LCOV - code coverage report
Current view: top level - contrib/pg_stat_statements - pg_stat_statements.c (source / functions) Hit Total Coverage
Test: PostgreSQL 17devel Lines: 702 949 74.0 %
Date: 2024-04-25 00:11:39 Functions: 42 49 85.7 %
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) 2008-2024, 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/hashfn.h"
      53             : #include "common/int.h"
      54             : #include "executor/instrument.h"
      55             : #include "funcapi.h"
      56             : #include "jit/jit.h"
      57             : #include "mb/pg_wchar.h"
      58             : #include "miscadmin.h"
      59             : #include "nodes/queryjumble.h"
      60             : #include "optimizer/planner.h"
      61             : #include "parser/analyze.h"
      62             : #include "parser/parsetree.h"
      63             : #include "parser/scanner.h"
      64             : #include "parser/scansup.h"
      65             : #include "pgstat.h"
      66             : #include "storage/fd.h"
      67             : #include "storage/ipc.h"
      68             : #include "storage/lwlock.h"
      69             : #include "storage/shmem.h"
      70             : #include "storage/spin.h"
      71             : #include "tcop/utility.h"
      72             : #include "utils/acl.h"
      73             : #include "utils/builtins.h"
      74             : #include "utils/memutils.h"
      75             : #include "utils/timestamp.h"
      76             : 
      77          14 : PG_MODULE_MAGIC;
      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             : } pgssVersion;
     117             : 
     118             : typedef enum pgssStoreKind
     119             : {
     120             :     PGSS_INVALID = -1,
     121             : 
     122             :     /*
     123             :      * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
     124             :      * reference the underlying values in the arrays in the Counters struct,
     125             :      * and this order is required in pg_stat_statements_internal().
     126             :      */
     127             :     PGSS_PLAN = 0,
     128             :     PGSS_EXEC,
     129             : 
     130             :     PGSS_NUMKIND                /* Must be last value of this enum */
     131             : } pgssStoreKind;
     132             : 
     133             : /*
     134             :  * Hashtable key that defines the identity of a hashtable entry.  We separate
     135             :  * queries by user and by database even if they are otherwise identical.
     136             :  *
     137             :  * If you add a new key to this struct, make sure to teach pgss_store() to
     138             :  * zero the padding bytes.  Otherwise, things will break, because pgss_hash is
     139             :  * created using HASH_BLOBS, and thus tag_hash is used to hash this.
     140             : 
     141             :  */
     142             : typedef struct pgssHashKey
     143             : {
     144             :     Oid         userid;         /* user OID */
     145             :     Oid         dbid;           /* database OID */
     146             :     uint64      queryid;        /* query identifier */
     147             :     bool        toplevel;       /* query executed at top level */
     148             : } pgssHashKey;
     149             : 
     150             : /*
     151             :  * The actual stats counters kept within pgssEntry.
     152             :  */
     153             : typedef struct Counters
     154             : {
     155             :     int64       calls[PGSS_NUMKIND];    /* # of times planned/executed */
     156             :     double      total_time[PGSS_NUMKIND];   /* total planning/execution time,
     157             :                                              * in msec */
     158             :     double      min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
     159             :                                          * msec since min/max reset */
     160             :     double      max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
     161             :                                          * msec since min/max reset */
     162             :     double      mean_time[PGSS_NUMKIND];    /* mean planning/execution time in
     163             :                                              * msec */
     164             :     double      sum_var_time[PGSS_NUMKIND]; /* sum of variances in
     165             :                                              * planning/execution time in msec */
     166             :     int64       rows;           /* total # of retrieved or affected rows */
     167             :     int64       shared_blks_hit;    /* # of shared buffer hits */
     168             :     int64       shared_blks_read;   /* # of shared disk blocks read */
     169             :     int64       shared_blks_dirtied;    /* # of shared disk blocks dirtied */
     170             :     int64       shared_blks_written;    /* # of shared disk blocks written */
     171             :     int64       local_blks_hit; /* # of local buffer hits */
     172             :     int64       local_blks_read;    /* # of local disk blocks read */
     173             :     int64       local_blks_dirtied; /* # of local disk blocks dirtied */
     174             :     int64       local_blks_written; /* # of local disk blocks written */
     175             :     int64       temp_blks_read; /* # of temp blocks read */
     176             :     int64       temp_blks_written;  /* # of temp blocks written */
     177             :     double      shared_blk_read_time;   /* time spent reading shared blocks,
     178             :                                          * in msec */
     179             :     double      shared_blk_write_time;  /* time spent writing shared blocks,
     180             :                                          * in msec */
     181             :     double      local_blk_read_time;    /* time spent reading local blocks, in
     182             :                                          * msec */
     183             :     double      local_blk_write_time;   /* time spent writing local blocks, in
     184             :                                          * msec */
     185             :     double      temp_blk_read_time; /* time spent reading temp blocks, in msec */
     186             :     double      temp_blk_write_time;    /* time spent writing temp blocks, in
     187             :                                          * msec */
     188             :     double      usage;          /* usage factor */
     189             :     int64       wal_records;    /* # of WAL records generated */
     190             :     int64       wal_fpi;        /* # of WAL full page images generated */
     191             :     uint64      wal_bytes;      /* total amount of WAL generated in bytes */
     192             :     int64       jit_functions;  /* total number of JIT functions emitted */
     193             :     double      jit_generation_time;    /* total time to generate jit code */
     194             :     int64       jit_inlining_count; /* number of times inlining time has been
     195             :                                      * > 0 */
     196             :     double      jit_deform_time;    /* total time to deform tuples in jit code */
     197             :     int64       jit_deform_count;   /* number of times deform time has been >
     198             :                                      * 0 */
     199             : 
     200             :     double      jit_inlining_time;  /* total time to inline jit code */
     201             :     int64       jit_optimization_count; /* number of times optimization time
     202             :                                          * has been > 0 */
     203             :     double      jit_optimization_time;  /* total time to optimize jit code */
     204             :     int64       jit_emission_count; /* number of times emission time has been
     205             :                                      * > 0 */
     206             :     double      jit_emission_time;  /* total time to emit jit code */
     207             : } Counters;
     208             : 
     209             : /*
     210             :  * Global statistics for pg_stat_statements
     211             :  */
     212             : typedef struct pgssGlobalStats
     213             : {
     214             :     int64       dealloc;        /* # of times entries were deallocated */
     215             :     TimestampTz stats_reset;    /* timestamp with all stats reset */
     216             : } pgssGlobalStats;
     217             : 
     218             : /*
     219             :  * Statistics per statement
     220             :  *
     221             :  * Note: in event of a failure in garbage collection of the query text file,
     222             :  * we reset query_offset to zero and query_len to -1.  This will be seen as
     223             :  * an invalid state by qtext_fetch().
     224             :  */
     225             : typedef struct pgssEntry
     226             : {
     227             :     pgssHashKey key;            /* hash key of entry - MUST BE FIRST */
     228             :     Counters    counters;       /* the statistics for this query */
     229             :     Size        query_offset;   /* query text offset in external file */
     230             :     int         query_len;      /* # of valid bytes in query string, or -1 */
     231             :     int         encoding;       /* query text encoding */
     232             :     TimestampTz stats_since;    /* timestamp of entry allocation */
     233             :     TimestampTz minmax_stats_since; /* timestamp of last min/max values reset */
     234             :     slock_t     mutex;          /* protects the counters only */
     235             : } pgssEntry;
     236             : 
     237             : /*
     238             :  * Global shared state
     239             :  */
     240             : typedef struct pgssSharedState
     241             : {
     242             :     LWLock     *lock;           /* protects hashtable search/modification */
     243             :     double      cur_median_usage;   /* current median usage in hashtable */
     244             :     Size        mean_query_len; /* current mean entry text length */
     245             :     slock_t     mutex;          /* protects following fields only: */
     246             :     Size        extent;         /* current extent of query file */
     247             :     int         n_writers;      /* number of active writers to query file */
     248             :     int         gc_count;       /* query file garbage collection cycle count */
     249             :     pgssGlobalStats stats;      /* global statistics for pgss */
     250             : } pgssSharedState;
     251             : 
     252             : /*---- Local variables ----*/
     253             : 
     254             : /* Current nesting depth of planner/ExecutorRun/ProcessUtility calls */
     255             : static int  nesting_level = 0;
     256             : 
     257             : /* Saved hook values in case of unload */
     258             : static shmem_request_hook_type prev_shmem_request_hook = NULL;
     259             : static shmem_startup_hook_type prev_shmem_startup_hook = NULL;
     260             : static post_parse_analyze_hook_type prev_post_parse_analyze_hook = NULL;
     261             : static planner_hook_type prev_planner_hook = NULL;
     262             : static ExecutorStart_hook_type prev_ExecutorStart = NULL;
     263             : static ExecutorRun_hook_type prev_ExecutorRun = NULL;
     264             : static ExecutorFinish_hook_type prev_ExecutorFinish = NULL;
     265             : static ExecutorEnd_hook_type prev_ExecutorEnd = NULL;
     266             : static ProcessUtility_hook_type prev_ProcessUtility = NULL;
     267             : 
     268             : /* Links to shared memory state */
     269             : static pgssSharedState *pgss = NULL;
     270             : static HTAB *pgss_hash = NULL;
     271             : 
     272             : /*---- GUC variables ----*/
     273             : 
     274             : typedef enum
     275             : {
     276             :     PGSS_TRACK_NONE,            /* track no statements */
     277             :     PGSS_TRACK_TOP,             /* only top level statements */
     278             :     PGSS_TRACK_ALL,             /* all statements, including nested ones */
     279             : }           PGSSTrackLevel;
     280             : 
     281             : static const struct config_enum_entry track_options[] =
     282             : {
     283             :     {"none", PGSS_TRACK_NONE, false},
     284             :     {"top", PGSS_TRACK_TOP, false},
     285             :     {"all", PGSS_TRACK_ALL, false},
     286             :     {NULL, 0, false}
     287             : };
     288             : 
     289             : static int  pgss_max = 5000;    /* max # statements to track */
     290             : static int  pgss_track = PGSS_TRACK_TOP;    /* tracking level */
     291             : static bool pgss_track_utility = true;  /* whether to track utility commands */
     292             : static bool pgss_track_planning = false;    /* whether to track planning
     293             :                                              * duration */
     294             : static bool pgss_save = true;   /* whether to save stats across shutdown */
     295             : 
     296             : 
     297             : #define pgss_enabled(level) \
     298             :     (!IsParallelWorker() && \
     299             :     (pgss_track == PGSS_TRACK_ALL || \
     300             :     (pgss_track == PGSS_TRACK_TOP && (level) == 0)))
     301             : 
     302             : #define record_gc_qtexts() \
     303             :     do { \
     304             :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss; \
     305             :         SpinLockAcquire(&s->mutex); \
     306             :         s->gc_count++; \
     307             :         SpinLockRelease(&s->mutex); \
     308             :     } while(0)
     309             : 
     310             : /*---- Function declarations ----*/
     311             : 
     312          12 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset);
     313          12 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_7);
     314          30 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_11);
     315           0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_2);
     316          12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_3);
     317          12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_8);
     318          12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_9);
     319          12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_10);
     320          38 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_11);
     321           0 : PG_FUNCTION_INFO_V1(pg_stat_statements);
     322          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_info);
     323             : 
     324             : static void pgss_shmem_request(void);
     325             : static void pgss_shmem_startup(void);
     326             : static void pgss_shmem_shutdown(int code, Datum arg);
     327             : static void pgss_post_parse_analyze(ParseState *pstate, Query *query,
     328             :                                     JumbleState *jstate);
     329             : static PlannedStmt *pgss_planner(Query *parse,
     330             :                                  const char *query_string,
     331             :                                  int cursorOptions,
     332             :                                  ParamListInfo boundParams);
     333             : static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
     334             : static void pgss_ExecutorRun(QueryDesc *queryDesc,
     335             :                              ScanDirection direction,
     336             :                              uint64 count, bool execute_once);
     337             : static void pgss_ExecutorFinish(QueryDesc *queryDesc);
     338             : static void pgss_ExecutorEnd(QueryDesc *queryDesc);
     339             : static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
     340             :                                 bool readOnlyTree,
     341             :                                 ProcessUtilityContext context, ParamListInfo params,
     342             :                                 QueryEnvironment *queryEnv,
     343             :                                 DestReceiver *dest, QueryCompletion *qc);
     344             : static void pgss_store(const char *query, uint64 queryId,
     345             :                        int query_location, int query_len,
     346             :                        pgssStoreKind kind,
     347             :                        double total_time, uint64 rows,
     348             :                        const BufferUsage *bufusage,
     349             :                        const WalUsage *walusage,
     350             :                        const struct JitInstrumentation *jitusage,
     351             :                        JumbleState *jstate);
     352             : static void pg_stat_statements_internal(FunctionCallInfo fcinfo,
     353             :                                         pgssVersion api_version,
     354             :                                         bool showtext);
     355             : static Size pgss_memsize(void);
     356             : static pgssEntry *entry_alloc(pgssHashKey *key, Size query_offset, int query_len,
     357             :                               int encoding, bool sticky);
     358             : static void entry_dealloc(void);
     359             : static bool qtext_store(const char *query, int query_len,
     360             :                         Size *query_offset, int *gc_count);
     361             : static char *qtext_load_file(Size *buffer_size);
     362             : static char *qtext_fetch(Size query_offset, int query_len,
     363             :                          char *buffer, Size buffer_size);
     364             : static bool need_gc_qtexts(void);
     365             : static void gc_qtexts(void);
     366             : static TimestampTz entry_reset(Oid userid, Oid dbid, uint64 queryid, bool minmax_only);
     367             : static char *generate_normalized_query(JumbleState *jstate, const char *query,
     368             :                                        int query_loc, int *query_len_p);
     369             : static void fill_in_constant_lengths(JumbleState *jstate, const char *query,
     370             :                                      int query_loc);
     371             : static int  comp_location(const void *a, const void *b);
     372             : 
     373             : 
     374             : /*
     375             :  * Module load callback
     376             :  */
     377             : void
     378          14 : _PG_init(void)
     379             : {
     380             :     /*
     381             :      * In order to create our shared memory area, we have to be loaded via
     382             :      * shared_preload_libraries.  If not, fall out without hooking into any of
     383             :      * the main system.  (We don't throw error here because it seems useful to
     384             :      * allow the pg_stat_statements functions to be created even when the
     385             :      * module isn't active.  The functions must protect themselves against
     386             :      * being called then, however.)
     387             :      */
     388          14 :     if (!process_shared_preload_libraries_in_progress)
     389           2 :         return;
     390             : 
     391             :     /*
     392             :      * Inform the postmaster that we want to enable query_id calculation if
     393             :      * compute_query_id is set to auto.
     394             :      */
     395          12 :     EnableQueryId();
     396             : 
     397             :     /*
     398             :      * Define (or redefine) custom GUC variables.
     399             :      */
     400          12 :     DefineCustomIntVariable("pg_stat_statements.max",
     401             :                             "Sets the maximum number of statements tracked by pg_stat_statements.",
     402             :                             NULL,
     403             :                             &pgss_max,
     404             :                             5000,
     405             :                             100,
     406             :                             INT_MAX / 2,
     407             :                             PGC_POSTMASTER,
     408             :                             0,
     409             :                             NULL,
     410             :                             NULL,
     411             :                             NULL);
     412             : 
     413          12 :     DefineCustomEnumVariable("pg_stat_statements.track",
     414             :                              "Selects which statements are tracked by pg_stat_statements.",
     415             :                              NULL,
     416             :                              &pgss_track,
     417             :                              PGSS_TRACK_TOP,
     418             :                              track_options,
     419             :                              PGC_SUSET,
     420             :                              0,
     421             :                              NULL,
     422             :                              NULL,
     423             :                              NULL);
     424             : 
     425          12 :     DefineCustomBoolVariable("pg_stat_statements.track_utility",
     426             :                              "Selects whether utility commands are tracked by pg_stat_statements.",
     427             :                              NULL,
     428             :                              &pgss_track_utility,
     429             :                              true,
     430             :                              PGC_SUSET,
     431             :                              0,
     432             :                              NULL,
     433             :                              NULL,
     434             :                              NULL);
     435             : 
     436          12 :     DefineCustomBoolVariable("pg_stat_statements.track_planning",
     437             :                              "Selects whether planning duration is tracked by pg_stat_statements.",
     438             :                              NULL,
     439             :                              &pgss_track_planning,
     440             :                              false,
     441             :                              PGC_SUSET,
     442             :                              0,
     443             :                              NULL,
     444             :                              NULL,
     445             :                              NULL);
     446             : 
     447          12 :     DefineCustomBoolVariable("pg_stat_statements.save",
     448             :                              "Save pg_stat_statements statistics across server shutdowns.",
     449             :                              NULL,
     450             :                              &pgss_save,
     451             :                              true,
     452             :                              PGC_SIGHUP,
     453             :                              0,
     454             :                              NULL,
     455             :                              NULL,
     456             :                              NULL);
     457             : 
     458          12 :     MarkGUCPrefixReserved("pg_stat_statements");
     459             : 
     460             :     /*
     461             :      * Install hooks.
     462             :      */
     463          12 :     prev_shmem_request_hook = shmem_request_hook;
     464          12 :     shmem_request_hook = pgss_shmem_request;
     465          12 :     prev_shmem_startup_hook = shmem_startup_hook;
     466          12 :     shmem_startup_hook = pgss_shmem_startup;
     467          12 :     prev_post_parse_analyze_hook = post_parse_analyze_hook;
     468          12 :     post_parse_analyze_hook = pgss_post_parse_analyze;
     469          12 :     prev_planner_hook = planner_hook;
     470          12 :     planner_hook = pgss_planner;
     471          12 :     prev_ExecutorStart = ExecutorStart_hook;
     472          12 :     ExecutorStart_hook = pgss_ExecutorStart;
     473          12 :     prev_ExecutorRun = ExecutorRun_hook;
     474          12 :     ExecutorRun_hook = pgss_ExecutorRun;
     475          12 :     prev_ExecutorFinish = ExecutorFinish_hook;
     476          12 :     ExecutorFinish_hook = pgss_ExecutorFinish;
     477          12 :     prev_ExecutorEnd = ExecutorEnd_hook;
     478          12 :     ExecutorEnd_hook = pgss_ExecutorEnd;
     479          12 :     prev_ProcessUtility = ProcessUtility_hook;
     480          12 :     ProcessUtility_hook = pgss_ProcessUtility;
     481             : }
     482             : 
     483             : /*
     484             :  * shmem_request hook: request additional shared resources.  We'll allocate or
     485             :  * attach to the shared resources in pgss_shmem_startup().
     486             :  */
     487             : static void
     488          12 : pgss_shmem_request(void)
     489             : {
     490          12 :     if (prev_shmem_request_hook)
     491           0 :         prev_shmem_request_hook();
     492             : 
     493          12 :     RequestAddinShmemSpace(pgss_memsize());
     494          12 :     RequestNamedLWLockTranche("pg_stat_statements", 1);
     495          12 : }
     496             : 
     497             : /*
     498             :  * shmem_startup hook: allocate or attach to shared memory,
     499             :  * then load any pre-existing statistics from file.
     500             :  * Also create and load the query-texts file, which is expected to exist
     501             :  * (even if empty) while the module is enabled.
     502             :  */
     503             : static void
     504          12 : pgss_shmem_startup(void)
     505             : {
     506             :     bool        found;
     507             :     HASHCTL     info;
     508          12 :     FILE       *file = NULL;
     509          12 :     FILE       *qfile = NULL;
     510             :     uint32      header;
     511             :     int32       num;
     512             :     int32       pgver;
     513             :     int32       i;
     514             :     int         buffer_size;
     515          12 :     char       *buffer = NULL;
     516             : 
     517          12 :     if (prev_shmem_startup_hook)
     518           0 :         prev_shmem_startup_hook();
     519             : 
     520             :     /* reset in case this is a restart within the postmaster */
     521          12 :     pgss = NULL;
     522          12 :     pgss_hash = NULL;
     523             : 
     524             :     /*
     525             :      * Create or attach to the shared memory state, including hash table
     526             :      */
     527          12 :     LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
     528             : 
     529          12 :     pgss = ShmemInitStruct("pg_stat_statements",
     530             :                            sizeof(pgssSharedState),
     531             :                            &found);
     532             : 
     533          12 :     if (!found)
     534             :     {
     535             :         /* First time through ... */
     536          12 :         pgss->lock = &(GetNamedLWLockTranche("pg_stat_statements"))->lock;
     537          12 :         pgss->cur_median_usage = ASSUMED_MEDIAN_INIT;
     538          12 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
     539          12 :         SpinLockInit(&pgss->mutex);
     540          12 :         pgss->extent = 0;
     541          12 :         pgss->n_writers = 0;
     542          12 :         pgss->gc_count = 0;
     543          12 :         pgss->stats.dealloc = 0;
     544          12 :         pgss->stats.stats_reset = GetCurrentTimestamp();
     545             :     }
     546             : 
     547          12 :     info.keysize = sizeof(pgssHashKey);
     548          12 :     info.entrysize = sizeof(pgssEntry);
     549          12 :     pgss_hash = ShmemInitHash("pg_stat_statements hash",
     550             :                               pgss_max, pgss_max,
     551             :                               &info,
     552             :                               HASH_ELEM | HASH_BLOBS);
     553             : 
     554          12 :     LWLockRelease(AddinShmemInitLock);
     555             : 
     556             :     /*
     557             :      * If we're in the postmaster (or a standalone backend...), set up a shmem
     558             :      * exit hook to dump the statistics to disk.
     559             :      */
     560          12 :     if (!IsUnderPostmaster)
     561          12 :         on_shmem_exit(pgss_shmem_shutdown, (Datum) 0);
     562             : 
     563             :     /*
     564             :      * Done if some other process already completed our initialization.
     565             :      */
     566          12 :     if (found)
     567          12 :         return;
     568             : 
     569             :     /*
     570             :      * Note: we don't bother with locks here, because there should be no other
     571             :      * processes running when this code is reached.
     572             :      */
     573             : 
     574             :     /* Unlink query text file possibly left over from crash */
     575          12 :     unlink(PGSS_TEXT_FILE);
     576             : 
     577             :     /* Allocate new query text temp file */
     578          12 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
     579          12 :     if (qfile == NULL)
     580           0 :         goto write_error;
     581             : 
     582             :     /*
     583             :      * If we were told not to load old statistics, we're done.  (Note we do
     584             :      * not try to unlink any old dump file in this case.  This seems a bit
     585             :      * questionable but it's the historical behavior.)
     586             :      */
     587          12 :     if (!pgss_save)
     588             :     {
     589           2 :         FreeFile(qfile);
     590           2 :         return;
     591             :     }
     592             : 
     593             :     /*
     594             :      * Attempt to load old statistics from the dump file.
     595             :      */
     596          10 :     file = AllocateFile(PGSS_DUMP_FILE, PG_BINARY_R);
     597          10 :     if (file == NULL)
     598             :     {
     599           8 :         if (errno != ENOENT)
     600           0 :             goto read_error;
     601             :         /* No existing persisted stats file, so we're done */
     602           8 :         FreeFile(qfile);
     603           8 :         return;
     604             :     }
     605             : 
     606           2 :     buffer_size = 2048;
     607           2 :     buffer = (char *) palloc(buffer_size);
     608             : 
     609           4 :     if (fread(&header, sizeof(uint32), 1, file) != 1 ||
     610           4 :         fread(&pgver, sizeof(uint32), 1, file) != 1 ||
     611           2 :         fread(&num, sizeof(int32), 1, file) != 1)
     612           0 :         goto read_error;
     613             : 
     614           2 :     if (header != PGSS_FILE_HEADER ||
     615           2 :         pgver != PGSS_PG_MAJOR_VERSION)
     616           0 :         goto data_error;
     617             : 
     618          10 :     for (i = 0; i < num; i++)
     619             :     {
     620             :         pgssEntry   temp;
     621             :         pgssEntry  *entry;
     622             :         Size        query_offset;
     623             : 
     624           8 :         if (fread(&temp, sizeof(pgssEntry), 1, file) != 1)
     625           0 :             goto read_error;
     626             : 
     627             :         /* Encoding is the only field we can easily sanity-check */
     628           8 :         if (!PG_VALID_BE_ENCODING(temp.encoding))
     629           0 :             goto data_error;
     630             : 
     631             :         /* Resize buffer as needed */
     632           8 :         if (temp.query_len >= buffer_size)
     633             :         {
     634           0 :             buffer_size = Max(buffer_size * 2, temp.query_len + 1);
     635           0 :             buffer = repalloc(buffer, buffer_size);
     636             :         }
     637             : 
     638           8 :         if (fread(buffer, 1, temp.query_len + 1, file) != temp.query_len + 1)
     639           0 :             goto read_error;
     640             : 
     641             :         /* Should have a trailing null, but let's make sure */
     642           8 :         buffer[temp.query_len] = '\0';
     643             : 
     644             :         /* Skip loading "sticky" entries */
     645           8 :         if (IS_STICKY(temp.counters))
     646           0 :             continue;
     647             : 
     648             :         /* Store the query text */
     649           8 :         query_offset = pgss->extent;
     650           8 :         if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
     651           0 :             goto write_error;
     652           8 :         pgss->extent += temp.query_len + 1;
     653             : 
     654             :         /* make the hashtable entry (discards old entries if too many) */
     655           8 :         entry = entry_alloc(&temp.key, query_offset, temp.query_len,
     656             :                             temp.encoding,
     657             :                             false);
     658             : 
     659             :         /* copy in the actual stats */
     660           8 :         entry->counters = temp.counters;
     661           8 :         entry->stats_since = temp.stats_since;
     662           8 :         entry->minmax_stats_since = temp.minmax_stats_since;
     663             :     }
     664             : 
     665             :     /* Read global statistics for pg_stat_statements */
     666           2 :     if (fread(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
     667           0 :         goto read_error;
     668             : 
     669           2 :     pfree(buffer);
     670           2 :     FreeFile(file);
     671           2 :     FreeFile(qfile);
     672             : 
     673             :     /*
     674             :      * Remove the persisted stats file so it's not included in
     675             :      * backups/replication standbys, etc.  A new file will be written on next
     676             :      * shutdown.
     677             :      *
     678             :      * Note: it's okay if the PGSS_TEXT_FILE is included in a basebackup,
     679             :      * because we remove that file on startup; it acts inversely to
     680             :      * PGSS_DUMP_FILE, in that it is only supposed to be around when the
     681             :      * server is running, whereas PGSS_DUMP_FILE is only supposed to be around
     682             :      * when the server is not running.  Leaving the file creates no danger of
     683             :      * a newly restored database having a spurious record of execution costs,
     684             :      * which is what we're really concerned about here.
     685             :      */
     686           2 :     unlink(PGSS_DUMP_FILE);
     687             : 
     688           2 :     return;
     689             : 
     690           0 : read_error:
     691           0 :     ereport(LOG,
     692             :             (errcode_for_file_access(),
     693             :              errmsg("could not read file \"%s\": %m",
     694             :                     PGSS_DUMP_FILE)));
     695           0 :     goto fail;
     696           0 : data_error:
     697           0 :     ereport(LOG,
     698             :             (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
     699             :              errmsg("ignoring invalid data in file \"%s\"",
     700             :                     PGSS_DUMP_FILE)));
     701           0 :     goto fail;
     702           0 : write_error:
     703           0 :     ereport(LOG,
     704             :             (errcode_for_file_access(),
     705             :              errmsg("could not write file \"%s\": %m",
     706             :                     PGSS_TEXT_FILE)));
     707           0 : fail:
     708           0 :     if (buffer)
     709           0 :         pfree(buffer);
     710           0 :     if (file)
     711           0 :         FreeFile(file);
     712           0 :     if (qfile)
     713           0 :         FreeFile(qfile);
     714             :     /* If possible, throw away the bogus file; ignore any error */
     715           0 :     unlink(PGSS_DUMP_FILE);
     716             : 
     717             :     /*
     718             :      * Don't unlink PGSS_TEXT_FILE here; it should always be around while the
     719             :      * server is running with pg_stat_statements enabled
     720             :      */
     721             : }
     722             : 
     723             : /*
     724             :  * shmem_shutdown hook: Dump statistics into file.
     725             :  *
     726             :  * Note: we don't bother with acquiring lock, because there should be no
     727             :  * other processes running when this is called.
     728             :  */
     729             : static void
     730          12 : pgss_shmem_shutdown(int code, Datum arg)
     731             : {
     732             :     FILE       *file;
     733          12 :     char       *qbuffer = NULL;
     734          12 :     Size        qbuffer_size = 0;
     735             :     HASH_SEQ_STATUS hash_seq;
     736             :     int32       num_entries;
     737             :     pgssEntry  *entry;
     738             : 
     739             :     /* Don't try to dump during a crash. */
     740          12 :     if (code)
     741          12 :         return;
     742             : 
     743             :     /* Safety check ... shouldn't get here unless shmem is set up. */
     744          12 :     if (!pgss || !pgss_hash)
     745           0 :         return;
     746             : 
     747             :     /* Don't dump if told not to. */
     748          12 :     if (!pgss_save)
     749           4 :         return;
     750             : 
     751           8 :     file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
     752           8 :     if (file == NULL)
     753           0 :         goto error;
     754             : 
     755           8 :     if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
     756           0 :         goto error;
     757           8 :     if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
     758           0 :         goto error;
     759           8 :     num_entries = hash_get_num_entries(pgss_hash);
     760           8 :     if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
     761           0 :         goto error;
     762             : 
     763           8 :     qbuffer = qtext_load_file(&qbuffer_size);
     764           8 :     if (qbuffer == NULL)
     765           0 :         goto error;
     766             : 
     767             :     /*
     768             :      * When serializing to disk, we store query texts immediately after their
     769             :      * entry data.  Any orphaned query texts are thereby excluded.
     770             :      */
     771           8 :     hash_seq_init(&hash_seq, pgss_hash);
     772       52580 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
     773             :     {
     774       52572 :         int         len = entry->query_len;
     775       52572 :         char       *qstr = qtext_fetch(entry->query_offset, len,
     776             :                                        qbuffer, qbuffer_size);
     777             : 
     778       52572 :         if (qstr == NULL)
     779           0 :             continue;           /* Ignore any entries with bogus texts */
     780             : 
     781       52572 :         if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
     782       52572 :             fwrite(qstr, 1, len + 1, file) != len + 1)
     783             :         {
     784             :             /* note: we assume hash_seq_term won't change errno */
     785           0 :             hash_seq_term(&hash_seq);
     786           0 :             goto error;
     787             :         }
     788             :     }
     789             : 
     790             :     /* Dump global statistics for pg_stat_statements */
     791           8 :     if (fwrite(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
     792           0 :         goto error;
     793             : 
     794           8 :     free(qbuffer);
     795           8 :     qbuffer = NULL;
     796             : 
     797           8 :     if (FreeFile(file))
     798             :     {
     799           0 :         file = NULL;
     800           0 :         goto error;
     801             :     }
     802             : 
     803             :     /*
     804             :      * Rename file into place, so we atomically replace any old one.
     805             :      */
     806           8 :     (void) durable_rename(PGSS_DUMP_FILE ".tmp", PGSS_DUMP_FILE, LOG);
     807             : 
     808             :     /* Unlink query-texts file; it's not needed while shutdown */
     809           8 :     unlink(PGSS_TEXT_FILE);
     810             : 
     811           8 :     return;
     812             : 
     813           0 : error:
     814           0 :     ereport(LOG,
     815             :             (errcode_for_file_access(),
     816             :              errmsg("could not write file \"%s\": %m",
     817             :                     PGSS_DUMP_FILE ".tmp")));
     818           0 :     free(qbuffer);
     819           0 :     if (file)
     820           0 :         FreeFile(file);
     821           0 :     unlink(PGSS_DUMP_FILE ".tmp");
     822           0 :     unlink(PGSS_TEXT_FILE);
     823             : }
     824             : 
     825             : /*
     826             :  * Post-parse-analysis hook: mark query with a queryId
     827             :  */
     828             : static void
     829      132668 : pgss_post_parse_analyze(ParseState *pstate, Query *query, JumbleState *jstate)
     830             : {
     831      132668 :     if (prev_post_parse_analyze_hook)
     832           0 :         prev_post_parse_analyze_hook(pstate, query, jstate);
     833             : 
     834             :     /* Safety check... */
     835      132668 :     if (!pgss || !pgss_hash || !pgss_enabled(nesting_level))
     836       24642 :         return;
     837             : 
     838             :     /*
     839             :      * If it's EXECUTE, clear the queryId so that stats will accumulate for
     840             :      * the underlying PREPARE.  But don't do this if we're not tracking
     841             :      * utility statements, to avoid messing up another extension that might be
     842             :      * tracking them.
     843             :      */
     844      108026 :     if (query->utilityStmt)
     845             :     {
     846       49772 :         if (pgss_track_utility && IsA(query->utilityStmt, ExecuteStmt))
     847             :         {
     848        2774 :             query->queryId = UINT64CONST(0);
     849        2774 :             return;
     850             :         }
     851             :     }
     852             : 
     853             :     /*
     854             :      * If query jumbling were able to identify any ignorable constants, we
     855             :      * immediately create a hash table entry for the query, so that we can
     856             :      * record the normalized form of the query string.  If there were no such
     857             :      * constants, the normalized string would be the same as the query text
     858             :      * anyway, so there's no need for an early entry.
     859             :      */
     860      105252 :     if (jstate && jstate->clocations_count > 0)
     861       53786 :         pgss_store(pstate->p_sourcetext,
     862             :                    query->queryId,
     863             :                    query->stmt_location,
     864             :                    query->stmt_len,
     865             :                    PGSS_INVALID,
     866             :                    0,
     867             :                    0,
     868             :                    NULL,
     869             :                    NULL,
     870             :                    NULL,
     871             :                    jstate);
     872             : }
     873             : 
     874             : /*
     875             :  * Planner hook: forward to regular planner, but measure planning time
     876             :  * if needed.
     877             :  */
     878             : static PlannedStmt *
     879       81602 : pgss_planner(Query *parse,
     880             :              const char *query_string,
     881             :              int cursorOptions,
     882             :              ParamListInfo boundParams)
     883             : {
     884             :     PlannedStmt *result;
     885             : 
     886             :     /*
     887             :      * We can't process the query if no query_string is provided, as
     888             :      * pgss_store needs it.  We also ignore query without queryid, as it would
     889             :      * be treated as a utility statement, which may not be the case.
     890             :      *
     891             :      * Note that planner_hook can be called from the planner itself, so we
     892             :      * have a specific nesting level for the planner.  However, utility
     893             :      * commands containing optimizable statements can also call the planner,
     894             :      * same for regular DML (for instance for underlying foreign key queries).
     895             :      * So testing the planner nesting level only is not enough to detect real
     896             :      * top level planner call.
     897             :      */
     898       81602 :     if (pgss_enabled(nesting_level)
     899       58472 :         && pgss_track_planning && query_string
     900         146 :         && parse->queryId != UINT64CONST(0))
     901         146 :     {
     902             :         instr_time  start;
     903             :         instr_time  duration;
     904             :         BufferUsage bufusage_start,
     905             :                     bufusage;
     906             :         WalUsage    walusage_start,
     907             :                     walusage;
     908             : 
     909             :         /* We need to track buffer usage as the planner can access them. */
     910         146 :         bufusage_start = pgBufferUsage;
     911             : 
     912             :         /*
     913             :          * Similarly the planner could write some WAL records in some cases
     914             :          * (e.g. setting a hint bit with those being WAL-logged)
     915             :          */
     916         146 :         walusage_start = pgWalUsage;
     917         146 :         INSTR_TIME_SET_CURRENT(start);
     918             : 
     919         146 :         nesting_level++;
     920         146 :         PG_TRY();
     921             :         {
     922         146 :             if (prev_planner_hook)
     923           0 :                 result = prev_planner_hook(parse, query_string, cursorOptions,
     924             :                                            boundParams);
     925             :             else
     926         146 :                 result = standard_planner(parse, query_string, cursorOptions,
     927             :                                           boundParams);
     928             :         }
     929           0 :         PG_FINALLY();
     930             :         {
     931         146 :             nesting_level--;
     932             :         }
     933         146 :         PG_END_TRY();
     934             : 
     935         146 :         INSTR_TIME_SET_CURRENT(duration);
     936         146 :         INSTR_TIME_SUBTRACT(duration, start);
     937             : 
     938             :         /* calc differences of buffer counters. */
     939         146 :         memset(&bufusage, 0, sizeof(BufferUsage));
     940         146 :         BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
     941             : 
     942             :         /* calc differences of WAL counters. */
     943         146 :         memset(&walusage, 0, sizeof(WalUsage));
     944         146 :         WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
     945             : 
     946         146 :         pgss_store(query_string,
     947             :                    parse->queryId,
     948             :                    parse->stmt_location,
     949             :                    parse->stmt_len,
     950             :                    PGSS_PLAN,
     951         146 :                    INSTR_TIME_GET_MILLISEC(duration),
     952             :                    0,
     953             :                    &bufusage,
     954             :                    &walusage,
     955             :                    NULL,
     956             :                    NULL);
     957             :     }
     958             :     else
     959             :     {
     960             :         /*
     961             :          * Even though we're not tracking plan time for this statement, we
     962             :          * must still increment the nesting level, to ensure that functions
     963             :          * evaluated during planning are not seen as top-level calls.
     964             :          */
     965       81456 :         nesting_level++;
     966       81456 :         PG_TRY();
     967             :         {
     968       81456 :             if (prev_planner_hook)
     969           0 :                 result = prev_planner_hook(parse, query_string, cursorOptions,
     970             :                                            boundParams);
     971             :             else
     972       81456 :                 result = standard_planner(parse, query_string, cursorOptions,
     973             :                                           boundParams);
     974             :         }
     975        1204 :         PG_FINALLY();
     976             :         {
     977       81456 :             nesting_level--;
     978             :         }
     979       81456 :         PG_END_TRY();
     980             :     }
     981             : 
     982       80398 :     return result;
     983             : }
     984             : 
     985             : /*
     986             :  * ExecutorStart hook: start up tracking if needed
     987             :  */
     988             : static void
     989       94728 : pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
     990             : {
     991       94728 :     if (prev_ExecutorStart)
     992           0 :         prev_ExecutorStart(queryDesc, eflags);
     993             :     else
     994       94728 :         standard_ExecutorStart(queryDesc, eflags);
     995             : 
     996             :     /*
     997             :      * If query has queryId zero, don't track it.  This prevents double
     998             :      * counting of optimizable statements that are directly contained in
     999             :      * utility statements.
    1000             :      */
    1001       94110 :     if (pgss_enabled(nesting_level) && queryDesc->plannedstmt->queryId != UINT64CONST(0))
    1002             :     {
    1003             :         /*
    1004             :          * Set up to track total elapsed time in ExecutorRun.  Make sure the
    1005             :          * space is allocated in the per-query context so it will go away at
    1006             :          * ExecutorEnd.
    1007             :          */
    1008       58996 :         if (queryDesc->totaltime == NULL)
    1009             :         {
    1010             :             MemoryContext oldcxt;
    1011             : 
    1012       58996 :             oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
    1013       58996 :             queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL, false);
    1014       58996 :             MemoryContextSwitchTo(oldcxt);
    1015             :         }
    1016             :     }
    1017       94110 : }
    1018             : 
    1019             : /*
    1020             :  * ExecutorRun hook: all we need do is track nesting depth
    1021             :  */
    1022             : static void
    1023       92128 : pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count,
    1024             :                  bool execute_once)
    1025             : {
    1026       92128 :     nesting_level++;
    1027       92128 :     PG_TRY();
    1028             :     {
    1029       92128 :         if (prev_ExecutorRun)
    1030           0 :             prev_ExecutorRun(queryDesc, direction, count, execute_once);
    1031             :         else
    1032       92128 :             standard_ExecutorRun(queryDesc, direction, count, execute_once);
    1033             :     }
    1034        6784 :     PG_FINALLY();
    1035             :     {
    1036       92128 :         nesting_level--;
    1037             :     }
    1038       92128 :     PG_END_TRY();
    1039       85344 : }
    1040             : 
    1041             : /*
    1042             :  * ExecutorFinish hook: all we need do is track nesting depth
    1043             :  */
    1044             : static void
    1045       82090 : pgss_ExecutorFinish(QueryDesc *queryDesc)
    1046             : {
    1047       82090 :     nesting_level++;
    1048       82090 :     PG_TRY();
    1049             :     {
    1050       82090 :         if (prev_ExecutorFinish)
    1051           0 :             prev_ExecutorFinish(queryDesc);
    1052             :         else
    1053       82090 :             standard_ExecutorFinish(queryDesc);
    1054             :     }
    1055         314 :     PG_FINALLY();
    1056             :     {
    1057       82090 :         nesting_level--;
    1058             :     }
    1059       82090 :     PG_END_TRY();
    1060       81776 : }
    1061             : 
    1062             : /*
    1063             :  * ExecutorEnd hook: store results if needed
    1064             :  */
    1065             : static void
    1066       86934 : pgss_ExecutorEnd(QueryDesc *queryDesc)
    1067             : {
    1068       86934 :     uint64      queryId = queryDesc->plannedstmt->queryId;
    1069             : 
    1070       86934 :     if (queryId != UINT64CONST(0) && queryDesc->totaltime &&
    1071       56230 :         pgss_enabled(nesting_level))
    1072             :     {
    1073             :         /*
    1074             :          * Make sure stats accumulation is done.  (Note: it's okay if several
    1075             :          * levels of hook all do this.)
    1076             :          */
    1077       56230 :         InstrEndLoop(queryDesc->totaltime);
    1078             : 
    1079       56032 :         pgss_store(queryDesc->sourceText,
    1080             :                    queryId,
    1081       56230 :                    queryDesc->plannedstmt->stmt_location,
    1082       56230 :                    queryDesc->plannedstmt->stmt_len,
    1083             :                    PGSS_EXEC,
    1084       56230 :                    queryDesc->totaltime->total * 1000.0,  /* convert to msec */
    1085       56230 :                    queryDesc->estate->es_total_processed,
    1086       56230 :                    &queryDesc->totaltime->bufusage,
    1087       56230 :                    &queryDesc->totaltime->walusage,
    1088       56230 :                    queryDesc->estate->es_jit ? &queryDesc->estate->es_jit->instr : NULL,
    1089             :                    NULL);
    1090             :     }
    1091             : 
    1092       86934 :     if (prev_ExecutorEnd)
    1093           0 :         prev_ExecutorEnd(queryDesc);
    1094             :     else
    1095       86934 :         standard_ExecutorEnd(queryDesc);
    1096       86934 : }
    1097             : 
    1098             : /*
    1099             :  * ProcessUtility hook
    1100             :  */
    1101             : static void
    1102       59974 : pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
    1103             :                     bool readOnlyTree,
    1104             :                     ProcessUtilityContext context,
    1105             :                     ParamListInfo params, QueryEnvironment *queryEnv,
    1106             :                     DestReceiver *dest, QueryCompletion *qc)
    1107             : {
    1108       59974 :     Node       *parsetree = pstmt->utilityStmt;
    1109       59974 :     uint64      saved_queryId = pstmt->queryId;
    1110       59974 :     int         saved_stmt_location = pstmt->stmt_location;
    1111       59974 :     int         saved_stmt_len = pstmt->stmt_len;
    1112       59974 :     bool        enabled = pgss_track_utility && pgss_enabled(nesting_level);
    1113             : 
    1114             :     /*
    1115             :      * Force utility statements to get queryId zero.  We do this even in cases
    1116             :      * where the statement contains an optimizable statement for which a
    1117             :      * queryId could be derived (such as EXPLAIN or DECLARE CURSOR).  For such
    1118             :      * cases, runtime control will first go through ProcessUtility and then
    1119             :      * the executor, and we don't want the executor hooks to do anything,
    1120             :      * since we are already measuring the statement's costs at the utility
    1121             :      * level.
    1122             :      *
    1123             :      * Note that this is only done if pg_stat_statements is enabled and
    1124             :      * configured to track utility statements, in the unlikely possibility
    1125             :      * that user configured another extension to handle utility statements
    1126             :      * only.
    1127             :      */
    1128       59974 :     if (enabled)
    1129       49670 :         pstmt->queryId = UINT64CONST(0);
    1130             : 
    1131             :     /*
    1132             :      * If it's an EXECUTE statement, we don't track it and don't increment the
    1133             :      * nesting level.  This allows the cycles to be charged to the underlying
    1134             :      * PREPARE instead (by the Executor hooks), which is much more useful.
    1135             :      *
    1136             :      * We also don't track execution of PREPARE.  If we did, we would get one
    1137             :      * hash table entry for the PREPARE (with hash calculated from the query
    1138             :      * string), and then a different one with the same query string (but hash
    1139             :      * calculated from the query tree) would be used to accumulate costs of
    1140             :      * ensuing EXECUTEs.  This would be confusing.  Since PREPARE doesn't
    1141             :      * actually run the planner (only parse+rewrite), its costs are generally
    1142             :      * pretty negligible and it seems okay to just ignore it.
    1143             :      */
    1144       59974 :     if (enabled &&
    1145       49670 :         !IsA(parsetree, ExecuteStmt) &&
    1146       46898 :         !IsA(parsetree, PrepareStmt))
    1147       42334 :     {
    1148             :         instr_time  start;
    1149             :         instr_time  duration;
    1150             :         uint64      rows;
    1151             :         BufferUsage bufusage_start,
    1152             :                     bufusage;
    1153             :         WalUsage    walusage_start,
    1154             :                     walusage;
    1155             : 
    1156       46698 :         bufusage_start = pgBufferUsage;
    1157       46698 :         walusage_start = pgWalUsage;
    1158       46698 :         INSTR_TIME_SET_CURRENT(start);
    1159             : 
    1160       46698 :         nesting_level++;
    1161       46698 :         PG_TRY();
    1162             :         {
    1163       46698 :             if (prev_ProcessUtility)
    1164           0 :                 prev_ProcessUtility(pstmt, queryString, readOnlyTree,
    1165             :                                     context, params, queryEnv,
    1166             :                                     dest, qc);
    1167             :             else
    1168       46698 :                 standard_ProcessUtility(pstmt, queryString, readOnlyTree,
    1169             :                                         context, params, queryEnv,
    1170             :                                         dest, qc);
    1171             :         }
    1172        4364 :         PG_FINALLY();
    1173             :         {
    1174       46698 :             nesting_level--;
    1175             :         }
    1176       46698 :         PG_END_TRY();
    1177             : 
    1178             :         /*
    1179             :          * CAUTION: do not access the *pstmt data structure again below here.
    1180             :          * If it was a ROLLBACK or similar, that data structure may have been
    1181             :          * freed.  We must copy everything we still need into local variables,
    1182             :          * which we did above.
    1183             :          *
    1184             :          * For the same reason, we can't risk restoring pstmt->queryId to its
    1185             :          * former value, which'd otherwise be a good idea.
    1186             :          */
    1187             : 
    1188       42334 :         INSTR_TIME_SET_CURRENT(duration);
    1189       42334 :         INSTR_TIME_SUBTRACT(duration, start);
    1190             : 
    1191             :         /*
    1192             :          * Track the total number of rows retrieved or affected by the utility
    1193             :          * statements of COPY, FETCH, CREATE TABLE AS, CREATE MATERIALIZED
    1194             :          * VIEW, REFRESH MATERIALIZED VIEW and SELECT INTO.
    1195             :          */
    1196       42334 :         rows = (qc && (qc->commandTag == CMDTAG_COPY ||
    1197       39616 :                        qc->commandTag == CMDTAG_FETCH ||
    1198       39156 :                        qc->commandTag == CMDTAG_SELECT ||
    1199       38800 :                        qc->commandTag == CMDTAG_REFRESH_MATERIALIZED_VIEW)) ?
    1200       84668 :             qc->nprocessed : 0;
    1201             : 
    1202             :         /* calc differences of buffer counters. */
    1203       42334 :         memset(&bufusage, 0, sizeof(BufferUsage));
    1204       42334 :         BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
    1205             : 
    1206             :         /* calc differences of WAL counters. */
    1207       42334 :         memset(&walusage, 0, sizeof(WalUsage));
    1208       42334 :         WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
    1209             : 
    1210       42334 :         pgss_store(queryString,
    1211             :                    saved_queryId,
    1212             :                    saved_stmt_location,
    1213             :                    saved_stmt_len,
    1214             :                    PGSS_EXEC,
    1215       42334 :                    INSTR_TIME_GET_MILLISEC(duration),
    1216             :                    rows,
    1217             :                    &bufusage,
    1218             :                    &walusage,
    1219             :                    NULL,
    1220             :                    NULL);
    1221             :     }
    1222             :     else
    1223             :     {
    1224             :         /*
    1225             :          * Even though we're not tracking execution time for this statement,
    1226             :          * we must still increment the nesting level, to ensure that functions
    1227             :          * evaluated within it are not seen as top-level calls.  But don't do
    1228             :          * so for EXECUTE; that way, when control reaches pgss_planner or
    1229             :          * pgss_ExecutorStart, we will treat the costs as top-level if
    1230             :          * appropriate.  Likewise, don't bump for PREPARE, so that parse
    1231             :          * analysis will treat the statement as top-level if appropriate.
    1232             :          *
    1233             :          * To be absolutely certain we don't mess up the nesting level,
    1234             :          * evaluate the bump_level condition just once.
    1235             :          */
    1236       13276 :         bool        bump_level =
    1237       23778 :             !IsA(parsetree, ExecuteStmt) &&
    1238       10502 :             !IsA(parsetree, PrepareStmt);
    1239             : 
    1240       13276 :         if (bump_level)
    1241       10300 :             nesting_level++;
    1242       13276 :         PG_TRY();
    1243             :         {
    1244       13276 :             if (prev_ProcessUtility)
    1245           0 :                 prev_ProcessUtility(pstmt, queryString, readOnlyTree,
    1246             :                                     context, params, queryEnv,
    1247             :                                     dest, qc);
    1248             :             else
    1249       13276 :                 standard_ProcessUtility(pstmt, queryString, readOnlyTree,
    1250             :                                         context, params, queryEnv,
    1251             :                                         dest, qc);
    1252             :         }
    1253         240 :         PG_FINALLY();
    1254             :         {
    1255       13276 :             if (bump_level)
    1256       10300 :                 nesting_level--;
    1257             :         }
    1258       13276 :         PG_END_TRY();
    1259             :     }
    1260       55370 : }
    1261             : 
    1262             : /*
    1263             :  * Store some statistics for a statement.
    1264             :  *
    1265             :  * If jstate is not NULL then we're trying to create an entry for which
    1266             :  * we have no statistics as yet; we just want to record the normalized
    1267             :  * query string.  total_time, rows, bufusage and walusage are ignored in this
    1268             :  * case.
    1269             :  *
    1270             :  * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
    1271             :  * for the arrays in the Counters field.
    1272             :  */
    1273             : static void
    1274      152496 : pgss_store(const char *query, uint64 queryId,
    1275             :            int query_location, int query_len,
    1276             :            pgssStoreKind kind,
    1277             :            double total_time, uint64 rows,
    1278             :            const BufferUsage *bufusage,
    1279             :            const WalUsage *walusage,
    1280             :            const struct JitInstrumentation *jitusage,
    1281             :            JumbleState *jstate)
    1282             : {
    1283             :     pgssHashKey key;
    1284             :     pgssEntry  *entry;
    1285      152496 :     char       *norm_query = NULL;
    1286      152496 :     int         encoding = GetDatabaseEncoding();
    1287             : 
    1288             :     Assert(query != NULL);
    1289             : 
    1290             :     /* Safety check... */
    1291      152496 :     if (!pgss || !pgss_hash)
    1292           0 :         return;
    1293             : 
    1294             :     /*
    1295             :      * Nothing to do if compute_query_id isn't enabled and no other module
    1296             :      * computed a query identifier.
    1297             :      */
    1298      152496 :     if (queryId == UINT64CONST(0))
    1299           0 :         return;
    1300             : 
    1301             :     /*
    1302             :      * Confine our attention to the relevant part of the string, if the query
    1303             :      * is a portion of a multi-statement source string, and update query
    1304             :      * location and length if needed.
    1305             :      */
    1306      152496 :     query = CleanQuerytext(query, &query_location, &query_len);
    1307             : 
    1308             :     /* Set up key for hashtable search */
    1309             : 
    1310             :     /* clear padding */
    1311      152496 :     memset(&key, 0, sizeof(pgssHashKey));
    1312             : 
    1313      152496 :     key.userid = GetUserId();
    1314      152496 :     key.dbid = MyDatabaseId;
    1315      152496 :     key.queryid = queryId;
    1316      152496 :     key.toplevel = (nesting_level == 0);
    1317             : 
    1318             :     /* Lookup the hash table entry with shared lock. */
    1319      152496 :     LWLockAcquire(pgss->lock, LW_SHARED);
    1320             : 
    1321      152496 :     entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    1322             : 
    1323             :     /* Create new entry, if not present */
    1324      152496 :     if (!entry)
    1325             :     {
    1326             :         Size        query_offset;
    1327             :         int         gc_count;
    1328             :         bool        stored;
    1329             :         bool        do_gc;
    1330             : 
    1331             :         /*
    1332             :          * Create a new, normalized query string if caller asked.  We don't
    1333             :          * need to hold the lock while doing this work.  (Note: in any case,
    1334             :          * it's possible that someone else creates a duplicate hashtable entry
    1335             :          * in the interval where we don't hold the lock below.  That case is
    1336             :          * handled by entry_alloc.)
    1337             :          */
    1338       53310 :         if (jstate)
    1339             :         {
    1340       19312 :             LWLockRelease(pgss->lock);
    1341       19312 :             norm_query = generate_normalized_query(jstate, query,
    1342             :                                                    query_location,
    1343             :                                                    &query_len);
    1344       19312 :             LWLockAcquire(pgss->lock, LW_SHARED);
    1345             :         }
    1346             : 
    1347             :         /* Append new query text to file with only shared lock held */
    1348       53310 :         stored = qtext_store(norm_query ? norm_query : query, query_len,
    1349             :                              &query_offset, &gc_count);
    1350             : 
    1351             :         /*
    1352             :          * Determine whether we need to garbage collect external query texts
    1353             :          * while the shared lock is still held.  This micro-optimization
    1354             :          * avoids taking the time to decide this while holding exclusive lock.
    1355             :          */
    1356       53310 :         do_gc = need_gc_qtexts();
    1357             : 
    1358             :         /* Need exclusive lock to make a new hashtable entry - promote */
    1359       53310 :         LWLockRelease(pgss->lock);
    1360       53310 :         LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
    1361             : 
    1362             :         /*
    1363             :          * A garbage collection may have occurred while we weren't holding the
    1364             :          * lock.  In the unlikely event that this happens, the query text we
    1365             :          * stored above will have been garbage collected, so write it again.
    1366             :          * This should be infrequent enough that doing it while holding
    1367             :          * exclusive lock isn't a performance problem.
    1368             :          */
    1369       53310 :         if (!stored || pgss->gc_count != gc_count)
    1370           0 :             stored = qtext_store(norm_query ? norm_query : query, query_len,
    1371             :                                  &query_offset, NULL);
    1372             : 
    1373             :         /* If we failed to write to the text file, give up */
    1374       53310 :         if (!stored)
    1375           0 :             goto done;
    1376             : 
    1377             :         /* OK to create a new hashtable entry */
    1378       53310 :         entry = entry_alloc(&key, query_offset, query_len, encoding,
    1379             :                             jstate != NULL);
    1380             : 
    1381             :         /* If needed, perform garbage collection while exclusive lock held */
    1382       53310 :         if (do_gc)
    1383           0 :             gc_qtexts();
    1384             :     }
    1385             : 
    1386             :     /* Increment the counts, except when jstate is not NULL */
    1387      152496 :     if (!jstate)
    1388             :     {
    1389             :         /*
    1390             :          * Grab the spinlock while updating the counters (see comment about
    1391             :          * locking rules at the head of the file)
    1392             :          */
    1393       98710 :         volatile pgssEntry *e = (volatile pgssEntry *) entry;
    1394             : 
    1395             :         Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
    1396             : 
    1397       98710 :         SpinLockAcquire(&e->mutex);
    1398             : 
    1399             :         /* "Unstick" entry if it was previously sticky */
    1400       98710 :         if (IS_STICKY(e->counters))
    1401       52002 :             e->counters.usage = USAGE_INIT;
    1402             : 
    1403       98710 :         e->counters.calls[kind] += 1;
    1404       98710 :         e->counters.total_time[kind] += total_time;
    1405             : 
    1406       98710 :         if (e->counters.calls[kind] == 1)
    1407             :         {
    1408       52110 :             e->counters.min_time[kind] = total_time;
    1409       52110 :             e->counters.max_time[kind] = total_time;
    1410       52110 :             e->counters.mean_time[kind] = total_time;
    1411             :         }
    1412             :         else
    1413             :         {
    1414             :             /*
    1415             :              * Welford's method for accurately computing variance. See
    1416             :              * <http://www.johndcook.com/blog/standard_deviation/>
    1417             :              */
    1418       46600 :             double      old_mean = e->counters.mean_time[kind];
    1419             : 
    1420       46600 :             e->counters.mean_time[kind] +=
    1421       46600 :                 (total_time - old_mean) / e->counters.calls[kind];
    1422       46600 :             e->counters.sum_var_time[kind] +=
    1423       46600 :                 (total_time - old_mean) * (total_time - e->counters.mean_time[kind]);
    1424             : 
    1425             :             /*
    1426             :              * Calculate min and max time. min = 0 and max = 0 means that the
    1427             :              * min/max statistics were reset
    1428             :              */
    1429       46600 :             if (e->counters.min_time[kind] == 0
    1430           4 :                 && e->counters.max_time[kind] == 0)
    1431             :             {
    1432           4 :                 e->counters.min_time[kind] = total_time;
    1433           4 :                 e->counters.max_time[kind] = total_time;
    1434             :             }
    1435             :             else
    1436             :             {
    1437       46596 :                 if (e->counters.min_time[kind] > total_time)
    1438       11940 :                     e->counters.min_time[kind] = total_time;
    1439       46596 :                 if (e->counters.max_time[kind] < total_time)
    1440        5254 :                     e->counters.max_time[kind] = total_time;
    1441             :             }
    1442             :         }
    1443       98710 :         e->counters.rows += rows;
    1444       98710 :         e->counters.shared_blks_hit += bufusage->shared_blks_hit;
    1445       98710 :         e->counters.shared_blks_read += bufusage->shared_blks_read;
    1446       98710 :         e->counters.shared_blks_dirtied += bufusage->shared_blks_dirtied;
    1447       98710 :         e->counters.shared_blks_written += bufusage->shared_blks_written;
    1448       98710 :         e->counters.local_blks_hit += bufusage->local_blks_hit;
    1449       98710 :         e->counters.local_blks_read += bufusage->local_blks_read;
    1450       98710 :         e->counters.local_blks_dirtied += bufusage->local_blks_dirtied;
    1451       98710 :         e->counters.local_blks_written += bufusage->local_blks_written;
    1452       98710 :         e->counters.temp_blks_read += bufusage->temp_blks_read;
    1453       98710 :         e->counters.temp_blks_written += bufusage->temp_blks_written;
    1454       98710 :         e->counters.shared_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_read_time);
    1455       98710 :         e->counters.shared_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_write_time);
    1456       98710 :         e->counters.local_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_read_time);
    1457       98710 :         e->counters.local_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_write_time);
    1458       98710 :         e->counters.temp_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_read_time);
    1459       98710 :         e->counters.temp_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_write_time);
    1460       98710 :         e->counters.usage += USAGE_EXEC(total_time);
    1461       98710 :         e->counters.wal_records += walusage->wal_records;
    1462       98710 :         e->counters.wal_fpi += walusage->wal_fpi;
    1463       98710 :         e->counters.wal_bytes += walusage->wal_bytes;
    1464       98710 :         if (jitusage)
    1465             :         {
    1466         198 :             e->counters.jit_functions += jitusage->created_functions;
    1467         198 :             e->counters.jit_generation_time += INSTR_TIME_GET_MILLISEC(jitusage->generation_counter);
    1468             : 
    1469         198 :             if (INSTR_TIME_GET_MILLISEC(jitusage->deform_counter))
    1470         182 :                 e->counters.jit_deform_count++;
    1471         198 :             e->counters.jit_deform_time += INSTR_TIME_GET_MILLISEC(jitusage->deform_counter);
    1472             : 
    1473         198 :             if (INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter))
    1474         136 :                 e->counters.jit_inlining_count++;
    1475         198 :             e->counters.jit_inlining_time += INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter);
    1476             : 
    1477         198 :             if (INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter))
    1478         194 :                 e->counters.jit_optimization_count++;
    1479         198 :             e->counters.jit_optimization_time += INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter);
    1480             : 
    1481         198 :             if (INSTR_TIME_GET_MILLISEC(jitusage->emission_counter))
    1482         194 :                 e->counters.jit_emission_count++;
    1483         198 :             e->counters.jit_emission_time += INSTR_TIME_GET_MILLISEC(jitusage->emission_counter);
    1484             :         }
    1485             : 
    1486       98710 :         SpinLockRelease(&e->mutex);
    1487             :     }
    1488             : 
    1489       53786 : done:
    1490      152496 :     LWLockRelease(pgss->lock);
    1491             : 
    1492             :     /* We postpone this clean-up until we're out of the lock */
    1493      152496 :     if (norm_query)
    1494       19312 :         pfree(norm_query);
    1495             : }
    1496             : 
    1497             : /*
    1498             :  * Reset statement statistics corresponding to userid, dbid, and queryid.
    1499             :  */
    1500             : Datum
    1501           2 : pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
    1502             : {
    1503             :     Oid         userid;
    1504             :     Oid         dbid;
    1505             :     uint64      queryid;
    1506             : 
    1507           2 :     userid = PG_GETARG_OID(0);
    1508           2 :     dbid = PG_GETARG_OID(1);
    1509           2 :     queryid = (uint64) PG_GETARG_INT64(2);
    1510             : 
    1511           2 :     entry_reset(userid, dbid, queryid, false);
    1512             : 
    1513           2 :     PG_RETURN_VOID();
    1514             : }
    1515             : 
    1516             : Datum
    1517          98 : pg_stat_statements_reset_1_11(PG_FUNCTION_ARGS)
    1518             : {
    1519             :     Oid         userid;
    1520             :     Oid         dbid;
    1521             :     uint64      queryid;
    1522             :     bool        minmax_only;
    1523             : 
    1524          98 :     userid = PG_GETARG_OID(0);
    1525          98 :     dbid = PG_GETARG_OID(1);
    1526          98 :     queryid = (uint64) PG_GETARG_INT64(2);
    1527          98 :     minmax_only = PG_GETARG_BOOL(3);
    1528             : 
    1529          98 :     PG_RETURN_TIMESTAMPTZ(entry_reset(userid, dbid, queryid, minmax_only));
    1530             : }
    1531             : 
    1532             : /*
    1533             :  * Reset statement statistics.
    1534             :  */
    1535             : Datum
    1536           2 : pg_stat_statements_reset(PG_FUNCTION_ARGS)
    1537             : {
    1538           2 :     entry_reset(0, 0, 0, false);
    1539             : 
    1540           2 :     PG_RETURN_VOID();
    1541             : }
    1542             : 
    1543             : /* Number of output arguments (columns) for various API versions */
    1544             : #define PG_STAT_STATEMENTS_COLS_V1_0    14
    1545             : #define PG_STAT_STATEMENTS_COLS_V1_1    18
    1546             : #define PG_STAT_STATEMENTS_COLS_V1_2    19
    1547             : #define PG_STAT_STATEMENTS_COLS_V1_3    23
    1548             : #define PG_STAT_STATEMENTS_COLS_V1_8    32
    1549             : #define PG_STAT_STATEMENTS_COLS_V1_9    33
    1550             : #define PG_STAT_STATEMENTS_COLS_V1_10   43
    1551             : #define PG_STAT_STATEMENTS_COLS_V1_11   49
    1552             : #define PG_STAT_STATEMENTS_COLS         49  /* maximum of above */
    1553             : 
    1554             : /*
    1555             :  * Retrieve statement statistics.
    1556             :  *
    1557             :  * The SQL API of this function has changed multiple times, and will likely
    1558             :  * do so again in future.  To support the case where a newer version of this
    1559             :  * loadable module is being used with an old SQL declaration of the function,
    1560             :  * we continue to support the older API versions.  For 1.2 and later, the
    1561             :  * expected API version is identified by embedding it in the C name of the
    1562             :  * function.  Unfortunately we weren't bright enough to do that for 1.1.
    1563             :  */
    1564             : Datum
    1565         116 : pg_stat_statements_1_11(PG_FUNCTION_ARGS)
    1566             : {
    1567         116 :     bool        showtext = PG_GETARG_BOOL(0);
    1568             : 
    1569         116 :     pg_stat_statements_internal(fcinfo, PGSS_V1_11, showtext);
    1570             : 
    1571         116 :     return (Datum) 0;
    1572             : }
    1573             : 
    1574             : Datum
    1575           2 : pg_stat_statements_1_10(PG_FUNCTION_ARGS)
    1576             : {
    1577           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1578             : 
    1579           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_10, showtext);
    1580             : 
    1581           2 :     return (Datum) 0;
    1582             : }
    1583             : 
    1584             : Datum
    1585           2 : pg_stat_statements_1_9(PG_FUNCTION_ARGS)
    1586             : {
    1587           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1588             : 
    1589           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_9, showtext);
    1590             : 
    1591           2 :     return (Datum) 0;
    1592             : }
    1593             : 
    1594             : Datum
    1595           2 : pg_stat_statements_1_8(PG_FUNCTION_ARGS)
    1596             : {
    1597           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1598             : 
    1599           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
    1600             : 
    1601           2 :     return (Datum) 0;
    1602             : }
    1603             : 
    1604             : Datum
    1605           2 : pg_stat_statements_1_3(PG_FUNCTION_ARGS)
    1606             : {
    1607           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1608             : 
    1609           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
    1610             : 
    1611           2 :     return (Datum) 0;
    1612             : }
    1613             : 
    1614             : Datum
    1615           0 : pg_stat_statements_1_2(PG_FUNCTION_ARGS)
    1616             : {
    1617           0 :     bool        showtext = PG_GETARG_BOOL(0);
    1618             : 
    1619           0 :     pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
    1620             : 
    1621           0 :     return (Datum) 0;
    1622             : }
    1623             : 
    1624             : /*
    1625             :  * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
    1626             :  * This can be removed someday, perhaps.
    1627             :  */
    1628             : Datum
    1629           0 : pg_stat_statements(PG_FUNCTION_ARGS)
    1630             : {
    1631             :     /* If it's really API 1.1, we'll figure that out below */
    1632           0 :     pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
    1633             : 
    1634           0 :     return (Datum) 0;
    1635             : }
    1636             : 
    1637             : /* Common code for all versions of pg_stat_statements() */
    1638             : static void
    1639         124 : pg_stat_statements_internal(FunctionCallInfo fcinfo,
    1640             :                             pgssVersion api_version,
    1641             :                             bool showtext)
    1642             : {
    1643         124 :     ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
    1644         124 :     Oid         userid = GetUserId();
    1645         124 :     bool        is_allowed_role = false;
    1646         124 :     char       *qbuffer = NULL;
    1647         124 :     Size        qbuffer_size = 0;
    1648         124 :     Size        extent = 0;
    1649         124 :     int         gc_count = 0;
    1650             :     HASH_SEQ_STATUS hash_seq;
    1651             :     pgssEntry  *entry;
    1652             : 
    1653             :     /*
    1654             :      * Superusers or roles with the privileges of pg_read_all_stats members
    1655             :      * are allowed
    1656             :      */
    1657         124 :     is_allowed_role = has_privs_of_role(userid, ROLE_PG_READ_ALL_STATS);
    1658             : 
    1659             :     /* hash table must exist already */
    1660         124 :     if (!pgss || !pgss_hash)
    1661           0 :         ereport(ERROR,
    1662             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    1663             :                  errmsg("pg_stat_statements must be loaded via shared_preload_libraries")));
    1664             : 
    1665         124 :     InitMaterializedSRF(fcinfo, 0);
    1666             : 
    1667             :     /*
    1668             :      * Check we have the expected number of output arguments.  Aside from
    1669             :      * being a good safety check, we need a kluge here to detect API version
    1670             :      * 1.1, which was wedged into the code in an ill-considered way.
    1671             :      */
    1672         124 :     switch (rsinfo->setDesc->natts)
    1673             :     {
    1674           0 :         case PG_STAT_STATEMENTS_COLS_V1_0:
    1675           0 :             if (api_version != PGSS_V1_0)
    1676           0 :                 elog(ERROR, "incorrect number of output arguments");
    1677           0 :             break;
    1678           0 :         case PG_STAT_STATEMENTS_COLS_V1_1:
    1679             :             /* pg_stat_statements() should have told us 1.0 */
    1680           0 :             if (api_version != PGSS_V1_0)
    1681           0 :                 elog(ERROR, "incorrect number of output arguments");
    1682           0 :             api_version = PGSS_V1_1;
    1683           0 :             break;
    1684           0 :         case PG_STAT_STATEMENTS_COLS_V1_2:
    1685           0 :             if (api_version != PGSS_V1_2)
    1686           0 :                 elog(ERROR, "incorrect number of output arguments");
    1687           0 :             break;
    1688           2 :         case PG_STAT_STATEMENTS_COLS_V1_3:
    1689           2 :             if (api_version != PGSS_V1_3)
    1690           0 :                 elog(ERROR, "incorrect number of output arguments");
    1691           2 :             break;
    1692           2 :         case PG_STAT_STATEMENTS_COLS_V1_8:
    1693           2 :             if (api_version != PGSS_V1_8)
    1694           0 :                 elog(ERROR, "incorrect number of output arguments");
    1695           2 :             break;
    1696           2 :         case PG_STAT_STATEMENTS_COLS_V1_9:
    1697           2 :             if (api_version != PGSS_V1_9)
    1698           0 :                 elog(ERROR, "incorrect number of output arguments");
    1699           2 :             break;
    1700           2 :         case PG_STAT_STATEMENTS_COLS_V1_10:
    1701           2 :             if (api_version != PGSS_V1_10)
    1702           0 :                 elog(ERROR, "incorrect number of output arguments");
    1703           2 :             break;
    1704         116 :         case PG_STAT_STATEMENTS_COLS_V1_11:
    1705         116 :             if (api_version != PGSS_V1_11)
    1706           0 :                 elog(ERROR, "incorrect number of output arguments");
    1707         116 :             break;
    1708           0 :         default:
    1709           0 :             elog(ERROR, "incorrect number of output arguments");
    1710             :     }
    1711             : 
    1712             :     /*
    1713             :      * We'd like to load the query text file (if needed) while not holding any
    1714             :      * lock on pgss->lock.  In the worst case we'll have to do this again
    1715             :      * after we have the lock, but it's unlikely enough to make this a win
    1716             :      * despite occasional duplicated work.  We need to reload if anybody
    1717             :      * writes to the file (either a retail qtext_store(), or a garbage
    1718             :      * collection) between this point and where we've gotten shared lock.  If
    1719             :      * a qtext_store is actually in progress when we look, we might as well
    1720             :      * skip the speculative load entirely.
    1721             :      */
    1722         124 :     if (showtext)
    1723             :     {
    1724             :         int         n_writers;
    1725             : 
    1726             :         /* Take the mutex so we can examine variables */
    1727             :         {
    1728         124 :             volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    1729             : 
    1730         124 :             SpinLockAcquire(&s->mutex);
    1731         124 :             extent = s->extent;
    1732         124 :             n_writers = s->n_writers;
    1733         124 :             gc_count = s->gc_count;
    1734         124 :             SpinLockRelease(&s->mutex);
    1735             :         }
    1736             : 
    1737             :         /* No point in loading file now if there are active writers */
    1738         124 :         if (n_writers == 0)
    1739         124 :             qbuffer = qtext_load_file(&qbuffer_size);
    1740             :     }
    1741             : 
    1742             :     /*
    1743             :      * Get shared lock, load or reload the query text file if we must, and
    1744             :      * iterate over the hashtable entries.
    1745             :      *
    1746             :      * With a large hash table, we might be holding the lock rather longer
    1747             :      * than one could wish.  However, this only blocks creation of new hash
    1748             :      * table entries, and the larger the hash table the less likely that is to
    1749             :      * be needed.  So we can hope this is okay.  Perhaps someday we'll decide
    1750             :      * we need to partition the hash table to limit the time spent holding any
    1751             :      * one lock.
    1752             :      */
    1753         124 :     LWLockAcquire(pgss->lock, LW_SHARED);
    1754             : 
    1755         124 :     if (showtext)
    1756             :     {
    1757             :         /*
    1758             :          * Here it is safe to examine extent and gc_count without taking the
    1759             :          * mutex.  Note that although other processes might change
    1760             :          * pgss->extent just after we look at it, the strings they then write
    1761             :          * into the file cannot yet be referenced in the hashtable, so we
    1762             :          * don't care whether we see them or not.
    1763             :          *
    1764             :          * If qtext_load_file fails, we just press on; we'll return NULL for
    1765             :          * every query text.
    1766             :          */
    1767         124 :         if (qbuffer == NULL ||
    1768         124 :             pgss->extent != extent ||
    1769         124 :             pgss->gc_count != gc_count)
    1770             :         {
    1771           0 :             free(qbuffer);
    1772           0 :             qbuffer = qtext_load_file(&qbuffer_size);
    1773             :         }
    1774             :     }
    1775             : 
    1776         124 :     hash_seq_init(&hash_seq, pgss_hash);
    1777       51990 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    1778             :     {
    1779             :         Datum       values[PG_STAT_STATEMENTS_COLS];
    1780             :         bool        nulls[PG_STAT_STATEMENTS_COLS];
    1781       51866 :         int         i = 0;
    1782             :         Counters    tmp;
    1783             :         double      stddev;
    1784       51866 :         int64       queryid = entry->key.queryid;
    1785             :         TimestampTz stats_since;
    1786             :         TimestampTz minmax_stats_since;
    1787             : 
    1788       51866 :         memset(values, 0, sizeof(values));
    1789       51866 :         memset(nulls, 0, sizeof(nulls));
    1790             : 
    1791       51866 :         values[i++] = ObjectIdGetDatum(entry->key.userid);
    1792       51866 :         values[i++] = ObjectIdGetDatum(entry->key.dbid);
    1793       51866 :         if (api_version >= PGSS_V1_9)
    1794       51840 :             values[i++] = BoolGetDatum(entry->key.toplevel);
    1795             : 
    1796       51866 :         if (is_allowed_role || entry->key.userid == userid)
    1797             :         {
    1798       51866 :             if (api_version >= PGSS_V1_2)
    1799       51866 :                 values[i++] = Int64GetDatumFast(queryid);
    1800             : 
    1801       51866 :             if (showtext)
    1802             :             {
    1803       51866 :                 char       *qstr = qtext_fetch(entry->query_offset,
    1804             :                                                entry->query_len,
    1805             :                                                qbuffer,
    1806             :                                                qbuffer_size);
    1807             : 
    1808       51866 :                 if (qstr)
    1809             :                 {
    1810             :                     char       *enc;
    1811             : 
    1812       51866 :                     enc = pg_any_to_server(qstr,
    1813             :                                            entry->query_len,
    1814             :                                            entry->encoding);
    1815             : 
    1816       51866 :                     values[i++] = CStringGetTextDatum(enc);
    1817             : 
    1818       51866 :                     if (enc != qstr)
    1819           0 :                         pfree(enc);
    1820             :                 }
    1821             :                 else
    1822             :                 {
    1823             :                     /* Just return a null if we fail to find the text */
    1824           0 :                     nulls[i++] = true;
    1825             :                 }
    1826             :             }
    1827             :             else
    1828             :             {
    1829             :                 /* Query text not requested */
    1830           0 :                 nulls[i++] = true;
    1831             :             }
    1832             :         }
    1833             :         else
    1834             :         {
    1835             :             /* Don't show queryid */
    1836           0 :             if (api_version >= PGSS_V1_2)
    1837           0 :                 nulls[i++] = true;
    1838             : 
    1839             :             /*
    1840             :              * Don't show query text, but hint as to the reason for not doing
    1841             :              * so if it was requested
    1842             :              */
    1843           0 :             if (showtext)
    1844           0 :                 values[i++] = CStringGetTextDatum("<insufficient privilege>");
    1845             :             else
    1846           0 :                 nulls[i++] = true;
    1847             :         }
    1848             : 
    1849             :         /* copy counters to a local variable to keep locking time short */
    1850             :         {
    1851       51866 :             volatile pgssEntry *e = (volatile pgssEntry *) entry;
    1852             : 
    1853       51866 :             SpinLockAcquire(&e->mutex);
    1854       51866 :             tmp = e->counters;
    1855       51866 :             stats_since = e->stats_since;
    1856       51866 :             minmax_stats_since = e->minmax_stats_since;
    1857       51866 :             SpinLockRelease(&e->mutex);
    1858             :         }
    1859             : 
    1860             :         /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
    1861       51866 :         if (IS_STICKY(tmp))
    1862        1336 :             continue;
    1863             : 
    1864             :         /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
    1865      151590 :         for (int kind = 0; kind < PGSS_NUMKIND; kind++)
    1866             :         {
    1867      101060 :             if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
    1868             :             {
    1869      101052 :                 values[i++] = Int64GetDatumFast(tmp.calls[kind]);
    1870      101052 :                 values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
    1871             :             }
    1872             : 
    1873      101060 :             if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
    1874             :                 api_version >= PGSS_V1_8)
    1875             :             {
    1876      101052 :                 values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
    1877      101052 :                 values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
    1878      101052 :                 values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
    1879             : 
    1880             :                 /*
    1881             :                  * Note we are calculating the population variance here, not
    1882             :                  * the sample variance, as we have data for the whole
    1883             :                  * population, so Bessel's correction is not used, and we
    1884             :                  * don't divide by tmp.calls - 1.
    1885             :                  */
    1886      101052 :                 if (tmp.calls[kind] > 1)
    1887        9018 :                     stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
    1888             :                 else
    1889       92034 :                     stddev = 0.0;
    1890      101052 :                 values[i++] = Float8GetDatumFast(stddev);
    1891             :             }
    1892             :         }
    1893       50530 :         values[i++] = Int64GetDatumFast(tmp.rows);
    1894       50530 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
    1895       50530 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
    1896       50530 :         if (api_version >= PGSS_V1_1)
    1897       50530 :             values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
    1898       50530 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
    1899       50530 :         values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
    1900       50530 :         values[i++] = Int64GetDatumFast(tmp.local_blks_read);
    1901       50530 :         if (api_version >= PGSS_V1_1)
    1902       50530 :             values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
    1903       50530 :         values[i++] = Int64GetDatumFast(tmp.local_blks_written);
    1904       50530 :         values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
    1905       50530 :         values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
    1906       50530 :         if (api_version >= PGSS_V1_1)
    1907             :         {
    1908       50530 :             values[i++] = Float8GetDatumFast(tmp.shared_blk_read_time);
    1909       50530 :             values[i++] = Float8GetDatumFast(tmp.shared_blk_write_time);
    1910             :         }
    1911       50530 :         if (api_version >= PGSS_V1_11)
    1912             :         {
    1913       50468 :             values[i++] = Float8GetDatumFast(tmp.local_blk_read_time);
    1914       50468 :             values[i++] = Float8GetDatumFast(tmp.local_blk_write_time);
    1915             :         }
    1916       50530 :         if (api_version >= PGSS_V1_10)
    1917             :         {
    1918       50490 :             values[i++] = Float8GetDatumFast(tmp.temp_blk_read_time);
    1919       50490 :             values[i++] = Float8GetDatumFast(tmp.temp_blk_write_time);
    1920             :         }
    1921       50530 :         if (api_version >= PGSS_V1_8)
    1922             :         {
    1923             :             char        buf[256];
    1924             :             Datum       wal_bytes;
    1925             : 
    1926       50522 :             values[i++] = Int64GetDatumFast(tmp.wal_records);
    1927       50522 :             values[i++] = Int64GetDatumFast(tmp.wal_fpi);
    1928             : 
    1929       50522 :             snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
    1930             : 
    1931             :             /* Convert to numeric. */
    1932       50522 :             wal_bytes = DirectFunctionCall3(numeric_in,
    1933             :                                             CStringGetDatum(buf),
    1934             :                                             ObjectIdGetDatum(0),
    1935             :                                             Int32GetDatum(-1));
    1936       50522 :             values[i++] = wal_bytes;
    1937             :         }
    1938       50530 :         if (api_version >= PGSS_V1_10)
    1939             :         {
    1940       50490 :             values[i++] = Int64GetDatumFast(tmp.jit_functions);
    1941       50490 :             values[i++] = Float8GetDatumFast(tmp.jit_generation_time);
    1942       50490 :             values[i++] = Int64GetDatumFast(tmp.jit_inlining_count);
    1943       50490 :             values[i++] = Float8GetDatumFast(tmp.jit_inlining_time);
    1944       50490 :             values[i++] = Int64GetDatumFast(tmp.jit_optimization_count);
    1945       50490 :             values[i++] = Float8GetDatumFast(tmp.jit_optimization_time);
    1946       50490 :             values[i++] = Int64GetDatumFast(tmp.jit_emission_count);
    1947       50490 :             values[i++] = Float8GetDatumFast(tmp.jit_emission_time);
    1948             :         }
    1949       50530 :         if (api_version >= PGSS_V1_11)
    1950             :         {
    1951       50468 :             values[i++] = Int64GetDatumFast(tmp.jit_deform_count);
    1952       50468 :             values[i++] = Float8GetDatumFast(tmp.jit_deform_time);
    1953       50468 :             values[i++] = TimestampTzGetDatum(stats_since);
    1954       50468 :             values[i++] = TimestampTzGetDatum(minmax_stats_since);
    1955             :         }
    1956             : 
    1957             :         Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
    1958             :                      api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
    1959             :                      api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
    1960             :                      api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
    1961             :                      api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
    1962             :                      api_version == PGSS_V1_9 ? PG_STAT_STATEMENTS_COLS_V1_9 :
    1963             :                      api_version == PGSS_V1_10 ? PG_STAT_STATEMENTS_COLS_V1_10 :
    1964             :                      api_version == PGSS_V1_11 ? PG_STAT_STATEMENTS_COLS_V1_11 :
    1965             :                      -1 /* fail if you forget to update this assert */ ));
    1966             : 
    1967       50530 :         tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
    1968             :     }
    1969             : 
    1970         124 :     LWLockRelease(pgss->lock);
    1971             : 
    1972         124 :     free(qbuffer);
    1973         124 : }
    1974             : 
    1975             : /* Number of output arguments (columns) for pg_stat_statements_info */
    1976             : #define PG_STAT_STATEMENTS_INFO_COLS    2
    1977             : 
    1978             : /*
    1979             :  * Return statistics of pg_stat_statements.
    1980             :  */
    1981             : Datum
    1982           4 : pg_stat_statements_info(PG_FUNCTION_ARGS)
    1983             : {
    1984             :     pgssGlobalStats stats;
    1985             :     TupleDesc   tupdesc;
    1986           4 :     Datum       values[PG_STAT_STATEMENTS_INFO_COLS] = {0};
    1987           4 :     bool        nulls[PG_STAT_STATEMENTS_INFO_COLS] = {0};
    1988             : 
    1989           4 :     if (!pgss || !pgss_hash)
    1990           0 :         ereport(ERROR,
    1991             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    1992             :                  errmsg("pg_stat_statements must be loaded via shared_preload_libraries")));
    1993             : 
    1994             :     /* Build a tuple descriptor for our result type */
    1995           4 :     if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
    1996           0 :         elog(ERROR, "return type must be a row type");
    1997             : 
    1998             :     /* Read global statistics for pg_stat_statements */
    1999             :     {
    2000           4 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2001             : 
    2002           4 :         SpinLockAcquire(&s->mutex);
    2003           4 :         stats = s->stats;
    2004           4 :         SpinLockRelease(&s->mutex);
    2005             :     }
    2006             : 
    2007           4 :     values[0] = Int64GetDatum(stats.dealloc);
    2008           4 :     values[1] = TimestampTzGetDatum(stats.stats_reset);
    2009             : 
    2010           4 :     PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
    2011             : }
    2012             : 
    2013             : /*
    2014             :  * Estimate shared memory space needed.
    2015             :  */
    2016             : static Size
    2017          12 : pgss_memsize(void)
    2018             : {
    2019             :     Size        size;
    2020             : 
    2021          12 :     size = MAXALIGN(sizeof(pgssSharedState));
    2022          12 :     size = add_size(size, hash_estimate_size(pgss_max, sizeof(pgssEntry)));
    2023             : 
    2024          12 :     return size;
    2025             : }
    2026             : 
    2027             : /*
    2028             :  * Allocate a new hashtable entry.
    2029             :  * caller must hold an exclusive lock on pgss->lock
    2030             :  *
    2031             :  * "query" need not be null-terminated; we rely on query_len instead
    2032             :  *
    2033             :  * If "sticky" is true, make the new entry artificially sticky so that it will
    2034             :  * probably still be there when the query finishes execution.  We do this by
    2035             :  * giving it a median usage value rather than the normal value.  (Strictly
    2036             :  * speaking, query strings are normalized on a best effort basis, though it
    2037             :  * would be difficult to demonstrate this even under artificial conditions.)
    2038             :  *
    2039             :  * Note: despite needing exclusive lock, it's not an error for the target
    2040             :  * entry to already exist.  This is because pgss_store releases and
    2041             :  * reacquires lock after failing to find a match; so someone else could
    2042             :  * have made the entry while we waited to get exclusive lock.
    2043             :  */
    2044             : static pgssEntry *
    2045       53318 : entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
    2046             :             bool sticky)
    2047             : {
    2048             :     pgssEntry  *entry;
    2049             :     bool        found;
    2050             : 
    2051             :     /* Make space if needed */
    2052       53318 :     while (hash_get_num_entries(pgss_hash) >= pgss_max)
    2053           0 :         entry_dealloc();
    2054             : 
    2055             :     /* Find or create an entry with desired hash code */
    2056       53318 :     entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
    2057             : 
    2058       53318 :     if (!found)
    2059             :     {
    2060             :         /* New entry, initialize it */
    2061             : 
    2062             :         /* reset the statistics */
    2063       53318 :         memset(&entry->counters, 0, sizeof(Counters));
    2064             :         /* set the appropriate initial usage count */
    2065       53318 :         entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
    2066             :         /* re-initialize the mutex each time ... we assume no one using it */
    2067       53318 :         SpinLockInit(&entry->mutex);
    2068             :         /* ... and don't forget the query text metadata */
    2069             :         Assert(query_len >= 0);
    2070       53318 :         entry->query_offset = query_offset;
    2071       53318 :         entry->query_len = query_len;
    2072       53318 :         entry->encoding = encoding;
    2073       53318 :         entry->stats_since = GetCurrentTimestamp();
    2074       53318 :         entry->minmax_stats_since = entry->stats_since;
    2075             :     }
    2076             : 
    2077       53318 :     return entry;
    2078             : }
    2079             : 
    2080             : /*
    2081             :  * qsort comparator for sorting into increasing usage order
    2082             :  */
    2083             : static int
    2084           0 : entry_cmp(const void *lhs, const void *rhs)
    2085             : {
    2086           0 :     double      l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
    2087           0 :     double      r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
    2088             : 
    2089           0 :     if (l_usage < r_usage)
    2090           0 :         return -1;
    2091           0 :     else if (l_usage > r_usage)
    2092           0 :         return +1;
    2093             :     else
    2094           0 :         return 0;
    2095             : }
    2096             : 
    2097             : /*
    2098             :  * Deallocate least-used entries.
    2099             :  *
    2100             :  * Caller must hold an exclusive lock on pgss->lock.
    2101             :  */
    2102             : static void
    2103           0 : entry_dealloc(void)
    2104             : {
    2105             :     HASH_SEQ_STATUS hash_seq;
    2106             :     pgssEntry **entries;
    2107             :     pgssEntry  *entry;
    2108             :     int         nvictims;
    2109             :     int         i;
    2110             :     Size        tottextlen;
    2111             :     int         nvalidtexts;
    2112             : 
    2113             :     /*
    2114             :      * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
    2115             :      * While we're scanning the table, apply the decay factor to the usage
    2116             :      * values, and update the mean query length.
    2117             :      *
    2118             :      * Note that the mean query length is almost immediately obsolete, since
    2119             :      * we compute it before not after discarding the least-used entries.
    2120             :      * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
    2121             :      * making two passes to get a more current result.  Likewise, the new
    2122             :      * cur_median_usage includes the entries we're about to zap.
    2123             :      */
    2124             : 
    2125           0 :     entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
    2126             : 
    2127           0 :     i = 0;
    2128           0 :     tottextlen = 0;
    2129           0 :     nvalidtexts = 0;
    2130             : 
    2131           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2132           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2133             :     {
    2134           0 :         entries[i++] = entry;
    2135             :         /* "Sticky" entries get a different usage decay rate. */
    2136           0 :         if (IS_STICKY(entry->counters))
    2137           0 :             entry->counters.usage *= STICKY_DECREASE_FACTOR;
    2138             :         else
    2139           0 :             entry->counters.usage *= USAGE_DECREASE_FACTOR;
    2140             :         /* In the mean length computation, ignore dropped texts. */
    2141           0 :         if (entry->query_len >= 0)
    2142             :         {
    2143           0 :             tottextlen += entry->query_len + 1;
    2144           0 :             nvalidtexts++;
    2145             :         }
    2146             :     }
    2147             : 
    2148             :     /* Sort into increasing order by usage */
    2149           0 :     qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
    2150             : 
    2151             :     /* Record the (approximate) median usage */
    2152           0 :     if (i > 0)
    2153           0 :         pgss->cur_median_usage = entries[i / 2]->counters.usage;
    2154             :     /* Record the mean query length */
    2155           0 :     if (nvalidtexts > 0)
    2156           0 :         pgss->mean_query_len = tottextlen / nvalidtexts;
    2157             :     else
    2158           0 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2159             : 
    2160             :     /* Now zap an appropriate fraction of lowest-usage entries */
    2161           0 :     nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
    2162           0 :     nvictims = Min(nvictims, i);
    2163             : 
    2164           0 :     for (i = 0; i < nvictims; i++)
    2165             :     {
    2166           0 :         hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
    2167             :     }
    2168             : 
    2169           0 :     pfree(entries);
    2170             : 
    2171             :     /* Increment the number of times entries are deallocated */
    2172             :     {
    2173           0 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2174             : 
    2175           0 :         SpinLockAcquire(&s->mutex);
    2176           0 :         s->stats.dealloc += 1;
    2177           0 :         SpinLockRelease(&s->mutex);
    2178             :     }
    2179           0 : }
    2180             : 
    2181             : /*
    2182             :  * Given a query string (not necessarily null-terminated), allocate a new
    2183             :  * entry in the external query text file and store the string there.
    2184             :  *
    2185             :  * If successful, returns true, and stores the new entry's offset in the file
    2186             :  * into *query_offset.  Also, if gc_count isn't NULL, *gc_count is set to the
    2187             :  * number of garbage collections that have occurred so far.
    2188             :  *
    2189             :  * On failure, returns false.
    2190             :  *
    2191             :  * At least a shared lock on pgss->lock must be held by the caller, so as
    2192             :  * to prevent a concurrent garbage collection.  Share-lock-holding callers
    2193             :  * should pass a gc_count pointer to obtain the number of garbage collections,
    2194             :  * so that they can recheck the count after obtaining exclusive lock to
    2195             :  * detect whether a garbage collection occurred (and removed this entry).
    2196             :  */
    2197             : static bool
    2198       53310 : qtext_store(const char *query, int query_len,
    2199             :             Size *query_offset, int *gc_count)
    2200             : {
    2201             :     Size        off;
    2202             :     int         fd;
    2203             : 
    2204             :     /*
    2205             :      * We use a spinlock to protect extent/n_writers/gc_count, so that
    2206             :      * multiple processes may execute this function concurrently.
    2207             :      */
    2208             :     {
    2209       53310 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2210             : 
    2211       53310 :         SpinLockAcquire(&s->mutex);
    2212       53310 :         off = s->extent;
    2213       53310 :         s->extent += query_len + 1;
    2214       53310 :         s->n_writers++;
    2215       53310 :         if (gc_count)
    2216       53310 :             *gc_count = s->gc_count;
    2217       53310 :         SpinLockRelease(&s->mutex);
    2218             :     }
    2219             : 
    2220       53310 :     *query_offset = off;
    2221             : 
    2222             :     /*
    2223             :      * Don't allow the file to grow larger than what qtext_load_file can
    2224             :      * (theoretically) handle.  This has been seen to be reachable on 32-bit
    2225             :      * platforms.
    2226             :      */
    2227       53310 :     if (unlikely(query_len >= MaxAllocHugeSize - off))
    2228             :     {
    2229           0 :         errno = EFBIG;          /* not quite right, but it'll do */
    2230           0 :         fd = -1;
    2231           0 :         goto error;
    2232             :     }
    2233             : 
    2234             :     /* Now write the data into the successfully-reserved part of the file */
    2235       53310 :     fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
    2236       53310 :     if (fd < 0)
    2237           0 :         goto error;
    2238             : 
    2239       53310 :     if (pg_pwrite(fd, query, query_len, off) != query_len)
    2240           0 :         goto error;
    2241       53310 :     if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
    2242           0 :         goto error;
    2243             : 
    2244       53310 :     CloseTransientFile(fd);
    2245             : 
    2246             :     /* Mark our write complete */
    2247             :     {
    2248       53310 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2249             : 
    2250       53310 :         SpinLockAcquire(&s->mutex);
    2251       53310 :         s->n_writers--;
    2252       53310 :         SpinLockRelease(&s->mutex);
    2253             :     }
    2254             : 
    2255       53310 :     return true;
    2256             : 
    2257           0 : error:
    2258           0 :     ereport(LOG,
    2259             :             (errcode_for_file_access(),
    2260             :              errmsg("could not write file \"%s\": %m",
    2261             :                     PGSS_TEXT_FILE)));
    2262             : 
    2263           0 :     if (fd >= 0)
    2264           0 :         CloseTransientFile(fd);
    2265             : 
    2266             :     /* Mark our write complete */
    2267             :     {
    2268           0 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2269             : 
    2270           0 :         SpinLockAcquire(&s->mutex);
    2271           0 :         s->n_writers--;
    2272           0 :         SpinLockRelease(&s->mutex);
    2273             :     }
    2274             : 
    2275           0 :     return false;
    2276             : }
    2277             : 
    2278             : /*
    2279             :  * Read the external query text file into a malloc'd buffer.
    2280             :  *
    2281             :  * Returns NULL (without throwing an error) if unable to read, eg
    2282             :  * file not there or insufficient memory.
    2283             :  *
    2284             :  * On success, the buffer size is also returned into *buffer_size.
    2285             :  *
    2286             :  * This can be called without any lock on pgss->lock, but in that case
    2287             :  * the caller is responsible for verifying that the result is sane.
    2288             :  */
    2289             : static char *
    2290         132 : qtext_load_file(Size *buffer_size)
    2291             : {
    2292             :     char       *buf;
    2293             :     int         fd;
    2294             :     struct stat stat;
    2295             :     Size        nread;
    2296             : 
    2297         132 :     fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
    2298         132 :     if (fd < 0)
    2299             :     {
    2300           0 :         if (errno != ENOENT)
    2301           0 :             ereport(LOG,
    2302             :                     (errcode_for_file_access(),
    2303             :                      errmsg("could not read file \"%s\": %m",
    2304             :                             PGSS_TEXT_FILE)));
    2305           0 :         return NULL;
    2306             :     }
    2307             : 
    2308             :     /* Get file length */
    2309         132 :     if (fstat(fd, &stat))
    2310             :     {
    2311           0 :         ereport(LOG,
    2312             :                 (errcode_for_file_access(),
    2313             :                  errmsg("could not stat file \"%s\": %m",
    2314             :                         PGSS_TEXT_FILE)));
    2315           0 :         CloseTransientFile(fd);
    2316           0 :         return NULL;
    2317             :     }
    2318             : 
    2319             :     /* Allocate buffer; beware that off_t might be wider than size_t */
    2320         132 :     if (stat.st_size <= MaxAllocHugeSize)
    2321         132 :         buf = (char *) malloc(stat.st_size);
    2322             :     else
    2323           0 :         buf = NULL;
    2324         132 :     if (buf == NULL)
    2325             :     {
    2326           0 :         ereport(LOG,
    2327             :                 (errcode(ERRCODE_OUT_OF_MEMORY),
    2328             :                  errmsg("out of memory"),
    2329             :                  errdetail("Could not allocate enough memory to read file \"%s\".",
    2330             :                            PGSS_TEXT_FILE)));
    2331           0 :         CloseTransientFile(fd);
    2332           0 :         return NULL;
    2333             :     }
    2334             : 
    2335             :     /*
    2336             :      * OK, slurp in the file.  Windows fails if we try to read more than
    2337             :      * INT_MAX bytes at once, and other platforms might not like that either,
    2338             :      * so read a very large file in 1GB segments.
    2339             :      */
    2340         132 :     nread = 0;
    2341         262 :     while (nread < stat.st_size)
    2342             :     {
    2343         130 :         int         toread = Min(1024 * 1024 * 1024, stat.st_size - nread);
    2344             : 
    2345             :         /*
    2346             :          * If we get a short read and errno doesn't get set, the reason is
    2347             :          * probably that garbage collection truncated the file since we did
    2348             :          * the fstat(), so we don't log a complaint --- but we don't return
    2349             :          * the data, either, since it's most likely corrupt due to concurrent
    2350             :          * writes from garbage collection.
    2351             :          */
    2352         130 :         errno = 0;
    2353         130 :         if (read(fd, buf + nread, toread) != toread)
    2354             :         {
    2355           0 :             if (errno)
    2356           0 :                 ereport(LOG,
    2357             :                         (errcode_for_file_access(),
    2358             :                          errmsg("could not read file \"%s\": %m",
    2359             :                                 PGSS_TEXT_FILE)));
    2360           0 :             free(buf);
    2361           0 :             CloseTransientFile(fd);
    2362           0 :             return NULL;
    2363             :         }
    2364         130 :         nread += toread;
    2365             :     }
    2366             : 
    2367         132 :     if (CloseTransientFile(fd) != 0)
    2368           0 :         ereport(LOG,
    2369             :                 (errcode_for_file_access(),
    2370             :                  errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
    2371             : 
    2372         132 :     *buffer_size = nread;
    2373         132 :     return buf;
    2374             : }
    2375             : 
    2376             : /*
    2377             :  * Locate a query text in the file image previously read by qtext_load_file().
    2378             :  *
    2379             :  * We validate the given offset/length, and return NULL if bogus.  Otherwise,
    2380             :  * the result points to a null-terminated string within the buffer.
    2381             :  */
    2382             : static char *
    2383      104438 : qtext_fetch(Size query_offset, int query_len,
    2384             :             char *buffer, Size buffer_size)
    2385             : {
    2386             :     /* File read failed? */
    2387      104438 :     if (buffer == NULL)
    2388           0 :         return NULL;
    2389             :     /* Bogus offset/length? */
    2390      104438 :     if (query_len < 0 ||
    2391      104438 :         query_offset + query_len >= buffer_size)
    2392           0 :         return NULL;
    2393             :     /* As a further sanity check, make sure there's a trailing null */
    2394      104438 :     if (buffer[query_offset + query_len] != '\0')
    2395           0 :         return NULL;
    2396             :     /* Looks OK */
    2397      104438 :     return buffer + query_offset;
    2398             : }
    2399             : 
    2400             : /*
    2401             :  * Do we need to garbage-collect the external query text file?
    2402             :  *
    2403             :  * Caller should hold at least a shared lock on pgss->lock.
    2404             :  */
    2405             : static bool
    2406       53310 : need_gc_qtexts(void)
    2407             : {
    2408             :     Size        extent;
    2409             : 
    2410             :     /* Read shared extent pointer */
    2411             :     {
    2412       53310 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2413             : 
    2414       53310 :         SpinLockAcquire(&s->mutex);
    2415       53310 :         extent = s->extent;
    2416       53310 :         SpinLockRelease(&s->mutex);
    2417             :     }
    2418             : 
    2419             :     /*
    2420             :      * Don't proceed if file does not exceed 512 bytes per possible entry.
    2421             :      *
    2422             :      * Here and in the next test, 32-bit machines have overflow hazards if
    2423             :      * pgss_max and/or mean_query_len are large.  Force the multiplications
    2424             :      * and comparisons to be done in uint64 arithmetic to forestall trouble.
    2425             :      */
    2426       53310 :     if ((uint64) extent < (uint64) 512 * pgss_max)
    2427       53310 :         return false;
    2428             : 
    2429             :     /*
    2430             :      * Don't proceed if file is less than about 50% bloat.  Nothing can or
    2431             :      * should be done in the event of unusually large query texts accounting
    2432             :      * for file's large size.  We go to the trouble of maintaining the mean
    2433             :      * query length in order to prevent garbage collection from thrashing
    2434             :      * uselessly.
    2435             :      */
    2436           0 :     if ((uint64) extent < (uint64) pgss->mean_query_len * pgss_max * 2)
    2437           0 :         return false;
    2438             : 
    2439           0 :     return true;
    2440             : }
    2441             : 
    2442             : /*
    2443             :  * Garbage-collect orphaned query texts in external file.
    2444             :  *
    2445             :  * This won't be called often in the typical case, since it's likely that
    2446             :  * there won't be too much churn, and besides, a similar compaction process
    2447             :  * occurs when serializing to disk at shutdown or as part of resetting.
    2448             :  * Despite this, it seems prudent to plan for the edge case where the file
    2449             :  * becomes unreasonably large, with no other method of compaction likely to
    2450             :  * occur in the foreseeable future.
    2451             :  *
    2452             :  * The caller must hold an exclusive lock on pgss->lock.
    2453             :  *
    2454             :  * At the first sign of trouble we unlink the query text file to get a clean
    2455             :  * slate (although existing statistics are retained), rather than risk
    2456             :  * thrashing by allowing the same problem case to recur indefinitely.
    2457             :  */
    2458             : static void
    2459           0 : gc_qtexts(void)
    2460             : {
    2461             :     char       *qbuffer;
    2462             :     Size        qbuffer_size;
    2463           0 :     FILE       *qfile = NULL;
    2464             :     HASH_SEQ_STATUS hash_seq;
    2465             :     pgssEntry  *entry;
    2466             :     Size        extent;
    2467             :     int         nentries;
    2468             : 
    2469             :     /*
    2470             :      * When called from pgss_store, some other session might have proceeded
    2471             :      * with garbage collection in the no-lock-held interim of lock strength
    2472             :      * escalation.  Check once more that this is actually necessary.
    2473             :      */
    2474           0 :     if (!need_gc_qtexts())
    2475           0 :         return;
    2476             : 
    2477             :     /*
    2478             :      * Load the old texts file.  If we fail (out of memory, for instance),
    2479             :      * invalidate query texts.  Hopefully this is rare.  It might seem better
    2480             :      * to leave things alone on an OOM failure, but the problem is that the
    2481             :      * file is only going to get bigger; hoping for a future non-OOM result is
    2482             :      * risky and can easily lead to complete denial of service.
    2483             :      */
    2484           0 :     qbuffer = qtext_load_file(&qbuffer_size);
    2485           0 :     if (qbuffer == NULL)
    2486           0 :         goto gc_fail;
    2487             : 
    2488             :     /*
    2489             :      * We overwrite the query texts file in place, so as to reduce the risk of
    2490             :      * an out-of-disk-space failure.  Since the file is guaranteed not to get
    2491             :      * larger, this should always work on traditional filesystems; though we
    2492             :      * could still lose on copy-on-write filesystems.
    2493             :      */
    2494           0 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2495           0 :     if (qfile == NULL)
    2496             :     {
    2497           0 :         ereport(LOG,
    2498             :                 (errcode_for_file_access(),
    2499             :                  errmsg("could not write file \"%s\": %m",
    2500             :                         PGSS_TEXT_FILE)));
    2501           0 :         goto gc_fail;
    2502             :     }
    2503             : 
    2504           0 :     extent = 0;
    2505           0 :     nentries = 0;
    2506             : 
    2507           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2508           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2509             :     {
    2510           0 :         int         query_len = entry->query_len;
    2511           0 :         char       *qry = qtext_fetch(entry->query_offset,
    2512             :                                       query_len,
    2513             :                                       qbuffer,
    2514             :                                       qbuffer_size);
    2515             : 
    2516           0 :         if (qry == NULL)
    2517             :         {
    2518             :             /* Trouble ... drop the text */
    2519           0 :             entry->query_offset = 0;
    2520           0 :             entry->query_len = -1;
    2521             :             /* entry will not be counted in mean query length computation */
    2522           0 :             continue;
    2523             :         }
    2524             : 
    2525           0 :         if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
    2526             :         {
    2527           0 :             ereport(LOG,
    2528             :                     (errcode_for_file_access(),
    2529             :                      errmsg("could not write file \"%s\": %m",
    2530             :                             PGSS_TEXT_FILE)));
    2531           0 :             hash_seq_term(&hash_seq);
    2532           0 :             goto gc_fail;
    2533             :         }
    2534             : 
    2535           0 :         entry->query_offset = extent;
    2536           0 :         extent += query_len + 1;
    2537           0 :         nentries++;
    2538             :     }
    2539             : 
    2540             :     /*
    2541             :      * Truncate away any now-unused space.  If this fails for some odd reason,
    2542             :      * we log it, but there's no need to fail.
    2543             :      */
    2544           0 :     if (ftruncate(fileno(qfile), extent) != 0)
    2545           0 :         ereport(LOG,
    2546             :                 (errcode_for_file_access(),
    2547             :                  errmsg("could not truncate file \"%s\": %m",
    2548             :                         PGSS_TEXT_FILE)));
    2549             : 
    2550           0 :     if (FreeFile(qfile))
    2551             :     {
    2552           0 :         ereport(LOG,
    2553             :                 (errcode_for_file_access(),
    2554             :                  errmsg("could not write file \"%s\": %m",
    2555             :                         PGSS_TEXT_FILE)));
    2556           0 :         qfile = NULL;
    2557           0 :         goto gc_fail;
    2558             :     }
    2559             : 
    2560           0 :     elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
    2561             :          pgss->extent, extent);
    2562             : 
    2563             :     /* Reset the shared extent pointer */
    2564           0 :     pgss->extent = extent;
    2565             : 
    2566             :     /*
    2567             :      * Also update the mean query length, to be sure that need_gc_qtexts()
    2568             :      * won't still think we have a problem.
    2569             :      */
    2570           0 :     if (nentries > 0)
    2571           0 :         pgss->mean_query_len = extent / nentries;
    2572             :     else
    2573           0 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2574             : 
    2575           0 :     free(qbuffer);
    2576             : 
    2577             :     /*
    2578             :      * OK, count a garbage collection cycle.  (Note: even though we have
    2579             :      * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
    2580             :      * other processes may examine gc_count while holding only the mutex.
    2581             :      * Also, we have to advance the count *after* we've rewritten the file,
    2582             :      * else other processes might not realize they read a stale file.)
    2583             :      */
    2584           0 :     record_gc_qtexts();
    2585             : 
    2586           0 :     return;
    2587             : 
    2588           0 : gc_fail:
    2589             :     /* clean up resources */
    2590           0 :     if (qfile)
    2591           0 :         FreeFile(qfile);
    2592           0 :     free(qbuffer);
    2593             : 
    2594             :     /*
    2595             :      * Since the contents of the external file are now uncertain, mark all
    2596             :      * hashtable entries as having invalid texts.
    2597             :      */
    2598           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2599           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2600             :     {
    2601           0 :         entry->query_offset = 0;
    2602           0 :         entry->query_len = -1;
    2603             :     }
    2604             : 
    2605             :     /*
    2606             :      * Destroy the query text file and create a new, empty one
    2607             :      */
    2608           0 :     (void) unlink(PGSS_TEXT_FILE);
    2609           0 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2610           0 :     if (qfile == NULL)
    2611           0 :         ereport(LOG,
    2612             :                 (errcode_for_file_access(),
    2613             :                  errmsg("could not recreate file \"%s\": %m",
    2614             :                         PGSS_TEXT_FILE)));
    2615             :     else
    2616           0 :         FreeFile(qfile);
    2617             : 
    2618             :     /* Reset the shared extent pointer */
    2619           0 :     pgss->extent = 0;
    2620             : 
    2621             :     /* Reset mean_query_len to match the new state */
    2622           0 :     pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2623             : 
    2624             :     /*
    2625             :      * Bump the GC count even though we failed.
    2626             :      *
    2627             :      * This is needed to make concurrent readers of file without any lock on
    2628             :      * pgss->lock notice existence of new version of file.  Once readers
    2629             :      * subsequently observe a change in GC count with pgss->lock held, that
    2630             :      * forces a safe reopen of file.  Writers also require that we bump here,
    2631             :      * of course.  (As required by locking protocol, readers and writers don't
    2632             :      * trust earlier file contents until gc_count is found unchanged after
    2633             :      * pgss->lock acquired in shared or exclusive mode respectively.)
    2634             :      */
    2635           0 :     record_gc_qtexts();
    2636             : }
    2637             : 
    2638             : #define SINGLE_ENTRY_RESET(e) \
    2639             : if (e) { \
    2640             :     if (minmax_only) { \
    2641             :         /* When requested reset only min/max statistics of an entry */ \
    2642             :         for (int kind = 0; kind < PGSS_NUMKIND; kind++) \
    2643             :         { \
    2644             :             e->counters.max_time[kind] = 0; \
    2645             :             e->counters.min_time[kind] = 0; \
    2646             :         } \
    2647             :         e->minmax_stats_since = stats_reset; \
    2648             :     } \
    2649             :     else \
    2650             :     { \
    2651             :         /* Remove the key otherwise  */ \
    2652             :         hash_search(pgss_hash, &e->key, HASH_REMOVE, NULL); \
    2653             :         num_remove++; \
    2654             :     } \
    2655             : }
    2656             : 
    2657             : /*
    2658             :  * Reset entries corresponding to parameters passed.
    2659             :  */
    2660             : static TimestampTz
    2661         102 : entry_reset(Oid userid, Oid dbid, uint64 queryid, bool minmax_only)
    2662             : {
    2663             :     HASH_SEQ_STATUS hash_seq;
    2664             :     pgssEntry  *entry;
    2665             :     FILE       *qfile;
    2666             :     long        num_entries;
    2667         102 :     long        num_remove = 0;
    2668             :     pgssHashKey key;
    2669             :     TimestampTz stats_reset;
    2670             : 
    2671         102 :     if (!pgss || !pgss_hash)
    2672           0 :         ereport(ERROR,
    2673             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    2674             :                  errmsg("pg_stat_statements must be loaded via shared_preload_libraries")));
    2675             : 
    2676         102 :     LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
    2677         102 :     num_entries = hash_get_num_entries(pgss_hash);
    2678             : 
    2679         102 :     stats_reset = GetCurrentTimestamp();
    2680             : 
    2681         102 :     if (userid != 0 && dbid != 0 && queryid != UINT64CONST(0))
    2682             :     {
    2683             :         /* If all the parameters are available, use the fast path. */
    2684           2 :         memset(&key, 0, sizeof(pgssHashKey));
    2685           2 :         key.userid = userid;
    2686           2 :         key.dbid = dbid;
    2687           2 :         key.queryid = queryid;
    2688             : 
    2689             :         /*
    2690             :          * Reset the entry if it exists, starting with the non-top-level
    2691             :          * entry.
    2692             :          */
    2693           2 :         key.toplevel = false;
    2694           2 :         entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    2695             : 
    2696           2 :         SINGLE_ENTRY_RESET(entry);
    2697             : 
    2698             :         /* Also reset the top-level entry if it exists. */
    2699           2 :         key.toplevel = true;
    2700           2 :         entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    2701             : 
    2702           2 :         SINGLE_ENTRY_RESET(entry);
    2703             :     }
    2704         100 :     else if (userid != 0 || dbid != 0 || queryid != UINT64CONST(0))
    2705             :     {
    2706             :         /* Reset entries corresponding to valid parameters. */
    2707           8 :         hash_seq_init(&hash_seq, pgss_hash);
    2708          96 :         while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2709             :         {
    2710          88 :             if ((!userid || entry->key.userid == userid) &&
    2711          68 :                 (!dbid || entry->key.dbid == dbid) &&
    2712          64 :                 (!queryid || entry->key.queryid == queryid))
    2713             :             {
    2714          14 :                 SINGLE_ENTRY_RESET(entry);
    2715             :             }
    2716             :         }
    2717             :     }
    2718             :     else
    2719             :     {
    2720             :         /* Reset all entries. */
    2721          92 :         hash_seq_init(&hash_seq, pgss_hash);
    2722         840 :         while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2723             :         {
    2724         792 :             SINGLE_ENTRY_RESET(entry);
    2725             :         }
    2726             :     }
    2727             : 
    2728             :     /* All entries are removed? */
    2729         102 :     if (num_entries != num_remove)
    2730          12 :         goto release_lock;
    2731             : 
    2732             :     /*
    2733             :      * Reset global statistics for pg_stat_statements since all entries are
    2734             :      * removed.
    2735             :      */
    2736             :     {
    2737          90 :         volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
    2738             : 
    2739          90 :         SpinLockAcquire(&s->mutex);
    2740          90 :         s->stats.dealloc = 0;
    2741          90 :         s->stats.stats_reset = stats_reset;
    2742          90 :         SpinLockRelease(&s->mutex);
    2743             :     }
    2744             : 
    2745             :     /*
    2746             :      * Write new empty query file, perhaps even creating a new one to recover
    2747             :      * if the file was missing.
    2748             :      */
    2749          90 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2750          90 :     if (qfile == NULL)
    2751             :     {
    2752           0 :         ereport(LOG,
    2753             :                 (errcode_for_file_access(),
    2754             :                  errmsg("could not create file \"%s\": %m",
    2755             :                         PGSS_TEXT_FILE)));
    2756           0 :         goto done;
    2757             :     }
    2758             : 
    2759             :     /* If ftruncate fails, log it, but it's not a fatal problem */
    2760          90 :     if (ftruncate(fileno(qfile), 0) != 0)
    2761           0 :         ereport(LOG,
    2762             :                 (errcode_for_file_access(),
    2763             :                  errmsg("could not truncate file \"%s\": %m",
    2764             :                         PGSS_TEXT_FILE)));
    2765             : 
    2766          90 :     FreeFile(qfile);
    2767             : 
    2768          90 : done:
    2769          90 :     pgss->extent = 0;
    2770             :     /* This counts as a query text garbage collection for our purposes */
    2771          90 :     record_gc_qtexts();
    2772             : 
    2773         102 : release_lock:
    2774         102 :     LWLockRelease(pgss->lock);
    2775             : 
    2776         102 :     return stats_reset;
    2777             : }
    2778             : 
    2779             : /*
    2780             :  * Generate a normalized version of the query string that will be used to
    2781             :  * represent all similar queries.
    2782             :  *
    2783             :  * Note that the normalized representation may well vary depending on
    2784             :  * just which "equivalent" query is used to create the hashtable entry.
    2785             :  * We assume this is OK.
    2786             :  *
    2787             :  * If query_loc > 0, then "query" has been advanced by that much compared to
    2788             :  * the original string start, so we need to translate the provided locations
    2789             :  * to compensate.  (This lets us avoid re-scanning statements before the one
    2790             :  * of interest, so it's worth doing.)
    2791             :  *
    2792             :  * *query_len_p contains the input string length, and is updated with
    2793             :  * the result string length on exit.  The resulting string might be longer
    2794             :  * or shorter depending on what happens with replacement of constants.
    2795             :  *
    2796             :  * Returns a palloc'd string.
    2797             :  */
    2798             : static char *
    2799       19312 : generate_normalized_query(JumbleState *jstate, const char *query,
    2800             :                           int query_loc, int *query_len_p)
    2801             : {
    2802             :     char       *norm_query;
    2803       19312 :     int         query_len = *query_len_p;
    2804             :     int         i,
    2805             :                 norm_query_buflen,  /* Space allowed for norm_query */
    2806             :                 len_to_wrt,     /* Length (in bytes) to write */
    2807       19312 :                 quer_loc = 0,   /* Source query byte location */
    2808       19312 :                 n_quer_loc = 0, /* Normalized query byte location */
    2809       19312 :                 last_off = 0,   /* Offset from start for previous tok */
    2810       19312 :                 last_tok_len = 0;   /* Length (in bytes) of that tok */
    2811             : 
    2812             :     /*
    2813             :      * Get constants' lengths (core system only gives us locations).  Note
    2814             :      * this also ensures the items are sorted by location.
    2815             :      */
    2816       19312 :     fill_in_constant_lengths(jstate, query, query_loc);
    2817             : 
    2818             :     /*
    2819             :      * Allow for $n symbols to be longer than the constants they replace.
    2820             :      * Constants must take at least one byte in text form, while a $n symbol
    2821             :      * certainly isn't more than 11 bytes, even if n reaches INT_MAX.  We
    2822             :      * could refine that limit based on the max value of n for the current
    2823             :      * query, but it hardly seems worth any extra effort to do so.
    2824             :      */
    2825       19312 :     norm_query_buflen = query_len + jstate->clocations_count * 10;
    2826             : 
    2827             :     /* Allocate result buffer */
    2828       19312 :     norm_query = palloc(norm_query_buflen + 1);
    2829             : 
    2830       80230 :     for (i = 0; i < jstate->clocations_count; i++)
    2831             :     {
    2832             :         int         off,        /* Offset from start for cur tok */
    2833             :                     tok_len;    /* Length (in bytes) of that tok */
    2834             : 
    2835       60918 :         off = jstate->clocations[i].location;
    2836             :         /* Adjust recorded location if we're dealing with partial string */
    2837       60918 :         off -= query_loc;
    2838             : 
    2839       60918 :         tok_len = jstate->clocations[i].length;
    2840             : 
    2841       60918 :         if (tok_len < 0)
    2842         432 :             continue;           /* ignore any duplicates */
    2843             : 
    2844             :         /* Copy next chunk (what precedes the next constant) */
    2845       60486 :         len_to_wrt = off - last_off;
    2846       60486 :         len_to_wrt -= last_tok_len;
    2847             : 
    2848             :         Assert(len_to_wrt >= 0);
    2849       60486 :         memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2850       60486 :         n_quer_loc += len_to_wrt;
    2851             : 
    2852             :         /* And insert a param symbol in place of the constant token */
    2853      120972 :         n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d",
    2854       60486 :                               i + 1 + jstate->highest_extern_param_id);
    2855             : 
    2856       60486 :         quer_loc = off + tok_len;
    2857       60486 :         last_off = off;
    2858       60486 :         last_tok_len = tok_len;
    2859             :     }
    2860             : 
    2861             :     /*
    2862             :      * We've copied up until the last ignorable constant.  Copy over the
    2863             :      * remaining bytes of the original query string.
    2864             :      */
    2865       19312 :     len_to_wrt = query_len - quer_loc;
    2866             : 
    2867             :     Assert(len_to_wrt >= 0);
    2868       19312 :     memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2869       19312 :     n_quer_loc += len_to_wrt;
    2870             : 
    2871             :     Assert(n_quer_loc <= norm_query_buflen);
    2872       19312 :     norm_query[n_quer_loc] = '\0';
    2873             : 
    2874       19312 :     *query_len_p = n_quer_loc;
    2875       19312 :     return norm_query;
    2876             : }
    2877             : 
    2878             : /*
    2879             :  * Given a valid SQL string and an array of constant-location records,
    2880             :  * fill in the textual lengths of those constants.
    2881             :  *
    2882             :  * The constants may use any allowed constant syntax, such as float literals,
    2883             :  * bit-strings, single-quoted strings and dollar-quoted strings.  This is
    2884             :  * accomplished by using the public API for the core scanner.
    2885             :  *
    2886             :  * It is the caller's job to ensure that the string is a valid SQL statement
    2887             :  * with constants at the indicated locations.  Since in practice the string
    2888             :  * has already been parsed, and the locations that the caller provides will
    2889             :  * have originated from within the authoritative parser, this should not be
    2890             :  * a problem.
    2891             :  *
    2892             :  * Duplicate constant pointers are possible, and will have their lengths
    2893             :  * marked as '-1', so that they are later ignored.  (Actually, we assume the
    2894             :  * lengths were initialized as -1 to start with, and don't change them here.)
    2895             :  *
    2896             :  * If query_loc > 0, then "query" has been advanced by that much compared to
    2897             :  * the original string start, so we need to translate the provided locations
    2898             :  * to compensate.  (This lets us avoid re-scanning statements before the one
    2899             :  * of interest, so it's worth doing.)
    2900             :  *
    2901             :  * N.B. There is an assumption that a '-' character at a Const location begins
    2902             :  * a negative numeric constant.  This precludes there ever being another
    2903             :  * reason for a constant to start with a '-'.
    2904             :  */
    2905             : static void
    2906       19312 : fill_in_constant_lengths(JumbleState *jstate, const char *query,
    2907             :                          int query_loc)
    2908             : {
    2909             :     LocationLen *locs;
    2910             :     core_yyscan_t yyscanner;
    2911             :     core_yy_extra_type yyextra;
    2912             :     core_YYSTYPE yylval;
    2913             :     YYLTYPE     yylloc;
    2914       19312 :     int         last_loc = -1;
    2915             :     int         i;
    2916             : 
    2917             :     /*
    2918             :      * Sort the records by location so that we can process them in order while
    2919             :      * scanning the query text.
    2920             :      */
    2921       19312 :     if (jstate->clocations_count > 1)
    2922       12812 :         qsort(jstate->clocations, jstate->clocations_count,
    2923             :               sizeof(LocationLen), comp_location);
    2924       19312 :     locs = jstate->clocations;
    2925             : 
    2926             :     /* initialize the flex scanner --- should match raw_parser() */
    2927       19312 :     yyscanner = scanner_init(query,
    2928             :                              &yyextra,
    2929             :                              &ScanKeywords,
    2930             :                              ScanKeywordTokens);
    2931             : 
    2932             :     /* we don't want to re-emit any escape string warnings */
    2933       19312 :     yyextra.escape_string_warning = false;
    2934             : 
    2935             :     /* Search for each constant, in sequence */
    2936       80230 :     for (i = 0; i < jstate->clocations_count; i++)
    2937             :     {
    2938       60918 :         int         loc = locs[i].location;
    2939             :         int         tok;
    2940             : 
    2941             :         /* Adjust recorded location if we're dealing with partial string */
    2942       60918 :         loc -= query_loc;
    2943             : 
    2944             :         Assert(loc >= 0);
    2945             : 
    2946       60918 :         if (loc <= last_loc)
    2947         432 :             continue;           /* Duplicate constant, ignore */
    2948             : 
    2949             :         /* Lex tokens until we find the desired constant */
    2950             :         for (;;)
    2951             :         {
    2952      455414 :             tok = core_yylex(&yylval, &yylloc, yyscanner);
    2953             : 
    2954             :             /* We should not hit end-of-string, but if we do, behave sanely */
    2955      455414 :             if (tok == 0)
    2956           0 :                 break;          /* out of inner for-loop */
    2957             : 
    2958             :             /*
    2959             :              * We should find the token position exactly, but if we somehow
    2960             :              * run past it, work with that.
    2961             :              */
    2962      455414 :             if (yylloc >= loc)
    2963             :             {
    2964       60486 :                 if (query[loc] == '-')
    2965             :                 {
    2966             :                     /*
    2967             :                      * It's a negative value - this is the one and only case
    2968             :                      * where we replace more than a single token.
    2969             :                      *
    2970             :                      * Do not compensate for the core system's special-case
    2971             :                      * adjustment of location to that of the leading '-'
    2972             :                      * operator in the event of a negative constant.  It is
    2973             :                      * also useful for our purposes to start from the minus
    2974             :                      * symbol.  In this way, queries like "select * from foo
    2975             :                      * where bar = 1" and "select * from foo where bar = -2"
    2976             :                      * will have identical normalized query strings.
    2977             :                      */
    2978         804 :                     tok = core_yylex(&yylval, &yylloc, yyscanner);
    2979         804 :                     if (tok == 0)
    2980           0 :                         break;  /* out of inner for-loop */
    2981             :                 }
    2982             : 
    2983             :                 /*
    2984             :                  * We now rely on the assumption that flex has placed a zero
    2985             :                  * byte after the text of the current token in scanbuf.
    2986             :                  */
    2987       60486 :                 locs[i].length = strlen(yyextra.scanbuf + loc);
    2988       60486 :                 break;          /* out of inner for-loop */
    2989             :             }
    2990             :         }
    2991             : 
    2992             :         /* If we hit end-of-string, give up, leaving remaining lengths -1 */
    2993       60486 :         if (tok == 0)
    2994           0 :             break;
    2995             : 
    2996       60486 :         last_loc = loc;
    2997             :     }
    2998             : 
    2999       19312 :     scanner_finish(yyscanner);
    3000       19312 : }
    3001             : 
    3002             : /*
    3003             :  * comp_location: comparator for qsorting LocationLen structs by location
    3004             :  */
    3005             : static int
    3006       73044 : comp_location(const void *a, const void *b)
    3007             : {
    3008       73044 :     int         l = ((const LocationLen *) a)->location;
    3009       73044 :     int         r = ((const LocationLen *) b)->location;
    3010             : 
    3011       73044 :     return pg_cmp_s32(l, r);
    3012             : }

Generated by: LCOV version 1.14