fs/nilfs2/super.c

   1 /*
   2  * super.c - NILFS module and super block management.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * Written by Ryusuke Konishi.
  17  */
  18 /*
  19  *  linux/fs/ext2/super.c
  20  *
  21  * Copyright (C) 1992, 1993, 1994, 1995
  22  * Remy Card (card@masi.ibp.fr)
  23  * Laboratoire MASI - Institut Blaise Pascal
  24  * Universite Pierre et Marie Curie (Paris VI)
  25  *
  26  *  from
  27  *
  28  *  linux/fs/minix/inode.c
  29  *
  30  *  Copyright (C) 1991, 1992  Linus Torvalds
  31  *
  32  *  Big-endian to little-endian byte-swapping/bitmaps by
  33  *        David S. Miller (davem@caip.rutgers.edu), 1995
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/slab.h>
  39 #include <linux/init.h>
  40 #include <linux/blkdev.h>
  41 #include <linux/parser.h>
  42 #include <linux/crc32.h>
  43 #include <linux/vfs.h>
  44 #include <linux/writeback.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mount.h>
  47 #include "nilfs.h"
  48 #include "export.h"
  49 #include "mdt.h"
  50 #include "alloc.h"
  51 #include "btree.h"
  52 #include "btnode.h"
  53 #include "page.h"
  54 #include "cpfile.h"
  55 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  56 #include "ifile.h"
  57 #include "dat.h"
  58 #include "segment.h"
  59 #include "segbuf.h"
  60
  61 MODULE_AUTHOR("NTT Corp.");
  62 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  63                    "(NILFS)");
  64 MODULE_LICENSE("GPL");
  65
  66 static struct kmem_cache *nilfs_inode_cachep;
  67 struct kmem_cache *nilfs_transaction_cachep;
  68 struct kmem_cache *nilfs_segbuf_cachep;
  69 struct kmem_cache *nilfs_btree_path_cache;
  70
  71 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  72 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  73
  74 static void nilfs_set_error(struct super_block *sb)
  75 {
  76         struct the_nilfs *nilfs = sb->s_fs_info;
  77         struct nilfs_super_block **sbp;
  78
  79         down_write(&nilfs->ns_sem);
  80         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  81                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  82                 sbp = nilfs_prepare_super(sb, 0);
  83                 if (likely(sbp)) {
  84                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  85                         if (sbp[1])
  86                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  87                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
  88                 }
  89         }
  90         up_write(&nilfs->ns_sem);
  91 }
  92
  93 /**
  94  * __nilfs_error() - report failure condition on a filesystem
  95  *
  96  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
  97  * reporting an error message.  This function should be called when
  98  * NILFS detects incoherences or defects of meta data on disk.
  99  *
 100  * This implements the body of nilfs_error() macro.  Normally,
 101  * nilfs_error() should be used.  As for sustainable errors such as a
 102  * single-shot I/O error, nilfs_warning() or printk() should be used
 103  * instead.
 104  *
 105  * Callers should not add a trailing newline since this will do it.
 106  */
 107 void __nilfs_error(struct super_block *sb, const char *function,
 108                    const char *fmt, ...)
 109 {
 110         struct the_nilfs *nilfs = sb->s_fs_info;
 111         struct va_format vaf;
 112         va_list args;
 113
 114         va_start(args, fmt);
 115
 116         vaf.fmt = fmt;
 117         vaf.va = &args;
 118
 119         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 120                sb->s_id, function, &vaf);
 121
 122         va_end(args);
 123
 124         if (!(sb->s_flags & MS_RDONLY)) {
 125                 nilfs_set_error(sb);
 126
 127                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 128                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 129                         sb->s_flags |= MS_RDONLY;
 130                 }
 131         }
 132
 133         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 134                 panic("NILFS (device %s): panic forced after error\n",
 135                       sb->s_id);
 136 }
 137
 138 void nilfs_warning(struct super_block *sb, const char *function,
 139                    const char *fmt, ...)
 140 {
 141         struct va_format vaf;
 142         va_list args;
 143
 144         va_start(args, fmt);
 145
 146         vaf.fmt = fmt;
 147         vaf.va = &args;
 148
 149         printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
 150                sb->s_id, function, &vaf);
 151
 152         va_end(args);
 153 }
 154
 155
 156 struct inode *nilfs_alloc_inode(struct super_block *sb)
 157 {
 158         struct nilfs_inode_info *ii;
 159
 160         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 161         if (!ii)
 162                 return NULL;
 163         ii->i_bh = NULL;
 164         ii->i_state = 0;
 165         ii->i_cno = 0;
 166         ii->vfs_inode.i_version = 1;
 167         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
 168         return &ii->vfs_inode;
 169 }
 170
 171 static void nilfs_i_callback(struct rcu_head *head)
 172 {
 173         struct inode *inode = container_of(head, struct inode, i_rcu);
 174
 175         if (nilfs_is_metadata_file_inode(inode))
 176                 nilfs_mdt_destroy(inode);
 177
 178         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 179 }
 180
 181 void nilfs_destroy_inode(struct inode *inode)
 182 {
 183         call_rcu(&inode->i_rcu, nilfs_i_callback);
 184 }
 185
 186 static int nilfs_sync_super(struct super_block *sb, int flag)
 187 {
 188         struct the_nilfs *nilfs = sb->s_fs_info;
 189         int err;
 190
 191  retry:
 192         set_buffer_dirty(nilfs->ns_sbh[0]);
 193         if (nilfs_test_opt(nilfs, BARRIER)) {
 194                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 195                                           WRITE_SYNC | WRITE_FLUSH_FUA);
 196         } else {
 197                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 198         }
 199
 200         if (unlikely(err)) {
 201                 printk(KERN_ERR
 202                        "NILFS: unable to write superblock (err=%d)\n", err);
 203                 if (err == -EIO && nilfs->ns_sbh[1]) {
 204                         /*
 205                          * sbp[0] points to newer log than sbp[1],
 206                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 207                          */
 208                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 209                                nilfs->ns_sbsize);
 210                         nilfs_fall_back_super_block(nilfs);
 211                         goto retry;
 212                 }
 213         } else {
 214                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 215
 216                 nilfs->ns_sbwcount++;
 217
 218                 /*
 219                  * The latest segment becomes trailable from the position
 220                  * written in superblock.
 221                  */
 222                 clear_nilfs_discontinued(nilfs);
 223
 224                 /* update GC protection for recent segments */
 225                 if (nilfs->ns_sbh[1]) {
 226                         if (flag == NILFS_SB_COMMIT_ALL) {
 227                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 228                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 229                                         goto out;
 230                         }
 231                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 232                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 233                                 sbp = nilfs->ns_sbp[1];
 234                 }
 235
 236                 spin_lock(&nilfs->ns_last_segment_lock);
 237                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 238                 spin_unlock(&nilfs->ns_last_segment_lock);
 239         }
 240  out:
 241         return err;
 242 }
 243
 244 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 245                           struct the_nilfs *nilfs)
 246 {
 247         sector_t nfreeblocks;
 248
 249         /* nilfs->ns_sem must be locked by the caller. */
 250         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 251         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 252
 253         spin_lock(&nilfs->ns_last_segment_lock);
 254         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 255         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 256         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 257         spin_unlock(&nilfs->ns_last_segment_lock);
 258 }
 259
 260 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 261                                                int flip)
 262 {
 263         struct the_nilfs *nilfs = sb->s_fs_info;
 264         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 265
 266         /* nilfs->ns_sem must be locked by the caller. */
 267         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 268                 if (sbp[1] &&
 269                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 270                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 271                 } else {
 272                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 273                                sb->s_id);
 274                         return NULL;
 275                 }
 276         } else if (sbp[1] &&
 277                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 278                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 279         }
 280
 281         if (flip && sbp[1])
 282                 nilfs_swap_super_block(nilfs);
 283
 284         return sbp;
 285 }
 286
 287 int nilfs_commit_super(struct super_block *sb, int flag)
 288 {
 289         struct the_nilfs *nilfs = sb->s_fs_info;
 290         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 291         time_t t;
 292
 293         /* nilfs->ns_sem must be locked by the caller. */
 294         t = get_seconds();
 295         nilfs->ns_sbwtime = t;
 296         sbp[0]->s_wtime = cpu_to_le64(t);
 297         sbp[0]->s_sum = 0;
 298         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 299                                              (unsigned char *)sbp[0],
 300                                              nilfs->ns_sbsize));
 301         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 302                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 303                 sbp[1]->s_sum = 0;
 304                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 305                                             (unsigned char *)sbp[1],
 306                                             nilfs->ns_sbsize));
 307         }
 308         clear_nilfs_sb_dirty(nilfs);
 309         nilfs->ns_flushed_device = 1;
 310         /* make sure store to ns_flushed_device cannot be reordered */
 311         smp_wmb();
 312         return nilfs_sync_super(sb, flag);
 313 }
 314
 315 /**
 316  * nilfs_cleanup_super() - write filesystem state for cleanup
 317  * @sb: super block instance to be unmounted or degraded to read-only
 318  *
 319  * This function restores state flags in the on-disk super block.
 320  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 321  * filesystem was not clean previously.
 322  */
 323 int nilfs_cleanup_super(struct super_block *sb)
 324 {
 325         struct the_nilfs *nilfs = sb->s_fs_info;
 326         struct nilfs_super_block **sbp;
 327         int flag = NILFS_SB_COMMIT;
 328         int ret = -EIO;
 329
 330         sbp = nilfs_prepare_super(sb, 0);
 331         if (sbp) {
 332                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 333                 nilfs_set_log_cursor(sbp[0], nilfs);
 334                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 335                         /*
 336                          * make the "clean" flag also to the opposite
 337                          * super block if both super blocks point to
 338                          * the same checkpoint.
 339                          */
 340                         sbp[1]->s_state = sbp[0]->s_state;
 341                         flag = NILFS_SB_COMMIT_ALL;
 342                 }
 343                 ret = nilfs_commit_super(sb, flag);
 344         }
 345         return ret;
 346 }
 347
 348 /**
 349  * nilfs_move_2nd_super - relocate secondary super block
 350  * @sb: super block instance
 351  * @sb2off: new offset of the secondary super block (in bytes)
 352  */
 353 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 354 {
 355         struct the_nilfs *nilfs = sb->s_fs_info;
 356         struct buffer_head *nsbh;
 357         struct nilfs_super_block *nsbp;
 358         sector_t blocknr, newblocknr;
 359         unsigned long offset;
 360         int sb2i;  /* array index of the secondary superblock */
 361         int ret = 0;
 362
 363         /* nilfs->ns_sem must be locked by the caller. */
 364         if (nilfs->ns_sbh[1] &&
 365             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 366                 sb2i = 1;
 367                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 368         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 369                 sb2i = 0;
 370                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 371         } else {
 372                 sb2i = -1;
 373                 blocknr = 0;
 374         }
 375         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 376                 goto out;  /* super block location is unchanged */
 377
 378         /* Get new super block buffer */
 379         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 380         offset = sb2off & (nilfs->ns_blocksize - 1);
 381         nsbh = sb_getblk(sb, newblocknr);
 382         if (!nsbh) {
 383                 printk(KERN_WARNING
 384                        "NILFS warning: unable to move secondary superblock "
 385                        "to block %llu\n", (unsigned long long)newblocknr);
 386                 ret = -EIO;
 387                 goto out;
 388         }
 389         nsbp = (void *)nsbh->b_data + offset;
 390         memset(nsbp, 0, nilfs->ns_blocksize);
 391
 392         if (sb2i >= 0) {
 393                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 394                 brelse(nilfs->ns_sbh[sb2i]);
 395                 nilfs->ns_sbh[sb2i] = nsbh;
 396                 nilfs->ns_sbp[sb2i] = nsbp;
 397         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 398                 /* secondary super block will be restored to index 1 */
 399                 nilfs->ns_sbh[1] = nsbh;
 400                 nilfs->ns_sbp[1] = nsbp;
 401         } else {
 402                 brelse(nsbh);
 403         }
 404 out:
 405         return ret;
 406 }
 407
 408 /**
 409  * nilfs_resize_fs - resize the filesystem
 410  * @sb: super block instance
 411  * @newsize: new size of the filesystem (in bytes)
 412  */
 413 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 414 {
 415         struct the_nilfs *nilfs = sb->s_fs_info;
 416         struct nilfs_super_block **sbp;
 417         __u64 devsize, newnsegs;
 418         loff_t sb2off;
 419         int ret;
 420
 421         ret = -ERANGE;
 422         devsize = i_size_read(sb->s_bdev->bd_inode);
 423         if (newsize > devsize)
 424                 goto out;
 425
 426         /*
 427          * Write lock is required to protect some functions depending
 428          * on the number of segments, the number of reserved segments,
 429          * and so forth.
 430          */
 431         down_write(&nilfs->ns_segctor_sem);
 432
 433         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 434         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 435         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 436
 437         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 438         up_write(&nilfs->ns_segctor_sem);
 439         if (ret < 0)
 440                 goto out;
 441
 442         ret = nilfs_construct_segment(sb);
 443         if (ret < 0)
 444                 goto out;
 445
 446         down_write(&nilfs->ns_sem);
 447         nilfs_move_2nd_super(sb, sb2off);
 448         ret = -EIO;
 449         sbp = nilfs_prepare_super(sb, 0);
 450         if (likely(sbp)) {
 451                 nilfs_set_log_cursor(sbp[0], nilfs);
 452                 /*
 453                  * Drop NILFS_RESIZE_FS flag for compatibility with
 454                  * mount-time resize which may be implemented in a
 455                  * future release.
 456                  */
 457                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 458                                               ~NILFS_RESIZE_FS);
 459                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 460                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 461                 if (sbp[1])
 462                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 463                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 464         }
 465         up_write(&nilfs->ns_sem);
 466
 467         /*
 468          * Reset the range of allocatable segments last.  This order
 469          * is important in the case of expansion because the secondary
 470          * superblock must be protected from log write until migration
 471          * completes.
 472          */
 473         if (!ret)
 474                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 475 out:
 476         return ret;
 477 }
 478
 479 static void nilfs_put_super(struct super_block *sb)
 480 {
 481         struct the_nilfs *nilfs = sb->s_fs_info;
 482
 483         nilfs_detach_log_writer(sb);
 484
 485         if (!(sb->s_flags & MS_RDONLY)) {
 486                 down_write(&nilfs->ns_sem);
 487                 nilfs_cleanup_super(sb);
 488                 up_write(&nilfs->ns_sem);
 489         }
 490
 491         iput(nilfs->ns_sufile);
 492         iput(nilfs->ns_cpfile);
 493         iput(nilfs->ns_dat);
 494
 495         destroy_nilfs(nilfs);
 496         sb->s_fs_info = NULL;
 497 }
 498
 499 static int nilfs_sync_fs(struct super_block *sb, int wait)
 500 {
 501         struct the_nilfs *nilfs = sb->s_fs_info;
 502         struct nilfs_super_block **sbp;
 503         int err = 0;
 504
 505         /* This function is called when super block should be written back */
 506         if (wait)
 507                 err = nilfs_construct_segment(sb);
 508
 509         down_write(&nilfs->ns_sem);
 510         if (nilfs_sb_dirty(nilfs)) {
 511                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 512                 if (likely(sbp)) {
 513                         nilfs_set_log_cursor(sbp[0], nilfs);
 514                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 515                 }
 516         }
 517         up_write(&nilfs->ns_sem);
 518
 519         if (!err)
 520                 err = nilfs_flush_device(nilfs);
 521
 522         return err;
 523 }
 524
 525 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 526                             struct nilfs_root **rootp)
 527 {
 528         struct the_nilfs *nilfs = sb->s_fs_info;
 529         struct nilfs_root *root;
 530         struct nilfs_checkpoint *raw_cp;
 531         struct buffer_head *bh_cp;
 532         int err = -ENOMEM;
 533
 534         root = nilfs_find_or_create_root(
 535                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 536         if (!root)
 537                 return err;
 538
 539         if (root->ifile)
 540                 goto reuse; /* already attached checkpoint */
 541
 542         down_read(&nilfs->ns_segctor_sem);
 543         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 544                                           &bh_cp);
 545         up_read(&nilfs->ns_segctor_sem);
 546         if (unlikely(err)) {
 547                 if (err == -ENOENT || err == -EINVAL) {
 548                         printk(KERN_ERR
 549                                "NILFS: Invalid checkpoint "
 550                                "(checkpoint number=%llu)\n",
 551                                (unsigned long long)cno);
 552                         err = -EINVAL;
 553                 }
 554                 goto failed;
 555         }
 556
 557         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 558                                &raw_cp->cp_ifile_inode, &root->ifile);
 559         if (err)
 560                 goto failed_bh;
 561
 562         atomic64_set(&root->inodes_count,
 563                         le64_to_cpu(raw_cp->cp_inodes_count));
 564         atomic64_set(&root->blocks_count,
 565                         le64_to_cpu(raw_cp->cp_blocks_count));
 566
 567         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 568
 569  reuse:
 570         *rootp = root;
 571         return 0;
 572
 573  failed_bh:
 574         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 575  failed:
 576         nilfs_put_root(root);
 577
 578         return err;
 579 }
 580
 581 static int nilfs_freeze(struct super_block *sb)
 582 {
 583         struct the_nilfs *nilfs = sb->s_fs_info;
 584         int err;
 585
 586         if (sb->s_flags & MS_RDONLY)
 587                 return 0;
 588
 589         /* Mark super block clean */
 590         down_write(&nilfs->ns_sem);
 591         err = nilfs_cleanup_super(sb);
 592         up_write(&nilfs->ns_sem);
 593         return err;
 594 }
 595
 596 static int nilfs_unfreeze(struct super_block *sb)
 597 {
 598         struct the_nilfs *nilfs = sb->s_fs_info;
 599
 600         if (sb->s_flags & MS_RDONLY)
 601                 return 0;
 602
 603         down_write(&nilfs->ns_sem);
 604         nilfs_setup_super(sb, false);
 605         up_write(&nilfs->ns_sem);
 606         return 0;
 607 }
 608
 609 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 610 {
 611         struct super_block *sb = dentry->d_sb;
 612         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 613         struct the_nilfs *nilfs = root->nilfs;
 614         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 615         unsigned long long blocks;
 616         unsigned long overhead;
 617         unsigned long nrsvblocks;
 618         sector_t nfreeblocks;
 619         u64 nmaxinodes, nfreeinodes;
 620         int err;
 621
 622         /*
 623          * Compute all of the segment blocks
 624          *
 625          * The blocks before first segment and after last segment
 626          * are excluded.
 627          */
 628         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 629                 - nilfs->ns_first_data_block;
 630         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 631
 632         /*
 633          * Compute the overhead
 634          *
 635          * When distributing meta data blocks outside segment structure,
 636          * We must count them as the overhead.
 637          */
 638         overhead = 0;
 639
 640         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 641         if (unlikely(err))
 642                 return err;
 643
 644         err = nilfs_ifile_count_free_inodes(root->ifile,
 645                                             &nmaxinodes, &nfreeinodes);
 646         if (unlikely(err)) {
 647                 printk(KERN_WARNING
 648                         "NILFS warning: fail to count free inodes: err %d.\n",
 649                         err);
 650                 if (err == -ERANGE) {
 651                         /*
 652                          * If nilfs_palloc_count_max_entries() returns
 653                          * -ERANGE error code then we simply treat
 654                          * curent inodes count as maximum possible and
 655                          * zero as free inodes value.
 656                          */
 657                         nmaxinodes = atomic64_read(&root->inodes_count);
 658                         nfreeinodes = 0;
 659                         err = 0;
 660                 } else
 661                         return err;
 662         }
 663
 664         buf->f_type = NILFS_SUPER_MAGIC;
 665         buf->f_bsize = sb->s_blocksize;
 666         buf->f_blocks = blocks - overhead;
 667         buf->f_bfree = nfreeblocks;
 668         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 669                 (buf->f_bfree - nrsvblocks) : 0;
 670         buf->f_files = nmaxinodes;
 671         buf->f_ffree = nfreeinodes;
 672         buf->f_namelen = NILFS_NAME_LEN;
 673         buf->f_fsid.val[0] = (u32)id;
 674         buf->f_fsid.val[1] = (u32)(id >> 32);
 675
 676         return 0;
 677 }
 678
 679 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 680 {
 681         struct super_block *sb = dentry->d_sb;
 682         struct the_nilfs *nilfs = sb->s_fs_info;
 683         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 684
 685         if (!nilfs_test_opt(nilfs, BARRIER))
 686                 seq_puts(seq, ",nobarrier");
 687         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 688                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 689         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 690                 seq_puts(seq, ",errors=panic");
 691         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 692                 seq_puts(seq, ",errors=continue");
 693         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 694                 seq_puts(seq, ",order=strict");
 695         if (nilfs_test_opt(nilfs, NORECOVERY))
 696                 seq_puts(seq, ",norecovery");
 697         if (nilfs_test_opt(nilfs, DISCARD))
 698                 seq_puts(seq, ",discard");
 699
 700         return 0;
 701 }
 702
 703 static const struct super_operations nilfs_sops = {
 704         .alloc_inode    = nilfs_alloc_inode,
 705         .destroy_inode  = nilfs_destroy_inode,
 706         .dirty_inode    = nilfs_dirty_inode,
 707         .evict_inode    = nilfs_evict_inode,
 708         .put_super      = nilfs_put_super,
 709         .sync_fs        = nilfs_sync_fs,
 710         .freeze_fs      = nilfs_freeze,
 711         .unfreeze_fs    = nilfs_unfreeze,
 712         .statfs         = nilfs_statfs,
 713         .remount_fs     = nilfs_remount,
 714         .show_options = nilfs_show_options
 715 };
 716
 717 enum {
 718         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 719         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 720         Opt_discard, Opt_nodiscard, Opt_err,
 721 };
 722
 723 static match_table_t tokens = {
 724         {Opt_err_cont, "errors=continue"},
 725         {Opt_err_panic, "errors=panic"},
 726         {Opt_err_ro, "errors=remount-ro"},
 727         {Opt_barrier, "barrier"},
 728         {Opt_nobarrier, "nobarrier"},
 729         {Opt_snapshot, "cp=%u"},
 730         {Opt_order, "order=%s"},
 731         {Opt_norecovery, "norecovery"},
 732         {Opt_discard, "discard"},
 733         {Opt_nodiscard, "nodiscard"},
 734         {Opt_err, NULL}
 735 };
 736
 737 static int parse_options(char *options, struct super_block *sb, int is_remount)
 738 {
 739         struct the_nilfs *nilfs = sb->s_fs_info;
 740         char *p;
 741         substring_t args[MAX_OPT_ARGS];
 742
 743         if (!options)
 744                 return 1;
 745
 746         while ((p = strsep(&options, ",")) != NULL) {
 747                 int token;
 748
 749                 if (!*p)
 750                         continue;
 751
 752                 token = match_token(p, tokens, args);
 753                 switch (token) {
 754                 case Opt_barrier:
 755                         nilfs_set_opt(nilfs, BARRIER);
 756                         break;
 757                 case Opt_nobarrier:
 758                         nilfs_clear_opt(nilfs, BARRIER);
 759                         break;
 760                 case Opt_order:
 761                         if (strcmp(args[0].from, "relaxed") == 0)
 762                                 /* Ordered data semantics */
 763                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 764                         else if (strcmp(args[0].from, "strict") == 0)
 765                                 /* Strict in-order semantics */
 766                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 767                         else
 768                                 return 0;
 769                         break;
 770                 case Opt_err_panic:
 771                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 772                         break;
 773                 case Opt_err_ro:
 774                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 775                         break;
 776                 case Opt_err_cont:
 777                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 778                         break;
 779                 case Opt_snapshot:
 780                         if (is_remount) {
 781                                 printk(KERN_ERR
 782                                        "NILFS: \"%s\" option is invalid "
 783                                        "for remount.\n", p);
 784                                 return 0;
 785                         }
 786                         break;
 787                 case Opt_norecovery:
 788                         nilfs_set_opt(nilfs, NORECOVERY);
 789                         break;
 790                 case Opt_discard:
 791                         nilfs_set_opt(nilfs, DISCARD);
 792                         break;
 793                 case Opt_nodiscard:
 794                         nilfs_clear_opt(nilfs, DISCARD);
 795                         break;
 796                 default:
 797                         printk(KERN_ERR
 798                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 799                         return 0;
 800                 }
 801         }
 802         return 1;
 803 }
 804
 805 static inline void
 806 nilfs_set_default_options(struct super_block *sb,
 807                           struct nilfs_super_block *sbp)
 808 {
 809         struct the_nilfs *nilfs = sb->s_fs_info;
 810
 811         nilfs->ns_mount_opt =
 812                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 813 }
 814
 815 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 816 {
 817         struct the_nilfs *nilfs = sb->s_fs_info;
 818         struct nilfs_super_block **sbp;
 819         int max_mnt_count;
 820         int mnt_count;
 821
 822         /* nilfs->ns_sem must be locked by the caller. */
 823         sbp = nilfs_prepare_super(sb, 0);
 824         if (!sbp)
 825                 return -EIO;
 826
 827         if (!is_mount)
 828                 goto skip_mount_setup;
 829
 830         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 831         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 832
 833         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 834                 printk(KERN_WARNING
 835                        "NILFS warning: mounting fs with errors\n");
 836 #if 0
 837         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 838                 printk(KERN_WARNING
 839                        "NILFS warning: maximal mount count reached\n");
 840 #endif
 841         }
 842         if (!max_mnt_count)
 843                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 844
 845         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 846         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 847
 848 skip_mount_setup:
 849         sbp[0]->s_state =
 850                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 851         /* synchronize sbp[1] with sbp[0] */
 852         if (sbp[1])
 853                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 854         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 855 }
 856
 857 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 858                                                  u64 pos, int blocksize,
 859                                                  struct buffer_head **pbh)
 860 {
 861         unsigned long long sb_index = pos;
 862         unsigned long offset;
 863
 864         offset = do_div(sb_index, blocksize);
 865         *pbh = sb_bread(sb, sb_index);
 866         if (!*pbh)
 867                 return NULL;
 868         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 869 }
 870
 871 int nilfs_store_magic_and_option(struct super_block *sb,
 872                                  struct nilfs_super_block *sbp,
 873                                  char *data)
 874 {
 875         struct the_nilfs *nilfs = sb->s_fs_info;
 876
 877         sb->s_magic = le16_to_cpu(sbp->s_magic);
 878
 879         /* FS independent flags */
 880 #ifdef NILFS_ATIME_DISABLE
 881         sb->s_flags |= MS_NOATIME;
 882 #endif
 883
 884         nilfs_set_default_options(sb, sbp);
 885
 886         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 887         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 888         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 889         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 890
 891         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 892 }
 893
 894 int nilfs_check_feature_compatibility(struct super_block *sb,
 895                                       struct nilfs_super_block *sbp)
 896 {
 897         __u64 features;
 898
 899         features = le64_to_cpu(sbp->s_feature_incompat) &
 900                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 901         if (features) {
 902                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 903                        "optional features (%llx)\n",
 904                        (unsigned long long)features);
 905                 return -EINVAL;
 906         }
 907         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 908                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 909         if (!(sb->s_flags & MS_RDONLY) && features) {
 910                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
 911                        "unsupported optional features (%llx)\n",
 912                        (unsigned long long)features);
 913                 return -EINVAL;
 914         }
 915         return 0;
 916 }
 917
 918 static int nilfs_get_root_dentry(struct super_block *sb,
 919                                  struct nilfs_root *root,
 920                                  struct dentry **root_dentry)
 921 {
 922         struct inode *inode;
 923         struct dentry *dentry;
 924         int ret = 0;
 925
 926         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 927         if (IS_ERR(inode)) {
 928                 printk(KERN_ERR "NILFS: get root inode failed\n");
 929                 ret = PTR_ERR(inode);
 930                 goto out;
 931         }
 932         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 933                 iput(inode);
 934                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 935                 ret = -EINVAL;
 936                 goto out;
 937         }
 938
 939         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 940                 dentry = d_find_alias(inode);
 941                 if (!dentry) {
 942                         dentry = d_make_root(inode);
 943                         if (!dentry) {
 944                                 ret = -ENOMEM;
 945                                 goto failed_dentry;
 946                         }
 947                 } else {
 948                         iput(inode);
 949                 }
 950         } else {
 951                 dentry = d_obtain_root(inode);
 952                 if (IS_ERR(dentry)) {
 953                         ret = PTR_ERR(dentry);
 954                         goto failed_dentry;
 955                 }
 956         }
 957         *root_dentry = dentry;
 958  out:
 959         return ret;
 960
 961  failed_dentry:
 962         printk(KERN_ERR "NILFS: get root dentry failed\n");
 963         goto out;
 964 }
 965
 966 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 967                                  struct dentry **root_dentry)
 968 {
 969         struct the_nilfs *nilfs = s->s_fs_info;
 970         struct nilfs_root *root;
 971         int ret;
 972
 973         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 974
 975         down_read(&nilfs->ns_segctor_sem);
 976         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 977         up_read(&nilfs->ns_segctor_sem);
 978         if (ret < 0) {
 979                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 980                 goto out;
 981         } else if (!ret) {
 982                 printk(KERN_ERR "NILFS: The specified checkpoint is "
 983                        "not a snapshot (checkpoint number=%llu).\n",
 984                        (unsigned long long)cno);
 985                 ret = -EINVAL;
 986                 goto out;
 987         }
 988
 989         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 990         if (ret) {
 991                 printk(KERN_ERR "NILFS: error loading snapshot "
 992                        "(checkpoint number=%llu).\n",
 993                (unsigned long long)cno);
 994                 goto out;
 995         }
 996         ret = nilfs_get_root_dentry(s, root, root_dentry);
 997         nilfs_put_root(root);
 998  out:
 999         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1000         return ret;
1001 }
1002
1003 /**
1004  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1005  * @root_dentry: root dentry of the tree to be shrunk
1006  *
1007  * This function returns true if the tree was in-use.
1008  */
1009 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1010 {
1011         shrink_dcache_parent(root_dentry);
1012         return d_count(root_dentry) > 1;
1013 }
1014
1015 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1016 {
1017         struct the_nilfs *nilfs = sb->s_fs_info;
1018         struct nilfs_root *root;
1019         struct inode *inode;
1020         struct dentry *dentry;
1021         int ret;
1022
1023         if (cno > nilfs->ns_cno)
1024                 return false;
1025
1026         if (cno >= nilfs_last_cno(nilfs))
1027                 return true;    /* protect recent checkpoints */
1028
1029         ret = false;
1030         root = nilfs_lookup_root(nilfs, cno);
1031         if (root) {
1032                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1033                 if (inode) {
1034                         dentry = d_find_alias(inode);
1035                         if (dentry) {
1036                                 ret = nilfs_tree_is_busy(dentry);
1037                                 dput(dentry);
1038                         }
1039                         iput(inode);
1040                 }
1041                 nilfs_put_root(root);
1042         }
1043         return ret;
1044 }
1045
1046 /**
1047  * nilfs_fill_super() - initialize a super block instance
1048  * @sb: super_block
1049  * @data: mount options
1050  * @silent: silent mode flag
1051  *
1052  * This function is called exclusively by nilfs->ns_mount_mutex.
1053  * So, the recovery process is protected from other simultaneous mounts.
1054  */
1055 static int
1056 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1057 {
1058         struct the_nilfs *nilfs;
1059         struct nilfs_root *fsroot;
1060         __u64 cno;
1061         int err;
1062
1063         nilfs = alloc_nilfs(sb->s_bdev);
1064         if (!nilfs)
1065                 return -ENOMEM;
1066
1067         sb->s_fs_info = nilfs;
1068
1069         err = init_nilfs(nilfs, sb, (char *)data);
1070         if (err)
1071                 goto failed_nilfs;
1072
1073         sb->s_op = &nilfs_sops;
1074         sb->s_export_op = &nilfs_export_ops;
1075         sb->s_root = NULL;
1076         sb->s_time_gran = 1;
1077         sb->s_max_links = NILFS_LINK_MAX;
1078
1079         sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
1080
1081         err = load_nilfs(nilfs, sb);
1082         if (err)
1083                 goto failed_nilfs;
1084
1085         cno = nilfs_last_cno(nilfs);
1086         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1087         if (err) {
1088                 printk(KERN_ERR "NILFS: error loading last checkpoint "
1089                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
1090                 goto failed_unload;
1091         }
1092
1093         if (!(sb->s_flags & MS_RDONLY)) {
1094                 err = nilfs_attach_log_writer(sb, fsroot);
1095                 if (err)
1096                         goto failed_checkpoint;
1097         }
1098
1099         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1100         if (err)
1101                 goto failed_segctor;
1102
1103         nilfs_put_root(fsroot);
1104
1105         if (!(sb->s_flags & MS_RDONLY)) {
1106                 down_write(&nilfs->ns_sem);
1107                 nilfs_setup_super(sb, true);
1108                 up_write(&nilfs->ns_sem);
1109         }
1110
1111         return 0;
1112
1113  failed_segctor:
1114         nilfs_detach_log_writer(sb);
1115
1116  failed_checkpoint:
1117         nilfs_put_root(fsroot);
1118
1119  failed_unload:
1120         iput(nilfs->ns_sufile);
1121         iput(nilfs->ns_cpfile);
1122         iput(nilfs->ns_dat);
1123
1124  failed_nilfs:
1125         destroy_nilfs(nilfs);
1126         return err;
1127 }
1128
1129 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1130 {
1131         struct the_nilfs *nilfs = sb->s_fs_info;
1132         unsigned long old_sb_flags;
1133         unsigned long old_mount_opt;
1134         int err;
1135
1136         sync_filesystem(sb);
1137         old_sb_flags = sb->s_flags;
1138         old_mount_opt = nilfs->ns_mount_opt;
1139
1140         if (!parse_options(data, sb, 1)) {
1141                 err = -EINVAL;
1142                 goto restore_opts;
1143         }
1144         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1145
1146         err = -EINVAL;
1147
1148         if (!nilfs_valid_fs(nilfs)) {
1149                 printk(KERN_WARNING "NILFS (device %s): couldn't "
1150                        "remount because the filesystem is in an "
1151                        "incomplete recovery state.\n", sb->s_id);
1152                 goto restore_opts;
1153         }
1154
1155         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1156                 goto out;
1157         if (*flags & MS_RDONLY) {
1158                 /* Shutting down log writer */
1159                 nilfs_detach_log_writer(sb);
1160                 sb->s_flags |= MS_RDONLY;
1161
1162                 /*
1163                  * Remounting a valid RW partition RDONLY, so set
1164                  * the RDONLY flag and then mark the partition as valid again.
1165                  */
1166                 down_write(&nilfs->ns_sem);
1167                 nilfs_cleanup_super(sb);
1168                 up_write(&nilfs->ns_sem);
1169         } else {
1170                 __u64 features;
1171                 struct nilfs_root *root;
1172
1173                 /*
1174                  * Mounting a RDONLY partition read-write, so reread and
1175                  * store the current valid flag.  (It may have been changed
1176                  * by fsck since we originally mounted the partition.)
1177                  */
1178                 down_read(&nilfs->ns_sem);
1179                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1180                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1181                 up_read(&nilfs->ns_sem);
1182                 if (features) {
1183                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1184                                "remount RDWR because of unsupported optional "
1185                                "features (%llx)\n",
1186                                sb->s_id, (unsigned long long)features);
1187                         err = -EROFS;
1188                         goto restore_opts;
1189                 }
1190
1191                 sb->s_flags &= ~MS_RDONLY;
1192
1193                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1194                 err = nilfs_attach_log_writer(sb, root);
1195                 if (err)
1196                         goto restore_opts;
1197
1198                 down_write(&nilfs->ns_sem);
1199                 nilfs_setup_super(sb, true);
1200                 up_write(&nilfs->ns_sem);
1201         }
1202  out:
1203         return 0;
1204
1205  restore_opts:
1206         sb->s_flags = old_sb_flags;
1207         nilfs->ns_mount_opt = old_mount_opt;
1208         return err;
1209 }
1210
1211 struct nilfs_super_data {
1212         struct block_device *bdev;
1213         __u64 cno;
1214         int flags;
1215 };
1216
1217 /**
1218  * nilfs_identify - pre-read mount options needed to identify mount instance
1219  * @data: mount options
1220  * @sd: nilfs_super_data
1221  */
1222 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1223 {
1224         char *p, *options = data;
1225         substring_t args[MAX_OPT_ARGS];
1226         int token;
1227         int ret = 0;
1228
1229         do {
1230                 p = strsep(&options, ",");
1231                 if (p != NULL && *p) {
1232                         token = match_token(p, tokens, args);
1233                         if (token == Opt_snapshot) {
1234                                 if (!(sd->flags & MS_RDONLY)) {
1235                                         ret++;
1236                                 } else {
1237                                         sd->cno = simple_strtoull(args[0].from,
1238                                                                   NULL, 0);
1239                                         /*
1240                                          * No need to see the end pointer;
1241                                          * match_token() has done syntax
1242                                          * checking.
1243                                          */
1244                                         if (sd->cno == 0)
1245                                                 ret++;
1246                                 }
1247                         }
1248                         if (ret)
1249                                 printk(KERN_ERR
1250                                        "NILFS: invalid mount option: %s\n", p);
1251                 }
1252                 if (!options)
1253                         break;
1254                 BUG_ON(options == data);
1255                 *(options - 1) = ',';
1256         } while (!ret);
1257         return ret;
1258 }
1259
1260 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1261 {
1262         s->s_bdev = data;
1263         s->s_dev = s->s_bdev->bd_dev;
1264         return 0;
1265 }
1266
1267 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1268 {
1269         return (void *)s->s_bdev == data;
1270 }
1271
1272 static struct dentry *
1273 nilfs_mount(struct file_system_type *fs_type, int flags,
1274              const char *dev_name, void *data)
1275 {
1276         struct nilfs_super_data sd;
1277         struct super_block *s;
1278         fmode_t mode = FMODE_READ | FMODE_EXCL;
1279         struct dentry *root_dentry;
1280         int err, s_new = false;
1281
1282         if (!(flags & MS_RDONLY))
1283                 mode |= FMODE_WRITE;
1284
1285         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1286         if (IS_ERR(sd.bdev))
1287                 return ERR_CAST(sd.bdev);
1288
1289         sd.cno = 0;
1290         sd.flags = flags;
1291         if (nilfs_identify((char *)data, &sd)) {
1292                 err = -EINVAL;
1293                 goto failed;
1294         }
1295
1296         /*
1297          * once the super is inserted into the list by sget, s_umount
1298          * will protect the lockfs code from trying to start a snapshot
1299          * while we are mounting
1300          */
1301         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1302         if (sd.bdev->bd_fsfreeze_count > 0) {
1303                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1304                 err = -EBUSY;
1305                 goto failed;
1306         }
1307         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1308                  sd.bdev);
1309         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1310         if (IS_ERR(s)) {
1311                 err = PTR_ERR(s);
1312                 goto failed;
1313         }
1314
1315         if (!s->s_root) {
1316                 s_new = true;
1317
1318                 /* New superblock instance created */
1319                 s->s_mode = mode;
1320                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1321                 sb_set_blocksize(s, block_size(sd.bdev));
1322
1323                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1324                 if (err)
1325                         goto failed_super;
1326
1327                 s->s_flags |= MS_ACTIVE;
1328         } else if (!sd.cno) {
1329                 if (nilfs_tree_is_busy(s->s_root)) {
1330                         if ((flags ^ s->s_flags) & MS_RDONLY) {
1331                                 printk(KERN_ERR "NILFS: the device already "
1332                                        "has a %s mount.\n",
1333                                        (s->s_flags & MS_RDONLY) ?
1334                                        "read-only" : "read/write");
1335                                 err = -EBUSY;
1336                                 goto failed_super;
1337                         }
1338                 } else {
1339                         /*
1340                          * Try remount to setup mount states if the current
1341                          * tree is not mounted and only snapshots use this sb.
1342                          */
1343                         err = nilfs_remount(s, &flags, data);
1344                         if (err)
1345                                 goto failed_super;
1346                 }
1347         }
1348
1349         if (sd.cno) {
1350                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1351                 if (err)
1352                         goto failed_super;
1353         } else {
1354                 root_dentry = dget(s->s_root);
1355         }
1356
1357         if (!s_new)
1358                 blkdev_put(sd.bdev, mode);
1359
1360         return root_dentry;
1361
1362  failed_super:
1363         deactivate_locked_super(s);
1364
1365  failed:
1366         if (!s_new)
1367                 blkdev_put(sd.bdev, mode);
1368         return ERR_PTR(err);
1369 }
1370
1371 struct file_system_type nilfs_fs_type = {
1372         .owner    = THIS_MODULE,
1373         .name     = "nilfs2",
1374         .mount    = nilfs_mount,
1375         .kill_sb  = kill_block_super,
1376         .fs_flags = FS_REQUIRES_DEV,
1377 };
1378 MODULE_ALIAS_FS("nilfs2");
1379
1380 static void nilfs_inode_init_once(void *obj)
1381 {
1382         struct nilfs_inode_info *ii = obj;
1383
1384         INIT_LIST_HEAD(&ii->i_dirty);
1385 #ifdef CONFIG_NILFS_XATTR
1386         init_rwsem(&ii->xattr_sem);
1387 #endif
1388         address_space_init_once(&ii->i_btnode_cache);
1389         ii->i_bmap = &ii->i_bmap_data;
1390         inode_init_once(&ii->vfs_inode);
1391 }
1392
1393 static void nilfs_segbuf_init_once(void *obj)
1394 {
1395         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1396 }
1397
1398 static void nilfs_destroy_cachep(void)
1399 {
1400         /*
1401          * Make sure all delayed rcu free inodes are flushed before we
1402          * destroy cache.
1403          */
1404         rcu_barrier();
1405
1406         kmem_cache_destroy(nilfs_inode_cachep);
1407         kmem_cache_destroy(nilfs_transaction_cachep);
1408         kmem_cache_destroy(nilfs_segbuf_cachep);
1409         kmem_cache_destroy(nilfs_btree_path_cache);
1410 }
1411
1412 static int __init nilfs_init_cachep(void)
1413 {
1414         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1415                         sizeof(struct nilfs_inode_info), 0,
1416                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1417                         nilfs_inode_init_once);
1418         if (!nilfs_inode_cachep)
1419                 goto fail;
1420
1421         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1422                         sizeof(struct nilfs_transaction_info), 0,
1423                         SLAB_RECLAIM_ACCOUNT, NULL);
1424         if (!nilfs_transaction_cachep)
1425                 goto fail;
1426
1427         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1428                         sizeof(struct nilfs_segment_buffer), 0,
1429                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1430         if (!nilfs_segbuf_cachep)
1431                 goto fail;
1432
1433         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1434                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1435                         0, 0, NULL);
1436         if (!nilfs_btree_path_cache)
1437                 goto fail;
1438
1439         return 0;
1440
1441 fail:
1442         nilfs_destroy_cachep();
1443         return -ENOMEM;
1444 }
1445
1446 static int __init init_nilfs_fs(void)
1447 {
1448         int err;
1449
1450         err = nilfs_init_cachep();
1451         if (err)
1452                 goto fail;
1453
1454         err = nilfs_sysfs_init();
1455         if (err)
1456                 goto free_cachep;
1457
1458         err = register_filesystem(&nilfs_fs_type);
1459         if (err)
1460                 goto deinit_sysfs_entry;
1461
1462         printk(KERN_INFO "NILFS version 2 loaded\n");
1463         return 0;
1464
1465 deinit_sysfs_entry:
1466         nilfs_sysfs_exit();
1467 free_cachep:
1468         nilfs_destroy_cachep();
1469 fail:
1470         return err;
1471 }
1472
1473 static void __exit exit_nilfs_fs(void)
1474 {
1475         nilfs_destroy_cachep();
1476         nilfs_sysfs_exit();
1477         unregister_filesystem(&nilfs_fs_type);
1478 }
1479
1480 module_init(init_nilfs_fs)
1481 module_exit(exit_nilfs_fs)