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