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                         flush = 1;
 197                 }
 198         }
 199
 200         /* Prepare superblock for write */
 201         channel = data->real;
 202         block_size = channel->block_size;
 203
 204         io_channel_set_blksize(channel, SUPERBLOCK_OFFSET);
 205         retval = io_channel_read_blk64(channel, 1, -SUPERBLOCK_SIZE, &super);
 206         if (retval)
 207                 goto err_out;
 208         sb_crc = ext2fs_crc32c_le(~0, (unsigned char *)&super, SUPERBLOCK_SIZE);
 209         super.s_magic = ~super.s_magic;
 210
 211         /* Write the undo header to disk. */
 212         memcpy(data->hdr.magic, E2UNDO_MAGIC, sizeof(data->hdr.magic));
 213         data->hdr.num_keys = ext2fs_cpu_to_le64(data->num_keys);
 214         data->hdr.super_offset = ext2fs_cpu_to_le64(data->super_blk_num);
 215         data->hdr.key_offset = ext2fs_cpu_to_le64(data->first_key_blk);
 216         data->hdr.fs_block_size = ext2fs_cpu_to_le32(block_size);
 217         data->hdr.sb_crc = ext2fs_cpu_to_le32(sb_crc);
 218         hdr_crc = ext2fs_crc32c_le(~0, (unsigned char *)&data->hdr,
 219                                    sizeof(data->hdr) -
 220                                    sizeof(data->hdr.header_crc));
 221         data->hdr.header_crc = ext2fs_cpu_to_le32(hdr_crc);
 222         retval = io_channel_write_blk64(data->undo_file, 0,
 223                                         -(int)sizeof(data->hdr),
 224                                         &data->hdr);
 225         if (retval)
 226                 goto err_out;
 227
 228         /*
 229          * Record the entire superblock (in FS byte order) so that we can't
 230          * apply e2undo files to the wrong FS or out of order.
 231          */
 232         dbg_printf("Writing superblock to block %llu\n", data->super_blk_num);
 233         retval = io_channel_write_blk64(data->undo_file, data->super_blk_num,
 234                                         -SUPERBLOCK_SIZE, &super);
 235         if (retval)
 236                 goto err_out;
 237
 238         if (flush)
 239                 retval = io_channel_flush(data->undo_file);
 240 err_out:
 241         io_channel_set_blksize(channel, block_size);
 242         return retval;
 243 }
 244
 245 static errcode_t undo_setup_tdb(struct undo_private_data *data)
 246 {
 247         int i;
 248         errcode_t retval;
 249
 250         if (data->tdb_written == 1)
 251                 return 0;
 252
 253         data->tdb_written = 1;
 254
 255         /* Make a bitmap to track what we've written */
 256         memset(&data->fake_fs, 0, sizeof(data->fake_fs));
 257         data->fake_fs.blocksize = data->tdb_data_size;
 258         retval = ext2fs_alloc_generic_bmap(&data->fake_fs,
 259                                 EXT2_ET_MAGIC_BLOCK_BITMAP64,
 260                                 EXT2FS_BMAP64_RBTREE,
 261                                 0, ~1ULL, ~1ULL,
 262                                 "undo block map", &data->written_block_map);
 263         if (retval)
 264                 return retval;
 265
 266         /* Allocate key block */
 267         retval = ext2fs_get_mem(data->tdb_data_size, &data->keyb);
 268         if (retval)
 269                 return retval;
 270         data->key_blk_num = data->first_key_blk;
 271
 272         /* Record block size */
 273         dbg_printf("Undo block size %llu\n", data->tdb_data_size);
 274         dbg_printf("Keys per block %llu\n", KEYS_PER_BLOCK(data));
 275         data->hdr.block_size = ext2fs_cpu_to_le32(data->tdb_data_size);
 276         io_channel_set_blksize(data->undo_file, data->tdb_data_size);
 277
 278         /* Ensure that we have space for header blocks */
 279         for (i = 0; i <= 2; i++) {
 280                 retval = io_channel_read_blk64(data->undo_file, i, 1,
 281                                                data->keyb);
 282                 if (retval)
 283                         memset(data->keyb, 0, data->tdb_data_size);
 284                 retval = io_channel_write_blk64(data->undo_file, i, 1,
 285                                                 data->keyb);
 286                 if (retval)
 287                         return retval;
 288                 retval = io_channel_flush(data->undo_file);
 289                 if (retval)
 290                         return retval;
 291         }
 292         memset(data->keyb, 0, data->tdb_data_size);
 293         return 0;
 294 }
 295
 296 static errcode_t undo_write_tdb(io_channel channel,
 297                                 unsigned long long block, int count)
 298
 299 {
 300         int size, sz;
 301         unsigned long long block_num, backing_blk_num;
 302         errcode_t retval = 0;
 303         ext2_loff_t offset;
 304         struct undo_private_data *data;
 305         unsigned char *read_ptr;
 306         unsigned long long end_block;
 307         unsigned long long data_size;
 308         void *data_ptr;
 309         struct undo_key *key;
 310         __u32 blk_crc;
 311
 312         data = (struct undo_private_data *) channel->private_data;
 313
 314         if (data->undo_file == NULL) {
 315                 /*
 316                  * Transaction database not initialized
 317                  */
 318                 return 0;
 319         }
 320
 321         if (count == 1)
 322                 size = channel->block_size;
 323         else {
 324                 if (count < 0)
 325                         size = -count;
 326                 else
 327                         size = count * channel->block_size;
 328         }
 329
 330         retval = undo_setup_tdb(data);
 331         if (retval)
 332                 return retval;
 333         /*
 334          * Data is stored in tdb database as blocks of tdb_data_size size
 335          * This helps in efficient lookup further.
 336          *
 337          * We divide the disk to blocks of tdb_data_size.
 338          */
 339         offset = (block * channel->block_size) + data->offset ;
 340         block_num = offset / data->tdb_data_size;
 341         end_block = (offset + size - 1) / data->tdb_data_size;
 342
 343         while (block_num <= end_block) {
 344                 __u32 keysz;
 345
 346                 /*
 347                  * Check if we have the record already
 348                  */
 349                 if (ext2fs_test_block_bitmap2(data->written_block_map,
 350                                                    block_num)) {
 351                         /* Try the next block */
 352                         block_num++;
 353                         continue;
 354                 }
 355                 ext2fs_mark_block_bitmap2(data->written_block_map, block_num);
 356
 357                 /*
 358                  * Read one block using the backing I/O manager
 359                  * The backing I/O manager block size may be
 360                  * different from the tdb_data_size.
 361                  * Also we need to recalcuate the block number with respect
 362                  * to the backing I/O manager.
 363                  */
 364                 offset = block_num * data->tdb_data_size;
 365                 backing_blk_num = (offset - data->offset) / channel->block_size;
 366
 367                 count = data->tdb_data_size +
 368                                 ((offset - data->offset) % channel->block_size);
 369                 retval = ext2fs_get_mem(count, &read_ptr);
 370                 if (retval) {
 371                         return retval;
 372                 }
 373
 374                 memset(read_ptr, 0, count);
 375                 actual_size = 0;
 376                 if ((count % channel->block_size) == 0)
 377                         sz = count / channel->block_size;
 378                 else
 379                         sz = -count;
 380                 retval = io_channel_read_blk64(data->real, backing_blk_num,
 381                                              sz, read_ptr);
 382                 if (retval) {
 383                         if (retval != EXT2_ET_SHORT_READ) {
 384                                 free(read_ptr);
 385                                 return retval;
 386                         }
 387                         /*
 388                          * short read so update the record size
 389                          * accordingly
 390                          */
 391                         data_size = actual_size;
 392                 } else {
 393                         data_size = data->tdb_data_size;
 394                 }
 395                 if (data_size == 0) {
 396                         free(read_ptr);
 397                         block_num++;
 398                         continue;
 399                 }
 400                 dbg_printf("Read %llu bytes from FS block %llu (blk=%llu cnt=%u)\n",
 401                        data_size, backing_blk_num, block, count);
 402                 if ((data_size % data->undo_file->block_size) == 0)
 403                         sz = data_size / data->undo_file->block_size;
 404                 else
 405                         sz = -actual_size;
 406                 data_ptr = read_ptr + ((offset - data->offset) %
 407                                        data->undo_file->block_size);
 408                 /* extend this key? */
 409                 if (data->keys_in_block) {
 410                         key = data->keyb->keys + data->keys_in_block - 1;
 411                         keysz = ext2fs_le32_to_cpu(key->size);
 412                 } else {
 413                         key = NULL;
 414                         keysz = 0;
 415                 }
 416                 if (key != NULL &&
 417                     ext2fs_le64_to_cpu(key->fsblk) +
 418                     ((keysz + data->tdb_data_size - 1) /
 419                      data->tdb_data_size) == backing_blk_num &&
 420                     E2UNDO_MAX_EXTENT_BLOCKS * data->tdb_data_size >
 421                     keysz + sz) {
 422                         blk_crc = ext2fs_le32_to_cpu(key->blk_crc);
 423                         blk_crc = ext2fs_crc32c_le(blk_crc,
 424                                                    (unsigned char *)data_ptr,
 425                                                    data_size);
 426                         key->blk_crc = ext2fs_cpu_to_le32(blk_crc);
 427                         key->size = ext2fs_cpu_to_le32(keysz + data_size);
 428                 } else {
 429                         data->num_keys++;
 430                         key = data->keyb->keys + data->keys_in_block;
 431                         data->keys_in_block++;
 432                         key->fsblk = ext2fs_cpu_to_le64(backing_blk_num);
 433                         blk_crc = ext2fs_crc32c_le(~0,
 434                                                    (unsigned char *)data_ptr,
 435                                                    data_size);
 436                         key->blk_crc = ext2fs_cpu_to_le32(blk_crc);
 437                         key->size = ext2fs_cpu_to_le32(data_size);
 438                 }
 439                 dbg_printf("Writing block %llu to offset %llu size %d key %zu\n",
 440                        block_num,
 441                        data->undo_blk_num,
 442                        sz, data->num_keys - 1);
 443                 retval = io_channel_write_blk64(data->undo_file,
 444                                         data->undo_blk_num, sz, data_ptr);
 445                 if (retval) {
 446                         free(read_ptr);
 447                         return retval;
 448                 }
 449                 data->undo_blk_num++;
 450                 free(read_ptr);
 451
 452                 /* Write out the key block */
 453                 retval = write_undo_indexes(data, 0);
 454                 if (retval)
 455                         return retval;
 456
 457                 /* Next block */
 458                 block_num++;
 459         }
 460
 461         return retval;
 462 }
 463
 464 static errcode_t undo_io_read_error(io_channel channel ATTR((unused)),
 465                                     unsigned long block ATTR((unused)),
 466                                     int count ATTR((unused)),
 467                                     void *data ATTR((unused)),
 468                                     size_t size ATTR((unused)),
 469                                     int actual,
 470                                     errcode_t error ATTR((unused)))
 471 {
 472         actual_size = actual;
 473         return error;
 474 }
 475
 476 static void undo_err_handler_init(io_channel channel)
 477 {
 478         channel->read_error = undo_io_read_error;
 479 }
 480
 481 static int check_filesystem(struct undo_header *hdr, io_channel undo_file,
 482                             unsigned int blocksize, blk64_t super_block,
 483                             io_channel channel)
 484 {
 485         struct ext2_super_block super, *sb;
 486         char *buf;
 487         __u32 sb_crc;
 488         errcode_t retval;
 489
 490         io_channel_set_blksize(channel, SUPERBLOCK_OFFSET);
 491         retval = io_channel_read_blk64(channel, 1, -SUPERBLOCK_SIZE, &super);
 492         if (retval)
 493                 return retval;
 494
 495         /*
 496          * Compare the FS and the undo file superblock so that we don't
 497          * append to something that doesn't match this FS.
 498          */
 499         retval = ext2fs_get_mem(blocksize, &buf);
 500         if (retval)
 501                 return retval;
 502         retval = io_channel_read_blk64(undo_file, super_block,
 503                                        -SUPERBLOCK_SIZE, buf);
 504         if (retval)
 505                 goto out;
 506         sb = (struct ext2_super_block *)buf;
 507         sb->s_magic = ~sb->s_magic;
 508         if (memcmp(&super, buf, sizeof(super))) {
 509                 retval = -1;
 510                 goto out;
 511         }
 512         sb_crc = ext2fs_crc32c_le(~0, (unsigned char *)buf, SUPERBLOCK_SIZE);
 513         if (ext2fs_le32_to_cpu(hdr->sb_crc) != sb_crc) {
 514                 retval = -1;
 515                 goto out;
 516         }
 517
 518 out:
 519         ext2fs_free_mem(&buf);
 520         return retval;
 521 }
 522
 523 /*
 524  * Try to re-open the undo file, so that we can resume where we left off.
 525  * That way, the user can pass the same undo file to various programs as
 526  * part of an FS upgrade instead of having to create multiple files and
 527  * then apply them in correct order.
 528  */
 529 static errcode_t try_reopen_undo_file(int undo_fd,
 530                                       struct undo_private_data *data)
 531 {
 532         struct undo_header hdr;
 533         struct undo_key *dkey;
 534         ext2fs_struct_stat statbuf;
 535         unsigned int blocksize, fs_blocksize;
 536         blk64_t super_block, lblk;
 537         size_t num_keys, keys_per_block, i;
 538         __u32 hdr_crc, key_crc;
 539         errcode_t retval;
 540
 541         /* Zero size already? */
 542         retval = ext2fs_fstat(undo_fd, &statbuf);
 543         if (retval)
 544                 goto bad_file;
 545         if (statbuf.st_size == 0)
 546                 goto out;
 547
 548         /* check the file header */
 549         retval = io_channel_read_blk64(data->undo_file, 0, -(int)sizeof(hdr),
 550                                        &hdr);
 551         if (retval)
 552                 goto bad_file;
 553
 554         if (memcmp(hdr.magic, E2UNDO_MAGIC,
 555                     sizeof(hdr.magic)))
 556                 goto bad_file;
 557         hdr_crc = ext2fs_crc32c_le(~0, (unsigned char *)&hdr,
 558                                    sizeof(struct undo_header) -
 559                                    sizeof(__u32));
 560         if (ext2fs_le32_to_cpu(hdr.header_crc) != hdr_crc)
 561                 goto bad_file;
 562         blocksize = ext2fs_le32_to_cpu(hdr.block_size);
 563         fs_blocksize = ext2fs_le32_to_cpu(hdr.fs_block_size);
 564         if (blocksize > E2UNDO_MAX_BLOCK_SIZE ||
 565             blocksize < E2UNDO_MIN_BLOCK_SIZE ||
 566             !blocksize || !fs_blocksize)
 567                 goto bad_file;
 568         super_block = ext2fs_le64_to_cpu(hdr.super_offset);
 569         num_keys = ext2fs_le64_to_cpu(hdr.num_keys);
 570         io_channel_set_blksize(data->undo_file, blocksize);
 571         if (hdr.f_compat || hdr.f_incompat || hdr.f_rocompat)
 572                 goto bad_file;
 573
 574         /* Superblock matches this FS? */
 575         if (check_filesystem(&hdr, data->undo_file, blocksize, super_block,
 576                              data->real) != 0) {
 577                 retval = EXT2_ET_UNDO_FILE_WRONG;
 578                 goto out;
 579         }
 580
 581         /* Try to set ourselves up */
 582         data->tdb_data_size = blocksize;
 583         retval = undo_setup_tdb(data);
 584         if (retval)
 585                 goto bad_file;
 586         data->num_keys = num_keys;
 587         data->super_blk_num = super_block;
 588         data->first_key_blk = ext2fs_le64_to_cpu(hdr.key_offset);
 589
 590         /* load the written block map */
 591         keys_per_block = KEYS_PER_BLOCK(data);
 592         lblk = data->first_key_blk;
 593         dbg_printf("nr_keys=%lu, kpb=%zu, blksz=%u\n",
 594                    num_keys, keys_per_block, blocksize);
 595         for (i = 0; i < num_keys; i += keys_per_block) {
 596                 size_t j, max_j;
 597                 __le32 crc;
 598
 599                 data->key_blk_num = lblk;
 600                 retval = io_channel_read_blk64(data->undo_file,
 601                                                lblk, 1, data->keyb);
 602                 if (retval)
 603                         goto bad_key_replay;
 604
 605                 /* check keys */
 606                 if (ext2fs_le32_to_cpu(data->keyb->magic) != KEYBLOCK_MAGIC) {
 607                         retval = EXT2_ET_UNDO_FILE_CORRUPT;
 608                         goto bad_key_replay;
 609                 }
 610                 crc = data->keyb->crc;
 611                 data->keyb->crc = 0;
 612                 key_crc = ext2fs_crc32c_le(~0, (unsigned char *)data->keyb,
 613                                            blocksize);
 614                 if (ext2fs_le32_to_cpu(crc) != key_crc) {
 615                         retval = EXT2_ET_UNDO_FILE_CORRUPT;
 616                         goto bad_key_replay;
 617                 }
 618
 619                 /* load keys from key block */
 620                 lblk++;
 621                 max_j = data->num_keys - i;
 622                 if (max_j > keys_per_block)
 623                         max_j = keys_per_block;
 624                 for (j = 0, dkey = data->keyb->keys;
 625                      j < max_j;
 626                      j++, dkey++) {
 627                         blk64_t fsblk = ext2fs_le64_to_cpu(dkey->fsblk);
 628                         blk64_t undo_blk = fsblk * fs_blocksize / blocksize;
 629                         size_t size = ext2fs_le32_to_cpu(dkey->size);
 630
 631                         ext2fs_mark_block_bitmap_range2(data->written_block_map,
 632                                          undo_blk,
 633                                         (size + blocksize - 1) / blocksize);
 634                         lblk += (size + blocksize - 1) / blocksize;
 635                         data->undo_blk_num = lblk;
 636                         data->keys_in_block = j + 1;
 637                 }
 638         }
 639         dbg_printf("Reopen undo, keyblk=%llu undoblk=%llu nrkeys=%zu kib=%zu\n",
 640                    data->key_blk_num, data->undo_blk_num, data->num_keys,
 641                    data->keys_in_block);
 642
 643         data->hdr.state = hdr.state & ~E2UNDO_STATE_FINISHED;
 644         data->hdr.f_compat = hdr.f_compat;
 645         data->hdr.f_incompat = hdr.f_incompat;
 646         data->hdr.f_rocompat = hdr.f_rocompat;
 647         return retval;
 648
 649 bad_key_replay:
 650         data->key_blk_num = data->undo_blk_num = 0;
 651         data->keys_in_block = 0;
 652         ext2fs_free_mem(&data->keyb);
 653         ext2fs_free_generic_bitmap(data->written_block_map);
 654         data->tdb_written = 0;
 655         goto out;
 656 bad_file:
 657         retval = EXT2_ET_UNDO_FILE_CORRUPT;
 658 out:
 659         return retval;
 660 }
 661
 662 static void undo_atexit(void *p)
 663 {
 664         struct undo_private_data *data = p;
 665         errcode_t err;
 666
 667         err = write_undo_indexes(data, 1);
 668         io_channel_close(data->undo_file);
 669
 670         com_err(data->tdb_file, err, "while force-closing undo file");
 671 }
 672
 673 static errcode_t undo_open(const char *name, int flags, io_channel *channel)
 674 {
 675         io_channel      io = NULL;
 676         struct undo_private_data *data = NULL;
 677         int             undo_fd = -1;
 678         errcode_t       retval;
 679
 680         if (name == 0)
 681                 return EXT2_ET_BAD_DEVICE_NAME;
 682         retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io);
 683         if (retval)
 684                 goto cleanup;
 685         memset(io, 0, sizeof(struct struct_io_channel));
 686         io->magic = EXT2_ET_MAGIC_IO_CHANNEL;
 687         retval = ext2fs_get_mem(sizeof(struct undo_private_data), &data);
 688         if (retval)
 689                 goto cleanup;
 690
 691         io->manager = undo_io_manager;
 692         retval = ext2fs_get_mem(strlen(name)+1, &io->name);
 693         if (retval)
 694                 goto cleanup;
 695
 696         strcpy(io->name, name);
 697         io->private_data = data;
 698         io->block_size = 1024;
 699         io->read_error = 0;
 700         io->write_error = 0;
 701         io->refcount = 1;
 702
 703         memset(data, 0, sizeof(struct undo_private_data));
 704         data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL;
 705         data->super_blk_num = 1;
 706         data->first_key_blk = 2;
 707         data->undo_blk_num = 3;
 708
 709         if (undo_io_backing_manager) {
 710                 retval = undo_io_backing_manager->open(name, flags,
 711                                                        &data->real);
 712                 if (retval)
 713                         goto cleanup;
 714
 715                 data->tdb_file = strdup(tdb_file);
 716                 if (data->tdb_file == NULL)
 717                         goto cleanup;
 718                 undo_fd = ext2fs_open_file(data->tdb_file, O_RDWR | O_CREAT,
 719                                            0600);
 720                 if (undo_fd < 0)
 721                         goto cleanup;
 722
 723                 retval = undo_io_backing_manager->open(data->tdb_file,
 724                                                        IO_FLAG_RW,
 725                                                        &data->undo_file);
 726                 if (retval)
 727                         goto cleanup;
 728         } else {
 729                 data->real = NULL;
 730                 data->undo_file = NULL;
 731         }
 732
 733         if (data->real)
 734                 io->flags = (io->flags & ~CHANNEL_FLAGS_DISCARD_ZEROES) |
 735                             (data->real->flags & CHANNEL_FLAGS_DISCARD_ZEROES);
 736
 737         /*
 738          * setup err handler for read so that we know
 739          * when the backing manager fails do short read
 740          */
 741         if (data->real)
 742                 undo_err_handler_init(data->real);
 743
 744         if (data->undo_file) {
 745                 retval = try_reopen_undo_file(undo_fd, data);
 746                 if (retval)
 747                         goto cleanup;
 748         }
 749         retval = ext2fs_add_exit_fn(undo_atexit, data);
 750         if (retval)
 751                 goto cleanup;
 752
 753         *channel = io;
 754         if (undo_fd >= 0)
 755                 close(undo_fd);
 756         return retval;
 757
 758 cleanup:
 759         ext2fs_remove_exit_fn(undo_atexit, data);
 760         if (undo_fd >= 0)
 761                 close(undo_fd);
 762         if (data && data->undo_file)
 763                 io_channel_close(data->undo_file);
 764         if (data && data->tdb_file)
 765                 free(data->tdb_file);
 766         if (data && data->real)
 767                 io_channel_close(data->real);
 768         if (data)
 769                 ext2fs_free_mem(&data);
 770         if (io)
 771                 ext2fs_free_mem(&io);
 772         return retval;
 773 }
 774
 775 static errcode_t undo_close(io_channel channel)
 776 {
 777         struct undo_private_data *data;
 778         errcode_t       err, retval = 0;
 779
 780         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 781         data = (struct undo_private_data *) channel->private_data;
 782         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 783
 784         if (--channel->refcount > 0)
 785                 return 0;
 786         /* Before closing write the file system identity */
 787         if (!getenv("UNDO_IO_SIMULATE_UNFINISHED"))
 788                 data->hdr.state = ext2fs_cpu_to_le32(E2UNDO_STATE_FINISHED);
 789         err = write_undo_indexes(data, 1);
 790         ext2fs_remove_exit_fn(undo_atexit, data);
 791         if (data->real)
 792                 retval = io_channel_close(data->real);
 793         if (data->tdb_file)
 794                 free(data->tdb_file);
 795         if (data->undo_file)
 796                 io_channel_close(data->undo_file);
 797         ext2fs_free_mem(&data->keyb);
 798         if (data->written_block_map)
 799                 ext2fs_free_generic_bitmap(data->written_block_map);
 800         ext2fs_free_mem(&channel->private_data);
 801         if (channel->name)
 802                 ext2fs_free_mem(&channel->name);
 803         ext2fs_free_mem(&channel);
 804
 805         if (err)
 806                 return err;
 807         return retval;
 808 }
 809
 810 static errcode_t undo_set_blksize(io_channel channel, int blksize)
 811 {
 812         struct undo_private_data *data;
 813         errcode_t               retval = 0;
 814
 815         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 816         data = (struct undo_private_data *) channel->private_data;
 817         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 818
 819         if (blksize > E2UNDO_MAX_BLOCK_SIZE || blksize < E2UNDO_MIN_BLOCK_SIZE)
 820                 return EXT2_ET_INVALID_ARGUMENT;
 821
 822         if (data->real)
 823                 retval = io_channel_set_blksize(data->real, blksize);
 824         /*
 825          * Set the block size used for tdb
 826          */
 827         if (!data->tdb_data_size || !data->tdb_written)
 828                 data->tdb_data_size = blksize;
 829         channel->block_size = blksize;
 830         return retval;
 831 }
 832
 833 static errcode_t undo_read_blk64(io_channel channel, unsigned long long block,
 834                                int count, void *buf)
 835 {
 836         errcode_t       retval = 0;
 837         struct undo_private_data *data;
 838
 839         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 840         data = (struct undo_private_data *) channel->private_data;
 841         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 842
 843         if (data->real)
 844                 retval = io_channel_read_blk64(data->real, block, count, buf);
 845
 846         return retval;
 847 }
 848
 849 static errcode_t undo_read_blk(io_channel channel, unsigned long block,
 850                                int count, void *buf)
 851 {
 852         return undo_read_blk64(channel, block, count, buf);
 853 }
 854
 855 static errcode_t undo_write_blk64(io_channel channel, unsigned long long block,
 856                                 int count, const void *buf)
 857 {
 858         struct undo_private_data *data;
 859         errcode_t       retval = 0;
 860
 861         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 862         data = (struct undo_private_data *) channel->private_data;
 863         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 864         /*
 865          * First write the existing content into database
 866          */
 867         retval = undo_write_tdb(channel, block, count);
 868         if (retval)
 869                  return retval;
 870         if (data->real)
 871                 retval = io_channel_write_blk64(data->real, block, count, buf);
 872
 873         return retval;
 874 }
 875
 876 static errcode_t undo_write_blk(io_channel channel, unsigned long block,
 877                                 int count, const void *buf)
 878 {
 879         return undo_write_blk64(channel, block, count, buf);
 880 }
 881
 882 static errcode_t undo_write_byte(io_channel channel, unsigned long offset,
 883                                  int size, const void *buf)
 884 {
 885         struct undo_private_data *data;
 886         errcode_t       retval = 0;
 887         ext2_loff_t     location;
 888         unsigned long blk_num, count;;
 889
 890         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 891         data = (struct undo_private_data *) channel->private_data;
 892         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 893
 894         location = offset + data->offset;
 895         blk_num = location/channel->block_size;
 896         /*
 897          * the size specified may spread across multiple blocks
 898          * also make sure we account for the fact that block start
 899          * offset for tdb is different from the backing I/O manager
 900          * due to possible different block size
 901          */
 902         count = (size + (location % channel->block_size) +
 903                         channel->block_size  -1)/channel->block_size;
 904         retval = undo_write_tdb(channel, blk_num, count);
 905         if (retval)
 906                 return retval;
 907         if (data->real && data->real->manager->write_byte)
 908                 retval = io_channel_write_byte(data->real, offset, size, buf);
 909
 910         return retval;
 911 }
 912
 913 static errcode_t undo_discard(io_channel channel, unsigned long long block,
 914                               unsigned long long count)
 915 {
 916         struct undo_private_data *data;
 917         errcode_t       retval = 0;
 918         int icount;
 919
 920         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 921         data = (struct undo_private_data *) channel->private_data;
 922         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 923
 924         if (count > INT_MAX)
 925                 return EXT2_ET_UNIMPLEMENTED;
 926         icount = count;
 927
 928         /*
 929          * First write the existing content into database
 930          */
 931         retval = undo_write_tdb(channel, block, icount);
 932         if (retval)
 933                 return retval;
 934         if (data->real)
 935                 retval = io_channel_discard(data->real, block, count);
 936
 937         return retval;
 938 }
 939
 940 static errcode_t undo_zeroout(io_channel channel, unsigned long long block,
 941                               unsigned long long count)
 942 {
 943         struct undo_private_data *data;
 944         errcode_t       retval = 0;
 945         int icount;
 946
 947         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 948         data = (struct undo_private_data *) channel->private_data;
 949         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 950
 951         if (count > INT_MAX)
 952                 return EXT2_ET_UNIMPLEMENTED;
 953         icount = count;
 954
 955         /*
 956          * First write the existing content into database
 957          */
 958         retval = undo_write_tdb(channel, block, icount);
 959         if (retval)
 960                 return retval;
 961         if (data->real)
 962                 retval = io_channel_zeroout(data->real, block, count);
 963
 964         return retval;
 965 }
 966
 967 static errcode_t undo_cache_readahead(io_channel channel,
 968                                       unsigned long long block,
 969                                       unsigned long long count)
 970 {
 971         struct undo_private_data *data;
 972         errcode_t       retval = 0;
 973
 974         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 975         data = (struct undo_private_data *) channel->private_data;
 976         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 977
 978         if (data->real)
 979                 retval = io_channel_cache_readahead(data->real, block, count);
 980
 981         return retval;
 982 }
 983
 984 /*
 985  * Flush data buffers to disk.
 986  */
 987 static errcode_t undo_flush(io_channel channel)
 988 {
 989         errcode_t       retval = 0;
 990         struct undo_private_data *data;
 991
 992         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
 993         data = (struct undo_private_data *) channel->private_data;
 994         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
 995
 996         if (data->real)
 997                 retval = io_channel_flush(data->real);
 998
 999         return retval;
1000 }
1001
1002 static errcode_t undo_set_option(io_channel channel, const char *option,
1003                                  const char *arg)
1004 {
1005         errcode_t       retval = 0;
1006         struct undo_private_data *data;
1007         unsigned long tmp;
1008         char *end;
1009
1010         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1011         data = (struct undo_private_data *) channel->private_data;
1012         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1013
1014         if (!strcmp(option, "tdb_data_size")) {
1015                 if (!arg)
1016                         return EXT2_ET_INVALID_ARGUMENT;
1017
1018                 tmp = strtoul(arg, &end, 0);
1019                 if (*end)
1020                         return EXT2_ET_INVALID_ARGUMENT;
1021                 if (tmp > E2UNDO_MAX_BLOCK_SIZE || tmp < E2UNDO_MIN_BLOCK_SIZE)
1022                         return EXT2_ET_INVALID_ARGUMENT;
1023                 if (!data->tdb_data_size || !data->tdb_written) {
1024                         data->tdb_written = -1;
1025                         data->tdb_data_size = tmp;
1026                 }
1027                 return 0;
1028         }
1029         /*
1030          * Need to support offset option to work with
1031          * Unix I/O manager
1032          */
1033         if (data->real && data->real->manager->set_option) {
1034                 retval = data->real->manager->set_option(data->real,
1035                                                         option, arg);
1036         }
1037         if (!retval && !strcmp(option, "offset")) {
1038                 if (!arg)
1039                         return EXT2_ET_INVALID_ARGUMENT;
1040
1041                 tmp = strtoul(arg, &end, 0);
1042                 if (*end)
1043                         return EXT2_ET_INVALID_ARGUMENT;
1044                 data->offset = tmp;
1045         }
1046         return retval;
1047 }
1048
1049 static errcode_t undo_get_stats(io_channel channel, io_stats *stats)
1050 {
1051         errcode_t       retval = 0;
1052         struct undo_private_data *data;
1053
1054         EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1055         data = (struct undo_private_data *) channel->private_data;
1056         EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1057
1058         if (data->real)
1059                 retval = (data->real->manager->get_stats)(data->real, stats);
1060
1061         return retval;
1062 }
1063
1064 static struct struct_io_manager struct_undo_manager = {
1065         .magic          = EXT2_ET_MAGIC_IO_MANAGER,
1066         .name           = "Undo I/O Manager",
1067         .open           = undo_open,
1068         .close          = undo_close,
1069         .set_blksize    = undo_set_blksize,
1070         .read_blk       = undo_read_blk,
1071         .write_blk      = undo_write_blk,
1072         .flush          = undo_flush,
1073         .write_byte     = undo_write_byte,
1074         .set_option     = undo_set_option,
1075         .get_stats      = undo_get_stats,
1076         .read_blk64     = undo_read_blk64,
1077         .write_blk64    = undo_write_blk64,
1078         .discard        = undo_discard,
1079         .zeroout        = undo_zeroout,
1080         .cache_readahead        = undo_cache_readahead,
1081 };
1082
1083 io_manager undo_io_manager = &struct_undo_manager;