1 | /***************************************************************************** |
2 | |
3 | Copyright (c) 2007, 2015, Oracle and/or its affiliates. All Rights Reserved. |
4 | Copyright (c) 2017, MariaDB Corporation. |
5 | |
6 | This program is free software; you can redistribute it and/or modify it under |
7 | the terms of the GNU General Public License as published by the Free Software |
8 | Foundation; version 2 of the License. |
9 | |
10 | This program is distributed in the hope that it will be useful, but WITHOUT |
11 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
12 | FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU General Public License along with |
15 | this program; if not, write to the Free Software Foundation, Inc., |
16 | 51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA |
17 | |
18 | *****************************************************************************/ |
19 | |
20 | /**************************************************//** |
21 | @file trx/trx0i_s.cc |
22 | INFORMATION SCHEMA innodb_trx, innodb_locks and |
23 | innodb_lock_waits tables fetch code. |
24 | |
25 | The code below fetches information needed to fill those |
26 | 3 dynamic tables and uploads it into a "transactions |
27 | table cache" for later retrieval. |
28 | |
29 | Created July 17, 2007 Vasil Dimov |
30 | *******************************************************/ |
31 | |
32 | /* Found during the build of 5.5.3 on Linux 2.4 and early 2.6 kernels: |
33 | The includes "univ.i" -> "my_global.h" cause a different path |
34 | to be taken further down with pthread functions and types, |
35 | so they must come first. |
36 | From the symptoms, this is related to bug#46587 in the MySQL bug DB. |
37 | */ |
38 | |
39 | #include "ha_prototypes.h" |
40 | #include <sql_class.h> |
41 | |
42 | #include "buf0buf.h" |
43 | #include "dict0dict.h" |
44 | #include "ha0storage.h" |
45 | #include "hash0hash.h" |
46 | #include "lock0iter.h" |
47 | #include "lock0lock.h" |
48 | #include "mem0mem.h" |
49 | #include "page0page.h" |
50 | #include "rem0rec.h" |
51 | #include "row0row.h" |
52 | #include "srv0srv.h" |
53 | #include "sync0rw.h" |
54 | #include "sync0sync.h" |
55 | #include "trx0i_s.h" |
56 | #include "trx0sys.h" |
57 | #include "trx0trx.h" |
58 | #include "ut0mem.h" |
59 | #include "que0que.h" |
60 | #include "trx0purge.h" |
61 | |
62 | /** Initial number of rows in the table cache */ |
63 | #define TABLE_CACHE_INITIAL_ROWSNUM 1024 |
64 | |
65 | /** @brief The maximum number of chunks to allocate for a table cache. |
66 | |
67 | The rows of a table cache are stored in a set of chunks. When a new |
68 | row is added a new chunk is allocated if necessary. Assuming that the |
69 | first one is 1024 rows (TABLE_CACHE_INITIAL_ROWSNUM) and each |
70 | subsequent is N/2 where N is the number of rows we have allocated till |
71 | now, then 39th chunk would accommodate 1677416425 rows and all chunks |
72 | would accommodate 3354832851 rows. */ |
73 | #define MEM_CHUNKS_IN_TABLE_CACHE 39 |
74 | |
75 | /** The following are some testing auxiliary macros. Do not enable them |
76 | in a production environment. */ |
77 | /* @{ */ |
78 | |
79 | #if 0 |
80 | /** If this is enabled then lock folds will always be different |
81 | resulting in equal rows being put in a different cells of the hash |
82 | table. Checking for duplicates will be flawed because different |
83 | fold will be calculated when a row is searched in the hash table. */ |
84 | #define TEST_LOCK_FOLD_ALWAYS_DIFFERENT |
85 | #endif |
86 | |
87 | #if 0 |
88 | /** This effectively kills the search-for-duplicate-before-adding-a-row |
89 | function, but searching in the hash is still performed. It will always |
90 | be assumed that lock is not present and insertion will be performed in |
91 | the hash table. */ |
92 | #define TEST_NO_LOCKS_ROW_IS_EVER_EQUAL_TO_LOCK_T |
93 | #endif |
94 | |
95 | #if 0 |
96 | /** This aggressively repeats adding each row many times. Depending on |
97 | the above settings this may be noop or may result in lots of rows being |
98 | added. */ |
99 | #define TEST_ADD_EACH_LOCKS_ROW_MANY_TIMES |
100 | #endif |
101 | |
102 | #if 0 |
103 | /** Very similar to TEST_NO_LOCKS_ROW_IS_EVER_EQUAL_TO_LOCK_T but hash |
104 | table search is not performed at all. */ |
105 | #define TEST_DO_NOT_CHECK_FOR_DUPLICATE_ROWS |
106 | #endif |
107 | |
108 | #if 0 |
109 | /** Do not insert each row into the hash table, duplicates may appear |
110 | if this is enabled, also if this is enabled searching into the hash is |
111 | noop because it will be empty. */ |
112 | #define TEST_DO_NOT_INSERT_INTO_THE_HASH_TABLE |
113 | #endif |
114 | /* @} */ |
115 | |
116 | /** Memory limit passed to ha_storage_put_memlim(). |
117 | @param cache hash storage |
118 | @return maximum allowed allocation size */ |
119 | #define MAX_ALLOWED_FOR_STORAGE(cache) \ |
120 | (TRX_I_S_MEM_LIMIT \ |
121 | - (cache)->mem_allocd) |
122 | |
123 | /** Memory limit in table_cache_create_empty_row(). |
124 | @param cache hash storage |
125 | @return maximum allowed allocation size */ |
126 | #define MAX_ALLOWED_FOR_ALLOC(cache) \ |
127 | (TRX_I_S_MEM_LIMIT \ |
128 | - (cache)->mem_allocd \ |
129 | - ha_storage_get_size((cache)->storage)) |
130 | |
131 | /** Memory for each table in the intermediate buffer is allocated in |
132 | separate chunks. These chunks are considered to be concatenated to |
133 | represent one flat array of rows. */ |
134 | struct i_s_mem_chunk_t { |
135 | ulint offset; /*!< offset, in number of rows */ |
136 | ulint rows_allocd; /*!< the size of this chunk, in number |
137 | of rows */ |
138 | void* base; /*!< start of the chunk */ |
139 | }; |
140 | |
141 | /** This represents one table's cache. */ |
142 | struct i_s_table_cache_t { |
143 | ulint rows_used; /*!< number of used rows */ |
144 | ulint rows_allocd; /*!< number of allocated rows */ |
145 | ulint row_size; /*!< size of a single row */ |
146 | i_s_mem_chunk_t chunks[MEM_CHUNKS_IN_TABLE_CACHE]; /*!< array of |
147 | memory chunks that stores the |
148 | rows */ |
149 | }; |
150 | |
151 | /** This structure describes the intermediate buffer */ |
152 | struct trx_i_s_cache_t { |
153 | rw_lock_t* rw_lock; /*!< read-write lock protecting |
154 | the rest of this structure */ |
155 | uintmax_t last_read; /*!< last time the cache was read; |
156 | measured in microseconds since |
157 | epoch */ |
158 | ib_mutex_t last_read_mutex;/*!< mutex protecting the |
159 | last_read member - it is updated |
160 | inside a shared lock of the |
161 | rw_lock member */ |
162 | i_s_table_cache_t innodb_trx; /*!< innodb_trx table */ |
163 | i_s_table_cache_t innodb_locks; /*!< innodb_locks table */ |
164 | i_s_table_cache_t innodb_lock_waits;/*!< innodb_lock_waits table */ |
165 | /** the hash table size is LOCKS_HASH_CELLS_NUM * sizeof(void*) bytes */ |
166 | #define LOCKS_HASH_CELLS_NUM 10000 |
167 | hash_table_t* locks_hash; /*!< hash table used to eliminate |
168 | duplicate entries in the |
169 | innodb_locks table */ |
170 | /** Initial size of the cache storage */ |
171 | #define CACHE_STORAGE_INITIAL_SIZE 1024 |
172 | /** Number of hash cells in the cache storage */ |
173 | #define CACHE_STORAGE_HASH_CELLS 2048 |
174 | ha_storage_t* storage; /*!< storage for external volatile |
175 | data that may become unavailable |
176 | when we release |
177 | lock_sys.mutex or trx_sys.mutex */ |
178 | ulint mem_allocd; /*!< the amount of memory |
179 | allocated with mem_alloc*() */ |
180 | bool is_truncated; /*!< this is true if the memory |
181 | limit was hit and thus the data |
182 | in the cache is truncated */ |
183 | }; |
184 | |
185 | /** This is the intermediate buffer where data needed to fill the |
186 | INFORMATION SCHEMA tables is fetched and later retrieved by the C++ |
187 | code in handler/i_s.cc. */ |
188 | static trx_i_s_cache_t trx_i_s_cache_static; |
189 | /** This is the intermediate buffer where data needed to fill the |
190 | INFORMATION SCHEMA tables is fetched and later retrieved by the C++ |
191 | code in handler/i_s.cc. */ |
192 | trx_i_s_cache_t* trx_i_s_cache = &trx_i_s_cache_static; |
193 | |
194 | /*******************************************************************//** |
195 | For a record lock that is in waiting state retrieves the only bit that |
196 | is set, for a table lock returns ULINT_UNDEFINED. |
197 | @return record number within the heap */ |
198 | static |
199 | ulint |
200 | wait_lock_get_heap_no( |
201 | /*==================*/ |
202 | const lock_t* lock) /*!< in: lock */ |
203 | { |
204 | ulint ret; |
205 | |
206 | switch (lock_get_type(lock)) { |
207 | case LOCK_REC: |
208 | ret = lock_rec_find_set_bit(lock); |
209 | ut_a(ret != ULINT_UNDEFINED); |
210 | break; |
211 | case LOCK_TABLE: |
212 | ret = ULINT_UNDEFINED; |
213 | break; |
214 | default: |
215 | ut_error; |
216 | } |
217 | |
218 | return(ret); |
219 | } |
220 | |
221 | /*******************************************************************//** |
222 | Initializes the members of a table cache. */ |
223 | static |
224 | void |
225 | table_cache_init( |
226 | /*=============*/ |
227 | i_s_table_cache_t* table_cache, /*!< out: table cache */ |
228 | size_t row_size) /*!< in: the size of a |
229 | row */ |
230 | { |
231 | ulint i; |
232 | |
233 | table_cache->rows_used = 0; |
234 | table_cache->rows_allocd = 0; |
235 | table_cache->row_size = row_size; |
236 | |
237 | for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) { |
238 | |
239 | /* the memory is actually allocated in |
240 | table_cache_create_empty_row() */ |
241 | table_cache->chunks[i].base = NULL; |
242 | } |
243 | } |
244 | |
245 | /*******************************************************************//** |
246 | Frees a table cache. */ |
247 | static |
248 | void |
249 | table_cache_free( |
250 | /*=============*/ |
251 | i_s_table_cache_t* table_cache) /*!< in/out: table cache */ |
252 | { |
253 | ulint i; |
254 | |
255 | for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) { |
256 | |
257 | /* the memory is actually allocated in |
258 | table_cache_create_empty_row() */ |
259 | if (table_cache->chunks[i].base) { |
260 | ut_free(table_cache->chunks[i].base); |
261 | table_cache->chunks[i].base = NULL; |
262 | } |
263 | } |
264 | } |
265 | |
266 | /*******************************************************************//** |
267 | Returns an empty row from a table cache. The row is allocated if no more |
268 | empty rows are available. The number of used rows is incremented. |
269 | If the memory limit is hit then NULL is returned and nothing is |
270 | allocated. |
271 | @return empty row, or NULL if out of memory */ |
272 | static |
273 | void* |
274 | table_cache_create_empty_row( |
275 | /*=========================*/ |
276 | i_s_table_cache_t* table_cache, /*!< in/out: table cache */ |
277 | trx_i_s_cache_t* cache) /*!< in/out: cache to record |
278 | how many bytes are |
279 | allocated */ |
280 | { |
281 | ulint i; |
282 | void* row; |
283 | |
284 | ut_a(table_cache->rows_used <= table_cache->rows_allocd); |
285 | |
286 | if (table_cache->rows_used == table_cache->rows_allocd) { |
287 | |
288 | /* rows_used == rows_allocd means that new chunk needs |
289 | to be allocated: either no more empty rows in the |
290 | last allocated chunk or nothing has been allocated yet |
291 | (rows_num == rows_allocd == 0); */ |
292 | |
293 | i_s_mem_chunk_t* chunk; |
294 | ulint req_bytes; |
295 | ulint got_bytes; |
296 | ulint req_rows; |
297 | ulint got_rows; |
298 | |
299 | /* find the first not allocated chunk */ |
300 | for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) { |
301 | |
302 | if (table_cache->chunks[i].base == NULL) { |
303 | |
304 | break; |
305 | } |
306 | } |
307 | |
308 | /* i == MEM_CHUNKS_IN_TABLE_CACHE means that all chunks |
309 | have been allocated :-X */ |
310 | ut_a(i < MEM_CHUNKS_IN_TABLE_CACHE); |
311 | |
312 | /* allocate the chunk we just found */ |
313 | |
314 | if (i == 0) { |
315 | |
316 | /* first chunk, nothing is allocated yet */ |
317 | req_rows = TABLE_CACHE_INITIAL_ROWSNUM; |
318 | } else { |
319 | |
320 | /* Memory is increased by the formula |
321 | new = old + old / 2; We are trying not to be |
322 | aggressive here (= using the common new = old * 2) |
323 | because the allocated memory will not be freed |
324 | until InnoDB exit (it is reused). So it is better |
325 | to once allocate the memory in more steps, but |
326 | have less unused/wasted memory than to use less |
327 | steps in allocation (which is done once in a |
328 | lifetime) but end up with lots of unused/wasted |
329 | memory. */ |
330 | req_rows = table_cache->rows_allocd / 2; |
331 | } |
332 | req_bytes = req_rows * table_cache->row_size; |
333 | |
334 | if (req_bytes > MAX_ALLOWED_FOR_ALLOC(cache)) { |
335 | |
336 | return(NULL); |
337 | } |
338 | |
339 | chunk = &table_cache->chunks[i]; |
340 | |
341 | got_bytes = req_bytes; |
342 | chunk->base = ut_malloc_nokey(req_bytes); |
343 | |
344 | got_rows = got_bytes / table_cache->row_size; |
345 | |
346 | cache->mem_allocd += got_bytes; |
347 | |
348 | #if 0 |
349 | printf("allocating chunk %d req bytes=%lu, got bytes=%lu," |
350 | " row size=%lu," |
351 | " req rows=%lu, got rows=%lu\n" , |
352 | i, req_bytes, got_bytes, |
353 | table_cache->row_size, |
354 | req_rows, got_rows); |
355 | #endif |
356 | |
357 | chunk->rows_allocd = got_rows; |
358 | |
359 | table_cache->rows_allocd += got_rows; |
360 | |
361 | /* adjust the offset of the next chunk */ |
362 | if (i < MEM_CHUNKS_IN_TABLE_CACHE - 1) { |
363 | |
364 | table_cache->chunks[i + 1].offset |
365 | = chunk->offset + chunk->rows_allocd; |
366 | } |
367 | |
368 | /* return the first empty row in the newly allocated |
369 | chunk */ |
370 | row = chunk->base; |
371 | } else { |
372 | |
373 | char* chunk_start; |
374 | ulint offset; |
375 | |
376 | /* there is an empty row, no need to allocate new |
377 | chunks */ |
378 | |
379 | /* find the first chunk that contains allocated but |
380 | empty/unused rows */ |
381 | for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) { |
382 | |
383 | if (table_cache->chunks[i].offset |
384 | + table_cache->chunks[i].rows_allocd |
385 | > table_cache->rows_used) { |
386 | |
387 | break; |
388 | } |
389 | } |
390 | |
391 | /* i == MEM_CHUNKS_IN_TABLE_CACHE means that all chunks |
392 | are full, but |
393 | table_cache->rows_used != table_cache->rows_allocd means |
394 | exactly the opposite - there are allocated but |
395 | empty/unused rows :-X */ |
396 | ut_a(i < MEM_CHUNKS_IN_TABLE_CACHE); |
397 | |
398 | chunk_start = (char*) table_cache->chunks[i].base; |
399 | offset = table_cache->rows_used |
400 | - table_cache->chunks[i].offset; |
401 | |
402 | row = chunk_start + offset * table_cache->row_size; |
403 | } |
404 | |
405 | table_cache->rows_used++; |
406 | |
407 | return(row); |
408 | } |
409 | |
410 | #ifdef UNIV_DEBUG |
411 | /*******************************************************************//** |
412 | Validates a row in the locks cache. |
413 | @return TRUE if valid */ |
414 | static |
415 | ibool |
416 | i_s_locks_row_validate( |
417 | /*===================*/ |
418 | const i_s_locks_row_t* row) /*!< in: row to validate */ |
419 | { |
420 | ut_ad(row->lock_mode != NULL); |
421 | ut_ad(row->lock_type != NULL); |
422 | ut_ad(row->lock_table != NULL); |
423 | ut_ad(row->lock_table_id != 0); |
424 | |
425 | if (row->lock_space == ULINT_UNDEFINED) { |
426 | /* table lock */ |
427 | ut_ad(!strcmp("TABLE" , row->lock_type)); |
428 | ut_ad(row->lock_index == NULL); |
429 | ut_ad(row->lock_data == NULL); |
430 | ut_ad(row->lock_page == ULINT_UNDEFINED); |
431 | ut_ad(row->lock_rec == ULINT_UNDEFINED); |
432 | } else { |
433 | /* record lock */ |
434 | ut_ad(!strcmp("RECORD" , row->lock_type)); |
435 | ut_ad(row->lock_index != NULL); |
436 | /* row->lock_data == NULL if buf_page_try_get() == NULL */ |
437 | ut_ad(row->lock_page != ULINT_UNDEFINED); |
438 | ut_ad(row->lock_rec != ULINT_UNDEFINED); |
439 | } |
440 | |
441 | return(TRUE); |
442 | } |
443 | #endif /* UNIV_DEBUG */ |
444 | |
445 | /*******************************************************************//** |
446 | Fills i_s_trx_row_t object. |
447 | If memory can not be allocated then FALSE is returned. |
448 | @return FALSE if allocation fails */ |
449 | static |
450 | ibool |
451 | fill_trx_row( |
452 | /*=========*/ |
453 | i_s_trx_row_t* row, /*!< out: result object |
454 | that's filled */ |
455 | const trx_t* trx, /*!< in: transaction to |
456 | get data from */ |
457 | const i_s_locks_row_t* requested_lock_row,/*!< in: pointer to the |
458 | corresponding row in |
459 | innodb_locks if trx is |
460 | waiting or NULL if trx |
461 | is not waiting */ |
462 | trx_i_s_cache_t* cache) /*!< in/out: cache into |
463 | which to copy volatile |
464 | strings */ |
465 | { |
466 | size_t stmt_len; |
467 | const char* s; |
468 | |
469 | ut_ad(lock_mutex_own()); |
470 | |
471 | row->trx_id = trx_get_id_for_print(trx); |
472 | row->trx_started = (ib_time_t) trx->start_time; |
473 | row->trx_state = trx_get_que_state_str(trx); |
474 | row->requested_lock_row = requested_lock_row; |
475 | ut_ad(requested_lock_row == NULL |
476 | || i_s_locks_row_validate(requested_lock_row)); |
477 | |
478 | if (trx->lock.wait_lock != NULL) { |
479 | |
480 | ut_a(requested_lock_row != NULL); |
481 | row->trx_wait_started = (ib_time_t) trx->lock.wait_started; |
482 | } else { |
483 | ut_a(requested_lock_row == NULL); |
484 | row->trx_wait_started = 0; |
485 | } |
486 | |
487 | row->trx_weight = static_cast<uintmax_t>(TRX_WEIGHT(trx)); |
488 | |
489 | if (trx->mysql_thd == NULL) { |
490 | /* For internal transactions e.g., purge and transactions |
491 | being recovered at startup there is no associated MySQL |
492 | thread data structure. */ |
493 | row->trx_mysql_thread_id = 0; |
494 | row->trx_query = NULL; |
495 | goto thd_done; |
496 | } |
497 | |
498 | row->trx_mysql_thread_id = thd_get_thread_id(trx->mysql_thd); |
499 | |
500 | char query[TRX_I_S_TRX_QUERY_MAX_LEN + 1]; |
501 | stmt_len = innobase_get_stmt_safe(trx->mysql_thd, query, sizeof(query)); |
502 | |
503 | if (stmt_len > 0) { |
504 | |
505 | row->trx_query = static_cast<const char*>( |
506 | ha_storage_put_memlim( |
507 | cache->storage, query, stmt_len + 1, |
508 | MAX_ALLOWED_FOR_STORAGE(cache))); |
509 | |
510 | row->trx_query_cs = innobase_get_charset(trx->mysql_thd); |
511 | |
512 | if (row->trx_query == NULL) { |
513 | |
514 | return(FALSE); |
515 | } |
516 | } else { |
517 | |
518 | row->trx_query = NULL; |
519 | } |
520 | |
521 | thd_done: |
522 | s = trx->op_info; |
523 | |
524 | if (s != NULL && s[0] != '\0') { |
525 | |
526 | TRX_I_S_STRING_COPY(s, row->trx_operation_state, |
527 | TRX_I_S_TRX_OP_STATE_MAX_LEN, cache); |
528 | |
529 | if (row->trx_operation_state == NULL) { |
530 | |
531 | return(FALSE); |
532 | } |
533 | } else { |
534 | |
535 | row->trx_operation_state = NULL; |
536 | } |
537 | |
538 | row->trx_tables_in_use = trx->n_mysql_tables_in_use; |
539 | |
540 | row->trx_tables_locked = lock_number_of_tables_locked(&trx->lock); |
541 | |
542 | /* These are protected by both trx->mutex or lock_sys.mutex, |
543 | or just lock_sys.mutex. For reading, it suffices to hold |
544 | lock_sys.mutex. */ |
545 | |
546 | row->trx_lock_structs = UT_LIST_GET_LEN(trx->lock.trx_locks); |
547 | |
548 | row->trx_lock_memory_bytes = mem_heap_get_size(trx->lock.lock_heap); |
549 | |
550 | row->trx_rows_locked = lock_number_of_rows_locked(&trx->lock); |
551 | |
552 | row->trx_rows_modified = trx->undo_no; |
553 | |
554 | row->trx_concurrency_tickets = trx->n_tickets_to_enter_innodb; |
555 | |
556 | switch (trx->isolation_level) { |
557 | case TRX_ISO_READ_UNCOMMITTED: |
558 | row->trx_isolation_level = "READ UNCOMMITTED" ; |
559 | break; |
560 | case TRX_ISO_READ_COMMITTED: |
561 | row->trx_isolation_level = "READ COMMITTED" ; |
562 | break; |
563 | case TRX_ISO_REPEATABLE_READ: |
564 | row->trx_isolation_level = "REPEATABLE READ" ; |
565 | break; |
566 | case TRX_ISO_SERIALIZABLE: |
567 | row->trx_isolation_level = "SERIALIZABLE" ; |
568 | break; |
569 | /* Should not happen as TRX_ISO_READ_COMMITTED is default */ |
570 | default: |
571 | row->trx_isolation_level = "UNKNOWN" ; |
572 | } |
573 | |
574 | row->trx_unique_checks = (ibool) trx->check_unique_secondary; |
575 | |
576 | row->trx_foreign_key_checks = (ibool) trx->check_foreigns; |
577 | |
578 | s = trx->detailed_error; |
579 | |
580 | if (s != NULL && s[0] != '\0') { |
581 | |
582 | TRX_I_S_STRING_COPY(s, |
583 | row->trx_foreign_key_error, |
584 | TRX_I_S_TRX_FK_ERROR_MAX_LEN, cache); |
585 | |
586 | if (row->trx_foreign_key_error == NULL) { |
587 | |
588 | return(FALSE); |
589 | } |
590 | } else { |
591 | row->trx_foreign_key_error = NULL; |
592 | } |
593 | |
594 | row->trx_is_read_only = trx->read_only; |
595 | |
596 | row->trx_is_autocommit_non_locking = trx_is_autocommit_non_locking(trx); |
597 | |
598 | return(TRUE); |
599 | } |
600 | |
601 | /*******************************************************************//** |
602 | Format the nth field of "rec" and put it in "buf". The result is always |
603 | NUL-terminated. Returns the number of bytes that were written to "buf" |
604 | (including the terminating NUL). |
605 | @return end of the result */ |
606 | static |
607 | ulint |
608 | put_nth_field( |
609 | /*==========*/ |
610 | char* buf, /*!< out: buffer */ |
611 | ulint buf_size,/*!< in: buffer size in bytes */ |
612 | ulint n, /*!< in: number of field */ |
613 | const dict_index_t* index, /*!< in: index */ |
614 | const rec_t* rec, /*!< in: record */ |
615 | const ulint* offsets)/*!< in: record offsets, returned |
616 | by rec_get_offsets() */ |
617 | { |
618 | const byte* data; |
619 | ulint data_len; |
620 | dict_field_t* dict_field; |
621 | ulint ret; |
622 | |
623 | ut_ad(rec_offs_validate(rec, NULL, offsets)); |
624 | |
625 | if (buf_size == 0) { |
626 | |
627 | return(0); |
628 | } |
629 | |
630 | ret = 0; |
631 | |
632 | if (n > 0) { |
633 | /* we must append ", " before the actual data */ |
634 | |
635 | if (buf_size < 3) { |
636 | |
637 | buf[0] = '\0'; |
638 | return(1); |
639 | } |
640 | |
641 | memcpy(buf, ", " , 3); |
642 | |
643 | buf += 2; |
644 | buf_size -= 2; |
645 | ret += 2; |
646 | } |
647 | |
648 | /* now buf_size >= 1 */ |
649 | |
650 | data = rec_get_nth_field(rec, offsets, n, &data_len); |
651 | |
652 | dict_field = dict_index_get_nth_field(index, n); |
653 | |
654 | ret += row_raw_format((const char*) data, data_len, |
655 | dict_field, buf, buf_size); |
656 | |
657 | return(ret); |
658 | } |
659 | |
660 | /*******************************************************************//** |
661 | Fills the "lock_data" member of i_s_locks_row_t object. |
662 | If memory can not be allocated then FALSE is returned. |
663 | @return FALSE if allocation fails */ |
664 | static |
665 | ibool |
666 | fill_lock_data( |
667 | /*===========*/ |
668 | const char** lock_data,/*!< out: "lock_data" to fill */ |
669 | const lock_t* lock, /*!< in: lock used to find the data */ |
670 | ulint heap_no,/*!< in: rec num used to find the data */ |
671 | trx_i_s_cache_t* cache) /*!< in/out: cache where to store |
672 | volatile data */ |
673 | { |
674 | ut_a(lock_get_type(lock) == LOCK_REC); |
675 | |
676 | switch (heap_no) { |
677 | case PAGE_HEAP_NO_INFIMUM: |
678 | case PAGE_HEAP_NO_SUPREMUM: |
679 | *lock_data = ha_storage_put_str_memlim( |
680 | cache->storage, |
681 | heap_no == PAGE_HEAP_NO_INFIMUM |
682 | ? "infimum pseudo-record" |
683 | : "supremum pseudo-record" , |
684 | MAX_ALLOWED_FOR_STORAGE(cache)); |
685 | return(*lock_data != NULL); |
686 | } |
687 | |
688 | mtr_t mtr; |
689 | |
690 | const buf_block_t* block; |
691 | const page_t* page; |
692 | const rec_t* rec; |
693 | const dict_index_t* index; |
694 | ulint n_fields; |
695 | mem_heap_t* heap; |
696 | ulint offsets_onstack[REC_OFFS_NORMAL_SIZE]; |
697 | ulint* offsets; |
698 | char buf[TRX_I_S_LOCK_DATA_MAX_LEN]; |
699 | ulint buf_used; |
700 | ulint i; |
701 | |
702 | mtr_start(&mtr); |
703 | |
704 | block = buf_page_try_get(page_id_t(lock_rec_get_space_id(lock), |
705 | lock_rec_get_page_no(lock)), |
706 | &mtr); |
707 | |
708 | if (block == NULL) { |
709 | |
710 | *lock_data = NULL; |
711 | |
712 | mtr_commit(&mtr); |
713 | |
714 | return(TRUE); |
715 | } |
716 | |
717 | page = reinterpret_cast<const page_t*>(buf_block_get_frame(block)); |
718 | |
719 | rec_offs_init(offsets_onstack); |
720 | offsets = offsets_onstack; |
721 | |
722 | rec = page_find_rec_with_heap_no(page, heap_no); |
723 | |
724 | index = lock_rec_get_index(lock); |
725 | |
726 | n_fields = dict_index_get_n_unique(index); |
727 | |
728 | ut_a(n_fields > 0); |
729 | |
730 | heap = NULL; |
731 | offsets = rec_get_offsets(rec, index, offsets, true, n_fields, &heap); |
732 | |
733 | /* format and store the data */ |
734 | |
735 | buf_used = 0; |
736 | for (i = 0; i < n_fields; i++) { |
737 | |
738 | buf_used += put_nth_field( |
739 | buf + buf_used, sizeof(buf) - buf_used, |
740 | i, index, rec, offsets) - 1; |
741 | } |
742 | |
743 | *lock_data = (const char*) ha_storage_put_memlim( |
744 | cache->storage, buf, buf_used + 1, |
745 | MAX_ALLOWED_FOR_STORAGE(cache)); |
746 | |
747 | if (heap != NULL) { |
748 | |
749 | /* this means that rec_get_offsets() has created a new |
750 | heap and has stored offsets in it; check that this is |
751 | really the case and free the heap */ |
752 | ut_a(offsets != offsets_onstack); |
753 | mem_heap_free(heap); |
754 | } |
755 | |
756 | mtr_commit(&mtr); |
757 | |
758 | if (*lock_data == NULL) { |
759 | |
760 | return(FALSE); |
761 | } |
762 | |
763 | return(TRUE); |
764 | } |
765 | |
766 | /*******************************************************************//** |
767 | Fills i_s_locks_row_t object. Returns its first argument. |
768 | If memory can not be allocated then FALSE is returned. |
769 | @return FALSE if allocation fails */ |
770 | static |
771 | ibool |
772 | fill_locks_row( |
773 | /*===========*/ |
774 | i_s_locks_row_t* row, /*!< out: result object that's filled */ |
775 | const lock_t* lock, /*!< in: lock to get data from */ |
776 | ulint heap_no,/*!< in: lock's record number |
777 | or ULINT_UNDEFINED if the lock |
778 | is a table lock */ |
779 | trx_i_s_cache_t* cache) /*!< in/out: cache into which to copy |
780 | volatile strings */ |
781 | { |
782 | row->lock_trx_id = lock_get_trx_id(lock); |
783 | row->lock_mode = lock_get_mode_str(lock); |
784 | row->lock_type = lock_get_type_str(lock); |
785 | |
786 | row->lock_table = ha_storage_put_str_memlim( |
787 | cache->storage, lock_get_table_name(lock).m_name, |
788 | MAX_ALLOWED_FOR_STORAGE(cache)); |
789 | |
790 | /* memory could not be allocated */ |
791 | if (row->lock_table == NULL) { |
792 | |
793 | return(FALSE); |
794 | } |
795 | |
796 | switch (lock_get_type(lock)) { |
797 | case LOCK_REC: |
798 | row->lock_index = ha_storage_put_str_memlim( |
799 | cache->storage, lock_rec_get_index_name(lock), |
800 | MAX_ALLOWED_FOR_STORAGE(cache)); |
801 | |
802 | /* memory could not be allocated */ |
803 | if (row->lock_index == NULL) { |
804 | |
805 | return(FALSE); |
806 | } |
807 | |
808 | row->lock_space = lock_rec_get_space_id(lock); |
809 | row->lock_page = lock_rec_get_page_no(lock); |
810 | row->lock_rec = heap_no; |
811 | |
812 | if (!fill_lock_data(&row->lock_data, lock, heap_no, cache)) { |
813 | |
814 | /* memory could not be allocated */ |
815 | return(FALSE); |
816 | } |
817 | |
818 | break; |
819 | case LOCK_TABLE: |
820 | row->lock_index = NULL; |
821 | |
822 | row->lock_space = ULINT_UNDEFINED; |
823 | row->lock_page = ULINT_UNDEFINED; |
824 | row->lock_rec = ULINT_UNDEFINED; |
825 | |
826 | row->lock_data = NULL; |
827 | |
828 | break; |
829 | default: |
830 | ut_error; |
831 | } |
832 | |
833 | row->lock_table_id = lock_get_table_id(lock); |
834 | |
835 | row->hash_chain.value = row; |
836 | ut_ad(i_s_locks_row_validate(row)); |
837 | |
838 | return(TRUE); |
839 | } |
840 | |
841 | /*******************************************************************//** |
842 | Fills i_s_lock_waits_row_t object. Returns its first argument. |
843 | @return result object that's filled */ |
844 | static |
845 | i_s_lock_waits_row_t* |
846 | fill_lock_waits_row( |
847 | /*================*/ |
848 | i_s_lock_waits_row_t* row, /*!< out: result object |
849 | that's filled */ |
850 | const i_s_locks_row_t* requested_lock_row,/*!< in: pointer to the |
851 | relevant requested lock |
852 | row in innodb_locks */ |
853 | const i_s_locks_row_t* blocking_lock_row)/*!< in: pointer to the |
854 | relevant blocking lock |
855 | row in innodb_locks */ |
856 | { |
857 | ut_ad(i_s_locks_row_validate(requested_lock_row)); |
858 | ut_ad(i_s_locks_row_validate(blocking_lock_row)); |
859 | |
860 | row->requested_lock_row = requested_lock_row; |
861 | row->blocking_lock_row = blocking_lock_row; |
862 | |
863 | return(row); |
864 | } |
865 | |
866 | /*******************************************************************//** |
867 | Calculates a hash fold for a lock. For a record lock the fold is |
868 | calculated from 4 elements, which uniquely identify a lock at a given |
869 | point in time: transaction id, space id, page number, record number. |
870 | For a table lock the fold is table's id. |
871 | @return fold */ |
872 | static |
873 | ulint |
874 | fold_lock( |
875 | /*======*/ |
876 | const lock_t* lock, /*!< in: lock object to fold */ |
877 | ulint heap_no)/*!< in: lock's record number |
878 | or ULINT_UNDEFINED if the lock |
879 | is a table lock */ |
880 | { |
881 | #ifdef TEST_LOCK_FOLD_ALWAYS_DIFFERENT |
882 | static ulint fold = 0; |
883 | |
884 | return(fold++); |
885 | #else |
886 | ulint ret; |
887 | |
888 | switch (lock_get_type(lock)) { |
889 | case LOCK_REC: |
890 | ut_a(heap_no != ULINT_UNDEFINED); |
891 | |
892 | ret = ut_fold_ulint_pair((ulint) lock_get_trx_id(lock), |
893 | lock_rec_get_space_id(lock)); |
894 | |
895 | ret = ut_fold_ulint_pair(ret, |
896 | lock_rec_get_page_no(lock)); |
897 | |
898 | ret = ut_fold_ulint_pair(ret, heap_no); |
899 | |
900 | break; |
901 | case LOCK_TABLE: |
902 | /* this check is actually not necessary for continuing |
903 | correct operation, but something must have gone wrong if |
904 | it fails. */ |
905 | ut_a(heap_no == ULINT_UNDEFINED); |
906 | |
907 | ret = (ulint) lock_get_table_id(lock); |
908 | |
909 | break; |
910 | default: |
911 | ut_error; |
912 | } |
913 | |
914 | return(ret); |
915 | #endif |
916 | } |
917 | |
918 | /*******************************************************************//** |
919 | Checks whether i_s_locks_row_t object represents a lock_t object. |
920 | @return TRUE if they match */ |
921 | static |
922 | ibool |
923 | locks_row_eq_lock( |
924 | /*==============*/ |
925 | const i_s_locks_row_t* row, /*!< in: innodb_locks row */ |
926 | const lock_t* lock, /*!< in: lock object */ |
927 | ulint heap_no)/*!< in: lock's record number |
928 | or ULINT_UNDEFINED if the lock |
929 | is a table lock */ |
930 | { |
931 | ut_ad(i_s_locks_row_validate(row)); |
932 | #ifdef TEST_NO_LOCKS_ROW_IS_EVER_EQUAL_TO_LOCK_T |
933 | return(0); |
934 | #else |
935 | switch (lock_get_type(lock)) { |
936 | case LOCK_REC: |
937 | ut_a(heap_no != ULINT_UNDEFINED); |
938 | |
939 | return(row->lock_trx_id == lock_get_trx_id(lock) |
940 | && row->lock_space == lock_rec_get_space_id(lock) |
941 | && row->lock_page == lock_rec_get_page_no(lock) |
942 | && row->lock_rec == heap_no); |
943 | |
944 | case LOCK_TABLE: |
945 | /* this check is actually not necessary for continuing |
946 | correct operation, but something must have gone wrong if |
947 | it fails. */ |
948 | ut_a(heap_no == ULINT_UNDEFINED); |
949 | |
950 | return(row->lock_trx_id == lock_get_trx_id(lock) |
951 | && row->lock_table_id == lock_get_table_id(lock)); |
952 | |
953 | default: |
954 | ut_error; |
955 | return(FALSE); |
956 | } |
957 | #endif |
958 | } |
959 | |
960 | /*******************************************************************//** |
961 | Searches for a row in the innodb_locks cache that has a specified id. |
962 | This happens in O(1) time since a hash table is used. Returns pointer to |
963 | the row or NULL if none is found. |
964 | @return row or NULL */ |
965 | static |
966 | i_s_locks_row_t* |
967 | search_innodb_locks( |
968 | /*================*/ |
969 | trx_i_s_cache_t* cache, /*!< in: cache */ |
970 | const lock_t* lock, /*!< in: lock to search for */ |
971 | ulint heap_no)/*!< in: lock's record number |
972 | or ULINT_UNDEFINED if the lock |
973 | is a table lock */ |
974 | { |
975 | i_s_hash_chain_t* hash_chain; |
976 | |
977 | HASH_SEARCH( |
978 | /* hash_chain->"next" */ |
979 | next, |
980 | /* the hash table */ |
981 | cache->locks_hash, |
982 | /* fold */ |
983 | fold_lock(lock, heap_no), |
984 | /* the type of the next variable */ |
985 | i_s_hash_chain_t*, |
986 | /* auxiliary variable */ |
987 | hash_chain, |
988 | /* assertion on every traversed item */ |
989 | ut_ad(i_s_locks_row_validate(hash_chain->value)), |
990 | /* this determines if we have found the lock */ |
991 | locks_row_eq_lock(hash_chain->value, lock, heap_no)); |
992 | |
993 | if (hash_chain == NULL) { |
994 | |
995 | return(NULL); |
996 | } |
997 | /* else */ |
998 | |
999 | return(hash_chain->value); |
1000 | } |
1001 | |
1002 | /*******************************************************************//** |
1003 | Adds new element to the locks cache, enlarging it if necessary. |
1004 | Returns a pointer to the added row. If the row is already present then |
1005 | no row is added and a pointer to the existing row is returned. |
1006 | If row can not be allocated then NULL is returned. |
1007 | @return row */ |
1008 | static |
1009 | i_s_locks_row_t* |
1010 | add_lock_to_cache( |
1011 | /*==============*/ |
1012 | trx_i_s_cache_t* cache, /*!< in/out: cache */ |
1013 | const lock_t* lock, /*!< in: the element to add */ |
1014 | ulint heap_no)/*!< in: lock's record number |
1015 | or ULINT_UNDEFINED if the lock |
1016 | is a table lock */ |
1017 | { |
1018 | i_s_locks_row_t* dst_row; |
1019 | |
1020 | #ifdef TEST_ADD_EACH_LOCKS_ROW_MANY_TIMES |
1021 | ulint i; |
1022 | for (i = 0; i < 10000; i++) { |
1023 | #endif |
1024 | #ifndef TEST_DO_NOT_CHECK_FOR_DUPLICATE_ROWS |
1025 | /* quit if this lock is already present */ |
1026 | dst_row = search_innodb_locks(cache, lock, heap_no); |
1027 | if (dst_row != NULL) { |
1028 | |
1029 | ut_ad(i_s_locks_row_validate(dst_row)); |
1030 | return(dst_row); |
1031 | } |
1032 | #endif |
1033 | |
1034 | dst_row = (i_s_locks_row_t*) |
1035 | table_cache_create_empty_row(&cache->innodb_locks, cache); |
1036 | |
1037 | /* memory could not be allocated */ |
1038 | if (dst_row == NULL) { |
1039 | |
1040 | return(NULL); |
1041 | } |
1042 | |
1043 | if (!fill_locks_row(dst_row, lock, heap_no, cache)) { |
1044 | |
1045 | /* memory could not be allocated */ |
1046 | cache->innodb_locks.rows_used--; |
1047 | return(NULL); |
1048 | } |
1049 | |
1050 | #ifndef TEST_DO_NOT_INSERT_INTO_THE_HASH_TABLE |
1051 | HASH_INSERT( |
1052 | /* the type used in the hash chain */ |
1053 | i_s_hash_chain_t, |
1054 | /* hash_chain->"next" */ |
1055 | next, |
1056 | /* the hash table */ |
1057 | cache->locks_hash, |
1058 | /* fold */ |
1059 | fold_lock(lock, heap_no), |
1060 | /* add this data to the hash */ |
1061 | &dst_row->hash_chain); |
1062 | #endif |
1063 | #ifdef TEST_ADD_EACH_LOCKS_ROW_MANY_TIMES |
1064 | } /* for()-loop */ |
1065 | #endif |
1066 | |
1067 | ut_ad(i_s_locks_row_validate(dst_row)); |
1068 | return(dst_row); |
1069 | } |
1070 | |
1071 | /*******************************************************************//** |
1072 | Adds new pair of locks to the lock waits cache. |
1073 | If memory can not be allocated then FALSE is returned. |
1074 | @return FALSE if allocation fails */ |
1075 | static |
1076 | ibool |
1077 | add_lock_wait_to_cache( |
1078 | /*===================*/ |
1079 | trx_i_s_cache_t* cache, /*!< in/out: cache */ |
1080 | const i_s_locks_row_t* requested_lock_row,/*!< in: pointer to the |
1081 | relevant requested lock |
1082 | row in innodb_locks */ |
1083 | const i_s_locks_row_t* blocking_lock_row)/*!< in: pointer to the |
1084 | relevant blocking lock |
1085 | row in innodb_locks */ |
1086 | { |
1087 | i_s_lock_waits_row_t* dst_row; |
1088 | |
1089 | dst_row = (i_s_lock_waits_row_t*) |
1090 | table_cache_create_empty_row(&cache->innodb_lock_waits, |
1091 | cache); |
1092 | |
1093 | /* memory could not be allocated */ |
1094 | if (dst_row == NULL) { |
1095 | |
1096 | return(FALSE); |
1097 | } |
1098 | |
1099 | fill_lock_waits_row(dst_row, requested_lock_row, blocking_lock_row); |
1100 | |
1101 | return(TRUE); |
1102 | } |
1103 | |
1104 | /*******************************************************************//** |
1105 | Adds transaction's relevant (important) locks to cache. |
1106 | If the transaction is waiting, then the wait lock is added to |
1107 | innodb_locks and a pointer to the added row is returned in |
1108 | requested_lock_row, otherwise requested_lock_row is set to NULL. |
1109 | If rows can not be allocated then FALSE is returned and the value of |
1110 | requested_lock_row is undefined. |
1111 | @return FALSE if allocation fails */ |
1112 | static |
1113 | ibool |
1114 | add_trx_relevant_locks_to_cache( |
1115 | /*============================*/ |
1116 | trx_i_s_cache_t* cache, /*!< in/out: cache */ |
1117 | const trx_t* trx, /*!< in: transaction */ |
1118 | i_s_locks_row_t** requested_lock_row)/*!< out: pointer to the |
1119 | requested lock row, or NULL or |
1120 | undefined */ |
1121 | { |
1122 | ut_ad(lock_mutex_own()); |
1123 | |
1124 | /* If transaction is waiting we add the wait lock and all locks |
1125 | from another transactions that are blocking the wait lock. */ |
1126 | if (trx->lock.que_state == TRX_QUE_LOCK_WAIT) { |
1127 | |
1128 | const lock_t* curr_lock; |
1129 | ulint wait_lock_heap_no; |
1130 | i_s_locks_row_t* blocking_lock_row; |
1131 | lock_queue_iterator_t iter; |
1132 | |
1133 | ut_a(trx->lock.wait_lock != NULL); |
1134 | |
1135 | wait_lock_heap_no |
1136 | = wait_lock_get_heap_no(trx->lock.wait_lock); |
1137 | |
1138 | /* add the requested lock */ |
1139 | *requested_lock_row |
1140 | = add_lock_to_cache(cache, trx->lock.wait_lock, |
1141 | wait_lock_heap_no); |
1142 | |
1143 | /* memory could not be allocated */ |
1144 | if (*requested_lock_row == NULL) { |
1145 | |
1146 | return(FALSE); |
1147 | } |
1148 | |
1149 | /* then iterate over the locks before the wait lock and |
1150 | add the ones that are blocking it */ |
1151 | |
1152 | lock_queue_iterator_reset(&iter, trx->lock.wait_lock, |
1153 | ULINT_UNDEFINED); |
1154 | |
1155 | for (curr_lock = lock_queue_iterator_get_prev(&iter); |
1156 | curr_lock != NULL; |
1157 | curr_lock = lock_queue_iterator_get_prev(&iter)) { |
1158 | |
1159 | if (lock_has_to_wait(trx->lock.wait_lock, |
1160 | curr_lock)) { |
1161 | |
1162 | /* add the lock that is |
1163 | blocking trx->lock.wait_lock */ |
1164 | blocking_lock_row |
1165 | = add_lock_to_cache( |
1166 | cache, curr_lock, |
1167 | /* heap_no is the same |
1168 | for the wait and waited |
1169 | locks */ |
1170 | wait_lock_heap_no); |
1171 | |
1172 | /* memory could not be allocated */ |
1173 | if (blocking_lock_row == NULL) { |
1174 | |
1175 | return(FALSE); |
1176 | } |
1177 | |
1178 | /* add the relation between both locks |
1179 | to innodb_lock_waits */ |
1180 | if (!add_lock_wait_to_cache( |
1181 | cache, *requested_lock_row, |
1182 | blocking_lock_row)) { |
1183 | |
1184 | /* memory could not be allocated */ |
1185 | return(FALSE); |
1186 | } |
1187 | } |
1188 | } |
1189 | } else { |
1190 | |
1191 | *requested_lock_row = NULL; |
1192 | } |
1193 | |
1194 | return(TRUE); |
1195 | } |
1196 | |
1197 | /** The minimum time that a cache must not be updated after it has been |
1198 | read for the last time; measured in microseconds. We use this technique |
1199 | to ensure that SELECTs which join several INFORMATION SCHEMA tables read |
1200 | the same version of the cache. */ |
1201 | #define CACHE_MIN_IDLE_TIME_US 100000 /* 0.1 sec */ |
1202 | |
1203 | /*******************************************************************//** |
1204 | Checks if the cache can safely be updated. |
1205 | @return TRUE if can be updated */ |
1206 | static |
1207 | ibool |
1208 | can_cache_be_updated( |
1209 | /*=================*/ |
1210 | trx_i_s_cache_t* cache) /*!< in: cache */ |
1211 | { |
1212 | uintmax_t now; |
1213 | |
1214 | /* Here we read cache->last_read without acquiring its mutex |
1215 | because last_read is only updated when a shared rw lock on the |
1216 | whole cache is being held (see trx_i_s_cache_end_read()) and |
1217 | we are currently holding an exclusive rw lock on the cache. |
1218 | So it is not possible for last_read to be updated while we are |
1219 | reading it. */ |
1220 | |
1221 | ut_ad(rw_lock_own(cache->rw_lock, RW_LOCK_X)); |
1222 | |
1223 | now = ut_time_us(NULL); |
1224 | if (now - cache->last_read > CACHE_MIN_IDLE_TIME_US) { |
1225 | |
1226 | return(TRUE); |
1227 | } |
1228 | |
1229 | return(FALSE); |
1230 | } |
1231 | |
1232 | /*******************************************************************//** |
1233 | Declare a cache empty, preparing it to be filled up. Not all resources |
1234 | are freed because they can be reused. */ |
1235 | static |
1236 | void |
1237 | trx_i_s_cache_clear( |
1238 | /*================*/ |
1239 | trx_i_s_cache_t* cache) /*!< out: cache to clear */ |
1240 | { |
1241 | cache->innodb_trx.rows_used = 0; |
1242 | cache->innodb_locks.rows_used = 0; |
1243 | cache->innodb_lock_waits.rows_used = 0; |
1244 | |
1245 | hash_table_clear(cache->locks_hash); |
1246 | |
1247 | ha_storage_empty(&cache->storage); |
1248 | } |
1249 | |
1250 | |
1251 | /** |
1252 | Add transactions to innodb_trx's cache. |
1253 | |
1254 | We also add all locks that are relevant to each transaction into |
1255 | innodb_locks' and innodb_lock_waits' caches. |
1256 | */ |
1257 | |
1258 | static void fetch_data_into_cache_low(trx_i_s_cache_t *cache, const trx_t *trx) |
1259 | { |
1260 | i_s_locks_row_t *requested_lock_row; |
1261 | |
1262 | assert_trx_nonlocking_or_in_list(trx); |
1263 | |
1264 | if (add_trx_relevant_locks_to_cache(cache, trx, &requested_lock_row)) |
1265 | { |
1266 | if (i_s_trx_row_t *trx_row= reinterpret_cast<i_s_trx_row_t*>( |
1267 | table_cache_create_empty_row(&cache->innodb_trx, cache))) |
1268 | { |
1269 | if (fill_trx_row(trx_row, trx, requested_lock_row, cache)) |
1270 | return; |
1271 | --cache->innodb_trx.rows_used; |
1272 | } |
1273 | } |
1274 | |
1275 | /* memory could not be allocated */ |
1276 | cache->is_truncated= true; |
1277 | } |
1278 | |
1279 | |
1280 | /** |
1281 | Fetches the data needed to fill the 3 INFORMATION SCHEMA tables into the |
1282 | table cache buffer. Cache must be locked for write. |
1283 | */ |
1284 | |
1285 | static void fetch_data_into_cache(trx_i_s_cache_t *cache) |
1286 | { |
1287 | ut_ad(lock_mutex_own()); |
1288 | trx_i_s_cache_clear(cache); |
1289 | |
1290 | /* Capture the state of transactions */ |
1291 | mutex_enter(&trx_sys.mutex); |
1292 | for (const trx_t *trx= UT_LIST_GET_FIRST(trx_sys.trx_list); |
1293 | trx != NULL; |
1294 | trx= UT_LIST_GET_NEXT(trx_list, trx)) |
1295 | { |
1296 | if (trx_is_started(trx) && trx != purge_sys.query->trx) |
1297 | { |
1298 | fetch_data_into_cache_low(cache, trx); |
1299 | if (cache->is_truncated) |
1300 | break; |
1301 | } |
1302 | } |
1303 | mutex_exit(&trx_sys.mutex); |
1304 | cache->is_truncated= false; |
1305 | } |
1306 | |
1307 | |
1308 | /*******************************************************************//** |
1309 | Update the transactions cache if it has not been read for some time. |
1310 | Called from handler/i_s.cc. |
1311 | @return 0 - fetched, 1 - not */ |
1312 | int |
1313 | trx_i_s_possibly_fetch_data_into_cache( |
1314 | /*===================================*/ |
1315 | trx_i_s_cache_t* cache) /*!< in/out: cache */ |
1316 | { |
1317 | if (!can_cache_be_updated(cache)) { |
1318 | |
1319 | return(1); |
1320 | } |
1321 | |
1322 | /* We need to read trx_sys and record/table lock queues */ |
1323 | |
1324 | lock_mutex_enter(); |
1325 | fetch_data_into_cache(cache); |
1326 | lock_mutex_exit(); |
1327 | |
1328 | /* update cache last read time */ |
1329 | time_t now = ut_time_us(NULL); |
1330 | cache->last_read = now; |
1331 | |
1332 | return(0); |
1333 | } |
1334 | |
1335 | /*******************************************************************//** |
1336 | Returns TRUE if the data in the cache is truncated due to the memory |
1337 | limit posed by TRX_I_S_MEM_LIMIT. |
1338 | @return TRUE if truncated */ |
1339 | bool |
1340 | trx_i_s_cache_is_truncated( |
1341 | /*=======================*/ |
1342 | trx_i_s_cache_t* cache) /*!< in: cache */ |
1343 | { |
1344 | return(cache->is_truncated); |
1345 | } |
1346 | |
1347 | /*******************************************************************//** |
1348 | Initialize INFORMATION SCHEMA trx related cache. */ |
1349 | void |
1350 | trx_i_s_cache_init( |
1351 | /*===============*/ |
1352 | trx_i_s_cache_t* cache) /*!< out: cache to init */ |
1353 | { |
1354 | /* The latching is done in the following order: |
1355 | acquire trx_i_s_cache_t::rw_lock, X |
1356 | acquire lock mutex |
1357 | release lock mutex |
1358 | release trx_i_s_cache_t::rw_lock |
1359 | acquire trx_i_s_cache_t::rw_lock, S |
1360 | acquire trx_i_s_cache_t::last_read_mutex |
1361 | release trx_i_s_cache_t::last_read_mutex |
1362 | release trx_i_s_cache_t::rw_lock */ |
1363 | |
1364 | cache->rw_lock = static_cast<rw_lock_t*>( |
1365 | ut_malloc_nokey(sizeof(*cache->rw_lock))); |
1366 | |
1367 | rw_lock_create(trx_i_s_cache_lock_key, cache->rw_lock, |
1368 | SYNC_TRX_I_S_RWLOCK); |
1369 | |
1370 | cache->last_read = 0; |
1371 | |
1372 | mutex_create(LATCH_ID_CACHE_LAST_READ, &cache->last_read_mutex); |
1373 | |
1374 | table_cache_init(&cache->innodb_trx, sizeof(i_s_trx_row_t)); |
1375 | table_cache_init(&cache->innodb_locks, sizeof(i_s_locks_row_t)); |
1376 | table_cache_init(&cache->innodb_lock_waits, |
1377 | sizeof(i_s_lock_waits_row_t)); |
1378 | |
1379 | cache->locks_hash = hash_create(LOCKS_HASH_CELLS_NUM); |
1380 | |
1381 | cache->storage = ha_storage_create(CACHE_STORAGE_INITIAL_SIZE, |
1382 | CACHE_STORAGE_HASH_CELLS); |
1383 | |
1384 | cache->mem_allocd = 0; |
1385 | |
1386 | cache->is_truncated = false; |
1387 | } |
1388 | |
1389 | /*******************************************************************//** |
1390 | Free the INFORMATION SCHEMA trx related cache. */ |
1391 | void |
1392 | trx_i_s_cache_free( |
1393 | /*===============*/ |
1394 | trx_i_s_cache_t* cache) /*!< in, own: cache to free */ |
1395 | { |
1396 | rw_lock_free(cache->rw_lock); |
1397 | ut_free(cache->rw_lock); |
1398 | cache->rw_lock = NULL; |
1399 | |
1400 | mutex_free(&cache->last_read_mutex); |
1401 | |
1402 | hash_table_free(cache->locks_hash); |
1403 | ha_storage_free(cache->storage); |
1404 | table_cache_free(&cache->innodb_trx); |
1405 | table_cache_free(&cache->innodb_locks); |
1406 | table_cache_free(&cache->innodb_lock_waits); |
1407 | } |
1408 | |
1409 | /*******************************************************************//** |
1410 | Issue a shared/read lock on the tables cache. */ |
1411 | void |
1412 | trx_i_s_cache_start_read( |
1413 | /*=====================*/ |
1414 | trx_i_s_cache_t* cache) /*!< in: cache */ |
1415 | { |
1416 | rw_lock_s_lock(cache->rw_lock); |
1417 | } |
1418 | |
1419 | /*******************************************************************//** |
1420 | Release a shared/read lock on the tables cache. */ |
1421 | void |
1422 | trx_i_s_cache_end_read( |
1423 | /*===================*/ |
1424 | trx_i_s_cache_t* cache) /*!< in: cache */ |
1425 | { |
1426 | uintmax_t now; |
1427 | |
1428 | ut_ad(rw_lock_own(cache->rw_lock, RW_LOCK_S)); |
1429 | |
1430 | /* update cache last read time */ |
1431 | now = ut_time_us(NULL); |
1432 | mutex_enter(&cache->last_read_mutex); |
1433 | cache->last_read = now; |
1434 | mutex_exit(&cache->last_read_mutex); |
1435 | |
1436 | rw_lock_s_unlock(cache->rw_lock); |
1437 | } |
1438 | |
1439 | /*******************************************************************//** |
1440 | Issue an exclusive/write lock on the tables cache. */ |
1441 | void |
1442 | trx_i_s_cache_start_write( |
1443 | /*======================*/ |
1444 | trx_i_s_cache_t* cache) /*!< in: cache */ |
1445 | { |
1446 | rw_lock_x_lock(cache->rw_lock); |
1447 | } |
1448 | |
1449 | /*******************************************************************//** |
1450 | Release an exclusive/write lock on the tables cache. */ |
1451 | void |
1452 | trx_i_s_cache_end_write( |
1453 | /*====================*/ |
1454 | trx_i_s_cache_t* cache) /*!< in: cache */ |
1455 | { |
1456 | ut_ad(rw_lock_own(cache->rw_lock, RW_LOCK_X)); |
1457 | |
1458 | rw_lock_x_unlock(cache->rw_lock); |
1459 | } |
1460 | |
1461 | /*******************************************************************//** |
1462 | Selects a INFORMATION SCHEMA table cache from the whole cache. |
1463 | @return table cache */ |
1464 | static |
1465 | i_s_table_cache_t* |
1466 | cache_select_table( |
1467 | /*===============*/ |
1468 | trx_i_s_cache_t* cache, /*!< in: whole cache */ |
1469 | enum i_s_table table) /*!< in: which table */ |
1470 | { |
1471 | i_s_table_cache_t* table_cache; |
1472 | |
1473 | ut_ad(rw_lock_own(cache->rw_lock, RW_LOCK_S) |
1474 | || rw_lock_own(cache->rw_lock, RW_LOCK_X)); |
1475 | |
1476 | switch (table) { |
1477 | case I_S_INNODB_TRX: |
1478 | table_cache = &cache->innodb_trx; |
1479 | break; |
1480 | case I_S_INNODB_LOCKS: |
1481 | table_cache = &cache->innodb_locks; |
1482 | break; |
1483 | case I_S_INNODB_LOCK_WAITS: |
1484 | table_cache = &cache->innodb_lock_waits; |
1485 | break; |
1486 | default: |
1487 | ut_error; |
1488 | } |
1489 | |
1490 | return(table_cache); |
1491 | } |
1492 | |
1493 | /*******************************************************************//** |
1494 | Retrieves the number of used rows in the cache for a given |
1495 | INFORMATION SCHEMA table. |
1496 | @return number of rows */ |
1497 | ulint |
1498 | trx_i_s_cache_get_rows_used( |
1499 | /*========================*/ |
1500 | trx_i_s_cache_t* cache, /*!< in: cache */ |
1501 | enum i_s_table table) /*!< in: which table */ |
1502 | { |
1503 | i_s_table_cache_t* table_cache; |
1504 | |
1505 | table_cache = cache_select_table(cache, table); |
1506 | |
1507 | return(table_cache->rows_used); |
1508 | } |
1509 | |
1510 | /*******************************************************************//** |
1511 | Retrieves the nth row (zero-based) in the cache for a given |
1512 | INFORMATION SCHEMA table. |
1513 | @return row */ |
1514 | void* |
1515 | trx_i_s_cache_get_nth_row( |
1516 | /*======================*/ |
1517 | trx_i_s_cache_t* cache, /*!< in: cache */ |
1518 | enum i_s_table table, /*!< in: which table */ |
1519 | ulint n) /*!< in: row number */ |
1520 | { |
1521 | i_s_table_cache_t* table_cache; |
1522 | ulint i; |
1523 | void* row; |
1524 | |
1525 | table_cache = cache_select_table(cache, table); |
1526 | |
1527 | ut_a(n < table_cache->rows_used); |
1528 | |
1529 | row = NULL; |
1530 | |
1531 | for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) { |
1532 | |
1533 | if (table_cache->chunks[i].offset |
1534 | + table_cache->chunks[i].rows_allocd > n) { |
1535 | |
1536 | row = (char*) table_cache->chunks[i].base |
1537 | + (n - table_cache->chunks[i].offset) |
1538 | * table_cache->row_size; |
1539 | break; |
1540 | } |
1541 | } |
1542 | |
1543 | ut_a(row != NULL); |
1544 | |
1545 | return(row); |
1546 | } |
1547 | |
1548 | /*******************************************************************//** |
1549 | Crafts a lock id string from a i_s_locks_row_t object. Returns its |
1550 | second argument. This function aborts if there is not enough space in |
1551 | lock_id. Be sure to provide at least TRX_I_S_LOCK_ID_MAX_LEN + 1 if you |
1552 | want to be 100% sure that it will not abort. |
1553 | @return resulting lock id */ |
1554 | char* |
1555 | trx_i_s_create_lock_id( |
1556 | /*===================*/ |
1557 | const i_s_locks_row_t* row, /*!< in: innodb_locks row */ |
1558 | char* lock_id,/*!< out: resulting lock_id */ |
1559 | ulint lock_id_size)/*!< in: size of the lock id |
1560 | buffer */ |
1561 | { |
1562 | int res_len; |
1563 | |
1564 | /* please adjust TRX_I_S_LOCK_ID_MAX_LEN if you change this */ |
1565 | |
1566 | if (row->lock_space != ULINT_UNDEFINED) { |
1567 | /* record lock */ |
1568 | res_len = snprintf(lock_id, lock_id_size, |
1569 | TRX_ID_FMT |
1570 | ":" ULINTPF ":" ULINTPF ":" ULINTPF, |
1571 | row->lock_trx_id, row->lock_space, |
1572 | row->lock_page, row->lock_rec); |
1573 | } else { |
1574 | /* table lock */ |
1575 | res_len = snprintf(lock_id, lock_id_size, |
1576 | TRX_ID_FMT":" UINT64PF, |
1577 | row->lock_trx_id, |
1578 | row->lock_table_id); |
1579 | } |
1580 | |
1581 | /* the typecast is safe because snprintf(3) never returns |
1582 | negative result */ |
1583 | ut_a(res_len >= 0); |
1584 | ut_a((ulint) res_len < lock_id_size); |
1585 | |
1586 | return(lock_id); |
1587 | } |
1588 | |