fs/nilfs2/the_nilfs.c

   1 /*
   2  * the_nilfs.c - the_nilfs shared structure.
   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  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/slab.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/crc32.h>
  29 #include "nilfs.h"
  30 #include "segment.h"
  31 #include "alloc.h"
  32 #include "cpfile.h"
  33 #include "sufile.h"
  34 #include "dat.h"
  35 #include "segbuf.h"
  36
  37
  38 static LIST_HEAD(nilfs_objects);
  39 static DEFINE_SPINLOCK(nilfs_lock);
  40
  41 void nilfs_set_last_segment(struct the_nilfs *nilfs,
  42                             sector_t start_blocknr, u64 seq, __u64 cno)
  43 {
  44         spin_lock(&nilfs->ns_last_segment_lock);
  45         nilfs->ns_last_pseg = start_blocknr;
  46         nilfs->ns_last_seq = seq;
  47         nilfs->ns_last_cno = cno;
  48         spin_unlock(&nilfs->ns_last_segment_lock);
  49 }
  50
  51 /**
  52  * alloc_nilfs - allocate the_nilfs structure
  53  * @bdev: block device to which the_nilfs is related
  54  *
  55  * alloc_nilfs() allocates memory for the_nilfs and
  56  * initializes its reference count and locks.
  57  *
  58  * Return Value: On success, pointer to the_nilfs is returned.
  59  * On error, NULL is returned.
  60  */
  61 static struct the_nilfs *alloc_nilfs(struct block_device *bdev)
  62 {
  63         struct the_nilfs *nilfs;
  64
  65         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
  66         if (!nilfs)
  67                 return NULL;
  68
  69         nilfs->ns_bdev = bdev;
  70         atomic_set(&nilfs->ns_count, 1);
  71         atomic_set(&nilfs->ns_ndirtyblks, 0);
  72         init_rwsem(&nilfs->ns_sem);
  73         init_rwsem(&nilfs->ns_super_sem);
  74         mutex_init(&nilfs->ns_mount_mutex);
  75         init_rwsem(&nilfs->ns_writer_sem);
  76         INIT_LIST_HEAD(&nilfs->ns_list);
  77         INIT_LIST_HEAD(&nilfs->ns_supers);
  78         spin_lock_init(&nilfs->ns_last_segment_lock);
  79         nilfs->ns_gc_inodes_h = NULL;
  80         init_rwsem(&nilfs->ns_segctor_sem);
  81
  82         return nilfs;
  83 }
  84
  85 /**
  86  * find_or_create_nilfs - find or create nilfs object
  87  * @bdev: block device to which the_nilfs is related
  88  *
  89  * find_nilfs() looks up an existent nilfs object created on the
  90  * device and gets the reference count of the object.  If no nilfs object
  91  * is found on the device, a new nilfs object is allocated.
  92  *
  93  * Return Value: On success, pointer to the nilfs object is returned.
  94  * On error, NULL is returned.
  95  */
  96 struct the_nilfs *find_or_create_nilfs(struct block_device *bdev)
  97 {
  98         struct the_nilfs *nilfs, *new = NULL;
  99
 100  retry:
 101         spin_lock(&nilfs_lock);
 102         list_for_each_entry(nilfs, &nilfs_objects, ns_list) {
 103                 if (nilfs->ns_bdev == bdev) {
 104                         get_nilfs(nilfs);
 105                         spin_unlock(&nilfs_lock);
 106                         if (new)
 107                                 put_nilfs(new);
 108                         return nilfs; /* existing object */
 109                 }
 110         }
 111         if (new) {
 112                 list_add_tail(&new->ns_list, &nilfs_objects);
 113                 spin_unlock(&nilfs_lock);
 114                 return new; /* new object */
 115         }
 116         spin_unlock(&nilfs_lock);
 117
 118         new = alloc_nilfs(bdev);
 119         if (new)
 120                 goto retry;
 121         return NULL; /* insufficient memory */
 122 }
 123
 124 /**
 125  * put_nilfs - release a reference to the_nilfs
 126  * @nilfs: the_nilfs structure to be released
 127  *
 128  * put_nilfs() decrements a reference counter of the_nilfs.
 129  * If the reference count reaches zero, the_nilfs is freed.
 130  */
 131 void put_nilfs(struct the_nilfs *nilfs)
 132 {
 133         spin_lock(&nilfs_lock);
 134         if (!atomic_dec_and_test(&nilfs->ns_count)) {
 135                 spin_unlock(&nilfs_lock);
 136                 return;
 137         }
 138         list_del_init(&nilfs->ns_list);
 139         spin_unlock(&nilfs_lock);
 140
 141         /*
 142          * Increment of ns_count never occurs below because the caller
 143          * of get_nilfs() holds at least one reference to the_nilfs.
 144          * Thus its exclusion control is not required here.
 145          */
 146
 147         might_sleep();
 148         if (nilfs_loaded(nilfs)) {
 149                 nilfs_mdt_destroy(nilfs->ns_sufile);
 150                 nilfs_mdt_destroy(nilfs->ns_cpfile);
 151                 nilfs_mdt_destroy(nilfs->ns_dat);
 152                 nilfs_mdt_destroy(nilfs->ns_gc_dat);
 153         }
 154         if (nilfs_init(nilfs)) {
 155                 nilfs_destroy_gccache(nilfs);
 156                 brelse(nilfs->ns_sbh[0]);
 157                 brelse(nilfs->ns_sbh[1]);
 158         }
 159         kfree(nilfs);
 160 }
 161
 162 static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block)
 163 {
 164         struct buffer_head *bh_sr;
 165         struct nilfs_super_root *raw_sr;
 166         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 167         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
 168         unsigned inode_size;
 169         int err;
 170
 171         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
 172         if (unlikely(err))
 173                 return err;
 174
 175         down_read(&nilfs->ns_sem);
 176         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 177         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 178         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 179         up_read(&nilfs->ns_sem);
 180
 181         inode_size = nilfs->ns_inode_size;
 182
 183         err = -ENOMEM;
 184         nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size);
 185         if (unlikely(!nilfs->ns_dat))
 186                 goto failed;
 187
 188         nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size);
 189         if (unlikely(!nilfs->ns_gc_dat))
 190                 goto failed_dat;
 191
 192         nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size);
 193         if (unlikely(!nilfs->ns_cpfile))
 194                 goto failed_gc_dat;
 195
 196         nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
 197         if (unlikely(!nilfs->ns_sufile))
 198                 goto failed_cpfile;
 199
 200         nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
 201
 202         err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
 203                              NILFS_SR_DAT_OFFSET(inode_size));
 204         if (unlikely(err))
 205                 goto failed_sufile;
 206
 207         err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
 208                                 NILFS_SR_CPFILE_OFFSET(inode_size));
 209         if (unlikely(err))
 210                 goto failed_sufile;
 211
 212         err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
 213                                 NILFS_SR_SUFILE_OFFSET(inode_size));
 214         if (unlikely(err))
 215                 goto failed_sufile;
 216
 217         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 218         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 219
 220  failed:
 221         brelse(bh_sr);
 222         return err;
 223
 224  failed_sufile:
 225         nilfs_mdt_destroy(nilfs->ns_sufile);
 226
 227  failed_cpfile:
 228         nilfs_mdt_destroy(nilfs->ns_cpfile);
 229
 230  failed_gc_dat:
 231         nilfs_mdt_destroy(nilfs->ns_gc_dat);
 232
 233  failed_dat:
 234         nilfs_mdt_destroy(nilfs->ns_dat);
 235         goto failed;
 236 }
 237
 238 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 239 {
 240         memset(ri, 0, sizeof(*ri));
 241         INIT_LIST_HEAD(&ri->ri_used_segments);
 242 }
 243
 244 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 245 {
 246         nilfs_dispose_segment_list(&ri->ri_used_segments);
 247 }
 248
 249 /**
 250  * load_nilfs - load and recover the nilfs
 251  * @nilfs: the_nilfs structure to be released
 252  * @sbi: nilfs_sb_info used to recover past segment
 253  *
 254  * load_nilfs() searches and load the latest super root,
 255  * attaches the last segment, and does recovery if needed.
 256  * The caller must call this exclusively for simultaneous mounts.
 257  */
 258 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
 259 {
 260         struct nilfs_recovery_info ri;
 261         unsigned int s_flags = sbi->s_super->s_flags;
 262         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 263         int valid_fs = nilfs_valid_fs(nilfs);
 264         int err;
 265
 266         if (nilfs_loaded(nilfs)) {
 267                 if (valid_fs ||
 268                     ((s_flags & MS_RDONLY) && nilfs_test_opt(sbi, NORECOVERY)))
 269                         return 0;
 270                 printk(KERN_ERR "NILFS: the filesystem is in an incomplete "
 271                        "recovery state.\n");
 272                 return -EINVAL;
 273         }
 274
 275         if (!valid_fs) {
 276                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 277                 if (s_flags & MS_RDONLY) {
 278                         printk(KERN_INFO "NILFS: INFO: recovery "
 279                                "required for readonly filesystem.\n");
 280                         printk(KERN_INFO "NILFS: write access will "
 281                                "be enabled during recovery.\n");
 282                 }
 283         }
 284
 285         nilfs_init_recovery_info(&ri);
 286
 287         err = nilfs_search_super_root(nilfs, &ri);
 288         if (unlikely(err)) {
 289                 printk(KERN_ERR "NILFS: error searching super root.\n");
 290                 goto failed;
 291         }
 292
 293         err = nilfs_load_super_root(nilfs, ri.ri_super_root);
 294         if (unlikely(err)) {
 295                 printk(KERN_ERR "NILFS: error loading super root.\n");
 296                 goto failed;
 297         }
 298
 299         if (valid_fs)
 300                 goto skip_recovery;
 301
 302         if (s_flags & MS_RDONLY) {
 303                 if (nilfs_test_opt(sbi, NORECOVERY)) {
 304                         printk(KERN_INFO "NILFS: norecovery option specified. "
 305                                "skipping roll-forward recovery\n");
 306                         goto skip_recovery;
 307                 }
 308                 if (really_read_only) {
 309                         printk(KERN_ERR "NILFS: write access "
 310                                "unavailable, cannot proceed.\n");
 311                         err = -EROFS;
 312                         goto failed_unload;
 313                 }
 314                 sbi->s_super->s_flags &= ~MS_RDONLY;
 315         } else if (nilfs_test_opt(sbi, NORECOVERY)) {
 316                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
 317                        "option was specified for a read/write mount\n");
 318                 err = -EINVAL;
 319                 goto failed_unload;
 320         }
 321
 322         err = nilfs_salvage_orphan_logs(nilfs, sbi, &ri);
 323         if (err)
 324                 goto failed_unload;
 325
 326         down_write(&nilfs->ns_sem);
 327         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
 328         err = nilfs_cleanup_super(sbi);
 329         up_write(&nilfs->ns_sem);
 330
 331         if (err) {
 332                 printk(KERN_ERR "NILFS: failed to update super block. "
 333                        "recovery unfinished.\n");
 334                 goto failed_unload;
 335         }
 336         printk(KERN_INFO "NILFS: recovery complete.\n");
 337
 338  skip_recovery:
 339         set_nilfs_loaded(nilfs);
 340         nilfs_clear_recovery_info(&ri);
 341         sbi->s_super->s_flags = s_flags;
 342         return 0;
 343
 344  failed_unload:
 345         nilfs_mdt_destroy(nilfs->ns_cpfile);
 346         nilfs_mdt_destroy(nilfs->ns_sufile);
 347         nilfs_mdt_destroy(nilfs->ns_dat);
 348
 349  failed:
 350         nilfs_clear_recovery_info(&ri);
 351         sbi->s_super->s_flags = s_flags;
 352         return err;
 353 }
 354
 355 static unsigned long long nilfs_max_size(unsigned int blkbits)
 356 {
 357         unsigned int max_bits;
 358         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 359
 360         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 361         if (max_bits < 64)
 362                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 363         return res;
 364 }
 365
 366 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 367                                    struct nilfs_super_block *sbp)
 368 {
 369         if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
 370                 printk(KERN_ERR "NILFS: revision mismatch "
 371                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
 372                        "Please check the version of mkfs.nilfs.\n",
 373                        le32_to_cpu(sbp->s_rev_level),
 374                        le16_to_cpu(sbp->s_minor_rev_level),
 375                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
 376                 return -EINVAL;
 377         }
 378         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 379         if (nilfs->ns_sbsize > BLOCK_SIZE)
 380                 return -EINVAL;
 381
 382         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 383         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 384
 385         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 386         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 387                 printk(KERN_ERR "NILFS: too short segment.\n");
 388                 return -EINVAL;
 389         }
 390
 391         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 392         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
 393         nilfs->ns_r_segments_percentage =
 394                 le32_to_cpu(sbp->s_r_segments_percentage);
 395         nilfs->ns_nrsvsegs =
 396                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 397                       DIV_ROUND_UP(nilfs->ns_nsegments *
 398                                    nilfs->ns_r_segments_percentage, 100));
 399         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 400         return 0;
 401 }
 402
 403 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 404 {
 405         static unsigned char sum[4];
 406         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 407         size_t bytes;
 408         u32 crc;
 409
 410         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 411                 return 0;
 412         bytes = le16_to_cpu(sbp->s_bytes);
 413         if (bytes > BLOCK_SIZE)
 414                 return 0;
 415         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 416                        sumoff);
 417         crc = crc32_le(crc, sum, 4);
 418         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 419                        bytes - sumoff - 4);
 420         return crc == le32_to_cpu(sbp->s_sum);
 421 }
 422
 423 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 424 {
 425         return offset < ((le64_to_cpu(sbp->s_nsegments) *
 426                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
 427                          (le32_to_cpu(sbp->s_log_block_size) + 10));
 428 }
 429
 430 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 431 {
 432         int i;
 433
 434         for (i = 0; i < 2; i++) {
 435                 if (nilfs->ns_sbp[i]) {
 436                         brelse(nilfs->ns_sbh[i]);
 437                         nilfs->ns_sbh[i] = NULL;
 438                         nilfs->ns_sbp[i] = NULL;
 439                 }
 440         }
 441 }
 442
 443 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 444 {
 445         brelse(nilfs->ns_sbh[0]);
 446         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 447         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 448         nilfs->ns_sbh[1] = NULL;
 449         nilfs->ns_sbp[1] = NULL;
 450 }
 451
 452 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 453 {
 454         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 455         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 456
 457         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 458         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 459         nilfs->ns_sbh[1] = tsbh;
 460         nilfs->ns_sbp[1] = tsbp;
 461 }
 462
 463 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 464                                   struct super_block *sb, int blocksize,
 465                                   struct nilfs_super_block **sbpp)
 466 {
 467         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 468         struct buffer_head **sbh = nilfs->ns_sbh;
 469         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
 470         int valid[2], swp = 0;
 471
 472         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 473                                         &sbh[0]);
 474         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 475
 476         if (!sbp[0]) {
 477                 if (!sbp[1]) {
 478                         printk(KERN_ERR "NILFS: unable to read superblock\n");
 479                         return -EIO;
 480                 }
 481                 printk(KERN_WARNING
 482                        "NILFS warning: unable to read primary superblock\n");
 483         } else if (!sbp[1])
 484                 printk(KERN_WARNING
 485                        "NILFS warning: unable to read secondary superblock\n");
 486
 487         /*
 488          * Compare two super blocks and set 1 in swp if the secondary
 489          * super block is valid and newer.  Otherwise, set 0 in swp.
 490          */
 491         valid[0] = nilfs_valid_sb(sbp[0]);
 492         valid[1] = nilfs_valid_sb(sbp[1]);
 493         swp = valid[1] && (!valid[0] ||
 494                            le64_to_cpu(sbp[1]->s_last_cno) >
 495                            le64_to_cpu(sbp[0]->s_last_cno));
 496
 497         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 498                 brelse(sbh[1]);
 499                 sbh[1] = NULL;
 500                 sbp[1] = NULL;
 501                 swp = 0;
 502         }
 503         if (!valid[swp]) {
 504                 nilfs_release_super_block(nilfs);
 505                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
 506                        sb->s_id);
 507                 return -EINVAL;
 508         }
 509
 510         if (swp) {
 511                 printk(KERN_WARNING "NILFS warning: broken superblock. "
 512                        "using spare superblock.\n");
 513                 nilfs_swap_super_block(nilfs);
 514         }
 515
 516         nilfs->ns_sbwcount = 0;
 517         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 518         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 519         *sbpp = sbp[0];
 520         return 0;
 521 }
 522
 523 /**
 524  * init_nilfs - initialize a NILFS instance.
 525  * @nilfs: the_nilfs structure
 526  * @sbi: nilfs_sb_info
 527  * @sb: super block
 528  * @data: mount options
 529  *
 530  * init_nilfs() performs common initialization per block device (e.g.
 531  * reading the super block, getting disk layout information, initializing
 532  * shared fields in the_nilfs). It takes on some portion of the jobs
 533  * typically done by a fill_super() routine. This division arises from
 534  * the nature that multiple NILFS instances may be simultaneously
 535  * mounted on a device.
 536  * For multiple mounts on the same device, only the first mount
 537  * invokes these tasks.
 538  *
 539  * Return Value: On success, 0 is returned. On error, a negative error
 540  * code is returned.
 541  */
 542 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 543 {
 544         struct super_block *sb = sbi->s_super;
 545         struct nilfs_super_block *sbp;
 546         struct backing_dev_info *bdi;
 547         int blocksize;
 548         int err;
 549
 550         down_write(&nilfs->ns_sem);
 551         if (nilfs_init(nilfs)) {
 552                 /* Load values from existing the_nilfs */
 553                 sbp = nilfs->ns_sbp[0];
 554                 err = nilfs_store_magic_and_option(sb, sbp, data);
 555                 if (err)
 556                         goto out;
 557
 558                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 559                 if (sb->s_blocksize != blocksize &&
 560                     !sb_set_blocksize(sb, blocksize)) {
 561                         printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
 562                                blocksize);
 563                         err = -EINVAL;
 564                 }
 565                 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 566                 goto out;
 567         }
 568
 569         blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
 570         if (!blocksize) {
 571                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
 572                 err = -EINVAL;
 573                 goto out;
 574         }
 575         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 576         if (err)
 577                 goto out;
 578
 579         err = nilfs_store_magic_and_option(sb, sbp, data);
 580         if (err)
 581                 goto failed_sbh;
 582
 583         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 584         if (sb->s_blocksize != blocksize) {
 585                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 586
 587                 if (blocksize < hw_blocksize) {
 588                         printk(KERN_ERR
 589                                "NILFS: blocksize %d too small for device "
 590                                "(sector-size = %d).\n",
 591                                blocksize, hw_blocksize);
 592                         err = -EINVAL;
 593                         goto failed_sbh;
 594                 }
 595                 nilfs_release_super_block(nilfs);
 596                 sb_set_blocksize(sb, blocksize);
 597
 598                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 599                 if (err)
 600                         goto out;
 601                         /* not failed_sbh; sbh is released automatically
 602                            when reloading fails. */
 603         }
 604         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 605         nilfs->ns_blocksize = blocksize;
 606
 607         err = nilfs_store_disk_layout(nilfs, sbp);
 608         if (err)
 609                 goto failed_sbh;
 610
 611         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 612
 613         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 614
 615         bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
 616         nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
 617
 618         /* Finding last segment */
 619         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 620         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 621         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 622
 623         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 624         nilfs->ns_segnum =
 625                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 626         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 627         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 628                 printk(KERN_ERR "NILFS invalid last segment number.\n");
 629                 err = -EINVAL;
 630                 goto failed_sbh;
 631         }
 632
 633         /* Initialize gcinode cache */
 634         err = nilfs_init_gccache(nilfs);
 635         if (err)
 636                 goto failed_sbh;
 637
 638         set_nilfs_init(nilfs);
 639         err = 0;
 640  out:
 641         up_write(&nilfs->ns_sem);
 642         return err;
 643
 644  failed_sbh:
 645         nilfs_release_super_block(nilfs);
 646         goto out;
 647 }
 648
 649 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
 650                             size_t nsegs)
 651 {
 652         sector_t seg_start, seg_end;
 653         sector_t start = 0, nblocks = 0;
 654         unsigned int sects_per_block;
 655         __u64 *sn;
 656         int ret = 0;
 657
 658         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
 659                 bdev_logical_block_size(nilfs->ns_bdev);
 660         for (sn = segnump; sn < segnump + nsegs; sn++) {
 661                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
 662
 663                 if (!nblocks) {
 664                         start = seg_start;
 665                         nblocks = seg_end - seg_start + 1;
 666                 } else if (start + nblocks == seg_start) {
 667                         nblocks += seg_end - seg_start + 1;
 668                 } else {
 669                         ret = blkdev_issue_discard(nilfs->ns_bdev,
 670                                                    start * sects_per_block,
 671                                                    nblocks * sects_per_block,
 672                                                    GFP_NOFS,
 673                                                    BLKDEV_IFL_BARRIER);
 674                         if (ret < 0)
 675                                 return ret;
 676                         nblocks = 0;
 677                 }
 678         }
 679         if (nblocks)
 680                 ret = blkdev_issue_discard(nilfs->ns_bdev,
 681                                            start * sects_per_block,
 682                                            nblocks * sects_per_block,
 683                                            GFP_NOFS, BLKDEV_IFL_BARRIER);
 684         return ret;
 685 }
 686
 687 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 688 {
 689         struct inode *dat = nilfs_dat_inode(nilfs);
 690         unsigned long ncleansegs;
 691
 692         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 693         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 694         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 695         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 696         return 0;
 697 }
 698
 699 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 700 {
 701         unsigned long ncleansegs, nincsegs;
 702
 703         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 704         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 705                 nilfs->ns_blocks_per_segment + 1;
 706
 707         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
 708 }
 709
 710 /**
 711  * nilfs_find_sbinfo - find existing nilfs_sb_info structure
 712  * @nilfs: nilfs object
 713  * @rw_mount: mount type (non-zero value for read/write mount)
 714  * @cno: checkpoint number (zero for read-only mount)
 715  *
 716  * nilfs_find_sbinfo() returns the nilfs_sb_info structure which
 717  * @rw_mount and @cno (in case of snapshots) matched.  If no instance
 718  * was found, NULL is returned.  Although the super block instance can
 719  * be unmounted after this function returns, the nilfs_sb_info struct
 720  * is kept on memory until nilfs_put_sbinfo() is called.
 721  */
 722 struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *nilfs,
 723                                         int rw_mount, __u64 cno)
 724 {
 725         struct nilfs_sb_info *sbi;
 726
 727         down_read(&nilfs->ns_super_sem);
 728         /*
 729          * The SNAPSHOT flag and sb->s_flags are supposed to be
 730          * protected with nilfs->ns_super_sem.
 731          */
 732         sbi = nilfs->ns_current;
 733         if (rw_mount) {
 734                 if (sbi && !(sbi->s_super->s_flags & MS_RDONLY))
 735                         goto found; /* read/write mount */
 736                 else
 737                         goto out;
 738         } else if (cno == 0) {
 739                 if (sbi && (sbi->s_super->s_flags & MS_RDONLY))
 740                         goto found; /* read-only mount */
 741                 else
 742                         goto out;
 743         }
 744
 745         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
 746                 if (nilfs_test_opt(sbi, SNAPSHOT) &&
 747                     sbi->s_snapshot_cno == cno)
 748                         goto found; /* snapshot mount */
 749         }
 750  out:
 751         up_read(&nilfs->ns_super_sem);
 752         return NULL;
 753
 754  found:
 755         atomic_inc(&sbi->s_count);
 756         up_read(&nilfs->ns_super_sem);
 757         return sbi;
 758 }
 759
 760 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
 761                                 int snapshot_mount)
 762 {
 763         struct nilfs_sb_info *sbi;
 764         int ret = 0;
 765
 766         down_read(&nilfs->ns_super_sem);
 767         if (cno == 0 || cno > nilfs->ns_cno)
 768                 goto out_unlock;
 769
 770         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
 771                 if (sbi->s_snapshot_cno == cno &&
 772                     (!snapshot_mount || nilfs_test_opt(sbi, SNAPSHOT))) {
 773                                         /* exclude read-only mounts */
 774                         ret++;
 775                         break;
 776                 }
 777         }
 778         /* for protecting recent checkpoints */
 779         if (cno >= nilfs_last_cno(nilfs))
 780                 ret++;
 781
 782  out_unlock:
 783         up_read(&nilfs->ns_super_sem);
 784         return ret;
 785 }