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