| 1 | /*------------------------------------------------------------------------- |
| 2 | * |
| 3 | * proc.h |
| 4 | * per-process shared memory data structures |
| 5 | * |
| 6 | * |
| 7 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
| 8 | * Portions Copyright (c) 1994, Regents of the University of California |
| 9 | * |
| 10 | * src/include/storage/proc.h |
| 11 | * |
| 12 | *------------------------------------------------------------------------- |
| 13 | */ |
| 14 | #ifndef _PROC_H_ |
| 15 | #define _PROC_H_ |
| 16 | |
| 17 | #include "access/clog.h" |
| 18 | #include "access/xlogdefs.h" |
| 19 | #include "lib/ilist.h" |
| 20 | #include "storage/latch.h" |
| 21 | #include "storage/lock.h" |
| 22 | #include "storage/pg_sema.h" |
| 23 | #include "storage/proclist_types.h" |
| 24 | |
| 25 | /* |
| 26 | * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds |
| 27 | * for non-aborted subtransactions of its current top transaction. These |
| 28 | * have to be treated as running XIDs by other backends. |
| 29 | * |
| 30 | * We also keep track of whether the cache overflowed (ie, the transaction has |
| 31 | * generated at least one subtransaction that didn't fit in the cache). |
| 32 | * If none of the caches have overflowed, we can assume that an XID that's not |
| 33 | * listed anywhere in the PGPROC array is not a running transaction. Else we |
| 34 | * have to look at pg_subtrans. |
| 35 | */ |
| 36 | #define PGPROC_MAX_CACHED_SUBXIDS 64 /* XXX guessed-at value */ |
| 37 | |
| 38 | struct XidCache |
| 39 | { |
| 40 | TransactionId xids[PGPROC_MAX_CACHED_SUBXIDS]; |
| 41 | }; |
| 42 | |
| 43 | /* |
| 44 | * Flags for PGXACT->vacuumFlags |
| 45 | * |
| 46 | * Note: If you modify these flags, you need to modify PROCARRAY_XXX flags |
| 47 | * in src/include/storage/procarray.h. |
| 48 | * |
| 49 | * PROC_RESERVED may later be assigned for use in vacuumFlags, but its value is |
| 50 | * used for PROCARRAY_SLOTS_XMIN in procarray.h, so GetOldestXmin won't be able |
| 51 | * to match and ignore processes with this flag set. |
| 52 | */ |
| 53 | #define PROC_IS_AUTOVACUUM 0x01 /* is it an autovac worker? */ |
| 54 | #define PROC_IN_VACUUM 0x02 /* currently running lazy vacuum */ |
| 55 | #define PROC_IN_ANALYZE 0x04 /* currently running analyze */ |
| 56 | #define PROC_VACUUM_FOR_WRAPAROUND 0x08 /* set by autovac only */ |
| 57 | #define PROC_IN_LOGICAL_DECODING 0x10 /* currently doing logical |
| 58 | * decoding outside xact */ |
| 59 | #define PROC_RESERVED 0x20 /* reserved for procarray */ |
| 60 | |
| 61 | /* flags reset at EOXact */ |
| 62 | #define PROC_VACUUM_STATE_MASK \ |
| 63 | (PROC_IN_VACUUM | PROC_IN_ANALYZE | PROC_VACUUM_FOR_WRAPAROUND) |
| 64 | |
| 65 | /* |
| 66 | * We allow a small number of "weak" relation locks (AccessShareLock, |
| 67 | * RowShareLock, RowExclusiveLock) to be recorded in the PGPROC structure |
| 68 | * rather than the main lock table. This eases contention on the lock |
| 69 | * manager LWLocks. See storage/lmgr/README for additional details. |
| 70 | */ |
| 71 | #define FP_LOCK_SLOTS_PER_BACKEND 16 |
| 72 | |
| 73 | /* |
| 74 | * An invalid pgprocno. Must be larger than the maximum number of PGPROC |
| 75 | * structures we could possibly have. See comments for MAX_BACKENDS. |
| 76 | */ |
| 77 | #define INVALID_PGPROCNO PG_INT32_MAX |
| 78 | |
| 79 | /* |
| 80 | * Each backend has a PGPROC struct in shared memory. There is also a list of |
| 81 | * currently-unused PGPROC structs that will be reallocated to new backends. |
| 82 | * |
| 83 | * links: list link for any list the PGPROC is in. When waiting for a lock, |
| 84 | * the PGPROC is linked into that lock's waitProcs queue. A recycled PGPROC |
| 85 | * is linked into ProcGlobal's freeProcs list. |
| 86 | * |
| 87 | * Note: twophase.c also sets up a dummy PGPROC struct for each currently |
| 88 | * prepared transaction. These PGPROCs appear in the ProcArray data structure |
| 89 | * so that the prepared transactions appear to be still running and are |
| 90 | * correctly shown as holding locks. A prepared transaction PGPROC can be |
| 91 | * distinguished from a real one at need by the fact that it has pid == 0. |
| 92 | * The semaphore and lock-activity fields in a prepared-xact PGPROC are unused, |
| 93 | * but its myProcLocks[] lists are valid. |
| 94 | */ |
| 95 | struct PGPROC |
| 96 | { |
| 97 | /* proc->links MUST BE FIRST IN STRUCT (see ProcSleep,ProcWakeup,etc) */ |
| 98 | SHM_QUEUE links; /* list link if process is in a list */ |
| 99 | PGPROC **procgloballist; /* procglobal list that owns this PGPROC */ |
| 100 | |
| 101 | PGSemaphore sem; /* ONE semaphore to sleep on */ |
| 102 | int waitStatus; /* STATUS_WAITING, STATUS_OK or STATUS_ERROR */ |
| 103 | |
| 104 | Latch procLatch; /* generic latch for process */ |
| 105 | |
| 106 | LocalTransactionId lxid; /* local id of top-level transaction currently |
| 107 | * being executed by this proc, if running; |
| 108 | * else InvalidLocalTransactionId */ |
| 109 | int pid; /* Backend's process ID; 0 if prepared xact */ |
| 110 | int pgprocno; |
| 111 | |
| 112 | /* These fields are zero while a backend is still starting up: */ |
| 113 | BackendId backendId; /* This backend's backend ID (if assigned) */ |
| 114 | Oid databaseId; /* OID of database this backend is using */ |
| 115 | Oid roleId; /* OID of role using this backend */ |
| 116 | |
| 117 | Oid tempNamespaceId; /* OID of temp schema this backend is |
| 118 | * using */ |
| 119 | |
| 120 | bool isBackgroundWorker; /* true if background worker. */ |
| 121 | |
| 122 | /* |
| 123 | * While in hot standby mode, shows that a conflict signal has been sent |
| 124 | * for the current transaction. Set/cleared while holding ProcArrayLock, |
| 125 | * though not required. Accessed without lock, if needed. |
| 126 | */ |
| 127 | bool recoveryConflictPending; |
| 128 | |
| 129 | /* Info about LWLock the process is currently waiting for, if any. */ |
| 130 | bool lwWaiting; /* true if waiting for an LW lock */ |
| 131 | uint8 lwWaitMode; /* lwlock mode being waited for */ |
| 132 | proclist_node lwWaitLink; /* position in LW lock wait list */ |
| 133 | |
| 134 | /* Support for condition variables. */ |
| 135 | proclist_node cvWaitLink; /* position in CV wait list */ |
| 136 | |
| 137 | /* Info about lock the process is currently waiting for, if any. */ |
| 138 | /* waitLock and waitProcLock are NULL if not currently waiting. */ |
| 139 | LOCK *waitLock; /* Lock object we're sleeping on ... */ |
| 140 | PROCLOCK *waitProcLock; /* Per-holder info for awaited lock */ |
| 141 | LOCKMODE waitLockMode; /* type of lock we're waiting for */ |
| 142 | LOCKMASK heldLocks; /* bitmask for lock types already held on this |
| 143 | * lock object by this backend */ |
| 144 | |
| 145 | /* |
| 146 | * Info to allow us to wait for synchronous replication, if needed. |
| 147 | * waitLSN is InvalidXLogRecPtr if not waiting; set only by user backend. |
| 148 | * syncRepState must not be touched except by owning process or WALSender. |
| 149 | * syncRepLinks used only while holding SyncRepLock. |
| 150 | */ |
| 151 | XLogRecPtr waitLSN; /* waiting for this LSN or higher */ |
| 152 | int syncRepState; /* wait state for sync rep */ |
| 153 | SHM_QUEUE syncRepLinks; /* list link if process is in syncrep queue */ |
| 154 | |
| 155 | /* |
| 156 | * All PROCLOCK objects for locks held or awaited by this backend are |
| 157 | * linked into one of these lists, according to the partition number of |
| 158 | * their lock. |
| 159 | */ |
| 160 | SHM_QUEUE myProcLocks[NUM_LOCK_PARTITIONS]; |
| 161 | |
| 162 | struct XidCache subxids; /* cache for subtransaction XIDs */ |
| 163 | |
| 164 | /* Support for group XID clearing. */ |
| 165 | /* true, if member of ProcArray group waiting for XID clear */ |
| 166 | bool procArrayGroupMember; |
| 167 | /* next ProcArray group member waiting for XID clear */ |
| 168 | pg_atomic_uint32 procArrayGroupNext; |
| 169 | |
| 170 | /* |
| 171 | * latest transaction id among the transaction's main XID and |
| 172 | * subtransactions |
| 173 | */ |
| 174 | TransactionId procArrayGroupMemberXid; |
| 175 | |
| 176 | uint32 wait_event_info; /* proc's wait information */ |
| 177 | |
| 178 | /* Support for group transaction status update. */ |
| 179 | bool clogGroupMember; /* true, if member of clog group */ |
| 180 | pg_atomic_uint32 clogGroupNext; /* next clog group member */ |
| 181 | TransactionId clogGroupMemberXid; /* transaction id of clog group member */ |
| 182 | XidStatus clogGroupMemberXidStatus; /* transaction status of clog |
| 183 | * group member */ |
| 184 | int clogGroupMemberPage; /* clog page corresponding to |
| 185 | * transaction id of clog group member */ |
| 186 | XLogRecPtr clogGroupMemberLsn; /* WAL location of commit record for clog |
| 187 | * group member */ |
| 188 | |
| 189 | /* Per-backend LWLock. Protects fields below (but not group fields). */ |
| 190 | LWLock backendLock; |
| 191 | |
| 192 | /* Lock manager data, recording fast-path locks taken by this backend. */ |
| 193 | uint64 fpLockBits; /* lock modes held for each fast-path slot */ |
| 194 | Oid fpRelId[FP_LOCK_SLOTS_PER_BACKEND]; /* slots for rel oids */ |
| 195 | bool fpVXIDLock; /* are we holding a fast-path VXID lock? */ |
| 196 | LocalTransactionId fpLocalTransactionId; /* lxid for fast-path VXID |
| 197 | * lock */ |
| 198 | |
| 199 | /* |
| 200 | * Support for lock groups. Use LockHashPartitionLockByProc on the group |
| 201 | * leader to get the LWLock protecting these fields. |
| 202 | */ |
| 203 | PGPROC *lockGroupLeader; /* lock group leader, if I'm a member */ |
| 204 | dlist_head lockGroupMembers; /* list of members, if I'm a leader */ |
| 205 | dlist_node lockGroupLink; /* my member link, if I'm a member */ |
| 206 | }; |
| 207 | |
| 208 | /* NOTE: "typedef struct PGPROC PGPROC" appears in storage/lock.h. */ |
| 209 | |
| 210 | |
| 211 | extern PGDLLIMPORT PGPROC *MyProc; |
| 212 | extern PGDLLIMPORT struct PGXACT *MyPgXact; |
| 213 | |
| 214 | /* |
| 215 | * Prior to PostgreSQL 9.2, the fields below were stored as part of the |
| 216 | * PGPROC. However, benchmarking revealed that packing these particular |
| 217 | * members into a separate array as tightly as possible sped up GetSnapshotData |
| 218 | * considerably on systems with many CPU cores, by reducing the number of |
| 219 | * cache lines needing to be fetched. Thus, think very carefully before adding |
| 220 | * anything else here. |
| 221 | */ |
| 222 | typedef struct PGXACT |
| 223 | { |
| 224 | TransactionId xid; /* id of top-level transaction currently being |
| 225 | * executed by this proc, if running and XID |
| 226 | * is assigned; else InvalidTransactionId */ |
| 227 | |
| 228 | TransactionId xmin; /* minimal running XID as it was when we were |
| 229 | * starting our xact, excluding LAZY VACUUM: |
| 230 | * vacuum must not remove tuples deleted by |
| 231 | * xid >= xmin ! */ |
| 232 | |
| 233 | uint8 vacuumFlags; /* vacuum-related flags, see above */ |
| 234 | bool overflowed; |
| 235 | bool delayChkpt; /* true if this proc delays checkpoint start; |
| 236 | * previously called InCommit */ |
| 237 | |
| 238 | uint8 nxids; |
| 239 | } PGXACT; |
| 240 | |
| 241 | /* |
| 242 | * There is one ProcGlobal struct for the whole database cluster. |
| 243 | */ |
| 244 | typedef struct PROC_HDR |
| 245 | { |
| 246 | /* Array of PGPROC structures (not including dummies for prepared txns) */ |
| 247 | PGPROC *allProcs; |
| 248 | /* Array of PGXACT structures (not including dummies for prepared txns) */ |
| 249 | PGXACT *allPgXact; |
| 250 | /* Length of allProcs array */ |
| 251 | uint32 allProcCount; |
| 252 | /* Head of list of free PGPROC structures */ |
| 253 | PGPROC *freeProcs; |
| 254 | /* Head of list of autovacuum's free PGPROC structures */ |
| 255 | PGPROC *autovacFreeProcs; |
| 256 | /* Head of list of bgworker free PGPROC structures */ |
| 257 | PGPROC *bgworkerFreeProcs; |
| 258 | /* Head of list of walsender free PGPROC structures */ |
| 259 | PGPROC *walsenderFreeProcs; |
| 260 | /* First pgproc waiting for group XID clear */ |
| 261 | pg_atomic_uint32 procArrayGroupFirst; |
| 262 | /* First pgproc waiting for group transaction status update */ |
| 263 | pg_atomic_uint32 clogGroupFirst; |
| 264 | /* WALWriter process's latch */ |
| 265 | Latch *walwriterLatch; |
| 266 | /* Checkpointer process's latch */ |
| 267 | Latch *checkpointerLatch; |
| 268 | /* Current shared estimate of appropriate spins_per_delay value */ |
| 269 | int spins_per_delay; |
| 270 | /* The proc of the Startup process, since not in ProcArray */ |
| 271 | PGPROC *startupProc; |
| 272 | int startupProcPid; |
| 273 | /* Buffer id of the buffer that Startup process waits for pin on, or -1 */ |
| 274 | int startupBufferPinWaitBufId; |
| 275 | } PROC_HDR; |
| 276 | |
| 277 | extern PGDLLIMPORT PROC_HDR *ProcGlobal; |
| 278 | |
| 279 | extern PGPROC *PreparedXactProcs; |
| 280 | |
| 281 | /* Accessor for PGPROC given a pgprocno. */ |
| 282 | #define GetPGProcByNumber(n) (&ProcGlobal->allProcs[(n)]) |
| 283 | |
| 284 | /* |
| 285 | * We set aside some extra PGPROC structures for auxiliary processes, |
| 286 | * ie things that aren't full-fledged backends but need shmem access. |
| 287 | * |
| 288 | * Background writer, checkpointer and WAL writer run during normal operation. |
| 289 | * Startup process and WAL receiver also consume 2 slots, but WAL writer is |
| 290 | * launched only after startup has exited, so we only need 4 slots. |
| 291 | */ |
| 292 | #define NUM_AUXILIARY_PROCS 4 |
| 293 | |
| 294 | /* configurable options */ |
| 295 | extern PGDLLIMPORT int DeadlockTimeout; |
| 296 | extern int StatementTimeout; |
| 297 | extern int LockTimeout; |
| 298 | extern int IdleInTransactionSessionTimeout; |
| 299 | extern bool log_lock_waits; |
| 300 | |
| 301 | |
| 302 | /* |
| 303 | * Function Prototypes |
| 304 | */ |
| 305 | extern int ProcGlobalSemas(void); |
| 306 | extern Size ProcGlobalShmemSize(void); |
| 307 | extern void InitProcGlobal(void); |
| 308 | extern void InitProcess(void); |
| 309 | extern void InitProcessPhase2(void); |
| 310 | extern void InitAuxiliaryProcess(void); |
| 311 | |
| 312 | extern void PublishStartupProcessInformation(void); |
| 313 | extern void SetStartupBufferPinWaitBufId(int bufid); |
| 314 | extern int GetStartupBufferPinWaitBufId(void); |
| 315 | |
| 316 | extern bool HaveNFreeProcs(int n); |
| 317 | extern void ProcReleaseLocks(bool isCommit); |
| 318 | |
| 319 | extern void ProcQueueInit(PROC_QUEUE *queue); |
| 320 | extern int ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable); |
| 321 | extern PGPROC *ProcWakeup(PGPROC *proc, int waitStatus); |
| 322 | extern void ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock); |
| 323 | extern void CheckDeadLockAlert(void); |
| 324 | extern bool IsWaitingForLock(void); |
| 325 | extern void LockErrorCleanup(void); |
| 326 | |
| 327 | extern void ProcWaitForSignal(uint32 wait_event_info); |
| 328 | extern void ProcSendSignal(int pid); |
| 329 | |
| 330 | extern PGPROC *AuxiliaryPidGetProc(int pid); |
| 331 | |
| 332 | extern void BecomeLockGroupLeader(void); |
| 333 | extern bool BecomeLockGroupMember(PGPROC *leader, int pid); |
| 334 | |
| 335 | #endif /* PROC_H */ |
| 336 | |