src/ext4_journal.c

   1 /*
   2  * Copyright (c) 2015 Grzegorz Kostka (kostka.grzegorz@gmail.com)
   3  * Copyright (c) 2015 Kaho Ng (ngkaho1234@gmail.com)
   4  * All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  *
  10  * - Redistributions of source code must retain the above copyright
  11  *   notice, this list of conditions and the following disclaimer.
  12  * - Redistributions in binary form must reproduce the above copyright
  13  *   notice, this list of conditions and the following disclaimer in the
  14  *   documentation and/or other materials provided with the distribution.
  15  * - The name of the author may not be used to endorse or promote products
  16  *   derived from this software without specific prior written permission.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28  */
  29
  30 /** @addtogroup lwext4
  31  * @{
  32  */
  33 /**
  34  * @file  ext4_journal.c
  35  * @brief Journal handle functions
  36  */
  37
  38 #include "ext4_config.h"
  39 #include "ext4_types.h"
  40 #include "ext4_misc.h"
  41 #include "ext4_errno.h"
  42 #include "ext4_debug.h"
  43
  44 #include "ext4_fs.h"
  45 #include "ext4_super.h"
  46 #include "ext4_journal.h"
  47 #include "ext4_blockdev.h"
  48 #include "ext4_crc32.h"
  49 #include "ext4_journal.h"
  50
  51 #include <string.h>
  52 #include <stdlib.h>
  53
  54 /**@brief  Revoke entry during journal replay.*/
  55 struct revoke_entry {
  56         /**@brief  Block number not to be replayed.*/
  57         ext4_fsblk_t block;
  58
  59         /**@brief  For any transaction id smaller
  60          *         than trans_id, records of @block
  61          *         in those transactions should not
  62          *         be replayed.*/
  63         uint32_t trans_id;
  64
  65         /**@brief  Revoke tree node.*/
  66         RB_ENTRY(revoke_entry) revoke_node;
  67 };
  68
  69 /**@brief  Valid journal replay information.*/
  70 struct recover_info {
  71         /**@brief  Starting transaction id.*/
  72         uint32_t start_trans_id;
  73
  74         /**@brief  Ending transaction id.*/
  75         uint32_t last_trans_id;
  76
  77         /**@brief  Used as internal argument.*/
  78         uint32_t this_trans_id;
  79
  80         /**@brief  No of transactions went through.*/
  81         uint32_t trans_cnt;
  82
  83         /**@brief  RB-Tree storing revoke entries.*/
  84         RB_HEAD(jbd_revoke, revoke_entry) revoke_root;
  85 };
  86
  87 /**@brief  Journal replay internal arguments.*/
  88 struct replay_arg {
  89         /**@brief  Journal replay information.*/
  90         struct recover_info *info;
  91
  92         /**@brief  Current block we are on.*/
  93         uint32_t *this_block;
  94
  95         /**@brief  Current trans_id we are on.*/
  96         uint32_t this_trans_id;
  97 };
  98
  99 /* Make sure we wrap around the log correctly! */
 100 #define wrap(sb, var)                                           \
 101 do {                                                                    \
 102         if (var >= jbd_get32((sb), maxlen))                                     \
 103                 var -= (jbd_get32((sb), maxlen) - jbd_get32((sb), first));      \
 104 } while (0)
 105
 106 static inline int32_t
 107 trans_id_diff(uint32_t x, uint32_t y)
 108 {
 109         int32_t diff = x - y;
 110         return diff;
 111 }
 112
 113 static int
 114 jbd_revoke_entry_cmp(struct revoke_entry *a, struct revoke_entry *b)
 115 {
 116         if (a->block > b->block)
 117                 return 1;
 118         else if (a->block < b->block)
 119                 return -1;
 120         return 0;
 121 }
 122
 123 static int
 124 jbd_block_rec_cmp(struct jbd_block_rec *a, struct jbd_block_rec *b)
 125 {
 126         if (a->lba > b->lba)
 127                 return 1;
 128         else if (a->lba < b->lba)
 129                 return -1;
 130         return 0;
 131 }
 132
 133 static int
 134 jbd_revoke_rec_cmp(struct jbd_revoke_rec *a, struct jbd_revoke_rec *b)
 135 {
 136         if (a->lba > b->lba)
 137                 return 1;
 138         else if (a->lba < b->lba)
 139                 return -1;
 140         return 0;
 141 }
 142
 143 RB_GENERATE_INTERNAL(jbd_revoke, revoke_entry, revoke_node,
 144                      jbd_revoke_entry_cmp, static inline)
 145 RB_GENERATE_INTERNAL(jbd_block, jbd_block_rec, block_rec_node,
 146                      jbd_block_rec_cmp, static inline)
 147 RB_GENERATE_INTERNAL(jbd_revoke_tree, jbd_revoke_rec, revoke_node,
 148                      jbd_revoke_rec_cmp, static inline)
 149
 150 #define jbd_alloc_revoke_entry() ext4_calloc(1, sizeof(struct revoke_entry))
 151 #define jbd_free_revoke_entry(addr) ext4_free(addr)
 152
 153 static int jbd_has_csum(struct jbd_sb *jbd_sb)
 154 {
 155         if (JBD_HAS_INCOMPAT_FEATURE(jbd_sb, JBD_FEATURE_INCOMPAT_CSUM_V2))
 156                 return 2;
 157
 158         if (JBD_HAS_INCOMPAT_FEATURE(jbd_sb, JBD_FEATURE_INCOMPAT_CSUM_V3))
 159                 return 3;
 160
 161         return 0;
 162 }
 163
 164 #if CONFIG_META_CSUM_ENABLE
 165 static uint32_t jbd_sb_csum(struct jbd_sb *jbd_sb)
 166 {
 167         uint32_t checksum = 0;
 168
 169         if (jbd_has_csum(jbd_sb)) {
 170                 uint32_t orig_checksum = jbd_sb->checksum;
 171                 jbd_set32(jbd_sb, checksum, 0);
 172                 /* Calculate crc32c checksum against tho whole superblock */
 173                 checksum = ext4_crc32c(EXT4_CRC32_INIT, jbd_sb,
 174                                 JBD_SUPERBLOCK_SIZE);
 175                 jbd_sb->checksum = orig_checksum;
 176         }
 177         return checksum;
 178 }
 179 #else
 180 #define jbd_sb_csum(...) 0
 181 #endif
 182
 183 static void jbd_sb_csum_set(struct jbd_sb *jbd_sb)
 184 {
 185         if (!jbd_has_csum(jbd_sb))
 186                 return;
 187
 188         jbd_set32(jbd_sb, checksum, jbd_sb_csum(jbd_sb));
 189 }
 190
 191 #if CONFIG_META_CSUM_ENABLE
 192 static bool
 193 jbd_verify_sb_csum(struct jbd_sb *jbd_sb)
 194 {
 195         if (!jbd_has_csum(jbd_sb))
 196                 return true;
 197
 198         return jbd_sb_csum(jbd_sb) == jbd_get32(jbd_sb, checksum);
 199 }
 200 #else
 201 #define jbd_verify_sb_csum(...) true
 202 #endif
 203
 204 #if CONFIG_META_CSUM_ENABLE
 205 static uint32_t jbd_meta_csum(struct jbd_fs *jbd_fs,
 206                               struct jbd_bhdr *bhdr)
 207 {
 208         uint32_t checksum = 0;
 209
 210         if (jbd_has_csum(&jbd_fs->sb)) {
 211                 uint32_t block_size = jbd_get32(&jbd_fs->sb, blocksize);
 212                 struct jbd_block_tail *tail =
 213                         (struct jbd_block_tail *)((char *)bhdr + block_size -
 214                                 sizeof(struct jbd_block_tail));
 215                 uint32_t orig_checksum = tail->checksum;
 216                 tail->checksum = 0;
 217
 218                 /* First calculate crc32c checksum against fs uuid */
 219                 checksum = ext4_crc32c(EXT4_CRC32_INIT, jbd_fs->sb.uuid,
 220                                        sizeof(jbd_fs->sb.uuid));
 221                 /* Calculate crc32c checksum against tho whole block */
 222                 checksum = ext4_crc32c(checksum, bhdr,
 223                                 block_size);
 224                 tail->checksum = orig_checksum;
 225         }
 226         return checksum;
 227 }
 228 #else
 229 #define jbd_meta_csum(...) 0
 230 #endif
 231
 232 static void jbd_meta_csum_set(struct jbd_fs *jbd_fs,
 233                               struct jbd_bhdr *bhdr)
 234 {
 235         uint32_t block_size = jbd_get32(&jbd_fs->sb, blocksize);
 236         struct jbd_block_tail *tail = (struct jbd_block_tail *)
 237                                 ((char *)bhdr + block_size -
 238                                 sizeof(struct jbd_block_tail));
 239         if (!jbd_has_csum(&jbd_fs->sb))
 240                 return;
 241
 242         tail->checksum = to_be32(jbd_meta_csum(jbd_fs, bhdr));
 243 }
 244
 245 #if CONFIG_META_CSUM_ENABLE
 246 static bool
 247 jbd_verify_meta_csum(struct jbd_fs *jbd_fs,
 248                      struct jbd_bhdr *bhdr)
 249 {
 250         uint32_t block_size = jbd_get32(&jbd_fs->sb, blocksize);
 251         struct jbd_block_tail *tail = (struct jbd_block_tail *)
 252                                 ((char *)bhdr + block_size -
 253                                 sizeof(struct jbd_block_tail));
 254         if (!jbd_has_csum(&jbd_fs->sb))
 255                 return true;
 256
 257         return jbd_meta_csum(jbd_fs, bhdr) == to_be32(tail->checksum);
 258 }
 259 #else
 260 #define jbd_verify_meta_csum(...) true
 261 #endif
 262
 263 #if CONFIG_META_CSUM_ENABLE
 264 static uint32_t jbd_commit_csum(struct jbd_fs *jbd_fs,
 265                               struct jbd_commit_header *header)
 266 {
 267         uint32_t checksum = 0;
 268
 269         if (jbd_has_csum(&jbd_fs->sb)) {
 270                 uint32_t orig_checksum_type = header->chksum_type,
 271                          orig_checksum_size = header->chksum_size,
 272                          orig_checksum = header->chksum[0];
 273                 uint32_t block_size = jbd_get32(&jbd_fs->sb, blocksize);
 274                 header->chksum_type = 0;
 275                 header->chksum_size = 0;
 276                 header->chksum[0] = 0;
 277
 278                 /* First calculate crc32c checksum against fs uuid */
 279                 checksum = ext4_crc32c(EXT4_CRC32_INIT, jbd_fs->sb.uuid,
 280                                        sizeof(jbd_fs->sb.uuid));
 281                 /* Calculate crc32c checksum against tho whole block */
 282                 checksum = ext4_crc32c(checksum, header,
 283                                 block_size);
 284
 285                 header->chksum_type = orig_checksum_type;
 286                 header->chksum_size = orig_checksum_size;
 287                 header->chksum[0] = orig_checksum;
 288         }
 289         return checksum;
 290 }
 291 #else
 292 #define jbd_commit_csum(...) 0
 293 #endif
 294
 295 static void jbd_commit_csum_set(struct jbd_fs *jbd_fs,
 296                               struct jbd_commit_header *header)
 297 {
 298         if (!jbd_has_csum(&jbd_fs->sb))
 299                 return;
 300
 301         header->chksum_type = 0;
 302         header->chksum_size = 0;
 303         header->chksum[0] = jbd_commit_csum(jbd_fs, header);
 304 }
 305
 306 #if CONFIG_META_CSUM_ENABLE
 307 static bool jbd_verify_commit_csum(struct jbd_fs *jbd_fs,
 308                                    struct jbd_commit_header *header)
 309 {
 310         if (!jbd_has_csum(&jbd_fs->sb))
 311                 return true;
 312
 313         return header->chksum[0] == to_be32(jbd_commit_csum(jbd_fs,
 314                                             header));
 315 }
 316 #else
 317 #define jbd_verify_commit_csum(...) true
 318 #endif
 319
 320 #if CONFIG_META_CSUM_ENABLE
 321 /*
 322  * NOTE: We only make use of @csum parameter when
 323  *       JBD_FEATURE_COMPAT_CHECKSUM is enabled.
 324  */
 325 static uint32_t jbd_block_csum(struct jbd_fs *jbd_fs, const void *buf,
 326                                uint32_t csum,
 327                                uint32_t sequence)
 328 {
 329         uint32_t checksum = 0;
 330
 331         if (jbd_has_csum(&jbd_fs->sb)) {
 332                 uint32_t block_size = jbd_get32(&jbd_fs->sb, blocksize);
 333                 /* First calculate crc32c checksum against fs uuid */
 334                 checksum = ext4_crc32c(EXT4_CRC32_INIT, jbd_fs->sb.uuid,
 335                                        sizeof(jbd_fs->sb.uuid));
 336                 /* Then calculate crc32c checksum against sequence no. */
 337                 checksum = ext4_crc32c(checksum, &sequence,
 338                                 sizeof(uint32_t));
 339                 /* Calculate crc32c checksum against tho whole block */
 340                 checksum = ext4_crc32c(checksum, buf,
 341                                 block_size);
 342         } else if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 343                                      JBD_FEATURE_COMPAT_CHECKSUM)) {
 344                 uint32_t block_size = jbd_get32(&jbd_fs->sb, blocksize);
 345                 /* Calculate crc32c checksum against tho whole block */
 346                 checksum = ext4_crc32(csum, buf,
 347                                 block_size);
 348         }
 349         return checksum;
 350 }
 351 #else
 352 #define jbd_block_csum(...) 0
 353 #endif
 354
 355 static void jbd_block_tag_csum_set(struct jbd_fs *jbd_fs, void *__tag,
 356                                    uint32_t checksum)
 357 {
 358         int ver = jbd_has_csum(&jbd_fs->sb);
 359         if (!ver)
 360                 return;
 361
 362         if (ver == 2) {
 363                 struct jbd_block_tag *tag = __tag;
 364                 tag->checksum = (uint16_t)to_be32(checksum);
 365         } else {
 366                 struct jbd_block_tag3 *tag = __tag;
 367                 tag->checksum = to_be32(checksum);
 368         }
 369 }
 370
 371 /**@brief  Write jbd superblock to disk.
 372  * @param  jbd_fs jbd filesystem
 373  * @param  s jbd superblock
 374  * @return standard error code*/
 375 static int jbd_sb_write(struct jbd_fs *jbd_fs, struct jbd_sb *s)
 376 {
 377         int rc;
 378         struct ext4_fs *fs = jbd_fs->inode_ref.fs;
 379         uint64_t offset;
 380         ext4_fsblk_t fblock;
 381         rc = jbd_inode_bmap(jbd_fs, 0, &fblock);
 382         if (rc != EOK)
 383                 return rc;
 384
 385         jbd_sb_csum_set(s);
 386         offset = fblock * ext4_sb_get_block_size(&fs->sb);
 387         return ext4_block_writebytes(fs->bdev, offset, s,
 388                                      EXT4_SUPERBLOCK_SIZE);
 389 }
 390
 391 /**@brief  Read jbd superblock from disk.
 392  * @param  jbd_fs jbd filesystem
 393  * @param  s jbd superblock
 394  * @return standard error code*/
 395 static int jbd_sb_read(struct jbd_fs *jbd_fs, struct jbd_sb *s)
 396 {
 397         int rc;
 398         struct ext4_fs *fs = jbd_fs->inode_ref.fs;
 399         uint64_t offset;
 400         ext4_fsblk_t fblock;
 401         rc = jbd_inode_bmap(jbd_fs, 0, &fblock);
 402         if (rc != EOK)
 403                 return rc;
 404
 405         offset = fblock * ext4_sb_get_block_size(&fs->sb);
 406         return ext4_block_readbytes(fs->bdev, offset, s,
 407                                     EXT4_SUPERBLOCK_SIZE);
 408 }
 409
 410 /**@brief  Verify jbd superblock.
 411  * @param  sb jbd superblock
 412  * @return true if jbd superblock is valid */
 413 static bool jbd_verify_sb(struct jbd_sb *sb)
 414 {
 415         struct jbd_bhdr *header = &sb->header;
 416         if (jbd_get32(header, magic) != JBD_MAGIC_NUMBER)
 417                 return false;
 418
 419         if (jbd_get32(header, blocktype) != JBD_SUPERBLOCK &&
 420             jbd_get32(header, blocktype) != JBD_SUPERBLOCK_V2)
 421                 return false;
 422
 423         return jbd_verify_sb_csum(sb);
 424 }
 425
 426 /**@brief  Write back dirty jbd superblock to disk.
 427  * @param  jbd_fs jbd filesystem
 428  * @return standard error code*/
 429 static int jbd_write_sb(struct jbd_fs *jbd_fs)
 430 {
 431         int rc = EOK;
 432         if (jbd_fs->dirty) {
 433                 rc = jbd_sb_write(jbd_fs, &jbd_fs->sb);
 434                 if (rc != EOK)
 435                         return rc;
 436
 437                 jbd_fs->dirty = false;
 438         }
 439         return rc;
 440 }
 441
 442 /**@brief  Get reference to jbd filesystem.
 443  * @param  fs Filesystem to load journal of
 444  * @param  jbd_fs jbd filesystem
 445  * @return standard error code*/
 446 int jbd_get_fs(struct ext4_fs *fs,
 447                struct jbd_fs *jbd_fs)
 448 {
 449         int rc;
 450         uint32_t journal_ino;
 451
 452         memset(jbd_fs, 0, sizeof(struct jbd_fs));
 453         /* See if there is journal inode on this filesystem.*/
 454         /* FIXME: detection on existance ofbkejournal bdev is
 455          *        missing.*/
 456         journal_ino = ext4_get32(&fs->sb, journal_inode_number);
 457
 458         rc = ext4_fs_get_inode_ref(fs,
 459                                    journal_ino,
 460                                    &jbd_fs->inode_ref);
 461         if (rc != EOK) {
 462                 memset(jbd_fs, 0, sizeof(struct jbd_fs));
 463                 return rc;
 464         }
 465         rc = jbd_sb_read(jbd_fs, &jbd_fs->sb);
 466         if (rc != EOK) {
 467                 memset(jbd_fs, 0, sizeof(struct jbd_fs));
 468                 ext4_fs_put_inode_ref(&jbd_fs->inode_ref);
 469                 return rc;
 470         }
 471         if (!jbd_verify_sb(&jbd_fs->sb)) {
 472                 memset(jbd_fs, 0, sizeof(struct jbd_fs));
 473                 ext4_fs_put_inode_ref(&jbd_fs->inode_ref);
 474                 rc = EIO;
 475         }
 476
 477         if (rc == EOK)
 478                 jbd_fs->bdev = fs->bdev;
 479
 480         return rc;
 481 }
 482
 483 /**@brief  Put reference of jbd filesystem.
 484  * @param  jbd_fs jbd filesystem
 485  * @return standard error code*/
 486 int jbd_put_fs(struct jbd_fs *jbd_fs)
 487 {
 488         int rc = EOK;
 489         rc = jbd_write_sb(jbd_fs);
 490
 491         ext4_fs_put_inode_ref(&jbd_fs->inode_ref);
 492         return rc;
 493 }
 494
 495 /**@brief  Data block lookup helper.
 496  * @param  jbd_fs jbd filesystem
 497  * @param  iblock block index
 498  * @param  fblock logical block address
 499  * @return standard error code*/
 500 int jbd_inode_bmap(struct jbd_fs *jbd_fs,
 501                    ext4_lblk_t iblock,
 502                    ext4_fsblk_t *fblock)
 503 {
 504         int rc = ext4_fs_get_inode_dblk_idx(
 505                         &jbd_fs->inode_ref,
 506                         iblock,
 507                         fblock,
 508                         false);
 509         return rc;
 510 }
 511
 512 /**@brief   jbd block get function (through cache).
 513  * @param   jbd_fs jbd filesystem
 514  * @param   block block descriptor
 515  * @param   fblock jbd logical block address
 516  * @return  standard error code*/
 517 static int jbd_block_get(struct jbd_fs *jbd_fs,
 518                   struct ext4_block *block,
 519                   ext4_fsblk_t fblock)
 520 {
 521         /* TODO: journal device. */
 522         int rc;
 523         struct ext4_blockdev *bdev = jbd_fs->bdev;
 524         ext4_lblk_t iblock = (ext4_lblk_t)fblock;
 525
 526         /* Lookup the logical block address of
 527          * fblock.*/
 528         rc = jbd_inode_bmap(jbd_fs, iblock,
 529                             &fblock);
 530         if (rc != EOK)
 531                 return rc;
 532
 533         rc = ext4_block_get(bdev, block, fblock);
 534
 535         /* If succeeded, mark buffer as BC_FLUSH to indicate
 536          * that data should be written to disk immediately.*/
 537         if (rc == EOK) {
 538                 ext4_bcache_set_flag(block->buf, BC_FLUSH);
 539                 /* As we don't want to occupy too much space
 540                  * in block cache, we set this buffer BC_TMP.*/
 541                 ext4_bcache_set_flag(block->buf, BC_TMP);
 542         }
 543
 544         return rc;
 545 }
 546
 547 /**@brief   jbd block get function (through cache, don't read).
 548  * @param   jbd_fs jbd filesystem
 549  * @param   block block descriptor
 550  * @param   fblock jbd logical block address
 551  * @return  standard error code*/
 552 static int jbd_block_get_noread(struct jbd_fs *jbd_fs,
 553                          struct ext4_block *block,
 554                          ext4_fsblk_t fblock)
 555 {
 556         /* TODO: journal device. */
 557         int rc;
 558         struct ext4_blockdev *bdev = jbd_fs->bdev;
 559         ext4_lblk_t iblock = (ext4_lblk_t)fblock;
 560         rc = jbd_inode_bmap(jbd_fs, iblock,
 561                             &fblock);
 562         if (rc != EOK)
 563                 return rc;
 564
 565         rc = ext4_block_get_noread(bdev, block, fblock);
 566         if (rc == EOK)
 567                 ext4_bcache_set_flag(block->buf, BC_FLUSH);
 568
 569         return rc;
 570 }
 571
 572 /**@brief   jbd block set procedure (through cache).
 573  * @param   jbd_fs jbd filesystem
 574  * @param   block block descriptor
 575  * @return  standard error code*/
 576 static int jbd_block_set(struct jbd_fs *jbd_fs,
 577                   struct ext4_block *block)
 578 {
 579         struct ext4_blockdev *bdev = jbd_fs->bdev;
 580         return ext4_block_set(bdev, block);
 581 }
 582
 583 /**@brief  helper functions to calculate
 584  *         block tag size, not including UUID part.
 585  * @param  jbd_fs jbd filesystem
 586  * @return tag size in bytes*/
 587 static int jbd_tag_bytes(struct jbd_fs *jbd_fs)
 588 {
 589         int size;
 590
 591         /* It is very easy to deal with the case which
 592          * JBD_FEATURE_INCOMPAT_CSUM_V3 is enabled.*/
 593         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 594                                      JBD_FEATURE_INCOMPAT_CSUM_V3))
 595                 return sizeof(struct jbd_block_tag3);
 596
 597         size = sizeof(struct jbd_block_tag);
 598
 599         /* If JBD_FEATURE_INCOMPAT_CSUM_V2 is enabled,
 600          * add 2 bytes to size.*/
 601         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 602                                      JBD_FEATURE_INCOMPAT_CSUM_V2))
 603                 size += sizeof(uint16_t);
 604
 605         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 606                                      JBD_FEATURE_INCOMPAT_64BIT))
 607                 return size;
 608
 609         /* If block number is 4 bytes in size,
 610          * minus 4 bytes from size */
 611         return size - sizeof(uint32_t);
 612 }
 613
 614 /**@brief  Tag information. */
 615 struct tag_info {
 616         /**@brief  Tag size in bytes, including UUID part.*/
 617         int tag_bytes;
 618
 619         /**@brief  block number stored in this tag.*/
 620         ext4_fsblk_t block;
 621
 622         /**@brief  Is the first 4 bytes of block equals to
 623          *         JBD_MAGIC_NUMBER? */
 624         bool is_escape;
 625
 626         /**@brief  whether UUID part exists or not.*/
 627         bool uuid_exist;
 628
 629         /**@brief  UUID content if UUID part exists.*/
 630         uint8_t uuid[UUID_SIZE];
 631
 632         /**@brief  Is this the last tag? */
 633         bool last_tag;
 634
 635         /**@brief  crc32c checksum. */
 636         uint32_t checksum;
 637 };
 638
 639 /**@brief  Extract information from a block tag.
 640  * @param  __tag pointer to the block tag
 641  * @param  tag_bytes block tag size of this jbd filesystem
 642  * @param  remaining size in buffer containing the block tag
 643  * @param  tag_info information of this tag.
 644  * @return  EOK when succeed, otherwise return EINVAL.*/
 645 static int
 646 jbd_extract_block_tag(struct jbd_fs *jbd_fs,
 647                       void *__tag,
 648                       int tag_bytes,
 649                       int32_t remain_buf_size,
 650                       struct tag_info *tag_info)
 651 {
 652         char *uuid_start;
 653         tag_info->tag_bytes = tag_bytes;
 654         tag_info->uuid_exist = false;
 655         tag_info->last_tag = false;
 656         tag_info->is_escape = false;
 657
 658         /* See whether it is possible to hold a valid block tag.*/
 659         if (remain_buf_size - tag_bytes < 0)
 660                 return EINVAL;
 661
 662         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 663                                      JBD_FEATURE_INCOMPAT_CSUM_V3)) {
 664                 struct jbd_block_tag3 *tag = __tag;
 665                 tag_info->block = jbd_get32(tag, blocknr);
 666                 if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 667                                              JBD_FEATURE_INCOMPAT_64BIT))
 668                          tag_info->block |=
 669                                  (uint64_t)jbd_get32(tag, blocknr_high) << 32;
 670
 671                 if (jbd_get32(tag, flags) & JBD_FLAG_ESCAPE)
 672                         tag_info->is_escape = true;
 673
 674                 if (!(jbd_get32(tag, flags) & JBD_FLAG_SAME_UUID)) {
 675                         /* See whether it is possible to hold UUID part.*/
 676                         if (remain_buf_size - tag_bytes < UUID_SIZE)
 677                                 return EINVAL;
 678
 679                         uuid_start = (char *)tag + tag_bytes;
 680                         tag_info->uuid_exist = true;
 681                         tag_info->tag_bytes += UUID_SIZE;
 682                         memcpy(tag_info->uuid, uuid_start, UUID_SIZE);
 683                 }
 684
 685                 if (jbd_get32(tag, flags) & JBD_FLAG_LAST_TAG)
 686                         tag_info->last_tag = true;
 687
 688         } else {
 689                 struct jbd_block_tag *tag = __tag;
 690                 tag_info->block = jbd_get32(tag, blocknr);
 691                 if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 692                                              JBD_FEATURE_INCOMPAT_64BIT))
 693                          tag_info->block |=
 694                                  (uint64_t)jbd_get32(tag, blocknr_high) << 32;
 695
 696                 if (jbd_get16(tag, flags) & JBD_FLAG_ESCAPE)
 697                         tag_info->is_escape = true;
 698
 699                 if (!(jbd_get16(tag, flags) & JBD_FLAG_SAME_UUID)) {
 700                         /* See whether it is possible to hold UUID part.*/
 701                         if (remain_buf_size - tag_bytes < UUID_SIZE)
 702                                 return EINVAL;
 703
 704                         uuid_start = (char *)tag + tag_bytes;
 705                         tag_info->uuid_exist = true;
 706                         tag_info->tag_bytes += UUID_SIZE;
 707                         memcpy(tag_info->uuid, uuid_start, UUID_SIZE);
 708                 }
 709
 710                 if (jbd_get16(tag, flags) & JBD_FLAG_LAST_TAG)
 711                         tag_info->last_tag = true;
 712
 713         }
 714         return EOK;
 715 }
 716
 717 /**@brief  Write information to a block tag.
 718  * @param  __tag pointer to the block tag
 719  * @param  remaining size in buffer containing the block tag
 720  * @param  tag_info information of this tag.
 721  * @return  EOK when succeed, otherwise return EINVAL.*/
 722 static int
 723 jbd_write_block_tag(struct jbd_fs *jbd_fs,
 724                     void *__tag,
 725                     int32_t remain_buf_size,
 726                     struct tag_info *tag_info)
 727 {
 728         char *uuid_start;
 729         int tag_bytes = jbd_tag_bytes(jbd_fs);
 730
 731         tag_info->tag_bytes = tag_bytes;
 732
 733         /* See whether it is possible to hold a valid block tag.*/
 734         if (remain_buf_size - tag_bytes < 0)
 735                 return EINVAL;
 736
 737         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 738                                      JBD_FEATURE_INCOMPAT_CSUM_V3)) {
 739                 struct jbd_block_tag3 *tag = __tag;
 740                 memset(tag, 0, sizeof(struct jbd_block_tag3));
 741                 jbd_set32(tag, blocknr, (uint32_t)tag_info->block);
 742                 if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 743                                              JBD_FEATURE_INCOMPAT_64BIT))
 744                         jbd_set32(tag, blocknr_high, tag_info->block >> 32);
 745
 746                 if (tag_info->uuid_exist) {
 747                         /* See whether it is possible to hold UUID part.*/
 748                         if (remain_buf_size - tag_bytes < UUID_SIZE)
 749                                 return EINVAL;
 750
 751                         uuid_start = (char *)tag + tag_bytes;
 752                         tag_info->tag_bytes += UUID_SIZE;
 753                         memcpy(uuid_start, tag_info->uuid, UUID_SIZE);
 754                 } else
 755                         jbd_set32(tag, flags,
 756                                   jbd_get32(tag, flags) | JBD_FLAG_SAME_UUID);
 757
 758                 jbd_block_tag_csum_set(jbd_fs, __tag, tag_info->checksum);
 759
 760                 if (tag_info->last_tag)
 761                         jbd_set32(tag, flags,
 762                                   jbd_get32(tag, flags) | JBD_FLAG_LAST_TAG);
 763
 764                 if (tag_info->is_escape)
 765                         jbd_set32(tag, flags,
 766                                   jbd_get32(tag, flags) | JBD_FLAG_ESCAPE);
 767
 768         } else {
 769                 struct jbd_block_tag *tag = __tag;
 770                 memset(tag, 0, sizeof(struct jbd_block_tag));
 771                 jbd_set32(tag, blocknr, (uint32_t)tag_info->block);
 772                 if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 773                                              JBD_FEATURE_INCOMPAT_64BIT))
 774                         jbd_set32(tag, blocknr_high, tag_info->block >> 32);
 775
 776                 if (tag_info->uuid_exist) {
 777                         /* See whether it is possible to hold UUID part.*/
 778                         if (remain_buf_size - tag_bytes < UUID_SIZE)
 779                                 return EINVAL;
 780
 781                         uuid_start = (char *)tag + tag_bytes;
 782                         tag_info->tag_bytes += UUID_SIZE;
 783                         memcpy(uuid_start, tag_info->uuid, UUID_SIZE);
 784                 } else
 785                         jbd_set16(tag, flags,
 786                                   jbd_get16(tag, flags) | JBD_FLAG_SAME_UUID);
 787
 788                 jbd_block_tag_csum_set(jbd_fs, __tag, tag_info->checksum);
 789
 790                 if (tag_info->last_tag)
 791                         jbd_set16(tag, flags,
 792                                   jbd_get16(tag, flags) | JBD_FLAG_LAST_TAG);
 793
 794
 795                 if (tag_info->is_escape)
 796                         jbd_set16(tag, flags,
 797                                   jbd_get16(tag, flags) | JBD_FLAG_ESCAPE);
 798
 799         }
 800         return EOK;
 801 }
 802
 803 /**@brief  Iterate all block tags in a block.
 804  * @param  jbd_fs jbd filesystem
 805  * @param  __tag_start pointer to the block
 806  * @param  tag_tbl_size size of the block
 807  * @param  func callback routine to indicate that
 808  *         a block tag is found
 809  * @param  arg additional argument to be passed to func */
 810 static void
 811 jbd_iterate_block_table(struct jbd_fs *jbd_fs,
 812                         void *__tag_start,
 813                         int32_t tag_tbl_size,
 814                         void (*func)(struct jbd_fs * jbd_fs,
 815                                      struct tag_info *tag_info,
 816                                      void *arg),
 817                         void *arg)
 818 {
 819         char *tag_start, *tag_ptr;
 820         int tag_bytes = jbd_tag_bytes(jbd_fs);
 821         tag_start = __tag_start;
 822         tag_ptr = tag_start;
 823
 824         /* Cut off the size of block tail storing checksum. */
 825         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 826                                      JBD_FEATURE_INCOMPAT_CSUM_V2) ||
 827             JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
 828                                      JBD_FEATURE_INCOMPAT_CSUM_V3))
 829                 tag_tbl_size -= sizeof(struct jbd_block_tail);
 830
 831         while (tag_tbl_size) {
 832                 struct tag_info tag_info;
 833                 int rc = jbd_extract_block_tag(jbd_fs,
 834                                       tag_ptr,
 835                                       tag_bytes,
 836                                       tag_tbl_size,
 837                                       &tag_info);
 838                 if (rc != EOK)
 839                         break;
 840
 841                 if (func)
 842                         func(jbd_fs, &tag_info, arg);
 843
 844                 /* Stop the iteration when we reach the last tag. */
 845                 if (tag_info.last_tag)
 846                         break;
 847
 848                 tag_ptr += tag_info.tag_bytes;
 849                 tag_tbl_size -= tag_info.tag_bytes;
 850         }
 851 }
 852
 853 static void jbd_display_block_tags(struct jbd_fs *jbd_fs,
 854                                    struct tag_info *tag_info,
 855                                    void *arg)
 856 {
 857         uint32_t *iblock = arg;
 858         ext4_dbg(DEBUG_JBD, "Block in block_tag: %" PRIu64 "\n", tag_info->block);
 859         (*iblock)++;
 860         wrap(&jbd_fs->sb, *iblock);
 861         (void)jbd_fs;
 862         return;
 863 }
 864
 865 static struct revoke_entry *
 866 jbd_revoke_entry_lookup(struct recover_info *info, ext4_fsblk_t block)
 867 {
 868         struct revoke_entry tmp = {
 869                 .block = block
 870         };
 871
 872         return RB_FIND(jbd_revoke, &info->revoke_root, &tmp);
 873 }
 874
 875 /**@brief  Replay a block in a transaction.
 876  * @param  jbd_fs jbd filesystem
 877  * @param  tag_info tag_info of the logged block.*/
 878 static void jbd_replay_block_tags(struct jbd_fs *jbd_fs,
 879                                   struct tag_info *tag_info,
 880                                   void *__arg)
 881 {
 882         int r;
 883         struct replay_arg *arg = __arg;
 884         struct recover_info *info = arg->info;
 885         uint32_t *this_block = arg->this_block;
 886         struct revoke_entry *revoke_entry;
 887         struct ext4_block journal_block, ext4_block;
 888         struct ext4_fs *fs = jbd_fs->inode_ref.fs;
 889
 890         (*this_block)++;
 891         wrap(&jbd_fs->sb, *this_block);
 892
 893         /* We replay this block only if the current transaction id
 894          * is equal or greater than that in revoke entry.*/
 895         revoke_entry = jbd_revoke_entry_lookup(info, tag_info->block);
 896         if (revoke_entry &&
 897             trans_id_diff(arg->this_trans_id, revoke_entry->trans_id) <= 0)
 898                 return;
 899
 900         ext4_dbg(DEBUG_JBD,
 901                  "Replaying block in block_tag: %" PRIu64 "\n",
 902                  tag_info->block);
 903
 904         r = jbd_block_get(jbd_fs, &journal_block, *this_block);
 905         if (r != EOK)
 906                 return;
 907
 908         /* We need special treatment for ext4 superblock. */
 909         if (tag_info->block) {
 910                 r = ext4_block_get_noread(fs->bdev, &ext4_block, tag_info->block);
 911                 if (r != EOK) {
 912                         jbd_block_set(jbd_fs, &journal_block);
 913                         return;
 914                 }
 915
 916                 memcpy(ext4_block.data,
 917                         journal_block.data,
 918                         jbd_get32(&jbd_fs->sb, blocksize));
 919
 920                 if (tag_info->is_escape)
 921                         ((struct jbd_bhdr *)ext4_block.data)->magic =
 922                                         to_be32(JBD_MAGIC_NUMBER);
 923
 924                 ext4_bcache_set_dirty(ext4_block.buf);
 925                 ext4_block_set(fs->bdev, &ext4_block);
 926         } else {
 927                 uint16_t mount_count, state;
 928                 mount_count = ext4_get16(&fs->sb, mount_count);
 929                 state = ext4_get16(&fs->sb, state);
 930
 931                 memcpy(&fs->sb,
 932                         journal_block.data + EXT4_SUPERBLOCK_OFFSET,
 933                         EXT4_SUPERBLOCK_SIZE);
 934
 935                 /* Mark system as mounted */
 936                 ext4_set16(&fs->sb, state, state);
 937                 r = ext4_sb_write(fs->bdev, &fs->sb);
 938                 if (r != EOK)
 939                         return;
 940
 941                 /*Update mount count*/
 942                 ext4_set16(&fs->sb, mount_count, mount_count);
 943         }
 944
 945         jbd_block_set(jbd_fs, &journal_block);
 946
 947         return;
 948 }
 949
 950 /**@brief  Add block address to revoke tree, along with
 951  *         its transaction id.
 952  * @param  info  journal replay info
 953  * @param  block  block address to be replayed.*/
 954 static void jbd_add_revoke_block_tags(struct recover_info *info,
 955                                       ext4_fsblk_t block)
 956 {
 957         struct revoke_entry *revoke_entry;
 958
 959         ext4_dbg(DEBUG_JBD, "Add block %" PRIu64 " to revoke tree\n", block);
 960         /* If the revoke entry with respect to the block address
 961          * exists already, update its transaction id.*/
 962         revoke_entry = jbd_revoke_entry_lookup(info, block);
 963         if (revoke_entry) {
 964                 revoke_entry->trans_id = info->this_trans_id;
 965                 return;
 966         }
 967
 968         revoke_entry = jbd_alloc_revoke_entry();
 969         ext4_assert(revoke_entry);
 970         revoke_entry->block = block;
 971         revoke_entry->trans_id = info->this_trans_id;
 972         RB_INSERT(jbd_revoke, &info->revoke_root, revoke_entry);
 973
 974         return;
 975 }
 976
 977 static void jbd_destroy_revoke_tree(struct recover_info *info)
 978 {
 979         while (!RB_EMPTY(&info->revoke_root)) {
 980                 struct revoke_entry *revoke_entry =
 981                         RB_MIN(jbd_revoke, &info->revoke_root);
 982                 ext4_assert(revoke_entry);
 983                 RB_REMOVE(jbd_revoke, &info->revoke_root, revoke_entry);
 984                 jbd_free_revoke_entry(revoke_entry);
 985         }
 986 }
 987
 988
 989 #define ACTION_SCAN 0
 990 #define ACTION_REVOKE 1
 991 #define ACTION_RECOVER 2
 992
 993 /**@brief  Add entries in a revoke block to revoke tree.
 994  * @param  jbd_fs jbd filesystem
 995  * @param  header revoke block header
 996  * @param  recover_info  journal replay info*/
 997 static void jbd_build_revoke_tree(struct jbd_fs *jbd_fs,
 998                                   struct jbd_bhdr *header,
 999                                   struct recover_info *info)
1000 {
1001         char *blocks_entry;
1002         struct jbd_revoke_header *revoke_hdr =
1003                 (struct jbd_revoke_header *)header;
1004         uint32_t i, nr_entries, record_len = 4;
1005
1006         /* If we are working on a 64bit jbd filesystem, */
1007         if (JBD_HAS_INCOMPAT_FEATURE(&jbd_fs->sb,
1008                                      JBD_FEATURE_INCOMPAT_64BIT))
1009                 record_len = 8;
1010
1011         nr_entries = (jbd_get32(revoke_hdr, count) -
1012                         sizeof(struct jbd_revoke_header)) /
1013                         record_len;
1014
1015         blocks_entry = (char *)(revoke_hdr + 1);
1016
1017         for (i = 0;i < nr_entries;i++) {
1018                 if (record_len == 8) {
1019                         uint64_t *blocks =
1020                                 (uint64_t *)blocks_entry;
1021                         jbd_add_revoke_block_tags(info, to_be64(*blocks));
1022                 } else {
1023                         uint32_t *blocks =
1024                                 (uint32_t *)blocks_entry;
1025                         jbd_add_revoke_block_tags(info, to_be32(*blocks));
1026                 }
1027                 blocks_entry += record_len;
1028         }
1029 }
1030
1031 static void jbd_debug_descriptor_block(struct jbd_fs *jbd_fs,
1032                                        struct jbd_bhdr *header,
1033                                        uint32_t *iblock)
1034 {
1035         jbd_iterate_block_table(jbd_fs,
1036                                 header + 1,
1037                                 jbd_get32(&jbd_fs->sb, blocksize) -
1038                                         sizeof(struct jbd_bhdr),
1039                                 jbd_display_block_tags,
1040                                 iblock);
1041 }
1042
1043 static void jbd_replay_descriptor_block(struct jbd_fs *jbd_fs,
1044                                         struct jbd_bhdr *header,
1045                                         struct replay_arg *arg)
1046 {
1047         jbd_iterate_block_table(jbd_fs,
1048                                 header + 1,
1049                                 jbd_get32(&jbd_fs->sb, blocksize) -
1050                                         sizeof(struct jbd_bhdr),
1051                                 jbd_replay_block_tags,
1052                                 arg);
1053 }
1054
1055 /**@brief  The core routine of journal replay.
1056  * @param  jbd_fs jbd filesystem
1057  * @param  recover_info  journal replay info
1058  * @param  action action needed to be taken
1059  * @return standard error code*/
1060 static int jbd_iterate_log(struct jbd_fs *jbd_fs,
1061                            struct recover_info *info,
1062                            int action)
1063 {
1064         int r = EOK;
1065         bool log_end = false;
1066         struct jbd_sb *sb = &jbd_fs->sb;
1067         uint32_t start_trans_id, this_trans_id;
1068         uint32_t start_block, this_block;
1069
1070         /* We start iterating valid blocks in the whole journal.*/
1071         start_trans_id = this_trans_id = jbd_get32(sb, sequence);
1072         start_block = this_block = jbd_get32(sb, start);
1073         if (action == ACTION_SCAN)
1074                 info->trans_cnt = 0;
1075         else if (!info->trans_cnt)
1076                 log_end = true;
1077
1078         ext4_dbg(DEBUG_JBD, "Start of journal at trans id: %" PRIu32 "\n",
1079                             start_trans_id);
1080
1081         while (!log_end) {
1082                 struct ext4_block block;
1083                 struct jbd_bhdr *header;
1084                 /* If we are not scanning for the last
1085                  * valid transaction in the journal,
1086                  * we will stop when we reach the end of
1087                  * the journal.*/
1088                 if (action != ACTION_SCAN)
1089                         if (trans_id_diff(this_trans_id, info->last_trans_id) > 0) {
1090                                 log_end = true;
1091                                 continue;
1092                         }
1093
1094                 r = jbd_block_get(jbd_fs, &block, this_block);
1095                 if (r != EOK)
1096                         break;
1097
1098                 header = (struct jbd_bhdr *)block.data;
1099                 /* This block does not have a valid magic number,
1100                  * so we have reached the end of the journal.*/
1101                 if (jbd_get32(header, magic) != JBD_MAGIC_NUMBER) {
1102                         jbd_block_set(jbd_fs, &block);
1103                         log_end = true;
1104                         continue;
1105                 }
1106
1107                 /* If the transaction id we found is not expected,
1108                  * we may have reached the end of the journal.
1109                  *
1110                  * If we are not scanning the journal, something
1111                  * bad might have taken place. :-( */
1112                 if (jbd_get32(header, sequence) != this_trans_id) {
1113                         if (action != ACTION_SCAN)
1114                                 r = EIO;
1115
1116                         jbd_block_set(jbd_fs, &block);
1117                         log_end = true;
1118                         continue;
1119                 }
1120
1121                 switch (jbd_get32(header, blocktype)) {
1122                 case JBD_DESCRIPTOR_BLOCK:
1123                         if (!jbd_verify_meta_csum(jbd_fs, header)) {
1124                                 ext4_dbg(DEBUG_JBD,
1125                                         DBG_WARN "Descriptor block checksum failed."
1126                                                 "Journal block: %" PRIu32"\n",
1127                                                 this_block);
1128                                 log_end = true;
1129                                 break;
1130                         }
1131                         ext4_dbg(DEBUG_JBD, "Descriptor block: %" PRIu32", "
1132                                             "trans_id: %" PRIu32"\n",
1133                                             this_block, this_trans_id);
1134                         if (action == ACTION_RECOVER) {
1135                                 struct replay_arg replay_arg;
1136                                 replay_arg.info = info;
1137                                 replay_arg.this_block = &this_block;
1138                                 replay_arg.this_trans_id = this_trans_id;
1139
1140                                 jbd_replay_descriptor_block(jbd_fs,
1141                                                 header, &replay_arg);
1142                         } else
1143                                 jbd_debug_descriptor_block(jbd_fs,
1144                                                 header, &this_block);
1145
1146                         break;
1147                 case JBD_COMMIT_BLOCK:
1148                         if (!jbd_verify_commit_csum(jbd_fs,
1149                                         (struct jbd_commit_header *)header)) {
1150                                 ext4_dbg(DEBUG_JBD,
1151                                         DBG_WARN "Commit block checksum failed."
1152                                                 "Journal block: %" PRIu32"\n",
1153                                                 this_block);
1154                                 log_end = true;
1155                                 break;
1156                         }
1157                         ext4_dbg(DEBUG_JBD, "Commit block: %" PRIu32", "
1158                                             "trans_id: %" PRIu32"\n",
1159                                             this_block, this_trans_id);
1160                         /*
1161                          * This is the end of a transaction,
1162                          * we may now proceed to the next transaction.
1163                          */
1164                         this_trans_id++;
1165                         if (action == ACTION_SCAN)
1166                                 info->trans_cnt++;
1167                         break;
1168                 case JBD_REVOKE_BLOCK:
1169                         if (!jbd_verify_meta_csum(jbd_fs, header)) {
1170                                 ext4_dbg(DEBUG_JBD,
1171                                         DBG_WARN "Revoke block checksum failed."
1172                                                 "Journal block: %" PRIu32"\n",
1173                                                 this_block);
1174                                 log_end = true;
1175                                 break;
1176                         }
1177                         ext4_dbg(DEBUG_JBD, "Revoke block: %" PRIu32", "
1178                                             "trans_id: %" PRIu32"\n",
1179                                             this_block, this_trans_id);
1180                         if (action == ACTION_REVOKE) {
1181                                 info->this_trans_id = this_trans_id;
1182                                 jbd_build_revoke_tree(jbd_fs,
1183                                                 header, info);
1184                         }
1185                         break;
1186                 default:
1187                         log_end = true;
1188                         break;
1189                 }
1190                 jbd_block_set(jbd_fs, &block);
1191                 this_block++;
1192                 wrap(sb, this_block);
1193                 if (this_block == start_block)
1194                         log_end = true;
1195
1196         }
1197         ext4_dbg(DEBUG_JBD, "End of journal.\n");
1198         if (r == EOK && action == ACTION_SCAN) {
1199                 /* We have finished scanning the journal. */
1200                 info->start_trans_id = start_trans_id;
1201                 if (trans_id_diff(this_trans_id, start_trans_id) > 0)
1202                         info->last_trans_id = this_trans_id - 1;
1203                 else
1204                         info->last_trans_id = this_trans_id;
1205         }
1206
1207         return r;
1208 }
1209
1210 /**@brief  Replay journal.
1211  * @param  jbd_fs jbd filesystem
1212  * @return standard error code*/
1213 int jbd_recover(struct jbd_fs *jbd_fs)
1214 {
1215         int r;
1216         struct recover_info info;
1217         struct jbd_sb *sb = &jbd_fs->sb;
1218         if (!sb->start)
1219                 return EOK;
1220
1221         RB_INIT(&info.revoke_root);
1222
1223         r = jbd_iterate_log(jbd_fs, &info, ACTION_SCAN);
1224         if (r != EOK)
1225                 return r;
1226
1227         r = jbd_iterate_log(jbd_fs, &info, ACTION_REVOKE);
1228         if (r != EOK)
1229                 return r;
1230
1231         r = jbd_iterate_log(jbd_fs, &info, ACTION_RECOVER);
1232         if (r == EOK) {
1233                 /* If we successfully replay the journal,
1234                  * clear EXT4_FINCOM_RECOVER flag on the
1235                  * ext4 superblock, and set the start of
1236                  * journal to 0.*/
1237                 uint32_t features_incompatible =
1238                         ext4_get32(&jbd_fs->inode_ref.fs->sb,
1239                                    features_incompatible);
1240                 jbd_set32(&jbd_fs->sb, start, 0);
1241                 jbd_set32(&jbd_fs->sb, sequence, info.last_trans_id);
1242                 features_incompatible &= ~EXT4_FINCOM_RECOVER;
1243                 ext4_set32(&jbd_fs->inode_ref.fs->sb,
1244                            features_incompatible,
1245                            features_incompatible);
1246                 jbd_fs->dirty = true;
1247                 r = ext4_sb_write(jbd_fs->bdev,
1248                                   &jbd_fs->inode_ref.fs->sb);
1249         }
1250         jbd_destroy_revoke_tree(&info);
1251         return r;
1252 }
1253
1254 static void jbd_journal_write_sb(struct jbd_journal *journal)
1255 {
1256         struct jbd_fs *jbd_fs = journal->jbd_fs;
1257         jbd_set32(&jbd_fs->sb, start, journal->start);
1258         jbd_set32(&jbd_fs->sb, sequence, journal->trans_id);
1259         jbd_fs->dirty = true;
1260 }
1261
1262 /**@brief  Start accessing the journal.
1263  * @param  jbd_fs jbd filesystem
1264  * @param  journal current journal session
1265  * @return standard error code*/
1266 int jbd_journal_start(struct jbd_fs *jbd_fs,
1267                       struct jbd_journal *journal)
1268 {
1269         int r;
1270         uint32_t features_incompatible =
1271                         ext4_get32(&jbd_fs->inode_ref.fs->sb,
1272                                    features_incompatible);
1273         features_incompatible |= EXT4_FINCOM_RECOVER;
1274         ext4_set32(&jbd_fs->inode_ref.fs->sb,
1275                         features_incompatible,
1276                         features_incompatible);
1277         r = ext4_sb_write(jbd_fs->bdev,
1278                         &jbd_fs->inode_ref.fs->sb);
1279         if (r != EOK)
1280                 return r;
1281
1282         journal->first = jbd_get32(&jbd_fs->sb, first);
1283         journal->start = journal->first;
1284         journal->last = journal->first;
1285         /*
1286          * To invalidate any stale records we need to start from
1287          * the checkpoint transaction ID of the previous journalling session
1288          * plus 1.
1289          */
1290         journal->trans_id = jbd_get32(&jbd_fs->sb, sequence) + 1;
1291         journal->alloc_trans_id = journal->trans_id;
1292
1293         journal->block_size = jbd_get32(&jbd_fs->sb, blocksize);
1294
1295         TAILQ_INIT(&journal->cp_queue);
1296         RB_INIT(&journal->block_rec_root);
1297         journal->jbd_fs = jbd_fs;
1298         jbd_journal_write_sb(journal);
1299         r = jbd_write_sb(jbd_fs);
1300         if (r != EOK)
1301                 return r;
1302
1303         jbd_fs->bdev->journal = journal;
1304         return EOK;
1305 }
1306
1307 static void jbd_trans_end_write(struct ext4_bcache *bc __unused,
1308                           struct ext4_buf *buf __unused,
1309                           int res,
1310                           void *arg);
1311
1312 /*
1313  * This routine is only suitable to committed transactions. */
1314 static void jbd_journal_flush_trans(struct jbd_trans *trans)
1315 {
1316         struct jbd_buf *jbd_buf, *tmp;
1317         struct jbd_journal *journal = trans->journal;
1318         struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
1319         void *tmp_data = ext4_malloc(journal->block_size);
1320         ext4_assert(tmp_data);
1321
1322         TAILQ_FOREACH_SAFE(jbd_buf, &trans->buf_queue, buf_node,
1323                         tmp) {
1324                 struct ext4_buf *buf;
1325                 struct ext4_block block;
1326                 /* The buffer is not yet flushed. */
1327                 buf = ext4_bcache_find_get(fs->bdev->bc, &block,
1328                                            jbd_buf->block_rec->lba);
1329                 if (!(buf && ext4_bcache_test_flag(buf, BC_UPTODATE) &&
1330                       jbd_buf->block_rec->trans == trans)) {
1331                         int r;
1332                         struct ext4_block jbd_block = EXT4_BLOCK_ZERO();
1333                         ext4_assert(jbd_block_get(journal->jbd_fs,
1334                                                 &jbd_block,
1335                                                 jbd_buf->jbd_lba) == EOK);
1336                         memcpy(tmp_data, jbd_block.data,
1337                                         journal->block_size);
1338                         ext4_block_set(fs->bdev, &jbd_block);
1339                         r = ext4_blocks_set_direct(fs->bdev, tmp_data,
1340                                         jbd_buf->block_rec->lba, 1);
1341                         jbd_trans_end_write(fs->bdev->bc, buf, r, jbd_buf);
1342                 } else
1343                         ext4_block_flush_buf(fs->bdev, buf);
1344
1345                 if (buf)
1346                         ext4_block_set(fs->bdev, &block);
1347         }
1348
1349         ext4_free(tmp_data);
1350 }
1351
1352 static void
1353 jbd_journal_skip_pure_revoke(struct jbd_journal *journal,
1354                              struct jbd_trans *trans)
1355 {
1356         journal->start = trans->start_iblock +
1357                 trans->alloc_blocks;
1358         wrap(&journal->jbd_fs->sb, journal->start);
1359         journal->trans_id = trans->trans_id + 1;
1360         jbd_journal_free_trans(journal,
1361                         trans, false);
1362         jbd_journal_write_sb(journal);
1363 }
1364
1365 void
1366 jbd_journal_purge_cp_trans(struct jbd_journal *journal,
1367                            bool flush,
1368                            bool once)
1369 {
1370         struct jbd_trans *trans;
1371         while ((trans = TAILQ_FIRST(&journal->cp_queue))) {
1372                 if (!trans->data_cnt) {
1373                         TAILQ_REMOVE(&journal->cp_queue,
1374                                         trans,
1375                                         trans_node);
1376                         jbd_journal_skip_pure_revoke(journal, trans);
1377                 } else {
1378                         if (trans->data_cnt ==
1379                                         trans->written_cnt) {
1380                                 journal->start =
1381                                         trans->start_iblock +
1382                                         trans->alloc_blocks;
1383                                 wrap(&journal->jbd_fs->sb,
1384                                                 journal->start);
1385                                 journal->trans_id =
1386                                         trans->trans_id + 1;
1387                                 TAILQ_REMOVE(&journal->cp_queue,
1388                                                 trans,
1389                                                 trans_node);
1390                                 jbd_journal_free_trans(journal,
1391                                                 trans,
1392                                                 false);
1393                                 jbd_journal_write_sb(journal);
1394                         } else if (!flush) {
1395                                 journal->start =
1396                                         trans->start_iblock;
1397                                 wrap(&journal->jbd_fs->sb,
1398                                                 journal->start);
1399                                 journal->trans_id =
1400                                         trans->trans_id;
1401                                 jbd_journal_write_sb(journal);
1402                                 break;
1403                         } else
1404                                 jbd_journal_flush_trans(trans);
1405                 }
1406                 if (once)
1407                         break;
1408         }
1409 }
1410
1411 /**@brief  Stop accessing the journal.
1412  * @param  journal current journal session
1413  * @return standard error code*/
1414 int jbd_journal_stop(struct jbd_journal *journal)
1415 {
1416         int r;
1417         struct jbd_fs *jbd_fs = journal->jbd_fs;
1418         uint32_t features_incompatible;
1419
1420         /* Make sure that journalled content have reached
1421          * the disk.*/
1422         jbd_journal_purge_cp_trans(journal, true, false);
1423
1424         /* There should be no block record in this journal
1425          * session. */
1426         if (!RB_EMPTY(&journal->block_rec_root))
1427                 ext4_dbg(DEBUG_JBD,
1428                          DBG_WARN "There are still block records "
1429                                   "in this journal session!\n");
1430
1431         features_incompatible =
1432                 ext4_get32(&jbd_fs->inode_ref.fs->sb,
1433                            features_incompatible);
1434         features_incompatible &= ~EXT4_FINCOM_RECOVER;
1435         ext4_set32(&jbd_fs->inode_ref.fs->sb,
1436                         features_incompatible,
1437                         features_incompatible);
1438         r = ext4_sb_write(jbd_fs->bdev,
1439                         &jbd_fs->inode_ref.fs->sb);
1440         if (r != EOK)
1441                 return r;
1442
1443         journal->start = 0;
1444         journal->trans_id = 0;
1445         jbd_journal_write_sb(journal);
1446         return jbd_write_sb(journal->jbd_fs);
1447 }
1448
1449 /**@brief  Allocate a block in the journal.
1450  * @param  journal current journal session
1451  * @param  trans transaction
1452  * @return allocated block address*/
1453 static uint32_t jbd_journal_alloc_block(struct jbd_journal *journal,
1454                                         struct jbd_trans *trans)
1455 {
1456         uint32_t start_block;
1457
1458         start_block = journal->last++;
1459         trans->alloc_blocks++;
1460         wrap(&journal->jbd_fs->sb, journal->last);
1461
1462         /* If there is no space left, flush all journalled
1463          * blocks to disk first.*/
1464         if (journal->last == journal->start)
1465                 jbd_journal_purge_cp_trans(journal, true, false);
1466
1467         return start_block;
1468 }
1469
1470 static struct jbd_block_rec *
1471 jbd_trans_block_rec_lookup(struct jbd_journal *journal,
1472                            ext4_fsblk_t lba)
1473 {
1474         struct jbd_block_rec tmp = {
1475                 .lba = lba
1476         };
1477
1478         return RB_FIND(jbd_block,
1479                        &journal->block_rec_root,
1480                        &tmp);
1481 }
1482
1483 static void
1484 jbd_trans_change_ownership(struct jbd_block_rec *block_rec,
1485                            struct jbd_trans *new_trans)
1486 {
1487         LIST_REMOVE(block_rec, tbrec_node);
1488         if (new_trans) {
1489                 /* Now this block record belongs to this transaction. */
1490                 LIST_INSERT_HEAD(&new_trans->tbrec_list, block_rec, tbrec_node);
1491         }
1492         block_rec->trans = new_trans;
1493 }
1494
1495 static inline struct jbd_block_rec *
1496 jbd_trans_insert_block_rec(struct jbd_trans *trans,
1497                            ext4_fsblk_t lba)
1498 {
1499         struct jbd_block_rec *block_rec;
1500         block_rec = jbd_trans_block_rec_lookup(trans->journal, lba);
1501         if (block_rec) {
1502                 jbd_trans_change_ownership(block_rec, trans);
1503                 return block_rec;
1504         }
1505         block_rec = ext4_calloc(1, sizeof(struct jbd_block_rec));
1506         if (!block_rec)
1507                 return NULL;
1508
1509         block_rec->lba = lba;
1510         block_rec->trans = trans;
1511         TAILQ_INIT(&block_rec->dirty_buf_queue);
1512         LIST_INSERT_HEAD(&trans->tbrec_list, block_rec, tbrec_node);
1513         RB_INSERT(jbd_block, &trans->journal->block_rec_root, block_rec);
1514         return block_rec;
1515 }
1516
1517 /*
1518  * This routine will do the dirty works.
1519  */
1520 static void
1521 jbd_trans_finish_callback(struct jbd_journal *journal,
1522                           const struct jbd_trans *trans,
1523                           struct jbd_block_rec *block_rec,
1524                           bool abort,
1525                           bool revoke)
1526 {
1527         struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
1528         if (block_rec->trans != trans)
1529                 return;
1530
1531         if (!abort) {
1532                 struct jbd_buf *jbd_buf, *tmp;
1533                 TAILQ_FOREACH_SAFE(jbd_buf,
1534                                 &block_rec->dirty_buf_queue,
1535                                 dirty_buf_node,
1536                                 tmp) {
1537                         jbd_trans_end_write(fs->bdev->bc,
1538                                         NULL,
1539                                         EOK,
1540                                         jbd_buf);
1541                 }
1542         } else {
1543                 /*
1544                  * We have to roll back data if the block is going to be
1545                  * aborted.
1546                  */
1547                 struct jbd_buf *jbd_buf;
1548                 struct ext4_block jbd_block = EXT4_BLOCK_ZERO(),
1549                                   block = EXT4_BLOCK_ZERO();
1550                 jbd_buf = TAILQ_LAST(&block_rec->dirty_buf_queue,
1551                                 jbd_buf_dirty);
1552                 if (jbd_buf) {
1553                         if (!revoke) {
1554                                 ext4_assert(ext4_block_get_noread(fs->bdev,
1555                                                         &block,
1556                                                         block_rec->lba) == EOK);
1557                                 ext4_assert(jbd_block_get(journal->jbd_fs,
1558                                                         &jbd_block,
1559                                                         jbd_buf->jbd_lba) == EOK);
1560                                 memcpy(block.data, jbd_block.data,
1561                                                 journal->block_size);
1562
1563                                 jbd_trans_change_ownership(block_rec,
1564                                                 jbd_buf->trans);
1565
1566                                 block.buf->end_write = jbd_trans_end_write;
1567                                 block.buf->end_write_arg = jbd_buf;
1568
1569                                 ext4_bcache_set_flag(jbd_block.buf, BC_TMP);
1570                                 ext4_bcache_set_dirty(block.buf);
1571
1572                                 ext4_block_set(fs->bdev, &jbd_block);
1573                                 ext4_block_set(fs->bdev, &block);
1574                                 return;
1575                         } else {
1576                                 /* The revoked buffer is yet written. */
1577                                 jbd_trans_change_ownership(block_rec,
1578                                                 jbd_buf->trans);
1579                         }
1580                 }
1581         }
1582 }
1583
1584 static inline void
1585 jbd_trans_remove_block_rec(struct jbd_journal *journal,
1586                            struct jbd_block_rec *block_rec,
1587                            struct jbd_trans *trans)
1588 {
1589         /* If this block record doesn't belong to this transaction,
1590          * give up.*/
1591         if (block_rec->trans == trans) {
1592                 LIST_REMOVE(block_rec, tbrec_node);
1593                 RB_REMOVE(jbd_block,
1594                                 &journal->block_rec_root,
1595                                 block_rec);
1596                 ext4_free(block_rec);
1597         }
1598 }
1599
1600 /**@brief  Add block to a transaction and mark it dirty.
1601  * @param  trans transaction
1602  * @param  block block descriptor
1603  * @return standard error code*/
1604 int jbd_trans_set_block_dirty(struct jbd_trans *trans,
1605                               struct ext4_block *block)
1606 {
1607         struct jbd_buf *jbd_buf;
1608         struct jbd_revoke_rec *rec, tmp_rec = {
1609                 .lba = block->lb_id
1610         };
1611         struct jbd_block_rec *block_rec;
1612
1613         if (block->buf->end_write == jbd_trans_end_write) {
1614                 jbd_buf = block->buf->end_write_arg;
1615                 if (jbd_buf && jbd_buf->trans == trans)
1616                         return EOK;
1617         }
1618         jbd_buf = ext4_calloc(1, sizeof(struct jbd_buf));
1619         if (!jbd_buf)
1620                 return ENOMEM;
1621
1622         if ((block_rec = jbd_trans_insert_block_rec(trans,
1623                                         block->lb_id)) == NULL) {
1624                 ext4_free(jbd_buf);
1625                 return ENOMEM;
1626         }
1627
1628         TAILQ_INSERT_TAIL(&block_rec->dirty_buf_queue,
1629                         jbd_buf,
1630                         dirty_buf_node);
1631
1632         jbd_buf->block_rec = block_rec;
1633         jbd_buf->trans = trans;
1634         jbd_buf->block = *block;
1635         ext4_bcache_inc_ref(block->buf);
1636
1637         /* If the content reach the disk, notify us
1638          * so that we may do a checkpoint. */
1639         block->buf->end_write = jbd_trans_end_write;
1640         block->buf->end_write_arg = jbd_buf;
1641
1642         trans->data_cnt++;
1643         TAILQ_INSERT_HEAD(&trans->buf_queue, jbd_buf, buf_node);
1644
1645         ext4_bcache_set_dirty(block->buf);
1646         rec = RB_FIND(jbd_revoke_tree,
1647                         &trans->revoke_root,
1648                         &tmp_rec);
1649         if (rec) {
1650                 RB_REMOVE(jbd_revoke_tree, &trans->revoke_root,
1651                           rec);
1652                 ext4_free(rec);
1653         }
1654
1655         return EOK;
1656 }
1657
1658 /**@brief  Add block to be revoked to a transaction
1659  * @param  trans transaction
1660  * @param  lba logical block address
1661  * @return standard error code*/
1662 int jbd_trans_revoke_block(struct jbd_trans *trans,
1663                            ext4_fsblk_t lba)
1664 {
1665         struct jbd_revoke_rec tmp_rec = {
1666                 .lba = lba
1667         }, *rec;
1668         rec = RB_FIND(jbd_revoke_tree,
1669                       &trans->revoke_root,
1670                       &tmp_rec);
1671         if (rec)
1672                 return EOK;
1673
1674         rec = ext4_calloc(1, sizeof(struct jbd_revoke_rec));
1675         if (!rec)
1676                 return ENOMEM;
1677
1678         rec->lba = lba;
1679         RB_INSERT(jbd_revoke_tree, &trans->revoke_root, rec);
1680         return EOK;
1681 }
1682
1683 /**@brief  Try to add block to be revoked to a transaction.
1684  *         If @lba still remains in an transaction on checkpoint
1685  *         queue, add @lba as a revoked block to the transaction.
1686  * @param  trans transaction
1687  * @param  lba logical block address
1688  * @return standard error code*/
1689 int jbd_trans_try_revoke_block(struct jbd_trans *trans,
1690                                ext4_fsblk_t lba)
1691 {
1692         struct jbd_journal *journal = trans->journal;
1693         struct jbd_block_rec *block_rec =
1694                 jbd_trans_block_rec_lookup(journal, lba);
1695
1696         if (block_rec) {
1697                 if (block_rec->trans == trans) {
1698                         struct jbd_buf *jbd_buf =
1699                                 TAILQ_LAST(&block_rec->dirty_buf_queue,
1700                                         jbd_buf_dirty);
1701                         /* If there are still unwritten buffers. */
1702                         if (TAILQ_FIRST(&block_rec->dirty_buf_queue) !=
1703                             jbd_buf)
1704                                 jbd_trans_revoke_block(trans, lba);
1705
1706                 } else
1707                         jbd_trans_revoke_block(trans, lba);
1708         }
1709
1710         return EOK;
1711 }
1712
1713 /**@brief  Free a transaction
1714  * @param  journal current journal session
1715  * @param  trans transaction
1716  * @param  abort discard all the modifications on the block?
1717  * @return standard error code*/
1718 void jbd_journal_free_trans(struct jbd_journal *journal,
1719                             struct jbd_trans *trans,
1720                             bool abort)
1721 {
1722         struct jbd_buf *jbd_buf, *tmp;
1723         struct jbd_revoke_rec *rec, *tmp2;
1724         struct jbd_block_rec *block_rec, *tmp3;
1725         struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
1726         TAILQ_FOREACH_SAFE(jbd_buf, &trans->buf_queue, buf_node,
1727                           tmp) {
1728                 block_rec = jbd_buf->block_rec;
1729                 if (abort) {
1730                         jbd_buf->block.buf->end_write = NULL;
1731                         jbd_buf->block.buf->end_write_arg = NULL;
1732                         ext4_bcache_clear_dirty(jbd_buf->block.buf);
1733                         ext4_block_set(fs->bdev, &jbd_buf->block);
1734                 }
1735
1736                 TAILQ_REMOVE(&jbd_buf->block_rec->dirty_buf_queue,
1737                         jbd_buf,
1738                         dirty_buf_node);
1739                 jbd_trans_finish_callback(journal,
1740                                 trans,
1741                                 block_rec,
1742                                 abort,
1743                                 false);
1744                 TAILQ_REMOVE(&trans->buf_queue, jbd_buf, buf_node);
1745                 ext4_free(jbd_buf);
1746         }
1747         RB_FOREACH_SAFE(rec, jbd_revoke_tree, &trans->revoke_root,
1748                           tmp2) {
1749                 RB_REMOVE(jbd_revoke_tree, &trans->revoke_root, rec);
1750                 ext4_free(rec);
1751         }
1752         LIST_FOREACH_SAFE(block_rec, &trans->tbrec_list, tbrec_node,
1753                           tmp3) {
1754                 jbd_trans_remove_block_rec(journal, block_rec, trans);
1755         }
1756
1757         ext4_free(trans);
1758 }
1759
1760 /**@brief  Write commit block for a transaction
1761  * @param  trans transaction
1762  * @return standard error code*/
1763 static int jbd_trans_write_commit_block(struct jbd_trans *trans)
1764 {
1765         int rc;
1766         struct ext4_block block;
1767         struct jbd_commit_header *header;
1768         uint32_t commit_iblock;
1769         struct jbd_journal *journal = trans->journal;
1770
1771         commit_iblock = jbd_journal_alloc_block(journal, trans);
1772
1773         rc = jbd_block_get_noread(journal->jbd_fs, &block, commit_iblock);
1774         if (rc != EOK)
1775                 return rc;
1776
1777         header = (struct jbd_commit_header *)block.data;
1778         jbd_set32(&header->header, magic, JBD_MAGIC_NUMBER);
1779         jbd_set32(&header->header, blocktype, JBD_COMMIT_BLOCK);
1780         jbd_set32(&header->header, sequence, trans->trans_id);
1781
1782         if (JBD_HAS_INCOMPAT_FEATURE(&journal->jbd_fs->sb,
1783                                 JBD_FEATURE_COMPAT_CHECKSUM)) {
1784                 jbd_set32(header, chksum_type, JBD_CRC32_CHKSUM);
1785                 jbd_set32(header, chksum_size, JBD_CRC32_CHKSUM_SIZE);
1786                 jbd_set32(header, chksum[0], trans->data_csum);
1787         }
1788         jbd_commit_csum_set(journal->jbd_fs, header);
1789         ext4_bcache_set_dirty(block.buf);
1790         ext4_bcache_set_flag(block.buf, BC_TMP);
1791         rc = jbd_block_set(journal->jbd_fs, &block);
1792         return rc;
1793 }
1794
1795 /**@brief  Write descriptor block for a transaction
1796  * @param  journal current journal session
1797  * @param  trans transaction
1798  * @return standard error code*/
1799 static int jbd_journal_prepare(struct jbd_journal *journal,
1800                                struct jbd_trans *trans)
1801 {
1802         int rc = EOK, i = 0;
1803         struct ext4_block desc_block = EXT4_BLOCK_ZERO(),
1804                           data_block = EXT4_BLOCK_ZERO();
1805         int32_t tag_tbl_size = 0;
1806         uint32_t desc_iblock = 0;
1807         uint32_t data_iblock = 0;
1808         char *tag_start = NULL, *tag_ptr = NULL;
1809         struct jbd_buf *jbd_buf, *tmp;
1810         struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
1811         uint32_t checksum = EXT4_CRC32_INIT;
1812         struct jbd_bhdr *bhdr = NULL;
1813         void *data;
1814
1815         /* Try to remove any non-dirty buffers from the tail of
1816          * buf_queue. */
1817         TAILQ_FOREACH_REVERSE_SAFE(jbd_buf, &trans->buf_queue,
1818                         jbd_trans_buf, buf_node, tmp) {
1819                 struct jbd_revoke_rec tmp_rec = {
1820                         .lba = jbd_buf->block_rec->lba
1821                 };
1822                 /* We stop the iteration when we find a dirty buffer. */
1823                 if (ext4_bcache_test_flag(jbd_buf->block.buf,
1824                                         BC_DIRTY))
1825                         break;
1826
1827                 TAILQ_REMOVE(&jbd_buf->block_rec->dirty_buf_queue,
1828                         jbd_buf,
1829                         dirty_buf_node);
1830
1831                 jbd_buf->block.buf->end_write = NULL;
1832                 jbd_buf->block.buf->end_write_arg = NULL;
1833                 jbd_trans_finish_callback(journal,
1834                                 trans,
1835                                 jbd_buf->block_rec,
1836                                 true,
1837                                 RB_FIND(jbd_revoke_tree,
1838                                         &trans->revoke_root,
1839                                         &tmp_rec));
1840                 jbd_trans_remove_block_rec(journal,
1841                                         jbd_buf->block_rec, trans);
1842                 trans->data_cnt--;
1843
1844                 ext4_block_set(fs->bdev, &jbd_buf->block);
1845                 TAILQ_REMOVE(&trans->buf_queue, jbd_buf, buf_node);
1846                 ext4_free(jbd_buf);
1847         }
1848
1849         TAILQ_FOREACH_SAFE(jbd_buf, &trans->buf_queue, buf_node, tmp) {
1850                 struct tag_info tag_info;
1851                 bool uuid_exist = false;
1852                 bool is_escape = false;
1853                 struct jbd_revoke_rec tmp_rec = {
1854                         .lba = jbd_buf->block_rec->lba
1855                 };
1856                 if (!ext4_bcache_test_flag(jbd_buf->block.buf,
1857                                            BC_DIRTY)) {
1858                         TAILQ_REMOVE(&jbd_buf->block_rec->dirty_buf_queue,
1859                                         jbd_buf,
1860                                         dirty_buf_node);
1861
1862                         jbd_buf->block.buf->end_write = NULL;
1863                         jbd_buf->block.buf->end_write_arg = NULL;
1864
1865                         /* The buffer has not been modified, just release
1866                          * that jbd_buf. */
1867                         jbd_trans_finish_callback(journal,
1868                                         trans,
1869                                         jbd_buf->block_rec,
1870                                         true,
1871                                         RB_FIND(jbd_revoke_tree,
1872                                                 &trans->revoke_root,
1873                                                 &tmp_rec));
1874                         jbd_trans_remove_block_rec(journal,
1875                                         jbd_buf->block_rec, trans);
1876                         trans->data_cnt--;
1877
1878                         ext4_block_set(fs->bdev, &jbd_buf->block);
1879                         TAILQ_REMOVE(&trans->buf_queue, jbd_buf, buf_node);
1880                         ext4_free(jbd_buf);
1881                         continue;
1882                 }
1883                 checksum = jbd_block_csum(journal->jbd_fs,
1884                                           jbd_buf->block.data,
1885                                           checksum,
1886                                           trans->trans_id);
1887                 if (((struct jbd_bhdr *)jbd_buf->block.data)->magic ==
1888                                 to_be32(JBD_MAGIC_NUMBER))
1889                         is_escape = true;
1890
1891 again:
1892                 if (!desc_iblock) {
1893                         desc_iblock = jbd_journal_alloc_block(journal, trans);
1894                         rc = jbd_block_get_noread(journal->jbd_fs, &desc_block, desc_iblock);
1895                         if (rc != EOK)
1896                                 break;
1897
1898                         bhdr = (struct jbd_bhdr *)desc_block.data;
1899                         jbd_set32(bhdr, magic, JBD_MAGIC_NUMBER);
1900                         jbd_set32(bhdr, blocktype, JBD_DESCRIPTOR_BLOCK);
1901                         jbd_set32(bhdr, sequence, trans->trans_id);
1902
1903                         tag_start = (char *)(bhdr + 1);
1904                         tag_ptr = tag_start;
1905                         uuid_exist = true;
1906                         tag_tbl_size = journal->block_size -
1907                                 sizeof(struct jbd_bhdr);
1908
1909                         if (jbd_has_csum(&journal->jbd_fs->sb))
1910                                 tag_tbl_size -= sizeof(struct jbd_block_tail);
1911
1912                         if (!trans->start_iblock)
1913                                 trans->start_iblock = desc_iblock;
1914
1915                         ext4_bcache_set_dirty(desc_block.buf);
1916                         ext4_bcache_set_flag(desc_block.buf, BC_TMP);
1917                 }
1918                 tag_info.block = jbd_buf->block.lb_id;
1919                 tag_info.uuid_exist = uuid_exist;
1920                 tag_info.is_escape = is_escape;
1921                 if (i == trans->data_cnt - 1)
1922                         tag_info.last_tag = true;
1923                 else
1924                         tag_info.last_tag = false;
1925
1926                 tag_info.checksum = checksum;
1927
1928                 if (uuid_exist)
1929                         memcpy(tag_info.uuid, journal->jbd_fs->sb.uuid,
1930                                         UUID_SIZE);
1931
1932                 rc = jbd_write_block_tag(journal->jbd_fs,
1933                                 tag_ptr,
1934                                 tag_tbl_size,
1935                                 &tag_info);
1936                 if (rc != EOK) {
1937                         jbd_meta_csum_set(journal->jbd_fs, bhdr);
1938                         desc_iblock = 0;
1939                         rc = jbd_block_set(journal->jbd_fs, &desc_block);
1940                         if (rc != EOK)
1941                                 break;
1942
1943                         goto again;
1944                 }
1945
1946                 data_iblock = jbd_journal_alloc_block(journal, trans);
1947                 rc = jbd_block_get_noread(journal->jbd_fs, &data_block, data_iblock);
1948                 if (rc != EOK) {
1949                         desc_iblock = 0;
1950                         ext4_bcache_clear_dirty(desc_block.buf);
1951                         jbd_block_set(journal->jbd_fs, &desc_block);
1952                         break;
1953                 }
1954
1955                 data = data_block.data;
1956                 memcpy(data, jbd_buf->block.data,
1957                         journal->block_size);
1958                 if (is_escape)
1959                         ((struct jbd_bhdr *)data)->magic = 0;
1960
1961                 ext4_bcache_set_dirty(data_block.buf);
1962                 ext4_bcache_set_flag(data_block.buf, BC_TMP);
1963                 rc = jbd_block_set(journal->jbd_fs, &data_block);
1964                 if (rc != EOK) {
1965                         desc_iblock = 0;
1966                         ext4_bcache_clear_dirty(desc_block.buf);
1967                         jbd_block_set(journal->jbd_fs, &desc_block);
1968                         break;
1969                 }
1970                 jbd_buf->jbd_lba = data_iblock;
1971
1972                 tag_ptr += tag_info.tag_bytes;
1973                 tag_tbl_size -= tag_info.tag_bytes;
1974
1975                 i++;
1976         }
1977         if (rc == EOK && desc_iblock) {
1978                 jbd_meta_csum_set(journal->jbd_fs,
1979                                 (struct jbd_bhdr *)bhdr);
1980                 trans->data_csum = checksum;
1981                 rc = jbd_block_set(journal->jbd_fs, &desc_block);
1982         }
1983
1984         return rc;
1985 }
1986
1987 /**@brief  Write revoke block for a transaction
1988  * @param  journal current journal session
1989  * @param  trans transaction
1990  * @return standard error code*/
1991 static int
1992 jbd_journal_prepare_revoke(struct jbd_journal *journal,
1993                            struct jbd_trans *trans)
1994 {
1995         int rc = EOK, i = 0;
1996         struct ext4_block desc_block = EXT4_BLOCK_ZERO();
1997         int32_t tag_tbl_size = 0;
1998         uint32_t desc_iblock = 0;
1999         char *blocks_entry = NULL;
2000         struct jbd_revoke_rec *rec, *tmp;
2001         struct jbd_revoke_header *header = NULL;
2002         int32_t record_len = 4;
2003         struct jbd_bhdr *bhdr = NULL;
2004
2005         if (JBD_HAS_INCOMPAT_FEATURE(&journal->jbd_fs->sb,
2006                                      JBD_FEATURE_INCOMPAT_64BIT))
2007                 record_len = 8;
2008
2009         RB_FOREACH_SAFE(rec, jbd_revoke_tree, &trans->revoke_root,
2010                           tmp) {
2011 again:
2012                 if (!desc_iblock) {
2013                         desc_iblock = jbd_journal_alloc_block(journal, trans);
2014                         rc = jbd_block_get_noread(journal->jbd_fs, &desc_block,
2015                                                   desc_iblock);
2016                         if (rc != EOK)
2017                                 break;
2018
2019                         bhdr = (struct jbd_bhdr *)desc_block.data;
2020                         jbd_set32(bhdr, magic, JBD_MAGIC_NUMBER);
2021                         jbd_set32(bhdr, blocktype, JBD_REVOKE_BLOCK);
2022                         jbd_set32(bhdr, sequence, trans->trans_id);
2023
2024                         header = (struct jbd_revoke_header *)bhdr;
2025                         blocks_entry = (char *)(header + 1);
2026                         tag_tbl_size = journal->block_size -
2027                                 sizeof(struct jbd_revoke_header);
2028
2029                         if (jbd_has_csum(&journal->jbd_fs->sb))
2030                                 tag_tbl_size -= sizeof(struct jbd_block_tail);
2031
2032                         if (!trans->start_iblock)
2033                                 trans->start_iblock = desc_iblock;
2034
2035                         ext4_bcache_set_dirty(desc_block.buf);
2036                         ext4_bcache_set_flag(desc_block.buf, BC_TMP);
2037                 }
2038
2039                 if (tag_tbl_size < record_len) {
2040                         jbd_set32(header, count,
2041                                   journal->block_size - tag_tbl_size);
2042                         jbd_meta_csum_set(journal->jbd_fs, bhdr);
2043                         bhdr = NULL;
2044                         desc_iblock = 0;
2045                         header = NULL;
2046                         rc = jbd_block_set(journal->jbd_fs, &desc_block);
2047                         if (rc != EOK)
2048                                 break;
2049
2050                         goto again;
2051                 }
2052                 if (record_len == 8) {
2053                         uint64_t *blocks =
2054                                 (uint64_t *)blocks_entry;
2055                         *blocks = to_be64(rec->lba);
2056                 } else {
2057                         uint32_t *blocks =
2058                                 (uint32_t *)blocks_entry;
2059                         *blocks = to_be32((uint32_t)rec->lba);
2060                 }
2061                 blocks_entry += record_len;
2062                 tag_tbl_size -= record_len;
2063
2064                 i++;
2065         }
2066         if (rc == EOK && desc_iblock) {
2067                 if (header != NULL)
2068                         jbd_set32(header, count,
2069                                   journal->block_size - tag_tbl_size);
2070
2071                 jbd_meta_csum_set(journal->jbd_fs, bhdr);
2072                 rc = jbd_block_set(journal->jbd_fs, &desc_block);
2073         }
2074
2075         return rc;
2076 }
2077
2078 /**@brief  Put references of block descriptors in a transaction.
2079  * @param  journal current journal session
2080  * @param  trans transaction*/
2081 void jbd_journal_cp_trans(struct jbd_journal *journal, struct jbd_trans *trans)
2082 {
2083         struct jbd_buf *jbd_buf, *tmp;
2084         struct ext4_fs *fs = journal->jbd_fs->inode_ref.fs;
2085         TAILQ_FOREACH_SAFE(jbd_buf, &trans->buf_queue, buf_node,
2086                         tmp) {
2087                 struct ext4_block block = jbd_buf->block;
2088                 ext4_block_set(fs->bdev, &block);
2089         }
2090 }
2091
2092 /**@brief  Update the start block of the journal when
2093  *         all the contents in a transaction reach the disk.*/
2094 static void jbd_trans_end_write(struct ext4_bcache *bc __unused,
2095                           struct ext4_buf *buf,
2096                           int res,
2097                           void *arg)
2098 {
2099         struct jbd_buf *jbd_buf = arg;
2100         struct jbd_trans *trans = jbd_buf->trans;
2101         struct jbd_block_rec *block_rec = jbd_buf->block_rec;
2102         struct jbd_journal *journal = trans->journal;
2103         bool first_in_queue =
2104                 trans == TAILQ_FIRST(&journal->cp_queue);
2105         if (res != EOK)
2106                 trans->error = res;
2107
2108         TAILQ_REMOVE(&trans->buf_queue, jbd_buf, buf_node);
2109         TAILQ_REMOVE(&block_rec->dirty_buf_queue,
2110                         jbd_buf,
2111                         dirty_buf_node);
2112
2113         jbd_trans_finish_callback(journal,
2114                         trans,
2115                         jbd_buf->block_rec,
2116                         false,
2117                         false);
2118         if (block_rec->trans == trans && buf) {
2119                 /* Clear the end_write and end_write_arg fields. */
2120                 buf->end_write = NULL;
2121                 buf->end_write_arg = NULL;
2122         }
2123
2124         ext4_free(jbd_buf);
2125
2126         trans->written_cnt++;
2127         if (trans->written_cnt == trans->data_cnt) {
2128                 /* If it is the first transaction on checkpoint queue,
2129                  * we will shift the start of the journal to the next
2130                  * transaction, and remove subsequent written
2131                  * transactions from checkpoint queue until we find
2132                  * an unwritten one. */
2133                 if (first_in_queue) {
2134                         journal->start = trans->start_iblock +
2135                                 trans->alloc_blocks;
2136                         wrap(&journal->jbd_fs->sb, journal->start);
2137                         journal->trans_id = trans->trans_id + 1;
2138                         TAILQ_REMOVE(&journal->cp_queue, trans, trans_node);
2139                         jbd_journal_free_trans(journal, trans, false);
2140
2141                         jbd_journal_purge_cp_trans(journal, false, true);
2142                         jbd_journal_write_sb(journal);
2143                         jbd_write_sb(journal->jbd_fs);
2144                 }
2145         }
2146 }
2147
2148 /**@brief  Commit a transaction to the journal immediately.
2149  * @param  journal current journal session
2150  * @param  trans transaction
2151  * @return standard error code*/
2152 static int __jbd_journal_commit_trans(struct jbd_journal *journal,
2153                                       struct jbd_trans *trans)
2154 {
2155         int rc = EOK;
2156         uint32_t last = journal->last;
2157         struct jbd_revoke_rec *rec, *tmp;
2158
2159         trans->trans_id = journal->alloc_trans_id;
2160         rc = jbd_journal_prepare(journal, trans);
2161         if (rc != EOK)
2162                 goto Finish;
2163
2164         rc = jbd_journal_prepare_revoke(journal, trans);
2165         if (rc != EOK)
2166                 goto Finish;
2167
2168         if (TAILQ_EMPTY(&trans->buf_queue) &&
2169             RB_EMPTY(&trans->revoke_root)) {
2170                 /* Since there are no entries in both buffer list
2171                  * and revoke entry list, we do not consider trans as
2172                  * complete transaction and just return EOK.*/
2173                 jbd_journal_free_trans(journal, trans, false);
2174                 goto Finish;
2175         }
2176
2177         rc = jbd_trans_write_commit_block(trans);
2178         if (rc != EOK)
2179                 goto Finish;
2180
2181         journal->alloc_trans_id++;
2182
2183         /* Complete the checkpoint of buffers which are revoked. */
2184         RB_FOREACH_SAFE(rec, jbd_revoke_tree, &trans->revoke_root,
2185                         tmp) {
2186                 struct jbd_block_rec *block_rec =
2187                         jbd_trans_block_rec_lookup(journal, rec->lba);
2188                 struct jbd_buf *jbd_buf = NULL;
2189                 if (block_rec)
2190                         jbd_buf = TAILQ_LAST(&block_rec->dirty_buf_queue,
2191                                         jbd_buf_dirty);
2192                 if (jbd_buf) {
2193                         struct ext4_buf *buf;
2194                         struct ext4_block block = EXT4_BLOCK_ZERO();
2195                         /*
2196                          * We do this to reset the ext4_buf::end_write and
2197                          * ext4_buf::end_write_arg fields so that the checkpoint
2198                          * callback won't be triggered again.
2199                          */
2200                         buf = ext4_bcache_find_get(journal->jbd_fs->bdev->bc,
2201                                         &block,
2202                                         jbd_buf->block_rec->lba);
2203                         jbd_trans_end_write(journal->jbd_fs->bdev->bc,
2204                                         buf,
2205                                         EOK,
2206                                         jbd_buf);
2207                         if (buf)
2208                                 ext4_block_set(journal->jbd_fs->bdev, &block);
2209                 }
2210         }
2211
2212         if (TAILQ_EMPTY(&journal->cp_queue)) {
2213                 /*
2214                  * This transaction is going to be the first object in the
2215                  * checkpoint queue.
2216                  * When the first transaction in checkpoint queue is completely
2217                  * written to disk, we shift the tail of the log to right.
2218                  */
2219                 if (trans->data_cnt) {
2220                         journal->start = trans->start_iblock;
2221                         wrap(&journal->jbd_fs->sb, journal->start);
2222                         journal->trans_id = trans->trans_id;
2223                         jbd_journal_write_sb(journal);
2224                         jbd_write_sb(journal->jbd_fs);
2225                         TAILQ_INSERT_TAIL(&journal->cp_queue, trans,
2226                                         trans_node);
2227                         jbd_journal_cp_trans(journal, trans);
2228                 } else {
2229                         journal->start = trans->start_iblock +
2230                                 trans->alloc_blocks;
2231                         wrap(&journal->jbd_fs->sb, journal->start);
2232                         journal->trans_id = trans->trans_id + 1;
2233                         jbd_journal_write_sb(journal);
2234                         jbd_journal_free_trans(journal, trans, false);
2235                 }
2236         } else {
2237                 /* No need to do anything to the JBD superblock. */
2238                 TAILQ_INSERT_TAIL(&journal->cp_queue, trans,
2239                                 trans_node);
2240                 if (trans->data_cnt)
2241                         jbd_journal_cp_trans(journal, trans);
2242         }
2243 Finish:
2244         if (rc != EOK && rc != ENOSPC) {
2245                 journal->last = last;
2246                 jbd_journal_free_trans(journal, trans, true);
2247         }
2248         return rc;
2249 }
2250
2251 /**@brief  Allocate a new transaction
2252  * @param  journal current journal session
2253  * @return transaction allocated*/
2254 struct jbd_trans *
2255 jbd_journal_new_trans(struct jbd_journal *journal)
2256 {
2257         struct jbd_trans *trans = NULL;
2258         trans = ext4_calloc(1, sizeof(struct jbd_trans));
2259         if (!trans)
2260                 return NULL;
2261
2262         /* We will assign a trans_id to this transaction,
2263          * once it has been committed.*/
2264         trans->journal = journal;
2265         trans->data_csum = EXT4_CRC32_INIT;
2266         trans->error = EOK;
2267         TAILQ_INIT(&trans->buf_queue);
2268         return trans;
2269 }
2270
2271 /**@brief  Commit a transaction to the journal immediately.
2272  * @param  journal current journal session
2273  * @param  trans transaction
2274  * @return standard error code*/
2275 int jbd_journal_commit_trans(struct jbd_journal *journal,
2276                              struct jbd_trans *trans)
2277 {
2278         int r = EOK;
2279         r = __jbd_journal_commit_trans(journal, trans);
2280         return r;
2281 }
2282
2283 /**
2284  * @}
2285  */