PostgreSQL Source Code  git master
multixact.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * multixact.c
4  * PostgreSQL multi-transaction-log manager
5  *
6  * The pg_multixact manager is a pg_xact-like manager that stores an array of
7  * MultiXactMember for each MultiXactId. It is a fundamental part of the
8  * shared-row-lock implementation. Each MultiXactMember is comprised of a
9  * TransactionId and a set of flag bits. The name is a bit historical:
10  * originally, a MultiXactId consisted of more than one TransactionId (except
11  * in rare corner cases), hence "multi". Nowadays, however, it's perfectly
12  * legitimate to have MultiXactIds that only include a single Xid.
13  *
14  * The meaning of the flag bits is opaque to this module, but they are mostly
15  * used in heapam.c to identify lock modes that each of the member transactions
16  * is holding on any given tuple. This module just contains support to store
17  * and retrieve the arrays.
18  *
19  * We use two SLRU areas, one for storing the offsets at which the data
20  * starts for each MultiXactId in the other one. This trick allows us to
21  * store variable length arrays of TransactionIds. (We could alternatively
22  * use one area containing counts and TransactionIds, with valid MultiXactId
23  * values pointing at slots containing counts; but that way seems less robust
24  * since it would get completely confused if someone inquired about a bogus
25  * MultiXactId that pointed to an intermediate slot containing an XID.)
26  *
27  * XLOG interactions: this module generates a record whenever a new OFFSETs or
28  * MEMBERs page is initialized to zeroes, as well as an
29  * XLOG_MULTIXACT_CREATE_ID record whenever a new MultiXactId is defined.
30  * This module ignores the WAL rule "write xlog before data," because it
31  * suffices that actions recording a MultiXactId in a heap xmax do follow that
32  * rule. The only way for the MXID to be referenced from any data page is for
33  * heap_lock_tuple() or heap_update() to have put it there, and each generates
34  * an XLOG record that must follow ours. The normal LSN interlock between the
35  * data page and that XLOG record will ensure that our XLOG record reaches
36  * disk first. If the SLRU members/offsets data reaches disk sooner than the
37  * XLOG records, we do not care; after recovery, no xmax will refer to it. On
38  * the flip side, to ensure that all referenced entries _do_ reach disk, this
39  * module's XLOG records completely rebuild the data entered since the last
40  * checkpoint. We flush and sync all dirty OFFSETs and MEMBERs pages to disk
41  * before each checkpoint is considered complete.
42  *
43  * Like clog.c, and unlike subtrans.c, we have to preserve state across
44  * crashes and ensure that MXID and offset numbering increases monotonically
45  * across a crash. We do this in the same way as it's done for transaction
46  * IDs: the WAL record is guaranteed to contain evidence of every MXID we
47  * could need to worry about, and we just make sure that at the end of
48  * replay, the next-MXID and next-offset counters are at least as large as
49  * anything we saw during replay.
50  *
51  * We are able to remove segments no longer necessary by carefully tracking
52  * each table's used values: during vacuum, any multixact older than a certain
53  * value is removed; the cutoff value is stored in pg_class. The minimum value
54  * across all tables in each database is stored in pg_database, and the global
55  * minimum across all databases is part of pg_control and is kept in shared
56  * memory. Whenever that minimum is advanced, the SLRUs are truncated.
57  *
58  * When new multixactid values are to be created, care is taken that the
59  * counter does not fall within the wraparound horizon considering the global
60  * minimum value.
61  *
62  * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
63  * Portions Copyright (c) 1994, Regents of the University of California
64  *
65  * src/backend/access/transam/multixact.c
66  *
67  *-------------------------------------------------------------------------
68  */
69 #include "postgres.h"
70 
71 #include "access/multixact.h"
72 #include "access/slru.h"
73 #include "access/transam.h"
74 #include "access/twophase.h"
75 #include "access/twophase_rmgr.h"
76 #include "access/xact.h"
77 #include "access/xloginsert.h"
78 #include "access/xlogutils.h"
79 #include "catalog/pg_type.h"
80 #include "commands/dbcommands.h"
81 #include "funcapi.h"
82 #include "lib/ilist.h"
83 #include "miscadmin.h"
84 #include "pg_trace.h"
85 #include "postmaster/autovacuum.h"
86 #include "storage/lmgr.h"
87 #include "storage/pmsignal.h"
88 #include "storage/proc.h"
89 #include "storage/procarray.h"
90 #include "utils/builtins.h"
91 #include "utils/memutils.h"
92 #include "utils/snapmgr.h"
93 
94 
95 /*
96  * Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
97  * used everywhere else in Postgres.
98  *
99  * Note: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
100  * MultiXact page numbering also wraps around at
101  * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
102  * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need
103  * take no explicit notice of that fact in this module, except when comparing
104  * segment and page numbers in TruncateMultiXact (see
105  * MultiXactOffsetPagePrecedes).
106  */
107 
108 /* We need four bytes per offset */
109 #define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
110 
111 #define MultiXactIdToOffsetPage(xid) \
112  ((xid) / (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
113 #define MultiXactIdToOffsetEntry(xid) \
114  ((xid) % (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
115 #define MultiXactIdToOffsetSegment(xid) (MultiXactIdToOffsetPage(xid) / SLRU_PAGES_PER_SEGMENT)
116 
117 /*
118  * The situation for members is a bit more complex: we store one byte of
119  * additional flag bits for each TransactionId. To do this without getting
120  * into alignment issues, we store four bytes of flags, and then the
121  * corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
122  * are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
123  * per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
124  * performance) trumps space efficiency here.
125  *
126  * Note that the "offset" macros work with byte offset, not array indexes, so
127  * arithmetic must be done using "char *" pointers.
128  */
129 /* We need eight bits per xact, so one xact fits in a byte */
130 #define MXACT_MEMBER_BITS_PER_XACT 8
131 #define MXACT_MEMBER_FLAGS_PER_BYTE 1
132 #define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
133 
134 /* how many full bytes of flags are there in a group? */
135 #define MULTIXACT_FLAGBYTES_PER_GROUP 4
136 #define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
137  (MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
138 /* size in bytes of a complete group */
139 #define MULTIXACT_MEMBERGROUP_SIZE \
140  (sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
141 #define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
142 #define MULTIXACT_MEMBERS_PER_PAGE \
143  (MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
144 
145 /*
146  * Because the number of items per page is not a divisor of the last item
147  * number (member 0xFFFFFFFF), the last segment does not use the maximum number
148  * of pages, and moreover the last used page therein does not use the same
149  * number of items as previous pages. (Another way to say it is that the
150  * 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
151  * has some empty space after that item.)
152  *
153  * This constant is the number of members in the last page of the last segment.
154  */
155 #define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
156  ((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
157 
158 /* page in which a member is to be found */
159 #define MXOffsetToMemberPage(xid) ((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_PAGE)
160 #define MXOffsetToMemberSegment(xid) (MXOffsetToMemberPage(xid) / SLRU_PAGES_PER_SEGMENT)
161 
162 /* Location (byte offset within page) of flag word for a given member */
163 #define MXOffsetToFlagsOffset(xid) \
164  ((((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) % \
165  (TransactionId) MULTIXACT_MEMBERGROUPS_PER_PAGE) * \
166  (TransactionId) MULTIXACT_MEMBERGROUP_SIZE)
167 #define MXOffsetToFlagsBitShift(xid) \
168  (((xid) % (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) * \
169  MXACT_MEMBER_BITS_PER_XACT)
170 
171 /* Location (byte offset within page) of TransactionId of given member */
172 #define MXOffsetToMemberOffset(xid) \
173  (MXOffsetToFlagsOffset(xid) + MULTIXACT_FLAGBYTES_PER_GROUP + \
174  ((xid) % MULTIXACT_MEMBERS_PER_MEMBERGROUP) * sizeof(TransactionId))
175 
176 /* Multixact members wraparound thresholds. */
177 #define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
178 #define MULTIXACT_MEMBER_DANGER_THRESHOLD \
179  (MaxMultiXactOffset - MaxMultiXactOffset / 4)
180 
181 #define PreviousMultiXactId(xid) \
182  ((xid) == FirstMultiXactId ? MaxMultiXactId : (xid) - 1)
183 
184 /*
185  * Links to shared-memory data structures for MultiXact control
186  */
189 
190 #define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
191 #define MultiXactMemberCtl (&MultiXactMemberCtlData)
192 
193 /*
194  * MultiXact state shared across all backends. All this state is protected
195  * by MultiXactGenLock. (We also use MultiXactOffsetSLRULock and
196  * MultiXactMemberSLRULock to guard accesses to the two sets of SLRU
197  * buffers. For concurrency's sake, we avoid holding more than one of these
198  * locks at a time.)
199  */
200 typedef struct MultiXactStateData
201 {
202  /* next-to-be-assigned MultiXactId */
204 
205  /* next-to-be-assigned offset */
207 
208  /* Have we completed multixact startup? */
210 
211  /*
212  * Oldest multixact that is still potentially referenced by a relation.
213  * Anything older than this should not be consulted. These values are
214  * updated by vacuum.
215  */
218 
219  /*
220  * Oldest multixact offset that is potentially referenced by a multixact
221  * referenced by a relation. We don't always know this value, so there's
222  * a flag here to indicate whether or not we currently do.
223  */
226 
227  /* support for anti-wraparound measures */
232 
233  /* support for members anti-wraparound measures */
234  MultiXactOffset offsetStopLimit; /* known if oldestOffsetKnown */
235 
236  /*
237  * Per-backend data starts here. We have two arrays stored in the area
238  * immediately following the MultiXactStateData struct. Each is indexed by
239  * BackendId.
240  *
241  * In both arrays, there's a slot for all normal backends (1..MaxBackends)
242  * followed by a slot for max_prepared_xacts prepared transactions. Valid
243  * BackendIds start from 1; element zero of each array is never used.
244  *
245  * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
246  * transaction(s) could possibly be a member of, or InvalidMultiXactId
247  * when the backend has no live transaction that could possibly be a
248  * member of a MultiXact. Each backend sets its entry to the current
249  * nextMXact counter just before first acquiring a shared lock in a given
250  * transaction, and clears it at transaction end. (This works because only
251  * during or after acquiring a shared lock could an XID possibly become a
252  * member of a MultiXact, and that MultiXact would have to be created
253  * during or after the lock acquisition.)
254  *
255  * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
256  * current transaction(s) think is potentially live, or InvalidMultiXactId
257  * when not in a transaction or not in a transaction that's paid any
258  * attention to MultiXacts yet. This is computed when first needed in a
259  * given transaction, and cleared at transaction end. We can compute it
260  * as the minimum of the valid OldestMemberMXactId[] entries at the time
261  * we compute it (using nextMXact if none are valid). Each backend is
262  * required not to attempt to access any SLRU data for MultiXactIds older
263  * than its own OldestVisibleMXactId[] setting; this is necessary because
264  * the checkpointer could truncate away such data at any instant.
265  *
266  * The oldest valid value among all of the OldestMemberMXactId[] and
267  * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
268  * possible value still having any live member transaction. Subtracting
269  * vacuum_multixact_freeze_min_age from that value we obtain the freezing
270  * point for multixacts for that table. Any value older than that is
271  * removed from tuple headers (or "frozen"; see FreezeMultiXactId. Note
272  * that multis that have member xids that are older than the cutoff point
273  * for xids must also be frozen, even if the multis themselves are newer
274  * than the multixid cutoff point). Whenever a full table vacuum happens,
275  * the freezing point so computed is used as the new pg_class.relminmxid
276  * value. The minimum of all those values in a database is stored as
277  * pg_database.datminmxid. In turn, the minimum of all of those values is
278  * stored in pg_control and used as truncation point for pg_multixact. At
279  * checkpoint or restartpoint, unneeded segments are removed.
280  */
283 
284 /*
285  * Last element of OldestMemberMXactId and OldestVisibleMXactId arrays.
286  * Valid elements are (1..MaxOldestSlot); element 0 is never used.
287  */
288 #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
289 
290 /* Pointers to the state data in shared memory */
294 
295 
296 /*
297  * Definitions for the backend-local MultiXactId cache.
298  *
299  * We use this cache to store known MultiXacts, so we don't need to go to
300  * SLRU areas every time.
301  *
302  * The cache lasts for the duration of a single transaction, the rationale
303  * for this being that most entries will contain our own TransactionId and
304  * so they will be uninteresting by the time our next transaction starts.
305  * (XXX not clear that this is correct --- other members of the MultiXact
306  * could hang around longer than we did. However, it's not clear what a
307  * better policy for flushing old cache entries would be.) FIXME actually
308  * this is plain wrong now that multixact's may contain update Xids.
309  *
310  * We allocate the cache entries in a memory context that is deleted at
311  * transaction end, so we don't need to do retail freeing of entries.
312  */
313 typedef struct mXactCacheEnt
314 {
316  int nmembers;
320 
321 #define MAX_CACHE_ENTRIES 256
324 
325 #ifdef MULTIXACT_DEBUG
326 #define debug_elog2(a,b) elog(a,b)
327 #define debug_elog3(a,b,c) elog(a,b,c)
328 #define debug_elog4(a,b,c,d) elog(a,b,c,d)
329 #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
330 #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
331 #else
332 #define debug_elog2(a,b)
333 #define debug_elog3(a,b,c)
334 #define debug_elog4(a,b,c,d)
335 #define debug_elog5(a,b,c,d,e)
336 #define debug_elog6(a,b,c,d,e,f)
337 #endif
338 
339 /* internal MultiXactId management */
340 static void MultiXactIdSetOldestVisible(void);
341 static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
342  int nmembers, MultiXactMember *members);
343 static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
344 
345 /* MultiXact cache management */
346 static int mxactMemberComparator(const void *arg1, const void *arg2);
347 static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
348 static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
349 static void mXactCachePut(MultiXactId multi, int nmembers,
350  MultiXactMember *members);
351 
353 
354 /* management of SLRU infrastructure */
355 static int ZeroMultiXactOffsetPage(int pageno, bool writeXlog);
356 static int ZeroMultiXactMemberPage(int pageno, bool writeXlog);
357 static bool MultiXactOffsetPagePrecedes(int page1, int page2);
358 static bool MultiXactMemberPagePrecedes(int page1, int page2);
359 static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
360  MultiXactOffset offset2);
361 static void ExtendMultiXactOffset(MultiXactId multi);
362 static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
363 static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
364  MultiXactOffset start, uint32 distance);
365 static bool SetOffsetVacuumLimit(bool is_startup);
366 static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
367 static void WriteMZeroPageXlogRec(int pageno, uint8 info);
368 static void WriteMTruncateXlogRec(Oid oldestMultiDB,
369  MultiXactId startTruncOff,
370  MultiXactId endTruncOff,
371  MultiXactOffset startTruncMemb,
372  MultiXactOffset endTruncMemb);
373 
374 
375 /*
376  * MultiXactIdCreate
377  * Construct a MultiXactId representing two TransactionIds.
378  *
379  * The two XIDs must be different, or be requesting different statuses.
380  *
381  * NB - we don't worry about our local MultiXactId cache here, because that
382  * is handled by the lower-level routines.
383  */
386  TransactionId xid2, MultiXactStatus status2)
387 {
388  MultiXactId newMulti;
389  MultiXactMember members[2];
390 
393 
394  Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
395 
396  /* MultiXactIdSetOldestMember() must have been called already. */
398 
399  /*
400  * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
401  * are still running. In typical usage, xid2 will be our own XID and the
402  * caller just did a check on xid1, so it'd be wasted effort.
403  */
404 
405  members[0].xid = xid1;
406  members[0].status = status1;
407  members[1].xid = xid2;
408  members[1].status = status2;
409 
410  newMulti = MultiXactIdCreateFromMembers(2, members);
411 
412  debug_elog3(DEBUG2, "Create: %s",
413  mxid_to_string(newMulti, 2, members));
414 
415  return newMulti;
416 }
417 
418 /*
419  * MultiXactIdExpand
420  * Add a TransactionId to a pre-existing MultiXactId.
421  *
422  * If the TransactionId is already a member of the passed MultiXactId with the
423  * same status, just return it as-is.
424  *
425  * Note that we do NOT actually modify the membership of a pre-existing
426  * MultiXactId; instead we create a new one. This is necessary to avoid
427  * a race condition against code trying to wait for one MultiXactId to finish;
428  * see notes in heapam.c.
429  *
430  * NB - we don't worry about our local MultiXactId cache here, because that
431  * is handled by the lower-level routines.
432  *
433  * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
434  * one upgraded by pg_upgrade from a cluster older than this feature) are not
435  * passed in.
436  */
439 {
440  MultiXactId newMulti;
441  MultiXactMember *members;
442  MultiXactMember *newMembers;
443  int nmembers;
444  int i;
445  int j;
446 
447  Assert(MultiXactIdIsValid(multi));
449 
450  /* MultiXactIdSetOldestMember() must have been called already. */
452 
453  debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
454  multi, xid, mxstatus_to_string(status));
455 
456  /*
457  * Note: we don't allow for old multis here. The reason is that the only
458  * caller of this function does a check that the multixact is no longer
459  * running.
460  */
461  nmembers = GetMultiXactIdMembers(multi, &members, false, false);
462 
463  if (nmembers < 0)
464  {
465  MultiXactMember member;
466 
467  /*
468  * The MultiXactId is obsolete. This can only happen if all the
469  * MultiXactId members stop running between the caller checking and
470  * passing it to us. It would be better to return that fact to the
471  * caller, but it would complicate the API and it's unlikely to happen
472  * too often, so just deal with it by creating a singleton MultiXact.
473  */
474  member.xid = xid;
475  member.status = status;
476  newMulti = MultiXactIdCreateFromMembers(1, &member);
477 
478  debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
479  multi, newMulti);
480  return newMulti;
481  }
482 
483  /*
484  * If the TransactionId is already a member of the MultiXactId with the
485  * same status, just return the existing MultiXactId.
486  */
487  for (i = 0; i < nmembers; i++)
488  {
489  if (TransactionIdEquals(members[i].xid, xid) &&
490  (members[i].status == status))
491  {
492  debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
493  xid, multi);
494  pfree(members);
495  return multi;
496  }
497  }
498 
499  /*
500  * Determine which of the members of the MultiXactId are still of
501  * interest. This is any running transaction, and also any transaction
502  * that grabbed something stronger than just a lock and was committed. (An
503  * update that aborted is of no interest here; and having more than one
504  * update Xid in a multixact would cause errors elsewhere.)
505  *
506  * Removing dead members is not just an optimization: freezing of tuples
507  * whose Xmax are multis depends on this behavior.
508  *
509  * Note we have the same race condition here as above: j could be 0 at the
510  * end of the loop.
511  */
512  newMembers = (MultiXactMember *)
513  palloc(sizeof(MultiXactMember) * (nmembers + 1));
514 
515  for (i = 0, j = 0; i < nmembers; i++)
516  {
517  if (TransactionIdIsInProgress(members[i].xid) ||
518  (ISUPDATE_from_mxstatus(members[i].status) &&
519  TransactionIdDidCommit(members[i].xid)))
520  {
521  newMembers[j].xid = members[i].xid;
522  newMembers[j++].status = members[i].status;
523  }
524  }
525 
526  newMembers[j].xid = xid;
527  newMembers[j++].status = status;
528  newMulti = MultiXactIdCreateFromMembers(j, newMembers);
529 
530  pfree(members);
531  pfree(newMembers);
532 
533  debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
534 
535  return newMulti;
536 }
537 
538 /*
539  * MultiXactIdIsRunning
540  * Returns whether a MultiXactId is "running".
541  *
542  * We return true if at least one member of the given MultiXactId is still
543  * running. Note that a "false" result is certain not to change,
544  * because it is not legal to add members to an existing MultiXactId.
545  *
546  * Caller is expected to have verified that the multixact does not come from
547  * a pg_upgraded share-locked tuple.
548  */
549 bool
550 MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
551 {
552  MultiXactMember *members;
553  int nmembers;
554  int i;
555 
556  debug_elog3(DEBUG2, "IsRunning %u?", multi);
557 
558  /*
559  * "false" here means we assume our callers have checked that the given
560  * multi cannot possibly come from a pg_upgraded database.
561  */
562  nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
563 
564  if (nmembers <= 0)
565  {
566  debug_elog2(DEBUG2, "IsRunning: no members");
567  return false;
568  }
569 
570  /*
571  * Checking for myself is cheap compared to looking in shared memory;
572  * return true if any live subtransaction of the current top-level
573  * transaction is a member.
574  *
575  * This is not needed for correctness, it's just a fast path.
576  */
577  for (i = 0; i < nmembers; i++)
578  {
579  if (TransactionIdIsCurrentTransactionId(members[i].xid))
580  {
581  debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
582  pfree(members);
583  return true;
584  }
585  }
586 
587  /*
588  * This could be made faster by having another entry point in procarray.c,
589  * walking the PGPROC array only once for all the members. But in most
590  * cases nmembers should be small enough that it doesn't much matter.
591  */
592  for (i = 0; i < nmembers; i++)
593  {
594  if (TransactionIdIsInProgress(members[i].xid))
595  {
596  debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
597  i, members[i].xid);
598  pfree(members);
599  return true;
600  }
601  }
602 
603  pfree(members);
604 
605  debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
606 
607  return false;
608 }
609 
610 /*
611  * MultiXactIdSetOldestMember
612  * Save the oldest MultiXactId this transaction could be a member of.
613  *
614  * We set the OldestMemberMXactId for a given transaction the first time it's
615  * going to do some operation that might require a MultiXactId (tuple lock,
616  * update or delete). We need to do this even if we end up using a
617  * TransactionId instead of a MultiXactId, because there is a chance that
618  * another transaction would add our XID to a MultiXactId.
619  *
620  * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
621  * be called just before doing any such possibly-MultiXactId-able operation.
622  */
623 void
625 {
627  {
628  MultiXactId nextMXact;
629 
630  /*
631  * You might think we don't need to acquire a lock here, since
632  * fetching and storing of TransactionIds is probably atomic, but in
633  * fact we do: suppose we pick up nextMXact and then lose the CPU for
634  * a long time. Someone else could advance nextMXact, and then
635  * another someone else could compute an OldestVisibleMXactId that
636  * would be after the value we are going to store when we get control
637  * back. Which would be wrong.
638  *
639  * Note that a shared lock is sufficient, because it's enough to stop
640  * someone from advancing nextMXact; and nobody else could be trying
641  * to write to our OldestMember entry, only reading (and we assume
642  * storing it is atomic.)
643  */
644  LWLockAcquire(MultiXactGenLock, LW_SHARED);
645 
646  /*
647  * We have to beware of the possibility that nextMXact is in the
648  * wrapped-around state. We don't fix the counter itself here, but we
649  * must be sure to store a valid value in our array entry.
650  */
651  nextMXact = MultiXactState->nextMXact;
652  if (nextMXact < FirstMultiXactId)
653  nextMXact = FirstMultiXactId;
654 
655  OldestMemberMXactId[MyBackendId] = nextMXact;
656 
657  LWLockRelease(MultiXactGenLock);
658 
659  debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
660  MyBackendId, nextMXact);
661  }
662 }
663 
664 /*
665  * MultiXactIdSetOldestVisible
666  * Save the oldest MultiXactId this transaction considers possibly live.
667  *
668  * We set the OldestVisibleMXactId for a given transaction the first time
669  * it's going to inspect any MultiXactId. Once we have set this, we are
670  * guaranteed that the checkpointer won't truncate off SLRU data for
671  * MultiXactIds at or after our OldestVisibleMXactId.
672  *
673  * The value to set is the oldest of nextMXact and all the valid per-backend
674  * OldestMemberMXactId[] entries. Because of the locking we do, we can be
675  * certain that no subsequent call to MultiXactIdSetOldestMember can set
676  * an OldestMemberMXactId[] entry older than what we compute here. Therefore
677  * there is no live transaction, now or later, that can be a member of any
678  * MultiXactId older than the OldestVisibleMXactId we compute here.
679  */
680 static void
682 {
684  {
685  MultiXactId oldestMXact;
686  int i;
687 
688  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
689 
690  /*
691  * We have to beware of the possibility that nextMXact is in the
692  * wrapped-around state. We don't fix the counter itself here, but we
693  * must be sure to store a valid value in our array entry.
694  */
695  oldestMXact = MultiXactState->nextMXact;
696  if (oldestMXact < FirstMultiXactId)
697  oldestMXact = FirstMultiXactId;
698 
699  for (i = 1; i <= MaxOldestSlot; i++)
700  {
701  MultiXactId thisoldest = OldestMemberMXactId[i];
702 
703  if (MultiXactIdIsValid(thisoldest) &&
704  MultiXactIdPrecedes(thisoldest, oldestMXact))
705  oldestMXact = thisoldest;
706  }
707 
708  OldestVisibleMXactId[MyBackendId] = oldestMXact;
709 
710  LWLockRelease(MultiXactGenLock);
711 
712  debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
713  MyBackendId, oldestMXact);
714  }
715 }
716 
717 /*
718  * ReadNextMultiXactId
719  * Return the next MultiXactId to be assigned, but don't allocate it
720  */
723 {
724  MultiXactId mxid;
725 
726  /* XXX we could presumably do this without a lock. */
727  LWLockAcquire(MultiXactGenLock, LW_SHARED);
728  mxid = MultiXactState->nextMXact;
729  LWLockRelease(MultiXactGenLock);
730 
731  if (mxid < FirstMultiXactId)
732  mxid = FirstMultiXactId;
733 
734  return mxid;
735 }
736 
737 /*
738  * ReadMultiXactIdRange
739  * Get the range of IDs that may still be referenced by a relation.
740  */
741 void
743 {
744  LWLockAcquire(MultiXactGenLock, LW_SHARED);
747  LWLockRelease(MultiXactGenLock);
748 
749  if (*oldest < FirstMultiXactId)
750  *oldest = FirstMultiXactId;
751  if (*next < FirstMultiXactId)
753 }
754 
755 
756 /*
757  * MultiXactIdCreateFromMembers
758  * Make a new MultiXactId from the specified set of members
759  *
760  * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
761  * given TransactionIds as members. Returns the newly created MultiXactId.
762  *
763  * NB: the passed members[] array will be sorted in-place.
764  */
767 {
768  MultiXactId multi;
769  MultiXactOffset offset;
770  xl_multixact_create xlrec;
771 
772  debug_elog3(DEBUG2, "Create: %s",
773  mxid_to_string(InvalidMultiXactId, nmembers, members));
774 
775  /*
776  * See if the same set of members already exists in our cache; if so, just
777  * re-use that MultiXactId. (Note: it might seem that looking in our
778  * cache is insufficient, and we ought to search disk to see if a
779  * duplicate definition already exists. But since we only ever create
780  * MultiXacts containing our own XID, in most cases any such MultiXacts
781  * were in fact created by us, and so will be in our cache. There are
782  * corner cases where someone else added us to a MultiXact without our
783  * knowledge, but it's not worth checking for.)
784  */
785  multi = mXactCacheGetBySet(nmembers, members);
786  if (MultiXactIdIsValid(multi))
787  {
788  debug_elog2(DEBUG2, "Create: in cache!");
789  return multi;
790  }
791 
792  /* Verify that there is a single update Xid among the given members. */
793  {
794  int i;
795  bool has_update = false;
796 
797  for (i = 0; i < nmembers; i++)
798  {
799  if (ISUPDATE_from_mxstatus(members[i].status))
800  {
801  if (has_update)
802  elog(ERROR, "new multixact has more than one updating member: %s",
803  mxid_to_string(InvalidMultiXactId, nmembers, members));
804  has_update = true;
805  }
806  }
807  }
808 
809  /*
810  * Assign the MXID and offsets range to use, and make sure there is space
811  * in the OFFSETs and MEMBERs files. NB: this routine does
812  * START_CRIT_SECTION().
813  *
814  * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
815  * that we've called MultiXactIdSetOldestMember here. This is because
816  * this routine is used in some places to create new MultiXactIds of which
817  * the current backend is not a member, notably during freezing of multis
818  * in vacuum. During vacuum, in particular, it would be unacceptable to
819  * keep OldestMulti set, in case it runs for long.
820  */
821  multi = GetNewMultiXactId(nmembers, &offset);
822 
823  /* Make an XLOG entry describing the new MXID. */
824  xlrec.mid = multi;
825  xlrec.moff = offset;
826  xlrec.nmembers = nmembers;
827 
828  /*
829  * XXX Note: there's a lot of padding space in MultiXactMember. We could
830  * find a more compact representation of this Xlog record -- perhaps all
831  * the status flags in one XLogRecData, then all the xids in another one?
832  * Not clear that it's worth the trouble though.
833  */
834  XLogBeginInsert();
835  XLogRegisterData((char *) (&xlrec), SizeOfMultiXactCreate);
836  XLogRegisterData((char *) members, nmembers * sizeof(MultiXactMember));
837 
838  (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
839 
840  /* Now enter the information into the OFFSETs and MEMBERs logs */
841  RecordNewMultiXact(multi, offset, nmembers, members);
842 
843  /* Done with critical section */
845 
846  /* Store the new MultiXactId in the local cache, too */
847  mXactCachePut(multi, nmembers, members);
848 
849  debug_elog2(DEBUG2, "Create: all done");
850 
851  return multi;
852 }
853 
854 /*
855  * RecordNewMultiXact
856  * Write info about a new multixact into the offsets and members files
857  *
858  * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
859  * use it.
860  */
861 static void
863  int nmembers, MultiXactMember *members)
864 {
865  int pageno;
866  int prev_pageno;
867  int entryno;
868  int slotno;
869  MultiXactOffset *offptr;
870  int i;
871 
872  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
873 
874  pageno = MultiXactIdToOffsetPage(multi);
875  entryno = MultiXactIdToOffsetEntry(multi);
876 
877  /*
878  * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
879  * to complain about if there's any I/O error. This is kinda bogus, but
880  * since the errors will always give the full pathname, it should be clear
881  * enough that a MultiXactId is really involved. Perhaps someday we'll
882  * take the trouble to generalize the slru.c error reporting code.
883  */
884  slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
885  offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
886  offptr += entryno;
887 
888  *offptr = offset;
889 
890  MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
891 
892  /* Exchange our lock */
893  LWLockRelease(MultiXactOffsetSLRULock);
894 
895  LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
896 
897  prev_pageno = -1;
898 
899  for (i = 0; i < nmembers; i++, offset++)
900  {
901  TransactionId *memberptr;
902  uint32 *flagsptr;
903  uint32 flagsval;
904  int bshift;
905  int flagsoff;
906  int memberoff;
907 
908  Assert(members[i].status <= MultiXactStatusUpdate);
909 
910  pageno = MXOffsetToMemberPage(offset);
911  memberoff = MXOffsetToMemberOffset(offset);
912  flagsoff = MXOffsetToFlagsOffset(offset);
913  bshift = MXOffsetToFlagsBitShift(offset);
914 
915  if (pageno != prev_pageno)
916  {
917  slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
918  prev_pageno = pageno;
919  }
920 
921  memberptr = (TransactionId *)
922  (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
923 
924  *memberptr = members[i].xid;
925 
926  flagsptr = (uint32 *)
927  (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
928 
929  flagsval = *flagsptr;
930  flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
931  flagsval |= (members[i].status << bshift);
932  *flagsptr = flagsval;
933 
934  MultiXactMemberCtl->shared->page_dirty[slotno] = true;
935  }
936 
937  LWLockRelease(MultiXactMemberSLRULock);
938 }
939 
940 /*
941  * GetNewMultiXactId
942  * Get the next MultiXactId.
943  *
944  * Also, reserve the needed amount of space in the "members" area. The
945  * starting offset of the reserved space is returned in *offset.
946  *
947  * This may generate XLOG records for expansion of the offsets and/or members
948  * files. Unfortunately, we have to do that while holding MultiXactGenLock
949  * to avoid race conditions --- the XLOG record for zeroing a page must appear
950  * before any backend can possibly try to store data in that page!
951  *
952  * We start a critical section before advancing the shared counters. The
953  * caller must end the critical section after writing SLRU data.
954  */
955 static MultiXactId
956 GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
957 {
958  MultiXactId result;
959  MultiXactOffset nextOffset;
960 
961  debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
962 
963  /* safety check, we should never get this far in a HS standby */
964  if (RecoveryInProgress())
965  elog(ERROR, "cannot assign MultiXactIds during recovery");
966 
967  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
968 
969  /* Handle wraparound of the nextMXact counter */
972 
973  /* Assign the MXID */
974  result = MultiXactState->nextMXact;
975 
976  /*----------
977  * Check to see if it's safe to assign another MultiXactId. This protects
978  * against catastrophic data loss due to multixact wraparound. The basic
979  * rules are:
980  *
981  * If we're past multiVacLimit or the safe threshold for member storage
982  * space, or we don't know what the safe threshold for member storage is,
983  * start trying to force autovacuum cycles.
984  * If we're past multiWarnLimit, start issuing warnings.
985  * If we're past multiStopLimit, refuse to create new MultiXactIds.
986  *
987  * Note these are pretty much the same protections in GetNewTransactionId.
988  *----------
989  */
991  {
992  /*
993  * For safety's sake, we release MultiXactGenLock while sending
994  * signals, warnings, etc. This is not so much because we care about
995  * preserving concurrency in this situation, as to avoid any
996  * possibility of deadlock while doing get_database_name(). First,
997  * copy all the shared values we'll need in this path.
998  */
999  MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
1000  MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
1001  MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
1002  Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
1003 
1004  LWLockRelease(MultiXactGenLock);
1005 
1006  if (IsUnderPostmaster &&
1007  !MultiXactIdPrecedes(result, multiStopLimit))
1008  {
1009  char *oldest_datname = get_database_name(oldest_datoid);
1010 
1011  /*
1012  * Immediately kick autovacuum into action as we're already in
1013  * ERROR territory.
1014  */
1016 
1017  /* complain even if that DB has disappeared */
1018  if (oldest_datname)
1019  ereport(ERROR,
1020  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1021  errmsg("database is not accepting commands that generate new MultiXactIds to avoid wraparound data loss in database \"%s\"",
1022  oldest_datname),
1023  errhint("Execute a database-wide VACUUM in that database.\n"
1024  "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1025  else
1026  ereport(ERROR,
1027  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1028  errmsg("database is not accepting commands that generate new MultiXactIds to avoid wraparound data loss in database with OID %u",
1029  oldest_datoid),
1030  errhint("Execute a database-wide VACUUM in that database.\n"
1031  "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1032  }
1033 
1034  /*
1035  * To avoid swamping the postmaster with signals, we issue the autovac
1036  * request only once per 64K multis generated. This still gives
1037  * plenty of chances before we get into real trouble.
1038  */
1039  if (IsUnderPostmaster && (result % 65536) == 0)
1041 
1042  if (!MultiXactIdPrecedes(result, multiWarnLimit))
1043  {
1044  char *oldest_datname = get_database_name(oldest_datoid);
1045 
1046  /* complain even if that DB has disappeared */
1047  if (oldest_datname)
1048  ereport(WARNING,
1049  (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1050  "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1051  multiWrapLimit - result,
1052  oldest_datname,
1053  multiWrapLimit - result),
1054  errhint("Execute a database-wide VACUUM in that database.\n"
1055  "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1056  else
1057  ereport(WARNING,
1058  (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1059  "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1060  multiWrapLimit - result,
1061  oldest_datoid,
1062  multiWrapLimit - result),
1063  errhint("Execute a database-wide VACUUM in that database.\n"
1064  "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1065  }
1066 
1067  /* Re-acquire lock and start over */
1068  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1069  result = MultiXactState->nextMXact;
1070  if (result < FirstMultiXactId)
1071  result = FirstMultiXactId;
1072  }
1073 
1074  /* Make sure there is room for the MXID in the file. */
1075  ExtendMultiXactOffset(result);
1076 
1077  /*
1078  * Reserve the members space, similarly to above. Also, be careful not to
1079  * return zero as the starting offset for any multixact. See
1080  * GetMultiXactIdMembers() for motivation.
1081  */
1082  nextOffset = MultiXactState->nextOffset;
1083  if (nextOffset == 0)
1084  {
1085  *offset = 1;
1086  nmembers++; /* allocate member slot 0 too */
1087  }
1088  else
1089  *offset = nextOffset;
1090 
1091  /*----------
1092  * Protect against overrun of the members space as well, with the
1093  * following rules:
1094  *
1095  * If we're past offsetStopLimit, refuse to generate more multis.
1096  * If we're close to offsetStopLimit, emit a warning.
1097  *
1098  * Arbitrarily, we start emitting warnings when we're 20 segments or less
1099  * from offsetStopLimit.
1100  *
1101  * Note we haven't updated the shared state yet, so if we fail at this
1102  * point, the multixact ID we grabbed can still be used by the next guy.
1103  *
1104  * Note that there is no point in forcing autovacuum runs here: the
1105  * multixact freeze settings would have to be reduced for that to have any
1106  * effect.
1107  *----------
1108  */
1109 #define OFFSET_WARN_SEGMENTS 20
1112  nmembers))
1113  {
1114  /* see comment in the corresponding offsets wraparound case */
1116 
1117  ereport(ERROR,
1118  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1119  errmsg("multixact \"members\" limit exceeded"),
1120  errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
1121  "This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
1122  MultiXactState->offsetStopLimit - nextOffset - 1,
1123  nmembers,
1124  MultiXactState->offsetStopLimit - nextOffset - 1),
1125  errhint("Execute a database-wide VACUUM in database with OID %u with reduced vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age settings.",
1127  }
1128 
1129  /*
1130  * Check whether we should kick autovacuum into action, to prevent members
1131  * wraparound. NB we use a much larger window to trigger autovacuum than
1132  * just the warning limit. The warning is just a measure of last resort -
1133  * this is in line with GetNewTransactionId's behaviour.
1134  */
1138  {
1139  /*
1140  * To avoid swamping the postmaster with signals, we issue the autovac
1141  * request only when crossing a segment boundary. With default
1142  * compilation settings that's roughly after 50k members. This still
1143  * gives plenty of chances before we get into real trouble.
1144  */
1145  if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
1146  (MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
1148  }
1149 
1152  nextOffset,
1154  ereport(WARNING,
1155  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1156  errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
1157  "database with OID %u must be vacuumed before %d more multixact members are used",
1158  MultiXactState->offsetStopLimit - nextOffset + nmembers,
1160  MultiXactState->offsetStopLimit - nextOffset + nmembers),
1161  errhint("Execute a database-wide VACUUM in that database with reduced vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age settings.")));
1162 
1163  ExtendMultiXactMember(nextOffset, nmembers);
1164 
1165  /*
1166  * Critical section from here until caller has written the data into the
1167  * just-reserved SLRU space; we don't want to error out with a partly
1168  * written MultiXact structure. (In particular, failing to write our
1169  * start offset after advancing nextMXact would effectively corrupt the
1170  * previous MultiXact.)
1171  */
1173 
1174  /*
1175  * Advance counters. As in GetNewTransactionId(), this must not happen
1176  * until after file extension has succeeded!
1177  *
1178  * We don't care about MultiXactId wraparound here; it will be handled by
1179  * the next iteration. But note that nextMXact may be InvalidMultiXactId
1180  * or the first value on a segment-beginning page after this routine
1181  * exits, so anyone else looking at the variable must be prepared to deal
1182  * with either case. Similarly, nextOffset may be zero, but we won't use
1183  * that as the actual start offset of the next multixact.
1184  */
1186 
1187  MultiXactState->nextOffset += nmembers;
1188 
1189  LWLockRelease(MultiXactGenLock);
1190 
1191  debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
1192  return result;
1193 }
1194 
1195 /*
1196  * GetMultiXactIdMembers
1197  * Return the set of MultiXactMembers that make up a MultiXactId
1198  *
1199  * Return value is the number of members found, or -1 if there are none,
1200  * and *members is set to a newly palloc'ed array of members. It's the
1201  * caller's responsibility to free it when done with it.
1202  *
1203  * from_pgupgrade must be passed as true if and only if only the multixact
1204  * corresponds to a value from a tuple that was locked in a 9.2-or-older
1205  * installation and later pg_upgrade'd (that is, the infomask is
1206  * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1207  * can still be running, so we return -1 just like for an empty multixact
1208  * without any further checking. It would be wrong to try to resolve such a
1209  * multixact: either the multixact is within the current valid multixact
1210  * range, in which case the returned result would be bogus, or outside that
1211  * range, in which case an error would be raised.
1212  *
1213  * In all other cases, the passed multixact must be within the known valid
1214  * range, that is, greater to or equal than oldestMultiXactId, and less than
1215  * nextMXact. Otherwise, an error is raised.
1216  *
1217  * isLockOnly must be set to true if caller is certain that the given multi
1218  * is used only to lock tuples; can be false without loss of correctness,
1219  * but passing a true means we can return quickly without checking for
1220  * old updates.
1221  */
1222 int
1224  bool from_pgupgrade, bool isLockOnly)
1225 {
1226  int pageno;
1227  int prev_pageno;
1228  int entryno;
1229  int slotno;
1230  MultiXactOffset *offptr;
1231  MultiXactOffset offset;
1232  int length;
1233  int truelength;
1234  int i;
1235  MultiXactId oldestMXact;
1236  MultiXactId nextMXact;
1237  MultiXactId tmpMXact;
1238  MultiXactOffset nextOffset;
1239  MultiXactMember *ptr;
1240 
1241  debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1242 
1243  if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1244  {
1245  *members = NULL;
1246  return -1;
1247  }
1248 
1249  /* See if the MultiXactId is in the local cache */
1250  length = mXactCacheGetById(multi, members);
1251  if (length >= 0)
1252  {
1253  debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1254  mxid_to_string(multi, length, *members));
1255  return length;
1256  }
1257 
1258  /* Set our OldestVisibleMXactId[] entry if we didn't already */
1260 
1261  /*
1262  * If we know the multi is used only for locking and not for updates, then
1263  * we can skip checking if the value is older than our oldest visible
1264  * multi. It cannot possibly still be running.
1265  */
1266  if (isLockOnly &&
1268  {
1269  debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1270  *members = NULL;
1271  return -1;
1272  }
1273 
1274  /*
1275  * We check known limits on MultiXact before resorting to the SLRU area.
1276  *
1277  * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1278  * useful; it has already been removed, or will be removed shortly, by
1279  * truncation. If one is passed, an error is raised.
1280  *
1281  * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1282  * implies undetected ID wraparound has occurred. This raises a hard
1283  * error.
1284  *
1285  * Shared lock is enough here since we aren't modifying any global state.
1286  * Acquire it just long enough to grab the current counter values. We may
1287  * need both nextMXact and nextOffset; see below.
1288  */
1289  LWLockAcquire(MultiXactGenLock, LW_SHARED);
1290 
1291  oldestMXact = MultiXactState->oldestMultiXactId;
1292  nextMXact = MultiXactState->nextMXact;
1293  nextOffset = MultiXactState->nextOffset;
1294 
1295  LWLockRelease(MultiXactGenLock);
1296 
1297  if (MultiXactIdPrecedes(multi, oldestMXact))
1298  ereport(ERROR,
1299  (errcode(ERRCODE_INTERNAL_ERROR),
1300  errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1301  multi)));
1302 
1303  if (!MultiXactIdPrecedes(multi, nextMXact))
1304  ereport(ERROR,
1305  (errcode(ERRCODE_INTERNAL_ERROR),
1306  errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1307  multi)));
1308 
1309  /*
1310  * Find out the offset at which we need to start reading MultiXactMembers
1311  * and the number of members in the multixact. We determine the latter as
1312  * the difference between this multixact's starting offset and the next
1313  * one's. However, there are some corner cases to worry about:
1314  *
1315  * 1. This multixact may be the latest one created, in which case there is
1316  * no next one to look at. In this case the nextOffset value we just
1317  * saved is the correct endpoint.
1318  *
1319  * 2. The next multixact may still be in process of being filled in: that
1320  * is, another process may have done GetNewMultiXactId but not yet written
1321  * the offset entry for that ID. In that scenario, it is guaranteed that
1322  * the offset entry for that multixact exists (because GetNewMultiXactId
1323  * won't release MultiXactGenLock until it does) but contains zero
1324  * (because we are careful to pre-zero offset pages). Because
1325  * GetNewMultiXactId will never return zero as the starting offset for a
1326  * multixact, when we read zero as the next multixact's offset, we know we
1327  * have this case. We sleep for a bit and try again.
1328  *
1329  * 3. Because GetNewMultiXactId increments offset zero to offset one to
1330  * handle case #2, there is an ambiguity near the point of offset
1331  * wraparound. If we see next multixact's offset is one, is that our
1332  * multixact's actual endpoint, or did it end at zero with a subsequent
1333  * increment? We handle this using the knowledge that if the zero'th
1334  * member slot wasn't filled, it'll contain zero, and zero isn't a valid
1335  * transaction ID so it can't be a multixact member. Therefore, if we
1336  * read a zero from the members array, just ignore it.
1337  *
1338  * This is all pretty messy, but the mess occurs only in infrequent corner
1339  * cases, so it seems better than holding the MultiXactGenLock for a long
1340  * time on every multixact creation.
1341  */
1342 retry:
1343  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
1344 
1345  pageno = MultiXactIdToOffsetPage(multi);
1346  entryno = MultiXactIdToOffsetEntry(multi);
1347 
1348  slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1349  offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1350  offptr += entryno;
1351  offset = *offptr;
1352 
1353  Assert(offset != 0);
1354 
1355  /*
1356  * Use the same increment rule as GetNewMultiXactId(), that is, don't
1357  * handle wraparound explicitly until needed.
1358  */
1359  tmpMXact = multi + 1;
1360 
1361  if (nextMXact == tmpMXact)
1362  {
1363  /* Corner case 1: there is no next multixact */
1364  length = nextOffset - offset;
1365  }
1366  else
1367  {
1368  MultiXactOffset nextMXOffset;
1369 
1370  /* handle wraparound if needed */
1371  if (tmpMXact < FirstMultiXactId)
1372  tmpMXact = FirstMultiXactId;
1373 
1374  prev_pageno = pageno;
1375 
1376  pageno = MultiXactIdToOffsetPage(tmpMXact);
1377  entryno = MultiXactIdToOffsetEntry(tmpMXact);
1378 
1379  if (pageno != prev_pageno)
1380  slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1381 
1382  offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1383  offptr += entryno;
1384  nextMXOffset = *offptr;
1385 
1386  if (nextMXOffset == 0)
1387  {
1388  /* Corner case 2: next multixact is still being filled in */
1389  LWLockRelease(MultiXactOffsetSLRULock);
1391  pg_usleep(1000L);
1392  goto retry;
1393  }
1394 
1395  length = nextMXOffset - offset;
1396  }
1397 
1398  LWLockRelease(MultiXactOffsetSLRULock);
1399 
1400  ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1401 
1402  /* Now get the members themselves. */
1403  LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
1404 
1405  truelength = 0;
1406  prev_pageno = -1;
1407  for (i = 0; i < length; i++, offset++)
1408  {
1409  TransactionId *xactptr;
1410  uint32 *flagsptr;
1411  int flagsoff;
1412  int bshift;
1413  int memberoff;
1414 
1415  pageno = MXOffsetToMemberPage(offset);
1416  memberoff = MXOffsetToMemberOffset(offset);
1417 
1418  if (pageno != prev_pageno)
1419  {
1420  slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1421  prev_pageno = pageno;
1422  }
1423 
1424  xactptr = (TransactionId *)
1425  (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1426 
1427  if (!TransactionIdIsValid(*xactptr))
1428  {
1429  /* Corner case 3: we must be looking at unused slot zero */
1430  Assert(offset == 0);
1431  continue;
1432  }
1433 
1434  flagsoff = MXOffsetToFlagsOffset(offset);
1435  bshift = MXOffsetToFlagsBitShift(offset);
1436  flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1437 
1438  ptr[truelength].xid = *xactptr;
1439  ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1440  truelength++;
1441  }
1442 
1443  LWLockRelease(MultiXactMemberSLRULock);
1444 
1445  /* A multixid with zero members should not happen */
1446  Assert(truelength > 0);
1447 
1448  /*
1449  * Copy the result into the local cache.
1450  */
1451  mXactCachePut(multi, truelength, ptr);
1452 
1453  debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1454  mxid_to_string(multi, truelength, ptr));
1455  *members = ptr;
1456  return truelength;
1457 }
1458 
1459 /*
1460  * mxactMemberComparator
1461  * qsort comparison function for MultiXactMember
1462  *
1463  * We can't use wraparound comparison for XIDs because that does not respect
1464  * the triangle inequality! Any old sort order will do.
1465  */
1466 static int
1467 mxactMemberComparator(const void *arg1, const void *arg2)
1468 {
1469  MultiXactMember member1 = *(const MultiXactMember *) arg1;
1470  MultiXactMember member2 = *(const MultiXactMember *) arg2;
1471 
1472  if (member1.xid > member2.xid)
1473  return 1;
1474  if (member1.xid < member2.xid)
1475  return -1;
1476  if (member1.status > member2.status)
1477  return 1;
1478  if (member1.status < member2.status)
1479  return -1;
1480  return 0;
1481 }
1482 
1483 /*
1484  * mXactCacheGetBySet
1485  * returns a MultiXactId from the cache based on the set of
1486  * TransactionIds that compose it, or InvalidMultiXactId if
1487  * none matches.
1488  *
1489  * This is helpful, for example, if two transactions want to lock a huge
1490  * table. By using the cache, the second will use the same MultiXactId
1491  * for the majority of tuples, thus keeping MultiXactId usage low (saving
1492  * both I/O and wraparound issues).
1493  *
1494  * NB: the passed members array will be sorted in-place.
1495  */
1496 static MultiXactId
1497 mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1498 {
1499  dlist_iter iter;
1500 
1501  debug_elog3(DEBUG2, "CacheGet: looking for %s",
1502  mxid_to_string(InvalidMultiXactId, nmembers, members));
1503 
1504  /* sort the array so comparison is easy */
1505  qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1506 
1507  dclist_foreach(iter, &MXactCache)
1508  {
1510  iter.cur);
1511 
1512  if (entry->nmembers != nmembers)
1513  continue;
1514 
1515  /*
1516  * We assume the cache entries are sorted, and that the unused bits in
1517  * "status" are zeroed.
1518  */
1519  if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1520  {
1521  debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1523  return entry->multi;
1524  }
1525  }
1526 
1527  debug_elog2(DEBUG2, "CacheGet: not found :-(");
1528  return InvalidMultiXactId;
1529 }
1530 
1531 /*
1532  * mXactCacheGetById
1533  * returns the composing MultiXactMember set from the cache for a
1534  * given MultiXactId, if present.
1535  *
1536  * If successful, *xids is set to the address of a palloc'd copy of the
1537  * MultiXactMember set. Return value is number of members, or -1 on failure.
1538  */
1539 static int
1541 {
1542  dlist_iter iter;
1543 
1544  debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1545 
1546  dclist_foreach(iter, &MXactCache)
1547  {
1549  iter.cur);
1550 
1551  if (entry->multi == multi)
1552  {
1553  MultiXactMember *ptr;
1554  Size size;
1555 
1556  size = sizeof(MultiXactMember) * entry->nmembers;
1557  ptr = (MultiXactMember *) palloc(size);
1558 
1559  memcpy(ptr, entry->members, size);
1560 
1561  debug_elog3(DEBUG2, "CacheGet: found %s",
1562  mxid_to_string(multi,
1563  entry->nmembers,
1564  entry->members));
1565 
1566  /*
1567  * Note we modify the list while not using a modifiable iterator.
1568  * This is acceptable only because we exit the iteration
1569  * immediately afterwards.
1570  */
1572 
1573  *members = ptr;
1574  return entry->nmembers;
1575  }
1576  }
1577 
1578  debug_elog2(DEBUG2, "CacheGet: not found");
1579  return -1;
1580 }
1581 
1582 /*
1583  * mXactCachePut
1584  * Add a new MultiXactId and its composing set into the local cache.
1585  */
1586 static void
1587 mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1588 {
1589  mXactCacheEnt *entry;
1590 
1591  debug_elog3(DEBUG2, "CachePut: storing %s",
1592  mxid_to_string(multi, nmembers, members));
1593 
1594  if (MXactContext == NULL)
1595  {
1596  /* The cache only lives as long as the current transaction */
1597  debug_elog2(DEBUG2, "CachePut: initializing memory context");
1599  "MultiXact cache context",
1601  }
1602 
1603  entry = (mXactCacheEnt *)
1605  offsetof(mXactCacheEnt, members) +
1606  nmembers * sizeof(MultiXactMember));
1607 
1608  entry->multi = multi;
1609  entry->nmembers = nmembers;
1610  memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1611 
1612  /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1613  qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1614 
1615  dclist_push_head(&MXactCache, &entry->node);
1617  {
1618  dlist_node *node;
1619 
1620  node = dclist_tail_node(&MXactCache);
1622 
1623  entry = dclist_container(mXactCacheEnt, node, node);
1624  debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1625  entry->multi);
1626 
1627  pfree(entry);
1628  }
1629 }
1630 
1631 static char *
1633 {
1634  switch (status)
1635  {
1637  return "keysh";
1639  return "sh";
1641  return "fornokeyupd";
1643  return "forupd";
1645  return "nokeyupd";
1646  case MultiXactStatusUpdate:
1647  return "upd";
1648  default:
1649  elog(ERROR, "unrecognized multixact status %d", status);
1650  return "";
1651  }
1652 }
1653 
1654 char *
1655 mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1656 {
1657  static char *str = NULL;
1659  int i;
1660 
1661  if (str != NULL)
1662  pfree(str);
1663 
1664  initStringInfo(&buf);
1665 
1666  appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1667  mxstatus_to_string(members[0].status));
1668 
1669  for (i = 1; i < nmembers; i++)
1670  appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1671  mxstatus_to_string(members[i].status));
1672 
1673  appendStringInfoChar(&buf, ']');
1675  pfree(buf.data);
1676  return str;
1677 }
1678 
1679 /*
1680  * AtEOXact_MultiXact
1681  * Handle transaction end for MultiXact
1682  *
1683  * This is called at top transaction commit or abort (we don't care which).
1684  */
1685 void
1687 {
1688  /*
1689  * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1690  * which should only be valid while within a transaction.
1691  *
1692  * We assume that storing a MultiXactId is atomic and so we need not take
1693  * MultiXactGenLock to do this.
1694  */
1697 
1698  /*
1699  * Discard the local MultiXactId cache. Since MXactContext was created as
1700  * a child of TopTransactionContext, we needn't delete it explicitly.
1701  */
1702  MXactContext = NULL;
1704 }
1705 
1706 /*
1707  * AtPrepare_MultiXact
1708  * Save multixact state at 2PC transaction prepare
1709  *
1710  * In this phase, we only store our OldestMemberMXactId value in the two-phase
1711  * state file.
1712  */
1713 void
1715 {
1716  MultiXactId myOldestMember = OldestMemberMXactId[MyBackendId];
1717 
1718  if (MultiXactIdIsValid(myOldestMember))
1720  &myOldestMember, sizeof(MultiXactId));
1721 }
1722 
1723 /*
1724  * PostPrepare_MultiXact
1725  * Clean up after successful PREPARE TRANSACTION
1726  */
1727 void
1729 {
1730  MultiXactId myOldestMember;
1731 
1732  /*
1733  * Transfer our OldestMemberMXactId value to the slot reserved for the
1734  * prepared transaction.
1735  */
1736  myOldestMember = OldestMemberMXactId[MyBackendId];
1737  if (MultiXactIdIsValid(myOldestMember))
1738  {
1739  BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, false);
1740 
1741  /*
1742  * Even though storing MultiXactId is atomic, acquire lock to make
1743  * sure others see both changes, not just the reset of the slot of the
1744  * current backend. Using a volatile pointer might suffice, but this
1745  * isn't a hot spot.
1746  */
1747  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1748 
1749  OldestMemberMXactId[dummyBackendId] = myOldestMember;
1751 
1752  LWLockRelease(MultiXactGenLock);
1753  }
1754 
1755  /*
1756  * We don't need to transfer OldestVisibleMXactId value, because the
1757  * transaction is not going to be looking at any more multixacts once it's
1758  * prepared.
1759  *
1760  * We assume that storing a MultiXactId is atomic and so we need not take
1761  * MultiXactGenLock to do this.
1762  */
1764 
1765  /*
1766  * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1767  */
1768  MXactContext = NULL;
1770 }
1771 
1772 /*
1773  * multixact_twophase_recover
1774  * Recover the state of a prepared transaction at startup
1775  */
1776 void
1778  void *recdata, uint32 len)
1779 {
1780  BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, false);
1781  MultiXactId oldestMember;
1782 
1783  /*
1784  * Get the oldest member XID from the state file record, and set it in the
1785  * OldestMemberMXactId slot reserved for this prepared transaction.
1786  */
1787  Assert(len == sizeof(MultiXactId));
1788  oldestMember = *((MultiXactId *) recdata);
1789 
1790  OldestMemberMXactId[dummyBackendId] = oldestMember;
1791 }
1792 
1793 /*
1794  * multixact_twophase_postcommit
1795  * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1796  */
1797 void
1799  void *recdata, uint32 len)
1800 {
1801  BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, true);
1802 
1803  Assert(len == sizeof(MultiXactId));
1804 
1805  OldestMemberMXactId[dummyBackendId] = InvalidMultiXactId;
1806 }
1807 
1808 /*
1809  * multixact_twophase_postabort
1810  * This is actually just the same as the COMMIT case.
1811  */
1812 void
1814  void *recdata, uint32 len)
1815 {
1816  multixact_twophase_postcommit(xid, info, recdata, len);
1817 }
1818 
1819 /*
1820  * Initialization of shared memory for MultiXact. We use two SLRU areas,
1821  * thus double memory. Also, reserve space for the shared MultiXactState
1822  * struct and the per-backend MultiXactId arrays (two of those, too).
1823  */
1824 Size
1826 {
1827  Size size;
1828 
1829  /* We need 2*MaxOldestSlot + 1 perBackendXactIds[] entries */
1830 #define SHARED_MULTIXACT_STATE_SIZE \
1831  add_size(offsetof(MultiXactStateData, perBackendXactIds) + sizeof(MultiXactId), \
1832  mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1833 
1837 
1838  return size;
1839 }
1840 
1841 void
1843 {
1844  bool found;
1845 
1846  debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1847 
1850 
1852  "MultiXactOffset", NUM_MULTIXACTOFFSET_BUFFERS, 0,
1853  MultiXactOffsetSLRULock, "pg_multixact/offsets",
1858  "MultiXactMember", NUM_MULTIXACTMEMBER_BUFFERS, 0,
1859  MultiXactMemberSLRULock, "pg_multixact/members",
1862  /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1863 
1864  /* Initialize our shared state struct */
1865  MultiXactState = ShmemInitStruct("Shared MultiXact State",
1867  &found);
1868  if (!IsUnderPostmaster)
1869  {
1870  Assert(!found);
1871 
1872  /* Make sure we zero out the per-backend state */
1874  }
1875  else
1876  Assert(found);
1877 
1878  /*
1879  * Set up array pointers. Note that perBackendXactIds[0] is wasted space
1880  * since we only use indexes 1..MaxOldestSlot in each array.
1881  */
1884 }
1885 
1886 /*
1887  * This func must be called ONCE on system install. It creates the initial
1888  * MultiXact segments. (The MultiXacts directories are assumed to have been
1889  * created by initdb, and MultiXactShmemInit must have been called already.)
1890  */
1891 void
1893 {
1894  int slotno;
1895 
1896  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
1897 
1898  /* Create and zero the first page of the offsets log */
1899  slotno = ZeroMultiXactOffsetPage(0, false);
1900 
1901  /* Make sure it's written out */
1903  Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
1904 
1905  LWLockRelease(MultiXactOffsetSLRULock);
1906 
1907  LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
1908 
1909  /* Create and zero the first page of the members log */
1910  slotno = ZeroMultiXactMemberPage(0, false);
1911 
1912  /* Make sure it's written out */
1914  Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
1915 
1916  LWLockRelease(MultiXactMemberSLRULock);
1917 }
1918 
1919 /*
1920  * Initialize (or reinitialize) a page of MultiXactOffset to zeroes.
1921  * If writeXlog is true, also emit an XLOG record saying we did this.
1922  *
1923  * The page is not actually written, just set up in shared memory.
1924  * The slot number of the new page is returned.
1925  *
1926  * Control lock must be held at entry, and will be held at exit.
1927  */
1928 static int
1929 ZeroMultiXactOffsetPage(int pageno, bool writeXlog)
1930 {
1931  int slotno;
1932 
1933  slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
1934 
1935  if (writeXlog)
1937 
1938  return slotno;
1939 }
1940 
1941 /*
1942  * Ditto, for MultiXactMember
1943  */
1944 static int
1945 ZeroMultiXactMemberPage(int pageno, bool writeXlog)
1946 {
1947  int slotno;
1948 
1949  slotno = SimpleLruZeroPage(MultiXactMemberCtl, pageno);
1950 
1951  if (writeXlog)
1953 
1954  return slotno;
1955 }
1956 
1957 /*
1958  * MaybeExtendOffsetSlru
1959  * Extend the offsets SLRU area, if necessary
1960  *
1961  * After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
1962  * contain files that are shorter than necessary; this would occur if the old
1963  * installation had used multixacts beyond the first page (files cannot be
1964  * copied, because the on-disk representation is different). pg_upgrade would
1965  * update pg_control to set the next offset value to be at that position, so
1966  * that tuples marked as locked by such MultiXacts would be seen as visible
1967  * without having to consult multixact. However, trying to create and use a
1968  * new MultiXactId would result in an error because the page on which the new
1969  * value would reside does not exist. This routine is in charge of creating
1970  * such pages.
1971  */
1972 static void
1974 {
1975  int pageno;
1976 
1978 
1979  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
1980 
1982  {
1983  int slotno;
1984 
1985  /*
1986  * Fortunately for us, SimpleLruWritePage is already prepared to deal
1987  * with creating a new segment file even if the page we're writing is
1988  * not the first in it, so this is enough.
1989  */
1990  slotno = ZeroMultiXactOffsetPage(pageno, false);
1992  }
1993 
1994  LWLockRelease(MultiXactOffsetSLRULock);
1995 }
1996 
1997 /*
1998  * This must be called ONCE during postmaster or standalone-backend startup.
1999  *
2000  * StartupXLOG has already established nextMXact/nextOffset by calling
2001  * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
2002  * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
2003  * replayed WAL.
2004  */
2005 void
2007 {
2010  int pageno;
2011 
2012  /*
2013  * Initialize offset's idea of the latest page number.
2014  */
2015  pageno = MultiXactIdToOffsetPage(multi);
2016  MultiXactOffsetCtl->shared->latest_page_number = pageno;
2017 
2018  /*
2019  * Initialize member's idea of the latest page number.
2020  */
2021  pageno = MXOffsetToMemberPage(offset);
2022  MultiXactMemberCtl->shared->latest_page_number = pageno;
2023 }
2024 
2025 /*
2026  * This must be called ONCE at the end of startup/recovery.
2027  */
2028 void
2030 {
2031  MultiXactId nextMXact;
2032  MultiXactOffset offset;
2033  MultiXactId oldestMXact;
2034  Oid oldestMXactDB;
2035  int pageno;
2036  int entryno;
2037  int flagsoff;
2038 
2039  LWLockAcquire(MultiXactGenLock, LW_SHARED);
2040  nextMXact = MultiXactState->nextMXact;
2041  offset = MultiXactState->nextOffset;
2042  oldestMXact = MultiXactState->oldestMultiXactId;
2043  oldestMXactDB = MultiXactState->oldestMultiXactDB;
2044  LWLockRelease(MultiXactGenLock);
2045 
2046  /* Clean up offsets state */
2047  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
2048 
2049  /*
2050  * (Re-)Initialize our idea of the latest page number for offsets.
2051  */
2052  pageno = MultiXactIdToOffsetPage(nextMXact);
2053  MultiXactOffsetCtl->shared->latest_page_number = pageno;
2054 
2055  /*
2056  * Zero out the remainder of the current offsets page. See notes in
2057  * TrimCLOG() for background. Unlike CLOG, some WAL record covers every
2058  * pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
2059  * rule "write xlog before data," nextMXact successors may carry obsolete,
2060  * nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
2061  * operates normally.
2062  */
2063  entryno = MultiXactIdToOffsetEntry(nextMXact);
2064  if (entryno != 0)
2065  {
2066  int slotno;
2067  MultiXactOffset *offptr;
2068 
2069  slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
2070  offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2071  offptr += entryno;
2072 
2073  MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
2074 
2075  MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
2076  }
2077 
2078  LWLockRelease(MultiXactOffsetSLRULock);
2079 
2080  /* And the same for members */
2081  LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
2082 
2083  /*
2084  * (Re-)Initialize our idea of the latest page number for members.
2085  */
2086  pageno = MXOffsetToMemberPage(offset);
2087  MultiXactMemberCtl->shared->latest_page_number = pageno;
2088 
2089  /*
2090  * Zero out the remainder of the current members page. See notes in
2091  * TrimCLOG() for motivation.
2092  */
2093  flagsoff = MXOffsetToFlagsOffset(offset);
2094  if (flagsoff != 0)
2095  {
2096  int slotno;
2097  TransactionId *xidptr;
2098  int memberoff;
2099 
2100  memberoff = MXOffsetToMemberOffset(offset);
2101  slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
2102  xidptr = (TransactionId *)
2103  (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
2104 
2105  MemSet(xidptr, 0, BLCKSZ - memberoff);
2106 
2107  /*
2108  * Note: we don't need to zero out the flag bits in the remaining
2109  * members of the current group, because they are always reset before
2110  * writing.
2111  */
2112 
2113  MultiXactMemberCtl->shared->page_dirty[slotno] = true;
2114  }
2115 
2116  LWLockRelease(MultiXactMemberSLRULock);
2117 
2118  /* signal that we're officially up */
2119  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2121  LWLockRelease(MultiXactGenLock);
2122 
2123  /* Now compute how far away the next members wraparound is. */
2124  SetMultiXactIdLimit(oldestMXact, oldestMXactDB, true);
2125 }
2126 
2127 /*
2128  * Get the MultiXact data to save in a checkpoint record
2129  */
2130 void
2131 MultiXactGetCheckptMulti(bool is_shutdown,
2132  MultiXactId *nextMulti,
2133  MultiXactOffset *nextMultiOffset,
2134  MultiXactId *oldestMulti,
2135  Oid *oldestMultiDB)
2136 {
2137  LWLockAcquire(MultiXactGenLock, LW_SHARED);
2138  *nextMulti = MultiXactState->nextMXact;
2139  *nextMultiOffset = MultiXactState->nextOffset;
2140  *oldestMulti = MultiXactState->oldestMultiXactId;
2141  *oldestMultiDB = MultiXactState->oldestMultiXactDB;
2142  LWLockRelease(MultiXactGenLock);
2143 
2145  "MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
2146  *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
2147 }
2148 
2149 /*
2150  * Perform a checkpoint --- either during shutdown, or on-the-fly
2151  */
2152 void
2154 {
2155  TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
2156 
2157  /*
2158  * Write dirty MultiXact pages to disk. This may result in sync requests
2159  * queued for later handling by ProcessSyncRequests(), as part of the
2160  * checkpoint.
2161  */
2164 
2165  TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
2166 }
2167 
2168 /*
2169  * Set the next-to-be-assigned MultiXactId and offset
2170  *
2171  * This is used when we can determine the correct next ID/offset exactly
2172  * from a checkpoint record. Although this is only called during bootstrap
2173  * and XLog replay, we take the lock in case any hot-standby backends are
2174  * examining the values.
2175  */
2176 void
2178  MultiXactOffset nextMultiOffset)
2179 {
2180  debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
2181  nextMulti, nextMultiOffset);
2182  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2183  MultiXactState->nextMXact = nextMulti;
2184  MultiXactState->nextOffset = nextMultiOffset;
2185  LWLockRelease(MultiXactGenLock);
2186 
2187  /*
2188  * During a binary upgrade, make sure that the offsets SLRU is large
2189  * enough to contain the next value that would be created.
2190  *
2191  * We need to do this pretty early during the first startup in binary
2192  * upgrade mode: before StartupMultiXact() in fact, because this routine
2193  * is called even before that by StartupXLOG(). And we can't do it
2194  * earlier than at this point, because during that first call of this
2195  * routine we determine the MultiXactState->nextMXact value that
2196  * MaybeExtendOffsetSlru needs.
2197  */
2198  if (IsBinaryUpgrade)
2200 }
2201 
2202 /*
2203  * Determine the last safe MultiXactId to allocate given the currently oldest
2204  * datminmxid (ie, the oldest MultiXactId that might exist in any database
2205  * of our cluster), and the OID of the (or a) database with that value.
2206  *
2207  * is_startup is true when we are just starting the cluster, false when we
2208  * are updating state in a running cluster. This only affects log messages.
2209  */
2210 void
2211 SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
2212  bool is_startup)
2213 {
2214  MultiXactId multiVacLimit;
2215  MultiXactId multiWarnLimit;
2216  MultiXactId multiStopLimit;
2217  MultiXactId multiWrapLimit;
2218  MultiXactId curMulti;
2219  bool needs_offset_vacuum;
2220 
2221  Assert(MultiXactIdIsValid(oldest_datminmxid));
2222 
2223  /*
2224  * We pretend that a wrap will happen halfway through the multixact ID
2225  * space, but that's not really true, because multixacts wrap differently
2226  * from transaction IDs. Note that, separately from any concern about
2227  * multixact IDs wrapping, we must ensure that multixact members do not
2228  * wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
2229  */
2230  multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2231  if (multiWrapLimit < FirstMultiXactId)
2232  multiWrapLimit += FirstMultiXactId;
2233 
2234  /*
2235  * We'll refuse to continue assigning MultiXactIds once we get within 3M
2236  * multi of data loss. See SetTransactionIdLimit.
2237  */
2238  multiStopLimit = multiWrapLimit - 3000000;
2239  if (multiStopLimit < FirstMultiXactId)
2240  multiStopLimit -= FirstMultiXactId;
2241 
2242  /*
2243  * We'll start complaining loudly when we get within 40M multis of data
2244  * loss. This is kind of arbitrary, but if you let your gas gauge get
2245  * down to 2% of full, would you be looking for the next gas station? We
2246  * need to be fairly liberal about this number because there are lots of
2247  * scenarios where most transactions are done by automatic clients that
2248  * won't pay attention to warnings. (No, we're not gonna make this
2249  * configurable. If you know enough to configure it, you know enough to
2250  * not get in this kind of trouble in the first place.)
2251  */
2252  multiWarnLimit = multiWrapLimit - 40000000;
2253  if (multiWarnLimit < FirstMultiXactId)
2254  multiWarnLimit -= FirstMultiXactId;
2255 
2256  /*
2257  * We'll start trying to force autovacuums when oldest_datminmxid gets to
2258  * be more than autovacuum_multixact_freeze_max_age mxids old.
2259  *
2260  * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2261  * so that we don't have to worry about dealing with on-the-fly changes in
2262  * its value. See SetTransactionIdLimit.
2263  */
2264  multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2265  if (multiVacLimit < FirstMultiXactId)
2266  multiVacLimit += FirstMultiXactId;
2267 
2268  /* Grab lock for just long enough to set the new limit values */
2269  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2270  MultiXactState->oldestMultiXactId = oldest_datminmxid;
2271  MultiXactState->oldestMultiXactDB = oldest_datoid;
2272  MultiXactState->multiVacLimit = multiVacLimit;
2273  MultiXactState->multiWarnLimit = multiWarnLimit;
2274  MultiXactState->multiStopLimit = multiStopLimit;
2275  MultiXactState->multiWrapLimit = multiWrapLimit;
2276  curMulti = MultiXactState->nextMXact;
2277  LWLockRelease(MultiXactGenLock);
2278 
2279  /* Log the info */
2280  ereport(DEBUG1,
2281  (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2282  multiWrapLimit, oldest_datoid)));
2283 
2284  /*
2285  * Computing the actual limits is only possible once the data directory is
2286  * in a consistent state. There's no need to compute the limits while
2287  * still replaying WAL - no decisions about new multis are made even
2288  * though multixact creations might be replayed. So we'll only do further
2289  * checks after TrimMultiXact() has been called.
2290  */
2292  return;
2293 
2294  Assert(!InRecovery);
2295 
2296  /* Set limits for offset vacuum. */
2297  needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
2298 
2299  /*
2300  * If past the autovacuum force point, immediately signal an autovac
2301  * request. The reason for this is that autovac only processes one
2302  * database per invocation. Once it's finished cleaning up the oldest
2303  * database, it'll call here, and we'll signal the postmaster to start
2304  * another iteration immediately if there are still any old databases.
2305  */
2306  if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
2307  needs_offset_vacuum) && IsUnderPostmaster)
2309 
2310  /* Give an immediate warning if past the wrap warn point */
2311  if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2312  {
2313  char *oldest_datname;
2314 
2315  /*
2316  * We can be called when not inside a transaction, for example during
2317  * StartupXLOG(). In such a case we cannot do database access, so we
2318  * must just report the oldest DB's OID.
2319  *
2320  * Note: it's also possible that get_database_name fails and returns
2321  * NULL, for example because the database just got dropped. We'll
2322  * still warn, even though the warning might now be unnecessary.
2323  */
2324  if (IsTransactionState())
2325  oldest_datname = get_database_name(oldest_datoid);
2326  else
2327  oldest_datname = NULL;
2328 
2329  if (oldest_datname)
2330  ereport(WARNING,
2331  (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2332  "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2333  multiWrapLimit - curMulti,
2334  oldest_datname,
2335  multiWrapLimit - curMulti),
2336  errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
2337  "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2338  else
2339  ereport(WARNING,
2340  (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2341  "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2342  multiWrapLimit - curMulti,
2343  oldest_datoid,
2344  multiWrapLimit - curMulti),
2345  errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
2346  "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2347  }
2348 }
2349 
2350 /*
2351  * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2352  * and similarly nextOffset is at least minMultiOffset.
2353  *
2354  * This is used when we can determine minimum safe values from an XLog
2355  * record (either an on-line checkpoint or an mxact creation log entry).
2356  * Although this is only called during XLog replay, we take the lock in case
2357  * any hot-standby backends are examining the values.
2358  */
2359 void
2361  MultiXactOffset minMultiOffset)
2362 {
2363  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2365  {
2366  debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2367  MultiXactState->nextMXact = minMulti;
2368  }
2369  if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
2370  {
2371  debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
2372  minMultiOffset);
2373  MultiXactState->nextOffset = minMultiOffset;
2374  }
2375  LWLockRelease(MultiXactGenLock);
2376 }
2377 
2378 /*
2379  * Update our oldestMultiXactId value, but only if it's more recent than what
2380  * we had.
2381  *
2382  * This may only be called during WAL replay.
2383  */
2384 void
2385 MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2386 {
2387  Assert(InRecovery);
2388 
2390  SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
2391 }
2392 
2393 /*
2394  * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2395  *
2396  * NB: this is called while holding MultiXactGenLock. We want it to be very
2397  * fast most of the time; even when it's not so fast, no actual I/O need
2398  * happen unless we're forced to write out a dirty log or xlog page to make
2399  * room in shared memory.
2400  */
2401 static void
2403 {
2404  int pageno;
2405 
2406  /*
2407  * No work except at first MultiXactId of a page. But beware: just after
2408  * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2409  */
2410  if (MultiXactIdToOffsetEntry(multi) != 0 &&
2411  multi != FirstMultiXactId)
2412  return;
2413 
2414  pageno = MultiXactIdToOffsetPage(multi);
2415 
2416  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
2417 
2418  /* Zero the page and make an XLOG entry about it */
2419  ZeroMultiXactOffsetPage(pageno, true);
2420 
2421  LWLockRelease(MultiXactOffsetSLRULock);
2422 }
2423 
2424 /*
2425  * Make sure that MultiXactMember has room for the members of a newly-
2426  * allocated MultiXactId.
2427  *
2428  * Like the above routine, this is called while holding MultiXactGenLock;
2429  * same comments apply.
2430  */
2431 static void
2433 {
2434  /*
2435  * It's possible that the members span more than one page of the members
2436  * file, so we loop to ensure we consider each page. The coding is not
2437  * optimal if the members span several pages, but that seems unusual
2438  * enough to not worry much about.
2439  */
2440  while (nmembers > 0)
2441  {
2442  int flagsoff;
2443  int flagsbit;
2445 
2446  /*
2447  * Only zero when at first entry of a page.
2448  */
2449  flagsoff = MXOffsetToFlagsOffset(offset);
2450  flagsbit = MXOffsetToFlagsBitShift(offset);
2451  if (flagsoff == 0 && flagsbit == 0)
2452  {
2453  int pageno;
2454 
2455  pageno = MXOffsetToMemberPage(offset);
2456 
2457  LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
2458 
2459  /* Zero the page and make an XLOG entry about it */
2460  ZeroMultiXactMemberPage(pageno, true);
2461 
2462  LWLockRelease(MultiXactMemberSLRULock);
2463  }
2464 
2465  /*
2466  * Compute the number of items till end of current page. Careful: if
2467  * addition of unsigned ints wraps around, we're at the last page of
2468  * the last segment; since that page holds a different number of items
2469  * than other pages, we need to do it differently.
2470  */
2471  if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
2472  {
2473  /*
2474  * This is the last page of the last segment; we can compute the
2475  * number of items left to allocate in it without modulo
2476  * arithmetic.
2477  */
2478  difference = MaxMultiXactOffset - offset + 1;
2479  }
2480  else
2482 
2483  /*
2484  * Advance to next page, taking care to properly handle the wraparound
2485  * case. OK if nmembers goes negative.
2486  */
2487  nmembers -= difference;
2488  offset += difference;
2489  }
2490 }
2491 
2492 /*
2493  * GetOldestMultiXactId
2494  *
2495  * Return the oldest MultiXactId that's still possibly still seen as live by
2496  * any running transaction. Older ones might still exist on disk, but they no
2497  * longer have any running member transaction.
2498  *
2499  * It's not safe to truncate MultiXact SLRU segments on the value returned by
2500  * this function; however, it can be used by a full-table vacuum to set the
2501  * point at which it will be possible to truncate SLRU for that table.
2502  */
2505 {
2506  MultiXactId oldestMXact;
2507  MultiXactId nextMXact;
2508  int i;
2509 
2510  /*
2511  * This is the oldest valid value among all the OldestMemberMXactId[] and
2512  * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2513  */
2514  LWLockAcquire(MultiXactGenLock, LW_SHARED);
2515 
2516  /*
2517  * We have to beware of the possibility that nextMXact is in the
2518  * wrapped-around state. We don't fix the counter itself here, but we
2519  * must be sure to use a valid value in our calculation.
2520  */
2521  nextMXact = MultiXactState->nextMXact;
2522  if (nextMXact < FirstMultiXactId)
2523  nextMXact = FirstMultiXactId;
2524 
2525  oldestMXact = nextMXact;
2526  for (i = 1; i <= MaxOldestSlot; i++)
2527  {
2528  MultiXactId thisoldest;
2529 
2530  thisoldest = OldestMemberMXactId[i];
2531  if (MultiXactIdIsValid(thisoldest) &&
2532  MultiXactIdPrecedes(thisoldest, oldestMXact))
2533  oldestMXact = thisoldest;
2534  thisoldest = OldestVisibleMXactId[i];
2535  if (MultiXactIdIsValid(thisoldest) &&
2536  MultiXactIdPrecedes(thisoldest, oldestMXact))
2537  oldestMXact = thisoldest;
2538  }
2539 
2540  LWLockRelease(MultiXactGenLock);
2541 
2542  return oldestMXact;
2543 }
2544 
2545 /*
2546  * Determine how aggressively we need to vacuum in order to prevent member
2547  * wraparound.
2548  *
2549  * To do so determine what's the oldest member offset and install the limit
2550  * info in MultiXactState, where it can be used to prevent overrun of old data
2551  * in the members SLRU area.
2552  *
2553  * The return value is true if emergency autovacuum is required and false
2554  * otherwise.
2555  */
2556 static bool
2557 SetOffsetVacuumLimit(bool is_startup)
2558 {
2559  MultiXactId oldestMultiXactId;
2560  MultiXactId nextMXact;
2561  MultiXactOffset oldestOffset = 0; /* placate compiler */
2562  MultiXactOffset prevOldestOffset;
2563  MultiXactOffset nextOffset;
2564  bool oldestOffsetKnown = false;
2565  bool prevOldestOffsetKnown;
2566  MultiXactOffset offsetStopLimit = 0;
2567  MultiXactOffset prevOffsetStopLimit;
2568 
2569  /*
2570  * NB: Have to prevent concurrent truncation, we might otherwise try to
2571  * lookup an oldestMulti that's concurrently getting truncated away.
2572  */
2573  LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2574 
2575  /* Read relevant fields from shared memory. */
2576  LWLockAcquire(MultiXactGenLock, LW_SHARED);
2577  oldestMultiXactId = MultiXactState->oldestMultiXactId;
2578  nextMXact = MultiXactState->nextMXact;
2579  nextOffset = MultiXactState->nextOffset;
2580  prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2581  prevOldestOffset = MultiXactState->oldestOffset;
2582  prevOffsetStopLimit = MultiXactState->offsetStopLimit;
2584  LWLockRelease(MultiXactGenLock);
2585 
2586  /*
2587  * Determine the offset of the oldest multixact. Normally, we can read
2588  * the offset from the multixact itself, but there's an important special
2589  * case: if there are no multixacts in existence at all, oldestMXact
2590  * obviously can't point to one. It will instead point to the multixact
2591  * ID that will be assigned the next time one is needed.
2592  */
2593  if (oldestMultiXactId == nextMXact)
2594  {
2595  /*
2596  * When the next multixact gets created, it will be stored at the next
2597  * offset.
2598  */
2599  oldestOffset = nextOffset;
2600  oldestOffsetKnown = true;
2601  }
2602  else
2603  {
2604  /*
2605  * Figure out where the oldest existing multixact's offsets are
2606  * stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
2607  * the supposedly-earliest multixact might not really exist. We are
2608  * careful not to fail in that case.
2609  */
2610  oldestOffsetKnown =
2611  find_multixact_start(oldestMultiXactId, &oldestOffset);
2612 
2613  if (oldestOffsetKnown)
2614  ereport(DEBUG1,
2615  (errmsg_internal("oldest MultiXactId member is at offset %u",
2616  oldestOffset)));
2617  else
2618  ereport(LOG,
2619  (errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
2620  oldestMultiXactId)));
2621  }
2622 
2623  LWLockRelease(MultiXactTruncationLock);
2624 
2625  /*
2626  * If we can, compute limits (and install them MultiXactState) to prevent
2627  * overrun of old data in the members SLRU area. We can only do so if the
2628  * oldest offset is known though.
2629  */
2630  if (oldestOffsetKnown)
2631  {
2632  /* move back to start of the corresponding segment */
2633  offsetStopLimit = oldestOffset - (oldestOffset %
2635 
2636  /* always leave one segment before the wraparound point */
2637  offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
2638 
2639  if (!prevOldestOffsetKnown && !is_startup)
2640  ereport(LOG,
2641  (errmsg("MultiXact member wraparound protections are now enabled")));
2642 
2643  ereport(DEBUG1,
2644  (errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
2645  offsetStopLimit, oldestMultiXactId)));
2646  }
2647  else if (prevOldestOffsetKnown)
2648  {
2649  /*
2650  * If we failed to get the oldest offset this time, but we have a
2651  * value from a previous pass through this function, use the old
2652  * values rather than automatically forcing an emergency autovacuum
2653  * cycle again.
2654  */
2655  oldestOffset = prevOldestOffset;
2656  oldestOffsetKnown = true;
2657  offsetStopLimit = prevOffsetStopLimit;
2658  }
2659 
2660  /* Install the computed values */
2661  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2662  MultiXactState->oldestOffset = oldestOffset;
2663  MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
2664  MultiXactState->offsetStopLimit = offsetStopLimit;
2665  LWLockRelease(MultiXactGenLock);
2666 
2667  /*
2668  * Do we need an emergency autovacuum? If we're not sure, assume yes.
2669  */
2670  return !oldestOffsetKnown ||
2671  (nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
2672 }
2673 
2674 /*
2675  * Return whether adding "distance" to "start" would move past "boundary".
2676  *
2677  * We use this to determine whether the addition is "wrapping around" the
2678  * boundary point, hence the name. The reason we don't want to use the regular
2679  * 2^31-modulo arithmetic here is that we want to be able to use the whole of
2680  * the 2^32-1 space here, allowing for more multixacts than would fit
2681  * otherwise.
2682  */
2683 static bool
2685  uint32 distance)
2686 {
2687  MultiXactOffset finish;
2688 
2689  /*
2690  * Note that offset number 0 is not used (see GetMultiXactIdMembers), so
2691  * if the addition wraps around the UINT_MAX boundary, skip that value.
2692  */
2693  finish = start + distance;
2694  if (finish < start)
2695  finish++;
2696 
2697  /*-----------------------------------------------------------------------
2698  * When the boundary is numerically greater than the starting point, any
2699  * value numerically between the two is not wrapped:
2700  *
2701  * <----S----B---->
2702  * [---) = F wrapped past B (and UINT_MAX)
2703  * [---) = F not wrapped
2704  * [----] = F wrapped past B
2705  *
2706  * When the boundary is numerically less than the starting point (i.e. the
2707  * UINT_MAX wraparound occurs somewhere in between) then all values in
2708  * between are wrapped:
2709  *
2710  * <----B----S---->
2711  * [---) = F not wrapped past B (but wrapped past UINT_MAX)
2712  * [---) = F wrapped past B (and UINT_MAX)
2713  * [----] = F not wrapped
2714  *-----------------------------------------------------------------------
2715  */
2716  if (start < boundary)
2717  return finish >= boundary || finish < start;
2718  else
2719  return finish >= boundary && finish < start;
2720 }
2721 
2722 /*
2723  * Find the starting offset of the given MultiXactId.
2724  *
2725  * Returns false if the file containing the multi does not exist on disk.
2726  * Otherwise, returns true and sets *result to the starting member offset.
2727  *
2728  * This function does not prevent concurrent truncation, so if that's
2729  * required, the caller has to protect against that.
2730  */
2731 static bool
2733 {
2734  MultiXactOffset offset;
2735  int pageno;
2736  int entryno;
2737  int slotno;
2738  MultiXactOffset *offptr;
2739 
2741 
2742  pageno = MultiXactIdToOffsetPage(multi);
2743  entryno = MultiXactIdToOffsetEntry(multi);
2744 
2745  /*
2746  * Write out dirty data, so PhysicalPageExists can work correctly.
2747  */
2750 
2752  return false;
2753 
2754  /* lock is acquired by SimpleLruReadPage_ReadOnly */
2755  slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2756  offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2757  offptr += entryno;
2758  offset = *offptr;
2759  LWLockRelease(MultiXactOffsetSLRULock);
2760 
2761  *result = offset;
2762  return true;
2763 }
2764 
2765 /*
2766  * Determine how many multixacts, and how many multixact members, currently
2767  * exist. Return false if unable to determine.
2768  */
2769 static bool
2771 {
2772  MultiXactOffset nextOffset;
2773  MultiXactOffset oldestOffset;
2774  MultiXactId oldestMultiXactId;
2775  MultiXactId nextMultiXactId;
2776  bool oldestOffsetKnown;
2777 
2778  LWLockAcquire(MultiXactGenLock, LW_SHARED);
2779  nextOffset = MultiXactState->nextOffset;
2780  oldestMultiXactId = MultiXactState->oldestMultiXactId;
2781  nextMultiXactId = MultiXactState->nextMXact;
2782  oldestOffset = MultiXactState->oldestOffset;
2783  oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2784  LWLockRelease(MultiXactGenLock);
2785 
2786  if (!oldestOffsetKnown)
2787  return false;
2788 
2789  *members = nextOffset - oldestOffset;
2790  *multixacts = nextMultiXactId - oldestMultiXactId;
2791  return true;
2792 }
2793 
2794 /*
2795  * Multixact members can be removed once the multixacts that refer to them
2796  * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2797  * vacuum_multixact_freeze_table_age work together to make sure we never have
2798  * too many multixacts; we hope that, at least under normal circumstances,
2799  * this will also be sufficient to keep us from using too many offsets.
2800  * However, if the average multixact has many members, we might exhaust the
2801  * members space while still using few enough members that these limits fail
2802  * to trigger full table scans for relminmxid advancement. At that point,
2803  * we'd have no choice but to start failing multixact-creating operations
2804  * with an error.
2805  *
2806  * To prevent that, if more than a threshold portion of the members space is
2807  * used, we effectively reduce autovacuum_multixact_freeze_max_age and
2808  * to a value just less than the number of multixacts in use. We hope that
2809  * this will quickly trigger autovacuuming on the table or tables with the
2810  * oldest relminmxid, thus allowing datminmxid values to advance and removing
2811  * some members.
2812  *
2813  * As the fraction of the member space currently in use grows, we become
2814  * more aggressive in clamping this value. That not only causes autovacuum
2815  * to ramp up, but also makes any manual vacuums the user issues more
2816  * aggressive. This happens because vacuum_set_xid_limits() clamps the
2817  * freeze table and the minimum freeze age based on the effective
2818  * autovacuum_multixact_freeze_max_age this function returns. In the worst
2819  * case, we'll claim the freeze_max_age to zero, and every vacuum of any
2820  * table will try to freeze every multixact.
2821  *
2822  * It's possible that these thresholds should be user-tunable, but for now
2823  * we keep it simple.
2824  */
2825 int
2827 {
2828  MultiXactOffset members;
2829  uint32 multixacts;
2830  uint32 victim_multixacts;
2831  double fraction;
2832 
2833  /* If we can't determine member space utilization, assume the worst. */
2834  if (!ReadMultiXactCounts(&multixacts, &members))
2835  return 0;
2836 
2837  /* If member space utilization is low, no special action is required. */
2838  if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
2840 
2841  /*
2842  * Compute a target for relminmxid advancement. The number of multixacts
2843  * we try to eliminate from the system is based on how far we are past
2844  * MULTIXACT_MEMBER_SAFE_THRESHOLD.
2845  */
2846  fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
2848  victim_multixacts = multixacts * fraction;
2849 
2850  /* fraction could be > 1.0, but lowest possible freeze age is zero */
2851  if (victim_multixacts > multixacts)
2852  return 0;
2853  return multixacts - victim_multixacts;
2854 }
2855 
2856 typedef struct mxtruncinfo
2857 {
2860 
2861 /*
2862  * SlruScanDirectory callback
2863  * This callback determines the earliest existing page number.
2864  */
2865 static bool
2866 SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int segpage, void *data)
2867 {
2868  mxtruncinfo *trunc = (mxtruncinfo *) data;
2869 
2870  if (trunc->earliestExistingPage == -1 ||
2871  ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
2872  {
2873  trunc->earliestExistingPage = segpage;
2874  }
2875 
2876  return false; /* keep going */
2877 }
2878 
2879 
2880 /*
2881  * Delete members segments [oldest, newOldest)
2882  *
2883  * The members SLRU can, in contrast to the offsets one, be filled to almost
2884  * the full range at once. This means SimpleLruTruncate() can't trivially be
2885  * used - instead the to-be-deleted range is computed using the offsets
2886  * SLRU. C.f. TruncateMultiXact().
2887  */
2888 static void
2890 {
2891  const int maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
2892  int startsegment = MXOffsetToMemberSegment(oldestOffset);
2893  int endsegment = MXOffsetToMemberSegment(newOldestOffset);
2894  int segment = startsegment;
2895 
2896  /*
2897  * Delete all the segments but the last one. The last segment can still
2898  * contain, possibly partially, valid data.
2899  */
2900  while (segment != endsegment)
2901  {
2902  elog(DEBUG2, "truncating multixact members segment %x", segment);
2904 
2905  /* move to next segment, handling wraparound correctly */
2906  if (segment == maxsegment)
2907  segment = 0;
2908  else
2909  segment += 1;
2910  }
2911 }
2912 
2913 /*
2914  * Delete offsets segments [oldest, newOldest)
2915  */
2916 static void
2918 {
2919  /*
2920  * We step back one multixact to avoid passing a cutoff page that hasn't
2921  * been created yet in the rare case that oldestMulti would be the first
2922  * item on a page and oldestMulti == nextMulti. In that case, if we
2923  * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
2924  * detection.
2925  */
2927  MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
2928 }
2929 
2930 /*
2931  * Remove all MultiXactOffset and MultiXactMember segments before the oldest
2932  * ones still of interest.
2933  *
2934  * This is only called on a primary as part of vacuum (via
2935  * vac_truncate_clog()). During recovery truncation is done by replaying
2936  * truncation WAL records logged here.
2937  *
2938  * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
2939  * is one of the databases preventing newOldestMulti from increasing.
2940  */
2941 void
2942 TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
2943 {
2944  MultiXactId oldestMulti;
2945  MultiXactId nextMulti;
2946  MultiXactOffset newOldestOffset;
2947  MultiXactOffset oldestOffset;
2948  MultiXactOffset nextOffset;
2949  mxtruncinfo trunc;
2950  MultiXactId earliest;
2951 
2954 
2955  /*
2956  * We can only allow one truncation to happen at once. Otherwise parts of
2957  * members might vanish while we're doing lookups or similar. There's no
2958  * need to have an interlock with creating new multis or such, since those
2959  * are constrained by the limits (which only grow, never shrink).
2960  */
2961  LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
2962 
2963  LWLockAcquire(MultiXactGenLock, LW_SHARED);
2964  nextMulti = MultiXactState->nextMXact;
2965  nextOffset = MultiXactState->nextOffset;
2966  oldestMulti = MultiXactState->oldestMultiXactId;
2967  LWLockRelease(MultiXactGenLock);
2968  Assert(MultiXactIdIsValid(oldestMulti));
2969 
2970  /*
2971  * Make sure to only attempt truncation if there's values to truncate
2972  * away. In normal processing values shouldn't go backwards, but there's
2973  * some corner cases (due to bugs) where that's possible.
2974  */
2975  if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
2976  {
2977  LWLockRelease(MultiXactTruncationLock);
2978  return;
2979  }
2980 
2981  /*
2982  * Note we can't just plow ahead with the truncation; it's possible that
2983  * there are no segments to truncate, which is a problem because we are
2984  * going to attempt to read the offsets page to determine where to
2985  * truncate the members SLRU. So we first scan the directory to determine
2986  * the earliest offsets page number that we can read without error.
2987  *
2988  * When nextMXact is less than one segment away from multiWrapLimit,
2989  * SlruScanDirCbFindEarliest can find some early segment other than the
2990  * actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
2991  * returns false, because not all pairs of entries have the same answer.)
2992  * That can also arise when an earlier truncation attempt failed unlink()
2993  * or returned early from this function. The only consequence is
2994  * returning early, which wastes space that we could have liberated.
2995  *
2996  * NB: It's also possible that the page that oldestMulti is on has already
2997  * been truncated away, and we crashed before updating oldestMulti.
2998  */
2999  trunc.earliestExistingPage = -1;
3002  if (earliest < FirstMultiXactId)
3003  earliest = FirstMultiXactId;
3004 
3005  /* If there's nothing to remove, we can bail out early. */
3006  if (MultiXactIdPrecedes(oldestMulti, earliest))
3007  {
3008  LWLockRelease(MultiXactTruncationLock);
3009  return;
3010  }
3011 
3012  /*
3013  * First, compute the safe truncation point for MultiXactMember. This is
3014  * the starting offset of the oldest multixact.
3015  *
3016  * Hopefully, find_multixact_start will always work here, because we've
3017  * already checked that it doesn't precede the earliest MultiXact on disk.
3018  * But if it fails, don't truncate anything, and log a message.
3019  */
3020  if (oldestMulti == nextMulti)
3021  {
3022  /* there are NO MultiXacts */
3023  oldestOffset = nextOffset;
3024  }
3025  else if (!find_multixact_start(oldestMulti, &oldestOffset))
3026  {
3027  ereport(LOG,
3028  (errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
3029  oldestMulti, earliest)));
3030  LWLockRelease(MultiXactTruncationLock);
3031  return;
3032  }
3033 
3034  /*
3035  * Secondly compute up to where to truncate. Lookup the corresponding
3036  * member offset for newOldestMulti for that.
3037  */
3038  if (newOldestMulti == nextMulti)
3039  {
3040  /* there are NO MultiXacts */
3041  newOldestOffset = nextOffset;
3042  }
3043  else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
3044  {
3045  ereport(LOG,
3046  (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
3047  newOldestMulti)));
3048  LWLockRelease(MultiXactTruncationLock);
3049  return;
3050  }
3051 
3052  elog(DEBUG1, "performing multixact truncation: "
3053  "offsets [%u, %u), offsets segments [%x, %x), "
3054  "members [%u, %u), members segments [%x, %x)",
3055  oldestMulti, newOldestMulti,
3056  MultiXactIdToOffsetSegment(oldestMulti),
3057  MultiXactIdToOffsetSegment(newOldestMulti),
3058  oldestOffset, newOldestOffset,
3059  MXOffsetToMemberSegment(oldestOffset),
3060  MXOffsetToMemberSegment(newOldestOffset));
3061 
3062  /*
3063  * Do truncation, and the WAL logging of the truncation, in a critical
3064  * section. That way offsets/members cannot get out of sync anymore, i.e.
3065  * once consistent the newOldestMulti will always exist in members, even
3066  * if we crashed in the wrong moment.
3067  */
3069 
3070  /*
3071  * Prevent checkpoints from being scheduled concurrently. This is critical
3072  * because otherwise a truncation record might not be replayed after a
3073  * crash/basebackup, even though the state of the data directory would
3074  * require it.
3075  */
3078 
3079  /* WAL log truncation */
3080  WriteMTruncateXlogRec(newOldestMultiDB,
3081  oldestMulti, newOldestMulti,
3082  oldestOffset, newOldestOffset);
3083 
3084  /*
3085  * Update in-memory limits before performing the truncation, while inside
3086  * the critical section: Have to do it before truncation, to prevent
3087  * concurrent lookups of those values. Has to be inside the critical
3088  * section as otherwise a future call to this function would error out,
3089  * while looking up the oldest member in offsets, if our caller crashes
3090  * before updating the limits.
3091  */
3092  LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
3093  MultiXactState->oldestMultiXactId = newOldestMulti;
3094  MultiXactState->oldestMultiXactDB = newOldestMultiDB;
3095  LWLockRelease(MultiXactGenLock);
3096 
3097  /* First truncate members */
3098  PerformMembersTruncation(oldestOffset, newOldestOffset);
3099 
3100  /* Then offsets */
3101  PerformOffsetsTruncation(oldestMulti, newOldestMulti);
3102 
3104 
3105  END_CRIT_SECTION();
3106  LWLockRelease(MultiXactTruncationLock);
3107 }
3108 
3109 /*
3110  * Decide whether a MultiXactOffset page number is "older" for truncation
3111  * purposes. Analogous to CLOGPagePrecedes().
3112  *
3113  * Offsetting the values is optional, because MultiXactIdPrecedes() has
3114  * translational symmetry.
3115  */
3116 static bool
3117 MultiXactOffsetPagePrecedes(int page1, int page2)
3118 {
3119  MultiXactId multi1;
3120  MultiXactId multi2;
3121 
3122  multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
3123  multi1 += FirstMultiXactId + 1;
3124  multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
3125  multi2 += FirstMultiXactId + 1;
3126 
3127  return (MultiXactIdPrecedes(multi1, multi2) &&
3128  MultiXactIdPrecedes(multi1,
3129  multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
3130 }
3131 
3132 /*
3133  * Decide whether a MultiXactMember page number is "older" for truncation
3134  * purposes. There is no "invalid offset number" so use the numbers verbatim.
3135  */
3136 static bool
3137 MultiXactMemberPagePrecedes(int page1, int page2)
3138 {
3139  MultiXactOffset offset1;
3140  MultiXactOffset offset2;
3141 
3142  offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
3143  offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
3144 
3145  return (MultiXactOffsetPrecedes(offset1, offset2) &&
3146  MultiXactOffsetPrecedes(offset1,
3147  offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
3148 }
3149 
3150 /*
3151  * Decide which of two MultiXactIds is earlier.
3152  *
3153  * XXX do we need to do something special for InvalidMultiXactId?
3154  * (Doesn't look like it.)
3155  */
3156 bool
3158 {
3159  int32 diff = (int32) (multi1 - multi2);
3160 
3161  return (diff < 0);
3162 }
3163 
3164 /*
3165  * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
3166  *
3167  * XXX do we need to do something special for InvalidMultiXactId?
3168  * (Doesn't look like it.)
3169  */
3170 bool
3172 {
3173  int32 diff = (int32) (multi1 - multi2);
3174 
3175  return (diff <= 0);
3176 }
3177 
3178 
3179 /*
3180  * Decide which of two offsets is earlier.
3181  */
3182 static bool
3184 {
3185  int32 diff = (int32) (offset1 - offset2);
3186 
3187  return (diff < 0);
3188 }
3189 
3190 /*
3191  * Write an xlog record reflecting the zeroing of either a MEMBERs or
3192  * OFFSETs page (info shows which)
3193  */
3194 static void
3195 WriteMZeroPageXlogRec(int pageno, uint8 info)
3196 {
3197  XLogBeginInsert();
3198  XLogRegisterData((char *) (&pageno), sizeof(int));
3199  (void) XLogInsert(RM_MULTIXACT_ID, info);
3200 }
3201 
3202 /*
3203  * Write a TRUNCATE xlog record
3204  *
3205  * We must flush the xlog record to disk before returning --- see notes in
3206  * TruncateCLOG().
3207  */
3208 static void
3210  MultiXactId startTruncOff, MultiXactId endTruncOff,
3211  MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
3212 {
3213  XLogRecPtr recptr;
3214  xl_multixact_truncate xlrec;
3215 
3216  xlrec.oldestMultiDB = oldestMultiDB;
3217 
3218  xlrec.startTruncOff = startTruncOff;
3219  xlrec.endTruncOff = endTruncOff;
3220 
3221  xlrec.startTruncMemb = startTruncMemb;
3222  xlrec.endTruncMemb = endTruncMemb;
3223 
3224  XLogBeginInsert();
3225  XLogRegisterData((char *) (&xlrec), SizeOfMultiXactTruncate);
3226  recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
3227  XLogFlush(recptr);
3228 }
3229 
3230 /*
3231  * MULTIXACT resource manager's routines
3232  */
3233 void
3235 {
3236  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3237 
3238  /* Backup blocks are not used in multixact records */
3239  Assert(!XLogRecHasAnyBlockRefs(record));
3240 
3241  if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
3242  {
3243  int pageno;
3244  int slotno;
3245 
3246  memcpy(&pageno, XLogRecGetData(record), sizeof(int));
3247 
3248  LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
3249 
3250  slotno = ZeroMultiXactOffsetPage(pageno, false);
3252  Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
3253 
3254  LWLockRelease(MultiXactOffsetSLRULock);
3255  }
3256  else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
3257  {
3258  int pageno;
3259  int slotno;
3260 
3261  memcpy(&pageno, XLogRecGetData(record), sizeof(int));
3262 
3263  LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
3264 
3265  slotno = ZeroMultiXactMemberPage(pageno, false);
3267  Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
3268 
3269  LWLockRelease(MultiXactMemberSLRULock);
3270  }
3271  else if (info == XLOG_MULTIXACT_CREATE_ID)
3272  {
3273  xl_multixact_create *xlrec =
3274  (xl_multixact_create *) XLogRecGetData(record);
3275  TransactionId max_xid;
3276  int i;
3277 
3278  /* Store the data back into the SLRU files */
3279  RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
3280  xlrec->members);
3281 
3282  /* Make sure nextMXact/nextOffset are beyond what this record has */
3283  MultiXactAdvanceNextMXact(xlrec->mid + 1,
3284  xlrec->moff + xlrec->nmembers);
3285 
3286  /*
3287  * Make sure nextXid is beyond any XID mentioned in the record. This
3288  * should be unnecessary, since any XID found here ought to have other
3289  * evidence in the XLOG, but let's be safe.
3290  */
3291  max_xid = XLogRecGetXid(record);
3292  for (i = 0; i < xlrec->nmembers; i++)
3293  {
3294  if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
3295  max_xid = xlrec->members[i].xid;
3296  }
3297 
3299  }
3300  else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
3301  {
3302  xl_multixact_truncate xlrec;
3303  int pageno;
3304 
3305  memcpy(&xlrec, XLogRecGetData(record),
3307 
3308  elog(DEBUG1, "replaying multixact truncation: "
3309  "offsets [%u, %u), offsets segments [%x, %x), "
3310  "members [%u, %u), members segments [%x, %x)",
3311  xlrec.startTruncOff, xlrec.endTruncOff,
3314  xlrec.startTruncMemb, xlrec.endTruncMemb,
3317 
3318  /* should not be required, but more than cheap enough */
3319  LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3320 
3321  /*
3322  * Advance the horizon values, so they're current at the end of
3323  * recovery.
3324  */
3325  SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
3326 
3328 
3329  /*
3330  * During XLOG replay, latest_page_number isn't necessarily set up
3331  * yet; insert a suitable value to bypass the sanity test in
3332  * SimpleLruTruncate.
3333  */
3334  pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
3335  MultiXactOffsetCtl->shared->latest_page_number = pageno;
3337 
3338  LWLockRelease(MultiXactTruncationLock);
3339  }
3340  else
3341  elog(PANIC, "multixact_redo: unknown op code %u", info);
3342 }
3343 
3344 Datum
3346 {
3347  typedef struct
3348  {
3349  MultiXactMember *members;
3350  int nmembers;
3351  int iter;
3352  } mxact;
3354  mxact *multi;
3355  FuncCallContext *funccxt;
3356 
3357  if (mxid < FirstMultiXactId)
3358  ereport(ERROR,
3359  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3360  errmsg("invalid MultiXactId: %u", mxid)));
3361 
3362  if (SRF_IS_FIRSTCALL())
3363  {
3364  MemoryContext oldcxt;
3365  TupleDesc tupdesc;
3366 
3367  funccxt = SRF_FIRSTCALL_INIT();
3368  oldcxt = MemoryContextSwitchTo(funccxt->multi_call_memory_ctx);
3369 
3370  multi = palloc(sizeof(mxact));
3371  /* no need to allow for old values here */
3372  multi->nmembers = GetMultiXactIdMembers(mxid, &multi->members, false,
3373  false);
3374  multi->iter = 0;
3375 
3376  tupdesc = CreateTemplateTupleDesc(2);
3377  TupleDescInitEntry(tupdesc, (AttrNumber) 1, "xid",
3378  XIDOID, -1, 0);
3379  TupleDescInitEntry(tupdesc, (AttrNumber) 2, "mode",
3380  TEXTOID, -1, 0);
3381 
3382  funccxt->attinmeta = TupleDescGetAttInMetadata(tupdesc);
3383  funccxt->user_fctx = multi;
3384 
3385  MemoryContextSwitchTo(oldcxt);
3386  }
3387 
3388  funccxt = SRF_PERCALL_SETUP();
3389  multi = (mxact *) funccxt->user_fctx;
3390 
3391  while (multi->iter < multi->nmembers)
3392  {
3393  HeapTuple tuple;
3394  char *values[2];
3395 
3396  values[0] = psprintf("%u", multi->members[multi->iter].xid);
3397  values[1] = mxstatus_to_string(multi->members[multi->iter].status);
3398 
3399  tuple = BuildTupleFromCStrings(funccxt->attinmeta, values);
3400 
3401  multi->iter++;
3402  pfree(values[0]);
3403  SRF_RETURN_NEXT(funccxt, HeapTupleGetDatum(tuple));
3404  }
3405 
3406  SRF_RETURN_DONE(funccxt);
3407 }
3408 
3409 /*
3410  * Entrypoint for sync.c to sync offsets files.
3411  */
3412 int
3413 multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
3414 {
3415  return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
3416 }
3417 
3418 /*
3419  * Entrypoint for sync.c to sync members files.
3420  */
3421 int
3422 multixactmemberssyncfiletag(const FileTag *ftag, char *path)
3423 {
3424  return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
3425 }
int16 AttrNumber
Definition: attnum.h:21
int autovacuum_multixact_freeze_max_age
Definition: autovacuum.c:127
int BackendId
Definition: backendid.h:21
static int32 next
Definition: blutils.c:219
static Datum values[MAXATTR]
Definition: bootstrap.c:156
unsigned short uint16
Definition: c.h:441
unsigned int uint32
Definition: c.h:442
signed int int32
Definition: c.h:430
uint32 MultiXactOffset
Definition: c.h:600
TransactionId MultiXactId
Definition: c.h:598
#define FLEXIBLE_ARRAY_MEMBER
Definition: c.h:362
unsigned char uint8
Definition: c.h:440
#define MemSet(start, val, len)
Definition: c.h:953
uint32 TransactionId
Definition: c.h:588
size_t Size
Definition: c.h:541
char * get_database_name(Oid dbid)
Definition: dbcommands.c:2981
int errmsg_plural(const char *fmt_singular, const char *fmt_plural, unsigned long n,...)
Definition: elog.c:1016
int errmsg_internal(const char *fmt,...)
Definition: elog.c:993
int errdetail_plural(const char *fmt_singular, const char *fmt_plural, unsigned long n,...)
Definition: elog.c:1131
int errhint(const char *fmt,...)
Definition: elog.c:1153
int errcode(int sqlerrcode)
Definition: elog.c:695
int errmsg(const char *fmt,...)
Definition: elog.c:906
#define LOG
Definition: elog.h:27
#define WARNING
Definition: elog.h:32
#define DEBUG2
Definition: elog.h:25
#define PANIC
Definition: elog.h:38
#define DEBUG1
Definition: elog.h:26
#define ERROR
Definition: elog.h:35
#define ereport(elevel,...)
Definition: elog.h:145
HeapTuple BuildTupleFromCStrings(AttInMetadata *attinmeta, char **values)
Definition: execTuples.c:2135
AttInMetadata * TupleDescGetAttInMetadata(TupleDesc tupdesc)
Definition: execTuples.c:2086
#define PG_GETARG_TRANSACTIONID(n)
Definition: fmgr.h:279
#define PG_FUNCTION_ARGS
Definition: fmgr.h:193
#define SRF_IS_FIRSTCALL()
Definition: funcapi.h:303
#define SRF_PERCALL_SETUP()
Definition: funcapi.h:307
#define SRF_RETURN_NEXT(_funcctx, _result)
Definition: funcapi.h:309
#define SRF_FIRSTCALL_INIT()
Definition: funcapi.h:305
static Datum HeapTupleGetDatum(const HeapTupleData *tuple)
Definition: funcapi.h:230
#define SRF_RETURN_DONE(_funcctx)
Definition: funcapi.h:327
Datum difference(PG_FUNCTION_ARGS)
bool IsBinaryUpgrade
Definition: globals.c:114
bool IsUnderPostmaster
Definition: globals.c:113
BackendId MyBackendId
Definition: globals.c:85
#define dclist_container(type, membername, ptr)
Definition: ilist.h:884
static dlist_node * dclist_tail_node(dclist_head *head)
Definition: ilist.h:857
static void dclist_move_head(dclist_head *head, dlist_node *node)
Definition: ilist.h:745
static uint32 dclist_count(dclist_head *head)
Definition: ilist.h:869
static void dclist_delete_from(dclist_head *head, dlist_node *node)
Definition: ilist.h:713
#define DCLIST_STATIC_INIT(name)
Definition: ilist.h:282
static void dclist_push_head(dclist_head *head, dlist_node *node)
Definition: ilist.h:643
static void dclist_init(dclist_head *head)
Definition: ilist.h:621
#define dclist_foreach(iter, lhead)
Definition: ilist.h:907
int j
Definition: isn.c:74
int i
Definition: isn.c:73
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:77
Assert(fmt[strlen(fmt) - 1] !='\n')
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1194
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1802
@ LWTRANCHE_MULTIXACTMEMBER_BUFFER
Definition: lwlock.h:181
@ LWTRANCHE_MULTIXACTOFFSET_BUFFER
Definition: lwlock.h:180
@ LW_SHARED
Definition: lwlock.h:113
@ LW_EXCLUSIVE
Definition: lwlock.h:112
MemoryContext TopTransactionContext
Definition: mcxt.c:135
void pfree(void *pointer)
Definition: mcxt.c:1306
MemoryContext TopMemoryContext
Definition: mcxt.c:130
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:994
char * MemoryContextStrdup(MemoryContext context, const char *string)
Definition: mcxt.c:1470
void * palloc(Size size)
Definition: mcxt.c:1199
#define AllocSetContextCreate
Definition: memutils.h:129
#define ALLOCSET_SMALL_SIZES
Definition: memutils.h:163
#define START_CRIT_SECTION()
Definition: miscadmin.h:148
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:121
#define END_CRIT_SECTION()
Definition: miscadmin.h:150
#define MXOffsetToMemberPage(xid)
Definition: multixact.c:159
static void WriteMTruncateXlogRec(Oid oldestMultiDB, MultiXactId startTruncOff, MultiXactId endTruncOff, MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
Definition: multixact.c:3209
static SlruCtlData MultiXactOffsetCtlData
Definition: multixact.c:187
void MultiXactShmemInit(void)
Definition: multixact.c:1842
#define MULTIXACT_MEMBER_SAFE_THRESHOLD
Definition: multixact.c:177
static int ZeroMultiXactMemberPage(int pageno, bool writeXlog)
Definition: multixact.c:1945
static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
Definition: multixact.c:956
static int ZeroMultiXactOffsetPage(int pageno, bool writeXlog)
Definition: multixact.c:1929
static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
Definition: multixact.c:1540
MultiXactId MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
Definition: multixact.c:438
#define MXACT_MEMBER_BITS_PER_XACT
Definition: multixact.c:130
static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
Definition: multixact.c:2432
void ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
Definition: multixact.c:742
static void PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
Definition: multixact.c:2917
#define MXACT_MEMBER_XACT_BITMASK
Definition: multixact.c:132
#define MultiXactIdToOffsetEntry(xid)
Definition: multixact.c:113
bool MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
Definition: multixact.c:3157
void multixact_redo(XLogReaderState *record)
Definition: multixact.c:3234
#define MULTIXACT_OFFSETS_PER_PAGE
Definition: multixact.c:109
#define debug_elog5(a, b, c, d, e)
Definition: multixact.c:335
static void MultiXactIdSetOldestVisible(void)
Definition: multixact.c:681
int multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
Definition: multixact.c:3413
void multixact_twophase_postcommit(TransactionId xid, uint16 info, void *recdata, uint32 len)
Definition: multixact.c:1798
static bool MultiXactOffsetPagePrecedes(int page1, int page2)
Definition: multixact.c:3117
static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result)
Definition: multixact.c:2732
void MultiXactSetNextMXact(MultiXactId nextMulti, MultiXactOffset nextMultiOffset)
Definition: multixact.c:2177
void multixact_twophase_recover(TransactionId xid, uint16 info, void *recdata, uint32 len)
Definition: multixact.c:1777
#define MultiXactMemberCtl
Definition: multixact.c:191
void AtPrepare_MultiXact(void)
Definition: multixact.c:1714
static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start, uint32 distance)
Definition: multixact.c:2684
bool MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
Definition: multixact.c:3171
#define MXOffsetToFlagsBitShift(xid)
Definition: multixact.c:167
void MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
Definition: multixact.c:2385
static bool SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: multixact.c:2866
static void mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
Definition: multixact.c:1587
#define MXOffsetToFlagsOffset(xid)
Definition: multixact.c:163
static void MaybeExtendOffsetSlru(void)
Definition: multixact.c:1973
bool MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
Definition: multixact.c:550
void MultiXactIdSetOldestMember(void)
Definition: multixact.c:624
static void PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
Definition: multixact.c:2889
static MemoryContext MXactContext
Definition: multixact.c:323
#define SHARED_MULTIXACT_STATE_SIZE
static MultiXactId * OldestVisibleMXactId
Definition: multixact.c:293
struct mxtruncinfo mxtruncinfo
static int mxactMemberComparator(const void *arg1, const void *arg2)
Definition: multixact.c:1467
struct MultiXactStateData MultiXactStateData
static void ExtendMultiXactOffset(MultiXactId multi)
Definition: multixact.c:2402
void PostPrepare_MultiXact(TransactionId xid)
Definition: multixact.c:1728
Size MultiXactShmemSize(void)
Definition: multixact.c:1825
#define MultiXactOffsetCtl
Definition: multixact.c:190
void multixact_twophase_postabort(TransactionId xid, uint16 info, void *recdata, uint32 len)
Definition: multixact.c:1813
#define PreviousMultiXactId(xid)
Definition: multixact.c:181
void MultiXactGetCheckptMulti(bool is_shutdown, MultiXactId *nextMulti, MultiXactOffset *nextMultiOffset, MultiXactId *oldestMulti, Oid *oldestMultiDB)
Definition: multixact.c:2131
void SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid, bool is_startup)
Definition: multixact.c:2211
static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset, int nmembers, MultiXactMember *members)
Definition: multixact.c:862
int multixactmemberssyncfiletag(const FileTag *ftag, char *path)
Definition: multixact.c:3422
#define MAX_CACHE_ENTRIES
Definition: multixact.c:321
#define MXOffsetToMemberOffset(xid)
Definition: multixact.c:172
static void WriteMZeroPageXlogRec(int pageno, uint8 info)
Definition: multixact.c:3195
MultiXactId GetOldestMultiXactId(void)
Definition: multixact.c:2504
void CheckPointMultiXact(void)
Definition: multixact.c:2153
#define MaxOldestSlot
Definition: multixact.c:288
MultiXactId MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
Definition: multixact.c:766
static bool ReadMultiXactCounts(uint32 *multixacts, MultiXactOffset *members)
Definition: multixact.c:2770
struct mXactCacheEnt mXactCacheEnt
static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members)
Definition: multixact.c:1497
static dclist_head MXactCache
Definition: multixact.c:322
#define MXOffsetToMemberSegment(xid)
Definition: multixact.c:160
void TrimMultiXact(void)
Definition: multixact.c:2029
char * mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
Definition: multixact.c:1655
#define debug_elog3(a, b, c)
Definition: multixact.c:333
#define debug_elog4(a, b, c, d)
Definition: multixact.c:334
static bool SetOffsetVacuumLimit(bool is_startup)
Definition: multixact.c:2557
int MultiXactMemberFreezeThreshold(void)
Definition: multixact.c:2826
void MultiXactAdvanceNextMXact(MultiXactId minMulti, MultiXactOffset minMultiOffset)
Definition: multixact.c:2360
static MultiXactId * OldestMemberMXactId
Definition: multixact.c:292
#define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE
Definition: multixact.c:155
static MultiXactStateData * MultiXactState
Definition: multixact.c:291
static char * mxstatus_to_string(MultiXactStatus status)
Definition: multixact.c:1632
#define OFFSET_WARN_SEGMENTS
Datum pg_get_multixact_members(PG_FUNCTION_ARGS)
Definition: multixact.c:3345
MultiXactId ReadNextMultiXactId(void)
Definition: multixact.c:722
void BootStrapMultiXact(void)
Definition: multixact.c:1892
#define debug_elog6(a, b, c, d, e, f)
Definition: multixact.c:336
#define MULTIXACT_MEMBERS_PER_PAGE
Definition: multixact.c:142
MultiXactId MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1, TransactionId xid2, MultiXactStatus status2)
Definition: multixact.c:385
static bool MultiXactMemberPagePrecedes(int page1, int page2)
Definition: multixact.c:3137
void TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
Definition: multixact.c:2942
#define MULTIXACT_MEMBER_DANGER_THRESHOLD
Definition: multixact.c:178
static bool MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
Definition: multixact.c:3183
#define MultiXactIdToOffsetSegment(xid)
Definition: multixact.c:115
void AtEOXact_MultiXact(void)
Definition: multixact.c:1686
#define MultiXactIdToOffsetPage(xid)
Definition: multixact.c:111
static SlruCtlData MultiXactMemberCtlData
Definition: multixact.c:188
#define debug_elog2(a, b)
Definition: multixact.c:332
void StartupMultiXact(void)
Definition: multixact.c:2006
int GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members, bool from_pgupgrade, bool isLockOnly)
Definition: multixact.c:1223
#define MultiXactIdIsValid(multi)
Definition: multixact.h:28
#define XLOG_MULTIXACT_ZERO_MEM_PAGE
Definition: multixact.h:73
#define XLOG_MULTIXACT_ZERO_OFF_PAGE
Definition: multixact.h:72
#define FirstMultiXactId
Definition: multixact.h:25
#define NUM_MULTIXACTOFFSET_BUFFERS
Definition: multixact.h:33
MultiXactStatus
Definition: multixact.h:42
@ MultiXactStatusForShare
Definition: multixact.h:44
@ MultiXactStatusForNoKeyUpdate
Definition: multixact.h:45
@ MultiXactStatusNoKeyUpdate
Definition: multixact.h:48
@ MultiXactStatusUpdate
Definition: multixact.h:50
@ MultiXactStatusForUpdate
Definition: multixact.h:46
@ MultiXactStatusForKeyShare
Definition: multixact.h:43
#define ISUPDATE_from_mxstatus(status)
Definition: multixact.h:56
#define InvalidMultiXactId
Definition: multixact.h:24
#define XLOG_MULTIXACT_TRUNCATE_ID
Definition: multixact.h:75
#define SizeOfMultiXactCreate
Definition: multixact.h:85
#define SizeOfMultiXactTruncate
Definition: multixact.h:100
#define XLOG_MULTIXACT_CREATE_ID
Definition: multixact.h:74
#define MaxMultiXactOffset
Definition: multixact.h:30
#define MaxMultiXactId
Definition: multixact.h:26
#define NUM_MULTIXACTMEMBER_BUFFERS
Definition: multixact.h:34
struct MultiXactMember MultiXactMember
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:135
const void size_t len
const void * data
while(p+4<=pend)
static char * filename
Definition: pg_dumpall.c:119
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:225
static char * buf
Definition: pg_test_fsync.c:67
void SendPostmasterSignal(PMSignalReason reason)
Definition: pmsignal.c:181
@ PMSIGNAL_START_AUTOVAC_LAUNCHER
Definition: pmsignal.h:38
#define qsort(a, b, c, d)
Definition: port.h:445
uintptr_t Datum
Definition: postgres.h:412
unsigned int Oid
Definition: postgres_ext.h:31
#define DELAY_CHKPT_START
Definition: proc.h:119
bool TransactionIdIsInProgress(TransactionId xid)
Definition: procarray.c:1386
char * psprintf(const char *fmt,...)
Definition: psprintf.c:46
Size add_size(Size s1, Size s2)
Definition: shmem.c:502
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:396
void pg_usleep(long microsec)
Definition: signal.c:53
void SimpleLruWritePage(SlruCtl ctl, int slotno)
Definition: slru.c:615
void SimpleLruWriteAll(SlruCtl ctl, bool allow_redirtied)
Definition: slru.c:1157
void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, LWLock *ctllock, const char *subdir, int tranche_id, SyncRequestHandler sync_handler)
Definition: slru.c:188
void SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
Definition: slru.c:1227
bool SlruScanDirectory(SlruCtl ctl, SlruScanCallback callback, void *data)
Definition: slru.c:1554
int SlruSyncFileTag(SlruCtl ctl, const FileTag *ftag, char *path)
Definition: slru.c:1594
int SimpleLruZeroPage(SlruCtl ctl, int pageno)
Definition: slru.c:281
void SlruDeleteSegment(SlruCtl ctl, int segno)
Definition: slru.c:1328
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int pageno, TransactionId xid)
Definition: slru.c:496
bool SimpleLruDoesPhysicalPageExist(SlruCtl ctl, int pageno)
Definition: slru.c:627
int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok, TransactionId xid)
Definition: slru.c:396
Size SimpleLruShmemSize(int nslots, int nlsns)
Definition: slru.c:156
#define SlruPagePrecedesUnitTests(ctl, per_page)
Definition: slru.h:156
#define SLRU_PAGES_PER_SEGMENT
Definition: slru.h:34
PGPROC * MyProc
Definition: proc.c:68
void appendStringInfo(StringInfo str, const char *fmt,...)
Definition: stringinfo.c:91
void appendStringInfoChar(StringInfo str, char ch)
Definition: stringinfo.c:188
void initStringInfo(StringInfo str)
Definition: stringinfo.c:59
Definition: sync.h:51
void * user_fctx
Definition: funcapi.h:82
AttInMetadata * attinmeta
Definition: funcapi.h:91
MemoryContext multi_call_memory_ctx
Definition: funcapi.h:101
TransactionId xid
Definition: multixact.h:62
MultiXactStatus status
Definition: multixact.h:63
MultiXactId multiWrapLimit
Definition: multixact.c:231
MultiXactId multiStopLimit
Definition: multixact.c:230
MultiXactId multiWarnLimit
Definition: multixact.c:229
MultiXactId multiVacLimit
Definition: multixact.c:228
MultiXactOffset offsetStopLimit
Definition: multixact.c:234
MultiXactOffset nextOffset
Definition: multixact.c:206
MultiXactId nextMXact
Definition: multixact.c:203
MultiXactId oldestMultiXactId
Definition: multixact.c:216
MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER]
Definition: multixact.c:281
MultiXactOffset oldestOffset
Definition: multixact.c:224
int delayChkptFlags
Definition: proc.h:231
bool(* PagePrecedes)(int, int)
Definition: slru.h:130
dlist_node * cur
Definition: ilist.h:179
MultiXactId multi
Definition: multixact.c:315
dlist_node node
Definition: multixact.c:317
MultiXactMember members[FLEXIBLE_ARRAY_MEMBER]
Definition: multixact.c:318
int earliestExistingPage
Definition: multixact.c:2858
MultiXactId mid
Definition: multixact.h:79
MultiXactMember members[FLEXIBLE_ARRAY_MEMBER]
Definition: multixact.h:82
MultiXactOffset moff
Definition: multixact.h:80
MultiXactId endTruncOff
Definition: multixact.h:93
MultiXactOffset startTruncMemb
Definition: multixact.h:96
MultiXactOffset endTruncMemb
Definition: multixact.h:97
MultiXactId startTruncOff
Definition: multixact.h:92
@ SYNC_HANDLER_MULTIXACT_MEMBER
Definition: sync.h:41
@ SYNC_HANDLER_MULTIXACT_OFFSET
Definition: sync.h:40
bool TransactionIdDidCommit(TransactionId transactionId)
Definition: transam.c:125
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:273
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
#define TransactionIdIsValid(xid)
Definition: transam.h:41
TupleDesc CreateTemplateTupleDesc(int natts)
Definition: tupdesc.c:45
void TupleDescInitEntry(TupleDesc desc, AttrNumber attributeNumber, const char *attributeName, Oid oidtypeid, int32 typmod, int attdim)
Definition: tupdesc.c:583
void RegisterTwoPhaseRecord(TwoPhaseRmgrId rmid, uint16 info, const void *data, uint32 len)
Definition: twophase.c:1257
BackendId TwoPhaseGetDummyBackendId(TransactionId xid, bool lock_held)
Definition: twophase.c:918
#define TWOPHASE_RM_MULTIXACT_ID
Definition: twophase_rmgr.h:27
void AdvanceNextFullTransactionIdPastXid(TransactionId xid)
Definition: varsup.c:277
bool IsTransactionState(void)
Definition: xact.c:377
bool TransactionIdIsCurrentTransactionId(TransactionId xid)
Definition: xact.c:925
bool RecoveryInProgress(void)
Definition: xlog.c:5912
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2512
uint64 XLogRecPtr
Definition: xlogdefs.h:21
void XLogRegisterData(char *data, uint32 len)
Definition: xloginsert.c:351
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:451
void XLogBeginInsert(void)
Definition: xloginsert.c:150
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:411
#define XLogRecGetData(decoder)
Definition: xlogreader.h:416
#define XLogRecGetXid(decoder)
Definition: xlogreader.h:413
#define XLogRecHasAnyBlockRefs(decoder)
Definition: xlogreader.h:418
#define XLR_INFO_MASK
Definition: xlogrecord.h:62
bool InRecovery
Definition: xlogutils.c:53