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  * Written by Ryusuke Konishi.
  17  *
  18  */
  19
  20 #include <linux/buffer_head.h>
  21 #include <linux/slab.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/backing-dev.h>
  24 #include <linux/random.h>
  25 #include <linux/log2.h>
  26 #include <linux/crc32.h>
  27 #include "nilfs.h"
  28 #include "segment.h"
  29 #include "alloc.h"
  30 #include "cpfile.h"
  31 #include "sufile.h"
  32 #include "dat.h"
  33 #include "segbuf.h"
  34
  35
  36 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
  37
  38 void nilfs_set_last_segment(struct the_nilfs *nilfs,
  39                             sector_t start_blocknr, u64 seq, __u64 cno)
  40 {
  41         spin_lock(&nilfs->ns_last_segment_lock);
  42         nilfs->ns_last_pseg = start_blocknr;
  43         nilfs->ns_last_seq = seq;
  44         nilfs->ns_last_cno = cno;
  45
  46         if (!nilfs_sb_dirty(nilfs)) {
  47                 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
  48                         goto stay_cursor;
  49
  50                 set_nilfs_sb_dirty(nilfs);
  51         }
  52         nilfs->ns_prev_seq = nilfs->ns_last_seq;
  53
  54  stay_cursor:
  55         spin_unlock(&nilfs->ns_last_segment_lock);
  56 }
  57
  58 /**
  59  * alloc_nilfs - allocate a nilfs object
  60  * @sb: super block instance
  61  *
  62  * Return Value: On success, pointer to the_nilfs is returned.
  63  * On error, NULL is returned.
  64  */
  65 struct the_nilfs *alloc_nilfs(struct super_block *sb)
  66 {
  67         struct the_nilfs *nilfs;
  68
  69         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
  70         if (!nilfs)
  71                 return NULL;
  72
  73         nilfs->ns_sb = sb;
  74         nilfs->ns_bdev = sb->s_bdev;
  75         atomic_set(&nilfs->ns_ndirtyblks, 0);
  76         init_rwsem(&nilfs->ns_sem);
  77         mutex_init(&nilfs->ns_snapshot_mount_mutex);
  78         INIT_LIST_HEAD(&nilfs->ns_dirty_files);
  79         INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
  80         spin_lock_init(&nilfs->ns_inode_lock);
  81         spin_lock_init(&nilfs->ns_next_gen_lock);
  82         spin_lock_init(&nilfs->ns_last_segment_lock);
  83         nilfs->ns_cptree = RB_ROOT;
  84         spin_lock_init(&nilfs->ns_cptree_lock);
  85         init_rwsem(&nilfs->ns_segctor_sem);
  86         nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
  87
  88         return nilfs;
  89 }
  90
  91 /**
  92  * destroy_nilfs - destroy nilfs object
  93  * @nilfs: nilfs object to be released
  94  */
  95 void destroy_nilfs(struct the_nilfs *nilfs)
  96 {
  97         might_sleep();
  98         if (nilfs_init(nilfs)) {
  99                 brelse(nilfs->ns_sbh[0]);
 100                 brelse(nilfs->ns_sbh[1]);
 101         }
 102         kfree(nilfs);
 103 }
 104
 105 static int nilfs_load_super_root(struct the_nilfs *nilfs,
 106                                  struct super_block *sb, sector_t sr_block)
 107 {
 108         struct buffer_head *bh_sr;
 109         struct nilfs_super_root *raw_sr;
 110         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 111         struct nilfs_inode *rawi;
 112         unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
 113         unsigned int inode_size;
 114         int err;
 115
 116         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
 117         if (unlikely(err))
 118                 return err;
 119
 120         down_read(&nilfs->ns_sem);
 121         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 122         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 123         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 124         up_read(&nilfs->ns_sem);
 125
 126         inode_size = nilfs->ns_inode_size;
 127
 128         rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
 129         err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
 130         if (err)
 131                 goto failed;
 132
 133         rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
 134         err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
 135         if (err)
 136                 goto failed_dat;
 137
 138         rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
 139         err = nilfs_sufile_read(sb, segment_usage_size, rawi,
 140                                 &nilfs->ns_sufile);
 141         if (err)
 142                 goto failed_cpfile;
 143
 144         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 145         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 146
 147  failed:
 148         brelse(bh_sr);
 149         return err;
 150
 151  failed_cpfile:
 152         iput(nilfs->ns_cpfile);
 153
 154  failed_dat:
 155         iput(nilfs->ns_dat);
 156         goto failed;
 157 }
 158
 159 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 160 {
 161         memset(ri, 0, sizeof(*ri));
 162         INIT_LIST_HEAD(&ri->ri_used_segments);
 163 }
 164
 165 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 166 {
 167         nilfs_dispose_segment_list(&ri->ri_used_segments);
 168 }
 169
 170 /**
 171  * nilfs_store_log_cursor - load log cursor from a super block
 172  * @nilfs: nilfs object
 173  * @sbp: buffer storing super block to be read
 174  *
 175  * nilfs_store_log_cursor() reads the last position of the log
 176  * containing a super root from a given super block, and initializes
 177  * relevant information on the nilfs object preparatory for log
 178  * scanning and recovery.
 179  */
 180 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
 181                                   struct nilfs_super_block *sbp)
 182 {
 183         int ret = 0;
 184
 185         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 186         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 187         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 188
 189         nilfs->ns_prev_seq = nilfs->ns_last_seq;
 190         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 191         nilfs->ns_segnum =
 192                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 193         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 194         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 195                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 196                           "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
 197                           (unsigned long long)nilfs->ns_segnum,
 198                           nilfs->ns_nsegments);
 199                 ret = -EINVAL;
 200         }
 201         return ret;
 202 }
 203
 204 /**
 205  * load_nilfs - load and recover the nilfs
 206  * @nilfs: the_nilfs structure to be released
 207  * @sb: super block isntance used to recover past segment
 208  *
 209  * load_nilfs() searches and load the latest super root,
 210  * attaches the last segment, and does recovery if needed.
 211  * The caller must call this exclusively for simultaneous mounts.
 212  */
 213 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
 214 {
 215         struct nilfs_recovery_info ri;
 216         unsigned int s_flags = sb->s_flags;
 217         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 218         int valid_fs = nilfs_valid_fs(nilfs);
 219         int err;
 220
 221         if (!valid_fs) {
 222                 nilfs_msg(sb, KERN_WARNING, "mounting unchecked fs");
 223                 if (s_flags & MS_RDONLY) {
 224                         nilfs_msg(sb, KERN_INFO,
 225                                   "recovery required for readonly filesystem");
 226                         nilfs_msg(sb, KERN_INFO,
 227                                   "write access will be enabled during recovery");
 228                 }
 229         }
 230
 231         nilfs_init_recovery_info(&ri);
 232
 233         err = nilfs_search_super_root(nilfs, &ri);
 234         if (unlikely(err)) {
 235                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
 236                 int blocksize;
 237
 238                 if (err != -EINVAL)
 239                         goto scan_error;
 240
 241                 if (!nilfs_valid_sb(sbp[1])) {
 242                         nilfs_msg(sb, KERN_WARNING,
 243                                   "unable to fall back to spare super block");
 244                         goto scan_error;
 245                 }
 246                 nilfs_msg(sb, KERN_INFO,
 247                           "trying rollback from an earlier position");
 248
 249                 /*
 250                  * restore super block with its spare and reconfigure
 251                  * relevant states of the nilfs object.
 252                  */
 253                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 254                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
 255                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 256
 257                 /* verify consistency between two super blocks */
 258                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
 259                 if (blocksize != nilfs->ns_blocksize) {
 260                         nilfs_msg(sb, KERN_WARNING,
 261                                   "blocksize differs between two super blocks (%d != %d)",
 262                                   blocksize, nilfs->ns_blocksize);
 263                         goto scan_error;
 264                 }
 265
 266                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
 267                 if (err)
 268                         goto scan_error;
 269
 270                 /* drop clean flag to allow roll-forward and recovery */
 271                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
 272                 valid_fs = 0;
 273
 274                 err = nilfs_search_super_root(nilfs, &ri);
 275                 if (err)
 276                         goto scan_error;
 277         }
 278
 279         err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
 280         if (unlikely(err)) {
 281                 nilfs_msg(sb, KERN_ERR, "error %d while loading super root",
 282                           err);
 283                 goto failed;
 284         }
 285
 286         err = nilfs_sysfs_create_device_group(sb);
 287         if (unlikely(err))
 288                 goto sysfs_error;
 289
 290         if (valid_fs)
 291                 goto skip_recovery;
 292
 293         if (s_flags & MS_RDONLY) {
 294                 __u64 features;
 295
 296                 if (nilfs_test_opt(nilfs, NORECOVERY)) {
 297                         nilfs_msg(sb, KERN_INFO,
 298                                   "norecovery option specified, skipping roll-forward recovery");
 299                         goto skip_recovery;
 300                 }
 301                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
 302                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
 303                 if (features) {
 304                         nilfs_msg(sb, KERN_ERR,
 305                                   "couldn't proceed with recovery because of unsupported optional features (%llx)",
 306                                   (unsigned long long)features);
 307                         err = -EROFS;
 308                         goto failed_unload;
 309                 }
 310                 if (really_read_only) {
 311                         nilfs_msg(sb, KERN_ERR,
 312                                   "write access unavailable, cannot proceed");
 313                         err = -EROFS;
 314                         goto failed_unload;
 315                 }
 316                 sb->s_flags &= ~MS_RDONLY;
 317         } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
 318                 nilfs_msg(sb, KERN_ERR,
 319                           "recovery cancelled because norecovery option was specified for a read/write mount");
 320                 err = -EINVAL;
 321                 goto failed_unload;
 322         }
 323
 324         err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
 325         if (err)
 326                 goto failed_unload;
 327
 328         down_write(&nilfs->ns_sem);
 329         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
 330         err = nilfs_cleanup_super(sb);
 331         up_write(&nilfs->ns_sem);
 332
 333         if (err) {
 334                 nilfs_msg(sb, KERN_ERR,
 335                           "error %d updating super block. recovery unfinished.",
 336                           err);
 337                 goto failed_unload;
 338         }
 339         nilfs_msg(sb, KERN_INFO, "recovery complete");
 340
 341  skip_recovery:
 342         nilfs_clear_recovery_info(&ri);
 343         sb->s_flags = s_flags;
 344         return 0;
 345
 346  scan_error:
 347         nilfs_msg(sb, KERN_ERR, "error %d while searching super root", err);
 348         goto failed;
 349
 350  failed_unload:
 351         nilfs_sysfs_delete_device_group(nilfs);
 352
 353  sysfs_error:
 354         iput(nilfs->ns_cpfile);
 355         iput(nilfs->ns_sufile);
 356         iput(nilfs->ns_dat);
 357
 358  failed:
 359         nilfs_clear_recovery_info(&ri);
 360         sb->s_flags = s_flags;
 361         return err;
 362 }
 363
 364 static unsigned long long nilfs_max_size(unsigned int blkbits)
 365 {
 366         unsigned int max_bits;
 367         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 368
 369         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 370         if (max_bits < 64)
 371                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 372         return res;
 373 }
 374
 375 /**
 376  * nilfs_nrsvsegs - calculate the number of reserved segments
 377  * @nilfs: nilfs object
 378  * @nsegs: total number of segments
 379  */
 380 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
 381 {
 382         return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 383                      DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
 384                                   100));
 385 }
 386
 387 /**
 388  * nilfs_max_segment_count - calculate the maximum number of segments
 389  * @nilfs: nilfs object
 390  */
 391 static u64 nilfs_max_segment_count(struct the_nilfs *nilfs)
 392 {
 393         u64 max_count = U64_MAX;
 394
 395         do_div(max_count, nilfs->ns_blocks_per_segment);
 396         return min_t(u64, max_count, ULONG_MAX);
 397 }
 398
 399 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
 400 {
 401         nilfs->ns_nsegments = nsegs;
 402         nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
 403 }
 404
 405 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 406                                    struct nilfs_super_block *sbp)
 407 {
 408         u64 nsegments, nblocks;
 409
 410         if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
 411                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 412                           "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
 413                           le32_to_cpu(sbp->s_rev_level),
 414                           le16_to_cpu(sbp->s_minor_rev_level),
 415                           NILFS_CURRENT_REV, NILFS_MINOR_REV);
 416                 return -EINVAL;
 417         }
 418         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 419         if (nilfs->ns_sbsize > BLOCK_SIZE)
 420                 return -EINVAL;
 421
 422         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 423         if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
 424                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 425                           "too large inode size: %d bytes",
 426                           nilfs->ns_inode_size);
 427                 return -EINVAL;
 428         } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
 429                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 430                           "too small inode size: %d bytes",
 431                           nilfs->ns_inode_size);
 432                 return -EINVAL;
 433         }
 434
 435         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 436
 437         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 438         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 439                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 440                           "too short segment: %lu blocks",
 441                           nilfs->ns_blocks_per_segment);
 442                 return -EINVAL;
 443         }
 444
 445         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 446         nilfs->ns_r_segments_percentage =
 447                 le32_to_cpu(sbp->s_r_segments_percentage);
 448         if (nilfs->ns_r_segments_percentage < 1 ||
 449             nilfs->ns_r_segments_percentage > 99) {
 450                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 451                           "invalid reserved segments percentage: %lu",
 452                           nilfs->ns_r_segments_percentage);
 453                 return -EINVAL;
 454         }
 455
 456         nsegments = le64_to_cpu(sbp->s_nsegments);
 457         if (nsegments > nilfs_max_segment_count(nilfs)) {
 458                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 459                           "segment count %llu exceeds upper limit (%llu segments)",
 460                           (unsigned long long)nsegments,
 461                           (unsigned long long)nilfs_max_segment_count(nilfs));
 462                 return -EINVAL;
 463         }
 464
 465         nblocks = (u64)i_size_read(nilfs->ns_sb->s_bdev->bd_inode) >>
 466                 nilfs->ns_sb->s_blocksize_bits;
 467         if (nblocks) {
 468                 u64 min_block_count = nsegments * nilfs->ns_blocks_per_segment;
 469                 /*
 470                  * To avoid failing to mount early device images without a
 471                  * second superblock, exclude that block count from the
 472                  * "min_block_count" calculation.
 473                  */
 474
 475                 if (nblocks < min_block_count) {
 476                         nilfs_msg(nilfs->ns_sb, KERN_ERR,
 477                                   "total number of segment blocks %llu exceeds device size (%llu blocks)",
 478                                   (unsigned long long)min_block_count,
 479                                   (unsigned long long)nblocks);
 480                         return -EINVAL;
 481                 }
 482         }
 483
 484         nilfs_set_nsegments(nilfs, nsegments);
 485         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 486         return 0;
 487 }
 488
 489 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 490 {
 491         static unsigned char sum[4];
 492         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 493         size_t bytes;
 494         u32 crc;
 495
 496         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 497                 return 0;
 498         bytes = le16_to_cpu(sbp->s_bytes);
 499         if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
 500                 return 0;
 501         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 502                        sumoff);
 503         crc = crc32_le(crc, sum, 4);
 504         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 505                        bytes - sumoff - 4);
 506         return crc == le32_to_cpu(sbp->s_sum);
 507 }
 508
 509 /**
 510  * nilfs_sb2_bad_offset - check the location of the second superblock
 511  * @sbp: superblock raw data buffer
 512  * @offset: byte offset of second superblock calculated from device size
 513  *
 514  * nilfs_sb2_bad_offset() checks if the position on the second
 515  * superblock is valid or not based on the filesystem parameters
 516  * stored in @sbp.  If @offset points to a location within the segment
 517  * area, or if the parameters themselves are not normal, it is
 518  * determined to be invalid.
 519  *
 520  * Return Value: true if invalid, false if valid.
 521  */
 522 static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 523 {
 524         unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
 525         u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 526         u64 nsegments = le64_to_cpu(sbp->s_nsegments);
 527         u64 index;
 528
 529         if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
 530             shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
 531                 return true;
 532
 533         index = offset >> (shift_bits + BLOCK_SIZE_BITS);
 534         do_div(index, blocks_per_segment);
 535         return index < nsegments;
 536 }
 537
 538 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 539 {
 540         int i;
 541
 542         for (i = 0; i < 2; i++) {
 543                 if (nilfs->ns_sbp[i]) {
 544                         brelse(nilfs->ns_sbh[i]);
 545                         nilfs->ns_sbh[i] = NULL;
 546                         nilfs->ns_sbp[i] = NULL;
 547                 }
 548         }
 549 }
 550
 551 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 552 {
 553         brelse(nilfs->ns_sbh[0]);
 554         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 555         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 556         nilfs->ns_sbh[1] = NULL;
 557         nilfs->ns_sbp[1] = NULL;
 558 }
 559
 560 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 561 {
 562         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 563         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 564
 565         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 566         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 567         nilfs->ns_sbh[1] = tsbh;
 568         nilfs->ns_sbp[1] = tsbp;
 569 }
 570
 571 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 572                                   struct super_block *sb, int blocksize,
 573                                   struct nilfs_super_block **sbpp)
 574 {
 575         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 576         struct buffer_head **sbh = nilfs->ns_sbh;
 577         u64 sb2off, devsize = nilfs->ns_bdev->bd_inode->i_size;
 578         int valid[2], swp = 0;
 579
 580         if (devsize < NILFS_SEG_MIN_BLOCKS * NILFS_MIN_BLOCK_SIZE + 4096) {
 581                 nilfs_msg(sb, KERN_ERR, "device size too small");
 582                 return -EINVAL;
 583         }
 584         sb2off = NILFS_SB2_OFFSET_BYTES(devsize);
 585
 586         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 587                                         &sbh[0]);
 588         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 589
 590         if (!sbp[0]) {
 591                 if (!sbp[1]) {
 592                         nilfs_msg(sb, KERN_ERR, "unable to read superblock");
 593                         return -EIO;
 594                 }
 595                 nilfs_msg(sb, KERN_WARNING,
 596                           "unable to read primary superblock (blocksize = %d)",
 597                           blocksize);
 598         } else if (!sbp[1]) {
 599                 nilfs_msg(sb, KERN_WARNING,
 600                           "unable to read secondary superblock (blocksize = %d)",
 601                           blocksize);
 602         }
 603
 604         /*
 605          * Compare two super blocks and set 1 in swp if the secondary
 606          * super block is valid and newer.  Otherwise, set 0 in swp.
 607          */
 608         valid[0] = nilfs_valid_sb(sbp[0]);
 609         valid[1] = nilfs_valid_sb(sbp[1]);
 610         swp = valid[1] && (!valid[0] ||
 611                            le64_to_cpu(sbp[1]->s_last_cno) >
 612                            le64_to_cpu(sbp[0]->s_last_cno));
 613
 614         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 615                 brelse(sbh[1]);
 616                 sbh[1] = NULL;
 617                 sbp[1] = NULL;
 618                 valid[1] = 0;
 619                 swp = 0;
 620         }
 621         if (!valid[swp]) {
 622                 nilfs_release_super_block(nilfs);
 623                 nilfs_msg(sb, KERN_ERR, "couldn't find nilfs on the device");
 624                 return -EINVAL;
 625         }
 626
 627         if (!valid[!swp])
 628                 nilfs_msg(sb, KERN_WARNING,
 629                           "broken superblock, retrying with spare superblock (blocksize = %d)",
 630                           blocksize);
 631         if (swp)
 632                 nilfs_swap_super_block(nilfs);
 633
 634         nilfs->ns_sbwcount = 0;
 635         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 636         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 637         *sbpp = sbp[0];
 638         return 0;
 639 }
 640
 641 /**
 642  * init_nilfs - initialize a NILFS instance.
 643  * @nilfs: the_nilfs structure
 644  * @sb: super block
 645  * @data: mount options
 646  *
 647  * init_nilfs() performs common initialization per block device (e.g.
 648  * reading the super block, getting disk layout information, initializing
 649  * shared fields in the_nilfs).
 650  *
 651  * Return Value: On success, 0 is returned. On error, a negative error
 652  * code is returned.
 653  */
 654 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
 655 {
 656         struct nilfs_super_block *sbp;
 657         int blocksize;
 658         int err;
 659
 660         down_write(&nilfs->ns_sem);
 661
 662         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
 663         if (!blocksize) {
 664                 nilfs_msg(sb, KERN_ERR, "unable to set blocksize");
 665                 err = -EINVAL;
 666                 goto out;
 667         }
 668         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 669         if (err)
 670                 goto out;
 671
 672         err = nilfs_store_magic_and_option(sb, sbp, data);
 673         if (err)
 674                 goto failed_sbh;
 675
 676         err = nilfs_check_feature_compatibility(sb, sbp);
 677         if (err)
 678                 goto failed_sbh;
 679
 680         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 681         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
 682             blocksize > NILFS_MAX_BLOCK_SIZE) {
 683                 nilfs_msg(sb, KERN_ERR,
 684                           "couldn't mount because of unsupported filesystem blocksize %d",
 685                           blocksize);
 686                 err = -EINVAL;
 687                 goto failed_sbh;
 688         }
 689         if (sb->s_blocksize != blocksize) {
 690                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 691
 692                 if (blocksize < hw_blocksize) {
 693                         nilfs_msg(sb, KERN_ERR,
 694                                   "blocksize %d too small for device (sector-size = %d)",
 695                                   blocksize, hw_blocksize);
 696                         err = -EINVAL;
 697                         goto failed_sbh;
 698                 }
 699                 nilfs_release_super_block(nilfs);
 700                 sb_set_blocksize(sb, blocksize);
 701
 702                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 703                 if (err)
 704                         goto out;
 705                         /*
 706                          * Not to failed_sbh; sbh is released automatically
 707                          * when reloading fails.
 708                          */
 709         }
 710         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 711         nilfs->ns_blocksize = blocksize;
 712
 713         get_random_bytes(&nilfs->ns_next_generation,
 714                          sizeof(nilfs->ns_next_generation));
 715
 716         err = nilfs_store_disk_layout(nilfs, sbp);
 717         if (err)
 718                 goto failed_sbh;
 719
 720         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 721
 722         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 723
 724         err = nilfs_store_log_cursor(nilfs, sbp);
 725         if (err)
 726                 goto failed_sbh;
 727
 728         set_nilfs_init(nilfs);
 729         err = 0;
 730  out:
 731         up_write(&nilfs->ns_sem);
 732         return err;
 733
 734  failed_sbh:
 735         nilfs_release_super_block(nilfs);
 736         goto out;
 737 }
 738
 739 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
 740                             size_t nsegs)
 741 {
 742         sector_t seg_start, seg_end;
 743         sector_t start = 0, nblocks = 0;
 744         unsigned int sects_per_block;
 745         __u64 *sn;
 746         int ret = 0;
 747
 748         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
 749                 bdev_logical_block_size(nilfs->ns_bdev);
 750         for (sn = segnump; sn < segnump + nsegs; sn++) {
 751                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
 752
 753                 if (!nblocks) {
 754                         start = seg_start;
 755                         nblocks = seg_end - seg_start + 1;
 756                 } else if (start + nblocks == seg_start) {
 757                         nblocks += seg_end - seg_start + 1;
 758                 } else {
 759                         ret = blkdev_issue_discard(nilfs->ns_bdev,
 760                                                    start * sects_per_block,
 761                                                    nblocks * sects_per_block,
 762                                                    GFP_NOFS, 0);
 763                         if (ret < 0)
 764                                 return ret;
 765                         nblocks = 0;
 766                 }
 767         }
 768         if (nblocks)
 769                 ret = blkdev_issue_discard(nilfs->ns_bdev,
 770                                            start * sects_per_block,
 771                                            nblocks * sects_per_block,
 772                                            GFP_NOFS, 0);
 773         return ret;
 774 }
 775
 776 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 777 {
 778         unsigned long ncleansegs;
 779
 780         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 781         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 782         return 0;
 783 }
 784
 785 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 786 {
 787         unsigned long ncleansegs, nincsegs;
 788
 789         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 790         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 791                 nilfs->ns_blocks_per_segment + 1;
 792
 793         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
 794 }
 795
 796 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
 797 {
 798         struct rb_node *n;
 799         struct nilfs_root *root;
 800
 801         spin_lock(&nilfs->ns_cptree_lock);
 802         n = nilfs->ns_cptree.rb_node;
 803         while (n) {
 804                 root = rb_entry(n, struct nilfs_root, rb_node);
 805
 806                 if (cno < root->cno) {
 807                         n = n->rb_left;
 808                 } else if (cno > root->cno) {
 809                         n = n->rb_right;
 810                 } else {
 811                         atomic_inc(&root->count);
 812                         spin_unlock(&nilfs->ns_cptree_lock);
 813                         return root;
 814                 }
 815         }
 816         spin_unlock(&nilfs->ns_cptree_lock);
 817
 818         return NULL;
 819 }
 820
 821 struct nilfs_root *
 822 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
 823 {
 824         struct rb_node **p, *parent;
 825         struct nilfs_root *root, *new;
 826         int err;
 827
 828         root = nilfs_lookup_root(nilfs, cno);
 829         if (root)
 830                 return root;
 831
 832         new = kzalloc(sizeof(*root), GFP_KERNEL);
 833         if (!new)
 834                 return NULL;
 835
 836         spin_lock(&nilfs->ns_cptree_lock);
 837
 838         p = &nilfs->ns_cptree.rb_node;
 839         parent = NULL;
 840
 841         while (*p) {
 842                 parent = *p;
 843                 root = rb_entry(parent, struct nilfs_root, rb_node);
 844
 845                 if (cno < root->cno) {
 846                         p = &(*p)->rb_left;
 847                 } else if (cno > root->cno) {
 848                         p = &(*p)->rb_right;
 849                 } else {
 850                         atomic_inc(&root->count);
 851                         spin_unlock(&nilfs->ns_cptree_lock);
 852                         kfree(new);
 853                         return root;
 854                 }
 855         }
 856
 857         new->cno = cno;
 858         new->ifile = NULL;
 859         new->nilfs = nilfs;
 860         atomic_set(&new->count, 1);
 861         atomic64_set(&new->inodes_count, 0);
 862         atomic64_set(&new->blocks_count, 0);
 863
 864         rb_link_node(&new->rb_node, parent, p);
 865         rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
 866
 867         spin_unlock(&nilfs->ns_cptree_lock);
 868
 869         err = nilfs_sysfs_create_snapshot_group(new);
 870         if (err) {
 871                 kfree(new);
 872                 new = NULL;
 873         }
 874
 875         return new;
 876 }
 877
 878 void nilfs_put_root(struct nilfs_root *root)
 879 {
 880         if (atomic_dec_and_test(&root->count)) {
 881                 struct the_nilfs *nilfs = root->nilfs;
 882
 883                 nilfs_sysfs_delete_snapshot_group(root);
 884
 885                 spin_lock(&nilfs->ns_cptree_lock);
 886                 rb_erase(&root->rb_node, &nilfs->ns_cptree);
 887                 spin_unlock(&nilfs->ns_cptree_lock);
 888                 iput(root->ifile);
 889
 890                 kfree(root);
 891         }
 892 }