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         init_rwsem(&nilfs->ns_super_sem);
  86         mutex_init(&nilfs->ns_mount_mutex);
  87         init_rwsem(&nilfs->ns_writer_sem);
  88         INIT_LIST_HEAD(&nilfs->ns_list);
  89         INIT_LIST_HEAD(&nilfs->ns_supers);
  90         spin_lock_init(&nilfs->ns_last_segment_lock);
  91         nilfs->ns_gc_inodes_h = NULL;
  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                 nilfs_destroy_gccache(nilfs);
 168                 brelse(nilfs->ns_sbh[0]);
 169                 brelse(nilfs->ns_sbh[1]);
 170         }
 171         kfree(nilfs);
 172 }
 173
 174 static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block)
 175 {
 176         struct buffer_head *bh_sr;
 177         struct nilfs_super_root *raw_sr;
 178         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 179         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
 180         unsigned inode_size;
 181         int err;
 182
 183         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
 184         if (unlikely(err))
 185                 return err;
 186
 187         down_read(&nilfs->ns_sem);
 188         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
 189         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
 190         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
 191         up_read(&nilfs->ns_sem);
 192
 193         inode_size = nilfs->ns_inode_size;
 194
 195         err = -ENOMEM;
 196         nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size);
 197         if (unlikely(!nilfs->ns_dat))
 198                 goto failed;
 199
 200         nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size);
 201         if (unlikely(!nilfs->ns_gc_dat))
 202                 goto failed_dat;
 203
 204         nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size);
 205         if (unlikely(!nilfs->ns_cpfile))
 206                 goto failed_gc_dat;
 207
 208         nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
 209         if (unlikely(!nilfs->ns_sufile))
 210                 goto failed_cpfile;
 211
 212         nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
 213
 214         err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
 215                              NILFS_SR_DAT_OFFSET(inode_size));
 216         if (unlikely(err))
 217                 goto failed_sufile;
 218
 219         err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
 220                                 NILFS_SR_CPFILE_OFFSET(inode_size));
 221         if (unlikely(err))
 222                 goto failed_sufile;
 223
 224         err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
 225                                 NILFS_SR_SUFILE_OFFSET(inode_size));
 226         if (unlikely(err))
 227                 goto failed_sufile;
 228
 229         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 230         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 231
 232  failed:
 233         brelse(bh_sr);
 234         return err;
 235
 236  failed_sufile:
 237         nilfs_mdt_destroy(nilfs->ns_sufile);
 238
 239  failed_cpfile:
 240         nilfs_mdt_destroy(nilfs->ns_cpfile);
 241
 242  failed_gc_dat:
 243         nilfs_mdt_destroy(nilfs->ns_gc_dat);
 244
 245  failed_dat:
 246         nilfs_mdt_destroy(nilfs->ns_dat);
 247         goto failed;
 248 }
 249
 250 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 251 {
 252         memset(ri, 0, sizeof(*ri));
 253         INIT_LIST_HEAD(&ri->ri_used_segments);
 254 }
 255
 256 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 257 {
 258         nilfs_dispose_segment_list(&ri->ri_used_segments);
 259 }
 260
 261 /**
 262  * nilfs_store_log_cursor - load log cursor from a super block
 263  * @nilfs: nilfs object
 264  * @sbp: buffer storing super block to be read
 265  *
 266  * nilfs_store_log_cursor() reads the last position of the log
 267  * containing a super root from a given super block, and initializes
 268  * relevant information on the nilfs object preparatory for log
 269  * scanning and recovery.
 270  */
 271 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
 272                                   struct nilfs_super_block *sbp)
 273 {
 274         int ret = 0;
 275
 276         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 277         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 278         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 279
 280         nilfs->ns_prev_seq = nilfs->ns_last_seq;
 281         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 282         nilfs->ns_segnum =
 283                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 284         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 285         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 286                 printk(KERN_ERR "NILFS invalid last segment number.\n");
 287                 ret = -EINVAL;
 288         }
 289         return ret;
 290 }
 291
 292 /**
 293  * load_nilfs - load and recover the nilfs
 294  * @nilfs: the_nilfs structure to be released
 295  * @sbi: nilfs_sb_info used to recover past segment
 296  *
 297  * load_nilfs() searches and load the latest super root,
 298  * attaches the last segment, and does recovery if needed.
 299  * The caller must call this exclusively for simultaneous mounts.
 300  */
 301 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
 302 {
 303         struct nilfs_recovery_info ri;
 304         unsigned int s_flags = sbi->s_super->s_flags;
 305         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 306         int valid_fs = nilfs_valid_fs(nilfs);
 307         int err;
 308
 309         if (nilfs_loaded(nilfs)) {
 310                 if (valid_fs ||
 311                     ((s_flags & MS_RDONLY) && nilfs_test_opt(sbi, NORECOVERY)))
 312                         return 0;
 313                 printk(KERN_ERR "NILFS: the filesystem is in an incomplete "
 314                        "recovery state.\n");
 315                 return -EINVAL;
 316         }
 317
 318         if (!valid_fs) {
 319                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
 320                 if (s_flags & MS_RDONLY) {
 321                         printk(KERN_INFO "NILFS: INFO: recovery "
 322                                "required for readonly filesystem.\n");
 323                         printk(KERN_INFO "NILFS: write access will "
 324                                "be enabled during recovery.\n");
 325                 }
 326         }
 327
 328         nilfs_init_recovery_info(&ri);
 329
 330         err = nilfs_search_super_root(nilfs, &ri);
 331         if (unlikely(err)) {
 332                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
 333                 int blocksize;
 334
 335                 if (err != -EINVAL)
 336                         goto scan_error;
 337
 338                 if (!nilfs_valid_sb(sbp[1])) {
 339                         printk(KERN_WARNING
 340                                "NILFS warning: unable to fall back to spare"
 341                                "super block\n");
 342                         goto scan_error;
 343                 }
 344                 printk(KERN_INFO
 345                        "NILFS: try rollback from an earlier position\n");
 346
 347                 /*
 348                  * restore super block with its spare and reconfigure
 349                  * relevant states of the nilfs object.
 350                  */
 351                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 352                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
 353                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 354
 355                 /* verify consistency between two super blocks */
 356                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
 357                 if (blocksize != nilfs->ns_blocksize) {
 358                         printk(KERN_WARNING
 359                                "NILFS warning: blocksize differs between "
 360                                "two super blocks (%d != %d)\n",
 361                                blocksize, nilfs->ns_blocksize);
 362                         goto scan_error;
 363                 }
 364
 365                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
 366                 if (err)
 367                         goto scan_error;
 368
 369                 /* drop clean flag to allow roll-forward and recovery */
 370                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
 371                 valid_fs = 0;
 372
 373                 err = nilfs_search_super_root(nilfs, &ri);
 374                 if (err)
 375                         goto scan_error;
 376         }
 377
 378         err = nilfs_load_super_root(nilfs, ri.ri_super_root);
 379         if (unlikely(err)) {
 380                 printk(KERN_ERR "NILFS: error loading super root.\n");
 381                 goto failed;
 382         }
 383
 384         if (valid_fs)
 385                 goto skip_recovery;
 386
 387         if (s_flags & MS_RDONLY) {
 388                 __u64 features;
 389
 390                 if (nilfs_test_opt(sbi, NORECOVERY)) {
 391                         printk(KERN_INFO "NILFS: norecovery option specified. "
 392                                "skipping roll-forward recovery\n");
 393                         goto skip_recovery;
 394                 }
 395                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
 396                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
 397                 if (features) {
 398                         printk(KERN_ERR "NILFS: couldn't proceed with "
 399                                "recovery because of unsupported optional "
 400                                "features (%llx)\n",
 401                                (unsigned long long)features);
 402                         err = -EROFS;
 403                         goto failed_unload;
 404                 }
 405                 if (really_read_only) {
 406                         printk(KERN_ERR "NILFS: write access "
 407                                "unavailable, cannot proceed.\n");
 408                         err = -EROFS;
 409                         goto failed_unload;
 410                 }
 411                 sbi->s_super->s_flags &= ~MS_RDONLY;
 412         } else if (nilfs_test_opt(sbi, NORECOVERY)) {
 413                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
 414                        "option was specified for a read/write mount\n");
 415                 err = -EINVAL;
 416                 goto failed_unload;
 417         }
 418
 419         err = nilfs_salvage_orphan_logs(nilfs, sbi, &ri);
 420         if (err)
 421                 goto failed_unload;
 422
 423         down_write(&nilfs->ns_sem);
 424         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
 425         err = nilfs_cleanup_super(sbi);
 426         up_write(&nilfs->ns_sem);
 427
 428         if (err) {
 429                 printk(KERN_ERR "NILFS: failed to update super block. "
 430                        "recovery unfinished.\n");
 431                 goto failed_unload;
 432         }
 433         printk(KERN_INFO "NILFS: recovery complete.\n");
 434
 435  skip_recovery:
 436         set_nilfs_loaded(nilfs);
 437         nilfs_clear_recovery_info(&ri);
 438         sbi->s_super->s_flags = s_flags;
 439         return 0;
 440
 441  scan_error:
 442         printk(KERN_ERR "NILFS: error searching super root.\n");
 443         goto failed;
 444
 445  failed_unload:
 446         nilfs_mdt_destroy(nilfs->ns_cpfile);
 447         nilfs_mdt_destroy(nilfs->ns_sufile);
 448         nilfs_mdt_destroy(nilfs->ns_dat);
 449
 450  failed:
 451         nilfs_clear_recovery_info(&ri);
 452         sbi->s_super->s_flags = s_flags;
 453         return err;
 454 }
 455
 456 static unsigned long long nilfs_max_size(unsigned int blkbits)
 457 {
 458         unsigned int max_bits;
 459         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 460
 461         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 462         if (max_bits < 64)
 463                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 464         return res;
 465 }
 466
 467 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
 468                                    struct nilfs_super_block *sbp)
 469 {
 470         if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
 471                 printk(KERN_ERR "NILFS: revision mismatch "
 472                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
 473                        "Please check the version of mkfs.nilfs.\n",
 474                        le32_to_cpu(sbp->s_rev_level),
 475                        le16_to_cpu(sbp->s_minor_rev_level),
 476                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
 477                 return -EINVAL;
 478         }
 479         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
 480         if (nilfs->ns_sbsize > BLOCK_SIZE)
 481                 return -EINVAL;
 482
 483         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 484         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 485
 486         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 487         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 488                 printk(KERN_ERR "NILFS: too short segment.\n");
 489                 return -EINVAL;
 490         }
 491
 492         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 493         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
 494         nilfs->ns_r_segments_percentage =
 495                 le32_to_cpu(sbp->s_r_segments_percentage);
 496         nilfs->ns_nrsvsegs =
 497                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 498                       DIV_ROUND_UP(nilfs->ns_nsegments *
 499                                    nilfs->ns_r_segments_percentage, 100));
 500         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 501         return 0;
 502 }
 503
 504 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 505 {
 506         static unsigned char sum[4];
 507         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
 508         size_t bytes;
 509         u32 crc;
 510
 511         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
 512                 return 0;
 513         bytes = le16_to_cpu(sbp->s_bytes);
 514         if (bytes > BLOCK_SIZE)
 515                 return 0;
 516         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
 517                        sumoff);
 518         crc = crc32_le(crc, sum, 4);
 519         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
 520                        bytes - sumoff - 4);
 521         return crc == le32_to_cpu(sbp->s_sum);
 522 }
 523
 524 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 525 {
 526         return offset < ((le64_to_cpu(sbp->s_nsegments) *
 527                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
 528                          (le32_to_cpu(sbp->s_log_block_size) + 10));
 529 }
 530
 531 static void nilfs_release_super_block(struct the_nilfs *nilfs)
 532 {
 533         int i;
 534
 535         for (i = 0; i < 2; i++) {
 536                 if (nilfs->ns_sbp[i]) {
 537                         brelse(nilfs->ns_sbh[i]);
 538                         nilfs->ns_sbh[i] = NULL;
 539                         nilfs->ns_sbp[i] = NULL;
 540                 }
 541         }
 542 }
 543
 544 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
 545 {
 546         brelse(nilfs->ns_sbh[0]);
 547         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 548         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 549         nilfs->ns_sbh[1] = NULL;
 550         nilfs->ns_sbp[1] = NULL;
 551 }
 552
 553 void nilfs_swap_super_block(struct the_nilfs *nilfs)
 554 {
 555         struct buffer_head *tsbh = nilfs->ns_sbh[0];
 556         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
 557
 558         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
 559         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
 560         nilfs->ns_sbh[1] = tsbh;
 561         nilfs->ns_sbp[1] = tsbp;
 562 }
 563
 564 static int nilfs_load_super_block(struct the_nilfs *nilfs,
 565                                   struct super_block *sb, int blocksize,
 566                                   struct nilfs_super_block **sbpp)
 567 {
 568         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 569         struct buffer_head **sbh = nilfs->ns_sbh;
 570         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
 571         int valid[2], swp = 0;
 572
 573         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
 574                                         &sbh[0]);
 575         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
 576
 577         if (!sbp[0]) {
 578                 if (!sbp[1]) {
 579                         printk(KERN_ERR "NILFS: unable to read superblock\n");
 580                         return -EIO;
 581                 }
 582                 printk(KERN_WARNING
 583                        "NILFS warning: unable to read primary superblock\n");
 584         } else if (!sbp[1])
 585                 printk(KERN_WARNING
 586                        "NILFS warning: unable to read secondary superblock\n");
 587
 588         /*
 589          * Compare two super blocks and set 1 in swp if the secondary
 590          * super block is valid and newer.  Otherwise, set 0 in swp.
 591          */
 592         valid[0] = nilfs_valid_sb(sbp[0]);
 593         valid[1] = nilfs_valid_sb(sbp[1]);
 594         swp = valid[1] && (!valid[0] ||
 595                            le64_to_cpu(sbp[1]->s_last_cno) >
 596                            le64_to_cpu(sbp[0]->s_last_cno));
 597
 598         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
 599                 brelse(sbh[1]);
 600                 sbh[1] = NULL;
 601                 sbp[1] = NULL;
 602                 swp = 0;
 603         }
 604         if (!valid[swp]) {
 605                 nilfs_release_super_block(nilfs);
 606                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
 607                        sb->s_id);
 608                 return -EINVAL;
 609         }
 610
 611         if (!valid[!swp])
 612                 printk(KERN_WARNING "NILFS warning: broken superblock. "
 613                        "using spare superblock.\n");
 614         if (swp)
 615                 nilfs_swap_super_block(nilfs);
 616
 617         nilfs->ns_sbwcount = 0;
 618         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
 619         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
 620         *sbpp = sbp[0];
 621         return 0;
 622 }
 623
 624 /**
 625  * init_nilfs - initialize a NILFS instance.
 626  * @nilfs: the_nilfs structure
 627  * @sbi: nilfs_sb_info
 628  * @sb: super block
 629  * @data: mount options
 630  *
 631  * init_nilfs() performs common initialization per block device (e.g.
 632  * reading the super block, getting disk layout information, initializing
 633  * shared fields in the_nilfs). It takes on some portion of the jobs
 634  * typically done by a fill_super() routine. This division arises from
 635  * the nature that multiple NILFS instances may be simultaneously
 636  * mounted on a device.
 637  * For multiple mounts on the same device, only the first mount
 638  * invokes these tasks.
 639  *
 640  * Return Value: On success, 0 is returned. On error, a negative error
 641  * code is returned.
 642  */
 643 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 644 {
 645         struct super_block *sb = sbi->s_super;
 646         struct nilfs_super_block *sbp;
 647         struct backing_dev_info *bdi;
 648         int blocksize;
 649         int err;
 650
 651         down_write(&nilfs->ns_sem);
 652         if (nilfs_init(nilfs)) {
 653                 /* Load values from existing the_nilfs */
 654                 sbp = nilfs->ns_sbp[0];
 655                 err = nilfs_store_magic_and_option(sb, sbp, data);
 656                 if (err)
 657                         goto out;
 658
 659                 err = nilfs_check_feature_compatibility(sb, sbp);
 660                 if (err)
 661                         goto out;
 662
 663                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 664                 if (sb->s_blocksize != blocksize &&
 665                     !sb_set_blocksize(sb, blocksize)) {
 666                         printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
 667                                blocksize);
 668                         err = -EINVAL;
 669                 }
 670                 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 671                 goto out;
 672         }
 673
 674         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
 675         if (!blocksize) {
 676                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
 677                 err = -EINVAL;
 678                 goto out;
 679         }
 680         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 681         if (err)
 682                 goto out;
 683
 684         err = nilfs_store_magic_and_option(sb, sbp, data);
 685         if (err)
 686                 goto failed_sbh;
 687
 688         err = nilfs_check_feature_compatibility(sb, sbp);
 689         if (err)
 690                 goto failed_sbh;
 691
 692         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 693         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
 694             blocksize > NILFS_MAX_BLOCK_SIZE) {
 695                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 696                        "filesystem blocksize %d\n", blocksize);
 697                 err = -EINVAL;
 698                 goto failed_sbh;
 699         }
 700         if (sb->s_blocksize != blocksize) {
 701                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
 702
 703                 if (blocksize < hw_blocksize) {
 704                         printk(KERN_ERR
 705                                "NILFS: blocksize %d too small for device "
 706                                "(sector-size = %d).\n",
 707                                blocksize, hw_blocksize);
 708                         err = -EINVAL;
 709                         goto failed_sbh;
 710                 }
 711                 nilfs_release_super_block(nilfs);
 712                 sb_set_blocksize(sb, blocksize);
 713
 714                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
 715                 if (err)
 716                         goto out;
 717                         /* not failed_sbh; sbh is released automatically
 718                            when reloading fails. */
 719         }
 720         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 721         nilfs->ns_blocksize = blocksize;
 722
 723         err = nilfs_store_disk_layout(nilfs, sbp);
 724         if (err)
 725                 goto failed_sbh;
 726
 727         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 728
 729         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 730
 731         bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
 732         nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
 733
 734         err = nilfs_store_log_cursor(nilfs, sbp);
 735         if (err)
 736                 goto failed_sbh;
 737
 738         /* Initialize gcinode cache */
 739         err = nilfs_init_gccache(nilfs);
 740         if (err)
 741                 goto failed_sbh;
 742
 743         set_nilfs_init(nilfs);
 744         err = 0;
 745  out:
 746         up_write(&nilfs->ns_sem);
 747         return err;
 748
 749  failed_sbh:
 750         nilfs_release_super_block(nilfs);
 751         goto out;
 752 }
 753
 754 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
 755                             size_t nsegs)
 756 {
 757         sector_t seg_start, seg_end;
 758         sector_t start = 0, nblocks = 0;
 759         unsigned int sects_per_block;
 760         __u64 *sn;
 761         int ret = 0;
 762
 763         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
 764                 bdev_logical_block_size(nilfs->ns_bdev);
 765         for (sn = segnump; sn < segnump + nsegs; sn++) {
 766                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
 767
 768                 if (!nblocks) {
 769                         start = seg_start;
 770                         nblocks = seg_end - seg_start + 1;
 771                 } else if (start + nblocks == seg_start) {
 772                         nblocks += seg_end - seg_start + 1;
 773                 } else {
 774                         ret = blkdev_issue_discard(nilfs->ns_bdev,
 775                                                    start * sects_per_block,
 776                                                    nblocks * sects_per_block,
 777                                                    GFP_NOFS,
 778                                                    BLKDEV_IFL_WAIT |
 779                                                    BLKDEV_IFL_BARRIER);
 780                         if (ret < 0)
 781                                 return ret;
 782                         nblocks = 0;
 783                 }
 784         }
 785         if (nblocks)
 786                 ret = blkdev_issue_discard(nilfs->ns_bdev,
 787                                            start * sects_per_block,
 788                                            nblocks * sects_per_block,
 789                                            GFP_NOFS,
 790                                           BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
 791         return ret;
 792 }
 793
 794 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 795 {
 796         struct inode *dat = nilfs_dat_inode(nilfs);
 797         unsigned long ncleansegs;
 798
 799         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 800         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 801         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 802         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 803         return 0;
 804 }
 805
 806 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 807 {
 808         unsigned long ncleansegs, nincsegs;
 809
 810         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
 811         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 812                 nilfs->ns_blocks_per_segment + 1;
 813
 814         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
 815 }
 816
 817 /**
 818  * nilfs_find_sbinfo - find existing nilfs_sb_info structure
 819  * @nilfs: nilfs object
 820  * @rw_mount: mount type (non-zero value for read/write mount)
 821  * @cno: checkpoint number (zero for read-only mount)
 822  *
 823  * nilfs_find_sbinfo() returns the nilfs_sb_info structure which
 824  * @rw_mount and @cno (in case of snapshots) matched.  If no instance
 825  * was found, NULL is returned.  Although the super block instance can
 826  * be unmounted after this function returns, the nilfs_sb_info struct
 827  * is kept on memory until nilfs_put_sbinfo() is called.
 828  */
 829 struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *nilfs,
 830                                         int rw_mount, __u64 cno)
 831 {
 832         struct nilfs_sb_info *sbi;
 833
 834         down_read(&nilfs->ns_super_sem);
 835         /*
 836          * The SNAPSHOT flag and sb->s_flags are supposed to be
 837          * protected with nilfs->ns_super_sem.
 838          */
 839         sbi = nilfs->ns_current;
 840         if (rw_mount) {
 841                 if (sbi && !(sbi->s_super->s_flags & MS_RDONLY))
 842                         goto found; /* read/write mount */
 843                 else
 844                         goto out;
 845         } else if (cno == 0) {
 846                 if (sbi && (sbi->s_super->s_flags & MS_RDONLY))
 847                         goto found; /* read-only mount */
 848                 else
 849                         goto out;
 850         }
 851
 852         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
 853                 if (nilfs_test_opt(sbi, SNAPSHOT) &&
 854                     sbi->s_snapshot_cno == cno)
 855                         goto found; /* snapshot mount */
 856         }
 857  out:
 858         up_read(&nilfs->ns_super_sem);
 859         return NULL;
 860
 861  found:
 862         atomic_inc(&sbi->s_count);
 863         up_read(&nilfs->ns_super_sem);
 864         return sbi;
 865 }
 866
 867 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
 868                                 int snapshot_mount)
 869 {
 870         struct nilfs_sb_info *sbi;
 871         int ret = 0;
 872
 873         down_read(&nilfs->ns_super_sem);
 874         if (cno == 0 || cno > nilfs->ns_cno)
 875                 goto out_unlock;
 876
 877         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
 878                 if (sbi->s_snapshot_cno == cno &&
 879                     (!snapshot_mount || nilfs_test_opt(sbi, SNAPSHOT))) {
 880                                         /* exclude read-only mounts */
 881                         ret++;
 882                         break;
 883                 }
 884         }
 885         /* for protecting recent checkpoints */
 886         if (cno >= nilfs_last_cno(nilfs))
 887                 ret++;
 888
 889  out_unlock:
 890         up_read(&nilfs->ns_super_sem);
 891         return ret;
 892 }