| 1 | /* Copyright (C) 2006 MySQL AB & MySQL Finland AB & TCX DataKonsult AB |
| 2 | |
| 3 | This program is free software; you can redistribute it and/or modify |
| 4 | it under the terms of the GNU General Public License as published by |
| 5 | the Free Software Foundation; version 2 of the License. |
| 6 | |
| 7 | This program is distributed in the hope that it will be useful, |
| 8 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 9 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 10 | GNU General Public License for more details. |
| 11 | |
| 12 | You should have received a copy of the GNU General Public License |
| 13 | along with this program; if not, write to the Free Software |
| 14 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111-1301 USA */ |
| 15 | |
| 16 | /* |
| 17 | Functions to handle space-packed-records and blobs |
| 18 | |
| 19 | A row may be stored in one or more linked blocks. |
| 20 | The block size is between MARIA_MIN_BLOCK_LENGTH and MARIA_MAX_BLOCK_LENGTH. |
| 21 | Each block is aligned on MARIA_DYN_ALIGN_SIZE. |
| 22 | The reson for the max block size is to not have too many different types |
| 23 | of blocks. For the differnet block types, look at _ma_get_block_info() |
| 24 | */ |
| 25 | |
| 26 | #include "maria_def.h" |
| 27 | |
| 28 | static my_bool write_dynamic_record(MARIA_HA *info,const uchar *record, |
| 29 | ulong reclength); |
| 30 | static int _ma_find_writepos(MARIA_HA *info,ulong reclength,my_off_t *filepos, |
| 31 | ulong *length); |
| 32 | static my_bool update_dynamic_record(MARIA_HA *info, MARIA_RECORD_POS filepos, |
| 33 | uchar *record, ulong reclength); |
| 34 | static my_bool delete_dynamic_record(MARIA_HA *info,MARIA_RECORD_POS filepos, |
| 35 | uint second_read); |
| 36 | static my_bool _ma_cmp_buffer(File file, const uchar *buff, my_off_t filepos, |
| 37 | uint length); |
| 38 | |
| 39 | /* Interface function from MARIA_HA */ |
| 40 | |
| 41 | #ifdef HAVE_MMAP |
| 42 | |
| 43 | /* |
| 44 | Create mmaped area for MARIA handler |
| 45 | |
| 46 | SYNOPSIS |
| 47 | _ma_dynmap_file() |
| 48 | info MARIA handler |
| 49 | |
| 50 | RETURN |
| 51 | 0 ok |
| 52 | 1 error. |
| 53 | */ |
| 54 | |
| 55 | my_bool _ma_dynmap_file(MARIA_HA *info, my_off_t size) |
| 56 | { |
| 57 | DBUG_ENTER("_ma_dynmap_file" ); |
| 58 | if (size > (my_off_t) (~((size_t) 0)) - MEMMAP_EXTRA_MARGIN) |
| 59 | { |
| 60 | DBUG_PRINT("warning" , ("File is too large for mmap" )); |
| 61 | DBUG_RETURN(1); |
| 62 | } |
| 63 | /* |
| 64 | Ingo wonders if it is good to use MAP_NORESERVE. From the Linux man page: |
| 65 | MAP_NORESERVE |
| 66 | Do not reserve swap space for this mapping. When swap space is |
| 67 | reserved, one has the guarantee that it is possible to modify the |
| 68 | mapping. When swap space is not reserved one might get SIGSEGV |
| 69 | upon a write if no physical memory is available. |
| 70 | */ |
| 71 | info->s->file_map= (uchar*) |
| 72 | my_mmap(0, (size_t)(size + MEMMAP_EXTRA_MARGIN), |
| 73 | info->s->mode==O_RDONLY ? PROT_READ : |
| 74 | PROT_READ | PROT_WRITE, |
| 75 | MAP_SHARED | MAP_NORESERVE, |
| 76 | info->dfile.file, 0L); |
| 77 | if (info->s->file_map == (uchar*) MAP_FAILED) |
| 78 | { |
| 79 | info->s->file_map= NULL; |
| 80 | DBUG_RETURN(1); |
| 81 | } |
| 82 | #if defined(HAVE_MADVISE) |
| 83 | madvise((char*) info->s->file_map, size, MADV_RANDOM); |
| 84 | #endif |
| 85 | info->s->mmaped_length= size; |
| 86 | DBUG_RETURN(0); |
| 87 | } |
| 88 | |
| 89 | |
| 90 | /* |
| 91 | Resize mmaped area for MARIA handler |
| 92 | |
| 93 | SYNOPSIS |
| 94 | _ma_remap_file() |
| 95 | info MARIA handler |
| 96 | |
| 97 | RETURN |
| 98 | */ |
| 99 | |
| 100 | void _ma_remap_file(MARIA_HA *info, my_off_t size) |
| 101 | { |
| 102 | if (info->s->file_map) |
| 103 | { |
| 104 | my_munmap((char*) info->s->file_map, |
| 105 | (size_t) info->s->mmaped_length + MEMMAP_EXTRA_MARGIN); |
| 106 | _ma_dynmap_file(info, size); |
| 107 | } |
| 108 | } |
| 109 | #endif |
| 110 | |
| 111 | |
| 112 | /* |
| 113 | Read bytes from MySAM handler, using mmap or pread |
| 114 | |
| 115 | SYNOPSIS |
| 116 | _ma_mmap_pread() |
| 117 | info MARIA handler |
| 118 | Buffer Input buffer |
| 119 | Count Count of bytes for read |
| 120 | offset Start position |
| 121 | MyFlags |
| 122 | |
| 123 | RETURN |
| 124 | 0 ok |
| 125 | */ |
| 126 | |
| 127 | size_t _ma_mmap_pread(MARIA_HA *info, uchar *Buffer, |
| 128 | size_t Count, my_off_t offset, myf MyFlags) |
| 129 | { |
| 130 | DBUG_PRINT("info" , ("maria_read with mmap %d\n" , info->dfile.file)); |
| 131 | if (info->s->lock_key_trees) |
| 132 | mysql_rwlock_rdlock(&info->s->mmap_lock); |
| 133 | |
| 134 | /* |
| 135 | The following test may fail in the following cases: |
| 136 | - We failed to remap a memory area (fragmented memory?) |
| 137 | - This thread has done some writes, but not yet extended the |
| 138 | memory mapped area. |
| 139 | */ |
| 140 | |
| 141 | if (info->s->mmaped_length >= offset + Count) |
| 142 | { |
| 143 | memcpy(Buffer, info->s->file_map + offset, Count); |
| 144 | if (info->s->lock_key_trees) |
| 145 | mysql_rwlock_unlock(&info->s->mmap_lock); |
| 146 | return 0; |
| 147 | } |
| 148 | else |
| 149 | { |
| 150 | if (info->s->lock_key_trees) |
| 151 | mysql_rwlock_unlock(&info->s->mmap_lock); |
| 152 | return mysql_file_pread(info->dfile.file, Buffer, Count, offset, MyFlags); |
| 153 | } |
| 154 | } |
| 155 | |
| 156 | |
| 157 | /* wrapper for my_pread in case if mmap isn't used */ |
| 158 | |
| 159 | size_t _ma_nommap_pread(MARIA_HA *info, uchar *Buffer, |
| 160 | size_t Count, my_off_t offset, myf MyFlags) |
| 161 | { |
| 162 | return mysql_file_pread(info->dfile.file, Buffer, Count, offset, MyFlags); |
| 163 | } |
| 164 | |
| 165 | |
| 166 | /* |
| 167 | Write bytes to MySAM handler, using mmap or pwrite |
| 168 | |
| 169 | SYNOPSIS |
| 170 | _ma_mmap_pwrite() |
| 171 | info MARIA handler |
| 172 | Buffer Output buffer |
| 173 | Count Count of bytes for write |
| 174 | offset Start position |
| 175 | MyFlags |
| 176 | |
| 177 | RETURN |
| 178 | 0 ok |
| 179 | !=0 error. In this case return error from pwrite |
| 180 | */ |
| 181 | |
| 182 | size_t _ma_mmap_pwrite(MARIA_HA *info, const uchar *Buffer, |
| 183 | size_t Count, my_off_t offset, myf MyFlags) |
| 184 | { |
| 185 | DBUG_PRINT("info" , ("maria_write with mmap %d\n" , info->dfile.file)); |
| 186 | if (info->s->lock_key_trees) |
| 187 | mysql_rwlock_rdlock(&info->s->mmap_lock); |
| 188 | |
| 189 | /* |
| 190 | The following test may fail in the following cases: |
| 191 | - We failed to remap a memory area (fragmented memory?) |
| 192 | - This thread has done some writes, but not yet extended the |
| 193 | memory mapped area. |
| 194 | */ |
| 195 | |
| 196 | if (info->s->mmaped_length >= offset + Count) |
| 197 | { |
| 198 | memcpy(info->s->file_map + offset, Buffer, Count); |
| 199 | if (info->s->lock_key_trees) |
| 200 | mysql_rwlock_unlock(&info->s->mmap_lock); |
| 201 | return 0; |
| 202 | } |
| 203 | else |
| 204 | { |
| 205 | info->s->nonmmaped_inserts++; |
| 206 | if (info->s->lock_key_trees) |
| 207 | mysql_rwlock_unlock(&info->s->mmap_lock); |
| 208 | return my_pwrite(info->dfile.file, Buffer, Count, offset, MyFlags); |
| 209 | } |
| 210 | |
| 211 | } |
| 212 | |
| 213 | |
| 214 | /* wrapper for my_pwrite in case if mmap isn't used */ |
| 215 | |
| 216 | size_t _ma_nommap_pwrite(MARIA_HA *info, const uchar *Buffer, |
| 217 | size_t Count, my_off_t offset, myf MyFlags) |
| 218 | { |
| 219 | return my_pwrite(info->dfile.file, Buffer, Count, offset, MyFlags); |
| 220 | } |
| 221 | |
| 222 | |
| 223 | my_bool _ma_write_dynamic_record(MARIA_HA *info, const uchar *record) |
| 224 | { |
| 225 | ulong reclength= _ma_rec_pack(info,info->rec_buff + MARIA_REC_BUFF_OFFSET, |
| 226 | record); |
| 227 | return (write_dynamic_record(info,info->rec_buff + MARIA_REC_BUFF_OFFSET, |
| 228 | reclength)); |
| 229 | } |
| 230 | |
| 231 | my_bool _ma_update_dynamic_record(MARIA_HA *info, MARIA_RECORD_POS pos, |
| 232 | const uchar *oldrec __attribute__ ((unused)), |
| 233 | const uchar *record) |
| 234 | { |
| 235 | uint length= _ma_rec_pack(info, info->rec_buff + MARIA_REC_BUFF_OFFSET, |
| 236 | record); |
| 237 | return (update_dynamic_record(info, pos, |
| 238 | info->rec_buff + MARIA_REC_BUFF_OFFSET, |
| 239 | length)); |
| 240 | } |
| 241 | |
| 242 | |
| 243 | my_bool _ma_write_blob_record(MARIA_HA *info, const uchar *record) |
| 244 | { |
| 245 | uchar *rec_buff; |
| 246 | int error; |
| 247 | ulong reclength,reclength2,; |
| 248 | |
| 249 | extra= (ALIGN_SIZE(MARIA_MAX_DYN_BLOCK_HEADER)+MARIA_SPLIT_LENGTH+ |
| 250 | MARIA_DYN_DELETE_BLOCK_HEADER+1); |
| 251 | reclength= (info->s->base.pack_reclength + |
| 252 | _ma_calc_total_blob_length(info,record)+ extra); |
| 253 | if (!(rec_buff=(uchar*) my_safe_alloca(reclength))) |
| 254 | { |
| 255 | my_errno= HA_ERR_OUT_OF_MEM; /* purecov: inspected */ |
| 256 | return(1); |
| 257 | } |
| 258 | reclength2= _ma_rec_pack(info, |
| 259 | rec_buff+ALIGN_SIZE(MARIA_MAX_DYN_BLOCK_HEADER), |
| 260 | record); |
| 261 | DBUG_PRINT("info" ,("reclength: %lu reclength2: %lu" , |
| 262 | reclength, reclength2)); |
| 263 | DBUG_ASSERT(reclength2 <= reclength); |
| 264 | error= write_dynamic_record(info, |
| 265 | rec_buff+ALIGN_SIZE(MARIA_MAX_DYN_BLOCK_HEADER), |
| 266 | reclength2); |
| 267 | my_safe_afree(rec_buff, reclength); |
| 268 | return(error != 0); |
| 269 | } |
| 270 | |
| 271 | |
| 272 | my_bool _ma_update_blob_record(MARIA_HA *info, MARIA_RECORD_POS pos, |
| 273 | const uchar *oldrec __attribute__ ((unused)), |
| 274 | const uchar *record) |
| 275 | { |
| 276 | uchar *rec_buff; |
| 277 | int error; |
| 278 | ulong reclength,reclength2,; |
| 279 | |
| 280 | extra= (ALIGN_SIZE(MARIA_MAX_DYN_BLOCK_HEADER)+MARIA_SPLIT_LENGTH+ |
| 281 | MARIA_DYN_DELETE_BLOCK_HEADER); |
| 282 | reclength= (info->s->base.pack_reclength+ |
| 283 | _ma_calc_total_blob_length(info,record)+ extra); |
| 284 | #ifdef NOT_USED /* We now support big rows */ |
| 285 | if (reclength > MARIA_DYN_MAX_ROW_LENGTH) |
| 286 | { |
| 287 | my_errno=HA_ERR_TO_BIG_ROW; |
| 288 | return 1; |
| 289 | } |
| 290 | #endif |
| 291 | if (!(rec_buff=(uchar*) my_safe_alloca(reclength))) |
| 292 | { |
| 293 | my_errno= HA_ERR_OUT_OF_MEM; /* purecov: inspected */ |
| 294 | return(1); |
| 295 | } |
| 296 | reclength2= _ma_rec_pack(info,rec_buff+ALIGN_SIZE(MARIA_MAX_DYN_BLOCK_HEADER), |
| 297 | record); |
| 298 | DBUG_ASSERT(reclength2 <= reclength); |
| 299 | error=update_dynamic_record(info,pos, |
| 300 | rec_buff+ALIGN_SIZE(MARIA_MAX_DYN_BLOCK_HEADER), |
| 301 | reclength2); |
| 302 | my_safe_afree(rec_buff, reclength); |
| 303 | return(error != 0); |
| 304 | } |
| 305 | |
| 306 | |
| 307 | my_bool _ma_delete_dynamic_record(MARIA_HA *info, |
| 308 | const uchar *record __attribute__ ((unused))) |
| 309 | { |
| 310 | return delete_dynamic_record(info, info->cur_row.lastpos, 0); |
| 311 | } |
| 312 | |
| 313 | |
| 314 | /** |
| 315 | Write record to data-file. |
| 316 | |
| 317 | @todo it's cheating: it casts "const uchar*" to uchar*. |
| 318 | */ |
| 319 | |
| 320 | static my_bool write_dynamic_record(MARIA_HA *info, const uchar *record, |
| 321 | ulong reclength) |
| 322 | { |
| 323 | int flag; |
| 324 | ulong length; |
| 325 | my_off_t filepos; |
| 326 | DBUG_ENTER("write_dynamic_record" ); |
| 327 | |
| 328 | flag=0; |
| 329 | |
| 330 | /* |
| 331 | Check if we have enough room for the new record. |
| 332 | First we do simplified check to make usual case faster. |
| 333 | Then we do more precise check for the space left. |
| 334 | Though it still is not absolutely precise, as |
| 335 | we always use MARIA_MAX_DYN_BLOCK_HEADER while it can be |
| 336 | less in the most of the cases. |
| 337 | */ |
| 338 | |
| 339 | if (unlikely(info->s->base.max_data_file_length - |
| 340 | info->state->data_file_length < |
| 341 | reclength + MARIA_MAX_DYN_BLOCK_HEADER)) |
| 342 | { |
| 343 | if (info->s->base.max_data_file_length - info->state->data_file_length + |
| 344 | info->state->empty - info->state->del * MARIA_MAX_DYN_BLOCK_HEADER < |
| 345 | reclength + MARIA_MAX_DYN_BLOCK_HEADER) |
| 346 | { |
| 347 | my_errno=HA_ERR_RECORD_FILE_FULL; |
| 348 | DBUG_RETURN(1); |
| 349 | } |
| 350 | } |
| 351 | |
| 352 | do |
| 353 | { |
| 354 | if (_ma_find_writepos(info,reclength,&filepos,&length)) |
| 355 | goto err; |
| 356 | if (_ma_write_part_record(info,filepos,length, |
| 357 | (info->append_insert_at_end ? |
| 358 | HA_OFFSET_ERROR : info->s->state.dellink), |
| 359 | (uchar**) &record,&reclength,&flag)) |
| 360 | goto err; |
| 361 | } while (reclength); |
| 362 | |
| 363 | DBUG_RETURN(0); |
| 364 | err: |
| 365 | DBUG_RETURN(1); |
| 366 | } |
| 367 | |
| 368 | |
| 369 | /* Get a block for data ; The given data-area must be used !! */ |
| 370 | |
| 371 | static int _ma_find_writepos(MARIA_HA *info, |
| 372 | ulong reclength, /* record length */ |
| 373 | my_off_t *filepos, /* Return file pos */ |
| 374 | ulong *length) /* length of block at filepos */ |
| 375 | { |
| 376 | MARIA_BLOCK_INFO block_info; |
| 377 | ulong tmp; |
| 378 | DBUG_ENTER("_ma_find_writepos" ); |
| 379 | |
| 380 | if (info->s->state.dellink != HA_OFFSET_ERROR && |
| 381 | !info->append_insert_at_end) |
| 382 | { |
| 383 | /* Deleted blocks exists; Get last used block */ |
| 384 | *filepos=info->s->state.dellink; |
| 385 | block_info.second_read=0; |
| 386 | info->rec_cache.seek_not_done=1; |
| 387 | if (!(_ma_get_block_info(info, &block_info, info->dfile.file, |
| 388 | info->s->state.dellink) & |
| 389 | BLOCK_DELETED)) |
| 390 | { |
| 391 | DBUG_PRINT("error" ,("Delete link crashed" )); |
| 392 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 393 | DBUG_RETURN(-1); |
| 394 | } |
| 395 | info->s->state.dellink=block_info.next_filepos; |
| 396 | info->state->del--; |
| 397 | info->state->empty-= block_info.block_len; |
| 398 | *length= block_info.block_len; |
| 399 | } |
| 400 | else |
| 401 | { |
| 402 | /* No deleted blocks; Allocate a new block */ |
| 403 | *filepos=info->state->data_file_length; |
| 404 | if ((tmp= reclength + 3 + MY_TEST(reclength >= (65520 - 3))) < |
| 405 | info->s->base.min_block_length) |
| 406 | tmp= info->s->base.min_block_length; |
| 407 | else |
| 408 | tmp= ((tmp+MARIA_DYN_ALIGN_SIZE-1) & |
| 409 | (~ (ulong) (MARIA_DYN_ALIGN_SIZE-1))); |
| 410 | if (info->state->data_file_length > |
| 411 | (info->s->base.max_data_file_length - tmp)) |
| 412 | { |
| 413 | my_errno=HA_ERR_RECORD_FILE_FULL; |
| 414 | DBUG_RETURN(-1); |
| 415 | } |
| 416 | if (tmp > MARIA_MAX_BLOCK_LENGTH) |
| 417 | tmp=MARIA_MAX_BLOCK_LENGTH; |
| 418 | *length= tmp; |
| 419 | info->state->data_file_length+= tmp; |
| 420 | info->s->state.split++; |
| 421 | info->update|=HA_STATE_WRITE_AT_END; |
| 422 | } |
| 423 | DBUG_RETURN(0); |
| 424 | } /* _ma_find_writepos */ |
| 425 | |
| 426 | |
| 427 | |
| 428 | /* |
| 429 | Unlink a deleted block from the deleted list. |
| 430 | This block will be combined with the preceding or next block to form |
| 431 | a big block. |
| 432 | */ |
| 433 | |
| 434 | static my_bool unlink_deleted_block(MARIA_HA *info, |
| 435 | MARIA_BLOCK_INFO *block_info) |
| 436 | { |
| 437 | DBUG_ENTER("unlink_deleted_block" ); |
| 438 | if (block_info->filepos == info->s->state.dellink) |
| 439 | { |
| 440 | /* First deleted block; We can just use this ! */ |
| 441 | info->s->state.dellink=block_info->next_filepos; |
| 442 | } |
| 443 | else |
| 444 | { |
| 445 | MARIA_BLOCK_INFO tmp; |
| 446 | tmp.second_read=0; |
| 447 | /* Unlink block from the previous block */ |
| 448 | if (!(_ma_get_block_info(info, &tmp, info->dfile.file, |
| 449 | block_info->prev_filepos) |
| 450 | & BLOCK_DELETED)) |
| 451 | DBUG_RETURN(1); /* Something is wrong */ |
| 452 | mi_sizestore(tmp.header+4,block_info->next_filepos); |
| 453 | if (info->s->file_write(info, tmp.header+4,8, |
| 454 | block_info->prev_filepos+4, MYF(MY_NABP))) |
| 455 | DBUG_RETURN(1); |
| 456 | /* Unlink block from next block */ |
| 457 | if (block_info->next_filepos != HA_OFFSET_ERROR) |
| 458 | { |
| 459 | if (!(_ma_get_block_info(info, &tmp, info->dfile.file, |
| 460 | block_info->next_filepos) |
| 461 | & BLOCK_DELETED)) |
| 462 | DBUG_RETURN(1); /* Something is wrong */ |
| 463 | mi_sizestore(tmp.header+12,block_info->prev_filepos); |
| 464 | if (info->s->file_write(info, tmp.header+12,8, |
| 465 | block_info->next_filepos+12, |
| 466 | MYF(MY_NABP))) |
| 467 | DBUG_RETURN(1); |
| 468 | } |
| 469 | } |
| 470 | /* We now have one less deleted block */ |
| 471 | info->state->del--; |
| 472 | info->state->empty-= block_info->block_len; |
| 473 | info->s->state.split--; |
| 474 | |
| 475 | /* |
| 476 | If this was a block that we where accessing through table scan |
| 477 | (maria_rrnd() or maria_scan(), then ensure that we skip over this block |
| 478 | when doing next maria_rrnd() or maria_scan(). |
| 479 | */ |
| 480 | if (info->cur_row.nextpos == block_info->filepos) |
| 481 | info->cur_row.nextpos+= block_info->block_len; |
| 482 | DBUG_RETURN(0); |
| 483 | } |
| 484 | |
| 485 | |
| 486 | /* |
| 487 | Add a backward link to delete block |
| 488 | |
| 489 | SYNOPSIS |
| 490 | update_backward_delete_link() |
| 491 | info MARIA handler |
| 492 | delete_block Position to delete block to update. |
| 493 | If this is 'HA_OFFSET_ERROR', nothing will be done |
| 494 | filepos Position to block that 'delete_block' should point to |
| 495 | |
| 496 | RETURN |
| 497 | 0 ok |
| 498 | 1 error. In this case my_error is set. |
| 499 | */ |
| 500 | |
| 501 | static my_bool update_backward_delete_link(MARIA_HA *info, |
| 502 | my_off_t delete_block, |
| 503 | MARIA_RECORD_POS filepos) |
| 504 | { |
| 505 | MARIA_BLOCK_INFO block_info; |
| 506 | DBUG_ENTER("update_backward_delete_link" ); |
| 507 | |
| 508 | if (delete_block != HA_OFFSET_ERROR) |
| 509 | { |
| 510 | block_info.second_read=0; |
| 511 | if (_ma_get_block_info(info, &block_info, info->dfile.file, delete_block) |
| 512 | & BLOCK_DELETED) |
| 513 | { |
| 514 | uchar buff[8]; |
| 515 | mi_sizestore(buff,filepos); |
| 516 | if (info->s->file_write(info,buff, 8, delete_block+12, MYF(MY_NABP))) |
| 517 | DBUG_RETURN(1); /* Error on write */ |
| 518 | } |
| 519 | else |
| 520 | { |
| 521 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 522 | DBUG_RETURN(1); /* Wrong delete link */ |
| 523 | } |
| 524 | } |
| 525 | DBUG_RETURN(0); |
| 526 | } |
| 527 | |
| 528 | /* Delete datarecord from database */ |
| 529 | /* info->rec_cache.seek_not_done is updated in cmp_record */ |
| 530 | |
| 531 | static my_bool delete_dynamic_record(MARIA_HA *info, MARIA_RECORD_POS filepos, |
| 532 | uint second_read) |
| 533 | { |
| 534 | uint length,b_type; |
| 535 | MARIA_BLOCK_INFO block_info,del_block; |
| 536 | int error; |
| 537 | my_bool remove_next_block; |
| 538 | DBUG_ENTER("delete_dynamic_record" ); |
| 539 | |
| 540 | /* First add a link from the last block to the new one */ |
| 541 | error= update_backward_delete_link(info, info->s->state.dellink, filepos); |
| 542 | |
| 543 | block_info.second_read=second_read; |
| 544 | do |
| 545 | { |
| 546 | /* Remove block at 'filepos' */ |
| 547 | if ((b_type= _ma_get_block_info(info, &block_info, info->dfile.file, |
| 548 | filepos)) |
| 549 | & (BLOCK_DELETED | BLOCK_ERROR | BLOCK_SYNC_ERROR | |
| 550 | BLOCK_FATAL_ERROR) || |
| 551 | (length=(uint) (block_info.filepos-filepos) +block_info.block_len) < |
| 552 | MARIA_MIN_BLOCK_LENGTH) |
| 553 | { |
| 554 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 555 | DBUG_RETURN(1); |
| 556 | } |
| 557 | /* Check if next block is a delete block */ |
| 558 | del_block.second_read=0; |
| 559 | remove_next_block=0; |
| 560 | if (_ma_get_block_info(info, &del_block, info->dfile.file, |
| 561 | filepos + length) & |
| 562 | BLOCK_DELETED && del_block.block_len+length < |
| 563 | MARIA_DYN_MAX_BLOCK_LENGTH) |
| 564 | { |
| 565 | /* We can't remove this yet as this block may be the head block */ |
| 566 | remove_next_block=1; |
| 567 | length+=del_block.block_len; |
| 568 | } |
| 569 | |
| 570 | block_info.header[0]=0; |
| 571 | mi_int3store(block_info.header+1,length); |
| 572 | mi_sizestore(block_info.header+4,info->s->state.dellink); |
| 573 | if (b_type & BLOCK_LAST) |
| 574 | bfill(block_info.header+12,8,255); |
| 575 | else |
| 576 | mi_sizestore(block_info.header+12,block_info.next_filepos); |
| 577 | if (info->s->file_write(info, block_info.header, 20, filepos, |
| 578 | MYF(MY_NABP))) |
| 579 | DBUG_RETURN(1); |
| 580 | info->s->state.dellink = filepos; |
| 581 | info->state->del++; |
| 582 | info->state->empty+=length; |
| 583 | filepos=block_info.next_filepos; |
| 584 | |
| 585 | /* Now it's safe to unlink the deleted block directly after this one */ |
| 586 | if (remove_next_block && unlink_deleted_block(info,&del_block)) |
| 587 | error=1; |
| 588 | } while (!(b_type & BLOCK_LAST)); |
| 589 | |
| 590 | DBUG_RETURN(error); |
| 591 | } |
| 592 | |
| 593 | |
| 594 | /* Write a block to datafile */ |
| 595 | |
| 596 | int _ma_write_part_record(MARIA_HA *info, |
| 597 | my_off_t filepos, /* points at empty block */ |
| 598 | ulong length, /* length of block */ |
| 599 | my_off_t next_filepos,/* Next empty block */ |
| 600 | uchar **record, /* pointer to record ptr */ |
| 601 | ulong *reclength, /* length of *record */ |
| 602 | int *flag) /* *flag == 0 if header */ |
| 603 | { |
| 604 | ulong head_length,res_length,,long_block,del_length; |
| 605 | uchar *pos,*record_end; |
| 606 | my_off_t next_delete_block; |
| 607 | uchar temp[MARIA_SPLIT_LENGTH+MARIA_DYN_DELETE_BLOCK_HEADER]; |
| 608 | DBUG_ENTER("_ma_write_part_record" ); |
| 609 | |
| 610 | next_delete_block=HA_OFFSET_ERROR; |
| 611 | |
| 612 | res_length=extra_length=0; |
| 613 | if (length > *reclength + MARIA_SPLIT_LENGTH) |
| 614 | { /* Splitt big block */ |
| 615 | res_length=MY_ALIGN(length- *reclength - MARIA_EXTEND_BLOCK_LENGTH, |
| 616 | MARIA_DYN_ALIGN_SIZE); |
| 617 | length-= res_length; /* Use this for first part */ |
| 618 | } |
| 619 | long_block= (length < 65520L && *reclength < 65520L) ? 0 : 1; |
| 620 | if (length == *reclength+ 3 + long_block) |
| 621 | { |
| 622 | /* Block is exactly of the right length */ |
| 623 | temp[0]=(uchar) (1+ *flag)+(uchar) long_block; /* Flag is 0 or 6 */ |
| 624 | if (long_block) |
| 625 | { |
| 626 | mi_int3store(temp+1,*reclength); |
| 627 | head_length=4; |
| 628 | } |
| 629 | else |
| 630 | { |
| 631 | mi_int2store(temp+1,*reclength); |
| 632 | head_length=3; |
| 633 | } |
| 634 | } |
| 635 | else if (length-long_block < *reclength+4) |
| 636 | { /* To short block */ |
| 637 | if (next_filepos == HA_OFFSET_ERROR) |
| 638 | next_filepos= (info->s->state.dellink != HA_OFFSET_ERROR && |
| 639 | !info->append_insert_at_end ? |
| 640 | info->s->state.dellink : info->state->data_file_length); |
| 641 | if (*flag == 0) /* First block */ |
| 642 | { |
| 643 | if (*reclength > MARIA_MAX_BLOCK_LENGTH) |
| 644 | { |
| 645 | head_length= 16; |
| 646 | temp[0]=13; |
| 647 | mi_int4store(temp+1,*reclength); |
| 648 | mi_int3store(temp+5,length-head_length); |
| 649 | mi_sizestore(temp+8,next_filepos); |
| 650 | } |
| 651 | else |
| 652 | { |
| 653 | head_length=5+8+long_block*2; |
| 654 | temp[0]=5+(uchar) long_block; |
| 655 | if (long_block) |
| 656 | { |
| 657 | mi_int3store(temp+1,*reclength); |
| 658 | mi_int3store(temp+4,length-head_length); |
| 659 | mi_sizestore(temp+7,next_filepos); |
| 660 | } |
| 661 | else |
| 662 | { |
| 663 | mi_int2store(temp+1,*reclength); |
| 664 | mi_int2store(temp+3,length-head_length); |
| 665 | mi_sizestore(temp+5,next_filepos); |
| 666 | } |
| 667 | } |
| 668 | } |
| 669 | else |
| 670 | { |
| 671 | head_length=3+8+long_block; |
| 672 | temp[0]=11+(uchar) long_block; |
| 673 | if (long_block) |
| 674 | { |
| 675 | mi_int3store(temp+1,length-head_length); |
| 676 | mi_sizestore(temp+4,next_filepos); |
| 677 | } |
| 678 | else |
| 679 | { |
| 680 | mi_int2store(temp+1,length-head_length); |
| 681 | mi_sizestore(temp+3,next_filepos); |
| 682 | } |
| 683 | } |
| 684 | } |
| 685 | else |
| 686 | { /* Block with empty info last */ |
| 687 | head_length=4+long_block; |
| 688 | extra_length= length- *reclength-head_length; |
| 689 | temp[0]= (uchar) (3+ *flag)+(uchar) long_block; /* 3,4 or 9,10 */ |
| 690 | if (long_block) |
| 691 | { |
| 692 | mi_int3store(temp+1,*reclength); |
| 693 | temp[4]= (uchar) (extra_length); |
| 694 | } |
| 695 | else |
| 696 | { |
| 697 | mi_int2store(temp+1,*reclength); |
| 698 | temp[3]= (uchar) (extra_length); |
| 699 | } |
| 700 | length= *reclength+head_length; /* Write only what is needed */ |
| 701 | } |
| 702 | DBUG_DUMP("header" , temp, head_length); |
| 703 | |
| 704 | /* Make a long block for one write */ |
| 705 | record_end= *record+length-head_length; |
| 706 | del_length=(res_length ? MARIA_DYN_DELETE_BLOCK_HEADER : 0); |
| 707 | bmove((*record-head_length), temp, head_length); |
| 708 | memcpy(temp,record_end,(size_t) (extra_length+del_length)); |
| 709 | bzero(record_end, extra_length); |
| 710 | |
| 711 | if (res_length) |
| 712 | { |
| 713 | /* Check first if we can join this block with the next one */ |
| 714 | MARIA_BLOCK_INFO del_block; |
| 715 | my_off_t next_block=filepos+length+extra_length+res_length; |
| 716 | |
| 717 | del_block.second_read=0; |
| 718 | if (next_block < info->state->data_file_length && |
| 719 | info->s->state.dellink != HA_OFFSET_ERROR) |
| 720 | { |
| 721 | if ((_ma_get_block_info(info, &del_block, info->dfile.file, next_block) |
| 722 | & BLOCK_DELETED) && |
| 723 | res_length + del_block.block_len < MARIA_DYN_MAX_BLOCK_LENGTH) |
| 724 | { |
| 725 | if (unlink_deleted_block(info,&del_block)) |
| 726 | goto err; |
| 727 | res_length+=del_block.block_len; |
| 728 | } |
| 729 | } |
| 730 | |
| 731 | /* Create a delete link of the last part of the block */ |
| 732 | pos=record_end+extra_length; |
| 733 | pos[0]= '\0'; |
| 734 | mi_int3store(pos+1,res_length); |
| 735 | mi_sizestore(pos+4,info->s->state.dellink); |
| 736 | bfill(pos+12,8,255); /* End link */ |
| 737 | next_delete_block=info->s->state.dellink; |
| 738 | info->s->state.dellink= filepos+length+extra_length; |
| 739 | info->state->del++; |
| 740 | info->state->empty+=res_length; |
| 741 | info->s->state.split++; |
| 742 | } |
| 743 | if (info->opt_flag & WRITE_CACHE_USED && |
| 744 | info->update & HA_STATE_WRITE_AT_END) |
| 745 | { |
| 746 | if (info->update & HA_STATE_EXTEND_BLOCK) |
| 747 | { |
| 748 | info->update&= ~HA_STATE_EXTEND_BLOCK; |
| 749 | if (my_block_write(&info->rec_cache, *record-head_length, |
| 750 | length+extra_length+del_length,filepos)) |
| 751 | goto err; |
| 752 | } |
| 753 | else if (my_b_write(&info->rec_cache, *record-head_length, |
| 754 | length+extra_length+del_length)) |
| 755 | goto err; |
| 756 | } |
| 757 | else |
| 758 | { |
| 759 | info->rec_cache.seek_not_done=1; |
| 760 | if (info->s->file_write(info, *record-head_length, |
| 761 | length+extra_length+ |
| 762 | del_length,filepos,info->s->write_flag)) |
| 763 | goto err; |
| 764 | } |
| 765 | memcpy(record_end,temp,(size_t) (extra_length+del_length)); |
| 766 | *record=record_end; |
| 767 | *reclength-=(length-head_length); |
| 768 | *flag=6; |
| 769 | |
| 770 | if (del_length) |
| 771 | { |
| 772 | /* link the next delete block to this */ |
| 773 | if (update_backward_delete_link(info, next_delete_block, |
| 774 | info->s->state.dellink)) |
| 775 | goto err; |
| 776 | } |
| 777 | |
| 778 | DBUG_RETURN(0); |
| 779 | err: |
| 780 | DBUG_PRINT("exit" ,("errno: %d" ,my_errno)); |
| 781 | DBUG_RETURN(1); |
| 782 | } /* _ma_write_part_record */ |
| 783 | |
| 784 | |
| 785 | /* update record from datafile */ |
| 786 | |
| 787 | static my_bool update_dynamic_record(MARIA_HA *info, MARIA_RECORD_POS filepos, |
| 788 | uchar *record, ulong reclength) |
| 789 | { |
| 790 | int flag; |
| 791 | uint error; |
| 792 | ulong length; |
| 793 | MARIA_BLOCK_INFO block_info; |
| 794 | DBUG_ENTER("update_dynamic_record" ); |
| 795 | |
| 796 | flag=block_info.second_read=0; |
| 797 | /* |
| 798 | Check if we have enough room for the record. |
| 799 | First we do simplified check to make usual case faster. |
| 800 | Then we do more precise check for the space left. |
| 801 | Though it still is not absolutely precise, as |
| 802 | we always use MARIA_MAX_DYN_BLOCK_HEADER while it can be |
| 803 | less in the most of the cases. |
| 804 | */ |
| 805 | |
| 806 | /* |
| 807 | compare with just the reclength as we're going |
| 808 | to get some space from the old replaced record |
| 809 | */ |
| 810 | if (unlikely(info->s->base.max_data_file_length - |
| 811 | info->state->data_file_length < reclength)) |
| 812 | { |
| 813 | /* If new record isn't longer, we can go on safely */ |
| 814 | if (info->cur_row.total_length < reclength) |
| 815 | { |
| 816 | if (info->s->base.max_data_file_length - info->state->data_file_length + |
| 817 | info->state->empty - info->state->del * MARIA_MAX_DYN_BLOCK_HEADER < |
| 818 | reclength - info->cur_row.total_length + MARIA_MAX_DYN_BLOCK_HEADER) |
| 819 | { |
| 820 | my_errno=HA_ERR_RECORD_FILE_FULL; |
| 821 | goto err; |
| 822 | } |
| 823 | } |
| 824 | } |
| 825 | /* Remember length for updated row if it's updated again */ |
| 826 | info->cur_row.total_length= reclength; |
| 827 | |
| 828 | while (reclength > 0) |
| 829 | { |
| 830 | if (filepos != info->s->state.dellink) |
| 831 | { |
| 832 | block_info.next_filepos= HA_OFFSET_ERROR; |
| 833 | if ((error= _ma_get_block_info(info, &block_info, info->dfile.file, |
| 834 | filepos)) |
| 835 | & (BLOCK_DELETED | BLOCK_ERROR | BLOCK_SYNC_ERROR | |
| 836 | BLOCK_FATAL_ERROR)) |
| 837 | { |
| 838 | DBUG_PRINT("error" ,("Got wrong block info" )); |
| 839 | if (!(error & BLOCK_FATAL_ERROR)) |
| 840 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 841 | goto err; |
| 842 | } |
| 843 | length=(ulong) (block_info.filepos-filepos) + block_info.block_len; |
| 844 | if (length < reclength) |
| 845 | { |
| 846 | uint tmp=MY_ALIGN(reclength - length + 3 + |
| 847 | MY_TEST(reclength >= 65520L), MARIA_DYN_ALIGN_SIZE); |
| 848 | /* Don't create a block bigger than MARIA_MAX_BLOCK_LENGTH */ |
| 849 | tmp= MY_MIN(length+tmp, MARIA_MAX_BLOCK_LENGTH)-length; |
| 850 | /* Check if we can extend this block */ |
| 851 | if (block_info.filepos + block_info.block_len == |
| 852 | info->state->data_file_length && |
| 853 | info->state->data_file_length < |
| 854 | info->s->base.max_data_file_length-tmp) |
| 855 | { |
| 856 | /* extend file */ |
| 857 | DBUG_PRINT("info" ,("Extending file with %d bytes" ,tmp)); |
| 858 | if (info->cur_row.nextpos == info->state->data_file_length) |
| 859 | info->cur_row.nextpos+= tmp; |
| 860 | info->state->data_file_length+= tmp; |
| 861 | info->update|= HA_STATE_WRITE_AT_END | HA_STATE_EXTEND_BLOCK; |
| 862 | length+=tmp; |
| 863 | } |
| 864 | else if (length < MARIA_MAX_BLOCK_LENGTH - MARIA_MIN_BLOCK_LENGTH) |
| 865 | { |
| 866 | /* |
| 867 | Check if next block is a deleted block |
| 868 | Above we have MARIA_MIN_BLOCK_LENGTH to avoid the problem where |
| 869 | the next block is so small it can't be splited which could |
| 870 | casue problems |
| 871 | */ |
| 872 | |
| 873 | MARIA_BLOCK_INFO del_block; |
| 874 | del_block.second_read=0; |
| 875 | if (_ma_get_block_info(info, &del_block, info->dfile.file, |
| 876 | block_info.filepos + block_info.block_len) & |
| 877 | BLOCK_DELETED) |
| 878 | { |
| 879 | /* Use; Unlink it and extend the current block */ |
| 880 | DBUG_PRINT("info" ,("Extending current block" )); |
| 881 | if (unlink_deleted_block(info,&del_block)) |
| 882 | goto err; |
| 883 | if ((length+=del_block.block_len) > MARIA_MAX_BLOCK_LENGTH) |
| 884 | { |
| 885 | /* |
| 886 | New block was too big, link overflow part back to |
| 887 | delete list |
| 888 | */ |
| 889 | my_off_t next_pos; |
| 890 | ulong rest_length= length-MARIA_MAX_BLOCK_LENGTH; |
| 891 | set_if_bigger(rest_length, MARIA_MIN_BLOCK_LENGTH); |
| 892 | next_pos= del_block.filepos+ del_block.block_len - rest_length; |
| 893 | |
| 894 | if (update_backward_delete_link(info, info->s->state.dellink, |
| 895 | next_pos)) |
| 896 | DBUG_RETURN(1); |
| 897 | |
| 898 | /* create delete link for data that didn't fit into the page */ |
| 899 | del_block.header[0]=0; |
| 900 | mi_int3store(del_block.header+1, rest_length); |
| 901 | mi_sizestore(del_block.header+4,info->s->state.dellink); |
| 902 | bfill(del_block.header+12,8,255); |
| 903 | if (info->s->file_write(info, del_block.header, 20, |
| 904 | next_pos, MYF(MY_NABP))) |
| 905 | DBUG_RETURN(1); |
| 906 | info->s->state.dellink= next_pos; |
| 907 | info->s->state.split++; |
| 908 | info->state->del++; |
| 909 | info->state->empty+= rest_length; |
| 910 | length-= rest_length; |
| 911 | } |
| 912 | } |
| 913 | } |
| 914 | } |
| 915 | } |
| 916 | else |
| 917 | { |
| 918 | if (_ma_find_writepos(info,reclength,&filepos,&length)) |
| 919 | goto err; |
| 920 | } |
| 921 | if (_ma_write_part_record(info,filepos,length,block_info.next_filepos, |
| 922 | &record,&reclength,&flag)) |
| 923 | goto err; |
| 924 | if ((filepos=block_info.next_filepos) == HA_OFFSET_ERROR) |
| 925 | { |
| 926 | /* Start writing data on deleted blocks */ |
| 927 | filepos=info->s->state.dellink; |
| 928 | } |
| 929 | } |
| 930 | |
| 931 | if (block_info.next_filepos != HA_OFFSET_ERROR) |
| 932 | if (delete_dynamic_record(info,block_info.next_filepos,1)) |
| 933 | goto err; |
| 934 | |
| 935 | DBUG_RETURN(0); |
| 936 | err: |
| 937 | DBUG_RETURN(1); |
| 938 | } |
| 939 | |
| 940 | |
| 941 | /* Pack a record. Return new reclength */ |
| 942 | |
| 943 | uint _ma_rec_pack(MARIA_HA *info, register uchar *to, |
| 944 | register const uchar *from) |
| 945 | { |
| 946 | uint length,new_length,flag,bit,i; |
| 947 | const uchar *pos,*end; |
| 948 | uchar *startpos,*packpos; |
| 949 | enum en_fieldtype type; |
| 950 | reg3 MARIA_COLUMNDEF *column; |
| 951 | MARIA_BLOB *blob; |
| 952 | DBUG_ENTER("_ma_rec_pack" ); |
| 953 | |
| 954 | flag= 0; |
| 955 | bit= 1; |
| 956 | startpos= packpos=to; |
| 957 | to+= info->s->base.pack_bytes; |
| 958 | blob= info->blobs; |
| 959 | column= info->s->columndef; |
| 960 | if (info->s->base.null_bytes) |
| 961 | { |
| 962 | memcpy(to, from, info->s->base.null_bytes); |
| 963 | from+= info->s->base.null_bytes; |
| 964 | to+= info->s->base.null_bytes; |
| 965 | } |
| 966 | |
| 967 | for (i=info->s->base.fields ; i-- > 0; from+= length, column++) |
| 968 | { |
| 969 | length=(uint) column->length; |
| 970 | if ((type = (enum en_fieldtype) column->type) != FIELD_NORMAL) |
| 971 | { |
| 972 | if (type == FIELD_BLOB) |
| 973 | { |
| 974 | if (!blob->length) |
| 975 | flag|=bit; |
| 976 | else |
| 977 | { |
| 978 | char *temp_pos; |
| 979 | size_t tmp_length=length-portable_sizeof_char_ptr; |
| 980 | memcpy(to,from,tmp_length); |
| 981 | memcpy(&temp_pos,from+tmp_length,sizeof(char*)); |
| 982 | memcpy(to+tmp_length,temp_pos,(size_t) blob->length); |
| 983 | to+=tmp_length+blob->length; |
| 984 | } |
| 985 | blob++; |
| 986 | } |
| 987 | else if (type == FIELD_SKIP_ZERO) |
| 988 | { |
| 989 | if (memcmp(from, maria_zero_string, length) == 0) |
| 990 | flag|=bit; |
| 991 | else |
| 992 | { |
| 993 | memcpy(to, from, (size_t) length); |
| 994 | to+=length; |
| 995 | } |
| 996 | } |
| 997 | else if (type == FIELD_SKIP_ENDSPACE || |
| 998 | type == FIELD_SKIP_PRESPACE) |
| 999 | { |
| 1000 | pos= from; end= from + length; |
| 1001 | if (type == FIELD_SKIP_ENDSPACE) |
| 1002 | { /* Pack trailing spaces */ |
| 1003 | while (end > from && *(end-1) == ' ') |
| 1004 | end--; |
| 1005 | } |
| 1006 | else |
| 1007 | { /* Pack pref-spaces */ |
| 1008 | while (pos < end && *pos == ' ') |
| 1009 | pos++; |
| 1010 | } |
| 1011 | new_length=(uint) (end-pos); |
| 1012 | if (new_length + 1 + MY_TEST(column->length > 255 && new_length > 127) |
| 1013 | < length) |
| 1014 | { |
| 1015 | if (column->length > 255 && new_length > 127) |
| 1016 | { |
| 1017 | to[0]= (uchar) ((new_length & 127) + 128); |
| 1018 | to[1]= (uchar) (new_length >> 7); |
| 1019 | to+=2; |
| 1020 | } |
| 1021 | else |
| 1022 | *to++= (uchar) new_length; |
| 1023 | memcpy(to, pos, (size_t) new_length); to+=new_length; |
| 1024 | flag|=bit; |
| 1025 | } |
| 1026 | else |
| 1027 | { |
| 1028 | memcpy(to,from,(size_t) length); to+=length; |
| 1029 | } |
| 1030 | } |
| 1031 | else if (type == FIELD_VARCHAR) |
| 1032 | { |
| 1033 | uint pack_length= HA_VARCHAR_PACKLENGTH(column->length -1); |
| 1034 | uint tmp_length; |
| 1035 | if (pack_length == 1) |
| 1036 | { |
| 1037 | tmp_length= (uint) *from; |
| 1038 | *to++= *from; |
| 1039 | } |
| 1040 | else |
| 1041 | { |
| 1042 | tmp_length= uint2korr(from); |
| 1043 | store_key_length_inc(to,tmp_length); |
| 1044 | } |
| 1045 | memcpy(to, from+pack_length,tmp_length); |
| 1046 | to+= tmp_length; |
| 1047 | continue; |
| 1048 | } |
| 1049 | else |
| 1050 | { |
| 1051 | memcpy(to,from,(size_t) length); to+=length; |
| 1052 | continue; /* Normal field */ |
| 1053 | } |
| 1054 | if ((bit= bit << 1) >= 256) |
| 1055 | { |
| 1056 | *packpos++ = (uchar) flag; |
| 1057 | bit=1; flag=0; |
| 1058 | } |
| 1059 | } |
| 1060 | else |
| 1061 | { |
| 1062 | memcpy(to,from,(size_t) length); to+=length; |
| 1063 | } |
| 1064 | } |
| 1065 | if (bit != 1) |
| 1066 | *packpos= (uchar) flag; |
| 1067 | if (info->s->calc_checksum) |
| 1068 | *to++= (uchar) info->cur_row.checksum; |
| 1069 | DBUG_PRINT("exit" ,("packed length: %d" ,(int) (to-startpos))); |
| 1070 | DBUG_RETURN((uint) (to-startpos)); |
| 1071 | } /* _ma_rec_pack */ |
| 1072 | |
| 1073 | |
| 1074 | |
| 1075 | /* |
| 1076 | Check if a record was correctly packed. Used only by maria_chk |
| 1077 | Returns 0 if record is ok. |
| 1078 | */ |
| 1079 | |
| 1080 | my_bool _ma_rec_check(MARIA_HA *info,const uchar *record, uchar *rec_buff, |
| 1081 | ulong packed_length, my_bool with_checksum, |
| 1082 | ha_checksum checksum) |
| 1083 | { |
| 1084 | uint length,new_length,flag,bit,i; |
| 1085 | const uchar *pos,*end; |
| 1086 | uchar *packpos,*to; |
| 1087 | enum en_fieldtype type; |
| 1088 | reg3 MARIA_COLUMNDEF *column; |
| 1089 | DBUG_ENTER("_ma_rec_check" ); |
| 1090 | |
| 1091 | packpos=rec_buff; to= rec_buff+info->s->base.pack_bytes; |
| 1092 | column= info->s->columndef; |
| 1093 | flag= *packpos; bit=1; |
| 1094 | record+= info->s->base.null_bytes; |
| 1095 | to+= info->s->base.null_bytes; |
| 1096 | |
| 1097 | for (i=info->s->base.fields ; i-- > 0; record+= length, column++) |
| 1098 | { |
| 1099 | length=(uint) column->length; |
| 1100 | if ((type = (enum en_fieldtype) column->type) != FIELD_NORMAL) |
| 1101 | { |
| 1102 | if (type == FIELD_BLOB) |
| 1103 | { |
| 1104 | uint blob_length= |
| 1105 | _ma_calc_blob_length(length-portable_sizeof_char_ptr,record); |
| 1106 | if (!blob_length && !(flag & bit)) |
| 1107 | goto err; |
| 1108 | if (blob_length) |
| 1109 | to+=length - portable_sizeof_char_ptr+ blob_length; |
| 1110 | } |
| 1111 | else if (type == FIELD_SKIP_ZERO) |
| 1112 | { |
| 1113 | if (memcmp(record, maria_zero_string, length) == 0) |
| 1114 | { |
| 1115 | if (!(flag & bit)) |
| 1116 | goto err; |
| 1117 | } |
| 1118 | else |
| 1119 | to+=length; |
| 1120 | } |
| 1121 | else if (type == FIELD_SKIP_ENDSPACE || |
| 1122 | type == FIELD_SKIP_PRESPACE) |
| 1123 | { |
| 1124 | pos= record; end= record + length; |
| 1125 | if (type == FIELD_SKIP_ENDSPACE) |
| 1126 | { /* Pack trailing spaces */ |
| 1127 | while (end > record && *(end-1) == ' ') |
| 1128 | end--; |
| 1129 | } |
| 1130 | else |
| 1131 | { /* Pack pre-spaces */ |
| 1132 | while (pos < end && *pos == ' ') |
| 1133 | pos++; |
| 1134 | } |
| 1135 | new_length=(uint) (end-pos); |
| 1136 | if (new_length + 1 + MY_TEST(column->length > 255 && new_length > 127) |
| 1137 | < length) |
| 1138 | { |
| 1139 | if (!(flag & bit)) |
| 1140 | goto err; |
| 1141 | if (column->length > 255 && new_length > 127) |
| 1142 | { |
| 1143 | /* purecov: begin inspected */ |
| 1144 | if (to[0] != (uchar) ((new_length & 127) + 128) || |
| 1145 | to[1] != (uchar) (new_length >> 7)) |
| 1146 | goto err; |
| 1147 | to+=2; |
| 1148 | /* purecov: end */ |
| 1149 | } |
| 1150 | else if (*to++ != (uchar) new_length) |
| 1151 | goto err; |
| 1152 | to+=new_length; |
| 1153 | } |
| 1154 | else |
| 1155 | to+=length; |
| 1156 | } |
| 1157 | else if (type == FIELD_VARCHAR) |
| 1158 | { |
| 1159 | uint pack_length= HA_VARCHAR_PACKLENGTH(column->length -1); |
| 1160 | uint tmp_length; |
| 1161 | if (pack_length == 1) |
| 1162 | { |
| 1163 | tmp_length= (uint) *record; |
| 1164 | to+= 1+ tmp_length; |
| 1165 | continue; |
| 1166 | } |
| 1167 | else |
| 1168 | { |
| 1169 | tmp_length= uint2korr(record); |
| 1170 | to+= get_pack_length(tmp_length)+tmp_length; |
| 1171 | } |
| 1172 | continue; |
| 1173 | } |
| 1174 | else |
| 1175 | { |
| 1176 | to+=length; |
| 1177 | continue; /* Normal field */ |
| 1178 | } |
| 1179 | if ((bit= bit << 1) >= 256) |
| 1180 | { |
| 1181 | flag= *++packpos; |
| 1182 | bit=1; |
| 1183 | } |
| 1184 | } |
| 1185 | else |
| 1186 | to+= length; |
| 1187 | } |
| 1188 | if (packed_length != (uint) (to - rec_buff) + |
| 1189 | MY_TEST(info->s->calc_checksum) || (bit != 1 && (flag & ~(bit - 1)))) |
| 1190 | goto err; |
| 1191 | if (with_checksum && ((uchar) checksum != (uchar) *to)) |
| 1192 | { |
| 1193 | DBUG_PRINT("error" ,("wrong checksum for row" )); |
| 1194 | goto err; |
| 1195 | } |
| 1196 | DBUG_RETURN(0); |
| 1197 | |
| 1198 | err: |
| 1199 | DBUG_RETURN(1); |
| 1200 | } |
| 1201 | |
| 1202 | |
| 1203 | /* |
| 1204 | @brief Unpacks a record |
| 1205 | |
| 1206 | @return Recordlength |
| 1207 | @retval >0 ok |
| 1208 | @retval MY_FILE_ERROR (== -1) Error. |
| 1209 | my_errno is set to HA_ERR_WRONG_IN_RECORD |
| 1210 | */ |
| 1211 | |
| 1212 | size_t _ma_rec_unpack(register MARIA_HA *info, register uchar *to, uchar *from, |
| 1213 | size_t found_length) |
| 1214 | { |
| 1215 | uint flag,bit,length,min_pack_length, column_length; |
| 1216 | enum en_fieldtype type; |
| 1217 | uchar *from_end,*to_end,*packpos; |
| 1218 | reg3 MARIA_COLUMNDEF *column, *end_column; |
| 1219 | DBUG_ENTER("_ma_rec_unpack" ); |
| 1220 | |
| 1221 | to_end=to + info->s->base.reclength; |
| 1222 | from_end=from+found_length; |
| 1223 | flag= (uchar) *from; bit=1; packpos=from; |
| 1224 | if (found_length < info->s->base.min_pack_length) |
| 1225 | goto err; |
| 1226 | from+= info->s->base.pack_bytes; |
| 1227 | min_pack_length= info->s->base.min_pack_length - info->s->base.pack_bytes; |
| 1228 | |
| 1229 | if ((length= info->s->base.null_bytes)) |
| 1230 | { |
| 1231 | memcpy(to, from, length); |
| 1232 | from+= length; |
| 1233 | to+= length; |
| 1234 | min_pack_length-= length; |
| 1235 | } |
| 1236 | |
| 1237 | for (column= info->s->columndef, end_column= column + info->s->base.fields; |
| 1238 | column < end_column ; to+= column_length, column++) |
| 1239 | { |
| 1240 | column_length= column->length; |
| 1241 | if ((type = (enum en_fieldtype) column->type) != FIELD_NORMAL && |
| 1242 | (type != FIELD_CHECK)) |
| 1243 | { |
| 1244 | if (type == FIELD_VARCHAR) |
| 1245 | { |
| 1246 | uint pack_length= HA_VARCHAR_PACKLENGTH(column_length-1); |
| 1247 | if (pack_length == 1) |
| 1248 | { |
| 1249 | length= (uint) *(uchar*) from; |
| 1250 | if (length > column_length-1) |
| 1251 | goto err; |
| 1252 | *to= *from++; |
| 1253 | } |
| 1254 | else |
| 1255 | { |
| 1256 | get_key_length(length, from); |
| 1257 | if (length > column_length-2) |
| 1258 | goto err; |
| 1259 | int2store(to,length); |
| 1260 | } |
| 1261 | if (from+length > from_end) |
| 1262 | goto err; |
| 1263 | memcpy(to+pack_length, from, length); |
| 1264 | from+= length; |
| 1265 | min_pack_length--; |
| 1266 | continue; |
| 1267 | } |
| 1268 | if (flag & bit) |
| 1269 | { |
| 1270 | if (type == FIELD_BLOB || type == FIELD_SKIP_ZERO) |
| 1271 | bzero(to, column_length); |
| 1272 | else if (type == FIELD_SKIP_ENDSPACE || |
| 1273 | type == FIELD_SKIP_PRESPACE) |
| 1274 | { |
| 1275 | if (column->length > 255 && *from & 128) |
| 1276 | { |
| 1277 | if (from + 1 >= from_end) |
| 1278 | goto err; |
| 1279 | length= (*from & 127)+ ((uint) (uchar) *(from+1) << 7); from+=2; |
| 1280 | } |
| 1281 | else |
| 1282 | { |
| 1283 | if (from == from_end) |
| 1284 | goto err; |
| 1285 | length= (uchar) *from++; |
| 1286 | } |
| 1287 | min_pack_length--; |
| 1288 | if (length >= column_length || |
| 1289 | min_pack_length + length > (uint) (from_end - from)) |
| 1290 | goto err; |
| 1291 | if (type == FIELD_SKIP_ENDSPACE) |
| 1292 | { |
| 1293 | memcpy(to, from, (size_t) length); |
| 1294 | bfill(to+length, column_length-length, ' '); |
| 1295 | } |
| 1296 | else |
| 1297 | { |
| 1298 | bfill(to, column_length-length, ' '); |
| 1299 | memcpy(to+column_length-length, from, (size_t) length); |
| 1300 | } |
| 1301 | from+=length; |
| 1302 | } |
| 1303 | } |
| 1304 | else if (type == FIELD_BLOB) |
| 1305 | { |
| 1306 | uint size_length=column_length- portable_sizeof_char_ptr; |
| 1307 | ulong blob_length= _ma_calc_blob_length(size_length,from); |
| 1308 | ulong from_left= (ulong) (from_end - from); |
| 1309 | if (from_left < size_length || |
| 1310 | from_left - size_length < blob_length || |
| 1311 | from_left - size_length - blob_length < min_pack_length) |
| 1312 | goto err; |
| 1313 | memcpy(to, from, (size_t) size_length); |
| 1314 | from+=size_length; |
| 1315 | memcpy(to+size_length,(uchar*) &from,sizeof(char*)); |
| 1316 | from+=blob_length; |
| 1317 | } |
| 1318 | else |
| 1319 | { |
| 1320 | if (type == FIELD_SKIP_ENDSPACE || type == FIELD_SKIP_PRESPACE) |
| 1321 | min_pack_length--; |
| 1322 | if (min_pack_length + column_length > (uint) (from_end - from)) |
| 1323 | goto err; |
| 1324 | memcpy(to, from, (size_t) column_length); from+=column_length; |
| 1325 | } |
| 1326 | if ((bit= bit << 1) >= 256) |
| 1327 | { |
| 1328 | flag= (uchar) *++packpos; bit=1; |
| 1329 | } |
| 1330 | } |
| 1331 | else |
| 1332 | { |
| 1333 | if (min_pack_length > (uint) (from_end - from)) |
| 1334 | goto err; |
| 1335 | min_pack_length-=column_length; |
| 1336 | memcpy(to, from, (size_t) column_length); |
| 1337 | from+=column_length; |
| 1338 | } |
| 1339 | } |
| 1340 | if (info->s->calc_checksum) |
| 1341 | info->cur_row.checksum= (uint) (uchar) *from++; |
| 1342 | if (to == to_end && from == from_end && (bit == 1 || !(flag & ~(bit-1)))) |
| 1343 | DBUG_RETURN(found_length); |
| 1344 | |
| 1345 | err: |
| 1346 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 1347 | DBUG_PRINT("error" ,("to_end: %p -> %p from_end: %p -> %p" , |
| 1348 | to, to_end, from, from_end)); |
| 1349 | DBUG_DUMP("from" , info->rec_buff, info->s->base.min_pack_length); |
| 1350 | DBUG_RETURN(MY_FILE_ERROR); |
| 1351 | } /* _ma_rec_unpack */ |
| 1352 | |
| 1353 | |
| 1354 | /* Calc length of blob. Update info in blobs->length */ |
| 1355 | |
| 1356 | ulong _ma_calc_total_blob_length(MARIA_HA *info, const uchar *record) |
| 1357 | { |
| 1358 | ulong length; |
| 1359 | MARIA_BLOB *blob,*end; |
| 1360 | |
| 1361 | for (length=0, blob= info->blobs, end=blob+info->s->base.blobs ; |
| 1362 | blob != end; |
| 1363 | blob++) |
| 1364 | { |
| 1365 | blob->length= _ma_calc_blob_length(blob->pack_length, |
| 1366 | record + blob->offset); |
| 1367 | length+=blob->length; |
| 1368 | } |
| 1369 | return length; |
| 1370 | } |
| 1371 | |
| 1372 | |
| 1373 | ulong _ma_calc_blob_length(uint length, const uchar *pos) |
| 1374 | { |
| 1375 | switch (length) { |
| 1376 | case 1: |
| 1377 | return (uint) (uchar) *pos; |
| 1378 | case 2: |
| 1379 | return (uint) uint2korr(pos); |
| 1380 | case 3: |
| 1381 | return uint3korr(pos); |
| 1382 | case 4: |
| 1383 | return uint4korr(pos); |
| 1384 | default: |
| 1385 | break; |
| 1386 | } |
| 1387 | return 0; /* Impossible */ |
| 1388 | } |
| 1389 | |
| 1390 | |
| 1391 | void _ma_store_blob_length(uchar *pos,uint pack_length,uint length) |
| 1392 | { |
| 1393 | switch (pack_length) { |
| 1394 | case 1: |
| 1395 | *pos= (uchar) length; |
| 1396 | break; |
| 1397 | case 2: |
| 1398 | int2store(pos,length); |
| 1399 | break; |
| 1400 | case 3: |
| 1401 | int3store(pos,length); |
| 1402 | break; |
| 1403 | case 4: |
| 1404 | int4store(pos,length); |
| 1405 | default: |
| 1406 | break; |
| 1407 | } |
| 1408 | return; |
| 1409 | } |
| 1410 | |
| 1411 | |
| 1412 | /* |
| 1413 | Read record from datafile. |
| 1414 | |
| 1415 | SYNOPSIS |
| 1416 | _ma_read_dynamic_record() |
| 1417 | info MARIA_HA pointer to table. |
| 1418 | filepos From where to read the record. |
| 1419 | buf Destination for record. |
| 1420 | |
| 1421 | NOTE |
| 1422 | If a write buffer is active, it needs to be flushed if its contents |
| 1423 | intersects with the record to read. We always check if the position |
| 1424 | of the first uchar of the write buffer is lower than the position |
| 1425 | past the last uchar to read. In theory this is also true if the write |
| 1426 | buffer is completely below the read segment. That is, if there is no |
| 1427 | intersection. But this case is unusual. We flush anyway. Only if the |
| 1428 | first uchar in the write buffer is above the last uchar to read, we do |
| 1429 | not flush. |
| 1430 | |
| 1431 | A dynamic record may need several reads. So this check must be done |
| 1432 | before every read. Reading a dynamic record starts with reading the |
| 1433 | block header. If the record does not fit into the free space of the |
| 1434 | header, the block may be longer than the header. In this case a |
| 1435 | second read is necessary. These one or two reads repeat for every |
| 1436 | part of the record. |
| 1437 | |
| 1438 | RETURN |
| 1439 | 0 OK |
| 1440 | # Error number |
| 1441 | */ |
| 1442 | |
| 1443 | int _ma_read_dynamic_record(MARIA_HA *info, uchar *buf, |
| 1444 | MARIA_RECORD_POS filepos) |
| 1445 | { |
| 1446 | int block_of_record; |
| 1447 | uint b_type; |
| 1448 | MARIA_BLOCK_INFO block_info; |
| 1449 | File file; |
| 1450 | uchar *UNINIT_VAR(to); |
| 1451 | uint UNINIT_VAR(left_length); |
| 1452 | DBUG_ENTER("_ma_read_dynamic_record" ); |
| 1453 | |
| 1454 | if (filepos == HA_OFFSET_ERROR) |
| 1455 | goto err; |
| 1456 | |
| 1457 | file= info->dfile.file; |
| 1458 | block_of_record= 0; /* First block of record is numbered as zero. */ |
| 1459 | block_info.second_read= 0; |
| 1460 | do |
| 1461 | { |
| 1462 | /* A corrupted table can have wrong pointers. (Bug# 19835) */ |
| 1463 | if (filepos == HA_OFFSET_ERROR) |
| 1464 | goto panic; |
| 1465 | if (info->opt_flag & WRITE_CACHE_USED && |
| 1466 | (info->rec_cache.pos_in_file < filepos + |
| 1467 | MARIA_BLOCK_INFO_HEADER_LENGTH) && |
| 1468 | flush_io_cache(&info->rec_cache)) |
| 1469 | goto err; |
| 1470 | info->rec_cache.seek_not_done=1; |
| 1471 | if ((b_type= _ma_get_block_info(info, &block_info, file, filepos)) & |
| 1472 | (BLOCK_DELETED | BLOCK_ERROR | BLOCK_SYNC_ERROR | |
| 1473 | BLOCK_FATAL_ERROR)) |
| 1474 | { |
| 1475 | if (b_type & (BLOCK_SYNC_ERROR | BLOCK_DELETED)) |
| 1476 | my_errno=HA_ERR_RECORD_DELETED; |
| 1477 | goto err; |
| 1478 | } |
| 1479 | if (block_of_record++ == 0) /* First block */ |
| 1480 | { |
| 1481 | info->cur_row.total_length= block_info.rec_len; |
| 1482 | if (block_info.rec_len > (uint) info->s->base.max_pack_length) |
| 1483 | goto panic; |
| 1484 | if (info->s->base.blobs) |
| 1485 | { |
| 1486 | if (_ma_alloc_buffer(&info->rec_buff, &info->rec_buff_size, |
| 1487 | block_info.rec_len + |
| 1488 | info->s->base.extra_rec_buff_size)) |
| 1489 | goto err; |
| 1490 | } |
| 1491 | to= info->rec_buff; |
| 1492 | left_length=block_info.rec_len; |
| 1493 | } |
| 1494 | if (left_length < block_info.data_len || ! block_info.data_len) |
| 1495 | goto panic; /* Wrong linked record */ |
| 1496 | /* copy information that is already read */ |
| 1497 | { |
| 1498 | uint offset= (uint) (block_info.filepos - filepos); |
| 1499 | uint prefetch_len= (sizeof(block_info.header) - offset); |
| 1500 | filepos+= sizeof(block_info.header); |
| 1501 | |
| 1502 | if (prefetch_len > block_info.data_len) |
| 1503 | prefetch_len= block_info.data_len; |
| 1504 | if (prefetch_len) |
| 1505 | { |
| 1506 | memcpy(to, block_info.header + offset, prefetch_len); |
| 1507 | block_info.data_len-= prefetch_len; |
| 1508 | left_length-= prefetch_len; |
| 1509 | to+= prefetch_len; |
| 1510 | } |
| 1511 | } |
| 1512 | /* read rest of record from file */ |
| 1513 | if (block_info.data_len) |
| 1514 | { |
| 1515 | if (info->opt_flag & WRITE_CACHE_USED && |
| 1516 | info->rec_cache.pos_in_file < filepos + block_info.data_len && |
| 1517 | flush_io_cache(&info->rec_cache)) |
| 1518 | goto err; |
| 1519 | /* |
| 1520 | What a pity that this method is not called 'file_pread' and that |
| 1521 | there is no equivalent without seeking. We are at the right |
| 1522 | position already. :( |
| 1523 | */ |
| 1524 | if (info->s->file_read(info, to, block_info.data_len, |
| 1525 | filepos, MYF(MY_NABP))) |
| 1526 | goto panic; |
| 1527 | left_length-=block_info.data_len; |
| 1528 | to+=block_info.data_len; |
| 1529 | } |
| 1530 | filepos= block_info.next_filepos; |
| 1531 | } while (left_length); |
| 1532 | |
| 1533 | info->update|= HA_STATE_AKTIV; /* We have a aktive record */ |
| 1534 | fast_ma_writeinfo(info); |
| 1535 | DBUG_RETURN(_ma_rec_unpack(info,buf,info->rec_buff,block_info.rec_len) != |
| 1536 | MY_FILE_ERROR ? 0 : my_errno); |
| 1537 | |
| 1538 | err: |
| 1539 | fast_ma_writeinfo(info); |
| 1540 | DBUG_RETURN(my_errno); |
| 1541 | |
| 1542 | panic: |
| 1543 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 1544 | goto err; |
| 1545 | } |
| 1546 | |
| 1547 | /* compare unique constraint between stored rows */ |
| 1548 | |
| 1549 | my_bool _ma_cmp_dynamic_unique(MARIA_HA *info, MARIA_UNIQUEDEF *def, |
| 1550 | const uchar *record, MARIA_RECORD_POS pos) |
| 1551 | { |
| 1552 | uchar *old_rec_buff,*old_record; |
| 1553 | size_t old_rec_buff_size; |
| 1554 | my_bool error; |
| 1555 | DBUG_ENTER("_ma_cmp_dynamic_unique" ); |
| 1556 | |
| 1557 | if (!(old_record= my_safe_alloca(info->s->base.reclength))) |
| 1558 | DBUG_RETURN(1); |
| 1559 | |
| 1560 | /* Don't let the compare destroy blobs that may be in use */ |
| 1561 | old_rec_buff= info->rec_buff; |
| 1562 | old_rec_buff_size= info->rec_buff_size; |
| 1563 | |
| 1564 | if (info->s->base.blobs) |
| 1565 | { |
| 1566 | info->rec_buff= 0; |
| 1567 | info->rec_buff_size= 0; |
| 1568 | } |
| 1569 | error= _ma_read_dynamic_record(info, old_record, pos) != 0; |
| 1570 | if (!error) |
| 1571 | error=_ma_unique_comp(def, record, old_record, def->null_are_equal) != 0; |
| 1572 | if (info->s->base.blobs) |
| 1573 | { |
| 1574 | my_free(info->rec_buff); |
| 1575 | info->rec_buff= old_rec_buff; |
| 1576 | info->rec_buff_size= old_rec_buff_size; |
| 1577 | } |
| 1578 | my_safe_afree(old_record, info->s->base.reclength); |
| 1579 | DBUG_RETURN(error); |
| 1580 | } |
| 1581 | |
| 1582 | |
| 1583 | /* Compare of record on disk with packed record in memory */ |
| 1584 | |
| 1585 | my_bool _ma_cmp_dynamic_record(register MARIA_HA *info, |
| 1586 | register const uchar *record) |
| 1587 | { |
| 1588 | uint flag, reclength, b_type,cmp_length; |
| 1589 | my_off_t filepos; |
| 1590 | uchar *buffer; |
| 1591 | MARIA_BLOCK_INFO block_info; |
| 1592 | my_bool error= 1; |
| 1593 | size_t UNINIT_VAR(buffer_length); |
| 1594 | DBUG_ENTER("_ma_cmp_dynamic_record" ); |
| 1595 | |
| 1596 | if (info->opt_flag & WRITE_CACHE_USED) |
| 1597 | { |
| 1598 | info->update&= ~(HA_STATE_WRITE_AT_END | HA_STATE_EXTEND_BLOCK); |
| 1599 | if (flush_io_cache(&info->rec_cache)) |
| 1600 | DBUG_RETURN(1); |
| 1601 | } |
| 1602 | info->rec_cache.seek_not_done=1; |
| 1603 | |
| 1604 | /* If nobody have touched the database we don't have to test rec */ |
| 1605 | |
| 1606 | buffer=info->rec_buff; |
| 1607 | if ((info->opt_flag & READ_CHECK_USED)) |
| 1608 | { /* If check isn't disabled */ |
| 1609 | if (info->s->base.blobs) |
| 1610 | { |
| 1611 | buffer_length= (info->s->base.pack_reclength + |
| 1612 | _ma_calc_total_blob_length(info,record)); |
| 1613 | if (!(buffer=(uchar*) my_safe_alloca(buffer_length))) |
| 1614 | DBUG_RETURN(1); |
| 1615 | } |
| 1616 | reclength= _ma_rec_pack(info,buffer,record); |
| 1617 | record= buffer; |
| 1618 | |
| 1619 | filepos= info->cur_row.lastpos; |
| 1620 | flag=block_info.second_read=0; |
| 1621 | block_info.next_filepos=filepos; |
| 1622 | while (reclength > 0) |
| 1623 | { |
| 1624 | if ((b_type= _ma_get_block_info(info, &block_info, info->dfile.file, |
| 1625 | block_info.next_filepos)) |
| 1626 | & (BLOCK_DELETED | BLOCK_ERROR | BLOCK_SYNC_ERROR | |
| 1627 | BLOCK_FATAL_ERROR)) |
| 1628 | { |
| 1629 | if (b_type & (BLOCK_SYNC_ERROR | BLOCK_DELETED)) |
| 1630 | my_errno=HA_ERR_RECORD_CHANGED; |
| 1631 | goto err; |
| 1632 | } |
| 1633 | if (flag == 0) /* First block */ |
| 1634 | { |
| 1635 | flag=1; |
| 1636 | if (reclength != block_info.rec_len) |
| 1637 | { |
| 1638 | my_errno=HA_ERR_RECORD_CHANGED; |
| 1639 | goto err; |
| 1640 | } |
| 1641 | } else if (reclength < block_info.data_len) |
| 1642 | { |
| 1643 | _ma_set_fatal_error(info->s, HA_ERR_WRONG_IN_RECORD); |
| 1644 | goto err; |
| 1645 | } |
| 1646 | reclength-= block_info.data_len; |
| 1647 | cmp_length= block_info.data_len; |
| 1648 | if (!reclength && info->s->calc_checksum) |
| 1649 | cmp_length--; /* 'record' may not contain checksum */ |
| 1650 | |
| 1651 | if (_ma_cmp_buffer(info->dfile.file, record, block_info.filepos, |
| 1652 | cmp_length)) |
| 1653 | { |
| 1654 | my_errno=HA_ERR_RECORD_CHANGED; |
| 1655 | goto err; |
| 1656 | } |
| 1657 | flag=1; |
| 1658 | record+=block_info.data_len; |
| 1659 | } |
| 1660 | } |
| 1661 | my_errno=0; |
| 1662 | error= 0; |
| 1663 | err: |
| 1664 | if (buffer != info->rec_buff) |
| 1665 | my_safe_afree(buffer, buffer_length); |
| 1666 | DBUG_PRINT("exit" , ("result: %d" , error)); |
| 1667 | DBUG_RETURN(error); |
| 1668 | } |
| 1669 | |
| 1670 | |
| 1671 | /* Compare file to buffert */ |
| 1672 | |
| 1673 | static my_bool _ma_cmp_buffer(File file, const uchar *buff, my_off_t filepos, |
| 1674 | uint length) |
| 1675 | { |
| 1676 | uint next_length; |
| 1677 | uchar temp_buff[IO_SIZE*2]; |
| 1678 | DBUG_ENTER("_ma_cmp_buffer" ); |
| 1679 | |
| 1680 | next_length= IO_SIZE*2 - (uint) (filepos & (IO_SIZE-1)); |
| 1681 | |
| 1682 | while (length > IO_SIZE*2) |
| 1683 | { |
| 1684 | if (mysql_file_pread(file,temp_buff,next_length,filepos, MYF(MY_NABP)) || |
| 1685 | memcmp(buff, temp_buff, next_length)) |
| 1686 | goto err; |
| 1687 | filepos+=next_length; |
| 1688 | buff+=next_length; |
| 1689 | length-= next_length; |
| 1690 | next_length=IO_SIZE*2; |
| 1691 | } |
| 1692 | if (mysql_file_pread(file,temp_buff,length,filepos,MYF(MY_NABP))) |
| 1693 | goto err; |
| 1694 | DBUG_RETURN(memcmp(buff, temp_buff, length) != 0); |
| 1695 | err: |
| 1696 | DBUG_RETURN(1); |
| 1697 | } |
| 1698 | |
| 1699 | |
| 1700 | /* |
| 1701 | Read next record from datafile during table scan. |
| 1702 | |
| 1703 | SYNOPSIS |
| 1704 | _ma_read_rnd_dynamic_record() |
| 1705 | info MARIA_HA pointer to table. |
| 1706 | buf Destination for record. |
| 1707 | filepos From where to read the record. |
| 1708 | skip_deleted_blocks If to repeat reading until a non-deleted |
| 1709 | record is found. |
| 1710 | |
| 1711 | NOTE |
| 1712 | This is identical to _ma_read_dynamic_record(), except the following |
| 1713 | cases: |
| 1714 | |
| 1715 | - If there is no active row at 'filepos', continue scanning for |
| 1716 | an active row. (This is becasue the previous |
| 1717 | _ma_read_rnd_dynamic_record() call stored the next block position |
| 1718 | in filepos, but this position may not be a start block for a row |
| 1719 | - We may have READ_CACHING enabled, in which case we use the cache |
| 1720 | to read rows. |
| 1721 | |
| 1722 | For other comments, check _ma_read_dynamic_record() |
| 1723 | |
| 1724 | RETURN |
| 1725 | 0 OK |
| 1726 | != 0 Error number |
| 1727 | */ |
| 1728 | |
| 1729 | int _ma_read_rnd_dynamic_record(MARIA_HA *info, |
| 1730 | uchar *buf, |
| 1731 | MARIA_RECORD_POS filepos, |
| 1732 | my_bool skip_deleted_blocks) |
| 1733 | { |
| 1734 | int block_of_record; |
| 1735 | #ifdef MARIA_EXTERNAL_LOCKING |
| 1736 | int info_read; |
| 1737 | #endif |
| 1738 | uint left_len,b_type; |
| 1739 | uchar *UNINIT_VAR(to); |
| 1740 | MARIA_BLOCK_INFO block_info; |
| 1741 | MARIA_SHARE *share= info->s; |
| 1742 | DBUG_ENTER("_ma_read_rnd_dynamic_record" ); |
| 1743 | |
| 1744 | #ifdef MARIA_EXTERNAL_LOCKING |
| 1745 | info_read=0; |
| 1746 | #endif |
| 1747 | |
| 1748 | if (info->lock_type == F_UNLCK) |
| 1749 | { |
| 1750 | #ifndef UNSAFE_LOCKING |
| 1751 | #else |
| 1752 | info->tmp_lock_type=F_RDLCK; |
| 1753 | #endif |
| 1754 | } |
| 1755 | #ifdef MARIA_EXTERNAL_LOCKING |
| 1756 | else |
| 1757 | info_read=1; /* memory-keyinfoblock is ok */ |
| 1758 | #endif |
| 1759 | |
| 1760 | block_of_record= 0; /* First block of record is numbered as zero. */ |
| 1761 | block_info.second_read= 0; |
| 1762 | left_len=1; |
| 1763 | do |
| 1764 | { |
| 1765 | if (filepos >= info->state->data_file_length) |
| 1766 | { |
| 1767 | #ifdef MARIA_EXTERNAL_LOCKING |
| 1768 | if (!info_read) |
| 1769 | { /* Check if changed */ |
| 1770 | info_read=1; |
| 1771 | info->rec_cache.seek_not_done=1; |
| 1772 | if (_ma_state_info_read_dsk(share->kfile.file, &share->state)) |
| 1773 | goto panic; |
| 1774 | } |
| 1775 | if (filepos >= info->state->data_file_length) |
| 1776 | { |
| 1777 | my_errno= HA_ERR_END_OF_FILE; |
| 1778 | goto err; |
| 1779 | } |
| 1780 | #else |
| 1781 | my_errno= HA_ERR_END_OF_FILE; |
| 1782 | goto err; |
| 1783 | #endif |
| 1784 | } |
| 1785 | if (info->opt_flag & READ_CACHE_USED) |
| 1786 | { |
| 1787 | if (_ma_read_cache(info, &info->rec_cache, block_info.header, filepos, |
| 1788 | sizeof(block_info.header), |
| 1789 | (!block_of_record && skip_deleted_blocks ? |
| 1790 | READING_NEXT : 0) | READING_HEADER)) |
| 1791 | goto panic; |
| 1792 | b_type= _ma_get_block_info(info, &block_info,-1,filepos); |
| 1793 | } |
| 1794 | else |
| 1795 | { |
| 1796 | if (info->opt_flag & WRITE_CACHE_USED && |
| 1797 | info->rec_cache.pos_in_file < filepos + MARIA_BLOCK_INFO_HEADER_LENGTH && |
| 1798 | flush_io_cache(&info->rec_cache)) |
| 1799 | DBUG_RETURN(my_errno); |
| 1800 | info->rec_cache.seek_not_done=1; |
| 1801 | b_type= _ma_get_block_info(info, &block_info, info->dfile.file, filepos); |
| 1802 | } |
| 1803 | |
| 1804 | if (b_type & (BLOCK_DELETED | BLOCK_ERROR | BLOCK_SYNC_ERROR | |
| 1805 | BLOCK_FATAL_ERROR)) |
| 1806 | { |
| 1807 | if ((b_type & (BLOCK_DELETED | BLOCK_SYNC_ERROR)) |
| 1808 | && skip_deleted_blocks) |
| 1809 | { |
| 1810 | filepos=block_info.filepos+block_info.block_len; |
| 1811 | block_info.second_read=0; |
| 1812 | continue; /* Search after next_record */ |
| 1813 | } |
| 1814 | if (b_type & (BLOCK_DELETED | BLOCK_SYNC_ERROR)) |
| 1815 | { |
| 1816 | my_errno= HA_ERR_RECORD_DELETED; |
| 1817 | info->cur_row.lastpos= block_info.filepos; |
| 1818 | info->cur_row.nextpos= block_info.filepos+block_info.block_len; |
| 1819 | } |
| 1820 | goto err; |
| 1821 | } |
| 1822 | if (block_of_record == 0) /* First block */ |
| 1823 | { |
| 1824 | info->cur_row.total_length= block_info.rec_len; |
| 1825 | if (block_info.rec_len > (uint) share->base.max_pack_length) |
| 1826 | goto panic; |
| 1827 | info->cur_row.lastpos= filepos; |
| 1828 | if (share->base.blobs) |
| 1829 | { |
| 1830 | if (_ma_alloc_buffer(&info->rec_buff, &info->rec_buff_size, |
| 1831 | block_info.rec_len + |
| 1832 | info->s->base.extra_rec_buff_size)) |
| 1833 | goto err; |
| 1834 | } |
| 1835 | to= info->rec_buff; |
| 1836 | left_len=block_info.rec_len; |
| 1837 | } |
| 1838 | if (left_len < block_info.data_len) |
| 1839 | goto panic; /* Wrong linked record */ |
| 1840 | |
| 1841 | /* copy information that is already read */ |
| 1842 | { |
| 1843 | uint offset=(uint) (block_info.filepos - filepos); |
| 1844 | uint tmp_length= (sizeof(block_info.header) - offset); |
| 1845 | filepos=block_info.filepos; |
| 1846 | |
| 1847 | if (tmp_length > block_info.data_len) |
| 1848 | tmp_length= block_info.data_len; |
| 1849 | if (tmp_length) |
| 1850 | { |
| 1851 | memcpy(to, block_info.header+offset, tmp_length); |
| 1852 | block_info.data_len-=tmp_length; |
| 1853 | left_len-=tmp_length; |
| 1854 | to+=tmp_length; |
| 1855 | filepos+=tmp_length; |
| 1856 | } |
| 1857 | } |
| 1858 | /* read rest of record from file */ |
| 1859 | if (block_info.data_len) |
| 1860 | { |
| 1861 | if (info->opt_flag & READ_CACHE_USED) |
| 1862 | { |
| 1863 | if (_ma_read_cache(info, &info->rec_cache, to,filepos, |
| 1864 | block_info.data_len, |
| 1865 | (!block_of_record && skip_deleted_blocks) ? |
| 1866 | READING_NEXT : 0)) |
| 1867 | goto panic; |
| 1868 | } |
| 1869 | else |
| 1870 | { |
| 1871 | if (info->opt_flag & WRITE_CACHE_USED && |
| 1872 | info->rec_cache.pos_in_file < |
| 1873 | block_info.filepos + block_info.data_len && |
| 1874 | flush_io_cache(&info->rec_cache)) |
| 1875 | goto err; |
| 1876 | /* VOID(my_seek(info->dfile.file, filepos, MY_SEEK_SET, MYF(0))); */ |
| 1877 | if (mysql_file_read(info->dfile.file, to, block_info.data_len, MYF(MY_NABP))) |
| 1878 | { |
| 1879 | if (my_errno == HA_ERR_FILE_TOO_SHORT) |
| 1880 | { |
| 1881 | /* Unexpected end of file */ |
| 1882 | _ma_set_fatal_error(share, HA_ERR_WRONG_IN_RECORD); |
| 1883 | } |
| 1884 | goto err; |
| 1885 | } |
| 1886 | } |
| 1887 | } |
| 1888 | /* |
| 1889 | Increment block-of-record counter. If it was the first block, |
| 1890 | remember the position behind the block for the next call. |
| 1891 | */ |
| 1892 | if (block_of_record++ == 0) |
| 1893 | { |
| 1894 | info->cur_row.nextpos= block_info.filepos+block_info.block_len; |
| 1895 | skip_deleted_blocks=0; |
| 1896 | } |
| 1897 | left_len-=block_info.data_len; |
| 1898 | to+=block_info.data_len; |
| 1899 | filepos=block_info.next_filepos; |
| 1900 | } while (left_len); |
| 1901 | |
| 1902 | info->update|= HA_STATE_AKTIV | HA_STATE_KEY_CHANGED; |
| 1903 | fast_ma_writeinfo(info); |
| 1904 | if (_ma_rec_unpack(info,buf,info->rec_buff,block_info.rec_len) != |
| 1905 | MY_FILE_ERROR) |
| 1906 | DBUG_RETURN(0); |
| 1907 | DBUG_RETURN(my_errno); /* Wrong record */ |
| 1908 | |
| 1909 | panic: |
| 1910 | /* Something is fatal wrong */ |
| 1911 | _ma_set_fatal_error(share, HA_ERR_WRONG_IN_RECORD); |
| 1912 | err: |
| 1913 | fast_ma_writeinfo(info); |
| 1914 | DBUG_RETURN(my_errno); |
| 1915 | } |
| 1916 | |
| 1917 | |
| 1918 | /* Read and process header from a dynamic-record-file */ |
| 1919 | |
| 1920 | uint _ma_get_block_info(MARIA_HA *handler, MARIA_BLOCK_INFO *info, File file, |
| 1921 | my_off_t filepos) |
| 1922 | { |
| 1923 | uint return_val=0; |
| 1924 | uchar *=info->header; |
| 1925 | |
| 1926 | if (file >= 0) |
| 1927 | { |
| 1928 | /* |
| 1929 | We do not use my_pread() here because we want to have the file |
| 1930 | pointer set to the end of the header after this function. |
| 1931 | my_pread() may leave the file pointer untouched. |
| 1932 | */ |
| 1933 | mysql_file_seek(file,filepos,MY_SEEK_SET,MYF(0)); |
| 1934 | if (mysql_file_read(file, header, sizeof(info->header),MYF(0)) != |
| 1935 | sizeof(info->header)) |
| 1936 | { |
| 1937 | /* |
| 1938 | This is either an error or just reading at end of file. |
| 1939 | Don't give a fatal error for this case. |
| 1940 | */ |
| 1941 | my_errno= HA_ERR_WRONG_IN_RECORD; |
| 1942 | return BLOCK_ERROR; |
| 1943 | } |
| 1944 | } |
| 1945 | DBUG_DUMP("header" ,header,MARIA_BLOCK_INFO_HEADER_LENGTH); |
| 1946 | if (info->second_read) |
| 1947 | { |
| 1948 | if (info->header[0] <= 6 || info->header[0] == 13) |
| 1949 | return_val=BLOCK_SYNC_ERROR; |
| 1950 | } |
| 1951 | else |
| 1952 | { |
| 1953 | if (info->header[0] > 6 && info->header[0] != 13) |
| 1954 | return_val=BLOCK_SYNC_ERROR; |
| 1955 | } |
| 1956 | info->next_filepos= HA_OFFSET_ERROR; /* Dummy if no next block */ |
| 1957 | |
| 1958 | switch (info->header[0]) { |
| 1959 | case 0: |
| 1960 | if ((info->block_len=(uint) mi_uint3korr(header+1)) < |
| 1961 | MARIA_MIN_BLOCK_LENGTH || |
| 1962 | (info->block_len & (MARIA_DYN_ALIGN_SIZE -1))) |
| 1963 | goto err; |
| 1964 | info->filepos=filepos; |
| 1965 | info->next_filepos=mi_sizekorr(header+4); |
| 1966 | info->prev_filepos=mi_sizekorr(header+12); |
| 1967 | #if SIZEOF_OFF_T == 4 |
| 1968 | if ((mi_uint4korr(header+4) != 0 && |
| 1969 | (mi_uint4korr(header+4) != (ulong) ~0 || |
| 1970 | info->next_filepos != (ulong) ~0)) || |
| 1971 | (mi_uint4korr(header+12) != 0 && |
| 1972 | (mi_uint4korr(header+12) != (ulong) ~0 || |
| 1973 | info->prev_filepos != (ulong) ~0))) |
| 1974 | goto err; |
| 1975 | #endif |
| 1976 | return return_val | BLOCK_DELETED; /* Deleted block */ |
| 1977 | |
| 1978 | case 1: |
| 1979 | info->rec_len=info->data_len=info->block_len=mi_uint2korr(header+1); |
| 1980 | info->filepos=filepos+3; |
| 1981 | return return_val | BLOCK_FIRST | BLOCK_LAST; |
| 1982 | case 2: |
| 1983 | info->rec_len=info->data_len=info->block_len=mi_uint3korr(header+1); |
| 1984 | info->filepos=filepos+4; |
| 1985 | return return_val | BLOCK_FIRST | BLOCK_LAST; |
| 1986 | |
| 1987 | case 13: |
| 1988 | info->rec_len=mi_uint4korr(header+1); |
| 1989 | info->block_len=info->data_len=mi_uint3korr(header+5); |
| 1990 | info->next_filepos=mi_sizekorr(header+8); |
| 1991 | info->second_read=1; |
| 1992 | info->filepos=filepos+16; |
| 1993 | return return_val | BLOCK_FIRST; |
| 1994 | |
| 1995 | case 3: |
| 1996 | info->rec_len=info->data_len=mi_uint2korr(header+1); |
| 1997 | info->block_len=info->rec_len+ (uint) header[3]; |
| 1998 | info->filepos=filepos+4; |
| 1999 | return return_val | BLOCK_FIRST | BLOCK_LAST; |
| 2000 | case 4: |
| 2001 | info->rec_len=info->data_len=mi_uint3korr(header+1); |
| 2002 | info->block_len=info->rec_len+ (uint) header[4]; |
| 2003 | info->filepos=filepos+5; |
| 2004 | return return_val | BLOCK_FIRST | BLOCK_LAST; |
| 2005 | |
| 2006 | case 5: |
| 2007 | info->rec_len=mi_uint2korr(header+1); |
| 2008 | info->block_len=info->data_len=mi_uint2korr(header+3); |
| 2009 | info->next_filepos=mi_sizekorr(header+5); |
| 2010 | info->second_read=1; |
| 2011 | info->filepos=filepos+13; |
| 2012 | return return_val | BLOCK_FIRST; |
| 2013 | case 6: |
| 2014 | info->rec_len=mi_uint3korr(header+1); |
| 2015 | info->block_len=info->data_len=mi_uint3korr(header+4); |
| 2016 | info->next_filepos=mi_sizekorr(header+7); |
| 2017 | info->second_read=1; |
| 2018 | info->filepos=filepos+15; |
| 2019 | return return_val | BLOCK_FIRST; |
| 2020 | |
| 2021 | /* The following blocks are identical to 1-6 without rec_len */ |
| 2022 | case 7: |
| 2023 | info->data_len=info->block_len=mi_uint2korr(header+1); |
| 2024 | info->filepos=filepos+3; |
| 2025 | return return_val | BLOCK_LAST; |
| 2026 | case 8: |
| 2027 | info->data_len=info->block_len=mi_uint3korr(header+1); |
| 2028 | info->filepos=filepos+4; |
| 2029 | return return_val | BLOCK_LAST; |
| 2030 | |
| 2031 | case 9: |
| 2032 | info->data_len=mi_uint2korr(header+1); |
| 2033 | info->block_len=info->data_len+ (uint) header[3]; |
| 2034 | info->filepos=filepos+4; |
| 2035 | return return_val | BLOCK_LAST; |
| 2036 | case 10: |
| 2037 | info->data_len=mi_uint3korr(header+1); |
| 2038 | info->block_len=info->data_len+ (uint) header[4]; |
| 2039 | info->filepos=filepos+5; |
| 2040 | return return_val | BLOCK_LAST; |
| 2041 | |
| 2042 | case 11: |
| 2043 | info->data_len=info->block_len=mi_uint2korr(header+1); |
| 2044 | info->next_filepos=mi_sizekorr(header+3); |
| 2045 | info->second_read=1; |
| 2046 | info->filepos=filepos+11; |
| 2047 | return return_val; |
| 2048 | case 12: |
| 2049 | info->data_len=info->block_len=mi_uint3korr(header+1); |
| 2050 | info->next_filepos=mi_sizekorr(header+4); |
| 2051 | info->second_read=1; |
| 2052 | info->filepos=filepos+12; |
| 2053 | return return_val; |
| 2054 | } |
| 2055 | |
| 2056 | err: |
| 2057 | if (!handler->in_check_table) |
| 2058 | { |
| 2059 | /* We may be scanning the table for new rows; Don't give an error */ |
| 2060 | _ma_set_fatal_error(handler->s, HA_ERR_WRONG_IN_RECORD); |
| 2061 | } |
| 2062 | return BLOCK_ERROR; |
| 2063 | } |
| 2064 | |