debugfs/do_journal.c

   1 /*
   2  * do_journal.c --- Scribble onto the journal!
   3  *
   4  * Copyright (C) 2014 Oracle.  This file may be redistributed
   5  * under the terms of the GNU Public License.
   6  */
   7
   8 #include "config.h"
   9 #include <stdio.h>
  10 #ifdef HAVE_GETOPT_H
  11 #include <getopt.h>
  12 #else
  13 extern int optind;
  14 extern char *optarg;
  15 #endif
  16 #include <ctype.h>
  17 #include <unistd.h>
  18 #ifdef HAVE_SYS_TIME_H
  19 #include <sys/time.h>
  20 #endif
  21
  22 #include "debugfs.h"
  23 #include "ext2fs/kernel-jbd.h"
  24 #include "journal.h"
  25
  26 #undef DEBUG
  27
  28 #ifdef DEBUG
  29 # define dbg_printf(f, a...)  do {printf("JFS DEBUG: " f, ## a); \
  30         fflush(stdout); \
  31 } while (0)
  32 #else
  33 # define dbg_printf(f, a...)
  34 #endif
  35
  36 #define JOURNAL_CHECK_TRANS_MAGIC(x)    \
  37         do { \
  38                 if ((x)->magic != J_TRANS_MAGIC) \
  39                         return EXT2_ET_INVALID_ARGUMENT; \
  40         } while (0)
  41
  42 #define J_TRANS_MAGIC           0xD15EA5ED
  43 #define J_TRANS_OPEN            1
  44 #define J_TRANS_COMMITTED       2
  45 struct journal_transaction_s {
  46         unsigned int magic;
  47         ext2_filsys fs;
  48         journal_t *journal;
  49         blk64_t block;
  50         blk64_t start, end;
  51         tid_t tid;
  52         int flags;
  53 };
  54
  55 typedef struct journal_transaction_s journal_transaction_t;
  56
  57 static journal_t *current_journal = NULL;
  58
  59 static void journal_dump_trans(journal_transaction_t *trans EXT2FS_ATTR((unused)),
  60                                const char *tag EXT2FS_ATTR((unused)))
  61 {
  62         dbg_printf("TRANS %p(%s): tid=%d start=%llu block=%llu end=%llu "
  63                    "flags=0x%x\n", trans, tag, trans->tid, trans->start,
  64                    trans->block, trans->end, trans->flags);
  65 }
  66
  67 static errcode_t journal_commit_trans(journal_transaction_t *trans)
  68 {
  69         struct buffer_head *bh, *cbh = NULL;
  70         struct commit_header *commit;
  71 #ifdef HAVE_SYS_TIME_H
  72         struct timeval tv;
  73 #endif
  74         errcode_t err;
  75
  76         JOURNAL_CHECK_TRANS_MAGIC(trans);
  77
  78         if ((trans->flags & J_TRANS_COMMITTED) ||
  79             !(trans->flags & J_TRANS_OPEN))
  80                 return EXT2_ET_INVALID_ARGUMENT;
  81
  82         bh = getblk(trans->journal->j_dev, 0, trans->journal->j_blocksize);
  83         if (bh == NULL)
  84                 return ENOMEM;
  85
  86         /* write the descriptor block header */
  87         commit = (struct commit_header *)bh->b_data;
  88         commit->h_magic = ext2fs_cpu_to_be32(JBD2_MAGIC_NUMBER);
  89         commit->h_blocktype = ext2fs_cpu_to_be32(JBD2_COMMIT_BLOCK);
  90         commit->h_sequence = ext2fs_cpu_to_be32(trans->tid);
  91         if (jbd2_has_feature_checksum(trans->journal)) {
  92                 __u32 csum_v1 = ~0;
  93                 blk64_t cblk;
  94
  95                 cbh = getblk(trans->journal->j_dev, 0,
  96                              trans->journal->j_blocksize);
  97                 if (cbh == NULL) {
  98                         err = ENOMEM;
  99                         goto error;
 100                 }
 101
 102                 for (cblk = trans->start; cblk < trans->block; cblk++) {
 103                         err = jbd2_journal_bmap(trans->journal, cblk,
 104                                                 &cbh->b_blocknr);
 105                         if (err)
 106                                 goto error;
 107                         mark_buffer_uptodate(cbh, 0);
 108                         ll_rw_block(REQ_OP_READ, 0, 1, &cbh);
 109                         err = cbh->b_err;
 110                         if (err)
 111                                 goto error;
 112                         csum_v1 = ext2fs_crc32_be(csum_v1,
 113                                         (unsigned char const *)cbh->b_data,
 114                                         cbh->b_size);
 115                 }
 116
 117                 commit->h_chksum_type = JBD2_CRC32_CHKSUM;
 118                 commit->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
 119                 commit->h_chksum[0] = ext2fs_cpu_to_be32(csum_v1);
 120         } else {
 121                 commit->h_chksum_type = 0;
 122                 commit->h_chksum_size = 0;
 123                 commit->h_chksum[0] = 0;
 124         }
 125 #ifdef HAVE_SYS_TIME_H
 126         gettimeofday(&tv, NULL);
 127         commit->h_commit_sec = ext2fs_cpu_to_be32(tv.tv_sec);
 128         commit->h_commit_nsec = ext2fs_cpu_to_be32(tv.tv_usec * 1000);
 129 #else
 130         commit->h_commit_sec = 0;
 131         commit->h_commit_nsec = 0;
 132 #endif
 133
 134         /* Write block */
 135         jbd2_commit_block_csum_set(trans->journal, bh);
 136         err = jbd2_journal_bmap(trans->journal, trans->block, &bh->b_blocknr);
 137         if (err)
 138                 goto error;
 139
 140         dbg_printf("Writing commit block at %llu:%llu\n", trans->block,
 141                    bh->b_blocknr);
 142         mark_buffer_dirty(bh);
 143         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 144         err = bh->b_err;
 145         if (err)
 146                 goto error;
 147         trans->flags |= J_TRANS_COMMITTED;
 148         trans->flags &= ~J_TRANS_OPEN;
 149         trans->block++;
 150
 151         ext2fs_set_feature_journal_needs_recovery(trans->fs->super);
 152         ext2fs_mark_super_dirty(trans->fs);
 153 error:
 154         if (cbh)
 155                 brelse(cbh);
 156         brelse(bh);
 157         return err;
 158 }
 159
 160 static errcode_t journal_add_revoke_to_trans(journal_transaction_t *trans,
 161                                              blk64_t *revoke_list,
 162                                              size_t revoke_len)
 163 {
 164         jbd2_journal_revoke_header_t *jrb;
 165         void *buf;
 166         size_t i, offset;
 167         blk64_t curr_blk;
 168         unsigned int sz;
 169         unsigned csum_size = 0;
 170         struct buffer_head *bh;
 171         errcode_t err;
 172
 173         JOURNAL_CHECK_TRANS_MAGIC(trans);
 174
 175         if ((trans->flags & J_TRANS_COMMITTED) ||
 176             !(trans->flags & J_TRANS_OPEN))
 177                 return EXT2_ET_INVALID_ARGUMENT;
 178
 179         if (revoke_len == 0)
 180                 return 0;
 181
 182         /* Do we need to leave space at the end for a checksum? */
 183         if (jbd2_journal_has_csum_v2or3(trans->journal))
 184                 csum_size = sizeof(struct jbd2_journal_block_tail);
 185
 186         curr_blk = trans->block;
 187
 188         bh = getblk(trans->journal->j_dev, curr_blk,
 189                     trans->journal->j_blocksize);
 190         if (bh == NULL)
 191                 return ENOMEM;
 192         jrb = buf = bh->b_data;
 193         jrb->r_header.h_magic = ext2fs_cpu_to_be32(JBD2_MAGIC_NUMBER);
 194         jrb->r_header.h_blocktype = ext2fs_cpu_to_be32(JBD2_REVOKE_BLOCK);
 195         jrb->r_header.h_sequence = ext2fs_cpu_to_be32(trans->tid);
 196         offset = sizeof(*jrb);
 197
 198         if (jbd2_has_feature_64bit(trans->journal))
 199                 sz = 8;
 200         else
 201                 sz = 4;
 202
 203         for (i = 0; i < revoke_len; i++) {
 204                 /* Block full, write to journal */
 205                 if (offset + sz > trans->journal->j_blocksize - csum_size) {
 206                         jrb->r_count = ext2fs_cpu_to_be32(offset);
 207                         jbd2_revoke_csum_set(trans->journal, bh);
 208
 209                         err = jbd2_journal_bmap(trans->journal, curr_blk,
 210                                                 &bh->b_blocknr);
 211                         if (err)
 212                                 goto error;
 213                         dbg_printf("Writing revoke block at %llu:%llu\n",
 214                                    curr_blk, bh->b_blocknr);
 215                         mark_buffer_dirty(bh);
 216                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 217                         err = bh->b_err;
 218                         if (err)
 219                                 goto error;
 220
 221                         offset = sizeof(*jrb);
 222                         curr_blk++;
 223                 }
 224
 225                 if (revoke_list[i] >=
 226                     ext2fs_blocks_count(trans->journal->j_fs_dev->k_fs->super)) {
 227                         err = EXT2_ET_BAD_BLOCK_NUM;
 228                         goto error;
 229                 }
 230
 231                 if (jbd2_has_feature_64bit(trans->journal))
 232                         *((__u64 *)(&((char *)buf)[offset])) =
 233                                 ext2fs_cpu_to_be64(revoke_list[i]);
 234                 else
 235                         *((__u32 *)(&((char *)buf)[offset])) =
 236                                 ext2fs_cpu_to_be32(revoke_list[i]);
 237                 offset += sz;
 238         }
 239
 240         if (offset > 0) {
 241                 jrb->r_count = ext2fs_cpu_to_be32(offset);
 242                 jbd2_revoke_csum_set(trans->journal, bh);
 243
 244                 err = jbd2_journal_bmap(trans->journal, curr_blk,
 245                                         &bh->b_blocknr);
 246                 if (err)
 247                         goto error;
 248                 dbg_printf("Writing revoke block at %llu:%llu\n",
 249                            curr_blk, bh->b_blocknr);
 250                 mark_buffer_dirty(bh);
 251                 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 252                 err = bh->b_err;
 253                 if (err)
 254                         goto error;
 255                 curr_blk++;
 256         }
 257
 258 error:
 259         trans->block = curr_blk;
 260         brelse(bh);
 261         return err;
 262 }
 263
 264 static errcode_t journal_add_blocks_to_trans(journal_transaction_t *trans,
 265                                       blk64_t *block_list, size_t block_len,
 266                                       FILE *fp)
 267 {
 268         blk64_t curr_blk, jdb_blk;
 269         size_t i, j;
 270         int csum_size = 0;
 271         journal_header_t *jdb;
 272         journal_block_tag_t *jdbt;
 273         int tag_bytes;
 274         void *buf = NULL, *jdb_buf = NULL;
 275         struct buffer_head *bh = NULL, *data_bh;
 276         errcode_t err;
 277
 278         JOURNAL_CHECK_TRANS_MAGIC(trans);
 279
 280         if ((trans->flags & J_TRANS_COMMITTED) ||
 281             !(trans->flags & J_TRANS_OPEN))
 282                 return EXT2_ET_INVALID_ARGUMENT;
 283
 284         if (block_len == 0)
 285                 return 0;
 286
 287         /* Do we need to leave space at the end for a checksum? */
 288         if (jbd2_journal_has_csum_v2or3(trans->journal))
 289                 csum_size = sizeof(struct jbd2_journal_block_tail);
 290
 291         curr_blk = jdb_blk = trans->block;
 292
 293         data_bh = getblk(trans->journal->j_dev, curr_blk,
 294                          trans->journal->j_blocksize);
 295         if (data_bh == NULL)
 296                 return ENOMEM;
 297         buf = data_bh->b_data;
 298
 299         /* write the descriptor block header */
 300         bh = getblk(trans->journal->j_dev, curr_blk,
 301                     trans->journal->j_blocksize);
 302         if (bh == NULL) {
 303                 err = ENOMEM;
 304                 goto error;
 305         }
 306         jdb = jdb_buf = bh->b_data;
 307         jdb->h_magic = ext2fs_cpu_to_be32(JBD2_MAGIC_NUMBER);
 308         jdb->h_blocktype = ext2fs_cpu_to_be32(JBD2_DESCRIPTOR_BLOCK);
 309         jdb->h_sequence = ext2fs_cpu_to_be32(trans->tid);
 310         jdbt = (journal_block_tag_t *)(jdb + 1);
 311
 312         curr_blk++;
 313         for (i = 0; i < block_len; i++) {
 314                 j = fread(data_bh->b_data, trans->journal->j_blocksize, 1, fp);
 315                 if (j != 1) {
 316                         err = errno;
 317                         goto error;
 318                 }
 319
 320                 tag_bytes = journal_tag_bytes(trans->journal);
 321
 322                 /* No space left in descriptor block, write it out */
 323                 if ((char *)jdbt + tag_bytes >
 324                     (char *)jdb_buf + trans->journal->j_blocksize - csum_size) {
 325                         jbd2_descr_block_csum_set(trans->journal, bh);
 326                         err = jbd2_journal_bmap(trans->journal, jdb_blk,
 327                                            &bh->b_blocknr);
 328                         if (err)
 329                                 goto error;
 330                         dbg_printf("Writing descriptor block at %llu:%llu\n",
 331                                    jdb_blk, bh->b_blocknr);
 332                         mark_buffer_dirty(bh);
 333                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 334                         err = bh->b_err;
 335                         if (err)
 336                                 goto error;
 337
 338                         jdbt = (journal_block_tag_t *)(jdb + 1);
 339                         jdb_blk = curr_blk;
 340                         curr_blk++;
 341                 }
 342
 343                 if (block_list[i] >=
 344                     ext2fs_blocks_count(trans->journal->j_fs_dev->k_fs->super)) {
 345                         err = EXT2_ET_BAD_BLOCK_NUM;
 346                         goto error;
 347                 }
 348
 349                 /* Fill out the block tag */
 350                 jdbt->t_blocknr = ext2fs_cpu_to_be32(block_list[i] & 0xFFFFFFFF);
 351                 jdbt->t_flags = 0;
 352                 if (jdbt != (journal_block_tag_t *)(jdb + 1))
 353                         jdbt->t_flags |= ext2fs_cpu_to_be16(JBD2_FLAG_SAME_UUID);
 354                 else {
 355                         memcpy(jdbt + tag_bytes,
 356                                trans->journal->j_superblock->s_uuid,
 357                                sizeof(trans->journal->j_superblock->s_uuid));
 358                         tag_bytes += 16;
 359                 }
 360                 if (i == block_len - 1)
 361                         jdbt->t_flags |= ext2fs_cpu_to_be16(JBD2_FLAG_LAST_TAG);
 362                 if (*((__u32 *)buf) == ext2fs_cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 363                         *((__u32 *)buf) = 0;
 364                         jdbt->t_flags |= ext2fs_cpu_to_be16(JBD2_FLAG_ESCAPE);
 365                 }
 366                 if (jbd2_has_feature_64bit(trans->journal))
 367                         jdbt->t_blocknr_high = ext2fs_cpu_to_be32(block_list[i] >> 32);
 368                 jbd2_block_tag_csum_set(trans->journal, jdbt, data_bh,
 369                                         trans->tid);
 370
 371                 /* Write the data block */
 372                 err = jbd2_journal_bmap(trans->journal, curr_blk,
 373                                         &data_bh->b_blocknr);
 374                 if (err)
 375                         goto error;
 376                 dbg_printf("Writing data block %llu at %llu:%llu tag %d\n",
 377                            block_list[i], curr_blk, data_bh->b_blocknr,
 378                            tag_bytes);
 379                 mark_buffer_dirty(data_bh);
 380                 ll_rw_block(REQ_OP_WRITE, 0, 1, &data_bh);
 381                 err = data_bh->b_err;
 382                 if (err)
 383                         goto error;
 384
 385                 curr_blk++;
 386                 jdbt = (journal_block_tag_t *)(((char *)jdbt) + tag_bytes);
 387         }
 388
 389         /* Write out the last descriptor block */
 390         if (jdbt != (journal_block_tag_t *)(jdb + 1)) {
 391                 jbd2_descr_block_csum_set(trans->journal, bh);
 392                 err = jbd2_journal_bmap(trans->journal, jdb_blk,
 393                                         &bh->b_blocknr);
 394                 if (err)
 395                         goto error;
 396                 dbg_printf("Writing descriptor block at %llu:%llu\n",
 397                            jdb_blk, bh->b_blocknr);
 398                 mark_buffer_dirty(bh);
 399                 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 400                 err = bh->b_err;
 401                 if (err)
 402                         goto error;
 403         }
 404
 405 error:
 406         trans->block = curr_blk;
 407         if (bh)
 408                 brelse(bh);
 409         brelse(data_bh);
 410         return err;
 411 }
 412
 413 static blk64_t journal_guess_blocks(journal_t *journal, blk64_t data_blocks,
 414                                     blk64_t revoke_blocks)
 415 {
 416         blk64_t ret = 1;
 417         unsigned int bs, sz;
 418
 419         /* Estimate # of revoke blocks */
 420         bs = journal->j_blocksize;
 421         if (jbd2_journal_has_csum_v2or3(journal))
 422                 bs -= sizeof(struct jbd2_journal_block_tail);
 423         sz = jbd2_has_feature_64bit(journal) ? sizeof(__u64) : sizeof(__u32);
 424         ret += revoke_blocks * sz / bs;
 425
 426         /* Estimate # of data blocks */
 427         bs = journal->j_blocksize - 16;
 428         if (jbd2_journal_has_csum_v2or3(journal))
 429                 bs -= sizeof(struct jbd2_journal_block_tail);
 430         sz = journal_tag_bytes(journal);
 431         ret += data_blocks * sz / bs;
 432
 433         ret += data_blocks;
 434
 435         return ret;
 436 }
 437
 438 static errcode_t journal_open_trans(journal_t *journal,
 439                                     journal_transaction_t *trans,
 440                                     blk64_t blocks)
 441 {
 442         trans->fs = journal->j_fs_dev->k_fs;
 443         trans->journal = journal;
 444         trans->flags = J_TRANS_OPEN;
 445
 446         if (journal->j_tail == 0) {
 447                 /* Clean journal, start at the tail */
 448                 trans->tid = journal->j_tail_sequence;
 449                 trans->start = journal->j_first;
 450         } else {
 451                 /* Put new transaction at the head of the list */
 452                 trans->tid = journal->j_transaction_sequence;
 453                 trans->start = journal->j_head;
 454         }
 455
 456         trans->block = trans->start;
 457         if (trans->start + blocks > journal->j_last)
 458                 return ENOSPC;
 459         trans->end = trans->block + blocks;
 460         journal_dump_trans(trans, "new transaction");
 461
 462         trans->magic = J_TRANS_MAGIC;
 463         return 0;
 464 }
 465
 466 static errcode_t journal_close_trans(journal_transaction_t *trans)
 467 {
 468         journal_t *journal;
 469
 470         JOURNAL_CHECK_TRANS_MAGIC(trans);
 471
 472         if (!(trans->flags & J_TRANS_COMMITTED))
 473                 return 0;
 474
 475         journal = trans->journal;
 476         if (journal->j_tail == 0) {
 477                 /* Update the tail */
 478                 journal->j_tail_sequence = trans->tid;
 479                 journal->j_tail = trans->start;
 480                 journal->j_superblock->s_start = ext2fs_cpu_to_be32(trans->start);
 481         }
 482
 483         /* Update the head */
 484         journal->j_head = trans->end + 1;
 485         journal->j_transaction_sequence = trans->tid + 1;
 486
 487         trans->magic = 0;
 488
 489         /* Mark ourselves as needing recovery */
 490         if (!ext2fs_has_feature_journal_needs_recovery(trans->fs->super)) {
 491                 ext2fs_set_feature_journal_needs_recovery(trans->fs->super);
 492                 ext2fs_mark_super_dirty(trans->fs);
 493         }
 494
 495         return 0;
 496 }
 497
 498 #define JOURNAL_WRITE_NO_COMMIT         1
 499 static errcode_t journal_write(journal_t *journal,
 500                                int flags, blk64_t *block_list,
 501                                size_t block_len, blk64_t *revoke_list,
 502                                size_t revoke_len, FILE *fp)
 503 {
 504         blk64_t blocks;
 505         journal_transaction_t trans;
 506         errcode_t err;
 507
 508         if (revoke_len > 0) {
 509                 jbd2_set_feature_revoke(journal);
 510                 mark_buffer_dirty(journal->j_sb_buffer);
 511         }
 512
 513         blocks = journal_guess_blocks(journal, block_len, revoke_len);
 514         err = journal_open_trans(journal, &trans, blocks);
 515         if (err)
 516                 goto error;
 517
 518         err = journal_add_blocks_to_trans(&trans, block_list, block_len, fp);
 519         if (err)
 520                 goto error;
 521
 522         err = journal_add_revoke_to_trans(&trans, revoke_list, revoke_len);
 523         if (err)
 524                 goto error;
 525
 526         if (!(flags & JOURNAL_WRITE_NO_COMMIT)) {
 527                 err = journal_commit_trans(&trans);
 528                 if (err)
 529                         goto error;
 530         }
 531
 532         err = journal_close_trans(&trans);
 533         if (err)
 534                 goto error;
 535 error:
 536         return err;
 537 }
 538
 539 void do_journal_write(int argc, char *argv[], int sci_idx EXT2FS_ATTR((unused)),
 540                       void *infop EXT2FS_ATTR((unused)))
 541 {
 542         blk64_t *blist = NULL, *rlist = NULL;
 543         size_t bn = 0, rn = 0;
 544         FILE *fp = NULL;
 545         int opt;
 546         int flags = 0;
 547         errcode_t err;
 548
 549         if (current_journal == NULL) {
 550                 printf("Journal not open.\n");
 551                 return;
 552         }
 553
 554         reset_getopt();
 555         while ((opt = getopt(argc, argv, "b:r:c")) != -1) {
 556                 switch (opt) {
 557                 case 'b':
 558                         err = read_list(optarg, &blist, &bn);
 559                         if (err)
 560                                 com_err(argv[0], err,
 561                                         "while reading block list");
 562                         break;
 563                 case 'r':
 564                         err = read_list(optarg, &rlist, &rn);
 565                         if (err)
 566                                 com_err(argv[0], err,
 567                                         "while reading revoke list");
 568                         break;
 569                 case 'c':
 570                         flags |= JOURNAL_WRITE_NO_COMMIT;
 571                         break;
 572                 default:
 573                         printf("%s [-b blocks] [-r revoke] [-c] file\n",
 574                                argv[0]);
 575                         printf("-b: Write these blocks into transaction.\n");
 576                         printf("-c: Do not commit transaction.\n");
 577                         printf("-r: Revoke these blocks from transaction.\n");
 578
 579                         goto out;
 580                 }
 581         }
 582
 583         if (bn > 0 && optind != argc - 1) {
 584                 printf("Need a file to read blocks from.\n");
 585                 return;
 586         }
 587
 588         if (bn > 0) {
 589                 fp = fopen(argv[optind], "r");
 590                 if (fp == NULL) {
 591                         com_err(argv[0], errno,
 592                                 "while opening journal data file");
 593                         goto out;
 594                 }
 595         }
 596
 597         err = journal_write(current_journal, flags, blist, bn,
 598                             rlist, rn, fp);
 599         if (err)
 600                 com_err("journal_write", err, "while writing journal");
 601
 602         if (fp)
 603                 fclose(fp);
 604 out:
 605         if (blist)
 606                 free(blist);
 607         if (rlist)
 608                 free(rlist);
 609 }
 610
 611 /* Make sure we wrap around the log correctly! */
 612 #define wrap(journal, var)                                              \
 613 do {                                                                    \
 614         if (var >= (journal)->j_last)                                   \
 615                 var -= ((journal)->j_last - (journal)->j_first);        \
 616 } while (0)
 617
 618 /*
 619  * Count the number of in-use tags in a journal descriptor block.
 620  */
 621
 622 static int count_tags(journal_t *journal, char *buf)
 623 {
 624         char                    *tagp;
 625         journal_block_tag_t     *tag;
 626         int                     nr = 0, size = journal->j_blocksize;
 627         int                     tag_bytes = journal_tag_bytes(journal);
 628
 629         if (jbd2_journal_has_csum_v2or3(journal))
 630                 size -= sizeof(struct jbd2_journal_block_tail);
 631
 632         tagp = buf + sizeof(journal_header_t);
 633
 634         while ((tagp - buf + tag_bytes) <= size) {
 635                 tag = (journal_block_tag_t *) tagp;
 636
 637                 nr++;
 638                 tagp += tag_bytes;
 639                 if (!(tag->t_flags & ext2fs_cpu_to_be16(JBD2_FLAG_SAME_UUID)))
 640                         tagp += 16;
 641
 642                 if (tag->t_flags & ext2fs_cpu_to_be16(JBD2_FLAG_LAST_TAG))
 643                         break;
 644         }
 645
 646         return nr;
 647 }
 648
 649 static errcode_t journal_find_head(journal_t *journal)
 650 {
 651         unsigned int            next_commit_ID;
 652         blk64_t                 next_log_block, head_block;
 653         int                     err;
 654         journal_superblock_t    *sb;
 655         journal_header_t        *tmp;
 656         struct buffer_head      *bh;
 657         unsigned int            sequence;
 658         int                     blocktype;
 659
 660         /*
 661          * First thing is to establish what we expect to find in the log
 662          * (in terms of transaction IDs), and where (in terms of log
 663          * block offsets): query the superblock.
 664          */
 665
 666         sb = journal->j_superblock;
 667         next_commit_ID = ext2fs_be32_to_cpu(sb->s_sequence);
 668         next_log_block = ext2fs_be32_to_cpu(sb->s_start);
 669         head_block = next_log_block;
 670
 671         if (next_log_block == 0)
 672                 return 0;
 673
 674         bh = getblk(journal->j_dev, 0, journal->j_blocksize);
 675         if (bh == NULL)
 676                 return ENOMEM;
 677
 678         /*
 679          * Now we walk through the log, transaction by transaction,
 680          * making sure that each transaction has a commit block in the
 681          * expected place.  Each complete transaction gets replayed back
 682          * into the main filesystem.
 683          */
 684         while (1) {
 685                 dbg_printf("Scanning for sequence ID %u at %lu/%lu\n",
 686                           next_commit_ID, (unsigned long)next_log_block,
 687                           journal->j_last);
 688
 689                 /* Skip over each chunk of the transaction looking
 690                  * either the next descriptor block or the final commit
 691                  * record. */
 692                 err = jbd2_journal_bmap(journal, next_log_block,
 693                                         &bh->b_blocknr);
 694                 if (err)
 695                         goto err;
 696                 mark_buffer_uptodate(bh, 0);
 697                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
 698                 err = bh->b_err;
 699                 if (err)
 700                         goto err;
 701
 702                 next_log_block++;
 703                 wrap(journal, next_log_block);
 704
 705                 /* What kind of buffer is it?
 706                  *
 707                  * If it is a descriptor block, check that it has the
 708                  * expected sequence number.  Otherwise, we're all done
 709                  * here. */
 710
 711                 tmp = (journal_header_t *)bh->b_data;
 712
 713                 if (tmp->h_magic != ext2fs_cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 714                         dbg_printf("JBD2: wrong magic 0x%x\n", tmp->h_magic);
 715                         goto err;
 716                 }
 717
 718                 blocktype = ext2fs_be32_to_cpu(tmp->h_blocktype);
 719                 sequence = ext2fs_be32_to_cpu(tmp->h_sequence);
 720                 dbg_printf("Found magic %d, sequence %d\n",
 721                           blocktype, sequence);
 722
 723                 if (sequence != next_commit_ID) {
 724                         dbg_printf("JBD2: Wrong sequence %d (wanted %d)\n",
 725                                    sequence, next_commit_ID);
 726                         goto err;
 727                 }
 728
 729                 /* OK, we have a valid descriptor block which matches
 730                  * all of the sequence number checks.  What are we going
 731                  * to do with it?  That depends on the pass... */
 732
 733                 switch (blocktype) {
 734                 case JBD2_DESCRIPTOR_BLOCK:
 735                         next_log_block += count_tags(journal, bh->b_data);
 736                         wrap(journal, next_log_block);
 737                         continue;
 738
 739                 case JBD2_COMMIT_BLOCK:
 740                         head_block = next_log_block;
 741                         next_commit_ID++;
 742                         continue;
 743
 744                 case JBD2_REVOKE_BLOCK:
 745                         continue;
 746
 747                 default:
 748                         dbg_printf("Unrecognised magic %d, end of scan.\n",
 749                                   blocktype);
 750                         err = -EINVAL;
 751                         goto err;
 752                 }
 753         }
 754
 755 err:
 756         if (err == 0) {
 757                 dbg_printf("head seq=%d blk=%llu\n", next_commit_ID,
 758                            head_block);
 759                 journal->j_transaction_sequence = next_commit_ID;
 760                 journal->j_head = head_block;
 761         }
 762         brelse(bh);
 763         return err;
 764 }
 765
 766 static void update_journal_csum(journal_t *journal, int ver)
 767 {
 768         journal_superblock_t *jsb;
 769
 770         if (journal->j_format_version < 2)
 771                 return;
 772
 773         if (journal->j_tail != 0 ||
 774             ext2fs_has_feature_journal_needs_recovery(
 775                                         journal->j_fs_dev->k_fs->super)) {
 776                 printf("Journal needs recovery, will not add csums.\n");
 777                 return;
 778         }
 779
 780         /* metadata_csum implies journal csum v3 */
 781         jsb = journal->j_superblock;
 782         if (ext2fs_has_feature_metadata_csum(journal->j_fs_dev->k_fs->super)) {
 783                 printf("Setting csum v%d\n", ver);
 784                 switch (ver) {
 785                 case 2:
 786                         jbd2_clear_feature_csum3(journal);
 787                         jbd2_set_feature_csum2(journal);
 788                         jbd2_clear_feature_checksum(journal);
 789                         break;
 790                 case 3:
 791                         jbd2_set_feature_csum3(journal);
 792                         jbd2_clear_feature_csum2(journal);
 793                         jbd2_clear_feature_checksum(journal);
 794                         break;
 795                 default:
 796                         printf("Unknown checksum v%d\n", ver);
 797                         break;
 798                 }
 799                 journal->j_superblock->s_checksum_type = JBD2_CRC32C_CHKSUM;
 800                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
 801                                                    sizeof(jsb->s_uuid));
 802         } else {
 803                 jbd2_clear_feature_csum3(journal);
 804                 jbd2_clear_feature_csum2(journal);
 805                 jbd2_set_feature_checksum(journal);
 806         }
 807 }
 808
 809 static void update_uuid(journal_t *journal)
 810 {
 811         size_t z;
 812         ext2_filsys fs;
 813
 814         if (journal->j_format_version < 2)
 815                 return;
 816
 817         for (z = 0; z < sizeof(journal->j_superblock->s_uuid); z++)
 818                 if (journal->j_superblock->s_uuid[z])
 819                         break;
 820         if (z == 0)
 821                 return;
 822
 823         fs = journal->j_fs_dev->k_fs;
 824         if (!ext2fs_has_feature_64bit(fs->super))
 825                 return;
 826
 827         if (jbd2_has_feature_64bit(journal) &&
 828             ext2fs_has_feature_64bit(fs->super))
 829                 return;
 830
 831         if (journal->j_tail != 0 ||
 832             ext2fs_has_feature_journal_needs_recovery(fs->super)) {
 833                 printf("Journal needs recovery, will not set 64bit.\n");
 834                 return;
 835         }
 836
 837         memcpy(journal->j_superblock->s_uuid, fs->super->s_uuid,
 838                sizeof(fs->super->s_uuid));
 839 }
 840
 841 static void update_64bit_flag(journal_t *journal)
 842 {
 843         if (journal->j_format_version < 2)
 844                 return;
 845
 846         if (!ext2fs_has_feature_64bit(journal->j_fs_dev->k_fs->super))
 847                 return;
 848
 849         if (jbd2_has_feature_64bit(journal) &&
 850             ext2fs_has_feature_64bit(journal->j_fs_dev->k_fs->super))
 851                 return;
 852
 853         if (journal->j_tail != 0 ||
 854             ext2fs_has_feature_journal_needs_recovery(
 855                                 journal->j_fs_dev->k_fs->super)) {
 856                 printf("Journal needs recovery, will not set 64bit.\n");
 857                 return;
 858         }
 859
 860         jbd2_set_feature_64bit(journal);
 861 }
 862
 863 void do_journal_open(int argc, char *argv[], int sci_idx EXT2FS_ATTR((unused)),
 864                      void *infop EXT2FS_ATTR((unused)))
 865 {
 866         int opt, enable_csum = 0, csum_ver = 3;
 867         journal_t *journal;
 868         errcode_t err;
 869
 870         if (check_fs_open(argv[0]))
 871                 return;
 872         if (check_fs_read_write(argv[0]))
 873                 return;
 874         if (check_fs_bitmaps(argv[0]))
 875                 return;
 876         if (current_journal) {
 877                 printf("Journal is already open.\n");
 878                 return;
 879         }
 880         if (!ext2fs_has_feature_journal(current_fs->super)) {
 881                 printf("Journalling is not enabled on this filesystem.\n");
 882                 return;
 883         }
 884
 885         reset_getopt();
 886         while ((opt = getopt(argc, argv, "cv:f:")) != -1) {
 887                 switch (opt) {
 888                 case 'c':
 889                         enable_csum = 1;
 890                         break;
 891                 case 'f':
 892                         if (current_fs->journal_name)
 893                                 free(current_fs->journal_name);
 894                         current_fs->journal_name = strdup(optarg);
 895                         break;
 896                 case 'v':
 897                         csum_ver = atoi(optarg);
 898                         if (csum_ver != 2 && csum_ver != 3) {
 899                                 printf("Unknown journal csum v%d\n", csum_ver);
 900                                 csum_ver = 3;
 901                         }
 902                         break;
 903                 default:
 904                         printf("%s: [-c] [-v ver] [-f ext_jnl]\n", argv[0]);
 905                         printf("-c: Enable journal checksumming.\n");
 906                         printf("-v: Use this version checksum format.\n");
 907                         printf("-f: Load this external journal.\n");
 908                 }
 909         }
 910
 911         err = ext2fs_open_journal(current_fs, &current_journal);
 912         if (err) {
 913                 com_err(argv[0], err, "while opening journal");
 914                 return;
 915         }
 916         journal = current_journal;
 917
 918         dbg_printf("JOURNAL: seq=%d tailseq=%d start=%lu first=%lu "
 919                    "maxlen=%lu\n", journal->j_tail_sequence,
 920                    journal->j_transaction_sequence, journal->j_tail,
 921                    journal->j_first, journal->j_last);
 922
 923         update_uuid(journal);
 924         update_64bit_flag(journal);
 925         if (enable_csum)
 926                 update_journal_csum(journal, csum_ver);
 927
 928         err = journal_find_head(journal);
 929         if (err)
 930                 com_err(argv[0], err, "while examining journal");
 931 }
 932
 933 void do_journal_close(int argc EXT2FS_ATTR((unused)),
 934                       char *argv[] EXT2FS_ATTR((unused)),
 935                       int sci_idx EXT2FS_ATTR((unused)),
 936                       void *infop EXT2FS_ATTR((unused)))
 937 {
 938         if (current_journal == NULL) {
 939                 printf("Journal not open.\n");
 940                 return;
 941         }
 942
 943         ext2fs_close_journal(current_fs, &current_journal);
 944 }
 945
 946 void do_journal_run(int argc EXT2FS_ATTR((unused)), char *argv[],
 947                     int sci_idx EXT2FS_ATTR((unused)),
 948                     void *infop EXT2FS_ATTR((unused)))
 949 {
 950         errcode_t err;
 951
 952         if (check_fs_open(argv[0]))
 953                 return;
 954         if (check_fs_read_write(argv[0]))
 955                 return;
 956         if (check_fs_bitmaps(argv[0]))
 957                 return;
 958         if (current_journal) {
 959                 printf("Please close the journal before recovering it.\n");
 960                 return;
 961         }
 962
 963         err = ext2fs_run_ext3_journal(&current_fs);
 964         if (err)
 965                 com_err("journal_run", err, "while recovering journal");
 966         else {
 967                 ext2fs_clear_feature_journal_needs_recovery(current_fs->super);
 968                 ext2fs_mark_super_dirty(current_fs);
 969         }
 970 }