lib/ext2fs/undo_io.c

   1 /*
   2  * undo_io.c --- This is the undo io manager that copies the old data that
   3  * copies the old data being overwritten into a tdb database
   4  *
   5  * Copyright IBM Corporation, 2007
   6  * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
   7  *
   8  * %Begin-Header%
   9  * This file may be redistributed under the terms of the GNU Library
  10  * General Public License, version 2.
  11  * %End-Header%
  12  */
  13
  14 #define _LARGEFILE_SOURCE
  15 #define _LARGEFILE64_SOURCE
  16
  17 #include "config.h"
  18 #include <stdio.h>
  19 #include <string.h>
  20 #if HAVE_UNISTD_H
  21 #include <unistd.h>
  22 #endif
  23 #if HAVE_ERRNO_H
  24 #include <errno.h>
  25 #endif
  26 #include <fcntl.h>
  27 #include <time.h>
  28 #ifdef __linux__
  29 #include <sys/utsname.h>
  30 #endif
  31 #if HAVE_SYS_STAT_H
  32 #include <sys/stat.h>
  33 #endif
  34 #if HAVE_SYS_TYPES_H
  35 #include <sys/types.h>
  36 #endif
  37 #if HAVE_SYS_RESOURCE_H
  38 #include <sys/resource.h>
  39 #endif
  40 #include <limits.h>
  41
  42 #include "ext2_fs.h"
  43 #include "ext2fs.h"
  44 #include "ext2fsP.h"
  45
  46 #ifdef __GNUC__
  47 #define ATTR(x) __attribute__(x)
  48 #else
  49 #define ATTR(x)
  50 #endif
  51
  52 #undef DEBUG
  53
  54 #ifdef DEBUG
  55 # define dbg_printf(f, a...)  do {printf(f, ## a); fflush(stdout); } while (0)
  56 #else
  57 # define dbg_printf(f, a...)
  58 #endif
  59
  60 /*
  61  * For checking structure magic numbers...
  62  */
  63
  64 #define EXT2_CHECK_MAGIC(struct, code) \
  65           if ((struct)->magic != (code)) return (code)
  66 /*
  67  * Undo file format: The file is cut up into undo_header.block_size blocks.
  68  * The first block contains the header.
  69  * The second block contains the superblock.
  70  * There is then a repeating series of blocks as follows:
  71  *   A key block, which contains undo_keys to map the following data blocks.
  72  *   Data blocks
  73  * (Note that there are pointers to the first key block and the sb, so this
  74  * order isn't strictly necessary.)
  75  */
  76 #define E2UNDO_MAGIC "E2UNDO02"
  77 #define KEYBLOCK_MAGIC 0xCADECADE
  78
  79 #define E2UNDO_STATE_FINISHED   0x1     /* undo file is complete */
  80
  81 #define E2UNDO_MIN_BLOCK_SIZE   1024    /* undo blocks are no less than 1KB */
  82 #define E2UNDO_MAX_BLOCK_SIZE   1048576 /* undo blocks are no more than 1MB */
  83
  84 struct undo_header {
  85         char magic[8];          /* "E2UNDO02" */
  86         __le64 num_keys;        /* how many keys? */
  87         __le64 super_offset;    /* where in the file is the superblock copy? */
  88         __le64 key_offset;      /* where do the key/data block chunks start? */
  89         __le32 block_size;      /* block size of the undo file */
  90         __le32 fs_block_size;   /* block size of the target device */
  91         __le32 sb_crc;          /* crc32c of the superblock */
  92         __le32 state;           /* e2undo state flags */
  93         __le32 f_compat;        /* compatible features (none so far) */
  94         __le32 f_incompat;      /* incompatible features (none so far) */
  95         __le32 f_rocompat;      /* ro compatible features (none so far) */
  96         __u8 padding[448];      /* padding */
  97         __le32 header_crc;      /* crc32c of this header (but not this field) */
  98 };
  99
 100 #define E2UNDO_MAX_EXTENT_BLOCKS        512     /* max extent size, in blocks */
 101
 102 struct undo_key {
 103         __le64 fsblk;           /* where in the fs does the block go */
 104         __le32 blk_crc;         /* crc32c of the block */
 105         __le32 size;            /* how many bytes in this block? */
 106 };
 107
 108 struct undo_key_block {
 109         __le32 magic;           /* KEYBLOCK_MAGIC number */
 110         __le32 crc;             /* block checksum */
 111         __le64 reserved;        /* zero */
 112
 113         struct undo_key keys[0];        /* keys, which come immediately after */
 114 };
 115
 116 struct undo_private_data {
 117         int     magic;
 118
 119         /* the undo file io channel */
 120         io_channel undo_file;
 121         blk64_t undo_blk_num;                   /* next free block */
 122         blk64_t key_blk_num;                    /* current key block location */
 123         blk64_t super_blk_num;                  /* superblock location */
 124         blk64_t first_key_blk;                  /* first key block location */
 125         struct undo_key_block *keyb;
 126         size_t num_keys, keys_in_block;
 127
 128         /* The backing io channel */
 129         io_channel real;
 130
 131         unsigned long long tdb_data_size;
 132         int tdb_written;
 133
 134         /* to support offset in unix I/O manager */
 135         ext2_loff_t offset;
 136
 137         ext2fs_block_bitmap written_block_map;
 138         struct struct_ext2_filsys fake_fs;
 139         char *tdb_file;
 140         struct undo_header hdr;
 141 };
 142 #define KEYS_PER_BLOCK(d) (((d)->tdb_data_size / sizeof(struct undo_key)) - 1)
 143
 144 static io_manager undo_io_backing_manager;
 145 static char *tdb_file;
 146 static int actual_size;
 147
 148 errcode_t set_undo_io_backing_manager(io_manager manager)
 149 {
 150         /*
 151          * We may want to do some validation later
 152          */
 153         undo_io_backing_manager = manager;
 154         return 0;
 155 }
 156
 157 errcode_t set_undo_io_backup_file(char *file_name)
 158 {
 159         tdb_file = strdup(file_name);
 160
 161         if (tdb_file == NULL) {
 162                 return EXT2_ET_NO_MEMORY;
 163         }
 164
 165         return 0;
 166 }
 167
 168 static errcode_t write_undo_indexes(struct undo_private_data *data, int flush)
 169 {
 170         errcode_t retval;
 171         struct ext2_super_block super;
 172         io_channel channel;
 173         int block_size;
 174         __u32 sb_crc, hdr_crc;
 175
 176         /* Spit out a key block, if there's any data */
 177         if (data->keys_in_block) {
 178                 data->keyb->magic = ext2fs_cpu_to_le32(KEYBLOCK_MAGIC);
 179                 data->keyb->crc = 0;
 180                 data->keyb->crc = ext2fs_cpu_to_le32(
 181                                          ext2fs_crc32c_le(~0,
 182                                          (unsigned char *)data->keyb,
 183                                          data->tdb_data_size));
 184                 dbg_printf("Writing keyblock to blk %llu\n", data->key_blk_num);
 185                 retval = io_channel_write_blk64(data->undo_file,
 186                                                 data->key_blk_num,
 187                                                 1, data->keyb);
 188                 if (retval)
 189                         return retval;
 190                 /* Move on to the next key block if it's full. */
 191                 if (data->keys_in_block == KEYS_PER_BLOCK(data)) {
 192                         memset(data->keyb, 0, data->tdb_data_size);
 193                         data->keys_in_block = 0;
 194                         data->key_blk_num = data->undo_blk_num;
 195                         data->undo_blk_num++;
 196                 }
 197         }
 198
 199         /* Prepare superblock for write */
 200         channel = data->real;
 201         block_size = channel->block_size;
 202
 203         io_channel_set_blksize(channel, SUPERBLOCK_OFFSET);
 204         retval = io_channel_read_blk64(channel, 1, -SUPERBLOCK_SIZE, &super);
 205         if (retval)
 206                 goto err_out;
 207         sb_crc = ext2fs_crc32c_le(~0, (unsigned char *)&super, SUPERBLOCK_SIZE);
 208         super.s_magic = ~super.s_magic;
 209
 210         /* Write the undo header to disk. */
 211         memcpy(data->hdr.magic, E2UNDO_MAGIC, sizeof(data->hdr.magic));
 212         data->hdr.num_keys = ext2fs_cpu_to_le64(data->num_keys);
 213         data->hdr.super_offset = ext2fs_cpu_to_le64(data->super_blk_num);
 214         data->hdr.key_offset = ext2fs_cpu_to_le64(data->first_key_blk);
 215         data->hdr.fs_block_size = ext2fs_cpu_to_le32(block_size);
 216         data->hdr.sb_crc = ext2fs_cpu_to_le32(sb_crc);
 217         hdr_crc = ext2fs_crc32c_le(~0, (unsigned char *)&data->hdr,
 218                                    sizeof(data->hdr) -
 219                                    sizeof(data->hdr.header_crc));
 220         data->hdr.header_crc = ext2fs_cpu_to_le32(hdr_crc);
 221         retval = io_channel_write_blk64(data->undo_file, 0,
 222                                         -(int)sizeof(data->hdr),
 223                                         &data->hdr);
 224         if (retval)
 225                 goto err_out;
 226
 227         /*
 228          * Record the entire superblock (in FS byte order) so that we can't
 229          * apply e2undo files to the wrong FS or out of order.
 230          */
 231         dbg_printf("Writing superblock to block %llu\n", data->super_blk_num);
 232         retval = io_channel_write_blk64(data->undo_file, data->super_blk_num,
 233                                         -SUPERBLOCK_SIZE, &super);
 234         if (retval)
 235                 goto err_out;
 236
 237         if (flush)
 238                 retval = io_channel_flush(data->undo_file);
 239 err_out:
 240         io_channel_set_blksize(channel, block_size);
 241         return retval;
 242 }
 243
 244 static errcode_t undo_setup_tdb(struct undo_private_data *data)
 245 {
 246         int i;
 247         errcode_t retval;
 248
 249         if (data->tdb_written == 1)
 250                 return 0;
 251
 252         data->tdb_written = 1;
 253
 254         /* Make a bitmap to track what we've written */
 255         memset(&data->fake_fs, 0, sizeof(data->fake_fs));
 256         data->fake_fs.blocksize = data->tdb_data_size;
 257         retval = ext2fs_alloc_generic_bmap(&data->fake_fs,
 258                                 EXT2_ET_MAGIC_BLOCK_BITMAP64,
 259                                 EXT2FS_BMAP64_RBTREE,
 260                                 0, ~1ULL, ~1ULL,
 261                                 "undo block map", &data->written_block_map);
 262         if (retval)
 263                 return retval;
 264
 265         /* Allocate key block */
 266         retval = ext2fs_get_mem(data->tdb_data_size, &data->keyb);
 267         if (retval)
 268                 return retval;
 269         data->key_blk_num = data->first_key_blk;
 270
 271         /* Record block size */
 272         dbg_printf("Undo block size %llu\n", data->tdb_data_size);
 273         dbg_printf("Keys per block %llu\n", KEYS_PER_BLOCK(data));
 274         data->hdr.block_size = ext2fs_cpu_to_le32(data->tdb_data_size);
 275         io_channel_set_blksize(data->undo_file, data->tdb_data_size);
 276
 277         /* Ensure that we have space for header blocks */
 278         for (i = 0; i <= 2; i++) {
 279                 retval = io_channel_read_blk64(data->undo_file, i, 1,
 280                                                data->keyb);
 281                 if (retval)
 282                         memset(data->keyb, 0, data->tdb_data_size);
 283                 retval = io_channel_write_blk64(data->undo_file, i, 1,
 284                                                 data->keyb);
 285                 if (retval)
 286                         return retval;
 287                 retval = io_channel_flush(data->undo_file);
 288                 if (retval)
 289                         return retval;
 290         }
 291         memset(data->keyb, 0, data->tdb_data_size);
 292         return 0;
 293 }
 294
 295 static errcode_t undo_write_tdb(io_channel channel,
 296                                 unsigned long long block, int count)
 297
 298 {
 299         int size, sz;
 300         unsigned long long block_num, backing_blk_num;
 301         errcode_t retval = 0;
 302         ext2_loff_t offset;
 303         struct undo_private_data *data;
 304         unsigned char *read_ptr;
 305         unsigned long long end_block;
 306         unsigned long long data_size;
 307         void *data_ptr;
 308         struct undo_key *key;
 309         __u32 blk_crc;
 310
 311         data = (struct undo_private_data *) channel->private_data;
 312
 313         if (data->undo_file == NULL) {
 314                 /*
 315                  * Transaction database not initialized
 316                  */
 317                 return 0;
 318         }
 319
 320         if (count == 1)
 321                 size = channel->block_size;
 322         else {
 323                 if (count < 0)
 324                         size = -count;
 325                 else
 326                         size = count * channel->block_size;
 327         }
 328
 329         retval = undo_setup_tdb(data);
 330         if (retval)
 331                 return retval;
 332         /*
 333          * Data is stored in tdb database as blocks of tdb_data_size size
 334          * This helps in efficient lookup further.
 335          *
 336          * We divide the disk to blocks of tdb_data_size.
 337          */
 338         offset = (block * channel->block_size) + data->offset ;
 339         block_num = offset / data->tdb_data_size;
 340         end_block = (offset + size - 1) / data->tdb_data_size;
 341
 342         while (block_num <= end_block) {
 343                 __u32 keysz;
 344
 345                 /*
 346                  * Check if we have the record already
 347                  */
 348                 if (ext2fs_test_block_bitmap2(data->written_block_map,
 349                                                    block_num)) {
 350                         /* Try the next block */
 351                         block_num++;
 352                         continue;
 353                 }
 354                 ext2fs_mark_block_bitmap2(data->written_block_map, block_num);
 355
 356                 /*
 357                  * Read one block using the backing I/O manager
 358                  * The backing I/O manager block size may be
 359                  * different from the tdb_data_size.
 360                  * Also we need to recalcuate the block number with respect
 361                  * to the backing I/O manager.
 362                  */
 363                 offset = block_num * data->tdb_data_size;
 364                 backing_blk_num = (offset - data->offset) / channel->block_size;
 365
 366                 count = data->tdb_data_size +
 367                                 ((offset - data->offset) % channel->block_size);
 368                 retval = ext2fs_get_mem(count, &read_ptr);
 369                 if (retval) {
 370                         return retval;
 371                 }
 372
 373                 memset(read_ptr, 0, count);
 374                 actual_size = 0;
 375                 if ((count % channel->block_size) == 0)
 376                         sz = count / channel->block_size;
 377                 else
 378                         sz = -count;
 379                 retval = io_channel_read_blk64(data->real, backing_blk_num,
 380                                              sz, read_ptr);
 381                 if (retval) {
 382                         if (retval != EXT2_ET_SHORT_READ) {
 383                                 free(read_ptr);
 384                                 return retval;
 385                         }
 386                         /*
 387                          * short read so update the record size
 388                          * accordingly
 389                          */
 390                         data_size = actual_size;
 391                 } else {
 392                         data_size = data->tdb_data_size;
 393                 }
 394                 if (data_size == 0) {
 395                         free(read_ptr);
 396                         block_num++;
 397                         continue;
 398                 }
 399                 dbg_printf("Read %llu bytes from FS block %llu (blk=%llu cnt=%u)\n",
 400                        data_size, backing_blk_num, block, count);
 401                 if ((data_size % data->undo_file->block_size) == 0)
 402                         sz = data_size / data->undo_file->block_size;
 403                 else
 404                         sz = -actual_size;
 405                 data_ptr = read_ptr + ((offset - data->offset) %
 406                                        data->undo_file->block_size);
 407                 /* extend this key? */
 408                 if (data->keys_in_block) {
 409                         key = data->keyb->keys + data->keys_in_block - 1;
 410                         keysz = ext2fs_le32_to_cpu(key->size);
 411                 } else {
 412                         key = NULL;
 413                         keysz = 0;
 414                 }
 415                 if (key != NULL &&
 416                     ext2fs_le64_to_cpu(key->fsblk) +
 417                     ((keysz + data->tdb_data_size - 1) /
 418                      data->tdb_data_size) == backing_blk_num &&
 419                     E2UNDO_MAX_EXTENT_BLOCKS * data->tdb_data_size >
 420                     keysz + sz) {
 421                         blk_crc = ext2fs_le32_to_cpu(key->blk_crc);
 422                         blk_crc = ext2fs_crc32c_le(blk_crc,
 423                                                    (unsigned char *)data_ptr,
 424                                                    data_size);
 425                         key->blk_crc = ext2fs_cpu_to_le32(blk_crc);
 426                         key->size = ext2fs_cpu_to_le32(keysz + data_size);
 427                 } else {
 428                         data->num_keys++;
 429                         key = data->keyb->keys + data->keys_in_block;
 430                         data->keys_in_block++;
 431                         key->fsblk = ext2fs_cpu_to_le64(backing_blk_num);
 432                         blk_crc = ext2fs_crc32c_le(~0,
 433                                                    (unsigned char *)data_ptr,
 434                                                    data_size);
 435                         key->blk_crc = ext2fs_cpu_to_le32(blk_crc);
 436                         key->size = ext2fs_cpu_to_le32(data_size);
 437                 }
 438                 dbg_printf("Writing block %llu to offset %llu size %d key %zu\n",
 439                        block_num,
 440                        data->undo_blk_num,
 441                        sz, data->num_keys - 1);
 442                 retval = io_channel_write_blk64(data->undo_file,
 443                                         data->undo_blk_num, sz, data_ptr);
 444                 if (retval) {
 445                         free(read_ptr);
 446                         return retval;
 447                 }
 448                 data->undo_blk_num++;
 449                 free(read_ptr);
 450
 451                 /* Write out the key block */
 452                 retval = write_undo_indexes(data, 0);
 453                 if (retval)
 454                         return retval;
 455
 456                 /* Next block */
 457                 block_num++;
 458         }
 459
 460         return retval;
 461 }
 462
 463 static errcode_t undo_io_read_error(io_channel channel ATTR((unused)),
 464                                     unsigned long block ATTR((unused)),
 465                                     int count ATTR((unused)),
 466                                     void *data ATTR((unused)),
 467                                     size_t size ATTR((unused)),
 468                                     int actual,
 469                                     errcode_t error ATTR((unused)))
 470 {
 471         actual_size = actual;
 472         return error;
 473 }
 474
 475 static void undo_err_handler_init(io_channel channel)
 476 {
 477         channel->read_error = undo_io_read_error;
 478 }
 479
 480 static int check_filesystem(struct undo_header *hdr, io_channel undo_file,
 481                             unsigned int blocksize, blk64_t super_block,
 482                             io_channel channel)
 483 {
 484         struct ext2_super_block super, *sb;
 485         char *buf;
 486         __u32 sb_crc;
 487         errcode_t retval;
 488
 489         io_channel_set_blksize(channel, SUPERBLOCK_OFFSET);
 490         retval = io_channel_read_blk64(channel, 1, -SUPERBLOCK_SIZE, &super);
 491         if (retval)
 492                 return retval;
 493
 494         /*
 495          * Compare the FS and the undo file superblock so that we don't
 496          * append to something that doesn't match this FS.
 497          */
 498         retval = ext2fs_get_mem(blocksize, &buf);
 499         if (retval)
 500                 return retval;
 501         retval = io_channel_read_blk64(undo_file, super_block,
 502                                        -SUPERBLOCK_SIZE, buf);
 503         if (retval)
 504                 goto out;
 505         sb = (struct ext2_super_block *)buf;
 506         sb->s_magic = ~sb->s_magic;
 507         if (memcmp(&super, buf, sizeof(super))) {
 508                 retval = -1;
 509                 goto out;
 510         }
 511         sb_crc = ext2fs_crc32c_le(~0, (unsigned char *)buf, SUPERBLOCK_SIZE);
 512         if (ext2fs_le32_to_cpu(hdr->sb_crc) != sb_crc) {
 513                 retval = -1;
 514                 goto out;
 515         }
 516
 517 out:
 518         ext2fs_free_mem(&buf);
 519         return retval;
 520 }
 521
 522 /*
 523  * Try to re-open the undo file, so that we can resume where we left off.
 524  * That way, the user can pass the same undo file to various programs as
 525  * part of an FS upgrade instead of having to create multiple files and
 526  * then apply them in correct order.
 527  */
 528 static errcode_t try_reopen_undo_file(int undo_fd,
 529                                       struct undo_private_data *data)
 530 {
 531         struct undo_header hdr;
 532         struct undo_key *dkey;
 533         ext2fs_struct_stat statbuf;
 534         unsigned int blocksize, fs_blocksize;
 535         blk64_t super_block, lblk;
 536         size_t num_keys, keys_per_block, i;
 537         __u32 hdr_crc, key_crc;
 538         errcode_t retval;
 539
 540         /* Zero size already? */
 541         retval = ext2fs_fstat(undo_fd, &statbuf);
 542         if (retval)
 543                 goto bad_file;
 544         if (statbuf.st_size == 0)
 545                 goto out;
 546
 547         /* check the file header */
 548         retval = io_channel_read_blk64(data->undo_file, 0, -(int)sizeof(hdr),
 549                                        &hdr);
 550         if (retval)
 551                 goto bad_file;
 552
 553         if (memcmp(hdr.magic, E2UNDO_MAGIC,
 554                     sizeof(hdr.magic)))
 555                 goto bad_file;
 556         hdr_crc = ext2fs_crc32c_le(~0, (unsigned char *)&hdr,
 557                                    sizeof(struct undo_header) -
 558                                    sizeof(__u32));
 559         if (ext2fs_le32_to_cpu(hdr.header_crc) != hdr_crc)
 560                 goto bad_file;
 561         blocksize = ext2fs_le32_to_cpu(hdr.block_size);
 562         fs_blocksize = ext2fs_le32_to_cpu(hdr.fs_block_size);
 563         if (blocksize > E2UNDO_MAX_BLOCK_SIZE ||
 564             blocksize < E2UNDO_MIN_BLOCK_SIZE ||
 565             !blocksize || !fs_blocksize)
 566                 goto bad_file;
 567         super_block = ext2fs_le64_to_cpu(hdr.super_offset);
 568         num_keys = ext2fs_le64_to_cpu(hdr.num_keys);
 569         io_channel_set_blksize(data->undo_file, blocksize);
 570         if (hdr.f_compat || hdr.f_incompat || hdr.f_rocompat)
 571                 goto bad_file;
 572
 573         /* Superblock matches this FS? */
 574         if (check_filesystem(&hdr, data->undo_file, blocksize, super_block,
 575                              data->real) != 0) {
 576                 retval = EXT2_ET_UNDO_FILE_WRONG;
 577                 goto out;
 578         }
 579
 580         /* Try to set ourselves up */
 581         data->tdb_data_size = blocksize;
 582         retval = undo_setup_tdb(data);
 583         if (retval)
 584                 goto bad_file;
 585         data->num_keys = num_keys;
 586         data->super_blk_num = super_block;
 587         data->first_key_blk = ext2fs_le64_to_cpu(hdr.key_offset);
 588
 589         /* load the written block map */
 590         keys_per_block = KEYS_PER_BLOCK(data);
 591         lblk = data->first_key_blk;
 592         dbg_printf("nr_keys=%lu, kpb=%zu, blksz=%u\n",
 593                    num_keys, keys_per_block, blocksize);
 594         for (i = 0; i < num_keys; i += keys_per_block) {
 595                 size_t j, max_j;
 596                 __le32 crc;
 597
 598                 data->key_blk_num = lblk;
 599                 retval = io_channel_read_blk64(data->undo_file,
 600                                                lblk, 1, data->keyb);
 601                 if (retval)
 602                         goto bad_key_replay;
 603
 604                 /* check keys */
 605                 if (ext2fs_le32_to_cpu(data->keyb->magic) != KEYBLOCK_MAGIC) {
 606                         retval = EXT2_ET_UNDO_FILE_CORRUPT;
 607                         goto bad_key_replay;
 608                 }
 609                 crc = data->keyb->crc;
 610                 data->keyb->crc = 0;
 611                 key_crc = ext2fs_crc32c_le(~0, (unsigned char *)data->keyb,
 612                                            blocksize);
 613                 if (ext2fs_le32_to_cpu(crc) != key_crc) {
 614                         retval = EXT2_ET_UNDO_FILE_CORRUPT;
 615                         goto bad_key_replay;
 616                 }
 617
 618                 /* load keys from key block */
 619                 lblk++;
 620                 max_j = data->num_keys - i;
 621                 if (max_j > keys_per_block)
 622                         max_j = keys_per_block;
 623                 for (j = 0, dkey = data->keyb->keys;
 624                      j < max_j;
 625                      j++, dkey++) {
 626                         blk64_t fsblk = ext2fs_le64_to_cpu(dkey->fsblk);
 627                         blk64_t undo_blk = fsblk * fs_blocksize / blocksize;
 628                         size_t size = ext2fs_le32_to_cpu(dkey->size);
 629
 630                         ext2fs_mark_block_bitmap_range2(data->written_block_map,
 631                                          undo_blk,
 632                                         (size + blocksize - 1) / blocksize);
 633                         lblk += (size + blocksize - 1) / blocksize;
 634                         data->undo_blk_num = lblk;
 635                         data->keys_in_block = j + 1;
 636                 }
 637         }
 638         dbg_printf("Reopen undo, keyblk=%llu undoblk=%llu nrkeys=%zu kib=%zu\n",
 639                    data->key_blk_num, data->undo_blk_num, data->num_keys,
 640                    data->keys_in_block);
 641
 642         data->hdr.state = hdr.state & ~E2UNDO_STATE_FINISHED;
 643         data->hdr.f_compat = hdr.f_compat;
 644         data->hdr.f_incompat = hdr.f_incompat;
 645         data->hdr.f_rocompat = hdr.f_rocompat;
 646         return retval;
 647
 648 bad_key_replay:
 649         data->key_blk_num = data->undo_blk_num = 0;
 650         data->keys_in_block = 0;
 651         ext2fs_free_mem(&data->keyb);
 652         ext2fs_free_generic_bitmap(data->written_block_map);
 653         data->tdb_written = 0;
 654         goto out;
 655 bad_file:
 656         retval = EXT2_ET_UNDO_FILE_CORRUPT;
 657 out:
 658         return retval;
 659 }
 660
 661 static void undo_atexit(void *p)
 662 {
 663         struct undo_private_data *data = p;
 664         errcode_t err;
 665
 666         err = write_undo_indexes(data, 1);
 667         io_channel_close(data->undo_file);
 668
 669         com_err(data->tdb_file, err, "while force-closing undo file");
 670 }
 671
 672 static errcode_t undo_open(const char *name, int flags, io_channel *channel)
 673 {
 674         io_channel      io = NULL;
 675         struct undo_private_data *data = NULL;
 676         int             undo_fd = -1;
 677         errcode_t       retval;
 678
 679         if (name == 0)
 680                 return EXT2_ET_BAD_DEVICE_NAME;
 681         retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io);
 682         if (retval)
 683                 goto cleanup;
 684         memset(io, 0, sizeof(struct struct_io_channel));
 685         io->magic = EXT2_ET_MAGIC_IO_CHANNEL;
 686         retval = ext2fs_get_mem(sizeof(struct undo_private_data), &data);
 687         if (retval)
 688                 goto cleanup;
 689
 690         io->manager = undo_io_manager;
 691         retval = ext2fs_get_mem(strlen(name)+1, &io->name);
 692         if (retval)
 693                 goto cleanup;
 694
 695         strcpy(io->name, name);
 696         io->private_data = data;
 697         io->block_size = 1024;
 698         io->read_error = 0;
 699         io->write_error = 0;
 700         io->refcount = 1;
 701
 702         memset(data, 0, sizeof(struct undo_private_data));
 703         data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL;
 704         data->super_blk_num = 1;
 705         data->first_key_blk = 2;
 706         data->undo_blk_num = 3;
 707
 708         if (undo_io_backing_manager) {
 709                 retval = undo_io_backing_manager->open(name, flags,
 710                                                        &data->real);
 711                 if (retval)
 712                         goto cleanup;
 713
 714                 data->tdb_file = strdup(tdb_file);
 715                 if (data->tdb_file == NULL)
 716                         goto cleanup;
 717                 undo_fd = ext2fs_open_file(data->tdb_file, O_RDWR | O_CREAT,
 718                                            0600);
 719                 if (undo_fd < 0)
 720                         goto cleanup;
 721
 722                 retval = undo_io_backing_manager->open(data->tdb_file,
 723                                                        IO_FLAG_RW,
 724                                                        &data->undo_file);
 725                 if (retval)
 726                         goto cleanup;
 727         } else {
 728                 data->real = NULL;
 729                 data->undo_file = NULL;
 730         }
 731
 732         if (data->real)
 733                 io->flags = (io->flags & ~CHANNEL_FLAGS_DISCARD_ZEROES) |
 734                             (data->real->flags & CHANNEL_FLAGS_DISCARD_ZEROES);
 735
 736         /*
 737          * setup err handler for read so that we know
 738          * when the backing manager fails do short read
 739          */
 740         if (data->real)
 741                 undo_err_handler_init(data->real);
 742
 743         if (data->undo_file) {
 744                 retval = try_reopen_undo_file(undo_fd, data);
 745                 if (retval)
 746                         goto cleanup;
 747         }
 748         retval = ext2fs_add_exit_fn(undo_atexit, data);
 749         if (retval)
 750                 goto cleanup;
 751
 752         *channel = io;
 753         if (undo_fd >= 0)
 754                 close(undo_fd);
 755         return retval;
 756
 757 cleanup:
 758         ext2fs_remove_exit_fn(undo_atexit, data);
 759         if (undo_fd >= 0)
 760                 close(undo_fd);
 761         if (data && data->undo_file)
 762                 io_channel_close(data->undo_file);
 763         if (data && data->tdb_file)
 764                 free(data->tdb_file);
 765         if (data && data->real)
 766                 io_channel_close(data->real);
 767         if (data)
 768                 ext2fs_free_mem(&data);
 769         if (io)
 770                 ext2fs_free_mem(&io);
 771         return retval;
 772 }
 773
 774 static errcode_t undo_close(io_channel channel)
 775 {
 776         struct undo_private_data *data;
 777         errcode_t       err, retval = 0;
 778
 779         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 780         data = (struct undo_private_data *) channel->private_data;
 781         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 782
 783         if (--channel->refcount > 0)
 784                 return 0;
 785         /* Before closing write the file system identity */
 786         if (!getenv("UNDO_IO_SIMULATE_UNFINISHED"))
 787                 data->hdr.state = ext2fs_cpu_to_le32(E2UNDO_STATE_FINISHED);
 788         err = write_undo_indexes(data, 1);
 789         ext2fs_remove_exit_fn(undo_atexit, data);
 790         if (data->real)
 791                 retval = io_channel_close(data->real);
 792         if (data->tdb_file)
 793                 free(data->tdb_file);
 794         if (data->undo_file)
 795                 io_channel_close(data->undo_file);
 796         ext2fs_free_mem(&data->keyb);
 797         if (data->written_block_map)
 798                 ext2fs_free_generic_bitmap(data->written_block_map);
 799         ext2fs_free_mem(&channel->private_data);
 800         if (channel->name)
 801                 ext2fs_free_mem(&channel->name);
 802         ext2fs_free_mem(&channel);
 803
 804         if (err)
 805                 return err;
 806         return retval;
 807 }
 808
 809 static errcode_t undo_set_blksize(io_channel channel, int blksize)
 810 {
 811         struct undo_private_data *data;
 812         errcode_t               retval = 0;
 813
 814         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 815         data = (struct undo_private_data *) channel->private_data;
 816         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 817
 818         if (blksize > E2UNDO_MAX_BLOCK_SIZE || blksize < E2UNDO_MIN_BLOCK_SIZE)
 819                 return EXT2_ET_INVALID_ARGUMENT;
 820
 821         if (data->real)
 822                 retval = io_channel_set_blksize(data->real, blksize);
 823         /*
 824          * Set the block size used for tdb
 825          */
 826         if (!data->tdb_data_size || !data->tdb_written)
 827                 data->tdb_data_size = blksize;
 828         channel->block_size = blksize;
 829         return retval;
 830 }
 831
 832 static errcode_t undo_read_blk64(io_channel channel, unsigned long long block,
 833                                int count, void *buf)
 834 {
 835         errcode_t       retval = 0;
 836         struct undo_private_data *data;
 837
 838         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 839         data = (struct undo_private_data *) channel->private_data;
 840         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 841
 842         if (data->real)
 843                 retval = io_channel_read_blk64(data->real, block, count, buf);
 844
 845         return retval;
 846 }
 847
 848 static errcode_t undo_read_blk(io_channel channel, unsigned long block,
 849                                int count, void *buf)
 850 {
 851         return undo_read_blk64(channel, block, count, buf);
 852 }
 853
 854 static errcode_t undo_write_blk64(io_channel channel, unsigned long long block,
 855                                 int count, const void *buf)
 856 {
 857         struct undo_private_data *data;
 858         errcode_t       retval = 0;
 859
 860         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 861         data = (struct undo_private_data *) channel->private_data;
 862         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 863         /*
 864          * First write the existing content into database
 865          */
 866         retval = undo_write_tdb(channel, block, count);
 867         if (retval)
 868                  return retval;
 869         if (data->real)
 870                 retval = io_channel_write_blk64(data->real, block, count, buf);
 871
 872         return retval;
 873 }
 874
 875 static errcode_t undo_write_blk(io_channel channel, unsigned long block,
 876                                 int count, const void *buf)
 877 {
 878         return undo_write_blk64(channel, block, count, buf);
 879 }
 880
 881 static errcode_t undo_write_byte(io_channel channel, unsigned long offset,
 882                                  int size, const void *buf)
 883 {
 884         struct undo_private_data *data;
 885         errcode_t       retval = 0;
 886         ext2_loff_t     location;
 887         unsigned long blk_num, count;;
 888
 889         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 890         data = (struct undo_private_data *) channel->private_data;
 891         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 892
 893         location = offset + data->offset;
 894         blk_num = location/channel->block_size;
 895         /*
 896          * the size specified may spread across multiple blocks
 897          * also make sure we account for the fact that block start
 898          * offset for tdb is different from the backing I/O manager
 899          * due to possible different block size
 900          */
 901         count = (size + (location % channel->block_size) +
 902                         channel->block_size  -1)/channel->block_size;
 903         retval = undo_write_tdb(channel, blk_num, count);
 904         if (retval)
 905                 return retval;
 906         if (data->real && data->real->manager->write_byte)
 907                 retval = io_channel_write_byte(data->real, offset, size, buf);
 908
 909         return retval;
 910 }
 911
 912 static errcode_t undo_discard(io_channel channel, unsigned long long block,
 913                               unsigned long long count)
 914 {
 915         struct undo_private_data *data;
 916         errcode_t       retval = 0;
 917         int icount;
 918
 919         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 920         data = (struct undo_private_data *) channel->private_data;
 921         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 922
 923         if (count > INT_MAX)
 924                 return EXT2_ET_UNIMPLEMENTED;
 925         icount = count;
 926
 927         /*
 928          * First write the existing content into database
 929          */
 930         retval = undo_write_tdb(channel, block, icount);
 931         if (retval)
 932                 return retval;
 933         if (data->real)
 934                 retval = io_channel_discard(data->real, block, count);
 935
 936         return retval;
 937 }
 938
 939 static errcode_t undo_zeroout(io_channel channel, unsigned long long block,
 940                               unsigned long long count)
 941 {
 942         struct undo_private_data *data;
 943         errcode_t       retval = 0;
 944         int icount;
 945
 946         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 947         data = (struct undo_private_data *) channel->private_data;
 948         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 949
 950         if (count > INT_MAX)
 951                 return EXT2_ET_UNIMPLEMENTED;
 952         icount = count;
 953
 954         /*
 955          * First write the existing content into database
 956          */
 957         retval = undo_write_tdb(channel, block, icount);
 958         if (retval)
 959                 return retval;
 960         if (data->real)
 961                 retval = io_channel_zeroout(data->real, block, count);
 962
 963         return retval;
 964 }
 965
 966 static errcode_t undo_cache_readahead(io_channel channel,
 967                                       unsigned long long block,
 968                                       unsigned long long count)
 969 {
 970         struct undo_private_data *data;
 971         errcode_t       retval = 0;
 972
 973         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 974         data = (struct undo_private_data *) channel->private_data;
 975         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 976
 977         if (data->real)
 978                 retval = io_channel_cache_readahead(data->real, block, count);
 979
 980         return retval;
 981 }
 982
 983 /*
 984  * Flush data buffers to disk.
 985  */
 986 static errcode_t undo_flush(io_channel channel)
 987 {
 988         errcode_t       retval = 0;
 989         struct undo_private_data *data;
 990
 991         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 992         data = (struct undo_private_data *) channel->private_data;
 993         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 994
 995         if (data->real)
 996                 retval = io_channel_flush(data->real);
 997
 998         return retval;
 999 }
1000
1001 static errcode_t undo_set_option(io_channel channel, const char *option,
1002                                  const char *arg)
1003 {
1004         errcode_t       retval = 0;
1005         struct undo_private_data *data;
1006         unsigned long tmp;
1007         char *end;
1008
1009         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1010         data = (struct undo_private_data *) channel->private_data;
1011         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1012
1013         if (!strcmp(option, "tdb_data_size")) {
1014                 if (!arg)
1015                         return EXT2_ET_INVALID_ARGUMENT;
1016
1017                 tmp = strtoul(arg, &end, 0);
1018                 if (*end)
1019                         return EXT2_ET_INVALID_ARGUMENT;
1020                 if (tmp > E2UNDO_MAX_BLOCK_SIZE || tmp < E2UNDO_MIN_BLOCK_SIZE)
1021                         return EXT2_ET_INVALID_ARGUMENT;
1022                 if (!data->tdb_data_size || !data->tdb_written) {
1023                         data->tdb_written = -1;
1024                         data->tdb_data_size = tmp;
1025                 }
1026                 return 0;
1027         }
1028         /*
1029          * Need to support offset option to work with
1030          * Unix I/O manager
1031          */
1032         if (data->real && data->real->manager->set_option) {
1033                 retval = data->real->manager->set_option(data->real,
1034                                                         option, arg);
1035         }
1036         if (!retval && !strcmp(option, "offset")) {
1037                 if (!arg)
1038                         return EXT2_ET_INVALID_ARGUMENT;
1039
1040                 tmp = strtoul(arg, &end, 0);
1041                 if (*end)
1042                         return EXT2_ET_INVALID_ARGUMENT;
1043                 data->offset = tmp;
1044         }
1045         return retval;
1046 }
1047
1048 static errcode_t undo_get_stats(io_channel channel, io_stats *stats)
1049 {
1050         errcode_t       retval = 0;
1051         struct undo_private_data *data;
1052
1053         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1054         data = (struct undo_private_data *) channel->private_data;
1055         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1056
1057         if (data->real)
1058                 retval = (data->real->manager->get_stats)(data->real, stats);
1059
1060         return retval;
1061 }
1062
1063 static struct struct_io_manager struct_undo_manager = {
1064         .magic          = EXT2_ET_MAGIC_IO_MANAGER,
1065         .name           = "Undo I/O Manager",
1066         .open           = undo_open,
1067         .close          = undo_close,
1068         .set_blksize    = undo_set_blksize,
1069         .read_blk       = undo_read_blk,
1070         .write_blk      = undo_write_blk,
1071         .flush          = undo_flush,
1072         .write_byte     = undo_write_byte,
1073         .set_option     = undo_set_option,
1074         .get_stats      = undo_get_stats,
1075         .read_blk64     = undo_read_blk64,
1076         .write_blk64    = undo_write_blk64,
1077         .discard        = undo_discard,
1078         .zeroout        = undo_zeroout,
1079         .cache_readahead        = undo_cache_readahead,
1080 };
1081
1082 io_manager undo_io_manager = &struct_undo_manager;