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/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 *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         memset(nsbp, 0, nilfs->ns_blocksize);
 376
 377         if (sb2i >= 0) {
 378                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 379                 brelse(nilfs->ns_sbh[sb2i]);
 380                 nilfs->ns_sbh[sb2i] = nsbh;
 381                 nilfs->ns_sbp[sb2i] = nsbp;
 382         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 383                 /* secondary super block will be restored to index 1 */
 384                 nilfs->ns_sbh[1] = nsbh;
 385                 nilfs->ns_sbp[1] = nsbp;
 386         } else {
 387                 brelse(nsbh);
 388         }
 389 out:
 390         return ret;
 391 }
 392
 393 /**
 394  * nilfs_resize_fs - resize the filesystem
 395  * @sb: super block instance
 396  * @newsize: new size of the filesystem (in bytes)
 397  */
 398 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 399 {
 400         struct the_nilfs *nilfs = sb->s_fs_info;
 401         struct nilfs_super_block **sbp;
 402         __u64 devsize, newnsegs;
 403         loff_t sb2off;
 404         int ret;
 405
 406         ret = -ERANGE;
 407         devsize = bdev_nr_bytes(sb->s_bdev);
 408         if (newsize > devsize)
 409                 goto out;
 410
 411         /*
 412          * Prevent underflow in second superblock position calculation.
 413          * The exact minimum size check is done in nilfs_sufile_resize().
 414          */
 415         if (newsize < 4096) {
 416                 ret = -ENOSPC;
 417                 goto out;
 418         }
 419
 420         /*
 421          * Write lock is required to protect some functions depending
 422          * on the number of segments, the number of reserved segments,
 423          * and so forth.
 424          */
 425         down_write(&nilfs->ns_segctor_sem);
 426
 427         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 428         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 429         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 430
 431         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 432         up_write(&nilfs->ns_segctor_sem);
 433         if (ret < 0)
 434                 goto out;
 435
 436         ret = nilfs_construct_segment(sb);
 437         if (ret < 0)
 438                 goto out;
 439
 440         down_write(&nilfs->ns_sem);
 441         nilfs_move_2nd_super(sb, sb2off);
 442         ret = -EIO;
 443         sbp = nilfs_prepare_super(sb, 0);
 444         if (likely(sbp)) {
 445                 nilfs_set_log_cursor(sbp[0], nilfs);
 446                 /*
 447                  * Drop NILFS_RESIZE_FS flag for compatibility with
 448                  * mount-time resize which may be implemented in a
 449                  * future release.
 450                  */
 451                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 452                                               ~NILFS_RESIZE_FS);
 453                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 454                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 455                 if (sbp[1])
 456                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 457                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 458         }
 459         up_write(&nilfs->ns_sem);
 460
 461         /*
 462          * Reset the range of allocatable segments last.  This order
 463          * is important in the case of expansion because the secondary
 464          * superblock must be protected from log write until migration
 465          * completes.
 466          */
 467         if (!ret)
 468                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 469 out:
 470         return ret;
 471 }
 472
 473 static void nilfs_put_super(struct super_block *sb)
 474 {
 475         struct the_nilfs *nilfs = sb->s_fs_info;
 476
 477         nilfs_detach_log_writer(sb);
 478
 479         if (!sb_rdonly(sb)) {
 480                 down_write(&nilfs->ns_sem);
 481                 nilfs_cleanup_super(sb);
 482                 up_write(&nilfs->ns_sem);
 483         }
 484
 485         iput(nilfs->ns_sufile);
 486         iput(nilfs->ns_cpfile);
 487         iput(nilfs->ns_dat);
 488
 489         destroy_nilfs(nilfs);
 490         sb->s_fs_info = NULL;
 491 }
 492
 493 static int nilfs_sync_fs(struct super_block *sb, int wait)
 494 {
 495         struct the_nilfs *nilfs = sb->s_fs_info;
 496         struct nilfs_super_block **sbp;
 497         int err = 0;
 498
 499         /* This function is called when super block should be written back */
 500         if (wait)
 501                 err = nilfs_construct_segment(sb);
 502
 503         down_write(&nilfs->ns_sem);
 504         if (nilfs_sb_dirty(nilfs)) {
 505                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 506                 if (likely(sbp)) {
 507                         nilfs_set_log_cursor(sbp[0], nilfs);
 508                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 509                 }
 510         }
 511         up_write(&nilfs->ns_sem);
 512
 513         if (!err)
 514                 err = nilfs_flush_device(nilfs);
 515
 516         return err;
 517 }
 518
 519 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 520                             struct nilfs_root **rootp)
 521 {
 522         struct the_nilfs *nilfs = sb->s_fs_info;
 523         struct nilfs_root *root;
 524         struct nilfs_checkpoint *raw_cp;
 525         struct buffer_head *bh_cp;
 526         int err = -ENOMEM;
 527
 528         root = nilfs_find_or_create_root(
 529                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 530         if (!root)
 531                 return err;
 532
 533         if (root->ifile)
 534                 goto reuse; /* already attached checkpoint */
 535
 536         down_read(&nilfs->ns_segctor_sem);
 537         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 538                                           &bh_cp);
 539         up_read(&nilfs->ns_segctor_sem);
 540         if (unlikely(err)) {
 541                 if (err == -ENOENT || err == -EINVAL) {
 542                         nilfs_err(sb,
 543                                   "Invalid checkpoint (checkpoint number=%llu)",
 544                                   (unsigned long long)cno);
 545                         err = -EINVAL;
 546                 }
 547                 goto failed;
 548         }
 549
 550         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 551                                &raw_cp->cp_ifile_inode, &root->ifile);
 552         if (err)
 553                 goto failed_bh;
 554
 555         atomic64_set(&root->inodes_count,
 556                         le64_to_cpu(raw_cp->cp_inodes_count));
 557         atomic64_set(&root->blocks_count,
 558                         le64_to_cpu(raw_cp->cp_blocks_count));
 559
 560         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 561
 562  reuse:
 563         *rootp = root;
 564         return 0;
 565
 566  failed_bh:
 567         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 568  failed:
 569         nilfs_put_root(root);
 570
 571         return err;
 572 }
 573
 574 static int nilfs_freeze(struct super_block *sb)
 575 {
 576         struct the_nilfs *nilfs = sb->s_fs_info;
 577         int err;
 578
 579         if (sb_rdonly(sb))
 580                 return 0;
 581
 582         /* Mark super block clean */
 583         down_write(&nilfs->ns_sem);
 584         err = nilfs_cleanup_super(sb);
 585         up_write(&nilfs->ns_sem);
 586         return err;
 587 }
 588
 589 static int nilfs_unfreeze(struct super_block *sb)
 590 {
 591         struct the_nilfs *nilfs = sb->s_fs_info;
 592
 593         if (sb_rdonly(sb))
 594                 return 0;
 595
 596         down_write(&nilfs->ns_sem);
 597         nilfs_setup_super(sb, false);
 598         up_write(&nilfs->ns_sem);
 599         return 0;
 600 }
 601
 602 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 603 {
 604         struct super_block *sb = dentry->d_sb;
 605         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 606         struct the_nilfs *nilfs = root->nilfs;
 607         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 608         unsigned long long blocks;
 609         unsigned long overhead;
 610         unsigned long nrsvblocks;
 611         sector_t nfreeblocks;
 612         u64 nmaxinodes, nfreeinodes;
 613         int err;
 614
 615         /*
 616          * Compute all of the segment blocks
 617          *
 618          * The blocks before first segment and after last segment
 619          * are excluded.
 620          */
 621         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 622                 - nilfs->ns_first_data_block;
 623         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 624
 625         /*
 626          * Compute the overhead
 627          *
 628          * When distributing meta data blocks outside segment structure,
 629          * We must count them as the overhead.
 630          */
 631         overhead = 0;
 632
 633         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 634         if (unlikely(err))
 635                 return err;
 636
 637         err = nilfs_ifile_count_free_inodes(root->ifile,
 638                                             &nmaxinodes, &nfreeinodes);
 639         if (unlikely(err)) {
 640                 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
 641                 if (err == -ERANGE) {
 642                         /*
 643                          * If nilfs_palloc_count_max_entries() returns
 644                          * -ERANGE error code then we simply treat
 645                          * curent inodes count as maximum possible and
 646                          * zero as free inodes value.
 647                          */
 648                         nmaxinodes = atomic64_read(&root->inodes_count);
 649                         nfreeinodes = 0;
 650                         err = 0;
 651                 } else
 652                         return err;
 653         }
 654
 655         buf->f_type = NILFS_SUPER_MAGIC;
 656         buf->f_bsize = sb->s_blocksize;
 657         buf->f_blocks = blocks - overhead;
 658         buf->f_bfree = nfreeblocks;
 659         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 660                 (buf->f_bfree - nrsvblocks) : 0;
 661         buf->f_files = nmaxinodes;
 662         buf->f_ffree = nfreeinodes;
 663         buf->f_namelen = NILFS_NAME_LEN;
 664         buf->f_fsid = u64_to_fsid(id);
 665
 666         return 0;
 667 }
 668
 669 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 670 {
 671         struct super_block *sb = dentry->d_sb;
 672         struct the_nilfs *nilfs = sb->s_fs_info;
 673         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 674
 675         if (!nilfs_test_opt(nilfs, BARRIER))
 676                 seq_puts(seq, ",nobarrier");
 677         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 678                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 679         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 680                 seq_puts(seq, ",errors=panic");
 681         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 682                 seq_puts(seq, ",errors=continue");
 683         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 684                 seq_puts(seq, ",order=strict");
 685         if (nilfs_test_opt(nilfs, NORECOVERY))
 686                 seq_puts(seq, ",norecovery");
 687         if (nilfs_test_opt(nilfs, DISCARD))
 688                 seq_puts(seq, ",discard");
 689
 690         return 0;
 691 }
 692
 693 static const struct super_operations nilfs_sops = {
 694         .alloc_inode    = nilfs_alloc_inode,
 695         .free_inode     = nilfs_free_inode,
 696         .dirty_inode    = nilfs_dirty_inode,
 697         .evict_inode    = nilfs_evict_inode,
 698         .put_super      = nilfs_put_super,
 699         .sync_fs        = nilfs_sync_fs,
 700         .freeze_fs      = nilfs_freeze,
 701         .unfreeze_fs    = nilfs_unfreeze,
 702         .statfs         = nilfs_statfs,
 703         .remount_fs     = nilfs_remount,
 704         .show_options = nilfs_show_options
 705 };
 706
 707 enum {
 708         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 709         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 710         Opt_discard, Opt_nodiscard, Opt_err,
 711 };
 712
 713 static match_table_t tokens = {
 714         {Opt_err_cont, "errors=continue"},
 715         {Opt_err_panic, "errors=panic"},
 716         {Opt_err_ro, "errors=remount-ro"},
 717         {Opt_barrier, "barrier"},
 718         {Opt_nobarrier, "nobarrier"},
 719         {Opt_snapshot, "cp=%u"},
 720         {Opt_order, "order=%s"},
 721         {Opt_norecovery, "norecovery"},
 722         {Opt_discard, "discard"},
 723         {Opt_nodiscard, "nodiscard"},
 724         {Opt_err, NULL}
 725 };
 726
 727 static int parse_options(char *options, struct super_block *sb, int is_remount)
 728 {
 729         struct the_nilfs *nilfs = sb->s_fs_info;
 730         char *p;
 731         substring_t args[MAX_OPT_ARGS];
 732
 733         if (!options)
 734                 return 1;
 735
 736         while ((p = strsep(&options, ",")) != NULL) {
 737                 int token;
 738
 739                 if (!*p)
 740                         continue;
 741
 742                 token = match_token(p, tokens, args);
 743                 switch (token) {
 744                 case Opt_barrier:
 745                         nilfs_set_opt(nilfs, BARRIER);
 746                         break;
 747                 case Opt_nobarrier:
 748                         nilfs_clear_opt(nilfs, BARRIER);
 749                         break;
 750                 case Opt_order:
 751                         if (strcmp(args[0].from, "relaxed") == 0)
 752                                 /* Ordered data semantics */
 753                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 754                         else if (strcmp(args[0].from, "strict") == 0)
 755                                 /* Strict in-order semantics */
 756                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 757                         else
 758                                 return 0;
 759                         break;
 760                 case Opt_err_panic:
 761                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 762                         break;
 763                 case Opt_err_ro:
 764                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 765                         break;
 766                 case Opt_err_cont:
 767                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 768                         break;
 769                 case Opt_snapshot:
 770                         if (is_remount) {
 771                                 nilfs_err(sb,
 772                                           "\"%s\" option is invalid for remount",
 773                                           p);
 774                                 return 0;
 775                         }
 776                         break;
 777                 case Opt_norecovery:
 778                         nilfs_set_opt(nilfs, NORECOVERY);
 779                         break;
 780                 case Opt_discard:
 781                         nilfs_set_opt(nilfs, DISCARD);
 782                         break;
 783                 case Opt_nodiscard:
 784                         nilfs_clear_opt(nilfs, DISCARD);
 785                         break;
 786                 default:
 787                         nilfs_err(sb, "unrecognized mount option \"%s\"", p);
 788                         return 0;
 789                 }
 790         }
 791         return 1;
 792 }
 793
 794 static inline void
 795 nilfs_set_default_options(struct super_block *sb,
 796                           struct nilfs_super_block *sbp)
 797 {
 798         struct the_nilfs *nilfs = sb->s_fs_info;
 799
 800         nilfs->ns_mount_opt =
 801                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 802 }
 803
 804 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 805 {
 806         struct the_nilfs *nilfs = sb->s_fs_info;
 807         struct nilfs_super_block **sbp;
 808         int max_mnt_count;
 809         int mnt_count;
 810
 811         /* nilfs->ns_sem must be locked by the caller. */
 812         sbp = nilfs_prepare_super(sb, 0);
 813         if (!sbp)
 814                 return -EIO;
 815
 816         if (!is_mount)
 817                 goto skip_mount_setup;
 818
 819         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 820         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 821
 822         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 823                 nilfs_warn(sb, "mounting fs with errors");
 824 #if 0
 825         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 826                 nilfs_warn(sb, "maximal mount count reached");
 827 #endif
 828         }
 829         if (!max_mnt_count)
 830                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 831
 832         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 833         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 834
 835 skip_mount_setup:
 836         sbp[0]->s_state =
 837                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 838         /* synchronize sbp[1] with sbp[0] */
 839         if (sbp[1])
 840                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 841         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 842 }
 843
 844 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 845                                                  u64 pos, int blocksize,
 846                                                  struct buffer_head **pbh)
 847 {
 848         unsigned long long sb_index = pos;
 849         unsigned long offset;
 850
 851         offset = do_div(sb_index, blocksize);
 852         *pbh = sb_bread(sb, sb_index);
 853         if (!*pbh)
 854                 return NULL;
 855         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 856 }
 857
 858 int nilfs_store_magic_and_option(struct super_block *sb,
 859                                  struct nilfs_super_block *sbp,
 860                                  char *data)
 861 {
 862         struct the_nilfs *nilfs = sb->s_fs_info;
 863
 864         sb->s_magic = le16_to_cpu(sbp->s_magic);
 865
 866         /* FS independent flags */
 867 #ifdef NILFS_ATIME_DISABLE
 868         sb->s_flags |= SB_NOATIME;
 869 #endif
 870
 871         nilfs_set_default_options(sb, sbp);
 872
 873         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 874         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 875         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 876         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 877
 878         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 879 }
 880
 881 int nilfs_check_feature_compatibility(struct super_block *sb,
 882                                       struct nilfs_super_block *sbp)
 883 {
 884         __u64 features;
 885
 886         features = le64_to_cpu(sbp->s_feature_incompat) &
 887                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 888         if (features) {
 889                 nilfs_err(sb,
 890                           "couldn't mount because of unsupported optional features (%llx)",
 891                           (unsigned long long)features);
 892                 return -EINVAL;
 893         }
 894         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 895                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 896         if (!sb_rdonly(sb) && features) {
 897                 nilfs_err(sb,
 898                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 899                           (unsigned long long)features);
 900                 return -EINVAL;
 901         }
 902         return 0;
 903 }
 904
 905 static int nilfs_get_root_dentry(struct super_block *sb,
 906                                  struct nilfs_root *root,
 907                                  struct dentry **root_dentry)
 908 {
 909         struct inode *inode;
 910         struct dentry *dentry;
 911         int ret = 0;
 912
 913         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 914         if (IS_ERR(inode)) {
 915                 ret = PTR_ERR(inode);
 916                 nilfs_err(sb, "error %d getting root inode", ret);
 917                 goto out;
 918         }
 919         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 920                 iput(inode);
 921                 nilfs_err(sb, "corrupt root inode");
 922                 ret = -EINVAL;
 923                 goto out;
 924         }
 925
 926         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 927                 dentry = d_find_alias(inode);
 928                 if (!dentry) {
 929                         dentry = d_make_root(inode);
 930                         if (!dentry) {
 931                                 ret = -ENOMEM;
 932                                 goto failed_dentry;
 933                         }
 934                 } else {
 935                         iput(inode);
 936                 }
 937         } else {
 938                 dentry = d_obtain_root(inode);
 939                 if (IS_ERR(dentry)) {
 940                         ret = PTR_ERR(dentry);
 941                         goto failed_dentry;
 942                 }
 943         }
 944         *root_dentry = dentry;
 945  out:
 946         return ret;
 947
 948  failed_dentry:
 949         nilfs_err(sb, "error %d getting root dentry", ret);
 950         goto out;
 951 }
 952
 953 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 954                                  struct dentry **root_dentry)
 955 {
 956         struct the_nilfs *nilfs = s->s_fs_info;
 957         struct nilfs_root *root;
 958         int ret;
 959
 960         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 961
 962         down_read(&nilfs->ns_segctor_sem);
 963         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 964         up_read(&nilfs->ns_segctor_sem);
 965         if (ret < 0) {
 966                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 967                 goto out;
 968         } else if (!ret) {
 969                 nilfs_err(s,
 970                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 971                           (unsigned long long)cno);
 972                 ret = -EINVAL;
 973                 goto out;
 974         }
 975
 976         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 977         if (ret) {
 978                 nilfs_err(s,
 979                           "error %d while loading snapshot (checkpoint number=%llu)",
 980                           ret, (unsigned long long)cno);
 981                 goto out;
 982         }
 983         ret = nilfs_get_root_dentry(s, root, root_dentry);
 984         nilfs_put_root(root);
 985  out:
 986         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 987         return ret;
 988 }
 989
 990 /**
 991  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 992  * @root_dentry: root dentry of the tree to be shrunk
 993  *
 994  * This function returns true if the tree was in-use.
 995  */
 996 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
 997 {
 998         shrink_dcache_parent(root_dentry);
 999         return d_count(root_dentry) > 1;
1000 }
1001
1002 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1003 {
1004         struct the_nilfs *nilfs = sb->s_fs_info;
1005         struct nilfs_root *root;
1006         struct inode *inode;
1007         struct dentry *dentry;
1008         int ret;
1009
1010         if (cno > nilfs->ns_cno)
1011                 return false;
1012
1013         if (cno >= nilfs_last_cno(nilfs))
1014                 return true;    /* protect recent checkpoints */
1015
1016         ret = false;
1017         root = nilfs_lookup_root(nilfs, cno);
1018         if (root) {
1019                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1020                 if (inode) {
1021                         dentry = d_find_alias(inode);
1022                         if (dentry) {
1023                                 ret = nilfs_tree_is_busy(dentry);
1024                                 dput(dentry);
1025                         }
1026                         iput(inode);
1027                 }
1028                 nilfs_put_root(root);
1029         }
1030         return ret;
1031 }
1032
1033 /**
1034  * nilfs_fill_super() - initialize a super block instance
1035  * @sb: super_block
1036  * @data: mount options
1037  * @silent: silent mode flag
1038  *
1039  * This function is called exclusively by nilfs->ns_mount_mutex.
1040  * So, the recovery process is protected from other simultaneous mounts.
1041  */
1042 static int
1043 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1044 {
1045         struct the_nilfs *nilfs;
1046         struct nilfs_root *fsroot;
1047         __u64 cno;
1048         int err;
1049
1050         nilfs = alloc_nilfs(sb);
1051         if (!nilfs)
1052                 return -ENOMEM;
1053
1054         sb->s_fs_info = nilfs;
1055
1056         err = init_nilfs(nilfs, sb, (char *)data);
1057         if (err)
1058                 goto failed_nilfs;
1059
1060         sb->s_op = &nilfs_sops;
1061         sb->s_export_op = &nilfs_export_ops;
1062         sb->s_root = NULL;
1063         sb->s_time_gran = 1;
1064         sb->s_max_links = NILFS_LINK_MAX;
1065
1066         sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1067
1068         err = load_nilfs(nilfs, sb);
1069         if (err)
1070                 goto failed_nilfs;
1071
1072         cno = nilfs_last_cno(nilfs);
1073         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1074         if (err) {
1075                 nilfs_err(sb,
1076                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1077                           err, (unsigned long long)cno);
1078                 goto failed_unload;
1079         }
1080
1081         if (!sb_rdonly(sb)) {
1082                 err = nilfs_attach_log_writer(sb, fsroot);
1083                 if (err)
1084                         goto failed_checkpoint;
1085         }
1086
1087         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1088         if (err)
1089                 goto failed_segctor;
1090
1091         nilfs_put_root(fsroot);
1092
1093         if (!sb_rdonly(sb)) {
1094                 down_write(&nilfs->ns_sem);
1095                 nilfs_setup_super(sb, true);
1096                 up_write(&nilfs->ns_sem);
1097         }
1098
1099         return 0;
1100
1101  failed_segctor:
1102         nilfs_detach_log_writer(sb);
1103
1104  failed_checkpoint:
1105         nilfs_put_root(fsroot);
1106
1107  failed_unload:
1108         iput(nilfs->ns_sufile);
1109         iput(nilfs->ns_cpfile);
1110         iput(nilfs->ns_dat);
1111
1112  failed_nilfs:
1113         destroy_nilfs(nilfs);
1114         return err;
1115 }
1116
1117 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1118 {
1119         struct the_nilfs *nilfs = sb->s_fs_info;
1120         unsigned long old_sb_flags;
1121         unsigned long old_mount_opt;
1122         int err;
1123
1124         sync_filesystem(sb);
1125         old_sb_flags = sb->s_flags;
1126         old_mount_opt = nilfs->ns_mount_opt;
1127
1128         if (!parse_options(data, sb, 1)) {
1129                 err = -EINVAL;
1130                 goto restore_opts;
1131         }
1132         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1133
1134         err = -EINVAL;
1135
1136         if (!nilfs_valid_fs(nilfs)) {
1137                 nilfs_warn(sb,
1138                            "couldn't remount because the filesystem is in an incomplete recovery state");
1139                 goto restore_opts;
1140         }
1141
1142         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1143                 goto out;
1144         if (*flags & SB_RDONLY) {
1145                 sb->s_flags |= SB_RDONLY;
1146
1147                 /*
1148                  * Remounting a valid RW partition RDONLY, so set
1149                  * the RDONLY flag and then mark the partition as valid again.
1150                  */
1151                 down_write(&nilfs->ns_sem);
1152                 nilfs_cleanup_super(sb);
1153                 up_write(&nilfs->ns_sem);
1154         } else {
1155                 __u64 features;
1156                 struct nilfs_root *root;
1157
1158                 /*
1159                  * Mounting a RDONLY partition read-write, so reread and
1160                  * store the current valid flag.  (It may have been changed
1161                  * by fsck since we originally mounted the partition.)
1162                  */
1163                 down_read(&nilfs->ns_sem);
1164                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1165                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1166                 up_read(&nilfs->ns_sem);
1167                 if (features) {
1168                         nilfs_warn(sb,
1169                                    "couldn't remount RDWR because of unsupported optional features (%llx)",
1170                                    (unsigned long long)features);
1171                         err = -EROFS;
1172                         goto restore_opts;
1173                 }
1174
1175                 sb->s_flags &= ~SB_RDONLY;
1176
1177                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1178                 err = nilfs_attach_log_writer(sb, root);
1179                 if (err)
1180                         goto restore_opts;
1181
1182                 down_write(&nilfs->ns_sem);
1183                 nilfs_setup_super(sb, true);
1184                 up_write(&nilfs->ns_sem);
1185         }
1186  out:
1187         return 0;
1188
1189  restore_opts:
1190         sb->s_flags = old_sb_flags;
1191         nilfs->ns_mount_opt = old_mount_opt;
1192         return err;
1193 }
1194
1195 struct nilfs_super_data {
1196         struct block_device *bdev;
1197         __u64 cno;
1198         int flags;
1199 };
1200
1201 static int nilfs_parse_snapshot_option(const char *option,
1202                                        const substring_t *arg,
1203                                        struct nilfs_super_data *sd)
1204 {
1205         unsigned long long val;
1206         const char *msg = NULL;
1207         int err;
1208
1209         if (!(sd->flags & SB_RDONLY)) {
1210                 msg = "read-only option is not specified";
1211                 goto parse_error;
1212         }
1213
1214         err = kstrtoull(arg->from, 0, &val);
1215         if (err) {
1216                 if (err == -ERANGE)
1217                         msg = "too large checkpoint number";
1218                 else
1219                         msg = "malformed argument";
1220                 goto parse_error;
1221         } else if (val == 0) {
1222                 msg = "invalid checkpoint number 0";
1223                 goto parse_error;
1224         }
1225         sd->cno = val;
1226         return 0;
1227
1228 parse_error:
1229         nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1230         return 1;
1231 }
1232
1233 /**
1234  * nilfs_identify - pre-read mount options needed to identify mount instance
1235  * @data: mount options
1236  * @sd: nilfs_super_data
1237  */
1238 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1239 {
1240         char *p, *options = data;
1241         substring_t args[MAX_OPT_ARGS];
1242         int token;
1243         int ret = 0;
1244
1245         do {
1246                 p = strsep(&options, ",");
1247                 if (p != NULL && *p) {
1248                         token = match_token(p, tokens, args);
1249                         if (token == Opt_snapshot)
1250                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1251                                                                   sd);
1252                 }
1253                 if (!options)
1254                         break;
1255                 BUG_ON(options == data);
1256                 *(options - 1) = ',';
1257         } while (!ret);
1258         return ret;
1259 }
1260
1261 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1262 {
1263         s->s_bdev = data;
1264         s->s_dev = s->s_bdev->bd_dev;
1265         return 0;
1266 }
1267
1268 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1269 {
1270         return (void *)s->s_bdev == data;
1271 }
1272
1273 static struct dentry *
1274 nilfs_mount(struct file_system_type *fs_type, int flags,
1275              const char *dev_name, void *data)
1276 {
1277         struct nilfs_super_data sd;
1278         struct super_block *s;
1279         fmode_t mode = FMODE_READ | FMODE_EXCL;
1280         struct dentry *root_dentry;
1281         int err, s_new = false;
1282
1283         if (!(flags & SB_RDONLY))
1284                 mode |= FMODE_WRITE;
1285
1286         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1287         if (IS_ERR(sd.bdev))
1288                 return ERR_CAST(sd.bdev);
1289
1290         sd.cno = 0;
1291         sd.flags = flags;
1292         if (nilfs_identify((char *)data, &sd)) {
1293                 err = -EINVAL;
1294                 goto failed;
1295         }
1296
1297         /*
1298          * once the super is inserted into the list by sget, s_umount
1299          * will protect the lockfs code from trying to start a snapshot
1300          * while we are mounting
1301          */
1302         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1303         if (sd.bdev->bd_fsfreeze_count > 0) {
1304                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1305                 err = -EBUSY;
1306                 goto failed;
1307         }
1308         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1309                  sd.bdev);
1310         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1311         if (IS_ERR(s)) {
1312                 err = PTR_ERR(s);
1313                 goto failed;
1314         }
1315
1316         if (!s->s_root) {
1317                 s_new = true;
1318
1319                 /* New superblock instance created */
1320                 s->s_mode = mode;
1321                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1322                 sb_set_blocksize(s, block_size(sd.bdev));
1323
1324                 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1325                 if (err)
1326                         goto failed_super;
1327
1328                 s->s_flags |= SB_ACTIVE;
1329         } else if (!sd.cno) {
1330                 if (nilfs_tree_is_busy(s->s_root)) {
1331                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1332                                 nilfs_err(s,
1333                                           "the device already has a %s mount.",
1334                                           sb_rdonly(s) ? "read-only" : "read/write");
1335                                 err = -EBUSY;
1336                                 goto failed_super;
1337                         }
1338                 } else {
1339                         /*
1340                          * Try remount to setup mount states if the current
1341                          * tree is not mounted and only snapshots use this sb.
1342                          */
1343                         err = nilfs_remount(s, &flags, data);
1344                         if (err)
1345                                 goto failed_super;
1346                 }
1347         }
1348
1349         if (sd.cno) {
1350                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1351                 if (err)
1352                         goto failed_super;
1353         } else {
1354                 root_dentry = dget(s->s_root);
1355         }
1356
1357         if (!s_new)
1358                 blkdev_put(sd.bdev, mode);
1359
1360         return root_dentry;
1361
1362  failed_super:
1363         deactivate_locked_super(s);
1364
1365  failed:
1366         if (!s_new)
1367                 blkdev_put(sd.bdev, mode);
1368         return ERR_PTR(err);
1369 }
1370
1371 struct file_system_type nilfs_fs_type = {
1372         .owner    = THIS_MODULE,
1373         .name     = "nilfs2",
1374         .mount    = nilfs_mount,
1375         .kill_sb  = kill_block_super,
1376         .fs_flags = FS_REQUIRES_DEV,
1377 };
1378 MODULE_ALIAS_FS("nilfs2");
1379
1380 static void nilfs_inode_init_once(void *obj)
1381 {
1382         struct nilfs_inode_info *ii = obj;
1383
1384         INIT_LIST_HEAD(&ii->i_dirty);
1385 #ifdef CONFIG_NILFS_XATTR
1386         init_rwsem(&ii->xattr_sem);
1387 #endif
1388         inode_init_once(&ii->vfs_inode);
1389 }
1390
1391 static void nilfs_segbuf_init_once(void *obj)
1392 {
1393         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1394 }
1395
1396 static void nilfs_destroy_cachep(void)
1397 {
1398         /*
1399          * Make sure all delayed rcu free inodes are flushed before we
1400          * destroy cache.
1401          */
1402         rcu_barrier();
1403
1404         kmem_cache_destroy(nilfs_inode_cachep);
1405         kmem_cache_destroy(nilfs_transaction_cachep);
1406         kmem_cache_destroy(nilfs_segbuf_cachep);
1407         kmem_cache_destroy(nilfs_btree_path_cache);
1408 }
1409
1410 static int __init nilfs_init_cachep(void)
1411 {
1412         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1413                         sizeof(struct nilfs_inode_info), 0,
1414                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1415                         nilfs_inode_init_once);
1416         if (!nilfs_inode_cachep)
1417                 goto fail;
1418
1419         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1420                         sizeof(struct nilfs_transaction_info), 0,
1421                         SLAB_RECLAIM_ACCOUNT, NULL);
1422         if (!nilfs_transaction_cachep)
1423                 goto fail;
1424
1425         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1426                         sizeof(struct nilfs_segment_buffer), 0,
1427                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1428         if (!nilfs_segbuf_cachep)
1429                 goto fail;
1430
1431         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1432                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1433                         0, 0, NULL);
1434         if (!nilfs_btree_path_cache)
1435                 goto fail;
1436
1437         return 0;
1438
1439 fail:
1440         nilfs_destroy_cachep();
1441         return -ENOMEM;
1442 }
1443
1444 static int __init init_nilfs_fs(void)
1445 {
1446         int err;
1447
1448         err = nilfs_init_cachep();
1449         if (err)
1450                 goto fail;
1451
1452         err = nilfs_sysfs_init();
1453         if (err)
1454                 goto free_cachep;
1455
1456         err = register_filesystem(&nilfs_fs_type);
1457         if (err)
1458                 goto deinit_sysfs_entry;
1459
1460         printk(KERN_INFO "NILFS version 2 loaded\n");
1461         return 0;
1462
1463 deinit_sysfs_entry:
1464         nilfs_sysfs_exit();
1465 free_cachep:
1466         nilfs_destroy_cachep();
1467 fail:
1468         return err;
1469 }
1470
1471 static void __exit exit_nilfs_fs(void)
1472 {
1473         nilfs_destroy_cachep();
1474         nilfs_sysfs_exit();
1475         unregister_filesystem(&nilfs_fs_type);
1476 }
1477
1478 module_init(init_nilfs_fs)
1479 module_exit(exit_nilfs_fs)