Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pg_stat_statements.c
4 : * Track statement planning and execution times as well as resource
5 : * usage across a whole database cluster.
6 : *
7 : * Execution costs are totaled for each distinct source query, and kept in
8 : * a shared hashtable. (We track only as many distinct queries as will fit
9 : * in the designated amount of shared memory.)
10 : *
11 : * Starting in Postgres 9.2, this module normalized query entries. As of
12 : * Postgres 14, the normalization is done by the core if compute_query_id is
13 : * enabled, or optionally by third-party modules.
14 : *
15 : * To facilitate presenting entries to users, we create "representative" query
16 : * strings in which constants are replaced with parameter symbols ($n), to
17 : * make it clearer what a normalized entry can represent. To save on shared
18 : * memory, and to avoid having to truncate oversized query strings, we store
19 : * these strings in a temporary external query-texts file. Offsets into this
20 : * file are kept in shared memory.
21 : *
22 : * Note about locking issues: to create or delete an entry in the shared
23 : * hashtable, one must hold pgss->lock exclusively. Modifying any field
24 : * in an entry except the counters requires the same. To look up an entry,
25 : * one must hold the lock shared. To read or update the counters within
26 : * an entry, one must hold the lock shared or exclusive (so the entry doesn't
27 : * disappear!) and also take the entry's mutex spinlock.
28 : * The shared state variable pgss->extent (the next free spot in the external
29 : * query-text file) should be accessed only while holding either the
30 : * pgss->mutex spinlock, or exclusive lock on pgss->lock. We use the mutex to
31 : * allow reserving file space while holding only shared lock on pgss->lock.
32 : * Rewriting the entire external query-text file, eg for garbage collection,
33 : * requires holding pgss->lock exclusively; this allows individual entries
34 : * in the file to be read or written while holding only shared lock.
35 : *
36 : *
37 : * Copyright (c) 2008-2026, PostgreSQL Global Development Group
38 : *
39 : * IDENTIFICATION
40 : * contrib/pg_stat_statements/pg_stat_statements.c
41 : *
42 : *-------------------------------------------------------------------------
43 : */
44 : #include "postgres.h"
45 :
46 : #include <math.h>
47 : #include <sys/stat.h>
48 : #include <unistd.h>
49 :
50 : #include "access/htup_details.h"
51 : #include "access/parallel.h"
52 : #include "catalog/pg_authid.h"
53 : #include "common/int.h"
54 : #include "executor/instrument.h"
55 : #include "funcapi.h"
56 : #include "jit/jit.h"
57 : #include "mb/pg_wchar.h"
58 : #include "miscadmin.h"
59 : #include "nodes/queryjumble.h"
60 : #include "optimizer/planner.h"
61 : #include "parser/analyze.h"
62 : #include "parser/scanner.h"
63 : #include "pgstat.h"
64 : #include "storage/fd.h"
65 : #include "storage/ipc.h"
66 : #include "storage/lwlock.h"
67 : #include "storage/shmem.h"
68 : #include "storage/spin.h"
69 : #include "tcop/utility.h"
70 : #include "utils/acl.h"
71 : #include "utils/builtins.h"
72 : #include "utils/memutils.h"
73 : #include "utils/timestamp.h"
74 :
75 0 : PG_MODULE_MAGIC_EXT(
76 : .name = "pg_stat_statements",
77 : .version = PG_VERSION
78 : );
79 :
80 : /* Location of permanent stats file (valid when database is shut down) */
81 : #define PGSS_DUMP_FILE PGSTAT_STAT_PERMANENT_DIRECTORY "/pg_stat_statements.stat"
82 :
83 : /*
84 : * Location of external query text file.
85 : */
86 : #define PGSS_TEXT_FILE PG_STAT_TMP_DIR "/pgss_query_texts.stat"
87 :
88 : /* Magic number identifying the stats file format */
89 : static const uint32 PGSS_FILE_HEADER = 0x20250731;
90 :
91 : /* PostgreSQL major version number, changes in which invalidate all entries */
92 : static const uint32 PGSS_PG_MAJOR_VERSION = PG_VERSION_NUM / 100;
93 :
94 : /* XXX: Should USAGE_EXEC reflect execution time and/or buffer usage? */
95 : #define USAGE_EXEC(duration) (1.0)
96 : #define USAGE_INIT (1.0) /* including initial planning */
97 : #define ASSUMED_MEDIAN_INIT (10.0) /* initial assumed median usage */
98 : #define ASSUMED_LENGTH_INIT 1024 /* initial assumed mean query length */
99 : #define USAGE_DECREASE_FACTOR (0.99) /* decreased every entry_dealloc */
100 : #define STICKY_DECREASE_FACTOR (0.50) /* factor for sticky entries */
101 : #define USAGE_DEALLOC_PERCENT 5 /* free this % of entries at once */
102 : #define IS_STICKY(c) ((c.calls[PGSS_PLAN] + c.calls[PGSS_EXEC]) == 0)
103 :
104 : /*
105 : * Extension version number, for supporting older extension versions' objects
106 : */
107 : typedef enum pgssVersion
108 : {
109 : PGSS_V1_0 = 0,
110 : PGSS_V1_1,
111 : PGSS_V1_2,
112 : PGSS_V1_3,
113 : PGSS_V1_8,
114 : PGSS_V1_9,
115 : PGSS_V1_10,
116 : PGSS_V1_11,
117 : PGSS_V1_12,
118 : PGSS_V1_13,
119 : } pgssVersion;
120 :
121 : typedef enum pgssStoreKind
122 : {
123 : PGSS_INVALID = -1,
124 :
125 : /*
126 : * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
127 : * reference the underlying values in the arrays in the Counters struct,
128 : * and this order is required in pg_stat_statements_internal().
129 : */
130 : PGSS_PLAN = 0,
131 : PGSS_EXEC,
132 : } pgssStoreKind;
133 :
134 : #define PGSS_NUMKIND (PGSS_EXEC + 1)
135 :
136 : /*
137 : * Hashtable key that defines the identity of a hashtable entry. We separate
138 : * queries by user and by database even if they are otherwise identical.
139 : *
140 : * If you add a new key to this struct, make sure to teach pgss_store() to
141 : * zero the padding bytes. Otherwise, things will break, because pgss_hash is
142 : * created using HASH_BLOBS, and thus tag_hash is used to hash this.
143 : */
144 : typedef struct pgssHashKey
145 : {
146 : Oid userid; /* user OID */
147 : Oid dbid; /* database OID */
148 : int64 queryid; /* query identifier */
149 : bool toplevel; /* query executed at top level */
150 : } pgssHashKey;
151 :
152 : /*
153 : * The actual stats counters kept within pgssEntry.
154 : */
155 : typedef struct Counters
156 : {
157 : int64 calls[PGSS_NUMKIND]; /* # of times planned/executed */
158 : double total_time[PGSS_NUMKIND]; /* total planning/execution time,
159 : * in msec */
160 : double min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
161 : * msec since min/max reset */
162 : double max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
163 : * msec since min/max reset */
164 : double mean_time[PGSS_NUMKIND]; /* mean planning/execution time in
165 : * msec */
166 : double sum_var_time[PGSS_NUMKIND]; /* sum of variances in
167 : * planning/execution time in msec */
168 : int64 rows; /* total # of retrieved or affected rows */
169 : int64 shared_blks_hit; /* # of shared buffer hits */
170 : int64 shared_blks_read; /* # of shared disk blocks read */
171 : int64 shared_blks_dirtied; /* # of shared disk blocks dirtied */
172 : int64 shared_blks_written; /* # of shared disk blocks written */
173 : int64 local_blks_hit; /* # of local buffer hits */
174 : int64 local_blks_read; /* # of local disk blocks read */
175 : int64 local_blks_dirtied; /* # of local disk blocks dirtied */
176 : int64 local_blks_written; /* # of local disk blocks written */
177 : int64 temp_blks_read; /* # of temp blocks read */
178 : int64 temp_blks_written; /* # of temp blocks written */
179 : double shared_blk_read_time; /* time spent reading shared blocks,
180 : * in msec */
181 : double shared_blk_write_time; /* time spent writing shared blocks,
182 : * in msec */
183 : double local_blk_read_time; /* time spent reading local blocks, in
184 : * msec */
185 : double local_blk_write_time; /* time spent writing local blocks, in
186 : * msec */
187 : double temp_blk_read_time; /* time spent reading temp blocks, in msec */
188 : double temp_blk_write_time; /* time spent writing temp blocks, in
189 : * msec */
190 : double usage; /* usage factor */
191 : int64 wal_records; /* # of WAL records generated */
192 : int64 wal_fpi; /* # of WAL full page images generated */
193 : uint64 wal_bytes; /* total amount of WAL generated in bytes */
194 : int64 wal_buffers_full; /* # of times the WAL buffers became full */
195 : int64 jit_functions; /* total number of JIT functions emitted */
196 : double jit_generation_time; /* total time to generate jit code */
197 : int64 jit_inlining_count; /* number of times inlining time has been
198 : * > 0 */
199 : double jit_deform_time; /* total time to deform tuples in jit code */
200 : int64 jit_deform_count; /* number of times deform time has been >
201 : * 0 */
202 :
203 : double jit_inlining_time; /* total time to inline jit code */
204 : int64 jit_optimization_count; /* number of times optimization time
205 : * has been > 0 */
206 : double jit_optimization_time; /* total time to optimize jit code */
207 : int64 jit_emission_count; /* number of times emission time has been
208 : * > 0 */
209 : double jit_emission_time; /* total time to emit jit code */
210 : int64 parallel_workers_to_launch; /* # of parallel workers planned
211 : * to be launched */
212 : int64 parallel_workers_launched; /* # of parallel workers actually
213 : * launched */
214 : int64 generic_plan_calls; /* number of calls using a generic plan */
215 : int64 custom_plan_calls; /* number of calls using a custom plan */
216 : } Counters;
217 :
218 : /*
219 : * Global statistics for pg_stat_statements
220 : */
221 : typedef struct pgssGlobalStats
222 : {
223 : int64 dealloc; /* # of times entries were deallocated */
224 : TimestampTz stats_reset; /* timestamp with all stats reset */
225 : } pgssGlobalStats;
226 :
227 : /*
228 : * Statistics per statement
229 : *
230 : * Note: in event of a failure in garbage collection of the query text file,
231 : * we reset query_offset to zero and query_len to -1. This will be seen as
232 : * an invalid state by qtext_fetch().
233 : */
234 : typedef struct pgssEntry
235 : {
236 : pgssHashKey key; /* hash key of entry - MUST BE FIRST */
237 : Counters counters; /* the statistics for this query */
238 : Size query_offset; /* query text offset in external file */
239 : int query_len; /* # of valid bytes in query string, or -1 */
240 : int encoding; /* query text encoding */
241 : TimestampTz stats_since; /* timestamp of entry allocation */
242 : TimestampTz minmax_stats_since; /* timestamp of last min/max values reset */
243 : slock_t mutex; /* protects the counters only */
244 : } pgssEntry;
245 :
246 : /*
247 : * Global shared state
248 : */
249 : typedef struct pgssSharedState
250 : {
251 : LWLock *lock; /* protects hashtable search/modification */
252 : double cur_median_usage; /* current median usage in hashtable */
253 : Size mean_query_len; /* current mean entry text length */
254 : slock_t mutex; /* protects following fields only: */
255 : Size extent; /* current extent of query file */
256 : int n_writers; /* number of active writers to query file */
257 : int gc_count; /* query file garbage collection cycle count */
258 : pgssGlobalStats stats; /* global statistics for pgss */
259 : } pgssSharedState;
260 :
261 : /*---- Local variables ----*/
262 :
263 : /* Current nesting depth of planner/ExecutorRun/ProcessUtility calls */
264 : static int nesting_level = 0;
265 :
266 : /* Saved hook values */
267 : static shmem_request_hook_type prev_shmem_request_hook = NULL;
268 : static shmem_startup_hook_type prev_shmem_startup_hook = NULL;
269 : static post_parse_analyze_hook_type prev_post_parse_analyze_hook = NULL;
270 : static planner_hook_type prev_planner_hook = NULL;
271 : static ExecutorStart_hook_type prev_ExecutorStart = NULL;
272 : static ExecutorRun_hook_type prev_ExecutorRun = NULL;
273 : static ExecutorFinish_hook_type prev_ExecutorFinish = NULL;
274 : static ExecutorEnd_hook_type prev_ExecutorEnd = NULL;
275 : static ProcessUtility_hook_type prev_ProcessUtility = NULL;
276 :
277 : /* Links to shared memory state */
278 : static pgssSharedState *pgss = NULL;
279 : static HTAB *pgss_hash = NULL;
280 :
281 : /*---- GUC variables ----*/
282 :
283 : typedef enum
284 : {
285 : PGSS_TRACK_NONE, /* track no statements */
286 : PGSS_TRACK_TOP, /* only top level statements */
287 : PGSS_TRACK_ALL, /* all statements, including nested ones */
288 : } PGSSTrackLevel;
289 :
290 : static const struct config_enum_entry track_options[] =
291 : {
292 : {"none", PGSS_TRACK_NONE, false},
293 : {"top", PGSS_TRACK_TOP, false},
294 : {"all", PGSS_TRACK_ALL, false},
295 : {NULL, 0, false}
296 : };
297 :
298 : static int pgss_max = 5000; /* max # statements to track */
299 : static int pgss_track = PGSS_TRACK_TOP; /* tracking level */
300 : static bool pgss_track_utility = true; /* whether to track utility commands */
301 : static bool pgss_track_planning = false; /* whether to track planning
302 : * duration */
303 : static bool pgss_save = true; /* whether to save stats across shutdown */
304 :
305 : #define pgss_enabled(level) \
306 : (!IsParallelWorker() && \
307 : (pgss_track == PGSS_TRACK_ALL || \
308 : (pgss_track == PGSS_TRACK_TOP && (level) == 0)))
309 :
310 : #define record_gc_qtexts() \
311 : do { \
312 : SpinLockAcquire(&pgss->mutex); \
313 : pgss->gc_count++; \
314 : SpinLockRelease(&pgss->mutex); \
315 : } while(0)
316 :
317 : /*---- Function declarations ----*/
318 :
319 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset);
320 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_7);
321 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_11);
322 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_2);
323 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_3);
324 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_8);
325 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_9);
326 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_10);
327 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_11);
328 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_12);
329 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_13);
330 0 : PG_FUNCTION_INFO_V1(pg_stat_statements);
331 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_info);
332 :
333 : static void pgss_shmem_request(void);
334 : static void pgss_shmem_startup(void);
335 : static void pgss_shmem_shutdown(int code, Datum arg);
336 : static void pgss_post_parse_analyze(ParseState *pstate, Query *query,
337 : JumbleState *jstate);
338 : static PlannedStmt *pgss_planner(Query *parse,
339 : const char *query_string,
340 : int cursorOptions,
341 : ParamListInfo boundParams,
342 : ExplainState *es);
343 : static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
344 : static void pgss_ExecutorRun(QueryDesc *queryDesc,
345 : ScanDirection direction,
346 : uint64 count);
347 : static void pgss_ExecutorFinish(QueryDesc *queryDesc);
348 : static void pgss_ExecutorEnd(QueryDesc *queryDesc);
349 : static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
350 : bool readOnlyTree,
351 : ProcessUtilityContext context, ParamListInfo params,
352 : QueryEnvironment *queryEnv,
353 : DestReceiver *dest, QueryCompletion *qc);
354 : static void pgss_store(const char *query, int64 queryId,
355 : int query_location, int query_len,
356 : pgssStoreKind kind,
357 : double total_time, uint64 rows,
358 : const BufferUsage *bufusage,
359 : const WalUsage *walusage,
360 : const struct JitInstrumentation *jitusage,
361 : JumbleState *jstate,
362 : int parallel_workers_to_launch,
363 : int parallel_workers_launched,
364 : PlannedStmtOrigin planOrigin);
365 : static void pg_stat_statements_internal(FunctionCallInfo fcinfo,
366 : pgssVersion api_version,
367 : bool showtext);
368 : static Size pgss_memsize(void);
369 : static pgssEntry *entry_alloc(pgssHashKey *key, Size query_offset, int query_len,
370 : int encoding, bool sticky);
371 : static void entry_dealloc(void);
372 : static bool qtext_store(const char *query, int query_len,
373 : Size *query_offset, int *gc_count);
374 : static char *qtext_load_file(Size *buffer_size);
375 : static char *qtext_fetch(Size query_offset, int query_len,
376 : char *buffer, Size buffer_size);
377 : static bool need_gc_qtexts(void);
378 : static void gc_qtexts(void);
379 : static TimestampTz entry_reset(Oid userid, Oid dbid, int64 queryid, bool minmax_only);
380 : static char *generate_normalized_query(JumbleState *jstate, const char *query,
381 : int query_loc, int *query_len_p);
382 : static void fill_in_constant_lengths(JumbleState *jstate, const char *query,
383 : int query_loc);
384 : static int comp_location(const void *a, const void *b);
385 :
386 :
387 : /*
388 : * Module load callback
389 : */
390 : void
391 0 : _PG_init(void)
392 : {
393 : /*
394 : * In order to create our shared memory area, we have to be loaded via
395 : * shared_preload_libraries. If not, fall out without hooking into any of
396 : * the main system. (We don't throw error here because it seems useful to
397 : * allow the pg_stat_statements functions to be created even when the
398 : * module isn't active. The functions must protect themselves against
399 : * being called then, however.)
400 : */
401 0 : if (!process_shared_preload_libraries_in_progress)
402 0 : return;
403 :
404 : /*
405 : * Inform the postmaster that we want to enable query_id calculation if
406 : * compute_query_id is set to auto.
407 : */
408 0 : EnableQueryId();
409 :
410 : /*
411 : * Define (or redefine) custom GUC variables.
412 : */
413 0 : DefineCustomIntVariable("pg_stat_statements.max",
414 : "Sets the maximum number of statements tracked by pg_stat_statements.",
415 : NULL,
416 : &pgss_max,
417 : 5000,
418 : 100,
419 : INT_MAX / 2,
420 : PGC_POSTMASTER,
421 : 0,
422 : NULL,
423 : NULL,
424 : NULL);
425 :
426 0 : DefineCustomEnumVariable("pg_stat_statements.track",
427 : "Selects which statements are tracked by pg_stat_statements.",
428 : NULL,
429 : &pgss_track,
430 : PGSS_TRACK_TOP,
431 : track_options,
432 : PGC_SUSET,
433 : 0,
434 : NULL,
435 : NULL,
436 : NULL);
437 :
438 0 : DefineCustomBoolVariable("pg_stat_statements.track_utility",
439 : "Selects whether utility commands are tracked by pg_stat_statements.",
440 : NULL,
441 : &pgss_track_utility,
442 : true,
443 : PGC_SUSET,
444 : 0,
445 : NULL,
446 : NULL,
447 : NULL);
448 :
449 0 : DefineCustomBoolVariable("pg_stat_statements.track_planning",
450 : "Selects whether planning duration is tracked by pg_stat_statements.",
451 : NULL,
452 : &pgss_track_planning,
453 : false,
454 : PGC_SUSET,
455 : 0,
456 : NULL,
457 : NULL,
458 : NULL);
459 :
460 0 : DefineCustomBoolVariable("pg_stat_statements.save",
461 : "Save pg_stat_statements statistics across server shutdowns.",
462 : NULL,
463 : &pgss_save,
464 : true,
465 : PGC_SIGHUP,
466 : 0,
467 : NULL,
468 : NULL,
469 : NULL);
470 :
471 0 : MarkGUCPrefixReserved("pg_stat_statements");
472 :
473 : /*
474 : * Install hooks.
475 : */
476 0 : prev_shmem_request_hook = shmem_request_hook;
477 0 : shmem_request_hook = pgss_shmem_request;
478 0 : prev_shmem_startup_hook = shmem_startup_hook;
479 0 : shmem_startup_hook = pgss_shmem_startup;
480 0 : prev_post_parse_analyze_hook = post_parse_analyze_hook;
481 0 : post_parse_analyze_hook = pgss_post_parse_analyze;
482 0 : prev_planner_hook = planner_hook;
483 0 : planner_hook = pgss_planner;
484 0 : prev_ExecutorStart = ExecutorStart_hook;
485 0 : ExecutorStart_hook = pgss_ExecutorStart;
486 0 : prev_ExecutorRun = ExecutorRun_hook;
487 0 : ExecutorRun_hook = pgss_ExecutorRun;
488 0 : prev_ExecutorFinish = ExecutorFinish_hook;
489 0 : ExecutorFinish_hook = pgss_ExecutorFinish;
490 0 : prev_ExecutorEnd = ExecutorEnd_hook;
491 0 : ExecutorEnd_hook = pgss_ExecutorEnd;
492 0 : prev_ProcessUtility = ProcessUtility_hook;
493 0 : ProcessUtility_hook = pgss_ProcessUtility;
494 0 : }
495 :
496 : /*
497 : * shmem_request hook: request additional shared resources. We'll allocate or
498 : * attach to the shared resources in pgss_shmem_startup().
499 : */
500 : static void
501 0 : pgss_shmem_request(void)
502 : {
503 0 : if (prev_shmem_request_hook)
504 0 : prev_shmem_request_hook();
505 :
506 0 : RequestAddinShmemSpace(pgss_memsize());
507 0 : RequestNamedLWLockTranche("pg_stat_statements", 1);
508 0 : }
509 :
510 : /*
511 : * shmem_startup hook: allocate or attach to shared memory,
512 : * then load any pre-existing statistics from file.
513 : * Also create and load the query-texts file, which is expected to exist
514 : * (even if empty) while the module is enabled.
515 : */
516 : static void
517 0 : pgss_shmem_startup(void)
518 : {
519 0 : bool found;
520 0 : HASHCTL info;
521 0 : FILE *file = NULL;
522 0 : FILE *qfile = NULL;
523 0 : uint32 header;
524 0 : int32 num;
525 0 : int32 pgver;
526 0 : int32 i;
527 0 : int buffer_size;
528 0 : char *buffer = NULL;
529 :
530 0 : if (prev_shmem_startup_hook)
531 0 : prev_shmem_startup_hook();
532 :
533 : /* reset in case this is a restart within the postmaster */
534 0 : pgss = NULL;
535 0 : pgss_hash = NULL;
536 :
537 : /*
538 : * Create or attach to the shared memory state, including hash table
539 : */
540 0 : LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
541 :
542 0 : pgss = ShmemInitStruct("pg_stat_statements",
543 : sizeof(pgssSharedState),
544 : &found);
545 :
546 0 : if (!found)
547 : {
548 : /* First time through ... */
549 0 : pgss->lock = &(GetNamedLWLockTranche("pg_stat_statements"))->lock;
550 0 : pgss->cur_median_usage = ASSUMED_MEDIAN_INIT;
551 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
552 0 : SpinLockInit(&pgss->mutex);
553 0 : pgss->extent = 0;
554 0 : pgss->n_writers = 0;
555 0 : pgss->gc_count = 0;
556 0 : pgss->stats.dealloc = 0;
557 0 : pgss->stats.stats_reset = GetCurrentTimestamp();
558 0 : }
559 :
560 0 : info.keysize = sizeof(pgssHashKey);
561 0 : info.entrysize = sizeof(pgssEntry);
562 0 : pgss_hash = ShmemInitHash("pg_stat_statements hash",
563 0 : pgss_max, pgss_max,
564 : &info,
565 : HASH_ELEM | HASH_BLOBS);
566 :
567 0 : LWLockRelease(AddinShmemInitLock);
568 :
569 : /*
570 : * If we're in the postmaster (or a standalone backend...), set up a shmem
571 : * exit hook to dump the statistics to disk.
572 : */
573 0 : if (!IsUnderPostmaster)
574 0 : on_shmem_exit(pgss_shmem_shutdown, (Datum) 0);
575 :
576 : /*
577 : * Done if some other process already completed our initialization.
578 : */
579 0 : if (found)
580 0 : return;
581 :
582 : /*
583 : * Note: we don't bother with locks here, because there should be no other
584 : * processes running when this code is reached.
585 : */
586 :
587 : /* Unlink query text file possibly left over from crash */
588 0 : unlink(PGSS_TEXT_FILE);
589 :
590 : /* Allocate new query text temp file */
591 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
592 0 : if (qfile == NULL)
593 0 : goto write_error;
594 :
595 : /*
596 : * If we were told not to load old statistics, we're done. (Note we do
597 : * not try to unlink any old dump file in this case. This seems a bit
598 : * questionable but it's the historical behavior.)
599 : */
600 0 : if (!pgss_save)
601 : {
602 0 : FreeFile(qfile);
603 0 : return;
604 : }
605 :
606 : /*
607 : * Attempt to load old statistics from the dump file.
608 : */
609 0 : file = AllocateFile(PGSS_DUMP_FILE, PG_BINARY_R);
610 0 : if (file == NULL)
611 : {
612 0 : if (errno != ENOENT)
613 0 : goto read_error;
614 : /* No existing persisted stats file, so we're done */
615 0 : FreeFile(qfile);
616 0 : return;
617 : }
618 :
619 0 : buffer_size = 2048;
620 0 : buffer = (char *) palloc(buffer_size);
621 :
622 0 : if (fread(&header, sizeof(uint32), 1, file) != 1 ||
623 0 : fread(&pgver, sizeof(uint32), 1, file) != 1 ||
624 0 : fread(&num, sizeof(int32), 1, file) != 1)
625 0 : goto read_error;
626 :
627 0 : if (header != PGSS_FILE_HEADER ||
628 0 : pgver != PGSS_PG_MAJOR_VERSION)
629 0 : goto data_error;
630 :
631 0 : for (i = 0; i < num; i++)
632 : {
633 0 : pgssEntry temp;
634 0 : pgssEntry *entry;
635 0 : Size query_offset;
636 :
637 0 : if (fread(&temp, sizeof(pgssEntry), 1, file) != 1)
638 0 : goto read_error;
639 :
640 : /* Encoding is the only field we can easily sanity-check */
641 0 : if (!PG_VALID_BE_ENCODING(temp.encoding))
642 0 : goto data_error;
643 :
644 : /* Resize buffer as needed */
645 0 : if (temp.query_len >= buffer_size)
646 : {
647 0 : buffer_size = Max(buffer_size * 2, temp.query_len + 1);
648 0 : buffer = repalloc(buffer, buffer_size);
649 0 : }
650 :
651 0 : if (fread(buffer, 1, temp.query_len + 1, file) != temp.query_len + 1)
652 0 : goto read_error;
653 :
654 : /* Should have a trailing null, but let's make sure */
655 0 : buffer[temp.query_len] = '\0';
656 :
657 : /* Skip loading "sticky" entries */
658 0 : if (IS_STICKY(temp.counters))
659 0 : continue;
660 :
661 : /* Store the query text */
662 0 : query_offset = pgss->extent;
663 0 : if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
664 0 : goto write_error;
665 0 : pgss->extent += temp.query_len + 1;
666 :
667 : /* make the hashtable entry (discards old entries if too many) */
668 0 : entry = entry_alloc(&temp.key, query_offset, temp.query_len,
669 0 : temp.encoding,
670 : false);
671 :
672 : /* copy in the actual stats */
673 0 : entry->counters = temp.counters;
674 0 : entry->stats_since = temp.stats_since;
675 0 : entry->minmax_stats_since = temp.minmax_stats_since;
676 0 : }
677 :
678 : /* Read global statistics for pg_stat_statements */
679 0 : if (fread(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
680 0 : goto read_error;
681 :
682 0 : pfree(buffer);
683 0 : FreeFile(file);
684 0 : FreeFile(qfile);
685 :
686 : /*
687 : * Remove the persisted stats file so it's not included in
688 : * backups/replication standbys, etc. A new file will be written on next
689 : * shutdown.
690 : *
691 : * Note: it's okay if the PGSS_TEXT_FILE is included in a basebackup,
692 : * because we remove that file on startup; it acts inversely to
693 : * PGSS_DUMP_FILE, in that it is only supposed to be around when the
694 : * server is running, whereas PGSS_DUMP_FILE is only supposed to be around
695 : * when the server is not running. Leaving the file creates no danger of
696 : * a newly restored database having a spurious record of execution costs,
697 : * which is what we're really concerned about here.
698 : */
699 0 : unlink(PGSS_DUMP_FILE);
700 :
701 0 : return;
702 :
703 : read_error:
704 0 : ereport(LOG,
705 : (errcode_for_file_access(),
706 : errmsg("could not read file \"%s\": %m",
707 : PGSS_DUMP_FILE)));
708 0 : goto fail;
709 : data_error:
710 0 : ereport(LOG,
711 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
712 : errmsg("ignoring invalid data in file \"%s\"",
713 : PGSS_DUMP_FILE)));
714 0 : goto fail;
715 : write_error:
716 0 : ereport(LOG,
717 : (errcode_for_file_access(),
718 : errmsg("could not write file \"%s\": %m",
719 : PGSS_TEXT_FILE)));
720 : fail:
721 0 : if (buffer)
722 0 : pfree(buffer);
723 0 : if (file)
724 0 : FreeFile(file);
725 0 : if (qfile)
726 0 : FreeFile(qfile);
727 : /* If possible, throw away the bogus file; ignore any error */
728 0 : unlink(PGSS_DUMP_FILE);
729 :
730 : /*
731 : * Don't unlink PGSS_TEXT_FILE here; it should always be around while the
732 : * server is running with pg_stat_statements enabled
733 : */
734 0 : }
735 :
736 : /*
737 : * shmem_shutdown hook: Dump statistics into file.
738 : *
739 : * Note: we don't bother with acquiring lock, because there should be no
740 : * other processes running when this is called.
741 : */
742 : static void
743 0 : pgss_shmem_shutdown(int code, Datum arg)
744 : {
745 0 : FILE *file;
746 0 : char *qbuffer = NULL;
747 0 : Size qbuffer_size = 0;
748 0 : HASH_SEQ_STATUS hash_seq;
749 0 : int32 num_entries;
750 0 : pgssEntry *entry;
751 :
752 : /* Don't try to dump during a crash. */
753 0 : if (code)
754 0 : return;
755 :
756 : /* Safety check ... shouldn't get here unless shmem is set up. */
757 0 : if (!pgss || !pgss_hash)
758 0 : return;
759 :
760 : /* Don't dump if told not to. */
761 0 : if (!pgss_save)
762 0 : return;
763 :
764 0 : file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
765 0 : if (file == NULL)
766 0 : goto error;
767 :
768 0 : if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
769 0 : goto error;
770 0 : if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
771 0 : goto error;
772 0 : num_entries = hash_get_num_entries(pgss_hash);
773 0 : if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
774 0 : goto error;
775 :
776 0 : qbuffer = qtext_load_file(&qbuffer_size);
777 0 : if (qbuffer == NULL)
778 0 : goto error;
779 :
780 : /*
781 : * When serializing to disk, we store query texts immediately after their
782 : * entry data. Any orphaned query texts are thereby excluded.
783 : */
784 0 : hash_seq_init(&hash_seq, pgss_hash);
785 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
786 : {
787 0 : int len = entry->query_len;
788 0 : char *qstr = qtext_fetch(entry->query_offset, len,
789 0 : qbuffer, qbuffer_size);
790 :
791 0 : if (qstr == NULL)
792 0 : continue; /* Ignore any entries with bogus texts */
793 :
794 0 : if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
795 0 : fwrite(qstr, 1, len + 1, file) != len + 1)
796 : {
797 : /* note: we assume hash_seq_term won't change errno */
798 0 : hash_seq_term(&hash_seq);
799 0 : goto error;
800 : }
801 0 : }
802 :
803 : /* Dump global statistics for pg_stat_statements */
804 0 : if (fwrite(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
805 0 : goto error;
806 :
807 0 : free(qbuffer);
808 0 : qbuffer = NULL;
809 :
810 0 : if (FreeFile(file))
811 : {
812 0 : file = NULL;
813 0 : goto error;
814 : }
815 :
816 : /*
817 : * Rename file into place, so we atomically replace any old one.
818 : */
819 0 : (void) durable_rename(PGSS_DUMP_FILE ".tmp", PGSS_DUMP_FILE, LOG);
820 :
821 : /* Unlink query-texts file; it's not needed while shutdown */
822 0 : unlink(PGSS_TEXT_FILE);
823 :
824 0 : return;
825 :
826 : error:
827 0 : ereport(LOG,
828 : (errcode_for_file_access(),
829 : errmsg("could not write file \"%s\": %m",
830 : PGSS_DUMP_FILE ".tmp")));
831 0 : free(qbuffer);
832 0 : if (file)
833 0 : FreeFile(file);
834 0 : unlink(PGSS_DUMP_FILE ".tmp");
835 0 : unlink(PGSS_TEXT_FILE);
836 0 : }
837 :
838 : /*
839 : * Post-parse-analysis hook: mark query with a queryId
840 : */
841 : static void
842 0 : pgss_post_parse_analyze(ParseState *pstate, Query *query, JumbleState *jstate)
843 : {
844 0 : if (prev_post_parse_analyze_hook)
845 0 : prev_post_parse_analyze_hook(pstate, query, jstate);
846 :
847 : /* Safety check... */
848 0 : if (!pgss || !pgss_hash || !pgss_enabled(nesting_level))
849 0 : return;
850 :
851 : /*
852 : * If it's EXECUTE, clear the queryId so that stats will accumulate for
853 : * the underlying PREPARE. But don't do this if we're not tracking
854 : * utility statements, to avoid messing up another extension that might be
855 : * tracking them.
856 : */
857 0 : if (query->utilityStmt)
858 : {
859 0 : if (pgss_track_utility && IsA(query->utilityStmt, ExecuteStmt))
860 : {
861 0 : query->queryId = INT64CONST(0);
862 0 : return;
863 : }
864 0 : }
865 :
866 : /*
867 : * If query jumbling were able to identify any ignorable constants, we
868 : * immediately create a hash table entry for the query, so that we can
869 : * record the normalized form of the query string. If there were no such
870 : * constants, the normalized string would be the same as the query text
871 : * anyway, so there's no need for an early entry.
872 : */
873 0 : if (jstate && jstate->clocations_count > 0)
874 0 : pgss_store(pstate->p_sourcetext,
875 0 : query->queryId,
876 0 : query->stmt_location,
877 0 : query->stmt_len,
878 : PGSS_INVALID,
879 : 0,
880 : 0,
881 : NULL,
882 : NULL,
883 : NULL,
884 0 : jstate,
885 : 0,
886 : 0,
887 : PLAN_STMT_UNKNOWN);
888 0 : }
889 :
890 : /*
891 : * Planner hook: forward to regular planner, but measure planning time
892 : * if needed.
893 : */
894 : static PlannedStmt *
895 0 : pgss_planner(Query *parse,
896 : const char *query_string,
897 : int cursorOptions,
898 : ParamListInfo boundParams,
899 : ExplainState *es)
900 : {
901 0 : PlannedStmt *result;
902 :
903 : /*
904 : * We can't process the query if no query_string is provided, as
905 : * pgss_store needs it. We also ignore query without queryid, as it would
906 : * be treated as a utility statement, which may not be the case.
907 : */
908 0 : if (pgss_enabled(nesting_level)
909 0 : && pgss_track_planning && query_string
910 0 : && parse->queryId != INT64CONST(0))
911 : {
912 0 : instr_time start;
913 0 : instr_time duration;
914 0 : BufferUsage bufusage_start,
915 : bufusage;
916 0 : WalUsage walusage_start,
917 : walusage;
918 :
919 : /* We need to track buffer usage as the planner can access them. */
920 0 : bufusage_start = pgBufferUsage;
921 :
922 : /*
923 : * Similarly the planner could write some WAL records in some cases
924 : * (e.g. setting a hint bit with those being WAL-logged)
925 : */
926 0 : walusage_start = pgWalUsage;
927 0 : INSTR_TIME_SET_CURRENT(start);
928 :
929 0 : nesting_level++;
930 0 : PG_TRY();
931 : {
932 0 : if (prev_planner_hook)
933 0 : result = prev_planner_hook(parse, query_string, cursorOptions,
934 0 : boundParams, es);
935 : else
936 0 : result = standard_planner(parse, query_string, cursorOptions,
937 0 : boundParams, es);
938 : }
939 0 : PG_FINALLY();
940 : {
941 0 : nesting_level--;
942 : }
943 0 : PG_END_TRY();
944 :
945 0 : INSTR_TIME_SET_CURRENT(duration);
946 0 : INSTR_TIME_SUBTRACT(duration, start);
947 :
948 : /* calc differences of buffer counters. */
949 0 : memset(&bufusage, 0, sizeof(BufferUsage));
950 0 : BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
951 :
952 : /* calc differences of WAL counters. */
953 0 : memset(&walusage, 0, sizeof(WalUsage));
954 0 : WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
955 :
956 0 : pgss_store(query_string,
957 0 : parse->queryId,
958 0 : parse->stmt_location,
959 0 : parse->stmt_len,
960 : PGSS_PLAN,
961 0 : INSTR_TIME_GET_MILLISEC(duration),
962 : 0,
963 : &bufusage,
964 : &walusage,
965 : NULL,
966 : NULL,
967 : 0,
968 : 0,
969 0 : result->planOrigin);
970 0 : }
971 : else
972 : {
973 : /*
974 : * Even though we're not tracking plan time for this statement, we
975 : * must still increment the nesting level, to ensure that functions
976 : * evaluated during planning are not seen as top-level calls.
977 : */
978 0 : nesting_level++;
979 0 : PG_TRY();
980 : {
981 0 : if (prev_planner_hook)
982 0 : result = prev_planner_hook(parse, query_string, cursorOptions,
983 0 : boundParams, es);
984 : else
985 0 : result = standard_planner(parse, query_string, cursorOptions,
986 0 : boundParams, es);
987 : }
988 0 : PG_FINALLY();
989 : {
990 0 : nesting_level--;
991 : }
992 0 : PG_END_TRY();
993 : }
994 :
995 0 : return result;
996 0 : }
997 :
998 : /*
999 : * ExecutorStart hook: start up tracking if needed
1000 : */
1001 : static void
1002 0 : pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
1003 : {
1004 0 : if (prev_ExecutorStart)
1005 0 : prev_ExecutorStart(queryDesc, eflags);
1006 : else
1007 0 : standard_ExecutorStart(queryDesc, eflags);
1008 :
1009 : /*
1010 : * If query has queryId zero, don't track it. This prevents double
1011 : * counting of optimizable statements that are directly contained in
1012 : * utility statements.
1013 : */
1014 0 : if (pgss_enabled(nesting_level) && queryDesc->plannedstmt->queryId != INT64CONST(0))
1015 : {
1016 : /*
1017 : * Set up to track total elapsed time in ExecutorRun. Make sure the
1018 : * space is allocated in the per-query context so it will go away at
1019 : * ExecutorEnd.
1020 : */
1021 0 : if (queryDesc->totaltime == NULL)
1022 : {
1023 0 : MemoryContext oldcxt;
1024 :
1025 0 : oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
1026 0 : queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL, false);
1027 0 : MemoryContextSwitchTo(oldcxt);
1028 0 : }
1029 0 : }
1030 0 : }
1031 :
1032 : /*
1033 : * ExecutorRun hook: all we need do is track nesting depth
1034 : */
1035 : static void
1036 0 : pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count)
1037 : {
1038 0 : nesting_level++;
1039 0 : PG_TRY();
1040 : {
1041 0 : if (prev_ExecutorRun)
1042 0 : prev_ExecutorRun(queryDesc, direction, count);
1043 : else
1044 0 : standard_ExecutorRun(queryDesc, direction, count);
1045 : }
1046 0 : PG_FINALLY();
1047 : {
1048 0 : nesting_level--;
1049 : }
1050 0 : PG_END_TRY();
1051 0 : }
1052 :
1053 : /*
1054 : * ExecutorFinish hook: all we need do is track nesting depth
1055 : */
1056 : static void
1057 0 : pgss_ExecutorFinish(QueryDesc *queryDesc)
1058 : {
1059 0 : nesting_level++;
1060 0 : PG_TRY();
1061 : {
1062 0 : if (prev_ExecutorFinish)
1063 0 : prev_ExecutorFinish(queryDesc);
1064 : else
1065 0 : standard_ExecutorFinish(queryDesc);
1066 : }
1067 0 : PG_FINALLY();
1068 : {
1069 0 : nesting_level--;
1070 : }
1071 0 : PG_END_TRY();
1072 0 : }
1073 :
1074 : /*
1075 : * ExecutorEnd hook: store results if needed
1076 : */
1077 : static void
1078 0 : pgss_ExecutorEnd(QueryDesc *queryDesc)
1079 : {
1080 0 : int64 queryId = queryDesc->plannedstmt->queryId;
1081 :
1082 0 : if (queryId != INT64CONST(0) && queryDesc->totaltime &&
1083 0 : pgss_enabled(nesting_level))
1084 : {
1085 : /*
1086 : * Make sure stats accumulation is done. (Note: it's okay if several
1087 : * levels of hook all do this.)
1088 : */
1089 0 : InstrEndLoop(queryDesc->totaltime);
1090 :
1091 0 : pgss_store(queryDesc->sourceText,
1092 0 : queryId,
1093 0 : queryDesc->plannedstmt->stmt_location,
1094 0 : queryDesc->plannedstmt->stmt_len,
1095 : PGSS_EXEC,
1096 0 : INSTR_TIME_GET_MILLISEC(queryDesc->totaltime->total),
1097 0 : queryDesc->estate->es_total_processed,
1098 0 : &queryDesc->totaltime->bufusage,
1099 0 : &queryDesc->totaltime->walusage,
1100 0 : queryDesc->estate->es_jit ? &queryDesc->estate->es_jit->instr : NULL,
1101 : NULL,
1102 0 : queryDesc->estate->es_parallel_workers_to_launch,
1103 0 : queryDesc->estate->es_parallel_workers_launched,
1104 0 : queryDesc->plannedstmt->planOrigin);
1105 0 : }
1106 :
1107 0 : if (prev_ExecutorEnd)
1108 0 : prev_ExecutorEnd(queryDesc);
1109 : else
1110 0 : standard_ExecutorEnd(queryDesc);
1111 0 : }
1112 :
1113 : /*
1114 : * ProcessUtility hook
1115 : */
1116 : static void
1117 0 : pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
1118 : bool readOnlyTree,
1119 : ProcessUtilityContext context,
1120 : ParamListInfo params, QueryEnvironment *queryEnv,
1121 : DestReceiver *dest, QueryCompletion *qc)
1122 : {
1123 0 : Node *parsetree = pstmt->utilityStmt;
1124 0 : int64 saved_queryId = pstmt->queryId;
1125 0 : int saved_stmt_location = pstmt->stmt_location;
1126 0 : int saved_stmt_len = pstmt->stmt_len;
1127 0 : bool enabled = pgss_track_utility && pgss_enabled(nesting_level);
1128 :
1129 : /*
1130 : * Force utility statements to get queryId zero. We do this even in cases
1131 : * where the statement contains an optimizable statement for which a
1132 : * queryId could be derived (such as EXPLAIN or DECLARE CURSOR). For such
1133 : * cases, runtime control will first go through ProcessUtility and then
1134 : * the executor, and we don't want the executor hooks to do anything,
1135 : * since we are already measuring the statement's costs at the utility
1136 : * level.
1137 : *
1138 : * Note that this is only done if pg_stat_statements is enabled and
1139 : * configured to track utility statements, in the unlikely possibility
1140 : * that user configured another extension to handle utility statements
1141 : * only.
1142 : */
1143 0 : if (enabled)
1144 0 : pstmt->queryId = INT64CONST(0);
1145 :
1146 : /*
1147 : * If it's an EXECUTE statement, we don't track it and don't increment the
1148 : * nesting level. This allows the cycles to be charged to the underlying
1149 : * PREPARE instead (by the Executor hooks), which is much more useful.
1150 : *
1151 : * We also don't track execution of PREPARE. If we did, we would get one
1152 : * hash table entry for the PREPARE (with hash calculated from the query
1153 : * string), and then a different one with the same query string (but hash
1154 : * calculated from the query tree) would be used to accumulate costs of
1155 : * ensuing EXECUTEs. This would be confusing. Since PREPARE doesn't
1156 : * actually run the planner (only parse+rewrite), its costs are generally
1157 : * pretty negligible and it seems okay to just ignore it.
1158 : */
1159 0 : if (enabled &&
1160 0 : !IsA(parsetree, ExecuteStmt) &&
1161 0 : !IsA(parsetree, PrepareStmt))
1162 : {
1163 0 : instr_time start;
1164 0 : instr_time duration;
1165 0 : uint64 rows;
1166 0 : BufferUsage bufusage_start,
1167 : bufusage;
1168 0 : WalUsage walusage_start,
1169 : walusage;
1170 :
1171 0 : bufusage_start = pgBufferUsage;
1172 0 : walusage_start = pgWalUsage;
1173 0 : INSTR_TIME_SET_CURRENT(start);
1174 :
1175 0 : nesting_level++;
1176 0 : PG_TRY();
1177 : {
1178 0 : if (prev_ProcessUtility)
1179 0 : prev_ProcessUtility(pstmt, queryString, readOnlyTree,
1180 0 : context, params, queryEnv,
1181 0 : dest, qc);
1182 : else
1183 0 : standard_ProcessUtility(pstmt, queryString, readOnlyTree,
1184 0 : context, params, queryEnv,
1185 0 : dest, qc);
1186 : }
1187 0 : PG_FINALLY();
1188 : {
1189 0 : nesting_level--;
1190 : }
1191 0 : PG_END_TRY();
1192 :
1193 : /*
1194 : * CAUTION: do not access the *pstmt data structure again below here.
1195 : * If it was a ROLLBACK or similar, that data structure may have been
1196 : * freed. We must copy everything we still need into local variables,
1197 : * which we did above.
1198 : *
1199 : * For the same reason, we can't risk restoring pstmt->queryId to its
1200 : * former value, which'd otherwise be a good idea.
1201 : */
1202 :
1203 0 : INSTR_TIME_SET_CURRENT(duration);
1204 0 : INSTR_TIME_SUBTRACT(duration, start);
1205 :
1206 : /*
1207 : * Track the total number of rows retrieved or affected by the utility
1208 : * statements of COPY, FETCH, CREATE TABLE AS, CREATE MATERIALIZED
1209 : * VIEW, REFRESH MATERIALIZED VIEW and SELECT INTO.
1210 : */
1211 0 : rows = (qc && (qc->commandTag == CMDTAG_COPY ||
1212 0 : qc->commandTag == CMDTAG_FETCH ||
1213 0 : qc->commandTag == CMDTAG_SELECT ||
1214 0 : qc->commandTag == CMDTAG_REFRESH_MATERIALIZED_VIEW)) ?
1215 0 : qc->nprocessed : 0;
1216 :
1217 : /* calc differences of buffer counters. */
1218 0 : memset(&bufusage, 0, sizeof(BufferUsage));
1219 0 : BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
1220 :
1221 : /* calc differences of WAL counters. */
1222 0 : memset(&walusage, 0, sizeof(WalUsage));
1223 0 : WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
1224 :
1225 0 : pgss_store(queryString,
1226 0 : saved_queryId,
1227 0 : saved_stmt_location,
1228 0 : saved_stmt_len,
1229 : PGSS_EXEC,
1230 0 : INSTR_TIME_GET_MILLISEC(duration),
1231 0 : rows,
1232 : &bufusage,
1233 : &walusage,
1234 : NULL,
1235 : NULL,
1236 : 0,
1237 : 0,
1238 0 : pstmt->planOrigin);
1239 0 : }
1240 : else
1241 : {
1242 : /*
1243 : * Even though we're not tracking execution time for this statement,
1244 : * we must still increment the nesting level, to ensure that functions
1245 : * evaluated within it are not seen as top-level calls. But don't do
1246 : * so for EXECUTE; that way, when control reaches pgss_planner or
1247 : * pgss_ExecutorStart, we will treat the costs as top-level if
1248 : * appropriate. Likewise, don't bump for PREPARE, so that parse
1249 : * analysis will treat the statement as top-level if appropriate.
1250 : *
1251 : * To be absolutely certain we don't mess up the nesting level,
1252 : * evaluate the bump_level condition just once.
1253 : */
1254 0 : bool bump_level =
1255 0 : !IsA(parsetree, ExecuteStmt) &&
1256 0 : !IsA(parsetree, PrepareStmt);
1257 :
1258 0 : if (bump_level)
1259 0 : nesting_level++;
1260 0 : PG_TRY();
1261 : {
1262 0 : if (prev_ProcessUtility)
1263 0 : prev_ProcessUtility(pstmt, queryString, readOnlyTree,
1264 0 : context, params, queryEnv,
1265 0 : dest, qc);
1266 : else
1267 0 : standard_ProcessUtility(pstmt, queryString, readOnlyTree,
1268 0 : context, params, queryEnv,
1269 0 : dest, qc);
1270 : }
1271 0 : PG_FINALLY();
1272 : {
1273 0 : if (bump_level)
1274 0 : nesting_level--;
1275 : }
1276 0 : PG_END_TRY();
1277 0 : }
1278 0 : }
1279 :
1280 : /*
1281 : * Store some statistics for a statement.
1282 : *
1283 : * If jstate is not NULL then we're trying to create an entry for which
1284 : * we have no statistics as yet; we just want to record the normalized
1285 : * query string. total_time, rows, bufusage and walusage are ignored in this
1286 : * case.
1287 : *
1288 : * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
1289 : * for the arrays in the Counters field.
1290 : */
1291 : static void
1292 0 : pgss_store(const char *query, int64 queryId,
1293 : int query_location, int query_len,
1294 : pgssStoreKind kind,
1295 : double total_time, uint64 rows,
1296 : const BufferUsage *bufusage,
1297 : const WalUsage *walusage,
1298 : const struct JitInstrumentation *jitusage,
1299 : JumbleState *jstate,
1300 : int parallel_workers_to_launch,
1301 : int parallel_workers_launched,
1302 : PlannedStmtOrigin planOrigin)
1303 : {
1304 0 : pgssHashKey key;
1305 0 : pgssEntry *entry;
1306 0 : char *norm_query = NULL;
1307 0 : int encoding = GetDatabaseEncoding();
1308 :
1309 0 : Assert(query != NULL);
1310 :
1311 : /* Safety check... */
1312 0 : if (!pgss || !pgss_hash)
1313 0 : return;
1314 :
1315 : /*
1316 : * Nothing to do if compute_query_id isn't enabled and no other module
1317 : * computed a query identifier.
1318 : */
1319 0 : if (queryId == INT64CONST(0))
1320 0 : return;
1321 :
1322 : /*
1323 : * Confine our attention to the relevant part of the string, if the query
1324 : * is a portion of a multi-statement source string, and update query
1325 : * location and length if needed.
1326 : */
1327 0 : query = CleanQuerytext(query, &query_location, &query_len);
1328 :
1329 : /* Set up key for hashtable search */
1330 :
1331 : /* clear padding */
1332 0 : memset(&key, 0, sizeof(pgssHashKey));
1333 :
1334 0 : key.userid = GetUserId();
1335 0 : key.dbid = MyDatabaseId;
1336 0 : key.queryid = queryId;
1337 0 : key.toplevel = (nesting_level == 0);
1338 :
1339 : /* Lookup the hash table entry with shared lock. */
1340 0 : LWLockAcquire(pgss->lock, LW_SHARED);
1341 :
1342 0 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
1343 :
1344 : /* Create new entry, if not present */
1345 0 : if (!entry)
1346 : {
1347 0 : Size query_offset;
1348 0 : int gc_count;
1349 0 : bool stored;
1350 0 : bool do_gc;
1351 :
1352 : /*
1353 : * Create a new, normalized query string if caller asked. We don't
1354 : * need to hold the lock while doing this work. (Note: in any case,
1355 : * it's possible that someone else creates a duplicate hashtable entry
1356 : * in the interval where we don't hold the lock below. That case is
1357 : * handled by entry_alloc.)
1358 : */
1359 0 : if (jstate)
1360 : {
1361 0 : LWLockRelease(pgss->lock);
1362 0 : norm_query = generate_normalized_query(jstate, query,
1363 0 : query_location,
1364 : &query_len);
1365 0 : LWLockAcquire(pgss->lock, LW_SHARED);
1366 0 : }
1367 :
1368 : /* Append new query text to file with only shared lock held */
1369 0 : stored = qtext_store(norm_query ? norm_query : query, query_len,
1370 : &query_offset, &gc_count);
1371 :
1372 : /*
1373 : * Determine whether we need to garbage collect external query texts
1374 : * while the shared lock is still held. This micro-optimization
1375 : * avoids taking the time to decide this while holding exclusive lock.
1376 : */
1377 0 : do_gc = need_gc_qtexts();
1378 :
1379 : /* Need exclusive lock to make a new hashtable entry - promote */
1380 0 : LWLockRelease(pgss->lock);
1381 0 : LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
1382 :
1383 : /*
1384 : * A garbage collection may have occurred while we weren't holding the
1385 : * lock. In the unlikely event that this happens, the query text we
1386 : * stored above will have been garbage collected, so write it again.
1387 : * This should be infrequent enough that doing it while holding
1388 : * exclusive lock isn't a performance problem.
1389 : */
1390 0 : if (!stored || pgss->gc_count != gc_count)
1391 0 : stored = qtext_store(norm_query ? norm_query : query, query_len,
1392 : &query_offset, NULL);
1393 :
1394 : /* If we failed to write to the text file, give up */
1395 0 : if (!stored)
1396 0 : goto done;
1397 :
1398 : /* OK to create a new hashtable entry */
1399 0 : entry = entry_alloc(&key, query_offset, query_len, encoding,
1400 0 : jstate != NULL);
1401 :
1402 : /* If needed, perform garbage collection while exclusive lock held */
1403 0 : if (do_gc)
1404 0 : gc_qtexts();
1405 0 : }
1406 :
1407 : /* Increment the counts, except when jstate is not NULL */
1408 0 : if (!jstate)
1409 : {
1410 0 : Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
1411 :
1412 : /*
1413 : * Grab the spinlock while updating the counters (see comment about
1414 : * locking rules at the head of the file)
1415 : */
1416 0 : SpinLockAcquire(&entry->mutex);
1417 :
1418 : /* "Unstick" entry if it was previously sticky */
1419 0 : if (IS_STICKY(entry->counters))
1420 0 : entry->counters.usage = USAGE_INIT;
1421 :
1422 0 : entry->counters.calls[kind] += 1;
1423 0 : entry->counters.total_time[kind] += total_time;
1424 :
1425 0 : if (entry->counters.calls[kind] == 1)
1426 : {
1427 0 : entry->counters.min_time[kind] = total_time;
1428 0 : entry->counters.max_time[kind] = total_time;
1429 0 : entry->counters.mean_time[kind] = total_time;
1430 0 : }
1431 : else
1432 : {
1433 : /*
1434 : * Welford's method for accurately computing variance. See
1435 : * <http://www.johndcook.com/blog/standard_deviation/>
1436 : */
1437 0 : double old_mean = entry->counters.mean_time[kind];
1438 :
1439 0 : entry->counters.mean_time[kind] +=
1440 0 : (total_time - old_mean) / entry->counters.calls[kind];
1441 0 : entry->counters.sum_var_time[kind] +=
1442 0 : (total_time - old_mean) * (total_time - entry->counters.mean_time[kind]);
1443 :
1444 : /*
1445 : * Calculate min and max time. min = 0 and max = 0 means that the
1446 : * min/max statistics were reset
1447 : */
1448 0 : if (entry->counters.min_time[kind] == 0
1449 0 : && entry->counters.max_time[kind] == 0)
1450 : {
1451 0 : entry->counters.min_time[kind] = total_time;
1452 0 : entry->counters.max_time[kind] = total_time;
1453 0 : }
1454 : else
1455 : {
1456 0 : if (entry->counters.min_time[kind] > total_time)
1457 0 : entry->counters.min_time[kind] = total_time;
1458 0 : if (entry->counters.max_time[kind] < total_time)
1459 0 : entry->counters.max_time[kind] = total_time;
1460 : }
1461 0 : }
1462 0 : entry->counters.rows += rows;
1463 0 : entry->counters.shared_blks_hit += bufusage->shared_blks_hit;
1464 0 : entry->counters.shared_blks_read += bufusage->shared_blks_read;
1465 0 : entry->counters.shared_blks_dirtied += bufusage->shared_blks_dirtied;
1466 0 : entry->counters.shared_blks_written += bufusage->shared_blks_written;
1467 0 : entry->counters.local_blks_hit += bufusage->local_blks_hit;
1468 0 : entry->counters.local_blks_read += bufusage->local_blks_read;
1469 0 : entry->counters.local_blks_dirtied += bufusage->local_blks_dirtied;
1470 0 : entry->counters.local_blks_written += bufusage->local_blks_written;
1471 0 : entry->counters.temp_blks_read += bufusage->temp_blks_read;
1472 0 : entry->counters.temp_blks_written += bufusage->temp_blks_written;
1473 0 : entry->counters.shared_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_read_time);
1474 0 : entry->counters.shared_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_write_time);
1475 0 : entry->counters.local_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_read_time);
1476 0 : entry->counters.local_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_write_time);
1477 0 : entry->counters.temp_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_read_time);
1478 0 : entry->counters.temp_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_write_time);
1479 0 : entry->counters.usage += USAGE_EXEC(total_time);
1480 0 : entry->counters.wal_records += walusage->wal_records;
1481 0 : entry->counters.wal_fpi += walusage->wal_fpi;
1482 0 : entry->counters.wal_bytes += walusage->wal_bytes;
1483 0 : entry->counters.wal_buffers_full += walusage->wal_buffers_full;
1484 0 : if (jitusage)
1485 : {
1486 0 : entry->counters.jit_functions += jitusage->created_functions;
1487 0 : entry->counters.jit_generation_time += INSTR_TIME_GET_MILLISEC(jitusage->generation_counter);
1488 :
1489 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->deform_counter))
1490 0 : entry->counters.jit_deform_count++;
1491 0 : entry->counters.jit_deform_time += INSTR_TIME_GET_MILLISEC(jitusage->deform_counter);
1492 :
1493 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter))
1494 0 : entry->counters.jit_inlining_count++;
1495 0 : entry->counters.jit_inlining_time += INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter);
1496 :
1497 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter))
1498 0 : entry->counters.jit_optimization_count++;
1499 0 : entry->counters.jit_optimization_time += INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter);
1500 :
1501 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->emission_counter))
1502 0 : entry->counters.jit_emission_count++;
1503 0 : entry->counters.jit_emission_time += INSTR_TIME_GET_MILLISEC(jitusage->emission_counter);
1504 0 : }
1505 :
1506 : /* parallel worker counters */
1507 0 : entry->counters.parallel_workers_to_launch += parallel_workers_to_launch;
1508 0 : entry->counters.parallel_workers_launched += parallel_workers_launched;
1509 :
1510 : /* plan cache counters */
1511 0 : if (planOrigin == PLAN_STMT_CACHE_GENERIC)
1512 0 : entry->counters.generic_plan_calls++;
1513 0 : else if (planOrigin == PLAN_STMT_CACHE_CUSTOM)
1514 0 : entry->counters.custom_plan_calls++;
1515 :
1516 0 : SpinLockRelease(&entry->mutex);
1517 0 : }
1518 :
1519 : done:
1520 0 : LWLockRelease(pgss->lock);
1521 :
1522 : /* We postpone this clean-up until we're out of the lock */
1523 0 : if (norm_query)
1524 0 : pfree(norm_query);
1525 0 : }
1526 :
1527 : /*
1528 : * Reset statement statistics corresponding to userid, dbid, and queryid.
1529 : */
1530 : Datum
1531 0 : pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
1532 : {
1533 0 : Oid userid;
1534 0 : Oid dbid;
1535 0 : int64 queryid;
1536 :
1537 0 : userid = PG_GETARG_OID(0);
1538 0 : dbid = PG_GETARG_OID(1);
1539 0 : queryid = PG_GETARG_INT64(2);
1540 :
1541 0 : entry_reset(userid, dbid, queryid, false);
1542 :
1543 0 : PG_RETURN_VOID();
1544 0 : }
1545 :
1546 : Datum
1547 0 : pg_stat_statements_reset_1_11(PG_FUNCTION_ARGS)
1548 : {
1549 0 : Oid userid;
1550 0 : Oid dbid;
1551 0 : int64 queryid;
1552 0 : bool minmax_only;
1553 :
1554 0 : userid = PG_GETARG_OID(0);
1555 0 : dbid = PG_GETARG_OID(1);
1556 0 : queryid = PG_GETARG_INT64(2);
1557 0 : minmax_only = PG_GETARG_BOOL(3);
1558 :
1559 0 : PG_RETURN_TIMESTAMPTZ(entry_reset(userid, dbid, queryid, minmax_only));
1560 0 : }
1561 :
1562 : /*
1563 : * Reset statement statistics.
1564 : */
1565 : Datum
1566 0 : pg_stat_statements_reset(PG_FUNCTION_ARGS)
1567 : {
1568 0 : entry_reset(0, 0, 0, false);
1569 :
1570 0 : PG_RETURN_VOID();
1571 : }
1572 :
1573 : /* Number of output arguments (columns) for various API versions */
1574 : #define PG_STAT_STATEMENTS_COLS_V1_0 14
1575 : #define PG_STAT_STATEMENTS_COLS_V1_1 18
1576 : #define PG_STAT_STATEMENTS_COLS_V1_2 19
1577 : #define PG_STAT_STATEMENTS_COLS_V1_3 23
1578 : #define PG_STAT_STATEMENTS_COLS_V1_8 32
1579 : #define PG_STAT_STATEMENTS_COLS_V1_9 33
1580 : #define PG_STAT_STATEMENTS_COLS_V1_10 43
1581 : #define PG_STAT_STATEMENTS_COLS_V1_11 49
1582 : #define PG_STAT_STATEMENTS_COLS_V1_12 52
1583 : #define PG_STAT_STATEMENTS_COLS_V1_13 54
1584 : #define PG_STAT_STATEMENTS_COLS 54 /* maximum of above */
1585 :
1586 : /*
1587 : * Retrieve statement statistics.
1588 : *
1589 : * The SQL API of this function has changed multiple times, and will likely
1590 : * do so again in future. To support the case where a newer version of this
1591 : * loadable module is being used with an old SQL declaration of the function,
1592 : * we continue to support the older API versions. For 1.2 and later, the
1593 : * expected API version is identified by embedding it in the C name of the
1594 : * function. Unfortunately we weren't bright enough to do that for 1.1.
1595 : */
1596 : Datum
1597 0 : pg_stat_statements_1_13(PG_FUNCTION_ARGS)
1598 : {
1599 0 : bool showtext = PG_GETARG_BOOL(0);
1600 :
1601 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_13, showtext);
1602 :
1603 0 : return (Datum) 0;
1604 0 : }
1605 :
1606 : Datum
1607 0 : pg_stat_statements_1_12(PG_FUNCTION_ARGS)
1608 : {
1609 0 : bool showtext = PG_GETARG_BOOL(0);
1610 :
1611 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_12, showtext);
1612 :
1613 0 : return (Datum) 0;
1614 0 : }
1615 :
1616 : Datum
1617 0 : pg_stat_statements_1_11(PG_FUNCTION_ARGS)
1618 : {
1619 0 : bool showtext = PG_GETARG_BOOL(0);
1620 :
1621 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_11, showtext);
1622 :
1623 0 : return (Datum) 0;
1624 0 : }
1625 :
1626 : Datum
1627 0 : pg_stat_statements_1_10(PG_FUNCTION_ARGS)
1628 : {
1629 0 : bool showtext = PG_GETARG_BOOL(0);
1630 :
1631 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_10, showtext);
1632 :
1633 0 : return (Datum) 0;
1634 0 : }
1635 :
1636 : Datum
1637 0 : pg_stat_statements_1_9(PG_FUNCTION_ARGS)
1638 : {
1639 0 : bool showtext = PG_GETARG_BOOL(0);
1640 :
1641 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_9, showtext);
1642 :
1643 0 : return (Datum) 0;
1644 0 : }
1645 :
1646 : Datum
1647 0 : pg_stat_statements_1_8(PG_FUNCTION_ARGS)
1648 : {
1649 0 : bool showtext = PG_GETARG_BOOL(0);
1650 :
1651 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
1652 :
1653 0 : return (Datum) 0;
1654 0 : }
1655 :
1656 : Datum
1657 0 : pg_stat_statements_1_3(PG_FUNCTION_ARGS)
1658 : {
1659 0 : bool showtext = PG_GETARG_BOOL(0);
1660 :
1661 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
1662 :
1663 0 : return (Datum) 0;
1664 0 : }
1665 :
1666 : Datum
1667 0 : pg_stat_statements_1_2(PG_FUNCTION_ARGS)
1668 : {
1669 0 : bool showtext = PG_GETARG_BOOL(0);
1670 :
1671 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
1672 :
1673 0 : return (Datum) 0;
1674 0 : }
1675 :
1676 : /*
1677 : * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
1678 : * This can be removed someday, perhaps.
1679 : */
1680 : Datum
1681 0 : pg_stat_statements(PG_FUNCTION_ARGS)
1682 : {
1683 : /* If it's really API 1.1, we'll figure that out below */
1684 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
1685 :
1686 0 : return (Datum) 0;
1687 : }
1688 :
1689 : /* Common code for all versions of pg_stat_statements() */
1690 : static void
1691 0 : pg_stat_statements_internal(FunctionCallInfo fcinfo,
1692 : pgssVersion api_version,
1693 : bool showtext)
1694 : {
1695 0 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1696 0 : Oid userid = GetUserId();
1697 0 : bool is_allowed_role = false;
1698 0 : char *qbuffer = NULL;
1699 0 : Size qbuffer_size = 0;
1700 0 : Size extent = 0;
1701 0 : int gc_count = 0;
1702 0 : HASH_SEQ_STATUS hash_seq;
1703 0 : pgssEntry *entry;
1704 :
1705 : /*
1706 : * Superusers or roles with the privileges of pg_read_all_stats members
1707 : * are allowed
1708 : */
1709 0 : is_allowed_role = has_privs_of_role(userid, ROLE_PG_READ_ALL_STATS);
1710 :
1711 : /* hash table must exist already */
1712 0 : if (!pgss || !pgss_hash)
1713 0 : ereport(ERROR,
1714 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1715 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
1716 :
1717 0 : InitMaterializedSRF(fcinfo, 0);
1718 :
1719 : /*
1720 : * Check we have the expected number of output arguments. Aside from
1721 : * being a good safety check, we need a kluge here to detect API version
1722 : * 1.1, which was wedged into the code in an ill-considered way.
1723 : */
1724 0 : switch (rsinfo->setDesc->natts)
1725 : {
1726 : case PG_STAT_STATEMENTS_COLS_V1_0:
1727 0 : if (api_version != PGSS_V1_0)
1728 0 : elog(ERROR, "incorrect number of output arguments");
1729 0 : break;
1730 : case PG_STAT_STATEMENTS_COLS_V1_1:
1731 : /* pg_stat_statements() should have told us 1.0 */
1732 0 : if (api_version != PGSS_V1_0)
1733 0 : elog(ERROR, "incorrect number of output arguments");
1734 0 : api_version = PGSS_V1_1;
1735 0 : break;
1736 : case PG_STAT_STATEMENTS_COLS_V1_2:
1737 0 : if (api_version != PGSS_V1_2)
1738 0 : elog(ERROR, "incorrect number of output arguments");
1739 0 : break;
1740 : case PG_STAT_STATEMENTS_COLS_V1_3:
1741 0 : if (api_version != PGSS_V1_3)
1742 0 : elog(ERROR, "incorrect number of output arguments");
1743 0 : break;
1744 : case PG_STAT_STATEMENTS_COLS_V1_8:
1745 0 : if (api_version != PGSS_V1_8)
1746 0 : elog(ERROR, "incorrect number of output arguments");
1747 0 : break;
1748 : case PG_STAT_STATEMENTS_COLS_V1_9:
1749 0 : if (api_version != PGSS_V1_9)
1750 0 : elog(ERROR, "incorrect number of output arguments");
1751 0 : break;
1752 : case PG_STAT_STATEMENTS_COLS_V1_10:
1753 0 : if (api_version != PGSS_V1_10)
1754 0 : elog(ERROR, "incorrect number of output arguments");
1755 0 : break;
1756 : case PG_STAT_STATEMENTS_COLS_V1_11:
1757 0 : if (api_version != PGSS_V1_11)
1758 0 : elog(ERROR, "incorrect number of output arguments");
1759 0 : break;
1760 : case PG_STAT_STATEMENTS_COLS_V1_12:
1761 0 : if (api_version != PGSS_V1_12)
1762 0 : elog(ERROR, "incorrect number of output arguments");
1763 0 : break;
1764 : case PG_STAT_STATEMENTS_COLS_V1_13:
1765 0 : if (api_version != PGSS_V1_13)
1766 0 : elog(ERROR, "incorrect number of output arguments");
1767 0 : break;
1768 : default:
1769 0 : elog(ERROR, "incorrect number of output arguments");
1770 0 : }
1771 :
1772 : /*
1773 : * We'd like to load the query text file (if needed) while not holding any
1774 : * lock on pgss->lock. In the worst case we'll have to do this again
1775 : * after we have the lock, but it's unlikely enough to make this a win
1776 : * despite occasional duplicated work. We need to reload if anybody
1777 : * writes to the file (either a retail qtext_store(), or a garbage
1778 : * collection) between this point and where we've gotten shared lock. If
1779 : * a qtext_store is actually in progress when we look, we might as well
1780 : * skip the speculative load entirely.
1781 : */
1782 0 : if (showtext)
1783 : {
1784 0 : int n_writers;
1785 :
1786 : /* Take the mutex so we can examine variables */
1787 0 : SpinLockAcquire(&pgss->mutex);
1788 0 : extent = pgss->extent;
1789 0 : n_writers = pgss->n_writers;
1790 0 : gc_count = pgss->gc_count;
1791 0 : SpinLockRelease(&pgss->mutex);
1792 :
1793 : /* No point in loading file now if there are active writers */
1794 0 : if (n_writers == 0)
1795 0 : qbuffer = qtext_load_file(&qbuffer_size);
1796 0 : }
1797 :
1798 : /*
1799 : * Get shared lock, load or reload the query text file if we must, and
1800 : * iterate over the hashtable entries.
1801 : *
1802 : * With a large hash table, we might be holding the lock rather longer
1803 : * than one could wish. However, this only blocks creation of new hash
1804 : * table entries, and the larger the hash table the less likely that is to
1805 : * be needed. So we can hope this is okay. Perhaps someday we'll decide
1806 : * we need to partition the hash table to limit the time spent holding any
1807 : * one lock.
1808 : */
1809 0 : LWLockAcquire(pgss->lock, LW_SHARED);
1810 :
1811 0 : if (showtext)
1812 : {
1813 : /*
1814 : * Here it is safe to examine extent and gc_count without taking the
1815 : * mutex. Note that although other processes might change
1816 : * pgss->extent just after we look at it, the strings they then write
1817 : * into the file cannot yet be referenced in the hashtable, so we
1818 : * don't care whether we see them or not.
1819 : *
1820 : * If qtext_load_file fails, we just press on; we'll return NULL for
1821 : * every query text.
1822 : */
1823 0 : if (qbuffer == NULL ||
1824 0 : pgss->extent != extent ||
1825 0 : pgss->gc_count != gc_count)
1826 : {
1827 0 : free(qbuffer);
1828 0 : qbuffer = qtext_load_file(&qbuffer_size);
1829 0 : }
1830 0 : }
1831 :
1832 0 : hash_seq_init(&hash_seq, pgss_hash);
1833 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
1834 : {
1835 0 : Datum values[PG_STAT_STATEMENTS_COLS];
1836 0 : bool nulls[PG_STAT_STATEMENTS_COLS];
1837 0 : int i = 0;
1838 0 : Counters tmp;
1839 0 : double stddev;
1840 0 : int64 queryid = entry->key.queryid;
1841 0 : TimestampTz stats_since;
1842 0 : TimestampTz minmax_stats_since;
1843 :
1844 0 : memset(values, 0, sizeof(values));
1845 0 : memset(nulls, 0, sizeof(nulls));
1846 :
1847 0 : values[i++] = ObjectIdGetDatum(entry->key.userid);
1848 0 : values[i++] = ObjectIdGetDatum(entry->key.dbid);
1849 0 : if (api_version >= PGSS_V1_9)
1850 0 : values[i++] = BoolGetDatum(entry->key.toplevel);
1851 :
1852 0 : if (is_allowed_role || entry->key.userid == userid)
1853 : {
1854 0 : if (api_version >= PGSS_V1_2)
1855 0 : values[i++] = Int64GetDatumFast(queryid);
1856 :
1857 0 : if (showtext)
1858 : {
1859 0 : char *qstr = qtext_fetch(entry->query_offset,
1860 0 : entry->query_len,
1861 0 : qbuffer,
1862 0 : qbuffer_size);
1863 :
1864 0 : if (qstr)
1865 : {
1866 0 : char *enc;
1867 :
1868 0 : enc = pg_any_to_server(qstr,
1869 0 : entry->query_len,
1870 0 : entry->encoding);
1871 :
1872 0 : values[i++] = CStringGetTextDatum(enc);
1873 :
1874 0 : if (enc != qstr)
1875 0 : pfree(enc);
1876 0 : }
1877 : else
1878 : {
1879 : /* Just return a null if we fail to find the text */
1880 0 : nulls[i++] = true;
1881 : }
1882 0 : }
1883 : else
1884 : {
1885 : /* Query text not requested */
1886 0 : nulls[i++] = true;
1887 : }
1888 0 : }
1889 : else
1890 : {
1891 : /* Don't show queryid */
1892 0 : if (api_version >= PGSS_V1_2)
1893 0 : nulls[i++] = true;
1894 :
1895 : /*
1896 : * Don't show query text, but hint as to the reason for not doing
1897 : * so if it was requested
1898 : */
1899 0 : if (showtext)
1900 0 : values[i++] = CStringGetTextDatum("<insufficient privilege>");
1901 : else
1902 0 : nulls[i++] = true;
1903 : }
1904 :
1905 : /* copy counters to a local variable to keep locking time short */
1906 0 : SpinLockAcquire(&entry->mutex);
1907 0 : tmp = entry->counters;
1908 0 : SpinLockRelease(&entry->mutex);
1909 :
1910 : /*
1911 : * The spinlock is not required when reading these two as they are
1912 : * always updated when holding pgss->lock exclusively.
1913 : */
1914 0 : stats_since = entry->stats_since;
1915 0 : minmax_stats_since = entry->minmax_stats_since;
1916 :
1917 : /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
1918 0 : if (IS_STICKY(tmp))
1919 0 : continue;
1920 :
1921 : /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
1922 0 : for (int kind = 0; kind < PGSS_NUMKIND; kind++)
1923 : {
1924 0 : if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
1925 : {
1926 0 : values[i++] = Int64GetDatumFast(tmp.calls[kind]);
1927 0 : values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
1928 0 : }
1929 :
1930 0 : if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
1931 0 : api_version >= PGSS_V1_8)
1932 : {
1933 0 : values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
1934 0 : values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
1935 0 : values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
1936 :
1937 : /*
1938 : * Note we are calculating the population variance here, not
1939 : * the sample variance, as we have data for the whole
1940 : * population, so Bessel's correction is not used, and we
1941 : * don't divide by tmp.calls - 1.
1942 : */
1943 0 : if (tmp.calls[kind] > 1)
1944 0 : stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
1945 : else
1946 0 : stddev = 0.0;
1947 0 : values[i++] = Float8GetDatumFast(stddev);
1948 0 : }
1949 0 : }
1950 0 : values[i++] = Int64GetDatumFast(tmp.rows);
1951 0 : values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
1952 0 : values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
1953 0 : if (api_version >= PGSS_V1_1)
1954 0 : values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
1955 0 : values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
1956 0 : values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
1957 0 : values[i++] = Int64GetDatumFast(tmp.local_blks_read);
1958 0 : if (api_version >= PGSS_V1_1)
1959 0 : values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
1960 0 : values[i++] = Int64GetDatumFast(tmp.local_blks_written);
1961 0 : values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
1962 0 : values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
1963 0 : if (api_version >= PGSS_V1_1)
1964 : {
1965 0 : values[i++] = Float8GetDatumFast(tmp.shared_blk_read_time);
1966 0 : values[i++] = Float8GetDatumFast(tmp.shared_blk_write_time);
1967 0 : }
1968 0 : if (api_version >= PGSS_V1_11)
1969 : {
1970 0 : values[i++] = Float8GetDatumFast(tmp.local_blk_read_time);
1971 0 : values[i++] = Float8GetDatumFast(tmp.local_blk_write_time);
1972 0 : }
1973 0 : if (api_version >= PGSS_V1_10)
1974 : {
1975 0 : values[i++] = Float8GetDatumFast(tmp.temp_blk_read_time);
1976 0 : values[i++] = Float8GetDatumFast(tmp.temp_blk_write_time);
1977 0 : }
1978 0 : if (api_version >= PGSS_V1_8)
1979 : {
1980 0 : char buf[256];
1981 0 : Datum wal_bytes;
1982 :
1983 0 : values[i++] = Int64GetDatumFast(tmp.wal_records);
1984 0 : values[i++] = Int64GetDatumFast(tmp.wal_fpi);
1985 :
1986 0 : snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
1987 :
1988 : /* Convert to numeric. */
1989 0 : wal_bytes = DirectFunctionCall3(numeric_in,
1990 : CStringGetDatum(buf),
1991 : ObjectIdGetDatum(0),
1992 : Int32GetDatum(-1));
1993 0 : values[i++] = wal_bytes;
1994 0 : }
1995 0 : if (api_version >= PGSS_V1_12)
1996 : {
1997 0 : values[i++] = Int64GetDatumFast(tmp.wal_buffers_full);
1998 0 : }
1999 0 : if (api_version >= PGSS_V1_10)
2000 : {
2001 0 : values[i++] = Int64GetDatumFast(tmp.jit_functions);
2002 0 : values[i++] = Float8GetDatumFast(tmp.jit_generation_time);
2003 0 : values[i++] = Int64GetDatumFast(tmp.jit_inlining_count);
2004 0 : values[i++] = Float8GetDatumFast(tmp.jit_inlining_time);
2005 0 : values[i++] = Int64GetDatumFast(tmp.jit_optimization_count);
2006 0 : values[i++] = Float8GetDatumFast(tmp.jit_optimization_time);
2007 0 : values[i++] = Int64GetDatumFast(tmp.jit_emission_count);
2008 0 : values[i++] = Float8GetDatumFast(tmp.jit_emission_time);
2009 0 : }
2010 0 : if (api_version >= PGSS_V1_11)
2011 : {
2012 0 : values[i++] = Int64GetDatumFast(tmp.jit_deform_count);
2013 0 : values[i++] = Float8GetDatumFast(tmp.jit_deform_time);
2014 0 : }
2015 0 : if (api_version >= PGSS_V1_12)
2016 : {
2017 0 : values[i++] = Int64GetDatumFast(tmp.parallel_workers_to_launch);
2018 0 : values[i++] = Int64GetDatumFast(tmp.parallel_workers_launched);
2019 0 : }
2020 0 : if (api_version >= PGSS_V1_13)
2021 : {
2022 0 : values[i++] = Int64GetDatumFast(tmp.generic_plan_calls);
2023 0 : values[i++] = Int64GetDatumFast(tmp.custom_plan_calls);
2024 0 : }
2025 0 : if (api_version >= PGSS_V1_11)
2026 : {
2027 0 : values[i++] = TimestampTzGetDatum(stats_since);
2028 0 : values[i++] = TimestampTzGetDatum(minmax_stats_since);
2029 0 : }
2030 :
2031 0 : Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
2032 : api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
2033 : api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
2034 : api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
2035 : api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
2036 : api_version == PGSS_V1_9 ? PG_STAT_STATEMENTS_COLS_V1_9 :
2037 : api_version == PGSS_V1_10 ? PG_STAT_STATEMENTS_COLS_V1_10 :
2038 : api_version == PGSS_V1_11 ? PG_STAT_STATEMENTS_COLS_V1_11 :
2039 : api_version == PGSS_V1_12 ? PG_STAT_STATEMENTS_COLS_V1_12 :
2040 : api_version == PGSS_V1_13 ? PG_STAT_STATEMENTS_COLS_V1_13 :
2041 : -1 /* fail if you forget to update this assert */ ));
2042 :
2043 0 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
2044 0 : }
2045 :
2046 0 : LWLockRelease(pgss->lock);
2047 :
2048 0 : free(qbuffer);
2049 0 : }
2050 :
2051 : /* Number of output arguments (columns) for pg_stat_statements_info */
2052 : #define PG_STAT_STATEMENTS_INFO_COLS 2
2053 :
2054 : /*
2055 : * Return statistics of pg_stat_statements.
2056 : */
2057 : Datum
2058 0 : pg_stat_statements_info(PG_FUNCTION_ARGS)
2059 : {
2060 0 : pgssGlobalStats stats;
2061 0 : TupleDesc tupdesc;
2062 0 : Datum values[PG_STAT_STATEMENTS_INFO_COLS] = {0};
2063 0 : bool nulls[PG_STAT_STATEMENTS_INFO_COLS] = {0};
2064 :
2065 0 : if (!pgss || !pgss_hash)
2066 0 : ereport(ERROR,
2067 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2068 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
2069 :
2070 : /* Build a tuple descriptor for our result type */
2071 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
2072 0 : elog(ERROR, "return type must be a row type");
2073 :
2074 : /* Read global statistics for pg_stat_statements */
2075 0 : SpinLockAcquire(&pgss->mutex);
2076 0 : stats = pgss->stats;
2077 0 : SpinLockRelease(&pgss->mutex);
2078 :
2079 0 : values[0] = Int64GetDatum(stats.dealloc);
2080 0 : values[1] = TimestampTzGetDatum(stats.stats_reset);
2081 :
2082 0 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
2083 0 : }
2084 :
2085 : /*
2086 : * Estimate shared memory space needed.
2087 : */
2088 : static Size
2089 0 : pgss_memsize(void)
2090 : {
2091 0 : Size size;
2092 :
2093 0 : size = MAXALIGN(sizeof(pgssSharedState));
2094 0 : size = add_size(size, hash_estimate_size(pgss_max, sizeof(pgssEntry)));
2095 :
2096 0 : return size;
2097 0 : }
2098 :
2099 : /*
2100 : * Allocate a new hashtable entry.
2101 : * caller must hold an exclusive lock on pgss->lock
2102 : *
2103 : * "query" need not be null-terminated; we rely on query_len instead
2104 : *
2105 : * If "sticky" is true, make the new entry artificially sticky so that it will
2106 : * probably still be there when the query finishes execution. We do this by
2107 : * giving it a median usage value rather than the normal value. (Strictly
2108 : * speaking, query strings are normalized on a best effort basis, though it
2109 : * would be difficult to demonstrate this even under artificial conditions.)
2110 : *
2111 : * Note: despite needing exclusive lock, it's not an error for the target
2112 : * entry to already exist. This is because pgss_store releases and
2113 : * reacquires lock after failing to find a match; so someone else could
2114 : * have made the entry while we waited to get exclusive lock.
2115 : */
2116 : static pgssEntry *
2117 0 : entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
2118 : bool sticky)
2119 : {
2120 0 : pgssEntry *entry;
2121 0 : bool found;
2122 :
2123 : /* Make space if needed */
2124 0 : while (hash_get_num_entries(pgss_hash) >= pgss_max)
2125 0 : entry_dealloc();
2126 :
2127 : /* Find or create an entry with desired hash code */
2128 0 : entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
2129 :
2130 0 : if (!found)
2131 : {
2132 : /* New entry, initialize it */
2133 :
2134 : /* reset the statistics */
2135 0 : memset(&entry->counters, 0, sizeof(Counters));
2136 : /* set the appropriate initial usage count */
2137 0 : entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
2138 : /* re-initialize the mutex each time ... we assume no one using it */
2139 0 : SpinLockInit(&entry->mutex);
2140 : /* ... and don't forget the query text metadata */
2141 0 : Assert(query_len >= 0);
2142 0 : entry->query_offset = query_offset;
2143 0 : entry->query_len = query_len;
2144 0 : entry->encoding = encoding;
2145 0 : entry->stats_since = GetCurrentTimestamp();
2146 0 : entry->minmax_stats_since = entry->stats_since;
2147 0 : }
2148 :
2149 0 : return entry;
2150 0 : }
2151 :
2152 : /*
2153 : * qsort comparator for sorting into increasing usage order
2154 : */
2155 : static int
2156 0 : entry_cmp(const void *lhs, const void *rhs)
2157 : {
2158 0 : double l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
2159 0 : double r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
2160 :
2161 0 : if (l_usage < r_usage)
2162 0 : return -1;
2163 0 : else if (l_usage > r_usage)
2164 0 : return +1;
2165 : else
2166 0 : return 0;
2167 0 : }
2168 :
2169 : /*
2170 : * Deallocate least-used entries.
2171 : *
2172 : * Caller must hold an exclusive lock on pgss->lock.
2173 : */
2174 : static void
2175 0 : entry_dealloc(void)
2176 : {
2177 0 : HASH_SEQ_STATUS hash_seq;
2178 0 : pgssEntry **entries;
2179 0 : pgssEntry *entry;
2180 0 : int nvictims;
2181 0 : int i;
2182 0 : Size tottextlen;
2183 0 : int nvalidtexts;
2184 :
2185 : /*
2186 : * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
2187 : * While we're scanning the table, apply the decay factor to the usage
2188 : * values, and update the mean query length.
2189 : *
2190 : * Note that the mean query length is almost immediately obsolete, since
2191 : * we compute it before not after discarding the least-used entries.
2192 : * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
2193 : * making two passes to get a more current result. Likewise, the new
2194 : * cur_median_usage includes the entries we're about to zap.
2195 : */
2196 :
2197 0 : entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
2198 :
2199 0 : i = 0;
2200 0 : tottextlen = 0;
2201 0 : nvalidtexts = 0;
2202 :
2203 0 : hash_seq_init(&hash_seq, pgss_hash);
2204 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2205 : {
2206 0 : entries[i++] = entry;
2207 : /* "Sticky" entries get a different usage decay rate. */
2208 0 : if (IS_STICKY(entry->counters))
2209 0 : entry->counters.usage *= STICKY_DECREASE_FACTOR;
2210 : else
2211 0 : entry->counters.usage *= USAGE_DECREASE_FACTOR;
2212 : /* In the mean length computation, ignore dropped texts. */
2213 0 : if (entry->query_len >= 0)
2214 : {
2215 0 : tottextlen += entry->query_len + 1;
2216 0 : nvalidtexts++;
2217 0 : }
2218 : }
2219 :
2220 : /* Sort into increasing order by usage */
2221 0 : qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
2222 :
2223 : /* Record the (approximate) median usage */
2224 0 : if (i > 0)
2225 0 : pgss->cur_median_usage = entries[i / 2]->counters.usage;
2226 : /* Record the mean query length */
2227 0 : if (nvalidtexts > 0)
2228 0 : pgss->mean_query_len = tottextlen / nvalidtexts;
2229 : else
2230 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2231 :
2232 : /* Now zap an appropriate fraction of lowest-usage entries */
2233 0 : nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
2234 0 : nvictims = Min(nvictims, i);
2235 :
2236 0 : for (i = 0; i < nvictims; i++)
2237 : {
2238 0 : hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
2239 0 : }
2240 :
2241 0 : pfree(entries);
2242 :
2243 : /* Increment the number of times entries are deallocated */
2244 0 : SpinLockAcquire(&pgss->mutex);
2245 0 : pgss->stats.dealloc += 1;
2246 0 : SpinLockRelease(&pgss->mutex);
2247 0 : }
2248 :
2249 : /*
2250 : * Given a query string (not necessarily null-terminated), allocate a new
2251 : * entry in the external query text file and store the string there.
2252 : *
2253 : * If successful, returns true, and stores the new entry's offset in the file
2254 : * into *query_offset. Also, if gc_count isn't NULL, *gc_count is set to the
2255 : * number of garbage collections that have occurred so far.
2256 : *
2257 : * On failure, returns false.
2258 : *
2259 : * At least a shared lock on pgss->lock must be held by the caller, so as
2260 : * to prevent a concurrent garbage collection. Share-lock-holding callers
2261 : * should pass a gc_count pointer to obtain the number of garbage collections,
2262 : * so that they can recheck the count after obtaining exclusive lock to
2263 : * detect whether a garbage collection occurred (and removed this entry).
2264 : */
2265 : static bool
2266 0 : qtext_store(const char *query, int query_len,
2267 : Size *query_offset, int *gc_count)
2268 : {
2269 0 : Size off;
2270 0 : int fd;
2271 :
2272 : /*
2273 : * We use a spinlock to protect extent/n_writers/gc_count, so that
2274 : * multiple processes may execute this function concurrently.
2275 : */
2276 0 : SpinLockAcquire(&pgss->mutex);
2277 0 : off = pgss->extent;
2278 0 : pgss->extent += query_len + 1;
2279 0 : pgss->n_writers++;
2280 0 : if (gc_count)
2281 0 : *gc_count = pgss->gc_count;
2282 0 : SpinLockRelease(&pgss->mutex);
2283 :
2284 0 : *query_offset = off;
2285 :
2286 : /*
2287 : * Don't allow the file to grow larger than what qtext_load_file can
2288 : * (theoretically) handle. This has been seen to be reachable on 32-bit
2289 : * platforms.
2290 : */
2291 0 : if (unlikely(query_len >= MaxAllocHugeSize - off))
2292 : {
2293 0 : errno = EFBIG; /* not quite right, but it'll do */
2294 0 : fd = -1;
2295 0 : goto error;
2296 : }
2297 :
2298 : /* Now write the data into the successfully-reserved part of the file */
2299 0 : fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
2300 0 : if (fd < 0)
2301 0 : goto error;
2302 :
2303 0 : if (pg_pwrite(fd, query, query_len, off) != query_len)
2304 0 : goto error;
2305 0 : if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
2306 0 : goto error;
2307 :
2308 0 : CloseTransientFile(fd);
2309 :
2310 : /* Mark our write complete */
2311 0 : SpinLockAcquire(&pgss->mutex);
2312 0 : pgss->n_writers--;
2313 0 : SpinLockRelease(&pgss->mutex);
2314 :
2315 0 : return true;
2316 :
2317 : error:
2318 0 : ereport(LOG,
2319 : (errcode_for_file_access(),
2320 : errmsg("could not write file \"%s\": %m",
2321 : PGSS_TEXT_FILE)));
2322 :
2323 0 : if (fd >= 0)
2324 0 : CloseTransientFile(fd);
2325 :
2326 : /* Mark our write complete */
2327 0 : SpinLockAcquire(&pgss->mutex);
2328 0 : pgss->n_writers--;
2329 0 : SpinLockRelease(&pgss->mutex);
2330 :
2331 0 : return false;
2332 0 : }
2333 :
2334 : /*
2335 : * Read the external query text file into a malloc'd buffer.
2336 : *
2337 : * Returns NULL (without throwing an error) if unable to read, eg
2338 : * file not there or insufficient memory.
2339 : *
2340 : * On success, the buffer size is also returned into *buffer_size.
2341 : *
2342 : * This can be called without any lock on pgss->lock, but in that case
2343 : * the caller is responsible for verifying that the result is sane.
2344 : */
2345 : static char *
2346 0 : qtext_load_file(Size *buffer_size)
2347 : {
2348 0 : char *buf;
2349 0 : int fd;
2350 0 : struct stat stat;
2351 0 : Size nread;
2352 :
2353 0 : fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
2354 0 : if (fd < 0)
2355 : {
2356 0 : if (errno != ENOENT)
2357 0 : ereport(LOG,
2358 : (errcode_for_file_access(),
2359 : errmsg("could not read file \"%s\": %m",
2360 : PGSS_TEXT_FILE)));
2361 0 : return NULL;
2362 : }
2363 :
2364 : /* Get file length */
2365 0 : if (fstat(fd, &stat))
2366 : {
2367 0 : ereport(LOG,
2368 : (errcode_for_file_access(),
2369 : errmsg("could not stat file \"%s\": %m",
2370 : PGSS_TEXT_FILE)));
2371 0 : CloseTransientFile(fd);
2372 0 : return NULL;
2373 : }
2374 :
2375 : /* Allocate buffer; beware that off_t might be wider than size_t */
2376 0 : if (stat.st_size <= MaxAllocHugeSize)
2377 0 : buf = (char *) malloc(stat.st_size);
2378 : else
2379 0 : buf = NULL;
2380 0 : if (buf == NULL)
2381 : {
2382 0 : ereport(LOG,
2383 : (errcode(ERRCODE_OUT_OF_MEMORY),
2384 : errmsg("out of memory"),
2385 : errdetail("Could not allocate enough memory to read file \"%s\".",
2386 : PGSS_TEXT_FILE)));
2387 0 : CloseTransientFile(fd);
2388 0 : return NULL;
2389 : }
2390 :
2391 : /*
2392 : * OK, slurp in the file. Windows fails if we try to read more than
2393 : * INT_MAX bytes at once, and other platforms might not like that either,
2394 : * so read a very large file in 1GB segments.
2395 : */
2396 0 : nread = 0;
2397 0 : while (nread < stat.st_size)
2398 : {
2399 0 : int toread = Min(1024 * 1024 * 1024, stat.st_size - nread);
2400 :
2401 : /*
2402 : * If we get a short read and errno doesn't get set, the reason is
2403 : * probably that garbage collection truncated the file since we did
2404 : * the fstat(), so we don't log a complaint --- but we don't return
2405 : * the data, either, since it's most likely corrupt due to concurrent
2406 : * writes from garbage collection.
2407 : */
2408 0 : errno = 0;
2409 0 : if (read(fd, buf + nread, toread) != toread)
2410 : {
2411 0 : if (errno)
2412 0 : ereport(LOG,
2413 : (errcode_for_file_access(),
2414 : errmsg("could not read file \"%s\": %m",
2415 : PGSS_TEXT_FILE)));
2416 0 : free(buf);
2417 0 : CloseTransientFile(fd);
2418 0 : return NULL;
2419 : }
2420 0 : nread += toread;
2421 0 : }
2422 :
2423 0 : if (CloseTransientFile(fd) != 0)
2424 0 : ereport(LOG,
2425 : (errcode_for_file_access(),
2426 : errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
2427 :
2428 0 : *buffer_size = nread;
2429 0 : return buf;
2430 0 : }
2431 :
2432 : /*
2433 : * Locate a query text in the file image previously read by qtext_load_file().
2434 : *
2435 : * We validate the given offset/length, and return NULL if bogus. Otherwise,
2436 : * the result points to a null-terminated string within the buffer.
2437 : */
2438 : static char *
2439 0 : qtext_fetch(Size query_offset, int query_len,
2440 : char *buffer, Size buffer_size)
2441 : {
2442 : /* File read failed? */
2443 0 : if (buffer == NULL)
2444 0 : return NULL;
2445 : /* Bogus offset/length? */
2446 0 : if (query_len < 0 ||
2447 0 : query_offset + query_len >= buffer_size)
2448 0 : return NULL;
2449 : /* As a further sanity check, make sure there's a trailing null */
2450 0 : if (buffer[query_offset + query_len] != '\0')
2451 0 : return NULL;
2452 : /* Looks OK */
2453 0 : return buffer + query_offset;
2454 0 : }
2455 :
2456 : /*
2457 : * Do we need to garbage-collect the external query text file?
2458 : *
2459 : * Caller should hold at least a shared lock on pgss->lock.
2460 : */
2461 : static bool
2462 0 : need_gc_qtexts(void)
2463 : {
2464 0 : Size extent;
2465 :
2466 : /* Read shared extent pointer */
2467 0 : SpinLockAcquire(&pgss->mutex);
2468 0 : extent = pgss->extent;
2469 0 : SpinLockRelease(&pgss->mutex);
2470 :
2471 : /*
2472 : * Don't proceed if file does not exceed 512 bytes per possible entry.
2473 : *
2474 : * Here and in the next test, 32-bit machines have overflow hazards if
2475 : * pgss_max and/or mean_query_len are large. Force the multiplications
2476 : * and comparisons to be done in uint64 arithmetic to forestall trouble.
2477 : */
2478 0 : if ((uint64) extent < (uint64) 512 * pgss_max)
2479 0 : return false;
2480 :
2481 : /*
2482 : * Don't proceed if file is less than about 50% bloat. Nothing can or
2483 : * should be done in the event of unusually large query texts accounting
2484 : * for file's large size. We go to the trouble of maintaining the mean
2485 : * query length in order to prevent garbage collection from thrashing
2486 : * uselessly.
2487 : */
2488 0 : if ((uint64) extent < (uint64) pgss->mean_query_len * pgss_max * 2)
2489 0 : return false;
2490 :
2491 0 : return true;
2492 0 : }
2493 :
2494 : /*
2495 : * Garbage-collect orphaned query texts in external file.
2496 : *
2497 : * This won't be called often in the typical case, since it's likely that
2498 : * there won't be too much churn, and besides, a similar compaction process
2499 : * occurs when serializing to disk at shutdown or as part of resetting.
2500 : * Despite this, it seems prudent to plan for the edge case where the file
2501 : * becomes unreasonably large, with no other method of compaction likely to
2502 : * occur in the foreseeable future.
2503 : *
2504 : * The caller must hold an exclusive lock on pgss->lock.
2505 : *
2506 : * At the first sign of trouble we unlink the query text file to get a clean
2507 : * slate (although existing statistics are retained), rather than risk
2508 : * thrashing by allowing the same problem case to recur indefinitely.
2509 : */
2510 : static void
2511 0 : gc_qtexts(void)
2512 : {
2513 0 : char *qbuffer;
2514 0 : Size qbuffer_size;
2515 0 : FILE *qfile = NULL;
2516 0 : HASH_SEQ_STATUS hash_seq;
2517 0 : pgssEntry *entry;
2518 0 : Size extent;
2519 0 : int nentries;
2520 :
2521 : /*
2522 : * When called from pgss_store, some other session might have proceeded
2523 : * with garbage collection in the no-lock-held interim of lock strength
2524 : * escalation. Check once more that this is actually necessary.
2525 : */
2526 0 : if (!need_gc_qtexts())
2527 0 : return;
2528 :
2529 : /*
2530 : * Load the old texts file. If we fail (out of memory, for instance),
2531 : * invalidate query texts. Hopefully this is rare. It might seem better
2532 : * to leave things alone on an OOM failure, but the problem is that the
2533 : * file is only going to get bigger; hoping for a future non-OOM result is
2534 : * risky and can easily lead to complete denial of service.
2535 : */
2536 0 : qbuffer = qtext_load_file(&qbuffer_size);
2537 0 : if (qbuffer == NULL)
2538 0 : goto gc_fail;
2539 :
2540 : /*
2541 : * We overwrite the query texts file in place, so as to reduce the risk of
2542 : * an out-of-disk-space failure. Since the file is guaranteed not to get
2543 : * larger, this should always work on traditional filesystems; though we
2544 : * could still lose on copy-on-write filesystems.
2545 : */
2546 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2547 0 : if (qfile == NULL)
2548 : {
2549 0 : ereport(LOG,
2550 : (errcode_for_file_access(),
2551 : errmsg("could not write file \"%s\": %m",
2552 : PGSS_TEXT_FILE)));
2553 0 : goto gc_fail;
2554 : }
2555 :
2556 0 : extent = 0;
2557 0 : nentries = 0;
2558 :
2559 0 : hash_seq_init(&hash_seq, pgss_hash);
2560 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2561 : {
2562 0 : int query_len = entry->query_len;
2563 0 : char *qry = qtext_fetch(entry->query_offset,
2564 0 : query_len,
2565 0 : qbuffer,
2566 0 : qbuffer_size);
2567 :
2568 0 : if (qry == NULL)
2569 : {
2570 : /* Trouble ... drop the text */
2571 0 : entry->query_offset = 0;
2572 0 : entry->query_len = -1;
2573 : /* entry will not be counted in mean query length computation */
2574 0 : continue;
2575 : }
2576 :
2577 0 : if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
2578 : {
2579 0 : ereport(LOG,
2580 : (errcode_for_file_access(),
2581 : errmsg("could not write file \"%s\": %m",
2582 : PGSS_TEXT_FILE)));
2583 0 : hash_seq_term(&hash_seq);
2584 0 : goto gc_fail;
2585 : }
2586 :
2587 0 : entry->query_offset = extent;
2588 0 : extent += query_len + 1;
2589 0 : nentries++;
2590 0 : }
2591 :
2592 : /*
2593 : * Truncate away any now-unused space. If this fails for some odd reason,
2594 : * we log it, but there's no need to fail.
2595 : */
2596 0 : if (ftruncate(fileno(qfile), extent) != 0)
2597 0 : ereport(LOG,
2598 : (errcode_for_file_access(),
2599 : errmsg("could not truncate file \"%s\": %m",
2600 : PGSS_TEXT_FILE)));
2601 :
2602 0 : if (FreeFile(qfile))
2603 : {
2604 0 : ereport(LOG,
2605 : (errcode_for_file_access(),
2606 : errmsg("could not write file \"%s\": %m",
2607 : PGSS_TEXT_FILE)));
2608 0 : qfile = NULL;
2609 0 : goto gc_fail;
2610 : }
2611 :
2612 0 : elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
2613 : pgss->extent, extent);
2614 :
2615 : /* Reset the shared extent pointer */
2616 0 : pgss->extent = extent;
2617 :
2618 : /*
2619 : * Also update the mean query length, to be sure that need_gc_qtexts()
2620 : * won't still think we have a problem.
2621 : */
2622 0 : if (nentries > 0)
2623 0 : pgss->mean_query_len = extent / nentries;
2624 : else
2625 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2626 :
2627 0 : free(qbuffer);
2628 :
2629 : /*
2630 : * OK, count a garbage collection cycle. (Note: even though we have
2631 : * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
2632 : * other processes may examine gc_count while holding only the mutex.
2633 : * Also, we have to advance the count *after* we've rewritten the file,
2634 : * else other processes might not realize they read a stale file.)
2635 : */
2636 0 : record_gc_qtexts();
2637 :
2638 0 : return;
2639 :
2640 : gc_fail:
2641 : /* clean up resources */
2642 0 : if (qfile)
2643 0 : FreeFile(qfile);
2644 0 : free(qbuffer);
2645 :
2646 : /*
2647 : * Since the contents of the external file are now uncertain, mark all
2648 : * hashtable entries as having invalid texts.
2649 : */
2650 0 : hash_seq_init(&hash_seq, pgss_hash);
2651 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2652 : {
2653 0 : entry->query_offset = 0;
2654 0 : entry->query_len = -1;
2655 : }
2656 :
2657 : /*
2658 : * Destroy the query text file and create a new, empty one
2659 : */
2660 0 : (void) unlink(PGSS_TEXT_FILE);
2661 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2662 0 : if (qfile == NULL)
2663 0 : ereport(LOG,
2664 : (errcode_for_file_access(),
2665 : errmsg("could not recreate file \"%s\": %m",
2666 : PGSS_TEXT_FILE)));
2667 : else
2668 0 : FreeFile(qfile);
2669 :
2670 : /* Reset the shared extent pointer */
2671 0 : pgss->extent = 0;
2672 :
2673 : /* Reset mean_query_len to match the new state */
2674 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2675 :
2676 : /*
2677 : * Bump the GC count even though we failed.
2678 : *
2679 : * This is needed to make concurrent readers of file without any lock on
2680 : * pgss->lock notice existence of new version of file. Once readers
2681 : * subsequently observe a change in GC count with pgss->lock held, that
2682 : * forces a safe reopen of file. Writers also require that we bump here,
2683 : * of course. (As required by locking protocol, readers and writers don't
2684 : * trust earlier file contents until gc_count is found unchanged after
2685 : * pgss->lock acquired in shared or exclusive mode respectively.)
2686 : */
2687 0 : record_gc_qtexts();
2688 0 : }
2689 :
2690 : #define SINGLE_ENTRY_RESET(e) \
2691 : if (e) { \
2692 : if (minmax_only) { \
2693 : /* When requested reset only min/max statistics of an entry */ \
2694 : for (int kind = 0; kind < PGSS_NUMKIND; kind++) \
2695 : { \
2696 : e->counters.max_time[kind] = 0; \
2697 : e->counters.min_time[kind] = 0; \
2698 : } \
2699 : e->minmax_stats_since = stats_reset; \
2700 : } \
2701 : else \
2702 : { \
2703 : /* Remove the key otherwise */ \
2704 : hash_search(pgss_hash, &e->key, HASH_REMOVE, NULL); \
2705 : num_remove++; \
2706 : } \
2707 : }
2708 :
2709 : /*
2710 : * Reset entries corresponding to parameters passed.
2711 : */
2712 : static TimestampTz
2713 0 : entry_reset(Oid userid, Oid dbid, int64 queryid, bool minmax_only)
2714 : {
2715 0 : HASH_SEQ_STATUS hash_seq;
2716 0 : pgssEntry *entry;
2717 0 : FILE *qfile;
2718 0 : int64 num_entries;
2719 0 : int64 num_remove = 0;
2720 0 : pgssHashKey key;
2721 0 : TimestampTz stats_reset;
2722 :
2723 0 : if (!pgss || !pgss_hash)
2724 0 : ereport(ERROR,
2725 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2726 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
2727 :
2728 0 : LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
2729 0 : num_entries = hash_get_num_entries(pgss_hash);
2730 :
2731 0 : stats_reset = GetCurrentTimestamp();
2732 :
2733 0 : if (userid != 0 && dbid != 0 && queryid != INT64CONST(0))
2734 : {
2735 : /* If all the parameters are available, use the fast path. */
2736 0 : memset(&key, 0, sizeof(pgssHashKey));
2737 0 : key.userid = userid;
2738 0 : key.dbid = dbid;
2739 0 : key.queryid = queryid;
2740 :
2741 : /*
2742 : * Reset the entry if it exists, starting with the non-top-level
2743 : * entry.
2744 : */
2745 0 : key.toplevel = false;
2746 0 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
2747 :
2748 0 : SINGLE_ENTRY_RESET(entry);
2749 :
2750 : /* Also reset the top-level entry if it exists. */
2751 0 : key.toplevel = true;
2752 0 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
2753 :
2754 0 : SINGLE_ENTRY_RESET(entry);
2755 0 : }
2756 0 : else if (userid != 0 || dbid != 0 || queryid != INT64CONST(0))
2757 : {
2758 : /* Reset entries corresponding to valid parameters. */
2759 0 : hash_seq_init(&hash_seq, pgss_hash);
2760 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2761 : {
2762 0 : if ((!userid || entry->key.userid == userid) &&
2763 0 : (!dbid || entry->key.dbid == dbid) &&
2764 0 : (!queryid || entry->key.queryid == queryid))
2765 : {
2766 0 : SINGLE_ENTRY_RESET(entry);
2767 0 : }
2768 : }
2769 0 : }
2770 : else
2771 : {
2772 : /* Reset all entries. */
2773 0 : hash_seq_init(&hash_seq, pgss_hash);
2774 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2775 : {
2776 0 : SINGLE_ENTRY_RESET(entry);
2777 : }
2778 : }
2779 :
2780 : /* All entries are removed? */
2781 0 : if (num_entries != num_remove)
2782 0 : goto release_lock;
2783 :
2784 : /*
2785 : * Reset global statistics for pg_stat_statements since all entries are
2786 : * removed.
2787 : */
2788 0 : SpinLockAcquire(&pgss->mutex);
2789 0 : pgss->stats.dealloc = 0;
2790 0 : pgss->stats.stats_reset = stats_reset;
2791 0 : SpinLockRelease(&pgss->mutex);
2792 :
2793 : /*
2794 : * Write new empty query file, perhaps even creating a new one to recover
2795 : * if the file was missing.
2796 : */
2797 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2798 0 : if (qfile == NULL)
2799 : {
2800 0 : ereport(LOG,
2801 : (errcode_for_file_access(),
2802 : errmsg("could not create file \"%s\": %m",
2803 : PGSS_TEXT_FILE)));
2804 0 : goto done;
2805 : }
2806 :
2807 : /* If ftruncate fails, log it, but it's not a fatal problem */
2808 0 : if (ftruncate(fileno(qfile), 0) != 0)
2809 0 : ereport(LOG,
2810 : (errcode_for_file_access(),
2811 : errmsg("could not truncate file \"%s\": %m",
2812 : PGSS_TEXT_FILE)));
2813 :
2814 0 : FreeFile(qfile);
2815 :
2816 : done:
2817 0 : pgss->extent = 0;
2818 : /* This counts as a query text garbage collection for our purposes */
2819 0 : record_gc_qtexts();
2820 :
2821 : release_lock:
2822 0 : LWLockRelease(pgss->lock);
2823 :
2824 0 : return stats_reset;
2825 0 : }
2826 :
2827 : /*
2828 : * Generate a normalized version of the query string that will be used to
2829 : * represent all similar queries.
2830 : *
2831 : * Note that the normalized representation may well vary depending on
2832 : * just which "equivalent" query is used to create the hashtable entry.
2833 : * We assume this is OK.
2834 : *
2835 : * If query_loc > 0, then "query" has been advanced by that much compared to
2836 : * the original string start, so we need to translate the provided locations
2837 : * to compensate. (This lets us avoid re-scanning statements before the one
2838 : * of interest, so it's worth doing.)
2839 : *
2840 : * *query_len_p contains the input string length, and is updated with
2841 : * the result string length on exit. The resulting string might be longer
2842 : * or shorter depending on what happens with replacement of constants.
2843 : *
2844 : * Returns a palloc'd string.
2845 : */
2846 : static char *
2847 0 : generate_normalized_query(JumbleState *jstate, const char *query,
2848 : int query_loc, int *query_len_p)
2849 : {
2850 0 : char *norm_query;
2851 0 : int query_len = *query_len_p;
2852 0 : int norm_query_buflen, /* Space allowed for norm_query */
2853 : len_to_wrt, /* Length (in bytes) to write */
2854 0 : quer_loc = 0, /* Source query byte location */
2855 0 : n_quer_loc = 0, /* Normalized query byte location */
2856 0 : last_off = 0, /* Offset from start for previous tok */
2857 0 : last_tok_len = 0; /* Length (in bytes) of that tok */
2858 0 : int num_constants_replaced = 0;
2859 :
2860 : /*
2861 : * Get constants' lengths (core system only gives us locations). Note
2862 : * this also ensures the items are sorted by location.
2863 : */
2864 0 : fill_in_constant_lengths(jstate, query, query_loc);
2865 :
2866 : /*
2867 : * Allow for $n symbols to be longer than the constants they replace.
2868 : * Constants must take at least one byte in text form, while a $n symbol
2869 : * certainly isn't more than 11 bytes, even if n reaches INT_MAX. We
2870 : * could refine that limit based on the max value of n for the current
2871 : * query, but it hardly seems worth any extra effort to do so.
2872 : */
2873 0 : norm_query_buflen = query_len + jstate->clocations_count * 10;
2874 :
2875 : /* Allocate result buffer */
2876 0 : norm_query = palloc(norm_query_buflen + 1);
2877 :
2878 0 : for (int i = 0; i < jstate->clocations_count; i++)
2879 : {
2880 0 : int off, /* Offset from start for cur tok */
2881 : tok_len; /* Length (in bytes) of that tok */
2882 :
2883 : /*
2884 : * If we have an external param at this location, but no lists are
2885 : * being squashed across the query, then we skip here; this will make
2886 : * us print the characters found in the original query that represent
2887 : * the parameter in the next iteration (or after the loop is done),
2888 : * which is a bit odd but seems to work okay in most cases.
2889 : */
2890 0 : if (jstate->clocations[i].extern_param && !jstate->has_squashed_lists)
2891 0 : continue;
2892 :
2893 0 : off = jstate->clocations[i].location;
2894 :
2895 : /* Adjust recorded location if we're dealing with partial string */
2896 0 : off -= query_loc;
2897 :
2898 0 : tok_len = jstate->clocations[i].length;
2899 :
2900 0 : if (tok_len < 0)
2901 0 : continue; /* ignore any duplicates */
2902 :
2903 : /* Copy next chunk (what precedes the next constant) */
2904 0 : len_to_wrt = off - last_off;
2905 0 : len_to_wrt -= last_tok_len;
2906 0 : Assert(len_to_wrt >= 0);
2907 0 : memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
2908 0 : n_quer_loc += len_to_wrt;
2909 :
2910 : /*
2911 : * And insert a param symbol in place of the constant token; and, if
2912 : * we have a squashable list, insert a placeholder comment starting
2913 : * from the list's second value.
2914 : */
2915 0 : n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d%s",
2916 0 : num_constants_replaced + 1 + jstate->highest_extern_param_id,
2917 0 : jstate->clocations[i].squashed ? " /*, ... */" : "");
2918 0 : num_constants_replaced++;
2919 :
2920 : /* move forward */
2921 0 : quer_loc = off + tok_len;
2922 0 : last_off = off;
2923 0 : last_tok_len = tok_len;
2924 0 : }
2925 :
2926 : /*
2927 : * We've copied up until the last ignorable constant. Copy over the
2928 : * remaining bytes of the original query string.
2929 : */
2930 0 : len_to_wrt = query_len - quer_loc;
2931 :
2932 0 : Assert(len_to_wrt >= 0);
2933 0 : memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
2934 0 : n_quer_loc += len_to_wrt;
2935 :
2936 0 : Assert(n_quer_loc <= norm_query_buflen);
2937 0 : norm_query[n_quer_loc] = '\0';
2938 :
2939 0 : *query_len_p = n_quer_loc;
2940 0 : return norm_query;
2941 0 : }
2942 :
2943 : /*
2944 : * Given a valid SQL string and an array of constant-location records,
2945 : * fill in the textual lengths of those constants.
2946 : *
2947 : * The constants may use any allowed constant syntax, such as float literals,
2948 : * bit-strings, single-quoted strings and dollar-quoted strings. This is
2949 : * accomplished by using the public API for the core scanner.
2950 : *
2951 : * It is the caller's job to ensure that the string is a valid SQL statement
2952 : * with constants at the indicated locations. Since in practice the string
2953 : * has already been parsed, and the locations that the caller provides will
2954 : * have originated from within the authoritative parser, this should not be
2955 : * a problem.
2956 : *
2957 : * Multiple constants can have the same location. We reset lengths of those
2958 : * past the first to -1 so that they can later be ignored.
2959 : *
2960 : * If query_loc > 0, then "query" has been advanced by that much compared to
2961 : * the original string start, so we need to translate the provided locations
2962 : * to compensate. (This lets us avoid re-scanning statements before the one
2963 : * of interest, so it's worth doing.)
2964 : *
2965 : * N.B. There is an assumption that a '-' character at a Const location begins
2966 : * a negative numeric constant. This precludes there ever being another
2967 : * reason for a constant to start with a '-'.
2968 : */
2969 : static void
2970 0 : fill_in_constant_lengths(JumbleState *jstate, const char *query,
2971 : int query_loc)
2972 : {
2973 0 : LocationLen *locs;
2974 0 : core_yyscan_t yyscanner;
2975 0 : core_yy_extra_type yyextra;
2976 0 : core_YYSTYPE yylval;
2977 0 : YYLTYPE yylloc;
2978 :
2979 : /*
2980 : * Sort the records by location so that we can process them in order while
2981 : * scanning the query text.
2982 : */
2983 0 : if (jstate->clocations_count > 1)
2984 0 : qsort(jstate->clocations, jstate->clocations_count,
2985 : sizeof(LocationLen), comp_location);
2986 0 : locs = jstate->clocations;
2987 :
2988 : /* initialize the flex scanner --- should match raw_parser() */
2989 0 : yyscanner = scanner_init(query,
2990 : &yyextra,
2991 : &ScanKeywords,
2992 : ScanKeywordTokens);
2993 :
2994 : /* Search for each constant, in sequence */
2995 0 : for (int i = 0; i < jstate->clocations_count; i++)
2996 : {
2997 0 : int loc;
2998 0 : int tok;
2999 :
3000 : /* Ignore constants after the first one in the same location */
3001 0 : if (i > 0 && locs[i].location == locs[i - 1].location)
3002 : {
3003 0 : locs[i].length = -1;
3004 0 : continue;
3005 : }
3006 :
3007 0 : if (locs[i].squashed)
3008 0 : continue; /* squashable list, ignore */
3009 :
3010 : /* Adjust recorded location if we're dealing with partial string */
3011 0 : loc = locs[i].location - query_loc;
3012 0 : Assert(loc >= 0);
3013 :
3014 : /*
3015 : * We have a valid location for a constant that's not a dupe. Lex
3016 : * tokens until we find the desired constant.
3017 : */
3018 0 : for (;;)
3019 : {
3020 0 : tok = core_yylex(&yylval, &yylloc, yyscanner);
3021 :
3022 : /* We should not hit end-of-string, but if we do, behave sanely */
3023 0 : if (tok == 0)
3024 0 : break; /* out of inner for-loop */
3025 :
3026 : /*
3027 : * We should find the token position exactly, but if we somehow
3028 : * run past it, work with that.
3029 : */
3030 0 : if (yylloc >= loc)
3031 : {
3032 0 : if (query[loc] == '-')
3033 : {
3034 : /*
3035 : * It's a negative value - this is the one and only case
3036 : * where we replace more than a single token.
3037 : *
3038 : * Do not compensate for the core system's special-case
3039 : * adjustment of location to that of the leading '-'
3040 : * operator in the event of a negative constant. It is
3041 : * also useful for our purposes to start from the minus
3042 : * symbol. In this way, queries like "select * from foo
3043 : * where bar = 1" and "select * from foo where bar = -2"
3044 : * will have identical normalized query strings.
3045 : */
3046 0 : tok = core_yylex(&yylval, &yylloc, yyscanner);
3047 0 : if (tok == 0)
3048 0 : break; /* out of inner for-loop */
3049 0 : }
3050 :
3051 : /*
3052 : * We now rely on the assumption that flex has placed a zero
3053 : * byte after the text of the current token in scanbuf.
3054 : */
3055 0 : locs[i].length = strlen(yyextra.scanbuf + loc);
3056 0 : break; /* out of inner for-loop */
3057 : }
3058 : }
3059 :
3060 : /* If we hit end-of-string, give up, leaving remaining lengths -1 */
3061 0 : if (tok == 0)
3062 0 : break;
3063 0 : }
3064 :
3065 0 : scanner_finish(yyscanner);
3066 0 : }
3067 :
3068 : /*
3069 : * comp_location: comparator for qsorting LocationLen structs by location
3070 : */
3071 : static int
3072 0 : comp_location(const void *a, const void *b)
3073 : {
3074 0 : int l = ((const LocationLen *) a)->location;
3075 0 : int r = ((const LocationLen *) b)->location;
3076 :
3077 0 : return pg_cmp_s32(l, r);
3078 0 : }
|
