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 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
  75                  ...)
  76 {
  77         struct va_format vaf;
  78         va_list args;
  79
  80         va_start(args, fmt);
  81         vaf.fmt = fmt;
  82         vaf.va = &args;
  83         if (sb)
  84                 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
  85         else
  86                 printk("%sNILFS: %pV\n", level, &vaf);
  87         va_end(args);
  88 }
  89
  90 static void nilfs_set_error(struct super_block *sb)
  91 {
  92         struct the_nilfs *nilfs = sb->s_fs_info;
  93         struct nilfs_super_block **sbp;
  94
  95         down_write(&nilfs->ns_sem);
  96         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  97                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  98                 sbp = nilfs_prepare_super(sb, 0);
  99                 if (likely(sbp)) {
 100                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 101                         if (sbp[1])
 102                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 103                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 104                 }
 105         }
 106         up_write(&nilfs->ns_sem);
 107 }
 108
 109 /**
 110  * __nilfs_error() - report failure condition on a filesystem
 111  *
 112  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 113  * reporting an error message.  This function should be called when
 114  * NILFS detects incoherences or defects of meta data on disk.
 115  *
 116  * This implements the body of nilfs_error() macro.  Normally,
 117  * nilfs_error() should be used.  As for sustainable errors such as a
 118  * single-shot I/O error, nilfs_msg() should be used instead.
 119  *
 120  * Callers should not add a trailing newline since this will do it.
 121  */
 122 void __nilfs_error(struct super_block *sb, const char *function,
 123                    const char *fmt, ...)
 124 {
 125         struct the_nilfs *nilfs = sb->s_fs_info;
 126         struct va_format vaf;
 127         va_list args;
 128
 129         va_start(args, fmt);
 130
 131         vaf.fmt = fmt;
 132         vaf.va = &args;
 133
 134         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 135                sb->s_id, function, &vaf);
 136
 137         va_end(args);
 138
 139         if (!(sb->s_flags & MS_RDONLY)) {
 140                 nilfs_set_error(sb);
 141
 142                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 143                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 144                         sb->s_flags |= MS_RDONLY;
 145                 }
 146         }
 147
 148         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 149                 panic("NILFS (device %s): panic forced after error\n",
 150                       sb->s_id);
 151 }
 152
 153 struct inode *nilfs_alloc_inode(struct super_block *sb)
 154 {
 155         struct nilfs_inode_info *ii;
 156
 157         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 158         if (!ii)
 159                 return NULL;
 160         ii->i_bh = NULL;
 161         ii->i_state = 0;
 162         ii->i_cno = 0;
 163         ii->vfs_inode.i_version = 1;
 164         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
 165         return &ii->vfs_inode;
 166 }
 167
 168 static void nilfs_i_callback(struct rcu_head *head)
 169 {
 170         struct inode *inode = container_of(head, struct inode, i_rcu);
 171
 172         if (nilfs_is_metadata_file_inode(inode))
 173                 nilfs_mdt_destroy(inode);
 174
 175         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 176 }
 177
 178 void nilfs_destroy_inode(struct inode *inode)
 179 {
 180         call_rcu(&inode->i_rcu, nilfs_i_callback);
 181 }
 182
 183 static int nilfs_sync_super(struct super_block *sb, int flag)
 184 {
 185         struct the_nilfs *nilfs = sb->s_fs_info;
 186         int err;
 187
 188  retry:
 189         set_buffer_dirty(nilfs->ns_sbh[0]);
 190         if (nilfs_test_opt(nilfs, BARRIER)) {
 191                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 192                                           WRITE_SYNC | WRITE_FLUSH_FUA);
 193         } else {
 194                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 195         }
 196
 197         if (unlikely(err)) {
 198                 nilfs_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
 199                           err);
 200                 if (err == -EIO && nilfs->ns_sbh[1]) {
 201                         /*
 202                          * sbp[0] points to newer log than sbp[1],
 203                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 204                          */
 205                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 206                                nilfs->ns_sbsize);
 207                         nilfs_fall_back_super_block(nilfs);
 208                         goto retry;
 209                 }
 210         } else {
 211                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 212
 213                 nilfs->ns_sbwcount++;
 214
 215                 /*
 216                  * The latest segment becomes trailable from the position
 217                  * written in superblock.
 218                  */
 219                 clear_nilfs_discontinued(nilfs);
 220
 221                 /* update GC protection for recent segments */
 222                 if (nilfs->ns_sbh[1]) {
 223                         if (flag == NILFS_SB_COMMIT_ALL) {
 224                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 225                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 226                                         goto out;
 227                         }
 228                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 229                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 230                                 sbp = nilfs->ns_sbp[1];
 231                 }
 232
 233                 spin_lock(&nilfs->ns_last_segment_lock);
 234                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 235                 spin_unlock(&nilfs->ns_last_segment_lock);
 236         }
 237  out:
 238         return err;
 239 }
 240
 241 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 242                           struct the_nilfs *nilfs)
 243 {
 244         sector_t nfreeblocks;
 245
 246         /* nilfs->ns_sem must be locked by the caller. */
 247         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 248         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 249
 250         spin_lock(&nilfs->ns_last_segment_lock);
 251         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 252         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 253         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 254         spin_unlock(&nilfs->ns_last_segment_lock);
 255 }
 256
 257 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 258                                                int flip)
 259 {
 260         struct the_nilfs *nilfs = sb->s_fs_info;
 261         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 262
 263         /* nilfs->ns_sem must be locked by the caller. */
 264         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 265                 if (sbp[1] &&
 266                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 267                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 268                 } else {
 269                         nilfs_msg(sb, KERN_CRIT, "superblock broke");
 270                         return NULL;
 271                 }
 272         } else if (sbp[1] &&
 273                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 274                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 275         }
 276
 277         if (flip && sbp[1])
 278                 nilfs_swap_super_block(nilfs);
 279
 280         return sbp;
 281 }
 282
 283 int nilfs_commit_super(struct super_block *sb, int flag)
 284 {
 285         struct the_nilfs *nilfs = sb->s_fs_info;
 286         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 287         time_t t;
 288
 289         /* nilfs->ns_sem must be locked by the caller. */
 290         t = get_seconds();
 291         nilfs->ns_sbwtime = t;
 292         sbp[0]->s_wtime = cpu_to_le64(t);
 293         sbp[0]->s_sum = 0;
 294         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 295                                              (unsigned char *)sbp[0],
 296                                              nilfs->ns_sbsize));
 297         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 298                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 299                 sbp[1]->s_sum = 0;
 300                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 301                                             (unsigned char *)sbp[1],
 302                                             nilfs->ns_sbsize));
 303         }
 304         clear_nilfs_sb_dirty(nilfs);
 305         nilfs->ns_flushed_device = 1;
 306         /* make sure store to ns_flushed_device cannot be reordered */
 307         smp_wmb();
 308         return nilfs_sync_super(sb, flag);
 309 }
 310
 311 /**
 312  * nilfs_cleanup_super() - write filesystem state for cleanup
 313  * @sb: super block instance to be unmounted or degraded to read-only
 314  *
 315  * This function restores state flags in the on-disk super block.
 316  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 317  * filesystem was not clean previously.
 318  */
 319 int nilfs_cleanup_super(struct super_block *sb)
 320 {
 321         struct the_nilfs *nilfs = sb->s_fs_info;
 322         struct nilfs_super_block **sbp;
 323         int flag = NILFS_SB_COMMIT;
 324         int ret = -EIO;
 325
 326         sbp = nilfs_prepare_super(sb, 0);
 327         if (sbp) {
 328                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 329                 nilfs_set_log_cursor(sbp[0], nilfs);
 330                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 331                         /*
 332                          * make the "clean" flag also to the opposite
 333                          * super block if both super blocks point to
 334                          * the same checkpoint.
 335                          */
 336                         sbp[1]->s_state = sbp[0]->s_state;
 337                         flag = NILFS_SB_COMMIT_ALL;
 338                 }
 339                 ret = nilfs_commit_super(sb, flag);
 340         }
 341         return ret;
 342 }
 343
 344 /**
 345  * nilfs_move_2nd_super - relocate secondary super block
 346  * @sb: super block instance
 347  * @sb2off: new offset of the secondary super block (in bytes)
 348  */
 349 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 350 {
 351         struct the_nilfs *nilfs = sb->s_fs_info;
 352         struct buffer_head *nsbh;
 353         struct nilfs_super_block *nsbp;
 354         sector_t blocknr, newblocknr;
 355         unsigned long offset;
 356         int sb2i;  /* array index of the secondary superblock */
 357         int ret = 0;
 358
 359         /* nilfs->ns_sem must be locked by the caller. */
 360         if (nilfs->ns_sbh[1] &&
 361             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 362                 sb2i = 1;
 363                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 364         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 365                 sb2i = 0;
 366                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 367         } else {
 368                 sb2i = -1;
 369                 blocknr = 0;
 370         }
 371         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 372                 goto out;  /* super block location is unchanged */
 373
 374         /* Get new super block buffer */
 375         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 376         offset = sb2off & (nilfs->ns_blocksize - 1);
 377         nsbh = sb_getblk(sb, newblocknr);
 378         if (!nsbh) {
 379                 nilfs_msg(sb, KERN_WARNING,
 380                           "unable to move secondary superblock to block %llu",
 381                           (unsigned long long)newblocknr);
 382                 ret = -EIO;
 383                 goto out;
 384         }
 385         nsbp = (void *)nsbh->b_data + offset;
 386         memset(nsbp, 0, nilfs->ns_blocksize);
 387
 388         if (sb2i >= 0) {
 389                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 390                 brelse(nilfs->ns_sbh[sb2i]);
 391                 nilfs->ns_sbh[sb2i] = nsbh;
 392                 nilfs->ns_sbp[sb2i] = nsbp;
 393         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 394                 /* secondary super block will be restored to index 1 */
 395                 nilfs->ns_sbh[1] = nsbh;
 396                 nilfs->ns_sbp[1] = nsbp;
 397         } else {
 398                 brelse(nsbh);
 399         }
 400 out:
 401         return ret;
 402 }
 403
 404 /**
 405  * nilfs_resize_fs - resize the filesystem
 406  * @sb: super block instance
 407  * @newsize: new size of the filesystem (in bytes)
 408  */
 409 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 410 {
 411         struct the_nilfs *nilfs = sb->s_fs_info;
 412         struct nilfs_super_block **sbp;
 413         __u64 devsize, newnsegs;
 414         loff_t sb2off;
 415         int ret;
 416
 417         ret = -ERANGE;
 418         devsize = i_size_read(sb->s_bdev->bd_inode);
 419         if (newsize > devsize)
 420                 goto out;
 421
 422         /*
 423          * Write lock is required to protect some functions depending
 424          * on the number of segments, the number of reserved segments,
 425          * and so forth.
 426          */
 427         down_write(&nilfs->ns_segctor_sem);
 428
 429         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 430         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 431         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 432
 433         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 434         up_write(&nilfs->ns_segctor_sem);
 435         if (ret < 0)
 436                 goto out;
 437
 438         ret = nilfs_construct_segment(sb);
 439         if (ret < 0)
 440                 goto out;
 441
 442         down_write(&nilfs->ns_sem);
 443         nilfs_move_2nd_super(sb, sb2off);
 444         ret = -EIO;
 445         sbp = nilfs_prepare_super(sb, 0);
 446         if (likely(sbp)) {
 447                 nilfs_set_log_cursor(sbp[0], nilfs);
 448                 /*
 449                  * Drop NILFS_RESIZE_FS flag for compatibility with
 450                  * mount-time resize which may be implemented in a
 451                  * future release.
 452                  */
 453                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 454                                               ~NILFS_RESIZE_FS);
 455                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 456                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 457                 if (sbp[1])
 458                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 459                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 460         }
 461         up_write(&nilfs->ns_sem);
 462
 463         /*
 464          * Reset the range of allocatable segments last.  This order
 465          * is important in the case of expansion because the secondary
 466          * superblock must be protected from log write until migration
 467          * completes.
 468          */
 469         if (!ret)
 470                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 471 out:
 472         return ret;
 473 }
 474
 475 static void nilfs_put_super(struct super_block *sb)
 476 {
 477         struct the_nilfs *nilfs = sb->s_fs_info;
 478
 479         nilfs_detach_log_writer(sb);
 480
 481         if (!(sb->s_flags & MS_RDONLY)) {
 482                 down_write(&nilfs->ns_sem);
 483                 nilfs_cleanup_super(sb);
 484                 up_write(&nilfs->ns_sem);
 485         }
 486
 487         iput(nilfs->ns_sufile);
 488         iput(nilfs->ns_cpfile);
 489         iput(nilfs->ns_dat);
 490
 491         destroy_nilfs(nilfs);
 492         sb->s_fs_info = NULL;
 493 }
 494
 495 static int nilfs_sync_fs(struct super_block *sb, int wait)
 496 {
 497         struct the_nilfs *nilfs = sb->s_fs_info;
 498         struct nilfs_super_block **sbp;
 499         int err = 0;
 500
 501         /* This function is called when super block should be written back */
 502         if (wait)
 503                 err = nilfs_construct_segment(sb);
 504
 505         down_write(&nilfs->ns_sem);
 506         if (nilfs_sb_dirty(nilfs)) {
 507                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 508                 if (likely(sbp)) {
 509                         nilfs_set_log_cursor(sbp[0], nilfs);
 510                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 511                 }
 512         }
 513         up_write(&nilfs->ns_sem);
 514
 515         if (!err)
 516                 err = nilfs_flush_device(nilfs);
 517
 518         return err;
 519 }
 520
 521 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 522                             struct nilfs_root **rootp)
 523 {
 524         struct the_nilfs *nilfs = sb->s_fs_info;
 525         struct nilfs_root *root;
 526         struct nilfs_checkpoint *raw_cp;
 527         struct buffer_head *bh_cp;
 528         int err = -ENOMEM;
 529
 530         root = nilfs_find_or_create_root(
 531                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 532         if (!root)
 533                 return err;
 534
 535         if (root->ifile)
 536                 goto reuse; /* already attached checkpoint */
 537
 538         down_read(&nilfs->ns_segctor_sem);
 539         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 540                                           &bh_cp);
 541         up_read(&nilfs->ns_segctor_sem);
 542         if (unlikely(err)) {
 543                 if (err == -ENOENT || err == -EINVAL) {
 544                         nilfs_msg(sb, KERN_ERR,
 545                                   "Invalid checkpoint (checkpoint number=%llu)",
 546                                   (unsigned long long)cno);
 547                         err = -EINVAL;
 548                 }
 549                 goto failed;
 550         }
 551
 552         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 553                                &raw_cp->cp_ifile_inode, &root->ifile);
 554         if (err)
 555                 goto failed_bh;
 556
 557         atomic64_set(&root->inodes_count,
 558                         le64_to_cpu(raw_cp->cp_inodes_count));
 559         atomic64_set(&root->blocks_count,
 560                         le64_to_cpu(raw_cp->cp_blocks_count));
 561
 562         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 563
 564  reuse:
 565         *rootp = root;
 566         return 0;
 567
 568  failed_bh:
 569         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 570  failed:
 571         nilfs_put_root(root);
 572
 573         return err;
 574 }
 575
 576 static int nilfs_freeze(struct super_block *sb)
 577 {
 578         struct the_nilfs *nilfs = sb->s_fs_info;
 579         int err;
 580
 581         if (sb->s_flags & MS_RDONLY)
 582                 return 0;
 583
 584         /* Mark super block clean */
 585         down_write(&nilfs->ns_sem);
 586         err = nilfs_cleanup_super(sb);
 587         up_write(&nilfs->ns_sem);
 588         return err;
 589 }
 590
 591 static int nilfs_unfreeze(struct super_block *sb)
 592 {
 593         struct the_nilfs *nilfs = sb->s_fs_info;
 594
 595         if (sb->s_flags & MS_RDONLY)
 596                 return 0;
 597
 598         down_write(&nilfs->ns_sem);
 599         nilfs_setup_super(sb, false);
 600         up_write(&nilfs->ns_sem);
 601         return 0;
 602 }
 603
 604 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 605 {
 606         struct super_block *sb = dentry->d_sb;
 607         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 608         struct the_nilfs *nilfs = root->nilfs;
 609         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 610         unsigned long long blocks;
 611         unsigned long overhead;
 612         unsigned long nrsvblocks;
 613         sector_t nfreeblocks;
 614         u64 nmaxinodes, nfreeinodes;
 615         int err;
 616
 617         /*
 618          * Compute all of the segment blocks
 619          *
 620          * The blocks before first segment and after last segment
 621          * are excluded.
 622          */
 623         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 624                 - nilfs->ns_first_data_block;
 625         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 626
 627         /*
 628          * Compute the overhead
 629          *
 630          * When distributing meta data blocks outside segment structure,
 631          * We must count them as the overhead.
 632          */
 633         overhead = 0;
 634
 635         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 636         if (unlikely(err))
 637                 return err;
 638
 639         err = nilfs_ifile_count_free_inodes(root->ifile,
 640                                             &nmaxinodes, &nfreeinodes);
 641         if (unlikely(err)) {
 642                 nilfs_msg(sb, KERN_WARNING,
 643                           "failed to count free inodes: err=%d", err);
 644                 if (err == -ERANGE) {
 645                         /*
 646                          * If nilfs_palloc_count_max_entries() returns
 647                          * -ERANGE error code then we simply treat
 648                          * curent inodes count as maximum possible and
 649                          * zero as free inodes value.
 650                          */
 651                         nmaxinodes = atomic64_read(&root->inodes_count);
 652                         nfreeinodes = 0;
 653                         err = 0;
 654                 } else
 655                         return err;
 656         }
 657
 658         buf->f_type = NILFS_SUPER_MAGIC;
 659         buf->f_bsize = sb->s_blocksize;
 660         buf->f_blocks = blocks - overhead;
 661         buf->f_bfree = nfreeblocks;
 662         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 663                 (buf->f_bfree - nrsvblocks) : 0;
 664         buf->f_files = nmaxinodes;
 665         buf->f_ffree = nfreeinodes;
 666         buf->f_namelen = NILFS_NAME_LEN;
 667         buf->f_fsid.val[0] = (u32)id;
 668         buf->f_fsid.val[1] = (u32)(id >> 32);
 669
 670         return 0;
 671 }
 672
 673 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 674 {
 675         struct super_block *sb = dentry->d_sb;
 676         struct the_nilfs *nilfs = sb->s_fs_info;
 677         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 678
 679         if (!nilfs_test_opt(nilfs, BARRIER))
 680                 seq_puts(seq, ",nobarrier");
 681         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 682                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 683         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 684                 seq_puts(seq, ",errors=panic");
 685         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 686                 seq_puts(seq, ",errors=continue");
 687         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 688                 seq_puts(seq, ",order=strict");
 689         if (nilfs_test_opt(nilfs, NORECOVERY))
 690                 seq_puts(seq, ",norecovery");
 691         if (nilfs_test_opt(nilfs, DISCARD))
 692                 seq_puts(seq, ",discard");
 693
 694         return 0;
 695 }
 696
 697 static const struct super_operations nilfs_sops = {
 698         .alloc_inode    = nilfs_alloc_inode,
 699         .destroy_inode  = nilfs_destroy_inode,
 700         .dirty_inode    = nilfs_dirty_inode,
 701         .evict_inode    = nilfs_evict_inode,
 702         .put_super      = nilfs_put_super,
 703         .sync_fs        = nilfs_sync_fs,
 704         .freeze_fs      = nilfs_freeze,
 705         .unfreeze_fs    = nilfs_unfreeze,
 706         .statfs         = nilfs_statfs,
 707         .remount_fs     = nilfs_remount,
 708         .show_options = nilfs_show_options
 709 };
 710
 711 enum {
 712         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 713         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 714         Opt_discard, Opt_nodiscard, Opt_err,
 715 };
 716
 717 static match_table_t tokens = {
 718         {Opt_err_cont, "errors=continue"},
 719         {Opt_err_panic, "errors=panic"},
 720         {Opt_err_ro, "errors=remount-ro"},
 721         {Opt_barrier, "barrier"},
 722         {Opt_nobarrier, "nobarrier"},
 723         {Opt_snapshot, "cp=%u"},
 724         {Opt_order, "order=%s"},
 725         {Opt_norecovery, "norecovery"},
 726         {Opt_discard, "discard"},
 727         {Opt_nodiscard, "nodiscard"},
 728         {Opt_err, NULL}
 729 };
 730
 731 static int parse_options(char *options, struct super_block *sb, int is_remount)
 732 {
 733         struct the_nilfs *nilfs = sb->s_fs_info;
 734         char *p;
 735         substring_t args[MAX_OPT_ARGS];
 736
 737         if (!options)
 738                 return 1;
 739
 740         while ((p = strsep(&options, ",")) != NULL) {
 741                 int token;
 742
 743                 if (!*p)
 744                         continue;
 745
 746                 token = match_token(p, tokens, args);
 747                 switch (token) {
 748                 case Opt_barrier:
 749                         nilfs_set_opt(nilfs, BARRIER);
 750                         break;
 751                 case Opt_nobarrier:
 752                         nilfs_clear_opt(nilfs, BARRIER);
 753                         break;
 754                 case Opt_order:
 755                         if (strcmp(args[0].from, "relaxed") == 0)
 756                                 /* Ordered data semantics */
 757                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 758                         else if (strcmp(args[0].from, "strict") == 0)
 759                                 /* Strict in-order semantics */
 760                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 761                         else
 762                                 return 0;
 763                         break;
 764                 case Opt_err_panic:
 765                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 766                         break;
 767                 case Opt_err_ro:
 768                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 769                         break;
 770                 case Opt_err_cont:
 771                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 772                         break;
 773                 case Opt_snapshot:
 774                         if (is_remount) {
 775                                 nilfs_msg(sb, KERN_ERR,
 776                                           "\"%s\" option is invalid for remount",
 777                                           p);
 778                                 return 0;
 779                         }
 780                         break;
 781                 case Opt_norecovery:
 782                         nilfs_set_opt(nilfs, NORECOVERY);
 783                         break;
 784                 case Opt_discard:
 785                         nilfs_set_opt(nilfs, DISCARD);
 786                         break;
 787                 case Opt_nodiscard:
 788                         nilfs_clear_opt(nilfs, DISCARD);
 789                         break;
 790                 default:
 791                         nilfs_msg(sb, KERN_ERR,
 792                                   "unrecognized mount option \"%s\"", p);
 793                         return 0;
 794                 }
 795         }
 796         return 1;
 797 }
 798
 799 static inline void
 800 nilfs_set_default_options(struct super_block *sb,
 801                           struct nilfs_super_block *sbp)
 802 {
 803         struct the_nilfs *nilfs = sb->s_fs_info;
 804
 805         nilfs->ns_mount_opt =
 806                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 807 }
 808
 809 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 810 {
 811         struct the_nilfs *nilfs = sb->s_fs_info;
 812         struct nilfs_super_block **sbp;
 813         int max_mnt_count;
 814         int mnt_count;
 815
 816         /* nilfs->ns_sem must be locked by the caller. */
 817         sbp = nilfs_prepare_super(sb, 0);
 818         if (!sbp)
 819                 return -EIO;
 820
 821         if (!is_mount)
 822                 goto skip_mount_setup;
 823
 824         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 825         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 826
 827         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 828                 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
 829 #if 0
 830         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 831                 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
 832 #endif
 833         }
 834         if (!max_mnt_count)
 835                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 836
 837         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 838         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 839
 840 skip_mount_setup:
 841         sbp[0]->s_state =
 842                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 843         /* synchronize sbp[1] with sbp[0] */
 844         if (sbp[1])
 845                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 846         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 847 }
 848
 849 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 850                                                  u64 pos, int blocksize,
 851                                                  struct buffer_head **pbh)
 852 {
 853         unsigned long long sb_index = pos;
 854         unsigned long offset;
 855
 856         offset = do_div(sb_index, blocksize);
 857         *pbh = sb_bread(sb, sb_index);
 858         if (!*pbh)
 859                 return NULL;
 860         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 861 }
 862
 863 int nilfs_store_magic_and_option(struct super_block *sb,
 864                                  struct nilfs_super_block *sbp,
 865                                  char *data)
 866 {
 867         struct the_nilfs *nilfs = sb->s_fs_info;
 868
 869         sb->s_magic = le16_to_cpu(sbp->s_magic);
 870
 871         /* FS independent flags */
 872 #ifdef NILFS_ATIME_DISABLE
 873         sb->s_flags |= MS_NOATIME;
 874 #endif
 875
 876         nilfs_set_default_options(sb, sbp);
 877
 878         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 879         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 880         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 881         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 882
 883         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 884 }
 885
 886 int nilfs_check_feature_compatibility(struct super_block *sb,
 887                                       struct nilfs_super_block *sbp)
 888 {
 889         __u64 features;
 890
 891         features = le64_to_cpu(sbp->s_feature_incompat) &
 892                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 893         if (features) {
 894                 nilfs_msg(sb, KERN_ERR,
 895                           "couldn't mount because of unsupported optional features (%llx)",
 896                           (unsigned long long)features);
 897                 return -EINVAL;
 898         }
 899         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 900                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 901         if (!(sb->s_flags & MS_RDONLY) && features) {
 902                 nilfs_msg(sb, KERN_ERR,
 903                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 904                           (unsigned long long)features);
 905                 return -EINVAL;
 906         }
 907         return 0;
 908 }
 909
 910 static int nilfs_get_root_dentry(struct super_block *sb,
 911                                  struct nilfs_root *root,
 912                                  struct dentry **root_dentry)
 913 {
 914         struct inode *inode;
 915         struct dentry *dentry;
 916         int ret = 0;
 917
 918         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 919         if (IS_ERR(inode)) {
 920                 ret = PTR_ERR(inode);
 921                 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
 922                 goto out;
 923         }
 924         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 925                 iput(inode);
 926                 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
 927                 ret = -EINVAL;
 928                 goto out;
 929         }
 930
 931         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 932                 dentry = d_find_alias(inode);
 933                 if (!dentry) {
 934                         dentry = d_make_root(inode);
 935                         if (!dentry) {
 936                                 ret = -ENOMEM;
 937                                 goto failed_dentry;
 938                         }
 939                 } else {
 940                         iput(inode);
 941                 }
 942         } else {
 943                 dentry = d_obtain_root(inode);
 944                 if (IS_ERR(dentry)) {
 945                         ret = PTR_ERR(dentry);
 946                         goto failed_dentry;
 947                 }
 948         }
 949         *root_dentry = dentry;
 950  out:
 951         return ret;
 952
 953  failed_dentry:
 954         nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
 955         goto out;
 956 }
 957
 958 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 959                                  struct dentry **root_dentry)
 960 {
 961         struct the_nilfs *nilfs = s->s_fs_info;
 962         struct nilfs_root *root;
 963         int ret;
 964
 965         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 966
 967         down_read(&nilfs->ns_segctor_sem);
 968         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 969         up_read(&nilfs->ns_segctor_sem);
 970         if (ret < 0) {
 971                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 972                 goto out;
 973         } else if (!ret) {
 974                 nilfs_msg(s, KERN_ERR,
 975                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 976                           (unsigned long long)cno);
 977                 ret = -EINVAL;
 978                 goto out;
 979         }
 980
 981         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 982         if (ret) {
 983                 nilfs_msg(s, KERN_ERR,
 984                           "error %d while loading snapshot (checkpoint number=%llu)",
 985                           ret, (unsigned long long)cno);
 986                 goto out;
 987         }
 988         ret = nilfs_get_root_dentry(s, root, root_dentry);
 989         nilfs_put_root(root);
 990  out:
 991         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 992         return ret;
 993 }
 994
 995 /**
 996  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 997  * @root_dentry: root dentry of the tree to be shrunk
 998  *
 999  * This function returns true if the tree was in-use.
1000  */
1001 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1002 {
1003         shrink_dcache_parent(root_dentry);
1004         return d_count(root_dentry) > 1;
1005 }
1006
1007 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1008 {
1009         struct the_nilfs *nilfs = sb->s_fs_info;
1010         struct nilfs_root *root;
1011         struct inode *inode;
1012         struct dentry *dentry;
1013         int ret;
1014
1015         if (cno > nilfs->ns_cno)
1016                 return false;
1017
1018         if (cno >= nilfs_last_cno(nilfs))
1019                 return true;    /* protect recent checkpoints */
1020
1021         ret = false;
1022         root = nilfs_lookup_root(nilfs, cno);
1023         if (root) {
1024                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1025                 if (inode) {
1026                         dentry = d_find_alias(inode);
1027                         if (dentry) {
1028                                 ret = nilfs_tree_is_busy(dentry);
1029                                 dput(dentry);
1030                         }
1031                         iput(inode);
1032                 }
1033                 nilfs_put_root(root);
1034         }
1035         return ret;
1036 }
1037
1038 /**
1039  * nilfs_fill_super() - initialize a super block instance
1040  * @sb: super_block
1041  * @data: mount options
1042  * @silent: silent mode flag
1043  *
1044  * This function is called exclusively by nilfs->ns_mount_mutex.
1045  * So, the recovery process is protected from other simultaneous mounts.
1046  */
1047 static int
1048 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1049 {
1050         struct the_nilfs *nilfs;
1051         struct nilfs_root *fsroot;
1052         __u64 cno;
1053         int err;
1054
1055         nilfs = alloc_nilfs(sb);
1056         if (!nilfs)
1057                 return -ENOMEM;
1058
1059         sb->s_fs_info = nilfs;
1060
1061         err = init_nilfs(nilfs, sb, (char *)data);
1062         if (err)
1063                 goto failed_nilfs;
1064
1065         sb->s_op = &nilfs_sops;
1066         sb->s_export_op = &nilfs_export_ops;
1067         sb->s_root = NULL;
1068         sb->s_time_gran = 1;
1069         sb->s_max_links = NILFS_LINK_MAX;
1070
1071         sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
1072
1073         err = load_nilfs(nilfs, sb);
1074         if (err)
1075                 goto failed_nilfs;
1076
1077         cno = nilfs_last_cno(nilfs);
1078         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1079         if (err) {
1080                 nilfs_msg(sb, KERN_ERR,
1081                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1082                           err, (unsigned long long)cno);
1083                 goto failed_unload;
1084         }
1085
1086         if (!(sb->s_flags & MS_RDONLY)) {
1087                 err = nilfs_attach_log_writer(sb, fsroot);
1088                 if (err)
1089                         goto failed_checkpoint;
1090         }
1091
1092         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1093         if (err)
1094                 goto failed_segctor;
1095
1096         nilfs_put_root(fsroot);
1097
1098         if (!(sb->s_flags & MS_RDONLY)) {
1099                 down_write(&nilfs->ns_sem);
1100                 nilfs_setup_super(sb, true);
1101                 up_write(&nilfs->ns_sem);
1102         }
1103
1104         return 0;
1105
1106  failed_segctor:
1107         nilfs_detach_log_writer(sb);
1108
1109  failed_checkpoint:
1110         nilfs_put_root(fsroot);
1111
1112  failed_unload:
1113         iput(nilfs->ns_sufile);
1114         iput(nilfs->ns_cpfile);
1115         iput(nilfs->ns_dat);
1116
1117  failed_nilfs:
1118         destroy_nilfs(nilfs);
1119         return err;
1120 }
1121
1122 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1123 {
1124         struct the_nilfs *nilfs = sb->s_fs_info;
1125         unsigned long old_sb_flags;
1126         unsigned long old_mount_opt;
1127         int err;
1128
1129         sync_filesystem(sb);
1130         old_sb_flags = sb->s_flags;
1131         old_mount_opt = nilfs->ns_mount_opt;
1132
1133         if (!parse_options(data, sb, 1)) {
1134                 err = -EINVAL;
1135                 goto restore_opts;
1136         }
1137         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1138
1139         err = -EINVAL;
1140
1141         if (!nilfs_valid_fs(nilfs)) {
1142                 nilfs_msg(sb, KERN_WARNING,
1143                           "couldn't remount because the filesystem is in an incomplete recovery state");
1144                 goto restore_opts;
1145         }
1146
1147         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1148                 goto out;
1149         if (*flags & MS_RDONLY) {
1150                 /* Shutting down log writer */
1151                 nilfs_detach_log_writer(sb);
1152                 sb->s_flags |= MS_RDONLY;
1153
1154                 /*
1155                  * Remounting a valid RW partition RDONLY, so set
1156                  * the RDONLY flag and then mark the partition as valid again.
1157                  */
1158                 down_write(&nilfs->ns_sem);
1159                 nilfs_cleanup_super(sb);
1160                 up_write(&nilfs->ns_sem);
1161         } else {
1162                 __u64 features;
1163                 struct nilfs_root *root;
1164
1165                 /*
1166                  * Mounting a RDONLY partition read-write, so reread and
1167                  * store the current valid flag.  (It may have been changed
1168                  * by fsck since we originally mounted the partition.)
1169                  */
1170                 down_read(&nilfs->ns_sem);
1171                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1172                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1173                 up_read(&nilfs->ns_sem);
1174                 if (features) {
1175                         nilfs_msg(sb, KERN_WARNING,
1176                                   "couldn't remount RDWR because of unsupported optional features (%llx)",
1177                                   (unsigned long long)features);
1178                         err = -EROFS;
1179                         goto restore_opts;
1180                 }
1181
1182                 sb->s_flags &= ~MS_RDONLY;
1183
1184                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1185                 err = nilfs_attach_log_writer(sb, root);
1186                 if (err)
1187                         goto restore_opts;
1188
1189                 down_write(&nilfs->ns_sem);
1190                 nilfs_setup_super(sb, true);
1191                 up_write(&nilfs->ns_sem);
1192         }
1193  out:
1194         return 0;
1195
1196  restore_opts:
1197         sb->s_flags = old_sb_flags;
1198         nilfs->ns_mount_opt = old_mount_opt;
1199         return err;
1200 }
1201
1202 struct nilfs_super_data {
1203         struct block_device *bdev;
1204         __u64 cno;
1205         int flags;
1206 };
1207
1208 static int nilfs_parse_snapshot_option(const char *option,
1209                                        const substring_t *arg,
1210                                        struct nilfs_super_data *sd)
1211 {
1212         unsigned long long val;
1213         const char *msg = NULL;
1214         int err;
1215
1216         if (!(sd->flags & MS_RDONLY)) {
1217                 msg = "read-only option is not specified";
1218                 goto parse_error;
1219         }
1220
1221         err = kstrtoull(arg->from, 0, &val);
1222         if (err) {
1223                 if (err == -ERANGE)
1224                         msg = "too large checkpoint number";
1225                 else
1226                         msg = "malformed argument";
1227                 goto parse_error;
1228         } else if (val == 0) {
1229                 msg = "invalid checkpoint number 0";
1230                 goto parse_error;
1231         }
1232         sd->cno = val;
1233         return 0;
1234
1235 parse_error:
1236         nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1237         return 1;
1238 }
1239
1240 /**
1241  * nilfs_identify - pre-read mount options needed to identify mount instance
1242  * @data: mount options
1243  * @sd: nilfs_super_data
1244  */
1245 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1246 {
1247         char *p, *options = data;
1248         substring_t args[MAX_OPT_ARGS];
1249         int token;
1250         int ret = 0;
1251
1252         do {
1253                 p = strsep(&options, ",");
1254                 if (p != NULL && *p) {
1255                         token = match_token(p, tokens, args);
1256                         if (token == Opt_snapshot)
1257                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1258                                                                   sd);
1259                 }
1260                 if (!options)
1261                         break;
1262                 BUG_ON(options == data);
1263                 *(options - 1) = ',';
1264         } while (!ret);
1265         return ret;
1266 }
1267
1268 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1269 {
1270         s->s_bdev = data;
1271         s->s_dev = s->s_bdev->bd_dev;
1272         return 0;
1273 }
1274
1275 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1276 {
1277         return (void *)s->s_bdev == data;
1278 }
1279
1280 static struct dentry *
1281 nilfs_mount(struct file_system_type *fs_type, int flags,
1282              const char *dev_name, void *data)
1283 {
1284         struct nilfs_super_data sd;
1285         struct super_block *s;
1286         fmode_t mode = FMODE_READ | FMODE_EXCL;
1287         struct dentry *root_dentry;
1288         int err, s_new = false;
1289
1290         if (!(flags & MS_RDONLY))
1291                 mode |= FMODE_WRITE;
1292
1293         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1294         if (IS_ERR(sd.bdev))
1295                 return ERR_CAST(sd.bdev);
1296
1297         sd.cno = 0;
1298         sd.flags = flags;
1299         if (nilfs_identify((char *)data, &sd)) {
1300                 err = -EINVAL;
1301                 goto failed;
1302         }
1303
1304         /*
1305          * once the super is inserted into the list by sget, s_umount
1306          * will protect the lockfs code from trying to start a snapshot
1307          * while we are mounting
1308          */
1309         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1310         if (sd.bdev->bd_fsfreeze_count > 0) {
1311                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1312                 err = -EBUSY;
1313                 goto failed;
1314         }
1315         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1316                  sd.bdev);
1317         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1318         if (IS_ERR(s)) {
1319                 err = PTR_ERR(s);
1320                 goto failed;
1321         }
1322
1323         if (!s->s_root) {
1324                 s_new = true;
1325
1326                 /* New superblock instance created */
1327                 s->s_mode = mode;
1328                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1329                 sb_set_blocksize(s, block_size(sd.bdev));
1330
1331                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1332                 if (err)
1333                         goto failed_super;
1334
1335                 s->s_flags |= MS_ACTIVE;
1336         } else if (!sd.cno) {
1337                 if (nilfs_tree_is_busy(s->s_root)) {
1338                         if ((flags ^ s->s_flags) & MS_RDONLY) {
1339                                 nilfs_msg(s, KERN_ERR,
1340                                           "the device already has a %s mount.",
1341                                           (s->s_flags & MS_RDONLY) ?
1342                                           "read-only" : "read/write");
1343                                 err = -EBUSY;
1344                                 goto failed_super;
1345                         }
1346                 } else {
1347                         /*
1348                          * Try remount to setup mount states if the current
1349                          * tree is not mounted and only snapshots use this sb.
1350                          */
1351                         err = nilfs_remount(s, &flags, data);
1352                         if (err)
1353                                 goto failed_super;
1354                 }
1355         }
1356
1357         if (sd.cno) {
1358                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1359                 if (err)
1360                         goto failed_super;
1361         } else {
1362                 root_dentry = dget(s->s_root);
1363         }
1364
1365         if (!s_new)
1366                 blkdev_put(sd.bdev, mode);
1367
1368         return root_dentry;
1369
1370  failed_super:
1371         deactivate_locked_super(s);
1372
1373  failed:
1374         if (!s_new)
1375                 blkdev_put(sd.bdev, mode);
1376         return ERR_PTR(err);
1377 }
1378
1379 struct file_system_type nilfs_fs_type = {
1380         .owner    = THIS_MODULE,
1381         .name     = "nilfs2",
1382         .mount    = nilfs_mount,
1383         .kill_sb  = kill_block_super,
1384         .fs_flags = FS_REQUIRES_DEV,
1385 };
1386 MODULE_ALIAS_FS("nilfs2");
1387
1388 static void nilfs_inode_init_once(void *obj)
1389 {
1390         struct nilfs_inode_info *ii = obj;
1391
1392         INIT_LIST_HEAD(&ii->i_dirty);
1393 #ifdef CONFIG_NILFS_XATTR
1394         init_rwsem(&ii->xattr_sem);
1395 #endif
1396         address_space_init_once(&ii->i_btnode_cache);
1397         ii->i_bmap = &ii->i_bmap_data;
1398         inode_init_once(&ii->vfs_inode);
1399 }
1400
1401 static void nilfs_segbuf_init_once(void *obj)
1402 {
1403         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1404 }
1405
1406 static void nilfs_destroy_cachep(void)
1407 {
1408         /*
1409          * Make sure all delayed rcu free inodes are flushed before we
1410          * destroy cache.
1411          */
1412         rcu_barrier();
1413
1414         kmem_cache_destroy(nilfs_inode_cachep);
1415         kmem_cache_destroy(nilfs_transaction_cachep);
1416         kmem_cache_destroy(nilfs_segbuf_cachep);
1417         kmem_cache_destroy(nilfs_btree_path_cache);
1418 }
1419
1420 static int __init nilfs_init_cachep(void)
1421 {
1422         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1423                         sizeof(struct nilfs_inode_info), 0,
1424                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1425                         nilfs_inode_init_once);
1426         if (!nilfs_inode_cachep)
1427                 goto fail;
1428
1429         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1430                         sizeof(struct nilfs_transaction_info), 0,
1431                         SLAB_RECLAIM_ACCOUNT, NULL);
1432         if (!nilfs_transaction_cachep)
1433                 goto fail;
1434
1435         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1436                         sizeof(struct nilfs_segment_buffer), 0,
1437                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1438         if (!nilfs_segbuf_cachep)
1439                 goto fail;
1440
1441         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1442                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1443                         0, 0, NULL);
1444         if (!nilfs_btree_path_cache)
1445                 goto fail;
1446
1447         return 0;
1448
1449 fail:
1450         nilfs_destroy_cachep();
1451         return -ENOMEM;
1452 }
1453
1454 static int __init init_nilfs_fs(void)
1455 {
1456         int err;
1457
1458         err = nilfs_init_cachep();
1459         if (err)
1460                 goto fail;
1461
1462         err = nilfs_sysfs_init();
1463         if (err)
1464                 goto free_cachep;
1465
1466         err = register_filesystem(&nilfs_fs_type);
1467         if (err)
1468                 goto deinit_sysfs_entry;
1469
1470         printk(KERN_INFO "NILFS version 2 loaded\n");
1471         return 0;
1472
1473 deinit_sysfs_entry:
1474         nilfs_sysfs_exit();
1475 free_cachep:
1476         nilfs_destroy_cachep();
1477 fail:
1478         return err;
1479 }
1480
1481 static void __exit exit_nilfs_fs(void)
1482 {
1483         nilfs_destroy_cachep();
1484         nilfs_sysfs_exit();
1485         unregister_filesystem(&nilfs_fs_type);
1486 }
1487
1488 module_init(init_nilfs_fs)
1489 module_exit(exit_nilfs_fs)