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 (card@masi.ibp.fr)
  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 (davem@caip.rutgers.edu), 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 <linux/fs_context.h>
  39 #include "nilfs.h"
  40 #include "export.h"
  41 #include "mdt.h"
  42 #include "alloc.h"
  43 #include "btree.h"
  44 #include "btnode.h"
  45 #include "page.h"
  46 #include "cpfile.h"
  47 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  48 #include "ifile.h"
  49 #include "dat.h"
  50 #include "segment.h"
  51 #include "segbuf.h"
  52
  53 MODULE_AUTHOR("NTT Corp.");
  54 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  55                    "(NILFS)");
  56 MODULE_LICENSE("GPL");
  57
  58 static struct kmem_cache *nilfs_inode_cachep;
  59 struct kmem_cache *nilfs_transaction_cachep;
  60 struct kmem_cache *nilfs_segbuf_cachep;
  61 struct kmem_cache *nilfs_btree_path_cache;
  62
  63 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  64 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  65
  66 void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
  67 {
  68         struct va_format vaf;
  69         va_list args;
  70         int level;
  71
  72         va_start(args, fmt);
  73
  74         level = printk_get_level(fmt);
  75         vaf.fmt = printk_skip_level(fmt);
  76         vaf.va = &args;
  77
  78         if (sb)
  79                 printk("%c%cNILFS (%s): %pV\n",
  80                        KERN_SOH_ASCII, level, sb->s_id, &vaf);
  81         else
  82                 printk("%c%cNILFS: %pV\n",
  83                        KERN_SOH_ASCII, level, &vaf);
  84
  85         va_end(args);
  86 }
  87
  88 static void nilfs_set_error(struct super_block *sb)
  89 {
  90         struct the_nilfs *nilfs = sb->s_fs_info;
  91         struct nilfs_super_block **sbp;
  92
  93         down_write(&nilfs->ns_sem);
  94         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  95                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  96                 sbp = nilfs_prepare_super(sb, 0);
  97                 if (likely(sbp)) {
  98                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  99                         if (sbp[1])
 100                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 101                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 102                 }
 103         }
 104         up_write(&nilfs->ns_sem);
 105 }
 106
 107 /**
 108  * __nilfs_error() - report failure condition on a filesystem
 109  *
 110  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 111  * reporting an error message.  This function should be called when
 112  * NILFS detects incoherences or defects of meta data on disk.
 113  *
 114  * This implements the body of nilfs_error() macro.  Normally,
 115  * nilfs_error() should be used.  As for sustainable errors such as a
 116  * single-shot I/O error, nilfs_err() should be used instead.
 117  *
 118  * Callers should not add a trailing newline since this will do it.
 119  */
 120 void __nilfs_error(struct super_block *sb, const char *function,
 121                    const char *fmt, ...)
 122 {
 123         struct the_nilfs *nilfs = sb->s_fs_info;
 124         struct va_format vaf;
 125         va_list args;
 126
 127         va_start(args, fmt);
 128
 129         vaf.fmt = fmt;
 130         vaf.va = &args;
 131
 132         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 133                sb->s_id, function, &vaf);
 134
 135         va_end(args);
 136
 137         if (!sb_rdonly(sb)) {
 138                 nilfs_set_error(sb);
 139
 140                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 141                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 142                         sb->s_flags |= SB_RDONLY;
 143                 }
 144         }
 145
 146         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 147                 panic("NILFS (device %s): panic forced after error\n",
 148                       sb->s_id);
 149 }
 150
 151 struct inode *nilfs_alloc_inode(struct super_block *sb)
 152 {
 153         struct nilfs_inode_info *ii;
 154
 155         ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
 156         if (!ii)
 157                 return NULL;
 158         ii->i_bh = NULL;
 159         ii->i_state = 0;
 160         ii->i_cno = 0;
 161         ii->i_assoc_inode = NULL;
 162         ii->i_bmap = &ii->i_bmap_data;
 163         return &ii->vfs_inode;
 164 }
 165
 166 static void nilfs_free_inode(struct inode *inode)
 167 {
 168         if (nilfs_is_metadata_file_inode(inode))
 169                 nilfs_mdt_destroy(inode);
 170
 171         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 172 }
 173
 174 static int nilfs_sync_super(struct super_block *sb, int flag)
 175 {
 176         struct the_nilfs *nilfs = sb->s_fs_info;
 177         int err;
 178
 179  retry:
 180         set_buffer_dirty(nilfs->ns_sbh[0]);
 181         if (nilfs_test_opt(nilfs, BARRIER)) {
 182                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 183                                           REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
 184         } else {
 185                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 186         }
 187
 188         if (unlikely(err)) {
 189                 nilfs_err(sb, "unable to write superblock: err=%d", err);
 190                 if (err == -EIO && nilfs->ns_sbh[1]) {
 191                         /*
 192                          * sbp[0] points to newer log than sbp[1],
 193                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 194                          */
 195                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 196                                nilfs->ns_sbsize);
 197                         nilfs_fall_back_super_block(nilfs);
 198                         goto retry;
 199                 }
 200         } else {
 201                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 202
 203                 nilfs->ns_sbwcount++;
 204
 205                 /*
 206                  * The latest segment becomes trailable from the position
 207                  * written in superblock.
 208                  */
 209                 clear_nilfs_discontinued(nilfs);
 210
 211                 /* update GC protection for recent segments */
 212                 if (nilfs->ns_sbh[1]) {
 213                         if (flag == NILFS_SB_COMMIT_ALL) {
 214                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 215                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 216                                         goto out;
 217                         }
 218                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 219                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 220                                 sbp = nilfs->ns_sbp[1];
 221                 }
 222
 223                 spin_lock(&nilfs->ns_last_segment_lock);
 224                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 225                 spin_unlock(&nilfs->ns_last_segment_lock);
 226         }
 227  out:
 228         return err;
 229 }
 230
 231 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 232                           struct the_nilfs *nilfs)
 233 {
 234         sector_t nfreeblocks;
 235
 236         /* nilfs->ns_sem must be locked by the caller. */
 237         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 238         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 239
 240         spin_lock(&nilfs->ns_last_segment_lock);
 241         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 242         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 243         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 244         spin_unlock(&nilfs->ns_last_segment_lock);
 245 }
 246
 247 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 248                                                int flip)
 249 {
 250         struct the_nilfs *nilfs = sb->s_fs_info;
 251         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 252
 253         /* nilfs->ns_sem must be locked by the caller. */
 254         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 255                 if (sbp[1] &&
 256                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 257                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 258                 } else {
 259                         nilfs_crit(sb, "superblock broke");
 260                         return NULL;
 261                 }
 262         } else if (sbp[1] &&
 263                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 264                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 265         }
 266
 267         if (flip && sbp[1])
 268                 nilfs_swap_super_block(nilfs);
 269
 270         return sbp;
 271 }
 272
 273 int nilfs_commit_super(struct super_block *sb, int flag)
 274 {
 275         struct the_nilfs *nilfs = sb->s_fs_info;
 276         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 277         time64_t t;
 278
 279         /* nilfs->ns_sem must be locked by the caller. */
 280         t = ktime_get_real_seconds();
 281         nilfs->ns_sbwtime = t;
 282         sbp[0]->s_wtime = cpu_to_le64(t);
 283         sbp[0]->s_sum = 0;
 284         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 285                                              (unsigned char *)sbp[0],
 286                                              nilfs->ns_sbsize));
 287         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 288                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 289                 sbp[1]->s_sum = 0;
 290                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 291                                             (unsigned char *)sbp[1],
 292                                             nilfs->ns_sbsize));
 293         }
 294         clear_nilfs_sb_dirty(nilfs);
 295         nilfs->ns_flushed_device = 1;
 296         /* make sure store to ns_flushed_device cannot be reordered */
 297         smp_wmb();
 298         return nilfs_sync_super(sb, flag);
 299 }
 300
 301 /**
 302  * nilfs_cleanup_super() - write filesystem state for cleanup
 303  * @sb: super block instance to be unmounted or degraded to read-only
 304  *
 305  * This function restores state flags in the on-disk super block.
 306  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 307  * filesystem was not clean previously.
 308  */
 309 int nilfs_cleanup_super(struct super_block *sb)
 310 {
 311         struct the_nilfs *nilfs = sb->s_fs_info;
 312         struct nilfs_super_block **sbp;
 313         int flag = NILFS_SB_COMMIT;
 314         int ret = -EIO;
 315
 316         sbp = nilfs_prepare_super(sb, 0);
 317         if (sbp) {
 318                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 319                 nilfs_set_log_cursor(sbp[0], nilfs);
 320                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 321                         /*
 322                          * make the "clean" flag also to the opposite
 323                          * super block if both super blocks point to
 324                          * the same checkpoint.
 325                          */
 326                         sbp[1]->s_state = sbp[0]->s_state;
 327                         flag = NILFS_SB_COMMIT_ALL;
 328                 }
 329                 ret = nilfs_commit_super(sb, flag);
 330         }
 331         return ret;
 332 }
 333
 334 /**
 335  * nilfs_move_2nd_super - relocate secondary super block
 336  * @sb: super block instance
 337  * @sb2off: new offset of the secondary super block (in bytes)
 338  */
 339 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 340 {
 341         struct the_nilfs *nilfs = sb->s_fs_info;
 342         struct buffer_head *nsbh;
 343         struct nilfs_super_block *nsbp;
 344         sector_t blocknr, newblocknr;
 345         unsigned long offset;
 346         int sb2i;  /* array index of the secondary superblock */
 347         int ret = 0;
 348
 349         /* nilfs->ns_sem must be locked by the caller. */
 350         if (nilfs->ns_sbh[1] &&
 351             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 352                 sb2i = 1;
 353                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 354         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 355                 sb2i = 0;
 356                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 357         } else {
 358                 sb2i = -1;
 359                 blocknr = 0;
 360         }
 361         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 362                 goto out;  /* super block location is unchanged */
 363
 364         /* Get new super block buffer */
 365         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 366         offset = sb2off & (nilfs->ns_blocksize - 1);
 367         nsbh = sb_getblk(sb, newblocknr);
 368         if (!nsbh) {
 369                 nilfs_warn(sb,
 370                            "unable to move secondary superblock to block %llu",
 371                            (unsigned long long)newblocknr);
 372                 ret = -EIO;
 373                 goto out;
 374         }
 375         nsbp = (void *)nsbh->b_data + offset;
 376
 377         lock_buffer(nsbh);
 378         if (sb2i >= 0) {
 379                 /*
 380                  * The position of the second superblock only changes by 4KiB,
 381                  * which is larger than the maximum superblock data size
 382                  * (= 1KiB), so there is no need to use memmove() to allow
 383                  * overlap between source and destination.
 384                  */
 385                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 386
 387                 /*
 388                  * Zero fill after copy to avoid overwriting in case of move
 389                  * within the same block.
 390                  */
 391                 memset(nsbh->b_data, 0, offset);
 392                 memset((void *)nsbp + nilfs->ns_sbsize, 0,
 393                        nsbh->b_size - offset - nilfs->ns_sbsize);
 394         } else {
 395                 memset(nsbh->b_data, 0, nsbh->b_size);
 396         }
 397         set_buffer_uptodate(nsbh);
 398         unlock_buffer(nsbh);
 399
 400         if (sb2i >= 0) {
 401                 brelse(nilfs->ns_sbh[sb2i]);
 402                 nilfs->ns_sbh[sb2i] = nsbh;
 403                 nilfs->ns_sbp[sb2i] = nsbp;
 404         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 405                 /* secondary super block will be restored to index 1 */
 406                 nilfs->ns_sbh[1] = nsbh;
 407                 nilfs->ns_sbp[1] = nsbp;
 408         } else {
 409                 brelse(nsbh);
 410         }
 411 out:
 412         return ret;
 413 }
 414
 415 /**
 416  * nilfs_resize_fs - resize the filesystem
 417  * @sb: super block instance
 418  * @newsize: new size of the filesystem (in bytes)
 419  */
 420 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 421 {
 422         struct the_nilfs *nilfs = sb->s_fs_info;
 423         struct nilfs_super_block **sbp;
 424         __u64 devsize, newnsegs;
 425         loff_t sb2off;
 426         int ret;
 427
 428         ret = -ERANGE;
 429         devsize = bdev_nr_bytes(sb->s_bdev);
 430         if (newsize > devsize)
 431                 goto out;
 432
 433         /*
 434          * Prevent underflow in second superblock position calculation.
 435          * The exact minimum size check is done in nilfs_sufile_resize().
 436          */
 437         if (newsize < 4096) {
 438                 ret = -ENOSPC;
 439                 goto out;
 440         }
 441
 442         /*
 443          * Write lock is required to protect some functions depending
 444          * on the number of segments, the number of reserved segments,
 445          * and so forth.
 446          */
 447         down_write(&nilfs->ns_segctor_sem);
 448
 449         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 450         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 451         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 452
 453         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 454         up_write(&nilfs->ns_segctor_sem);
 455         if (ret < 0)
 456                 goto out;
 457
 458         ret = nilfs_construct_segment(sb);
 459         if (ret < 0)
 460                 goto out;
 461
 462         down_write(&nilfs->ns_sem);
 463         nilfs_move_2nd_super(sb, sb2off);
 464         ret = -EIO;
 465         sbp = nilfs_prepare_super(sb, 0);
 466         if (likely(sbp)) {
 467                 nilfs_set_log_cursor(sbp[0], nilfs);
 468                 /*
 469                  * Drop NILFS_RESIZE_FS flag for compatibility with
 470                  * mount-time resize which may be implemented in a
 471                  * future release.
 472                  */
 473                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 474                                               ~NILFS_RESIZE_FS);
 475                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 476                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 477                 if (sbp[1])
 478                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 479                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 480         }
 481         up_write(&nilfs->ns_sem);
 482
 483         /*
 484          * Reset the range of allocatable segments last.  This order
 485          * is important in the case of expansion because the secondary
 486          * superblock must be protected from log write until migration
 487          * completes.
 488          */
 489         if (!ret)
 490                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 491 out:
 492         return ret;
 493 }
 494
 495 static void nilfs_put_super(struct super_block *sb)
 496 {
 497         struct the_nilfs *nilfs = sb->s_fs_info;
 498
 499         nilfs_detach_log_writer(sb);
 500
 501         if (!sb_rdonly(sb)) {
 502                 down_write(&nilfs->ns_sem);
 503                 nilfs_cleanup_super(sb);
 504                 up_write(&nilfs->ns_sem);
 505         }
 506
 507         nilfs_sysfs_delete_device_group(nilfs);
 508         iput(nilfs->ns_sufile);
 509         iput(nilfs->ns_cpfile);
 510         iput(nilfs->ns_dat);
 511
 512         destroy_nilfs(nilfs);
 513         sb->s_fs_info = NULL;
 514 }
 515
 516 static int nilfs_sync_fs(struct super_block *sb, int wait)
 517 {
 518         struct the_nilfs *nilfs = sb->s_fs_info;
 519         struct nilfs_super_block **sbp;
 520         int err = 0;
 521
 522         /* This function is called when super block should be written back */
 523         if (wait)
 524                 err = nilfs_construct_segment(sb);
 525
 526         down_write(&nilfs->ns_sem);
 527         if (nilfs_sb_dirty(nilfs)) {
 528                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 529                 if (likely(sbp)) {
 530                         nilfs_set_log_cursor(sbp[0], nilfs);
 531                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 532                 }
 533         }
 534         up_write(&nilfs->ns_sem);
 535
 536         if (!err)
 537                 err = nilfs_flush_device(nilfs);
 538
 539         return err;
 540 }
 541
 542 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 543                             struct nilfs_root **rootp)
 544 {
 545         struct the_nilfs *nilfs = sb->s_fs_info;
 546         struct nilfs_root *root;
 547         struct nilfs_checkpoint *raw_cp;
 548         struct buffer_head *bh_cp;
 549         int err = -ENOMEM;
 550
 551         root = nilfs_find_or_create_root(
 552                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 553         if (!root)
 554                 return err;
 555
 556         if (root->ifile)
 557                 goto reuse; /* already attached checkpoint */
 558
 559         down_read(&nilfs->ns_segctor_sem);
 560         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 561                                           &bh_cp);
 562         up_read(&nilfs->ns_segctor_sem);
 563         if (unlikely(err)) {
 564                 if (err == -ENOENT || err == -EINVAL) {
 565                         nilfs_err(sb,
 566                                   "Invalid checkpoint (checkpoint number=%llu)",
 567                                   (unsigned long long)cno);
 568                         err = -EINVAL;
 569                 }
 570                 goto failed;
 571         }
 572
 573         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 574                                &raw_cp->cp_ifile_inode, &root->ifile);
 575         if (err)
 576                 goto failed_bh;
 577
 578         atomic64_set(&root->inodes_count,
 579                         le64_to_cpu(raw_cp->cp_inodes_count));
 580         atomic64_set(&root->blocks_count,
 581                         le64_to_cpu(raw_cp->cp_blocks_count));
 582
 583         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 584
 585  reuse:
 586         *rootp = root;
 587         return 0;
 588
 589  failed_bh:
 590         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 591  failed:
 592         nilfs_put_root(root);
 593
 594         return err;
 595 }
 596
 597 static int nilfs_freeze(struct super_block *sb)
 598 {
 599         struct the_nilfs *nilfs = sb->s_fs_info;
 600         int err;
 601
 602         if (sb_rdonly(sb))
 603                 return 0;
 604
 605         /* Mark super block clean */
 606         down_write(&nilfs->ns_sem);
 607         err = nilfs_cleanup_super(sb);
 608         up_write(&nilfs->ns_sem);
 609         return err;
 610 }
 611
 612 static int nilfs_unfreeze(struct super_block *sb)
 613 {
 614         struct the_nilfs *nilfs = sb->s_fs_info;
 615
 616         if (sb_rdonly(sb))
 617                 return 0;
 618
 619         down_write(&nilfs->ns_sem);
 620         nilfs_setup_super(sb, false);
 621         up_write(&nilfs->ns_sem);
 622         return 0;
 623 }
 624
 625 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 626 {
 627         struct super_block *sb = dentry->d_sb;
 628         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 629         struct the_nilfs *nilfs = root->nilfs;
 630         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 631         unsigned long long blocks;
 632         unsigned long overhead;
 633         unsigned long nrsvblocks;
 634         sector_t nfreeblocks;
 635         u64 nmaxinodes, nfreeinodes;
 636         int err;
 637
 638         /*
 639          * Compute all of the segment blocks
 640          *
 641          * The blocks before first segment and after last segment
 642          * are excluded.
 643          */
 644         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 645                 - nilfs->ns_first_data_block;
 646         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 647
 648         /*
 649          * Compute the overhead
 650          *
 651          * When distributing meta data blocks outside segment structure,
 652          * We must count them as the overhead.
 653          */
 654         overhead = 0;
 655
 656         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 657         if (unlikely(err))
 658                 return err;
 659
 660         err = nilfs_ifile_count_free_inodes(root->ifile,
 661                                             &nmaxinodes, &nfreeinodes);
 662         if (unlikely(err)) {
 663                 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
 664                 if (err == -ERANGE) {
 665                         /*
 666                          * If nilfs_palloc_count_max_entries() returns
 667                          * -ERANGE error code then we simply treat
 668                          * curent inodes count as maximum possible and
 669                          * zero as free inodes value.
 670                          */
 671                         nmaxinodes = atomic64_read(&root->inodes_count);
 672                         nfreeinodes = 0;
 673                         err = 0;
 674                 } else
 675                         return err;
 676         }
 677
 678         buf->f_type = NILFS_SUPER_MAGIC;
 679         buf->f_bsize = sb->s_blocksize;
 680         buf->f_blocks = blocks - overhead;
 681         buf->f_bfree = nfreeblocks;
 682         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 683                 (buf->f_bfree - nrsvblocks) : 0;
 684         buf->f_files = nmaxinodes;
 685         buf->f_ffree = nfreeinodes;
 686         buf->f_namelen = NILFS_NAME_LEN;
 687         buf->f_fsid = u64_to_fsid(id);
 688
 689         return 0;
 690 }
 691
 692 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 693 {
 694         struct super_block *sb = dentry->d_sb;
 695         struct the_nilfs *nilfs = sb->s_fs_info;
 696         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 697
 698         if (!nilfs_test_opt(nilfs, BARRIER))
 699                 seq_puts(seq, ",nobarrier");
 700         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 701                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 702         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 703                 seq_puts(seq, ",errors=panic");
 704         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 705                 seq_puts(seq, ",errors=continue");
 706         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 707                 seq_puts(seq, ",order=strict");
 708         if (nilfs_test_opt(nilfs, NORECOVERY))
 709                 seq_puts(seq, ",norecovery");
 710         if (nilfs_test_opt(nilfs, DISCARD))
 711                 seq_puts(seq, ",discard");
 712
 713         return 0;
 714 }
 715
 716 static const struct super_operations nilfs_sops = {
 717         .alloc_inode    = nilfs_alloc_inode,
 718         .free_inode     = nilfs_free_inode,
 719         .dirty_inode    = nilfs_dirty_inode,
 720         .evict_inode    = nilfs_evict_inode,
 721         .put_super      = nilfs_put_super,
 722         .sync_fs        = nilfs_sync_fs,
 723         .freeze_fs      = nilfs_freeze,
 724         .unfreeze_fs    = nilfs_unfreeze,
 725         .statfs         = nilfs_statfs,
 726         .remount_fs     = nilfs_remount,
 727         .show_options = nilfs_show_options
 728 };
 729
 730 enum {
 731         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 732         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 733         Opt_discard, Opt_nodiscard, Opt_err,
 734 };
 735
 736 static match_table_t tokens = {
 737         {Opt_err_cont, "errors=continue"},
 738         {Opt_err_panic, "errors=panic"},
 739         {Opt_err_ro, "errors=remount-ro"},
 740         {Opt_barrier, "barrier"},
 741         {Opt_nobarrier, "nobarrier"},
 742         {Opt_snapshot, "cp=%u"},
 743         {Opt_order, "order=%s"},
 744         {Opt_norecovery, "norecovery"},
 745         {Opt_discard, "discard"},
 746         {Opt_nodiscard, "nodiscard"},
 747         {Opt_err, NULL}
 748 };
 749
 750 static int parse_options(char *options, struct super_block *sb, int is_remount)
 751 {
 752         struct the_nilfs *nilfs = sb->s_fs_info;
 753         char *p;
 754         substring_t args[MAX_OPT_ARGS];
 755
 756         if (!options)
 757                 return 1;
 758
 759         while ((p = strsep(&options, ",")) != NULL) {
 760                 int token;
 761
 762                 if (!*p)
 763                         continue;
 764
 765                 token = match_token(p, tokens, args);
 766                 switch (token) {
 767                 case Opt_barrier:
 768                         nilfs_set_opt(nilfs, BARRIER);
 769                         break;
 770                 case Opt_nobarrier:
 771                         nilfs_clear_opt(nilfs, BARRIER);
 772                         break;
 773                 case Opt_order:
 774                         if (strcmp(args[0].from, "relaxed") == 0)
 775                                 /* Ordered data semantics */
 776                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 777                         else if (strcmp(args[0].from, "strict") == 0)
 778                                 /* Strict in-order semantics */
 779                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 780                         else
 781                                 return 0;
 782                         break;
 783                 case Opt_err_panic:
 784                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 785                         break;
 786                 case Opt_err_ro:
 787                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 788                         break;
 789                 case Opt_err_cont:
 790                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 791                         break;
 792                 case Opt_snapshot:
 793                         if (is_remount) {
 794                                 nilfs_err(sb,
 795                                           "\"%s\" option is invalid for remount",
 796                                           p);
 797                                 return 0;
 798                         }
 799                         break;
 800                 case Opt_norecovery:
 801                         nilfs_set_opt(nilfs, NORECOVERY);
 802                         break;
 803                 case Opt_discard:
 804                         nilfs_set_opt(nilfs, DISCARD);
 805                         break;
 806                 case Opt_nodiscard:
 807                         nilfs_clear_opt(nilfs, DISCARD);
 808                         break;
 809                 default:
 810                         nilfs_err(sb, "unrecognized mount option \"%s\"", p);
 811                         return 0;
 812                 }
 813         }
 814         return 1;
 815 }
 816
 817 static inline void
 818 nilfs_set_default_options(struct super_block *sb,
 819                           struct nilfs_super_block *sbp)
 820 {
 821         struct the_nilfs *nilfs = sb->s_fs_info;
 822
 823         nilfs->ns_mount_opt =
 824                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 825 }
 826
 827 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 828 {
 829         struct the_nilfs *nilfs = sb->s_fs_info;
 830         struct nilfs_super_block **sbp;
 831         int max_mnt_count;
 832         int mnt_count;
 833
 834         /* nilfs->ns_sem must be locked by the caller. */
 835         sbp = nilfs_prepare_super(sb, 0);
 836         if (!sbp)
 837                 return -EIO;
 838
 839         if (!is_mount)
 840                 goto skip_mount_setup;
 841
 842         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 843         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 844
 845         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 846                 nilfs_warn(sb, "mounting fs with errors");
 847 #if 0
 848         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 849                 nilfs_warn(sb, "maximal mount count reached");
 850 #endif
 851         }
 852         if (!max_mnt_count)
 853                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 854
 855         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 856         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 857
 858 skip_mount_setup:
 859         sbp[0]->s_state =
 860                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 861         /* synchronize sbp[1] with sbp[0] */
 862         if (sbp[1])
 863                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 864         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 865 }
 866
 867 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 868                                                  u64 pos, int blocksize,
 869                                                  struct buffer_head **pbh)
 870 {
 871         unsigned long long sb_index = pos;
 872         unsigned long offset;
 873
 874         offset = do_div(sb_index, blocksize);
 875         *pbh = sb_bread(sb, sb_index);
 876         if (!*pbh)
 877                 return NULL;
 878         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 879 }
 880
 881 int nilfs_store_magic_and_option(struct super_block *sb,
 882                                  struct nilfs_super_block *sbp,
 883                                  char *data)
 884 {
 885         struct the_nilfs *nilfs = sb->s_fs_info;
 886
 887         sb->s_magic = le16_to_cpu(sbp->s_magic);
 888
 889         /* FS independent flags */
 890 #ifdef NILFS_ATIME_DISABLE
 891         sb->s_flags |= SB_NOATIME;
 892 #endif
 893
 894         nilfs_set_default_options(sb, sbp);
 895
 896         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 897         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 898         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 899         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 900
 901         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 902 }
 903
 904 int nilfs_check_feature_compatibility(struct super_block *sb,
 905                                       struct nilfs_super_block *sbp)
 906 {
 907         __u64 features;
 908
 909         features = le64_to_cpu(sbp->s_feature_incompat) &
 910                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 911         if (features) {
 912                 nilfs_err(sb,
 913                           "couldn't mount because of unsupported optional features (%llx)",
 914                           (unsigned long long)features);
 915                 return -EINVAL;
 916         }
 917         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 918                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 919         if (!sb_rdonly(sb) && features) {
 920                 nilfs_err(sb,
 921                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 922                           (unsigned long long)features);
 923                 return -EINVAL;
 924         }
 925         return 0;
 926 }
 927
 928 static int nilfs_get_root_dentry(struct super_block *sb,
 929                                  struct nilfs_root *root,
 930                                  struct dentry **root_dentry)
 931 {
 932         struct inode *inode;
 933         struct dentry *dentry;
 934         int ret = 0;
 935
 936         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 937         if (IS_ERR(inode)) {
 938                 ret = PTR_ERR(inode);
 939                 nilfs_err(sb, "error %d getting root inode", ret);
 940                 goto out;
 941         }
 942         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 943                 iput(inode);
 944                 nilfs_err(sb, "corrupt root inode");
 945                 ret = -EINVAL;
 946                 goto out;
 947         }
 948
 949         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 950                 dentry = d_find_alias(inode);
 951                 if (!dentry) {
 952                         dentry = d_make_root(inode);
 953                         if (!dentry) {
 954                                 ret = -ENOMEM;
 955                                 goto failed_dentry;
 956                         }
 957                 } else {
 958                         iput(inode);
 959                 }
 960         } else {
 961                 dentry = d_obtain_root(inode);
 962                 if (IS_ERR(dentry)) {
 963                         ret = PTR_ERR(dentry);
 964                         goto failed_dentry;
 965                 }
 966         }
 967         *root_dentry = dentry;
 968  out:
 969         return ret;
 970
 971  failed_dentry:
 972         nilfs_err(sb, "error %d getting root dentry", ret);
 973         goto out;
 974 }
 975
 976 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 977                                  struct dentry **root_dentry)
 978 {
 979         struct the_nilfs *nilfs = s->s_fs_info;
 980         struct nilfs_root *root;
 981         int ret;
 982
 983         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 984
 985         down_read(&nilfs->ns_segctor_sem);
 986         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 987         up_read(&nilfs->ns_segctor_sem);
 988         if (ret < 0) {
 989                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 990                 goto out;
 991         } else if (!ret) {
 992                 nilfs_err(s,
 993                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 994                           (unsigned long long)cno);
 995                 ret = -EINVAL;
 996                 goto out;
 997         }
 998
 999         ret = nilfs_attach_checkpoint(s, cno, false, &root);
1000         if (ret) {
1001                 nilfs_err(s,
1002                           "error %d while loading snapshot (checkpoint number=%llu)",
1003                           ret, (unsigned long long)cno);
1004                 goto out;
1005         }
1006         ret = nilfs_get_root_dentry(s, root, root_dentry);
1007         nilfs_put_root(root);
1008  out:
1009         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1010         return ret;
1011 }
1012
1013 /**
1014  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1015  * @root_dentry: root dentry of the tree to be shrunk
1016  *
1017  * This function returns true if the tree was in-use.
1018  */
1019 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1020 {
1021         shrink_dcache_parent(root_dentry);
1022         return d_count(root_dentry) > 1;
1023 }
1024
1025 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1026 {
1027         struct the_nilfs *nilfs = sb->s_fs_info;
1028         struct nilfs_root *root;
1029         struct inode *inode;
1030         struct dentry *dentry;
1031         int ret;
1032
1033         if (cno > nilfs->ns_cno)
1034                 return false;
1035
1036         if (cno >= nilfs_last_cno(nilfs))
1037                 return true;    /* protect recent checkpoints */
1038
1039         ret = false;
1040         root = nilfs_lookup_root(nilfs, cno);
1041         if (root) {
1042                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1043                 if (inode) {
1044                         dentry = d_find_alias(inode);
1045                         if (dentry) {
1046                                 ret = nilfs_tree_is_busy(dentry);
1047                                 dput(dentry);
1048                         }
1049                         iput(inode);
1050                 }
1051                 nilfs_put_root(root);
1052         }
1053         return ret;
1054 }
1055
1056 /**
1057  * nilfs_fill_super() - initialize a super block instance
1058  * @sb: super_block
1059  * @data: mount options
1060  * @silent: silent mode flag
1061  *
1062  * This function is called exclusively by nilfs->ns_mount_mutex.
1063  * So, the recovery process is protected from other simultaneous mounts.
1064  */
1065 static int
1066 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1067 {
1068         struct the_nilfs *nilfs;
1069         struct nilfs_root *fsroot;
1070         __u64 cno;
1071         int err;
1072
1073         nilfs = alloc_nilfs(sb);
1074         if (!nilfs)
1075                 return -ENOMEM;
1076
1077         sb->s_fs_info = nilfs;
1078
1079         err = init_nilfs(nilfs, sb, (char *)data);
1080         if (err)
1081                 goto failed_nilfs;
1082
1083         sb->s_op = &nilfs_sops;
1084         sb->s_export_op = &nilfs_export_ops;
1085         sb->s_root = NULL;
1086         sb->s_time_gran = 1;
1087         sb->s_max_links = NILFS_LINK_MAX;
1088
1089         sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1090
1091         err = load_nilfs(nilfs, sb);
1092         if (err)
1093                 goto failed_nilfs;
1094
1095         cno = nilfs_last_cno(nilfs);
1096         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1097         if (err) {
1098                 nilfs_err(sb,
1099                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1100                           err, (unsigned long long)cno);
1101                 goto failed_unload;
1102         }
1103
1104         if (!sb_rdonly(sb)) {
1105                 err = nilfs_attach_log_writer(sb, fsroot);
1106                 if (err)
1107                         goto failed_checkpoint;
1108         }
1109
1110         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1111         if (err)
1112                 goto failed_segctor;
1113
1114         nilfs_put_root(fsroot);
1115
1116         if (!sb_rdonly(sb)) {
1117                 down_write(&nilfs->ns_sem);
1118                 nilfs_setup_super(sb, true);
1119                 up_write(&nilfs->ns_sem);
1120         }
1121
1122         return 0;
1123
1124  failed_segctor:
1125         nilfs_detach_log_writer(sb);
1126
1127  failed_checkpoint:
1128         nilfs_put_root(fsroot);
1129
1130  failed_unload:
1131         nilfs_sysfs_delete_device_group(nilfs);
1132         iput(nilfs->ns_sufile);
1133         iput(nilfs->ns_cpfile);
1134         iput(nilfs->ns_dat);
1135
1136  failed_nilfs:
1137         destroy_nilfs(nilfs);
1138         return err;
1139 }
1140
1141 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1142 {
1143         struct the_nilfs *nilfs = sb->s_fs_info;
1144         unsigned long old_sb_flags;
1145         unsigned long old_mount_opt;
1146         int err;
1147
1148         sync_filesystem(sb);
1149         old_sb_flags = sb->s_flags;
1150         old_mount_opt = nilfs->ns_mount_opt;
1151
1152         if (!parse_options(data, sb, 1)) {
1153                 err = -EINVAL;
1154                 goto restore_opts;
1155         }
1156         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1157
1158         err = -EINVAL;
1159
1160         if (!nilfs_valid_fs(nilfs)) {
1161                 nilfs_warn(sb,
1162                            "couldn't remount because the filesystem is in an incomplete recovery state");
1163                 goto restore_opts;
1164         }
1165
1166         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1167                 goto out;
1168         if (*flags & SB_RDONLY) {
1169                 sb->s_flags |= SB_RDONLY;
1170
1171                 /*
1172                  * Remounting a valid RW partition RDONLY, so set
1173                  * the RDONLY flag and then mark the partition as valid again.
1174                  */
1175                 down_write(&nilfs->ns_sem);
1176                 nilfs_cleanup_super(sb);
1177                 up_write(&nilfs->ns_sem);
1178         } else {
1179                 __u64 features;
1180                 struct nilfs_root *root;
1181
1182                 /*
1183                  * Mounting a RDONLY partition read-write, so reread and
1184                  * store the current valid flag.  (It may have been changed
1185                  * by fsck since we originally mounted the partition.)
1186                  */
1187                 down_read(&nilfs->ns_sem);
1188                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1189                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1190                 up_read(&nilfs->ns_sem);
1191                 if (features) {
1192                         nilfs_warn(sb,
1193                                    "couldn't remount RDWR because of unsupported optional features (%llx)",
1194                                    (unsigned long long)features);
1195                         err = -EROFS;
1196                         goto restore_opts;
1197                 }
1198
1199                 sb->s_flags &= ~SB_RDONLY;
1200
1201                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1202                 err = nilfs_attach_log_writer(sb, root);
1203                 if (err)
1204                         goto restore_opts;
1205
1206                 down_write(&nilfs->ns_sem);
1207                 nilfs_setup_super(sb, true);
1208                 up_write(&nilfs->ns_sem);
1209         }
1210  out:
1211         return 0;
1212
1213  restore_opts:
1214         sb->s_flags = old_sb_flags;
1215         nilfs->ns_mount_opt = old_mount_opt;
1216         return err;
1217 }
1218
1219 struct nilfs_super_data {
1220         __u64 cno;
1221         int flags;
1222 };
1223
1224 static int nilfs_parse_snapshot_option(const char *option,
1225                                        const substring_t *arg,
1226                                        struct nilfs_super_data *sd)
1227 {
1228         unsigned long long val;
1229         const char *msg = NULL;
1230         int err;
1231
1232         if (!(sd->flags & SB_RDONLY)) {
1233                 msg = "read-only option is not specified";
1234                 goto parse_error;
1235         }
1236
1237         err = kstrtoull(arg->from, 0, &val);
1238         if (err) {
1239                 if (err == -ERANGE)
1240                         msg = "too large checkpoint number";
1241                 else
1242                         msg = "malformed argument";
1243                 goto parse_error;
1244         } else if (val == 0) {
1245                 msg = "invalid checkpoint number 0";
1246                 goto parse_error;
1247         }
1248         sd->cno = val;
1249         return 0;
1250
1251 parse_error:
1252         nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1253         return 1;
1254 }
1255
1256 /**
1257  * nilfs_identify - pre-read mount options needed to identify mount instance
1258  * @data: mount options
1259  * @sd: nilfs_super_data
1260  */
1261 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1262 {
1263         char *p, *options = data;
1264         substring_t args[MAX_OPT_ARGS];
1265         int token;
1266         int ret = 0;
1267
1268         do {
1269                 p = strsep(&options, ",");
1270                 if (p != NULL && *p) {
1271                         token = match_token(p, tokens, args);
1272                         if (token == Opt_snapshot)
1273                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1274                                                                   sd);
1275                 }
1276                 if (!options)
1277                         break;
1278                 BUG_ON(options == data);
1279                 *(options - 1) = ',';
1280         } while (!ret);
1281         return ret;
1282 }
1283
1284 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1285 {
1286         s->s_dev = *(dev_t *)data;
1287         return 0;
1288 }
1289
1290 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1291 {
1292         return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1293 }
1294
1295 static struct dentry *
1296 nilfs_mount(struct file_system_type *fs_type, int flags,
1297              const char *dev_name, void *data)
1298 {
1299         struct nilfs_super_data sd = { .flags = flags };
1300         struct super_block *s;
1301         dev_t dev;
1302         int err;
1303
1304         if (nilfs_identify(data, &sd))
1305                 return ERR_PTR(-EINVAL);
1306
1307         err = lookup_bdev(dev_name, &dev);
1308         if (err)
1309                 return ERR_PTR(err);
1310
1311         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1312                  &dev);
1313         if (IS_ERR(s))
1314                 return ERR_CAST(s);
1315
1316         if (!s->s_root) {
1317                 /*
1318                  * We drop s_umount here because we need to open the bdev and
1319                  * bdev->open_mutex ranks above s_umount (blkdev_put() ->
1320                  * __invalidate_device()). It is safe because we have active sb
1321                  * reference and SB_BORN is not set yet.
1322                  */
1323                 up_write(&s->s_umount);
1324                 err = setup_bdev_super(s, flags, NULL);
1325                 down_write(&s->s_umount);
1326                 if (!err)
1327                         err = nilfs_fill_super(s, data,
1328                                                flags & SB_SILENT ? 1 : 0);
1329                 if (err)
1330                         goto failed_super;
1331
1332                 s->s_flags |= SB_ACTIVE;
1333         } else if (!sd.cno) {
1334                 if (nilfs_tree_is_busy(s->s_root)) {
1335                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1336                                 nilfs_err(s,
1337                                           "the device already has a %s mount.",
1338                                           sb_rdonly(s) ? "read-only" : "read/write");
1339                                 err = -EBUSY;
1340                                 goto failed_super;
1341                         }
1342                 } else {
1343                         /*
1344                          * Try remount to setup mount states if the current
1345                          * tree is not mounted and only snapshots use this sb.
1346                          */
1347                         err = nilfs_remount(s, &flags, data);
1348                         if (err)
1349                                 goto failed_super;
1350                 }
1351         }
1352
1353         if (sd.cno) {
1354                 struct dentry *root_dentry;
1355
1356                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1357                 if (err)
1358                         goto failed_super;
1359                 return root_dentry;
1360         }
1361
1362         return dget(s->s_root);
1363
1364  failed_super:
1365         deactivate_locked_super(s);
1366         return ERR_PTR(err);
1367 }
1368
1369 struct file_system_type nilfs_fs_type = {
1370         .owner    = THIS_MODULE,
1371         .name     = "nilfs2",
1372         .mount    = nilfs_mount,
1373         .kill_sb  = kill_block_super,
1374         .fs_flags = FS_REQUIRES_DEV,
1375 };
1376 MODULE_ALIAS_FS("nilfs2");
1377
1378 static void nilfs_inode_init_once(void *obj)
1379 {
1380         struct nilfs_inode_info *ii = obj;
1381
1382         INIT_LIST_HEAD(&ii->i_dirty);
1383 #ifdef CONFIG_NILFS_XATTR
1384         init_rwsem(&ii->xattr_sem);
1385 #endif
1386         inode_init_once(&ii->vfs_inode);
1387 }
1388
1389 static void nilfs_segbuf_init_once(void *obj)
1390 {
1391         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1392 }
1393
1394 static void nilfs_destroy_cachep(void)
1395 {
1396         /*
1397          * Make sure all delayed rcu free inodes are flushed before we
1398          * destroy cache.
1399          */
1400         rcu_barrier();
1401
1402         kmem_cache_destroy(nilfs_inode_cachep);
1403         kmem_cache_destroy(nilfs_transaction_cachep);
1404         kmem_cache_destroy(nilfs_segbuf_cachep);
1405         kmem_cache_destroy(nilfs_btree_path_cache);
1406 }
1407
1408 static int __init nilfs_init_cachep(void)
1409 {
1410         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1411                         sizeof(struct nilfs_inode_info), 0,
1412                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1413                         nilfs_inode_init_once);
1414         if (!nilfs_inode_cachep)
1415                 goto fail;
1416
1417         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1418                         sizeof(struct nilfs_transaction_info), 0,
1419                         SLAB_RECLAIM_ACCOUNT, NULL);
1420         if (!nilfs_transaction_cachep)
1421                 goto fail;
1422
1423         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1424                         sizeof(struct nilfs_segment_buffer), 0,
1425                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1426         if (!nilfs_segbuf_cachep)
1427                 goto fail;
1428
1429         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1430                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1431                         0, 0, NULL);
1432         if (!nilfs_btree_path_cache)
1433                 goto fail;
1434
1435         return 0;
1436
1437 fail:
1438         nilfs_destroy_cachep();
1439         return -ENOMEM;
1440 }
1441
1442 static int __init init_nilfs_fs(void)
1443 {
1444         int err;
1445
1446         err = nilfs_init_cachep();
1447         if (err)
1448                 goto fail;
1449
1450         err = nilfs_sysfs_init();
1451         if (err)
1452                 goto free_cachep;
1453
1454         err = register_filesystem(&nilfs_fs_type);
1455         if (err)
1456                 goto deinit_sysfs_entry;
1457
1458         printk(KERN_INFO "NILFS version 2 loaded\n");
1459         return 0;
1460
1461 deinit_sysfs_entry:
1462         nilfs_sysfs_exit();
1463 free_cachep:
1464         nilfs_destroy_cachep();
1465 fail:
1466         return err;
1467 }
1468
1469 static void __exit exit_nilfs_fs(void)
1470 {
1471         nilfs_destroy_cachep();
1472         nilfs_sysfs_exit();
1473         unregister_filesystem(&nilfs_fs_type);
1474 }
1475
1476 module_init(init_nilfs_fs)
1477 module_exit(exit_nilfs_fs)