
1445         return 0;
1446
1447 free_cachep:
1448         nilfs_destroy_cachep();
1449 fail:
1450         return err;
1451 }
1452
1453 static void __exit exit_nilfs_fs(void)
1454 {
1455         nilfs_destroy_cachep();
1456         unregister_filesystem(&nilfs_fs_type);
1457 }
1458
1459 module_init(init_nilfs_fs)
1460 module_exit(exit_nilfs_fs)