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