1
2// vim:sw=2:ai
3
4/*
5 * Copyright (C) 2010 DeNA Co.,Ltd.. All rights reserved.
6 * See COPYRIGHT.txt for details.
7 */
8
9#include <my_global.h>
10#include <string.h>
11
12#include "database.hpp"
13#include "string_util.hpp"
14#include "escape.hpp"
15#include "mysql_incl.hpp"
16
17#define DBG_KEY(x)
18#define DBG_SHUT(x)
19#define DBG_LOCK(x)
20#define DBG_THR(x)
21#define DBG_CMP(x)
22#define DBG_FLD(x)
23#define DBG_FILTER(x)
24#define DBG_REFCNT(x)
25#define DBG_KEYLEN(x)
26#define DBG_DELETED
27
28/* status variables */
29unsigned long long int open_tables_count;
30unsigned long long int close_tables_count;
31unsigned long long int lock_tables_count;
32unsigned long long int unlock_tables_count;
33unsigned long long int index_exec_count;
34
35namespace dena {
36
37prep_stmt::prep_stmt()
38 : dbctx(0), table_id(static_cast<size_t>(-1)),
39 idxnum(static_cast<size_t>(-1))
40{
41}
42prep_stmt::prep_stmt(dbcontext_i *c, size_t tbl, size_t idx,
43 const fields_type& rf, const fields_type& ff)
44 : dbctx(c), table_id(tbl), idxnum(idx), ret_fields(rf), filter_fields(ff)
45{
46 if (dbctx) {
47 dbctx->table_addref(table_id);
48 }
49}
50prep_stmt::~prep_stmt()
51{
52 if (dbctx) {
53 dbctx->table_release(table_id);
54 }
55}
56
57prep_stmt::prep_stmt(const prep_stmt& x)
58 : dbctx(x.dbctx), table_id(x.table_id), idxnum(x.idxnum),
59 ret_fields(x.ret_fields), filter_fields(x.filter_fields)
60{
61 if (dbctx) {
62 dbctx->table_addref(table_id);
63 }
64}
65
66prep_stmt&
67prep_stmt::operator =(const prep_stmt& x)
68{
69 if (this != &x) {
70 if (dbctx) {
71 dbctx->table_release(table_id);
72 }
73 dbctx = x.dbctx;
74 table_id = x.table_id;
75 idxnum = x.idxnum;
76 ret_fields = x.ret_fields;
77 filter_fields = x.filter_fields;
78 if (dbctx) {
79 dbctx->table_addref(table_id);
80 }
81 }
82 return *this;
83}
84
85struct database : public database_i, private noncopyable {
86 database(const config& c);
87 virtual ~database();
88 virtual dbcontext_ptr create_context(bool for_write) volatile;
89 virtual void stop() volatile;
90 virtual const config& get_conf() const volatile;
91 public:
92 int child_running;
93 private:
94 config conf;
95};
96
97struct tablevec_entry {
98 TABLE *table;
99 size_t refcount;
100 bool modified;
101 tablevec_entry() : table(0), refcount(0), modified(false) { }
102};
103
104struct expr_user_lock : private noncopyable {
105 expr_user_lock(THD *thd, int timeout)
106 : lck_key(thd, "handlersocket_wr", 16, &my_charset_latin1),
107 lck_timeout(thd, timeout),
108 lck_func_get_lock(thd, &lck_key, &lck_timeout),
109 lck_func_release_lock(thd, &lck_key)
110 {
111 lck_key.fix_fields(thd, 0);
112 lck_timeout.fix_fields(thd, 0);
113 lck_func_get_lock.fix_fields(thd, 0);
114 lck_func_release_lock.fix_fields(thd, 0);
115 }
116 long long get_lock() {
117 return lck_func_get_lock.val_int();
118 }
119 long long release_lock() {
120 return lck_func_release_lock.val_int();
121 }
122 private:
123 Item_string lck_key;
124 Item_int lck_timeout;
125 Item_func_get_lock lck_func_get_lock;
126 Item_func_release_lock lck_func_release_lock;
127};
128
129struct dbcontext : public dbcontext_i, private noncopyable {
130 dbcontext(volatile database *d, bool for_write);
131 virtual ~dbcontext();
132 virtual void init_thread(const void *stack_botton,
133 volatile int& shutdown_flag);
134 virtual void term_thread();
135 virtual bool check_alive();
136 virtual void lock_tables_if();
137 virtual void unlock_tables_if();
138 virtual bool get_commit_error();
139 virtual void clear_error();
140 virtual void close_tables_if();
141 virtual void table_addref(size_t tbl_id);
142 virtual void table_release(size_t tbl_id);
143 virtual void cmd_open(dbcallback_i& cb, const cmd_open_args& args);
144 virtual void cmd_exec(dbcallback_i& cb, const cmd_exec_args& args);
145 virtual void set_statistics(size_t num_conns, size_t num_active);
146 private:
147 int set_thread_message(const char *fmt, ...)
148 __attribute__((format (printf, 2, 3)));
149 bool parse_fields(TABLE *const table, const char *str,
150 prep_stmt::fields_type& flds);
151 void cmd_insert_internal(dbcallback_i& cb, const prep_stmt& pst,
152 const string_ref *fvals, size_t fvalslen);
153 void cmd_sql_internal(dbcallback_i& cb, const prep_stmt& pst,
154 const string_ref *fvals, size_t fvalslen);
155 void cmd_find_internal(dbcallback_i& cb, const prep_stmt& pst,
156 ha_rkey_function find_flag, const cmd_exec_args& args);
157 size_t calc_filter_buf_size(TABLE *table, const prep_stmt& pst,
158 const record_filter *filters);
159 bool fill_filter_buf(TABLE *table, const prep_stmt& pst,
160 const record_filter *filters, uchar *filter_buf, size_t len);
161 int check_filter(dbcallback_i& cb, TABLE *table, const prep_stmt& pst,
162 const record_filter *filters, const uchar *filter_buf);
163 void resp_record(dbcallback_i& cb, TABLE *const table, const prep_stmt& pst);
164 void dump_record(dbcallback_i& cb, TABLE *const table, const prep_stmt& pst);
165 int modify_record(dbcallback_i& cb, TABLE *const table,
166 const prep_stmt& pst, const cmd_exec_args& args, char mod_op,
167 size_t& modified_count);
168 private:
169 typedef std::vector<tablevec_entry> table_vec_type;
170 typedef std::pair<std::string, std::string> table_name_type;
171 typedef std::map<table_name_type, size_t> table_map_type;
172 private:
173 volatile database *const dbref;
174 bool for_write_flag;
175 THD *thd;
176 MYSQL_LOCK *lock;
177 bool lock_failed;
178 std::auto_ptr<expr_user_lock> user_lock;
179 int user_level_lock_timeout;
180 bool user_level_lock_locked;
181 bool commit_error;
182 std::vector<char> info_message_buf;
183 table_vec_type table_vec;
184 table_map_type table_map;
185};
186
187database::database(const config& c)
188 : child_running(1), conf(c)
189{
190}
191
192database::~database()
193{
194}
195
196dbcontext_ptr
197database::create_context(bool for_write) volatile
198{
199 return dbcontext_ptr(new dbcontext(this, for_write));
200}
201
202void
203database::stop() volatile
204{
205 child_running = false;
206}
207
208const config&
209database::get_conf() const volatile
210{
211 return const_cast<const config&>(conf);
212}
213
214database_ptr
215database_i::create(const config& conf)
216{
217 return database_ptr(new database(conf));
218}
219
220dbcontext::dbcontext(volatile database *d, bool for_write)
221 : dbref(d), for_write_flag(for_write), thd(0), lock(0), lock_failed(false),
222 user_level_lock_timeout(0), user_level_lock_locked(false),
223 commit_error(false)
224{
225 info_message_buf.resize(8192);
226 user_level_lock_timeout = d->get_conf().get_int("wrlock_timeout", 12);
227}
228
229dbcontext::~dbcontext()
230{
231}
232
233namespace {
234
235int
236wait_server_to_start(THD *thd, volatile int& shutdown_flag)
237{
238 int r = 0;
239 DBG_SHUT(fprintf(stderr, "HNDSOCK wsts\n"));
240 pthread_mutex_lock(&LOCK_server_started);
241 while (!mysqld_server_started) {
242 timespec abstime;
243 set_timespec(abstime, 1);
244 pthread_cond_timedwait(&COND_server_started, &LOCK_server_started,
245 &abstime);
246 pthread_mutex_unlock(&LOCK_server_started);
247 pthread_mutex_lock(&thd->mysys_var->mutex);
248 killed_state st = thd->killed;
249 pthread_mutex_unlock(&thd->mysys_var->mutex);
250 DBG_SHUT(fprintf(stderr, "HNDSOCK wsts kst %d\n", (int)st));
251 pthread_mutex_lock(&LOCK_server_started);
252 if (st != NOT_KILLED) {
253 DBG_SHUT(fprintf(stderr, "HNDSOCK wsts kst %d break\n", (int)st));
254 r = -1;
255 break;
256 }
257 if (shutdown_flag) {
258 DBG_SHUT(fprintf(stderr, "HNDSOCK wsts kst shut break\n"));
259 r = -1;
260 break;
261 }
262 }
263 pthread_mutex_unlock(&LOCK_server_started);
264 DBG_SHUT(fprintf(stderr, "HNDSOCK wsts done\n"));
265 return r;
266}
267
268}; // namespace
269
270#define DENA_THR_OFFSETOF(fld) ((char *)(&thd->fld) - (char *)thd)
271
272void
273dbcontext::init_thread(const void *stack_bottom, volatile int& shutdown_flag)
274{
275 DBG_THR(fprintf(stderr, "HNDSOCK init thread\n"));
276 {
277 my_thread_init();
278 thd = new THD(0);
279 thd->thread_stack = (char *)stack_bottom;
280 DBG_THR(fprintf(stderr,
281 "thread_stack = %p sizeof(THD)=%zu sizeof(mtx)=%zu "
282 "O: %zu %zu %zu %zu %zu %zu %zu\n",
283 thd->thread_stack, sizeof(THD), sizeof(LOCK_thread_count),
284 DENA_THR_OFFSETOF(mdl_context),
285 DENA_THR_OFFSETOF(net),
286 DENA_THR_OFFSETOF(LOCK_thd_data),
287 DENA_THR_OFFSETOF(mysys_var),
288 DENA_THR_OFFSETOF(stmt_arena),
289 DENA_THR_OFFSETOF(limit_found_rows),
290 DENA_THR_OFFSETOF(locked_tables_list)));
291 thd->store_globals();
292 thd->system_thread = static_cast<enum_thread_type>(1<<30UL);
293 memset(&thd->net, 0, sizeof(thd->net));
294 if (for_write_flag) {
295 #if MYSQL_VERSION_ID >= 50505
296 thd->variables.option_bits |= OPTION_BIN_LOG;
297 #else
298 thd->options |= OPTION_BIN_LOG;
299 #endif
300 safeFree((char*) thd->db.str);
301 thd->db.str= my_strdup("handlersocket", MYF(0));
302 thd->db.length= sizeof("handlersocket")-1;
303 }
304 thd->variables.option_bits |= OPTION_TABLE_LOCK;
305 my_pthread_setspecific_ptr(THR_THD, thd);
306 DBG_THR(fprintf(stderr, "HNDSOCK x0 %p\n", thd));
307 }
308 {
309 thd->thread_id = next_thread_id();
310 add_to_active_threads(thd);
311 }
312
313 DBG_THR(fprintf(stderr, "HNDSOCK init thread wsts\n"));
314 wait_server_to_start(thd, shutdown_flag);
315 DBG_THR(fprintf(stderr, "HNDSOCK init thread done\n"));
316
317 thd_proc_info(thd, &info_message_buf[0]);
318 set_thread_message("hs:listening");
319 DBG_THR(fprintf(stderr, "HNDSOCK x1 %p\n", thd));
320
321 lex_start(thd);
322
323 user_lock.reset(new expr_user_lock(thd, user_level_lock_timeout));
324}
325
326int
327dbcontext::set_thread_message(const char *fmt, ...)
328{
329 va_list ap;
330 va_start(ap, fmt);
331 const int n = vsnprintf(&info_message_buf[0], info_message_buf.size(),
332 fmt, ap);
333 va_end(ap);
334 return n;
335}
336
337void
338dbcontext::term_thread()
339{
340 DBG_THR(fprintf(stderr, "HNDSOCK thread end %p\n", thd));
341 close_tables_if();
342 my_pthread_setspecific_ptr(THR_THD, 0);
343 {
344 pthread_mutex_lock(&LOCK_thread_count);
345 delete thd;
346 thd = 0;
347 pthread_mutex_unlock(&LOCK_thread_count);
348 my_thread_end();
349 }
350}
351
352bool
353dbcontext::check_alive()
354{
355 pthread_mutex_lock(&thd->mysys_var->mutex);
356 killed_state st = thd->killed;
357 pthread_mutex_unlock(&thd->mysys_var->mutex);
358 DBG_SHUT(fprintf(stderr, "chk HNDSOCK kst %p %p %d %zu\n", thd, &thd->killed,
359 (int)st, sizeof(*thd)));
360 if (st != NOT_KILLED) {
361 DBG_SHUT(fprintf(stderr, "chk HNDSOCK kst %d break\n", (int)st));
362 return false;
363 }
364 return true;
365}
366
367void
368dbcontext::lock_tables_if()
369{
370 if (lock_failed) {
371 return;
372 }
373 if (for_write_flag && !user_level_lock_locked) {
374 if (user_lock->get_lock()) {
375 user_level_lock_locked = true;
376 } else {
377 lock_failed = true;
378 return;
379 }
380 }
381 if (lock == 0) {
382 const size_t num_max = table_vec.size();
383 TABLE **const tables = DENA_ALLOCA_ALLOCATE(TABLE *, num_max + 1);
384 size_t num_open = 0;
385 for (size_t i = 0; i < num_max; ++i) {
386 if (table_vec[i].refcount > 0) {
387 tables[num_open++] = table_vec[i].table;
388 }
389 table_vec[i].modified = false;
390 }
391 #if MYSQL_VERSION_ID >= 50505
392 lock = thd->lock = mysql_lock_tables(thd, &tables[0], num_open, 0);
393 #else
394 bool need_reopen= false;
395 lock = thd->lock = mysql_lock_tables(thd, &tables[0], num_open,
396 MYSQL_LOCK_NOTIFY_IF_NEED_REOPEN, &need_reopen);
397 #endif
398 statistic_increment(lock_tables_count, &LOCK_status);
399 thd_proc_info(thd, &info_message_buf[0]);
400 DENA_VERBOSE(100, fprintf(stderr, "HNDSOCK lock tables %p %p %zu %zu\n",
401 thd, lock, num_max, num_open));
402 if (lock == 0) {
403 lock_failed = true;
404 DENA_VERBOSE(10, fprintf(stderr, "HNDSOCK failed to lock tables %p\n",
405 thd));
406 }
407 if (for_write_flag) {
408 #if MYSQL_VERSION_ID >= 50505
409 thd->set_current_stmt_binlog_format_row();
410 #else
411 thd->current_stmt_binlog_row_based = 1;
412 #endif
413 }
414 DENA_ALLOCA_FREE(tables);
415 }
416 DBG_LOCK(fprintf(stderr, "HNDSOCK tblnum=%d\n", (int)tblnum));
417}
418
419void
420dbcontext::unlock_tables_if()
421{
422 if (lock != 0) {
423 DENA_VERBOSE(100, fprintf(stderr, "HNDSOCK unlock tables %p %p\n",
424 thd, thd->lock));
425 if (for_write_flag) {
426 for (size_t i = 0; i < table_vec.size(); ++i) {
427 if (table_vec[i].modified) {
428 query_cache_invalidate3(thd, table_vec[i].table, 1);
429 table_vec[i].table->file->ha_release_auto_increment();
430 }
431 }
432 }
433 {
434 bool suc = true;
435 #if MYSQL_VERSION_ID >= 50505
436 suc = (trans_commit_stmt(thd) == 0);
437 #else
438 suc = (ha_autocommit_or_rollback(thd, 0) == 0);
439 #endif
440 if (!suc) {
441 commit_error = true;
442 DENA_VERBOSE(10, fprintf(stderr,
443 "HNDSOCK unlock tables: commit failed\n"));
444 }
445 }
446 mysql_unlock_tables(thd, lock);
447 lock = thd->lock = 0;
448 statistic_increment(unlock_tables_count, &LOCK_status);
449 }
450 if (user_level_lock_locked) {
451 if (user_lock->release_lock()) {
452 user_level_lock_locked = false;
453 }
454 }
455}
456
457bool
458dbcontext::get_commit_error()
459{
460 return commit_error;
461}
462
463void
464dbcontext::clear_error()
465{
466 lock_failed = false;
467 commit_error = false;
468}
469
470void
471dbcontext::close_tables_if()
472{
473 unlock_tables_if();
474 DENA_VERBOSE(100, fprintf(stderr, "HNDSOCK close tables\n"));
475 close_thread_tables(thd);
476 #if MYSQL_VERSION_ID >= 50505
477 thd->mdl_context.release_transactional_locks();
478 #endif
479 if (!table_vec.empty()) {
480 statistic_increment(close_tables_count, &LOCK_status);
481 table_vec.clear();
482 table_map.clear();
483 }
484}
485
486void
487dbcontext::table_addref(size_t tbl_id)
488{
489 table_vec[tbl_id].refcount += 1;
490 DBG_REFCNT(fprintf(stderr, "%p %zu %zu addref\n", this, tbl_id,
491 table_vec[tbl_id].refcount));
492}
493
494void
495dbcontext::table_release(size_t tbl_id)
496{
497 table_vec[tbl_id].refcount -= 1;
498 DBG_REFCNT(fprintf(stderr, "%p %zu %zu release\n", this, tbl_id,
499 table_vec[tbl_id].refcount));
500}
501
502void
503dbcontext::resp_record(dbcallback_i& cb, TABLE *const table,
504 const prep_stmt& pst)
505{
506 char rwpstr_buf[64];
507 String rwpstr(rwpstr_buf, sizeof(rwpstr_buf), &my_charset_bin);
508 const prep_stmt::fields_type& rf = pst.get_ret_fields();
509 const size_t n = rf.size();
510 for (size_t i = 0; i < n; ++i) {
511 uint32_t fn = rf[i];
512 Field *const fld = table->field[fn];
513 DBG_FLD(fprintf(stderr, "fld=%p %zu\n", fld, fn));
514 if (fld->is_null()) {
515 /* null */
516 cb.dbcb_resp_entry(0, 0);
517 } else {
518 fld->val_str(&rwpstr, &rwpstr);
519 const size_t len = rwpstr.length();
520 if (len != 0) {
521 /* non-empty */
522 cb.dbcb_resp_entry(rwpstr.ptr(), rwpstr.length());
523 } else {
524 /* empty */
525 static const char empty_str[] = "";
526 cb.dbcb_resp_entry(empty_str, 0);
527 }
528 }
529 }
530}
531
532void
533dbcontext::dump_record(dbcallback_i& cb, TABLE *const table,
534 const prep_stmt& pst)
535{
536 char rwpstr_buf[64];
537 String rwpstr(rwpstr_buf, sizeof(rwpstr_buf), &my_charset_bin);
538 const prep_stmt::fields_type& rf = pst.get_ret_fields();
539 const size_t n = rf.size();
540 for (size_t i = 0; i < n; ++i) {
541 uint32_t fn = rf[i];
542 Field *const fld = table->field[fn];
543 if (fld->is_null()) {
544 /* null */
545 fprintf(stderr, "NULL");
546 } else {
547 fld->val_str(&rwpstr, &rwpstr);
548 const std::string s(rwpstr.ptr(), rwpstr.length());
549 fprintf(stderr, "[%s]", s.c_str());
550 }
551 }
552 fprintf(stderr, "\n");
553}
554
555int
556dbcontext::modify_record(dbcallback_i& cb, TABLE *const table,
557 const prep_stmt& pst, const cmd_exec_args& args, char mod_op,
558 size_t& modified_count)
559{
560 if (mod_op == 'U') {
561 /* update */
562 handler *const hnd = table->file;
563 uchar *const buf = table->record[0];
564 store_record(table, record[1]);
565 const prep_stmt::fields_type& rf = pst.get_ret_fields();
566 const size_t n = rf.size();
567 for (size_t i = 0; i < n; ++i) {
568 const string_ref& nv = args.uvals[i];
569 uint32_t fn = rf[i];
570 Field *const fld = table->field[fn];
571 if (nv.begin() == 0) {
572 fld->set_null();
573 } else {
574 fld->set_notnull();
575 fld->store(nv.begin(), nv.size(), &my_charset_bin);
576 }
577 }
578 table_vec[pst.get_table_id()].modified = true;
579 const int r = hnd->ha_update_row(table->record[1], buf);
580 if (r != 0 && r != HA_ERR_RECORD_IS_THE_SAME) {
581 return r;
582 }
583 ++modified_count; /* TODO: HA_ERR_RECORD_IS_THE_SAME? */
584 } else if (mod_op == 'D') {
585 /* delete */
586 handler *const hnd = table->file;
587 table_vec[pst.get_table_id()].modified = true;
588 const int r = hnd->ha_delete_row(table->record[0]);
589 if (r != 0) {
590 return r;
591 }
592 ++modified_count;
593 } else if (mod_op == '+' || mod_op == '-') {
594 /* increment/decrement */
595 handler *const hnd = table->file;
596 uchar *const buf = table->record[0];
597 store_record(table, record[1]);
598 const prep_stmt::fields_type& rf = pst.get_ret_fields();
599 const size_t n = rf.size();
600 size_t i = 0;
601 for (i = 0; i < n; ++i) {
602 const string_ref& nv = args.uvals[i];
603 uint32_t fn = rf[i];
604 Field *const fld = table->field[fn];
605 if (fld->is_null() || nv.begin() == 0) {
606 continue;
607 }
608 const long long pval = fld->val_int();
609 const long long llv = atoll_nocheck(nv.begin(), nv.end());
610 /* TODO: llv == 0? */
611 long long nval = 0;
612 if (mod_op == '+') {
613 /* increment */
614 nval = pval + llv;
615 } else {
616 /* decrement */
617 nval = pval - llv;
618 if ((pval < 0 && nval > 0) || (pval > 0 && nval < 0)) {
619 break; /* don't modify */
620 }
621 }
622 fld->store(nval, false);
623 }
624 if (i == n) {
625 /* modify */
626 table_vec[pst.get_table_id()].modified = true;
627 const int r = hnd->ha_update_row(table->record[1], buf);
628 if (r != 0 && r != HA_ERR_RECORD_IS_THE_SAME) {
629 return r;
630 }
631 ++modified_count;
632 }
633 }
634 return 0;
635}
636
637void
638dbcontext::cmd_insert_internal(dbcallback_i& cb, const prep_stmt& pst,
639 const string_ref *fvals, size_t fvalslen)
640{
641 if (!for_write_flag) {
642 return cb.dbcb_resp_short(2, "readonly");
643 }
644 lock_tables_if();
645 if (lock == 0) {
646 return cb.dbcb_resp_short(1, "lock_tables");
647 }
648 if (pst.get_table_id() >= table_vec.size()) {
649 return cb.dbcb_resp_short(2, "tblnum");
650 }
651 TABLE *const table = table_vec[pst.get_table_id()].table;
652 handler *const hnd = table->file;
653 uchar *const buf = table->record[0];
654 empty_record(table);
655 memset(buf, 0, table->s->null_bytes); /* clear null flags */
656 const prep_stmt::fields_type& rf = pst.get_ret_fields();
657 const size_t n = std::min(rf.size(), fvalslen);
658 for (size_t i = 0; i < n; ++i) {
659 uint32_t fn = rf[i];
660 Field *const fld = table->field[fn];
661 if (fvals[i].begin() == 0) {
662 fld->set_null();
663 } else {
664 fld->store(fvals[i].begin(), fvals[i].size(), &my_charset_bin);
665 }
666 }
667 table->next_number_field = table->found_next_number_field;
668 /* FIXME: test */
669 const int r = hnd->ha_write_row(buf);
670 const ulonglong insert_id = table->file->insert_id_for_cur_row;
671 table->next_number_field = 0;
672 table_vec[pst.get_table_id()].modified = true;
673 if (r == 0 && table->found_next_number_field != 0) {
674 return cb.dbcb_resp_short_num64(0, insert_id);
675 }
676 if (r != 0) {
677 return cb.dbcb_resp_short_num(1, r);
678 }
679 return cb.dbcb_resp_short(0, "");
680}
681
682void
683dbcontext::cmd_sql_internal(dbcallback_i& cb, const prep_stmt& pst,
684 const string_ref *fvals, size_t fvalslen)
685{
686 if (fvalslen < 1) {
687 return cb.dbcb_resp_short(2, "syntax");
688 }
689 return cb.dbcb_resp_short(2, "notimpl");
690}
691
692static size_t
693prepare_keybuf(const cmd_exec_args& args, uchar *key_buf, TABLE *table,
694 KEY& kinfo, size_t invalues_index)
695{
696 size_t kplen_sum = 0;
697 DBG_KEY(fprintf(stderr, "SLOW\n"));
698 for (size_t i = 0; i < args.kvalslen; ++i) {
699 const KEY_PART_INFO & kpt = kinfo.key_part[i];
700 string_ref kval = args.kvals[i];
701 if (args.invalues_keypart >= 0 &&
702 static_cast<size_t>(args.invalues_keypart) == i) {
703 kval = args.invalues[invalues_index];
704 }
705 if (kval.begin() == 0) {
706 kpt.field->set_null();
707 } else {
708 kpt.field->set_notnull();
709 }
710 kpt.field->store(kval.begin(), kval.size(), &my_charset_bin);
711 kplen_sum += kpt.store_length;
712 DBG_KEYLEN(fprintf(stderr, "l=%u sl=%zu\n", kpt.length,
713 kpt.store_length));
714 }
715 key_copy(key_buf, table->record[0], &kinfo, kplen_sum);
716 DBG_KEYLEN(fprintf(stderr, "sum=%zu flen=%u\n", kplen_sum,
717 kinfo.key_length));
718 return kplen_sum;
719}
720
721void
722dbcontext::cmd_find_internal(dbcallback_i& cb, const prep_stmt& pst,
723 ha_rkey_function find_flag, const cmd_exec_args& args)
724{
725 const bool debug_out = (verbose_level >= 100);
726 bool need_resp_record = true;
727 char mod_op = 0;
728 const string_ref& mod_op_str = args.mod_op;
729 if (mod_op_str.size() != 0) {
730 if (!for_write_flag) {
731 return cb.dbcb_resp_short(2, "readonly");
732 }
733 mod_op = mod_op_str.begin()[0];
734 need_resp_record = mod_op_str.size() > 1 && mod_op_str.begin()[1] == '?';
735 switch (mod_op) {
736 case 'U': /* update */
737 case 'D': /* delete */
738 case '+': /* increment */
739 case '-': /* decrement */
740 break;
741 default:
742 if (debug_out) {
743 fprintf(stderr, "unknown modop: %c\n", mod_op);
744 }
745 return cb.dbcb_resp_short(2, "modop");
746 }
747 }
748 lock_tables_if();
749 if (lock == 0) {
750 return cb.dbcb_resp_short(1, "lock_tables");
751 }
752 if (pst.get_table_id() >= table_vec.size()) {
753 return cb.dbcb_resp_short(2, "tblnum");
754 }
755 TABLE *const table = table_vec[pst.get_table_id()].table;
756 /* keys */
757 if (pst.get_idxnum() >= table->s->keys) {
758 return cb.dbcb_resp_short(2, "idxnum");
759 }
760 KEY& kinfo = table->key_info[pst.get_idxnum()];
761 if (args.kvalslen > kinfo.user_defined_key_parts) {
762 return cb.dbcb_resp_short(2, "kpnum");
763 }
764 uchar *const key_buf = DENA_ALLOCA_ALLOCATE(uchar, kinfo.key_length);
765 size_t invalues_idx = 0;
766 size_t kplen_sum = prepare_keybuf(args, key_buf, table, kinfo, invalues_idx);
767 /* filters */
768 uchar *filter_buf = 0;
769 if (args.filters != 0) {
770 const size_t filter_buf_len = calc_filter_buf_size(table, pst,
771 args.filters);
772 filter_buf = DENA_ALLOCA_ALLOCATE(uchar, filter_buf_len);
773 if (!fill_filter_buf(table, pst, args.filters, filter_buf,
774 filter_buf_len)) {
775 return cb.dbcb_resp_short(2, "filterblob");
776 }
777 }
778 /* handler */
779 table->read_set = &table->s->all_set;
780 handler *const hnd = table->file;
781 if (!for_write_flag) {
782 hnd->init_table_handle_for_HANDLER();
783 }
784 hnd->ha_index_or_rnd_end();
785 hnd->ha_index_init(pst.get_idxnum(), 1);
786 if (need_resp_record) {
787 cb.dbcb_resp_begin(pst.get_ret_fields().size());
788 }
789 const uint32_t limit = args.limit ? args.limit : 1;
790 uint32_t skip = args.skip;
791 size_t modified_count = 0;
792 int r = 0;
793 bool is_first = true;
794 for (uint32_t cnt = 0; cnt < limit + skip;) {
795 if (is_first) {
796 is_first = false;
797 const key_part_map kpm = (1U << args.kvalslen) - 1;
798 r = hnd->ha_index_read_map(table->record[0], key_buf, kpm, find_flag);
799 } else if (args.invalues_keypart >= 0) {
800 if (++invalues_idx >= args.invalueslen) {
801 break;
802 }
803 kplen_sum = prepare_keybuf(args, key_buf, table, kinfo, invalues_idx);
804 const key_part_map kpm = (1U << args.kvalslen) - 1;
805 r = hnd->ha_index_read_map(table->record[0], key_buf, kpm, find_flag);
806 } else {
807 switch (find_flag) {
808 case HA_READ_BEFORE_KEY:
809 case HA_READ_KEY_OR_PREV:
810 r = hnd->ha_index_prev(table->record[0]);
811 break;
812 case HA_READ_AFTER_KEY:
813 case HA_READ_KEY_OR_NEXT:
814 r = hnd->ha_index_next(table->record[0]);
815 break;
816 case HA_READ_KEY_EXACT:
817 r = hnd->ha_index_next_same(table->record[0], key_buf, kplen_sum);
818 break;
819 default:
820 r = HA_ERR_END_OF_FILE; /* to finish the loop */
821 break;
822 }
823 }
824 if (debug_out) {
825 fprintf(stderr, "r=%d\n", r);
826 if (r == 0 || r == HA_ERR_RECORD_DELETED) {
827 dump_record(cb, table, pst);
828 }
829 }
830 int filter_res = 0;
831 if (r != 0) {
832 /* no-count */
833 } else if (args.filters != 0 && (filter_res = check_filter(cb, table,
834 pst, args.filters, filter_buf)) != 0) {
835 if (filter_res < 0) {
836 break;
837 }
838 } else if (skip > 0) {
839 --skip;
840 } else {
841 /* hit */
842 if (need_resp_record) {
843 resp_record(cb, table, pst);
844 }
845 if (mod_op != 0) {
846 r = modify_record(cb, table, pst, args, mod_op, modified_count);
847 }
848 ++cnt;
849 }
850 if (args.invalues_keypart >= 0 && r == HA_ERR_KEY_NOT_FOUND) {
851 continue;
852 }
853 if (r != 0 && r != HA_ERR_RECORD_DELETED) {
854 break;
855 }
856 }
857 hnd->ha_index_or_rnd_end();
858 if (r != 0 && r != HA_ERR_RECORD_DELETED && r != HA_ERR_KEY_NOT_FOUND &&
859 r != HA_ERR_END_OF_FILE) {
860 /* failed */
861 if (need_resp_record) {
862 /* revert dbcb_resp_begin() and dbcb_resp_entry() */
863 cb.dbcb_resp_cancel();
864 }
865 cb.dbcb_resp_short_num(1, r);
866 } else {
867 /* succeeded */
868 if (need_resp_record) {
869 cb.dbcb_resp_end();
870 } else {
871 cb.dbcb_resp_short_num(0, modified_count);
872 }
873 }
874 DENA_ALLOCA_FREE(filter_buf);
875 DENA_ALLOCA_FREE(key_buf);
876}
877
878size_t
879dbcontext::calc_filter_buf_size(TABLE *table, const prep_stmt& pst,
880 const record_filter *filters)
881{
882 size_t filter_buf_len = 0;
883 for (const record_filter *f = filters; f->op.begin() != 0; ++f) {
884 if (f->val.begin() == 0) {
885 continue;
886 }
887 const uint32_t fn = pst.get_filter_fields()[f->ff_offset];
888 filter_buf_len += table->field[fn]->pack_length();
889 }
890 ++filter_buf_len;
891 /* Field_medium::cmp() calls uint3korr(), which may read 4 bytes.
892 Allocate 1 more byte for safety. */
893 return filter_buf_len;
894}
895
896bool
897dbcontext::fill_filter_buf(TABLE *table, const prep_stmt& pst,
898 const record_filter *filters, uchar *filter_buf, size_t len)
899{
900 memset(filter_buf, 0, len);
901 size_t pos = 0;
902 for (const record_filter *f = filters; f->op.begin() != 0; ++f) {
903 if (f->val.begin() == 0) {
904 continue;
905 }
906 const uint32_t fn = pst.get_filter_fields()[f->ff_offset];
907 Field *const fld = table->field[fn];
908 if ((fld->flags & BLOB_FLAG) != 0) {
909 return false;
910 }
911 fld->store(f->val.begin(), f->val.size(), &my_charset_bin);
912 const size_t packlen = fld->pack_length();
913 memcpy(filter_buf + pos, fld->ptr, packlen);
914 pos += packlen;
915 }
916 return true;
917}
918
919int
920dbcontext::check_filter(dbcallback_i& cb, TABLE *table, const prep_stmt& pst,
921 const record_filter *filters, const uchar *filter_buf)
922{
923 DBG_FILTER(fprintf(stderr, "check_filter\n"));
924 size_t pos = 0;
925 for (const record_filter *f = filters; f->op.begin() != 0; ++f) {
926 const string_ref& op = f->op;
927 const string_ref& val = f->val;
928 const uint32_t fn = pst.get_filter_fields()[f->ff_offset];
929 Field *const fld = table->field[fn];
930 const size_t packlen = fld->pack_length();
931 const uchar *const bval = filter_buf + pos;
932 int cv = 0;
933 if (fld->is_null()) {
934 cv = (val.begin() == 0) ? 0 : -1;
935 } else {
936 cv = (val.begin() == 0) ? 1 : fld->cmp(bval);
937 }
938 DBG_FILTER(fprintf(stderr, "check_filter cv=%d\n", cv));
939 bool cond = true;
940 if (op.size() == 1) {
941 switch (op.begin()[0]) {
942 case '>':
943 DBG_FILTER(fprintf(stderr, "check_filter op: >\n"));
944 cond = (cv > 0);
945 break;
946 case '<':
947 DBG_FILTER(fprintf(stderr, "check_filter op: <\n"));
948 cond = (cv < 0);
949 break;
950 case '=':
951 DBG_FILTER(fprintf(stderr, "check_filter op: =\n"));
952 cond = (cv == 0);
953 break;
954 default:
955 DBG_FILTER(fprintf(stderr, "check_filter op: unknown\n"));
956 cond = false; /* FIXME: error */
957 break;
958 }
959 } else if (op.size() == 2 && op.begin()[1] == '=') {
960 switch (op.begin()[0]) {
961 case '>':
962 DBG_FILTER(fprintf(stderr, "check_filter op: >=\n"));
963 cond = (cv >= 0);
964 break;
965 case '<':
966 DBG_FILTER(fprintf(stderr, "check_filter op: <=\n"));
967 cond = (cv <= 0);
968 break;
969 case '!':
970 DBG_FILTER(fprintf(stderr, "check_filter op: !=\n"));
971 cond = (cv != 0);
972 break;
973 default:
974 DBG_FILTER(fprintf(stderr, "check_filter op: unknown\n"));
975 cond = false; /* FIXME: error */
976 break;
977 }
978 }
979 DBG_FILTER(fprintf(stderr, "check_filter cond: %d\n", (int)cond));
980 if (!cond) {
981 return (f->filter_type == record_filter_type_skip) ? 1 : -1;
982 }
983 if (val.begin() != 0) {
984 pos += packlen;
985 }
986 }
987 return 0;
988}
989
990void
991dbcontext::cmd_open(dbcallback_i& cb, const cmd_open_args& arg)
992{
993 unlock_tables_if();
994 const table_name_type k = std::make_pair(std::string(arg.dbn),
995 std::string(arg.tbl));
996 const table_map_type::const_iterator iter = table_map.find(k);
997 uint32_t tblnum = 0;
998 if (iter != table_map.end()) {
999 tblnum = iter->second;
1000 DBG_CMP(fprintf(stderr, "HNDSOCK k=%s tblnum=%d\n", k.c_str(),
1001 (int)tblnum));
1002 } else {
1003 TABLE_LIST tables;
1004 TABLE *table = 0;
1005 bool refresh = true;
1006 const thr_lock_type lock_type = for_write_flag ? TL_WRITE : TL_READ;
1007 #if MYSQL_VERSION_ID >= 50505
1008 LEX_CSTRING db_name= { arg.dbn, strlen(arg.dbn) };
1009 LEX_CSTRING tbl_name= { arg.tbl, strlen(arg.tbl) };
1010 tables.init_one_table(&db_name, &tbl_name, 0, lock_type);
1011 tables.mdl_request.init(MDL_key::TABLE, arg.dbn, arg.tbl,
1012 for_write_flag ? MDL_SHARED_WRITE : MDL_SHARED_READ, MDL_TRANSACTION);
1013 Open_table_context ot_act(thd, 0);
1014 if (!open_table(thd, &tables, &ot_act)) {
1015 table = tables.table;
1016 }
1017 #else
1018 tables.init_one_table(arg.dbn, arg.tbl, lock_type);
1019 table = open_table(thd, &tables, thd->mem_root, &refresh,
1020 OPEN_VIEW_NO_PARSE);
1021 #endif
1022 if (table == 0) {
1023 DENA_VERBOSE(20, fprintf(stderr,
1024 "HNDSOCK failed to open %p [%s] [%s] [%d]\n",
1025 thd, arg.dbn, arg.tbl, static_cast<int>(refresh)));
1026 return cb.dbcb_resp_short(1, "open_table");
1027 }
1028 statistic_increment(open_tables_count, &LOCK_status);
1029 table->reginfo.lock_type = lock_type;
1030 table->use_all_columns();
1031 tblnum = table_vec.size();
1032 tablevec_entry e;
1033 e.table = table;
1034 table_vec.push_back(e);
1035 table_map[k] = tblnum;
1036 }
1037 size_t idxnum = static_cast<size_t>(-1);
1038 if (arg.idx[0] >= '0' && arg.idx[0] <= '9') {
1039 /* numeric */
1040 TABLE *const table = table_vec[tblnum].table;
1041 idxnum = atoi(arg.idx);
1042 if (idxnum >= table->s->keys) {
1043 return cb.dbcb_resp_short(2, "idxnum");
1044 }
1045 } else {
1046 const char *const idx_name_to_open =
1047 arg.idx[0] == '\0' ? "PRIMARY" : arg.idx;
1048 TABLE *const table = table_vec[tblnum].table;
1049 for (uint i = 0; i < table->s->keys; ++i) {
1050 KEY& kinfo = table->key_info[i];
1051 if (strcmp(kinfo.name.str, idx_name_to_open) == 0) {
1052 idxnum = i;
1053 break;
1054 }
1055 }
1056 }
1057 if (idxnum == size_t(-1)) {
1058 return cb.dbcb_resp_short(2, "idxnum");
1059 }
1060 prep_stmt::fields_type rf;
1061 prep_stmt::fields_type ff;
1062 if (!parse_fields(table_vec[tblnum].table, arg.retflds, rf)) {
1063 return cb.dbcb_resp_short(2, "fld");
1064 }
1065 if (!parse_fields(table_vec[tblnum].table, arg.filflds, ff)) {
1066 return cb.dbcb_resp_short(2, "fld");
1067 }
1068 prep_stmt p(this, tblnum, idxnum, rf, ff);
1069 cb.dbcb_set_prep_stmt(arg.pst_id, p);
1070 return cb.dbcb_resp_short(0, "");
1071}
1072
1073bool
1074dbcontext::parse_fields(TABLE *const table, const char *str,
1075 prep_stmt::fields_type& flds)
1076{
1077 string_ref flds_sr(str, strlen(str));
1078 std::vector<string_ref> fldnms;
1079 if (flds_sr.size() != 0) {
1080 split(',', flds_sr, fldnms);
1081 }
1082 for (size_t i = 0; i < fldnms.size(); ++i) {
1083 Field **fld = 0;
1084 size_t j = 0;
1085 for (fld = table->field; *fld; ++fld, ++j) {
1086 DBG_FLD(fprintf(stderr, "f %s\n", (*fld)->field_name.str));
1087 string_ref fn((*fld)->field_name.str, (*fld)->field_name.length);
1088 if (fn == fldnms[i]) {
1089 break;
1090 }
1091 }
1092 if (*fld == 0) {
1093 DBG_FLD(fprintf(stderr, "UNKNOWN FLD %s [%s]\n", retflds,
1094 std::string(fldnms[i].begin(), fldnms[i].size()).c_str()));
1095 return false;
1096 }
1097 DBG_FLD(fprintf(stderr, "FLD %s %zu\n", (*fld)->field_name.str, j));
1098 flds.push_back(j);
1099 }
1100 return true;
1101}
1102
1103enum db_write_op {
1104 db_write_op_none = 0,
1105 db_write_op_insert = 1,
1106 db_write_op_sql = 2,
1107};
1108
1109void
1110dbcontext::cmd_exec(dbcallback_i& cb, const cmd_exec_args& args)
1111{
1112 const prep_stmt& p = *args.pst;
1113 if (p.get_table_id() == static_cast<size_t>(-1)) {
1114 return cb.dbcb_resp_short(2, "stmtnum");
1115 }
1116 ha_rkey_function find_flag = HA_READ_KEY_EXACT;
1117 db_write_op wrop = db_write_op_none;
1118 if (args.op.size() == 1) {
1119 switch (args.op.begin()[0]) {
1120 case '=':
1121 find_flag = HA_READ_KEY_EXACT;
1122 break;
1123 case '>':
1124 find_flag = HA_READ_AFTER_KEY;
1125 break;
1126 case '<':
1127 find_flag = HA_READ_BEFORE_KEY;
1128 break;
1129 case '+':
1130 wrop = db_write_op_insert;
1131 break;
1132 case 'S':
1133 wrop = db_write_op_sql;
1134 break;
1135 default:
1136 return cb.dbcb_resp_short(2, "op");
1137 }
1138 } else if (args.op.size() == 2 && args.op.begin()[1] == '=') {
1139 switch (args.op.begin()[0]) {
1140 case '>':
1141 find_flag = HA_READ_KEY_OR_NEXT;
1142 break;
1143 case '<':
1144 find_flag = HA_READ_KEY_OR_PREV;
1145 break;
1146 default:
1147 return cb.dbcb_resp_short(2, "op");
1148 }
1149 } else {
1150 return cb.dbcb_resp_short(2, "op");
1151 }
1152 if (args.kvalslen <= 0) {
1153 return cb.dbcb_resp_short(2, "klen");
1154 }
1155 switch (wrop) {
1156 case db_write_op_none:
1157 return cmd_find_internal(cb, p, find_flag, args);
1158 case db_write_op_insert:
1159 return cmd_insert_internal(cb, p, args.kvals, args.kvalslen);
1160 case db_write_op_sql:
1161 return cmd_sql_internal(cb, p, args.kvals, args.kvalslen);
1162 }
1163}
1164
1165void
1166dbcontext::set_statistics(size_t num_conns, size_t num_active)
1167{
1168 if (for_write_flag) {
1169 set_thread_message("handlersocket: mode=wr, %zu conns, %zu active",
1170 num_conns, num_active);
1171 } else {
1172 set_thread_message("handlersocket: mode=rd, %zu conns, %zu active",
1173 num_conns, num_active);
1174 }
1175 /*
1176 Don't set message buf if it's already in use. This saves slow call to
1177 thd_proc_info() (if profiling is enabled)
1178 */
1179 if (thd->proc_info != &info_message_buf[0])
1180 thd_proc_info(thd, &info_message_buf[0]);
1181}
1182
1183};
1184
1185