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 LIST_HEAD(nilfs_objects);
  39 static DEFINE_SPINLOCK(nilfs_lock);
  40
  41 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
  42
  43 void nilfs_set_last_segment(struct the_nilfs *nilfs,
  44                             sector_t start_blocknr, u64 seq, __u64 cno)
  45 {
  46         spin_lock(&nilfs->ns_last_segment_lock);
  47         nilfs->ns_last_pseg = start_blocknr;
  48         nilfs->ns_last_seq = seq;
  49         nilfs->ns_last_cno = cno;
  50
  51         if (!nilfs_sb_dirty(nilfs)) {
  52                 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
  53                         goto stay_cursor;
  54
  55                 set_nilfs_sb_dirty(nilfs);
  56         }
  57         nilfs->ns_prev_seq = nilfs->ns_last_seq;
  58
  59  stay_cursor:
  60         spin_unlock(&nilfs->ns_last_segment_lock);
  61 }
  62
  63 /**
  64  * alloc_nilfs - allocate the_nilfs structure
  65  * @bdev: block device to which the_nilfs is related
  66  *
  67  * alloc_nilfs() allocates memory for the_nilfs and
  68  * initializes its reference count and locks.
  69  *
  70  * Return Value: On success, pointer to the_nilfs is returned.
  71  * On error, NULL is returned.
  72  */
  73 static struct the_nilfs *alloc_nilfs(struct block_device *bdev)
  74 {
  75         struct the_nilfs *nilfs;
  76
  77         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
  78         if (!nilfs)
  79                 return NULL;
  80
  81         nilfs->ns_bdev = bdev;
  82         atomic_set(&nilfs->ns_count, 1);
  83         atomic_set(&nilfs->ns_ndirtyblks, 0);
  84         init_rwsem(&nilfs->ns_sem);
  85         mutex_init(&nilfs->ns_mount_mutex);
  86         init_rwsem(&nilfs->ns_writer_sem);
  87         INIT_LIST_HEAD(&nilfs->ns_list);
  88         INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
  89         spin_lock_init(&nilfs->ns_last_segment_lock);
  90         nilfs->ns_cptree = RB_ROOT;
  91         spin_lock_init(&nilfs->ns_cptree_lock);
  92         init_rwsem(&nilfs->ns_segctor_sem);
  93
  94         return nilfs;
  95 }
  96
  97 /**
  98  * find_or_create_nilfs - find or create nilfs object
  99  * @bdev: block device to which the_nilfs is related
 100  *
 101  * find_nilfs() looks up an existent nilfs object created on the
 102  * device and gets the reference count of the object.  If no nilfs object
 103  * is found on the device, a new nilfs object is allocated.
 104  *
 105  * Return Value: On success, pointer to the nilfs object is returned.
 106  * On error, NULL is returned.
 107  */
 108 struct the_nilfs *find_or_create_nilfs(struct block_device *bdev)
 109 {
 110         struct the_nilfs *nilfs, *new = NULL;
 111
 112  retry:
 113         spin_lock(&nilfs_lock);
 114         list_for_each_entry(nilfs, &nilfs_objects, ns_list) {
 115                 if (nilfs->ns_bdev == bdev) {
 116                         get_nilfs(nilfs);
 117                         spin_unlock(&nilfs_lock);
 118                         if (new)
 119                                 put_nilfs(new);
 120                         return nilfs; /* existing object */
 121                 }
 122         }
 123         if (new) {
 124                 list_add_tail(&new->ns_list, &nilfs_objects);
 125                 spin_unlock(&nilfs_lock);
 126                 return new; /* new object */
 127         }
 128         spin_unlock(&nilfs_lock);
 129
 130         new = alloc_nilfs(bdev);
 131         if (new)
 132                 goto retry;
 133         return NULL; /* insufficient memory */
 134 }
 135
 136 /**
 137  * put_nilfs - release a reference to the_nilfs
 138  * @nilfs: the_nilfs structure to be released
 139  *
 140  * put_nilfs() decrements a reference counter of the_nilfs.
 141  * If the reference count reaches zero, the_nilfs is freed.
 142  */
 143 void put_nilfs(struct the_nilfs *nilfs)
 144 {
 145         spin_lock(&nilfs_lock);
 146         if (!atomic_dec_and_test(&nilfs->ns_count)) {
 147                 spin_unlock(&nilfs_lock);
 148                 return;
 149         }
 150         list_del_init(&nilfs->ns_list);
 151         spin_unlock(&nilfs_lock);
 152
 153         /*
 154          * Increment of ns_count never occurs below because the caller
 155          * of get_nilfs() holds at least one reference to the_nilfs.
 156          * Thus its exclusion control is not required here.
 157          */
 158
 159         might_sleep();
 160         if (nilfs_loaded(nilfs)) {
 161                 nilfs_mdt_destroy(nilfs->ns_sufile);
 162                 nilfs_mdt_destroy(nilfs->ns_cpfile);
 163                 nilfs_mdt_destroy(nilfs->ns_dat);
 164                 nilfs_mdt_destroy(nilfs->ns_gc_dat);
 165         }
 166         if (nilfs_init(nilfs)) {
 167                 brelse(nilfs->ns_sbh[0]);
 168                 brelse(nilfs->ns_sbh[1]);
 169         }
 170         kfree(nilfs);
 171 }
 172
 173 static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block)
 174 {
 175         struct buffer_head *bh_sr;
 176         struct nilfs_super_root *raw_sr;
 177         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 178         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
 179         unsigned inode_size;
 180         int err;
 181
 182         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
 183         if (unlikely(err))
 184                 return err;
 185
 186         down_read(&nilfs->ns_sem);
 187         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 188         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 189         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 190         up_read(&nilfs->ns_sem);
 191
 192         inode_size = nilfs->ns_inode_size;
 193
 194         err = -ENOMEM;
 195         nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size);
 196         if (unlikely(!nilfs->ns_dat))
 197                 goto failed;
 198
 199         nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size);
 200         if (unlikely(!nilfs->ns_gc_dat))
 201                 goto failed_dat;
 202
 203         nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size);
 204         if (unlikely(!nilfs->ns_cpfile))
 205                 goto failed_gc_dat;
 206
 207         nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
 208         if (unlikely(!nilfs->ns_sufile))
 209                 goto failed_cpfile;
 210
 211         nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
 212
 213         err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
 214                              NILFS_SR_DAT_OFFSET(inode_size));
 215         if (unlikely(err))
 216                 goto failed_sufile;
 217
 218         err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
 219                                 NILFS_SR_CPFILE_OFFSET(inode_size));
 220         if (unlikely(err))
 221                 goto failed_sufile;
 222
 223         err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
 224                                 NILFS_SR_SUFILE_OFFSET(inode_size));
 225         if (unlikely(err))
 226                 goto failed_sufile;
 227
 228         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 229         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 230
 231  failed:
 232         brelse(bh_sr);
 233         return err;
 234
 235  failed_sufile:
 236         nilfs_mdt_destroy(nilfs->ns_sufile);
 237
 238  failed_cpfile:
 239         nilfs_mdt_destroy(nilfs->ns_cpfile);
 240
 241  failed_gc_dat:
 242         nilfs_mdt_destroy(nilfs->ns_gc_dat);
 243
 244  failed_dat:
 245         nilfs_mdt_destroy(nilfs->ns_dat);
 246         goto failed;
 247 }
 248
 249 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 250 {
 251         memset(ri, 0, sizeof(*ri));
 252         INIT_LIST_HEAD(&ri->ri_used_segments);
 253 }
 254
 255 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 256 {
 257         nilfs_dispose_segment_list(&ri->ri_used_segments);
 258 }
 259
 260 /**
 261  * nilfs_store_log_cursor - load log cursor from a super block
 262  * @nilfs: nilfs object
 263  * @sbp: buffer storing super block to be read
 264  *
 265  * nilfs_store_log_cursor() reads the last position of the log
 266  * containing a super root from a given super block, and initializes
 267  * relevant information on the nilfs object preparatory for log
 268  * scanning and recovery.
 269  */
 270 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
 271                                   struct nilfs_super_block *sbp)
 272 {
 273         int ret = 0;
 274
 275         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 276         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 277         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 278
 279         nilfs->ns_prev_seq = nilfs->ns_last_seq;
 280         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 281         nilfs->ns_segnum =
 282                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 283         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 284         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 285                 printk(KERN_ERR "NILFS invalid last segment number.\n");
 286                 ret = -EINVAL;
 287         }
 288         return ret;
 289 }
 290
 291 /**
 292  * load_nilfs - load and recover the nilfs
 293  * @nilfs: the_nilfs structure to be released
 294  * @sbi: nilfs_sb_info used to recover past segment
 295  *
 296  * load_nilfs() searches and load the latest super root,
 297  * attaches the last segment, and does recovery if needed.
 298  * The caller must call this exclusively for simultaneous mounts.
 299  */
 300 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
 301 {
 302         struct nilfs_recovery_info ri;
 303         unsigned int s_flags = sbi->s_super->s_flags;
 304         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 305         int valid_fs = nilfs_valid_fs(nilfs);
 306         int err;
 307
 308         if (!valid_fs) {
 309                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 310                 if (s_flags & MS_RDONLY) {
 311                         printk(KERN_INFO "NILFS: INFO: recovery "
 312                                "required for readonly filesystem.\n");
 313                         printk(KERN_INFO "NILFS: write access will "
 314                                "be enabled during recovery.\n");
 315                 }
 316         }
 317
 318         nilfs_init_recovery_info(&ri);
 319
 320         err = nilfs_search_super_root(nilfs, &ri);
 321         if (unlikely(err)) {
 322                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
 323                 int blocksize;
 324
 325                 if (err != -EINVAL)
 326                         goto scan_error;
 327
 328                 if (!nilfs_valid_sb(sbp[1])) {
 329                         printk(KERN_WARNING
 330                                "NILFS warning: unable to fall back to spare"
 331                                "super block\n");
 332                         goto scan_error;
 333                 }
 334                 printk(KERN_INFO
 335                        "NILFS: try rollback from an earlier position\n");
 336
 337                 /*
 338                  * restore super block with its spare and reconfigure
 339                  * relevant states of the nilfs object.
 340                  */
 341                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 342                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
 343                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 344
 345                 /* verify consistency between two super blocks */
 346                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
 347                 if (blocksize != nilfs->ns_blocksize) {
 348                         printk(KERN_WARNING
 349                                "NILFS warning: blocksize differs between "
 350                                "two super blocks (%d != %d)\n",
 351                                blocksize, nilfs->ns_blocksize);
 352                         goto scan_error;
 353                 }
 354
 355                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
 356                 if (err)
 357                         goto scan_error;
 358
 359                 /* drop clean flag to allow roll-forward and recovery */
 360                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
 361                 valid_fs = 0;
 362
 363                 err = nilfs_search_super_root(nilfs, &ri);
 364                 if (err)
 365                         goto scan_error;
 366         }
 367
 368         err = nilfs_load_super_root(nilfs, ri.ri_super_root);
 369         if (unlikely(err)) {
 370                 printk(KERN_ERR "NILFS: error loading super root.\n");
 371                 goto failed;
 372         }
 373
 374         if (valid_fs)
 375                 goto skip_recovery;
 376
 377         if (s_flags & MS_RDONLY) {
 378                 __u64 features;
 379
 380                 if (nilfs_test_opt(sbi, NORECOVERY)) {
 381                         printk(KERN_INFO "NILFS: norecovery option specified. "
 382                                "skipping roll-forward recovery\n");
 383                         goto skip_recovery;
 384                 }
 385                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
 386                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
 387                 if (features) {
 388                         printk(KERN_ERR "NILFS: couldn't proceed with "
 389                                "recovery because of unsupported optional "
 390                                "features (%llx)\n",
 391                                (unsigned long long)features);
 392                         err = -EROFS;
 393                         goto failed_unload;
 394                 }
 395                 if (really_read_only) {
 396                         printk(KERN_ERR "NILFS: write access "
 397                                "unavailable, cannot proceed.\n");
 398                         err = -EROFS;
 399                         goto failed_unload;
 400                 }
 401                 sbi->s_super->s_flags &= ~MS_RDONLY;
 402         } else if (nilfs_test_opt(sbi, NORECOVERY)) {
 403                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
 404                        "option was specified for a read/write mount\n");
 405                 err = -EINVAL;
 406                 goto failed_unload;
 407         }
 408
 409         err = nilfs_salvage_orphan_logs(nilfs, sbi, &ri);
 410         if (err)
 411                 goto failed_unload;
 412
 413         down_write(&nilfs->ns_sem);
 414         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
 415         err = nilfs_cleanup_super(sbi);
 416         up_write(&nilfs->ns_sem);
 417
 418         if (err) {
 419                 printk(KERN_ERR "NILFS: failed to update super block. "
 420                        "recovery unfinished.\n");
 421                 goto failed_unload;
 422         }
 423         printk(KERN_INFO "NILFS: recovery complete.\n");
 424
 425  skip_recovery:
 426         set_nilfs_loaded(nilfs);
 427         nilfs_clear_recovery_info(&ri);
 428         sbi->s_super->s_flags = s_flags;
 429         return 0;
 430
 431  scan_error:
 432         printk(KERN_ERR "NILFS: error searching super root.\n");
 433         goto failed;
 434
 435  failed_unload:
 436         nilfs_mdt_destroy(nilfs->ns_cpfile);
 437         nilfs_mdt_destroy(nilfs->ns_sufile);
 438         nilfs_mdt_destroy(nilfs->ns_dat);
 439         nilfs_mdt_destroy(nilfs->ns_gc_dat);
 440
 441  failed:
 442         nilfs_clear_recovery_info(&ri);
 443         sbi->s_super->s_flags = s_flags;
 444         return err;
 445 }
 446
 447 static unsigned long long nilfs_max_size(unsigned int blkbits)
 448 {
 449         unsigned int max_bits;
 450         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 451
 452         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 453         if (max_bits < 64)
 454                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 455         return res;
 456 }
 457
 458 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 459                                    struct nilfs_super_block *sbp)
 460 {
 461         if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
 462                 printk(KERN_ERR "NILFS: unsupported revision "
 463                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
 464                        "Please check the version of mkfs.nilfs.\n",
 465                        le32_to_cpu(sbp->s_rev_level),
 466                        le16_to_cpu(sbp->s_minor_rev_level),
 467                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
 468                 return -EINVAL;
 469         }
 470         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 471         if (nilfs->ns_sbsize > BLOCK_SIZE)
 472                 return -EINVAL;
 473
 474         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 475         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 476
 477         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 478         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 479                 printk(KERN_ERR "NILFS: too short segment.\n");
 480                 return -EINVAL;
 481         }
 482
 483         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 484         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
 485         nilfs->ns_r_segments_percentage =
 486                 le32_to_cpu(sbp->s_r_segments_percentage);
 487         nilfs->ns_nrsvsegs =
 488                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 489                       DIV_ROUND_UP(nilfs->ns_nsegments *
 490                                    nilfs->ns_r_segments_percentage, 100));
 491         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 492         return 0;
 493 }
 494
 495 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 496 {
 497         static unsigned char sum[4];
 498         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 499         size_t bytes;
 500         u32 crc;
 501
 502         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 503                 return 0;
 504         bytes = le16_to_cpu(sbp->s_bytes);
 505         if (bytes > BLOCK_SIZE)
 506                 return 0;
 507         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 508                        sumoff);
 509         crc = crc32_le(crc, sum, 4);
 510         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 511                        bytes - sumoff - 4);
 512         return crc == le32_to_cpu(sbp->s_sum);
 513 }
 514
 515 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 516 {
 517         return offset < ((le64_to_cpu(sbp->s_nsegments) *
 518                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
 519                          (le32_to_cpu(sbp->s_log_block_size) + 10));
 520 }
 521
 522 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 523 {
 524         int i;
 525
 526         for (i = 0; i < 2; i++) {
 527                 if (nilfs->ns_sbp[i]) {
 528                         brelse(nilfs->ns_sbh[i]);
 529                         nilfs->ns_sbh[i] = NULL;
 530                         nilfs->ns_sbp[i] = NULL;
 531                 }
 532         }
 533 }
 534
 535 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 536 {
 537         brelse(nilfs->ns_sbh[0]);
 538         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 539         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 540         nilfs->ns_sbh[1] = NULL;
 541         nilfs->ns_sbp[1] = NULL;
 542 }
 543
 544 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 545 {
 546         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 547         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 548
 549         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 550         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 551         nilfs->ns_sbh[1] = tsbh;
 552         nilfs->ns_sbp[1] = tsbp;
 553 }
 554
 555 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 556                                   struct super_block *sb, int blocksize,
 557                                   struct nilfs_super_block **sbpp)
 558 {
 559         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 560         struct buffer_head **sbh = nilfs->ns_sbh;
 561         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
 562         int valid[2], swp = 0;
 563
 564         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 565                                         &sbh[0]);
 566         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 567
 568         if (!sbp[0]) {
 569                 if (!sbp[1]) {
 570                         printk(KERN_ERR "NILFS: unable to read superblock\n");
 571                         return -EIO;
 572                 }
 573                 printk(KERN_WARNING
 574                        "NILFS warning: unable to read primary superblock\n");
 575         } else if (!sbp[1])
 576                 printk(KERN_WARNING
 577                        "NILFS warning: unable to read secondary superblock\n");
 578
 579         /*
 580          * Compare two super blocks and set 1 in swp if the secondary
 581          * super block is valid and newer.  Otherwise, set 0 in swp.
 582          */
 583         valid[0] = nilfs_valid_sb(sbp[0]);
 584         valid[1] = nilfs_valid_sb(sbp[1]);
 585         swp = valid[1] && (!valid[0] ||
 586                            le64_to_cpu(sbp[1]->s_last_cno) >
 587                            le64_to_cpu(sbp[0]->s_last_cno));
 588
 589         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 590                 brelse(sbh[1]);
 591                 sbh[1] = NULL;
 592                 sbp[1] = NULL;
 593                 swp = 0;
 594         }
 595         if (!valid[swp]) {
 596                 nilfs_release_super_block(nilfs);
 597                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
 598                        sb->s_id);
 599                 return -EINVAL;
 600         }
 601
 602         if (!valid[!swp])
 603                 printk(KERN_WARNING "NILFS warning: broken superblock. "
 604                        "using spare superblock.\n");
 605         if (swp)
 606                 nilfs_swap_super_block(nilfs);
 607
 608         nilfs->ns_sbwcount = 0;
 609         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 610         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 611         *sbpp = sbp[0];
 612         return 0;
 613 }
 614
 615 /**
 616  * init_nilfs - initialize a NILFS instance.
 617  * @nilfs: the_nilfs structure
 618  * @sbi: nilfs_sb_info
 619  * @sb: super block
 620  * @data: mount options
 621  *
 622  * init_nilfs() performs common initialization per block device (e.g.
 623  * reading the super block, getting disk layout information, initializing
 624  * shared fields in the_nilfs).
 625  *
 626  * Return Value: On success, 0 is returned. On error, a negative error
 627  * code is returned.
 628  */
 629 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 630 {
 631         struct super_block *sb = sbi->s_super;
 632         struct nilfs_super_block *sbp;
 633         struct backing_dev_info *bdi;
 634         int blocksize;
 635         int err;
 636
 637         down_write(&nilfs->ns_sem);
 638
 639         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
 640         if (!blocksize) {
 641                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
 642                 err = -EINVAL;
 643                 goto out;
 644         }
 645         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 646         if (err)
 647                 goto out;
 648
 649         err = nilfs_store_magic_and_option(sb, sbp, data);
 650         if (err)
 651                 goto failed_sbh;
 652
 653         err = nilfs_check_feature_compatibility(sb, sbp);
 654         if (err)
 655                 goto failed_sbh;
 656
 657         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 658         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
 659             blocksize > NILFS_MAX_BLOCK_SIZE) {
 660                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 661                        "filesystem blocksize %d\n", blocksize);
 662                 err = -EINVAL;
 663                 goto failed_sbh;
 664         }
 665         if (sb->s_blocksize != blocksize) {
 666                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 667
 668                 if (blocksize < hw_blocksize) {
 669                         printk(KERN_ERR
 670                                "NILFS: blocksize %d too small for device "
 671                                "(sector-size = %d).\n",
 672                                blocksize, hw_blocksize);
 673                         err = -EINVAL;
 674                         goto failed_sbh;
 675                 }
 676                 nilfs_release_super_block(nilfs);
 677                 sb_set_blocksize(sb, blocksize);
 678
 679                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 680                 if (err)
 681                         goto out;
 682                         /* not failed_sbh; sbh is released automatically
 683                            when reloading fails. */
 684         }
 685         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 686         nilfs->ns_blocksize = blocksize;
 687
 688         err = nilfs_store_disk_layout(nilfs, sbp);
 689         if (err)
 690                 goto failed_sbh;
 691
 692         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 693
 694         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 695
 696         bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
 697         nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
 698
 699         err = nilfs_store_log_cursor(nilfs, sbp);
 700         if (err)
 701                 goto failed_sbh;
 702
 703         set_nilfs_init(nilfs);
 704         err = 0;
 705  out:
 706         up_write(&nilfs->ns_sem);
 707         return err;
 708
 709  failed_sbh:
 710         nilfs_release_super_block(nilfs);
 711         goto out;
 712 }
 713
 714 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
 715                             size_t nsegs)
 716 {
 717         sector_t seg_start, seg_end;
 718         sector_t start = 0, nblocks = 0;
 719         unsigned int sects_per_block;
 720         __u64 *sn;
 721         int ret = 0;
 722
 723         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
 724                 bdev_logical_block_size(nilfs->ns_bdev);
 725         for (sn = segnump; sn < segnump + nsegs; sn++) {
 726                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
 727
 728                 if (!nblocks) {
 729                         start = seg_start;
 730                         nblocks = seg_end - seg_start + 1;
 731                 } else if (start + nblocks == seg_start) {
 732                         nblocks += seg_end - seg_start + 1;
 733                 } else {
 734                         ret = blkdev_issue_discard(nilfs->ns_bdev,
 735                                                    start * sects_per_block,
 736                                                    nblocks * sects_per_block,
 737                                                    GFP_NOFS,
 738                                                    BLKDEV_IFL_WAIT |
 739                                                    BLKDEV_IFL_BARRIER);
 740                         if (ret < 0)
 741                                 return ret;
 742                         nblocks = 0;
 743                 }
 744         }
 745         if (nblocks)
 746                 ret = blkdev_issue_discard(nilfs->ns_bdev,
 747                                            start * sects_per_block,
 748                                            nblocks * sects_per_block,
 749                                            GFP_NOFS,
 750                                           BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
 751         return ret;
 752 }
 753
 754 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 755 {
 756         struct inode *dat = nilfs_dat_inode(nilfs);
 757         unsigned long ncleansegs;
 758
 759         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 760         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 761         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 762         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 763         return 0;
 764 }
 765
 766 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 767 {
 768         unsigned long ncleansegs, nincsegs;
 769
 770         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 771         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 772                 nilfs->ns_blocks_per_segment + 1;
 773
 774         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
 775 }
 776
 777 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
 778 {
 779         struct rb_node *n;
 780         struct nilfs_root *root;
 781
 782         spin_lock(&nilfs->ns_cptree_lock);
 783         n = nilfs->ns_cptree.rb_node;
 784         while (n) {
 785                 root = rb_entry(n, struct nilfs_root, rb_node);
 786
 787                 if (cno < root->cno) {
 788                         n = n->rb_left;
 789                 } else if (cno > root->cno) {
 790                         n = n->rb_right;
 791                 } else {
 792                         atomic_inc(&root->count);
 793                         spin_unlock(&nilfs->ns_cptree_lock);
 794                         return root;
 795                 }
 796         }
 797         spin_unlock(&nilfs->ns_cptree_lock);
 798
 799         return NULL;
 800 }
 801
 802 struct nilfs_root *
 803 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
 804 {
 805         struct rb_node **p, *parent;
 806         struct nilfs_root *root, *new;
 807
 808         root = nilfs_lookup_root(nilfs, cno);
 809         if (root)
 810                 return root;
 811
 812         new = kmalloc(sizeof(*root), GFP_KERNEL);
 813         if (!new)
 814                 return NULL;
 815
 816         spin_lock(&nilfs->ns_cptree_lock);
 817
 818         p = &nilfs->ns_cptree.rb_node;
 819         parent = NULL;
 820
 821         while (*p) {
 822                 parent = *p;
 823                 root = rb_entry(parent, struct nilfs_root, rb_node);
 824
 825                 if (cno < root->cno) {
 826                         p = &(*p)->rb_left;
 827                 } else if (cno > root->cno) {
 828                         p = &(*p)->rb_right;
 829                 } else {
 830                         atomic_inc(&root->count);
 831                         spin_unlock(&nilfs->ns_cptree_lock);
 832                         kfree(new);
 833                         return root;
 834                 }
 835         }
 836
 837         new->cno = cno;
 838         new->ifile = NULL;
 839         new->nilfs = nilfs;
 840         atomic_set(&new->count, 1);
 841         atomic_set(&new->inodes_count, 0);
 842         atomic_set(&new->blocks_count, 0);
 843
 844         rb_link_node(&new->rb_node, parent, p);
 845         rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
 846
 847         spin_unlock(&nilfs->ns_cptree_lock);
 848
 849         return new;
 850 }
 851
 852 void nilfs_put_root(struct nilfs_root *root)
 853 {
 854         if (atomic_dec_and_test(&root->count)) {
 855                 struct the_nilfs *nilfs = root->nilfs;
 856
 857                 spin_lock(&nilfs->ns_cptree_lock);
 858                 rb_erase(&root->rb_node, &nilfs->ns_cptree);
 859                 spin_unlock(&nilfs->ns_cptree_lock);
 860                 if (root->ifile)
 861                         nilfs_mdt_destroy(root->ifile);
 862
 863                 kfree(root);
 864         }
 865 }
 866
 867 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
 868                                 int snapshot_mount)
 869 {
 870         struct nilfs_root *root;
 871         int ret;
 872
 873         if (cno < 0 || cno > nilfs->ns_cno)
 874                 return false;
 875
 876         if (cno >= nilfs_last_cno(nilfs))
 877                 return true;    /* protect recent checkpoints */
 878
 879         ret = false;
 880         root = nilfs_lookup_root(nilfs, cno);
 881         if (root) {
 882                 ret = true;
 883                 nilfs_put_root(root);
 884         }
 885         return ret;
 886 }