fs/nilfs2/the_nilfs.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * the_nilfs.c - the_nilfs shared structure.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  *
   9  */
  10
  11 #include <linux/buffer_head.h>
  12 #include <linux/slab.h>
  13 #include <linux/blkdev.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/random.h>
  16 #include <linux/log2.h>
  17 #include <linux/crc32.h>
  18 #include "nilfs.h"
  19 #include "segment.h"
  20 #include "alloc.h"
  21 #include "cpfile.h"
  22 #include "sufile.h"
  23 #include "dat.h"
  24 #include "segbuf.h"
  25
  26
  27 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
  28
  29 void nilfs_set_last_segment(struct the_nilfs *nilfs,
  30                             sector_t start_blocknr, u64 seq, __u64 cno)
  31 {
  32         spin_lock(&nilfs->ns_last_segment_lock);
  33         nilfs->ns_last_pseg = start_blocknr;
  34         nilfs->ns_last_seq = seq;
  35         nilfs->ns_last_cno = cno;
  36
  37         if (!nilfs_sb_dirty(nilfs)) {
  38                 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
  39                         goto stay_cursor;
  40
  41                 set_nilfs_sb_dirty(nilfs);
  42         }
  43         nilfs->ns_prev_seq = nilfs->ns_last_seq;
  44
  45  stay_cursor:
  46         spin_unlock(&nilfs->ns_last_segment_lock);
  47 }
  48
  49 /**
  50  * alloc_nilfs - allocate a nilfs object
  51  * @sb: super block instance
  52  *
  53  * Return Value: On success, pointer to the_nilfs is returned.
  54  * On error, NULL is returned.
  55  */
  56 struct the_nilfs *alloc_nilfs(struct super_block *sb)
  57 {
  58         struct the_nilfs *nilfs;
  59
  60         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
  61         if (!nilfs)
  62                 return NULL;
  63
  64         nilfs->ns_sb = sb;
  65         nilfs->ns_bdev = sb->s_bdev;
  66         atomic_set(&nilfs->ns_ndirtyblks, 0);
  67         init_rwsem(&nilfs->ns_sem);
  68         mutex_init(&nilfs->ns_snapshot_mount_mutex);
  69         INIT_LIST_HEAD(&nilfs->ns_dirty_files);
  70         INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
  71         spin_lock_init(&nilfs->ns_inode_lock);
  72         spin_lock_init(&nilfs->ns_next_gen_lock);
  73         spin_lock_init(&nilfs->ns_last_segment_lock);
  74         nilfs->ns_cptree = RB_ROOT;
  75         spin_lock_init(&nilfs->ns_cptree_lock);
  76         init_rwsem(&nilfs->ns_segctor_sem);
  77         nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
  78
  79         return nilfs;
  80 }
  81
  82 /**
  83  * destroy_nilfs - destroy nilfs object
  84  * @nilfs: nilfs object to be released
  85  */
  86 void destroy_nilfs(struct the_nilfs *nilfs)
  87 {
  88         might_sleep();
  89         if (nilfs_init(nilfs)) {
  90                 brelse(nilfs->ns_sbh[0]);
  91                 brelse(nilfs->ns_sbh[1]);
  92         }
  93         kfree(nilfs);
  94 }
  95
  96 static int nilfs_load_super_root(struct the_nilfs *nilfs,
  97                                  struct super_block *sb, sector_t sr_block)
  98 {
  99         struct buffer_head *bh_sr;
 100         struct nilfs_super_root *raw_sr;
 101         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 102         struct nilfs_inode *rawi;
 103         unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
 104         unsigned int inode_size;
 105         int err;
 106
 107         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
 108         if (unlikely(err))
 109                 return err;
 110
 111         down_read(&nilfs->ns_sem);
 112         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 113         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 114         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 115         up_read(&nilfs->ns_sem);
 116
 117         inode_size = nilfs->ns_inode_size;
 118
 119         rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
 120         err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
 121         if (err)
 122                 goto failed;
 123
 124         rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
 125         err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
 126         if (err)
 127                 goto failed_dat;
 128
 129         rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
 130         err = nilfs_sufile_read(sb, segment_usage_size, rawi,
 131                                 &nilfs->ns_sufile);
 132         if (err)
 133                 goto failed_cpfile;
 134
 135         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 136         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 137
 138  failed:
 139         brelse(bh_sr);
 140         return err;
 141
 142  failed_cpfile:
 143         iput(nilfs->ns_cpfile);
 144
 145  failed_dat:
 146         iput(nilfs->ns_dat);
 147         goto failed;
 148 }
 149
 150 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 151 {
 152         memset(ri, 0, sizeof(*ri));
 153         INIT_LIST_HEAD(&ri->ri_used_segments);
 154 }
 155
 156 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 157 {
 158         nilfs_dispose_segment_list(&ri->ri_used_segments);
 159 }
 160
 161 /**
 162  * nilfs_store_log_cursor - load log cursor from a super block
 163  * @nilfs: nilfs object
 164  * @sbp: buffer storing super block to be read
 165  *
 166  * nilfs_store_log_cursor() reads the last position of the log
 167  * containing a super root from a given super block, and initializes
 168  * relevant information on the nilfs object preparatory for log
 169  * scanning and recovery.
 170  */
 171 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
 172                                   struct nilfs_super_block *sbp)
 173 {
 174         int ret = 0;
 175
 176         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 177         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 178         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 179
 180         nilfs->ns_prev_seq = nilfs->ns_last_seq;
 181         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 182         nilfs->ns_segnum =
 183                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 184         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 185         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 186                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 187                           "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
 188                           (unsigned long long)nilfs->ns_segnum,
 189                           nilfs->ns_nsegments);
 190                 ret = -EINVAL;
 191         }
 192         return ret;
 193 }
 194
 195 /**
 196  * load_nilfs - load and recover the nilfs
 197  * @nilfs: the_nilfs structure to be released
 198  * @sb: super block isntance used to recover past segment
 199  *
 200  * load_nilfs() searches and load the latest super root,
 201  * attaches the last segment, and does recovery if needed.
 202  * The caller must call this exclusively for simultaneous mounts.
 203  */
 204 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
 205 {
 206         struct nilfs_recovery_info ri;
 207         unsigned int s_flags = sb->s_flags;
 208         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 209         int valid_fs = nilfs_valid_fs(nilfs);
 210         int err;
 211
 212         if (!valid_fs) {
 213                 nilfs_msg(sb, KERN_WARNING, "mounting unchecked fs");
 214                 if (s_flags & SB_RDONLY) {
 215                         nilfs_msg(sb, KERN_INFO,
 216                                   "recovery required for readonly filesystem");
 217                         nilfs_msg(sb, KERN_INFO,
 218                                   "write access will be enabled during recovery");
 219                 }
 220         }
 221
 222         nilfs_init_recovery_info(&ri);
 223
 224         err = nilfs_search_super_root(nilfs, &ri);
 225         if (unlikely(err)) {
 226                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
 227                 int blocksize;
 228
 229                 if (err != -EINVAL)
 230                         goto scan_error;
 231
 232                 if (!nilfs_valid_sb(sbp[1])) {
 233                         nilfs_msg(sb, KERN_WARNING,
 234                                   "unable to fall back to spare super block");
 235                         goto scan_error;
 236                 }
 237                 nilfs_msg(sb, KERN_INFO,
 238                           "trying rollback from an earlier position");
 239
 240                 /*
 241                  * restore super block with its spare and reconfigure
 242                  * relevant states of the nilfs object.
 243                  */
 244                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 245                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
 246                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 247
 248                 /* verify consistency between two super blocks */
 249                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
 250                 if (blocksize != nilfs->ns_blocksize) {
 251                         nilfs_msg(sb, KERN_WARNING,
 252                                   "blocksize differs between two super blocks (%d != %d)",
 253                                   blocksize, nilfs->ns_blocksize);
 254                         goto scan_error;
 255                 }
 256
 257                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
 258                 if (err)
 259                         goto scan_error;
 260
 261                 /* drop clean flag to allow roll-forward and recovery */
 262                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
 263                 valid_fs = 0;
 264
 265                 err = nilfs_search_super_root(nilfs, &ri);
 266                 if (err)
 267                         goto scan_error;
 268         }
 269
 270         err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
 271         if (unlikely(err)) {
 272                 nilfs_msg(sb, KERN_ERR, "error %d while loading super root",
 273                           err);
 274                 goto failed;
 275         }
 276
 277         err = nilfs_sysfs_create_device_group(sb);
 278         if (unlikely(err))
 279                 goto sysfs_error;
 280
 281         if (valid_fs)
 282                 goto skip_recovery;
 283
 284         if (s_flags & SB_RDONLY) {
 285                 __u64 features;
 286
 287                 if (nilfs_test_opt(nilfs, NORECOVERY)) {
 288                         nilfs_msg(sb, KERN_INFO,
 289                                   "norecovery option specified, skipping roll-forward recovery");
 290                         goto skip_recovery;
 291                 }
 292                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
 293                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
 294                 if (features) {
 295                         nilfs_msg(sb, KERN_ERR,
 296                                   "couldn't proceed with recovery because of unsupported optional features (%llx)",
 297                                   (unsigned long long)features);
 298                         err = -EROFS;
 299                         goto failed_unload;
 300                 }
 301                 if (really_read_only) {
 302                         nilfs_msg(sb, KERN_ERR,
 303                                   "write access unavailable, cannot proceed");
 304                         err = -EROFS;
 305                         goto failed_unload;
 306                 }
 307                 sb->s_flags &= ~SB_RDONLY;
 308         } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
 309                 nilfs_msg(sb, KERN_ERR,
 310                           "recovery cancelled because norecovery option was specified for a read/write mount");
 311                 err = -EINVAL;
 312                 goto failed_unload;
 313         }
 314
 315         err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
 316         if (err)
 317                 goto failed_unload;
 318
 319         down_write(&nilfs->ns_sem);
 320         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
 321         err = nilfs_cleanup_super(sb);
 322         up_write(&nilfs->ns_sem);
 323
 324         if (err) {
 325                 nilfs_msg(sb, KERN_ERR,
 326                           "error %d updating super block. recovery unfinished.",
 327                           err);
 328                 goto failed_unload;
 329         }
 330         nilfs_msg(sb, KERN_INFO, "recovery complete");
 331
 332  skip_recovery:
 333         nilfs_clear_recovery_info(&ri);
 334         sb->s_flags = s_flags;
 335         return 0;
 336
 337  scan_error:
 338         nilfs_msg(sb, KERN_ERR, "error %d while searching super root", err);
 339         goto failed;
 340
 341  failed_unload:
 342         nilfs_sysfs_delete_device_group(nilfs);
 343
 344  sysfs_error:
 345         iput(nilfs->ns_cpfile);
 346         iput(nilfs->ns_sufile);
 347         iput(nilfs->ns_dat);
 348
 349  failed:
 350         nilfs_clear_recovery_info(&ri);
 351         sb->s_flags = s_flags;
 352         return err;
 353 }
 354
 355 static unsigned long long nilfs_max_size(unsigned int blkbits)
 356 {
 357         unsigned int max_bits;
 358         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 359
 360         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 361         if (max_bits < 64)
 362                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 363         return res;
 364 }
 365
 366 /**
 367  * nilfs_nrsvsegs - calculate the number of reserved segments
 368  * @nilfs: nilfs object
 369  * @nsegs: total number of segments
 370  */
 371 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
 372 {
 373         return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 374                      DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
 375                                   100));
 376 }
 377
 378 /**
 379  * nilfs_max_segment_count - calculate the maximum number of segments
 380  * @nilfs: nilfs object
 381  */
 382 static u64 nilfs_max_segment_count(struct the_nilfs *nilfs)
 383 {
 384         u64 max_count = U64_MAX;
 385
 386         do_div(max_count, nilfs->ns_blocks_per_segment);
 387         return min_t(u64, max_count, ULONG_MAX);
 388 }
 389
 390 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
 391 {
 392         nilfs->ns_nsegments = nsegs;
 393         nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
 394 }
 395
 396 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 397                                    struct nilfs_super_block *sbp)
 398 {
 399         u64 nsegments, nblocks;
 400
 401         if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
 402                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 403                           "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
 404                           le32_to_cpu(sbp->s_rev_level),
 405                           le16_to_cpu(sbp->s_minor_rev_level),
 406                           NILFS_CURRENT_REV, NILFS_MINOR_REV);
 407                 return -EINVAL;
 408         }
 409         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 410         if (nilfs->ns_sbsize > BLOCK_SIZE)
 411                 return -EINVAL;
 412
 413         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 414         if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
 415                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 416                           "too large inode size: %d bytes",
 417                           nilfs->ns_inode_size);
 418                 return -EINVAL;
 419         } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
 420                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 421                           "too small inode size: %d bytes",
 422                           nilfs->ns_inode_size);
 423                 return -EINVAL;
 424         }
 425
 426         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 427
 428         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 429         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 430                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 431                           "too short segment: %lu blocks",
 432                           nilfs->ns_blocks_per_segment);
 433                 return -EINVAL;
 434         }
 435
 436         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 437         nilfs->ns_r_segments_percentage =
 438                 le32_to_cpu(sbp->s_r_segments_percentage);
 439         if (nilfs->ns_r_segments_percentage < 1 ||
 440             nilfs->ns_r_segments_percentage > 99) {
 441                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 442                           "invalid reserved segments percentage: %lu",
 443                           nilfs->ns_r_segments_percentage);
 444                 return -EINVAL;
 445         }
 446
 447         nsegments = le64_to_cpu(sbp->s_nsegments);
 448         if (nsegments > nilfs_max_segment_count(nilfs)) {
 449                 nilfs_msg(nilfs->ns_sb, KERN_ERR,
 450                           "segment count %llu exceeds upper limit (%llu segments)",
 451                           (unsigned long long)nsegments,
 452                           (unsigned long long)nilfs_max_segment_count(nilfs));
 453                 return -EINVAL;
 454         }
 455
 456         nblocks = (u64)i_size_read(nilfs->ns_sb->s_bdev->bd_inode) >>
 457                 nilfs->ns_sb->s_blocksize_bits;
 458         if (nblocks) {
 459                 u64 min_block_count = nsegments * nilfs->ns_blocks_per_segment;
 460                 /*
 461                  * To avoid failing to mount early device images without a
 462                  * second superblock, exclude that block count from the
 463                  * "min_block_count" calculation.
 464                  */
 465
 466                 if (nblocks < min_block_count) {
 467                         nilfs_msg(nilfs->ns_sb, KERN_ERR,
 468                                   "total number of segment blocks %llu exceeds device size (%llu blocks)",
 469                                   (unsigned long long)min_block_count,
 470                                   (unsigned long long)nblocks);
 471                         return -EINVAL;
 472                 }
 473         }
 474
 475         nilfs_set_nsegments(nilfs, nsegments);
 476         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 477         return 0;
 478 }
 479
 480 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 481 {
 482         static unsigned char sum[4];
 483         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 484         size_t bytes;
 485         u32 crc;
 486
 487         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 488                 return 0;
 489         bytes = le16_to_cpu(sbp->s_bytes);
 490         if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
 491                 return 0;
 492         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 493                        sumoff);
 494         crc = crc32_le(crc, sum, 4);
 495         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 496                        bytes - sumoff - 4);
 497         return crc == le32_to_cpu(sbp->s_sum);
 498 }
 499
 500 /**
 501  * nilfs_sb2_bad_offset - check the location of the second superblock
 502  * @sbp: superblock raw data buffer
 503  * @offset: byte offset of second superblock calculated from device size
 504  *
 505  * nilfs_sb2_bad_offset() checks if the position on the second
 506  * superblock is valid or not based on the filesystem parameters
 507  * stored in @sbp.  If @offset points to a location within the segment
 508  * area, or if the parameters themselves are not normal, it is
 509  * determined to be invalid.
 510  *
 511  * Return Value: true if invalid, false if valid.
 512  */
 513 static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 514 {
 515         unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
 516         u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 517         u64 nsegments = le64_to_cpu(sbp->s_nsegments);
 518         u64 index;
 519
 520         if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
 521             shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
 522                 return true;
 523
 524         index = offset >> (shift_bits + BLOCK_SIZE_BITS);
 525         do_div(index, blocks_per_segment);
 526         return index < nsegments;
 527 }
 528
 529 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 530 {
 531         int i;
 532
 533         for (i = 0; i < 2; i++) {
 534                 if (nilfs->ns_sbp[i]) {
 535                         brelse(nilfs->ns_sbh[i]);
 536                         nilfs->ns_sbh[i] = NULL;
 537                         nilfs->ns_sbp[i] = NULL;
 538                 }
 539         }
 540 }
 541
 542 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 543 {
 544         brelse(nilfs->ns_sbh[0]);
 545         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 546         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 547         nilfs->ns_sbh[1] = NULL;
 548         nilfs->ns_sbp[1] = NULL;
 549 }
 550
 551 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 552 {
 553         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 554         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 555
 556         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 557         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 558         nilfs->ns_sbh[1] = tsbh;
 559         nilfs->ns_sbp[1] = tsbp;
 560 }
 561
 562 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 563                                   struct super_block *sb, int blocksize,
 564                                   struct nilfs_super_block **sbpp)
 565 {
 566         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 567         struct buffer_head **sbh = nilfs->ns_sbh;
 568         u64 sb2off, devsize = nilfs->ns_bdev->bd_inode->i_size;
 569         int valid[2], swp = 0;
 570
 571         if (devsize < NILFS_SEG_MIN_BLOCKS * NILFS_MIN_BLOCK_SIZE + 4096) {
 572                 nilfs_msg(sb, KERN_ERR, "device size too small");
 573                 return -EINVAL;
 574         }
 575         sb2off = NILFS_SB2_OFFSET_BYTES(devsize);
 576
 577         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 578                                         &sbh[0]);
 579         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 580
 581         if (!sbp[0]) {
 582                 if (!sbp[1]) {
 583                         nilfs_msg(sb, KERN_ERR, "unable to read superblock");
 584                         return -EIO;
 585                 }
 586                 nilfs_msg(sb, KERN_WARNING,
 587                           "unable to read primary superblock (blocksize = %d)",
 588                           blocksize);
 589         } else if (!sbp[1]) {
 590                 nilfs_msg(sb, KERN_WARNING,
 591                           "unable to read secondary superblock (blocksize = %d)",
 592                           blocksize);
 593         }
 594
 595         /*
 596          * Compare two super blocks and set 1 in swp if the secondary
 597          * super block is valid and newer.  Otherwise, set 0 in swp.
 598          */
 599         valid[0] = nilfs_valid_sb(sbp[0]);
 600         valid[1] = nilfs_valid_sb(sbp[1]);
 601         swp = valid[1] && (!valid[0] ||
 602                            le64_to_cpu(sbp[1]->s_last_cno) >
 603                            le64_to_cpu(sbp[0]->s_last_cno));
 604
 605         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 606                 brelse(sbh[1]);
 607                 sbh[1] = NULL;
 608                 sbp[1] = NULL;
 609                 valid[1] = 0;
 610                 swp = 0;
 611         }
 612         if (!valid[swp]) {
 613                 nilfs_release_super_block(nilfs);
 614                 nilfs_msg(sb, KERN_ERR, "couldn't find nilfs on the device");
 615                 return -EINVAL;
 616         }
 617
 618         if (!valid[!swp])
 619                 nilfs_msg(sb, KERN_WARNING,
 620                           "broken superblock, retrying with spare superblock (blocksize = %d)",
 621                           blocksize);
 622         if (swp)
 623                 nilfs_swap_super_block(nilfs);
 624
 625         nilfs->ns_sbwcount = 0;
 626         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 627         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 628         *sbpp = sbp[0];
 629         return 0;
 630 }
 631
 632 /**
 633  * init_nilfs - initialize a NILFS instance.
 634  * @nilfs: the_nilfs structure
 635  * @sb: super block
 636  * @data: mount options
 637  *
 638  * init_nilfs() performs common initialization per block device (e.g.
 639  * reading the super block, getting disk layout information, initializing
 640  * shared fields in the_nilfs).
 641  *
 642  * Return Value: On success, 0 is returned. On error, a negative error
 643  * code is returned.
 644  */
 645 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
 646 {
 647         struct nilfs_super_block *sbp;
 648         int blocksize;
 649         int err;
 650
 651         down_write(&nilfs->ns_sem);
 652
 653         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
 654         if (!blocksize) {
 655                 nilfs_msg(sb, KERN_ERR, "unable to set blocksize");
 656                 err = -EINVAL;
 657                 goto out;
 658         }
 659         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 660         if (err)
 661                 goto out;
 662
 663         err = nilfs_store_magic_and_option(sb, sbp, data);
 664         if (err)
 665                 goto failed_sbh;
 666
 667         err = nilfs_check_feature_compatibility(sb, sbp);
 668         if (err)
 669                 goto failed_sbh;
 670
 671         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 672         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
 673             blocksize > NILFS_MAX_BLOCK_SIZE) {
 674                 nilfs_msg(sb, KERN_ERR,
 675                           "couldn't mount because of unsupported filesystem blocksize %d",
 676                           blocksize);
 677                 err = -EINVAL;
 678                 goto failed_sbh;
 679         }
 680         if (sb->s_blocksize != blocksize) {
 681                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 682
 683                 if (blocksize < hw_blocksize) {
 684                         nilfs_msg(sb, KERN_ERR,
 685                                   "blocksize %d too small for device (sector-size = %d)",
 686                                   blocksize, hw_blocksize);
 687                         err = -EINVAL;
 688                         goto failed_sbh;
 689                 }
 690                 nilfs_release_super_block(nilfs);
 691                 sb_set_blocksize(sb, blocksize);
 692
 693                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 694                 if (err)
 695                         goto out;
 696                         /*
 697                          * Not to failed_sbh; sbh is released automatically
 698                          * when reloading fails.
 699                          */
 700         }
 701         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 702         nilfs->ns_blocksize = blocksize;
 703
 704         get_random_bytes(&nilfs->ns_next_generation,
 705                          sizeof(nilfs->ns_next_generation));
 706
 707         err = nilfs_store_disk_layout(nilfs, sbp);
 708         if (err)
 709                 goto failed_sbh;
 710
 711         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 712
 713         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 714
 715         err = nilfs_store_log_cursor(nilfs, sbp);
 716         if (err)
 717                 goto failed_sbh;
 718
 719         set_nilfs_init(nilfs);
 720         err = 0;
 721  out:
 722         up_write(&nilfs->ns_sem);
 723         return err;
 724
 725  failed_sbh:
 726         nilfs_release_super_block(nilfs);
 727         goto out;
 728 }
 729
 730 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
 731                             size_t nsegs)
 732 {
 733         sector_t seg_start, seg_end;
 734         sector_t start = 0, nblocks = 0;
 735         unsigned int sects_per_block;
 736         __u64 *sn;
 737         int ret = 0;
 738
 739         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
 740                 bdev_logical_block_size(nilfs->ns_bdev);
 741         for (sn = segnump; sn < segnump + nsegs; sn++) {
 742                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
 743
 744                 if (!nblocks) {
 745                         start = seg_start;
 746                         nblocks = seg_end - seg_start + 1;
 747                 } else if (start + nblocks == seg_start) {
 748                         nblocks += seg_end - seg_start + 1;
 749                 } else {
 750                         ret = blkdev_issue_discard(nilfs->ns_bdev,
 751                                                    start * sects_per_block,
 752                                                    nblocks * sects_per_block,
 753                                                    GFP_NOFS, 0);
 754                         if (ret < 0)
 755                                 return ret;
 756                         nblocks = 0;
 757                 }
 758         }
 759         if (nblocks)
 760                 ret = blkdev_issue_discard(nilfs->ns_bdev,
 761                                            start * sects_per_block,
 762                                            nblocks * sects_per_block,
 763                                            GFP_NOFS, 0);
 764         return ret;
 765 }
 766
 767 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 768 {
 769         unsigned long ncleansegs;
 770
 771         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 772         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 773         return 0;
 774 }
 775
 776 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 777 {
 778         unsigned long ncleansegs, nincsegs;
 779
 780         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 781         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 782                 nilfs->ns_blocks_per_segment + 1;
 783
 784         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
 785 }
 786
 787 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
 788 {
 789         struct rb_node *n;
 790         struct nilfs_root *root;
 791
 792         spin_lock(&nilfs->ns_cptree_lock);
 793         n = nilfs->ns_cptree.rb_node;
 794         while (n) {
 795                 root = rb_entry(n, struct nilfs_root, rb_node);
 796
 797                 if (cno < root->cno) {
 798                         n = n->rb_left;
 799                 } else if (cno > root->cno) {
 800                         n = n->rb_right;
 801                 } else {
 802                         refcount_inc(&root->count);
 803                         spin_unlock(&nilfs->ns_cptree_lock);
 804                         return root;
 805                 }
 806         }
 807         spin_unlock(&nilfs->ns_cptree_lock);
 808
 809         return NULL;
 810 }
 811
 812 struct nilfs_root *
 813 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
 814 {
 815         struct rb_node **p, *parent;
 816         struct nilfs_root *root, *new;
 817         int err;
 818
 819         root = nilfs_lookup_root(nilfs, cno);
 820         if (root)
 821                 return root;
 822
 823         new = kzalloc(sizeof(*root), GFP_KERNEL);
 824         if (!new)
 825                 return NULL;
 826
 827         spin_lock(&nilfs->ns_cptree_lock);
 828
 829         p = &nilfs->ns_cptree.rb_node;
 830         parent = NULL;
 831
 832         while (*p) {
 833                 parent = *p;
 834                 root = rb_entry(parent, struct nilfs_root, rb_node);
 835
 836                 if (cno < root->cno) {
 837                         p = &(*p)->rb_left;
 838                 } else if (cno > root->cno) {
 839                         p = &(*p)->rb_right;
 840                 } else {
 841                         refcount_inc(&root->count);
 842                         spin_unlock(&nilfs->ns_cptree_lock);
 843                         kfree(new);
 844                         return root;
 845                 }
 846         }
 847
 848         new->cno = cno;
 849         new->ifile = NULL;
 850         new->nilfs = nilfs;
 851         refcount_set(&new->count, 1);
 852         atomic64_set(&new->inodes_count, 0);
 853         atomic64_set(&new->blocks_count, 0);
 854
 855         rb_link_node(&new->rb_node, parent, p);
 856         rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
 857
 858         spin_unlock(&nilfs->ns_cptree_lock);
 859
 860         err = nilfs_sysfs_create_snapshot_group(new);
 861         if (err) {
 862                 kfree(new);
 863                 new = NULL;
 864         }
 865
 866         return new;
 867 }
 868
 869 void nilfs_put_root(struct nilfs_root *root)
 870 {
 871         struct the_nilfs *nilfs = root->nilfs;
 872
 873         if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
 874                 rb_erase(&root->rb_node, &nilfs->ns_cptree);
 875                 spin_unlock(&nilfs->ns_cptree_lock);
 876
 877                 nilfs_sysfs_delete_snapshot_group(root);
 878                 iput(root->ifile);
 879
 880                 kfree(root);
 881         }
 882 }