Line data Source code
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) 1996-2025, 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/twophase.h"
74 : #include "access/twophase_rmgr.h"
75 : #include "access/xlog.h"
76 : #include "access/xloginsert.h"
77 : #include "access/xlogutils.h"
78 : #include "miscadmin.h"
79 : #include "pg_trace.h"
80 : #include "pgstat.h"
81 : #include "postmaster/autovacuum.h"
82 : #include "storage/pmsignal.h"
83 : #include "storage/proc.h"
84 : #include "storage/procarray.h"
85 : #include "utils/guc_hooks.h"
86 : #include "utils/injection_point.h"
87 : #include "utils/lsyscache.h"
88 : #include "utils/memutils.h"
89 :
90 :
91 : /*
92 : * Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
93 : * used everywhere else in Postgres.
94 : *
95 : * Note: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
96 : * MultiXact page numbering also wraps around at
97 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
98 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need
99 : * take no explicit notice of that fact in this module, except when comparing
100 : * segment and page numbers in TruncateMultiXact (see
101 : * MultiXactOffsetPagePrecedes).
102 : */
103 :
104 : /* We need four bytes per offset */
105 : #define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
106 :
107 : static inline int64
108 4874 : MultiXactIdToOffsetPage(MultiXactId multi)
109 : {
110 4874 : return multi / MULTIXACT_OFFSETS_PER_PAGE;
111 : }
112 :
113 : static inline int
114 3412 : MultiXactIdToOffsetEntry(MultiXactId multi)
115 : {
116 3412 : return multi % MULTIXACT_OFFSETS_PER_PAGE;
117 : }
118 :
119 : static inline int64
120 0 : MultiXactIdToOffsetSegment(MultiXactId multi)
121 : {
122 0 : return MultiXactIdToOffsetPage(multi) / SLRU_PAGES_PER_SEGMENT;
123 : }
124 :
125 : /*
126 : * The situation for members is a bit more complex: we store one byte of
127 : * additional flag bits for each TransactionId. To do this without getting
128 : * into alignment issues, we store four bytes of flags, and then the
129 : * corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
130 : * are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
131 : * per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
132 : * performance) trumps space efficiency here.
133 : *
134 : * Note that the "offset" macros work with byte offset, not array indexes, so
135 : * arithmetic must be done using "char *" pointers.
136 : */
137 : /* We need eight bits per xact, so one xact fits in a byte */
138 : #define MXACT_MEMBER_BITS_PER_XACT 8
139 : #define MXACT_MEMBER_FLAGS_PER_BYTE 1
140 : #define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
141 :
142 : /* how many full bytes of flags are there in a group? */
143 : #define MULTIXACT_FLAGBYTES_PER_GROUP 4
144 : #define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
145 : (MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
146 : /* size in bytes of a complete group */
147 : #define MULTIXACT_MEMBERGROUP_SIZE \
148 : (sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
149 : #define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
150 : #define MULTIXACT_MEMBERS_PER_PAGE \
151 : (MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
152 :
153 : /*
154 : * Because the number of items per page is not a divisor of the last item
155 : * number (member 0xFFFFFFFF), the last segment does not use the maximum number
156 : * of pages, and moreover the last used page therein does not use the same
157 : * number of items as previous pages. (Another way to say it is that the
158 : * 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
159 : * has some empty space after that item.)
160 : *
161 : * This constant is the number of members in the last page of the last segment.
162 : */
163 : #define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
164 : ((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
165 :
166 : /* page in which a member is to be found */
167 : static inline int64
168 6072 : MXOffsetToMemberPage(MultiXactOffset offset)
169 : {
170 6072 : return offset / MULTIXACT_MEMBERS_PER_PAGE;
171 : }
172 :
173 : static inline int64
174 0 : MXOffsetToMemberSegment(MultiXactOffset offset)
175 : {
176 0 : return MXOffsetToMemberPage(offset) / SLRU_PAGES_PER_SEGMENT;
177 : }
178 :
179 : /* Location (byte offset within page) of flag word for a given member */
180 : static inline int
181 7114 : MXOffsetToFlagsOffset(MultiXactOffset offset)
182 : {
183 7114 : MultiXactOffset group = offset / MULTIXACT_MEMBERS_PER_MEMBERGROUP;
184 7114 : int grouponpg = group % MULTIXACT_MEMBERGROUPS_PER_PAGE;
185 7114 : int byteoff = grouponpg * MULTIXACT_MEMBERGROUP_SIZE;
186 :
187 7114 : return byteoff;
188 : }
189 :
190 : static inline int
191 2932 : MXOffsetToFlagsBitShift(MultiXactOffset offset)
192 : {
193 2932 : int member_in_group = offset % MULTIXACT_MEMBERS_PER_MEMBERGROUP;
194 2932 : int bshift = member_in_group * MXACT_MEMBER_BITS_PER_XACT;
195 :
196 2932 : return bshift;
197 : }
198 :
199 : /* Location (byte offset within page) of TransactionId of given member */
200 : static inline int
201 2410 : MXOffsetToMemberOffset(MultiXactOffset offset)
202 : {
203 2410 : int member_in_group = offset % MULTIXACT_MEMBERS_PER_MEMBERGROUP;
204 :
205 2410 : return MXOffsetToFlagsOffset(offset) +
206 2410 : MULTIXACT_FLAGBYTES_PER_GROUP +
207 : member_in_group * sizeof(TransactionId);
208 : }
209 :
210 : /* Multixact members wraparound thresholds. */
211 : #define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
212 : #define MULTIXACT_MEMBER_DANGER_THRESHOLD \
213 : (MaxMultiXactOffset - MaxMultiXactOffset / 4)
214 :
215 : static inline MultiXactId
216 0 : PreviousMultiXactId(MultiXactId multi)
217 : {
218 0 : return multi == FirstMultiXactId ? MaxMultiXactId : multi - 1;
219 : }
220 :
221 : /*
222 : * Links to shared-memory data structures for MultiXact control
223 : */
224 : static SlruCtlData MultiXactOffsetCtlData;
225 : static SlruCtlData MultiXactMemberCtlData;
226 :
227 : #define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
228 : #define MultiXactMemberCtl (&MultiXactMemberCtlData)
229 :
230 : /*
231 : * MultiXact state shared across all backends. All this state is protected
232 : * by MultiXactGenLock. (We also use SLRU bank's lock of MultiXactOffset and
233 : * MultiXactMember to guard accesses to the two sets of SLRU buffers. For
234 : * concurrency's sake, we avoid holding more than one of these locks at a
235 : * time.)
236 : */
237 : typedef struct MultiXactStateData
238 : {
239 : /* next-to-be-assigned MultiXactId */
240 : MultiXactId nextMXact;
241 :
242 : /* next-to-be-assigned offset */
243 : MultiXactOffset nextOffset;
244 :
245 : /* Have we completed multixact startup? */
246 : bool finishedStartup;
247 :
248 : /*
249 : * Oldest multixact that is still potentially referenced by a relation.
250 : * Anything older than this should not be consulted. These values are
251 : * updated by vacuum.
252 : */
253 : MultiXactId oldestMultiXactId;
254 : Oid oldestMultiXactDB;
255 :
256 : /*
257 : * Oldest multixact offset that is potentially referenced by a multixact
258 : * referenced by a relation. We don't always know this value, so there's
259 : * a flag here to indicate whether or not we currently do.
260 : */
261 : MultiXactOffset oldestOffset;
262 : bool oldestOffsetKnown;
263 :
264 : /* support for anti-wraparound measures */
265 : MultiXactId multiVacLimit;
266 : MultiXactId multiWarnLimit;
267 : MultiXactId multiStopLimit;
268 : MultiXactId multiWrapLimit;
269 :
270 : /* support for members anti-wraparound measures */
271 : MultiXactOffset offsetStopLimit; /* known if oldestOffsetKnown */
272 :
273 : /*
274 : * This is used to sleep until a multixact offset is written when we want
275 : * to create the next one.
276 : */
277 : ConditionVariable nextoff_cv;
278 :
279 : /*
280 : * Per-backend data starts here. We have two arrays stored in the area
281 : * immediately following the MultiXactStateData struct. Each is indexed by
282 : * ProcNumber.
283 : *
284 : * In both arrays, there's a slot for all normal backends
285 : * (0..MaxBackends-1) followed by a slot for max_prepared_xacts prepared
286 : * transactions.
287 : *
288 : * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
289 : * transaction(s) could possibly be a member of, or InvalidMultiXactId
290 : * when the backend has no live transaction that could possibly be a
291 : * member of a MultiXact. Each backend sets its entry to the current
292 : * nextMXact counter just before first acquiring a shared lock in a given
293 : * transaction, and clears it at transaction end. (This works because only
294 : * during or after acquiring a shared lock could an XID possibly become a
295 : * member of a MultiXact, and that MultiXact would have to be created
296 : * during or after the lock acquisition.)
297 : *
298 : * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
299 : * current transaction(s) think is potentially live, or InvalidMultiXactId
300 : * when not in a transaction or not in a transaction that's paid any
301 : * attention to MultiXacts yet. This is computed when first needed in a
302 : * given transaction, and cleared at transaction end. We can compute it
303 : * as the minimum of the valid OldestMemberMXactId[] entries at the time
304 : * we compute it (using nextMXact if none are valid). Each backend is
305 : * required not to attempt to access any SLRU data for MultiXactIds older
306 : * than its own OldestVisibleMXactId[] setting; this is necessary because
307 : * the relevant SLRU data can be concurrently truncated away.
308 : *
309 : * The oldest valid value among all of the OldestMemberMXactId[] and
310 : * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
311 : * possible value still having any live member transaction -- OldestMxact.
312 : * Any value older than that is typically removed from tuple headers, or
313 : * "frozen" via being replaced with a new xmax. VACUUM can sometimes even
314 : * remove an individual MultiXact xmax whose value is >= its OldestMxact
315 : * cutoff, though typically only when no individual member XID is still
316 : * running. See FreezeMultiXactId for full details.
317 : *
318 : * Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
319 : * or the oldest extant Multi remaining in the table is used as the new
320 : * pg_class.relminmxid value (whichever is earlier). The minimum of all
321 : * relminmxid values in each database is stored in pg_database.datminmxid.
322 : * In turn, the minimum of all of those values is stored in pg_control.
323 : * This is used as the truncation point for pg_multixact when unneeded
324 : * segments get removed by vac_truncate_clog() during vacuuming.
325 : */
326 : MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
327 : } MultiXactStateData;
328 :
329 : /*
330 : * Size of OldestMemberMXactId and OldestVisibleMXactId arrays.
331 : */
332 : #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
333 :
334 : /* Pointers to the state data in shared memory */
335 : static MultiXactStateData *MultiXactState;
336 : static MultiXactId *OldestMemberMXactId;
337 : static MultiXactId *OldestVisibleMXactId;
338 :
339 :
340 : /*
341 : * Definitions for the backend-local MultiXactId cache.
342 : *
343 : * We use this cache to store known MultiXacts, so we don't need to go to
344 : * SLRU areas every time.
345 : *
346 : * The cache lasts for the duration of a single transaction, the rationale
347 : * for this being that most entries will contain our own TransactionId and
348 : * so they will be uninteresting by the time our next transaction starts.
349 : * (XXX not clear that this is correct --- other members of the MultiXact
350 : * could hang around longer than we did. However, it's not clear what a
351 : * better policy for flushing old cache entries would be.) FIXME actually
352 : * this is plain wrong now that multixact's may contain update Xids.
353 : *
354 : * We allocate the cache entries in a memory context that is deleted at
355 : * transaction end, so we don't need to do retail freeing of entries.
356 : */
357 : typedef struct mXactCacheEnt
358 : {
359 : MultiXactId multi;
360 : int nmembers;
361 : dlist_node node;
362 : MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
363 : } mXactCacheEnt;
364 :
365 : #define MAX_CACHE_ENTRIES 256
366 : static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
367 : static MemoryContext MXactContext = NULL;
368 :
369 : #ifdef MULTIXACT_DEBUG
370 : #define debug_elog2(a,b) elog(a,b)
371 : #define debug_elog3(a,b,c) elog(a,b,c)
372 : #define debug_elog4(a,b,c,d) elog(a,b,c,d)
373 : #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
374 : #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
375 : #else
376 : #define debug_elog2(a,b)
377 : #define debug_elog3(a,b,c)
378 : #define debug_elog4(a,b,c,d)
379 : #define debug_elog5(a,b,c,d,e)
380 : #define debug_elog6(a,b,c,d,e,f)
381 : #endif
382 :
383 : /* internal MultiXactId management */
384 : static void MultiXactIdSetOldestVisible(void);
385 : static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
386 : int nmembers, MultiXactMember *members);
387 : static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
388 :
389 : /* MultiXact cache management */
390 : static int mxactMemberComparator(const void *arg1, const void *arg2);
391 : static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
392 : static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
393 : static void mXactCachePut(MultiXactId multi, int nmembers,
394 : MultiXactMember *members);
395 :
396 : /* management of SLRU infrastructure */
397 : static bool MultiXactOffsetPagePrecedes(int64 page1, int64 page2);
398 : static bool MultiXactMemberPagePrecedes(int64 page1, int64 page2);
399 : static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
400 : MultiXactOffset offset2);
401 : static void ExtendMultiXactOffset(MultiXactId multi);
402 : static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
403 : static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
404 : MultiXactOffset start, uint32 distance);
405 : static bool SetOffsetVacuumLimit(bool is_startup);
406 : static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
407 : static void WriteMTruncateXlogRec(Oid oldestMultiDB,
408 : MultiXactId startTruncOff,
409 : MultiXactId endTruncOff,
410 : MultiXactOffset startTruncMemb,
411 : MultiXactOffset endTruncMemb);
412 :
413 :
414 : /*
415 : * MultiXactIdCreate
416 : * Construct a MultiXactId representing two TransactionIds.
417 : *
418 : * The two XIDs must be different, or be requesting different statuses.
419 : *
420 : * NB - we don't worry about our local MultiXactId cache here, because that
421 : * is handled by the lower-level routines.
422 : */
423 : MultiXactId
424 2064 : MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
425 : TransactionId xid2, MultiXactStatus status2)
426 : {
427 : MultiXactId newMulti;
428 : MultiXactMember members[2];
429 :
430 : Assert(TransactionIdIsValid(xid1));
431 : Assert(TransactionIdIsValid(xid2));
432 :
433 : Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
434 :
435 : /* MultiXactIdSetOldestMember() must have been called already. */
436 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
437 :
438 : /*
439 : * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
440 : * are still running. In typical usage, xid2 will be our own XID and the
441 : * caller just did a check on xid1, so it'd be wasted effort.
442 : */
443 :
444 2064 : members[0].xid = xid1;
445 2064 : members[0].status = status1;
446 2064 : members[1].xid = xid2;
447 2064 : members[1].status = status2;
448 :
449 2064 : newMulti = MultiXactIdCreateFromMembers(2, members);
450 :
451 : debug_elog3(DEBUG2, "Create: %s",
452 : mxid_to_string(newMulti, 2, members));
453 :
454 2064 : return newMulti;
455 : }
456 :
457 : /*
458 : * MultiXactIdExpand
459 : * Add a TransactionId to a pre-existing MultiXactId.
460 : *
461 : * If the TransactionId is already a member of the passed MultiXactId with the
462 : * same status, just return it as-is.
463 : *
464 : * Note that we do NOT actually modify the membership of a pre-existing
465 : * MultiXactId; instead we create a new one. This is necessary to avoid
466 : * a race condition against code trying to wait for one MultiXactId to finish;
467 : * see notes in heapam.c.
468 : *
469 : * NB - we don't worry about our local MultiXactId cache here, because that
470 : * is handled by the lower-level routines.
471 : *
472 : * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
473 : * one upgraded by pg_upgrade from a cluster older than this feature) are not
474 : * passed in.
475 : */
476 : MultiXactId
477 178 : MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
478 : {
479 : MultiXactId newMulti;
480 : MultiXactMember *members;
481 : MultiXactMember *newMembers;
482 : int nmembers;
483 : int i;
484 : int j;
485 :
486 : Assert(MultiXactIdIsValid(multi));
487 : Assert(TransactionIdIsValid(xid));
488 :
489 : /* MultiXactIdSetOldestMember() must have been called already. */
490 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
491 :
492 : debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
493 : multi, xid, mxstatus_to_string(status));
494 :
495 : /*
496 : * Note: we don't allow for old multis here. The reason is that the only
497 : * caller of this function does a check that the multixact is no longer
498 : * running.
499 : */
500 178 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
501 :
502 178 : if (nmembers < 0)
503 : {
504 : MultiXactMember member;
505 :
506 : /*
507 : * The MultiXactId is obsolete. This can only happen if all the
508 : * MultiXactId members stop running between the caller checking and
509 : * passing it to us. It would be better to return that fact to the
510 : * caller, but it would complicate the API and it's unlikely to happen
511 : * too often, so just deal with it by creating a singleton MultiXact.
512 : */
513 0 : member.xid = xid;
514 0 : member.status = status;
515 0 : newMulti = MultiXactIdCreateFromMembers(1, &member);
516 :
517 : debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
518 : multi, newMulti);
519 0 : return newMulti;
520 : }
521 :
522 : /*
523 : * If the TransactionId is already a member of the MultiXactId with the
524 : * same status, just return the existing MultiXactId.
525 : */
526 548 : for (i = 0; i < nmembers; i++)
527 : {
528 370 : if (TransactionIdEquals(members[i].xid, xid) &&
529 108 : (members[i].status == status))
530 : {
531 : debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
532 : xid, multi);
533 0 : pfree(members);
534 0 : return multi;
535 : }
536 : }
537 :
538 : /*
539 : * Determine which of the members of the MultiXactId are still of
540 : * interest. This is any running transaction, and also any transaction
541 : * that grabbed something stronger than just a lock and was committed. (An
542 : * update that aborted is of no interest here; and having more than one
543 : * update Xid in a multixact would cause errors elsewhere.)
544 : *
545 : * Removing dead members is not just an optimization: freezing of tuples
546 : * whose Xmax are multis depends on this behavior.
547 : *
548 : * Note we have the same race condition here as above: j could be 0 at the
549 : * end of the loop.
550 : */
551 : newMembers = (MultiXactMember *)
552 178 : palloc(sizeof(MultiXactMember) * (nmembers + 1));
553 :
554 548 : for (i = 0, j = 0; i < nmembers; i++)
555 : {
556 370 : if (TransactionIdIsInProgress(members[i].xid) ||
557 78 : (ISUPDATE_from_mxstatus(members[i].status) &&
558 12 : TransactionIdDidCommit(members[i].xid)))
559 : {
560 304 : newMembers[j].xid = members[i].xid;
561 304 : newMembers[j++].status = members[i].status;
562 : }
563 : }
564 :
565 178 : newMembers[j].xid = xid;
566 178 : newMembers[j++].status = status;
567 178 : newMulti = MultiXactIdCreateFromMembers(j, newMembers);
568 :
569 178 : pfree(members);
570 178 : pfree(newMembers);
571 :
572 : debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
573 :
574 178 : return newMulti;
575 : }
576 :
577 : /*
578 : * MultiXactIdIsRunning
579 : * Returns whether a MultiXactId is "running".
580 : *
581 : * We return true if at least one member of the given MultiXactId is still
582 : * running. Note that a "false" result is certain not to change,
583 : * because it is not legal to add members to an existing MultiXactId.
584 : *
585 : * Caller is expected to have verified that the multixact does not come from
586 : * a pg_upgraded share-locked tuple.
587 : */
588 : bool
589 1942 : MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
590 : {
591 : MultiXactMember *members;
592 : int nmembers;
593 : int i;
594 :
595 : debug_elog3(DEBUG2, "IsRunning %u?", multi);
596 :
597 : /*
598 : * "false" here means we assume our callers have checked that the given
599 : * multi cannot possibly come from a pg_upgraded database.
600 : */
601 1942 : nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
602 :
603 1942 : if (nmembers <= 0)
604 : {
605 : debug_elog2(DEBUG2, "IsRunning: no members");
606 1332 : return false;
607 : }
608 :
609 : /*
610 : * Checking for myself is cheap compared to looking in shared memory;
611 : * return true if any live subtransaction of the current top-level
612 : * transaction is a member.
613 : *
614 : * This is not needed for correctness, it's just a fast path.
615 : */
616 1382 : for (i = 0; i < nmembers; i++)
617 : {
618 1084 : if (TransactionIdIsCurrentTransactionId(members[i].xid))
619 : {
620 : debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
621 312 : pfree(members);
622 312 : return true;
623 : }
624 : }
625 :
626 : /*
627 : * This could be made faster by having another entry point in procarray.c,
628 : * walking the PGPROC array only once for all the members. But in most
629 : * cases nmembers should be small enough that it doesn't much matter.
630 : */
631 548 : for (i = 0; i < nmembers; i++)
632 : {
633 448 : if (TransactionIdIsInProgress(members[i].xid))
634 : {
635 : debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
636 : i, members[i].xid);
637 198 : pfree(members);
638 198 : return true;
639 : }
640 : }
641 :
642 100 : pfree(members);
643 :
644 : debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
645 :
646 100 : return false;
647 : }
648 :
649 : /*
650 : * MultiXactIdSetOldestMember
651 : * Save the oldest MultiXactId this transaction could be a member of.
652 : *
653 : * We set the OldestMemberMXactId for a given transaction the first time it's
654 : * going to do some operation that might require a MultiXactId (tuple lock,
655 : * update or delete). We need to do this even if we end up using a
656 : * TransactionId instead of a MultiXactId, because there is a chance that
657 : * another transaction would add our XID to a MultiXactId.
658 : *
659 : * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
660 : * be called just before doing any such possibly-MultiXactId-able operation.
661 : */
662 : void
663 3742806 : MultiXactIdSetOldestMember(void)
664 : {
665 3742806 : if (!MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]))
666 : {
667 : MultiXactId nextMXact;
668 :
669 : /*
670 : * You might think we don't need to acquire a lock here, since
671 : * fetching and storing of TransactionIds is probably atomic, but in
672 : * fact we do: suppose we pick up nextMXact and then lose the CPU for
673 : * a long time. Someone else could advance nextMXact, and then
674 : * another someone else could compute an OldestVisibleMXactId that
675 : * would be after the value we are going to store when we get control
676 : * back. Which would be wrong.
677 : *
678 : * Note that a shared lock is sufficient, because it's enough to stop
679 : * someone from advancing nextMXact; and nobody else could be trying
680 : * to write to our OldestMember entry, only reading (and we assume
681 : * storing it is atomic.)
682 : */
683 138796 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
684 :
685 : /*
686 : * We have to beware of the possibility that nextMXact is in the
687 : * wrapped-around state. We don't fix the counter itself here, but we
688 : * must be sure to store a valid value in our array entry.
689 : */
690 138796 : nextMXact = MultiXactState->nextMXact;
691 138796 : if (nextMXact < FirstMultiXactId)
692 0 : nextMXact = FirstMultiXactId;
693 :
694 138796 : OldestMemberMXactId[MyProcNumber] = nextMXact;
695 :
696 138796 : LWLockRelease(MultiXactGenLock);
697 :
698 : debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
699 : MyProcNumber, nextMXact);
700 : }
701 3742806 : }
702 :
703 : /*
704 : * MultiXactIdSetOldestVisible
705 : * Save the oldest MultiXactId this transaction considers possibly live.
706 : *
707 : * We set the OldestVisibleMXactId for a given transaction the first time
708 : * it's going to inspect any MultiXactId. Once we have set this, we are
709 : * guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
710 : * won't be truncated away.
711 : *
712 : * The value to set is the oldest of nextMXact and all the valid per-backend
713 : * OldestMemberMXactId[] entries. Because of the locking we do, we can be
714 : * certain that no subsequent call to MultiXactIdSetOldestMember can set
715 : * an OldestMemberMXactId[] entry older than what we compute here. Therefore
716 : * there is no live transaction, now or later, that can be a member of any
717 : * MultiXactId older than the OldestVisibleMXactId we compute here.
718 : */
719 : static void
720 1792 : MultiXactIdSetOldestVisible(void)
721 : {
722 1792 : if (!MultiXactIdIsValid(OldestVisibleMXactId[MyProcNumber]))
723 : {
724 : MultiXactId oldestMXact;
725 : int i;
726 :
727 450 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
728 :
729 : /*
730 : * We have to beware of the possibility that nextMXact is in the
731 : * wrapped-around state. We don't fix the counter itself here, but we
732 : * must be sure to store a valid value in our array entry.
733 : */
734 450 : oldestMXact = MultiXactState->nextMXact;
735 450 : if (oldestMXact < FirstMultiXactId)
736 0 : oldestMXact = FirstMultiXactId;
737 :
738 60894 : for (i = 0; i < MaxOldestSlot; i++)
739 : {
740 60444 : MultiXactId thisoldest = OldestMemberMXactId[i];
741 :
742 60898 : if (MultiXactIdIsValid(thisoldest) &&
743 454 : MultiXactIdPrecedes(thisoldest, oldestMXact))
744 226 : oldestMXact = thisoldest;
745 : }
746 :
747 450 : OldestVisibleMXactId[MyProcNumber] = oldestMXact;
748 :
749 450 : LWLockRelease(MultiXactGenLock);
750 :
751 : debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
752 : MyProcNumber, oldestMXact);
753 : }
754 1792 : }
755 :
756 : /*
757 : * ReadNextMultiXactId
758 : * Return the next MultiXactId to be assigned, but don't allocate it
759 : */
760 : MultiXactId
761 362246 : ReadNextMultiXactId(void)
762 : {
763 : MultiXactId mxid;
764 :
765 : /* XXX we could presumably do this without a lock. */
766 362246 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
767 362246 : mxid = MultiXactState->nextMXact;
768 362246 : LWLockRelease(MultiXactGenLock);
769 :
770 362246 : if (mxid < FirstMultiXactId)
771 0 : mxid = FirstMultiXactId;
772 :
773 362246 : return mxid;
774 : }
775 :
776 : /*
777 : * ReadMultiXactIdRange
778 : * Get the range of IDs that may still be referenced by a relation.
779 : */
780 : void
781 2860 : ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
782 : {
783 2860 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
784 2860 : *oldest = MultiXactState->oldestMultiXactId;
785 2860 : *next = MultiXactState->nextMXact;
786 2860 : LWLockRelease(MultiXactGenLock);
787 :
788 2860 : if (*oldest < FirstMultiXactId)
789 0 : *oldest = FirstMultiXactId;
790 2860 : if (*next < FirstMultiXactId)
791 0 : *next = FirstMultiXactId;
792 2860 : }
793 :
794 :
795 : /*
796 : * MultiXactIdCreateFromMembers
797 : * Make a new MultiXactId from the specified set of members
798 : *
799 : * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
800 : * given TransactionIds as members. Returns the newly created MultiXactId.
801 : *
802 : * NB: the passed members[] array will be sorted in-place.
803 : */
804 : MultiXactId
805 2244 : MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
806 : {
807 : MultiXactId multi;
808 : MultiXactOffset offset;
809 : xl_multixact_create xlrec;
810 :
811 : debug_elog3(DEBUG2, "Create: %s",
812 : mxid_to_string(InvalidMultiXactId, nmembers, members));
813 :
814 : /*
815 : * See if the same set of members already exists in our cache; if so, just
816 : * re-use that MultiXactId. (Note: it might seem that looking in our
817 : * cache is insufficient, and we ought to search disk to see if a
818 : * duplicate definition already exists. But since we only ever create
819 : * MultiXacts containing our own XID, in most cases any such MultiXacts
820 : * were in fact created by us, and so will be in our cache. There are
821 : * corner cases where someone else added us to a MultiXact without our
822 : * knowledge, but it's not worth checking for.)
823 : */
824 2244 : multi = mXactCacheGetBySet(nmembers, members);
825 2244 : if (MultiXactIdIsValid(multi))
826 : {
827 : debug_elog2(DEBUG2, "Create: in cache!");
828 1700 : return multi;
829 : }
830 :
831 : /* Verify that there is a single update Xid among the given members. */
832 : {
833 : int i;
834 544 : bool has_update = false;
835 :
836 1746 : for (i = 0; i < nmembers; i++)
837 : {
838 1202 : if (ISUPDATE_from_mxstatus(members[i].status))
839 : {
840 264 : if (has_update)
841 0 : elog(ERROR, "new multixact has more than one updating member: %s",
842 : mxid_to_string(InvalidMultiXactId, nmembers, members));
843 264 : has_update = true;
844 : }
845 : }
846 : }
847 :
848 : /* Load the injection point before entering the critical section */
849 544 : INJECTION_POINT_LOAD("multixact-create-from-members");
850 :
851 : /*
852 : * Assign the MXID and offsets range to use, and make sure there is space
853 : * in the OFFSETs and MEMBERs files. NB: this routine does
854 : * START_CRIT_SECTION().
855 : *
856 : * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
857 : * that we've called MultiXactIdSetOldestMember here. This is because
858 : * this routine is used in some places to create new MultiXactIds of which
859 : * the current backend is not a member, notably during freezing of multis
860 : * in vacuum. During vacuum, in particular, it would be unacceptable to
861 : * keep OldestMulti set, in case it runs for long.
862 : */
863 544 : multi = GetNewMultiXactId(nmembers, &offset);
864 :
865 544 : INJECTION_POINT_CACHED("multixact-create-from-members", NULL);
866 :
867 : /* Make an XLOG entry describing the new MXID. */
868 544 : xlrec.mid = multi;
869 544 : xlrec.moff = offset;
870 544 : xlrec.nmembers = nmembers;
871 :
872 : /*
873 : * XXX Note: there's a lot of padding space in MultiXactMember. We could
874 : * find a more compact representation of this Xlog record -- perhaps all
875 : * the status flags in one XLogRecData, then all the xids in another one?
876 : * Not clear that it's worth the trouble though.
877 : */
878 544 : XLogBeginInsert();
879 544 : XLogRegisterData(&xlrec, SizeOfMultiXactCreate);
880 544 : XLogRegisterData(members, nmembers * sizeof(MultiXactMember));
881 :
882 544 : (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
883 :
884 : /* Now enter the information into the OFFSETs and MEMBERs logs */
885 544 : RecordNewMultiXact(multi, offset, nmembers, members);
886 :
887 : /* Done with critical section */
888 544 : END_CRIT_SECTION();
889 :
890 : /* Store the new MultiXactId in the local cache, too */
891 544 : mXactCachePut(multi, nmembers, members);
892 :
893 : debug_elog2(DEBUG2, "Create: all done");
894 :
895 544 : return multi;
896 : }
897 :
898 : /*
899 : * RecordNewMultiXact
900 : * Write info about a new multixact into the offsets and members files
901 : *
902 : * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
903 : * use it.
904 : */
905 : static void
906 548 : RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
907 : int nmembers, MultiXactMember *members)
908 : {
909 : int64 pageno;
910 : int64 prev_pageno;
911 : int entryno;
912 : int slotno;
913 : MultiXactOffset *offptr;
914 : int i;
915 : LWLock *lock;
916 548 : LWLock *prevlock = NULL;
917 :
918 548 : pageno = MultiXactIdToOffsetPage(multi);
919 548 : entryno = MultiXactIdToOffsetEntry(multi);
920 :
921 548 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
922 548 : LWLockAcquire(lock, LW_EXCLUSIVE);
923 :
924 : /*
925 : * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
926 : * to complain about if there's any I/O error. This is kinda bogus, but
927 : * since the errors will always give the full pathname, it should be clear
928 : * enough that a MultiXactId is really involved. Perhaps someday we'll
929 : * take the trouble to generalize the slru.c error reporting code.
930 : */
931 548 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
932 548 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
933 548 : offptr += entryno;
934 :
935 548 : *offptr = offset;
936 :
937 548 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
938 :
939 : /* Release MultiXactOffset SLRU lock. */
940 548 : LWLockRelease(lock);
941 :
942 : /*
943 : * If anybody was waiting to know the offset of this multixact ID we just
944 : * wrote, they can read it now, so wake them up.
945 : */
946 548 : ConditionVariableBroadcast(&MultiXactState->nextoff_cv);
947 :
948 548 : prev_pageno = -1;
949 :
950 1758 : for (i = 0; i < nmembers; i++, offset++)
951 : {
952 : TransactionId *memberptr;
953 : uint32 *flagsptr;
954 : uint32 flagsval;
955 : int bshift;
956 : int flagsoff;
957 : int memberoff;
958 :
959 : Assert(members[i].status <= MultiXactStatusUpdate);
960 :
961 1210 : pageno = MXOffsetToMemberPage(offset);
962 1210 : memberoff = MXOffsetToMemberOffset(offset);
963 1210 : flagsoff = MXOffsetToFlagsOffset(offset);
964 1210 : bshift = MXOffsetToFlagsBitShift(offset);
965 :
966 1210 : if (pageno != prev_pageno)
967 : {
968 : /*
969 : * MultiXactMember SLRU page is changed so check if this new page
970 : * fall into the different SLRU bank then release the old bank's
971 : * lock and acquire lock on the new bank.
972 : */
973 548 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
974 548 : if (lock != prevlock)
975 : {
976 548 : if (prevlock != NULL)
977 0 : LWLockRelease(prevlock);
978 :
979 548 : LWLockAcquire(lock, LW_EXCLUSIVE);
980 548 : prevlock = lock;
981 : }
982 548 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
983 548 : prev_pageno = pageno;
984 : }
985 :
986 1210 : memberptr = (TransactionId *)
987 1210 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
988 :
989 1210 : *memberptr = members[i].xid;
990 :
991 1210 : flagsptr = (uint32 *)
992 1210 : (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
993 :
994 1210 : flagsval = *flagsptr;
995 1210 : flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
996 1210 : flagsval |= (members[i].status << bshift);
997 1210 : *flagsptr = flagsval;
998 :
999 1210 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
1000 : }
1001 :
1002 548 : if (prevlock != NULL)
1003 548 : LWLockRelease(prevlock);
1004 548 : }
1005 :
1006 : /*
1007 : * GetNewMultiXactId
1008 : * Get the next MultiXactId.
1009 : *
1010 : * Also, reserve the needed amount of space in the "members" area. The
1011 : * starting offset of the reserved space is returned in *offset.
1012 : *
1013 : * This may generate XLOG records for expansion of the offsets and/or members
1014 : * files. Unfortunately, we have to do that while holding MultiXactGenLock
1015 : * to avoid race conditions --- the XLOG record for zeroing a page must appear
1016 : * before any backend can possibly try to store data in that page!
1017 : *
1018 : * We start a critical section before advancing the shared counters. The
1019 : * caller must end the critical section after writing SLRU data.
1020 : */
1021 : static MultiXactId
1022 544 : GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
1023 : {
1024 : MultiXactId result;
1025 : MultiXactOffset nextOffset;
1026 :
1027 : debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
1028 :
1029 : /* safety check, we should never get this far in a HS standby */
1030 544 : if (RecoveryInProgress())
1031 0 : elog(ERROR, "cannot assign MultiXactIds during recovery");
1032 :
1033 544 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1034 :
1035 : /* Handle wraparound of the nextMXact counter */
1036 544 : if (MultiXactState->nextMXact < FirstMultiXactId)
1037 0 : MultiXactState->nextMXact = FirstMultiXactId;
1038 :
1039 : /* Assign the MXID */
1040 544 : result = MultiXactState->nextMXact;
1041 :
1042 : /*----------
1043 : * Check to see if it's safe to assign another MultiXactId. This protects
1044 : * against catastrophic data loss due to multixact wraparound. The basic
1045 : * rules are:
1046 : *
1047 : * If we're past multiVacLimit or the safe threshold for member storage
1048 : * space, or we don't know what the safe threshold for member storage is,
1049 : * start trying to force autovacuum cycles.
1050 : * If we're past multiWarnLimit, start issuing warnings.
1051 : * If we're past multiStopLimit, refuse to create new MultiXactIds.
1052 : *
1053 : * Note these are pretty much the same protections in GetNewTransactionId.
1054 : *----------
1055 : */
1056 544 : if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
1057 : {
1058 : /*
1059 : * For safety's sake, we release MultiXactGenLock while sending
1060 : * signals, warnings, etc. This is not so much because we care about
1061 : * preserving concurrency in this situation, as to avoid any
1062 : * possibility of deadlock while doing get_database_name(). First,
1063 : * copy all the shared values we'll need in this path.
1064 : */
1065 0 : MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
1066 0 : MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
1067 0 : MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
1068 0 : Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
1069 :
1070 0 : LWLockRelease(MultiXactGenLock);
1071 :
1072 0 : if (IsUnderPostmaster &&
1073 0 : !MultiXactIdPrecedes(result, multiStopLimit))
1074 : {
1075 0 : char *oldest_datname = get_database_name(oldest_datoid);
1076 :
1077 : /*
1078 : * Immediately kick autovacuum into action as we're already in
1079 : * ERROR territory.
1080 : */
1081 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1082 :
1083 : /* complain even if that DB has disappeared */
1084 0 : if (oldest_datname)
1085 0 : ereport(ERROR,
1086 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1087 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database \"%s\"",
1088 : oldest_datname),
1089 : errhint("Execute a database-wide VACUUM in that database.\n"
1090 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1091 : else
1092 0 : ereport(ERROR,
1093 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1094 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database with OID %u",
1095 : oldest_datoid),
1096 : errhint("Execute a database-wide VACUUM in that database.\n"
1097 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1098 : }
1099 :
1100 : /*
1101 : * To avoid swamping the postmaster with signals, we issue the autovac
1102 : * request only once per 64K multis generated. This still gives
1103 : * plenty of chances before we get into real trouble.
1104 : */
1105 0 : if (IsUnderPostmaster && (result % 65536) == 0)
1106 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1107 :
1108 0 : if (!MultiXactIdPrecedes(result, multiWarnLimit))
1109 : {
1110 0 : char *oldest_datname = get_database_name(oldest_datoid);
1111 :
1112 : /* complain even if that DB has disappeared */
1113 0 : if (oldest_datname)
1114 0 : ereport(WARNING,
1115 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1116 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1117 : multiWrapLimit - result,
1118 : oldest_datname,
1119 : multiWrapLimit - result),
1120 : errhint("Execute a database-wide VACUUM in that database.\n"
1121 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1122 : else
1123 0 : ereport(WARNING,
1124 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1125 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1126 : multiWrapLimit - result,
1127 : oldest_datoid,
1128 : multiWrapLimit - result),
1129 : errhint("Execute a database-wide VACUUM in that database.\n"
1130 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1131 : }
1132 :
1133 : /* Re-acquire lock and start over */
1134 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1135 0 : result = MultiXactState->nextMXact;
1136 0 : if (result < FirstMultiXactId)
1137 0 : result = FirstMultiXactId;
1138 : }
1139 :
1140 : /* Make sure there is room for the MXID in the file. */
1141 544 : ExtendMultiXactOffset(result);
1142 :
1143 : /*
1144 : * Reserve the members space, similarly to above. Also, be careful not to
1145 : * return zero as the starting offset for any multixact. See
1146 : * GetMultiXactIdMembers() for motivation.
1147 : */
1148 544 : nextOffset = MultiXactState->nextOffset;
1149 544 : if (nextOffset == 0)
1150 : {
1151 22 : *offset = 1;
1152 22 : nmembers++; /* allocate member slot 0 too */
1153 : }
1154 : else
1155 522 : *offset = nextOffset;
1156 :
1157 : /*----------
1158 : * Protect against overrun of the members space as well, with the
1159 : * following rules:
1160 : *
1161 : * If we're past offsetStopLimit, refuse to generate more multis.
1162 : * If we're close to offsetStopLimit, emit a warning.
1163 : *
1164 : * Arbitrarily, we start emitting warnings when we're 20 segments or less
1165 : * from offsetStopLimit.
1166 : *
1167 : * Note we haven't updated the shared state yet, so if we fail at this
1168 : * point, the multixact ID we grabbed can still be used by the next guy.
1169 : *
1170 : * Note that there is no point in forcing autovacuum runs here: the
1171 : * multixact freeze settings would have to be reduced for that to have any
1172 : * effect.
1173 : *----------
1174 : */
1175 : #define OFFSET_WARN_SEGMENTS 20
1176 1088 : if (MultiXactState->oldestOffsetKnown &&
1177 544 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit, nextOffset,
1178 : nmembers))
1179 : {
1180 : /* see comment in the corresponding offsets wraparound case */
1181 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1182 :
1183 0 : ereport(ERROR,
1184 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1185 : errmsg("multixact \"members\" limit exceeded"),
1186 : errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
1187 : "This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
1188 : MultiXactState->offsetStopLimit - nextOffset - 1,
1189 : nmembers,
1190 : MultiXactState->offsetStopLimit - nextOffset - 1),
1191 : 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.",
1192 : MultiXactState->oldestMultiXactDB)));
1193 : }
1194 :
1195 : /*
1196 : * Check whether we should kick autovacuum into action, to prevent members
1197 : * wraparound. NB we use a much larger window to trigger autovacuum than
1198 : * just the warning limit. The warning is just a measure of last resort -
1199 : * this is in line with GetNewTransactionId's behaviour.
1200 : */
1201 544 : if (!MultiXactState->oldestOffsetKnown ||
1202 544 : (MultiXactState->nextOffset - MultiXactState->oldestOffset
1203 544 : > MULTIXACT_MEMBER_SAFE_THRESHOLD))
1204 : {
1205 : /*
1206 : * To avoid swamping the postmaster with signals, we issue the autovac
1207 : * request only when crossing a segment boundary. With default
1208 : * compilation settings that's roughly after 50k members. This still
1209 : * gives plenty of chances before we get into real trouble.
1210 : */
1211 0 : if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
1212 0 : (MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
1213 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1214 : }
1215 :
1216 1088 : if (MultiXactState->oldestOffsetKnown &&
1217 544 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit,
1218 : nextOffset,
1219 : nmembers + MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT * OFFSET_WARN_SEGMENTS))
1220 0 : ereport(WARNING,
1221 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1222 : errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
1223 : "database with OID %u must be vacuumed before %d more multixact members are used",
1224 : MultiXactState->offsetStopLimit - nextOffset + nmembers,
1225 : MultiXactState->oldestMultiXactDB,
1226 : MultiXactState->offsetStopLimit - nextOffset + nmembers),
1227 : errhint("Execute a database-wide VACUUM in that database with reduced \"vacuum_multixact_freeze_min_age\" and \"vacuum_multixact_freeze_table_age\" settings.")));
1228 :
1229 544 : ExtendMultiXactMember(nextOffset, nmembers);
1230 :
1231 : /*
1232 : * Critical section from here until caller has written the data into the
1233 : * just-reserved SLRU space; we don't want to error out with a partly
1234 : * written MultiXact structure. (In particular, failing to write our
1235 : * start offset after advancing nextMXact would effectively corrupt the
1236 : * previous MultiXact.)
1237 : */
1238 544 : START_CRIT_SECTION();
1239 :
1240 : /*
1241 : * Advance counters. As in GetNewTransactionId(), this must not happen
1242 : * until after file extension has succeeded!
1243 : *
1244 : * We don't care about MultiXactId wraparound here; it will be handled by
1245 : * the next iteration. But note that nextMXact may be InvalidMultiXactId
1246 : * or the first value on a segment-beginning page after this routine
1247 : * exits, so anyone else looking at the variable must be prepared to deal
1248 : * with either case. Similarly, nextOffset may be zero, but we won't use
1249 : * that as the actual start offset of the next multixact.
1250 : */
1251 544 : (MultiXactState->nextMXact)++;
1252 :
1253 544 : MultiXactState->nextOffset += nmembers;
1254 :
1255 544 : LWLockRelease(MultiXactGenLock);
1256 :
1257 : debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
1258 544 : return result;
1259 : }
1260 :
1261 : /*
1262 : * GetMultiXactIdMembers
1263 : * Return the set of MultiXactMembers that make up a MultiXactId
1264 : *
1265 : * Return value is the number of members found, or -1 if there are none,
1266 : * and *members is set to a newly palloc'ed array of members. It's the
1267 : * caller's responsibility to free it when done with it.
1268 : *
1269 : * from_pgupgrade must be passed as true if and only if only the multixact
1270 : * corresponds to a value from a tuple that was locked in a 9.2-or-older
1271 : * installation and later pg_upgrade'd (that is, the infomask is
1272 : * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1273 : * can still be running, so we return -1 just like for an empty multixact
1274 : * without any further checking. It would be wrong to try to resolve such a
1275 : * multixact: either the multixact is within the current valid multixact
1276 : * range, in which case the returned result would be bogus, or outside that
1277 : * range, in which case an error would be raised.
1278 : *
1279 : * In all other cases, the passed multixact must be within the known valid
1280 : * range, that is, greater than or equal to oldestMultiXactId, and less than
1281 : * nextMXact. Otherwise, an error is raised.
1282 : *
1283 : * isLockOnly must be set to true if caller is certain that the given multi
1284 : * is used only to lock tuples; can be false without loss of correctness,
1285 : * but passing a true means we can return quickly without checking for
1286 : * old updates.
1287 : */
1288 : int
1289 5940 : GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
1290 : bool from_pgupgrade, bool isLockOnly)
1291 : {
1292 : int64 pageno;
1293 : int64 prev_pageno;
1294 : int entryno;
1295 : int slotno;
1296 : MultiXactOffset *offptr;
1297 : MultiXactOffset offset;
1298 : int length;
1299 : int truelength;
1300 : MultiXactId oldestMXact;
1301 : MultiXactId nextMXact;
1302 : MultiXactId tmpMXact;
1303 : MultiXactOffset nextOffset;
1304 : MultiXactMember *ptr;
1305 : LWLock *lock;
1306 5940 : bool slept = false;
1307 :
1308 : debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1309 :
1310 5940 : if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1311 : {
1312 0 : *members = NULL;
1313 0 : return -1;
1314 : }
1315 :
1316 : /* See if the MultiXactId is in the local cache */
1317 5940 : length = mXactCacheGetById(multi, members);
1318 5940 : if (length >= 0)
1319 : {
1320 : debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1321 : mxid_to_string(multi, length, *members));
1322 4148 : return length;
1323 : }
1324 :
1325 : /* Set our OldestVisibleMXactId[] entry if we didn't already */
1326 1792 : MultiXactIdSetOldestVisible();
1327 :
1328 : /*
1329 : * If we know the multi is used only for locking and not for updates, then
1330 : * we can skip checking if the value is older than our oldest visible
1331 : * multi. It cannot possibly still be running.
1332 : */
1333 3280 : if (isLockOnly &&
1334 1488 : MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyProcNumber]))
1335 : {
1336 : debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1337 1334 : *members = NULL;
1338 1334 : return -1;
1339 : }
1340 :
1341 : /*
1342 : * We check known limits on MultiXact before resorting to the SLRU area.
1343 : *
1344 : * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1345 : * useful; it has already been removed, or will be removed shortly, by
1346 : * truncation. If one is passed, an error is raised.
1347 : *
1348 : * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1349 : * implies undetected ID wraparound has occurred. This raises a hard
1350 : * error.
1351 : *
1352 : * Shared lock is enough here since we aren't modifying any global state.
1353 : * Acquire it just long enough to grab the current counter values. We may
1354 : * need both nextMXact and nextOffset; see below.
1355 : */
1356 458 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1357 :
1358 458 : oldestMXact = MultiXactState->oldestMultiXactId;
1359 458 : nextMXact = MultiXactState->nextMXact;
1360 458 : nextOffset = MultiXactState->nextOffset;
1361 :
1362 458 : LWLockRelease(MultiXactGenLock);
1363 :
1364 458 : if (MultiXactIdPrecedes(multi, oldestMXact))
1365 0 : ereport(ERROR,
1366 : (errcode(ERRCODE_INTERNAL_ERROR),
1367 : errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1368 : multi)));
1369 :
1370 458 : if (!MultiXactIdPrecedes(multi, nextMXact))
1371 0 : ereport(ERROR,
1372 : (errcode(ERRCODE_INTERNAL_ERROR),
1373 : errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1374 : multi)));
1375 :
1376 : /*
1377 : * Find out the offset at which we need to start reading MultiXactMembers
1378 : * and the number of members in the multixact. We determine the latter as
1379 : * the difference between this multixact's starting offset and the next
1380 : * one's. However, there are some corner cases to worry about:
1381 : *
1382 : * 1. This multixact may be the latest one created, in which case there is
1383 : * no next one to look at. In this case the nextOffset value we just
1384 : * saved is the correct endpoint.
1385 : *
1386 : * 2. The next multixact may still be in process of being filled in: that
1387 : * is, another process may have done GetNewMultiXactId but not yet written
1388 : * the offset entry for that ID. In that scenario, it is guaranteed that
1389 : * the offset entry for that multixact exists (because GetNewMultiXactId
1390 : * won't release MultiXactGenLock until it does) but contains zero
1391 : * (because we are careful to pre-zero offset pages). Because
1392 : * GetNewMultiXactId will never return zero as the starting offset for a
1393 : * multixact, when we read zero as the next multixact's offset, we know we
1394 : * have this case. We handle this by sleeping on the condition variable
1395 : * we have just for this; the process in charge will signal the CV as soon
1396 : * as it has finished writing the multixact offset.
1397 : *
1398 : * 3. Because GetNewMultiXactId increments offset zero to offset one to
1399 : * handle case #2, there is an ambiguity near the point of offset
1400 : * wraparound. If we see next multixact's offset is one, is that our
1401 : * multixact's actual endpoint, or did it end at zero with a subsequent
1402 : * increment? We handle this using the knowledge that if the zero'th
1403 : * member slot wasn't filled, it'll contain zero, and zero isn't a valid
1404 : * transaction ID so it can't be a multixact member. Therefore, if we
1405 : * read a zero from the members array, just ignore it.
1406 : *
1407 : * This is all pretty messy, but the mess occurs only in infrequent corner
1408 : * cases, so it seems better than holding the MultiXactGenLock for a long
1409 : * time on every multixact creation.
1410 : */
1411 458 : retry:
1412 460 : pageno = MultiXactIdToOffsetPage(multi);
1413 460 : entryno = MultiXactIdToOffsetEntry(multi);
1414 :
1415 : /* Acquire the bank lock for the page we need. */
1416 460 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1417 460 : LWLockAcquire(lock, LW_EXCLUSIVE);
1418 :
1419 460 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1420 460 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1421 460 : offptr += entryno;
1422 460 : offset = *offptr;
1423 :
1424 : Assert(offset != 0);
1425 :
1426 : /*
1427 : * Use the same increment rule as GetNewMultiXactId(), that is, don't
1428 : * handle wraparound explicitly until needed.
1429 : */
1430 460 : tmpMXact = multi + 1;
1431 :
1432 460 : if (nextMXact == tmpMXact)
1433 : {
1434 : /* Corner case 1: there is no next multixact */
1435 400 : length = nextOffset - offset;
1436 : }
1437 : else
1438 : {
1439 : MultiXactOffset nextMXOffset;
1440 :
1441 : /* handle wraparound if needed */
1442 60 : if (tmpMXact < FirstMultiXactId)
1443 0 : tmpMXact = FirstMultiXactId;
1444 :
1445 60 : prev_pageno = pageno;
1446 :
1447 60 : pageno = MultiXactIdToOffsetPage(tmpMXact);
1448 60 : entryno = MultiXactIdToOffsetEntry(tmpMXact);
1449 :
1450 60 : if (pageno != prev_pageno)
1451 : {
1452 : LWLock *newlock;
1453 :
1454 : /*
1455 : * Since we're going to access a different SLRU page, if this page
1456 : * falls under a different bank, release the old bank's lock and
1457 : * acquire the lock of the new bank.
1458 : */
1459 0 : newlock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1460 0 : if (newlock != lock)
1461 : {
1462 0 : LWLockRelease(lock);
1463 0 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1464 0 : lock = newlock;
1465 : }
1466 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1467 : }
1468 :
1469 60 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1470 60 : offptr += entryno;
1471 60 : nextMXOffset = *offptr;
1472 :
1473 60 : if (nextMXOffset == 0)
1474 : {
1475 : /* Corner case 2: next multixact is still being filled in */
1476 2 : LWLockRelease(lock);
1477 2 : CHECK_FOR_INTERRUPTS();
1478 :
1479 2 : INJECTION_POINT("multixact-get-members-cv-sleep", NULL);
1480 :
1481 2 : ConditionVariableSleep(&MultiXactState->nextoff_cv,
1482 : WAIT_EVENT_MULTIXACT_CREATION);
1483 2 : slept = true;
1484 2 : goto retry;
1485 : }
1486 :
1487 58 : length = nextMXOffset - offset;
1488 : }
1489 :
1490 458 : LWLockRelease(lock);
1491 458 : lock = NULL;
1492 :
1493 : /*
1494 : * If we slept above, clean up state; it's no longer needed.
1495 : */
1496 458 : if (slept)
1497 2 : ConditionVariableCancelSleep();
1498 :
1499 458 : ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1500 :
1501 458 : truelength = 0;
1502 458 : prev_pageno = -1;
1503 1636 : for (int i = 0; i < length; i++, offset++)
1504 : {
1505 : TransactionId *xactptr;
1506 : uint32 *flagsptr;
1507 : int flagsoff;
1508 : int bshift;
1509 : int memberoff;
1510 :
1511 1178 : pageno = MXOffsetToMemberPage(offset);
1512 1178 : memberoff = MXOffsetToMemberOffset(offset);
1513 :
1514 1178 : if (pageno != prev_pageno)
1515 : {
1516 : LWLock *newlock;
1517 :
1518 : /*
1519 : * Since we're going to access a different SLRU page, if this page
1520 : * falls under a different bank, release the old bank's lock and
1521 : * acquire the lock of the new bank.
1522 : */
1523 458 : newlock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
1524 458 : if (newlock != lock)
1525 : {
1526 458 : if (lock)
1527 0 : LWLockRelease(lock);
1528 458 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1529 458 : lock = newlock;
1530 : }
1531 :
1532 458 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1533 458 : prev_pageno = pageno;
1534 : }
1535 :
1536 1178 : xactptr = (TransactionId *)
1537 1178 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1538 :
1539 1178 : if (!TransactionIdIsValid(*xactptr))
1540 : {
1541 : /* Corner case 3: we must be looking at unused slot zero */
1542 : Assert(offset == 0);
1543 0 : continue;
1544 : }
1545 :
1546 1178 : flagsoff = MXOffsetToFlagsOffset(offset);
1547 1178 : bshift = MXOffsetToFlagsBitShift(offset);
1548 1178 : flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1549 :
1550 1178 : ptr[truelength].xid = *xactptr;
1551 1178 : ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1552 1178 : truelength++;
1553 : }
1554 :
1555 458 : LWLockRelease(lock);
1556 :
1557 : /* A multixid with zero members should not happen */
1558 : Assert(truelength > 0);
1559 :
1560 : /*
1561 : * Copy the result into the local cache.
1562 : */
1563 458 : mXactCachePut(multi, truelength, ptr);
1564 :
1565 : debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1566 : mxid_to_string(multi, truelength, ptr));
1567 458 : *members = ptr;
1568 458 : return truelength;
1569 : }
1570 :
1571 : /*
1572 : * mxactMemberComparator
1573 : * qsort comparison function for MultiXactMember
1574 : *
1575 : * We can't use wraparound comparison for XIDs because that does not respect
1576 : * the triangle inequality! Any old sort order will do.
1577 : */
1578 : static int
1579 3770 : mxactMemberComparator(const void *arg1, const void *arg2)
1580 : {
1581 3770 : MultiXactMember member1 = *(const MultiXactMember *) arg1;
1582 3770 : MultiXactMember member2 = *(const MultiXactMember *) arg2;
1583 :
1584 3770 : if (member1.xid > member2.xid)
1585 54 : return 1;
1586 3716 : if (member1.xid < member2.xid)
1587 3370 : return -1;
1588 346 : if (member1.status > member2.status)
1589 4 : return 1;
1590 342 : if (member1.status < member2.status)
1591 342 : return -1;
1592 0 : return 0;
1593 : }
1594 :
1595 : /*
1596 : * mXactCacheGetBySet
1597 : * returns a MultiXactId from the cache based on the set of
1598 : * TransactionIds that compose it, or InvalidMultiXactId if
1599 : * none matches.
1600 : *
1601 : * This is helpful, for example, if two transactions want to lock a huge
1602 : * table. By using the cache, the second will use the same MultiXactId
1603 : * for the majority of tuples, thus keeping MultiXactId usage low (saving
1604 : * both I/O and wraparound issues).
1605 : *
1606 : * NB: the passed members array will be sorted in-place.
1607 : */
1608 : static MultiXactId
1609 2244 : mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1610 : {
1611 : dlist_iter iter;
1612 :
1613 : debug_elog3(DEBUG2, "CacheGet: looking for %s",
1614 : mxid_to_string(InvalidMultiXactId, nmembers, members));
1615 :
1616 : /* sort the array so comparison is easy */
1617 2244 : qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1618 :
1619 2706 : dclist_foreach(iter, &MXactCache)
1620 : {
1621 2162 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1622 : iter.cur);
1623 :
1624 2162 : if (entry->nmembers != nmembers)
1625 236 : continue;
1626 :
1627 : /*
1628 : * We assume the cache entries are sorted, and that the unused bits in
1629 : * "status" are zeroed.
1630 : */
1631 1926 : if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1632 : {
1633 : debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1634 1700 : dclist_move_head(&MXactCache, iter.cur);
1635 1700 : return entry->multi;
1636 : }
1637 : }
1638 :
1639 : debug_elog2(DEBUG2, "CacheGet: not found :-(");
1640 544 : return InvalidMultiXactId;
1641 : }
1642 :
1643 : /*
1644 : * mXactCacheGetById
1645 : * returns the composing MultiXactMember set from the cache for a
1646 : * given MultiXactId, if present.
1647 : *
1648 : * If successful, *xids is set to the address of a palloc'd copy of the
1649 : * MultiXactMember set. Return value is number of members, or -1 on failure.
1650 : */
1651 : static int
1652 5940 : mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
1653 : {
1654 : dlist_iter iter;
1655 :
1656 : debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1657 :
1658 6828 : dclist_foreach(iter, &MXactCache)
1659 : {
1660 5036 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1661 : iter.cur);
1662 :
1663 5036 : if (entry->multi == multi)
1664 : {
1665 : MultiXactMember *ptr;
1666 : Size size;
1667 :
1668 4148 : size = sizeof(MultiXactMember) * entry->nmembers;
1669 4148 : ptr = (MultiXactMember *) palloc(size);
1670 :
1671 4148 : memcpy(ptr, entry->members, size);
1672 :
1673 : debug_elog3(DEBUG2, "CacheGet: found %s",
1674 : mxid_to_string(multi,
1675 : entry->nmembers,
1676 : entry->members));
1677 :
1678 : /*
1679 : * Note we modify the list while not using a modifiable iterator.
1680 : * This is acceptable only because we exit the iteration
1681 : * immediately afterwards.
1682 : */
1683 4148 : dclist_move_head(&MXactCache, iter.cur);
1684 :
1685 4148 : *members = ptr;
1686 4148 : return entry->nmembers;
1687 : }
1688 : }
1689 :
1690 : debug_elog2(DEBUG2, "CacheGet: not found");
1691 1792 : return -1;
1692 : }
1693 :
1694 : /*
1695 : * mXactCachePut
1696 : * Add a new MultiXactId and its composing set into the local cache.
1697 : */
1698 : static void
1699 1002 : mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1700 : {
1701 : mXactCacheEnt *entry;
1702 :
1703 : debug_elog3(DEBUG2, "CachePut: storing %s",
1704 : mxid_to_string(multi, nmembers, members));
1705 :
1706 1002 : if (MXactContext == NULL)
1707 : {
1708 : /* The cache only lives as long as the current transaction */
1709 : debug_elog2(DEBUG2, "CachePut: initializing memory context");
1710 722 : MXactContext = AllocSetContextCreate(TopTransactionContext,
1711 : "MultiXact cache context",
1712 : ALLOCSET_SMALL_SIZES);
1713 : }
1714 :
1715 : entry = (mXactCacheEnt *)
1716 1002 : MemoryContextAlloc(MXactContext,
1717 1002 : offsetof(mXactCacheEnt, members) +
1718 : nmembers * sizeof(MultiXactMember));
1719 :
1720 1002 : entry->multi = multi;
1721 1002 : entry->nmembers = nmembers;
1722 1002 : memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1723 :
1724 : /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1725 1002 : qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1726 :
1727 1002 : dclist_push_head(&MXactCache, &entry->node);
1728 1002 : if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
1729 : {
1730 : dlist_node *node;
1731 :
1732 0 : node = dclist_tail_node(&MXactCache);
1733 0 : dclist_delete_from(&MXactCache, node);
1734 :
1735 0 : entry = dclist_container(mXactCacheEnt, node, node);
1736 : debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1737 : entry->multi);
1738 :
1739 0 : pfree(entry);
1740 : }
1741 1002 : }
1742 :
1743 : char *
1744 0 : mxstatus_to_string(MultiXactStatus status)
1745 : {
1746 0 : switch (status)
1747 : {
1748 0 : case MultiXactStatusForKeyShare:
1749 0 : return "keysh";
1750 0 : case MultiXactStatusForShare:
1751 0 : return "sh";
1752 0 : case MultiXactStatusForNoKeyUpdate:
1753 0 : return "fornokeyupd";
1754 0 : case MultiXactStatusForUpdate:
1755 0 : return "forupd";
1756 0 : case MultiXactStatusNoKeyUpdate:
1757 0 : return "nokeyupd";
1758 0 : case MultiXactStatusUpdate:
1759 0 : return "upd";
1760 0 : default:
1761 0 : elog(ERROR, "unrecognized multixact status %d", status);
1762 : return "";
1763 : }
1764 : }
1765 :
1766 : char *
1767 0 : mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1768 : {
1769 : static char *str = NULL;
1770 : StringInfoData buf;
1771 : int i;
1772 :
1773 0 : if (str != NULL)
1774 0 : pfree(str);
1775 :
1776 0 : initStringInfo(&buf);
1777 :
1778 0 : appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1779 : mxstatus_to_string(members[0].status));
1780 :
1781 0 : for (i = 1; i < nmembers; i++)
1782 0 : appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1783 0 : mxstatus_to_string(members[i].status));
1784 :
1785 0 : appendStringInfoChar(&buf, ']');
1786 0 : str = MemoryContextStrdup(TopMemoryContext, buf.data);
1787 0 : pfree(buf.data);
1788 0 : return str;
1789 : }
1790 :
1791 : /*
1792 : * AtEOXact_MultiXact
1793 : * Handle transaction end for MultiXact
1794 : *
1795 : * This is called at top transaction commit or abort (we don't care which).
1796 : */
1797 : void
1798 1019620 : AtEOXact_MultiXact(void)
1799 : {
1800 : /*
1801 : * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1802 : * which should only be valid while within a transaction.
1803 : *
1804 : * We assume that storing a MultiXactId is atomic and so we need not take
1805 : * MultiXactGenLock to do this.
1806 : */
1807 1019620 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1808 1019620 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1809 :
1810 : /*
1811 : * Discard the local MultiXactId cache. Since MXactContext was created as
1812 : * a child of TopTransactionContext, we needn't delete it explicitly.
1813 : */
1814 1019620 : MXactContext = NULL;
1815 1019620 : dclist_init(&MXactCache);
1816 1019620 : }
1817 :
1818 : /*
1819 : * AtPrepare_MultiXact
1820 : * Save multixact state at 2PC transaction prepare
1821 : *
1822 : * In this phase, we only store our OldestMemberMXactId value in the two-phase
1823 : * state file.
1824 : */
1825 : void
1826 584 : AtPrepare_MultiXact(void)
1827 : {
1828 584 : MultiXactId myOldestMember = OldestMemberMXactId[MyProcNumber];
1829 :
1830 584 : if (MultiXactIdIsValid(myOldestMember))
1831 98 : RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
1832 : &myOldestMember, sizeof(MultiXactId));
1833 584 : }
1834 :
1835 : /*
1836 : * PostPrepare_MultiXact
1837 : * Clean up after successful PREPARE TRANSACTION
1838 : */
1839 : void
1840 584 : PostPrepare_MultiXact(FullTransactionId fxid)
1841 : {
1842 : MultiXactId myOldestMember;
1843 :
1844 : /*
1845 : * Transfer our OldestMemberMXactId value to the slot reserved for the
1846 : * prepared transaction.
1847 : */
1848 584 : myOldestMember = OldestMemberMXactId[MyProcNumber];
1849 584 : if (MultiXactIdIsValid(myOldestMember))
1850 : {
1851 98 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1852 :
1853 : /*
1854 : * Even though storing MultiXactId is atomic, acquire lock to make
1855 : * sure others see both changes, not just the reset of the slot of the
1856 : * current backend. Using a volatile pointer might suffice, but this
1857 : * isn't a hot spot.
1858 : */
1859 98 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1860 :
1861 98 : OldestMemberMXactId[dummyProcNumber] = myOldestMember;
1862 98 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1863 :
1864 98 : LWLockRelease(MultiXactGenLock);
1865 : }
1866 :
1867 : /*
1868 : * We don't need to transfer OldestVisibleMXactId value, because the
1869 : * transaction is not going to be looking at any more multixacts once it's
1870 : * prepared.
1871 : *
1872 : * We assume that storing a MultiXactId is atomic and so we need not take
1873 : * MultiXactGenLock to do this.
1874 : */
1875 584 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1876 :
1877 : /*
1878 : * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1879 : */
1880 584 : MXactContext = NULL;
1881 584 : dclist_init(&MXactCache);
1882 584 : }
1883 :
1884 : /*
1885 : * multixact_twophase_recover
1886 : * Recover the state of a prepared transaction at startup
1887 : */
1888 : void
1889 16 : multixact_twophase_recover(FullTransactionId fxid, uint16 info,
1890 : void *recdata, uint32 len)
1891 : {
1892 16 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1893 : MultiXactId oldestMember;
1894 :
1895 : /*
1896 : * Get the oldest member XID from the state file record, and set it in the
1897 : * OldestMemberMXactId slot reserved for this prepared transaction.
1898 : */
1899 : Assert(len == sizeof(MultiXactId));
1900 16 : oldestMember = *((MultiXactId *) recdata);
1901 :
1902 16 : OldestMemberMXactId[dummyProcNumber] = oldestMember;
1903 16 : }
1904 :
1905 : /*
1906 : * multixact_twophase_postcommit
1907 : * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1908 : */
1909 : void
1910 108 : multixact_twophase_postcommit(FullTransactionId fxid, uint16 info,
1911 : void *recdata, uint32 len)
1912 : {
1913 108 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, true);
1914 :
1915 : Assert(len == sizeof(MultiXactId));
1916 :
1917 108 : OldestMemberMXactId[dummyProcNumber] = InvalidMultiXactId;
1918 108 : }
1919 :
1920 : /*
1921 : * multixact_twophase_postabort
1922 : * This is actually just the same as the COMMIT case.
1923 : */
1924 : void
1925 44 : multixact_twophase_postabort(FullTransactionId fxid, uint16 info,
1926 : void *recdata, uint32 len)
1927 : {
1928 44 : multixact_twophase_postcommit(fxid, info, recdata, len);
1929 44 : }
1930 :
1931 : /*
1932 : * Initialization of shared memory for MultiXact. We use two SLRU areas,
1933 : * thus double memory. Also, reserve space for the shared MultiXactState
1934 : * struct and the per-backend MultiXactId arrays (two of those, too).
1935 : */
1936 : Size
1937 4048 : MultiXactShmemSize(void)
1938 : {
1939 : Size size;
1940 :
1941 : /* We need 2*MaxOldestSlot perBackendXactIds[] entries */
1942 : #define SHARED_MULTIXACT_STATE_SIZE \
1943 : add_size(offsetof(MultiXactStateData, perBackendXactIds), \
1944 : mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1945 :
1946 4048 : size = SHARED_MULTIXACT_STATE_SIZE;
1947 4048 : size = add_size(size, SimpleLruShmemSize(multixact_offset_buffers, 0));
1948 4048 : size = add_size(size, SimpleLruShmemSize(multixact_member_buffers, 0));
1949 :
1950 4048 : return size;
1951 : }
1952 :
1953 : void
1954 2174 : MultiXactShmemInit(void)
1955 : {
1956 : bool found;
1957 :
1958 : debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1959 :
1960 2174 : MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
1961 2174 : MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
1962 :
1963 2174 : SimpleLruInit(MultiXactOffsetCtl,
1964 : "multixact_offset", multixact_offset_buffers, 0,
1965 : "pg_multixact/offsets", LWTRANCHE_MULTIXACTOFFSET_BUFFER,
1966 : LWTRANCHE_MULTIXACTOFFSET_SLRU,
1967 : SYNC_HANDLER_MULTIXACT_OFFSET,
1968 : false);
1969 : SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
1970 2174 : SimpleLruInit(MultiXactMemberCtl,
1971 : "multixact_member", multixact_member_buffers, 0,
1972 : "pg_multixact/members", LWTRANCHE_MULTIXACTMEMBER_BUFFER,
1973 : LWTRANCHE_MULTIXACTMEMBER_SLRU,
1974 : SYNC_HANDLER_MULTIXACT_MEMBER,
1975 : false);
1976 : /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1977 :
1978 : /* Initialize our shared state struct */
1979 2174 : MultiXactState = ShmemInitStruct("Shared MultiXact State",
1980 2174 : SHARED_MULTIXACT_STATE_SIZE,
1981 : &found);
1982 2174 : if (!IsUnderPostmaster)
1983 : {
1984 : Assert(!found);
1985 :
1986 : /* Make sure we zero out the per-backend state */
1987 2174 : MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
1988 2174 : ConditionVariableInit(&MultiXactState->nextoff_cv);
1989 : }
1990 : else
1991 : Assert(found);
1992 :
1993 : /*
1994 : * Set up array pointers.
1995 : */
1996 2174 : OldestMemberMXactId = MultiXactState->perBackendXactIds;
1997 2174 : OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
1998 2174 : }
1999 :
2000 : /*
2001 : * GUC check_hook for multixact_offset_buffers
2002 : */
2003 : bool
2004 2250 : check_multixact_offset_buffers(int *newval, void **extra, GucSource source)
2005 : {
2006 2250 : return check_slru_buffers("multixact_offset_buffers", newval);
2007 : }
2008 :
2009 : /*
2010 : * GUC check_hook for multixact_member_buffers
2011 : */
2012 : bool
2013 2250 : check_multixact_member_buffers(int *newval, void **extra, GucSource source)
2014 : {
2015 2250 : return check_slru_buffers("multixact_member_buffers", newval);
2016 : }
2017 :
2018 : /*
2019 : * This func must be called ONCE on system install. It creates the initial
2020 : * MultiXact segments. (The MultiXacts directories are assumed to have been
2021 : * created by initdb, and MultiXactShmemInit must have been called already.)
2022 : */
2023 : void
2024 100 : BootStrapMultiXact(void)
2025 : {
2026 : /* Zero the initial pages and flush them to disk */
2027 100 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, 0);
2028 100 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, 0);
2029 100 : }
2030 :
2031 : /*
2032 : * MaybeExtendOffsetSlru
2033 : * Extend the offsets SLRU area, if necessary
2034 : *
2035 : * After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
2036 : * contain files that are shorter than necessary; this would occur if the old
2037 : * installation had used multixacts beyond the first page (files cannot be
2038 : * copied, because the on-disk representation is different). pg_upgrade would
2039 : * update pg_control to set the next offset value to be at that position, so
2040 : * that tuples marked as locked by such MultiXacts would be seen as visible
2041 : * without having to consult multixact. However, trying to create and use a
2042 : * new MultiXactId would result in an error because the page on which the new
2043 : * value would reside does not exist. This routine is in charge of creating
2044 : * such pages.
2045 : */
2046 : static void
2047 94 : MaybeExtendOffsetSlru(void)
2048 : {
2049 : int64 pageno;
2050 : LWLock *lock;
2051 :
2052 94 : pageno = MultiXactIdToOffsetPage(MultiXactState->nextMXact);
2053 94 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2054 :
2055 94 : LWLockAcquire(lock, LW_EXCLUSIVE);
2056 :
2057 94 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2058 : {
2059 : int slotno;
2060 :
2061 : /*
2062 : * Fortunately for us, SimpleLruWritePage is already prepared to deal
2063 : * with creating a new segment file even if the page we're writing is
2064 : * not the first in it, so this is enough.
2065 : */
2066 0 : slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2067 0 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
2068 : }
2069 :
2070 94 : LWLockRelease(lock);
2071 94 : }
2072 :
2073 : /*
2074 : * This must be called ONCE during postmaster or standalone-backend startup.
2075 : *
2076 : * StartupXLOG has already established nextMXact/nextOffset by calling
2077 : * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
2078 : * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
2079 : * replayed WAL.
2080 : */
2081 : void
2082 1890 : StartupMultiXact(void)
2083 : {
2084 1890 : MultiXactId multi = MultiXactState->nextMXact;
2085 1890 : MultiXactOffset offset = MultiXactState->nextOffset;
2086 : int64 pageno;
2087 :
2088 : /*
2089 : * Initialize offset's idea of the latest page number.
2090 : */
2091 1890 : pageno = MultiXactIdToOffsetPage(multi);
2092 1890 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2093 : pageno);
2094 :
2095 : /*
2096 : * Initialize member's idea of the latest page number.
2097 : */
2098 1890 : pageno = MXOffsetToMemberPage(offset);
2099 1890 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2100 : pageno);
2101 1890 : }
2102 :
2103 : /*
2104 : * This must be called ONCE at the end of startup/recovery.
2105 : */
2106 : void
2107 1772 : TrimMultiXact(void)
2108 : {
2109 : MultiXactId nextMXact;
2110 : MultiXactOffset offset;
2111 : MultiXactId oldestMXact;
2112 : Oid oldestMXactDB;
2113 : int64 pageno;
2114 : int entryno;
2115 : int flagsoff;
2116 :
2117 1772 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2118 1772 : nextMXact = MultiXactState->nextMXact;
2119 1772 : offset = MultiXactState->nextOffset;
2120 1772 : oldestMXact = MultiXactState->oldestMultiXactId;
2121 1772 : oldestMXactDB = MultiXactState->oldestMultiXactDB;
2122 1772 : LWLockRelease(MultiXactGenLock);
2123 :
2124 : /* Clean up offsets state */
2125 :
2126 : /*
2127 : * (Re-)Initialize our idea of the latest page number for offsets.
2128 : */
2129 1772 : pageno = MultiXactIdToOffsetPage(nextMXact);
2130 1772 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2131 : pageno);
2132 :
2133 : /*
2134 : * Zero out the remainder of the current offsets page. See notes in
2135 : * TrimCLOG() for background. Unlike CLOG, some WAL record covers every
2136 : * pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
2137 : * rule "write xlog before data," nextMXact successors may carry obsolete,
2138 : * nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
2139 : * operates normally.
2140 : */
2141 1772 : entryno = MultiXactIdToOffsetEntry(nextMXact);
2142 1772 : if (entryno != 0)
2143 : {
2144 : int slotno;
2145 : MultiXactOffset *offptr;
2146 1770 : LWLock *lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2147 :
2148 1770 : LWLockAcquire(lock, LW_EXCLUSIVE);
2149 1770 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
2150 1770 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2151 1770 : offptr += entryno;
2152 :
2153 1770 : MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
2154 :
2155 1770 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
2156 1770 : LWLockRelease(lock);
2157 : }
2158 :
2159 : /*
2160 : * And the same for members.
2161 : *
2162 : * (Re-)Initialize our idea of the latest page number for members.
2163 : */
2164 1772 : pageno = MXOffsetToMemberPage(offset);
2165 1772 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2166 : pageno);
2167 :
2168 : /*
2169 : * Zero out the remainder of the current members page. See notes in
2170 : * TrimCLOG() for motivation.
2171 : */
2172 1772 : flagsoff = MXOffsetToFlagsOffset(offset);
2173 1772 : if (flagsoff != 0)
2174 : {
2175 : int slotno;
2176 : TransactionId *xidptr;
2177 : int memberoff;
2178 22 : LWLock *lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2179 :
2180 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2181 22 : memberoff = MXOffsetToMemberOffset(offset);
2182 22 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
2183 22 : xidptr = (TransactionId *)
2184 22 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
2185 :
2186 22 : MemSet(xidptr, 0, BLCKSZ - memberoff);
2187 :
2188 : /*
2189 : * Note: we don't need to zero out the flag bits in the remaining
2190 : * members of the current group, because they are always reset before
2191 : * writing.
2192 : */
2193 :
2194 22 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
2195 22 : LWLockRelease(lock);
2196 : }
2197 :
2198 : /* signal that we're officially up */
2199 1772 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2200 1772 : MultiXactState->finishedStartup = true;
2201 1772 : LWLockRelease(MultiXactGenLock);
2202 :
2203 : /* Now compute how far away the next members wraparound is. */
2204 1772 : SetMultiXactIdLimit(oldestMXact, oldestMXactDB, true);
2205 1772 : }
2206 :
2207 : /*
2208 : * Get the MultiXact data to save in a checkpoint record
2209 : */
2210 : void
2211 3046 : MultiXactGetCheckptMulti(bool is_shutdown,
2212 : MultiXactId *nextMulti,
2213 : MultiXactOffset *nextMultiOffset,
2214 : MultiXactId *oldestMulti,
2215 : Oid *oldestMultiDB)
2216 : {
2217 3046 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2218 3046 : *nextMulti = MultiXactState->nextMXact;
2219 3046 : *nextMultiOffset = MultiXactState->nextOffset;
2220 3046 : *oldestMulti = MultiXactState->oldestMultiXactId;
2221 3046 : *oldestMultiDB = MultiXactState->oldestMultiXactDB;
2222 3046 : LWLockRelease(MultiXactGenLock);
2223 :
2224 : debug_elog6(DEBUG2,
2225 : "MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
2226 : *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
2227 3046 : }
2228 :
2229 : /*
2230 : * Perform a checkpoint --- either during shutdown, or on-the-fly
2231 : */
2232 : void
2233 3422 : CheckPointMultiXact(void)
2234 : {
2235 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
2236 :
2237 : /*
2238 : * Write dirty MultiXact pages to disk. This may result in sync requests
2239 : * queued for later handling by ProcessSyncRequests(), as part of the
2240 : * checkpoint.
2241 : */
2242 3422 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2243 3422 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2244 :
2245 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
2246 3422 : }
2247 :
2248 : /*
2249 : * Set the next-to-be-assigned MultiXactId and offset
2250 : *
2251 : * This is used when we can determine the correct next ID/offset exactly
2252 : * from a checkpoint record. Although this is only called during bootstrap
2253 : * and XLog replay, we take the lock in case any hot-standby backends are
2254 : * examining the values.
2255 : */
2256 : void
2257 2060 : MultiXactSetNextMXact(MultiXactId nextMulti,
2258 : MultiXactOffset nextMultiOffset)
2259 : {
2260 : debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
2261 : nextMulti, nextMultiOffset);
2262 2060 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2263 2060 : MultiXactState->nextMXact = nextMulti;
2264 2060 : MultiXactState->nextOffset = nextMultiOffset;
2265 2060 : LWLockRelease(MultiXactGenLock);
2266 :
2267 : /*
2268 : * During a binary upgrade, make sure that the offsets SLRU is large
2269 : * enough to contain the next value that would be created.
2270 : *
2271 : * We need to do this pretty early during the first startup in binary
2272 : * upgrade mode: before StartupMultiXact() in fact, because this routine
2273 : * is called even before that by StartupXLOG(). And we can't do it
2274 : * earlier than at this point, because during that first call of this
2275 : * routine we determine the MultiXactState->nextMXact value that
2276 : * MaybeExtendOffsetSlru needs.
2277 : */
2278 2060 : if (IsBinaryUpgrade)
2279 94 : MaybeExtendOffsetSlru();
2280 2060 : }
2281 :
2282 : /*
2283 : * Determine the last safe MultiXactId to allocate given the currently oldest
2284 : * datminmxid (ie, the oldest MultiXactId that might exist in any database
2285 : * of our cluster), and the OID of the (or a) database with that value.
2286 : *
2287 : * is_startup is true when we are just starting the cluster, false when we
2288 : * are updating state in a running cluster. This only affects log messages.
2289 : */
2290 : void
2291 5746 : SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
2292 : bool is_startup)
2293 : {
2294 : MultiXactId multiVacLimit;
2295 : MultiXactId multiWarnLimit;
2296 : MultiXactId multiStopLimit;
2297 : MultiXactId multiWrapLimit;
2298 : MultiXactId curMulti;
2299 : bool needs_offset_vacuum;
2300 :
2301 : Assert(MultiXactIdIsValid(oldest_datminmxid));
2302 :
2303 : /*
2304 : * We pretend that a wrap will happen halfway through the multixact ID
2305 : * space, but that's not really true, because multixacts wrap differently
2306 : * from transaction IDs. Note that, separately from any concern about
2307 : * multixact IDs wrapping, we must ensure that multixact members do not
2308 : * wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
2309 : */
2310 5746 : multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2311 5746 : if (multiWrapLimit < FirstMultiXactId)
2312 0 : multiWrapLimit += FirstMultiXactId;
2313 :
2314 : /*
2315 : * We'll refuse to continue assigning MultiXactIds once we get within 3M
2316 : * multi of data loss. See SetTransactionIdLimit.
2317 : */
2318 5746 : multiStopLimit = multiWrapLimit - 3000000;
2319 5746 : if (multiStopLimit < FirstMultiXactId)
2320 0 : multiStopLimit -= FirstMultiXactId;
2321 :
2322 : /*
2323 : * We'll start complaining loudly when we get within 40M multis of data
2324 : * loss. This is kind of arbitrary, but if you let your gas gauge get
2325 : * down to 2% of full, would you be looking for the next gas station? We
2326 : * need to be fairly liberal about this number because there are lots of
2327 : * scenarios where most transactions are done by automatic clients that
2328 : * won't pay attention to warnings. (No, we're not gonna make this
2329 : * configurable. If you know enough to configure it, you know enough to
2330 : * not get in this kind of trouble in the first place.)
2331 : */
2332 5746 : multiWarnLimit = multiWrapLimit - 40000000;
2333 5746 : if (multiWarnLimit < FirstMultiXactId)
2334 0 : multiWarnLimit -= FirstMultiXactId;
2335 :
2336 : /*
2337 : * We'll start trying to force autovacuums when oldest_datminmxid gets to
2338 : * be more than autovacuum_multixact_freeze_max_age mxids old.
2339 : *
2340 : * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2341 : * so that we don't have to worry about dealing with on-the-fly changes in
2342 : * its value. See SetTransactionIdLimit.
2343 : */
2344 5746 : multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2345 5746 : if (multiVacLimit < FirstMultiXactId)
2346 0 : multiVacLimit += FirstMultiXactId;
2347 :
2348 : /* Grab lock for just long enough to set the new limit values */
2349 5746 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2350 5746 : MultiXactState->oldestMultiXactId = oldest_datminmxid;
2351 5746 : MultiXactState->oldestMultiXactDB = oldest_datoid;
2352 5746 : MultiXactState->multiVacLimit = multiVacLimit;
2353 5746 : MultiXactState->multiWarnLimit = multiWarnLimit;
2354 5746 : MultiXactState->multiStopLimit = multiStopLimit;
2355 5746 : MultiXactState->multiWrapLimit = multiWrapLimit;
2356 5746 : curMulti = MultiXactState->nextMXact;
2357 5746 : LWLockRelease(MultiXactGenLock);
2358 :
2359 : /* Log the info */
2360 5746 : ereport(DEBUG1,
2361 : (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2362 : multiWrapLimit, oldest_datoid)));
2363 :
2364 : /*
2365 : * Computing the actual limits is only possible once the data directory is
2366 : * in a consistent state. There's no need to compute the limits while
2367 : * still replaying WAL - no decisions about new multis are made even
2368 : * though multixact creations might be replayed. So we'll only do further
2369 : * checks after TrimMultiXact() has been called.
2370 : */
2371 5746 : if (!MultiXactState->finishedStartup)
2372 1992 : return;
2373 :
2374 : Assert(!InRecovery);
2375 :
2376 : /* Set limits for offset vacuum. */
2377 3754 : needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
2378 :
2379 : /*
2380 : * If past the autovacuum force point, immediately signal an autovac
2381 : * request. The reason for this is that autovac only processes one
2382 : * database per invocation. Once it's finished cleaning up the oldest
2383 : * database, it'll call here, and we'll signal the postmaster to start
2384 : * another iteration immediately if there are still any old databases.
2385 : */
2386 3754 : if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
2387 0 : needs_offset_vacuum) && IsUnderPostmaster)
2388 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
2389 :
2390 : /* Give an immediate warning if past the wrap warn point */
2391 3754 : if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2392 : {
2393 : char *oldest_datname;
2394 :
2395 : /*
2396 : * We can be called when not inside a transaction, for example during
2397 : * StartupXLOG(). In such a case we cannot do database access, so we
2398 : * must just report the oldest DB's OID.
2399 : *
2400 : * Note: it's also possible that get_database_name fails and returns
2401 : * NULL, for example because the database just got dropped. We'll
2402 : * still warn, even though the warning might now be unnecessary.
2403 : */
2404 0 : if (IsTransactionState())
2405 0 : oldest_datname = get_database_name(oldest_datoid);
2406 : else
2407 0 : oldest_datname = NULL;
2408 :
2409 0 : if (oldest_datname)
2410 0 : ereport(WARNING,
2411 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2412 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2413 : multiWrapLimit - curMulti,
2414 : oldest_datname,
2415 : multiWrapLimit - curMulti),
2416 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2417 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2418 : else
2419 0 : ereport(WARNING,
2420 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2421 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2422 : multiWrapLimit - curMulti,
2423 : oldest_datoid,
2424 : multiWrapLimit - curMulti),
2425 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2426 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2427 : }
2428 : }
2429 :
2430 : /*
2431 : * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2432 : * and similarly nextOffset is at least minMultiOffset.
2433 : *
2434 : * This is used when we can determine minimum safe values from an XLog
2435 : * record (either an on-line checkpoint or an mxact creation log entry).
2436 : * Although this is only called during XLog replay, we take the lock in case
2437 : * any hot-standby backends are examining the values.
2438 : */
2439 : void
2440 1356 : MultiXactAdvanceNextMXact(MultiXactId minMulti,
2441 : MultiXactOffset minMultiOffset)
2442 : {
2443 1356 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2444 1356 : if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
2445 : {
2446 : debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2447 4 : MultiXactState->nextMXact = minMulti;
2448 : }
2449 1356 : if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
2450 : {
2451 : debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
2452 : minMultiOffset);
2453 4 : MultiXactState->nextOffset = minMultiOffset;
2454 : }
2455 1356 : LWLockRelease(MultiXactGenLock);
2456 1356 : }
2457 :
2458 : /*
2459 : * Update our oldestMultiXactId value, but only if it's more recent than what
2460 : * we had.
2461 : *
2462 : * This may only be called during WAL replay.
2463 : */
2464 : void
2465 1420 : MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2466 : {
2467 : Assert(InRecovery);
2468 :
2469 1420 : if (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
2470 0 : SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
2471 1420 : }
2472 :
2473 : /*
2474 : * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2475 : *
2476 : * NB: this is called while holding MultiXactGenLock. We want it to be very
2477 : * fast most of the time; even when it's not so fast, no actual I/O need
2478 : * happen unless we're forced to write out a dirty log or xlog page to make
2479 : * room in shared memory.
2480 : */
2481 : static void
2482 544 : ExtendMultiXactOffset(MultiXactId multi)
2483 : {
2484 : int64 pageno;
2485 : LWLock *lock;
2486 :
2487 : /*
2488 : * No work except at first MultiXactId of a page. But beware: just after
2489 : * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2490 : */
2491 544 : if (MultiXactIdToOffsetEntry(multi) != 0 &&
2492 : multi != FirstMultiXactId)
2493 522 : return;
2494 :
2495 22 : pageno = MultiXactIdToOffsetPage(multi);
2496 22 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2497 :
2498 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2499 :
2500 : /* Zero the page and make a WAL entry about it */
2501 22 : SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2502 22 : XLogSimpleInsertInt64(RM_MULTIXACT_ID, XLOG_MULTIXACT_ZERO_OFF_PAGE,
2503 : pageno);
2504 :
2505 22 : LWLockRelease(lock);
2506 : }
2507 :
2508 : /*
2509 : * Make sure that MultiXactMember has room for the members of a newly-
2510 : * allocated MultiXactId.
2511 : *
2512 : * Like the above routine, this is called while holding MultiXactGenLock;
2513 : * same comments apply.
2514 : */
2515 : static void
2516 544 : ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
2517 : {
2518 : /*
2519 : * It's possible that the members span more than one page of the members
2520 : * file, so we loop to ensure we consider each page. The coding is not
2521 : * optimal if the members span several pages, but that seems unusual
2522 : * enough to not worry much about.
2523 : */
2524 1088 : while (nmembers > 0)
2525 : {
2526 : int flagsoff;
2527 : int flagsbit;
2528 : uint32 difference;
2529 :
2530 : /*
2531 : * Only zero when at first entry of a page.
2532 : */
2533 544 : flagsoff = MXOffsetToFlagsOffset(offset);
2534 544 : flagsbit = MXOffsetToFlagsBitShift(offset);
2535 544 : if (flagsoff == 0 && flagsbit == 0)
2536 : {
2537 : int64 pageno;
2538 : LWLock *lock;
2539 :
2540 22 : pageno = MXOffsetToMemberPage(offset);
2541 22 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2542 :
2543 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2544 :
2545 : /* Zero the page and make a WAL entry about it */
2546 22 : SimpleLruZeroPage(MultiXactMemberCtl, pageno);
2547 22 : XLogSimpleInsertInt64(RM_MULTIXACT_ID,
2548 : XLOG_MULTIXACT_ZERO_MEM_PAGE, pageno);
2549 :
2550 22 : LWLockRelease(lock);
2551 : }
2552 :
2553 : /*
2554 : * Compute the number of items till end of current page. Careful: if
2555 : * addition of unsigned ints wraps around, we're at the last page of
2556 : * the last segment; since that page holds a different number of items
2557 : * than other pages, we need to do it differently.
2558 : */
2559 544 : if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
2560 : {
2561 : /*
2562 : * This is the last page of the last segment; we can compute the
2563 : * number of items left to allocate in it without modulo
2564 : * arithmetic.
2565 : */
2566 0 : difference = MaxMultiXactOffset - offset + 1;
2567 : }
2568 : else
2569 544 : difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
2570 :
2571 : /*
2572 : * Advance to next page, taking care to properly handle the wraparound
2573 : * case. OK if nmembers goes negative.
2574 : */
2575 544 : nmembers -= difference;
2576 544 : offset += difference;
2577 : }
2578 544 : }
2579 :
2580 : /*
2581 : * GetOldestMultiXactId
2582 : *
2583 : * Return the oldest MultiXactId that's still possibly still seen as live by
2584 : * any running transaction. Older ones might still exist on disk, but they no
2585 : * longer have any running member transaction.
2586 : *
2587 : * It's not safe to truncate MultiXact SLRU segments on the value returned by
2588 : * this function; however, it can be set as the new relminmxid for any table
2589 : * that VACUUM knows has no remaining MXIDs < the same value. It is only safe
2590 : * to truncate SLRUs when no table can possibly still have a referencing MXID.
2591 : */
2592 : MultiXactId
2593 277996 : GetOldestMultiXactId(void)
2594 : {
2595 : MultiXactId oldestMXact;
2596 : MultiXactId nextMXact;
2597 : int i;
2598 :
2599 : /*
2600 : * This is the oldest valid value among all the OldestMemberMXactId[] and
2601 : * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2602 : */
2603 277996 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2604 :
2605 : /*
2606 : * We have to beware of the possibility that nextMXact is in the
2607 : * wrapped-around state. We don't fix the counter itself here, but we
2608 : * must be sure to use a valid value in our calculation.
2609 : */
2610 277996 : nextMXact = MultiXactState->nextMXact;
2611 277996 : if (nextMXact < FirstMultiXactId)
2612 0 : nextMXact = FirstMultiXactId;
2613 :
2614 277996 : oldestMXact = nextMXact;
2615 34610850 : for (i = 0; i < MaxOldestSlot; i++)
2616 : {
2617 : MultiXactId thisoldest;
2618 :
2619 34332854 : thisoldest = OldestMemberMXactId[i];
2620 34409028 : if (MultiXactIdIsValid(thisoldest) &&
2621 76174 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2622 14 : oldestMXact = thisoldest;
2623 34332854 : thisoldest = OldestVisibleMXactId[i];
2624 34333094 : if (MultiXactIdIsValid(thisoldest) &&
2625 240 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2626 4 : oldestMXact = thisoldest;
2627 : }
2628 :
2629 277996 : LWLockRelease(MultiXactGenLock);
2630 :
2631 277996 : return oldestMXact;
2632 : }
2633 :
2634 : /*
2635 : * Determine how aggressively we need to vacuum in order to prevent member
2636 : * wraparound.
2637 : *
2638 : * To do so determine what's the oldest member offset and install the limit
2639 : * info in MultiXactState, where it can be used to prevent overrun of old data
2640 : * in the members SLRU area.
2641 : *
2642 : * The return value is true if emergency autovacuum is required and false
2643 : * otherwise.
2644 : */
2645 : static bool
2646 3754 : SetOffsetVacuumLimit(bool is_startup)
2647 : {
2648 : MultiXactId oldestMultiXactId;
2649 : MultiXactId nextMXact;
2650 3754 : MultiXactOffset oldestOffset = 0; /* placate compiler */
2651 : MultiXactOffset prevOldestOffset;
2652 : MultiXactOffset nextOffset;
2653 3754 : bool oldestOffsetKnown = false;
2654 : bool prevOldestOffsetKnown;
2655 3754 : MultiXactOffset offsetStopLimit = 0;
2656 : MultiXactOffset prevOffsetStopLimit;
2657 :
2658 : /*
2659 : * NB: Have to prevent concurrent truncation, we might otherwise try to
2660 : * lookup an oldestMulti that's concurrently getting truncated away.
2661 : */
2662 3754 : LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2663 :
2664 : /* Read relevant fields from shared memory. */
2665 3754 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2666 3754 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2667 3754 : nextMXact = MultiXactState->nextMXact;
2668 3754 : nextOffset = MultiXactState->nextOffset;
2669 3754 : prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2670 3754 : prevOldestOffset = MultiXactState->oldestOffset;
2671 3754 : prevOffsetStopLimit = MultiXactState->offsetStopLimit;
2672 : Assert(MultiXactState->finishedStartup);
2673 3754 : LWLockRelease(MultiXactGenLock);
2674 :
2675 : /*
2676 : * Determine the offset of the oldest multixact. Normally, we can read
2677 : * the offset from the multixact itself, but there's an important special
2678 : * case: if there are no multixacts in existence at all, oldestMXact
2679 : * obviously can't point to one. It will instead point to the multixact
2680 : * ID that will be assigned the next time one is needed.
2681 : */
2682 3754 : if (oldestMultiXactId == nextMXact)
2683 : {
2684 : /*
2685 : * When the next multixact gets created, it will be stored at the next
2686 : * offset.
2687 : */
2688 3726 : oldestOffset = nextOffset;
2689 3726 : oldestOffsetKnown = true;
2690 : }
2691 : else
2692 : {
2693 : /*
2694 : * Figure out where the oldest existing multixact's offsets are
2695 : * stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
2696 : * the supposedly-earliest multixact might not really exist. We are
2697 : * careful not to fail in that case.
2698 : */
2699 : oldestOffsetKnown =
2700 28 : find_multixact_start(oldestMultiXactId, &oldestOffset);
2701 :
2702 28 : if (oldestOffsetKnown)
2703 28 : ereport(DEBUG1,
2704 : (errmsg_internal("oldest MultiXactId member is at offset %u",
2705 : oldestOffset)));
2706 : else
2707 0 : ereport(LOG,
2708 : (errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
2709 : oldestMultiXactId)));
2710 : }
2711 :
2712 3754 : LWLockRelease(MultiXactTruncationLock);
2713 :
2714 : /*
2715 : * If we can, compute limits (and install them MultiXactState) to prevent
2716 : * overrun of old data in the members SLRU area. We can only do so if the
2717 : * oldest offset is known though.
2718 : */
2719 3754 : if (oldestOffsetKnown)
2720 : {
2721 : /* move back to start of the corresponding segment */
2722 3754 : offsetStopLimit = oldestOffset - (oldestOffset %
2723 : (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT));
2724 :
2725 : /* always leave one segment before the wraparound point */
2726 3754 : offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
2727 :
2728 3754 : if (!prevOldestOffsetKnown && !is_startup)
2729 0 : ereport(LOG,
2730 : (errmsg("MultiXact member wraparound protections are now enabled")));
2731 :
2732 3754 : ereport(DEBUG1,
2733 : (errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
2734 : offsetStopLimit, oldestMultiXactId)));
2735 : }
2736 0 : else if (prevOldestOffsetKnown)
2737 : {
2738 : /*
2739 : * If we failed to get the oldest offset this time, but we have a
2740 : * value from a previous pass through this function, use the old
2741 : * values rather than automatically forcing an emergency autovacuum
2742 : * cycle again.
2743 : */
2744 0 : oldestOffset = prevOldestOffset;
2745 0 : oldestOffsetKnown = true;
2746 0 : offsetStopLimit = prevOffsetStopLimit;
2747 : }
2748 :
2749 : /* Install the computed values */
2750 3754 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2751 3754 : MultiXactState->oldestOffset = oldestOffset;
2752 3754 : MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
2753 3754 : MultiXactState->offsetStopLimit = offsetStopLimit;
2754 3754 : LWLockRelease(MultiXactGenLock);
2755 :
2756 : /*
2757 : * Do we need an emergency autovacuum? If we're not sure, assume yes.
2758 : */
2759 7508 : return !oldestOffsetKnown ||
2760 3754 : (nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
2761 : }
2762 :
2763 : /*
2764 : * Return whether adding "distance" to "start" would move past "boundary".
2765 : *
2766 : * We use this to determine whether the addition is "wrapping around" the
2767 : * boundary point, hence the name. The reason we don't want to use the regular
2768 : * 2^31-modulo arithmetic here is that we want to be able to use the whole of
2769 : * the 2^32-1 space here, allowing for more multixacts than would fit
2770 : * otherwise.
2771 : */
2772 : static bool
2773 1088 : MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start,
2774 : uint32 distance)
2775 : {
2776 : MultiXactOffset finish;
2777 :
2778 : /*
2779 : * Note that offset number 0 is not used (see GetMultiXactIdMembers), so
2780 : * if the addition wraps around the UINT_MAX boundary, skip that value.
2781 : */
2782 1088 : finish = start + distance;
2783 1088 : if (finish < start)
2784 0 : finish++;
2785 :
2786 : /*-----------------------------------------------------------------------
2787 : * When the boundary is numerically greater than the starting point, any
2788 : * value numerically between the two is not wrapped:
2789 : *
2790 : * <----S----B---->
2791 : * [---) = F wrapped past B (and UINT_MAX)
2792 : * [---) = F not wrapped
2793 : * [----] = F wrapped past B
2794 : *
2795 : * When the boundary is numerically less than the starting point (i.e. the
2796 : * UINT_MAX wraparound occurs somewhere in between) then all values in
2797 : * between are wrapped:
2798 : *
2799 : * <----B----S---->
2800 : * [---) = F not wrapped past B (but wrapped past UINT_MAX)
2801 : * [---) = F wrapped past B (and UINT_MAX)
2802 : * [----] = F not wrapped
2803 : *-----------------------------------------------------------------------
2804 : */
2805 1088 : if (start < boundary)
2806 1088 : return finish >= boundary || finish < start;
2807 : else
2808 0 : return finish >= boundary && finish < start;
2809 : }
2810 :
2811 : /*
2812 : * Find the starting offset of the given MultiXactId.
2813 : *
2814 : * Returns false if the file containing the multi does not exist on disk.
2815 : * Otherwise, returns true and sets *result to the starting member offset.
2816 : *
2817 : * This function does not prevent concurrent truncation, so if that's
2818 : * required, the caller has to protect against that.
2819 : */
2820 : static bool
2821 28 : find_multixact_start(MultiXactId multi, MultiXactOffset *result)
2822 : {
2823 : MultiXactOffset offset;
2824 : int64 pageno;
2825 : int entryno;
2826 : int slotno;
2827 : MultiXactOffset *offptr;
2828 :
2829 : Assert(MultiXactState->finishedStartup);
2830 :
2831 28 : pageno = MultiXactIdToOffsetPage(multi);
2832 28 : entryno = MultiXactIdToOffsetEntry(multi);
2833 :
2834 : /*
2835 : * Write out dirty data, so PhysicalPageExists can work correctly.
2836 : */
2837 28 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2838 28 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2839 :
2840 28 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2841 0 : return false;
2842 :
2843 : /* lock is acquired by SimpleLruReadPage_ReadOnly */
2844 28 : slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2845 28 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2846 28 : offptr += entryno;
2847 28 : offset = *offptr;
2848 28 : LWLockRelease(SimpleLruGetBankLock(MultiXactOffsetCtl, pageno));
2849 :
2850 28 : *result = offset;
2851 28 : return true;
2852 : }
2853 :
2854 : /*
2855 : * GetMultiXactInfo
2856 : *
2857 : * Returns information about the current MultiXact state, as of:
2858 : * multixacts: Number of MultiXacts (nextMultiXactId - oldestMultiXactId)
2859 : * members: Number of member entries (nextOffset - oldestOffset)
2860 : * oldestMultiXactId: Oldest MultiXact ID still in use
2861 : * oldestOffset: Oldest offset still in use
2862 : *
2863 : * Returns false if unable to determine, the oldest offset being unknown.
2864 : */
2865 : bool
2866 215588 : GetMultiXactInfo(uint32 *multixacts, MultiXactOffset *members,
2867 : MultiXactId *oldestMultiXactId, MultiXactOffset *oldestOffset)
2868 : {
2869 : MultiXactOffset nextOffset;
2870 : MultiXactId nextMultiXactId;
2871 : bool oldestOffsetKnown;
2872 :
2873 215588 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2874 215588 : nextOffset = MultiXactState->nextOffset;
2875 215588 : *oldestMultiXactId = MultiXactState->oldestMultiXactId;
2876 215588 : nextMultiXactId = MultiXactState->nextMXact;
2877 215588 : *oldestOffset = MultiXactState->oldestOffset;
2878 215588 : oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2879 215588 : LWLockRelease(MultiXactGenLock);
2880 :
2881 215588 : if (!oldestOffsetKnown)
2882 : {
2883 0 : *members = 0;
2884 0 : *multixacts = 0;
2885 0 : *oldestMultiXactId = InvalidMultiXactId;
2886 0 : *oldestOffset = 0;
2887 0 : return false;
2888 : }
2889 :
2890 215588 : *members = nextOffset - *oldestOffset;
2891 215588 : *multixacts = nextMultiXactId - *oldestMultiXactId;
2892 215588 : return true;
2893 : }
2894 :
2895 : /*
2896 : * Multixact members can be removed once the multixacts that refer to them
2897 : * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2898 : * vacuum_multixact_freeze_table_age work together to make sure we never have
2899 : * too many multixacts; we hope that, at least under normal circumstances,
2900 : * this will also be sufficient to keep us from using too many offsets.
2901 : * However, if the average multixact has many members, we might exhaust the
2902 : * members space while still using few enough members that these limits fail
2903 : * to trigger relminmxid advancement by VACUUM. At that point, we'd have no
2904 : * choice but to start failing multixact-creating operations with an error.
2905 : *
2906 : * To prevent that, if more than a threshold portion of the members space is
2907 : * used, we effectively reduce autovacuum_multixact_freeze_max_age and
2908 : * to a value just less than the number of multixacts in use. We hope that
2909 : * this will quickly trigger autovacuuming on the table or tables with the
2910 : * oldest relminmxid, thus allowing datminmxid values to advance and removing
2911 : * some members.
2912 : *
2913 : * As the fraction of the member space currently in use grows, we become
2914 : * more aggressive in clamping this value. That not only causes autovacuum
2915 : * to ramp up, but also makes any manual vacuums the user issues more
2916 : * aggressive. This happens because vacuum_get_cutoffs() will clamp the
2917 : * freeze table and the minimum freeze age cutoffs based on the effective
2918 : * autovacuum_multixact_freeze_max_age this function returns. In the worst
2919 : * case, we'll claim the freeze_max_age to zero, and every vacuum of any
2920 : * table will freeze every multixact.
2921 : */
2922 : int
2923 215588 : MultiXactMemberFreezeThreshold(void)
2924 : {
2925 : MultiXactOffset members;
2926 : uint32 multixacts;
2927 : uint32 victim_multixacts;
2928 : double fraction;
2929 : int result;
2930 : MultiXactId oldestMultiXactId;
2931 : MultiXactOffset oldestOffset;
2932 :
2933 : /* If we can't determine member space utilization, assume the worst. */
2934 215588 : if (!GetMultiXactInfo(&multixacts, &members, &oldestMultiXactId, &oldestOffset))
2935 0 : return 0;
2936 :
2937 : /* If member space utilization is low, no special action is required. */
2938 215588 : if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
2939 215588 : return autovacuum_multixact_freeze_max_age;
2940 :
2941 : /*
2942 : * Compute a target for relminmxid advancement. The number of multixacts
2943 : * we try to eliminate from the system is based on how far we are past
2944 : * MULTIXACT_MEMBER_SAFE_THRESHOLD.
2945 : */
2946 0 : fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
2947 : (MULTIXACT_MEMBER_DANGER_THRESHOLD - MULTIXACT_MEMBER_SAFE_THRESHOLD);
2948 0 : victim_multixacts = multixacts * fraction;
2949 :
2950 : /* fraction could be > 1.0, but lowest possible freeze age is zero */
2951 0 : if (victim_multixacts > multixacts)
2952 0 : return 0;
2953 0 : result = multixacts - victim_multixacts;
2954 :
2955 : /*
2956 : * Clamp to autovacuum_multixact_freeze_max_age, so that we never make
2957 : * autovacuum less aggressive than it would otherwise be.
2958 : */
2959 0 : return Min(result, autovacuum_multixact_freeze_max_age);
2960 : }
2961 :
2962 : typedef struct mxtruncinfo
2963 : {
2964 : int64 earliestExistingPage;
2965 : } mxtruncinfo;
2966 :
2967 : /*
2968 : * SlruScanDirectory callback
2969 : * This callback determines the earliest existing page number.
2970 : */
2971 : static bool
2972 0 : SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int64 segpage, void *data)
2973 : {
2974 0 : mxtruncinfo *trunc = (mxtruncinfo *) data;
2975 :
2976 0 : if (trunc->earliestExistingPage == -1 ||
2977 0 : ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
2978 : {
2979 0 : trunc->earliestExistingPage = segpage;
2980 : }
2981 :
2982 0 : return false; /* keep going */
2983 : }
2984 :
2985 :
2986 : /*
2987 : * Delete members segments [oldest, newOldest)
2988 : *
2989 : * The members SLRU can, in contrast to the offsets one, be filled to almost
2990 : * the full range at once. This means SimpleLruTruncate() can't trivially be
2991 : * used - instead the to-be-deleted range is computed using the offsets
2992 : * SLRU. C.f. TruncateMultiXact().
2993 : */
2994 : static void
2995 0 : PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
2996 : {
2997 0 : const int64 maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
2998 0 : int64 startsegment = MXOffsetToMemberSegment(oldestOffset);
2999 0 : int64 endsegment = MXOffsetToMemberSegment(newOldestOffset);
3000 0 : int64 segment = startsegment;
3001 :
3002 : /*
3003 : * Delete all the segments but the last one. The last segment can still
3004 : * contain, possibly partially, valid data.
3005 : */
3006 0 : while (segment != endsegment)
3007 : {
3008 0 : elog(DEBUG2, "truncating multixact members segment %" PRIx64,
3009 : segment);
3010 0 : SlruDeleteSegment(MultiXactMemberCtl, segment);
3011 :
3012 : /* move to next segment, handling wraparound correctly */
3013 0 : if (segment == maxsegment)
3014 0 : segment = 0;
3015 : else
3016 0 : segment += 1;
3017 : }
3018 0 : }
3019 :
3020 : /*
3021 : * Delete offsets segments [oldest, newOldest)
3022 : */
3023 : static void
3024 0 : PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
3025 : {
3026 : /*
3027 : * We step back one multixact to avoid passing a cutoff page that hasn't
3028 : * been created yet in the rare case that oldestMulti would be the first
3029 : * item on a page and oldestMulti == nextMulti. In that case, if we
3030 : * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
3031 : * detection.
3032 : */
3033 0 : SimpleLruTruncate(MultiXactOffsetCtl,
3034 : MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
3035 0 : }
3036 :
3037 : /*
3038 : * Remove all MultiXactOffset and MultiXactMember segments before the oldest
3039 : * ones still of interest.
3040 : *
3041 : * This is only called on a primary as part of vacuum (via
3042 : * vac_truncate_clog()). During recovery truncation is done by replaying
3043 : * truncation WAL records logged here.
3044 : *
3045 : * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
3046 : * is one of the databases preventing newOldestMulti from increasing.
3047 : */
3048 : void
3049 1982 : TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
3050 : {
3051 : MultiXactId oldestMulti;
3052 : MultiXactId nextMulti;
3053 : MultiXactOffset newOldestOffset;
3054 : MultiXactOffset oldestOffset;
3055 : MultiXactOffset nextOffset;
3056 : mxtruncinfo trunc;
3057 : MultiXactId earliest;
3058 :
3059 : Assert(!RecoveryInProgress());
3060 : Assert(MultiXactState->finishedStartup);
3061 :
3062 : /*
3063 : * We can only allow one truncation to happen at once. Otherwise parts of
3064 : * members might vanish while we're doing lookups or similar. There's no
3065 : * need to have an interlock with creating new multis or such, since those
3066 : * are constrained by the limits (which only grow, never shrink).
3067 : */
3068 1982 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3069 :
3070 1982 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
3071 1982 : nextMulti = MultiXactState->nextMXact;
3072 1982 : nextOffset = MultiXactState->nextOffset;
3073 1982 : oldestMulti = MultiXactState->oldestMultiXactId;
3074 1982 : LWLockRelease(MultiXactGenLock);
3075 : Assert(MultiXactIdIsValid(oldestMulti));
3076 :
3077 : /*
3078 : * Make sure to only attempt truncation if there's values to truncate
3079 : * away. In normal processing values shouldn't go backwards, but there's
3080 : * some corner cases (due to bugs) where that's possible.
3081 : */
3082 1982 : if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
3083 : {
3084 1982 : LWLockRelease(MultiXactTruncationLock);
3085 1982 : return;
3086 : }
3087 :
3088 : /*
3089 : * Note we can't just plow ahead with the truncation; it's possible that
3090 : * there are no segments to truncate, which is a problem because we are
3091 : * going to attempt to read the offsets page to determine where to
3092 : * truncate the members SLRU. So we first scan the directory to determine
3093 : * the earliest offsets page number that we can read without error.
3094 : *
3095 : * When nextMXact is less than one segment away from multiWrapLimit,
3096 : * SlruScanDirCbFindEarliest can find some early segment other than the
3097 : * actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
3098 : * returns false, because not all pairs of entries have the same answer.)
3099 : * That can also arise when an earlier truncation attempt failed unlink()
3100 : * or returned early from this function. The only consequence is
3101 : * returning early, which wastes space that we could have liberated.
3102 : *
3103 : * NB: It's also possible that the page that oldestMulti is on has already
3104 : * been truncated away, and we crashed before updating oldestMulti.
3105 : */
3106 0 : trunc.earliestExistingPage = -1;
3107 0 : SlruScanDirectory(MultiXactOffsetCtl, SlruScanDirCbFindEarliest, &trunc);
3108 0 : earliest = trunc.earliestExistingPage * MULTIXACT_OFFSETS_PER_PAGE;
3109 0 : if (earliest < FirstMultiXactId)
3110 0 : earliest = FirstMultiXactId;
3111 :
3112 : /* If there's nothing to remove, we can bail out early. */
3113 0 : if (MultiXactIdPrecedes(oldestMulti, earliest))
3114 : {
3115 0 : LWLockRelease(MultiXactTruncationLock);
3116 0 : return;
3117 : }
3118 :
3119 : /*
3120 : * First, compute the safe truncation point for MultiXactMember. This is
3121 : * the starting offset of the oldest multixact.
3122 : *
3123 : * Hopefully, find_multixact_start will always work here, because we've
3124 : * already checked that it doesn't precede the earliest MultiXact on disk.
3125 : * But if it fails, don't truncate anything, and log a message.
3126 : */
3127 0 : if (oldestMulti == nextMulti)
3128 : {
3129 : /* there are NO MultiXacts */
3130 0 : oldestOffset = nextOffset;
3131 : }
3132 0 : else if (!find_multixact_start(oldestMulti, &oldestOffset))
3133 : {
3134 0 : ereport(LOG,
3135 : (errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
3136 : oldestMulti, earliest)));
3137 0 : LWLockRelease(MultiXactTruncationLock);
3138 0 : return;
3139 : }
3140 :
3141 : /*
3142 : * Secondly compute up to where to truncate. Lookup the corresponding
3143 : * member offset for newOldestMulti for that.
3144 : */
3145 0 : if (newOldestMulti == nextMulti)
3146 : {
3147 : /* there are NO MultiXacts */
3148 0 : newOldestOffset = nextOffset;
3149 : }
3150 0 : else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
3151 : {
3152 0 : ereport(LOG,
3153 : (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
3154 : newOldestMulti)));
3155 0 : LWLockRelease(MultiXactTruncationLock);
3156 0 : return;
3157 : }
3158 :
3159 0 : elog(DEBUG1, "performing multixact truncation: "
3160 : "offsets [%u, %u), offsets segments [%" PRIx64 ", %" PRIx64 "), "
3161 : "members [%u, %u), members segments [%" PRIx64 ", %" PRIx64 ")",
3162 : oldestMulti, newOldestMulti,
3163 : MultiXactIdToOffsetSegment(oldestMulti),
3164 : MultiXactIdToOffsetSegment(newOldestMulti),
3165 : oldestOffset, newOldestOffset,
3166 : MXOffsetToMemberSegment(oldestOffset),
3167 : MXOffsetToMemberSegment(newOldestOffset));
3168 :
3169 : /*
3170 : * Do truncation, and the WAL logging of the truncation, in a critical
3171 : * section. That way offsets/members cannot get out of sync anymore, i.e.
3172 : * once consistent the newOldestMulti will always exist in members, even
3173 : * if we crashed in the wrong moment.
3174 : */
3175 0 : START_CRIT_SECTION();
3176 :
3177 : /*
3178 : * Prevent checkpoints from being scheduled concurrently. This is critical
3179 : * because otherwise a truncation record might not be replayed after a
3180 : * crash/basebackup, even though the state of the data directory would
3181 : * require it.
3182 : */
3183 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
3184 0 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
3185 :
3186 : /* WAL log truncation */
3187 0 : WriteMTruncateXlogRec(newOldestMultiDB,
3188 : oldestMulti, newOldestMulti,
3189 : oldestOffset, newOldestOffset);
3190 :
3191 : /*
3192 : * Update in-memory limits before performing the truncation, while inside
3193 : * the critical section: Have to do it before truncation, to prevent
3194 : * concurrent lookups of those values. Has to be inside the critical
3195 : * section as otherwise a future call to this function would error out,
3196 : * while looking up the oldest member in offsets, if our caller crashes
3197 : * before updating the limits.
3198 : */
3199 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
3200 0 : MultiXactState->oldestMultiXactId = newOldestMulti;
3201 0 : MultiXactState->oldestMultiXactDB = newOldestMultiDB;
3202 0 : LWLockRelease(MultiXactGenLock);
3203 :
3204 : /* First truncate members */
3205 0 : PerformMembersTruncation(oldestOffset, newOldestOffset);
3206 :
3207 : /* Then offsets */
3208 0 : PerformOffsetsTruncation(oldestMulti, newOldestMulti);
3209 :
3210 0 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
3211 :
3212 0 : END_CRIT_SECTION();
3213 0 : LWLockRelease(MultiXactTruncationLock);
3214 : }
3215 :
3216 : /*
3217 : * Decide whether a MultiXactOffset page number is "older" for truncation
3218 : * purposes. Analogous to CLOGPagePrecedes().
3219 : *
3220 : * Offsetting the values is optional, because MultiXactIdPrecedes() has
3221 : * translational symmetry.
3222 : */
3223 : static bool
3224 0 : MultiXactOffsetPagePrecedes(int64 page1, int64 page2)
3225 : {
3226 : MultiXactId multi1;
3227 : MultiXactId multi2;
3228 :
3229 0 : multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
3230 0 : multi1 += FirstMultiXactId + 1;
3231 0 : multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
3232 0 : multi2 += FirstMultiXactId + 1;
3233 :
3234 0 : return (MultiXactIdPrecedes(multi1, multi2) &&
3235 0 : MultiXactIdPrecedes(multi1,
3236 : multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
3237 : }
3238 :
3239 : /*
3240 : * Decide whether a MultiXactMember page number is "older" for truncation
3241 : * purposes. There is no "invalid offset number" so use the numbers verbatim.
3242 : */
3243 : static bool
3244 0 : MultiXactMemberPagePrecedes(int64 page1, int64 page2)
3245 : {
3246 : MultiXactOffset offset1;
3247 : MultiXactOffset offset2;
3248 :
3249 0 : offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
3250 0 : offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
3251 :
3252 0 : return (MultiXactOffsetPrecedes(offset1, offset2) &&
3253 0 : MultiXactOffsetPrecedes(offset1,
3254 : offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
3255 : }
3256 :
3257 : /*
3258 : * Decide which of two MultiXactIds is earlier.
3259 : *
3260 : * XXX do we need to do something special for InvalidMultiXactId?
3261 : * (Doesn't look like it.)
3262 : */
3263 : bool
3264 2587534 : MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
3265 : {
3266 2587534 : int32 diff = (int32) (multi1 - multi2);
3267 :
3268 2587534 : return (diff < 0);
3269 : }
3270 :
3271 : /*
3272 : * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
3273 : *
3274 : * XXX do we need to do something special for InvalidMultiXactId?
3275 : * (Doesn't look like it.)
3276 : */
3277 : bool
3278 13036 : MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
3279 : {
3280 13036 : int32 diff = (int32) (multi1 - multi2);
3281 :
3282 13036 : return (diff <= 0);
3283 : }
3284 :
3285 :
3286 : /*
3287 : * Decide which of two offsets is earlier.
3288 : */
3289 : static bool
3290 1356 : MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
3291 : {
3292 1356 : int32 diff = (int32) (offset1 - offset2);
3293 :
3294 1356 : return (diff < 0);
3295 : }
3296 :
3297 : /*
3298 : * Write a TRUNCATE xlog record
3299 : *
3300 : * We must flush the xlog record to disk before returning --- see notes in
3301 : * TruncateCLOG().
3302 : */
3303 : static void
3304 0 : WriteMTruncateXlogRec(Oid oldestMultiDB,
3305 : MultiXactId startTruncOff, MultiXactId endTruncOff,
3306 : MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
3307 : {
3308 : XLogRecPtr recptr;
3309 : xl_multixact_truncate xlrec;
3310 :
3311 0 : xlrec.oldestMultiDB = oldestMultiDB;
3312 :
3313 0 : xlrec.startTruncOff = startTruncOff;
3314 0 : xlrec.endTruncOff = endTruncOff;
3315 :
3316 0 : xlrec.startTruncMemb = startTruncMemb;
3317 0 : xlrec.endTruncMemb = endTruncMemb;
3318 :
3319 0 : XLogBeginInsert();
3320 0 : XLogRegisterData(&xlrec, SizeOfMultiXactTruncate);
3321 0 : recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
3322 0 : XLogFlush(recptr);
3323 0 : }
3324 :
3325 : /*
3326 : * MULTIXACT resource manager's routines
3327 : */
3328 : void
3329 8 : multixact_redo(XLogReaderState *record)
3330 : {
3331 8 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3332 :
3333 : /* Backup blocks are not used in multixact records */
3334 : Assert(!XLogRecHasAnyBlockRefs(record));
3335 :
3336 8 : if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
3337 : {
3338 : int64 pageno;
3339 :
3340 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3341 2 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, pageno);
3342 : }
3343 6 : else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
3344 : {
3345 : int64 pageno;
3346 :
3347 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3348 2 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, pageno);
3349 : }
3350 4 : else if (info == XLOG_MULTIXACT_CREATE_ID)
3351 : {
3352 4 : xl_multixact_create *xlrec =
3353 4 : (xl_multixact_create *) XLogRecGetData(record);
3354 : TransactionId max_xid;
3355 : int i;
3356 :
3357 : /* Store the data back into the SLRU files */
3358 4 : RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
3359 4 : xlrec->members);
3360 :
3361 : /* Make sure nextMXact/nextOffset are beyond what this record has */
3362 4 : MultiXactAdvanceNextMXact(xlrec->mid + 1,
3363 4 : xlrec->moff + xlrec->nmembers);
3364 :
3365 : /*
3366 : * Make sure nextXid is beyond any XID mentioned in the record. This
3367 : * should be unnecessary, since any XID found here ought to have other
3368 : * evidence in the XLOG, but let's be safe.
3369 : */
3370 4 : max_xid = XLogRecGetXid(record);
3371 12 : for (i = 0; i < xlrec->nmembers; i++)
3372 : {
3373 8 : if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
3374 0 : max_xid = xlrec->members[i].xid;
3375 : }
3376 :
3377 4 : AdvanceNextFullTransactionIdPastXid(max_xid);
3378 : }
3379 0 : else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
3380 : {
3381 : xl_multixact_truncate xlrec;
3382 : int64 pageno;
3383 :
3384 0 : memcpy(&xlrec, XLogRecGetData(record),
3385 : SizeOfMultiXactTruncate);
3386 :
3387 0 : elog(DEBUG1, "replaying multixact truncation: "
3388 : "offsets [%u, %u), offsets segments [%" PRIx64 ", %" PRIx64 "), "
3389 : "members [%u, %u), members segments [%" PRIx64 ", %" PRIx64 ")",
3390 : xlrec.startTruncOff, xlrec.endTruncOff,
3391 : MultiXactIdToOffsetSegment(xlrec.startTruncOff),
3392 : MultiXactIdToOffsetSegment(xlrec.endTruncOff),
3393 : xlrec.startTruncMemb, xlrec.endTruncMemb,
3394 : MXOffsetToMemberSegment(xlrec.startTruncMemb),
3395 : MXOffsetToMemberSegment(xlrec.endTruncMemb));
3396 :
3397 : /* should not be required, but more than cheap enough */
3398 0 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3399 :
3400 : /*
3401 : * Advance the horizon values, so they're current at the end of
3402 : * recovery.
3403 : */
3404 0 : SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
3405 :
3406 0 : PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
3407 :
3408 : /*
3409 : * During XLOG replay, latest_page_number isn't necessarily set up
3410 : * yet; insert a suitable value to bypass the sanity test in
3411 : * SimpleLruTruncate.
3412 : */
3413 0 : pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
3414 0 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
3415 : pageno);
3416 0 : PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
3417 :
3418 0 : LWLockRelease(MultiXactTruncationLock);
3419 : }
3420 : else
3421 0 : elog(PANIC, "multixact_redo: unknown op code %u", info);
3422 8 : }
3423 :
3424 : /*
3425 : * Entrypoint for sync.c to sync offsets files.
3426 : */
3427 : int
3428 0 : multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
3429 : {
3430 0 : return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
3431 : }
3432 :
3433 : /*
3434 : * Entrypoint for sync.c to sync members files.
3435 : */
3436 : int
3437 0 : multixactmemberssyncfiletag(const FileTag *ftag, char *path)
3438 : {
3439 0 : return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
3440 : }
|
