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