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,
 163                                  struct nilfs_sb_info *sbi, sector_t sr_block)
 164 {
 165         struct buffer_head *bh_sr;
 166         struct nilfs_super_root *raw_sr;
 167         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 168         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
 169         unsigned inode_size;
 170         int err;
 171
 172         err = nilfs_read_super_root_block(sbi->s_super, sr_block, &bh_sr, 1);
 173         if (unlikely(err))
 174                 return err;
 175
 176         down_read(&nilfs->ns_sem);
 177         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 178         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 179         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 180         up_read(&nilfs->ns_sem);
 181
 182         inode_size = nilfs->ns_inode_size;
 183
 184         err = -ENOMEM;
 185         nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size);
 186         if (unlikely(!nilfs->ns_dat))
 187                 goto failed;
 188
 189         nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size);
 190         if (unlikely(!nilfs->ns_gc_dat))
 191                 goto failed_dat;
 192
 193         nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size);
 194         if (unlikely(!nilfs->ns_cpfile))
 195                 goto failed_gc_dat;
 196
 197         nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
 198         if (unlikely(!nilfs->ns_sufile))
 199                 goto failed_cpfile;
 200
 201         nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
 202
 203         err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
 204                              NILFS_SR_DAT_OFFSET(inode_size));
 205         if (unlikely(err))
 206                 goto failed_sufile;
 207
 208         err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
 209                                 NILFS_SR_CPFILE_OFFSET(inode_size));
 210         if (unlikely(err))
 211                 goto failed_sufile;
 212
 213         err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
 214                                 NILFS_SR_SUFILE_OFFSET(inode_size));
 215         if (unlikely(err))
 216                 goto failed_sufile;
 217
 218         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 219         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 220
 221  failed:
 222         brelse(bh_sr);
 223         return err;
 224
 225  failed_sufile:
 226         nilfs_mdt_destroy(nilfs->ns_sufile);
 227
 228  failed_cpfile:
 229         nilfs_mdt_destroy(nilfs->ns_cpfile);
 230
 231  failed_gc_dat:
 232         nilfs_mdt_destroy(nilfs->ns_gc_dat);
 233
 234  failed_dat:
 235         nilfs_mdt_destroy(nilfs->ns_dat);
 236         goto failed;
 237 }
 238
 239 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 240 {
 241         memset(ri, 0, sizeof(*ri));
 242         INIT_LIST_HEAD(&ri->ri_used_segments);
 243 }
 244
 245 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 246 {
 247         nilfs_dispose_segment_list(&ri->ri_used_segments);
 248 }
 249
 250 /**
 251  * load_nilfs - load and recover the nilfs
 252  * @nilfs: the_nilfs structure to be released
 253  * @sbi: nilfs_sb_info used to recover past segment
 254  *
 255  * load_nilfs() searches and load the latest super root,
 256  * attaches the last segment, and does recovery if needed.
 257  * The caller must call this exclusively for simultaneous mounts.
 258  */
 259 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
 260 {
 261         struct nilfs_recovery_info ri;
 262         unsigned int s_flags = sbi->s_super->s_flags;
 263         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 264         int valid_fs = nilfs_valid_fs(nilfs);
 265         int err;
 266
 267         if (nilfs_loaded(nilfs)) {
 268                 if (valid_fs ||
 269                     ((s_flags & MS_RDONLY) && nilfs_test_opt(sbi, NORECOVERY)))
 270                         return 0;
 271                 printk(KERN_ERR "NILFS: the filesystem is in an incomplete "
 272                        "recovery state.\n");
 273                 return -EINVAL;
 274         }
 275
 276         if (!valid_fs) {
 277                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 278                 if (s_flags & MS_RDONLY) {
 279                         printk(KERN_INFO "NILFS: INFO: recovery "
 280                                "required for readonly filesystem.\n");
 281                         printk(KERN_INFO "NILFS: write access will "
 282                                "be enabled during recovery.\n");
 283                 }
 284         }
 285
 286         nilfs_init_recovery_info(&ri);
 287
 288         err = nilfs_search_super_root(nilfs, sbi, &ri);
 289         if (unlikely(err)) {
 290                 printk(KERN_ERR "NILFS: error searching super root.\n");
 291                 goto failed;
 292         }
 293
 294         err = nilfs_load_super_root(nilfs, sbi, ri.ri_super_root);
 295         if (unlikely(err)) {
 296                 printk(KERN_ERR "NILFS: error loading super root.\n");
 297                 goto failed;
 298         }
 299
 300         if (valid_fs)
 301                 goto skip_recovery;
 302
 303         if (s_flags & MS_RDONLY) {
 304                 if (nilfs_test_opt(sbi, NORECOVERY)) {
 305                         printk(KERN_INFO "NILFS: norecovery option specified. "
 306                                "skipping roll-forward recovery\n");
 307                         goto skip_recovery;
 308                 }
 309                 if (really_read_only) {
 310                         printk(KERN_ERR "NILFS: write access "
 311                                "unavailable, cannot proceed.\n");
 312                         err = -EROFS;
 313                         goto failed_unload;
 314                 }
 315                 sbi->s_super->s_flags &= ~MS_RDONLY;
 316         } else if (nilfs_test_opt(sbi, NORECOVERY)) {
 317                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
 318                        "option was specified for a read/write mount\n");
 319                 err = -EINVAL;
 320                 goto failed_unload;
 321         }
 322
 323         err = nilfs_recover_logical_segments(nilfs, sbi, &ri);
 324         if (err)
 325                 goto failed_unload;
 326
 327         down_write(&nilfs->ns_sem);
 328         nilfs->ns_mount_state |= NILFS_VALID_FS;
 329         nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 330         err = nilfs_commit_super(sbi, 1);
 331         up_write(&nilfs->ns_sem);
 332
 333         if (err) {
 334                 printk(KERN_ERR "NILFS: failed to update super block. "
 335                        "recovery unfinished.\n");
 336                 goto failed_unload;
 337         }
 338         printk(KERN_INFO "NILFS: recovery complete.\n");
 339
 340  skip_recovery:
 341         set_nilfs_loaded(nilfs);
 342         nilfs_clear_recovery_info(&ri);
 343         sbi->s_super->s_flags = s_flags;
 344         return 0;
 345
 346  failed_unload:
 347         nilfs_mdt_destroy(nilfs->ns_cpfile);
 348         nilfs_mdt_destroy(nilfs->ns_sufile);
 349         nilfs_mdt_destroy(nilfs->ns_dat);
 350
 351  failed:
 352         nilfs_clear_recovery_info(&ri);
 353         sbi->s_super->s_flags = s_flags;
 354         return err;
 355 }
 356
 357 static unsigned long long nilfs_max_size(unsigned int blkbits)
 358 {
 359         unsigned int max_bits;
 360         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 361
 362         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 363         if (max_bits < 64)
 364                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 365         return res;
 366 }
 367
 368 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 369                                    struct nilfs_super_block *sbp)
 370 {
 371         if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
 372                 printk(KERN_ERR "NILFS: revision mismatch "
 373                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
 374                        "Please check the version of mkfs.nilfs.\n",
 375                        le32_to_cpu(sbp->s_rev_level),
 376                        le16_to_cpu(sbp->s_minor_rev_level),
 377                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
 378                 return -EINVAL;
 379         }
 380         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 381         if (nilfs->ns_sbsize > BLOCK_SIZE)
 382                 return -EINVAL;
 383
 384         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 385         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 386
 387         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 388         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 389                 printk(KERN_ERR "NILFS: too short segment.\n");
 390                 return -EINVAL;
 391         }
 392
 393         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 394         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
 395         nilfs->ns_r_segments_percentage =
 396                 le32_to_cpu(sbp->s_r_segments_percentage);
 397         nilfs->ns_nrsvsegs =
 398                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 399                       DIV_ROUND_UP(nilfs->ns_nsegments *
 400                                    nilfs->ns_r_segments_percentage, 100));
 401         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 402         return 0;
 403 }
 404
 405 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 406 {
 407         static unsigned char sum[4];
 408         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 409         size_t bytes;
 410         u32 crc;
 411
 412         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 413                 return 0;
 414         bytes = le16_to_cpu(sbp->s_bytes);
 415         if (bytes > BLOCK_SIZE)
 416                 return 0;
 417         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 418                        sumoff);
 419         crc = crc32_le(crc, sum, 4);
 420         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 421                        bytes - sumoff - 4);
 422         return crc == le32_to_cpu(sbp->s_sum);
 423 }
 424
 425 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 426 {
 427         return offset < ((le64_to_cpu(sbp->s_nsegments) *
 428                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
 429                          (le32_to_cpu(sbp->s_log_block_size) + 10));
 430 }
 431
 432 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 433 {
 434         int i;
 435
 436         for (i = 0; i < 2; i++) {
 437                 if (nilfs->ns_sbp[i]) {
 438                         brelse(nilfs->ns_sbh[i]);
 439                         nilfs->ns_sbh[i] = NULL;
 440                         nilfs->ns_sbp[i] = NULL;
 441                 }
 442         }
 443 }
 444
 445 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 446 {
 447         brelse(nilfs->ns_sbh[0]);
 448         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 449         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 450         nilfs->ns_sbh[1] = NULL;
 451         nilfs->ns_sbp[1] = NULL;
 452 }
 453
 454 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 455 {
 456         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 457         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 458
 459         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 460         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 461         nilfs->ns_sbh[1] = tsbh;
 462         nilfs->ns_sbp[1] = tsbp;
 463 }
 464
 465 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 466                                   struct super_block *sb, int blocksize,
 467                                   struct nilfs_super_block **sbpp)
 468 {
 469         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 470         struct buffer_head **sbh = nilfs->ns_sbh;
 471         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
 472         int valid[2], swp = 0;
 473
 474         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 475                                         &sbh[0]);
 476         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 477
 478         if (!sbp[0]) {
 479                 if (!sbp[1]) {
 480                         printk(KERN_ERR "NILFS: unable to read superblock\n");
 481                         return -EIO;
 482                 }
 483                 printk(KERN_WARNING
 484                        "NILFS warning: unable to read primary superblock\n");
 485         } else if (!sbp[1])
 486                 printk(KERN_WARNING
 487                        "NILFS warning: unable to read secondary superblock\n");
 488
 489         valid[0] = nilfs_valid_sb(sbp[0]);
 490         valid[1] = nilfs_valid_sb(sbp[1]);
 491         swp = valid[1] &&
 492                 (!valid[0] ||
 493                  le64_to_cpu(sbp[1]->s_wtime) > le64_to_cpu(sbp[0]->s_wtime));
 494
 495         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 496                 brelse(sbh[1]);
 497                 sbh[1] = NULL;
 498                 sbp[1] = NULL;
 499                 swp = 0;
 500         }
 501         if (!valid[swp]) {
 502                 nilfs_release_super_block(nilfs);
 503                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
 504                        sb->s_id);
 505                 return -EINVAL;
 506         }
 507
 508         if (swp) {
 509                 printk(KERN_WARNING "NILFS warning: broken superblock. "
 510                        "using spare superblock.\n");
 511                 nilfs_swap_super_block(nilfs);
 512         }
 513
 514         nilfs->ns_sbwtime[0] = le64_to_cpu(sbp[0]->s_wtime);
 515         nilfs->ns_sbwtime[1] = valid[!swp] ? le64_to_cpu(sbp[1]->s_wtime) : 0;
 516         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 517         *sbpp = sbp[0];
 518         return 0;
 519 }
 520
 521 /**
 522  * init_nilfs - initialize a NILFS instance.
 523  * @nilfs: the_nilfs structure
 524  * @sbi: nilfs_sb_info
 525  * @sb: super block
 526  * @data: mount options
 527  *
 528  * init_nilfs() performs common initialization per block device (e.g.
 529  * reading the super block, getting disk layout information, initializing
 530  * shared fields in the_nilfs). It takes on some portion of the jobs
 531  * typically done by a fill_super() routine. This division arises from
 532  * the nature that multiple NILFS instances may be simultaneously
 533  * mounted on a device.
 534  * For multiple mounts on the same device, only the first mount
 535  * invokes these tasks.
 536  *
 537  * Return Value: On success, 0 is returned. On error, a negative error
 538  * code is returned.
 539  */
 540 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 541 {
 542         struct super_block *sb = sbi->s_super;
 543         struct nilfs_super_block *sbp;
 544         struct backing_dev_info *bdi;
 545         int blocksize;
 546         int err;
 547
 548         down_write(&nilfs->ns_sem);
 549         if (nilfs_init(nilfs)) {
 550                 /* Load values from existing the_nilfs */
 551                 sbp = nilfs->ns_sbp[0];
 552                 err = nilfs_store_magic_and_option(sb, sbp, data);
 553                 if (err)
 554                         goto out;
 555
 556                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 557                 if (sb->s_blocksize != blocksize &&
 558                     !sb_set_blocksize(sb, blocksize)) {
 559                         printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
 560                                blocksize);
 561                         err = -EINVAL;
 562                 }
 563                 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 564                 goto out;
 565         }
 566
 567         blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
 568         if (!blocksize) {
 569                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
 570                 err = -EINVAL;
 571                 goto out;
 572         }
 573         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 574         if (err)
 575                 goto out;
 576
 577         err = nilfs_store_magic_and_option(sb, sbp, data);
 578         if (err)
 579                 goto failed_sbh;
 580
 581         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 582         if (sb->s_blocksize != blocksize) {
 583                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 584
 585                 if (blocksize < hw_blocksize) {
 586                         printk(KERN_ERR
 587                                "NILFS: blocksize %d too small for device "
 588                                "(sector-size = %d).\n",
 589                                blocksize, hw_blocksize);
 590                         err = -EINVAL;
 591                         goto failed_sbh;
 592                 }
 593                 nilfs_release_super_block(nilfs);
 594                 sb_set_blocksize(sb, blocksize);
 595
 596                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 597                 if (err)
 598                         goto out;
 599                         /* not failed_sbh; sbh is released automatically
 600                            when reloading fails. */
 601         }
 602         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 603
 604         err = nilfs_store_disk_layout(nilfs, sbp);
 605         if (err)
 606                 goto failed_sbh;
 607
 608         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 609
 610         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 611
 612         bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
 613         nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
 614
 615         /* Finding last segment */
 616         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 617         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 618         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 619
 620         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 621         nilfs->ns_segnum =
 622                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 623         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 624         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 625                 printk(KERN_ERR "NILFS invalid last segment number.\n");
 626                 err = -EINVAL;
 627                 goto failed_sbh;
 628         }
 629         /* Dummy values  */
 630         nilfs->ns_free_segments_count =
 631                 nilfs->ns_nsegments - (nilfs->ns_segnum + 1);
 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                                                    DISCARD_FL_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, DISCARD_FL_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 }