fs/nilfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * NILFS module and super block management.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  */
   9 /*
  10  *  linux/fs/ext2/super.c
  11  *
  12  * Copyright (C) 1992, 1993, 1994, 1995
  13  * Remy Card ([email protected])
  14  * Laboratoire MASI - Institut Blaise Pascal
  15  * Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *  from
  18  *
  19  *  linux/fs/minix/inode.c
  20  *
  21  *  Copyright (C) 1991, 1992  Linus Torvalds
  22  *
  23  *  Big-endian to little-endian byte-swapping/bitmaps by
  24  *        David S. Miller ([email protected]), 1995
  25  */
  26
  27 #include <linux/module.h>
  28 #include <linux/string.h>
  29 #include <linux/slab.h>
  30 #include <linux/init.h>
  31 #include <linux/blkdev.h>
  32 #include <linux/crc32.h>
  33 #include <linux/vfs.h>
  34 #include <linux/writeback.h>
  35 #include <linux/seq_file.h>
  36 #include <linux/mount.h>
  37 #include <linux/fs_context.h>
  38 #include <linux/fs_parser.h>
  39 #include "nilfs.h"
  40 #include "export.h"
  41 #include "mdt.h"
  42 #include "alloc.h"
  43 #include "btree.h"
  44 #include "btnode.h"
  45 #include "page.h"
  46 #include "cpfile.h"
  47 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  48 #include "ifile.h"
  49 #include "dat.h"
  50 #include "segment.h"
  51 #include "segbuf.h"
  52
  53 MODULE_AUTHOR("NTT Corp.");
  54 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  55                    "(NILFS)");
  56 MODULE_LICENSE("GPL");
  57
  58 static struct kmem_cache *nilfs_inode_cachep;
  59 struct kmem_cache *nilfs_transaction_cachep;
  60 struct kmem_cache *nilfs_segbuf_cachep;
  61 struct kmem_cache *nilfs_btree_path_cache;
  62
  63 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  64
  65 void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
  66 {
  67         struct va_format vaf;
  68         va_list args;
  69         int level;
  70
  71         va_start(args, fmt);
  72
  73         level = printk_get_level(fmt);
  74         vaf.fmt = printk_skip_level(fmt);
  75         vaf.va = &args;
  76
  77         if (sb)
  78                 printk("%c%cNILFS (%s): %pV\n",
  79                        KERN_SOH_ASCII, level, sb->s_id, &vaf);
  80         else
  81                 printk("%c%cNILFS: %pV\n",
  82                        KERN_SOH_ASCII, level, &vaf);
  83
  84         va_end(args);
  85 }
  86
  87 static void nilfs_set_error(struct super_block *sb)
  88 {
  89         struct the_nilfs *nilfs = sb->s_fs_info;
  90         struct nilfs_super_block **sbp;
  91
  92         down_write(&nilfs->ns_sem);
  93         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  94                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  95                 sbp = nilfs_prepare_super(sb, 0);
  96                 if (likely(sbp)) {
  97                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  98                         if (sbp[1])
  99                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 100                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 101                 }
 102         }
 103         up_write(&nilfs->ns_sem);
 104 }
 105
 106 /**
 107  * __nilfs_error() - report failure condition on a filesystem
 108  *
 109  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 110  * reporting an error message.  This function should be called when
 111  * NILFS detects incoherences or defects of meta data on disk.
 112  *
 113  * This implements the body of nilfs_error() macro.  Normally,
 114  * nilfs_error() should be used.  As for sustainable errors such as a
 115  * single-shot I/O error, nilfs_err() should be used instead.
 116  *
 117  * Callers should not add a trailing newline since this will do it.
 118  */
 119 void __nilfs_error(struct super_block *sb, const char *function,
 120                    const char *fmt, ...)
 121 {
 122         struct the_nilfs *nilfs = sb->s_fs_info;
 123         struct va_format vaf;
 124         va_list args;
 125
 126         va_start(args, fmt);
 127
 128         vaf.fmt = fmt;
 129         vaf.va = &args;
 130
 131         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 132                sb->s_id, function, &vaf);
 133
 134         va_end(args);
 135
 136         if (!sb_rdonly(sb)) {
 137                 nilfs_set_error(sb);
 138
 139                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 140                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 141                         sb->s_flags |= SB_RDONLY;
 142                 }
 143         }
 144
 145         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 146                 panic("NILFS (device %s): panic forced after error\n",
 147                       sb->s_id);
 148 }
 149
 150 struct inode *nilfs_alloc_inode(struct super_block *sb)
 151 {
 152         struct nilfs_inode_info *ii;
 153
 154         ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
 155         if (!ii)
 156                 return NULL;
 157         ii->i_bh = NULL;
 158         ii->i_state = 0;
 159         ii->i_cno = 0;
 160         ii->i_assoc_inode = NULL;
 161         ii->i_bmap = &ii->i_bmap_data;
 162         return &ii->vfs_inode;
 163 }
 164
 165 static void nilfs_free_inode(struct inode *inode)
 166 {
 167         if (nilfs_is_metadata_file_inode(inode))
 168                 nilfs_mdt_destroy(inode);
 169
 170         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 171 }
 172
 173 static int nilfs_sync_super(struct super_block *sb, int flag)
 174 {
 175         struct the_nilfs *nilfs = sb->s_fs_info;
 176         int err;
 177
 178  retry:
 179         set_buffer_dirty(nilfs->ns_sbh[0]);
 180         if (nilfs_test_opt(nilfs, BARRIER)) {
 181                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 182                                           REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
 183         } else {
 184                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 185         }
 186
 187         if (unlikely(err)) {
 188                 nilfs_err(sb, "unable to write superblock: err=%d", err);
 189                 if (err == -EIO && nilfs->ns_sbh[1]) {
 190                         /*
 191                          * sbp[0] points to newer log than sbp[1],
 192                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 193                          */
 194                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 195                                nilfs->ns_sbsize);
 196                         nilfs_fall_back_super_block(nilfs);
 197                         goto retry;
 198                 }
 199         } else {
 200                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 201
 202                 nilfs->ns_sbwcount++;
 203
 204                 /*
 205                  * The latest segment becomes trailable from the position
 206                  * written in superblock.
 207                  */
 208                 clear_nilfs_discontinued(nilfs);
 209
 210                 /* update GC protection for recent segments */
 211                 if (nilfs->ns_sbh[1]) {
 212                         if (flag == NILFS_SB_COMMIT_ALL) {
 213                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 214                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 215                                         goto out;
 216                         }
 217                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 218                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 219                                 sbp = nilfs->ns_sbp[1];
 220                 }
 221
 222                 spin_lock(&nilfs->ns_last_segment_lock);
 223                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 224                 spin_unlock(&nilfs->ns_last_segment_lock);
 225         }
 226  out:
 227         return err;
 228 }
 229
 230 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 231                           struct the_nilfs *nilfs)
 232 {
 233         sector_t nfreeblocks;
 234
 235         /* nilfs->ns_sem must be locked by the caller. */
 236         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 237         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 238
 239         spin_lock(&nilfs->ns_last_segment_lock);
 240         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 241         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 242         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 243         spin_unlock(&nilfs->ns_last_segment_lock);
 244 }
 245
 246 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 247                                                int flip)
 248 {
 249         struct the_nilfs *nilfs = sb->s_fs_info;
 250         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 251
 252         /* nilfs->ns_sem must be locked by the caller. */
 253         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 254                 if (sbp[1] &&
 255                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 256                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 257                 } else {
 258                         nilfs_crit(sb, "superblock broke");
 259                         return NULL;
 260                 }
 261         } else if (sbp[1] &&
 262                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 263                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 264         }
 265
 266         if (flip && sbp[1])
 267                 nilfs_swap_super_block(nilfs);
 268
 269         return sbp;
 270 }
 271
 272 int nilfs_commit_super(struct super_block *sb, int flag)
 273 {
 274         struct the_nilfs *nilfs = sb->s_fs_info;
 275         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 276         time64_t t;
 277
 278         /* nilfs->ns_sem must be locked by the caller. */
 279         t = ktime_get_real_seconds();
 280         nilfs->ns_sbwtime = t;
 281         sbp[0]->s_wtime = cpu_to_le64(t);
 282         sbp[0]->s_sum = 0;
 283         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 284                                              (unsigned char *)sbp[0],
 285                                              nilfs->ns_sbsize));
 286         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 287                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 288                 sbp[1]->s_sum = 0;
 289                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 290                                             (unsigned char *)sbp[1],
 291                                             nilfs->ns_sbsize));
 292         }
 293         clear_nilfs_sb_dirty(nilfs);
 294         nilfs->ns_flushed_device = 1;
 295         /* make sure store to ns_flushed_device cannot be reordered */
 296         smp_wmb();
 297         return nilfs_sync_super(sb, flag);
 298 }
 299
 300 /**
 301  * nilfs_cleanup_super() - write filesystem state for cleanup
 302  * @sb: super block instance to be unmounted or degraded to read-only
 303  *
 304  * This function restores state flags in the on-disk super block.
 305  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 306  * filesystem was not clean previously.
 307  */
 308 int nilfs_cleanup_super(struct super_block *sb)
 309 {
 310         struct the_nilfs *nilfs = sb->s_fs_info;
 311         struct nilfs_super_block **sbp;
 312         int flag = NILFS_SB_COMMIT;
 313         int ret = -EIO;
 314
 315         sbp = nilfs_prepare_super(sb, 0);
 316         if (sbp) {
 317                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 318                 nilfs_set_log_cursor(sbp[0], nilfs);
 319                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 320                         /*
 321                          * make the "clean" flag also to the opposite
 322                          * super block if both super blocks point to
 323                          * the same checkpoint.
 324                          */
 325                         sbp[1]->s_state = sbp[0]->s_state;
 326                         flag = NILFS_SB_COMMIT_ALL;
 327                 }
 328                 ret = nilfs_commit_super(sb, flag);
 329         }
 330         return ret;
 331 }
 332
 333 /**
 334  * nilfs_move_2nd_super - relocate secondary super block
 335  * @sb: super block instance
 336  * @sb2off: new offset of the secondary super block (in bytes)
 337  */
 338 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 339 {
 340         struct the_nilfs *nilfs = sb->s_fs_info;
 341         struct buffer_head *nsbh;
 342         struct nilfs_super_block *nsbp;
 343         sector_t blocknr, newblocknr;
 344         unsigned long offset;
 345         int sb2i;  /* array index of the secondary superblock */
 346         int ret = 0;
 347
 348         /* nilfs->ns_sem must be locked by the caller. */
 349         if (nilfs->ns_sbh[1] &&
 350             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 351                 sb2i = 1;
 352                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 353         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 354                 sb2i = 0;
 355                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 356         } else {
 357                 sb2i = -1;
 358                 blocknr = 0;
 359         }
 360         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 361                 goto out;  /* super block location is unchanged */
 362
 363         /* Get new super block buffer */
 364         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 365         offset = sb2off & (nilfs->ns_blocksize - 1);
 366         nsbh = sb_getblk(sb, newblocknr);
 367         if (!nsbh) {
 368                 nilfs_warn(sb,
 369                            "unable to move secondary superblock to block %llu",
 370                            (unsigned long long)newblocknr);
 371                 ret = -EIO;
 372                 goto out;
 373         }
 374         nsbp = (void *)nsbh->b_data + offset;
 375
 376         lock_buffer(nsbh);
 377         if (sb2i >= 0) {
 378                 /*
 379                  * The position of the second superblock only changes by 4KiB,
 380                  * which is larger than the maximum superblock data size
 381                  * (= 1KiB), so there is no need to use memmove() to allow
 382                  * overlap between source and destination.
 383                  */
 384                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 385
 386                 /*
 387                  * Zero fill after copy to avoid overwriting in case of move
 388                  * within the same block.
 389                  */
 390                 memset(nsbh->b_data, 0, offset);
 391                 memset((void *)nsbp + nilfs->ns_sbsize, 0,
 392                        nsbh->b_size - offset - nilfs->ns_sbsize);
 393         } else {
 394                 memset(nsbh->b_data, 0, nsbh->b_size);
 395         }
 396         set_buffer_uptodate(nsbh);
 397         unlock_buffer(nsbh);
 398
 399         if (sb2i >= 0) {
 400                 brelse(nilfs->ns_sbh[sb2i]);
 401                 nilfs->ns_sbh[sb2i] = nsbh;
 402                 nilfs->ns_sbp[sb2i] = nsbp;
 403         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 404                 /* secondary super block will be restored to index 1 */
 405                 nilfs->ns_sbh[1] = nsbh;
 406                 nilfs->ns_sbp[1] = nsbp;
 407         } else {
 408                 brelse(nsbh);
 409         }
 410 out:
 411         return ret;
 412 }
 413
 414 /**
 415  * nilfs_resize_fs - resize the filesystem
 416  * @sb: super block instance
 417  * @newsize: new size of the filesystem (in bytes)
 418  */
 419 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 420 {
 421         struct the_nilfs *nilfs = sb->s_fs_info;
 422         struct nilfs_super_block **sbp;
 423         __u64 devsize, newnsegs;
 424         loff_t sb2off;
 425         int ret;
 426
 427         ret = -ERANGE;
 428         devsize = bdev_nr_bytes(sb->s_bdev);
 429         if (newsize > devsize)
 430                 goto out;
 431
 432         /*
 433          * Prevent underflow in second superblock position calculation.
 434          * The exact minimum size check is done in nilfs_sufile_resize().
 435          */
 436         if (newsize < 4096) {
 437                 ret = -ENOSPC;
 438                 goto out;
 439         }
 440
 441         /*
 442          * Write lock is required to protect some functions depending
 443          * on the number of segments, the number of reserved segments,
 444          * and so forth.
 445          */
 446         down_write(&nilfs->ns_segctor_sem);
 447
 448         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 449         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 450         newnsegs = div64_ul(newnsegs, nilfs->ns_blocks_per_segment);
 451
 452         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 453         up_write(&nilfs->ns_segctor_sem);
 454         if (ret < 0)
 455                 goto out;
 456
 457         ret = nilfs_construct_segment(sb);
 458         if (ret < 0)
 459                 goto out;
 460
 461         down_write(&nilfs->ns_sem);
 462         nilfs_move_2nd_super(sb, sb2off);
 463         ret = -EIO;
 464         sbp = nilfs_prepare_super(sb, 0);
 465         if (likely(sbp)) {
 466                 nilfs_set_log_cursor(sbp[0], nilfs);
 467                 /*
 468                  * Drop NILFS_RESIZE_FS flag for compatibility with
 469                  * mount-time resize which may be implemented in a
 470                  * future release.
 471                  */
 472                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 473                                               ~NILFS_RESIZE_FS);
 474                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 475                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 476                 if (sbp[1])
 477                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 478                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 479         }
 480         up_write(&nilfs->ns_sem);
 481
 482         /*
 483          * Reset the range of allocatable segments last.  This order
 484          * is important in the case of expansion because the secondary
 485          * superblock must be protected from log write until migration
 486          * completes.
 487          */
 488         if (!ret)
 489                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 490 out:
 491         return ret;
 492 }
 493
 494 static void nilfs_put_super(struct super_block *sb)
 495 {
 496         struct the_nilfs *nilfs = sb->s_fs_info;
 497
 498         nilfs_detach_log_writer(sb);
 499
 500         if (!sb_rdonly(sb)) {
 501                 down_write(&nilfs->ns_sem);
 502                 nilfs_cleanup_super(sb);
 503                 up_write(&nilfs->ns_sem);
 504         }
 505
 506         nilfs_sysfs_delete_device_group(nilfs);
 507         iput(nilfs->ns_sufile);
 508         iput(nilfs->ns_cpfile);
 509         iput(nilfs->ns_dat);
 510
 511         destroy_nilfs(nilfs);
 512         sb->s_fs_info = NULL;
 513 }
 514
 515 static int nilfs_sync_fs(struct super_block *sb, int wait)
 516 {
 517         struct the_nilfs *nilfs = sb->s_fs_info;
 518         struct nilfs_super_block **sbp;
 519         int err = 0;
 520
 521         /* This function is called when super block should be written back */
 522         if (wait)
 523                 err = nilfs_construct_segment(sb);
 524
 525         down_write(&nilfs->ns_sem);
 526         if (nilfs_sb_dirty(nilfs)) {
 527                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 528                 if (likely(sbp)) {
 529                         nilfs_set_log_cursor(sbp[0], nilfs);
 530                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 531                 }
 532         }
 533         up_write(&nilfs->ns_sem);
 534
 535         if (!err)
 536                 err = nilfs_flush_device(nilfs);
 537
 538         return err;
 539 }
 540
 541 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 542                             struct nilfs_root **rootp)
 543 {
 544         struct the_nilfs *nilfs = sb->s_fs_info;
 545         struct nilfs_root *root;
 546         int err = -ENOMEM;
 547
 548         root = nilfs_find_or_create_root(
 549                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 550         if (!root)
 551                 return err;
 552
 553         if (root->ifile)
 554                 goto reuse; /* already attached checkpoint */
 555
 556         down_read(&nilfs->ns_segctor_sem);
 557         err = nilfs_ifile_read(sb, root, cno, nilfs->ns_inode_size);
 558         up_read(&nilfs->ns_segctor_sem);
 559         if (unlikely(err))
 560                 goto failed;
 561
 562  reuse:
 563         *rootp = root;
 564         return 0;
 565
 566  failed:
 567         if (err == -EINVAL)
 568                 nilfs_err(sb, "Invalid checkpoint (checkpoint number=%llu)",
 569                           (unsigned long long)cno);
 570         nilfs_put_root(root);
 571
 572         return err;
 573 }
 574
 575 static int nilfs_freeze(struct super_block *sb)
 576 {
 577         struct the_nilfs *nilfs = sb->s_fs_info;
 578         int err;
 579
 580         if (sb_rdonly(sb))
 581                 return 0;
 582
 583         /* Mark super block clean */
 584         down_write(&nilfs->ns_sem);
 585         err = nilfs_cleanup_super(sb);
 586         up_write(&nilfs->ns_sem);
 587         return err;
 588 }
 589
 590 static int nilfs_unfreeze(struct super_block *sb)
 591 {
 592         struct the_nilfs *nilfs = sb->s_fs_info;
 593
 594         if (sb_rdonly(sb))
 595                 return 0;
 596
 597         down_write(&nilfs->ns_sem);
 598         nilfs_setup_super(sb, false);
 599         up_write(&nilfs->ns_sem);
 600         return 0;
 601 }
 602
 603 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 604 {
 605         struct super_block *sb = dentry->d_sb;
 606         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 607         struct the_nilfs *nilfs = root->nilfs;
 608         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 609         unsigned long long blocks;
 610         unsigned long overhead;
 611         unsigned long nrsvblocks;
 612         sector_t nfreeblocks;
 613         u64 nmaxinodes, nfreeinodes;
 614         int err;
 615
 616         /*
 617          * Compute all of the segment blocks
 618          *
 619          * The blocks before first segment and after last segment
 620          * are excluded.
 621          */
 622         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 623                 - nilfs->ns_first_data_block;
 624         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 625
 626         /*
 627          * Compute the overhead
 628          *
 629          * When distributing meta data blocks outside segment structure,
 630          * We must count them as the overhead.
 631          */
 632         overhead = 0;
 633
 634         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 635         if (unlikely(err))
 636                 return err;
 637
 638         err = nilfs_ifile_count_free_inodes(root->ifile,
 639                                             &nmaxinodes, &nfreeinodes);
 640         if (unlikely(err)) {
 641                 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
 642                 if (err == -ERANGE) {
 643                         /*
 644                          * If nilfs_palloc_count_max_entries() returns
 645                          * -ERANGE error code then we simply treat
 646                          * curent inodes count as maximum possible and
 647                          * zero as free inodes value.
 648                          */
 649                         nmaxinodes = atomic64_read(&root->inodes_count);
 650                         nfreeinodes = 0;
 651                         err = 0;
 652                 } else
 653                         return err;
 654         }
 655
 656         buf->f_type = NILFS_SUPER_MAGIC;
 657         buf->f_bsize = sb->s_blocksize;
 658         buf->f_blocks = blocks - overhead;
 659         buf->f_bfree = nfreeblocks;
 660         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 661                 (buf->f_bfree - nrsvblocks) : 0;
 662         buf->f_files = nmaxinodes;
 663         buf->f_ffree = nfreeinodes;
 664         buf->f_namelen = NILFS_NAME_LEN;
 665         buf->f_fsid = u64_to_fsid(id);
 666
 667         return 0;
 668 }
 669
 670 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 671 {
 672         struct super_block *sb = dentry->d_sb;
 673         struct the_nilfs *nilfs = sb->s_fs_info;
 674         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 675
 676         if (!nilfs_test_opt(nilfs, BARRIER))
 677                 seq_puts(seq, ",nobarrier");
 678         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 679                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 680         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 681                 seq_puts(seq, ",errors=panic");
 682         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 683                 seq_puts(seq, ",errors=continue");
 684         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 685                 seq_puts(seq, ",order=strict");
 686         if (nilfs_test_opt(nilfs, NORECOVERY))
 687                 seq_puts(seq, ",norecovery");
 688         if (nilfs_test_opt(nilfs, DISCARD))
 689                 seq_puts(seq, ",discard");
 690
 691         return 0;
 692 }
 693
 694 static const struct super_operations nilfs_sops = {
 695         .alloc_inode    = nilfs_alloc_inode,
 696         .free_inode     = nilfs_free_inode,
 697         .dirty_inode    = nilfs_dirty_inode,
 698         .evict_inode    = nilfs_evict_inode,
 699         .put_super      = nilfs_put_super,
 700         .sync_fs        = nilfs_sync_fs,
 701         .freeze_fs      = nilfs_freeze,
 702         .unfreeze_fs    = nilfs_unfreeze,
 703         .statfs         = nilfs_statfs,
 704         .show_options = nilfs_show_options
 705 };
 706
 707 enum {
 708         Opt_err, Opt_barrier, Opt_snapshot, Opt_order, Opt_norecovery,
 709         Opt_discard,
 710 };
 711
 712 static const struct constant_table nilfs_param_err[] = {
 713         {"continue",    NILFS_MOUNT_ERRORS_CONT},
 714         {"panic",       NILFS_MOUNT_ERRORS_PANIC},
 715         {"remount-ro",  NILFS_MOUNT_ERRORS_RO},
 716         {}
 717 };
 718
 719 static const struct fs_parameter_spec nilfs_param_spec[] = {
 720         fsparam_enum    ("errors", Opt_err, nilfs_param_err),
 721         fsparam_flag_no ("barrier", Opt_barrier),
 722         fsparam_u64     ("cp", Opt_snapshot),
 723         fsparam_string  ("order", Opt_order),
 724         fsparam_flag    ("norecovery", Opt_norecovery),
 725         fsparam_flag_no ("discard", Opt_discard),
 726         {}
 727 };
 728
 729 struct nilfs_fs_context {
 730         unsigned long ns_mount_opt;
 731         __u64 cno;
 732 };
 733
 734 static int nilfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
 735 {
 736         struct nilfs_fs_context *nilfs = fc->fs_private;
 737         int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
 738         struct fs_parse_result result;
 739         int opt;
 740
 741         opt = fs_parse(fc, nilfs_param_spec, param, &result);
 742         if (opt < 0)
 743                 return opt;
 744
 745         switch (opt) {
 746         case Opt_barrier:
 747                 if (result.negated)
 748                         nilfs_clear_opt(nilfs, BARRIER);
 749                 else
 750                         nilfs_set_opt(nilfs, BARRIER);
 751                 break;
 752         case Opt_order:
 753                 if (strcmp(param->string, "relaxed") == 0)
 754                         /* Ordered data semantics */
 755                         nilfs_clear_opt(nilfs, STRICT_ORDER);
 756                 else if (strcmp(param->string, "strict") == 0)
 757                         /* Strict in-order semantics */
 758                         nilfs_set_opt(nilfs, STRICT_ORDER);
 759                 else
 760                         return -EINVAL;
 761                 break;
 762         case Opt_err:
 763                 nilfs->ns_mount_opt &= ~NILFS_MOUNT_ERROR_MODE;
 764                 nilfs->ns_mount_opt |= result.uint_32;
 765                 break;
 766         case Opt_snapshot:
 767                 if (is_remount) {
 768                         struct super_block *sb = fc->root->d_sb;
 769
 770                         nilfs_err(sb,
 771                                   "\"%s\" option is invalid for remount",
 772                                   param->key);
 773                         return -EINVAL;
 774                 }
 775                 if (result.uint_64 == 0) {
 776                         nilfs_err(NULL,
 777                                   "invalid option \"cp=0\": invalid checkpoint number 0");
 778                         return -EINVAL;
 779                 }
 780                 nilfs->cno = result.uint_64;
 781                 break;
 782         case Opt_norecovery:
 783                 nilfs_set_opt(nilfs, NORECOVERY);
 784                 break;
 785         case Opt_discard:
 786                 if (result.negated)
 787                         nilfs_clear_opt(nilfs, DISCARD);
 788                 else
 789                         nilfs_set_opt(nilfs, DISCARD);
 790                 break;
 791         default:
 792                 return -EINVAL;
 793         }
 794
 795         return 0;
 796 }
 797
 798 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 799 {
 800         struct the_nilfs *nilfs = sb->s_fs_info;
 801         struct nilfs_super_block **sbp;
 802         int max_mnt_count;
 803         int mnt_count;
 804
 805         /* nilfs->ns_sem must be locked by the caller. */
 806         sbp = nilfs_prepare_super(sb, 0);
 807         if (!sbp)
 808                 return -EIO;
 809
 810         if (!is_mount)
 811                 goto skip_mount_setup;
 812
 813         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 814         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 815
 816         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 817                 nilfs_warn(sb, "mounting fs with errors");
 818 #if 0
 819         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 820                 nilfs_warn(sb, "maximal mount count reached");
 821 #endif
 822         }
 823         if (!max_mnt_count)
 824                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 825
 826         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 827         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 828
 829 skip_mount_setup:
 830         sbp[0]->s_state =
 831                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 832         /* synchronize sbp[1] with sbp[0] */
 833         if (sbp[1])
 834                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 835         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 836 }
 837
 838 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 839                                                  u64 pos, int blocksize,
 840                                                  struct buffer_head **pbh)
 841 {
 842         unsigned long long sb_index = pos;
 843         unsigned long offset;
 844
 845         offset = do_div(sb_index, blocksize);
 846         *pbh = sb_bread(sb, sb_index);
 847         if (!*pbh)
 848                 return NULL;
 849         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 850 }
 851
 852 int nilfs_store_magic(struct super_block *sb,
 853                       struct nilfs_super_block *sbp)
 854 {
 855         struct the_nilfs *nilfs = sb->s_fs_info;
 856
 857         sb->s_magic = le16_to_cpu(sbp->s_magic);
 858
 859         /* FS independent flags */
 860 #ifdef NILFS_ATIME_DISABLE
 861         sb->s_flags |= SB_NOATIME;
 862 #endif
 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 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                 nilfs_err(sb,
 881                           "couldn't mount because of unsupported optional features (%llx)",
 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_rdonly(sb) && features) {
 888                 nilfs_err(sb,
 889                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 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                 ret = PTR_ERR(inode);
 907                 nilfs_err(sb, "error %d getting root inode", ret);
 908                 goto out;
 909         }
 910         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 911                 iput(inode);
 912                 nilfs_err(sb, "corrupt root inode");
 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_root(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         nilfs_err(sb, "error %d getting root dentry", ret);
 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         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 952
 953         down_read(&nilfs->ns_segctor_sem);
 954         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 955         up_read(&nilfs->ns_segctor_sem);
 956         if (ret < 0) {
 957                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 958                 goto out;
 959         } else if (!ret) {
 960                 nilfs_err(s,
 961                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 962                           (unsigned long long)cno);
 963                 ret = -EINVAL;
 964                 goto out;
 965         }
 966
 967         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 968         if (ret) {
 969                 nilfs_err(s,
 970                           "error %d while loading snapshot (checkpoint number=%llu)",
 971                           ret, (unsigned long long)cno);
 972                 goto out;
 973         }
 974         ret = nilfs_get_root_dentry(s, root, root_dentry);
 975         nilfs_put_root(root);
 976  out:
 977         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 978         return ret;
 979 }
 980
 981 /**
 982  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 983  * @root_dentry: root dentry of the tree to be shrunk
 984  *
 985  * This function returns true if the tree was in-use.
 986  */
 987 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
 988 {
 989         shrink_dcache_parent(root_dentry);
 990         return d_count(root_dentry) > 1;
 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 > 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                                 ret = nilfs_tree_is_busy(dentry);
1015                                 dput(dentry);
1016                         }
1017                         iput(inode);
1018                 }
1019                 nilfs_put_root(root);
1020         }
1021         return ret;
1022 }
1023
1024 /**
1025  * nilfs_fill_super() - initialize a super block instance
1026  * @sb: super_block
1027  * @fc: filesystem context
1028  *
1029  * This function is called exclusively by nilfs->ns_mount_mutex.
1030  * So, the recovery process is protected from other simultaneous mounts.
1031  */
1032 static int
1033 nilfs_fill_super(struct super_block *sb, struct fs_context *fc)
1034 {
1035         struct the_nilfs *nilfs;
1036         struct nilfs_root *fsroot;
1037         struct nilfs_fs_context *ctx = fc->fs_private;
1038         __u64 cno;
1039         int err;
1040
1041         nilfs = alloc_nilfs(sb);
1042         if (!nilfs)
1043                 return -ENOMEM;
1044
1045         sb->s_fs_info = nilfs;
1046
1047         err = init_nilfs(nilfs, sb);
1048         if (err)
1049                 goto failed_nilfs;
1050
1051         /* Copy in parsed mount options */
1052         nilfs->ns_mount_opt = ctx->ns_mount_opt;
1053
1054         sb->s_op = &nilfs_sops;
1055         sb->s_export_op = &nilfs_export_ops;
1056         sb->s_root = NULL;
1057         sb->s_time_gran = 1;
1058         sb->s_max_links = NILFS_LINK_MAX;
1059
1060         sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
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                 nilfs_err(sb,
1070                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1071                           err, (unsigned long long)cno);
1072                 goto failed_unload;
1073         }
1074
1075         if (!sb_rdonly(sb)) {
1076                 err = nilfs_attach_log_writer(sb, fsroot);
1077                 if (err)
1078                         goto failed_checkpoint;
1079         }
1080
1081         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1082         if (err)
1083                 goto failed_segctor;
1084
1085         nilfs_put_root(fsroot);
1086
1087         if (!sb_rdonly(sb)) {
1088                 down_write(&nilfs->ns_sem);
1089                 nilfs_setup_super(sb, true);
1090                 up_write(&nilfs->ns_sem);
1091         }
1092
1093         return 0;
1094
1095  failed_segctor:
1096         nilfs_detach_log_writer(sb);
1097
1098  failed_checkpoint:
1099         nilfs_put_root(fsroot);
1100
1101  failed_unload:
1102         nilfs_sysfs_delete_device_group(nilfs);
1103         iput(nilfs->ns_sufile);
1104         iput(nilfs->ns_cpfile);
1105         iput(nilfs->ns_dat);
1106
1107  failed_nilfs:
1108         destroy_nilfs(nilfs);
1109         return err;
1110 }
1111
1112 static int nilfs_reconfigure(struct fs_context *fc)
1113 {
1114         struct nilfs_fs_context *ctx = fc->fs_private;
1115         struct super_block *sb = fc->root->d_sb;
1116         struct the_nilfs *nilfs = sb->s_fs_info;
1117         int err;
1118
1119         sync_filesystem(sb);
1120
1121         err = -EINVAL;
1122
1123         if (!nilfs_valid_fs(nilfs)) {
1124                 nilfs_warn(sb,
1125                            "couldn't remount because the filesystem is in an incomplete recovery state");
1126                 goto ignore_opts;
1127         }
1128         if ((bool)(fc->sb_flags & SB_RDONLY) == sb_rdonly(sb))
1129                 goto out;
1130         if (fc->sb_flags & SB_RDONLY) {
1131                 sb->s_flags |= SB_RDONLY;
1132
1133                 /*
1134                  * Remounting a valid RW partition RDONLY, so set
1135                  * the RDONLY flag and then mark the partition as valid again.
1136                  */
1137                 down_write(&nilfs->ns_sem);
1138                 nilfs_cleanup_super(sb);
1139                 up_write(&nilfs->ns_sem);
1140         } else {
1141                 __u64 features;
1142                 struct nilfs_root *root;
1143
1144                 /*
1145                  * Mounting a RDONLY partition read-write, so reread and
1146                  * store the current valid flag.  (It may have been changed
1147                  * by fsck since we originally mounted the partition.)
1148                  */
1149                 down_read(&nilfs->ns_sem);
1150                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1151                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1152                 up_read(&nilfs->ns_sem);
1153                 if (features) {
1154                         nilfs_warn(sb,
1155                                    "couldn't remount RDWR because of unsupported optional features (%llx)",
1156                                    (unsigned long long)features);
1157                         err = -EROFS;
1158                         goto ignore_opts;
1159                 }
1160
1161                 sb->s_flags &= ~SB_RDONLY;
1162
1163                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1164                 err = nilfs_attach_log_writer(sb, root);
1165                 if (err) {
1166                         sb->s_flags |= SB_RDONLY;
1167                         goto ignore_opts;
1168                 }
1169
1170                 down_write(&nilfs->ns_sem);
1171                 nilfs_setup_super(sb, true);
1172                 up_write(&nilfs->ns_sem);
1173         }
1174  out:
1175         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1176         /* Copy over parsed remount options */
1177         nilfs->ns_mount_opt = ctx->ns_mount_opt;
1178
1179         return 0;
1180
1181  ignore_opts:
1182         return err;
1183 }
1184
1185 static int
1186 nilfs_get_tree(struct fs_context *fc)
1187 {
1188         struct nilfs_fs_context *ctx = fc->fs_private;
1189         struct super_block *s;
1190         dev_t dev;
1191         int err;
1192
1193         if (ctx->cno && !(fc->sb_flags & SB_RDONLY)) {
1194                 nilfs_err(NULL,
1195                           "invalid option \"cp=%llu\": read-only option is not specified",
1196                           ctx->cno);
1197                 return -EINVAL;
1198         }
1199
1200         err = lookup_bdev(fc->source, &dev);
1201         if (err)
1202                 return err;
1203
1204         s = sget_dev(fc, dev);
1205         if (IS_ERR(s))
1206                 return PTR_ERR(s);
1207
1208         if (!s->s_root) {
1209                 err = setup_bdev_super(s, fc->sb_flags, fc);
1210                 if (!err)
1211                         err = nilfs_fill_super(s, fc);
1212                 if (err)
1213                         goto failed_super;
1214
1215                 s->s_flags |= SB_ACTIVE;
1216         } else if (!ctx->cno) {
1217                 if (nilfs_tree_is_busy(s->s_root)) {
1218                         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1219                                 nilfs_err(s,
1220                                           "the device already has a %s mount.",
1221                                           sb_rdonly(s) ? "read-only" : "read/write");
1222                                 err = -EBUSY;
1223                                 goto failed_super;
1224                         }
1225                 } else {
1226                         /*
1227                          * Try reconfigure to setup mount states if the current
1228                          * tree is not mounted and only snapshots use this sb.
1229                          *
1230                          * Since nilfs_reconfigure() requires fc->root to be
1231                          * set, set it first and release it on failure.
1232                          */
1233                         fc->root = dget(s->s_root);
1234                         err = nilfs_reconfigure(fc);
1235                         if (err) {
1236                                 dput(fc->root);
1237                                 fc->root = NULL;  /* prevent double release */
1238                                 goto failed_super;
1239                         }
1240                         return 0;
1241                 }
1242         }
1243
1244         if (ctx->cno) {
1245                 struct dentry *root_dentry;
1246
1247                 err = nilfs_attach_snapshot(s, ctx->cno, &root_dentry);
1248                 if (err)
1249                         goto failed_super;
1250                 fc->root = root_dentry;
1251                 return 0;
1252         }
1253
1254         fc->root = dget(s->s_root);
1255         return 0;
1256
1257  failed_super:
1258         deactivate_locked_super(s);
1259         return err;
1260 }
1261
1262 static void nilfs_free_fc(struct fs_context *fc)
1263 {
1264         kfree(fc->fs_private);
1265 }
1266
1267 static const struct fs_context_operations nilfs_context_ops = {
1268         .parse_param    = nilfs_parse_param,
1269         .get_tree       = nilfs_get_tree,
1270         .reconfigure    = nilfs_reconfigure,
1271         .free           = nilfs_free_fc,
1272 };
1273
1274 static int nilfs_init_fs_context(struct fs_context *fc)
1275 {
1276         struct nilfs_fs_context *ctx;
1277
1278         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1279         if (!ctx)
1280                 return -ENOMEM;
1281
1282         ctx->ns_mount_opt = NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
1283         fc->fs_private = ctx;
1284         fc->ops = &nilfs_context_ops;
1285
1286         return 0;
1287 }
1288
1289 struct file_system_type nilfs_fs_type = {
1290         .owner    = THIS_MODULE,
1291         .name     = "nilfs2",
1292         .kill_sb  = kill_block_super,
1293         .fs_flags = FS_REQUIRES_DEV,
1294         .init_fs_context = nilfs_init_fs_context,
1295         .parameters = nilfs_param_spec,
1296 };
1297 MODULE_ALIAS_FS("nilfs2");
1298
1299 static void nilfs_inode_init_once(void *obj)
1300 {
1301         struct nilfs_inode_info *ii = obj;
1302
1303         INIT_LIST_HEAD(&ii->i_dirty);
1304 #ifdef CONFIG_NILFS_XATTR
1305         init_rwsem(&ii->xattr_sem);
1306 #endif
1307         inode_init_once(&ii->vfs_inode);
1308 }
1309
1310 static void nilfs_segbuf_init_once(void *obj)
1311 {
1312         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1313 }
1314
1315 static void nilfs_destroy_cachep(void)
1316 {
1317         /*
1318          * Make sure all delayed rcu free inodes are flushed before we
1319          * destroy cache.
1320          */
1321         rcu_barrier();
1322
1323         kmem_cache_destroy(nilfs_inode_cachep);
1324         kmem_cache_destroy(nilfs_transaction_cachep);
1325         kmem_cache_destroy(nilfs_segbuf_cachep);
1326         kmem_cache_destroy(nilfs_btree_path_cache);
1327 }
1328
1329 static int __init nilfs_init_cachep(void)
1330 {
1331         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1332                         sizeof(struct nilfs_inode_info), 0,
1333                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1334                         nilfs_inode_init_once);
1335         if (!nilfs_inode_cachep)
1336                 goto fail;
1337
1338         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1339                         sizeof(struct nilfs_transaction_info), 0,
1340                         SLAB_RECLAIM_ACCOUNT, NULL);
1341         if (!nilfs_transaction_cachep)
1342                 goto fail;
1343
1344         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1345                         sizeof(struct nilfs_segment_buffer), 0,
1346                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1347         if (!nilfs_segbuf_cachep)
1348                 goto fail;
1349
1350         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1351                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1352                         0, 0, NULL);
1353         if (!nilfs_btree_path_cache)
1354                 goto fail;
1355
1356         return 0;
1357
1358 fail:
1359         nilfs_destroy_cachep();
1360         return -ENOMEM;
1361 }
1362
1363 static int __init init_nilfs_fs(void)
1364 {
1365         int err;
1366
1367         err = nilfs_init_cachep();
1368         if (err)
1369                 goto fail;
1370
1371         err = nilfs_sysfs_init();
1372         if (err)
1373                 goto free_cachep;
1374
1375         err = register_filesystem(&nilfs_fs_type);
1376         if (err)
1377                 goto deinit_sysfs_entry;
1378
1379         printk(KERN_INFO "NILFS version 2 loaded\n");
1380         return 0;
1381
1382 deinit_sysfs_entry:
1383         nilfs_sysfs_exit();
1384 free_cachep:
1385         nilfs_destroy_cachep();
1386 fail:
1387         return err;
1388 }
1389
1390 static void __exit exit_nilfs_fs(void)
1391 {
1392         nilfs_destroy_cachep();
1393         nilfs_sysfs_exit();
1394         unregister_filesystem(&nilfs_fs_type);
1395 }
1396
1397 module_init(init_nilfs_fs)
1398 module_exit(exit_nilfs_fs)