fs/nilfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * super.c - NILFS module and super block management.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  */
   9 /*
  10  *  linux/fs/ext2/super.c
  11  *
  12  * Copyright (C) 1992, 1993, 1994, 1995
  13  * Remy Card (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 "nilfs.h"
  39 #include "export.h"
  40 #include "mdt.h"
  41 #include "alloc.h"
  42 #include "btree.h"
  43 #include "btnode.h"
  44 #include "page.h"
  45 #include "cpfile.h"
  46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  47 #include "ifile.h"
  48 #include "dat.h"
  49 #include "segment.h"
  50 #include "segbuf.h"
  51
  52 MODULE_AUTHOR("NTT Corp.");
  53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  54                    "(NILFS)");
  55 MODULE_LICENSE("GPL");
  56
  57 static struct kmem_cache *nilfs_inode_cachep;
  58 struct kmem_cache *nilfs_transaction_cachep;
  59 struct kmem_cache *nilfs_segbuf_cachep;
  60 struct kmem_cache *nilfs_btree_path_cache;
  61
  62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  64
  65 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
  66                  ...)
  67 {
  68         struct va_format vaf;
  69         va_list args;
  70
  71         va_start(args, fmt);
  72         vaf.fmt = fmt;
  73         vaf.va = &args;
  74         if (sb)
  75                 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
  76         else
  77                 printk("%sNILFS: %pV\n", level, &vaf);
  78         va_end(args);
  79 }
  80
  81 static void nilfs_set_error(struct super_block *sb)
  82 {
  83         struct the_nilfs *nilfs = sb->s_fs_info;
  84         struct nilfs_super_block **sbp;
  85
  86         down_write(&nilfs->ns_sem);
  87         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  88                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  89                 sbp = nilfs_prepare_super(sb, 0);
  90                 if (likely(sbp)) {
  91                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  92                         if (sbp[1])
  93                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  94                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
  95                 }
  96         }
  97         up_write(&nilfs->ns_sem);
  98 }
  99
 100 /**
 101  * __nilfs_error() - report failure condition on a filesystem
 102  *
 103  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 104  * reporting an error message.  This function should be called when
 105  * NILFS detects incoherences or defects of meta data on disk.
 106  *
 107  * This implements the body of nilfs_error() macro.  Normally,
 108  * nilfs_error() should be used.  As for sustainable errors such as a
 109  * single-shot I/O error, nilfs_msg() should be used instead.
 110  *
 111  * Callers should not add a trailing newline since this will do it.
 112  */
 113 void __nilfs_error(struct super_block *sb, const char *function,
 114                    const char *fmt, ...)
 115 {
 116         struct the_nilfs *nilfs = sb->s_fs_info;
 117         struct va_format vaf;
 118         va_list args;
 119
 120         va_start(args, fmt);
 121
 122         vaf.fmt = fmt;
 123         vaf.va = &args;
 124
 125         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 126                sb->s_id, function, &vaf);
 127
 128         va_end(args);
 129
 130         if (!sb_rdonly(sb)) {
 131                 nilfs_set_error(sb);
 132
 133                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 134                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 135                         sb->s_flags |= SB_RDONLY;
 136                 }
 137         }
 138
 139         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 140                 panic("NILFS (device %s): panic forced after error\n",
 141                       sb->s_id);
 142 }
 143
 144 struct inode *nilfs_alloc_inode(struct super_block *sb)
 145 {
 146         struct nilfs_inode_info *ii;
 147
 148         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 149         if (!ii)
 150                 return NULL;
 151         ii->i_bh = NULL;
 152         ii->i_state = 0;
 153         ii->i_cno = 0;
 154         ii->i_assoc_inode = NULL;
 155         ii->i_bmap = &ii->i_bmap_data;
 156         return &ii->vfs_inode;
 157 }
 158
 159 static void nilfs_i_callback(struct rcu_head *head)
 160 {
 161         struct inode *inode = container_of(head, struct inode, i_rcu);
 162
 163         if (nilfs_is_metadata_file_inode(inode))
 164                 nilfs_mdt_destroy(inode);
 165
 166         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 167 }
 168
 169 void nilfs_destroy_inode(struct inode *inode)
 170 {
 171         call_rcu(&inode->i_rcu, nilfs_i_callback);
 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_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
 190                           err);
 191                 if (err == -EIO && nilfs->ns_sbh[1]) {
 192                         /*
 193                          * sbp[0] points to newer log than sbp[1],
 194                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 195                          */
 196                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 197                                nilfs->ns_sbsize);
 198                         nilfs_fall_back_super_block(nilfs);
 199                         goto retry;
 200                 }
 201         } else {
 202                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 203
 204                 nilfs->ns_sbwcount++;
 205
 206                 /*
 207                  * The latest segment becomes trailable from the position
 208                  * written in superblock.
 209                  */
 210                 clear_nilfs_discontinued(nilfs);
 211
 212                 /* update GC protection for recent segments */
 213                 if (nilfs->ns_sbh[1]) {
 214                         if (flag == NILFS_SB_COMMIT_ALL) {
 215                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 216                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 217                                         goto out;
 218                         }
 219                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 220                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 221                                 sbp = nilfs->ns_sbp[1];
 222                 }
 223
 224                 spin_lock(&nilfs->ns_last_segment_lock);
 225                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 226                 spin_unlock(&nilfs->ns_last_segment_lock);
 227         }
 228  out:
 229         return err;
 230 }
 231
 232 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 233                           struct the_nilfs *nilfs)
 234 {
 235         sector_t nfreeblocks;
 236
 237         /* nilfs->ns_sem must be locked by the caller. */
 238         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 239         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 240
 241         spin_lock(&nilfs->ns_last_segment_lock);
 242         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 243         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 244         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 245         spin_unlock(&nilfs->ns_last_segment_lock);
 246 }
 247
 248 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 249                                                int flip)
 250 {
 251         struct the_nilfs *nilfs = sb->s_fs_info;
 252         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 253
 254         /* nilfs->ns_sem must be locked by the caller. */
 255         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 256                 if (sbp[1] &&
 257                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 258                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 259                 } else {
 260                         nilfs_msg(sb, KERN_CRIT, "superblock broke");
 261                         return NULL;
 262                 }
 263         } else if (sbp[1] &&
 264                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 265                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 266         }
 267
 268         if (flip && sbp[1])
 269                 nilfs_swap_super_block(nilfs);
 270
 271         return sbp;
 272 }
 273
 274 int nilfs_commit_super(struct super_block *sb, int flag)
 275 {
 276         struct the_nilfs *nilfs = sb->s_fs_info;
 277         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 278         time64_t t;
 279
 280         /* nilfs->ns_sem must be locked by the caller. */
 281         t = ktime_get_real_seconds();
 282         nilfs->ns_sbwtime = t;
 283         sbp[0]->s_wtime = cpu_to_le64(t);
 284         sbp[0]->s_sum = 0;
 285         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 286                                              (unsigned char *)sbp[0],
 287                                              nilfs->ns_sbsize));
 288         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 289                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 290                 sbp[1]->s_sum = 0;
 291                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 292                                             (unsigned char *)sbp[1],
 293                                             nilfs->ns_sbsize));
 294         }
 295         clear_nilfs_sb_dirty(nilfs);
 296         nilfs->ns_flushed_device = 1;
 297         /* make sure store to ns_flushed_device cannot be reordered */
 298         smp_wmb();
 299         return nilfs_sync_super(sb, flag);
 300 }
 301
 302 /**
 303  * nilfs_cleanup_super() - write filesystem state for cleanup
 304  * @sb: super block instance to be unmounted or degraded to read-only
 305  *
 306  * This function restores state flags in the on-disk super block.
 307  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 308  * filesystem was not clean previously.
 309  */
 310 int nilfs_cleanup_super(struct super_block *sb)
 311 {
 312         struct the_nilfs *nilfs = sb->s_fs_info;
 313         struct nilfs_super_block **sbp;
 314         int flag = NILFS_SB_COMMIT;
 315         int ret = -EIO;
 316
 317         sbp = nilfs_prepare_super(sb, 0);
 318         if (sbp) {
 319                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 320                 nilfs_set_log_cursor(sbp[0], nilfs);
 321                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 322                         /*
 323                          * make the "clean" flag also to the opposite
 324                          * super block if both super blocks point to
 325                          * the same checkpoint.
 326                          */
 327                         sbp[1]->s_state = sbp[0]->s_state;
 328                         flag = NILFS_SB_COMMIT_ALL;
 329                 }
 330                 ret = nilfs_commit_super(sb, flag);
 331         }
 332         return ret;
 333 }
 334
 335 /**
 336  * nilfs_move_2nd_super - relocate secondary super block
 337  * @sb: super block instance
 338  * @sb2off: new offset of the secondary super block (in bytes)
 339  */
 340 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 341 {
 342         struct the_nilfs *nilfs = sb->s_fs_info;
 343         struct buffer_head *nsbh;
 344         struct nilfs_super_block *nsbp;
 345         sector_t blocknr, newblocknr;
 346         unsigned long offset;
 347         int sb2i;  /* array index of the secondary superblock */
 348         int ret = 0;
 349
 350         /* nilfs->ns_sem must be locked by the caller. */
 351         if (nilfs->ns_sbh[1] &&
 352             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 353                 sb2i = 1;
 354                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 355         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 356                 sb2i = 0;
 357                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 358         } else {
 359                 sb2i = -1;
 360                 blocknr = 0;
 361         }
 362         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 363                 goto out;  /* super block location is unchanged */
 364
 365         /* Get new super block buffer */
 366         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 367         offset = sb2off & (nilfs->ns_blocksize - 1);
 368         nsbh = sb_getblk(sb, newblocknr);
 369         if (!nsbh) {
 370                 nilfs_msg(sb, KERN_WARNING,
 371                           "unable to move secondary superblock to block %llu",
 372                           (unsigned long long)newblocknr);
 373                 ret = -EIO;
 374                 goto out;
 375         }
 376         nsbp = (void *)nsbh->b_data + offset;
 377         memset(nsbp, 0, nilfs->ns_blocksize);
 378
 379         if (sb2i >= 0) {
 380                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 381                 brelse(nilfs->ns_sbh[sb2i]);
 382                 nilfs->ns_sbh[sb2i] = nsbh;
 383                 nilfs->ns_sbp[sb2i] = nsbp;
 384         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 385                 /* secondary super block will be restored to index 1 */
 386                 nilfs->ns_sbh[1] = nsbh;
 387                 nilfs->ns_sbp[1] = nsbp;
 388         } else {
 389                 brelse(nsbh);
 390         }
 391 out:
 392         return ret;
 393 }
 394
 395 /**
 396  * nilfs_resize_fs - resize the filesystem
 397  * @sb: super block instance
 398  * @newsize: new size of the filesystem (in bytes)
 399  */
 400 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 401 {
 402         struct the_nilfs *nilfs = sb->s_fs_info;
 403         struct nilfs_super_block **sbp;
 404         __u64 devsize, newnsegs;
 405         loff_t sb2off;
 406         int ret;
 407
 408         ret = -ERANGE;
 409         devsize = i_size_read(sb->s_bdev->bd_inode);
 410         if (newsize > devsize)
 411                 goto out;
 412
 413         /*
 414          * Write lock is required to protect some functions depending
 415          * on the number of segments, the number of reserved segments,
 416          * and so forth.
 417          */
 418         down_write(&nilfs->ns_segctor_sem);
 419
 420         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 421         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 422         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 423
 424         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 425         up_write(&nilfs->ns_segctor_sem);
 426         if (ret < 0)
 427                 goto out;
 428
 429         ret = nilfs_construct_segment(sb);
 430         if (ret < 0)
 431                 goto out;
 432
 433         down_write(&nilfs->ns_sem);
 434         nilfs_move_2nd_super(sb, sb2off);
 435         ret = -EIO;
 436         sbp = nilfs_prepare_super(sb, 0);
 437         if (likely(sbp)) {
 438                 nilfs_set_log_cursor(sbp[0], nilfs);
 439                 /*
 440                  * Drop NILFS_RESIZE_FS flag for compatibility with
 441                  * mount-time resize which may be implemented in a
 442                  * future release.
 443                  */
 444                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 445                                               ~NILFS_RESIZE_FS);
 446                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 447                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 448                 if (sbp[1])
 449                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 450                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 451         }
 452         up_write(&nilfs->ns_sem);
 453
 454         /*
 455          * Reset the range of allocatable segments last.  This order
 456          * is important in the case of expansion because the secondary
 457          * superblock must be protected from log write until migration
 458          * completes.
 459          */
 460         if (!ret)
 461                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 462 out:
 463         return ret;
 464 }
 465
 466 static void nilfs_put_super(struct super_block *sb)
 467 {
 468         struct the_nilfs *nilfs = sb->s_fs_info;
 469
 470         nilfs_detach_log_writer(sb);
 471
 472         if (!sb_rdonly(sb)) {
 473                 down_write(&nilfs->ns_sem);
 474                 nilfs_cleanup_super(sb);
 475                 up_write(&nilfs->ns_sem);
 476         }
 477
 478         iput(nilfs->ns_sufile);
 479         iput(nilfs->ns_cpfile);
 480         iput(nilfs->ns_dat);
 481
 482         destroy_nilfs(nilfs);
 483         sb->s_fs_info = NULL;
 484 }
 485
 486 static int nilfs_sync_fs(struct super_block *sb, int wait)
 487 {
 488         struct the_nilfs *nilfs = sb->s_fs_info;
 489         struct nilfs_super_block **sbp;
 490         int err = 0;
 491
 492         /* This function is called when super block should be written back */
 493         if (wait)
 494                 err = nilfs_construct_segment(sb);
 495
 496         down_write(&nilfs->ns_sem);
 497         if (nilfs_sb_dirty(nilfs)) {
 498                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 499                 if (likely(sbp)) {
 500                         nilfs_set_log_cursor(sbp[0], nilfs);
 501                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 502                 }
 503         }
 504         up_write(&nilfs->ns_sem);
 505
 506         if (!err)
 507                 err = nilfs_flush_device(nilfs);
 508
 509         return err;
 510 }
 511
 512 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 513                             struct nilfs_root **rootp)
 514 {
 515         struct the_nilfs *nilfs = sb->s_fs_info;
 516         struct nilfs_root *root;
 517         struct nilfs_checkpoint *raw_cp;
 518         struct buffer_head *bh_cp;
 519         int err = -ENOMEM;
 520
 521         root = nilfs_find_or_create_root(
 522                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 523         if (!root)
 524                 return err;
 525
 526         if (root->ifile)
 527                 goto reuse; /* already attached checkpoint */
 528
 529         down_read(&nilfs->ns_segctor_sem);
 530         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 531                                           &bh_cp);
 532         up_read(&nilfs->ns_segctor_sem);
 533         if (unlikely(err)) {
 534                 if (err == -ENOENT || err == -EINVAL) {
 535                         nilfs_msg(sb, KERN_ERR,
 536                                   "Invalid checkpoint (checkpoint number=%llu)",
 537                                   (unsigned long long)cno);
 538                         err = -EINVAL;
 539                 }
 540                 goto failed;
 541         }
 542
 543         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 544                                &raw_cp->cp_ifile_inode, &root->ifile);
 545         if (err)
 546                 goto failed_bh;
 547
 548         atomic64_set(&root->inodes_count,
 549                         le64_to_cpu(raw_cp->cp_inodes_count));
 550         atomic64_set(&root->blocks_count,
 551                         le64_to_cpu(raw_cp->cp_blocks_count));
 552
 553         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 554
 555  reuse:
 556         *rootp = root;
 557         return 0;
 558
 559  failed_bh:
 560         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 561  failed:
 562         nilfs_put_root(root);
 563
 564         return err;
 565 }
 566
 567 static int nilfs_freeze(struct super_block *sb)
 568 {
 569         struct the_nilfs *nilfs = sb->s_fs_info;
 570         int err;
 571
 572         if (sb_rdonly(sb))
 573                 return 0;
 574
 575         /* Mark super block clean */
 576         down_write(&nilfs->ns_sem);
 577         err = nilfs_cleanup_super(sb);
 578         up_write(&nilfs->ns_sem);
 579         return err;
 580 }
 581
 582 static int nilfs_unfreeze(struct super_block *sb)
 583 {
 584         struct the_nilfs *nilfs = sb->s_fs_info;
 585
 586         if (sb_rdonly(sb))
 587                 return 0;
 588
 589         down_write(&nilfs->ns_sem);
 590         nilfs_setup_super(sb, false);
 591         up_write(&nilfs->ns_sem);
 592         return 0;
 593 }
 594
 595 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 596 {
 597         struct super_block *sb = dentry->d_sb;
 598         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 599         struct the_nilfs *nilfs = root->nilfs;
 600         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 601         unsigned long long blocks;
 602         unsigned long overhead;
 603         unsigned long nrsvblocks;
 604         sector_t nfreeblocks;
 605         u64 nmaxinodes, nfreeinodes;
 606         int err;
 607
 608         /*
 609          * Compute all of the segment blocks
 610          *
 611          * The blocks before first segment and after last segment
 612          * are excluded.
 613          */
 614         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 615                 - nilfs->ns_first_data_block;
 616         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 617
 618         /*
 619          * Compute the overhead
 620          *
 621          * When distributing meta data blocks outside segment structure,
 622          * We must count them as the overhead.
 623          */
 624         overhead = 0;
 625
 626         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 627         if (unlikely(err))
 628                 return err;
 629
 630         err = nilfs_ifile_count_free_inodes(root->ifile,
 631                                             &nmaxinodes, &nfreeinodes);
 632         if (unlikely(err)) {
 633                 nilfs_msg(sb, KERN_WARNING,
 634                           "failed to count free inodes: err=%d", err);
 635                 if (err == -ERANGE) {
 636                         /*
 637                          * If nilfs_palloc_count_max_entries() returns
 638                          * -ERANGE error code then we simply treat
 639                          * curent inodes count as maximum possible and
 640                          * zero as free inodes value.
 641                          */
 642                         nmaxinodes = atomic64_read(&root->inodes_count);
 643                         nfreeinodes = 0;
 644                         err = 0;
 645                 } else
 646                         return err;
 647         }
 648
 649         buf->f_type = NILFS_SUPER_MAGIC;
 650         buf->f_bsize = sb->s_blocksize;
 651         buf->f_blocks = blocks - overhead;
 652         buf->f_bfree = nfreeblocks;
 653         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 654                 (buf->f_bfree - nrsvblocks) : 0;
 655         buf->f_files = nmaxinodes;
 656         buf->f_ffree = nfreeinodes;
 657         buf->f_namelen = NILFS_NAME_LEN;
 658         buf->f_fsid.val[0] = (u32)id;
 659         buf->f_fsid.val[1] = (u32)(id >> 32);
 660
 661         return 0;
 662 }
 663
 664 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 665 {
 666         struct super_block *sb = dentry->d_sb;
 667         struct the_nilfs *nilfs = sb->s_fs_info;
 668         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 669
 670         if (!nilfs_test_opt(nilfs, BARRIER))
 671                 seq_puts(seq, ",nobarrier");
 672         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 673                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 674         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 675                 seq_puts(seq, ",errors=panic");
 676         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 677                 seq_puts(seq, ",errors=continue");
 678         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 679                 seq_puts(seq, ",order=strict");
 680         if (nilfs_test_opt(nilfs, NORECOVERY))
 681                 seq_puts(seq, ",norecovery");
 682         if (nilfs_test_opt(nilfs, DISCARD))
 683                 seq_puts(seq, ",discard");
 684
 685         return 0;
 686 }
 687
 688 static const struct super_operations nilfs_sops = {
 689         .alloc_inode    = nilfs_alloc_inode,
 690         .destroy_inode  = nilfs_destroy_inode,
 691         .dirty_inode    = nilfs_dirty_inode,
 692         .evict_inode    = nilfs_evict_inode,
 693         .put_super      = nilfs_put_super,
 694         .sync_fs        = nilfs_sync_fs,
 695         .freeze_fs      = nilfs_freeze,
 696         .unfreeze_fs    = nilfs_unfreeze,
 697         .statfs         = nilfs_statfs,
 698         .remount_fs     = nilfs_remount,
 699         .show_options = nilfs_show_options
 700 };
 701
 702 enum {
 703         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 704         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 705         Opt_discard, Opt_nodiscard, Opt_err,
 706 };
 707
 708 static match_table_t tokens = {
 709         {Opt_err_cont, "errors=continue"},
 710         {Opt_err_panic, "errors=panic"},
 711         {Opt_err_ro, "errors=remount-ro"},
 712         {Opt_barrier, "barrier"},
 713         {Opt_nobarrier, "nobarrier"},
 714         {Opt_snapshot, "cp=%u"},
 715         {Opt_order, "order=%s"},
 716         {Opt_norecovery, "norecovery"},
 717         {Opt_discard, "discard"},
 718         {Opt_nodiscard, "nodiscard"},
 719         {Opt_err, NULL}
 720 };
 721
 722 static int parse_options(char *options, struct super_block *sb, int is_remount)
 723 {
 724         struct the_nilfs *nilfs = sb->s_fs_info;
 725         char *p;
 726         substring_t args[MAX_OPT_ARGS];
 727
 728         if (!options)
 729                 return 1;
 730
 731         while ((p = strsep(&options, ",")) != NULL) {
 732                 int token;
 733
 734                 if (!*p)
 735                         continue;
 736
 737                 token = match_token(p, tokens, args);
 738                 switch (token) {
 739                 case Opt_barrier:
 740                         nilfs_set_opt(nilfs, BARRIER);
 741                         break;
 742                 case Opt_nobarrier:
 743                         nilfs_clear_opt(nilfs, BARRIER);
 744                         break;
 745                 case Opt_order:
 746                         if (strcmp(args[0].from, "relaxed") == 0)
 747                                 /* Ordered data semantics */
 748                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 749                         else if (strcmp(args[0].from, "strict") == 0)
 750                                 /* Strict in-order semantics */
 751                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 752                         else
 753                                 return 0;
 754                         break;
 755                 case Opt_err_panic:
 756                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 757                         break;
 758                 case Opt_err_ro:
 759                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 760                         break;
 761                 case Opt_err_cont:
 762                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 763                         break;
 764                 case Opt_snapshot:
 765                         if (is_remount) {
 766                                 nilfs_msg(sb, KERN_ERR,
 767                                           "\"%s\" option is invalid for remount",
 768                                           p);
 769                                 return 0;
 770                         }
 771                         break;
 772                 case Opt_norecovery:
 773                         nilfs_set_opt(nilfs, NORECOVERY);
 774                         break;
 775                 case Opt_discard:
 776                         nilfs_set_opt(nilfs, DISCARD);
 777                         break;
 778                 case Opt_nodiscard:
 779                         nilfs_clear_opt(nilfs, DISCARD);
 780                         break;
 781                 default:
 782                         nilfs_msg(sb, KERN_ERR,
 783                                   "unrecognized mount option \"%s\"", p);
 784                         return 0;
 785                 }
 786         }
 787         return 1;
 788 }
 789
 790 static inline void
 791 nilfs_set_default_options(struct super_block *sb,
 792                           struct nilfs_super_block *sbp)
 793 {
 794         struct the_nilfs *nilfs = sb->s_fs_info;
 795
 796         nilfs->ns_mount_opt =
 797                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 798 }
 799
 800 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 801 {
 802         struct the_nilfs *nilfs = sb->s_fs_info;
 803         struct nilfs_super_block **sbp;
 804         int max_mnt_count;
 805         int mnt_count;
 806
 807         /* nilfs->ns_sem must be locked by the caller. */
 808         sbp = nilfs_prepare_super(sb, 0);
 809         if (!sbp)
 810                 return -EIO;
 811
 812         if (!is_mount)
 813                 goto skip_mount_setup;
 814
 815         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 816         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 817
 818         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 819                 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
 820 #if 0
 821         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 822                 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
 823 #endif
 824         }
 825         if (!max_mnt_count)
 826                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 827
 828         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 829         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 830
 831 skip_mount_setup:
 832         sbp[0]->s_state =
 833                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 834         /* synchronize sbp[1] with sbp[0] */
 835         if (sbp[1])
 836                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 837         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 838 }
 839
 840 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 841                                                  u64 pos, int blocksize,
 842                                                  struct buffer_head **pbh)
 843 {
 844         unsigned long long sb_index = pos;
 845         unsigned long offset;
 846
 847         offset = do_div(sb_index, blocksize);
 848         *pbh = sb_bread(sb, sb_index);
 849         if (!*pbh)
 850                 return NULL;
 851         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 852 }
 853
 854 int nilfs_store_magic_and_option(struct super_block *sb,
 855                                  struct nilfs_super_block *sbp,
 856                                  char *data)
 857 {
 858         struct the_nilfs *nilfs = sb->s_fs_info;
 859
 860         sb->s_magic = le16_to_cpu(sbp->s_magic);
 861
 862         /* FS independent flags */
 863 #ifdef NILFS_ATIME_DISABLE
 864         sb->s_flags |= SB_NOATIME;
 865 #endif
 866
 867         nilfs_set_default_options(sb, sbp);
 868
 869         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 870         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 871         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 872         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 873
 874         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 875 }
 876
 877 int nilfs_check_feature_compatibility(struct super_block *sb,
 878                                       struct nilfs_super_block *sbp)
 879 {
 880         __u64 features;
 881
 882         features = le64_to_cpu(sbp->s_feature_incompat) &
 883                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 884         if (features) {
 885                 nilfs_msg(sb, KERN_ERR,
 886                           "couldn't mount because of unsupported optional features (%llx)",
 887                           (unsigned long long)features);
 888                 return -EINVAL;
 889         }
 890         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 891                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 892         if (!sb_rdonly(sb) && features) {
 893                 nilfs_msg(sb, KERN_ERR,
 894                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 895                           (unsigned long long)features);
 896                 return -EINVAL;
 897         }
 898         return 0;
 899 }
 900
 901 static int nilfs_get_root_dentry(struct super_block *sb,
 902                                  struct nilfs_root *root,
 903                                  struct dentry **root_dentry)
 904 {
 905         struct inode *inode;
 906         struct dentry *dentry;
 907         int ret = 0;
 908
 909         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 910         if (IS_ERR(inode)) {
 911                 ret = PTR_ERR(inode);
 912                 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
 913                 goto out;
 914         }
 915         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 916                 iput(inode);
 917                 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
 918                 ret = -EINVAL;
 919                 goto out;
 920         }
 921
 922         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 923                 dentry = d_find_alias(inode);
 924                 if (!dentry) {
 925                         dentry = d_make_root(inode);
 926                         if (!dentry) {
 927                                 ret = -ENOMEM;
 928                                 goto failed_dentry;
 929                         }
 930                 } else {
 931                         iput(inode);
 932                 }
 933         } else {
 934                 dentry = d_obtain_root(inode);
 935                 if (IS_ERR(dentry)) {
 936                         ret = PTR_ERR(dentry);
 937                         goto failed_dentry;
 938                 }
 939         }
 940         *root_dentry = dentry;
 941  out:
 942         return ret;
 943
 944  failed_dentry:
 945         nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
 946         goto out;
 947 }
 948
 949 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 950                                  struct dentry **root_dentry)
 951 {
 952         struct the_nilfs *nilfs = s->s_fs_info;
 953         struct nilfs_root *root;
 954         int ret;
 955
 956         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 957
 958         down_read(&nilfs->ns_segctor_sem);
 959         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 960         up_read(&nilfs->ns_segctor_sem);
 961         if (ret < 0) {
 962                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 963                 goto out;
 964         } else if (!ret) {
 965                 nilfs_msg(s, KERN_ERR,
 966                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 967                           (unsigned long long)cno);
 968                 ret = -EINVAL;
 969                 goto out;
 970         }
 971
 972         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 973         if (ret) {
 974                 nilfs_msg(s, KERN_ERR,
 975                           "error %d while loading snapshot (checkpoint number=%llu)",
 976                           ret, (unsigned long long)cno);
 977                 goto out;
 978         }
 979         ret = nilfs_get_root_dentry(s, root, root_dentry);
 980         nilfs_put_root(root);
 981  out:
 982         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 983         return ret;
 984 }
 985
 986 /**
 987  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 988  * @root_dentry: root dentry of the tree to be shrunk
 989  *
 990  * This function returns true if the tree was in-use.
 991  */
 992 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
 993 {
 994         shrink_dcache_parent(root_dentry);
 995         return d_count(root_dentry) > 1;
 996 }
 997
 998 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
 999 {
1000         struct the_nilfs *nilfs = sb->s_fs_info;
1001         struct nilfs_root *root;
1002         struct inode *inode;
1003         struct dentry *dentry;
1004         int ret;
1005
1006         if (cno > nilfs->ns_cno)
1007                 return false;
1008
1009         if (cno >= nilfs_last_cno(nilfs))
1010                 return true;    /* protect recent checkpoints */
1011
1012         ret = false;
1013         root = nilfs_lookup_root(nilfs, cno);
1014         if (root) {
1015                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1016                 if (inode) {
1017                         dentry = d_find_alias(inode);
1018                         if (dentry) {
1019                                 ret = nilfs_tree_is_busy(dentry);
1020                                 dput(dentry);
1021                         }
1022                         iput(inode);
1023                 }
1024                 nilfs_put_root(root);
1025         }
1026         return ret;
1027 }
1028
1029 /**
1030  * nilfs_fill_super() - initialize a super block instance
1031  * @sb: super_block
1032  * @data: mount options
1033  * @silent: silent mode flag
1034  *
1035  * This function is called exclusively by nilfs->ns_mount_mutex.
1036  * So, the recovery process is protected from other simultaneous mounts.
1037  */
1038 static int
1039 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1040 {
1041         struct the_nilfs *nilfs;
1042         struct nilfs_root *fsroot;
1043         __u64 cno;
1044         int err;
1045
1046         nilfs = alloc_nilfs(sb);
1047         if (!nilfs)
1048                 return -ENOMEM;
1049
1050         sb->s_fs_info = nilfs;
1051
1052         err = init_nilfs(nilfs, sb, (char *)data);
1053         if (err)
1054                 goto failed_nilfs;
1055
1056         sb->s_op = &nilfs_sops;
1057         sb->s_export_op = &nilfs_export_ops;
1058         sb->s_root = NULL;
1059         sb->s_time_gran = 1;
1060         sb->s_max_links = NILFS_LINK_MAX;
1061
1062         sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1063
1064         err = load_nilfs(nilfs, sb);
1065         if (err)
1066                 goto failed_nilfs;
1067
1068         cno = nilfs_last_cno(nilfs);
1069         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1070         if (err) {
1071                 nilfs_msg(sb, KERN_ERR,
1072                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1073                           err, (unsigned long long)cno);
1074                 goto failed_unload;
1075         }
1076
1077         if (!sb_rdonly(sb)) {
1078                 err = nilfs_attach_log_writer(sb, fsroot);
1079                 if (err)
1080                         goto failed_checkpoint;
1081         }
1082
1083         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1084         if (err)
1085                 goto failed_segctor;
1086
1087         nilfs_put_root(fsroot);
1088
1089         if (!sb_rdonly(sb)) {
1090                 down_write(&nilfs->ns_sem);
1091                 nilfs_setup_super(sb, true);
1092                 up_write(&nilfs->ns_sem);
1093         }
1094
1095         return 0;
1096
1097  failed_segctor:
1098         nilfs_detach_log_writer(sb);
1099
1100  failed_checkpoint:
1101         nilfs_put_root(fsroot);
1102
1103  failed_unload:
1104         iput(nilfs->ns_sufile);
1105         iput(nilfs->ns_cpfile);
1106         iput(nilfs->ns_dat);
1107
1108  failed_nilfs:
1109         destroy_nilfs(nilfs);
1110         return err;
1111 }
1112
1113 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1114 {
1115         struct the_nilfs *nilfs = sb->s_fs_info;
1116         unsigned long old_sb_flags;
1117         unsigned long old_mount_opt;
1118         int err;
1119
1120         sync_filesystem(sb);
1121         old_sb_flags = sb->s_flags;
1122         old_mount_opt = nilfs->ns_mount_opt;
1123
1124         if (!parse_options(data, sb, 1)) {
1125                 err = -EINVAL;
1126                 goto restore_opts;
1127         }
1128         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1129
1130         err = -EINVAL;
1131
1132         if (!nilfs_valid_fs(nilfs)) {
1133                 nilfs_msg(sb, KERN_WARNING,
1134                           "couldn't remount because the filesystem is in an incomplete recovery state");
1135                 goto restore_opts;
1136         }
1137
1138         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1139                 goto out;
1140         if (*flags & SB_RDONLY) {
1141                 /* Shutting down log writer */
1142                 nilfs_detach_log_writer(sb);
1143                 sb->s_flags |= SB_RDONLY;
1144
1145                 /*
1146                  * Remounting a valid RW partition RDONLY, so set
1147                  * the RDONLY flag and then mark the partition as valid again.
1148                  */
1149                 down_write(&nilfs->ns_sem);
1150                 nilfs_cleanup_super(sb);
1151                 up_write(&nilfs->ns_sem);
1152         } else {
1153                 __u64 features;
1154                 struct nilfs_root *root;
1155
1156                 /*
1157                  * Mounting a RDONLY partition read-write, so reread and
1158                  * store the current valid flag.  (It may have been changed
1159                  * by fsck since we originally mounted the partition.)
1160                  */
1161                 down_read(&nilfs->ns_sem);
1162                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1163                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1164                 up_read(&nilfs->ns_sem);
1165                 if (features) {
1166                         nilfs_msg(sb, KERN_WARNING,
1167                                   "couldn't remount RDWR because of unsupported optional features (%llx)",
1168                                   (unsigned long long)features);
1169                         err = -EROFS;
1170                         goto restore_opts;
1171                 }
1172
1173                 sb->s_flags &= ~SB_RDONLY;
1174
1175                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1176                 err = nilfs_attach_log_writer(sb, root);
1177                 if (err)
1178                         goto restore_opts;
1179
1180                 down_write(&nilfs->ns_sem);
1181                 nilfs_setup_super(sb, true);
1182                 up_write(&nilfs->ns_sem);
1183         }
1184  out:
1185         return 0;
1186
1187  restore_opts:
1188         sb->s_flags = old_sb_flags;
1189         nilfs->ns_mount_opt = old_mount_opt;
1190         return err;
1191 }
1192
1193 struct nilfs_super_data {
1194         struct block_device *bdev;
1195         __u64 cno;
1196         int flags;
1197 };
1198
1199 static int nilfs_parse_snapshot_option(const char *option,
1200                                        const substring_t *arg,
1201                                        struct nilfs_super_data *sd)
1202 {
1203         unsigned long long val;
1204         const char *msg = NULL;
1205         int err;
1206
1207         if (!(sd->flags & SB_RDONLY)) {
1208                 msg = "read-only option is not specified";
1209                 goto parse_error;
1210         }
1211
1212         err = kstrtoull(arg->from, 0, &val);
1213         if (err) {
1214                 if (err == -ERANGE)
1215                         msg = "too large checkpoint number";
1216                 else
1217                         msg = "malformed argument";
1218                 goto parse_error;
1219         } else if (val == 0) {
1220                 msg = "invalid checkpoint number 0";
1221                 goto parse_error;
1222         }
1223         sd->cno = val;
1224         return 0;
1225
1226 parse_error:
1227         nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1228         return 1;
1229 }
1230
1231 /**
1232  * nilfs_identify - pre-read mount options needed to identify mount instance
1233  * @data: mount options
1234  * @sd: nilfs_super_data
1235  */
1236 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1237 {
1238         char *p, *options = data;
1239         substring_t args[MAX_OPT_ARGS];
1240         int token;
1241         int ret = 0;
1242
1243         do {
1244                 p = strsep(&options, ",");
1245                 if (p != NULL && *p) {
1246                         token = match_token(p, tokens, args);
1247                         if (token == Opt_snapshot)
1248                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1249                                                                   sd);
1250                 }
1251                 if (!options)
1252                         break;
1253                 BUG_ON(options == data);
1254                 *(options - 1) = ',';
1255         } while (!ret);
1256         return ret;
1257 }
1258
1259 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1260 {
1261         s->s_bdev = data;
1262         s->s_dev = s->s_bdev->bd_dev;
1263         return 0;
1264 }
1265
1266 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1267 {
1268         return (void *)s->s_bdev == data;
1269 }
1270
1271 static struct dentry *
1272 nilfs_mount(struct file_system_type *fs_type, int flags,
1273              const char *dev_name, void *data)
1274 {
1275         struct nilfs_super_data sd;
1276         struct super_block *s;
1277         fmode_t mode = FMODE_READ | FMODE_EXCL;
1278         struct dentry *root_dentry;
1279         int err, s_new = false;
1280
1281         if (!(flags & SB_RDONLY))
1282                 mode |= FMODE_WRITE;
1283
1284         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1285         if (IS_ERR(sd.bdev))
1286                 return ERR_CAST(sd.bdev);
1287
1288         sd.cno = 0;
1289         sd.flags = flags;
1290         if (nilfs_identify((char *)data, &sd)) {
1291                 err = -EINVAL;
1292                 goto failed;
1293         }
1294
1295         /*
1296          * once the super is inserted into the list by sget, s_umount
1297          * will protect the lockfs code from trying to start a snapshot
1298          * while we are mounting
1299          */
1300         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1301         if (sd.bdev->bd_fsfreeze_count > 0) {
1302                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1303                 err = -EBUSY;
1304                 goto failed;
1305         }
1306         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1307                  sd.bdev);
1308         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1309         if (IS_ERR(s)) {
1310                 err = PTR_ERR(s);
1311                 goto failed;
1312         }
1313
1314         if (!s->s_root) {
1315                 s_new = true;
1316
1317                 /* New superblock instance created */
1318                 s->s_mode = mode;
1319                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1320                 sb_set_blocksize(s, block_size(sd.bdev));
1321
1322                 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1323                 if (err)
1324                         goto failed_super;
1325
1326                 s->s_flags |= SB_ACTIVE;
1327         } else if (!sd.cno) {
1328                 if (nilfs_tree_is_busy(s->s_root)) {
1329                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1330                                 nilfs_msg(s, KERN_ERR,
1331                                           "the device already has a %s mount.",
1332                                           sb_rdonly(s) ? "read-only" : "read/write");
1333                                 err = -EBUSY;
1334                                 goto failed_super;
1335                         }
1336                 } else {
1337                         /*
1338                          * Try remount to setup mount states if the current
1339                          * tree is not mounted and only snapshots use this sb.
1340                          */
1341                         err = nilfs_remount(s, &flags, data);
1342                         if (err)
1343                                 goto failed_super;
1344                 }
1345         }
1346
1347         if (sd.cno) {
1348                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1349                 if (err)
1350                         goto failed_super;
1351         } else {
1352                 root_dentry = dget(s->s_root);
1353         }
1354
1355         if (!s_new)
1356                 blkdev_put(sd.bdev, mode);
1357
1358         return root_dentry;
1359
1360  failed_super:
1361         deactivate_locked_super(s);
1362
1363  failed:
1364         if (!s_new)
1365                 blkdev_put(sd.bdev, mode);
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)