fs/nilfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * super.c - 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/parser.h>
  33 #include <linux/crc32.h>
  34 #include <linux/vfs.h>
  35 #include <linux/writeback.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/mount.h>
  38 #include "nilfs.h"
  39 #include "export.h"
  40 #include "mdt.h"
  41 #include "alloc.h"
  42 #include "btree.h"
  43 #include "btnode.h"
  44 #include "page.h"
  45 #include "cpfile.h"
  46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  47 #include "ifile.h"
  48 #include "dat.h"
  49 #include "segment.h"
  50 #include "segbuf.h"
  51
  52 MODULE_AUTHOR("NTT Corp.");
  53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  54                    "(NILFS)");
  55 MODULE_LICENSE("GPL");
  56
  57 static struct kmem_cache *nilfs_inode_cachep;
  58 struct kmem_cache *nilfs_transaction_cachep;
  59 struct kmem_cache *nilfs_segbuf_cachep;
  60 struct kmem_cache *nilfs_btree_path_cache;
  61
  62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  64
  65 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
  66                  ...)
  67 {
  68         struct va_format vaf;
  69         va_list args;
  70
  71         va_start(args, fmt);
  72         vaf.fmt = fmt;
  73         vaf.va = &args;
  74         if (sb)
  75                 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
  76         else
  77                 printk("%sNILFS: %pV\n", level, &vaf);
  78         va_end(args);
  79 }
  80
  81 static void nilfs_set_error(struct super_block *sb)
  82 {
  83         struct the_nilfs *nilfs = sb->s_fs_info;
  84         struct nilfs_super_block **sbp;
  85
  86         down_write(&nilfs->ns_sem);
  87         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  88                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  89                 sbp = nilfs_prepare_super(sb, 0);
  90                 if (likely(sbp)) {
  91                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  92                         if (sbp[1])
  93                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  94                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
  95                 }
  96         }
  97         up_write(&nilfs->ns_sem);
  98 }
  99
 100 /**
 101  * __nilfs_error() - report failure condition on a filesystem
 102  *
 103  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 104  * reporting an error message.  This function should be called when
 105  * NILFS detects incoherences or defects of meta data on disk.
 106  *
 107  * This implements the body of nilfs_error() macro.  Normally,
 108  * nilfs_error() should be used.  As for sustainable errors such as a
 109  * single-shot I/O error, nilfs_msg() should be used instead.
 110  *
 111  * Callers should not add a trailing newline since this will do it.
 112  */
 113 void __nilfs_error(struct super_block *sb, const char *function,
 114                    const char *fmt, ...)
 115 {
 116         struct the_nilfs *nilfs = sb->s_fs_info;
 117         struct va_format vaf;
 118         va_list args;
 119
 120         va_start(args, fmt);
 121
 122         vaf.fmt = fmt;
 123         vaf.va = &args;
 124
 125         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 126                sb->s_id, function, &vaf);
 127
 128         va_end(args);
 129
 130         if (!sb_rdonly(sb)) {
 131                 nilfs_set_error(sb);
 132
 133                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 134                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 135                         sb->s_flags |= SB_RDONLY;
 136                 }
 137         }
 138
 139         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 140                 panic("NILFS (device %s): panic forced after error\n",
 141                       sb->s_id);
 142 }
 143
 144 struct inode *nilfs_alloc_inode(struct super_block *sb)
 145 {
 146         struct nilfs_inode_info *ii;
 147
 148         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 149         if (!ii)
 150                 return NULL;
 151         ii->i_bh = NULL;
 152         ii->i_state = 0;
 153         ii->i_cno = 0;
 154         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
 155         return &ii->vfs_inode;
 156 }
 157
 158 static void nilfs_free_inode(struct inode *inode)
 159 {
 160         if (nilfs_is_metadata_file_inode(inode))
 161                 nilfs_mdt_destroy(inode);
 162
 163         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 164 }
 165
 166 static int nilfs_sync_super(struct super_block *sb, int flag)
 167 {
 168         struct the_nilfs *nilfs = sb->s_fs_info;
 169         int err;
 170
 171  retry:
 172         set_buffer_dirty(nilfs->ns_sbh[0]);
 173         if (nilfs_test_opt(nilfs, BARRIER)) {
 174                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 175                                           REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
 176         } else {
 177                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 178         }
 179
 180         if (unlikely(err)) {
 181                 nilfs_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
 182                           err);
 183                 if (err == -EIO && nilfs->ns_sbh[1]) {
 184                         /*
 185                          * sbp[0] points to newer log than sbp[1],
 186                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 187                          */
 188                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 189                                nilfs->ns_sbsize);
 190                         nilfs_fall_back_super_block(nilfs);
 191                         goto retry;
 192                 }
 193         } else {
 194                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 195
 196                 nilfs->ns_sbwcount++;
 197
 198                 /*
 199                  * The latest segment becomes trailable from the position
 200                  * written in superblock.
 201                  */
 202                 clear_nilfs_discontinued(nilfs);
 203
 204                 /* update GC protection for recent segments */
 205                 if (nilfs->ns_sbh[1]) {
 206                         if (flag == NILFS_SB_COMMIT_ALL) {
 207                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 208                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 209                                         goto out;
 210                         }
 211                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 212                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 213                                 sbp = nilfs->ns_sbp[1];
 214                 }
 215
 216                 spin_lock(&nilfs->ns_last_segment_lock);
 217                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 218                 spin_unlock(&nilfs->ns_last_segment_lock);
 219         }
 220  out:
 221         return err;
 222 }
 223
 224 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 225                           struct the_nilfs *nilfs)
 226 {
 227         sector_t nfreeblocks;
 228
 229         /* nilfs->ns_sem must be locked by the caller. */
 230         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 231         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 232
 233         spin_lock(&nilfs->ns_last_segment_lock);
 234         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 235         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 236         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 237         spin_unlock(&nilfs->ns_last_segment_lock);
 238 }
 239
 240 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 241                                                int flip)
 242 {
 243         struct the_nilfs *nilfs = sb->s_fs_info;
 244         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 245
 246         /* nilfs->ns_sem must be locked by the caller. */
 247         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 248                 if (sbp[1] &&
 249                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 250                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 251                 } else {
 252                         nilfs_msg(sb, KERN_CRIT, "superblock broke");
 253                         return NULL;
 254                 }
 255         } else if (sbp[1] &&
 256                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 257                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 258         }
 259
 260         if (flip && sbp[1])
 261                 nilfs_swap_super_block(nilfs);
 262
 263         return sbp;
 264 }
 265
 266 int nilfs_commit_super(struct super_block *sb, int flag)
 267 {
 268         struct the_nilfs *nilfs = sb->s_fs_info;
 269         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 270         time64_t t;
 271
 272         /* nilfs->ns_sem must be locked by the caller. */
 273         t = ktime_get_real_seconds();
 274         nilfs->ns_sbwtime = t;
 275         sbp[0]->s_wtime = cpu_to_le64(t);
 276         sbp[0]->s_sum = 0;
 277         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 278                                              (unsigned char *)sbp[0],
 279                                              nilfs->ns_sbsize));
 280         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 281                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 282                 sbp[1]->s_sum = 0;
 283                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 284                                             (unsigned char *)sbp[1],
 285                                             nilfs->ns_sbsize));
 286         }
 287         clear_nilfs_sb_dirty(nilfs);
 288         nilfs->ns_flushed_device = 1;
 289         /* make sure store to ns_flushed_device cannot be reordered */
 290         smp_wmb();
 291         return nilfs_sync_super(sb, flag);
 292 }
 293
 294 /**
 295  * nilfs_cleanup_super() - write filesystem state for cleanup
 296  * @sb: super block instance to be unmounted or degraded to read-only
 297  *
 298  * This function restores state flags in the on-disk super block.
 299  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 300  * filesystem was not clean previously.
 301  */
 302 int nilfs_cleanup_super(struct super_block *sb)
 303 {
 304         struct the_nilfs *nilfs = sb->s_fs_info;
 305         struct nilfs_super_block **sbp;
 306         int flag = NILFS_SB_COMMIT;
 307         int ret = -EIO;
 308
 309         sbp = nilfs_prepare_super(sb, 0);
 310         if (sbp) {
 311                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 312                 nilfs_set_log_cursor(sbp[0], nilfs);
 313                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 314                         /*
 315                          * make the "clean" flag also to the opposite
 316                          * super block if both super blocks point to
 317                          * the same checkpoint.
 318                          */
 319                         sbp[1]->s_state = sbp[0]->s_state;
 320                         flag = NILFS_SB_COMMIT_ALL;
 321                 }
 322                 ret = nilfs_commit_super(sb, flag);
 323         }
 324         return ret;
 325 }
 326
 327 /**
 328  * nilfs_move_2nd_super - relocate secondary super block
 329  * @sb: super block instance
 330  * @sb2off: new offset of the secondary super block (in bytes)
 331  */
 332 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 333 {
 334         struct the_nilfs *nilfs = sb->s_fs_info;
 335         struct buffer_head *nsbh;
 336         struct nilfs_super_block *nsbp;
 337         sector_t blocknr, newblocknr;
 338         unsigned long offset;
 339         int sb2i;  /* array index of the secondary superblock */
 340         int ret = 0;
 341
 342         /* nilfs->ns_sem must be locked by the caller. */
 343         if (nilfs->ns_sbh[1] &&
 344             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 345                 sb2i = 1;
 346                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 347         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 348                 sb2i = 0;
 349                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 350         } else {
 351                 sb2i = -1;
 352                 blocknr = 0;
 353         }
 354         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 355                 goto out;  /* super block location is unchanged */
 356
 357         /* Get new super block buffer */
 358         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 359         offset = sb2off & (nilfs->ns_blocksize - 1);
 360         nsbh = sb_getblk(sb, newblocknr);
 361         if (!nsbh) {
 362                 nilfs_msg(sb, KERN_WARNING,
 363                           "unable to move secondary superblock to block %llu",
 364                           (unsigned long long)newblocknr);
 365                 ret = -EIO;
 366                 goto out;
 367         }
 368         nsbp = (void *)nsbh->b_data + offset;
 369         memset(nsbp, 0, nilfs->ns_blocksize);
 370
 371         if (sb2i >= 0) {
 372                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 373                 brelse(nilfs->ns_sbh[sb2i]);
 374                 nilfs->ns_sbh[sb2i] = nsbh;
 375                 nilfs->ns_sbp[sb2i] = nsbp;
 376         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 377                 /* secondary super block will be restored to index 1 */
 378                 nilfs->ns_sbh[1] = nsbh;
 379                 nilfs->ns_sbp[1] = nsbp;
 380         } else {
 381                 brelse(nsbh);
 382         }
 383 out:
 384         return ret;
 385 }
 386
 387 /**
 388  * nilfs_resize_fs - resize the filesystem
 389  * @sb: super block instance
 390  * @newsize: new size of the filesystem (in bytes)
 391  */
 392 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 393 {
 394         struct the_nilfs *nilfs = sb->s_fs_info;
 395         struct nilfs_super_block **sbp;
 396         __u64 devsize, newnsegs;
 397         loff_t sb2off;
 398         int ret;
 399
 400         ret = -ERANGE;
 401         devsize = i_size_read(sb->s_bdev->bd_inode);
 402         if (newsize > devsize)
 403                 goto out;
 404
 405         /*
 406          * Write lock is required to protect some functions depending
 407          * on the number of segments, the number of reserved segments,
 408          * and so forth.
 409          */
 410         down_write(&nilfs->ns_segctor_sem);
 411
 412         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 413         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 414         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 415
 416         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 417         up_write(&nilfs->ns_segctor_sem);
 418         if (ret < 0)
 419                 goto out;
 420
 421         ret = nilfs_construct_segment(sb);
 422         if (ret < 0)
 423                 goto out;
 424
 425         down_write(&nilfs->ns_sem);
 426         nilfs_move_2nd_super(sb, sb2off);
 427         ret = -EIO;
 428         sbp = nilfs_prepare_super(sb, 0);
 429         if (likely(sbp)) {
 430                 nilfs_set_log_cursor(sbp[0], nilfs);
 431                 /*
 432                  * Drop NILFS_RESIZE_FS flag for compatibility with
 433                  * mount-time resize which may be implemented in a
 434                  * future release.
 435                  */
 436                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 437                                               ~NILFS_RESIZE_FS);
 438                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 439                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 440                 if (sbp[1])
 441                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 442                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 443         }
 444         up_write(&nilfs->ns_sem);
 445
 446         /*
 447          * Reset the range of allocatable segments last.  This order
 448          * is important in the case of expansion because the secondary
 449          * superblock must be protected from log write until migration
 450          * completes.
 451          */
 452         if (!ret)
 453                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 454 out:
 455         return ret;
 456 }
 457
 458 static void nilfs_put_super(struct super_block *sb)
 459 {
 460         struct the_nilfs *nilfs = sb->s_fs_info;
 461
 462         nilfs_detach_log_writer(sb);
 463
 464         if (!sb_rdonly(sb)) {
 465                 down_write(&nilfs->ns_sem);
 466                 nilfs_cleanup_super(sb);
 467                 up_write(&nilfs->ns_sem);
 468         }
 469
 470         iput(nilfs->ns_sufile);
 471         iput(nilfs->ns_cpfile);
 472         iput(nilfs->ns_dat);
 473
 474         destroy_nilfs(nilfs);
 475         sb->s_fs_info = NULL;
 476 }
 477
 478 static int nilfs_sync_fs(struct super_block *sb, int wait)
 479 {
 480         struct the_nilfs *nilfs = sb->s_fs_info;
 481         struct nilfs_super_block **sbp;
 482         int err = 0;
 483
 484         /* This function is called when super block should be written back */
 485         if (wait)
 486                 err = nilfs_construct_segment(sb);
 487
 488         down_write(&nilfs->ns_sem);
 489         if (nilfs_sb_dirty(nilfs)) {
 490                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 491                 if (likely(sbp)) {
 492                         nilfs_set_log_cursor(sbp[0], nilfs);
 493                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 494                 }
 495         }
 496         up_write(&nilfs->ns_sem);
 497
 498         if (!err)
 499                 err = nilfs_flush_device(nilfs);
 500
 501         return err;
 502 }
 503
 504 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 505                             struct nilfs_root **rootp)
 506 {
 507         struct the_nilfs *nilfs = sb->s_fs_info;
 508         struct nilfs_root *root;
 509         struct nilfs_checkpoint *raw_cp;
 510         struct buffer_head *bh_cp;
 511         int err = -ENOMEM;
 512
 513         root = nilfs_find_or_create_root(
 514                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 515         if (!root)
 516                 return err;
 517
 518         if (root->ifile)
 519                 goto reuse; /* already attached checkpoint */
 520
 521         down_read(&nilfs->ns_segctor_sem);
 522         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 523                                           &bh_cp);
 524         up_read(&nilfs->ns_segctor_sem);
 525         if (unlikely(err)) {
 526                 if (err == -ENOENT || err == -EINVAL) {
 527                         nilfs_msg(sb, KERN_ERR,
 528                                   "Invalid checkpoint (checkpoint number=%llu)",
 529                                   (unsigned long long)cno);
 530                         err = -EINVAL;
 531                 }
 532                 goto failed;
 533         }
 534
 535         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 536                                &raw_cp->cp_ifile_inode, &root->ifile);
 537         if (err)
 538                 goto failed_bh;
 539
 540         atomic64_set(&root->inodes_count,
 541                         le64_to_cpu(raw_cp->cp_inodes_count));
 542         atomic64_set(&root->blocks_count,
 543                         le64_to_cpu(raw_cp->cp_blocks_count));
 544
 545         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 546
 547  reuse:
 548         *rootp = root;
 549         return 0;
 550
 551  failed_bh:
 552         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 553  failed:
 554         nilfs_put_root(root);
 555
 556         return err;
 557 }
 558
 559 static int nilfs_freeze(struct super_block *sb)
 560 {
 561         struct the_nilfs *nilfs = sb->s_fs_info;
 562         int err;
 563
 564         if (sb_rdonly(sb))
 565                 return 0;
 566
 567         /* Mark super block clean */
 568         down_write(&nilfs->ns_sem);
 569         err = nilfs_cleanup_super(sb);
 570         up_write(&nilfs->ns_sem);
 571         return err;
 572 }
 573
 574 static int nilfs_unfreeze(struct super_block *sb)
 575 {
 576         struct the_nilfs *nilfs = sb->s_fs_info;
 577
 578         if (sb_rdonly(sb))
 579                 return 0;
 580
 581         down_write(&nilfs->ns_sem);
 582         nilfs_setup_super(sb, false);
 583         up_write(&nilfs->ns_sem);
 584         return 0;
 585 }
 586
 587 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 588 {
 589         struct super_block *sb = dentry->d_sb;
 590         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 591         struct the_nilfs *nilfs = root->nilfs;
 592         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 593         unsigned long long blocks;
 594         unsigned long overhead;
 595         unsigned long nrsvblocks;
 596         sector_t nfreeblocks;
 597         u64 nmaxinodes, nfreeinodes;
 598         int err;
 599
 600         /*
 601          * Compute all of the segment blocks
 602          *
 603          * The blocks before first segment and after last segment
 604          * are excluded.
 605          */
 606         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 607                 - nilfs->ns_first_data_block;
 608         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 609
 610         /*
 611          * Compute the overhead
 612          *
 613          * When distributing meta data blocks outside segment structure,
 614          * We must count them as the overhead.
 615          */
 616         overhead = 0;
 617
 618         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 619         if (unlikely(err))
 620                 return err;
 621
 622         err = nilfs_ifile_count_free_inodes(root->ifile,
 623                                             &nmaxinodes, &nfreeinodes);
 624         if (unlikely(err)) {
 625                 nilfs_msg(sb, KERN_WARNING,
 626                           "failed to count free inodes: err=%d", err);
 627                 if (err == -ERANGE) {
 628                         /*
 629                          * If nilfs_palloc_count_max_entries() returns
 630                          * -ERANGE error code then we simply treat
 631                          * curent inodes count as maximum possible and
 632                          * zero as free inodes value.
 633                          */
 634                         nmaxinodes = atomic64_read(&root->inodes_count);
 635                         nfreeinodes = 0;
 636                         err = 0;
 637                 } else
 638                         return err;
 639         }
 640
 641         buf->f_type = NILFS_SUPER_MAGIC;
 642         buf->f_bsize = sb->s_blocksize;
 643         buf->f_blocks = blocks - overhead;
 644         buf->f_bfree = nfreeblocks;
 645         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 646                 (buf->f_bfree - nrsvblocks) : 0;
 647         buf->f_files = nmaxinodes;
 648         buf->f_ffree = nfreeinodes;
 649         buf->f_namelen = NILFS_NAME_LEN;
 650         buf->f_fsid.val[0] = (u32)id;
 651         buf->f_fsid.val[1] = (u32)(id >> 32);
 652
 653         return 0;
 654 }
 655
 656 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 657 {
 658         struct super_block *sb = dentry->d_sb;
 659         struct the_nilfs *nilfs = sb->s_fs_info;
 660         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 661
 662         if (!nilfs_test_opt(nilfs, BARRIER))
 663                 seq_puts(seq, ",nobarrier");
 664         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 665                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 666         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 667                 seq_puts(seq, ",errors=panic");
 668         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 669                 seq_puts(seq, ",errors=continue");
 670         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 671                 seq_puts(seq, ",order=strict");
 672         if (nilfs_test_opt(nilfs, NORECOVERY))
 673                 seq_puts(seq, ",norecovery");
 674         if (nilfs_test_opt(nilfs, DISCARD))
 675                 seq_puts(seq, ",discard");
 676
 677         return 0;
 678 }
 679
 680 static const struct super_operations nilfs_sops = {
 681         .alloc_inode    = nilfs_alloc_inode,
 682         .free_inode     = nilfs_free_inode,
 683         .dirty_inode    = nilfs_dirty_inode,
 684         .evict_inode    = nilfs_evict_inode,
 685         .put_super      = nilfs_put_super,
 686         .sync_fs        = nilfs_sync_fs,
 687         .freeze_fs      = nilfs_freeze,
 688         .unfreeze_fs    = nilfs_unfreeze,
 689         .statfs         = nilfs_statfs,
 690         .remount_fs     = nilfs_remount,
 691         .show_options = nilfs_show_options
 692 };
 693
 694 enum {
 695         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 696         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 697         Opt_discard, Opt_nodiscard, Opt_err,
 698 };
 699
 700 static match_table_t tokens = {
 701         {Opt_err_cont, "errors=continue"},
 702         {Opt_err_panic, "errors=panic"},
 703         {Opt_err_ro, "errors=remount-ro"},
 704         {Opt_barrier, "barrier"},
 705         {Opt_nobarrier, "nobarrier"},
 706         {Opt_snapshot, "cp=%u"},
 707         {Opt_order, "order=%s"},
 708         {Opt_norecovery, "norecovery"},
 709         {Opt_discard, "discard"},
 710         {Opt_nodiscard, "nodiscard"},
 711         {Opt_err, NULL}
 712 };
 713
 714 static int parse_options(char *options, struct super_block *sb, int is_remount)
 715 {
 716         struct the_nilfs *nilfs = sb->s_fs_info;
 717         char *p;
 718         substring_t args[MAX_OPT_ARGS];
 719
 720         if (!options)
 721                 return 1;
 722
 723         while ((p = strsep(&options, ",")) != NULL) {
 724                 int token;
 725
 726                 if (!*p)
 727                         continue;
 728
 729                 token = match_token(p, tokens, args);
 730                 switch (token) {
 731                 case Opt_barrier:
 732                         nilfs_set_opt(nilfs, BARRIER);
 733                         break;
 734                 case Opt_nobarrier:
 735                         nilfs_clear_opt(nilfs, BARRIER);
 736                         break;
 737                 case Opt_order:
 738                         if (strcmp(args[0].from, "relaxed") == 0)
 739                                 /* Ordered data semantics */
 740                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 741                         else if (strcmp(args[0].from, "strict") == 0)
 742                                 /* Strict in-order semantics */
 743                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 744                         else
 745                                 return 0;
 746                         break;
 747                 case Opt_err_panic:
 748                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 749                         break;
 750                 case Opt_err_ro:
 751                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 752                         break;
 753                 case Opt_err_cont:
 754                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 755                         break;
 756                 case Opt_snapshot:
 757                         if (is_remount) {
 758                                 nilfs_msg(sb, KERN_ERR,
 759                                           "\"%s\" option is invalid for remount",
 760                                           p);
 761                                 return 0;
 762                         }
 763                         break;
 764                 case Opt_norecovery:
 765                         nilfs_set_opt(nilfs, NORECOVERY);
 766                         break;
 767                 case Opt_discard:
 768                         nilfs_set_opt(nilfs, DISCARD);
 769                         break;
 770                 case Opt_nodiscard:
 771                         nilfs_clear_opt(nilfs, DISCARD);
 772                         break;
 773                 default:
 774                         nilfs_msg(sb, KERN_ERR,
 775                                   "unrecognized mount option \"%s\"", p);
 776                         return 0;
 777                 }
 778         }
 779         return 1;
 780 }
 781
 782 static inline void
 783 nilfs_set_default_options(struct super_block *sb,
 784                           struct nilfs_super_block *sbp)
 785 {
 786         struct the_nilfs *nilfs = sb->s_fs_info;
 787
 788         nilfs->ns_mount_opt =
 789                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 790 }
 791
 792 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 793 {
 794         struct the_nilfs *nilfs = sb->s_fs_info;
 795         struct nilfs_super_block **sbp;
 796         int max_mnt_count;
 797         int mnt_count;
 798
 799         /* nilfs->ns_sem must be locked by the caller. */
 800         sbp = nilfs_prepare_super(sb, 0);
 801         if (!sbp)
 802                 return -EIO;
 803
 804         if (!is_mount)
 805                 goto skip_mount_setup;
 806
 807         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 808         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 809
 810         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 811                 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
 812 #if 0
 813         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 814                 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
 815 #endif
 816         }
 817         if (!max_mnt_count)
 818                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 819
 820         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 821         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 822
 823 skip_mount_setup:
 824         sbp[0]->s_state =
 825                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 826         /* synchronize sbp[1] with sbp[0] */
 827         if (sbp[1])
 828                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 829         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 830 }
 831
 832 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 833                                                  u64 pos, int blocksize,
 834                                                  struct buffer_head **pbh)
 835 {
 836         unsigned long long sb_index = pos;
 837         unsigned long offset;
 838
 839         offset = do_div(sb_index, blocksize);
 840         *pbh = sb_bread(sb, sb_index);
 841         if (!*pbh)
 842                 return NULL;
 843         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 844 }
 845
 846 int nilfs_store_magic_and_option(struct super_block *sb,
 847                                  struct nilfs_super_block *sbp,
 848                                  char *data)
 849 {
 850         struct the_nilfs *nilfs = sb->s_fs_info;
 851
 852         sb->s_magic = le16_to_cpu(sbp->s_magic);
 853
 854         /* FS independent flags */
 855 #ifdef NILFS_ATIME_DISABLE
 856         sb->s_flags |= SB_NOATIME;
 857 #endif
 858
 859         nilfs_set_default_options(sb, sbp);
 860
 861         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 862         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 863         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 864         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 865
 866         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 867 }
 868
 869 int nilfs_check_feature_compatibility(struct super_block *sb,
 870                                       struct nilfs_super_block *sbp)
 871 {
 872         __u64 features;
 873
 874         features = le64_to_cpu(sbp->s_feature_incompat) &
 875                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 876         if (features) {
 877                 nilfs_msg(sb, KERN_ERR,
 878                           "couldn't mount because of unsupported optional features (%llx)",
 879                           (unsigned long long)features);
 880                 return -EINVAL;
 881         }
 882         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 883                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 884         if (!sb_rdonly(sb) && features) {
 885                 nilfs_msg(sb, KERN_ERR,
 886                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 887                           (unsigned long long)features);
 888                 return -EINVAL;
 889         }
 890         return 0;
 891 }
 892
 893 static int nilfs_get_root_dentry(struct super_block *sb,
 894                                  struct nilfs_root *root,
 895                                  struct dentry **root_dentry)
 896 {
 897         struct inode *inode;
 898         struct dentry *dentry;
 899         int ret = 0;
 900
 901         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 902         if (IS_ERR(inode)) {
 903                 ret = PTR_ERR(inode);
 904                 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
 905                 goto out;
 906         }
 907         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 908                 iput(inode);
 909                 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
 910                 ret = -EINVAL;
 911                 goto out;
 912         }
 913
 914         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 915                 dentry = d_find_alias(inode);
 916                 if (!dentry) {
 917                         dentry = d_make_root(inode);
 918                         if (!dentry) {
 919                                 ret = -ENOMEM;
 920                                 goto failed_dentry;
 921                         }
 922                 } else {
 923                         iput(inode);
 924                 }
 925         } else {
 926                 dentry = d_obtain_root(inode);
 927                 if (IS_ERR(dentry)) {
 928                         ret = PTR_ERR(dentry);
 929                         goto failed_dentry;
 930                 }
 931         }
 932         *root_dentry = dentry;
 933  out:
 934         return ret;
 935
 936  failed_dentry:
 937         nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
 938         goto out;
 939 }
 940
 941 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 942                                  struct dentry **root_dentry)
 943 {
 944         struct the_nilfs *nilfs = s->s_fs_info;
 945         struct nilfs_root *root;
 946         int ret;
 947
 948         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 949
 950         down_read(&nilfs->ns_segctor_sem);
 951         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 952         up_read(&nilfs->ns_segctor_sem);
 953         if (ret < 0) {
 954                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 955                 goto out;
 956         } else if (!ret) {
 957                 nilfs_msg(s, KERN_ERR,
 958                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 959                           (unsigned long long)cno);
 960                 ret = -EINVAL;
 961                 goto out;
 962         }
 963
 964         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 965         if (ret) {
 966                 nilfs_msg(s, KERN_ERR,
 967                           "error %d while loading snapshot (checkpoint number=%llu)",
 968                           ret, (unsigned long long)cno);
 969                 goto out;
 970         }
 971         ret = nilfs_get_root_dentry(s, root, root_dentry);
 972         nilfs_put_root(root);
 973  out:
 974         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 975         return ret;
 976 }
 977
 978 /**
 979  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 980  * @root_dentry: root dentry of the tree to be shrunk
 981  *
 982  * This function returns true if the tree was in-use.
 983  */
 984 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
 985 {
 986         shrink_dcache_parent(root_dentry);
 987         return d_count(root_dentry) > 1;
 988 }
 989
 990 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
 991 {
 992         struct the_nilfs *nilfs = sb->s_fs_info;
 993         struct nilfs_root *root;
 994         struct inode *inode;
 995         struct dentry *dentry;
 996         int ret;
 997
 998         if (cno > nilfs->ns_cno)
 999                 return false;
1000
1001         if (cno >= nilfs_last_cno(nilfs))
1002                 return true;    /* protect recent checkpoints */
1003
1004         ret = false;
1005         root = nilfs_lookup_root(nilfs, cno);
1006         if (root) {
1007                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1008                 if (inode) {
1009                         dentry = d_find_alias(inode);
1010                         if (dentry) {
1011                                 ret = nilfs_tree_is_busy(dentry);
1012                                 dput(dentry);
1013                         }
1014                         iput(inode);
1015                 }
1016                 nilfs_put_root(root);
1017         }
1018         return ret;
1019 }
1020
1021 /**
1022  * nilfs_fill_super() - initialize a super block instance
1023  * @sb: super_block
1024  * @data: mount options
1025  * @silent: silent mode flag
1026  *
1027  * This function is called exclusively by nilfs->ns_mount_mutex.
1028  * So, the recovery process is protected from other simultaneous mounts.
1029  */
1030 static int
1031 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1032 {
1033         struct the_nilfs *nilfs;
1034         struct nilfs_root *fsroot;
1035         __u64 cno;
1036         int err;
1037
1038         nilfs = alloc_nilfs(sb);
1039         if (!nilfs)
1040                 return -ENOMEM;
1041
1042         sb->s_fs_info = nilfs;
1043
1044         err = init_nilfs(nilfs, sb, (char *)data);
1045         if (err)
1046                 goto failed_nilfs;
1047
1048         sb->s_op = &nilfs_sops;
1049         sb->s_export_op = &nilfs_export_ops;
1050         sb->s_root = NULL;
1051         sb->s_time_gran = 1;
1052         sb->s_max_links = NILFS_LINK_MAX;
1053
1054         sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1055
1056         err = load_nilfs(nilfs, sb);
1057         if (err)
1058                 goto failed_nilfs;
1059
1060         cno = nilfs_last_cno(nilfs);
1061         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1062         if (err) {
1063                 nilfs_msg(sb, KERN_ERR,
1064                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1065                           err, (unsigned long long)cno);
1066                 goto failed_unload;
1067         }
1068
1069         if (!sb_rdonly(sb)) {
1070                 err = nilfs_attach_log_writer(sb, fsroot);
1071                 if (err)
1072                         goto failed_checkpoint;
1073         }
1074
1075         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1076         if (err)
1077                 goto failed_segctor;
1078
1079         nilfs_put_root(fsroot);
1080
1081         if (!sb_rdonly(sb)) {
1082                 down_write(&nilfs->ns_sem);
1083                 nilfs_setup_super(sb, true);
1084                 up_write(&nilfs->ns_sem);
1085         }
1086
1087         return 0;
1088
1089  failed_segctor:
1090         nilfs_detach_log_writer(sb);
1091
1092  failed_checkpoint:
1093         nilfs_put_root(fsroot);
1094
1095  failed_unload:
1096         iput(nilfs->ns_sufile);
1097         iput(nilfs->ns_cpfile);
1098         iput(nilfs->ns_dat);
1099
1100  failed_nilfs:
1101         destroy_nilfs(nilfs);
1102         return err;
1103 }
1104
1105 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1106 {
1107         struct the_nilfs *nilfs = sb->s_fs_info;
1108         unsigned long old_sb_flags;
1109         unsigned long old_mount_opt;
1110         int err;
1111
1112         sync_filesystem(sb);
1113         old_sb_flags = sb->s_flags;
1114         old_mount_opt = nilfs->ns_mount_opt;
1115
1116         if (!parse_options(data, sb, 1)) {
1117                 err = -EINVAL;
1118                 goto restore_opts;
1119         }
1120         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1121
1122         err = -EINVAL;
1123
1124         if (!nilfs_valid_fs(nilfs)) {
1125                 nilfs_msg(sb, KERN_WARNING,
1126                           "couldn't remount because the filesystem is in an incomplete recovery state");
1127                 goto restore_opts;
1128         }
1129
1130         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1131                 goto out;
1132         if (*flags & SB_RDONLY) {
1133                 /* Shutting down log writer */
1134                 nilfs_detach_log_writer(sb);
1135                 sb->s_flags |= SB_RDONLY;
1136
1137                 /*
1138                  * Remounting a valid RW partition RDONLY, so set
1139                  * the RDONLY flag and then mark the partition as valid again.
1140                  */
1141                 down_write(&nilfs->ns_sem);
1142                 nilfs_cleanup_super(sb);
1143                 up_write(&nilfs->ns_sem);
1144         } else {
1145                 __u64 features;
1146                 struct nilfs_root *root;
1147
1148                 /*
1149                  * Mounting a RDONLY partition read-write, so reread and
1150                  * store the current valid flag.  (It may have been changed
1151                  * by fsck since we originally mounted the partition.)
1152                  */
1153                 down_read(&nilfs->ns_sem);
1154                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1155                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1156                 up_read(&nilfs->ns_sem);
1157                 if (features) {
1158                         nilfs_msg(sb, KERN_WARNING,
1159                                   "couldn't remount RDWR because of unsupported optional features (%llx)",
1160                                   (unsigned long long)features);
1161                         err = -EROFS;
1162                         goto restore_opts;
1163                 }
1164
1165                 sb->s_flags &= ~SB_RDONLY;
1166
1167                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1168                 err = nilfs_attach_log_writer(sb, root);
1169                 if (err)
1170                         goto restore_opts;
1171
1172                 down_write(&nilfs->ns_sem);
1173                 nilfs_setup_super(sb, true);
1174                 up_write(&nilfs->ns_sem);
1175         }
1176  out:
1177         return 0;
1178
1179  restore_opts:
1180         sb->s_flags = old_sb_flags;
1181         nilfs->ns_mount_opt = old_mount_opt;
1182         return err;
1183 }
1184
1185 struct nilfs_super_data {
1186         struct block_device *bdev;
1187         __u64 cno;
1188         int flags;
1189 };
1190
1191 static int nilfs_parse_snapshot_option(const char *option,
1192                                        const substring_t *arg,
1193                                        struct nilfs_super_data *sd)
1194 {
1195         unsigned long long val;
1196         const char *msg = NULL;
1197         int err;
1198
1199         if (!(sd->flags & SB_RDONLY)) {
1200                 msg = "read-only option is not specified";
1201                 goto parse_error;
1202         }
1203
1204         err = kstrtoull(arg->from, 0, &val);
1205         if (err) {
1206                 if (err == -ERANGE)
1207                         msg = "too large checkpoint number";
1208                 else
1209                         msg = "malformed argument";
1210                 goto parse_error;
1211         } else if (val == 0) {
1212                 msg = "invalid checkpoint number 0";
1213                 goto parse_error;
1214         }
1215         sd->cno = val;
1216         return 0;
1217
1218 parse_error:
1219         nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1220         return 1;
1221 }
1222
1223 /**
1224  * nilfs_identify - pre-read mount options needed to identify mount instance
1225  * @data: mount options
1226  * @sd: nilfs_super_data
1227  */
1228 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1229 {
1230         char *p, *options = data;
1231         substring_t args[MAX_OPT_ARGS];
1232         int token;
1233         int ret = 0;
1234
1235         do {
1236                 p = strsep(&options, ",");
1237                 if (p != NULL && *p) {
1238                         token = match_token(p, tokens, args);
1239                         if (token == Opt_snapshot)
1240                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1241                                                                   sd);
1242                 }
1243                 if (!options)
1244                         break;
1245                 BUG_ON(options == data);
1246                 *(options - 1) = ',';
1247         } while (!ret);
1248         return ret;
1249 }
1250
1251 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1252 {
1253         s->s_bdev = data;
1254         s->s_dev = s->s_bdev->bd_dev;
1255         return 0;
1256 }
1257
1258 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1259 {
1260         return (void *)s->s_bdev == data;
1261 }
1262
1263 static struct dentry *
1264 nilfs_mount(struct file_system_type *fs_type, int flags,
1265              const char *dev_name, void *data)
1266 {
1267         struct nilfs_super_data sd;
1268         struct super_block *s;
1269         fmode_t mode = FMODE_READ | FMODE_EXCL;
1270         struct dentry *root_dentry;
1271         int err, s_new = false;
1272
1273         if (!(flags & SB_RDONLY))
1274                 mode |= FMODE_WRITE;
1275
1276         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1277         if (IS_ERR(sd.bdev))
1278                 return ERR_CAST(sd.bdev);
1279
1280         sd.cno = 0;
1281         sd.flags = flags;
1282         if (nilfs_identify((char *)data, &sd)) {
1283                 err = -EINVAL;
1284                 goto failed;
1285         }
1286
1287         /*
1288          * once the super is inserted into the list by sget, s_umount
1289          * will protect the lockfs code from trying to start a snapshot
1290          * while we are mounting
1291          */
1292         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1293         if (sd.bdev->bd_fsfreeze_count > 0) {
1294                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1295                 err = -EBUSY;
1296                 goto failed;
1297         }
1298         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1299                  sd.bdev);
1300         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1301         if (IS_ERR(s)) {
1302                 err = PTR_ERR(s);
1303                 goto failed;
1304         }
1305
1306         if (!s->s_root) {
1307                 s_new = true;
1308
1309                 /* New superblock instance created */
1310                 s->s_mode = mode;
1311                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1312                 sb_set_blocksize(s, block_size(sd.bdev));
1313
1314                 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1315                 if (err)
1316                         goto failed_super;
1317
1318                 s->s_flags |= SB_ACTIVE;
1319         } else if (!sd.cno) {
1320                 if (nilfs_tree_is_busy(s->s_root)) {
1321                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1322                                 nilfs_msg(s, KERN_ERR,
1323                                           "the device already has a %s mount.",
1324                                           sb_rdonly(s) ? "read-only" : "read/write");
1325                                 err = -EBUSY;
1326                                 goto failed_super;
1327                         }
1328                 } else {
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 MODULE_ALIAS_FS("nilfs2");
1369
1370 static void nilfs_inode_init_once(void *obj)
1371 {
1372         struct nilfs_inode_info *ii = obj;
1373
1374         INIT_LIST_HEAD(&ii->i_dirty);
1375 #ifdef CONFIG_NILFS_XATTR
1376         init_rwsem(&ii->xattr_sem);
1377 #endif
1378         address_space_init_once(&ii->i_btnode_cache);
1379         ii->i_bmap = &ii->i_bmap_data;
1380         inode_init_once(&ii->vfs_inode);
1381 }
1382
1383 static void nilfs_segbuf_init_once(void *obj)
1384 {
1385         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1386 }
1387
1388 static void nilfs_destroy_cachep(void)
1389 {
1390         /*
1391          * Make sure all delayed rcu free inodes are flushed before we
1392          * destroy cache.
1393          */
1394         rcu_barrier();
1395
1396         kmem_cache_destroy(nilfs_inode_cachep);
1397         kmem_cache_destroy(nilfs_transaction_cachep);
1398         kmem_cache_destroy(nilfs_segbuf_cachep);
1399         kmem_cache_destroy(nilfs_btree_path_cache);
1400 }
1401
1402 static int __init nilfs_init_cachep(void)
1403 {
1404         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1405                         sizeof(struct nilfs_inode_info), 0,
1406                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1407                         nilfs_inode_init_once);
1408         if (!nilfs_inode_cachep)
1409                 goto fail;
1410
1411         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1412                         sizeof(struct nilfs_transaction_info), 0,
1413                         SLAB_RECLAIM_ACCOUNT, NULL);
1414         if (!nilfs_transaction_cachep)
1415                 goto fail;
1416
1417         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1418                         sizeof(struct nilfs_segment_buffer), 0,
1419                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1420         if (!nilfs_segbuf_cachep)
1421                 goto fail;
1422
1423         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1424                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1425                         0, 0, NULL);
1426         if (!nilfs_btree_path_cache)
1427                 goto fail;
1428
1429         return 0;
1430
1431 fail:
1432         nilfs_destroy_cachep();
1433         return -ENOMEM;
1434 }
1435
1436 static int __init init_nilfs_fs(void)
1437 {
1438         int err;
1439
1440         err = nilfs_init_cachep();
1441         if (err)
1442                 goto fail;
1443
1444         err = nilfs_sysfs_init();
1445         if (err)
1446                 goto free_cachep;
1447
1448         err = register_filesystem(&nilfs_fs_type);
1449         if (err)
1450                 goto deinit_sysfs_entry;
1451
1452         printk(KERN_INFO "NILFS version 2 loaded\n");
1453         return 0;
1454
1455 deinit_sysfs_entry:
1456         nilfs_sysfs_exit();
1457 free_cachep:
1458         nilfs_destroy_cachep();
1459 fail:
1460         return err;
1461 }
1462
1463 static void __exit exit_nilfs_fs(void)
1464 {
1465         nilfs_destroy_cachep();
1466         nilfs_sysfs_exit();
1467         unregister_filesystem(&nilfs_fs_type);
1468 }
1469
1470 module_init(init_nilfs_fs)
1471 module_exit(exit_nilfs_fs)