1
by brian
clean slate |
1 |
/******************************************************
|
2 |
Transaction undo log record
|
|
3 |
||
4 |
(c) 1996 Innobase Oy
|
|
5 |
||
6 |
Created 3/26/1996 Heikki Tuuri
|
|
7 |
*******************************************************/
|
|
8 |
||
9 |
#include "trx0rec.h" |
|
10 |
||
11 |
#ifdef UNIV_NONINL
|
|
12 |
#include "trx0rec.ic" |
|
13 |
#endif
|
|
14 |
||
15 |
#include "fsp0fsp.h" |
|
16 |
#include "mach0data.h" |
|
17 |
#include "trx0rseg.h" |
|
18 |
#include "trx0trx.h" |
|
19 |
#include "trx0undo.h" |
|
20 |
#include "dict0dict.h" |
|
21 |
#include "ut0mem.h" |
|
22 |
#include "row0upd.h" |
|
23 |
#include "que0que.h" |
|
24 |
#include "trx0purge.h" |
|
25 |
#include "row0row.h" |
|
26 |
||
27 |
/*=========== UNDO LOG RECORD CREATION AND DECODING ====================*/
|
|
28 |
||
29 |
/**************************************************************************
|
|
30 |
Writes the mtr log entry of the inserted undo log record on the undo log
|
|
31 |
page. */
|
|
32 |
UNIV_INLINE
|
|
33 |
void
|
|
34 |
trx_undof_page_add_undo_rec_log( |
|
35 |
/*============================*/
|
|
36 |
page_t* undo_page, /* in: undo log page */ |
|
37 |
ulint old_free, /* in: start offset of the inserted entry */ |
|
38 |
ulint new_free, /* in: end offset of the entry */ |
|
39 |
mtr_t* mtr) /* in: mtr */ |
|
40 |
{
|
|
41 |
byte* log_ptr; |
|
42 |
const byte* log_end; |
|
43 |
ulint len; |
|
44 |
||
45 |
log_ptr = mlog_open(mtr, 11 + 13 + MLOG_BUF_MARGIN); |
|
46 |
||
47 |
if (log_ptr == NULL) { |
|
48 |
||
49 |
return; |
|
50 |
}
|
|
51 |
||
52 |
log_end = &log_ptr[11 + 13 + MLOG_BUF_MARGIN]; |
|
53 |
log_ptr = mlog_write_initial_log_record_fast( |
|
54 |
undo_page, MLOG_UNDO_INSERT, log_ptr, mtr); |
|
55 |
len = new_free - old_free - 4; |
|
56 |
||
57 |
mach_write_to_2(log_ptr, len); |
|
58 |
log_ptr += 2; |
|
59 |
||
60 |
if (log_ptr + len <= log_end) { |
|
61 |
memcpy(log_ptr, undo_page + old_free + 2, len); |
|
62 |
mlog_close(mtr, log_ptr + len); |
|
63 |
} else { |
|
64 |
mlog_close(mtr, log_ptr); |
|
65 |
mlog_catenate_string(mtr, undo_page + old_free + 2, len); |
|
66 |
}
|
|
67 |
}
|
|
68 |
||
69 |
/***************************************************************
|
|
70 |
Parses a redo log record of adding an undo log record. */
|
|
71 |
||
72 |
byte* |
|
73 |
trx_undo_parse_add_undo_rec( |
|
74 |
/*========================*/
|
|
75 |
/* out: end of log record or NULL */
|
|
76 |
byte* ptr, /* in: buffer */ |
|
77 |
byte* end_ptr,/* in: buffer end */ |
|
78 |
page_t* page) /* in: page or NULL */ |
|
79 |
{
|
|
80 |
ulint len; |
|
81 |
byte* rec; |
|
82 |
ulint first_free; |
|
83 |
||
84 |
if (end_ptr < ptr + 2) { |
|
85 |
||
86 |
return(NULL); |
|
87 |
}
|
|
88 |
||
89 |
len = mach_read_from_2(ptr); |
|
90 |
ptr += 2; |
|
91 |
||
92 |
if (end_ptr < ptr + len) { |
|
93 |
||
94 |
return(NULL); |
|
95 |
}
|
|
96 |
||
97 |
if (page == NULL) { |
|
98 |
||
99 |
return(ptr + len); |
|
100 |
}
|
|
101 |
||
102 |
first_free = mach_read_from_2(page + TRX_UNDO_PAGE_HDR |
|
103 |
+ TRX_UNDO_PAGE_FREE); |
|
104 |
rec = page + first_free; |
|
105 |
||
106 |
mach_write_to_2(rec, first_free + 4 + len); |
|
107 |
mach_write_to_2(rec + 2 + len, first_free); |
|
108 |
||
109 |
mach_write_to_2(page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE, |
|
110 |
first_free + 4 + len); |
|
111 |
ut_memcpy(rec + 2, ptr, len); |
|
112 |
||
113 |
return(ptr + len); |
|
114 |
}
|
|
115 |
||
116 |
/**************************************************************************
|
|
117 |
Calculates the free space left for extending an undo log record. */
|
|
118 |
UNIV_INLINE
|
|
119 |
ulint
|
|
120 |
trx_undo_left( |
|
121 |
/*==========*/
|
|
122 |
/* out: bytes left */
|
|
123 |
page_t* page, /* in: undo log page */ |
|
124 |
byte* ptr) /* in: pointer to page */ |
|
125 |
{
|
|
126 |
/* The '- 10' is a safety margin, in case we have some small
|
|
127 |
calculation error below */
|
|
128 |
||
129 |
return(UNIV_PAGE_SIZE - (ptr - page) - 10 - FIL_PAGE_DATA_END); |
|
130 |
}
|
|
131 |
||
132 |
/**************************************************************************
|
|
133 |
Reports in the undo log of an insert of a clustered index record. */
|
|
134 |
static
|
|
135 |
ulint
|
|
136 |
trx_undo_page_report_insert( |
|
137 |
/*========================*/
|
|
138 |
/* out: offset of the inserted entry
|
|
139 |
on the page if succeed, 0 if fail */
|
|
140 |
page_t* undo_page, /* in: undo log page */ |
|
141 |
trx_t* trx, /* in: transaction */ |
|
142 |
dict_index_t* index, /* in: clustered index */ |
|
143 |
dtuple_t* clust_entry, /* in: index entry which will be |
|
144 |
inserted to the clustered index */
|
|
145 |
mtr_t* mtr) /* in: mtr */ |
|
146 |
{
|
|
147 |
ulint first_free; |
|
148 |
byte* ptr; |
|
149 |
ulint len; |
|
150 |
dfield_t* field; |
|
151 |
ulint flen; |
|
152 |
ulint i; |
|
153 |
||
154 |
ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR |
|
155 |
+ TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT); |
|
156 |
||
157 |
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR |
|
158 |
+ TRX_UNDO_PAGE_FREE); |
|
159 |
ptr = undo_page + first_free; |
|
160 |
||
161 |
ut_ad(first_free <= UNIV_PAGE_SIZE); |
|
162 |
||
163 |
if (trx_undo_left(undo_page, ptr) < 30) { |
|
164 |
||
165 |
/* NOTE: the value 30 must be big enough such that the general
|
|
166 |
fields written below fit on the undo log page */
|
|
167 |
||
168 |
return(0); |
|
169 |
}
|
|
170 |
||
171 |
/* Reserve 2 bytes for the pointer to the next undo log record */
|
|
172 |
ptr += 2; |
|
173 |
||
174 |
/* Store first some general parameters to the undo log */
|
|
175 |
mach_write_to_1(ptr, TRX_UNDO_INSERT_REC); |
|
176 |
ptr++; |
|
177 |
||
178 |
len = mach_dulint_write_much_compressed(ptr, trx->undo_no); |
|
179 |
ptr += len; |
|
180 |
||
181 |
len = mach_dulint_write_much_compressed(ptr, (index->table)->id); |
|
182 |
ptr += len; |
|
183 |
/*----------------------------------------*/
|
|
184 |
/* Store then the fields required to uniquely determine the record
|
|
185 |
to be inserted in the clustered index */
|
|
186 |
||
187 |
for (i = 0; i < dict_index_get_n_unique(index); i++) { |
|
188 |
||
189 |
field = dtuple_get_nth_field(clust_entry, i); |
|
190 |
||
191 |
flen = dfield_get_len(field); |
|
192 |
||
193 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
194 |
||
195 |
return(0); |
|
196 |
}
|
|
197 |
||
198 |
len = mach_write_compressed(ptr, flen); |
|
199 |
ptr += len; |
|
200 |
||
201 |
if (flen != UNIV_SQL_NULL) { |
|
202 |
if (trx_undo_left(undo_page, ptr) < flen) { |
|
203 |
||
204 |
return(0); |
|
205 |
}
|
|
206 |
||
207 |
ut_memcpy(ptr, dfield_get_data(field), flen); |
|
208 |
ptr += flen; |
|
209 |
}
|
|
210 |
}
|
|
211 |
||
212 |
if (trx_undo_left(undo_page, ptr) < 2) { |
|
213 |
||
214 |
return(0); |
|
215 |
}
|
|
216 |
||
217 |
/*----------------------------------------*/
|
|
218 |
/* Write pointers to the previous and the next undo log records */
|
|
219 |
||
220 |
if (trx_undo_left(undo_page, ptr) < 2) { |
|
221 |
||
222 |
return(0); |
|
223 |
}
|
|
224 |
||
225 |
mach_write_to_2(ptr, first_free); |
|
226 |
ptr += 2; |
|
227 |
||
228 |
mach_write_to_2(undo_page + first_free, ptr - undo_page); |
|
229 |
||
230 |
mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE, |
|
231 |
ptr - undo_page); |
|
232 |
||
233 |
/* Write the log entry to the REDO log of this change in the UNDO
|
|
234 |
log */
|
|
235 |
trx_undof_page_add_undo_rec_log(undo_page, first_free, |
|
236 |
ptr - undo_page, mtr); |
|
237 |
return(first_free); |
|
238 |
}
|
|
239 |
||
240 |
/**************************************************************************
|
|
241 |
Reads from an undo log record the general parameters. */
|
|
242 |
||
243 |
byte* |
|
244 |
trx_undo_rec_get_pars( |
|
245 |
/*==================*/
|
|
246 |
/* out: remaining part of undo log
|
|
247 |
record after reading these values */
|
|
248 |
trx_undo_rec_t* undo_rec, /* in: undo log record */ |
|
249 |
ulint* type, /* out: undo record type: |
|
250 |
TRX_UNDO_INSERT_REC, ... */
|
|
251 |
ulint* cmpl_info, /* out: compiler info, relevant only |
|
252 |
for update type records */
|
|
253 |
ibool* updated_extern, /* out: TRUE if we updated an |
|
254 |
externally stored fild */
|
|
255 |
dulint* undo_no, /* out: undo log record number */ |
|
256 |
dulint* table_id) /* out: table id */ |
|
257 |
{
|
|
258 |
byte* ptr; |
|
259 |
ulint len; |
|
260 |
ulint type_cmpl; |
|
261 |
||
262 |
ptr = undo_rec + 2; |
|
263 |
||
264 |
type_cmpl = mach_read_from_1(ptr); |
|
265 |
ptr++; |
|
266 |
||
267 |
if (type_cmpl & TRX_UNDO_UPD_EXTERN) { |
|
268 |
*updated_extern = TRUE; |
|
269 |
type_cmpl -= TRX_UNDO_UPD_EXTERN; |
|
270 |
} else { |
|
271 |
*updated_extern = FALSE; |
|
272 |
}
|
|
273 |
||
274 |
*type = type_cmpl & (TRX_UNDO_CMPL_INFO_MULT - 1); |
|
275 |
*cmpl_info = type_cmpl / TRX_UNDO_CMPL_INFO_MULT; |
|
276 |
||
277 |
*undo_no = mach_dulint_read_much_compressed(ptr); |
|
278 |
len = mach_dulint_get_much_compressed_size(*undo_no); |
|
279 |
ptr += len; |
|
280 |
||
281 |
*table_id = mach_dulint_read_much_compressed(ptr); |
|
282 |
len = mach_dulint_get_much_compressed_size(*table_id); |
|
283 |
ptr += len; |
|
284 |
||
285 |
return(ptr); |
|
286 |
}
|
|
287 |
||
288 |
/**************************************************************************
|
|
289 |
Reads from an undo log record a stored column value. */
|
|
290 |
static
|
|
291 |
byte* |
|
292 |
trx_undo_rec_get_col_val( |
|
293 |
/*=====================*/
|
|
294 |
/* out: remaining part of undo log record after
|
|
295 |
reading these values */
|
|
296 |
byte* ptr, /* in: pointer to remaining part of undo log record */ |
|
297 |
byte** field, /* out: pointer to stored field */ |
|
298 |
ulint* len) /* out: length of the field, or UNIV_SQL_NULL */ |
|
299 |
{
|
|
300 |
*len = mach_read_compressed(ptr); |
|
301 |
ptr += mach_get_compressed_size(*len); |
|
302 |
||
303 |
*field = ptr; |
|
304 |
||
305 |
if (*len != UNIV_SQL_NULL) { |
|
306 |
if (*len >= UNIV_EXTERN_STORAGE_FIELD) { |
|
307 |
ptr += (*len - UNIV_EXTERN_STORAGE_FIELD); |
|
308 |
} else { |
|
309 |
ptr += *len; |
|
310 |
}
|
|
311 |
}
|
|
312 |
||
313 |
return(ptr); |
|
314 |
}
|
|
315 |
||
316 |
/***********************************************************************
|
|
317 |
Builds a row reference from an undo log record. */
|
|
318 |
||
319 |
byte* |
|
320 |
trx_undo_rec_get_row_ref( |
|
321 |
/*=====================*/
|
|
322 |
/* out: pointer to remaining part of undo
|
|
323 |
record */
|
|
324 |
byte* ptr, /* in: remaining part of a copy of an undo log |
|
325 |
record, at the start of the row reference;
|
|
326 |
NOTE that this copy of the undo log record must
|
|
327 |
be preserved as long as the row reference is
|
|
328 |
used, as we do NOT copy the data in the
|
|
329 |
record! */
|
|
330 |
dict_index_t* index, /* in: clustered index */ |
|
331 |
dtuple_t** ref, /* out, own: row reference */ |
|
332 |
mem_heap_t* heap) /* in: memory heap from which the memory |
|
333 |
needed is allocated */
|
|
334 |
{
|
|
335 |
dfield_t* dfield; |
|
336 |
byte* field; |
|
337 |
ulint len; |
|
338 |
ulint ref_len; |
|
339 |
ulint i; |
|
340 |
||
341 |
ut_ad(index && ptr && ref && heap); |
|
342 |
ut_a(index->type & DICT_CLUSTERED); |
|
343 |
||
344 |
ref_len = dict_index_get_n_unique(index); |
|
345 |
||
346 |
*ref = dtuple_create(heap, ref_len); |
|
347 |
||
348 |
dict_index_copy_types(*ref, index, ref_len); |
|
349 |
||
350 |
for (i = 0; i < ref_len; i++) { |
|
351 |
dfield = dtuple_get_nth_field(*ref, i); |
|
352 |
||
353 |
ptr = trx_undo_rec_get_col_val(ptr, &field, &len); |
|
354 |
||
355 |
dfield_set_data(dfield, field, len); |
|
356 |
}
|
|
357 |
||
358 |
return(ptr); |
|
359 |
}
|
|
360 |
||
361 |
/***********************************************************************
|
|
362 |
Skips a row reference from an undo log record. */
|
|
363 |
||
364 |
byte* |
|
365 |
trx_undo_rec_skip_row_ref( |
|
366 |
/*======================*/
|
|
367 |
/* out: pointer to remaining part of undo
|
|
368 |
record */
|
|
369 |
byte* ptr, /* in: remaining part in update undo log |
|
370 |
record, at the start of the row reference */
|
|
371 |
dict_index_t* index) /* in: clustered index */ |
|
372 |
{
|
|
373 |
byte* field; |
|
374 |
ulint len; |
|
375 |
ulint ref_len; |
|
376 |
ulint i; |
|
377 |
||
378 |
ut_ad(index && ptr); |
|
379 |
ut_a(index->type & DICT_CLUSTERED); |
|
380 |
||
381 |
ref_len = dict_index_get_n_unique(index); |
|
382 |
||
383 |
for (i = 0; i < ref_len; i++) { |
|
384 |
ptr = trx_undo_rec_get_col_val(ptr, &field, &len); |
|
385 |
}
|
|
386 |
||
387 |
return(ptr); |
|
388 |
}
|
|
389 |
||
390 |
/**************************************************************************
|
|
391 |
Reports in the undo log of an update or delete marking of a clustered index
|
|
392 |
record. */
|
|
393 |
static
|
|
394 |
ulint
|
|
395 |
trx_undo_page_report_modify( |
|
396 |
/*========================*/
|
|
397 |
/* out: byte offset of the inserted
|
|
398 |
undo log entry on the page if succeed,
|
|
399 |
0 if fail */
|
|
400 |
page_t* undo_page, /* in: undo log page */ |
|
401 |
trx_t* trx, /* in: transaction */ |
|
402 |
dict_index_t* index, /* in: clustered index where update or |
|
403 |
delete marking is done */
|
|
404 |
rec_t* rec, /* in: clustered index record which |
|
405 |
has NOT yet been modified */
|
|
406 |
const ulint* offsets, /* in: rec_get_offsets(rec, index) */ |
|
407 |
upd_t* update, /* in: update vector which tells the |
|
408 |
columns to be updated; in the case of
|
|
409 |
a delete, this should be set to NULL */
|
|
410 |
ulint cmpl_info, /* in: compiler info on secondary |
|
411 |
index updates */
|
|
412 |
mtr_t* mtr) /* in: mtr */ |
|
413 |
{
|
|
414 |
dict_table_t* table; |
|
415 |
upd_field_t* upd_field; |
|
416 |
ulint first_free; |
|
417 |
byte* ptr; |
|
418 |
ulint len; |
|
419 |
byte* field; |
|
420 |
ulint flen; |
|
421 |
ulint pos; |
|
422 |
dulint roll_ptr; |
|
423 |
dulint trx_id; |
|
424 |
ulint bits; |
|
425 |
ulint col_no; |
|
426 |
byte* old_ptr; |
|
427 |
ulint type_cmpl; |
|
428 |
byte* type_cmpl_ptr; |
|
429 |
ulint i; |
|
430 |
||
431 |
ut_a(index->type & DICT_CLUSTERED); |
|
432 |
ut_ad(rec_offs_validate(rec, index, offsets)); |
|
433 |
ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR |
|
434 |
+ TRX_UNDO_PAGE_TYPE) == TRX_UNDO_UPDATE); |
|
435 |
table = index->table; |
|
436 |
||
437 |
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR |
|
438 |
+ TRX_UNDO_PAGE_FREE); |
|
439 |
ptr = undo_page + first_free; |
|
440 |
||
441 |
ut_ad(first_free <= UNIV_PAGE_SIZE); |
|
442 |
||
443 |
if (trx_undo_left(undo_page, ptr) < 50) { |
|
444 |
||
445 |
/* NOTE: the value 50 must be big enough so that the general
|
|
446 |
fields written below fit on the undo log page */
|
|
447 |
||
448 |
return(0); |
|
449 |
}
|
|
450 |
||
451 |
/* Reserve 2 bytes for the pointer to the next undo log record */
|
|
452 |
ptr += 2; |
|
453 |
||
454 |
/* Store first some general parameters to the undo log */
|
|
455 |
||
456 |
if (update) { |
|
457 |
if (rec_get_deleted_flag(rec, dict_table_is_comp(table))) { |
|
458 |
type_cmpl = TRX_UNDO_UPD_DEL_REC; |
|
459 |
} else { |
|
460 |
type_cmpl = TRX_UNDO_UPD_EXIST_REC; |
|
461 |
}
|
|
462 |
} else { |
|
463 |
type_cmpl = TRX_UNDO_DEL_MARK_REC; |
|
464 |
}
|
|
465 |
||
466 |
type_cmpl = type_cmpl | (cmpl_info * TRX_UNDO_CMPL_INFO_MULT); |
|
467 |
||
468 |
mach_write_to_1(ptr, type_cmpl); |
|
469 |
||
470 |
type_cmpl_ptr = ptr; |
|
471 |
||
472 |
ptr++; |
|
473 |
len = mach_dulint_write_much_compressed(ptr, trx->undo_no); |
|
474 |
ptr += len; |
|
475 |
||
476 |
len = mach_dulint_write_much_compressed(ptr, table->id); |
|
477 |
ptr += len; |
|
478 |
||
479 |
/*----------------------------------------*/
|
|
480 |
/* Store the state of the info bits */
|
|
481 |
||
482 |
bits = rec_get_info_bits(rec, dict_table_is_comp(table)); |
|
483 |
mach_write_to_1(ptr, bits); |
|
484 |
ptr += 1; |
|
485 |
||
486 |
/* Store the values of the system columns */
|
|
487 |
field = rec_get_nth_field(rec, offsets, |
|
488 |
dict_index_get_sys_col_pos( |
|
489 |
index, DATA_TRX_ID), &len); |
|
490 |
ut_ad(len == DATA_TRX_ID_LEN); |
|
491 |
trx_id = trx_read_trx_id(field); |
|
492 |
field = rec_get_nth_field(rec, offsets, |
|
493 |
dict_index_get_sys_col_pos( |
|
494 |
index, DATA_ROLL_PTR), &len); |
|
495 |
ut_ad(len == DATA_ROLL_PTR_LEN); |
|
496 |
roll_ptr = trx_read_roll_ptr(field); |
|
497 |
||
498 |
len = mach_dulint_write_compressed(ptr, trx_id); |
|
499 |
ptr += len; |
|
500 |
||
501 |
len = mach_dulint_write_compressed(ptr, roll_ptr); |
|
502 |
ptr += len; |
|
503 |
||
504 |
/*----------------------------------------*/
|
|
505 |
/* Store then the fields required to uniquely determine the
|
|
506 |
record which will be modified in the clustered index */
|
|
507 |
||
508 |
for (i = 0; i < dict_index_get_n_unique(index); i++) { |
|
509 |
||
510 |
field = rec_get_nth_field(rec, offsets, i, &flen); |
|
511 |
||
512 |
if (trx_undo_left(undo_page, ptr) < 4) { |
|
513 |
||
514 |
return(0); |
|
515 |
}
|
|
516 |
||
517 |
len = mach_write_compressed(ptr, flen); |
|
518 |
ptr += len; |
|
519 |
||
520 |
if (flen != UNIV_SQL_NULL) { |
|
521 |
if (trx_undo_left(undo_page, ptr) < flen) { |
|
522 |
||
523 |
return(0); |
|
524 |
}
|
|
525 |
||
526 |
ut_memcpy(ptr, field, flen); |
|
527 |
ptr += flen; |
|
528 |
}
|
|
529 |
}
|
|
530 |
||
531 |
/*----------------------------------------*/
|
|
532 |
/* Save to the undo log the old values of the columns to be updated. */
|
|
533 |
||
534 |
if (update) { |
|
535 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
536 |
||
537 |
return(0); |
|
538 |
}
|
|
539 |
||
540 |
len = mach_write_compressed(ptr, upd_get_n_fields(update)); |
|
541 |
ptr += len; |
|
542 |
||
543 |
for (i = 0; i < upd_get_n_fields(update); i++) { |
|
544 |
||
545 |
upd_field = upd_get_nth_field(update, i); |
|
546 |
pos = upd_field->field_no; |
|
547 |
||
548 |
/* Write field number to undo log */
|
|
549 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
550 |
||
551 |
return(0); |
|
552 |
}
|
|
553 |
||
554 |
len = mach_write_compressed(ptr, pos); |
|
555 |
ptr += len; |
|
556 |
||
557 |
/* Save the old value of field */
|
|
558 |
field = rec_get_nth_field(rec, offsets, pos, &flen); |
|
559 |
||
560 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
561 |
||
562 |
return(0); |
|
563 |
}
|
|
564 |
||
565 |
if (rec_offs_nth_extern(offsets, pos)) { |
|
566 |
/* If a field has external storage, we add
|
|
567 |
to flen the flag */
|
|
568 |
||
569 |
len = mach_write_compressed( |
|
570 |
ptr, |
|
571 |
UNIV_EXTERN_STORAGE_FIELD + flen); |
|
572 |
||
573 |
/* Notify purge that it eventually has to
|
|
574 |
free the old externally stored field */
|
|
575 |
||
576 |
trx->update_undo->del_marks = TRUE; |
|
577 |
||
578 |
*type_cmpl_ptr = *type_cmpl_ptr |
|
579 |
| TRX_UNDO_UPD_EXTERN; |
|
580 |
} else { |
|
581 |
len = mach_write_compressed(ptr, flen); |
|
582 |
}
|
|
583 |
||
584 |
ptr += len; |
|
585 |
||
586 |
if (flen != UNIV_SQL_NULL) { |
|
587 |
if (trx_undo_left(undo_page, ptr) < flen) { |
|
588 |
||
589 |
return(0); |
|
590 |
}
|
|
591 |
||
592 |
ut_memcpy(ptr, field, flen); |
|
593 |
ptr += flen; |
|
594 |
}
|
|
595 |
}
|
|
596 |
}
|
|
597 |
||
598 |
/*----------------------------------------*/
|
|
599 |
/* In the case of a delete marking, and also in the case of an update
|
|
600 |
where any ordering field of any index changes, store the values of all
|
|
601 |
columns which occur as ordering fields in any index. This info is used
|
|
602 |
in the purge of old versions where we use it to build and search the
|
|
603 |
delete marked index records, to look if we can remove them from the
|
|
604 |
index tree. Note that starting from 4.0.14 also externally stored
|
|
605 |
fields can be ordering in some index. But we always store at least
|
|
606 |
384 first bytes locally to the clustered index record, which means
|
|
607 |
we can construct the column prefix fields in the index from the
|
|
608 |
stored data. */
|
|
609 |
||
610 |
if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) { |
|
611 |
||
612 |
trx->update_undo->del_marks = TRUE; |
|
613 |
||
614 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
615 |
||
616 |
return(0); |
|
617 |
}
|
|
618 |
||
619 |
old_ptr = ptr; |
|
620 |
||
621 |
/* Reserve 2 bytes to write the number of bytes the stored
|
|
622 |
fields take in this undo record */
|
|
623 |
||
624 |
ptr += 2; |
|
625 |
||
626 |
for (col_no = 0; col_no < dict_table_get_n_cols(table); |
|
627 |
col_no++) { |
|
628 |
||
629 |
const dict_col_t* col |
|
630 |
= dict_table_get_nth_col(table, col_no); |
|
631 |
||
632 |
if (col->ord_part > 0) { |
|
633 |
||
634 |
pos = dict_index_get_nth_col_pos(index, |
|
635 |
col_no); |
|
636 |
||
637 |
/* Write field number to undo log */
|
|
638 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
639 |
||
640 |
return(0); |
|
641 |
}
|
|
642 |
||
643 |
len = mach_write_compressed(ptr, pos); |
|
644 |
ptr += len; |
|
645 |
||
646 |
/* Save the old value of field */
|
|
647 |
field = rec_get_nth_field(rec, offsets, pos, |
|
648 |
&flen); |
|
649 |
||
650 |
if (trx_undo_left(undo_page, ptr) < 5) { |
|
651 |
||
652 |
return(0); |
|
653 |
}
|
|
654 |
||
655 |
len = mach_write_compressed(ptr, flen); |
|
656 |
ptr += len; |
|
657 |
||
658 |
if (flen != UNIV_SQL_NULL) { |
|
659 |
if (trx_undo_left(undo_page, ptr) |
|
660 |
< flen) { |
|
661 |
||
662 |
return(0); |
|
663 |
}
|
|
664 |
||
665 |
ut_memcpy(ptr, field, flen); |
|
666 |
ptr += flen; |
|
667 |
}
|
|
668 |
}
|
|
669 |
}
|
|
670 |
||
671 |
mach_write_to_2(old_ptr, ptr - old_ptr); |
|
672 |
}
|
|
673 |
||
674 |
/*----------------------------------------*/
|
|
675 |
/* Write pointers to the previous and the next undo log records */
|
|
676 |
if (trx_undo_left(undo_page, ptr) < 2) { |
|
677 |
||
678 |
return(0); |
|
679 |
}
|
|
680 |
||
681 |
mach_write_to_2(ptr, first_free); |
|
682 |
ptr += 2; |
|
683 |
mach_write_to_2(undo_page + first_free, ptr - undo_page); |
|
684 |
||
685 |
mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE, |
|
686 |
ptr - undo_page); |
|
687 |
||
688 |
/* Write to the REDO log about this change in the UNDO log */
|
|
689 |
||
690 |
trx_undof_page_add_undo_rec_log(undo_page, first_free, |
|
691 |
ptr - undo_page, mtr); |
|
692 |
return(first_free); |
|
693 |
}
|
|
694 |
||
695 |
/**************************************************************************
|
|
696 |
Reads from an undo log update record the system field values of the old
|
|
697 |
version. */
|
|
698 |
||
699 |
byte* |
|
700 |
trx_undo_update_rec_get_sys_cols( |
|
701 |
/*=============================*/
|
|
702 |
/* out: remaining part of undo log
|
|
703 |
record after reading these values */
|
|
704 |
byte* ptr, /* in: remaining part of undo log |
|
705 |
record after reading general
|
|
706 |
parameters */
|
|
707 |
dulint* trx_id, /* out: trx id */ |
|
708 |
dulint* roll_ptr, /* out: roll ptr */ |
|
709 |
ulint* info_bits) /* out: info bits state */ |
|
710 |
{
|
|
711 |
ulint len; |
|
712 |
||
713 |
/* Read the state of the info bits */
|
|
714 |
*info_bits = mach_read_from_1(ptr); |
|
715 |
ptr += 1; |
|
716 |
||
717 |
/* Read the values of the system columns */
|
|
718 |
||
719 |
*trx_id = mach_dulint_read_compressed(ptr); |
|
720 |
len = mach_dulint_get_compressed_size(*trx_id); |
|
721 |
ptr += len; |
|
722 |
||
723 |
*roll_ptr = mach_dulint_read_compressed(ptr); |
|
724 |
len = mach_dulint_get_compressed_size(*roll_ptr); |
|
725 |
ptr += len; |
|
726 |
||
727 |
return(ptr); |
|
728 |
}
|
|
729 |
||
730 |
/**************************************************************************
|
|
731 |
Reads from an update undo log record the number of updated fields. */
|
|
732 |
UNIV_INLINE
|
|
733 |
byte* |
|
734 |
trx_undo_update_rec_get_n_upd_fields( |
|
735 |
/*=================================*/
|
|
736 |
/* out: remaining part of undo log record after
|
|
737 |
reading this value */
|
|
738 |
byte* ptr, /* in: pointer to remaining part of undo log record */ |
|
739 |
ulint* n) /* out: number of fields */ |
|
740 |
{
|
|
741 |
*n = mach_read_compressed(ptr); |
|
742 |
ptr += mach_get_compressed_size(*n); |
|
743 |
||
744 |
return(ptr); |
|
745 |
}
|
|
746 |
||
747 |
/**************************************************************************
|
|
748 |
Reads from an update undo log record a stored field number. */
|
|
749 |
UNIV_INLINE
|
|
750 |
byte* |
|
751 |
trx_undo_update_rec_get_field_no( |
|
752 |
/*=============================*/
|
|
753 |
/* out: remaining part of undo log record after
|
|
754 |
reading this value */
|
|
755 |
byte* ptr, /* in: pointer to remaining part of undo log record */ |
|
756 |
ulint* field_no)/* out: field number */ |
|
757 |
{
|
|
758 |
*field_no = mach_read_compressed(ptr); |
|
759 |
ptr += mach_get_compressed_size(*field_no); |
|
760 |
||
761 |
return(ptr); |
|
762 |
}
|
|
763 |
||
764 |
/***********************************************************************
|
|
765 |
Builds an update vector based on a remaining part of an undo log record. */
|
|
766 |
||
767 |
byte* |
|
768 |
trx_undo_update_rec_get_update( |
|
769 |
/*===========================*/
|
|
770 |
/* out: remaining part of the record,
|
|
771 |
NULL if an error detected, which means that
|
|
772 |
the record is corrupted */
|
|
773 |
byte* ptr, /* in: remaining part in update undo log |
|
774 |
record, after reading the row reference
|
|
775 |
NOTE that this copy of the undo log record must
|
|
776 |
be preserved as long as the update vector is
|
|
777 |
used, as we do NOT copy the data in the
|
|
778 |
record! */
|
|
779 |
dict_index_t* index, /* in: clustered index */ |
|
780 |
ulint type, /* in: TRX_UNDO_UPD_EXIST_REC, |
|
781 |
TRX_UNDO_UPD_DEL_REC, or
|
|
782 |
TRX_UNDO_DEL_MARK_REC; in the last case,
|
|
783 |
only trx id and roll ptr fields are added to
|
|
784 |
the update vector */
|
|
785 |
dulint trx_id, /* in: transaction id from this undo record */ |
|
786 |
dulint roll_ptr,/* in: roll pointer from this undo record */ |
|
787 |
ulint info_bits,/* in: info bits from this undo record */ |
|
788 |
trx_t* trx, /* in: transaction */ |
|
789 |
mem_heap_t* heap, /* in: memory heap from which the memory |
|
790 |
needed is allocated */
|
|
791 |
upd_t** upd) /* out, own: update vector */ |
|
792 |
{
|
|
793 |
upd_field_t* upd_field; |
|
794 |
upd_t* update; |
|
795 |
ulint n_fields; |
|
796 |
byte* buf; |
|
797 |
byte* field; |
|
798 |
ulint len; |
|
799 |
ulint field_no; |
|
800 |
ulint i; |
|
801 |
||
802 |
ut_a(index->type & DICT_CLUSTERED); |
|
803 |
||
804 |
if (type != TRX_UNDO_DEL_MARK_REC) { |
|
805 |
ptr = trx_undo_update_rec_get_n_upd_fields(ptr, &n_fields); |
|
806 |
} else { |
|
807 |
n_fields = 0; |
|
808 |
}
|
|
809 |
||
810 |
update = upd_create(n_fields + 2, heap); |
|
811 |
||
812 |
update->info_bits = info_bits; |
|
813 |
||
814 |
/* Store first trx id and roll ptr to update vector */
|
|
815 |
||
816 |
upd_field = upd_get_nth_field(update, n_fields); |
|
817 |
buf = mem_heap_alloc(heap, DATA_TRX_ID_LEN); |
|
818 |
trx_write_trx_id(buf, trx_id); |
|
819 |
||
820 |
upd_field_set_field_no(upd_field, |
|
821 |
dict_index_get_sys_col_pos(index, DATA_TRX_ID), |
|
822 |
index, trx); |
|
823 |
dfield_set_data(&(upd_field->new_val), buf, DATA_TRX_ID_LEN); |
|
824 |
||
825 |
upd_field = upd_get_nth_field(update, n_fields + 1); |
|
826 |
buf = mem_heap_alloc(heap, DATA_ROLL_PTR_LEN); |
|
827 |
trx_write_roll_ptr(buf, roll_ptr); |
|
828 |
||
829 |
upd_field_set_field_no( |
|
830 |
upd_field, dict_index_get_sys_col_pos(index, DATA_ROLL_PTR), |
|
831 |
index, trx); |
|
832 |
dfield_set_data(&(upd_field->new_val), buf, DATA_ROLL_PTR_LEN); |
|
833 |
||
834 |
/* Store then the updated ordinary columns to the update vector */
|
|
835 |
||
836 |
for (i = 0; i < n_fields; i++) { |
|
837 |
||
838 |
ptr = trx_undo_update_rec_get_field_no(ptr, &field_no); |
|
839 |
||
840 |
if (field_no >= dict_index_get_n_fields(index)) { |
|
841 |
fprintf(stderr, |
|
842 |
"InnoDB: Error: trying to access"
|
|
843 |
" update undo rec field %lu in ", |
|
844 |
(ulong) field_no); |
|
845 |
dict_index_name_print(stderr, trx, index); |
|
846 |
fprintf(stderr, "\n" |
|
847 |
"InnoDB: but index has only %lu fields\n" |
|
848 |
"InnoDB: Submit a detailed bug report"
|
|
849 |
" to http://bugs.mysql.com\n" |
|
850 |
"InnoDB: Run also CHECK TABLE ", |
|
851 |
(ulong) dict_index_get_n_fields(index)); |
|
852 |
ut_print_name(stderr, trx, TRUE, index->table_name); |
|
853 |
fprintf(stderr, "\n" |
|
854 |
"InnoDB: n_fields = %lu, i = %lu, ptr %p\n", |
|
855 |
(ulong) n_fields, (ulong) i, ptr); |
|
856 |
return(NULL); |
|
857 |
}
|
|
858 |
||
859 |
ptr = trx_undo_rec_get_col_val(ptr, &field, &len); |
|
860 |
||
861 |
upd_field = upd_get_nth_field(update, i); |
|
862 |
||
863 |
upd_field_set_field_no(upd_field, field_no, index, trx); |
|
864 |
||
865 |
if (len != UNIV_SQL_NULL && len >= UNIV_EXTERN_STORAGE_FIELD) { |
|
866 |
||
867 |
upd_field->extern_storage = TRUE; |
|
868 |
||
869 |
len -= UNIV_EXTERN_STORAGE_FIELD; |
|
870 |
}
|
|
871 |
||
872 |
dfield_set_data(&(upd_field->new_val), field, len); |
|
873 |
}
|
|
874 |
||
875 |
*upd = update; |
|
876 |
||
877 |
return(ptr); |
|
878 |
}
|
|
879 |
||
880 |
/***********************************************************************
|
|
881 |
Builds a partial row from an update undo log record. It contains the
|
|
882 |
columns which occur as ordering in any index of the table. */
|
|
883 |
||
884 |
byte* |
|
885 |
trx_undo_rec_get_partial_row( |
|
886 |
/*=========================*/
|
|
887 |
/* out: pointer to remaining part of undo
|
|
888 |
record */
|
|
889 |
byte* ptr, /* in: remaining part in update undo log |
|
890 |
record of a suitable type, at the start of
|
|
891 |
the stored index columns;
|
|
892 |
NOTE that this copy of the undo log record must
|
|
893 |
be preserved as long as the partial row is
|
|
894 |
used, as we do NOT copy the data in the
|
|
895 |
record! */
|
|
896 |
dict_index_t* index, /* in: clustered index */ |
|
897 |
dtuple_t** row, /* out, own: partial row */ |
|
898 |
mem_heap_t* heap) /* in: memory heap from which the memory |
|
899 |
needed is allocated */
|
|
900 |
{
|
|
901 |
dfield_t* dfield; |
|
902 |
byte* field; |
|
903 |
ulint len; |
|
904 |
ulint field_no; |
|
905 |
ulint col_no; |
|
906 |
ulint row_len; |
|
907 |
ulint total_len; |
|
908 |
byte* start_ptr; |
|
909 |
ulint i; |
|
910 |
||
911 |
ut_ad(index && ptr && row && heap); |
|
912 |
||
913 |
row_len = dict_table_get_n_cols(index->table); |
|
914 |
||
915 |
*row = dtuple_create(heap, row_len); |
|
916 |
||
917 |
dict_table_copy_types(*row, index->table); |
|
918 |
||
919 |
start_ptr = ptr; |
|
920 |
||
921 |
total_len = mach_read_from_2(ptr); |
|
922 |
ptr += 2; |
|
923 |
||
924 |
for (i = 0;; i++) { |
|
925 |
||
926 |
if (ptr == start_ptr + total_len) { |
|
927 |
||
928 |
break; |
|
929 |
}
|
|
930 |
||
931 |
ptr = trx_undo_update_rec_get_field_no(ptr, &field_no); |
|
932 |
||
933 |
col_no = dict_index_get_nth_col_no(index, field_no); |
|
934 |
||
935 |
ptr = trx_undo_rec_get_col_val(ptr, &field, &len); |
|
936 |
||
937 |
dfield = dtuple_get_nth_field(*row, col_no); |
|
938 |
||
939 |
dfield_set_data(dfield, field, len); |
|
940 |
}
|
|
941 |
||
942 |
return(ptr); |
|
943 |
}
|
|
944 |
||
945 |
/***************************************************************************
|
|
946 |
Erases the unused undo log page end. */
|
|
947 |
static
|
|
948 |
void
|
|
949 |
trx_undo_erase_page_end( |
|
950 |
/*====================*/
|
|
951 |
page_t* undo_page, /* in: undo page whose end to erase */ |
|
952 |
mtr_t* mtr) /* in: mtr */ |
|
953 |
{
|
|
954 |
ulint first_free; |
|
955 |
||
956 |
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR |
|
957 |
+ TRX_UNDO_PAGE_FREE); |
|
958 |
memset(undo_page + first_free, 0xff, |
|
959 |
(UNIV_PAGE_SIZE - FIL_PAGE_DATA_END) - first_free); |
|
960 |
||
961 |
mlog_write_initial_log_record(undo_page, MLOG_UNDO_ERASE_END, mtr); |
|
962 |
}
|
|
963 |
||
964 |
/***************************************************************
|
|
965 |
Parses a redo log record of erasing of an undo page end. */
|
|
966 |
||
967 |
byte* |
|
968 |
trx_undo_parse_erase_page_end( |
|
969 |
/*==========================*/
|
|
970 |
/* out: end of log record or NULL */
|
|
971 |
byte* ptr, /* in: buffer */ |
|
972 |
byte* end_ptr __attribute__((unused)), /* in: buffer end */ |
|
973 |
page_t* page, /* in: page or NULL */ |
|
974 |
mtr_t* mtr) /* in: mtr or NULL */ |
|
975 |
{
|
|
976 |
ut_ad(ptr && end_ptr); |
|
977 |
||
978 |
if (page == NULL) { |
|
979 |
||
980 |
return(ptr); |
|
981 |
}
|
|
982 |
||
983 |
trx_undo_erase_page_end(page, mtr); |
|
984 |
||
985 |
return(ptr); |
|
986 |
}
|
|
987 |
||
988 |
/***************************************************************************
|
|
989 |
Writes information to an undo log about an insert, update, or a delete marking
|
|
990 |
of a clustered index record. This information is used in a rollback of the
|
|
991 |
transaction and in consistent reads that must look to the history of this
|
|
992 |
transaction. */
|
|
993 |
||
994 |
ulint
|
|
995 |
trx_undo_report_row_operation( |
|
996 |
/*==========================*/
|
|
997 |
/* out: DB_SUCCESS or error code */
|
|
998 |
ulint flags, /* in: if BTR_NO_UNDO_LOG_FLAG bit is |
|
999 |
set, does nothing */
|
|
1000 |
ulint op_type, /* in: TRX_UNDO_INSERT_OP or |
|
1001 |
TRX_UNDO_MODIFY_OP */
|
|
1002 |
que_thr_t* thr, /* in: query thread */ |
|
1003 |
dict_index_t* index, /* in: clustered index */ |
|
1004 |
dtuple_t* clust_entry, /* in: in the case of an insert, |
|
1005 |
index entry to insert into the
|
|
1006 |
clustered index, otherwise NULL */
|
|
1007 |
upd_t* update, /* in: in the case of an update, |
|
1008 |
the update vector, otherwise NULL */
|
|
1009 |
ulint cmpl_info, /* in: compiler info on secondary |
|
1010 |
index updates */
|
|
1011 |
rec_t* rec, /* in: in case of an update or delete |
|
1012 |
marking, the record in the clustered
|
|
1013 |
index, otherwise NULL */
|
|
1014 |
dulint* roll_ptr) /* out: rollback pointer to the |
|
1015 |
inserted undo log record,
|
|
1016 |
ut_dulint_zero if BTR_NO_UNDO_LOG
|
|
1017 |
flag was specified */
|
|
1018 |
{
|
|
1019 |
trx_t* trx; |
|
1020 |
trx_undo_t* undo; |
|
1021 |
page_t* undo_page; |
|
1022 |
ulint offset; |
|
1023 |
ulint page_no; |
|
1024 |
ibool is_insert; |
|
1025 |
trx_rseg_t* rseg; |
|
1026 |
mtr_t mtr; |
|
1027 |
ulint err = DB_SUCCESS; |
|
1028 |
mem_heap_t* heap = NULL; |
|
1029 |
ulint offsets_[REC_OFFS_NORMAL_SIZE]; |
|
1030 |
ulint* offsets = offsets_; |
|
1031 |
*offsets_ = (sizeof offsets_) / sizeof *offsets_; |
|
1032 |
||
1033 |
ut_a(index->type & DICT_CLUSTERED); |
|
1034 |
||
1035 |
if (flags & BTR_NO_UNDO_LOG_FLAG) { |
|
1036 |
||
1037 |
*roll_ptr = ut_dulint_zero; |
|
1038 |
||
1039 |
return(err); |
|
1040 |
}
|
|
1041 |
||
1042 |
ut_ad(thr); |
|
1043 |
ut_ad((op_type != TRX_UNDO_INSERT_OP) |
|
1044 |
|| (clust_entry && !update && !rec)); |
|
1045 |
||
1046 |
trx = thr_get_trx(thr); |
|
1047 |
rseg = trx->rseg; |
|
1048 |
||
1049 |
mutex_enter(&(trx->undo_mutex)); |
|
1050 |
||
1051 |
/* If the undo log is not assigned yet, assign one */
|
|
1052 |
||
1053 |
if (op_type == TRX_UNDO_INSERT_OP) { |
|
1054 |
||
1055 |
if (trx->insert_undo == NULL) { |
|
1056 |
||
1057 |
err = trx_undo_assign_undo(trx, TRX_UNDO_INSERT); |
|
1058 |
}
|
|
1059 |
||
1060 |
undo = trx->insert_undo; |
|
1061 |
is_insert = TRUE; |
|
1062 |
} else { |
|
1063 |
ut_ad(op_type == TRX_UNDO_MODIFY_OP); |
|
1064 |
||
1065 |
if (trx->update_undo == NULL) { |
|
1066 |
||
1067 |
err = trx_undo_assign_undo(trx, TRX_UNDO_UPDATE); |
|
1068 |
||
1069 |
}
|
|
1070 |
||
1071 |
undo = trx->update_undo; |
|
1072 |
is_insert = FALSE; |
|
1073 |
}
|
|
1074 |
||
1075 |
if (err != DB_SUCCESS) { |
|
1076 |
/* Did not succeed: return the error encountered */
|
|
1077 |
mutex_exit(&(trx->undo_mutex)); |
|
1078 |
||
1079 |
return(err); |
|
1080 |
}
|
|
1081 |
||
1082 |
page_no = undo->last_page_no; |
|
1083 |
||
1084 |
mtr_start(&mtr); |
|
1085 |
||
1086 |
for (;;) { |
|
1087 |
undo_page = buf_page_get_gen(undo->space, page_no, |
|
1088 |
RW_X_LATCH, undo->guess_page, |
|
1089 |
BUF_GET, |
|
1090 |
__FILE__, __LINE__, |
|
1091 |
&mtr); |
|
1092 |
||
1093 |
#ifdef UNIV_SYNC_DEBUG
|
|
1094 |
buf_page_dbg_add_level(undo_page, SYNC_TRX_UNDO_PAGE); |
|
1095 |
#endif /* UNIV_SYNC_DEBUG */ |
|
1096 |
||
1097 |
if (op_type == TRX_UNDO_INSERT_OP) { |
|
1098 |
offset = trx_undo_page_report_insert( |
|
1099 |
undo_page, trx, index, clust_entry, &mtr); |
|
1100 |
} else { |
|
1101 |
offsets = rec_get_offsets(rec, index, offsets, |
|
1102 |
ULINT_UNDEFINED, &heap); |
|
1103 |
offset = trx_undo_page_report_modify( |
|
1104 |
undo_page, trx, index, rec, offsets, update, |
|
1105 |
cmpl_info, &mtr); |
|
1106 |
}
|
|
1107 |
||
1108 |
if (offset == 0) { |
|
1109 |
/* The record did not fit on the page. We erase the
|
|
1110 |
end segment of the undo log page and write a log
|
|
1111 |
record of it: this is to ensure that in the debug
|
|
1112 |
version the replicate page constructed using the log
|
|
1113 |
records stays identical to the original page */
|
|
1114 |
||
1115 |
trx_undo_erase_page_end(undo_page, &mtr); |
|
1116 |
}
|
|
1117 |
||
1118 |
mtr_commit(&mtr); |
|
1119 |
||
1120 |
if (offset != 0) { |
|
1121 |
/* Success */
|
|
1122 |
||
1123 |
break; |
|
1124 |
}
|
|
1125 |
||
1126 |
ut_ad(page_no == undo->last_page_no); |
|
1127 |
||
1128 |
/* We have to extend the undo log by one page */
|
|
1129 |
||
1130 |
mtr_start(&mtr); |
|
1131 |
||
1132 |
/* When we add a page to an undo log, this is analogous to
|
|
1133 |
a pessimistic insert in a B-tree, and we must reserve the
|
|
1134 |
counterpart of the tree latch, which is the rseg mutex. */
|
|
1135 |
||
1136 |
mutex_enter(&(rseg->mutex)); |
|
1137 |
||
1138 |
page_no = trx_undo_add_page(trx, undo, &mtr); |
|
1139 |
||
1140 |
mutex_exit(&(rseg->mutex)); |
|
1141 |
||
1142 |
if (page_no == FIL_NULL) { |
|
1143 |
/* Did not succeed: out of space */
|
|
1144 |
||
1145 |
mutex_exit(&(trx->undo_mutex)); |
|
1146 |
mtr_commit(&mtr); |
|
1147 |
if (UNIV_LIKELY_NULL(heap)) { |
|
1148 |
mem_heap_free(heap); |
|
1149 |
}
|
|
1150 |
return(DB_OUT_OF_FILE_SPACE); |
|
1151 |
}
|
|
1152 |
}
|
|
1153 |
||
1154 |
undo->empty = FALSE; |
|
1155 |
undo->top_page_no = page_no; |
|
1156 |
undo->top_offset = offset; |
|
1157 |
undo->top_undo_no = trx->undo_no; |
|
1158 |
undo->guess_page = undo_page; |
|
1159 |
||
1160 |
UT_DULINT_INC(trx->undo_no); |
|
1161 |
||
1162 |
mutex_exit(&(trx->undo_mutex)); |
|
1163 |
||
1164 |
*roll_ptr = trx_undo_build_roll_ptr(is_insert, rseg->id, page_no, |
|
1165 |
offset); |
|
1166 |
if (UNIV_LIKELY_NULL(heap)) { |
|
1167 |
mem_heap_free(heap); |
|
1168 |
}
|
|
1169 |
return(err); |
|
1170 |
}
|
|
1171 |
||
1172 |
/*============== BUILDING PREVIOUS VERSION OF A RECORD ===============*/
|
|
1173 |
||
1174 |
/**********************************************************************
|
|
1175 |
Copies an undo record to heap. This function can be called if we know that
|
|
1176 |
the undo log record exists. */
|
|
1177 |
||
1178 |
trx_undo_rec_t* |
|
1179 |
trx_undo_get_undo_rec_low( |
|
1180 |
/*======================*/
|
|
1181 |
/* out, own: copy of the record */
|
|
1182 |
dulint roll_ptr, /* in: roll pointer to record */ |
|
1183 |
mem_heap_t* heap) /* in: memory heap where copied */ |
|
1184 |
{
|
|
1185 |
trx_undo_rec_t* undo_rec; |
|
1186 |
ulint rseg_id; |
|
1187 |
ulint page_no; |
|
1188 |
ulint offset; |
|
1189 |
page_t* undo_page; |
|
1190 |
trx_rseg_t* rseg; |
|
1191 |
ibool is_insert; |
|
1192 |
mtr_t mtr; |
|
1193 |
||
1194 |
trx_undo_decode_roll_ptr(roll_ptr, &is_insert, &rseg_id, &page_no, |
|
1195 |
&offset); |
|
1196 |
rseg = trx_rseg_get_on_id(rseg_id); |
|
1197 |
||
1198 |
mtr_start(&mtr); |
|
1199 |
||
1200 |
undo_page = trx_undo_page_get_s_latched(rseg->space, page_no, &mtr); |
|
1201 |
||
1202 |
undo_rec = trx_undo_rec_copy(undo_page + offset, heap); |
|
1203 |
||
1204 |
mtr_commit(&mtr); |
|
1205 |
||
1206 |
return(undo_rec); |
|
1207 |
}
|
|
1208 |
||
1209 |
/**********************************************************************
|
|
1210 |
Copies an undo record to heap. */
|
|
1211 |
||
1212 |
ulint
|
|
1213 |
trx_undo_get_undo_rec( |
|
1214 |
/*==================*/
|
|
1215 |
/* out: DB_SUCCESS, or
|
|
1216 |
DB_MISSING_HISTORY if the undo log
|
|
1217 |
has been truncated and we cannot
|
|
1218 |
fetch the old version; NOTE: the
|
|
1219 |
caller must have latches on the
|
|
1220 |
clustered index page and purge_view */
|
|
1221 |
dulint roll_ptr, /* in: roll pointer to record */ |
|
1222 |
dulint trx_id, /* in: id of the trx that generated |
|
1223 |
the roll pointer: it points to an
|
|
1224 |
undo log of this transaction */
|
|
1225 |
trx_undo_rec_t** undo_rec, /* out, own: copy of the record */ |
|
1226 |
mem_heap_t* heap) /* in: memory heap where copied */ |
|
1227 |
{
|
|
1228 |
#ifdef UNIV_SYNC_DEBUG
|
|
1229 |
ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED)); |
|
1230 |
#endif /* UNIV_SYNC_DEBUG */ |
|
1231 |
||
1232 |
if (!trx_purge_update_undo_must_exist(trx_id)) { |
|
1233 |
||
1234 |
/* It may be that the necessary undo log has already been
|
|
1235 |
deleted */
|
|
1236 |
||
1237 |
return(DB_MISSING_HISTORY); |
|
1238 |
}
|
|
1239 |
||
1240 |
*undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap); |
|
1241 |
||
1242 |
return(DB_SUCCESS); |
|
1243 |
}
|
|
1244 |
||
1245 |
/***********************************************************************
|
|
1246 |
Build a previous version of a clustered index record. This function checks
|
|
1247 |
that the caller has a latch on the index page of the clustered index record
|
|
1248 |
and an s-latch on the purge_view. This guarantees that the stack of versions
|
|
1249 |
is locked. */
|
|
1250 |
||
1251 |
ulint
|
|
1252 |
trx_undo_prev_version_build( |
|
1253 |
/*========================*/
|
|
1254 |
/* out: DB_SUCCESS, or DB_MISSING_HISTORY if
|
|
1255 |
the previous version is not >= purge_view,
|
|
1256 |
which means that it may have been removed,
|
|
1257 |
DB_ERROR if corrupted record */
|
|
1258 |
rec_t* index_rec,/* in: clustered index record in the |
|
1259 |
index tree */
|
|
1260 |
mtr_t* index_mtr __attribute__((unused)), |
|
1261 |
/* in: mtr which contains the latch to
|
|
1262 |
index_rec page and purge_view */
|
|
1263 |
rec_t* rec, /* in: version of a clustered index record */ |
|
1264 |
dict_index_t* index, /* in: clustered index */ |
|
1265 |
ulint* offsets,/* in: rec_get_offsets(rec, index) */ |
|
1266 |
mem_heap_t* heap, /* in: memory heap from which the memory |
|
1267 |
needed is allocated */
|
|
1268 |
rec_t** old_vers)/* out, own: previous version, or NULL if |
|
1269 |
rec is the first inserted version, or if
|
|
1270 |
history data has been deleted */
|
|
1271 |
{
|
|
1272 |
trx_undo_rec_t* undo_rec; |
|
1273 |
dtuple_t* entry; |
|
1274 |
dulint rec_trx_id; |
|
1275 |
ulint type; |
|
1276 |
dulint undo_no; |
|
1277 |
dulint table_id; |
|
1278 |
dulint trx_id; |
|
1279 |
dulint roll_ptr; |
|
1280 |
dulint old_roll_ptr; |
|
1281 |
upd_t* update; |
|
1282 |
byte* ptr; |
|
1283 |
ulint info_bits; |
|
1284 |
ulint cmpl_info; |
|
1285 |
ibool dummy_extern; |
|
1286 |
byte* buf; |
|
1287 |
ulint err; |
|
1288 |
#ifdef UNIV_SYNC_DEBUG
|
|
1289 |
ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED)); |
|
1290 |
#endif /* UNIV_SYNC_DEBUG */ |
|
1291 |
ut_ad(mtr_memo_contains(index_mtr, buf_block_align(index_rec), |
|
1292 |
MTR_MEMO_PAGE_S_FIX) |
|
1293 |
|| mtr_memo_contains(index_mtr, buf_block_align(index_rec), |
|
1294 |
MTR_MEMO_PAGE_X_FIX)); |
|
1295 |
ut_ad(rec_offs_validate(rec, index, offsets)); |
|
1296 |
||
1297 |
if (!(index->type & DICT_CLUSTERED)) { |
|
1298 |
fprintf(stderr, "InnoDB: Error: trying to access" |
|
1299 |
" update undo rec for non-clustered index %s\n" |
|
1300 |
"InnoDB: Submit a detailed bug report to"
|
|
1301 |
" http://bugs.mysql.com\n" |
|
1302 |
"InnoDB: index record ", index->name); |
|
1303 |
rec_print(stderr, index_rec, index); |
|
1304 |
fputs("\n" |
|
1305 |
"InnoDB: record version ", stderr); |
|
1306 |
rec_print_new(stderr, rec, offsets); |
|
1307 |
putc('\n', stderr); |
|
1308 |
return(DB_ERROR); |
|
1309 |
}
|
|
1310 |
||
1311 |
roll_ptr = row_get_rec_roll_ptr(rec, index, offsets); |
|
1312 |
old_roll_ptr = roll_ptr; |
|
1313 |
||
1314 |
*old_vers = NULL; |
|
1315 |
||
1316 |
if (trx_undo_roll_ptr_is_insert(roll_ptr)) { |
|
1317 |
||
1318 |
/* The record rec is the first inserted version */
|
|
1319 |
||
1320 |
return(DB_SUCCESS); |
|
1321 |
}
|
|
1322 |
||
1323 |
rec_trx_id = row_get_rec_trx_id(rec, index, offsets); |
|
1324 |
||
1325 |
err = trx_undo_get_undo_rec(roll_ptr, rec_trx_id, &undo_rec, heap); |
|
1326 |
||
1327 |
if (err != DB_SUCCESS) { |
|
1328 |
||
1329 |
return(err); |
|
1330 |
}
|
|
1331 |
||
1332 |
ptr = trx_undo_rec_get_pars(undo_rec, &type, &cmpl_info, |
|
1333 |
&dummy_extern, &undo_no, &table_id); |
|
1334 |
||
1335 |
ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr, |
|
1336 |
&info_bits); |
|
1337 |
ptr = trx_undo_rec_skip_row_ref(ptr, index); |
|
1338 |
||
1339 |
ptr = trx_undo_update_rec_get_update(ptr, index, type, trx_id, |
|
1340 |
roll_ptr, info_bits, |
|
1341 |
NULL, heap, &update); |
|
1342 |
||
1343 |
if (ut_dulint_cmp(table_id, index->table->id) != 0) { |
|
1344 |
ptr = NULL; |
|
1345 |
||
1346 |
fprintf(stderr, |
|
1347 |
"InnoDB: Error: trying to access update undo rec"
|
|
1348 |
" for table %s\n" |
|
1349 |
"InnoDB: but the table id in the"
|
|
1350 |
" undo record is wrong\n" |
|
1351 |
"InnoDB: Submit a detailed bug report"
|
|
1352 |
" to http://bugs.mysql.com\n" |
|
1353 |
"InnoDB: Run also CHECK TABLE %s\n", |
|
1354 |
index->table_name, index->table_name); |
|
1355 |
}
|
|
1356 |
||
1357 |
if (ptr == NULL) { |
|
1358 |
/* The record was corrupted, return an error; these printfs
|
|
1359 |
should catch an elusive bug in row_vers_old_has_index_entry */
|
|
1360 |
||
1361 |
fprintf(stderr, |
|
1362 |
"InnoDB: table %s, index %s, n_uniq %lu\n" |
|
1363 |
"InnoDB: undo rec address %p, type %lu cmpl_info %lu\n" |
|
1364 |
"InnoDB: undo rec table id %lu %lu,"
|
|
1365 |
" index table id %lu %lu\n" |
|
1366 |
"InnoDB: dump of 150 bytes in undo rec: ", |
|
1367 |
index->table_name, index->name, |
|
1368 |
(ulong) dict_index_get_n_unique(index), |
|
1369 |
undo_rec, (ulong) type, (ulong) cmpl_info, |
|
1370 |
(ulong) ut_dulint_get_high(table_id), |
|
1371 |
(ulong) ut_dulint_get_low(table_id), |
|
1372 |
(ulong) ut_dulint_get_high(index->table->id), |
|
1373 |
(ulong) ut_dulint_get_low(index->table->id)); |
|
1374 |
ut_print_buf(stderr, undo_rec, 150); |
|
1375 |
fputs("\n" |
|
1376 |
"InnoDB: index record ", stderr); |
|
1377 |
rec_print(stderr, index_rec, index); |
|
1378 |
fputs("\n" |
|
1379 |
"InnoDB: record version ", stderr); |
|
1380 |
rec_print_new(stderr, rec, offsets); |
|
1381 |
fprintf(stderr, "\n" |
|
1382 |
"InnoDB: Record trx id %lu %lu, update rec"
|
|
1383 |
" trx id %lu %lu\n" |
|
1384 |
"InnoDB: Roll ptr in rec %lu %lu, in update rec"
|
|
1385 |
" %lu %lu\n", |
|
1386 |
(ulong) ut_dulint_get_high(rec_trx_id), |
|
1387 |
(ulong) ut_dulint_get_low(rec_trx_id), |
|
1388 |
(ulong) ut_dulint_get_high(trx_id), |
|
1389 |
(ulong) ut_dulint_get_low(trx_id), |
|
1390 |
(ulong) ut_dulint_get_high(old_roll_ptr), |
|
1391 |
(ulong) ut_dulint_get_low(old_roll_ptr), |
|
1392 |
(ulong) ut_dulint_get_high(roll_ptr), |
|
1393 |
(ulong) ut_dulint_get_low(roll_ptr)); |
|
1394 |
||
1395 |
trx_purge_sys_print(); |
|
1396 |
return(DB_ERROR); |
|
1397 |
}
|
|
1398 |
||
1399 |
if (row_upd_changes_field_size_or_external(index, offsets, update)) { |
|
1400 |
ulint* ext_vect; |
|
1401 |
ulint n_ext_vect; |
|
1402 |
||
1403 |
/* We have to set the appropriate extern storage bits in the
|
|
1404 |
old version of the record: the extern bits in rec for those
|
|
1405 |
fields that update does NOT update, as well as the the bits for
|
|
1406 |
those fields that update updates to become externally stored
|
|
1407 |
fields. Store the info to ext_vect: */
|
|
1408 |
||
1409 |
ext_vect = mem_alloc(sizeof(ulint) |
|
1410 |
* rec_offs_n_fields(offsets)); |
|
1411 |
n_ext_vect = btr_push_update_extern_fields(ext_vect, offsets, |
|
1412 |
update); |
|
1413 |
entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, |
|
1414 |
heap); |
|
1415 |
row_upd_index_replace_new_col_vals(entry, index, update, heap); |
|
1416 |
||
1417 |
buf = mem_heap_alloc(heap, |
|
1418 |
rec_get_converted_size(index, entry)); |
|
1419 |
||
1420 |
*old_vers = rec_convert_dtuple_to_rec(buf, index, entry); |
|
1421 |
||
1422 |
/* Now set the extern bits in the old version of the record */
|
|
1423 |
rec_set_field_extern_bits(*old_vers, index, |
|
1424 |
ext_vect, n_ext_vect, NULL); |
|
1425 |
mem_free(ext_vect); |
|
1426 |
} else { |
|
1427 |
buf = mem_heap_alloc(heap, rec_offs_size(offsets)); |
|
1428 |
*old_vers = rec_copy(buf, rec, offsets); |
|
1429 |
rec_offs_make_valid(*old_vers, index, offsets); |
|
1430 |
row_upd_rec_in_place(*old_vers, offsets, update); |
|
1431 |
}
|
|
1432 |
||
1433 |
return(DB_SUCCESS); |
|
1434 |
}
|