fs/nilfs2/the_nilfs.c

   1 /*
   2  * the_nilfs.c - the_nilfs shared structure.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/slab.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/crc32.h>
  29 #include "nilfs.h"
  30 #include "segment.h"
  31 #include "alloc.h"
  32 #include "cpfile.h"
  33 #include "sufile.h"
  34 #include "dat.h"
  35 #include "segbuf.h"
  36
  37
  38 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
  39
  40 void nilfs_set_last_segment(struct the_nilfs *nilfs,
  41                             sector_t start_blocknr, u64 seq, __u64 cno)
  42 {
  43         spin_lock(&nilfs->ns_last_segment_lock);
  44         nilfs->ns_last_pseg = start_blocknr;
  45         nilfs->ns_last_seq = seq;
  46         nilfs->ns_last_cno = cno;
  47
  48         if (!nilfs_sb_dirty(nilfs)) {
  49                 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
  50                         goto stay_cursor;
  51
  52                 set_nilfs_sb_dirty(nilfs);
  53         }
  54         nilfs->ns_prev_seq = nilfs->ns_last_seq;
  55
  56  stay_cursor:
  57         spin_unlock(&nilfs->ns_last_segment_lock);
  58 }
  59
  60 /**
  61  * alloc_nilfs - allocate a nilfs object
  62  * @bdev: block device to which the_nilfs is related
  63  *
  64  * Return Value: On success, pointer to the_nilfs is returned.
  65  * On error, NULL is returned.
  66  */
  67 struct the_nilfs *alloc_nilfs(struct block_device *bdev)
  68 {
  69         struct the_nilfs *nilfs;
  70
  71         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
  72         if (!nilfs)
  73                 return NULL;
  74
  75         nilfs->ns_bdev = bdev;
  76         atomic_set(&nilfs->ns_ndirtyblks, 0);
  77         init_rwsem(&nilfs->ns_sem);
  78         INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
  79         spin_lock_init(&nilfs->ns_last_segment_lock);
  80         nilfs->ns_cptree = RB_ROOT;
  81         spin_lock_init(&nilfs->ns_cptree_lock);
  82         init_rwsem(&nilfs->ns_segctor_sem);
  83
  84         return nilfs;
  85 }
  86
  87 /**
  88  * destroy_nilfs - destroy nilfs object
  89  * @nilfs: nilfs object to be released
  90  */
  91 void destroy_nilfs(struct the_nilfs *nilfs)
  92 {
  93         might_sleep();
  94         if (nilfs_init(nilfs)) {
  95                 brelse(nilfs->ns_sbh[0]);
  96                 brelse(nilfs->ns_sbh[1]);
  97         }
  98         kfree(nilfs);
  99 }
 100
 101 static int nilfs_load_super_root(struct the_nilfs *nilfs,
 102                                  struct super_block *sb, sector_t sr_block)
 103 {
 104         struct buffer_head *bh_sr;
 105         struct nilfs_super_root *raw_sr;
 106         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 107         struct nilfs_inode *rawi;
 108         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
 109         unsigned inode_size;
 110         int err;
 111
 112         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
 113         if (unlikely(err))
 114                 return err;
 115
 116         down_read(&nilfs->ns_sem);
 117         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 118         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 119         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 120         up_read(&nilfs->ns_sem);
 121
 122         inode_size = nilfs->ns_inode_size;
 123
 124         rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
 125         err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
 126         if (err)
 127                 goto failed;
 128
 129         rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
 130         err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
 131         if (err)
 132                 goto failed_dat;
 133
 134         rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
 135         err = nilfs_sufile_read(sb, segment_usage_size, rawi,
 136                                 &nilfs->ns_sufile);
 137         if (err)
 138                 goto failed_cpfile;
 139
 140         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 141         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 142
 143  failed:
 144         brelse(bh_sr);
 145         return err;
 146
 147  failed_cpfile:
 148         iput(nilfs->ns_cpfile);
 149
 150  failed_dat:
 151         iput(nilfs->ns_dat);
 152         goto failed;
 153 }
 154
 155 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 156 {
 157         memset(ri, 0, sizeof(*ri));
 158         INIT_LIST_HEAD(&ri->ri_used_segments);
 159 }
 160
 161 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 162 {
 163         nilfs_dispose_segment_list(&ri->ri_used_segments);
 164 }
 165
 166 /**
 167  * nilfs_store_log_cursor - load log cursor from a super block
 168  * @nilfs: nilfs object
 169  * @sbp: buffer storing super block to be read
 170  *
 171  * nilfs_store_log_cursor() reads the last position of the log
 172  * containing a super root from a given super block, and initializes
 173  * relevant information on the nilfs object preparatory for log
 174  * scanning and recovery.
 175  */
 176 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
 177                                   struct nilfs_super_block *sbp)
 178 {
 179         int ret = 0;
 180
 181         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 182         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 183         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 184
 185         nilfs->ns_prev_seq = nilfs->ns_last_seq;
 186         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 187         nilfs->ns_segnum =
 188                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 189         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 190         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 191                 printk(KERN_ERR "NILFS invalid last segment number.\n");
 192                 ret = -EINVAL;
 193         }
 194         return ret;
 195 }
 196
 197 /**
 198  * load_nilfs - load and recover the nilfs
 199  * @nilfs: the_nilfs structure to be released
 200  * @sbi: nilfs_sb_info used to recover past segment
 201  *
 202  * load_nilfs() searches and load the latest super root,
 203  * attaches the last segment, and does recovery if needed.
 204  * The caller must call this exclusively for simultaneous mounts.
 205  */
 206 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
 207 {
 208         struct nilfs_recovery_info ri;
 209         unsigned int s_flags = sbi->s_super->s_flags;
 210         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 211         int valid_fs = nilfs_valid_fs(nilfs);
 212         int err;
 213
 214         if (!valid_fs) {
 215                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 216                 if (s_flags & MS_RDONLY) {
 217                         printk(KERN_INFO "NILFS: INFO: recovery "
 218                                "required for readonly filesystem.\n");
 219                         printk(KERN_INFO "NILFS: write access will "
 220                                "be enabled during recovery.\n");
 221                 }
 222         }
 223
 224         nilfs_init_recovery_info(&ri);
 225
 226         err = nilfs_search_super_root(nilfs, &ri);
 227         if (unlikely(err)) {
 228                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
 229                 int blocksize;
 230
 231                 if (err != -EINVAL)
 232                         goto scan_error;
 233
 234                 if (!nilfs_valid_sb(sbp[1])) {
 235                         printk(KERN_WARNING
 236                                "NILFS warning: unable to fall back to spare"
 237                                "super block\n");
 238                         goto scan_error;
 239                 }
 240                 printk(KERN_INFO
 241                        "NILFS: try rollback from an earlier position\n");
 242
 243                 /*
 244                  * restore super block with its spare and reconfigure
 245                  * relevant states of the nilfs object.
 246                  */
 247                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 248                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
 249                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 250
 251                 /* verify consistency between two super blocks */
 252                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
 253                 if (blocksize != nilfs->ns_blocksize) {
 254                         printk(KERN_WARNING
 255                                "NILFS warning: blocksize differs between "
 256                                "two super blocks (%d != %d)\n",
 257                                blocksize, nilfs->ns_blocksize);
 258                         goto scan_error;
 259                 }
 260
 261                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
 262                 if (err)
 263                         goto scan_error;
 264
 265                 /* drop clean flag to allow roll-forward and recovery */
 266                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
 267                 valid_fs = 0;
 268
 269                 err = nilfs_search_super_root(nilfs, &ri);
 270                 if (err)
 271                         goto scan_error;
 272         }
 273
 274         err = nilfs_load_super_root(nilfs, sbi->s_super, ri.ri_super_root);
 275         if (unlikely(err)) {
 276                 printk(KERN_ERR "NILFS: error loading super root.\n");
 277                 goto failed;
 278         }
 279
 280         if (valid_fs)
 281                 goto skip_recovery;
 282
 283         if (s_flags & MS_RDONLY) {
 284                 __u64 features;
 285
 286                 if (nilfs_test_opt(sbi, NORECOVERY)) {
 287                         printk(KERN_INFO "NILFS: norecovery option specified. "
 288                                "skipping roll-forward recovery\n");
 289                         goto skip_recovery;
 290                 }
 291                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
 292                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
 293                 if (features) {
 294                         printk(KERN_ERR "NILFS: couldn't proceed with "
 295                                "recovery because of unsupported optional "
 296                                "features (%llx)\n",
 297                                (unsigned long long)features);
 298                         err = -EROFS;
 299                         goto failed_unload;
 300                 }
 301                 if (really_read_only) {
 302                         printk(KERN_ERR "NILFS: write access "
 303                                "unavailable, cannot proceed.\n");
 304                         err = -EROFS;
 305                         goto failed_unload;
 306                 }
 307                 sbi->s_super->s_flags &= ~MS_RDONLY;
 308         } else if (nilfs_test_opt(sbi, NORECOVERY)) {
 309                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
 310                        "option was specified for a read/write mount\n");
 311                 err = -EINVAL;
 312                 goto failed_unload;
 313         }
 314
 315         err = nilfs_salvage_orphan_logs(nilfs, sbi, &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(sbi);
 322         up_write(&nilfs->ns_sem);
 323
 324         if (err) {
 325                 printk(KERN_ERR "NILFS: failed to update super block. "
 326                        "recovery unfinished.\n");
 327                 goto failed_unload;
 328         }
 329         printk(KERN_INFO "NILFS: recovery complete.\n");
 330
 331  skip_recovery:
 332         set_nilfs_loaded(nilfs);
 333         nilfs_clear_recovery_info(&ri);
 334         sbi->s_super->s_flags = s_flags;
 335         return 0;
 336
 337  scan_error:
 338         printk(KERN_ERR "NILFS: error searching super root.\n");
 339         goto failed;
 340
 341  failed_unload:
 342         iput(nilfs->ns_cpfile);
 343         iput(nilfs->ns_sufile);
 344         iput(nilfs->ns_dat);
 345
 346  failed:
 347         nilfs_clear_recovery_info(&ri);
 348         sbi->s_super->s_flags = s_flags;
 349         return err;
 350 }
 351
 352 static unsigned long long nilfs_max_size(unsigned int blkbits)
 353 {
 354         unsigned int max_bits;
 355         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 356
 357         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 358         if (max_bits < 64)
 359                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 360         return res;
 361 }
 362
 363 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 364                                    struct nilfs_super_block *sbp)
 365 {
 366         if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
 367                 printk(KERN_ERR "NILFS: unsupported revision "
 368                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
 369                        "Please check the version of mkfs.nilfs.\n",
 370                        le32_to_cpu(sbp->s_rev_level),
 371                        le16_to_cpu(sbp->s_minor_rev_level),
 372                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
 373                 return -EINVAL;
 374         }
 375         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 376         if (nilfs->ns_sbsize > BLOCK_SIZE)
 377                 return -EINVAL;
 378
 379         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 380         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 381
 382         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 383         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 384                 printk(KERN_ERR "NILFS: too short segment.\n");
 385                 return -EINVAL;
 386         }
 387
 388         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 389         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
 390         nilfs->ns_r_segments_percentage =
 391                 le32_to_cpu(sbp->s_r_segments_percentage);
 392         nilfs->ns_nrsvsegs =
 393                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 394                       DIV_ROUND_UP(nilfs->ns_nsegments *
 395                                    nilfs->ns_r_segments_percentage, 100));
 396         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 397         return 0;
 398 }
 399
 400 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 401 {
 402         static unsigned char sum[4];
 403         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 404         size_t bytes;
 405         u32 crc;
 406
 407         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 408                 return 0;
 409         bytes = le16_to_cpu(sbp->s_bytes);
 410         if (bytes > BLOCK_SIZE)
 411                 return 0;
 412         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 413                        sumoff);
 414         crc = crc32_le(crc, sum, 4);
 415         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 416                        bytes - sumoff - 4);
 417         return crc == le32_to_cpu(sbp->s_sum);
 418 }
 419
 420 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 421 {
 422         return offset < ((le64_to_cpu(sbp->s_nsegments) *
 423                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
 424                          (le32_to_cpu(sbp->s_log_block_size) + 10));
 425 }
 426
 427 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 428 {
 429         int i;
 430
 431         for (i = 0; i < 2; i++) {
 432                 if (nilfs->ns_sbp[i]) {
 433                         brelse(nilfs->ns_sbh[i]);
 434                         nilfs->ns_sbh[i] = NULL;
 435                         nilfs->ns_sbp[i] = NULL;
 436                 }
 437         }
 438 }
 439
 440 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 441 {
 442         brelse(nilfs->ns_sbh[0]);
 443         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 444         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 445         nilfs->ns_sbh[1] = NULL;
 446         nilfs->ns_sbp[1] = NULL;
 447 }
 448
 449 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 450 {
 451         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 452         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 453
 454         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 455         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 456         nilfs->ns_sbh[1] = tsbh;
 457         nilfs->ns_sbp[1] = tsbp;
 458 }
 459
 460 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 461                                   struct super_block *sb, int blocksize,
 462                                   struct nilfs_super_block **sbpp)
 463 {
 464         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 465         struct buffer_head **sbh = nilfs->ns_sbh;
 466         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
 467         int valid[2], swp = 0;
 468
 469         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 470                                         &sbh[0]);
 471         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 472
 473         if (!sbp[0]) {
 474                 if (!sbp[1]) {
 475                         printk(KERN_ERR "NILFS: unable to read superblock\n");
 476                         return -EIO;
 477                 }
 478                 printk(KERN_WARNING
 479                        "NILFS warning: unable to read primary superblock\n");
 480         } else if (!sbp[1])
 481                 printk(KERN_WARNING
 482                        "NILFS warning: unable to read secondary superblock\n");
 483
 484         /*
 485          * Compare two super blocks and set 1 in swp if the secondary
 486          * super block is valid and newer.  Otherwise, set 0 in swp.
 487          */
 488         valid[0] = nilfs_valid_sb(sbp[0]);
 489         valid[1] = nilfs_valid_sb(sbp[1]);
 490         swp = valid[1] && (!valid[0] ||
 491                            le64_to_cpu(sbp[1]->s_last_cno) >
 492                            le64_to_cpu(sbp[0]->s_last_cno));
 493
 494         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 495                 brelse(sbh[1]);
 496                 sbh[1] = NULL;
 497                 sbp[1] = NULL;
 498                 swp = 0;
 499         }
 500         if (!valid[swp]) {
 501                 nilfs_release_super_block(nilfs);
 502                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
 503                        sb->s_id);
 504                 return -EINVAL;
 505         }
 506
 507         if (!valid[!swp])
 508                 printk(KERN_WARNING "NILFS warning: broken superblock. "
 509                        "using spare superblock.\n");
 510         if (swp)
 511                 nilfs_swap_super_block(nilfs);
 512
 513         nilfs->ns_sbwcount = 0;
 514         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 515         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 516         *sbpp = sbp[0];
 517         return 0;
 518 }
 519
 520 /**
 521  * init_nilfs - initialize a NILFS instance.
 522  * @nilfs: the_nilfs structure
 523  * @sbi: nilfs_sb_info
 524  * @sb: super block
 525  * @data: mount options
 526  *
 527  * init_nilfs() performs common initialization per block device (e.g.
 528  * reading the super block, getting disk layout information, initializing
 529  * shared fields in the_nilfs).
 530  *
 531  * Return Value: On success, 0 is returned. On error, a negative error
 532  * code is returned.
 533  */
 534 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 535 {
 536         struct super_block *sb = sbi->s_super;
 537         struct nilfs_super_block *sbp;
 538         int blocksize;
 539         int err;
 540
 541         down_write(&nilfs->ns_sem);
 542
 543         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
 544         if (!blocksize) {
 545                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
 546                 err = -EINVAL;
 547                 goto out;
 548         }
 549         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 550         if (err)
 551                 goto out;
 552
 553         err = nilfs_store_magic_and_option(sb, sbp, data);
 554         if (err)
 555                 goto failed_sbh;
 556
 557         err = nilfs_check_feature_compatibility(sb, sbp);
 558         if (err)
 559                 goto failed_sbh;
 560
 561         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 562         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
 563             blocksize > NILFS_MAX_BLOCK_SIZE) {
 564                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 565                        "filesystem blocksize %d\n", blocksize);
 566                 err = -EINVAL;
 567                 goto failed_sbh;
 568         }
 569         if (sb->s_blocksize != blocksize) {
 570                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 571
 572                 if (blocksize < hw_blocksize) {
 573                         printk(KERN_ERR
 574                                "NILFS: blocksize %d too small for device "
 575                                "(sector-size = %d).\n",
 576                                blocksize, hw_blocksize);
 577                         err = -EINVAL;
 578                         goto failed_sbh;
 579                 }
 580                 nilfs_release_super_block(nilfs);
 581                 sb_set_blocksize(sb, blocksize);
 582
 583                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 584                 if (err)
 585                         goto out;
 586                         /* not failed_sbh; sbh is released automatically
 587                            when reloading fails. */
 588         }
 589         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 590         nilfs->ns_blocksize = blocksize;
 591
 592         err = nilfs_store_disk_layout(nilfs, sbp);
 593         if (err)
 594                 goto failed_sbh;
 595
 596         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 597
 598         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 599
 600         err = nilfs_store_log_cursor(nilfs, sbp);
 601         if (err)
 602                 goto failed_sbh;
 603
 604         set_nilfs_init(nilfs);
 605         err = 0;
 606  out:
 607         up_write(&nilfs->ns_sem);
 608         return err;
 609
 610  failed_sbh:
 611         nilfs_release_super_block(nilfs);
 612         goto out;
 613 }
 614
 615 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
 616                             size_t nsegs)
 617 {
 618         sector_t seg_start, seg_end;
 619         sector_t start = 0, nblocks = 0;
 620         unsigned int sects_per_block;
 621         __u64 *sn;
 622         int ret = 0;
 623
 624         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
 625                 bdev_logical_block_size(nilfs->ns_bdev);
 626         for (sn = segnump; sn < segnump + nsegs; sn++) {
 627                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
 628
 629                 if (!nblocks) {
 630                         start = seg_start;
 631                         nblocks = seg_end - seg_start + 1;
 632                 } else if (start + nblocks == seg_start) {
 633                         nblocks += seg_end - seg_start + 1;
 634                 } else {
 635                         ret = blkdev_issue_discard(nilfs->ns_bdev,
 636                                                    start * sects_per_block,
 637                                                    nblocks * sects_per_block,
 638                                                    GFP_NOFS, 0);
 639                         if (ret < 0)
 640                                 return ret;
 641                         nblocks = 0;
 642                 }
 643         }
 644         if (nblocks)
 645                 ret = blkdev_issue_discard(nilfs->ns_bdev,
 646                                            start * sects_per_block,
 647                                            nblocks * sects_per_block,
 648                                            GFP_NOFS, 0);
 649         return ret;
 650 }
 651
 652 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 653 {
 654         struct inode *dat = nilfs_dat_inode(nilfs);
 655         unsigned long ncleansegs;
 656
 657         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 658         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 659         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 660         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 661         return 0;
 662 }
 663
 664 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 665 {
 666         unsigned long ncleansegs, nincsegs;
 667
 668         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 669         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 670                 nilfs->ns_blocks_per_segment + 1;
 671
 672         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
 673 }
 674
 675 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
 676 {
 677         struct rb_node *n;
 678         struct nilfs_root *root;
 679
 680         spin_lock(&nilfs->ns_cptree_lock);
 681         n = nilfs->ns_cptree.rb_node;
 682         while (n) {
 683                 root = rb_entry(n, struct nilfs_root, rb_node);
 684
 685                 if (cno < root->cno) {
 686                         n = n->rb_left;
 687                 } else if (cno > root->cno) {
 688                         n = n->rb_right;
 689                 } else {
 690                         atomic_inc(&root->count);
 691                         spin_unlock(&nilfs->ns_cptree_lock);
 692                         return root;
 693                 }
 694         }
 695         spin_unlock(&nilfs->ns_cptree_lock);
 696
 697         return NULL;
 698 }
 699
 700 struct nilfs_root *
 701 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
 702 {
 703         struct rb_node **p, *parent;
 704         struct nilfs_root *root, *new;
 705
 706         root = nilfs_lookup_root(nilfs, cno);
 707         if (root)
 708                 return root;
 709
 710         new = kmalloc(sizeof(*root), GFP_KERNEL);
 711         if (!new)
 712                 return NULL;
 713
 714         spin_lock(&nilfs->ns_cptree_lock);
 715
 716         p = &nilfs->ns_cptree.rb_node;
 717         parent = NULL;
 718
 719         while (*p) {
 720                 parent = *p;
 721                 root = rb_entry(parent, struct nilfs_root, rb_node);
 722
 723                 if (cno < root->cno) {
 724                         p = &(*p)->rb_left;
 725                 } else if (cno > root->cno) {
 726                         p = &(*p)->rb_right;
 727                 } else {
 728                         atomic_inc(&root->count);
 729                         spin_unlock(&nilfs->ns_cptree_lock);
 730                         kfree(new);
 731                         return root;
 732                 }
 733         }
 734
 735         new->cno = cno;
 736         new->ifile = NULL;
 737         new->nilfs = nilfs;
 738         atomic_set(&new->count, 1);
 739         atomic_set(&new->inodes_count, 0);
 740         atomic_set(&new->blocks_count, 0);
 741
 742         rb_link_node(&new->rb_node, parent, p);
 743         rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
 744
 745         spin_unlock(&nilfs->ns_cptree_lock);
 746
 747         return new;
 748 }
 749
 750 void nilfs_put_root(struct nilfs_root *root)
 751 {
 752         if (atomic_dec_and_test(&root->count)) {
 753                 struct the_nilfs *nilfs = root->nilfs;
 754
 755                 spin_lock(&nilfs->ns_cptree_lock);
 756                 rb_erase(&root->rb_node, &nilfs->ns_cptree);
 757                 spin_unlock(&nilfs->ns_cptree_lock);
 758                 if (root->ifile)
 759                         iput(root->ifile);
 760
 761                 kfree(root);
 762         }
 763 }