LCOV - code coverage report
Current view: top level - contrib/pg_stat_statements - pg_stat_statements.c (source / functions) Hit Total Coverage
Test: PostgreSQL 18devel Lines: 735 975 75.4 %
Date: 2025-04-01 15:15:16 Functions: 44 51 86.3 %
Legend: Lines: hit not hit

          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) 2008-2025, 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/parallel.h"
      51             : #include "catalog/pg_authid.h"
      52             : #include "common/int.h"
      53             : #include "executor/instrument.h"
      54             : #include "funcapi.h"
      55             : #include "jit/jit.h"
      56             : #include "mb/pg_wchar.h"
      57             : #include "miscadmin.h"
      58             : #include "nodes/queryjumble.h"
      59             : #include "optimizer/planner.h"
      60             : #include "parser/analyze.h"
      61             : #include "parser/scanner.h"
      62             : #include "pgstat.h"
      63             : #include "storage/fd.h"
      64             : #include "storage/ipc.h"
      65             : #include "storage/lwlock.h"
      66             : #include "storage/shmem.h"
      67             : #include "storage/spin.h"
      68             : #include "tcop/utility.h"
      69             : #include "utils/acl.h"
      70             : #include "utils/builtins.h"
      71             : #include "utils/memutils.h"
      72             : #include "utils/timestamp.h"
      73             : 
      74          16 : PG_MODULE_MAGIC_EXT(
      75             :                     .name = "pg_stat_statements",
      76             :                     .version = PG_VERSION
      77             : );
      78             : 
      79             : /* Location of permanent stats file (valid when database is shut down) */
      80             : #define PGSS_DUMP_FILE  PGSTAT_STAT_PERMANENT_DIRECTORY "/pg_stat_statements.stat"
      81             : 
      82             : /*
      83             :  * Location of external query text file.
      84             :  */
      85             : #define PGSS_TEXT_FILE  PG_STAT_TMP_DIR "/pgss_query_texts.stat"
      86             : 
      87             : /* Magic number identifying the stats file format */
      88             : static const uint32 PGSS_FILE_HEADER = 0x20220408;
      89             : 
      90             : /* PostgreSQL major version number, changes in which invalidate all entries */
      91             : static const uint32 PGSS_PG_MAJOR_VERSION = PG_VERSION_NUM / 100;
      92             : 
      93             : /* XXX: Should USAGE_EXEC reflect execution time and/or buffer usage? */
      94             : #define USAGE_EXEC(duration)    (1.0)
      95             : #define USAGE_INIT              (1.0)   /* including initial planning */
      96             : #define ASSUMED_MEDIAN_INIT     (10.0)  /* initial assumed median usage */
      97             : #define ASSUMED_LENGTH_INIT     1024    /* initial assumed mean query length */
      98             : #define USAGE_DECREASE_FACTOR   (0.99)  /* decreased every entry_dealloc */
      99             : #define STICKY_DECREASE_FACTOR  (0.50)  /* factor for sticky entries */
     100             : #define USAGE_DEALLOC_PERCENT   5   /* free this % of entries at once */
     101             : #define IS_STICKY(c)    ((c.calls[PGSS_PLAN] + c.calls[PGSS_EXEC]) == 0)
     102             : 
     103             : /*
     104             :  * Extension version number, for supporting older extension versions' objects
     105             :  */
     106             : typedef enum pgssVersion
     107             : {
     108             :     PGSS_V1_0 = 0,
     109             :     PGSS_V1_1,
     110             :     PGSS_V1_2,
     111             :     PGSS_V1_3,
     112             :     PGSS_V1_8,
     113             :     PGSS_V1_9,
     114             :     PGSS_V1_10,
     115             :     PGSS_V1_11,
     116             :     PGSS_V1_12,
     117             : } pgssVersion;
     118             : 
     119             : typedef enum pgssStoreKind
     120             : {
     121             :     PGSS_INVALID = -1,
     122             : 
     123             :     /*
     124             :      * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
     125             :      * reference the underlying values in the arrays in the Counters struct,
     126             :      * and this order is required in pg_stat_statements_internal().
     127             :      */
     128             :     PGSS_PLAN = 0,
     129             :     PGSS_EXEC,
     130             : } pgssStoreKind;
     131             : 
     132             : #define PGSS_NUMKIND (PGSS_EXEC + 1)
     133             : 
     134             : /*
     135             :  * Hashtable key that defines the identity of a hashtable entry.  We separate
     136             :  * queries by user and by database even if they are otherwise identical.
     137             :  *
     138             :  * If you add a new key to this struct, make sure to teach pgss_store() to
     139             :  * zero the padding bytes.  Otherwise, things will break, because pgss_hash is
     140             :  * created using HASH_BLOBS, and thus tag_hash is used to hash this.
     141             : 
     142             :  */
     143             : typedef struct pgssHashKey
     144             : {
     145             :     Oid         userid;         /* user OID */
     146             :     Oid         dbid;           /* database OID */
     147             :     uint64      queryid;        /* query identifier */
     148             :     bool        toplevel;       /* query executed at top level */
     149             : } pgssHashKey;
     150             : 
     151             : /*
     152             :  * The actual stats counters kept within pgssEntry.
     153             :  */
     154             : typedef struct Counters
     155             : {
     156             :     int64       calls[PGSS_NUMKIND];    /* # of times planned/executed */
     157             :     double      total_time[PGSS_NUMKIND];   /* total planning/execution time,
     158             :                                              * in msec */
     159             :     double      min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
     160             :                                          * msec since min/max reset */
     161             :     double      max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
     162             :                                          * msec since min/max reset */
     163             :     double      mean_time[PGSS_NUMKIND];    /* mean planning/execution time in
     164             :                                              * msec */
     165             :     double      sum_var_time[PGSS_NUMKIND]; /* sum of variances in
     166             :                                              * planning/execution time in msec */
     167             :     int64       rows;           /* total # of retrieved or affected rows */
     168             :     int64       shared_blks_hit;    /* # of shared buffer hits */
     169             :     int64       shared_blks_read;   /* # of shared disk blocks read */
     170             :     int64       shared_blks_dirtied;    /* # of shared disk blocks dirtied */
     171             :     int64       shared_blks_written;    /* # of shared disk blocks written */
     172             :     int64       local_blks_hit; /* # of local buffer hits */
     173             :     int64       local_blks_read;    /* # of local disk blocks read */
     174             :     int64       local_blks_dirtied; /* # of local disk blocks dirtied */
     175             :     int64       local_blks_written; /* # of local disk blocks written */
     176             :     int64       temp_blks_read; /* # of temp blocks read */
     177             :     int64       temp_blks_written;  /* # of temp blocks written */
     178             :     double      shared_blk_read_time;   /* time spent reading shared blocks,
     179             :                                          * in msec */
     180             :     double      shared_blk_write_time;  /* time spent writing shared blocks,
     181             :                                          * in msec */
     182             :     double      local_blk_read_time;    /* time spent reading local blocks, in
     183             :                                          * msec */
     184             :     double      local_blk_write_time;   /* time spent writing local blocks, in
     185             :                                          * msec */
     186             :     double      temp_blk_read_time; /* time spent reading temp blocks, in msec */
     187             :     double      temp_blk_write_time;    /* time spent writing temp blocks, in
     188             :                                          * msec */
     189             :     double      usage;          /* usage factor */
     190             :     int64       wal_records;    /* # of WAL records generated */
     191             :     int64       wal_fpi;        /* # of WAL full page images generated */
     192             :     uint64      wal_bytes;      /* total amount of WAL generated in bytes */
     193             :     int64       wal_buffers_full;   /* # of times the WAL buffers became full */
     194             :     int64       jit_functions;  /* total number of JIT functions emitted */
     195             :     double      jit_generation_time;    /* total time to generate jit code */
     196             :     int64       jit_inlining_count; /* number of times inlining time has been
     197             :                                      * > 0 */
     198             :     double      jit_deform_time;    /* total time to deform tuples in jit code */
     199             :     int64       jit_deform_count;   /* number of times deform time has been >
     200             :                                      * 0 */
     201             : 
     202             :     double      jit_inlining_time;  /* total time to inline jit code */
     203             :     int64       jit_optimization_count; /* number of times optimization time
     204             :                                          * has been > 0 */
     205             :     double      jit_optimization_time;  /* total time to optimize jit code */
     206             :     int64       jit_emission_count; /* number of times emission time has been
     207             :                                      * > 0 */
     208             :     double      jit_emission_time;  /* total time to emit jit code */
     209             :     int64       parallel_workers_to_launch; /* # of parallel workers planned
     210             :                                              * to be launched */
     211             :     int64       parallel_workers_launched;  /* # of parallel workers actually
     212             :                                              * launched */
     213             : } Counters;
     214             : 
     215             : /*
     216             :  * Global statistics for pg_stat_statements
     217             :  */
     218             : typedef struct pgssGlobalStats
     219             : {
     220             :     int64       dealloc;        /* # of times entries were deallocated */
     221             :     TimestampTz stats_reset;    /* timestamp with all stats reset */
     222             : } pgssGlobalStats;
     223             : 
     224             : /*
     225             :  * Statistics per statement
     226             :  *
     227             :  * Note: in event of a failure in garbage collection of the query text file,
     228             :  * we reset query_offset to zero and query_len to -1.  This will be seen as
     229             :  * an invalid state by qtext_fetch().
     230             :  */
     231             : typedef struct pgssEntry
     232             : {
     233             :     pgssHashKey key;            /* hash key of entry - MUST BE FIRST */
     234             :     Counters    counters;       /* the statistics for this query */
     235             :     Size        query_offset;   /* query text offset in external file */
     236             :     int         query_len;      /* # of valid bytes in query string, or -1 */
     237             :     int         encoding;       /* query text encoding */
     238             :     TimestampTz stats_since;    /* timestamp of entry allocation */
     239             :     TimestampTz minmax_stats_since; /* timestamp of last min/max values reset */
     240             :     slock_t     mutex;          /* protects the counters only */
     241             : } pgssEntry;
     242             : 
     243             : /*
     244             :  * Global shared state
     245             :  */
     246             : typedef struct pgssSharedState
     247             : {
     248             :     LWLock     *lock;           /* protects hashtable search/modification */
     249             :     double      cur_median_usage;   /* current median usage in hashtable */
     250             :     Size        mean_query_len; /* current mean entry text length */
     251             :     slock_t     mutex;          /* protects following fields only: */
     252             :     Size        extent;         /* current extent of query file */
     253             :     int         n_writers;      /* number of active writers to query file */
     254             :     int         gc_count;       /* query file garbage collection cycle count */
     255             :     pgssGlobalStats stats;      /* global statistics for pgss */
     256             : } pgssSharedState;
     257             : 
     258             : /*---- Local variables ----*/
     259             : 
     260             : /* Current nesting depth of planner/ExecutorRun/ProcessUtility calls */
     261             : static int  nesting_level = 0;
     262             : 
     263             : /* Saved hook values */
     264             : static shmem_request_hook_type prev_shmem_request_hook = NULL;
     265             : static shmem_startup_hook_type prev_shmem_startup_hook = NULL;
     266             : static post_parse_analyze_hook_type prev_post_parse_analyze_hook = NULL;
     267             : static planner_hook_type prev_planner_hook = NULL;
     268             : static ExecutorStart_hook_type prev_ExecutorStart = NULL;
     269             : static ExecutorRun_hook_type prev_ExecutorRun = NULL;
     270             : static ExecutorFinish_hook_type prev_ExecutorFinish = NULL;
     271             : static ExecutorEnd_hook_type prev_ExecutorEnd = NULL;
     272             : static ProcessUtility_hook_type prev_ProcessUtility = NULL;
     273             : 
     274             : /* Links to shared memory state */
     275             : static pgssSharedState *pgss = NULL;
     276             : static HTAB *pgss_hash = NULL;
     277             : 
     278             : /*---- GUC variables ----*/
     279             : 
     280             : typedef enum
     281             : {
     282             :     PGSS_TRACK_NONE,            /* track no statements */
     283             :     PGSS_TRACK_TOP,             /* only top level statements */
     284             :     PGSS_TRACK_ALL,             /* all statements, including nested ones */
     285             : }           PGSSTrackLevel;
     286             : 
     287             : static const struct config_enum_entry track_options[] =
     288             : {
     289             :     {"none", PGSS_TRACK_NONE, false},
     290             :     {"top", PGSS_TRACK_TOP, false},
     291             :     {"all", PGSS_TRACK_ALL, false},
     292             :     {NULL, 0, false}
     293             : };
     294             : 
     295             : static int  pgss_max = 5000;    /* max # statements to track */
     296             : static int  pgss_track = PGSS_TRACK_TOP;    /* tracking level */
     297             : static bool pgss_track_utility = true;  /* whether to track utility commands */
     298             : static bool pgss_track_planning = false;    /* whether to track planning
     299             :                                              * duration */
     300             : static bool pgss_save = true;   /* whether to save stats across shutdown */
     301             : 
     302             : #define pgss_enabled(level) \
     303             :     (!IsParallelWorker() && \
     304             :     (pgss_track == PGSS_TRACK_ALL || \
     305             :     (pgss_track == PGSS_TRACK_TOP && (level) == 0)))
     306             : 
     307             : #define record_gc_qtexts() \
     308             :     do { \
     309             :         SpinLockAcquire(&pgss->mutex); \
     310             :         pgss->gc_count++; \
     311             :         SpinLockRelease(&pgss->mutex); \
     312             :     } while(0)
     313             : 
     314             : /*---- Function declarations ----*/
     315             : 
     316          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset);
     317          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_7);
     318          40 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_11);
     319           0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_2);
     320          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_3);
     321          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_8);
     322          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_9);
     323          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_10);
     324          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_11);
     325          48 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_12);
     326           0 : PG_FUNCTION_INFO_V1(pg_stat_statements);
     327          16 : PG_FUNCTION_INFO_V1(pg_stat_statements_info);
     328             : 
     329             : static void pgss_shmem_request(void);
     330             : static void pgss_shmem_startup(void);
     331             : static void pgss_shmem_shutdown(int code, Datum arg);
     332             : static void pgss_post_parse_analyze(ParseState *pstate, Query *query,
     333             :                                     JumbleState *jstate);
     334             : static PlannedStmt *pgss_planner(Query *parse,
     335             :                                  const char *query_string,
     336             :                                  int cursorOptions,
     337             :                                  ParamListInfo boundParams);
     338             : static bool pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
     339             : static void pgss_ExecutorRun(QueryDesc *queryDesc,
     340             :                              ScanDirection direction,
     341             :                              uint64 count);
     342             : static void pgss_ExecutorFinish(QueryDesc *queryDesc);
     343             : static void pgss_ExecutorEnd(QueryDesc *queryDesc);
     344             : static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
     345             :                                 bool readOnlyTree,
     346             :                                 ProcessUtilityContext context, ParamListInfo params,
     347             :                                 QueryEnvironment *queryEnv,
     348             :                                 DestReceiver *dest, QueryCompletion *qc);
     349             : static void pgss_store(const char *query, uint64 queryId,
     350             :                        int query_location, int query_len,
     351             :                        pgssStoreKind kind,
     352             :                        double total_time, uint64 rows,
     353             :                        const BufferUsage *bufusage,
     354             :                        const WalUsage *walusage,
     355             :                        const struct JitInstrumentation *jitusage,
     356             :                        JumbleState *jstate,
     357             :                        int parallel_workers_to_launch,
     358             :                        int parallel_workers_launched);
     359             : static void pg_stat_statements_internal(FunctionCallInfo fcinfo,
     360             :                                         pgssVersion api_version,
     361             :                                         bool showtext);
     362             : static Size pgss_memsize(void);
     363             : static pgssEntry *entry_alloc(pgssHashKey *key, Size query_offset, int query_len,
     364             :                               int encoding, bool sticky);
     365             : static void entry_dealloc(void);
     366             : static bool qtext_store(const char *query, int query_len,
     367             :                         Size *query_offset, int *gc_count);
     368             : static char *qtext_load_file(Size *buffer_size);
     369             : static char *qtext_fetch(Size query_offset, int query_len,
     370             :                          char *buffer, Size buffer_size);
     371             : static bool need_gc_qtexts(void);
     372             : static void gc_qtexts(void);
     373             : static TimestampTz entry_reset(Oid userid, Oid dbid, uint64 queryid, bool minmax_only);
     374             : static char *generate_normalized_query(JumbleState *jstate, const char *query,
     375             :                                        int query_loc, int *query_len_p);
     376             : static void fill_in_constant_lengths(JumbleState *jstate, const char *query,
     377             :                                      int query_loc);
     378             : static int  comp_location(const void *a, const void *b);
     379             : 
     380             : 
     381             : /*
     382             :  * Module load callback
     383             :  */
     384             : void
     385          16 : _PG_init(void)
     386             : {
     387             :     /*
     388             :      * In order to create our shared memory area, we have to be loaded via
     389             :      * shared_preload_libraries.  If not, fall out without hooking into any of
     390             :      * the main system.  (We don't throw error here because it seems useful to
     391             :      * allow the pg_stat_statements functions to be created even when the
     392             :      * module isn't active.  The functions must protect themselves against
     393             :      * being called then, however.)
     394             :      */
     395          16 :     if (!process_shared_preload_libraries_in_progress)
     396           2 :         return;
     397             : 
     398             :     /*
     399             :      * Inform the postmaster that we want to enable query_id calculation if
     400             :      * compute_query_id is set to auto.
     401             :      */
     402          14 :     EnableQueryId();
     403             : 
     404             :     /*
     405             :      * Define (or redefine) custom GUC variables.
     406             :      */
     407          14 :     DefineCustomIntVariable("pg_stat_statements.max",
     408             :                             "Sets the maximum number of statements tracked by pg_stat_statements.",
     409             :                             NULL,
     410             :                             &pgss_max,
     411             :                             5000,
     412             :                             100,
     413             :                             INT_MAX / 2,
     414             :                             PGC_POSTMASTER,
     415             :                             0,
     416             :                             NULL,
     417             :                             NULL,
     418             :                             NULL);
     419             : 
     420          14 :     DefineCustomEnumVariable("pg_stat_statements.track",
     421             :                              "Selects which statements are tracked by pg_stat_statements.",
     422             :                              NULL,
     423             :                              &pgss_track,
     424             :                              PGSS_TRACK_TOP,
     425             :                              track_options,
     426             :                              PGC_SUSET,
     427             :                              0,
     428             :                              NULL,
     429             :                              NULL,
     430             :                              NULL);
     431             : 
     432          14 :     DefineCustomBoolVariable("pg_stat_statements.track_utility",
     433             :                              "Selects whether utility commands are tracked by pg_stat_statements.",
     434             :                              NULL,
     435             :                              &pgss_track_utility,
     436             :                              true,
     437             :                              PGC_SUSET,
     438             :                              0,
     439             :                              NULL,
     440             :                              NULL,
     441             :                              NULL);
     442             : 
     443          14 :     DefineCustomBoolVariable("pg_stat_statements.track_planning",
     444             :                              "Selects whether planning duration is tracked by pg_stat_statements.",
     445             :                              NULL,
     446             :                              &pgss_track_planning,
     447             :                              false,
     448             :                              PGC_SUSET,
     449             :                              0,
     450             :                              NULL,
     451             :                              NULL,
     452             :                              NULL);
     453             : 
     454          14 :     DefineCustomBoolVariable("pg_stat_statements.save",
     455             :                              "Save pg_stat_statements statistics across server shutdowns.",
     456             :                              NULL,
     457             :                              &pgss_save,
     458             :                              true,
     459             :                              PGC_SIGHUP,
     460             :                              0,
     461             :                              NULL,
     462             :                              NULL,
     463             :                              NULL);
     464             : 
     465          14 :     MarkGUCPrefixReserved("pg_stat_statements");
     466             : 
     467             :     /*
     468             :      * Install hooks.
     469             :      */
     470          14 :     prev_shmem_request_hook = shmem_request_hook;
     471          14 :     shmem_request_hook = pgss_shmem_request;
     472          14 :     prev_shmem_startup_hook = shmem_startup_hook;
     473          14 :     shmem_startup_hook = pgss_shmem_startup;
     474          14 :     prev_post_parse_analyze_hook = post_parse_analyze_hook;
     475          14 :     post_parse_analyze_hook = pgss_post_parse_analyze;
     476          14 :     prev_planner_hook = planner_hook;
     477          14 :     planner_hook = pgss_planner;
     478          14 :     prev_ExecutorStart = ExecutorStart_hook;
     479          14 :     ExecutorStart_hook = pgss_ExecutorStart;
     480          14 :     prev_ExecutorRun = ExecutorRun_hook;
     481          14 :     ExecutorRun_hook = pgss_ExecutorRun;
     482          14 :     prev_ExecutorFinish = ExecutorFinish_hook;
     483          14 :     ExecutorFinish_hook = pgss_ExecutorFinish;
     484          14 :     prev_ExecutorEnd = ExecutorEnd_hook;
     485          14 :     ExecutorEnd_hook = pgss_ExecutorEnd;
     486          14 :     prev_ProcessUtility = ProcessUtility_hook;
     487          14 :     ProcessUtility_hook = pgss_ProcessUtility;
     488             : }
     489             : 
     490             : /*
     491             :  * shmem_request hook: request additional shared resources.  We'll allocate or
     492             :  * attach to the shared resources in pgss_shmem_startup().
     493             :  */
     494             : static void
     495          14 : pgss_shmem_request(void)
     496             : {
     497          14 :     if (prev_shmem_request_hook)
     498           0 :         prev_shmem_request_hook();
     499             : 
     500          14 :     RequestAddinShmemSpace(pgss_memsize());
     501          14 :     RequestNamedLWLockTranche("pg_stat_statements", 1);
     502          14 : }
     503             : 
     504             : /*
     505             :  * shmem_startup hook: allocate or attach to shared memory,
     506             :  * then load any pre-existing statistics from file.
     507             :  * Also create and load the query-texts file, which is expected to exist
     508             :  * (even if empty) while the module is enabled.
     509             :  */
     510             : static void
     511          14 : pgss_shmem_startup(void)
     512             : {
     513             :     bool        found;
     514             :     HASHCTL     info;
     515          14 :     FILE       *file = NULL;
     516          14 :     FILE       *qfile = NULL;
     517             :     uint32      header;
     518             :     int32       num;
     519             :     int32       pgver;
     520             :     int32       i;
     521             :     int         buffer_size;
     522          14 :     char       *buffer = NULL;
     523             : 
     524          14 :     if (prev_shmem_startup_hook)
     525           0 :         prev_shmem_startup_hook();
     526             : 
     527             :     /* reset in case this is a restart within the postmaster */
     528          14 :     pgss = NULL;
     529          14 :     pgss_hash = NULL;
     530             : 
     531             :     /*
     532             :      * Create or attach to the shared memory state, including hash table
     533             :      */
     534          14 :     LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
     535             : 
     536          14 :     pgss = ShmemInitStruct("pg_stat_statements",
     537             :                            sizeof(pgssSharedState),
     538             :                            &found);
     539             : 
     540          14 :     if (!found)
     541             :     {
     542             :         /* First time through ... */
     543          14 :         pgss->lock = &(GetNamedLWLockTranche("pg_stat_statements"))->lock;
     544          14 :         pgss->cur_median_usage = ASSUMED_MEDIAN_INIT;
     545          14 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
     546          14 :         SpinLockInit(&pgss->mutex);
     547          14 :         pgss->extent = 0;
     548          14 :         pgss->n_writers = 0;
     549          14 :         pgss->gc_count = 0;
     550          14 :         pgss->stats.dealloc = 0;
     551          14 :         pgss->stats.stats_reset = GetCurrentTimestamp();
     552             :     }
     553             : 
     554          14 :     info.keysize = sizeof(pgssHashKey);
     555          14 :     info.entrysize = sizeof(pgssEntry);
     556          14 :     pgss_hash = ShmemInitHash("pg_stat_statements hash",
     557             :                               pgss_max, pgss_max,
     558             :                               &info,
     559             :                               HASH_ELEM | HASH_BLOBS);
     560             : 
     561          14 :     LWLockRelease(AddinShmemInitLock);
     562             : 
     563             :     /*
     564             :      * If we're in the postmaster (or a standalone backend...), set up a shmem
     565             :      * exit hook to dump the statistics to disk.
     566             :      */
     567          14 :     if (!IsUnderPostmaster)
     568          14 :         on_shmem_exit(pgss_shmem_shutdown, (Datum) 0);
     569             : 
     570             :     /*
     571             :      * Done if some other process already completed our initialization.
     572             :      */
     573          14 :     if (found)
     574          14 :         return;
     575             : 
     576             :     /*
     577             :      * Note: we don't bother with locks here, because there should be no other
     578             :      * processes running when this code is reached.
     579             :      */
     580             : 
     581             :     /* Unlink query text file possibly left over from crash */
     582          14 :     unlink(PGSS_TEXT_FILE);
     583             : 
     584             :     /* Allocate new query text temp file */
     585          14 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
     586          14 :     if (qfile == NULL)
     587           0 :         goto write_error;
     588             : 
     589             :     /*
     590             :      * If we were told not to load old statistics, we're done.  (Note we do
     591             :      * not try to unlink any old dump file in this case.  This seems a bit
     592             :      * questionable but it's the historical behavior.)
     593             :      */
     594          14 :     if (!pgss_save)
     595             :     {
     596           2 :         FreeFile(qfile);
     597           2 :         return;
     598             :     }
     599             : 
     600             :     /*
     601             :      * Attempt to load old statistics from the dump file.
     602             :      */
     603          12 :     file = AllocateFile(PGSS_DUMP_FILE, PG_BINARY_R);
     604          12 :     if (file == NULL)
     605             :     {
     606           8 :         if (errno != ENOENT)
     607           0 :             goto read_error;
     608             :         /* No existing persisted stats file, so we're done */
     609           8 :         FreeFile(qfile);
     610           8 :         return;
     611             :     }
     612             : 
     613           4 :     buffer_size = 2048;
     614           4 :     buffer = (char *) palloc(buffer_size);
     615             : 
     616           8 :     if (fread(&header, sizeof(uint32), 1, file) != 1 ||
     617           8 :         fread(&pgver, sizeof(uint32), 1, file) != 1 ||
     618           4 :         fread(&num, sizeof(int32), 1, file) != 1)
     619           0 :         goto read_error;
     620             : 
     621           4 :     if (header != PGSS_FILE_HEADER ||
     622           4 :         pgver != PGSS_PG_MAJOR_VERSION)
     623           0 :         goto data_error;
     624             : 
     625       51810 :     for (i = 0; i < num; i++)
     626             :     {
     627             :         pgssEntry   temp;
     628             :         pgssEntry  *entry;
     629             :         Size        query_offset;
     630             : 
     631       51806 :         if (fread(&temp, sizeof(pgssEntry), 1, file) != 1)
     632           0 :             goto read_error;
     633             : 
     634             :         /* Encoding is the only field we can easily sanity-check */
     635       51806 :         if (!PG_VALID_BE_ENCODING(temp.encoding))
     636           0 :             goto data_error;
     637             : 
     638             :         /* Resize buffer as needed */
     639       51806 :         if (temp.query_len >= buffer_size)
     640             :         {
     641           4 :             buffer_size = Max(buffer_size * 2, temp.query_len + 1);
     642           4 :             buffer = repalloc(buffer, buffer_size);
     643             :         }
     644             : 
     645       51806 :         if (fread(buffer, 1, temp.query_len + 1, file) != temp.query_len + 1)
     646           0 :             goto read_error;
     647             : 
     648             :         /* Should have a trailing null, but let's make sure */
     649       51806 :         buffer[temp.query_len] = '\0';
     650             : 
     651             :         /* Skip loading "sticky" entries */
     652       51806 :         if (IS_STICKY(temp.counters))
     653        1390 :             continue;
     654             : 
     655             :         /* Store the query text */
     656       50416 :         query_offset = pgss->extent;
     657       50416 :         if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
     658           0 :             goto write_error;
     659       50416 :         pgss->extent += temp.query_len + 1;
     660             : 
     661             :         /* make the hashtable entry (discards old entries if too many) */
     662       50416 :         entry = entry_alloc(&temp.key, query_offset, temp.query_len,
     663             :                             temp.encoding,
     664             :                             false);
     665             : 
     666             :         /* copy in the actual stats */
     667       50416 :         entry->counters = temp.counters;
     668       50416 :         entry->stats_since = temp.stats_since;
     669       50416 :         entry->minmax_stats_since = temp.minmax_stats_since;
     670             :     }
     671             : 
     672             :     /* Read global statistics for pg_stat_statements */
     673           4 :     if (fread(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
     674           0 :         goto read_error;
     675             : 
     676           4 :     pfree(buffer);
     677           4 :     FreeFile(file);
     678           4 :     FreeFile(qfile);
     679             : 
     680             :     /*
     681             :      * Remove the persisted stats file so it's not included in
     682             :      * backups/replication standbys, etc.  A new file will be written on next
     683             :      * shutdown.
     684             :      *
     685             :      * Note: it's okay if the PGSS_TEXT_FILE is included in a basebackup,
     686             :      * because we remove that file on startup; it acts inversely to
     687             :      * PGSS_DUMP_FILE, in that it is only supposed to be around when the
     688             :      * server is running, whereas PGSS_DUMP_FILE is only supposed to be around
     689             :      * when the server is not running.  Leaving the file creates no danger of
     690             :      * a newly restored database having a spurious record of execution costs,
     691             :      * which is what we're really concerned about here.
     692             :      */
     693           4 :     unlink(PGSS_DUMP_FILE);
     694             : 
     695           4 :     return;
     696             : 
     697           0 : read_error:
     698           0 :     ereport(LOG,
     699             :             (errcode_for_file_access(),
     700             :              errmsg("could not read file \"%s\": %m",
     701             :                     PGSS_DUMP_FILE)));
     702           0 :     goto fail;
     703           0 : data_error:
     704           0 :     ereport(LOG,
     705             :             (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
     706             :              errmsg("ignoring invalid data in file \"%s\"",
     707             :                     PGSS_DUMP_FILE)));
     708           0 :     goto fail;
     709           0 : write_error:
     710           0 :     ereport(LOG,
     711             :             (errcode_for_file_access(),
     712             :              errmsg("could not write file \"%s\": %m",
     713             :                     PGSS_TEXT_FILE)));
     714           0 : fail:
     715           0 :     if (buffer)
     716           0 :         pfree(buffer);
     717           0 :     if (file)
     718           0 :         FreeFile(file);
     719           0 :     if (qfile)
     720           0 :         FreeFile(qfile);
     721             :     /* If possible, throw away the bogus file; ignore any error */
     722           0 :     unlink(PGSS_DUMP_FILE);
     723             : 
     724             :     /*
     725             :      * Don't unlink PGSS_TEXT_FILE here; it should always be around while the
     726             :      * server is running with pg_stat_statements enabled
     727             :      */
     728             : }
     729             : 
     730             : /*
     731             :  * shmem_shutdown hook: Dump statistics into file.
     732             :  *
     733             :  * Note: we don't bother with acquiring lock, because there should be no
     734             :  * other processes running when this is called.
     735             :  */
     736             : static void
     737          14 : pgss_shmem_shutdown(int code, Datum arg)
     738             : {
     739             :     FILE       *file;
     740          14 :     char       *qbuffer = NULL;
     741          14 :     Size        qbuffer_size = 0;
     742             :     HASH_SEQ_STATUS hash_seq;
     743             :     int32       num_entries;
     744             :     pgssEntry  *entry;
     745             : 
     746             :     /* Don't try to dump during a crash. */
     747          14 :     if (code)
     748          14 :         return;
     749             : 
     750             :     /* Safety check ... shouldn't get here unless shmem is set up. */
     751          14 :     if (!pgss || !pgss_hash)
     752           0 :         return;
     753             : 
     754             :     /* Don't dump if told not to. */
     755          14 :     if (!pgss_save)
     756           4 :         return;
     757             : 
     758          10 :     file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
     759          10 :     if (file == NULL)
     760           0 :         goto error;
     761             : 
     762          10 :     if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
     763           0 :         goto error;
     764          10 :     if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
     765           0 :         goto error;
     766          10 :     num_entries = hash_get_num_entries(pgss_hash);
     767          10 :     if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
     768           0 :         goto error;
     769             : 
     770          10 :     qbuffer = qtext_load_file(&qbuffer_size);
     771          10 :     if (qbuffer == NULL)
     772           0 :         goto error;
     773             : 
     774             :     /*
     775             :      * When serializing to disk, we store query texts immediately after their
     776             :      * entry data.  Any orphaned query texts are thereby excluded.
     777             :      */
     778          10 :     hash_seq_init(&hash_seq, pgss_hash);
     779      104188 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
     780             :     {
     781      104178 :         int         len = entry->query_len;
     782      104178 :         char       *qstr = qtext_fetch(entry->query_offset, len,
     783             :                                        qbuffer, qbuffer_size);
     784             : 
     785      104178 :         if (qstr == NULL)
     786           0 :             continue;           /* Ignore any entries with bogus texts */
     787             : 
     788      104178 :         if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
     789      104178 :             fwrite(qstr, 1, len + 1, file) != len + 1)
     790             :         {
     791             :             /* note: we assume hash_seq_term won't change errno */
     792           0 :             hash_seq_term(&hash_seq);
     793           0 :             goto error;
     794             :         }
     795             :     }
     796             : 
     797             :     /* Dump global statistics for pg_stat_statements */
     798          10 :     if (fwrite(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
     799           0 :         goto error;
     800             : 
     801          10 :     free(qbuffer);
     802          10 :     qbuffer = NULL;
     803             : 
     804          10 :     if (FreeFile(file))
     805             :     {
     806           0 :         file = NULL;
     807           0 :         goto error;
     808             :     }
     809             : 
     810             :     /*
     811             :      * Rename file into place, so we atomically replace any old one.
     812             :      */
     813          10 :     (void) durable_rename(PGSS_DUMP_FILE ".tmp", PGSS_DUMP_FILE, LOG);
     814             : 
     815             :     /* Unlink query-texts file; it's not needed while shutdown */
     816          10 :     unlink(PGSS_TEXT_FILE);
     817             : 
     818          10 :     return;
     819             : 
     820           0 : error:
     821           0 :     ereport(LOG,
     822             :             (errcode_for_file_access(),
     823             :              errmsg("could not write file \"%s\": %m",
     824             :                     PGSS_DUMP_FILE ".tmp")));
     825           0 :     free(qbuffer);
     826           0 :     if (file)
     827           0 :         FreeFile(file);
     828           0 :     unlink(PGSS_DUMP_FILE ".tmp");
     829           0 :     unlink(PGSS_TEXT_FILE);
     830             : }
     831             : 
     832             : /*
     833             :  * Post-parse-analysis hook: mark query with a queryId
     834             :  */
     835             : static void
     836      154728 : pgss_post_parse_analyze(ParseState *pstate, Query *query, JumbleState *jstate)
     837             : {
     838      154728 :     if (prev_post_parse_analyze_hook)
     839           0 :         prev_post_parse_analyze_hook(pstate, query, jstate);
     840             : 
     841             :     /* Safety check... */
     842      154728 :     if (!pgss || !pgss_hash || !pgss_enabled(nesting_level))
     843       27080 :         return;
     844             : 
     845             :     /*
     846             :      * If it's EXECUTE, clear the queryId so that stats will accumulate for
     847             :      * the underlying PREPARE.  But don't do this if we're not tracking
     848             :      * utility statements, to avoid messing up another extension that might be
     849             :      * tracking them.
     850             :      */
     851      127648 :     if (query->utilityStmt)
     852             :     {
     853       57650 :         if (pgss_track_utility && IsA(query->utilityStmt, ExecuteStmt))
     854             :         {
     855        7166 :             query->queryId = UINT64CONST(0);
     856        7166 :             return;
     857             :         }
     858             :     }
     859             : 
     860             :     /*
     861             :      * If query jumbling were able to identify any ignorable constants, we
     862             :      * immediately create a hash table entry for the query, so that we can
     863             :      * record the normalized form of the query string.  If there were no such
     864             :      * constants, the normalized string would be the same as the query text
     865             :      * anyway, so there's no need for an early entry.
     866             :      */
     867      120482 :     if (jstate && jstate->clocations_count > 0)
     868       68864 :         pgss_store(pstate->p_sourcetext,
     869             :                    query->queryId,
     870             :                    query->stmt_location,
     871             :                    query->stmt_len,
     872             :                    PGSS_INVALID,
     873             :                    0,
     874             :                    0,
     875             :                    NULL,
     876             :                    NULL,
     877             :                    NULL,
     878             :                    jstate,
     879             :                    0,
     880             :                    0);
     881             : }
     882             : 
     883             : /*
     884             :  * Planner hook: forward to regular planner, but measure planning time
     885             :  * if needed.
     886             :  */
     887             : static PlannedStmt *
     888       93708 : pgss_planner(Query *parse,
     889             :              const char *query_string,
     890             :              int cursorOptions,
     891             :              ParamListInfo boundParams)
     892             : {
     893             :     PlannedStmt *result;
     894             : 
     895             :     /*
     896             :      * We can't process the query if no query_string is provided, as
     897             :      * pgss_store needs it.  We also ignore query without queryid, as it would
     898             :      * be treated as a utility statement, which may not be the case.
     899             :      */
     900       93708 :     if (pgss_enabled(nesting_level)
     901       70290 :         && pgss_track_planning && query_string
     902         242 :         && parse->queryId != UINT64CONST(0))
     903         242 :     {
     904             :         instr_time  start;
     905             :         instr_time  duration;
     906             :         BufferUsage bufusage_start,
     907             :                     bufusage;
     908             :         WalUsage    walusage_start,
     909             :                     walusage;
     910             : 
     911             :         /* We need to track buffer usage as the planner can access them. */
     912         242 :         bufusage_start = pgBufferUsage;
     913             : 
     914             :         /*
     915             :          * Similarly the planner could write some WAL records in some cases
     916             :          * (e.g. setting a hint bit with those being WAL-logged)
     917             :          */
     918         242 :         walusage_start = pgWalUsage;
     919         242 :         INSTR_TIME_SET_CURRENT(start);
     920             : 
     921         242 :         nesting_level++;
     922         242 :         PG_TRY();
     923             :         {
     924         242 :             if (prev_planner_hook)
     925           0 :                 result = prev_planner_hook(parse, query_string, cursorOptions,
     926             :                                            boundParams);
     927             :             else
     928         242 :                 result = standard_planner(parse, query_string, cursorOptions,
     929             :                                           boundParams);
     930             :         }
     931           0 :         PG_FINALLY();
     932             :         {
     933         242 :             nesting_level--;
     934             :         }
     935         242 :         PG_END_TRY();
     936             : 
     937         242 :         INSTR_TIME_SET_CURRENT(duration);
     938         242 :         INSTR_TIME_SUBTRACT(duration, start);
     939             : 
     940             :         /* calc differences of buffer counters. */
     941         242 :         memset(&bufusage, 0, sizeof(BufferUsage));
     942         242 :         BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
     943             : 
     944             :         /* calc differences of WAL counters. */
     945         242 :         memset(&walusage, 0, sizeof(WalUsage));
     946         242 :         WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
     947             : 
     948         242 :         pgss_store(query_string,
     949             :                    parse->queryId,
     950             :                    parse->stmt_location,
     951             :                    parse->stmt_len,
     952             :                    PGSS_PLAN,
     953         242 :                    INSTR_TIME_GET_MILLISEC(duration),
     954             :                    0,
     955             :                    &bufusage,
     956             :                    &walusage,
     957             :                    NULL,
     958             :                    NULL,
     959             :                    0,
     960             :                    0);
     961             :     }
     962             :     else
     963             :     {
     964             :         /*
     965             :          * Even though we're not tracking plan time for this statement, we
     966             :          * must still increment the nesting level, to ensure that functions
     967             :          * evaluated during planning are not seen as top-level calls.
     968             :          */
     969       93466 :         nesting_level++;
     970       93466 :         PG_TRY();
     971             :         {
     972       93466 :             if (prev_planner_hook)
     973           0 :                 result = prev_planner_hook(parse, query_string, cursorOptions,
     974             :                                            boundParams);
     975             :             else
     976       93466 :                 result = standard_planner(parse, query_string, cursorOptions,
     977             :                                           boundParams);
     978             :         }
     979        1342 :         PG_FINALLY();
     980             :         {
     981       93466 :             nesting_level--;
     982             :         }
     983       93466 :         PG_END_TRY();
     984             :     }
     985             : 
     986       92366 :     return result;
     987             : }
     988             : 
     989             : /*
     990             :  * ExecutorStart hook: start up tracking if needed
     991             :  */
     992             : static bool
     993      113332 : pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
     994             : {
     995             :     bool        plan_valid;
     996             : 
     997      113332 :     if (prev_ExecutorStart)
     998           0 :         plan_valid = prev_ExecutorStart(queryDesc, eflags);
     999             :     else
    1000      113332 :         plan_valid = standard_ExecutorStart(queryDesc, eflags);
    1001             : 
    1002             :     /* The plan may have become invalid during standard_ExecutorStart() */
    1003      112700 :     if (!plan_valid)
    1004           0 :         return false;
    1005             : 
    1006             :     /*
    1007             :      * If query has queryId zero, don't track it.  This prevents double
    1008             :      * counting of optimizable statements that are directly contained in
    1009             :      * utility statements.
    1010             :      */
    1011      112700 :     if (pgss_enabled(nesting_level) && queryDesc->plannedstmt->queryId != UINT64CONST(0))
    1012             :     {
    1013             :         /*
    1014             :          * Set up to track total elapsed time in ExecutorRun.  Make sure the
    1015             :          * space is allocated in the per-query context so it will go away at
    1016             :          * ExecutorEnd.
    1017             :          */
    1018       74886 :         if (queryDesc->totaltime == NULL)
    1019             :         {
    1020             :             MemoryContext oldcxt;
    1021             : 
    1022       74886 :             oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
    1023       74886 :             queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL, false);
    1024       74886 :             MemoryContextSwitchTo(oldcxt);
    1025             :         }
    1026             :     }
    1027             : 
    1028      112700 :     return true;
    1029             : }
    1030             : 
    1031             : /*
    1032             :  * ExecutorRun hook: all we need do is track nesting depth
    1033             :  */
    1034             : static void
    1035      110254 : pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count)
    1036             : {
    1037      110254 :     nesting_level++;
    1038      110254 :     PG_TRY();
    1039             :     {
    1040      110254 :         if (prev_ExecutorRun)
    1041           0 :             prev_ExecutorRun(queryDesc, direction, count);
    1042             :         else
    1043      110254 :             standard_ExecutorRun(queryDesc, direction, count);
    1044             :     }
    1045        6778 :     PG_FINALLY();
    1046             :     {
    1047      110254 :         nesting_level--;
    1048             :     }
    1049      110254 :     PG_END_TRY();
    1050      103476 : }
    1051             : 
    1052             : /*
    1053             :  * ExecutorFinish hook: all we need do is track nesting depth
    1054             :  */
    1055             : static void
    1056       99642 : pgss_ExecutorFinish(QueryDesc *queryDesc)
    1057             : {
    1058       99642 :     nesting_level++;
    1059       99642 :     PG_TRY();
    1060             :     {
    1061       99642 :         if (prev_ExecutorFinish)
    1062           0 :             prev_ExecutorFinish(queryDesc);
    1063             :         else
    1064       99642 :             standard_ExecutorFinish(queryDesc);
    1065             :     }
    1066         314 :     PG_FINALLY();
    1067             :     {
    1068       99642 :         nesting_level--;
    1069             :     }
    1070       99642 :     PG_END_TRY();
    1071       99328 : }
    1072             : 
    1073             : /*
    1074             :  * ExecutorEnd hook: store results if needed
    1075             :  */
    1076             : static void
    1077      104954 : pgss_ExecutorEnd(QueryDesc *queryDesc)
    1078             : {
    1079      104954 :     uint64      queryId = queryDesc->plannedstmt->queryId;
    1080             : 
    1081      104954 :     if (queryId != UINT64CONST(0) && queryDesc->totaltime &&
    1082       71872 :         pgss_enabled(nesting_level))
    1083             :     {
    1084             :         /*
    1085             :          * Make sure stats accumulation is done.  (Note: it's okay if several
    1086             :          * levels of hook all do this.)
    1087             :          */
    1088       71872 :         InstrEndLoop(queryDesc->totaltime);
    1089             : 
    1090       71714 :         pgss_store(queryDesc->sourceText,
    1091             :                    queryId,
    1092       71872 :                    queryDesc->plannedstmt->stmt_location,
    1093       71872 :                    queryDesc->plannedstmt->stmt_len,
    1094             :                    PGSS_EXEC,
    1095       71872 :                    queryDesc->totaltime->total * 1000.0,  /* convert to msec */
    1096       71872 :                    queryDesc->estate->es_total_processed,
    1097       71872 :                    &queryDesc->totaltime->bufusage,
    1098       71872 :                    &queryDesc->totaltime->walusage,
    1099         158 :                    queryDesc->estate->es_jit ? &queryDesc->estate->es_jit->instr : NULL,
    1100             :                    NULL,
    1101       71872 :                    queryDesc->estate->es_parallel_workers_to_launch,
    1102       71872 :                    queryDesc->estate->es_parallel_workers_launched);
    1103             :     }
    1104             : 
    1105      104954 :     if (prev_ExecutorEnd)
    1106           0 :         prev_ExecutorEnd(queryDesc);
    1107             :     else
    1108      104954 :         standard_ExecutorEnd(queryDesc);
    1109      104954 : }
    1110             : 
    1111             : /*
    1112             :  * ProcessUtility hook
    1113             :  */
    1114             : static void
    1115       68322 : pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
    1116             :                     bool readOnlyTree,
    1117             :                     ProcessUtilityContext context,
    1118             :                     ParamListInfo params, QueryEnvironment *queryEnv,
    1119             :                     DestReceiver *dest, QueryCompletion *qc)
    1120             : {
    1121       68322 :     Node       *parsetree = pstmt->utilityStmt;
    1122       68322 :     uint64      saved_queryId = pstmt->queryId;
    1123       68322 :     int         saved_stmt_location = pstmt->stmt_location;
    1124       68322 :     int         saved_stmt_len = pstmt->stmt_len;
    1125       68322 :     bool        enabled = pgss_track_utility && pgss_enabled(nesting_level);
    1126             : 
    1127             :     /*
    1128             :      * Force utility statements to get queryId zero.  We do this even in cases
    1129             :      * where the statement contains an optimizable statement for which a
    1130             :      * queryId could be derived (such as EXPLAIN or DECLARE CURSOR).  For such
    1131             :      * cases, runtime control will first go through ProcessUtility and then
    1132             :      * the executor, and we don't want the executor hooks to do anything,
    1133             :      * since we are already measuring the statement's costs at the utility
    1134             :      * level.
    1135             :      *
    1136             :      * Note that this is only done if pg_stat_statements is enabled and
    1137             :      * configured to track utility statements, in the unlikely possibility
    1138             :      * that user configured another extension to handle utility statements
    1139             :      * only.
    1140             :      */
    1141       68322 :     if (enabled)
    1142       57450 :         pstmt->queryId = UINT64CONST(0);
    1143             : 
    1144             :     /*
    1145             :      * If it's an EXECUTE statement, we don't track it and don't increment the
    1146             :      * nesting level.  This allows the cycles to be charged to the underlying
    1147             :      * PREPARE instead (by the Executor hooks), which is much more useful.
    1148             :      *
    1149             :      * We also don't track execution of PREPARE.  If we did, we would get one
    1150             :      * hash table entry for the PREPARE (with hash calculated from the query
    1151             :      * string), and then a different one with the same query string (but hash
    1152             :      * calculated from the query tree) would be used to accumulate costs of
    1153             :      * ensuing EXECUTEs.  This would be confusing.  Since PREPARE doesn't
    1154             :      * actually run the planner (only parse+rewrite), its costs are generally
    1155             :      * pretty negligible and it seems okay to just ignore it.
    1156             :      */
    1157       68322 :     if (enabled &&
    1158       57450 :         !IsA(parsetree, ExecuteStmt) &&
    1159       50288 :         !IsA(parsetree, PrepareStmt))
    1160       45338 :     {
    1161             :         instr_time  start;
    1162             :         instr_time  duration;
    1163             :         uint64      rows;
    1164             :         BufferUsage bufusage_start,
    1165             :                     bufusage;
    1166             :         WalUsage    walusage_start,
    1167             :                     walusage;
    1168             : 
    1169       50050 :         bufusage_start = pgBufferUsage;
    1170       50050 :         walusage_start = pgWalUsage;
    1171       50050 :         INSTR_TIME_SET_CURRENT(start);
    1172             : 
    1173       50050 :         nesting_level++;
    1174       50050 :         PG_TRY();
    1175             :         {
    1176       50050 :             if (prev_ProcessUtility)
    1177           0 :                 prev_ProcessUtility(pstmt, queryString, readOnlyTree,
    1178             :                                     context, params, queryEnv,
    1179             :                                     dest, qc);
    1180             :             else
    1181       50050 :                 standard_ProcessUtility(pstmt, queryString, readOnlyTree,
    1182             :                                         context, params, queryEnv,
    1183             :                                         dest, qc);
    1184             :         }
    1185        4712 :         PG_FINALLY();
    1186             :         {
    1187       50050 :             nesting_level--;
    1188             :         }
    1189       50050 :         PG_END_TRY();
    1190             : 
    1191             :         /*
    1192             :          * CAUTION: do not access the *pstmt data structure again below here.
    1193             :          * If it was a ROLLBACK or similar, that data structure may have been
    1194             :          * freed.  We must copy everything we still need into local variables,
    1195             :          * which we did above.
    1196             :          *
    1197             :          * For the same reason, we can't risk restoring pstmt->queryId to its
    1198             :          * former value, which'd otherwise be a good idea.
    1199             :          */
    1200             : 
    1201       45338 :         INSTR_TIME_SET_CURRENT(duration);
    1202       45338 :         INSTR_TIME_SUBTRACT(duration, start);
    1203             : 
    1204             :         /*
    1205             :          * Track the total number of rows retrieved or affected by the utility
    1206             :          * statements of COPY, FETCH, CREATE TABLE AS, CREATE MATERIALIZED
    1207             :          * VIEW, REFRESH MATERIALIZED VIEW and SELECT INTO.
    1208             :          */
    1209       45332 :         rows = (qc && (qc->commandTag == CMDTAG_COPY ||
    1210       42114 :                        qc->commandTag == CMDTAG_FETCH ||
    1211       41644 :                        qc->commandTag == CMDTAG_SELECT ||
    1212       41272 :                        qc->commandTag == CMDTAG_REFRESH_MATERIALIZED_VIEW)) ?
    1213       90670 :             qc->nprocessed : 0;
    1214             : 
    1215             :         /* calc differences of buffer counters. */
    1216       45338 :         memset(&bufusage, 0, sizeof(BufferUsage));
    1217       45338 :         BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
    1218             : 
    1219             :         /* calc differences of WAL counters. */
    1220       45338 :         memset(&walusage, 0, sizeof(WalUsage));
    1221       45338 :         WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
    1222             : 
    1223       45338 :         pgss_store(queryString,
    1224             :                    saved_queryId,
    1225             :                    saved_stmt_location,
    1226             :                    saved_stmt_len,
    1227             :                    PGSS_EXEC,
    1228       45338 :                    INSTR_TIME_GET_MILLISEC(duration),
    1229             :                    rows,
    1230             :                    &bufusage,
    1231             :                    &walusage,
    1232             :                    NULL,
    1233             :                    NULL,
    1234             :                    0,
    1235             :                    0);
    1236             :     }
    1237             :     else
    1238             :     {
    1239             :         /*
    1240             :          * Even though we're not tracking execution time for this statement,
    1241             :          * we must still increment the nesting level, to ensure that functions
    1242             :          * evaluated within it are not seen as top-level calls.  But don't do
    1243             :          * so for EXECUTE; that way, when control reaches pgss_planner or
    1244             :          * pgss_ExecutorStart, we will treat the costs as top-level if
    1245             :          * appropriate.  Likewise, don't bump for PREPARE, so that parse
    1246             :          * analysis will treat the statement as top-level if appropriate.
    1247             :          *
    1248             :          * To be absolutely certain we don't mess up the nesting level,
    1249             :          * evaluate the bump_level condition just once.
    1250             :          */
    1251       18272 :         bool        bump_level =
    1252       29380 :             !IsA(parsetree, ExecuteStmt) &&
    1253       11108 :             !IsA(parsetree, PrepareStmt);
    1254             : 
    1255       18272 :         if (bump_level)
    1256       10868 :             nesting_level++;
    1257       18272 :         PG_TRY();
    1258             :         {
    1259       18272 :             if (prev_ProcessUtility)
    1260           0 :                 prev_ProcessUtility(pstmt, queryString, readOnlyTree,
    1261             :                                     context, params, queryEnv,
    1262             :                                     dest, qc);
    1263             :             else
    1264       18272 :                 standard_ProcessUtility(pstmt, queryString, readOnlyTree,
    1265             :                                         context, params, queryEnv,
    1266             :                                         dest, qc);
    1267             :         }
    1268         260 :         PG_FINALLY();
    1269             :         {
    1270       18272 :             if (bump_level)
    1271       10868 :                 nesting_level--;
    1272             :         }
    1273       18272 :         PG_END_TRY();
    1274             :     }
    1275       63350 : }
    1276             : 
    1277             : /*
    1278             :  * Store some statistics for a statement.
    1279             :  *
    1280             :  * If jstate is not NULL then we're trying to create an entry for which
    1281             :  * we have no statistics as yet; we just want to record the normalized
    1282             :  * query string.  total_time, rows, bufusage and walusage are ignored in this
    1283             :  * case.
    1284             :  *
    1285             :  * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
    1286             :  * for the arrays in the Counters field.
    1287             :  */
    1288             : static void
    1289      186316 : pgss_store(const char *query, uint64 queryId,
    1290             :            int query_location, int query_len,
    1291             :            pgssStoreKind kind,
    1292             :            double total_time, uint64 rows,
    1293             :            const BufferUsage *bufusage,
    1294             :            const WalUsage *walusage,
    1295             :            const struct JitInstrumentation *jitusage,
    1296             :            JumbleState *jstate,
    1297             :            int parallel_workers_to_launch,
    1298             :            int parallel_workers_launched)
    1299             : {
    1300             :     pgssHashKey key;
    1301             :     pgssEntry  *entry;
    1302      186316 :     char       *norm_query = NULL;
    1303      186316 :     int         encoding = GetDatabaseEncoding();
    1304             : 
    1305             :     Assert(query != NULL);
    1306             : 
    1307             :     /* Safety check... */
    1308      186316 :     if (!pgss || !pgss_hash)
    1309           0 :         return;
    1310             : 
    1311             :     /*
    1312             :      * Nothing to do if compute_query_id isn't enabled and no other module
    1313             :      * computed a query identifier.
    1314             :      */
    1315      186316 :     if (queryId == UINT64CONST(0))
    1316           0 :         return;
    1317             : 
    1318             :     /*
    1319             :      * Confine our attention to the relevant part of the string, if the query
    1320             :      * is a portion of a multi-statement source string, and update query
    1321             :      * location and length if needed.
    1322             :      */
    1323      186316 :     query = CleanQuerytext(query, &query_location, &query_len);
    1324             : 
    1325             :     /* Set up key for hashtable search */
    1326             : 
    1327             :     /* clear padding */
    1328      186316 :     memset(&key, 0, sizeof(pgssHashKey));
    1329             : 
    1330      186316 :     key.userid = GetUserId();
    1331      186316 :     key.dbid = MyDatabaseId;
    1332      186316 :     key.queryid = queryId;
    1333      186316 :     key.toplevel = (nesting_level == 0);
    1334             : 
    1335             :     /* Lookup the hash table entry with shared lock. */
    1336      186316 :     LWLockAcquire(pgss->lock, LW_SHARED);
    1337             : 
    1338      186316 :     entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    1339             : 
    1340             :     /* Create new entry, if not present */
    1341      186316 :     if (!entry)
    1342             :     {
    1343             :         Size        query_offset;
    1344             :         int         gc_count;
    1345             :         bool        stored;
    1346             :         bool        do_gc;
    1347             : 
    1348             :         /*
    1349             :          * Create a new, normalized query string if caller asked.  We don't
    1350             :          * need to hold the lock while doing this work.  (Note: in any case,
    1351             :          * it's possible that someone else creates a duplicate hashtable entry
    1352             :          * in the interval where we don't hold the lock below.  That case is
    1353             :          * handled by entry_alloc.)
    1354             :          */
    1355       55252 :         if (jstate)
    1356             :         {
    1357       20424 :             LWLockRelease(pgss->lock);
    1358       20424 :             norm_query = generate_normalized_query(jstate, query,
    1359             :                                                    query_location,
    1360             :                                                    &query_len);
    1361       20424 :             LWLockAcquire(pgss->lock, LW_SHARED);
    1362             :         }
    1363             : 
    1364             :         /* Append new query text to file with only shared lock held */
    1365       55252 :         stored = qtext_store(norm_query ? norm_query : query, query_len,
    1366             :                              &query_offset, &gc_count);
    1367             : 
    1368             :         /*
    1369             :          * Determine whether we need to garbage collect external query texts
    1370             :          * while the shared lock is still held.  This micro-optimization
    1371             :          * avoids taking the time to decide this while holding exclusive lock.
    1372             :          */
    1373       55252 :         do_gc = need_gc_qtexts();
    1374             : 
    1375             :         /* Need exclusive lock to make a new hashtable entry - promote */
    1376       55252 :         LWLockRelease(pgss->lock);
    1377       55252 :         LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
    1378             : 
    1379             :         /*
    1380             :          * A garbage collection may have occurred while we weren't holding the
    1381             :          * lock.  In the unlikely event that this happens, the query text we
    1382             :          * stored above will have been garbage collected, so write it again.
    1383             :          * This should be infrequent enough that doing it while holding
    1384             :          * exclusive lock isn't a performance problem.
    1385             :          */
    1386       55252 :         if (!stored || pgss->gc_count != gc_count)
    1387           0 :             stored = qtext_store(norm_query ? norm_query : query, query_len,
    1388             :                                  &query_offset, NULL);
    1389             : 
    1390             :         /* If we failed to write to the text file, give up */
    1391       55252 :         if (!stored)
    1392           0 :             goto done;
    1393             : 
    1394             :         /* OK to create a new hashtable entry */
    1395       55252 :         entry = entry_alloc(&key, query_offset, query_len, encoding,
    1396             :                             jstate != NULL);
    1397             : 
    1398             :         /* If needed, perform garbage collection while exclusive lock held */
    1399       55252 :         if (do_gc)
    1400           0 :             gc_qtexts();
    1401             :     }
    1402             : 
    1403             :     /* Increment the counts, except when jstate is not NULL */
    1404      186316 :     if (!jstate)
    1405             :     {
    1406             :         Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
    1407             : 
    1408             :         /*
    1409             :          * Grab the spinlock while updating the counters (see comment about
    1410             :          * locking rules at the head of the file)
    1411             :          */
    1412      117452 :         SpinLockAcquire(&entry->mutex);
    1413             : 
    1414             :         /* "Unstick" entry if it was previously sticky */
    1415      117452 :         if (IS_STICKY(entry->counters))
    1416       53812 :             entry->counters.usage = USAGE_INIT;
    1417             : 
    1418      117452 :         entry->counters.calls[kind] += 1;
    1419      117452 :         entry->counters.total_time[kind] += total_time;
    1420             : 
    1421      117452 :         if (entry->counters.calls[kind] == 1)
    1422             :         {
    1423       53952 :             entry->counters.min_time[kind] = total_time;
    1424       53952 :             entry->counters.max_time[kind] = total_time;
    1425       53952 :             entry->counters.mean_time[kind] = total_time;
    1426             :         }
    1427             :         else
    1428             :         {
    1429             :             /*
    1430             :              * Welford's method for accurately computing variance. See
    1431             :              * <http://www.johndcook.com/blog/standard_deviation/>
    1432             :              */
    1433       63500 :             double      old_mean = entry->counters.mean_time[kind];
    1434             : 
    1435       63500 :             entry->counters.mean_time[kind] +=
    1436       63500 :                 (total_time - old_mean) / entry->counters.calls[kind];
    1437       63500 :             entry->counters.sum_var_time[kind] +=
    1438       63500 :                 (total_time - old_mean) * (total_time - entry->counters.mean_time[kind]);
    1439             : 
    1440             :             /*
    1441             :              * Calculate min and max time. min = 0 and max = 0 means that the
    1442             :              * min/max statistics were reset
    1443             :              */
    1444       63500 :             if (entry->counters.min_time[kind] == 0
    1445           4 :                 && entry->counters.max_time[kind] == 0)
    1446             :             {
    1447           4 :                 entry->counters.min_time[kind] = total_time;
    1448           4 :                 entry->counters.max_time[kind] = total_time;
    1449             :             }
    1450             :             else
    1451             :             {
    1452       63496 :                 if (entry->counters.min_time[kind] > total_time)
    1453       12870 :                     entry->counters.min_time[kind] = total_time;
    1454       63496 :                 if (entry->counters.max_time[kind] < total_time)
    1455        6118 :                     entry->counters.max_time[kind] = total_time;
    1456             :             }
    1457             :         }
    1458      117452 :         entry->counters.rows += rows;
    1459      117452 :         entry->counters.shared_blks_hit += bufusage->shared_blks_hit;
    1460      117452 :         entry->counters.shared_blks_read += bufusage->shared_blks_read;
    1461      117452 :         entry->counters.shared_blks_dirtied += bufusage->shared_blks_dirtied;
    1462      117452 :         entry->counters.shared_blks_written += bufusage->shared_blks_written;
    1463      117452 :         entry->counters.local_blks_hit += bufusage->local_blks_hit;
    1464      117452 :         entry->counters.local_blks_read += bufusage->local_blks_read;
    1465      117452 :         entry->counters.local_blks_dirtied += bufusage->local_blks_dirtied;
    1466      117452 :         entry->counters.local_blks_written += bufusage->local_blks_written;
    1467      117452 :         entry->counters.temp_blks_read += bufusage->temp_blks_read;
    1468      117452 :         entry->counters.temp_blks_written += bufusage->temp_blks_written;
    1469      117452 :         entry->counters.shared_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_read_time);
    1470      117452 :         entry->counters.shared_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_write_time);
    1471      117452 :         entry->counters.local_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_read_time);
    1472      117452 :         entry->counters.local_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_write_time);
    1473      117452 :         entry->counters.temp_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_read_time);
    1474      117452 :         entry->counters.temp_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_write_time);
    1475      117452 :         entry->counters.usage += USAGE_EXEC(total_time);
    1476      117452 :         entry->counters.wal_records += walusage->wal_records;
    1477      117452 :         entry->counters.wal_fpi += walusage->wal_fpi;
    1478      117452 :         entry->counters.wal_bytes += walusage->wal_bytes;
    1479      117452 :         entry->counters.wal_buffers_full += walusage->wal_buffers_full;
    1480      117452 :         if (jitusage)
    1481             :         {
    1482         158 :             entry->counters.jit_functions += jitusage->created_functions;
    1483         158 :             entry->counters.jit_generation_time += INSTR_TIME_GET_MILLISEC(jitusage->generation_counter);
    1484             : 
    1485         158 :             if (INSTR_TIME_GET_MILLISEC(jitusage->deform_counter))
    1486         154 :                 entry->counters.jit_deform_count++;
    1487         158 :             entry->counters.jit_deform_time += INSTR_TIME_GET_MILLISEC(jitusage->deform_counter);
    1488             : 
    1489         158 :             if (INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter))
    1490          74 :                 entry->counters.jit_inlining_count++;
    1491         158 :             entry->counters.jit_inlining_time += INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter);
    1492             : 
    1493         158 :             if (INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter))
    1494         154 :                 entry->counters.jit_optimization_count++;
    1495         158 :             entry->counters.jit_optimization_time += INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter);
    1496             : 
    1497         158 :             if (INSTR_TIME_GET_MILLISEC(jitusage->emission_counter))
    1498         154 :                 entry->counters.jit_emission_count++;
    1499         158 :             entry->counters.jit_emission_time += INSTR_TIME_GET_MILLISEC(jitusage->emission_counter);
    1500             :         }
    1501             : 
    1502             :         /* parallel worker counters */
    1503      117452 :         entry->counters.parallel_workers_to_launch += parallel_workers_to_launch;
    1504      117452 :         entry->counters.parallel_workers_launched += parallel_workers_launched;
    1505             : 
    1506      117452 :         SpinLockRelease(&entry->mutex);
    1507             :     }
    1508             : 
    1509       68864 : done:
    1510      186316 :     LWLockRelease(pgss->lock);
    1511             : 
    1512             :     /* We postpone this clean-up until we're out of the lock */
    1513      186316 :     if (norm_query)
    1514       20424 :         pfree(norm_query);
    1515             : }
    1516             : 
    1517             : /*
    1518             :  * Reset statement statistics corresponding to userid, dbid, and queryid.
    1519             :  */
    1520             : Datum
    1521           2 : pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
    1522             : {
    1523             :     Oid         userid;
    1524             :     Oid         dbid;
    1525             :     uint64      queryid;
    1526             : 
    1527           2 :     userid = PG_GETARG_OID(0);
    1528           2 :     dbid = PG_GETARG_OID(1);
    1529           2 :     queryid = (uint64) PG_GETARG_INT64(2);
    1530             : 
    1531           2 :     entry_reset(userid, dbid, queryid, false);
    1532             : 
    1533           2 :     PG_RETURN_VOID();
    1534             : }
    1535             : 
    1536             : Datum
    1537         188 : pg_stat_statements_reset_1_11(PG_FUNCTION_ARGS)
    1538             : {
    1539             :     Oid         userid;
    1540             :     Oid         dbid;
    1541             :     uint64      queryid;
    1542             :     bool        minmax_only;
    1543             : 
    1544         188 :     userid = PG_GETARG_OID(0);
    1545         188 :     dbid = PG_GETARG_OID(1);
    1546         188 :     queryid = (uint64) PG_GETARG_INT64(2);
    1547         188 :     minmax_only = PG_GETARG_BOOL(3);
    1548             : 
    1549         188 :     PG_RETURN_TIMESTAMPTZ(entry_reset(userid, dbid, queryid, minmax_only));
    1550             : }
    1551             : 
    1552             : /*
    1553             :  * Reset statement statistics.
    1554             :  */
    1555             : Datum
    1556           2 : pg_stat_statements_reset(PG_FUNCTION_ARGS)
    1557             : {
    1558           2 :     entry_reset(0, 0, 0, false);
    1559             : 
    1560           2 :     PG_RETURN_VOID();
    1561             : }
    1562             : 
    1563             : /* Number of output arguments (columns) for various API versions */
    1564             : #define PG_STAT_STATEMENTS_COLS_V1_0    14
    1565             : #define PG_STAT_STATEMENTS_COLS_V1_1    18
    1566             : #define PG_STAT_STATEMENTS_COLS_V1_2    19
    1567             : #define PG_STAT_STATEMENTS_COLS_V1_3    23
    1568             : #define PG_STAT_STATEMENTS_COLS_V1_8    32
    1569             : #define PG_STAT_STATEMENTS_COLS_V1_9    33
    1570             : #define PG_STAT_STATEMENTS_COLS_V1_10   43
    1571             : #define PG_STAT_STATEMENTS_COLS_V1_11   49
    1572             : #define PG_STAT_STATEMENTS_COLS_V1_12   52
    1573             : #define PG_STAT_STATEMENTS_COLS         52  /* maximum of above */
    1574             : 
    1575             : /*
    1576             :  * Retrieve statement statistics.
    1577             :  *
    1578             :  * The SQL API of this function has changed multiple times, and will likely
    1579             :  * do so again in future.  To support the case where a newer version of this
    1580             :  * loadable module is being used with an old SQL declaration of the function,
    1581             :  * we continue to support the older API versions.  For 1.2 and later, the
    1582             :  * expected API version is identified by embedding it in the C name of the
    1583             :  * function.  Unfortunately we weren't bright enough to do that for 1.1.
    1584             :  */
    1585             : Datum
    1586         210 : pg_stat_statements_1_12(PG_FUNCTION_ARGS)
    1587             : {
    1588         210 :     bool        showtext = PG_GETARG_BOOL(0);
    1589             : 
    1590         210 :     pg_stat_statements_internal(fcinfo, PGSS_V1_12, showtext);
    1591             : 
    1592         210 :     return (Datum) 0;
    1593             : }
    1594             : 
    1595             : Datum
    1596           2 : pg_stat_statements_1_11(PG_FUNCTION_ARGS)
    1597             : {
    1598           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1599             : 
    1600           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_11, showtext);
    1601             : 
    1602           2 :     return (Datum) 0;
    1603             : }
    1604             : 
    1605             : Datum
    1606           2 : pg_stat_statements_1_10(PG_FUNCTION_ARGS)
    1607             : {
    1608           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1609             : 
    1610           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_10, showtext);
    1611             : 
    1612           2 :     return (Datum) 0;
    1613             : }
    1614             : 
    1615             : Datum
    1616           2 : pg_stat_statements_1_9(PG_FUNCTION_ARGS)
    1617             : {
    1618           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1619             : 
    1620           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_9, showtext);
    1621             : 
    1622           2 :     return (Datum) 0;
    1623             : }
    1624             : 
    1625             : Datum
    1626           2 : pg_stat_statements_1_8(PG_FUNCTION_ARGS)
    1627             : {
    1628           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1629             : 
    1630           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
    1631             : 
    1632           2 :     return (Datum) 0;
    1633             : }
    1634             : 
    1635             : Datum
    1636           2 : pg_stat_statements_1_3(PG_FUNCTION_ARGS)
    1637             : {
    1638           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1639             : 
    1640           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
    1641             : 
    1642           2 :     return (Datum) 0;
    1643             : }
    1644             : 
    1645             : Datum
    1646           0 : pg_stat_statements_1_2(PG_FUNCTION_ARGS)
    1647             : {
    1648           0 :     bool        showtext = PG_GETARG_BOOL(0);
    1649             : 
    1650           0 :     pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
    1651             : 
    1652           0 :     return (Datum) 0;
    1653             : }
    1654             : 
    1655             : /*
    1656             :  * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
    1657             :  * This can be removed someday, perhaps.
    1658             :  */
    1659             : Datum
    1660           0 : pg_stat_statements(PG_FUNCTION_ARGS)
    1661             : {
    1662             :     /* If it's really API 1.1, we'll figure that out below */
    1663           0 :     pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
    1664             : 
    1665           0 :     return (Datum) 0;
    1666             : }
    1667             : 
    1668             : /* Common code for all versions of pg_stat_statements() */
    1669             : static void
    1670         220 : pg_stat_statements_internal(FunctionCallInfo fcinfo,
    1671             :                             pgssVersion api_version,
    1672             :                             bool showtext)
    1673             : {
    1674         220 :     ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
    1675         220 :     Oid         userid = GetUserId();
    1676         220 :     bool        is_allowed_role = false;
    1677         220 :     char       *qbuffer = NULL;
    1678         220 :     Size        qbuffer_size = 0;
    1679         220 :     Size        extent = 0;
    1680         220 :     int         gc_count = 0;
    1681             :     HASH_SEQ_STATUS hash_seq;
    1682             :     pgssEntry  *entry;
    1683             : 
    1684             :     /*
    1685             :      * Superusers or roles with the privileges of pg_read_all_stats members
    1686             :      * are allowed
    1687             :      */
    1688         220 :     is_allowed_role = has_privs_of_role(userid, ROLE_PG_READ_ALL_STATS);
    1689             : 
    1690             :     /* hash table must exist already */
    1691         220 :     if (!pgss || !pgss_hash)
    1692           0 :         ereport(ERROR,
    1693             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    1694             :                  errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
    1695             : 
    1696         220 :     InitMaterializedSRF(fcinfo, 0);
    1697             : 
    1698             :     /*
    1699             :      * Check we have the expected number of output arguments.  Aside from
    1700             :      * being a good safety check, we need a kluge here to detect API version
    1701             :      * 1.1, which was wedged into the code in an ill-considered way.
    1702             :      */
    1703         220 :     switch (rsinfo->setDesc->natts)
    1704             :     {
    1705           0 :         case PG_STAT_STATEMENTS_COLS_V1_0:
    1706           0 :             if (api_version != PGSS_V1_0)
    1707           0 :                 elog(ERROR, "incorrect number of output arguments");
    1708           0 :             break;
    1709           0 :         case PG_STAT_STATEMENTS_COLS_V1_1:
    1710             :             /* pg_stat_statements() should have told us 1.0 */
    1711           0 :             if (api_version != PGSS_V1_0)
    1712           0 :                 elog(ERROR, "incorrect number of output arguments");
    1713           0 :             api_version = PGSS_V1_1;
    1714           0 :             break;
    1715           0 :         case PG_STAT_STATEMENTS_COLS_V1_2:
    1716           0 :             if (api_version != PGSS_V1_2)
    1717           0 :                 elog(ERROR, "incorrect number of output arguments");
    1718           0 :             break;
    1719           2 :         case PG_STAT_STATEMENTS_COLS_V1_3:
    1720           2 :             if (api_version != PGSS_V1_3)
    1721           0 :                 elog(ERROR, "incorrect number of output arguments");
    1722           2 :             break;
    1723           2 :         case PG_STAT_STATEMENTS_COLS_V1_8:
    1724           2 :             if (api_version != PGSS_V1_8)
    1725           0 :                 elog(ERROR, "incorrect number of output arguments");
    1726           2 :             break;
    1727           2 :         case PG_STAT_STATEMENTS_COLS_V1_9:
    1728           2 :             if (api_version != PGSS_V1_9)
    1729           0 :                 elog(ERROR, "incorrect number of output arguments");
    1730           2 :             break;
    1731           2 :         case PG_STAT_STATEMENTS_COLS_V1_10:
    1732           2 :             if (api_version != PGSS_V1_10)
    1733           0 :                 elog(ERROR, "incorrect number of output arguments");
    1734           2 :             break;
    1735           2 :         case PG_STAT_STATEMENTS_COLS_V1_11:
    1736           2 :             if (api_version != PGSS_V1_11)
    1737           0 :                 elog(ERROR, "incorrect number of output arguments");
    1738           2 :             break;
    1739         210 :         case PG_STAT_STATEMENTS_COLS_V1_12:
    1740         210 :             if (api_version != PGSS_V1_12)
    1741           0 :                 elog(ERROR, "incorrect number of output arguments");
    1742         210 :             break;
    1743           0 :         default:
    1744           0 :             elog(ERROR, "incorrect number of output arguments");
    1745             :     }
    1746             : 
    1747             :     /*
    1748             :      * We'd like to load the query text file (if needed) while not holding any
    1749             :      * lock on pgss->lock.  In the worst case we'll have to do this again
    1750             :      * after we have the lock, but it's unlikely enough to make this a win
    1751             :      * despite occasional duplicated work.  We need to reload if anybody
    1752             :      * writes to the file (either a retail qtext_store(), or a garbage
    1753             :      * collection) between this point and where we've gotten shared lock.  If
    1754             :      * a qtext_store is actually in progress when we look, we might as well
    1755             :      * skip the speculative load entirely.
    1756             :      */
    1757         220 :     if (showtext)
    1758             :     {
    1759             :         int         n_writers;
    1760             : 
    1761             :         /* Take the mutex so we can examine variables */
    1762         220 :         SpinLockAcquire(&pgss->mutex);
    1763         220 :         extent = pgss->extent;
    1764         220 :         n_writers = pgss->n_writers;
    1765         220 :         gc_count = pgss->gc_count;
    1766         220 :         SpinLockRelease(&pgss->mutex);
    1767             : 
    1768             :         /* No point in loading file now if there are active writers */
    1769         220 :         if (n_writers == 0)
    1770         220 :             qbuffer = qtext_load_file(&qbuffer_size);
    1771             :     }
    1772             : 
    1773             :     /*
    1774             :      * Get shared lock, load or reload the query text file if we must, and
    1775             :      * iterate over the hashtable entries.
    1776             :      *
    1777             :      * With a large hash table, we might be holding the lock rather longer
    1778             :      * than one could wish.  However, this only blocks creation of new hash
    1779             :      * table entries, and the larger the hash table the less likely that is to
    1780             :      * be needed.  So we can hope this is okay.  Perhaps someday we'll decide
    1781             :      * we need to partition the hash table to limit the time spent holding any
    1782             :      * one lock.
    1783             :      */
    1784         220 :     LWLockAcquire(pgss->lock, LW_SHARED);
    1785             : 
    1786         220 :     if (showtext)
    1787             :     {
    1788             :         /*
    1789             :          * Here it is safe to examine extent and gc_count without taking the
    1790             :          * mutex.  Note that although other processes might change
    1791             :          * pgss->extent just after we look at it, the strings they then write
    1792             :          * into the file cannot yet be referenced in the hashtable, so we
    1793             :          * don't care whether we see them or not.
    1794             :          *
    1795             :          * If qtext_load_file fails, we just press on; we'll return NULL for
    1796             :          * every query text.
    1797             :          */
    1798         220 :         if (qbuffer == NULL ||
    1799         220 :             pgss->extent != extent ||
    1800         220 :             pgss->gc_count != gc_count)
    1801             :         {
    1802           0 :             free(qbuffer);
    1803           0 :             qbuffer = qtext_load_file(&qbuffer_size);
    1804             :         }
    1805             :     }
    1806             : 
    1807         220 :     hash_seq_init(&hash_seq, pgss_hash);
    1808       52338 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    1809             :     {
    1810             :         Datum       values[PG_STAT_STATEMENTS_COLS];
    1811             :         bool        nulls[PG_STAT_STATEMENTS_COLS];
    1812       52118 :         int         i = 0;
    1813             :         Counters    tmp;
    1814             :         double      stddev;
    1815       52118 :         int64       queryid = entry->key.queryid;
    1816             :         TimestampTz stats_since;
    1817             :         TimestampTz minmax_stats_since;
    1818             : 
    1819       52118 :         memset(values, 0, sizeof(values));
    1820       52118 :         memset(nulls, 0, sizeof(nulls));
    1821             : 
    1822       52118 :         values[i++] = ObjectIdGetDatum(entry->key.userid);
    1823       52118 :         values[i++] = ObjectIdGetDatum(entry->key.dbid);
    1824       52118 :         if (api_version >= PGSS_V1_9)
    1825       52094 :             values[i++] = BoolGetDatum(entry->key.toplevel);
    1826             : 
    1827       52118 :         if (is_allowed_role || entry->key.userid == userid)
    1828             :         {
    1829       52110 :             if (api_version >= PGSS_V1_2)
    1830       52110 :                 values[i++] = Int64GetDatumFast(queryid);
    1831             : 
    1832       52110 :             if (showtext)
    1833             :             {
    1834       52110 :                 char       *qstr = qtext_fetch(entry->query_offset,
    1835             :                                                entry->query_len,
    1836             :                                                qbuffer,
    1837             :                                                qbuffer_size);
    1838             : 
    1839       52110 :                 if (qstr)
    1840             :                 {
    1841             :                     char       *enc;
    1842             : 
    1843       52110 :                     enc = pg_any_to_server(qstr,
    1844             :                                            entry->query_len,
    1845             :                                            entry->encoding);
    1846             : 
    1847       52110 :                     values[i++] = CStringGetTextDatum(enc);
    1848             : 
    1849       52110 :                     if (enc != qstr)
    1850           0 :                         pfree(enc);
    1851             :                 }
    1852             :                 else
    1853             :                 {
    1854             :                     /* Just return a null if we fail to find the text */
    1855           0 :                     nulls[i++] = true;
    1856             :                 }
    1857             :             }
    1858             :             else
    1859             :             {
    1860             :                 /* Query text not requested */
    1861           0 :                 nulls[i++] = true;
    1862             :             }
    1863             :         }
    1864             :         else
    1865             :         {
    1866             :             /* Don't show queryid */
    1867           8 :             if (api_version >= PGSS_V1_2)
    1868           8 :                 nulls[i++] = true;
    1869             : 
    1870             :             /*
    1871             :              * Don't show query text, but hint as to the reason for not doing
    1872             :              * so if it was requested
    1873             :              */
    1874           8 :             if (showtext)
    1875           8 :                 values[i++] = CStringGetTextDatum("<insufficient privilege>");
    1876             :             else
    1877           0 :                 nulls[i++] = true;
    1878             :         }
    1879             : 
    1880             :         /* copy counters to a local variable to keep locking time short */
    1881       52118 :         SpinLockAcquire(&entry->mutex);
    1882       52118 :         tmp = entry->counters;
    1883       52118 :         SpinLockRelease(&entry->mutex);
    1884             : 
    1885             :         /*
    1886             :          * The spinlock is not required when reading these two as they are
    1887             :          * always updated when holding pgss->lock exclusively.
    1888             :          */
    1889       52118 :         stats_since = entry->stats_since;
    1890       52118 :         minmax_stats_since = entry->minmax_stats_since;
    1891             : 
    1892             :         /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
    1893       52118 :         if (IS_STICKY(tmp))
    1894          72 :             continue;
    1895             : 
    1896             :         /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
    1897      156138 :         for (int kind = 0; kind < PGSS_NUMKIND; kind++)
    1898             :         {
    1899      104092 :             if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
    1900             :             {
    1901      104084 :                 values[i++] = Int64GetDatumFast(tmp.calls[kind]);
    1902      104084 :                 values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
    1903             :             }
    1904             : 
    1905      104092 :             if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
    1906             :                 api_version >= PGSS_V1_8)
    1907             :             {
    1908      104084 :                 values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
    1909      104084 :                 values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
    1910      104084 :                 values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
    1911             : 
    1912             :                 /*
    1913             :                  * Note we are calculating the population variance here, not
    1914             :                  * the sample variance, as we have data for the whole
    1915             :                  * population, so Bessel's correction is not used, and we
    1916             :                  * don't divide by tmp.calls - 1.
    1917             :                  */
    1918      104084 :                 if (tmp.calls[kind] > 1)
    1919        9816 :                     stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
    1920             :                 else
    1921       94268 :                     stddev = 0.0;
    1922      104084 :                 values[i++] = Float8GetDatumFast(stddev);
    1923             :             }
    1924             :         }
    1925       52046 :         values[i++] = Int64GetDatumFast(tmp.rows);
    1926       52046 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
    1927       52046 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
    1928       52046 :         if (api_version >= PGSS_V1_1)
    1929       52046 :             values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
    1930       52046 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
    1931       52046 :         values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
    1932       52046 :         values[i++] = Int64GetDatumFast(tmp.local_blks_read);
    1933       52046 :         if (api_version >= PGSS_V1_1)
    1934       52046 :             values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
    1935       52046 :         values[i++] = Int64GetDatumFast(tmp.local_blks_written);
    1936       52046 :         values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
    1937       52046 :         values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
    1938       52046 :         if (api_version >= PGSS_V1_1)
    1939             :         {
    1940       52046 :             values[i++] = Float8GetDatumFast(tmp.shared_blk_read_time);
    1941       52046 :             values[i++] = Float8GetDatumFast(tmp.shared_blk_write_time);
    1942             :         }
    1943       52046 :         if (api_version >= PGSS_V1_11)
    1944             :         {
    1945       51990 :             values[i++] = Float8GetDatumFast(tmp.local_blk_read_time);
    1946       51990 :             values[i++] = Float8GetDatumFast(tmp.local_blk_write_time);
    1947             :         }
    1948       52046 :         if (api_version >= PGSS_V1_10)
    1949             :         {
    1950       52008 :             values[i++] = Float8GetDatumFast(tmp.temp_blk_read_time);
    1951       52008 :             values[i++] = Float8GetDatumFast(tmp.temp_blk_write_time);
    1952             :         }
    1953       52046 :         if (api_version >= PGSS_V1_8)
    1954             :         {
    1955             :             char        buf[256];
    1956             :             Datum       wal_bytes;
    1957             : 
    1958       52038 :             values[i++] = Int64GetDatumFast(tmp.wal_records);
    1959       52038 :             values[i++] = Int64GetDatumFast(tmp.wal_fpi);
    1960             : 
    1961       52038 :             snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
    1962             : 
    1963             :             /* Convert to numeric. */
    1964       52038 :             wal_bytes = DirectFunctionCall3(numeric_in,
    1965             :                                             CStringGetDatum(buf),
    1966             :                                             ObjectIdGetDatum(0),
    1967             :                                             Int32GetDatum(-1));
    1968       52038 :             values[i++] = wal_bytes;
    1969             :         }
    1970       52046 :         if (api_version >= PGSS_V1_12)
    1971             :         {
    1972       51970 :             values[i++] = Int64GetDatumFast(tmp.wal_buffers_full);
    1973             :         }
    1974       52046 :         if (api_version >= PGSS_V1_10)
    1975             :         {
    1976       52008 :             values[i++] = Int64GetDatumFast(tmp.jit_functions);
    1977       52008 :             values[i++] = Float8GetDatumFast(tmp.jit_generation_time);
    1978       52008 :             values[i++] = Int64GetDatumFast(tmp.jit_inlining_count);
    1979       52008 :             values[i++] = Float8GetDatumFast(tmp.jit_inlining_time);
    1980       52008 :             values[i++] = Int64GetDatumFast(tmp.jit_optimization_count);
    1981       52008 :             values[i++] = Float8GetDatumFast(tmp.jit_optimization_time);
    1982       52008 :             values[i++] = Int64GetDatumFast(tmp.jit_emission_count);
    1983       52008 :             values[i++] = Float8GetDatumFast(tmp.jit_emission_time);
    1984             :         }
    1985       52046 :         if (api_version >= PGSS_V1_11)
    1986             :         {
    1987       51990 :             values[i++] = Int64GetDatumFast(tmp.jit_deform_count);
    1988       51990 :             values[i++] = Float8GetDatumFast(tmp.jit_deform_time);
    1989             :         }
    1990       52046 :         if (api_version >= PGSS_V1_12)
    1991             :         {
    1992       51970 :             values[i++] = Int64GetDatumFast(tmp.parallel_workers_to_launch);
    1993       51970 :             values[i++] = Int64GetDatumFast(tmp.parallel_workers_launched);
    1994             :         }
    1995       52046 :         if (api_version >= PGSS_V1_11)
    1996             :         {
    1997       51990 :             values[i++] = TimestampTzGetDatum(stats_since);
    1998       51990 :             values[i++] = TimestampTzGetDatum(minmax_stats_since);
    1999             :         }
    2000             : 
    2001             :         Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
    2002             :                      api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
    2003             :                      api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
    2004             :                      api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
    2005             :                      api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
    2006             :                      api_version == PGSS_V1_9 ? PG_STAT_STATEMENTS_COLS_V1_9 :
    2007             :                      api_version == PGSS_V1_10 ? PG_STAT_STATEMENTS_COLS_V1_10 :
    2008             :                      api_version == PGSS_V1_11 ? PG_STAT_STATEMENTS_COLS_V1_11 :
    2009             :                      api_version == PGSS_V1_12 ? PG_STAT_STATEMENTS_COLS_V1_12 :
    2010             :                      -1 /* fail if you forget to update this assert */ ));
    2011             : 
    2012       52046 :         tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
    2013             :     }
    2014             : 
    2015         220 :     LWLockRelease(pgss->lock);
    2016             : 
    2017         220 :     free(qbuffer);
    2018         220 : }
    2019             : 
    2020             : /* Number of output arguments (columns) for pg_stat_statements_info */
    2021             : #define PG_STAT_STATEMENTS_INFO_COLS    2
    2022             : 
    2023             : /*
    2024             :  * Return statistics of pg_stat_statements.
    2025             :  */
    2026             : Datum
    2027           4 : pg_stat_statements_info(PG_FUNCTION_ARGS)
    2028             : {
    2029             :     pgssGlobalStats stats;
    2030             :     TupleDesc   tupdesc;
    2031           4 :     Datum       values[PG_STAT_STATEMENTS_INFO_COLS] = {0};
    2032           4 :     bool        nulls[PG_STAT_STATEMENTS_INFO_COLS] = {0};
    2033             : 
    2034           4 :     if (!pgss || !pgss_hash)
    2035           0 :         ereport(ERROR,
    2036             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    2037             :                  errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
    2038             : 
    2039             :     /* Build a tuple descriptor for our result type */
    2040           4 :     if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
    2041           0 :         elog(ERROR, "return type must be a row type");
    2042             : 
    2043             :     /* Read global statistics for pg_stat_statements */
    2044           4 :     SpinLockAcquire(&pgss->mutex);
    2045           4 :     stats = pgss->stats;
    2046           4 :     SpinLockRelease(&pgss->mutex);
    2047             : 
    2048           4 :     values[0] = Int64GetDatum(stats.dealloc);
    2049           4 :     values[1] = TimestampTzGetDatum(stats.stats_reset);
    2050             : 
    2051           4 :     PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
    2052             : }
    2053             : 
    2054             : /*
    2055             :  * Estimate shared memory space needed.
    2056             :  */
    2057             : static Size
    2058          14 : pgss_memsize(void)
    2059             : {
    2060             :     Size        size;
    2061             : 
    2062          14 :     size = MAXALIGN(sizeof(pgssSharedState));
    2063          14 :     size = add_size(size, hash_estimate_size(pgss_max, sizeof(pgssEntry)));
    2064             : 
    2065          14 :     return size;
    2066             : }
    2067             : 
    2068             : /*
    2069             :  * Allocate a new hashtable entry.
    2070             :  * caller must hold an exclusive lock on pgss->lock
    2071             :  *
    2072             :  * "query" need not be null-terminated; we rely on query_len instead
    2073             :  *
    2074             :  * If "sticky" is true, make the new entry artificially sticky so that it will
    2075             :  * probably still be there when the query finishes execution.  We do this by
    2076             :  * giving it a median usage value rather than the normal value.  (Strictly
    2077             :  * speaking, query strings are normalized on a best effort basis, though it
    2078             :  * would be difficult to demonstrate this even under artificial conditions.)
    2079             :  *
    2080             :  * Note: despite needing exclusive lock, it's not an error for the target
    2081             :  * entry to already exist.  This is because pgss_store releases and
    2082             :  * reacquires lock after failing to find a match; so someone else could
    2083             :  * have made the entry while we waited to get exclusive lock.
    2084             :  */
    2085             : static pgssEntry *
    2086      105668 : entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
    2087             :             bool sticky)
    2088             : {
    2089             :     pgssEntry  *entry;
    2090             :     bool        found;
    2091             : 
    2092             :     /* Make space if needed */
    2093      105668 :     while (hash_get_num_entries(pgss_hash) >= pgss_max)
    2094           0 :         entry_dealloc();
    2095             : 
    2096             :     /* Find or create an entry with desired hash code */
    2097      105668 :     entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
    2098             : 
    2099      105668 :     if (!found)
    2100             :     {
    2101             :         /* New entry, initialize it */
    2102             : 
    2103             :         /* reset the statistics */
    2104      105668 :         memset(&entry->counters, 0, sizeof(Counters));
    2105             :         /* set the appropriate initial usage count */
    2106      105668 :         entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
    2107             :         /* re-initialize the mutex each time ... we assume no one using it */
    2108      105668 :         SpinLockInit(&entry->mutex);
    2109             :         /* ... and don't forget the query text metadata */
    2110             :         Assert(query_len >= 0);
    2111      105668 :         entry->query_offset = query_offset;
    2112      105668 :         entry->query_len = query_len;
    2113      105668 :         entry->encoding = encoding;
    2114      105668 :         entry->stats_since = GetCurrentTimestamp();
    2115      105668 :         entry->minmax_stats_since = entry->stats_since;
    2116             :     }
    2117             : 
    2118      105668 :     return entry;
    2119             : }
    2120             : 
    2121             : /*
    2122             :  * qsort comparator for sorting into increasing usage order
    2123             :  */
    2124             : static int
    2125           0 : entry_cmp(const void *lhs, const void *rhs)
    2126             : {
    2127           0 :     double      l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
    2128           0 :     double      r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
    2129             : 
    2130           0 :     if (l_usage < r_usage)
    2131           0 :         return -1;
    2132           0 :     else if (l_usage > r_usage)
    2133           0 :         return +1;
    2134             :     else
    2135           0 :         return 0;
    2136             : }
    2137             : 
    2138             : /*
    2139             :  * Deallocate least-used entries.
    2140             :  *
    2141             :  * Caller must hold an exclusive lock on pgss->lock.
    2142             :  */
    2143             : static void
    2144           0 : entry_dealloc(void)
    2145             : {
    2146             :     HASH_SEQ_STATUS hash_seq;
    2147             :     pgssEntry **entries;
    2148             :     pgssEntry  *entry;
    2149             :     int         nvictims;
    2150             :     int         i;
    2151             :     Size        tottextlen;
    2152             :     int         nvalidtexts;
    2153             : 
    2154             :     /*
    2155             :      * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
    2156             :      * While we're scanning the table, apply the decay factor to the usage
    2157             :      * values, and update the mean query length.
    2158             :      *
    2159             :      * Note that the mean query length is almost immediately obsolete, since
    2160             :      * we compute it before not after discarding the least-used entries.
    2161             :      * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
    2162             :      * making two passes to get a more current result.  Likewise, the new
    2163             :      * cur_median_usage includes the entries we're about to zap.
    2164             :      */
    2165             : 
    2166           0 :     entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
    2167             : 
    2168           0 :     i = 0;
    2169           0 :     tottextlen = 0;
    2170           0 :     nvalidtexts = 0;
    2171             : 
    2172           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2173           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2174             :     {
    2175           0 :         entries[i++] = entry;
    2176             :         /* "Sticky" entries get a different usage decay rate. */
    2177           0 :         if (IS_STICKY(entry->counters))
    2178           0 :             entry->counters.usage *= STICKY_DECREASE_FACTOR;
    2179             :         else
    2180           0 :             entry->counters.usage *= USAGE_DECREASE_FACTOR;
    2181             :         /* In the mean length computation, ignore dropped texts. */
    2182           0 :         if (entry->query_len >= 0)
    2183             :         {
    2184           0 :             tottextlen += entry->query_len + 1;
    2185           0 :             nvalidtexts++;
    2186             :         }
    2187             :     }
    2188             : 
    2189             :     /* Sort into increasing order by usage */
    2190           0 :     qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
    2191             : 
    2192             :     /* Record the (approximate) median usage */
    2193           0 :     if (i > 0)
    2194           0 :         pgss->cur_median_usage = entries[i / 2]->counters.usage;
    2195             :     /* Record the mean query length */
    2196           0 :     if (nvalidtexts > 0)
    2197           0 :         pgss->mean_query_len = tottextlen / nvalidtexts;
    2198             :     else
    2199           0 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2200             : 
    2201             :     /* Now zap an appropriate fraction of lowest-usage entries */
    2202           0 :     nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
    2203           0 :     nvictims = Min(nvictims, i);
    2204             : 
    2205           0 :     for (i = 0; i < nvictims; i++)
    2206             :     {
    2207           0 :         hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
    2208             :     }
    2209             : 
    2210           0 :     pfree(entries);
    2211             : 
    2212             :     /* Increment the number of times entries are deallocated */
    2213           0 :     SpinLockAcquire(&pgss->mutex);
    2214           0 :     pgss->stats.dealloc += 1;
    2215           0 :     SpinLockRelease(&pgss->mutex);
    2216           0 : }
    2217             : 
    2218             : /*
    2219             :  * Given a query string (not necessarily null-terminated), allocate a new
    2220             :  * entry in the external query text file and store the string there.
    2221             :  *
    2222             :  * If successful, returns true, and stores the new entry's offset in the file
    2223             :  * into *query_offset.  Also, if gc_count isn't NULL, *gc_count is set to the
    2224             :  * number of garbage collections that have occurred so far.
    2225             :  *
    2226             :  * On failure, returns false.
    2227             :  *
    2228             :  * At least a shared lock on pgss->lock must be held by the caller, so as
    2229             :  * to prevent a concurrent garbage collection.  Share-lock-holding callers
    2230             :  * should pass a gc_count pointer to obtain the number of garbage collections,
    2231             :  * so that they can recheck the count after obtaining exclusive lock to
    2232             :  * detect whether a garbage collection occurred (and removed this entry).
    2233             :  */
    2234             : static bool
    2235       55252 : qtext_store(const char *query, int query_len,
    2236             :             Size *query_offset, int *gc_count)
    2237             : {
    2238             :     Size        off;
    2239             :     int         fd;
    2240             : 
    2241             :     /*
    2242             :      * We use a spinlock to protect extent/n_writers/gc_count, so that
    2243             :      * multiple processes may execute this function concurrently.
    2244             :      */
    2245       55252 :     SpinLockAcquire(&pgss->mutex);
    2246       55252 :     off = pgss->extent;
    2247       55252 :     pgss->extent += query_len + 1;
    2248       55252 :     pgss->n_writers++;
    2249       55252 :     if (gc_count)
    2250       55252 :         *gc_count = pgss->gc_count;
    2251       55252 :     SpinLockRelease(&pgss->mutex);
    2252             : 
    2253       55252 :     *query_offset = off;
    2254             : 
    2255             :     /*
    2256             :      * Don't allow the file to grow larger than what qtext_load_file can
    2257             :      * (theoretically) handle.  This has been seen to be reachable on 32-bit
    2258             :      * platforms.
    2259             :      */
    2260       55252 :     if (unlikely(query_len >= MaxAllocHugeSize - off))
    2261             :     {
    2262           0 :         errno = EFBIG;          /* not quite right, but it'll do */
    2263           0 :         fd = -1;
    2264           0 :         goto error;
    2265             :     }
    2266             : 
    2267             :     /* Now write the data into the successfully-reserved part of the file */
    2268       55252 :     fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
    2269       55252 :     if (fd < 0)
    2270           0 :         goto error;
    2271             : 
    2272       55252 :     if (pg_pwrite(fd, query, query_len, off) != query_len)
    2273           0 :         goto error;
    2274       55252 :     if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
    2275           0 :         goto error;
    2276             : 
    2277       55252 :     CloseTransientFile(fd);
    2278             : 
    2279             :     /* Mark our write complete */
    2280       55252 :     SpinLockAcquire(&pgss->mutex);
    2281       55252 :     pgss->n_writers--;
    2282       55252 :     SpinLockRelease(&pgss->mutex);
    2283             : 
    2284       55252 :     return true;
    2285             : 
    2286           0 : error:
    2287           0 :     ereport(LOG,
    2288             :             (errcode_for_file_access(),
    2289             :              errmsg("could not write file \"%s\": %m",
    2290             :                     PGSS_TEXT_FILE)));
    2291             : 
    2292           0 :     if (fd >= 0)
    2293           0 :         CloseTransientFile(fd);
    2294             : 
    2295             :     /* Mark our write complete */
    2296           0 :     SpinLockAcquire(&pgss->mutex);
    2297           0 :     pgss->n_writers--;
    2298           0 :     SpinLockRelease(&pgss->mutex);
    2299             : 
    2300           0 :     return false;
    2301             : }
    2302             : 
    2303             : /*
    2304             :  * Read the external query text file into a malloc'd buffer.
    2305             :  *
    2306             :  * Returns NULL (without throwing an error) if unable to read, eg
    2307             :  * file not there or insufficient memory.
    2308             :  *
    2309             :  * On success, the buffer size is also returned into *buffer_size.
    2310             :  *
    2311             :  * This can be called without any lock on pgss->lock, but in that case
    2312             :  * the caller is responsible for verifying that the result is sane.
    2313             :  */
    2314             : static char *
    2315         230 : qtext_load_file(Size *buffer_size)
    2316             : {
    2317             :     char       *buf;
    2318             :     int         fd;
    2319             :     struct stat stat;
    2320             :     Size        nread;
    2321             : 
    2322         230 :     fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
    2323         230 :     if (fd < 0)
    2324             :     {
    2325           0 :         if (errno != ENOENT)
    2326           0 :             ereport(LOG,
    2327             :                     (errcode_for_file_access(),
    2328             :                      errmsg("could not read file \"%s\": %m",
    2329             :                             PGSS_TEXT_FILE)));
    2330           0 :         return NULL;
    2331             :     }
    2332             : 
    2333             :     /* Get file length */
    2334         230 :     if (fstat(fd, &stat))
    2335             :     {
    2336           0 :         ereport(LOG,
    2337             :                 (errcode_for_file_access(),
    2338             :                  errmsg("could not stat file \"%s\": %m",
    2339             :                         PGSS_TEXT_FILE)));
    2340           0 :         CloseTransientFile(fd);
    2341           0 :         return NULL;
    2342             :     }
    2343             : 
    2344             :     /* Allocate buffer; beware that off_t might be wider than size_t */
    2345         230 :     if (stat.st_size <= MaxAllocHugeSize)
    2346         230 :         buf = (char *) malloc(stat.st_size);
    2347             :     else
    2348           0 :         buf = NULL;
    2349         230 :     if (buf == NULL)
    2350             :     {
    2351           0 :         ereport(LOG,
    2352             :                 (errcode(ERRCODE_OUT_OF_MEMORY),
    2353             :                  errmsg("out of memory"),
    2354             :                  errdetail("Could not allocate enough memory to read file \"%s\".",
    2355             :                            PGSS_TEXT_FILE)));
    2356           0 :         CloseTransientFile(fd);
    2357           0 :         return NULL;
    2358             :     }
    2359             : 
    2360             :     /*
    2361             :      * OK, slurp in the file.  Windows fails if we try to read more than
    2362             :      * INT_MAX bytes at once, and other platforms might not like that either,
    2363             :      * so read a very large file in 1GB segments.
    2364             :      */
    2365         230 :     nread = 0;
    2366         458 :     while (nread < stat.st_size)
    2367             :     {
    2368         228 :         int         toread = Min(1024 * 1024 * 1024, stat.st_size - nread);
    2369             : 
    2370             :         /*
    2371             :          * If we get a short read and errno doesn't get set, the reason is
    2372             :          * probably that garbage collection truncated the file since we did
    2373             :          * the fstat(), so we don't log a complaint --- but we don't return
    2374             :          * the data, either, since it's most likely corrupt due to concurrent
    2375             :          * writes from garbage collection.
    2376             :          */
    2377         228 :         errno = 0;
    2378         228 :         if (read(fd, buf + nread, toread) != toread)
    2379             :         {
    2380           0 :             if (errno)
    2381           0 :                 ereport(LOG,
    2382             :                         (errcode_for_file_access(),
    2383             :                          errmsg("could not read file \"%s\": %m",
    2384             :                                 PGSS_TEXT_FILE)));
    2385           0 :             free(buf);
    2386           0 :             CloseTransientFile(fd);
    2387           0 :             return NULL;
    2388             :         }
    2389         228 :         nread += toread;
    2390             :     }
    2391             : 
    2392         230 :     if (CloseTransientFile(fd) != 0)
    2393           0 :         ereport(LOG,
    2394             :                 (errcode_for_file_access(),
    2395             :                  errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
    2396             : 
    2397         230 :     *buffer_size = nread;
    2398         230 :     return buf;
    2399             : }
    2400             : 
    2401             : /*
    2402             :  * Locate a query text in the file image previously read by qtext_load_file().
    2403             :  *
    2404             :  * We validate the given offset/length, and return NULL if bogus.  Otherwise,
    2405             :  * the result points to a null-terminated string within the buffer.
    2406             :  */
    2407             : static char *
    2408      156288 : qtext_fetch(Size query_offset, int query_len,
    2409             :             char *buffer, Size buffer_size)
    2410             : {
    2411             :     /* File read failed? */
    2412      156288 :     if (buffer == NULL)
    2413           0 :         return NULL;
    2414             :     /* Bogus offset/length? */
    2415      156288 :     if (query_len < 0 ||
    2416      156288 :         query_offset + query_len >= buffer_size)
    2417           0 :         return NULL;
    2418             :     /* As a further sanity check, make sure there's a trailing null */
    2419      156288 :     if (buffer[query_offset + query_len] != '\0')
    2420           0 :         return NULL;
    2421             :     /* Looks OK */
    2422      156288 :     return buffer + query_offset;
    2423             : }
    2424             : 
    2425             : /*
    2426             :  * Do we need to garbage-collect the external query text file?
    2427             :  *
    2428             :  * Caller should hold at least a shared lock on pgss->lock.
    2429             :  */
    2430             : static bool
    2431       55252 : need_gc_qtexts(void)
    2432             : {
    2433             :     Size        extent;
    2434             : 
    2435             :     /* Read shared extent pointer */
    2436       55252 :     SpinLockAcquire(&pgss->mutex);
    2437       55252 :     extent = pgss->extent;
    2438       55252 :     SpinLockRelease(&pgss->mutex);
    2439             : 
    2440             :     /*
    2441             :      * Don't proceed if file does not exceed 512 bytes per possible entry.
    2442             :      *
    2443             :      * Here and in the next test, 32-bit machines have overflow hazards if
    2444             :      * pgss_max and/or mean_query_len are large.  Force the multiplications
    2445             :      * and comparisons to be done in uint64 arithmetic to forestall trouble.
    2446             :      */
    2447       55252 :     if ((uint64) extent < (uint64) 512 * pgss_max)
    2448       55252 :         return false;
    2449             : 
    2450             :     /*
    2451             :      * Don't proceed if file is less than about 50% bloat.  Nothing can or
    2452             :      * should be done in the event of unusually large query texts accounting
    2453             :      * for file's large size.  We go to the trouble of maintaining the mean
    2454             :      * query length in order to prevent garbage collection from thrashing
    2455             :      * uselessly.
    2456             :      */
    2457           0 :     if ((uint64) extent < (uint64) pgss->mean_query_len * pgss_max * 2)
    2458           0 :         return false;
    2459             : 
    2460           0 :     return true;
    2461             : }
    2462             : 
    2463             : /*
    2464             :  * Garbage-collect orphaned query texts in external file.
    2465             :  *
    2466             :  * This won't be called often in the typical case, since it's likely that
    2467             :  * there won't be too much churn, and besides, a similar compaction process
    2468             :  * occurs when serializing to disk at shutdown or as part of resetting.
    2469             :  * Despite this, it seems prudent to plan for the edge case where the file
    2470             :  * becomes unreasonably large, with no other method of compaction likely to
    2471             :  * occur in the foreseeable future.
    2472             :  *
    2473             :  * The caller must hold an exclusive lock on pgss->lock.
    2474             :  *
    2475             :  * At the first sign of trouble we unlink the query text file to get a clean
    2476             :  * slate (although existing statistics are retained), rather than risk
    2477             :  * thrashing by allowing the same problem case to recur indefinitely.
    2478             :  */
    2479             : static void
    2480           0 : gc_qtexts(void)
    2481             : {
    2482             :     char       *qbuffer;
    2483             :     Size        qbuffer_size;
    2484           0 :     FILE       *qfile = NULL;
    2485             :     HASH_SEQ_STATUS hash_seq;
    2486             :     pgssEntry  *entry;
    2487             :     Size        extent;
    2488             :     int         nentries;
    2489             : 
    2490             :     /*
    2491             :      * When called from pgss_store, some other session might have proceeded
    2492             :      * with garbage collection in the no-lock-held interim of lock strength
    2493             :      * escalation.  Check once more that this is actually necessary.
    2494             :      */
    2495           0 :     if (!need_gc_qtexts())
    2496           0 :         return;
    2497             : 
    2498             :     /*
    2499             :      * Load the old texts file.  If we fail (out of memory, for instance),
    2500             :      * invalidate query texts.  Hopefully this is rare.  It might seem better
    2501             :      * to leave things alone on an OOM failure, but the problem is that the
    2502             :      * file is only going to get bigger; hoping for a future non-OOM result is
    2503             :      * risky and can easily lead to complete denial of service.
    2504             :      */
    2505           0 :     qbuffer = qtext_load_file(&qbuffer_size);
    2506           0 :     if (qbuffer == NULL)
    2507           0 :         goto gc_fail;
    2508             : 
    2509             :     /*
    2510             :      * We overwrite the query texts file in place, so as to reduce the risk of
    2511             :      * an out-of-disk-space failure.  Since the file is guaranteed not to get
    2512             :      * larger, this should always work on traditional filesystems; though we
    2513             :      * could still lose on copy-on-write filesystems.
    2514             :      */
    2515           0 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2516           0 :     if (qfile == NULL)
    2517             :     {
    2518           0 :         ereport(LOG,
    2519             :                 (errcode_for_file_access(),
    2520             :                  errmsg("could not write file \"%s\": %m",
    2521             :                         PGSS_TEXT_FILE)));
    2522           0 :         goto gc_fail;
    2523             :     }
    2524             : 
    2525           0 :     extent = 0;
    2526           0 :     nentries = 0;
    2527             : 
    2528           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2529           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2530             :     {
    2531           0 :         int         query_len = entry->query_len;
    2532           0 :         char       *qry = qtext_fetch(entry->query_offset,
    2533             :                                       query_len,
    2534             :                                       qbuffer,
    2535             :                                       qbuffer_size);
    2536             : 
    2537           0 :         if (qry == NULL)
    2538             :         {
    2539             :             /* Trouble ... drop the text */
    2540           0 :             entry->query_offset = 0;
    2541           0 :             entry->query_len = -1;
    2542             :             /* entry will not be counted in mean query length computation */
    2543           0 :             continue;
    2544             :         }
    2545             : 
    2546           0 :         if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
    2547             :         {
    2548           0 :             ereport(LOG,
    2549             :                     (errcode_for_file_access(),
    2550             :                      errmsg("could not write file \"%s\": %m",
    2551             :                             PGSS_TEXT_FILE)));
    2552           0 :             hash_seq_term(&hash_seq);
    2553           0 :             goto gc_fail;
    2554             :         }
    2555             : 
    2556           0 :         entry->query_offset = extent;
    2557           0 :         extent += query_len + 1;
    2558           0 :         nentries++;
    2559             :     }
    2560             : 
    2561             :     /*
    2562             :      * Truncate away any now-unused space.  If this fails for some odd reason,
    2563             :      * we log it, but there's no need to fail.
    2564             :      */
    2565           0 :     if (ftruncate(fileno(qfile), extent) != 0)
    2566           0 :         ereport(LOG,
    2567             :                 (errcode_for_file_access(),
    2568             :                  errmsg("could not truncate file \"%s\": %m",
    2569             :                         PGSS_TEXT_FILE)));
    2570             : 
    2571           0 :     if (FreeFile(qfile))
    2572             :     {
    2573           0 :         ereport(LOG,
    2574             :                 (errcode_for_file_access(),
    2575             :                  errmsg("could not write file \"%s\": %m",
    2576             :                         PGSS_TEXT_FILE)));
    2577           0 :         qfile = NULL;
    2578           0 :         goto gc_fail;
    2579             :     }
    2580             : 
    2581           0 :     elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
    2582             :          pgss->extent, extent);
    2583             : 
    2584             :     /* Reset the shared extent pointer */
    2585           0 :     pgss->extent = extent;
    2586             : 
    2587             :     /*
    2588             :      * Also update the mean query length, to be sure that need_gc_qtexts()
    2589             :      * won't still think we have a problem.
    2590             :      */
    2591           0 :     if (nentries > 0)
    2592           0 :         pgss->mean_query_len = extent / nentries;
    2593             :     else
    2594           0 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2595             : 
    2596           0 :     free(qbuffer);
    2597             : 
    2598             :     /*
    2599             :      * OK, count a garbage collection cycle.  (Note: even though we have
    2600             :      * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
    2601             :      * other processes may examine gc_count while holding only the mutex.
    2602             :      * Also, we have to advance the count *after* we've rewritten the file,
    2603             :      * else other processes might not realize they read a stale file.)
    2604             :      */
    2605           0 :     record_gc_qtexts();
    2606             : 
    2607           0 :     return;
    2608             : 
    2609           0 : gc_fail:
    2610             :     /* clean up resources */
    2611           0 :     if (qfile)
    2612           0 :         FreeFile(qfile);
    2613           0 :     free(qbuffer);
    2614             : 
    2615             :     /*
    2616             :      * Since the contents of the external file are now uncertain, mark all
    2617             :      * hashtable entries as having invalid texts.
    2618             :      */
    2619           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2620           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2621             :     {
    2622           0 :         entry->query_offset = 0;
    2623           0 :         entry->query_len = -1;
    2624             :     }
    2625             : 
    2626             :     /*
    2627             :      * Destroy the query text file and create a new, empty one
    2628             :      */
    2629           0 :     (void) unlink(PGSS_TEXT_FILE);
    2630           0 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2631           0 :     if (qfile == NULL)
    2632           0 :         ereport(LOG,
    2633             :                 (errcode_for_file_access(),
    2634             :                  errmsg("could not recreate file \"%s\": %m",
    2635             :                         PGSS_TEXT_FILE)));
    2636             :     else
    2637           0 :         FreeFile(qfile);
    2638             : 
    2639             :     /* Reset the shared extent pointer */
    2640           0 :     pgss->extent = 0;
    2641             : 
    2642             :     /* Reset mean_query_len to match the new state */
    2643           0 :     pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2644             : 
    2645             :     /*
    2646             :      * Bump the GC count even though we failed.
    2647             :      *
    2648             :      * This is needed to make concurrent readers of file without any lock on
    2649             :      * pgss->lock notice existence of new version of file.  Once readers
    2650             :      * subsequently observe a change in GC count with pgss->lock held, that
    2651             :      * forces a safe reopen of file.  Writers also require that we bump here,
    2652             :      * of course.  (As required by locking protocol, readers and writers don't
    2653             :      * trust earlier file contents until gc_count is found unchanged after
    2654             :      * pgss->lock acquired in shared or exclusive mode respectively.)
    2655             :      */
    2656           0 :     record_gc_qtexts();
    2657             : }
    2658             : 
    2659             : #define SINGLE_ENTRY_RESET(e) \
    2660             : if (e) { \
    2661             :     if (minmax_only) { \
    2662             :         /* When requested reset only min/max statistics of an entry */ \
    2663             :         for (int kind = 0; kind < PGSS_NUMKIND; kind++) \
    2664             :         { \
    2665             :             e->counters.max_time[kind] = 0; \
    2666             :             e->counters.min_time[kind] = 0; \
    2667             :         } \
    2668             :         e->minmax_stats_since = stats_reset; \
    2669             :     } \
    2670             :     else \
    2671             :     { \
    2672             :         /* Remove the key otherwise  */ \
    2673             :         hash_search(pgss_hash, &e->key, HASH_REMOVE, NULL); \
    2674             :         num_remove++; \
    2675             :     } \
    2676             : }
    2677             : 
    2678             : /*
    2679             :  * Reset entries corresponding to parameters passed.
    2680             :  */
    2681             : static TimestampTz
    2682         192 : entry_reset(Oid userid, Oid dbid, uint64 queryid, bool minmax_only)
    2683             : {
    2684             :     HASH_SEQ_STATUS hash_seq;
    2685             :     pgssEntry  *entry;
    2686             :     FILE       *qfile;
    2687             :     long        num_entries;
    2688         192 :     long        num_remove = 0;
    2689             :     pgssHashKey key;
    2690             :     TimestampTz stats_reset;
    2691             : 
    2692         192 :     if (!pgss || !pgss_hash)
    2693           0 :         ereport(ERROR,
    2694             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    2695             :                  errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
    2696             : 
    2697         192 :     LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
    2698         192 :     num_entries = hash_get_num_entries(pgss_hash);
    2699             : 
    2700         192 :     stats_reset = GetCurrentTimestamp();
    2701             : 
    2702         192 :     if (userid != 0 && dbid != 0 && queryid != UINT64CONST(0))
    2703             :     {
    2704             :         /* If all the parameters are available, use the fast path. */
    2705           2 :         memset(&key, 0, sizeof(pgssHashKey));
    2706           2 :         key.userid = userid;
    2707           2 :         key.dbid = dbid;
    2708           2 :         key.queryid = queryid;
    2709             : 
    2710             :         /*
    2711             :          * Reset the entry if it exists, starting with the non-top-level
    2712             :          * entry.
    2713             :          */
    2714           2 :         key.toplevel = false;
    2715           2 :         entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    2716             : 
    2717           2 :         SINGLE_ENTRY_RESET(entry);
    2718             : 
    2719             :         /* Also reset the top-level entry if it exists. */
    2720           2 :         key.toplevel = true;
    2721           2 :         entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    2722             : 
    2723           2 :         SINGLE_ENTRY_RESET(entry);
    2724             :     }
    2725         190 :     else if (userid != 0 || dbid != 0 || queryid != UINT64CONST(0))
    2726             :     {
    2727             :         /* Reset entries corresponding to valid parameters. */
    2728           8 :         hash_seq_init(&hash_seq, pgss_hash);
    2729         102 :         while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2730             :         {
    2731          94 :             if ((!userid || entry->key.userid == userid) &&
    2732          72 :                 (!dbid || entry->key.dbid == dbid) &&
    2733          68 :                 (!queryid || entry->key.queryid == queryid))
    2734             :             {
    2735          14 :                 SINGLE_ENTRY_RESET(entry);
    2736             :             }
    2737             :         }
    2738             :     }
    2739             :     else
    2740             :     {
    2741             :         /* Reset all entries. */
    2742         182 :         hash_seq_init(&hash_seq, pgss_hash);
    2743        1674 :         while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2744             :         {
    2745        1536 :             SINGLE_ENTRY_RESET(entry);
    2746             :         }
    2747             :     }
    2748             : 
    2749             :     /* All entries are removed? */
    2750         192 :     if (num_entries != num_remove)
    2751          12 :         goto release_lock;
    2752             : 
    2753             :     /*
    2754             :      * Reset global statistics for pg_stat_statements since all entries are
    2755             :      * removed.
    2756             :      */
    2757         180 :     SpinLockAcquire(&pgss->mutex);
    2758         180 :     pgss->stats.dealloc = 0;
    2759         180 :     pgss->stats.stats_reset = stats_reset;
    2760         180 :     SpinLockRelease(&pgss->mutex);
    2761             : 
    2762             :     /*
    2763             :      * Write new empty query file, perhaps even creating a new one to recover
    2764             :      * if the file was missing.
    2765             :      */
    2766         180 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2767         180 :     if (qfile == NULL)
    2768             :     {
    2769           0 :         ereport(LOG,
    2770             :                 (errcode_for_file_access(),
    2771             :                  errmsg("could not create file \"%s\": %m",
    2772             :                         PGSS_TEXT_FILE)));
    2773           0 :         goto done;
    2774             :     }
    2775             : 
    2776             :     /* If ftruncate fails, log it, but it's not a fatal problem */
    2777         180 :     if (ftruncate(fileno(qfile), 0) != 0)
    2778           0 :         ereport(LOG,
    2779             :                 (errcode_for_file_access(),
    2780             :                  errmsg("could not truncate file \"%s\": %m",
    2781             :                         PGSS_TEXT_FILE)));
    2782             : 
    2783         180 :     FreeFile(qfile);
    2784             : 
    2785         180 : done:
    2786         180 :     pgss->extent = 0;
    2787             :     /* This counts as a query text garbage collection for our purposes */
    2788         180 :     record_gc_qtexts();
    2789             : 
    2790         192 : release_lock:
    2791         192 :     LWLockRelease(pgss->lock);
    2792             : 
    2793         192 :     return stats_reset;
    2794             : }
    2795             : 
    2796             : /*
    2797             :  * Generate a normalized version of the query string that will be used to
    2798             :  * represent all similar queries.
    2799             :  *
    2800             :  * Note that the normalized representation may well vary depending on
    2801             :  * just which "equivalent" query is used to create the hashtable entry.
    2802             :  * We assume this is OK.
    2803             :  *
    2804             :  * If query_loc > 0, then "query" has been advanced by that much compared to
    2805             :  * the original string start, so we need to translate the provided locations
    2806             :  * to compensate.  (This lets us avoid re-scanning statements before the one
    2807             :  * of interest, so it's worth doing.)
    2808             :  *
    2809             :  * *query_len_p contains the input string length, and is updated with
    2810             :  * the result string length on exit.  The resulting string might be longer
    2811             :  * or shorter depending on what happens with replacement of constants.
    2812             :  *
    2813             :  * Returns a palloc'd string.
    2814             :  */
    2815             : static char *
    2816       20424 : generate_normalized_query(JumbleState *jstate, const char *query,
    2817             :                           int query_loc, int *query_len_p)
    2818             : {
    2819             :     char       *norm_query;
    2820       20424 :     int         query_len = *query_len_p;
    2821             :     int         i,
    2822             :                 norm_query_buflen,  /* Space allowed for norm_query */
    2823             :                 len_to_wrt,     /* Length (in bytes) to write */
    2824       20424 :                 quer_loc = 0,   /* Source query byte location */
    2825       20424 :                 n_quer_loc = 0, /* Normalized query byte location */
    2826       20424 :                 last_off = 0,   /* Offset from start for previous tok */
    2827       20424 :                 last_tok_len = 0;   /* Length (in bytes) of that tok */
    2828       20424 :     bool        in_squashed = false;    /* in a run of squashed consts? */
    2829       20424 :     int         skipped_constants = 0;  /* Position adjustment of later
    2830             :                                          * constants after squashed ones */
    2831             : 
    2832             : 
    2833             :     /*
    2834             :      * Get constants' lengths (core system only gives us locations).  Note
    2835             :      * this also ensures the items are sorted by location.
    2836             :      */
    2837       20424 :     fill_in_constant_lengths(jstate, query, query_loc);
    2838             : 
    2839             :     /*
    2840             :      * Allow for $n symbols to be longer than the constants they replace.
    2841             :      * Constants must take at least one byte in text form, while a $n symbol
    2842             :      * certainly isn't more than 11 bytes, even if n reaches INT_MAX.  We
    2843             :      * could refine that limit based on the max value of n for the current
    2844             :      * query, but it hardly seems worth any extra effort to do so.
    2845             :      *
    2846             :      * Note this also gives enough room for the commented-out ", ..." list
    2847             :      * syntax used by constant squashing.
    2848             :      */
    2849       20424 :     norm_query_buflen = query_len + jstate->clocations_count * 10;
    2850             : 
    2851             :     /* Allocate result buffer */
    2852       20424 :     norm_query = palloc(norm_query_buflen + 1);
    2853             : 
    2854       82312 :     for (i = 0; i < jstate->clocations_count; i++)
    2855             :     {
    2856             :         int         off,        /* Offset from start for cur tok */
    2857             :                     tok_len;    /* Length (in bytes) of that tok */
    2858             : 
    2859       61888 :         off = jstate->clocations[i].location;
    2860             : 
    2861             :         /* Adjust recorded location if we're dealing with partial string */
    2862       61888 :         off -= query_loc;
    2863             : 
    2864       61888 :         tok_len = jstate->clocations[i].length;
    2865             : 
    2866       61888 :         if (tok_len < 0)
    2867         456 :             continue;           /* ignore any duplicates */
    2868             : 
    2869             :         /*
    2870             :          * What to do next depends on whether we're squashing constant lists,
    2871             :          * and whether we're already in a run of such constants.
    2872             :          */
    2873       61432 :         if (!jstate->clocations[i].squashed)
    2874             :         {
    2875             :             /*
    2876             :              * This location corresponds to a constant not to be squashed.
    2877             :              * Print what comes before the constant ...
    2878             :              */
    2879       59040 :             len_to_wrt = off - last_off;
    2880       59040 :             len_to_wrt -= last_tok_len;
    2881             : 
    2882             :             Assert(len_to_wrt >= 0);
    2883             : 
    2884       59040 :             memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2885       59040 :             n_quer_loc += len_to_wrt;
    2886             : 
    2887             :             /* ... and then a param symbol replacing the constant itself */
    2888      118080 :             n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d",
    2889       59040 :                                   i + 1 + jstate->highest_extern_param_id - skipped_constants);
    2890             : 
    2891             :             /* In case previous constants were merged away, stop doing that */
    2892       59040 :             in_squashed = false;
    2893             :         }
    2894        2392 :         else if (!in_squashed)
    2895             :         {
    2896             :             /*
    2897             :              * This location is the start position of a run of constants to be
    2898             :              * squashed, so we need to print the representation of starting a
    2899             :              * group of stashed constants.
    2900             :              *
    2901             :              * Print what comes before the constant ...
    2902             :              */
    2903        1196 :             len_to_wrt = off - last_off;
    2904        1196 :             len_to_wrt -= last_tok_len;
    2905             :             Assert(len_to_wrt >= 0);
    2906             :             Assert(i + 1 < jstate->clocations_count);
    2907             :             Assert(jstate->clocations[i + 1].squashed);
    2908        1196 :             memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2909        1196 :             n_quer_loc += len_to_wrt;
    2910             : 
    2911             :             /* ... and then start a run of squashed constants */
    2912        2392 :             n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d /*, ... */",
    2913        1196 :                                   i + 1 + jstate->highest_extern_param_id - skipped_constants);
    2914             : 
    2915             :             /* The next location will match the block below, to end the run */
    2916        1196 :             in_squashed = true;
    2917             : 
    2918        1196 :             skipped_constants++;
    2919             :         }
    2920             :         else
    2921             :         {
    2922             :             /*
    2923             :              * The second location of a run of squashable elements; this
    2924             :              * indicates its end.
    2925             :              */
    2926        1196 :             in_squashed = false;
    2927             :         }
    2928             : 
    2929             :         /* Otherwise the constant is squashed away -- move forward */
    2930       61432 :         quer_loc = off + tok_len;
    2931       61432 :         last_off = off;
    2932       61432 :         last_tok_len = tok_len;
    2933             :     }
    2934             : 
    2935             :     /*
    2936             :      * We've copied up until the last ignorable constant.  Copy over the
    2937             :      * remaining bytes of the original query string.
    2938             :      */
    2939       20424 :     len_to_wrt = query_len - quer_loc;
    2940             : 
    2941             :     Assert(len_to_wrt >= 0);
    2942       20424 :     memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2943       20424 :     n_quer_loc += len_to_wrt;
    2944             : 
    2945             :     Assert(n_quer_loc <= norm_query_buflen);
    2946       20424 :     norm_query[n_quer_loc] = '\0';
    2947             : 
    2948       20424 :     *query_len_p = n_quer_loc;
    2949       20424 :     return norm_query;
    2950             : }
    2951             : 
    2952             : /*
    2953             :  * Given a valid SQL string and an array of constant-location records,
    2954             :  * fill in the textual lengths of those constants.
    2955             :  *
    2956             :  * The constants may use any allowed constant syntax, such as float literals,
    2957             :  * bit-strings, single-quoted strings and dollar-quoted strings.  This is
    2958             :  * accomplished by using the public API for the core scanner.
    2959             :  *
    2960             :  * It is the caller's job to ensure that the string is a valid SQL statement
    2961             :  * with constants at the indicated locations.  Since in practice the string
    2962             :  * has already been parsed, and the locations that the caller provides will
    2963             :  * have originated from within the authoritative parser, this should not be
    2964             :  * a problem.
    2965             :  *
    2966             :  * Duplicate constant pointers are possible, and will have their lengths
    2967             :  * marked as '-1', so that they are later ignored.  (Actually, we assume the
    2968             :  * lengths were initialized as -1 to start with, and don't change them here.)
    2969             :  *
    2970             :  * If query_loc > 0, then "query" has been advanced by that much compared to
    2971             :  * the original string start, so we need to translate the provided locations
    2972             :  * to compensate.  (This lets us avoid re-scanning statements before the one
    2973             :  * of interest, so it's worth doing.)
    2974             :  *
    2975             :  * N.B. There is an assumption that a '-' character at a Const location begins
    2976             :  * a negative numeric constant.  This precludes there ever being another
    2977             :  * reason for a constant to start with a '-'.
    2978             :  */
    2979             : static void
    2980       20424 : fill_in_constant_lengths(JumbleState *jstate, const char *query,
    2981             :                          int query_loc)
    2982             : {
    2983             :     LocationLen *locs;
    2984             :     core_yyscan_t yyscanner;
    2985             :     core_yy_extra_type yyextra;
    2986             :     core_YYSTYPE yylval;
    2987             :     YYLTYPE     yylloc;
    2988       20424 :     int         last_loc = -1;
    2989             :     int         i;
    2990             : 
    2991             :     /*
    2992             :      * Sort the records by location so that we can process them in order while
    2993             :      * scanning the query text.
    2994             :      */
    2995       20424 :     if (jstate->clocations_count > 1)
    2996       13296 :         qsort(jstate->clocations, jstate->clocations_count,
    2997             :               sizeof(LocationLen), comp_location);
    2998       20424 :     locs = jstate->clocations;
    2999             : 
    3000             :     /* initialize the flex scanner --- should match raw_parser() */
    3001       20424 :     yyscanner = scanner_init(query,
    3002             :                              &yyextra,
    3003             :                              &ScanKeywords,
    3004             :                              ScanKeywordTokens);
    3005             : 
    3006             :     /* we don't want to re-emit any escape string warnings */
    3007       20424 :     yyextra.escape_string_warning = false;
    3008             : 
    3009             :     /* Search for each constant, in sequence */
    3010       82312 :     for (i = 0; i < jstate->clocations_count; i++)
    3011             :     {
    3012       61888 :         int         loc = locs[i].location;
    3013             :         int         tok;
    3014             : 
    3015             :         /* Adjust recorded location if we're dealing with partial string */
    3016       61888 :         loc -= query_loc;
    3017             : 
    3018             :         Assert(loc >= 0);
    3019             : 
    3020       61888 :         if (loc <= last_loc)
    3021         456 :             continue;           /* Duplicate constant, ignore */
    3022             : 
    3023             :         /* Lex tokens until we find the desired constant */
    3024             :         for (;;)
    3025             :         {
    3026      471846 :             tok = core_yylex(&yylval, &yylloc, yyscanner);
    3027             : 
    3028             :             /* We should not hit end-of-string, but if we do, behave sanely */
    3029      471846 :             if (tok == 0)
    3030           0 :                 break;          /* out of inner for-loop */
    3031             : 
    3032             :             /*
    3033             :              * We should find the token position exactly, but if we somehow
    3034             :              * run past it, work with that.
    3035             :              */
    3036      471846 :             if (yylloc >= loc)
    3037             :             {
    3038       61432 :                 if (query[loc] == '-')
    3039             :                 {
    3040             :                     /*
    3041             :                      * It's a negative value - this is the one and only case
    3042             :                      * where we replace more than a single token.
    3043             :                      *
    3044             :                      * Do not compensate for the core system's special-case
    3045             :                      * adjustment of location to that of the leading '-'
    3046             :                      * operator in the event of a negative constant.  It is
    3047             :                      * also useful for our purposes to start from the minus
    3048             :                      * symbol.  In this way, queries like "select * from foo
    3049             :                      * where bar = 1" and "select * from foo where bar = -2"
    3050             :                      * will have identical normalized query strings.
    3051             :                      */
    3052         754 :                     tok = core_yylex(&yylval, &yylloc, yyscanner);
    3053         754 :                     if (tok == 0)
    3054           0 :                         break;  /* out of inner for-loop */
    3055             :                 }
    3056             : 
    3057             :                 /*
    3058             :                  * We now rely on the assumption that flex has placed a zero
    3059             :                  * byte after the text of the current token in scanbuf.
    3060             :                  */
    3061       61432 :                 locs[i].length = strlen(yyextra.scanbuf + loc);
    3062       61432 :                 break;          /* out of inner for-loop */
    3063             :             }
    3064             :         }
    3065             : 
    3066             :         /* If we hit end-of-string, give up, leaving remaining lengths -1 */
    3067       61432 :         if (tok == 0)
    3068           0 :             break;
    3069             : 
    3070       61432 :         last_loc = loc;
    3071             :     }
    3072             : 
    3073       20424 :     scanner_finish(yyscanner);
    3074       20424 : }
    3075             : 
    3076             : /*
    3077             :  * comp_location: comparator for qsorting LocationLen structs by location
    3078             :  */
    3079             : static int
    3080       69616 : comp_location(const void *a, const void *b)
    3081             : {
    3082       69616 :     int         l = ((const LocationLen *) a)->location;
    3083       69616 :     int         r = ((const LocationLen *) b)->location;
    3084             : 
    3085       69616 :     return pg_cmp_s32(l, r);
    3086             : }

Generated by: LCOV version 1.14