LCOV - code coverage report
Current view: top level - contrib/pg_stat_statements - pg_stat_statements.c (source / functions) Hit Total Coverage
Test: PostgreSQL 19devel Lines: 743 983 75.6 %
Date: 2025-10-02 06:18:21 Functions: 46 53 86.8 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /*-------------------------------------------------------------------------
       2             :  *
       3             :  * pg_stat_statements.c
       4             :  *      Track statement planning and execution times as well as resource
       5             :  *      usage across a whole database cluster.
       6             :  *
       7             :  * Execution costs are totaled for each distinct source query, and kept in
       8             :  * a shared hashtable.  (We track only as many distinct queries as will fit
       9             :  * in the designated amount of shared memory.)
      10             :  *
      11             :  * Starting in Postgres 9.2, this module normalized query entries.  As of
      12             :  * Postgres 14, the normalization is done by the core if compute_query_id is
      13             :  * enabled, or optionally by third-party modules.
      14             :  *
      15             :  * To facilitate presenting entries to users, we create "representative" query
      16             :  * strings in which constants are replaced with parameter symbols ($n), to
      17             :  * make it clearer what a normalized entry can represent.  To save on shared
      18             :  * memory, and to avoid having to truncate oversized query strings, we store
      19             :  * these strings in a temporary external query-texts file.  Offsets into this
      20             :  * file are kept in shared memory.
      21             :  *
      22             :  * Note about locking issues: to create or delete an entry in the shared
      23             :  * hashtable, one must hold pgss->lock exclusively.  Modifying any field
      24             :  * in an entry except the counters requires the same.  To look up an entry,
      25             :  * one must hold the lock shared.  To read or update the counters within
      26             :  * an entry, one must hold the lock shared or exclusive (so the entry doesn't
      27             :  * disappear!) and also take the entry's mutex spinlock.
      28             :  * The shared state variable pgss->extent (the next free spot in the external
      29             :  * query-text file) should be accessed only while holding either the
      30             :  * pgss->mutex spinlock, or exclusive lock on pgss->lock.  We use the mutex to
      31             :  * allow reserving file space while holding only shared lock on pgss->lock.
      32             :  * Rewriting the entire external query-text file, eg for garbage collection,
      33             :  * requires holding pgss->lock exclusively; this allows individual entries
      34             :  * in the file to be read or written while holding only shared lock.
      35             :  *
      36             :  *
      37             :  * Copyright (c) 2008-2025, PostgreSQL Global Development Group
      38             :  *
      39             :  * IDENTIFICATION
      40             :  *    contrib/pg_stat_statements/pg_stat_statements.c
      41             :  *
      42             :  *-------------------------------------------------------------------------
      43             :  */
      44             : #include "postgres.h"
      45             : 
      46             : #include <math.h>
      47             : #include <sys/stat.h>
      48             : #include <unistd.h>
      49             : 
      50             : #include "access/parallel.h"
      51             : #include "catalog/pg_authid.h"
      52             : #include "common/int.h"
      53             : #include "executor/instrument.h"
      54             : #include "funcapi.h"
      55             : #include "jit/jit.h"
      56             : #include "mb/pg_wchar.h"
      57             : #include "miscadmin.h"
      58             : #include "nodes/queryjumble.h"
      59             : #include "optimizer/planner.h"
      60             : #include "parser/analyze.h"
      61             : #include "parser/scanner.h"
      62             : #include "pgstat.h"
      63             : #include "storage/fd.h"
      64             : #include "storage/ipc.h"
      65             : #include "storage/lwlock.h"
      66             : #include "storage/shmem.h"
      67             : #include "storage/spin.h"
      68             : #include "tcop/utility.h"
      69             : #include "utils/acl.h"
      70             : #include "utils/builtins.h"
      71             : #include "utils/memutils.h"
      72             : #include "utils/timestamp.h"
      73             : 
      74          16 : PG_MODULE_MAGIC_EXT(
      75             :                     .name = "pg_stat_statements",
      76             :                     .version = PG_VERSION
      77             : );
      78             : 
      79             : /* Location of permanent stats file (valid when database is shut down) */
      80             : #define PGSS_DUMP_FILE  PGSTAT_STAT_PERMANENT_DIRECTORY "/pg_stat_statements.stat"
      81             : 
      82             : /*
      83             :  * Location of external query text file.
      84             :  */
      85             : #define PGSS_TEXT_FILE  PG_STAT_TMP_DIR "/pgss_query_texts.stat"
      86             : 
      87             : /* Magic number identifying the stats file format */
      88             : static const uint32 PGSS_FILE_HEADER = 0x20250731;
      89             : 
      90             : /* PostgreSQL major version number, changes in which invalidate all entries */
      91             : static const uint32 PGSS_PG_MAJOR_VERSION = PG_VERSION_NUM / 100;
      92             : 
      93             : /* XXX: Should USAGE_EXEC reflect execution time and/or buffer usage? */
      94             : #define USAGE_EXEC(duration)    (1.0)
      95             : #define USAGE_INIT              (1.0)   /* including initial planning */
      96             : #define ASSUMED_MEDIAN_INIT     (10.0)  /* initial assumed median usage */
      97             : #define ASSUMED_LENGTH_INIT     1024    /* initial assumed mean query length */
      98             : #define USAGE_DECREASE_FACTOR   (0.99)  /* decreased every entry_dealloc */
      99             : #define STICKY_DECREASE_FACTOR  (0.50)  /* factor for sticky entries */
     100             : #define USAGE_DEALLOC_PERCENT   5   /* free this % of entries at once */
     101             : #define IS_STICKY(c)    ((c.calls[PGSS_PLAN] + c.calls[PGSS_EXEC]) == 0)
     102             : 
     103             : /*
     104             :  * Extension version number, for supporting older extension versions' objects
     105             :  */
     106             : typedef enum pgssVersion
     107             : {
     108             :     PGSS_V1_0 = 0,
     109             :     PGSS_V1_1,
     110             :     PGSS_V1_2,
     111             :     PGSS_V1_3,
     112             :     PGSS_V1_8,
     113             :     PGSS_V1_9,
     114             :     PGSS_V1_10,
     115             :     PGSS_V1_11,
     116             :     PGSS_V1_12,
     117             :     PGSS_V1_13,
     118             : } pgssVersion;
     119             : 
     120             : typedef enum pgssStoreKind
     121             : {
     122             :     PGSS_INVALID = -1,
     123             : 
     124             :     /*
     125             :      * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
     126             :      * reference the underlying values in the arrays in the Counters struct,
     127             :      * and this order is required in pg_stat_statements_internal().
     128             :      */
     129             :     PGSS_PLAN = 0,
     130             :     PGSS_EXEC,
     131             : } pgssStoreKind;
     132             : 
     133             : #define PGSS_NUMKIND (PGSS_EXEC + 1)
     134             : 
     135             : /*
     136             :  * Hashtable key that defines the identity of a hashtable entry.  We separate
     137             :  * queries by user and by database even if they are otherwise identical.
     138             :  *
     139             :  * If you add a new key to this struct, make sure to teach pgss_store() to
     140             :  * zero the padding bytes.  Otherwise, things will break, because pgss_hash is
     141             :  * created using HASH_BLOBS, and thus tag_hash is used to hash this.
     142             :  */
     143             : typedef struct pgssHashKey
     144             : {
     145             :     Oid         userid;         /* user OID */
     146             :     Oid         dbid;           /* database OID */
     147             :     int64       queryid;        /* query identifier */
     148             :     bool        toplevel;       /* query executed at top level */
     149             : } pgssHashKey;
     150             : 
     151             : /*
     152             :  * The actual stats counters kept within pgssEntry.
     153             :  */
     154             : typedef struct Counters
     155             : {
     156             :     int64       calls[PGSS_NUMKIND];    /* # of times planned/executed */
     157             :     double      total_time[PGSS_NUMKIND];   /* total planning/execution time,
     158             :                                              * in msec */
     159             :     double      min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
     160             :                                          * msec since min/max reset */
     161             :     double      max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
     162             :                                          * msec since min/max reset */
     163             :     double      mean_time[PGSS_NUMKIND];    /* mean planning/execution time in
     164             :                                              * msec */
     165             :     double      sum_var_time[PGSS_NUMKIND]; /* sum of variances in
     166             :                                              * planning/execution time in msec */
     167             :     int64       rows;           /* total # of retrieved or affected rows */
     168             :     int64       shared_blks_hit;    /* # of shared buffer hits */
     169             :     int64       shared_blks_read;   /* # of shared disk blocks read */
     170             :     int64       shared_blks_dirtied;    /* # of shared disk blocks dirtied */
     171             :     int64       shared_blks_written;    /* # of shared disk blocks written */
     172             :     int64       local_blks_hit; /* # of local buffer hits */
     173             :     int64       local_blks_read;    /* # of local disk blocks read */
     174             :     int64       local_blks_dirtied; /* # of local disk blocks dirtied */
     175             :     int64       local_blks_written; /* # of local disk blocks written */
     176             :     int64       temp_blks_read; /* # of temp blocks read */
     177             :     int64       temp_blks_written;  /* # of temp blocks written */
     178             :     double      shared_blk_read_time;   /* time spent reading shared blocks,
     179             :                                          * in msec */
     180             :     double      shared_blk_write_time;  /* time spent writing shared blocks,
     181             :                                          * in msec */
     182             :     double      local_blk_read_time;    /* time spent reading local blocks, in
     183             :                                          * msec */
     184             :     double      local_blk_write_time;   /* time spent writing local blocks, in
     185             :                                          * msec */
     186             :     double      temp_blk_read_time; /* time spent reading temp blocks, in msec */
     187             :     double      temp_blk_write_time;    /* time spent writing temp blocks, in
     188             :                                          * msec */
     189             :     double      usage;          /* usage factor */
     190             :     int64       wal_records;    /* # of WAL records generated */
     191             :     int64       wal_fpi;        /* # of WAL full page images generated */
     192             :     uint64      wal_bytes;      /* total amount of WAL generated in bytes */
     193             :     int64       wal_buffers_full;   /* # of times the WAL buffers became full */
     194             :     int64       jit_functions;  /* total number of JIT functions emitted */
     195             :     double      jit_generation_time;    /* total time to generate jit code */
     196             :     int64       jit_inlining_count; /* number of times inlining time has been
     197             :                                      * > 0 */
     198             :     double      jit_deform_time;    /* total time to deform tuples in jit code */
     199             :     int64       jit_deform_count;   /* number of times deform time has been >
     200             :                                      * 0 */
     201             : 
     202             :     double      jit_inlining_time;  /* total time to inline jit code */
     203             :     int64       jit_optimization_count; /* number of times optimization time
     204             :                                          * has been > 0 */
     205             :     double      jit_optimization_time;  /* total time to optimize jit code */
     206             :     int64       jit_emission_count; /* number of times emission time has been
     207             :                                      * > 0 */
     208             :     double      jit_emission_time;  /* total time to emit jit code */
     209             :     int64       parallel_workers_to_launch; /* # of parallel workers planned
     210             :                                              * to be launched */
     211             :     int64       parallel_workers_launched;  /* # of parallel workers actually
     212             :                                              * launched */
     213             :     int64       generic_plan_calls; /* number of calls using a generic plan */
     214             :     int64       custom_plan_calls;  /* number of calls using a custom plan */
     215             : } Counters;
     216             : 
     217             : /*
     218             :  * Global statistics for pg_stat_statements
     219             :  */
     220             : typedef struct pgssGlobalStats
     221             : {
     222             :     int64       dealloc;        /* # of times entries were deallocated */
     223             :     TimestampTz stats_reset;    /* timestamp with all stats reset */
     224             : } pgssGlobalStats;
     225             : 
     226             : /*
     227             :  * Statistics per statement
     228             :  *
     229             :  * Note: in event of a failure in garbage collection of the query text file,
     230             :  * we reset query_offset to zero and query_len to -1.  This will be seen as
     231             :  * an invalid state by qtext_fetch().
     232             :  */
     233             : typedef struct pgssEntry
     234             : {
     235             :     pgssHashKey key;            /* hash key of entry - MUST BE FIRST */
     236             :     Counters    counters;       /* the statistics for this query */
     237             :     Size        query_offset;   /* query text offset in external file */
     238             :     int         query_len;      /* # of valid bytes in query string, or -1 */
     239             :     int         encoding;       /* query text encoding */
     240             :     TimestampTz stats_since;    /* timestamp of entry allocation */
     241             :     TimestampTz minmax_stats_since; /* timestamp of last min/max values reset */
     242             :     slock_t     mutex;          /* protects the counters only */
     243             : } pgssEntry;
     244             : 
     245             : /*
     246             :  * Global shared state
     247             :  */
     248             : typedef struct pgssSharedState
     249             : {
     250             :     LWLock     *lock;           /* protects hashtable search/modification */
     251             :     double      cur_median_usage;   /* current median usage in hashtable */
     252             :     Size        mean_query_len; /* current mean entry text length */
     253             :     slock_t     mutex;          /* protects following fields only: */
     254             :     Size        extent;         /* current extent of query file */
     255             :     int         n_writers;      /* number of active writers to query file */
     256             :     int         gc_count;       /* query file garbage collection cycle count */
     257             :     pgssGlobalStats stats;      /* global statistics for pgss */
     258             : } pgssSharedState;
     259             : 
     260             : /*---- Local variables ----*/
     261             : 
     262             : /* Current nesting depth of planner/ExecutorRun/ProcessUtility calls */
     263             : static int  nesting_level = 0;
     264             : 
     265             : /* Saved hook values */
     266             : static shmem_request_hook_type prev_shmem_request_hook = NULL;
     267             : static shmem_startup_hook_type prev_shmem_startup_hook = NULL;
     268             : static post_parse_analyze_hook_type prev_post_parse_analyze_hook = NULL;
     269             : static planner_hook_type prev_planner_hook = NULL;
     270             : static ExecutorStart_hook_type prev_ExecutorStart = NULL;
     271             : static ExecutorRun_hook_type prev_ExecutorRun = NULL;
     272             : static ExecutorFinish_hook_type prev_ExecutorFinish = NULL;
     273             : static ExecutorEnd_hook_type prev_ExecutorEnd = NULL;
     274             : static ProcessUtility_hook_type prev_ProcessUtility = NULL;
     275             : 
     276             : /* Links to shared memory state */
     277             : static pgssSharedState *pgss = NULL;
     278             : static HTAB *pgss_hash = NULL;
     279             : 
     280             : /*---- GUC variables ----*/
     281             : 
     282             : typedef enum
     283             : {
     284             :     PGSS_TRACK_NONE,            /* track no statements */
     285             :     PGSS_TRACK_TOP,             /* only top level statements */
     286             :     PGSS_TRACK_ALL,             /* all statements, including nested ones */
     287             : }           PGSSTrackLevel;
     288             : 
     289             : static const struct config_enum_entry track_options[] =
     290             : {
     291             :     {"none", PGSS_TRACK_NONE, false},
     292             :     {"top", PGSS_TRACK_TOP, false},
     293             :     {"all", PGSS_TRACK_ALL, false},
     294             :     {NULL, 0, false}
     295             : };
     296             : 
     297             : static int  pgss_max = 5000;    /* max # statements to track */
     298             : static int  pgss_track = PGSS_TRACK_TOP;    /* tracking level */
     299             : static bool pgss_track_utility = true;  /* whether to track utility commands */
     300             : static bool pgss_track_planning = false;    /* whether to track planning
     301             :                                              * duration */
     302             : static bool pgss_save = true;   /* whether to save stats across shutdown */
     303             : 
     304             : #define pgss_enabled(level) \
     305             :     (!IsParallelWorker() && \
     306             :     (pgss_track == PGSS_TRACK_ALL || \
     307             :     (pgss_track == PGSS_TRACK_TOP && (level) == 0)))
     308             : 
     309             : #define record_gc_qtexts() \
     310             :     do { \
     311             :         SpinLockAcquire(&pgss->mutex); \
     312             :         pgss->gc_count++; \
     313             :         SpinLockRelease(&pgss->mutex); \
     314             :     } while(0)
     315             : 
     316             : /*---- Function declarations ----*/
     317             : 
     318          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset);
     319          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_7);
     320          42 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_11);
     321           0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_2);
     322          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_3);
     323          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_8);
     324          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_9);
     325          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_10);
     326          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_11);
     327          14 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_12);
     328          50 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_13);
     329           0 : PG_FUNCTION_INFO_V1(pg_stat_statements);
     330          16 : PG_FUNCTION_INFO_V1(pg_stat_statements_info);
     331             : 
     332             : static void pgss_shmem_request(void);
     333             : static void pgss_shmem_startup(void);
     334             : static void pgss_shmem_shutdown(int code, Datum arg);
     335             : static void pgss_post_parse_analyze(ParseState *pstate, Query *query,
     336             :                                     JumbleState *jstate);
     337             : static PlannedStmt *pgss_planner(Query *parse,
     338             :                                  const char *query_string,
     339             :                                  int cursorOptions,
     340             :                                  ParamListInfo boundParams);
     341             : static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
     342             : static void pgss_ExecutorRun(QueryDesc *queryDesc,
     343             :                              ScanDirection direction,
     344             :                              uint64 count);
     345             : static void pgss_ExecutorFinish(QueryDesc *queryDesc);
     346             : static void pgss_ExecutorEnd(QueryDesc *queryDesc);
     347             : static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
     348             :                                 bool readOnlyTree,
     349             :                                 ProcessUtilityContext context, ParamListInfo params,
     350             :                                 QueryEnvironment *queryEnv,
     351             :                                 DestReceiver *dest, QueryCompletion *qc);
     352             : static void pgss_store(const char *query, int64 queryId,
     353             :                        int query_location, int query_len,
     354             :                        pgssStoreKind kind,
     355             :                        double total_time, uint64 rows,
     356             :                        const BufferUsage *bufusage,
     357             :                        const WalUsage *walusage,
     358             :                        const struct JitInstrumentation *jitusage,
     359             :                        JumbleState *jstate,
     360             :                        int parallel_workers_to_launch,
     361             :                        int parallel_workers_launched,
     362             :                        PlannedStmtOrigin planOrigin);
     363             : static void pg_stat_statements_internal(FunctionCallInfo fcinfo,
     364             :                                         pgssVersion api_version,
     365             :                                         bool showtext);
     366             : static Size pgss_memsize(void);
     367             : static pgssEntry *entry_alloc(pgssHashKey *key, Size query_offset, int query_len,
     368             :                               int encoding, bool sticky);
     369             : static void entry_dealloc(void);
     370             : static bool qtext_store(const char *query, int query_len,
     371             :                         Size *query_offset, int *gc_count);
     372             : static char *qtext_load_file(Size *buffer_size);
     373             : static char *qtext_fetch(Size query_offset, int query_len,
     374             :                          char *buffer, Size buffer_size);
     375             : static bool need_gc_qtexts(void);
     376             : static void gc_qtexts(void);
     377             : static TimestampTz entry_reset(Oid userid, Oid dbid, int64 queryid, bool minmax_only);
     378             : static char *generate_normalized_query(JumbleState *jstate, const char *query,
     379             :                                        int query_loc, int *query_len_p);
     380             : static void fill_in_constant_lengths(JumbleState *jstate, const char *query,
     381             :                                      int query_loc);
     382             : static int  comp_location(const void *a, const void *b);
     383             : 
     384             : 
     385             : /*
     386             :  * Module load callback
     387             :  */
     388             : void
     389          16 : _PG_init(void)
     390             : {
     391             :     /*
     392             :      * In order to create our shared memory area, we have to be loaded via
     393             :      * shared_preload_libraries.  If not, fall out without hooking into any of
     394             :      * the main system.  (We don't throw error here because it seems useful to
     395             :      * allow the pg_stat_statements functions to be created even when the
     396             :      * module isn't active.  The functions must protect themselves against
     397             :      * being called then, however.)
     398             :      */
     399          16 :     if (!process_shared_preload_libraries_in_progress)
     400           2 :         return;
     401             : 
     402             :     /*
     403             :      * Inform the postmaster that we want to enable query_id calculation if
     404             :      * compute_query_id is set to auto.
     405             :      */
     406          14 :     EnableQueryId();
     407             : 
     408             :     /*
     409             :      * Define (or redefine) custom GUC variables.
     410             :      */
     411          14 :     DefineCustomIntVariable("pg_stat_statements.max",
     412             :                             "Sets the maximum number of statements tracked by pg_stat_statements.",
     413             :                             NULL,
     414             :                             &pgss_max,
     415             :                             5000,
     416             :                             100,
     417             :                             INT_MAX / 2,
     418             :                             PGC_POSTMASTER,
     419             :                             0,
     420             :                             NULL,
     421             :                             NULL,
     422             :                             NULL);
     423             : 
     424          14 :     DefineCustomEnumVariable("pg_stat_statements.track",
     425             :                              "Selects which statements are tracked by pg_stat_statements.",
     426             :                              NULL,
     427             :                              &pgss_track,
     428             :                              PGSS_TRACK_TOP,
     429             :                              track_options,
     430             :                              PGC_SUSET,
     431             :                              0,
     432             :                              NULL,
     433             :                              NULL,
     434             :                              NULL);
     435             : 
     436          14 :     DefineCustomBoolVariable("pg_stat_statements.track_utility",
     437             :                              "Selects whether utility commands are tracked by pg_stat_statements.",
     438             :                              NULL,
     439             :                              &pgss_track_utility,
     440             :                              true,
     441             :                              PGC_SUSET,
     442             :                              0,
     443             :                              NULL,
     444             :                              NULL,
     445             :                              NULL);
     446             : 
     447          14 :     DefineCustomBoolVariable("pg_stat_statements.track_planning",
     448             :                              "Selects whether planning duration is tracked by pg_stat_statements.",
     449             :                              NULL,
     450             :                              &pgss_track_planning,
     451             :                              false,
     452             :                              PGC_SUSET,
     453             :                              0,
     454             :                              NULL,
     455             :                              NULL,
     456             :                              NULL);
     457             : 
     458          14 :     DefineCustomBoolVariable("pg_stat_statements.save",
     459             :                              "Save pg_stat_statements statistics across server shutdowns.",
     460             :                              NULL,
     461             :                              &pgss_save,
     462             :                              true,
     463             :                              PGC_SIGHUP,
     464             :                              0,
     465             :                              NULL,
     466             :                              NULL,
     467             :                              NULL);
     468             : 
     469          14 :     MarkGUCPrefixReserved("pg_stat_statements");
     470             : 
     471             :     /*
     472             :      * Install hooks.
     473             :      */
     474          14 :     prev_shmem_request_hook = shmem_request_hook;
     475          14 :     shmem_request_hook = pgss_shmem_request;
     476          14 :     prev_shmem_startup_hook = shmem_startup_hook;
     477          14 :     shmem_startup_hook = pgss_shmem_startup;
     478          14 :     prev_post_parse_analyze_hook = post_parse_analyze_hook;
     479          14 :     post_parse_analyze_hook = pgss_post_parse_analyze;
     480          14 :     prev_planner_hook = planner_hook;
     481          14 :     planner_hook = pgss_planner;
     482          14 :     prev_ExecutorStart = ExecutorStart_hook;
     483          14 :     ExecutorStart_hook = pgss_ExecutorStart;
     484          14 :     prev_ExecutorRun = ExecutorRun_hook;
     485          14 :     ExecutorRun_hook = pgss_ExecutorRun;
     486          14 :     prev_ExecutorFinish = ExecutorFinish_hook;
     487          14 :     ExecutorFinish_hook = pgss_ExecutorFinish;
     488          14 :     prev_ExecutorEnd = ExecutorEnd_hook;
     489          14 :     ExecutorEnd_hook = pgss_ExecutorEnd;
     490          14 :     prev_ProcessUtility = ProcessUtility_hook;
     491          14 :     ProcessUtility_hook = pgss_ProcessUtility;
     492             : }
     493             : 
     494             : /*
     495             :  * shmem_request hook: request additional shared resources.  We'll allocate or
     496             :  * attach to the shared resources in pgss_shmem_startup().
     497             :  */
     498             : static void
     499          14 : pgss_shmem_request(void)
     500             : {
     501          14 :     if (prev_shmem_request_hook)
     502           0 :         prev_shmem_request_hook();
     503             : 
     504          14 :     RequestAddinShmemSpace(pgss_memsize());
     505          14 :     RequestNamedLWLockTranche("pg_stat_statements", 1);
     506          14 : }
     507             : 
     508             : /*
     509             :  * shmem_startup hook: allocate or attach to shared memory,
     510             :  * then load any pre-existing statistics from file.
     511             :  * Also create and load the query-texts file, which is expected to exist
     512             :  * (even if empty) while the module is enabled.
     513             :  */
     514             : static void
     515          14 : pgss_shmem_startup(void)
     516             : {
     517             :     bool        found;
     518             :     HASHCTL     info;
     519          14 :     FILE       *file = NULL;
     520          14 :     FILE       *qfile = NULL;
     521             :     uint32      header;
     522             :     int32       num;
     523             :     int32       pgver;
     524             :     int32       i;
     525             :     int         buffer_size;
     526          14 :     char       *buffer = NULL;
     527             : 
     528          14 :     if (prev_shmem_startup_hook)
     529           0 :         prev_shmem_startup_hook();
     530             : 
     531             :     /* reset in case this is a restart within the postmaster */
     532          14 :     pgss = NULL;
     533          14 :     pgss_hash = NULL;
     534             : 
     535             :     /*
     536             :      * Create or attach to the shared memory state, including hash table
     537             :      */
     538          14 :     LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
     539             : 
     540          14 :     pgss = ShmemInitStruct("pg_stat_statements",
     541             :                            sizeof(pgssSharedState),
     542             :                            &found);
     543             : 
     544          14 :     if (!found)
     545             :     {
     546             :         /* First time through ... */
     547          14 :         pgss->lock = &(GetNamedLWLockTranche("pg_stat_statements"))->lock;
     548          14 :         pgss->cur_median_usage = ASSUMED_MEDIAN_INIT;
     549          14 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
     550          14 :         SpinLockInit(&pgss->mutex);
     551          14 :         pgss->extent = 0;
     552          14 :         pgss->n_writers = 0;
     553          14 :         pgss->gc_count = 0;
     554          14 :         pgss->stats.dealloc = 0;
     555          14 :         pgss->stats.stats_reset = GetCurrentTimestamp();
     556             :     }
     557             : 
     558          14 :     info.keysize = sizeof(pgssHashKey);
     559          14 :     info.entrysize = sizeof(pgssEntry);
     560          14 :     pgss_hash = ShmemInitHash("pg_stat_statements hash",
     561             :                               pgss_max, pgss_max,
     562             :                               &info,
     563             :                               HASH_ELEM | HASH_BLOBS);
     564             : 
     565          14 :     LWLockRelease(AddinShmemInitLock);
     566             : 
     567             :     /*
     568             :      * If we're in the postmaster (or a standalone backend...), set up a shmem
     569             :      * exit hook to dump the statistics to disk.
     570             :      */
     571          14 :     if (!IsUnderPostmaster)
     572          14 :         on_shmem_exit(pgss_shmem_shutdown, (Datum) 0);
     573             : 
     574             :     /*
     575             :      * Done if some other process already completed our initialization.
     576             :      */
     577          14 :     if (found)
     578          14 :         return;
     579             : 
     580             :     /*
     581             :      * Note: we don't bother with locks here, because there should be no other
     582             :      * processes running when this code is reached.
     583             :      */
     584             : 
     585             :     /* Unlink query text file possibly left over from crash */
     586          14 :     unlink(PGSS_TEXT_FILE);
     587             : 
     588             :     /* Allocate new query text temp file */
     589          14 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
     590          14 :     if (qfile == NULL)
     591           0 :         goto write_error;
     592             : 
     593             :     /*
     594             :      * If we were told not to load old statistics, we're done.  (Note we do
     595             :      * not try to unlink any old dump file in this case.  This seems a bit
     596             :      * questionable but it's the historical behavior.)
     597             :      */
     598          14 :     if (!pgss_save)
     599             :     {
     600           2 :         FreeFile(qfile);
     601           2 :         return;
     602             :     }
     603             : 
     604             :     /*
     605             :      * Attempt to load old statistics from the dump file.
     606             :      */
     607          12 :     file = AllocateFile(PGSS_DUMP_FILE, PG_BINARY_R);
     608          12 :     if (file == NULL)
     609             :     {
     610           8 :         if (errno != ENOENT)
     611           0 :             goto read_error;
     612             :         /* No existing persisted stats file, so we're done */
     613           8 :         FreeFile(qfile);
     614           8 :         return;
     615             :     }
     616             : 
     617           4 :     buffer_size = 2048;
     618           4 :     buffer = (char *) palloc(buffer_size);
     619             : 
     620           8 :     if (fread(&header, sizeof(uint32), 1, file) != 1 ||
     621           8 :         fread(&pgver, sizeof(uint32), 1, file) != 1 ||
     622           4 :         fread(&num, sizeof(int32), 1, file) != 1)
     623           0 :         goto read_error;
     624             : 
     625           4 :     if (header != PGSS_FILE_HEADER ||
     626           4 :         pgver != PGSS_PG_MAJOR_VERSION)
     627           0 :         goto data_error;
     628             : 
     629       52988 :     for (i = 0; i < num; i++)
     630             :     {
     631             :         pgssEntry   temp;
     632             :         pgssEntry  *entry;
     633             :         Size        query_offset;
     634             : 
     635       52984 :         if (fread(&temp, sizeof(pgssEntry), 1, file) != 1)
     636           0 :             goto read_error;
     637             : 
     638             :         /* Encoding is the only field we can easily sanity-check */
     639       52984 :         if (!PG_VALID_BE_ENCODING(temp.encoding))
     640           0 :             goto data_error;
     641             : 
     642             :         /* Resize buffer as needed */
     643       52984 :         if (temp.query_len >= buffer_size)
     644             :         {
     645           6 :             buffer_size = Max(buffer_size * 2, temp.query_len + 1);
     646           6 :             buffer = repalloc(buffer, buffer_size);
     647             :         }
     648             : 
     649       52984 :         if (fread(buffer, 1, temp.query_len + 1, file) != temp.query_len + 1)
     650           0 :             goto read_error;
     651             : 
     652             :         /* Should have a trailing null, but let's make sure */
     653       52984 :         buffer[temp.query_len] = '\0';
     654             : 
     655             :         /* Skip loading "sticky" entries */
     656       52984 :         if (IS_STICKY(temp.counters))
     657        1474 :             continue;
     658             : 
     659             :         /* Store the query text */
     660       51510 :         query_offset = pgss->extent;
     661       51510 :         if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
     662           0 :             goto write_error;
     663       51510 :         pgss->extent += temp.query_len + 1;
     664             : 
     665             :         /* make the hashtable entry (discards old entries if too many) */
     666       51510 :         entry = entry_alloc(&temp.key, query_offset, temp.query_len,
     667             :                             temp.encoding,
     668             :                             false);
     669             : 
     670             :         /* copy in the actual stats */
     671       51510 :         entry->counters = temp.counters;
     672       51510 :         entry->stats_since = temp.stats_since;
     673       51510 :         entry->minmax_stats_since = temp.minmax_stats_since;
     674             :     }
     675             : 
     676             :     /* Read global statistics for pg_stat_statements */
     677           4 :     if (fread(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
     678           0 :         goto read_error;
     679             : 
     680           4 :     pfree(buffer);
     681           4 :     FreeFile(file);
     682           4 :     FreeFile(qfile);
     683             : 
     684             :     /*
     685             :      * Remove the persisted stats file so it's not included in
     686             :      * backups/replication standbys, etc.  A new file will be written on next
     687             :      * shutdown.
     688             :      *
     689             :      * Note: it's okay if the PGSS_TEXT_FILE is included in a basebackup,
     690             :      * because we remove that file on startup; it acts inversely to
     691             :      * PGSS_DUMP_FILE, in that it is only supposed to be around when the
     692             :      * server is running, whereas PGSS_DUMP_FILE is only supposed to be around
     693             :      * when the server is not running.  Leaving the file creates no danger of
     694             :      * a newly restored database having a spurious record of execution costs,
     695             :      * which is what we're really concerned about here.
     696             :      */
     697           4 :     unlink(PGSS_DUMP_FILE);
     698             : 
     699           4 :     return;
     700             : 
     701           0 : read_error:
     702           0 :     ereport(LOG,
     703             :             (errcode_for_file_access(),
     704             :              errmsg("could not read file \"%s\": %m",
     705             :                     PGSS_DUMP_FILE)));
     706           0 :     goto fail;
     707           0 : data_error:
     708           0 :     ereport(LOG,
     709             :             (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
     710             :              errmsg("ignoring invalid data in file \"%s\"",
     711             :                     PGSS_DUMP_FILE)));
     712           0 :     goto fail;
     713           0 : write_error:
     714           0 :     ereport(LOG,
     715             :             (errcode_for_file_access(),
     716             :              errmsg("could not write file \"%s\": %m",
     717             :                     PGSS_TEXT_FILE)));
     718           0 : fail:
     719           0 :     if (buffer)
     720           0 :         pfree(buffer);
     721           0 :     if (file)
     722           0 :         FreeFile(file);
     723           0 :     if (qfile)
     724           0 :         FreeFile(qfile);
     725             :     /* If possible, throw away the bogus file; ignore any error */
     726           0 :     unlink(PGSS_DUMP_FILE);
     727             : 
     728             :     /*
     729             :      * Don't unlink PGSS_TEXT_FILE here; it should always be around while the
     730             :      * server is running with pg_stat_statements enabled
     731             :      */
     732             : }
     733             : 
     734             : /*
     735             :  * shmem_shutdown hook: Dump statistics into file.
     736             :  *
     737             :  * Note: we don't bother with acquiring lock, because there should be no
     738             :  * other processes running when this is called.
     739             :  */
     740             : static void
     741          14 : pgss_shmem_shutdown(int code, Datum arg)
     742             : {
     743             :     FILE       *file;
     744          14 :     char       *qbuffer = NULL;
     745          14 :     Size        qbuffer_size = 0;
     746             :     HASH_SEQ_STATUS hash_seq;
     747             :     int32       num_entries;
     748             :     pgssEntry  *entry;
     749             : 
     750             :     /* Don't try to dump during a crash. */
     751          14 :     if (code)
     752          14 :         return;
     753             : 
     754             :     /* Safety check ... shouldn't get here unless shmem is set up. */
     755          14 :     if (!pgss || !pgss_hash)
     756           0 :         return;
     757             : 
     758             :     /* Don't dump if told not to. */
     759          14 :     if (!pgss_save)
     760           4 :         return;
     761             : 
     762          10 :     file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
     763          10 :     if (file == NULL)
     764           0 :         goto error;
     765             : 
     766          10 :     if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
     767           0 :         goto error;
     768          10 :     if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
     769           0 :         goto error;
     770          10 :     num_entries = hash_get_num_entries(pgss_hash);
     771          10 :     if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
     772           0 :         goto error;
     773             : 
     774          10 :     qbuffer = qtext_load_file(&qbuffer_size);
     775          10 :     if (qbuffer == NULL)
     776           0 :         goto error;
     777             : 
     778             :     /*
     779             :      * When serializing to disk, we store query texts immediately after their
     780             :      * entry data.  Any orphaned query texts are thereby excluded.
     781             :      */
     782          10 :     hash_seq_init(&hash_seq, pgss_hash);
     783      106518 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
     784             :     {
     785      106508 :         int         len = entry->query_len;
     786      106508 :         char       *qstr = qtext_fetch(entry->query_offset, len,
     787             :                                        qbuffer, qbuffer_size);
     788             : 
     789      106508 :         if (qstr == NULL)
     790           0 :             continue;           /* Ignore any entries with bogus texts */
     791             : 
     792      106508 :         if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
     793      106508 :             fwrite(qstr, 1, len + 1, file) != len + 1)
     794             :         {
     795             :             /* note: we assume hash_seq_term won't change errno */
     796           0 :             hash_seq_term(&hash_seq);
     797           0 :             goto error;
     798             :         }
     799             :     }
     800             : 
     801             :     /* Dump global statistics for pg_stat_statements */
     802          10 :     if (fwrite(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
     803           0 :         goto error;
     804             : 
     805          10 :     free(qbuffer);
     806          10 :     qbuffer = NULL;
     807             : 
     808          10 :     if (FreeFile(file))
     809             :     {
     810           0 :         file = NULL;
     811           0 :         goto error;
     812             :     }
     813             : 
     814             :     /*
     815             :      * Rename file into place, so we atomically replace any old one.
     816             :      */
     817          10 :     (void) durable_rename(PGSS_DUMP_FILE ".tmp", PGSS_DUMP_FILE, LOG);
     818             : 
     819             :     /* Unlink query-texts file; it's not needed while shutdown */
     820          10 :     unlink(PGSS_TEXT_FILE);
     821             : 
     822          10 :     return;
     823             : 
     824           0 : error:
     825           0 :     ereport(LOG,
     826             :             (errcode_for_file_access(),
     827             :              errmsg("could not write file \"%s\": %m",
     828             :                     PGSS_DUMP_FILE ".tmp")));
     829           0 :     free(qbuffer);
     830           0 :     if (file)
     831           0 :         FreeFile(file);
     832           0 :     unlink(PGSS_DUMP_FILE ".tmp");
     833           0 :     unlink(PGSS_TEXT_FILE);
     834             : }
     835             : 
     836             : /*
     837             :  * Post-parse-analysis hook: mark query with a queryId
     838             :  */
     839             : static void
     840      155668 : pgss_post_parse_analyze(ParseState *pstate, Query *query, JumbleState *jstate)
     841             : {
     842      155668 :     if (prev_post_parse_analyze_hook)
     843           0 :         prev_post_parse_analyze_hook(pstate, query, jstate);
     844             : 
     845             :     /* Safety check... */
     846      155668 :     if (!pgss || !pgss_hash || !pgss_enabled(nesting_level))
     847       24506 :         return;
     848             : 
     849             :     /*
     850             :      * If it's EXECUTE, clear the queryId so that stats will accumulate for
     851             :      * the underlying PREPARE.  But don't do this if we're not tracking
     852             :      * utility statements, to avoid messing up another extension that might be
     853             :      * tracking them.
     854             :      */
     855      131162 :     if (query->utilityStmt)
     856             :     {
     857       58364 :         if (pgss_track_utility && IsA(query->utilityStmt, ExecuteStmt))
     858             :         {
     859        6484 :             query->queryId = INT64CONST(0);
     860        6484 :             return;
     861             :         }
     862             :     }
     863             : 
     864             :     /*
     865             :      * If query jumbling were able to identify any ignorable constants, we
     866             :      * immediately create a hash table entry for the query, so that we can
     867             :      * record the normalized form of the query string.  If there were no such
     868             :      * constants, the normalized string would be the same as the query text
     869             :      * anyway, so there's no need for an early entry.
     870             :      */
     871      124678 :     if (jstate && jstate->clocations_count > 0)
     872       72344 :         pgss_store(pstate->p_sourcetext,
     873             :                    query->queryId,
     874             :                    query->stmt_location,
     875             :                    query->stmt_len,
     876             :                    PGSS_INVALID,
     877             :                    0,
     878             :                    0,
     879             :                    NULL,
     880             :                    NULL,
     881             :                    NULL,
     882             :                    jstate,
     883             :                    0,
     884             :                    0,
     885             :                    PLAN_STMT_UNKNOWN);
     886             : }
     887             : 
     888             : /*
     889             :  * Planner hook: forward to regular planner, but measure planning time
     890             :  * if needed.
     891             :  */
     892             : static PlannedStmt *
     893       94344 : pgss_planner(Query *parse,
     894             :              const char *query_string,
     895             :              int cursorOptions,
     896             :              ParamListInfo boundParams)
     897             : {
     898             :     PlannedStmt *result;
     899             : 
     900             :     /*
     901             :      * We can't process the query if no query_string is provided, as
     902             :      * pgss_store needs it.  We also ignore query without queryid, as it would
     903             :      * be treated as a utility statement, which may not be the case.
     904             :      */
     905       94344 :     if (pgss_enabled(nesting_level)
     906       73130 :         && pgss_track_planning && query_string
     907         262 :         && parse->queryId != INT64CONST(0))
     908         262 :     {
     909             :         instr_time  start;
     910             :         instr_time  duration;
     911             :         BufferUsage bufusage_start,
     912             :                     bufusage;
     913             :         WalUsage    walusage_start,
     914             :                     walusage;
     915             : 
     916             :         /* We need to track buffer usage as the planner can access them. */
     917         262 :         bufusage_start = pgBufferUsage;
     918             : 
     919             :         /*
     920             :          * Similarly the planner could write some WAL records in some cases
     921             :          * (e.g. setting a hint bit with those being WAL-logged)
     922             :          */
     923         262 :         walusage_start = pgWalUsage;
     924         262 :         INSTR_TIME_SET_CURRENT(start);
     925             : 
     926         262 :         nesting_level++;
     927         262 :         PG_TRY();
     928             :         {
     929         262 :             if (prev_planner_hook)
     930           0 :                 result = prev_planner_hook(parse, query_string, cursorOptions,
     931             :                                            boundParams);
     932             :             else
     933         262 :                 result = standard_planner(parse, query_string, cursorOptions,
     934             :                                           boundParams);
     935             :         }
     936           0 :         PG_FINALLY();
     937             :         {
     938         262 :             nesting_level--;
     939             :         }
     940         262 :         PG_END_TRY();
     941             : 
     942         262 :         INSTR_TIME_SET_CURRENT(duration);
     943         262 :         INSTR_TIME_SUBTRACT(duration, start);
     944             : 
     945             :         /* calc differences of buffer counters. */
     946         262 :         memset(&bufusage, 0, sizeof(BufferUsage));
     947         262 :         BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
     948             : 
     949             :         /* calc differences of WAL counters. */
     950         262 :         memset(&walusage, 0, sizeof(WalUsage));
     951         262 :         WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
     952             : 
     953         262 :         pgss_store(query_string,
     954             :                    parse->queryId,
     955             :                    parse->stmt_location,
     956             :                    parse->stmt_len,
     957             :                    PGSS_PLAN,
     958         262 :                    INSTR_TIME_GET_MILLISEC(duration),
     959             :                    0,
     960             :                    &bufusage,
     961             :                    &walusage,
     962             :                    NULL,
     963             :                    NULL,
     964             :                    0,
     965             :                    0,
     966             :                    result->planOrigin);
     967             :     }
     968             :     else
     969             :     {
     970             :         /*
     971             :          * Even though we're not tracking plan time for this statement, we
     972             :          * must still increment the nesting level, to ensure that functions
     973             :          * evaluated during planning are not seen as top-level calls.
     974             :          */
     975       94082 :         nesting_level++;
     976       94082 :         PG_TRY();
     977             :         {
     978       94082 :             if (prev_planner_hook)
     979           0 :                 result = prev_planner_hook(parse, query_string, cursorOptions,
     980             :                                            boundParams);
     981             :             else
     982       94082 :                 result = standard_planner(parse, query_string, cursorOptions,
     983             :                                           boundParams);
     984             :         }
     985        1496 :         PG_FINALLY();
     986             :         {
     987       94082 :             nesting_level--;
     988             :         }
     989       94082 :         PG_END_TRY();
     990             :     }
     991             : 
     992       92848 :     return result;
     993             : }
     994             : 
     995             : /*
     996             :  * ExecutorStart hook: start up tracking if needed
     997             :  */
     998             : static void
     999      115696 : pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
    1000             : {
    1001      115696 :     if (prev_ExecutorStart)
    1002           0 :         prev_ExecutorStart(queryDesc, eflags);
    1003             :     else
    1004      115696 :         standard_ExecutorStart(queryDesc, eflags);
    1005             : 
    1006             :     /*
    1007             :      * If query has queryId zero, don't track it.  This prevents double
    1008             :      * counting of optimizable statements that are directly contained in
    1009             :      * utility statements.
    1010             :      */
    1011      115162 :     if (pgss_enabled(nesting_level) && queryDesc->plannedstmt->queryId != INT64CONST(0))
    1012             :     {
    1013             :         /*
    1014             :          * Set up to track total elapsed time in ExecutorRun.  Make sure the
    1015             :          * space is allocated in the per-query context so it will go away at
    1016             :          * ExecutorEnd.
    1017             :          */
    1018       76968 :         if (queryDesc->totaltime == NULL)
    1019             :         {
    1020             :             MemoryContext oldcxt;
    1021             : 
    1022       76968 :             oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
    1023       76968 :             queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL, false);
    1024       76968 :             MemoryContextSwitchTo(oldcxt);
    1025             :         }
    1026             :     }
    1027      115162 : }
    1028             : 
    1029             : /*
    1030             :  * ExecutorRun hook: all we need do is track nesting depth
    1031             :  */
    1032             : static void
    1033      112608 : pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count)
    1034             : {
    1035      112608 :     nesting_level++;
    1036      112608 :     PG_TRY();
    1037             :     {
    1038      112608 :         if (prev_ExecutorRun)
    1039           0 :             prev_ExecutorRun(queryDesc, direction, count);
    1040             :         else
    1041      112608 :             standard_ExecutorRun(queryDesc, direction, count);
    1042             :     }
    1043        6860 :     PG_FINALLY();
    1044             :     {
    1045      112608 :         nesting_level--;
    1046             :     }
    1047      112608 :     PG_END_TRY();
    1048      105748 : }
    1049             : 
    1050             : /*
    1051             :  * ExecutorFinish hook: all we need do is track nesting depth
    1052             :  */
    1053             : static void
    1054      101830 : pgss_ExecutorFinish(QueryDesc *queryDesc)
    1055             : {
    1056      101830 :     nesting_level++;
    1057      101830 :     PG_TRY();
    1058             :     {
    1059      101830 :         if (prev_ExecutorFinish)
    1060           0 :             prev_ExecutorFinish(queryDesc);
    1061             :         else
    1062      101830 :             standard_ExecutorFinish(queryDesc);
    1063             :     }
    1064         328 :     PG_FINALLY();
    1065             :     {
    1066      101830 :         nesting_level--;
    1067             :     }
    1068      101830 :     PG_END_TRY();
    1069      101502 : }
    1070             : 
    1071             : /*
    1072             :  * ExecutorEnd hook: store results if needed
    1073             :  */
    1074             : static void
    1075      107320 : pgss_ExecutorEnd(QueryDesc *queryDesc)
    1076             : {
    1077      107320 :     int64       queryId = queryDesc->plannedstmt->queryId;
    1078             : 
    1079      107320 :     if (queryId != INT64CONST(0) && queryDesc->totaltime &&
    1080       73908 :         pgss_enabled(nesting_level))
    1081             :     {
    1082             :         /*
    1083             :          * Make sure stats accumulation is done.  (Note: it's okay if several
    1084             :          * levels of hook all do this.)
    1085             :          */
    1086       73908 :         InstrEndLoop(queryDesc->totaltime);
    1087             : 
    1088       73758 :         pgss_store(queryDesc->sourceText,
    1089             :                    queryId,
    1090       73908 :                    queryDesc->plannedstmt->stmt_location,
    1091       73908 :                    queryDesc->plannedstmt->stmt_len,
    1092             :                    PGSS_EXEC,
    1093       73908 :                    queryDesc->totaltime->total * 1000.0,  /* convert to msec */
    1094       73908 :                    queryDesc->estate->es_total_processed,
    1095       73908 :                    &queryDesc->totaltime->bufusage,
    1096       73908 :                    &queryDesc->totaltime->walusage,
    1097         150 :                    queryDesc->estate->es_jit ? &queryDesc->estate->es_jit->instr : NULL,
    1098             :                    NULL,
    1099       73908 :                    queryDesc->estate->es_parallel_workers_to_launch,
    1100       73908 :                    queryDesc->estate->es_parallel_workers_launched,
    1101       73908 :                    queryDesc->plannedstmt->planOrigin);
    1102             :     }
    1103             : 
    1104      107320 :     if (prev_ExecutorEnd)
    1105           0 :         prev_ExecutorEnd(queryDesc);
    1106             :     else
    1107      107320 :         standard_ExecutorEnd(queryDesc);
    1108      107320 : }
    1109             : 
    1110             : /*
    1111             :  * ProcessUtility hook
    1112             :  */
    1113             : static void
    1114       69138 : pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
    1115             :                     bool readOnlyTree,
    1116             :                     ProcessUtilityContext context,
    1117             :                     ParamListInfo params, QueryEnvironment *queryEnv,
    1118             :                     DestReceiver *dest, QueryCompletion *qc)
    1119             : {
    1120       69138 :     Node       *parsetree = pstmt->utilityStmt;
    1121       69138 :     int64       saved_queryId = pstmt->queryId;
    1122       69138 :     int         saved_stmt_location = pstmt->stmt_location;
    1123       69138 :     int         saved_stmt_len = pstmt->stmt_len;
    1124       69138 :     bool        enabled = pgss_track_utility && pgss_enabled(nesting_level);
    1125             : 
    1126             :     /*
    1127             :      * Force utility statements to get queryId zero.  We do this even in cases
    1128             :      * where the statement contains an optimizable statement for which a
    1129             :      * queryId could be derived (such as EXPLAIN or DECLARE CURSOR).  For such
    1130             :      * cases, runtime control will first go through ProcessUtility and then
    1131             :      * the executor, and we don't want the executor hooks to do anything,
    1132             :      * since we are already measuring the statement's costs at the utility
    1133             :      * level.
    1134             :      *
    1135             :      * Note that this is only done if pg_stat_statements is enabled and
    1136             :      * configured to track utility statements, in the unlikely possibility
    1137             :      * that user configured another extension to handle utility statements
    1138             :      * only.
    1139             :      */
    1140       69138 :     if (enabled)
    1141       58156 :         pstmt->queryId = INT64CONST(0);
    1142             : 
    1143             :     /*
    1144             :      * If it's an EXECUTE statement, we don't track it and don't increment the
    1145             :      * nesting level.  This allows the cycles to be charged to the underlying
    1146             :      * PREPARE instead (by the Executor hooks), which is much more useful.
    1147             :      *
    1148             :      * We also don't track execution of PREPARE.  If we did, we would get one
    1149             :      * hash table entry for the PREPARE (with hash calculated from the query
    1150             :      * string), and then a different one with the same query string (but hash
    1151             :      * calculated from the query tree) would be used to accumulate costs of
    1152             :      * ensuing EXECUTEs.  This would be confusing.  Since PREPARE doesn't
    1153             :      * actually run the planner (only parse+rewrite), its costs are generally
    1154             :      * pretty negligible and it seems okay to just ignore it.
    1155             :      */
    1156       69138 :     if (enabled &&
    1157       58156 :         !IsA(parsetree, ExecuteStmt) &&
    1158       51684 :         !IsA(parsetree, PrepareStmt))
    1159       46608 :     {
    1160             :         instr_time  start;
    1161             :         instr_time  duration;
    1162             :         uint64      rows;
    1163             :         BufferUsage bufusage_start,
    1164             :                     bufusage;
    1165             :         WalUsage    walusage_start,
    1166             :                     walusage;
    1167             : 
    1168       51438 :         bufusage_start = pgBufferUsage;
    1169       51438 :         walusage_start = pgWalUsage;
    1170       51438 :         INSTR_TIME_SET_CURRENT(start);
    1171             : 
    1172       51438 :         nesting_level++;
    1173       51438 :         PG_TRY();
    1174             :         {
    1175       51438 :             if (prev_ProcessUtility)
    1176           0 :                 prev_ProcessUtility(pstmt, queryString, readOnlyTree,
    1177             :                                     context, params, queryEnv,
    1178             :                                     dest, qc);
    1179             :             else
    1180       51438 :                 standard_ProcessUtility(pstmt, queryString, readOnlyTree,
    1181             :                                         context, params, queryEnv,
    1182             :                                         dest, qc);
    1183             :         }
    1184        4830 :         PG_FINALLY();
    1185             :         {
    1186       51438 :             nesting_level--;
    1187             :         }
    1188       51438 :         PG_END_TRY();
    1189             : 
    1190             :         /*
    1191             :          * CAUTION: do not access the *pstmt data structure again below here.
    1192             :          * If it was a ROLLBACK or similar, that data structure may have been
    1193             :          * freed.  We must copy everything we still need into local variables,
    1194             :          * which we did above.
    1195             :          *
    1196             :          * For the same reason, we can't risk restoring pstmt->queryId to its
    1197             :          * former value, which'd otherwise be a good idea.
    1198             :          */
    1199             : 
    1200       46608 :         INSTR_TIME_SET_CURRENT(duration);
    1201       46608 :         INSTR_TIME_SUBTRACT(duration, start);
    1202             : 
    1203             :         /*
    1204             :          * Track the total number of rows retrieved or affected by the utility
    1205             :          * statements of COPY, FETCH, CREATE TABLE AS, CREATE MATERIALIZED
    1206             :          * VIEW, REFRESH MATERIALIZED VIEW and SELECT INTO.
    1207             :          */
    1208       46602 :         rows = (qc && (qc->commandTag == CMDTAG_COPY ||
    1209       43302 :                        qc->commandTag == CMDTAG_FETCH ||
    1210       42782 :                        qc->commandTag == CMDTAG_SELECT ||
    1211       42408 :                        qc->commandTag == CMDTAG_REFRESH_MATERIALIZED_VIEW)) ?
    1212       93210 :             qc->nprocessed : 0;
    1213             : 
    1214             :         /* calc differences of buffer counters. */
    1215       46608 :         memset(&bufusage, 0, sizeof(BufferUsage));
    1216       46608 :         BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
    1217             : 
    1218             :         /* calc differences of WAL counters. */
    1219       46608 :         memset(&walusage, 0, sizeof(WalUsage));
    1220       46608 :         WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
    1221             : 
    1222       46608 :         pgss_store(queryString,
    1223             :                    saved_queryId,
    1224             :                    saved_stmt_location,
    1225             :                    saved_stmt_len,
    1226             :                    PGSS_EXEC,
    1227       46608 :                    INSTR_TIME_GET_MILLISEC(duration),
    1228             :                    rows,
    1229             :                    &bufusage,
    1230             :                    &walusage,
    1231             :                    NULL,
    1232             :                    NULL,
    1233             :                    0,
    1234             :                    0,
    1235             :                    pstmt->planOrigin);
    1236             :     }
    1237             :     else
    1238             :     {
    1239             :         /*
    1240             :          * Even though we're not tracking execution time for this statement,
    1241             :          * we must still increment the nesting level, to ensure that functions
    1242             :          * evaluated within it are not seen as top-level calls.  But don't do
    1243             :          * so for EXECUTE; that way, when control reaches pgss_planner or
    1244             :          * pgss_ExecutorStart, we will treat the costs as top-level if
    1245             :          * appropriate.  Likewise, don't bump for PREPARE, so that parse
    1246             :          * analysis will treat the statement as top-level if appropriate.
    1247             :          *
    1248             :          * To be absolutely certain we don't mess up the nesting level,
    1249             :          * evaluate the bump_level condition just once.
    1250             :          */
    1251       17700 :         bool        bump_level =
    1252       28926 :             !IsA(parsetree, ExecuteStmt) &&
    1253       11226 :             !IsA(parsetree, PrepareStmt);
    1254             : 
    1255       17700 :         if (bump_level)
    1256       10978 :             nesting_level++;
    1257       17700 :         PG_TRY();
    1258             :         {
    1259       17700 :             if (prev_ProcessUtility)
    1260           0 :                 prev_ProcessUtility(pstmt, queryString, readOnlyTree,
    1261             :                                     context, params, queryEnv,
    1262             :                                     dest, qc);
    1263             :             else
    1264       17700 :                 standard_ProcessUtility(pstmt, queryString, readOnlyTree,
    1265             :                                         context, params, queryEnv,
    1266             :                                         dest, qc);
    1267             :         }
    1268         260 :         PG_FINALLY();
    1269             :         {
    1270       17700 :             if (bump_level)
    1271       10978 :                 nesting_level--;
    1272             :         }
    1273       17700 :         PG_END_TRY();
    1274             :     }
    1275       64048 : }
    1276             : 
    1277             : /*
    1278             :  * Store some statistics for a statement.
    1279             :  *
    1280             :  * If jstate is not NULL then we're trying to create an entry for which
    1281             :  * we have no statistics as yet; we just want to record the normalized
    1282             :  * query string.  total_time, rows, bufusage and walusage are ignored in this
    1283             :  * case.
    1284             :  *
    1285             :  * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
    1286             :  * for the arrays in the Counters field.
    1287             :  */
    1288             : static void
    1289      193122 : pgss_store(const char *query, int64 queryId,
    1290             :            int query_location, int query_len,
    1291             :            pgssStoreKind kind,
    1292             :            double total_time, uint64 rows,
    1293             :            const BufferUsage *bufusage,
    1294             :            const WalUsage *walusage,
    1295             :            const struct JitInstrumentation *jitusage,
    1296             :            JumbleState *jstate,
    1297             :            int parallel_workers_to_launch,
    1298             :            int parallel_workers_launched,
    1299             :            PlannedStmtOrigin planOrigin)
    1300             : {
    1301             :     pgssHashKey key;
    1302             :     pgssEntry  *entry;
    1303      193122 :     char       *norm_query = NULL;
    1304      193122 :     int         encoding = GetDatabaseEncoding();
    1305             : 
    1306             :     Assert(query != NULL);
    1307             : 
    1308             :     /* Safety check... */
    1309      193122 :     if (!pgss || !pgss_hash)
    1310           0 :         return;
    1311             : 
    1312             :     /*
    1313             :      * Nothing to do if compute_query_id isn't enabled and no other module
    1314             :      * computed a query identifier.
    1315             :      */
    1316      193122 :     if (queryId == INT64CONST(0))
    1317           0 :         return;
    1318             : 
    1319             :     /*
    1320             :      * Confine our attention to the relevant part of the string, if the query
    1321             :      * is a portion of a multi-statement source string, and update query
    1322             :      * location and length if needed.
    1323             :      */
    1324      193122 :     query = CleanQuerytext(query, &query_location, &query_len);
    1325             : 
    1326             :     /* Set up key for hashtable search */
    1327             : 
    1328             :     /* clear padding */
    1329      193122 :     memset(&key, 0, sizeof(pgssHashKey));
    1330             : 
    1331      193122 :     key.userid = GetUserId();
    1332      193122 :     key.dbid = MyDatabaseId;
    1333      193122 :     key.queryid = queryId;
    1334      193122 :     key.toplevel = (nesting_level == 0);
    1335             : 
    1336             :     /* Lookup the hash table entry with shared lock. */
    1337      193122 :     LWLockAcquire(pgss->lock, LW_SHARED);
    1338             : 
    1339      193122 :     entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    1340             : 
    1341             :     /* Create new entry, if not present */
    1342      193122 :     if (!entry)
    1343             :     {
    1344             :         Size        query_offset;
    1345             :         int         gc_count;
    1346             :         bool        stored;
    1347             :         bool        do_gc;
    1348             : 
    1349             :         /*
    1350             :          * Create a new, normalized query string if caller asked.  We don't
    1351             :          * need to hold the lock while doing this work.  (Note: in any case,
    1352             :          * it's possible that someone else creates a duplicate hashtable entry
    1353             :          * in the interval where we don't hold the lock below.  That case is
    1354             :          * handled by entry_alloc.)
    1355             :          */
    1356       56756 :         if (jstate)
    1357             :         {
    1358       21218 :             LWLockRelease(pgss->lock);
    1359       21218 :             norm_query = generate_normalized_query(jstate, query,
    1360             :                                                    query_location,
    1361             :                                                    &query_len);
    1362       21218 :             LWLockAcquire(pgss->lock, LW_SHARED);
    1363             :         }
    1364             : 
    1365             :         /* Append new query text to file with only shared lock held */
    1366       56756 :         stored = qtext_store(norm_query ? norm_query : query, query_len,
    1367             :                              &query_offset, &gc_count);
    1368             : 
    1369             :         /*
    1370             :          * Determine whether we need to garbage collect external query texts
    1371             :          * while the shared lock is still held.  This micro-optimization
    1372             :          * avoids taking the time to decide this while holding exclusive lock.
    1373             :          */
    1374       56756 :         do_gc = need_gc_qtexts();
    1375             : 
    1376             :         /* Need exclusive lock to make a new hashtable entry - promote */
    1377       56756 :         LWLockRelease(pgss->lock);
    1378       56756 :         LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
    1379             : 
    1380             :         /*
    1381             :          * A garbage collection may have occurred while we weren't holding the
    1382             :          * lock.  In the unlikely event that this happens, the query text we
    1383             :          * stored above will have been garbage collected, so write it again.
    1384             :          * This should be infrequent enough that doing it while holding
    1385             :          * exclusive lock isn't a performance problem.
    1386             :          */
    1387       56756 :         if (!stored || pgss->gc_count != gc_count)
    1388           0 :             stored = qtext_store(norm_query ? norm_query : query, query_len,
    1389             :                                  &query_offset, NULL);
    1390             : 
    1391             :         /* If we failed to write to the text file, give up */
    1392       56756 :         if (!stored)
    1393           0 :             goto done;
    1394             : 
    1395             :         /* OK to create a new hashtable entry */
    1396       56756 :         entry = entry_alloc(&key, query_offset, query_len, encoding,
    1397             :                             jstate != NULL);
    1398             : 
    1399             :         /* If needed, perform garbage collection while exclusive lock held */
    1400       56756 :         if (do_gc)
    1401           0 :             gc_qtexts();
    1402             :     }
    1403             : 
    1404             :     /* Increment the counts, except when jstate is not NULL */
    1405      193122 :     if (!jstate)
    1406             :     {
    1407             :         Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
    1408             : 
    1409             :         /*
    1410             :          * Grab the spinlock while updating the counters (see comment about
    1411             :          * locking rules at the head of the file)
    1412             :          */
    1413      120778 :         SpinLockAcquire(&entry->mutex);
    1414             : 
    1415             :         /* "Unstick" entry if it was previously sticky */
    1416      120778 :         if (IS_STICKY(entry->counters))
    1417       55226 :             entry->counters.usage = USAGE_INIT;
    1418             : 
    1419      120778 :         entry->counters.calls[kind] += 1;
    1420      120778 :         entry->counters.total_time[kind] += total_time;
    1421             : 
    1422      120778 :         if (entry->counters.calls[kind] == 1)
    1423             :         {
    1424       55382 :             entry->counters.min_time[kind] = total_time;
    1425       55382 :             entry->counters.max_time[kind] = total_time;
    1426       55382 :             entry->counters.mean_time[kind] = total_time;
    1427             :         }
    1428             :         else
    1429             :         {
    1430             :             /*
    1431             :              * Welford's method for accurately computing variance. See
    1432             :              * <http://www.johndcook.com/blog/standard_deviation/>
    1433             :              */
    1434       65396 :             double      old_mean = entry->counters.mean_time[kind];
    1435             : 
    1436       65396 :             entry->counters.mean_time[kind] +=
    1437       65396 :                 (total_time - old_mean) / entry->counters.calls[kind];
    1438       65396 :             entry->counters.sum_var_time[kind] +=
    1439       65396 :                 (total_time - old_mean) * (total_time - entry->counters.mean_time[kind]);
    1440             : 
    1441             :             /*
    1442             :              * Calculate min and max time. min = 0 and max = 0 means that the
    1443             :              * min/max statistics were reset
    1444             :              */
    1445       65396 :             if (entry->counters.min_time[kind] == 0
    1446          12 :                 && entry->counters.max_time[kind] == 0)
    1447             :             {
    1448           6 :                 entry->counters.min_time[kind] = total_time;
    1449           6 :                 entry->counters.max_time[kind] = total_time;
    1450             :             }
    1451             :             else
    1452             :             {
    1453       65390 :                 if (entry->counters.min_time[kind] > total_time)
    1454       13314 :                     entry->counters.min_time[kind] = total_time;
    1455       65390 :                 if (entry->counters.max_time[kind] < total_time)
    1456        6190 :                     entry->counters.max_time[kind] = total_time;
    1457             :             }
    1458             :         }
    1459      120778 :         entry->counters.rows += rows;
    1460      120778 :         entry->counters.shared_blks_hit += bufusage->shared_blks_hit;
    1461      120778 :         entry->counters.shared_blks_read += bufusage->shared_blks_read;
    1462      120778 :         entry->counters.shared_blks_dirtied += bufusage->shared_blks_dirtied;
    1463      120778 :         entry->counters.shared_blks_written += bufusage->shared_blks_written;
    1464      120778 :         entry->counters.local_blks_hit += bufusage->local_blks_hit;
    1465      120778 :         entry->counters.local_blks_read += bufusage->local_blks_read;
    1466      120778 :         entry->counters.local_blks_dirtied += bufusage->local_blks_dirtied;
    1467      120778 :         entry->counters.local_blks_written += bufusage->local_blks_written;
    1468      120778 :         entry->counters.temp_blks_read += bufusage->temp_blks_read;
    1469      120778 :         entry->counters.temp_blks_written += bufusage->temp_blks_written;
    1470      120778 :         entry->counters.shared_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_read_time);
    1471      120778 :         entry->counters.shared_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_write_time);
    1472      120778 :         entry->counters.local_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_read_time);
    1473      120778 :         entry->counters.local_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_write_time);
    1474      120778 :         entry->counters.temp_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_read_time);
    1475      120778 :         entry->counters.temp_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_write_time);
    1476      120778 :         entry->counters.usage += USAGE_EXEC(total_time);
    1477      120778 :         entry->counters.wal_records += walusage->wal_records;
    1478      120778 :         entry->counters.wal_fpi += walusage->wal_fpi;
    1479      120778 :         entry->counters.wal_bytes += walusage->wal_bytes;
    1480      120778 :         entry->counters.wal_buffers_full += walusage->wal_buffers_full;
    1481      120778 :         if (jitusage)
    1482             :         {
    1483         150 :             entry->counters.jit_functions += jitusage->created_functions;
    1484         150 :             entry->counters.jit_generation_time += INSTR_TIME_GET_MILLISEC(jitusage->generation_counter);
    1485             : 
    1486         150 :             if (INSTR_TIME_GET_MILLISEC(jitusage->deform_counter))
    1487         146 :                 entry->counters.jit_deform_count++;
    1488         150 :             entry->counters.jit_deform_time += INSTR_TIME_GET_MILLISEC(jitusage->deform_counter);
    1489             : 
    1490         150 :             if (INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter))
    1491          66 :                 entry->counters.jit_inlining_count++;
    1492         150 :             entry->counters.jit_inlining_time += INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter);
    1493             : 
    1494         150 :             if (INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter))
    1495         146 :                 entry->counters.jit_optimization_count++;
    1496         150 :             entry->counters.jit_optimization_time += INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter);
    1497             : 
    1498         150 :             if (INSTR_TIME_GET_MILLISEC(jitusage->emission_counter))
    1499         146 :                 entry->counters.jit_emission_count++;
    1500         150 :             entry->counters.jit_emission_time += INSTR_TIME_GET_MILLISEC(jitusage->emission_counter);
    1501             :         }
    1502             : 
    1503             :         /* parallel worker counters */
    1504      120778 :         entry->counters.parallel_workers_to_launch += parallel_workers_to_launch;
    1505      120778 :         entry->counters.parallel_workers_launched += parallel_workers_launched;
    1506             : 
    1507             :         /* plan cache counters */
    1508      120778 :         if (planOrigin == PLAN_STMT_CACHE_GENERIC)
    1509        6092 :             entry->counters.generic_plan_calls++;
    1510      114686 :         else if (planOrigin == PLAN_STMT_CACHE_CUSTOM)
    1511         738 :             entry->counters.custom_plan_calls++;
    1512             : 
    1513      120778 :         SpinLockRelease(&entry->mutex);
    1514             :     }
    1515             : 
    1516       72344 : done:
    1517      193122 :     LWLockRelease(pgss->lock);
    1518             : 
    1519             :     /* We postpone this clean-up until we're out of the lock */
    1520      193122 :     if (norm_query)
    1521       21218 :         pfree(norm_query);
    1522             : }
    1523             : 
    1524             : /*
    1525             :  * Reset statement statistics corresponding to userid, dbid, and queryid.
    1526             :  */
    1527             : Datum
    1528           2 : pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
    1529             : {
    1530             :     Oid         userid;
    1531             :     Oid         dbid;
    1532             :     int64       queryid;
    1533             : 
    1534           2 :     userid = PG_GETARG_OID(0);
    1535           2 :     dbid = PG_GETARG_OID(1);
    1536           2 :     queryid = PG_GETARG_INT64(2);
    1537             : 
    1538           2 :     entry_reset(userid, dbid, queryid, false);
    1539             : 
    1540           2 :     PG_RETURN_VOID();
    1541             : }
    1542             : 
    1543             : Datum
    1544         232 : pg_stat_statements_reset_1_11(PG_FUNCTION_ARGS)
    1545             : {
    1546             :     Oid         userid;
    1547             :     Oid         dbid;
    1548             :     int64       queryid;
    1549             :     bool        minmax_only;
    1550             : 
    1551         232 :     userid = PG_GETARG_OID(0);
    1552         232 :     dbid = PG_GETARG_OID(1);
    1553         232 :     queryid = PG_GETARG_INT64(2);
    1554         232 :     minmax_only = PG_GETARG_BOOL(3);
    1555             : 
    1556         232 :     PG_RETURN_TIMESTAMPTZ(entry_reset(userid, dbid, queryid, minmax_only));
    1557             : }
    1558             : 
    1559             : /*
    1560             :  * Reset statement statistics.
    1561             :  */
    1562             : Datum
    1563           2 : pg_stat_statements_reset(PG_FUNCTION_ARGS)
    1564             : {
    1565           2 :     entry_reset(0, 0, 0, false);
    1566             : 
    1567           2 :     PG_RETURN_VOID();
    1568             : }
    1569             : 
    1570             : /* Number of output arguments (columns) for various API versions */
    1571             : #define PG_STAT_STATEMENTS_COLS_V1_0    14
    1572             : #define PG_STAT_STATEMENTS_COLS_V1_1    18
    1573             : #define PG_STAT_STATEMENTS_COLS_V1_2    19
    1574             : #define PG_STAT_STATEMENTS_COLS_V1_3    23
    1575             : #define PG_STAT_STATEMENTS_COLS_V1_8    32
    1576             : #define PG_STAT_STATEMENTS_COLS_V1_9    33
    1577             : #define PG_STAT_STATEMENTS_COLS_V1_10   43
    1578             : #define PG_STAT_STATEMENTS_COLS_V1_11   49
    1579             : #define PG_STAT_STATEMENTS_COLS_V1_12   52
    1580             : #define PG_STAT_STATEMENTS_COLS_V1_13   54
    1581             : #define PG_STAT_STATEMENTS_COLS         54  /* maximum of above */
    1582             : 
    1583             : /*
    1584             :  * Retrieve statement statistics.
    1585             :  *
    1586             :  * The SQL API of this function has changed multiple times, and will likely
    1587             :  * do so again in future.  To support the case where a newer version of this
    1588             :  * loadable module is being used with an old SQL declaration of the function,
    1589             :  * we continue to support the older API versions.  For 1.2 and later, the
    1590             :  * expected API version is identified by embedding it in the C name of the
    1591             :  * function.  Unfortunately we weren't bright enough to do that for 1.1.
    1592             :  */
    1593             : Datum
    1594         250 : pg_stat_statements_1_13(PG_FUNCTION_ARGS)
    1595             : {
    1596         250 :     bool        showtext = PG_GETARG_BOOL(0);
    1597             : 
    1598         250 :     pg_stat_statements_internal(fcinfo, PGSS_V1_13, showtext);
    1599             : 
    1600         250 :     return (Datum) 0;
    1601             : }
    1602             : 
    1603             : Datum
    1604           2 : pg_stat_statements_1_12(PG_FUNCTION_ARGS)
    1605             : {
    1606           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1607             : 
    1608           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_12, showtext);
    1609             : 
    1610           2 :     return (Datum) 0;
    1611             : }
    1612             : 
    1613             : Datum
    1614           2 : pg_stat_statements_1_11(PG_FUNCTION_ARGS)
    1615             : {
    1616           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1617             : 
    1618           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_11, showtext);
    1619             : 
    1620           2 :     return (Datum) 0;
    1621             : }
    1622             : 
    1623             : Datum
    1624           2 : pg_stat_statements_1_10(PG_FUNCTION_ARGS)
    1625             : {
    1626           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1627             : 
    1628           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_10, showtext);
    1629             : 
    1630           2 :     return (Datum) 0;
    1631             : }
    1632             : 
    1633             : Datum
    1634           2 : pg_stat_statements_1_9(PG_FUNCTION_ARGS)
    1635             : {
    1636           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1637             : 
    1638           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_9, showtext);
    1639             : 
    1640           2 :     return (Datum) 0;
    1641             : }
    1642             : 
    1643             : Datum
    1644           2 : pg_stat_statements_1_8(PG_FUNCTION_ARGS)
    1645             : {
    1646           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1647             : 
    1648           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
    1649             : 
    1650           2 :     return (Datum) 0;
    1651             : }
    1652             : 
    1653             : Datum
    1654           2 : pg_stat_statements_1_3(PG_FUNCTION_ARGS)
    1655             : {
    1656           2 :     bool        showtext = PG_GETARG_BOOL(0);
    1657             : 
    1658           2 :     pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
    1659             : 
    1660           2 :     return (Datum) 0;
    1661             : }
    1662             : 
    1663             : Datum
    1664           0 : pg_stat_statements_1_2(PG_FUNCTION_ARGS)
    1665             : {
    1666           0 :     bool        showtext = PG_GETARG_BOOL(0);
    1667             : 
    1668           0 :     pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
    1669             : 
    1670           0 :     return (Datum) 0;
    1671             : }
    1672             : 
    1673             : /*
    1674             :  * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
    1675             :  * This can be removed someday, perhaps.
    1676             :  */
    1677             : Datum
    1678           0 : pg_stat_statements(PG_FUNCTION_ARGS)
    1679             : {
    1680             :     /* If it's really API 1.1, we'll figure that out below */
    1681           0 :     pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
    1682             : 
    1683           0 :     return (Datum) 0;
    1684             : }
    1685             : 
    1686             : /* Common code for all versions of pg_stat_statements() */
    1687             : static void
    1688         262 : pg_stat_statements_internal(FunctionCallInfo fcinfo,
    1689             :                             pgssVersion api_version,
    1690             :                             bool showtext)
    1691             : {
    1692         262 :     ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
    1693         262 :     Oid         userid = GetUserId();
    1694         262 :     bool        is_allowed_role = false;
    1695         262 :     char       *qbuffer = NULL;
    1696         262 :     Size        qbuffer_size = 0;
    1697         262 :     Size        extent = 0;
    1698         262 :     int         gc_count = 0;
    1699             :     HASH_SEQ_STATUS hash_seq;
    1700             :     pgssEntry  *entry;
    1701             : 
    1702             :     /*
    1703             :      * Superusers or roles with the privileges of pg_read_all_stats members
    1704             :      * are allowed
    1705             :      */
    1706         262 :     is_allowed_role = has_privs_of_role(userid, ROLE_PG_READ_ALL_STATS);
    1707             : 
    1708             :     /* hash table must exist already */
    1709         262 :     if (!pgss || !pgss_hash)
    1710           0 :         ereport(ERROR,
    1711             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    1712             :                  errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
    1713             : 
    1714         262 :     InitMaterializedSRF(fcinfo, 0);
    1715             : 
    1716             :     /*
    1717             :      * Check we have the expected number of output arguments.  Aside from
    1718             :      * being a good safety check, we need a kluge here to detect API version
    1719             :      * 1.1, which was wedged into the code in an ill-considered way.
    1720             :      */
    1721         262 :     switch (rsinfo->setDesc->natts)
    1722             :     {
    1723           0 :         case PG_STAT_STATEMENTS_COLS_V1_0:
    1724           0 :             if (api_version != PGSS_V1_0)
    1725           0 :                 elog(ERROR, "incorrect number of output arguments");
    1726           0 :             break;
    1727           0 :         case PG_STAT_STATEMENTS_COLS_V1_1:
    1728             :             /* pg_stat_statements() should have told us 1.0 */
    1729           0 :             if (api_version != PGSS_V1_0)
    1730           0 :                 elog(ERROR, "incorrect number of output arguments");
    1731           0 :             api_version = PGSS_V1_1;
    1732           0 :             break;
    1733           0 :         case PG_STAT_STATEMENTS_COLS_V1_2:
    1734           0 :             if (api_version != PGSS_V1_2)
    1735           0 :                 elog(ERROR, "incorrect number of output arguments");
    1736           0 :             break;
    1737           2 :         case PG_STAT_STATEMENTS_COLS_V1_3:
    1738           2 :             if (api_version != PGSS_V1_3)
    1739           0 :                 elog(ERROR, "incorrect number of output arguments");
    1740           2 :             break;
    1741           2 :         case PG_STAT_STATEMENTS_COLS_V1_8:
    1742           2 :             if (api_version != PGSS_V1_8)
    1743           0 :                 elog(ERROR, "incorrect number of output arguments");
    1744           2 :             break;
    1745           2 :         case PG_STAT_STATEMENTS_COLS_V1_9:
    1746           2 :             if (api_version != PGSS_V1_9)
    1747           0 :                 elog(ERROR, "incorrect number of output arguments");
    1748           2 :             break;
    1749           2 :         case PG_STAT_STATEMENTS_COLS_V1_10:
    1750           2 :             if (api_version != PGSS_V1_10)
    1751           0 :                 elog(ERROR, "incorrect number of output arguments");
    1752           2 :             break;
    1753           2 :         case PG_STAT_STATEMENTS_COLS_V1_11:
    1754           2 :             if (api_version != PGSS_V1_11)
    1755           0 :                 elog(ERROR, "incorrect number of output arguments");
    1756           2 :             break;
    1757           2 :         case PG_STAT_STATEMENTS_COLS_V1_12:
    1758           2 :             if (api_version != PGSS_V1_12)
    1759           0 :                 elog(ERROR, "incorrect number of output arguments");
    1760           2 :             break;
    1761         250 :         case PG_STAT_STATEMENTS_COLS_V1_13:
    1762         250 :             if (api_version != PGSS_V1_13)
    1763           0 :                 elog(ERROR, "incorrect number of output arguments");
    1764         250 :             break;
    1765           0 :         default:
    1766           0 :             elog(ERROR, "incorrect number of output arguments");
    1767             :     }
    1768             : 
    1769             :     /*
    1770             :      * We'd like to load the query text file (if needed) while not holding any
    1771             :      * lock on pgss->lock.  In the worst case we'll have to do this again
    1772             :      * after we have the lock, but it's unlikely enough to make this a win
    1773             :      * despite occasional duplicated work.  We need to reload if anybody
    1774             :      * writes to the file (either a retail qtext_store(), or a garbage
    1775             :      * collection) between this point and where we've gotten shared lock.  If
    1776             :      * a qtext_store is actually in progress when we look, we might as well
    1777             :      * skip the speculative load entirely.
    1778             :      */
    1779         262 :     if (showtext)
    1780             :     {
    1781             :         int         n_writers;
    1782             : 
    1783             :         /* Take the mutex so we can examine variables */
    1784         262 :         SpinLockAcquire(&pgss->mutex);
    1785         262 :         extent = pgss->extent;
    1786         262 :         n_writers = pgss->n_writers;
    1787         262 :         gc_count = pgss->gc_count;
    1788         262 :         SpinLockRelease(&pgss->mutex);
    1789             : 
    1790             :         /* No point in loading file now if there are active writers */
    1791         262 :         if (n_writers == 0)
    1792         262 :             qbuffer = qtext_load_file(&qbuffer_size);
    1793             :     }
    1794             : 
    1795             :     /*
    1796             :      * Get shared lock, load or reload the query text file if we must, and
    1797             :      * iterate over the hashtable entries.
    1798             :      *
    1799             :      * With a large hash table, we might be holding the lock rather longer
    1800             :      * than one could wish.  However, this only blocks creation of new hash
    1801             :      * table entries, and the larger the hash table the less likely that is to
    1802             :      * be needed.  So we can hope this is okay.  Perhaps someday we'll decide
    1803             :      * we need to partition the hash table to limit the time spent holding any
    1804             :      * one lock.
    1805             :      */
    1806         262 :     LWLockAcquire(pgss->lock, LW_SHARED);
    1807             : 
    1808         262 :     if (showtext)
    1809             :     {
    1810             :         /*
    1811             :          * Here it is safe to examine extent and gc_count without taking the
    1812             :          * mutex.  Note that although other processes might change
    1813             :          * pgss->extent just after we look at it, the strings they then write
    1814             :          * into the file cannot yet be referenced in the hashtable, so we
    1815             :          * don't care whether we see them or not.
    1816             :          *
    1817             :          * If qtext_load_file fails, we just press on; we'll return NULL for
    1818             :          * every query text.
    1819             :          */
    1820         262 :         if (qbuffer == NULL ||
    1821         262 :             pgss->extent != extent ||
    1822         262 :             pgss->gc_count != gc_count)
    1823             :         {
    1824           0 :             free(qbuffer);
    1825           0 :             qbuffer = qtext_load_file(&qbuffer_size);
    1826             :         }
    1827             :     }
    1828             : 
    1829         262 :     hash_seq_init(&hash_seq, pgss_hash);
    1830       53672 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    1831             :     {
    1832             :         Datum       values[PG_STAT_STATEMENTS_COLS];
    1833             :         bool        nulls[PG_STAT_STATEMENTS_COLS];
    1834       53410 :         int         i = 0;
    1835             :         Counters    tmp;
    1836             :         double      stddev;
    1837       53410 :         int64       queryid = entry->key.queryid;
    1838             :         TimestampTz stats_since;
    1839             :         TimestampTz minmax_stats_since;
    1840             : 
    1841       53410 :         memset(values, 0, sizeof(values));
    1842       53410 :         memset(nulls, 0, sizeof(nulls));
    1843             : 
    1844       53410 :         values[i++] = ObjectIdGetDatum(entry->key.userid);
    1845       53410 :         values[i++] = ObjectIdGetDatum(entry->key.dbid);
    1846       53410 :         if (api_version >= PGSS_V1_9)
    1847       53386 :             values[i++] = BoolGetDatum(entry->key.toplevel);
    1848             : 
    1849       53410 :         if (is_allowed_role || entry->key.userid == userid)
    1850             :         {
    1851       53402 :             if (api_version >= PGSS_V1_2)
    1852       53402 :                 values[i++] = Int64GetDatumFast(queryid);
    1853             : 
    1854       53402 :             if (showtext)
    1855             :             {
    1856       53402 :                 char       *qstr = qtext_fetch(entry->query_offset,
    1857             :                                                entry->query_len,
    1858             :                                                qbuffer,
    1859             :                                                qbuffer_size);
    1860             : 
    1861       53402 :                 if (qstr)
    1862             :                 {
    1863             :                     char       *enc;
    1864             : 
    1865       53402 :                     enc = pg_any_to_server(qstr,
    1866             :                                            entry->query_len,
    1867             :                                            entry->encoding);
    1868             : 
    1869       53402 :                     values[i++] = CStringGetTextDatum(enc);
    1870             : 
    1871       53402 :                     if (enc != qstr)
    1872           0 :                         pfree(enc);
    1873             :                 }
    1874             :                 else
    1875             :                 {
    1876             :                     /* Just return a null if we fail to find the text */
    1877           0 :                     nulls[i++] = true;
    1878             :                 }
    1879             :             }
    1880             :             else
    1881             :             {
    1882             :                 /* Query text not requested */
    1883           0 :                 nulls[i++] = true;
    1884             :             }
    1885             :         }
    1886             :         else
    1887             :         {
    1888             :             /* Don't show queryid */
    1889           8 :             if (api_version >= PGSS_V1_2)
    1890           8 :                 nulls[i++] = true;
    1891             : 
    1892             :             /*
    1893             :              * Don't show query text, but hint as to the reason for not doing
    1894             :              * so if it was requested
    1895             :              */
    1896           8 :             if (showtext)
    1897           8 :                 values[i++] = CStringGetTextDatum("<insufficient privilege>");
    1898             :             else
    1899           0 :                 nulls[i++] = true;
    1900             :         }
    1901             : 
    1902             :         /* copy counters to a local variable to keep locking time short */
    1903       53410 :         SpinLockAcquire(&entry->mutex);
    1904       53410 :         tmp = entry->counters;
    1905       53410 :         SpinLockRelease(&entry->mutex);
    1906             : 
    1907             :         /*
    1908             :          * The spinlock is not required when reading these two as they are
    1909             :          * always updated when holding pgss->lock exclusively.
    1910             :          */
    1911       53410 :         stats_since = entry->stats_since;
    1912       53410 :         minmax_stats_since = entry->minmax_stats_since;
    1913             : 
    1914             :         /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
    1915       53410 :         if (IS_STICKY(tmp))
    1916          78 :             continue;
    1917             : 
    1918             :         /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
    1919      159996 :         for (int kind = 0; kind < PGSS_NUMKIND; kind++)
    1920             :         {
    1921      106664 :             if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
    1922             :             {
    1923      106656 :                 values[i++] = Int64GetDatumFast(tmp.calls[kind]);
    1924      106656 :                 values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
    1925             :             }
    1926             : 
    1927      106664 :             if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
    1928             :                 api_version >= PGSS_V1_8)
    1929             :             {
    1930      106656 :                 values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
    1931      106656 :                 values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
    1932      106656 :                 values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
    1933             : 
    1934             :                 /*
    1935             :                  * Note we are calculating the population variance here, not
    1936             :                  * the sample variance, as we have data for the whole
    1937             :                  * population, so Bessel's correction is not used, and we
    1938             :                  * don't divide by tmp.calls - 1.
    1939             :                  */
    1940      106656 :                 if (tmp.calls[kind] > 1)
    1941       10016 :                     stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
    1942             :                 else
    1943       96640 :                     stddev = 0.0;
    1944      106656 :                 values[i++] = Float8GetDatumFast(stddev);
    1945             :             }
    1946             :         }
    1947       53332 :         values[i++] = Int64GetDatumFast(tmp.rows);
    1948       53332 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
    1949       53332 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
    1950       53332 :         if (api_version >= PGSS_V1_1)
    1951       53332 :             values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
    1952       53332 :         values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
    1953       53332 :         values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
    1954       53332 :         values[i++] = Int64GetDatumFast(tmp.local_blks_read);
    1955       53332 :         if (api_version >= PGSS_V1_1)
    1956       53332 :             values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
    1957       53332 :         values[i++] = Int64GetDatumFast(tmp.local_blks_written);
    1958       53332 :         values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
    1959       53332 :         values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
    1960       53332 :         if (api_version >= PGSS_V1_1)
    1961             :         {
    1962       53332 :             values[i++] = Float8GetDatumFast(tmp.shared_blk_read_time);
    1963       53332 :             values[i++] = Float8GetDatumFast(tmp.shared_blk_write_time);
    1964             :         }
    1965       53332 :         if (api_version >= PGSS_V1_11)
    1966             :         {
    1967       53276 :             values[i++] = Float8GetDatumFast(tmp.local_blk_read_time);
    1968       53276 :             values[i++] = Float8GetDatumFast(tmp.local_blk_write_time);
    1969             :         }
    1970       53332 :         if (api_version >= PGSS_V1_10)
    1971             :         {
    1972       53294 :             values[i++] = Float8GetDatumFast(tmp.temp_blk_read_time);
    1973       53294 :             values[i++] = Float8GetDatumFast(tmp.temp_blk_write_time);
    1974             :         }
    1975       53332 :         if (api_version >= PGSS_V1_8)
    1976             :         {
    1977             :             char        buf[256];
    1978             :             Datum       wal_bytes;
    1979             : 
    1980       53324 :             values[i++] = Int64GetDatumFast(tmp.wal_records);
    1981       53324 :             values[i++] = Int64GetDatumFast(tmp.wal_fpi);
    1982             : 
    1983       53324 :             snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
    1984             : 
    1985             :             /* Convert to numeric. */
    1986       53324 :             wal_bytes = DirectFunctionCall3(numeric_in,
    1987             :                                             CStringGetDatum(buf),
    1988             :                                             ObjectIdGetDatum(0),
    1989             :                                             Int32GetDatum(-1));
    1990       53324 :             values[i++] = wal_bytes;
    1991             :         }
    1992       53332 :         if (api_version >= PGSS_V1_12)
    1993             :         {
    1994       53256 :             values[i++] = Int64GetDatumFast(tmp.wal_buffers_full);
    1995             :         }
    1996       53332 :         if (api_version >= PGSS_V1_10)
    1997             :         {
    1998       53294 :             values[i++] = Int64GetDatumFast(tmp.jit_functions);
    1999       53294 :             values[i++] = Float8GetDatumFast(tmp.jit_generation_time);
    2000       53294 :             values[i++] = Int64GetDatumFast(tmp.jit_inlining_count);
    2001       53294 :             values[i++] = Float8GetDatumFast(tmp.jit_inlining_time);
    2002       53294 :             values[i++] = Int64GetDatumFast(tmp.jit_optimization_count);
    2003       53294 :             values[i++] = Float8GetDatumFast(tmp.jit_optimization_time);
    2004       53294 :             values[i++] = Int64GetDatumFast(tmp.jit_emission_count);
    2005       53294 :             values[i++] = Float8GetDatumFast(tmp.jit_emission_time);
    2006             :         }
    2007       53332 :         if (api_version >= PGSS_V1_11)
    2008             :         {
    2009       53276 :             values[i++] = Int64GetDatumFast(tmp.jit_deform_count);
    2010       53276 :             values[i++] = Float8GetDatumFast(tmp.jit_deform_time);
    2011             :         }
    2012       53332 :         if (api_version >= PGSS_V1_12)
    2013             :         {
    2014       53256 :             values[i++] = Int64GetDatumFast(tmp.parallel_workers_to_launch);
    2015       53256 :             values[i++] = Int64GetDatumFast(tmp.parallel_workers_launched);
    2016             :         }
    2017       53332 :         if (api_version >= PGSS_V1_13)
    2018             :         {
    2019       53246 :             values[i++] = Int64GetDatumFast(tmp.generic_plan_calls);
    2020       53246 :             values[i++] = Int64GetDatumFast(tmp.custom_plan_calls);
    2021             :         }
    2022       53332 :         if (api_version >= PGSS_V1_11)
    2023             :         {
    2024       53276 :             values[i++] = TimestampTzGetDatum(stats_since);
    2025       53276 :             values[i++] = TimestampTzGetDatum(minmax_stats_since);
    2026             :         }
    2027             : 
    2028             :         Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
    2029             :                      api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
    2030             :                      api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
    2031             :                      api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
    2032             :                      api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
    2033             :                      api_version == PGSS_V1_9 ? PG_STAT_STATEMENTS_COLS_V1_9 :
    2034             :                      api_version == PGSS_V1_10 ? PG_STAT_STATEMENTS_COLS_V1_10 :
    2035             :                      api_version == PGSS_V1_11 ? PG_STAT_STATEMENTS_COLS_V1_11 :
    2036             :                      api_version == PGSS_V1_12 ? PG_STAT_STATEMENTS_COLS_V1_12 :
    2037             :                      api_version == PGSS_V1_13 ? PG_STAT_STATEMENTS_COLS_V1_13 :
    2038             :                      -1 /* fail if you forget to update this assert */ ));
    2039             : 
    2040       53332 :         tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
    2041             :     }
    2042             : 
    2043         262 :     LWLockRelease(pgss->lock);
    2044             : 
    2045         262 :     free(qbuffer);
    2046         262 : }
    2047             : 
    2048             : /* Number of output arguments (columns) for pg_stat_statements_info */
    2049             : #define PG_STAT_STATEMENTS_INFO_COLS    2
    2050             : 
    2051             : /*
    2052             :  * Return statistics of pg_stat_statements.
    2053             :  */
    2054             : Datum
    2055           4 : pg_stat_statements_info(PG_FUNCTION_ARGS)
    2056             : {
    2057             :     pgssGlobalStats stats;
    2058             :     TupleDesc   tupdesc;
    2059           4 :     Datum       values[PG_STAT_STATEMENTS_INFO_COLS] = {0};
    2060           4 :     bool        nulls[PG_STAT_STATEMENTS_INFO_COLS] = {0};
    2061             : 
    2062           4 :     if (!pgss || !pgss_hash)
    2063           0 :         ereport(ERROR,
    2064             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    2065             :                  errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
    2066             : 
    2067             :     /* Build a tuple descriptor for our result type */
    2068           4 :     if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
    2069           0 :         elog(ERROR, "return type must be a row type");
    2070             : 
    2071             :     /* Read global statistics for pg_stat_statements */
    2072           4 :     SpinLockAcquire(&pgss->mutex);
    2073           4 :     stats = pgss->stats;
    2074           4 :     SpinLockRelease(&pgss->mutex);
    2075             : 
    2076           4 :     values[0] = Int64GetDatum(stats.dealloc);
    2077           4 :     values[1] = TimestampTzGetDatum(stats.stats_reset);
    2078             : 
    2079           4 :     PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
    2080             : }
    2081             : 
    2082             : /*
    2083             :  * Estimate shared memory space needed.
    2084             :  */
    2085             : static Size
    2086          14 : pgss_memsize(void)
    2087             : {
    2088             :     Size        size;
    2089             : 
    2090          14 :     size = MAXALIGN(sizeof(pgssSharedState));
    2091          14 :     size = add_size(size, hash_estimate_size(pgss_max, sizeof(pgssEntry)));
    2092             : 
    2093          14 :     return size;
    2094             : }
    2095             : 
    2096             : /*
    2097             :  * Allocate a new hashtable entry.
    2098             :  * caller must hold an exclusive lock on pgss->lock
    2099             :  *
    2100             :  * "query" need not be null-terminated; we rely on query_len instead
    2101             :  *
    2102             :  * If "sticky" is true, make the new entry artificially sticky so that it will
    2103             :  * probably still be there when the query finishes execution.  We do this by
    2104             :  * giving it a median usage value rather than the normal value.  (Strictly
    2105             :  * speaking, query strings are normalized on a best effort basis, though it
    2106             :  * would be difficult to demonstrate this even under artificial conditions.)
    2107             :  *
    2108             :  * Note: despite needing exclusive lock, it's not an error for the target
    2109             :  * entry to already exist.  This is because pgss_store releases and
    2110             :  * reacquires lock after failing to find a match; so someone else could
    2111             :  * have made the entry while we waited to get exclusive lock.
    2112             :  */
    2113             : static pgssEntry *
    2114      108266 : entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
    2115             :             bool sticky)
    2116             : {
    2117             :     pgssEntry  *entry;
    2118             :     bool        found;
    2119             : 
    2120             :     /* Make space if needed */
    2121      108266 :     while (hash_get_num_entries(pgss_hash) >= pgss_max)
    2122           0 :         entry_dealloc();
    2123             : 
    2124             :     /* Find or create an entry with desired hash code */
    2125      108266 :     entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
    2126             : 
    2127      108266 :     if (!found)
    2128             :     {
    2129             :         /* New entry, initialize it */
    2130             : 
    2131             :         /* reset the statistics */
    2132      108266 :         memset(&entry->counters, 0, sizeof(Counters));
    2133             :         /* set the appropriate initial usage count */
    2134      108266 :         entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
    2135             :         /* re-initialize the mutex each time ... we assume no one using it */
    2136      108266 :         SpinLockInit(&entry->mutex);
    2137             :         /* ... and don't forget the query text metadata */
    2138             :         Assert(query_len >= 0);
    2139      108266 :         entry->query_offset = query_offset;
    2140      108266 :         entry->query_len = query_len;
    2141      108266 :         entry->encoding = encoding;
    2142      108266 :         entry->stats_since = GetCurrentTimestamp();
    2143      108266 :         entry->minmax_stats_since = entry->stats_since;
    2144             :     }
    2145             : 
    2146      108266 :     return entry;
    2147             : }
    2148             : 
    2149             : /*
    2150             :  * qsort comparator for sorting into increasing usage order
    2151             :  */
    2152             : static int
    2153           0 : entry_cmp(const void *lhs, const void *rhs)
    2154             : {
    2155           0 :     double      l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
    2156           0 :     double      r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
    2157             : 
    2158           0 :     if (l_usage < r_usage)
    2159           0 :         return -1;
    2160           0 :     else if (l_usage > r_usage)
    2161           0 :         return +1;
    2162             :     else
    2163           0 :         return 0;
    2164             : }
    2165             : 
    2166             : /*
    2167             :  * Deallocate least-used entries.
    2168             :  *
    2169             :  * Caller must hold an exclusive lock on pgss->lock.
    2170             :  */
    2171             : static void
    2172           0 : entry_dealloc(void)
    2173             : {
    2174             :     HASH_SEQ_STATUS hash_seq;
    2175             :     pgssEntry **entries;
    2176             :     pgssEntry  *entry;
    2177             :     int         nvictims;
    2178             :     int         i;
    2179             :     Size        tottextlen;
    2180             :     int         nvalidtexts;
    2181             : 
    2182             :     /*
    2183             :      * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
    2184             :      * While we're scanning the table, apply the decay factor to the usage
    2185             :      * values, and update the mean query length.
    2186             :      *
    2187             :      * Note that the mean query length is almost immediately obsolete, since
    2188             :      * we compute it before not after discarding the least-used entries.
    2189             :      * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
    2190             :      * making two passes to get a more current result.  Likewise, the new
    2191             :      * cur_median_usage includes the entries we're about to zap.
    2192             :      */
    2193             : 
    2194           0 :     entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
    2195             : 
    2196           0 :     i = 0;
    2197           0 :     tottextlen = 0;
    2198           0 :     nvalidtexts = 0;
    2199             : 
    2200           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2201           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2202             :     {
    2203           0 :         entries[i++] = entry;
    2204             :         /* "Sticky" entries get a different usage decay rate. */
    2205           0 :         if (IS_STICKY(entry->counters))
    2206           0 :             entry->counters.usage *= STICKY_DECREASE_FACTOR;
    2207             :         else
    2208           0 :             entry->counters.usage *= USAGE_DECREASE_FACTOR;
    2209             :         /* In the mean length computation, ignore dropped texts. */
    2210           0 :         if (entry->query_len >= 0)
    2211             :         {
    2212           0 :             tottextlen += entry->query_len + 1;
    2213           0 :             nvalidtexts++;
    2214             :         }
    2215             :     }
    2216             : 
    2217             :     /* Sort into increasing order by usage */
    2218           0 :     qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
    2219             : 
    2220             :     /* Record the (approximate) median usage */
    2221           0 :     if (i > 0)
    2222           0 :         pgss->cur_median_usage = entries[i / 2]->counters.usage;
    2223             :     /* Record the mean query length */
    2224           0 :     if (nvalidtexts > 0)
    2225           0 :         pgss->mean_query_len = tottextlen / nvalidtexts;
    2226             :     else
    2227           0 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2228             : 
    2229             :     /* Now zap an appropriate fraction of lowest-usage entries */
    2230           0 :     nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
    2231           0 :     nvictims = Min(nvictims, i);
    2232             : 
    2233           0 :     for (i = 0; i < nvictims; i++)
    2234             :     {
    2235           0 :         hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
    2236             :     }
    2237             : 
    2238           0 :     pfree(entries);
    2239             : 
    2240             :     /* Increment the number of times entries are deallocated */
    2241           0 :     SpinLockAcquire(&pgss->mutex);
    2242           0 :     pgss->stats.dealloc += 1;
    2243           0 :     SpinLockRelease(&pgss->mutex);
    2244           0 : }
    2245             : 
    2246             : /*
    2247             :  * Given a query string (not necessarily null-terminated), allocate a new
    2248             :  * entry in the external query text file and store the string there.
    2249             :  *
    2250             :  * If successful, returns true, and stores the new entry's offset in the file
    2251             :  * into *query_offset.  Also, if gc_count isn't NULL, *gc_count is set to the
    2252             :  * number of garbage collections that have occurred so far.
    2253             :  *
    2254             :  * On failure, returns false.
    2255             :  *
    2256             :  * At least a shared lock on pgss->lock must be held by the caller, so as
    2257             :  * to prevent a concurrent garbage collection.  Share-lock-holding callers
    2258             :  * should pass a gc_count pointer to obtain the number of garbage collections,
    2259             :  * so that they can recheck the count after obtaining exclusive lock to
    2260             :  * detect whether a garbage collection occurred (and removed this entry).
    2261             :  */
    2262             : static bool
    2263       56756 : qtext_store(const char *query, int query_len,
    2264             :             Size *query_offset, int *gc_count)
    2265             : {
    2266             :     Size        off;
    2267             :     int         fd;
    2268             : 
    2269             :     /*
    2270             :      * We use a spinlock to protect extent/n_writers/gc_count, so that
    2271             :      * multiple processes may execute this function concurrently.
    2272             :      */
    2273       56756 :     SpinLockAcquire(&pgss->mutex);
    2274       56756 :     off = pgss->extent;
    2275       56756 :     pgss->extent += query_len + 1;
    2276       56756 :     pgss->n_writers++;
    2277       56756 :     if (gc_count)
    2278       56756 :         *gc_count = pgss->gc_count;
    2279       56756 :     SpinLockRelease(&pgss->mutex);
    2280             : 
    2281       56756 :     *query_offset = off;
    2282             : 
    2283             :     /*
    2284             :      * Don't allow the file to grow larger than what qtext_load_file can
    2285             :      * (theoretically) handle.  This has been seen to be reachable on 32-bit
    2286             :      * platforms.
    2287             :      */
    2288       56756 :     if (unlikely(query_len >= MaxAllocHugeSize - off))
    2289             :     {
    2290           0 :         errno = EFBIG;          /* not quite right, but it'll do */
    2291           0 :         fd = -1;
    2292           0 :         goto error;
    2293             :     }
    2294             : 
    2295             :     /* Now write the data into the successfully-reserved part of the file */
    2296       56756 :     fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
    2297       56756 :     if (fd < 0)
    2298           0 :         goto error;
    2299             : 
    2300       56756 :     if (pg_pwrite(fd, query, query_len, off) != query_len)
    2301           0 :         goto error;
    2302       56756 :     if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
    2303           0 :         goto error;
    2304             : 
    2305       56756 :     CloseTransientFile(fd);
    2306             : 
    2307             :     /* Mark our write complete */
    2308       56756 :     SpinLockAcquire(&pgss->mutex);
    2309       56756 :     pgss->n_writers--;
    2310       56756 :     SpinLockRelease(&pgss->mutex);
    2311             : 
    2312       56756 :     return true;
    2313             : 
    2314           0 : error:
    2315           0 :     ereport(LOG,
    2316             :             (errcode_for_file_access(),
    2317             :              errmsg("could not write file \"%s\": %m",
    2318             :                     PGSS_TEXT_FILE)));
    2319             : 
    2320           0 :     if (fd >= 0)
    2321           0 :         CloseTransientFile(fd);
    2322             : 
    2323             :     /* Mark our write complete */
    2324           0 :     SpinLockAcquire(&pgss->mutex);
    2325           0 :     pgss->n_writers--;
    2326           0 :     SpinLockRelease(&pgss->mutex);
    2327             : 
    2328           0 :     return false;
    2329             : }
    2330             : 
    2331             : /*
    2332             :  * Read the external query text file into a malloc'd buffer.
    2333             :  *
    2334             :  * Returns NULL (without throwing an error) if unable to read, eg
    2335             :  * file not there or insufficient memory.
    2336             :  *
    2337             :  * On success, the buffer size is also returned into *buffer_size.
    2338             :  *
    2339             :  * This can be called without any lock on pgss->lock, but in that case
    2340             :  * the caller is responsible for verifying that the result is sane.
    2341             :  */
    2342             : static char *
    2343         272 : qtext_load_file(Size *buffer_size)
    2344             : {
    2345             :     char       *buf;
    2346             :     int         fd;
    2347             :     struct stat stat;
    2348             :     Size        nread;
    2349             : 
    2350         272 :     fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
    2351         272 :     if (fd < 0)
    2352             :     {
    2353           0 :         if (errno != ENOENT)
    2354           0 :             ereport(LOG,
    2355             :                     (errcode_for_file_access(),
    2356             :                      errmsg("could not read file \"%s\": %m",
    2357             :                             PGSS_TEXT_FILE)));
    2358           0 :         return NULL;
    2359             :     }
    2360             : 
    2361             :     /* Get file length */
    2362         272 :     if (fstat(fd, &stat))
    2363             :     {
    2364           0 :         ereport(LOG,
    2365             :                 (errcode_for_file_access(),
    2366             :                  errmsg("could not stat file \"%s\": %m",
    2367             :                         PGSS_TEXT_FILE)));
    2368           0 :         CloseTransientFile(fd);
    2369           0 :         return NULL;
    2370             :     }
    2371             : 
    2372             :     /* Allocate buffer; beware that off_t might be wider than size_t */
    2373         272 :     if (stat.st_size <= MaxAllocHugeSize)
    2374         272 :         buf = (char *) malloc(stat.st_size);
    2375             :     else
    2376           0 :         buf = NULL;
    2377         272 :     if (buf == NULL)
    2378             :     {
    2379           0 :         ereport(LOG,
    2380             :                 (errcode(ERRCODE_OUT_OF_MEMORY),
    2381             :                  errmsg("out of memory"),
    2382             :                  errdetail("Could not allocate enough memory to read file \"%s\".",
    2383             :                            PGSS_TEXT_FILE)));
    2384           0 :         CloseTransientFile(fd);
    2385           0 :         return NULL;
    2386             :     }
    2387             : 
    2388             :     /*
    2389             :      * OK, slurp in the file.  Windows fails if we try to read more than
    2390             :      * INT_MAX bytes at once, and other platforms might not like that either,
    2391             :      * so read a very large file in 1GB segments.
    2392             :      */
    2393         272 :     nread = 0;
    2394         542 :     while (nread < stat.st_size)
    2395             :     {
    2396         270 :         int         toread = Min(1024 * 1024 * 1024, stat.st_size - nread);
    2397             : 
    2398             :         /*
    2399             :          * If we get a short read and errno doesn't get set, the reason is
    2400             :          * probably that garbage collection truncated the file since we did
    2401             :          * the fstat(), so we don't log a complaint --- but we don't return
    2402             :          * the data, either, since it's most likely corrupt due to concurrent
    2403             :          * writes from garbage collection.
    2404             :          */
    2405         270 :         errno = 0;
    2406         270 :         if (read(fd, buf + nread, toread) != toread)
    2407             :         {
    2408           0 :             if (errno)
    2409           0 :                 ereport(LOG,
    2410             :                         (errcode_for_file_access(),
    2411             :                          errmsg("could not read file \"%s\": %m",
    2412             :                                 PGSS_TEXT_FILE)));
    2413           0 :             free(buf);
    2414           0 :             CloseTransientFile(fd);
    2415           0 :             return NULL;
    2416             :         }
    2417         270 :         nread += toread;
    2418             :     }
    2419             : 
    2420         272 :     if (CloseTransientFile(fd) != 0)
    2421           0 :         ereport(LOG,
    2422             :                 (errcode_for_file_access(),
    2423             :                  errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
    2424             : 
    2425         272 :     *buffer_size = nread;
    2426         272 :     return buf;
    2427             : }
    2428             : 
    2429             : /*
    2430             :  * Locate a query text in the file image previously read by qtext_load_file().
    2431             :  *
    2432             :  * We validate the given offset/length, and return NULL if bogus.  Otherwise,
    2433             :  * the result points to a null-terminated string within the buffer.
    2434             :  */
    2435             : static char *
    2436      159910 : qtext_fetch(Size query_offset, int query_len,
    2437             :             char *buffer, Size buffer_size)
    2438             : {
    2439             :     /* File read failed? */
    2440      159910 :     if (buffer == NULL)
    2441           0 :         return NULL;
    2442             :     /* Bogus offset/length? */
    2443      159910 :     if (query_len < 0 ||
    2444      159910 :         query_offset + query_len >= buffer_size)
    2445           0 :         return NULL;
    2446             :     /* As a further sanity check, make sure there's a trailing null */
    2447      159910 :     if (buffer[query_offset + query_len] != '\0')
    2448           0 :         return NULL;
    2449             :     /* Looks OK */
    2450      159910 :     return buffer + query_offset;
    2451             : }
    2452             : 
    2453             : /*
    2454             :  * Do we need to garbage-collect the external query text file?
    2455             :  *
    2456             :  * Caller should hold at least a shared lock on pgss->lock.
    2457             :  */
    2458             : static bool
    2459       56756 : need_gc_qtexts(void)
    2460             : {
    2461             :     Size        extent;
    2462             : 
    2463             :     /* Read shared extent pointer */
    2464       56756 :     SpinLockAcquire(&pgss->mutex);
    2465       56756 :     extent = pgss->extent;
    2466       56756 :     SpinLockRelease(&pgss->mutex);
    2467             : 
    2468             :     /*
    2469             :      * Don't proceed if file does not exceed 512 bytes per possible entry.
    2470             :      *
    2471             :      * Here and in the next test, 32-bit machines have overflow hazards if
    2472             :      * pgss_max and/or mean_query_len are large.  Force the multiplications
    2473             :      * and comparisons to be done in uint64 arithmetic to forestall trouble.
    2474             :      */
    2475       56756 :     if ((uint64) extent < (uint64) 512 * pgss_max)
    2476       56756 :         return false;
    2477             : 
    2478             :     /*
    2479             :      * Don't proceed if file is less than about 50% bloat.  Nothing can or
    2480             :      * should be done in the event of unusually large query texts accounting
    2481             :      * for file's large size.  We go to the trouble of maintaining the mean
    2482             :      * query length in order to prevent garbage collection from thrashing
    2483             :      * uselessly.
    2484             :      */
    2485           0 :     if ((uint64) extent < (uint64) pgss->mean_query_len * pgss_max * 2)
    2486           0 :         return false;
    2487             : 
    2488           0 :     return true;
    2489             : }
    2490             : 
    2491             : /*
    2492             :  * Garbage-collect orphaned query texts in external file.
    2493             :  *
    2494             :  * This won't be called often in the typical case, since it's likely that
    2495             :  * there won't be too much churn, and besides, a similar compaction process
    2496             :  * occurs when serializing to disk at shutdown or as part of resetting.
    2497             :  * Despite this, it seems prudent to plan for the edge case where the file
    2498             :  * becomes unreasonably large, with no other method of compaction likely to
    2499             :  * occur in the foreseeable future.
    2500             :  *
    2501             :  * The caller must hold an exclusive lock on pgss->lock.
    2502             :  *
    2503             :  * At the first sign of trouble we unlink the query text file to get a clean
    2504             :  * slate (although existing statistics are retained), rather than risk
    2505             :  * thrashing by allowing the same problem case to recur indefinitely.
    2506             :  */
    2507             : static void
    2508           0 : gc_qtexts(void)
    2509             : {
    2510             :     char       *qbuffer;
    2511             :     Size        qbuffer_size;
    2512           0 :     FILE       *qfile = NULL;
    2513             :     HASH_SEQ_STATUS hash_seq;
    2514             :     pgssEntry  *entry;
    2515             :     Size        extent;
    2516             :     int         nentries;
    2517             : 
    2518             :     /*
    2519             :      * When called from pgss_store, some other session might have proceeded
    2520             :      * with garbage collection in the no-lock-held interim of lock strength
    2521             :      * escalation.  Check once more that this is actually necessary.
    2522             :      */
    2523           0 :     if (!need_gc_qtexts())
    2524           0 :         return;
    2525             : 
    2526             :     /*
    2527             :      * Load the old texts file.  If we fail (out of memory, for instance),
    2528             :      * invalidate query texts.  Hopefully this is rare.  It might seem better
    2529             :      * to leave things alone on an OOM failure, but the problem is that the
    2530             :      * file is only going to get bigger; hoping for a future non-OOM result is
    2531             :      * risky and can easily lead to complete denial of service.
    2532             :      */
    2533           0 :     qbuffer = qtext_load_file(&qbuffer_size);
    2534           0 :     if (qbuffer == NULL)
    2535           0 :         goto gc_fail;
    2536             : 
    2537             :     /*
    2538             :      * We overwrite the query texts file in place, so as to reduce the risk of
    2539             :      * an out-of-disk-space failure.  Since the file is guaranteed not to get
    2540             :      * larger, this should always work on traditional filesystems; though we
    2541             :      * could still lose on copy-on-write filesystems.
    2542             :      */
    2543           0 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2544           0 :     if (qfile == NULL)
    2545             :     {
    2546           0 :         ereport(LOG,
    2547             :                 (errcode_for_file_access(),
    2548             :                  errmsg("could not write file \"%s\": %m",
    2549             :                         PGSS_TEXT_FILE)));
    2550           0 :         goto gc_fail;
    2551             :     }
    2552             : 
    2553           0 :     extent = 0;
    2554           0 :     nentries = 0;
    2555             : 
    2556           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2557           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2558             :     {
    2559           0 :         int         query_len = entry->query_len;
    2560           0 :         char       *qry = qtext_fetch(entry->query_offset,
    2561             :                                       query_len,
    2562             :                                       qbuffer,
    2563             :                                       qbuffer_size);
    2564             : 
    2565           0 :         if (qry == NULL)
    2566             :         {
    2567             :             /* Trouble ... drop the text */
    2568           0 :             entry->query_offset = 0;
    2569           0 :             entry->query_len = -1;
    2570             :             /* entry will not be counted in mean query length computation */
    2571           0 :             continue;
    2572             :         }
    2573             : 
    2574           0 :         if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
    2575             :         {
    2576           0 :             ereport(LOG,
    2577             :                     (errcode_for_file_access(),
    2578             :                      errmsg("could not write file \"%s\": %m",
    2579             :                             PGSS_TEXT_FILE)));
    2580           0 :             hash_seq_term(&hash_seq);
    2581           0 :             goto gc_fail;
    2582             :         }
    2583             : 
    2584           0 :         entry->query_offset = extent;
    2585           0 :         extent += query_len + 1;
    2586           0 :         nentries++;
    2587             :     }
    2588             : 
    2589             :     /*
    2590             :      * Truncate away any now-unused space.  If this fails for some odd reason,
    2591             :      * we log it, but there's no need to fail.
    2592             :      */
    2593           0 :     if (ftruncate(fileno(qfile), extent) != 0)
    2594           0 :         ereport(LOG,
    2595             :                 (errcode_for_file_access(),
    2596             :                  errmsg("could not truncate file \"%s\": %m",
    2597             :                         PGSS_TEXT_FILE)));
    2598             : 
    2599           0 :     if (FreeFile(qfile))
    2600             :     {
    2601           0 :         ereport(LOG,
    2602             :                 (errcode_for_file_access(),
    2603             :                  errmsg("could not write file \"%s\": %m",
    2604             :                         PGSS_TEXT_FILE)));
    2605           0 :         qfile = NULL;
    2606           0 :         goto gc_fail;
    2607             :     }
    2608             : 
    2609           0 :     elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
    2610             :          pgss->extent, extent);
    2611             : 
    2612             :     /* Reset the shared extent pointer */
    2613           0 :     pgss->extent = extent;
    2614             : 
    2615             :     /*
    2616             :      * Also update the mean query length, to be sure that need_gc_qtexts()
    2617             :      * won't still think we have a problem.
    2618             :      */
    2619           0 :     if (nentries > 0)
    2620           0 :         pgss->mean_query_len = extent / nentries;
    2621             :     else
    2622           0 :         pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2623             : 
    2624           0 :     free(qbuffer);
    2625             : 
    2626             :     /*
    2627             :      * OK, count a garbage collection cycle.  (Note: even though we have
    2628             :      * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
    2629             :      * other processes may examine gc_count while holding only the mutex.
    2630             :      * Also, we have to advance the count *after* we've rewritten the file,
    2631             :      * else other processes might not realize they read a stale file.)
    2632             :      */
    2633           0 :     record_gc_qtexts();
    2634             : 
    2635           0 :     return;
    2636             : 
    2637           0 : gc_fail:
    2638             :     /* clean up resources */
    2639           0 :     if (qfile)
    2640           0 :         FreeFile(qfile);
    2641           0 :     free(qbuffer);
    2642             : 
    2643             :     /*
    2644             :      * Since the contents of the external file are now uncertain, mark all
    2645             :      * hashtable entries as having invalid texts.
    2646             :      */
    2647           0 :     hash_seq_init(&hash_seq, pgss_hash);
    2648           0 :     while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2649             :     {
    2650           0 :         entry->query_offset = 0;
    2651           0 :         entry->query_len = -1;
    2652             :     }
    2653             : 
    2654             :     /*
    2655             :      * Destroy the query text file and create a new, empty one
    2656             :      */
    2657           0 :     (void) unlink(PGSS_TEXT_FILE);
    2658           0 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2659           0 :     if (qfile == NULL)
    2660           0 :         ereport(LOG,
    2661             :                 (errcode_for_file_access(),
    2662             :                  errmsg("could not recreate file \"%s\": %m",
    2663             :                         PGSS_TEXT_FILE)));
    2664             :     else
    2665           0 :         FreeFile(qfile);
    2666             : 
    2667             :     /* Reset the shared extent pointer */
    2668           0 :     pgss->extent = 0;
    2669             : 
    2670             :     /* Reset mean_query_len to match the new state */
    2671           0 :     pgss->mean_query_len = ASSUMED_LENGTH_INIT;
    2672             : 
    2673             :     /*
    2674             :      * Bump the GC count even though we failed.
    2675             :      *
    2676             :      * This is needed to make concurrent readers of file without any lock on
    2677             :      * pgss->lock notice existence of new version of file.  Once readers
    2678             :      * subsequently observe a change in GC count with pgss->lock held, that
    2679             :      * forces a safe reopen of file.  Writers also require that we bump here,
    2680             :      * of course.  (As required by locking protocol, readers and writers don't
    2681             :      * trust earlier file contents until gc_count is found unchanged after
    2682             :      * pgss->lock acquired in shared or exclusive mode respectively.)
    2683             :      */
    2684           0 :     record_gc_qtexts();
    2685             : }
    2686             : 
    2687             : #define SINGLE_ENTRY_RESET(e) \
    2688             : if (e) { \
    2689             :     if (minmax_only) { \
    2690             :         /* When requested reset only min/max statistics of an entry */ \
    2691             :         for (int kind = 0; kind < PGSS_NUMKIND; kind++) \
    2692             :         { \
    2693             :             e->counters.max_time[kind] = 0; \
    2694             :             e->counters.min_time[kind] = 0; \
    2695             :         } \
    2696             :         e->minmax_stats_since = stats_reset; \
    2697             :     } \
    2698             :     else \
    2699             :     { \
    2700             :         /* Remove the key otherwise  */ \
    2701             :         hash_search(pgss_hash, &e->key, HASH_REMOVE, NULL); \
    2702             :         num_remove++; \
    2703             :     } \
    2704             : }
    2705             : 
    2706             : /*
    2707             :  * Reset entries corresponding to parameters passed.
    2708             :  */
    2709             : static TimestampTz
    2710         236 : entry_reset(Oid userid, Oid dbid, int64 queryid, bool minmax_only)
    2711             : {
    2712             :     HASH_SEQ_STATUS hash_seq;
    2713             :     pgssEntry  *entry;
    2714             :     FILE       *qfile;
    2715             :     int64       num_entries;
    2716         236 :     int64       num_remove = 0;
    2717             :     pgssHashKey key;
    2718             :     TimestampTz stats_reset;
    2719             : 
    2720         236 :     if (!pgss || !pgss_hash)
    2721           0 :         ereport(ERROR,
    2722             :                 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
    2723             :                  errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
    2724             : 
    2725         236 :     LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
    2726         236 :     num_entries = hash_get_num_entries(pgss_hash);
    2727             : 
    2728         236 :     stats_reset = GetCurrentTimestamp();
    2729             : 
    2730         236 :     if (userid != 0 && dbid != 0 && queryid != INT64CONST(0))
    2731             :     {
    2732             :         /* If all the parameters are available, use the fast path. */
    2733           2 :         memset(&key, 0, sizeof(pgssHashKey));
    2734           2 :         key.userid = userid;
    2735           2 :         key.dbid = dbid;
    2736           2 :         key.queryid = queryid;
    2737             : 
    2738             :         /*
    2739             :          * Reset the entry if it exists, starting with the non-top-level
    2740             :          * entry.
    2741             :          */
    2742           2 :         key.toplevel = false;
    2743           2 :         entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    2744             : 
    2745           2 :         SINGLE_ENTRY_RESET(entry);
    2746             : 
    2747             :         /* Also reset the top-level entry if it exists. */
    2748           2 :         key.toplevel = true;
    2749           2 :         entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
    2750             : 
    2751           2 :         SINGLE_ENTRY_RESET(entry);
    2752             :     }
    2753         234 :     else if (userid != 0 || dbid != 0 || queryid != INT64CONST(0))
    2754             :     {
    2755             :         /* Reset entries corresponding to valid parameters. */
    2756           8 :         hash_seq_init(&hash_seq, pgss_hash);
    2757         102 :         while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2758             :         {
    2759          94 :             if ((!userid || entry->key.userid == userid) &&
    2760          72 :                 (!dbid || entry->key.dbid == dbid) &&
    2761          68 :                 (!queryid || entry->key.queryid == queryid))
    2762             :             {
    2763          14 :                 SINGLE_ENTRY_RESET(entry);
    2764             :             }
    2765             :         }
    2766             :     }
    2767             :     else
    2768             :     {
    2769             :         /* Reset all entries. */
    2770         226 :         hash_seq_init(&hash_seq, pgss_hash);
    2771        2212 :         while ((entry = hash_seq_search(&hash_seq)) != NULL)
    2772             :         {
    2773        1804 :             SINGLE_ENTRY_RESET(entry);
    2774             :         }
    2775             :     }
    2776             : 
    2777             :     /* All entries are removed? */
    2778         236 :     if (num_entries != num_remove)
    2779          12 :         goto release_lock;
    2780             : 
    2781             :     /*
    2782             :      * Reset global statistics for pg_stat_statements since all entries are
    2783             :      * removed.
    2784             :      */
    2785         224 :     SpinLockAcquire(&pgss->mutex);
    2786         224 :     pgss->stats.dealloc = 0;
    2787         224 :     pgss->stats.stats_reset = stats_reset;
    2788         224 :     SpinLockRelease(&pgss->mutex);
    2789             : 
    2790             :     /*
    2791             :      * Write new empty query file, perhaps even creating a new one to recover
    2792             :      * if the file was missing.
    2793             :      */
    2794         224 :     qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
    2795         224 :     if (qfile == NULL)
    2796             :     {
    2797           0 :         ereport(LOG,
    2798             :                 (errcode_for_file_access(),
    2799             :                  errmsg("could not create file \"%s\": %m",
    2800             :                         PGSS_TEXT_FILE)));
    2801           0 :         goto done;
    2802             :     }
    2803             : 
    2804             :     /* If ftruncate fails, log it, but it's not a fatal problem */
    2805         224 :     if (ftruncate(fileno(qfile), 0) != 0)
    2806           0 :         ereport(LOG,
    2807             :                 (errcode_for_file_access(),
    2808             :                  errmsg("could not truncate file \"%s\": %m",
    2809             :                         PGSS_TEXT_FILE)));
    2810             : 
    2811         224 :     FreeFile(qfile);
    2812             : 
    2813         224 : done:
    2814         224 :     pgss->extent = 0;
    2815             :     /* This counts as a query text garbage collection for our purposes */
    2816         224 :     record_gc_qtexts();
    2817             : 
    2818         236 : release_lock:
    2819         236 :     LWLockRelease(pgss->lock);
    2820             : 
    2821         236 :     return stats_reset;
    2822             : }
    2823             : 
    2824             : /*
    2825             :  * Generate a normalized version of the query string that will be used to
    2826             :  * represent all similar queries.
    2827             :  *
    2828             :  * Note that the normalized representation may well vary depending on
    2829             :  * just which "equivalent" query is used to create the hashtable entry.
    2830             :  * We assume this is OK.
    2831             :  *
    2832             :  * If query_loc > 0, then "query" has been advanced by that much compared to
    2833             :  * the original string start, so we need to translate the provided locations
    2834             :  * to compensate.  (This lets us avoid re-scanning statements before the one
    2835             :  * of interest, so it's worth doing.)
    2836             :  *
    2837             :  * *query_len_p contains the input string length, and is updated with
    2838             :  * the result string length on exit.  The resulting string might be longer
    2839             :  * or shorter depending on what happens with replacement of constants.
    2840             :  *
    2841             :  * Returns a palloc'd string.
    2842             :  */
    2843             : static char *
    2844       21218 : generate_normalized_query(JumbleState *jstate, const char *query,
    2845             :                           int query_loc, int *query_len_p)
    2846             : {
    2847             :     char       *norm_query;
    2848       21218 :     int         query_len = *query_len_p;
    2849             :     int         norm_query_buflen,  /* Space allowed for norm_query */
    2850             :                 len_to_wrt,     /* Length (in bytes) to write */
    2851       21218 :                 quer_loc = 0,   /* Source query byte location */
    2852       21218 :                 n_quer_loc = 0, /* Normalized query byte location */
    2853       21218 :                 last_off = 0,   /* Offset from start for previous tok */
    2854       21218 :                 last_tok_len = 0;   /* Length (in bytes) of that tok */
    2855       21218 :     int         num_constants_replaced = 0;
    2856             : 
    2857             :     /*
    2858             :      * Get constants' lengths (core system only gives us locations).  Note
    2859             :      * this also ensures the items are sorted by location.
    2860             :      */
    2861       21218 :     fill_in_constant_lengths(jstate, query, query_loc);
    2862             : 
    2863             :     /*
    2864             :      * Allow for $n symbols to be longer than the constants they replace.
    2865             :      * Constants must take at least one byte in text form, while a $n symbol
    2866             :      * certainly isn't more than 11 bytes, even if n reaches INT_MAX.  We
    2867             :      * could refine that limit based on the max value of n for the current
    2868             :      * query, but it hardly seems worth any extra effort to do so.
    2869             :      */
    2870       21218 :     norm_query_buflen = query_len + jstate->clocations_count * 10;
    2871             : 
    2872             :     /* Allocate result buffer */
    2873       21218 :     norm_query = palloc(norm_query_buflen + 1);
    2874             : 
    2875       83702 :     for (int i = 0; i < jstate->clocations_count; i++)
    2876             :     {
    2877             :         int         off,        /* Offset from start for cur tok */
    2878             :                     tok_len;    /* Length (in bytes) of that tok */
    2879             : 
    2880             :         /*
    2881             :          * If we have an external param at this location, but no lists are
    2882             :          * being squashed across the query, then we skip here; this will make
    2883             :          * us print the characters found in the original query that represent
    2884             :          * the parameter in the next iteration (or after the loop is done),
    2885             :          * which is a bit odd but seems to work okay in most cases.
    2886             :          */
    2887       62484 :         if (jstate->clocations[i].extern_param && !jstate->has_squashed_lists)
    2888         298 :             continue;
    2889             : 
    2890       62186 :         off = jstate->clocations[i].location;
    2891             : 
    2892             :         /* Adjust recorded location if we're dealing with partial string */
    2893       62186 :         off -= query_loc;
    2894             : 
    2895       62186 :         tok_len = jstate->clocations[i].length;
    2896             : 
    2897       62186 :         if (tok_len < 0)
    2898         478 :             continue;           /* ignore any duplicates */
    2899             : 
    2900             :         /* Copy next chunk (what precedes the next constant) */
    2901       61708 :         len_to_wrt = off - last_off;
    2902       61708 :         len_to_wrt -= last_tok_len;
    2903             :         Assert(len_to_wrt >= 0);
    2904       61708 :         memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2905       61708 :         n_quer_loc += len_to_wrt;
    2906             : 
    2907             :         /*
    2908             :          * And insert a param symbol in place of the constant token; and, if
    2909             :          * we have a squashable list, insert a placeholder comment starting
    2910             :          * from the list's second value.
    2911             :          */
    2912       61708 :         n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d%s",
    2913       61708 :                               num_constants_replaced + 1 + jstate->highest_extern_param_id,
    2914       61708 :                               jstate->clocations[i].squashed ? " /*, ... */" : "");
    2915       61708 :         num_constants_replaced++;
    2916             : 
    2917             :         /* move forward */
    2918       61708 :         quer_loc = off + tok_len;
    2919       61708 :         last_off = off;
    2920       61708 :         last_tok_len = tok_len;
    2921             :     }
    2922             : 
    2923             :     /*
    2924             :      * We've copied up until the last ignorable constant.  Copy over the
    2925             :      * remaining bytes of the original query string.
    2926             :      */
    2927       21218 :     len_to_wrt = query_len - quer_loc;
    2928             : 
    2929             :     Assert(len_to_wrt >= 0);
    2930       21218 :     memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
    2931       21218 :     n_quer_loc += len_to_wrt;
    2932             : 
    2933             :     Assert(n_quer_loc <= norm_query_buflen);
    2934       21218 :     norm_query[n_quer_loc] = '\0';
    2935             : 
    2936       21218 :     *query_len_p = n_quer_loc;
    2937       21218 :     return norm_query;
    2938             : }
    2939             : 
    2940             : /*
    2941             :  * Given a valid SQL string and an array of constant-location records,
    2942             :  * fill in the textual lengths of those constants.
    2943             :  *
    2944             :  * The constants may use any allowed constant syntax, such as float literals,
    2945             :  * bit-strings, single-quoted strings and dollar-quoted strings.  This is
    2946             :  * accomplished by using the public API for the core scanner.
    2947             :  *
    2948             :  * It is the caller's job to ensure that the string is a valid SQL statement
    2949             :  * with constants at the indicated locations.  Since in practice the string
    2950             :  * has already been parsed, and the locations that the caller provides will
    2951             :  * have originated from within the authoritative parser, this should not be
    2952             :  * a problem.
    2953             :  *
    2954             :  * Duplicate constant pointers are possible, and will have their lengths
    2955             :  * marked as '-1', so that they are later ignored.  (Actually, we assume the
    2956             :  * lengths were initialized as -1 to start with, and don't change them here.)
    2957             :  *
    2958             :  * If query_loc > 0, then "query" has been advanced by that much compared to
    2959             :  * the original string start, so we need to translate the provided locations
    2960             :  * to compensate.  (This lets us avoid re-scanning statements before the one
    2961             :  * of interest, so it's worth doing.)
    2962             :  *
    2963             :  * N.B. There is an assumption that a '-' character at a Const location begins
    2964             :  * a negative numeric constant.  This precludes there ever being another
    2965             :  * reason for a constant to start with a '-'.
    2966             :  */
    2967             : static void
    2968       21218 : fill_in_constant_lengths(JumbleState *jstate, const char *query,
    2969             :                          int query_loc)
    2970             : {
    2971             :     LocationLen *locs;
    2972             :     core_yyscan_t yyscanner;
    2973             :     core_yy_extra_type yyextra;
    2974             :     core_YYSTYPE yylval;
    2975             :     YYLTYPE     yylloc;
    2976       21218 :     int         last_loc = -1;
    2977             :     int         i;
    2978             : 
    2979             :     /*
    2980             :      * Sort the records by location so that we can process them in order while
    2981             :      * scanning the query text.
    2982             :      */
    2983       21218 :     if (jstate->clocations_count > 1)
    2984       13426 :         qsort(jstate->clocations, jstate->clocations_count,
    2985             :               sizeof(LocationLen), comp_location);
    2986       21218 :     locs = jstate->clocations;
    2987             : 
    2988             :     /* initialize the flex scanner --- should match raw_parser() */
    2989       21218 :     yyscanner = scanner_init(query,
    2990             :                              &yyextra,
    2991             :                              &ScanKeywords,
    2992             :                              ScanKeywordTokens);
    2993             : 
    2994             :     /* we don't want to re-emit any escape string warnings */
    2995       21218 :     yyextra.escape_string_warning = false;
    2996             : 
    2997             :     /* Search for each constant, in sequence */
    2998       83702 :     for (i = 0; i < jstate->clocations_count; i++)
    2999             :     {
    3000       62484 :         int         loc = locs[i].location;
    3001             :         int         tok;
    3002             : 
    3003             :         /* Adjust recorded location if we're dealing with partial string */
    3004       62484 :         loc -= query_loc;
    3005             : 
    3006             :         Assert(loc >= 0);
    3007             : 
    3008       62484 :         if (locs[i].squashed)
    3009        1254 :             continue;           /* squashable list, ignore */
    3010             : 
    3011       61230 :         if (loc <= last_loc)
    3012         482 :             continue;           /* Duplicate constant, ignore */
    3013             : 
    3014             :         /* Lex tokens until we find the desired constant */
    3015             :         for (;;)
    3016             :         {
    3017      475124 :             tok = core_yylex(&yylval, &yylloc, yyscanner);
    3018             : 
    3019             :             /* We should not hit end-of-string, but if we do, behave sanely */
    3020      475124 :             if (tok == 0)
    3021           0 :                 break;          /* out of inner for-loop */
    3022             : 
    3023             :             /*
    3024             :              * We should find the token position exactly, but if we somehow
    3025             :              * run past it, work with that.
    3026             :              */
    3027      475124 :             if (yylloc >= loc)
    3028             :             {
    3029       60748 :                 if (query[loc] == '-')
    3030             :                 {
    3031             :                     /*
    3032             :                      * It's a negative value - this is the one and only case
    3033             :                      * where we replace more than a single token.
    3034             :                      *
    3035             :                      * Do not compensate for the core system's special-case
    3036             :                      * adjustment of location to that of the leading '-'
    3037             :                      * operator in the event of a negative constant.  It is
    3038             :                      * also useful for our purposes to start from the minus
    3039             :                      * symbol.  In this way, queries like "select * from foo
    3040             :                      * where bar = 1" and "select * from foo where bar = -2"
    3041             :                      * will have identical normalized query strings.
    3042             :                      */
    3043         736 :                     tok = core_yylex(&yylval, &yylloc, yyscanner);
    3044         736 :                     if (tok == 0)
    3045           0 :                         break;  /* out of inner for-loop */
    3046             :                 }
    3047             : 
    3048             :                 /*
    3049             :                  * We now rely on the assumption that flex has placed a zero
    3050             :                  * byte after the text of the current token in scanbuf.
    3051             :                  */
    3052       60748 :                 locs[i].length = strlen(yyextra.scanbuf + loc);
    3053       60748 :                 break;          /* out of inner for-loop */
    3054             :             }
    3055             :         }
    3056             : 
    3057             :         /* If we hit end-of-string, give up, leaving remaining lengths -1 */
    3058       60748 :         if (tok == 0)
    3059           0 :             break;
    3060             : 
    3061       60748 :         last_loc = loc;
    3062             :     }
    3063             : 
    3064       21218 :     scanner_finish(yyscanner);
    3065       21218 : }
    3066             : 
    3067             : /*
    3068             :  * comp_location: comparator for qsorting LocationLen structs by location
    3069             :  */
    3070             : static int
    3071       69510 : comp_location(const void *a, const void *b)
    3072             : {
    3073       69510 :     int         l = ((const LocationLen *) a)->location;
    3074       69510 :     int         r = ((const LocationLen *) b)->location;
    3075             : 
    3076       69510 :     return pg_cmp_s32(l, r);
    3077             : }

Generated by: LCOV version 1.16