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/condition_variable.h"
83 : #include "storage/pmsignal.h"
84 : #include "storage/proc.h"
85 : #include "storage/procarray.h"
86 : #include "utils/guc_hooks.h"
87 : #include "utils/injection_point.h"
88 : #include "utils/lsyscache.h"
89 : #include "utils/memutils.h"
90 :
91 :
92 : /*
93 : * Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
94 : * used everywhere else in Postgres.
95 : *
96 : * Note: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
97 : * MultiXact page numbering also wraps around at
98 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
99 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need
100 : * take no explicit notice of that fact in this module, except when comparing
101 : * segment and page numbers in TruncateMultiXact (see
102 : * MultiXactOffsetPagePrecedes).
103 : */
104 :
105 : /* We need four bytes per offset */
106 : #define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
107 :
108 : static inline int64
109 4928 : MultiXactIdToOffsetPage(MultiXactId multi)
110 : {
111 4928 : return multi / MULTIXACT_OFFSETS_PER_PAGE;
112 : }
113 :
114 : static inline int
115 3486 : MultiXactIdToOffsetEntry(MultiXactId multi)
116 : {
117 3486 : return multi % MULTIXACT_OFFSETS_PER_PAGE;
118 : }
119 :
120 : static inline int64
121 0 : MultiXactIdToOffsetSegment(MultiXactId multi)
122 : {
123 0 : return MultiXactIdToOffsetPage(multi) / SLRU_PAGES_PER_SEGMENT;
124 : }
125 :
126 : /*
127 : * The situation for members is a bit more complex: we store one byte of
128 : * additional flag bits for each TransactionId. To do this without getting
129 : * into alignment issues, we store four bytes of flags, and then the
130 : * corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
131 : * are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
132 : * per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
133 : * performance) trumps space efficiency here.
134 : *
135 : * Note that the "offset" macros work with byte offset, not array indexes, so
136 : * arithmetic must be done using "char *" pointers.
137 : */
138 : /* We need eight bits per xact, so one xact fits in a byte */
139 : #define MXACT_MEMBER_BITS_PER_XACT 8
140 : #define MXACT_MEMBER_FLAGS_PER_BYTE 1
141 : #define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
142 :
143 : /* how many full bytes of flags are there in a group? */
144 : #define MULTIXACT_FLAGBYTES_PER_GROUP 4
145 : #define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
146 : (MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
147 : /* size in bytes of a complete group */
148 : #define MULTIXACT_MEMBERGROUP_SIZE \
149 : (sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
150 : #define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
151 : #define MULTIXACT_MEMBERS_PER_PAGE \
152 : (MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
153 :
154 : /*
155 : * Because the number of items per page is not a divisor of the last item
156 : * number (member 0xFFFFFFFF), the last segment does not use the maximum number
157 : * of pages, and moreover the last used page therein does not use the same
158 : * number of items as previous pages. (Another way to say it is that the
159 : * 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
160 : * has some empty space after that item.)
161 : *
162 : * This constant is the number of members in the last page of the last segment.
163 : */
164 : #define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
165 : ((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
166 :
167 : /* page in which a member is to be found */
168 : static inline int64
169 6190 : MXOffsetToMemberPage(MultiXactOffset offset)
170 : {
171 6190 : return offset / MULTIXACT_MEMBERS_PER_PAGE;
172 : }
173 :
174 : static inline int64
175 0 : MXOffsetToMemberSegment(MultiXactOffset offset)
176 : {
177 0 : return MXOffsetToMemberPage(offset) / SLRU_PAGES_PER_SEGMENT;
178 : }
179 :
180 : /* Location (byte offset within page) of flag word for a given member */
181 : static inline int
182 7360 : MXOffsetToFlagsOffset(MultiXactOffset offset)
183 : {
184 7360 : MultiXactOffset group = offset / MULTIXACT_MEMBERS_PER_MEMBERGROUP;
185 7360 : int grouponpg = group % MULTIXACT_MEMBERGROUPS_PER_PAGE;
186 7360 : int byteoff = grouponpg * MULTIXACT_MEMBERGROUP_SIZE;
187 :
188 7360 : return byteoff;
189 : }
190 :
191 : static inline int
192 3064 : MXOffsetToFlagsBitShift(MultiXactOffset offset)
193 : {
194 3064 : int member_in_group = offset % MULTIXACT_MEMBERS_PER_MEMBERGROUP;
195 3064 : int bshift = member_in_group * MXACT_MEMBER_BITS_PER_XACT;
196 :
197 3064 : return bshift;
198 : }
199 :
200 : /* Location (byte offset within page) of TransactionId of given member */
201 : static inline int
202 2518 : MXOffsetToMemberOffset(MultiXactOffset offset)
203 : {
204 2518 : int member_in_group = offset % MULTIXACT_MEMBERS_PER_MEMBERGROUP;
205 :
206 2518 : return MXOffsetToFlagsOffset(offset) +
207 2518 : MULTIXACT_FLAGBYTES_PER_GROUP +
208 : member_in_group * sizeof(TransactionId);
209 : }
210 :
211 : /* Multixact members wraparound thresholds. */
212 : #define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
213 : #define MULTIXACT_MEMBER_DANGER_THRESHOLD \
214 : (MaxMultiXactOffset - MaxMultiXactOffset / 4)
215 :
216 : static inline MultiXactId
217 0 : PreviousMultiXactId(MultiXactId multi)
218 : {
219 0 : return multi == FirstMultiXactId ? MaxMultiXactId : multi - 1;
220 : }
221 :
222 : /*
223 : * Links to shared-memory data structures for MultiXact control
224 : */
225 : static SlruCtlData MultiXactOffsetCtlData;
226 : static SlruCtlData MultiXactMemberCtlData;
227 :
228 : #define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
229 : #define MultiXactMemberCtl (&MultiXactMemberCtlData)
230 :
231 : /*
232 : * MultiXact state shared across all backends. All this state is protected
233 : * by MultiXactGenLock. (We also use SLRU bank's lock of MultiXactOffset and
234 : * MultiXactMember to guard accesses to the two sets of SLRU buffers. For
235 : * concurrency's sake, we avoid holding more than one of these locks at a
236 : * time.)
237 : */
238 : typedef struct MultiXactStateData
239 : {
240 : /* next-to-be-assigned MultiXactId */
241 : MultiXactId nextMXact;
242 :
243 : /* next-to-be-assigned offset */
244 : MultiXactOffset nextOffset;
245 :
246 : /* Have we completed multixact startup? */
247 : bool finishedStartup;
248 :
249 : /*
250 : * Oldest multixact that is still potentially referenced by a relation.
251 : * Anything older than this should not be consulted. These values are
252 : * updated by vacuum.
253 : */
254 : MultiXactId oldestMultiXactId;
255 : Oid oldestMultiXactDB;
256 :
257 : /*
258 : * Oldest multixact offset that is potentially referenced by a multixact
259 : * referenced by a relation. We don't always know this value, so there's
260 : * a flag here to indicate whether or not we currently do.
261 : */
262 : MultiXactOffset oldestOffset;
263 : bool oldestOffsetKnown;
264 :
265 : /* support for anti-wraparound measures */
266 : MultiXactId multiVacLimit;
267 : MultiXactId multiWarnLimit;
268 : MultiXactId multiStopLimit;
269 : MultiXactId multiWrapLimit;
270 :
271 : /* support for members anti-wraparound measures */
272 : MultiXactOffset offsetStopLimit; /* known if oldestOffsetKnown */
273 :
274 : /*
275 : * This is used to sleep until a multixact offset is written when we want
276 : * to create the next one.
277 : */
278 : ConditionVariable nextoff_cv;
279 :
280 : /*
281 : * Per-backend data starts here. We have two arrays stored in the area
282 : * immediately following the MultiXactStateData struct. Each is indexed by
283 : * ProcNumber.
284 : *
285 : * In both arrays, there's a slot for all normal backends
286 : * (0..MaxBackends-1) followed by a slot for max_prepared_xacts prepared
287 : * transactions.
288 : *
289 : * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
290 : * transaction(s) could possibly be a member of, or InvalidMultiXactId
291 : * when the backend has no live transaction that could possibly be a
292 : * member of a MultiXact. Each backend sets its entry to the current
293 : * nextMXact counter just before first acquiring a shared lock in a given
294 : * transaction, and clears it at transaction end. (This works because only
295 : * during or after acquiring a shared lock could an XID possibly become a
296 : * member of a MultiXact, and that MultiXact would have to be created
297 : * during or after the lock acquisition.)
298 : *
299 : * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
300 : * current transaction(s) think is potentially live, or InvalidMultiXactId
301 : * when not in a transaction or not in a transaction that's paid any
302 : * attention to MultiXacts yet. This is computed when first needed in a
303 : * given transaction, and cleared at transaction end. We can compute it
304 : * as the minimum of the valid OldestMemberMXactId[] entries at the time
305 : * we compute it (using nextMXact if none are valid). Each backend is
306 : * required not to attempt to access any SLRU data for MultiXactIds older
307 : * than its own OldestVisibleMXactId[] setting; this is necessary because
308 : * the relevant SLRU data can be concurrently truncated away.
309 : *
310 : * The oldest valid value among all of the OldestMemberMXactId[] and
311 : * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
312 : * possible value still having any live member transaction -- OldestMxact.
313 : * Any value older than that is typically removed from tuple headers, or
314 : * "frozen" via being replaced with a new xmax. VACUUM can sometimes even
315 : * remove an individual MultiXact xmax whose value is >= its OldestMxact
316 : * cutoff, though typically only when no individual member XID is still
317 : * running. See FreezeMultiXactId for full details.
318 : *
319 : * Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
320 : * or the oldest extant Multi remaining in the table is used as the new
321 : * pg_class.relminmxid value (whichever is earlier). The minimum of all
322 : * relminmxid values in each database is stored in pg_database.datminmxid.
323 : * In turn, the minimum of all of those values is stored in pg_control.
324 : * This is used as the truncation point for pg_multixact when unneeded
325 : * segments get removed by vac_truncate_clog() during vacuuming.
326 : */
327 : MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
328 : } MultiXactStateData;
329 :
330 : /*
331 : * Size of OldestMemberMXactId and OldestVisibleMXactId arrays.
332 : */
333 : #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
334 :
335 : /* Pointers to the state data in shared memory */
336 : static MultiXactStateData *MultiXactState;
337 : static MultiXactId *OldestMemberMXactId;
338 : static MultiXactId *OldestVisibleMXactId;
339 :
340 :
341 : /*
342 : * Definitions for the backend-local MultiXactId cache.
343 : *
344 : * We use this cache to store known MultiXacts, so we don't need to go to
345 : * SLRU areas every time.
346 : *
347 : * The cache lasts for the duration of a single transaction, the rationale
348 : * for this being that most entries will contain our own TransactionId and
349 : * so they will be uninteresting by the time our next transaction starts.
350 : * (XXX not clear that this is correct --- other members of the MultiXact
351 : * could hang around longer than we did. However, it's not clear what a
352 : * better policy for flushing old cache entries would be.) FIXME actually
353 : * this is plain wrong now that multixact's may contain update Xids.
354 : *
355 : * We allocate the cache entries in a memory context that is deleted at
356 : * transaction end, so we don't need to do retail freeing of entries.
357 : */
358 : typedef struct mXactCacheEnt
359 : {
360 : MultiXactId multi;
361 : int nmembers;
362 : dlist_node node;
363 : MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
364 : } mXactCacheEnt;
365 :
366 : #define MAX_CACHE_ENTRIES 256
367 : static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
368 : static MemoryContext MXactContext = NULL;
369 :
370 : #ifdef MULTIXACT_DEBUG
371 : #define debug_elog2(a,b) elog(a,b)
372 : #define debug_elog3(a,b,c) elog(a,b,c)
373 : #define debug_elog4(a,b,c,d) elog(a,b,c,d)
374 : #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
375 : #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
376 : #else
377 : #define debug_elog2(a,b)
378 : #define debug_elog3(a,b,c)
379 : #define debug_elog4(a,b,c,d)
380 : #define debug_elog5(a,b,c,d,e)
381 : #define debug_elog6(a,b,c,d,e,f)
382 : #endif
383 :
384 : /* internal MultiXactId management */
385 : static void MultiXactIdSetOldestVisible(void);
386 : static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
387 : int nmembers, MultiXactMember *members);
388 : static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
389 :
390 : /* MultiXact cache management */
391 : static int mxactMemberComparator(const void *arg1, const void *arg2);
392 : static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
393 : static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
394 : static void mXactCachePut(MultiXactId multi, int nmembers,
395 : MultiXactMember *members);
396 :
397 : /* management of SLRU infrastructure */
398 : static bool MultiXactOffsetPagePrecedes(int64 page1, int64 page2);
399 : static bool MultiXactMemberPagePrecedes(int64 page1, int64 page2);
400 : static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
401 : MultiXactOffset offset2);
402 : static void ExtendMultiXactOffset(MultiXactId multi);
403 : static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
404 : static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
405 : MultiXactOffset start, uint32 distance);
406 : static bool SetOffsetVacuumLimit(bool is_startup);
407 : static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
408 : static void WriteMTruncateXlogRec(Oid oldestMultiDB,
409 : MultiXactId startTruncOff,
410 : MultiXactId endTruncOff,
411 : MultiXactOffset startTruncMemb,
412 : MultiXactOffset endTruncMemb);
413 :
414 :
415 : /*
416 : * MultiXactIdCreate
417 : * Construct a MultiXactId representing two TransactionIds.
418 : *
419 : * The two XIDs must be different, or be requesting different statuses.
420 : *
421 : * NB - we don't worry about our local MultiXactId cache here, because that
422 : * is handled by the lower-level routines.
423 : */
424 : MultiXactId
425 2074 : MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
426 : TransactionId xid2, MultiXactStatus status2)
427 : {
428 : MultiXactId newMulti;
429 : MultiXactMember members[2];
430 :
431 : Assert(TransactionIdIsValid(xid1));
432 : Assert(TransactionIdIsValid(xid2));
433 :
434 : Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
435 :
436 : /* MultiXactIdSetOldestMember() must have been called already. */
437 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
438 :
439 : /*
440 : * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
441 : * are still running. In typical usage, xid2 will be our own XID and the
442 : * caller just did a check on xid1, so it'd be wasted effort.
443 : */
444 :
445 2074 : members[0].xid = xid1;
446 2074 : members[0].status = status1;
447 2074 : members[1].xid = xid2;
448 2074 : members[1].status = status2;
449 :
450 2074 : newMulti = MultiXactIdCreateFromMembers(2, members);
451 :
452 : debug_elog3(DEBUG2, "Create: %s",
453 : mxid_to_string(newMulti, 2, members));
454 :
455 2074 : return newMulti;
456 : }
457 :
458 : /*
459 : * MultiXactIdExpand
460 : * Add a TransactionId to a pre-existing MultiXactId.
461 : *
462 : * If the TransactionId is already a member of the passed MultiXactId with the
463 : * same status, just return it as-is.
464 : *
465 : * Note that we do NOT actually modify the membership of a pre-existing
466 : * MultiXactId; instead we create a new one. This is necessary to avoid
467 : * a race condition against code trying to wait for one MultiXactId to finish;
468 : * see notes in heapam.c.
469 : *
470 : * NB - we don't worry about our local MultiXactId cache here, because that
471 : * is handled by the lower-level routines.
472 : *
473 : * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
474 : * one upgraded by pg_upgrade from a cluster older than this feature) are not
475 : * passed in.
476 : */
477 : MultiXactId
478 192 : MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
479 : {
480 : MultiXactId newMulti;
481 : MultiXactMember *members;
482 : MultiXactMember *newMembers;
483 : int nmembers;
484 : int i;
485 : int j;
486 :
487 : Assert(MultiXactIdIsValid(multi));
488 : Assert(TransactionIdIsValid(xid));
489 :
490 : /* MultiXactIdSetOldestMember() must have been called already. */
491 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
492 :
493 : debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
494 : multi, xid, mxstatus_to_string(status));
495 :
496 : /*
497 : * Note: we don't allow for old multis here. The reason is that the only
498 : * caller of this function does a check that the multixact is no longer
499 : * running.
500 : */
501 192 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
502 :
503 192 : if (nmembers < 0)
504 : {
505 : MultiXactMember member;
506 :
507 : /*
508 : * The MultiXactId is obsolete. This can only happen if all the
509 : * MultiXactId members stop running between the caller checking and
510 : * passing it to us. It would be better to return that fact to the
511 : * caller, but it would complicate the API and it's unlikely to happen
512 : * too often, so just deal with it by creating a singleton MultiXact.
513 : */
514 0 : member.xid = xid;
515 0 : member.status = status;
516 0 : newMulti = MultiXactIdCreateFromMembers(1, &member);
517 :
518 : debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
519 : multi, newMulti);
520 0 : return newMulti;
521 : }
522 :
523 : /*
524 : * If the TransactionId is already a member of the MultiXactId with the
525 : * same status, just return the existing MultiXactId.
526 : */
527 596 : for (i = 0; i < nmembers; i++)
528 : {
529 404 : if (TransactionIdEquals(members[i].xid, xid) &&
530 108 : (members[i].status == status))
531 : {
532 : debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
533 : xid, multi);
534 0 : pfree(members);
535 0 : return multi;
536 : }
537 : }
538 :
539 : /*
540 : * Determine which of the members of the MultiXactId are still of
541 : * interest. This is any running transaction, and also any transaction
542 : * that grabbed something stronger than just a lock and was committed. (An
543 : * update that aborted is of no interest here; and having more than one
544 : * update Xid in a multixact would cause errors elsewhere.)
545 : *
546 : * Removing dead members is not just an optimization: freezing of tuples
547 : * whose Xmax are multis depends on this behavior.
548 : *
549 : * Note we have the same race condition here as above: j could be 0 at the
550 : * end of the loop.
551 : */
552 : newMembers = (MultiXactMember *)
553 192 : palloc(sizeof(MultiXactMember) * (nmembers + 1));
554 :
555 596 : for (i = 0, j = 0; i < nmembers; i++)
556 : {
557 404 : if (TransactionIdIsInProgress(members[i].xid) ||
558 88 : (ISUPDATE_from_mxstatus(members[i].status) &&
559 12 : TransactionIdDidCommit(members[i].xid)))
560 : {
561 328 : newMembers[j].xid = members[i].xid;
562 328 : newMembers[j++].status = members[i].status;
563 : }
564 : }
565 :
566 192 : newMembers[j].xid = xid;
567 192 : newMembers[j++].status = status;
568 192 : newMulti = MultiXactIdCreateFromMembers(j, newMembers);
569 :
570 192 : pfree(members);
571 192 : pfree(newMembers);
572 :
573 : debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
574 :
575 192 : return newMulti;
576 : }
577 :
578 : /*
579 : * MultiXactIdIsRunning
580 : * Returns whether a MultiXactId is "running".
581 : *
582 : * We return true if at least one member of the given MultiXactId is still
583 : * running. Note that a "false" result is certain not to change,
584 : * because it is not legal to add members to an existing MultiXactId.
585 : *
586 : * Caller is expected to have verified that the multixact does not come from
587 : * a pg_upgraded share-locked tuple.
588 : */
589 : bool
590 1982 : MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
591 : {
592 : MultiXactMember *members;
593 : int nmembers;
594 : int i;
595 :
596 : debug_elog3(DEBUG2, "IsRunning %u?", multi);
597 :
598 : /*
599 : * "false" here means we assume our callers have checked that the given
600 : * multi cannot possibly come from a pg_upgraded database.
601 : */
602 1982 : nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
603 :
604 1982 : if (nmembers <= 0)
605 : {
606 : debug_elog2(DEBUG2, "IsRunning: no members");
607 1336 : return false;
608 : }
609 :
610 : /*
611 : * Checking for myself is cheap compared to looking in shared memory;
612 : * return true if any live subtransaction of the current top-level
613 : * transaction is a member.
614 : *
615 : * This is not needed for correctness, it's just a fast path.
616 : */
617 1502 : for (i = 0; i < nmembers; i++)
618 : {
619 1168 : if (TransactionIdIsCurrentTransactionId(members[i].xid))
620 : {
621 : debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
622 312 : pfree(members);
623 312 : return true;
624 : }
625 : }
626 :
627 : /*
628 : * This could be made faster by having another entry point in procarray.c,
629 : * walking the PGPROC array only once for all the members. But in most
630 : * cases nmembers should be small enough that it doesn't much matter.
631 : */
632 608 : for (i = 0; i < nmembers; i++)
633 : {
634 500 : if (TransactionIdIsInProgress(members[i].xid))
635 : {
636 : debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
637 : i, members[i].xid);
638 226 : pfree(members);
639 226 : return true;
640 : }
641 : }
642 :
643 108 : pfree(members);
644 :
645 : debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
646 :
647 108 : return false;
648 : }
649 :
650 : /*
651 : * MultiXactIdSetOldestMember
652 : * Save the oldest MultiXactId this transaction could be a member of.
653 : *
654 : * We set the OldestMemberMXactId for a given transaction the first time it's
655 : * going to do some operation that might require a MultiXactId (tuple lock,
656 : * update or delete). We need to do this even if we end up using a
657 : * TransactionId instead of a MultiXactId, because there is a chance that
658 : * another transaction would add our XID to a MultiXactId.
659 : *
660 : * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
661 : * be called just before doing any such possibly-MultiXactId-able operation.
662 : */
663 : void
664 3753964 : MultiXactIdSetOldestMember(void)
665 : {
666 3753964 : if (!MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]))
667 : {
668 : MultiXactId nextMXact;
669 :
670 : /*
671 : * You might think we don't need to acquire a lock here, since
672 : * fetching and storing of TransactionIds is probably atomic, but in
673 : * fact we do: suppose we pick up nextMXact and then lose the CPU for
674 : * a long time. Someone else could advance nextMXact, and then
675 : * another someone else could compute an OldestVisibleMXactId that
676 : * would be after the value we are going to store when we get control
677 : * back. Which would be wrong.
678 : *
679 : * Note that a shared lock is sufficient, because it's enough to stop
680 : * someone from advancing nextMXact; and nobody else could be trying
681 : * to write to our OldestMember entry, only reading (and we assume
682 : * storing it is atomic.)
683 : */
684 139462 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
685 :
686 : /*
687 : * We have to beware of the possibility that nextMXact is in the
688 : * wrapped-around state. We don't fix the counter itself here, but we
689 : * must be sure to store a valid value in our array entry.
690 : */
691 139462 : nextMXact = MultiXactState->nextMXact;
692 139462 : if (nextMXact < FirstMultiXactId)
693 0 : nextMXact = FirstMultiXactId;
694 :
695 139462 : OldestMemberMXactId[MyProcNumber] = nextMXact;
696 :
697 139462 : LWLockRelease(MultiXactGenLock);
698 :
699 : debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
700 : MyProcNumber, nextMXact);
701 : }
702 3753964 : }
703 :
704 : /*
705 : * MultiXactIdSetOldestVisible
706 : * Save the oldest MultiXactId this transaction considers possibly live.
707 : *
708 : * We set the OldestVisibleMXactId for a given transaction the first time
709 : * it's going to inspect any MultiXactId. Once we have set this, we are
710 : * guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
711 : * won't be truncated away.
712 : *
713 : * The value to set is the oldest of nextMXact and all the valid per-backend
714 : * OldestMemberMXactId[] entries. Because of the locking we do, we can be
715 : * certain that no subsequent call to MultiXactIdSetOldestMember can set
716 : * an OldestMemberMXactId[] entry older than what we compute here. Therefore
717 : * there is no live transaction, now or later, that can be a member of any
718 : * MultiXactId older than the OldestVisibleMXactId we compute here.
719 : */
720 : static void
721 1818 : MultiXactIdSetOldestVisible(void)
722 : {
723 1818 : if (!MultiXactIdIsValid(OldestVisibleMXactId[MyProcNumber]))
724 : {
725 : MultiXactId oldestMXact;
726 : int i;
727 :
728 468 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
729 :
730 : /*
731 : * We have to beware of the possibility that nextMXact is in the
732 : * wrapped-around state. We don't fix the counter itself here, but we
733 : * must be sure to store a valid value in our array entry.
734 : */
735 468 : oldestMXact = MultiXactState->nextMXact;
736 468 : if (oldestMXact < FirstMultiXactId)
737 0 : oldestMXact = FirstMultiXactId;
738 :
739 63180 : for (i = 0; i < MaxOldestSlot; i++)
740 : {
741 62712 : MultiXactId thisoldest = OldestMemberMXactId[i];
742 :
743 63218 : if (MultiXactIdIsValid(thisoldest) &&
744 506 : MultiXactIdPrecedes(thisoldest, oldestMXact))
745 248 : oldestMXact = thisoldest;
746 : }
747 :
748 468 : OldestVisibleMXactId[MyProcNumber] = oldestMXact;
749 :
750 468 : LWLockRelease(MultiXactGenLock);
751 :
752 : debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
753 : MyProcNumber, oldestMXact);
754 : }
755 1818 : }
756 :
757 : /*
758 : * ReadNextMultiXactId
759 : * Return the next MultiXactId to be assigned, but don't allocate it
760 : */
761 : MultiXactId
762 417502 : ReadNextMultiXactId(void)
763 : {
764 : MultiXactId mxid;
765 :
766 : /* XXX we could presumably do this without a lock. */
767 417502 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
768 417502 : mxid = MultiXactState->nextMXact;
769 417502 : LWLockRelease(MultiXactGenLock);
770 :
771 417502 : if (mxid < FirstMultiXactId)
772 0 : mxid = FirstMultiXactId;
773 :
774 417502 : return mxid;
775 : }
776 :
777 : /*
778 : * ReadMultiXactIdRange
779 : * Get the range of IDs that may still be referenced by a relation.
780 : */
781 : void
782 2954 : ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
783 : {
784 2954 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
785 2954 : *oldest = MultiXactState->oldestMultiXactId;
786 2954 : *next = MultiXactState->nextMXact;
787 2954 : LWLockRelease(MultiXactGenLock);
788 :
789 2954 : if (*oldest < FirstMultiXactId)
790 0 : *oldest = FirstMultiXactId;
791 2954 : if (*next < FirstMultiXactId)
792 0 : *next = FirstMultiXactId;
793 2954 : }
794 :
795 :
796 : /*
797 : * MultiXactIdCreateFromMembers
798 : * Make a new MultiXactId from the specified set of members
799 : *
800 : * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
801 : * given TransactionIds as members. Returns the newly created MultiXactId.
802 : *
803 : * NB: the passed members[] array will be sorted in-place.
804 : */
805 : MultiXactId
806 2268 : MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
807 : {
808 : MultiXactId multi;
809 : MultiXactOffset offset;
810 : xl_multixact_create xlrec;
811 :
812 : debug_elog3(DEBUG2, "Create: %s",
813 : mxid_to_string(InvalidMultiXactId, nmembers, members));
814 :
815 : /*
816 : * See if the same set of members already exists in our cache; if so, just
817 : * re-use that MultiXactId. (Note: it might seem that looking in our
818 : * cache is insufficient, and we ought to search disk to see if a
819 : * duplicate definition already exists. But since we only ever create
820 : * MultiXacts containing our own XID, in most cases any such MultiXacts
821 : * were in fact created by us, and so will be in our cache. There are
822 : * corner cases where someone else added us to a MultiXact without our
823 : * knowledge, but it's not worth checking for.)
824 : */
825 2268 : multi = mXactCacheGetBySet(nmembers, members);
826 2268 : if (MultiXactIdIsValid(multi))
827 : {
828 : debug_elog2(DEBUG2, "Create: in cache!");
829 1700 : return multi;
830 : }
831 :
832 : /* Verify that there is a single update Xid among the given members. */
833 : {
834 : int i;
835 568 : bool has_update = false;
836 :
837 1828 : for (i = 0; i < nmembers; i++)
838 : {
839 1260 : if (ISUPDATE_from_mxstatus(members[i].status))
840 : {
841 264 : if (has_update)
842 0 : elog(ERROR, "new multixact has more than one updating member: %s",
843 : mxid_to_string(InvalidMultiXactId, nmembers, members));
844 264 : has_update = true;
845 : }
846 : }
847 : }
848 :
849 : /* Load the injection point before entering the critical section */
850 568 : INJECTION_POINT_LOAD("multixact-create-from-members");
851 :
852 : /*
853 : * Assign the MXID and offsets range to use, and make sure there is space
854 : * in the OFFSETs and MEMBERs files. NB: this routine does
855 : * START_CRIT_SECTION().
856 : *
857 : * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
858 : * that we've called MultiXactIdSetOldestMember here. This is because
859 : * this routine is used in some places to create new MultiXactIds of which
860 : * the current backend is not a member, notably during freezing of multis
861 : * in vacuum. During vacuum, in particular, it would be unacceptable to
862 : * keep OldestMulti set, in case it runs for long.
863 : */
864 568 : multi = GetNewMultiXactId(nmembers, &offset);
865 :
866 568 : INJECTION_POINT_CACHED("multixact-create-from-members", NULL);
867 :
868 : /* Make an XLOG entry describing the new MXID. */
869 568 : xlrec.mid = multi;
870 568 : xlrec.moff = offset;
871 568 : xlrec.nmembers = nmembers;
872 :
873 : /*
874 : * XXX Note: there's a lot of padding space in MultiXactMember. We could
875 : * find a more compact representation of this Xlog record -- perhaps all
876 : * the status flags in one XLogRecData, then all the xids in another one?
877 : * Not clear that it's worth the trouble though.
878 : */
879 568 : XLogBeginInsert();
880 568 : XLogRegisterData(&xlrec, SizeOfMultiXactCreate);
881 568 : XLogRegisterData(members, nmembers * sizeof(MultiXactMember));
882 :
883 568 : (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
884 :
885 : /* Now enter the information into the OFFSETs and MEMBERs logs */
886 568 : RecordNewMultiXact(multi, offset, nmembers, members);
887 :
888 : /* Done with critical section */
889 568 : END_CRIT_SECTION();
890 :
891 : /* Store the new MultiXactId in the local cache, too */
892 568 : mXactCachePut(multi, nmembers, members);
893 :
894 : debug_elog2(DEBUG2, "Create: all done");
895 :
896 568 : return multi;
897 : }
898 :
899 : /*
900 : * RecordNewMultiXact
901 : * Write info about a new multixact into the offsets and members files
902 : *
903 : * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
904 : * use it.
905 : */
906 : static void
907 572 : RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
908 : int nmembers, MultiXactMember *members)
909 : {
910 : int64 pageno;
911 : int64 prev_pageno;
912 : int entryno;
913 : int slotno;
914 : MultiXactOffset *offptr;
915 : int i;
916 : LWLock *lock;
917 572 : LWLock *prevlock = NULL;
918 :
919 572 : pageno = MultiXactIdToOffsetPage(multi);
920 572 : entryno = MultiXactIdToOffsetEntry(multi);
921 :
922 572 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
923 572 : LWLockAcquire(lock, LW_EXCLUSIVE);
924 :
925 : /*
926 : * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
927 : * to complain about if there's any I/O error. This is kinda bogus, but
928 : * since the errors will always give the full pathname, it should be clear
929 : * enough that a MultiXactId is really involved. Perhaps someday we'll
930 : * take the trouble to generalize the slru.c error reporting code.
931 : */
932 572 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
933 572 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
934 572 : offptr += entryno;
935 :
936 572 : *offptr = offset;
937 :
938 572 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
939 :
940 : /* Release MultiXactOffset SLRU lock. */
941 572 : LWLockRelease(lock);
942 :
943 : /*
944 : * If anybody was waiting to know the offset of this multixact ID we just
945 : * wrote, they can read it now, so wake them up.
946 : */
947 572 : ConditionVariableBroadcast(&MultiXactState->nextoff_cv);
948 :
949 572 : prev_pageno = -1;
950 :
951 1840 : for (i = 0; i < nmembers; i++, offset++)
952 : {
953 : TransactionId *memberptr;
954 : uint32 *flagsptr;
955 : uint32 flagsval;
956 : int bshift;
957 : int flagsoff;
958 : int memberoff;
959 :
960 : Assert(members[i].status <= MultiXactStatusUpdate);
961 :
962 1268 : pageno = MXOffsetToMemberPage(offset);
963 1268 : memberoff = MXOffsetToMemberOffset(offset);
964 1268 : flagsoff = MXOffsetToFlagsOffset(offset);
965 1268 : bshift = MXOffsetToFlagsBitShift(offset);
966 :
967 1268 : if (pageno != prev_pageno)
968 : {
969 : /*
970 : * MultiXactMember SLRU page is changed so check if this new page
971 : * fall into the different SLRU bank then release the old bank's
972 : * lock and acquire lock on the new bank.
973 : */
974 572 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
975 572 : if (lock != prevlock)
976 : {
977 572 : if (prevlock != NULL)
978 0 : LWLockRelease(prevlock);
979 :
980 572 : LWLockAcquire(lock, LW_EXCLUSIVE);
981 572 : prevlock = lock;
982 : }
983 572 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
984 572 : prev_pageno = pageno;
985 : }
986 :
987 1268 : memberptr = (TransactionId *)
988 1268 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
989 :
990 1268 : *memberptr = members[i].xid;
991 :
992 1268 : flagsptr = (uint32 *)
993 1268 : (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
994 :
995 1268 : flagsval = *flagsptr;
996 1268 : flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
997 1268 : flagsval |= (members[i].status << bshift);
998 1268 : *flagsptr = flagsval;
999 :
1000 1268 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
1001 : }
1002 :
1003 572 : if (prevlock != NULL)
1004 572 : LWLockRelease(prevlock);
1005 572 : }
1006 :
1007 : /*
1008 : * GetNewMultiXactId
1009 : * Get the next MultiXactId.
1010 : *
1011 : * Also, reserve the needed amount of space in the "members" area. The
1012 : * starting offset of the reserved space is returned in *offset.
1013 : *
1014 : * This may generate XLOG records for expansion of the offsets and/or members
1015 : * files. Unfortunately, we have to do that while holding MultiXactGenLock
1016 : * to avoid race conditions --- the XLOG record for zeroing a page must appear
1017 : * before any backend can possibly try to store data in that page!
1018 : *
1019 : * We start a critical section before advancing the shared counters. The
1020 : * caller must end the critical section after writing SLRU data.
1021 : */
1022 : static MultiXactId
1023 568 : GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
1024 : {
1025 : MultiXactId result;
1026 : MultiXactOffset nextOffset;
1027 :
1028 : debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
1029 :
1030 : /* safety check, we should never get this far in a HS standby */
1031 568 : if (RecoveryInProgress())
1032 0 : elog(ERROR, "cannot assign MultiXactIds during recovery");
1033 :
1034 568 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1035 :
1036 : /* Handle wraparound of the nextMXact counter */
1037 568 : if (MultiXactState->nextMXact < FirstMultiXactId)
1038 0 : MultiXactState->nextMXact = FirstMultiXactId;
1039 :
1040 : /* Assign the MXID */
1041 568 : result = MultiXactState->nextMXact;
1042 :
1043 : /*----------
1044 : * Check to see if it's safe to assign another MultiXactId. This protects
1045 : * against catastrophic data loss due to multixact wraparound. The basic
1046 : * rules are:
1047 : *
1048 : * If we're past multiVacLimit or the safe threshold for member storage
1049 : * space, or we don't know what the safe threshold for member storage is,
1050 : * start trying to force autovacuum cycles.
1051 : * If we're past multiWarnLimit, start issuing warnings.
1052 : * If we're past multiStopLimit, refuse to create new MultiXactIds.
1053 : *
1054 : * Note these are pretty much the same protections in GetNewTransactionId.
1055 : *----------
1056 : */
1057 568 : if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
1058 : {
1059 : /*
1060 : * For safety's sake, we release MultiXactGenLock while sending
1061 : * signals, warnings, etc. This is not so much because we care about
1062 : * preserving concurrency in this situation, as to avoid any
1063 : * possibility of deadlock while doing get_database_name(). First,
1064 : * copy all the shared values we'll need in this path.
1065 : */
1066 0 : MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
1067 0 : MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
1068 0 : MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
1069 0 : Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
1070 :
1071 0 : LWLockRelease(MultiXactGenLock);
1072 :
1073 0 : if (IsUnderPostmaster &&
1074 0 : !MultiXactIdPrecedes(result, multiStopLimit))
1075 : {
1076 0 : char *oldest_datname = get_database_name(oldest_datoid);
1077 :
1078 : /*
1079 : * Immediately kick autovacuum into action as we're already in
1080 : * ERROR territory.
1081 : */
1082 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1083 :
1084 : /* complain even if that DB has disappeared */
1085 0 : if (oldest_datname)
1086 0 : ereport(ERROR,
1087 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1088 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database \"%s\"",
1089 : oldest_datname),
1090 : errhint("Execute a database-wide VACUUM in that database.\n"
1091 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1092 : else
1093 0 : ereport(ERROR,
1094 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1095 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database with OID %u",
1096 : oldest_datoid),
1097 : errhint("Execute a database-wide VACUUM in that database.\n"
1098 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1099 : }
1100 :
1101 : /*
1102 : * To avoid swamping the postmaster with signals, we issue the autovac
1103 : * request only once per 64K multis generated. This still gives
1104 : * plenty of chances before we get into real trouble.
1105 : */
1106 0 : if (IsUnderPostmaster && (result % 65536) == 0)
1107 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1108 :
1109 0 : if (!MultiXactIdPrecedes(result, multiWarnLimit))
1110 : {
1111 0 : char *oldest_datname = get_database_name(oldest_datoid);
1112 :
1113 : /* complain even if that DB has disappeared */
1114 0 : if (oldest_datname)
1115 0 : ereport(WARNING,
1116 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1117 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1118 : multiWrapLimit - result,
1119 : oldest_datname,
1120 : multiWrapLimit - result),
1121 : errhint("Execute a database-wide VACUUM in that database.\n"
1122 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1123 : else
1124 0 : ereport(WARNING,
1125 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1126 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1127 : multiWrapLimit - result,
1128 : oldest_datoid,
1129 : multiWrapLimit - result),
1130 : errhint("Execute a database-wide VACUUM in that database.\n"
1131 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1132 : }
1133 :
1134 : /* Re-acquire lock and start over */
1135 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1136 0 : result = MultiXactState->nextMXact;
1137 0 : if (result < FirstMultiXactId)
1138 0 : result = FirstMultiXactId;
1139 : }
1140 :
1141 : /* Make sure there is room for the MXID in the file. */
1142 568 : ExtendMultiXactOffset(result);
1143 :
1144 : /*
1145 : * Reserve the members space, similarly to above. Also, be careful not to
1146 : * return zero as the starting offset for any multixact. See
1147 : * GetMultiXactIdMembers() for motivation.
1148 : */
1149 568 : nextOffset = MultiXactState->nextOffset;
1150 568 : if (nextOffset == 0)
1151 : {
1152 22 : *offset = 1;
1153 22 : nmembers++; /* allocate member slot 0 too */
1154 : }
1155 : else
1156 546 : *offset = nextOffset;
1157 :
1158 : /*----------
1159 : * Protect against overrun of the members space as well, with the
1160 : * following rules:
1161 : *
1162 : * If we're past offsetStopLimit, refuse to generate more multis.
1163 : * If we're close to offsetStopLimit, emit a warning.
1164 : *
1165 : * Arbitrarily, we start emitting warnings when we're 20 segments or less
1166 : * from offsetStopLimit.
1167 : *
1168 : * Note we haven't updated the shared state yet, so if we fail at this
1169 : * point, the multixact ID we grabbed can still be used by the next guy.
1170 : *
1171 : * Note that there is no point in forcing autovacuum runs here: the
1172 : * multixact freeze settings would have to be reduced for that to have any
1173 : * effect.
1174 : *----------
1175 : */
1176 : #define OFFSET_WARN_SEGMENTS 20
1177 1136 : if (MultiXactState->oldestOffsetKnown &&
1178 568 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit, nextOffset,
1179 : nmembers))
1180 : {
1181 : /* see comment in the corresponding offsets wraparound case */
1182 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1183 :
1184 0 : ereport(ERROR,
1185 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1186 : errmsg("multixact \"members\" limit exceeded"),
1187 : errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
1188 : "This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
1189 : MultiXactState->offsetStopLimit - nextOffset - 1,
1190 : nmembers,
1191 : MultiXactState->offsetStopLimit - nextOffset - 1),
1192 : 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.",
1193 : MultiXactState->oldestMultiXactDB)));
1194 : }
1195 :
1196 : /*
1197 : * Check whether we should kick autovacuum into action, to prevent members
1198 : * wraparound. NB we use a much larger window to trigger autovacuum than
1199 : * just the warning limit. The warning is just a measure of last resort -
1200 : * this is in line with GetNewTransactionId's behaviour.
1201 : */
1202 568 : if (!MultiXactState->oldestOffsetKnown ||
1203 568 : (MultiXactState->nextOffset - MultiXactState->oldestOffset
1204 568 : > MULTIXACT_MEMBER_SAFE_THRESHOLD))
1205 : {
1206 : /*
1207 : * To avoid swamping the postmaster with signals, we issue the autovac
1208 : * request only when crossing a segment boundary. With default
1209 : * compilation settings that's roughly after 50k members. This still
1210 : * gives plenty of chances before we get into real trouble.
1211 : */
1212 0 : if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
1213 0 : (MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
1214 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1215 : }
1216 :
1217 1136 : if (MultiXactState->oldestOffsetKnown &&
1218 568 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit,
1219 : nextOffset,
1220 : nmembers + MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT * OFFSET_WARN_SEGMENTS))
1221 0 : ereport(WARNING,
1222 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1223 : errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
1224 : "database with OID %u must be vacuumed before %d more multixact members are used",
1225 : MultiXactState->offsetStopLimit - nextOffset + nmembers,
1226 : MultiXactState->oldestMultiXactDB,
1227 : MultiXactState->offsetStopLimit - nextOffset + nmembers),
1228 : errhint("Execute a database-wide VACUUM in that database with reduced \"vacuum_multixact_freeze_min_age\" and \"vacuum_multixact_freeze_table_age\" settings.")));
1229 :
1230 568 : ExtendMultiXactMember(nextOffset, nmembers);
1231 :
1232 : /*
1233 : * Critical section from here until caller has written the data into the
1234 : * just-reserved SLRU space; we don't want to error out with a partly
1235 : * written MultiXact structure. (In particular, failing to write our
1236 : * start offset after advancing nextMXact would effectively corrupt the
1237 : * previous MultiXact.)
1238 : */
1239 568 : START_CRIT_SECTION();
1240 :
1241 : /*
1242 : * Advance counters. As in GetNewTransactionId(), this must not happen
1243 : * until after file extension has succeeded!
1244 : *
1245 : * We don't care about MultiXactId wraparound here; it will be handled by
1246 : * the next iteration. But note that nextMXact may be InvalidMultiXactId
1247 : * or the first value on a segment-beginning page after this routine
1248 : * exits, so anyone else looking at the variable must be prepared to deal
1249 : * with either case. Similarly, nextOffset may be zero, but we won't use
1250 : * that as the actual start offset of the next multixact.
1251 : */
1252 568 : (MultiXactState->nextMXact)++;
1253 :
1254 568 : MultiXactState->nextOffset += nmembers;
1255 :
1256 568 : LWLockRelease(MultiXactGenLock);
1257 :
1258 : debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
1259 568 : return result;
1260 : }
1261 :
1262 : /*
1263 : * GetMultiXactIdMembers
1264 : * Return the set of MultiXactMembers that make up a MultiXactId
1265 : *
1266 : * Return value is the number of members found, or -1 if there are none,
1267 : * and *members is set to a newly palloc'ed array of members. It's the
1268 : * caller's responsibility to free it when done with it.
1269 : *
1270 : * from_pgupgrade must be passed as true if and only if only the multixact
1271 : * corresponds to a value from a tuple that was locked in a 9.2-or-older
1272 : * installation and later pg_upgrade'd (that is, the infomask is
1273 : * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1274 : * can still be running, so we return -1 just like for an empty multixact
1275 : * without any further checking. It would be wrong to try to resolve such a
1276 : * multixact: either the multixact is within the current valid multixact
1277 : * range, in which case the returned result would be bogus, or outside that
1278 : * range, in which case an error would be raised.
1279 : *
1280 : * In all other cases, the passed multixact must be within the known valid
1281 : * range, that is, greater than or equal to oldestMultiXactId, and less than
1282 : * nextMXact. Otherwise, an error is raised.
1283 : *
1284 : * isLockOnly must be set to true if caller is certain that the given multi
1285 : * is used only to lock tuples; can be false without loss of correctness,
1286 : * but passing a true means we can return quickly without checking for
1287 : * old updates.
1288 : */
1289 : int
1290 6032 : GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
1291 : bool from_pgupgrade, bool isLockOnly)
1292 : {
1293 : int64 pageno;
1294 : int64 prev_pageno;
1295 : int entryno;
1296 : int slotno;
1297 : MultiXactOffset *offptr;
1298 : MultiXactOffset offset;
1299 : int length;
1300 : int truelength;
1301 : MultiXactId oldestMXact;
1302 : MultiXactId nextMXact;
1303 : MultiXactId tmpMXact;
1304 : MultiXactOffset nextOffset;
1305 : MultiXactMember *ptr;
1306 : LWLock *lock;
1307 6032 : bool slept = false;
1308 :
1309 : debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1310 :
1311 6032 : if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1312 : {
1313 0 : *members = NULL;
1314 0 : return -1;
1315 : }
1316 :
1317 : /* See if the MultiXactId is in the local cache */
1318 6032 : length = mXactCacheGetById(multi, members);
1319 6032 : if (length >= 0)
1320 : {
1321 : debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1322 : mxid_to_string(multi, length, *members));
1323 4214 : return length;
1324 : }
1325 :
1326 : /* Set our OldestVisibleMXactId[] entry if we didn't already */
1327 1818 : MultiXactIdSetOldestVisible();
1328 :
1329 : /*
1330 : * If we know the multi is used only for locking and not for updates, then
1331 : * we can skip checking if the value is older than our oldest visible
1332 : * multi. It cannot possibly still be running.
1333 : */
1334 3332 : if (isLockOnly &&
1335 1514 : MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyProcNumber]))
1336 : {
1337 : debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1338 1338 : *members = NULL;
1339 1338 : return -1;
1340 : }
1341 :
1342 : /*
1343 : * We check known limits on MultiXact before resorting to the SLRU area.
1344 : *
1345 : * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1346 : * useful; it has already been removed, or will be removed shortly, by
1347 : * truncation. If one is passed, an error is raised.
1348 : *
1349 : * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1350 : * implies undetected ID wraparound has occurred. This raises a hard
1351 : * error.
1352 : *
1353 : * Shared lock is enough here since we aren't modifying any global state.
1354 : * Acquire it just long enough to grab the current counter values. We may
1355 : * need both nextMXact and nextOffset; see below.
1356 : */
1357 480 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1358 :
1359 480 : oldestMXact = MultiXactState->oldestMultiXactId;
1360 480 : nextMXact = MultiXactState->nextMXact;
1361 480 : nextOffset = MultiXactState->nextOffset;
1362 :
1363 480 : LWLockRelease(MultiXactGenLock);
1364 :
1365 480 : if (MultiXactIdPrecedes(multi, oldestMXact))
1366 0 : ereport(ERROR,
1367 : (errcode(ERRCODE_INTERNAL_ERROR),
1368 : errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1369 : multi)));
1370 :
1371 480 : if (!MultiXactIdPrecedes(multi, nextMXact))
1372 0 : ereport(ERROR,
1373 : (errcode(ERRCODE_INTERNAL_ERROR),
1374 : errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1375 : multi)));
1376 :
1377 : /*
1378 : * Find out the offset at which we need to start reading MultiXactMembers
1379 : * and the number of members in the multixact. We determine the latter as
1380 : * the difference between this multixact's starting offset and the next
1381 : * one's. However, there are some corner cases to worry about:
1382 : *
1383 : * 1. This multixact may be the latest one created, in which case there is
1384 : * no next one to look at. In this case the nextOffset value we just
1385 : * saved is the correct endpoint.
1386 : *
1387 : * 2. The next multixact may still be in process of being filled in: that
1388 : * is, another process may have done GetNewMultiXactId but not yet written
1389 : * the offset entry for that ID. In that scenario, it is guaranteed that
1390 : * the offset entry for that multixact exists (because GetNewMultiXactId
1391 : * won't release MultiXactGenLock until it does) but contains zero
1392 : * (because we are careful to pre-zero offset pages). Because
1393 : * GetNewMultiXactId will never return zero as the starting offset for a
1394 : * multixact, when we read zero as the next multixact's offset, we know we
1395 : * have this case. We handle this by sleeping on the condition variable
1396 : * we have just for this; the process in charge will signal the CV as soon
1397 : * as it has finished writing the multixact offset.
1398 : *
1399 : * 3. Because GetNewMultiXactId increments offset zero to offset one to
1400 : * handle case #2, there is an ambiguity near the point of offset
1401 : * wraparound. If we see next multixact's offset is one, is that our
1402 : * multixact's actual endpoint, or did it end at zero with a subsequent
1403 : * increment? We handle this using the knowledge that if the zero'th
1404 : * member slot wasn't filled, it'll contain zero, and zero isn't a valid
1405 : * transaction ID so it can't be a multixact member. Therefore, if we
1406 : * read a zero from the members array, just ignore it.
1407 : *
1408 : * This is all pretty messy, but the mess occurs only in infrequent corner
1409 : * cases, so it seems better than holding the MultiXactGenLock for a long
1410 : * time on every multixact creation.
1411 : */
1412 480 : retry:
1413 482 : pageno = MultiXactIdToOffsetPage(multi);
1414 482 : entryno = MultiXactIdToOffsetEntry(multi);
1415 :
1416 : /* Acquire the bank lock for the page we need. */
1417 482 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1418 482 : LWLockAcquire(lock, LW_EXCLUSIVE);
1419 :
1420 482 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1421 482 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1422 482 : offptr += entryno;
1423 482 : offset = *offptr;
1424 :
1425 : Assert(offset != 0);
1426 :
1427 : /*
1428 : * Use the same increment rule as GetNewMultiXactId(), that is, don't
1429 : * handle wraparound explicitly until needed.
1430 : */
1431 482 : tmpMXact = multi + 1;
1432 :
1433 482 : if (nextMXact == tmpMXact)
1434 : {
1435 : /* Corner case 1: there is no next multixact */
1436 424 : length = nextOffset - offset;
1437 : }
1438 : else
1439 : {
1440 : MultiXactOffset nextMXOffset;
1441 :
1442 : /* handle wraparound if needed */
1443 58 : if (tmpMXact < FirstMultiXactId)
1444 0 : tmpMXact = FirstMultiXactId;
1445 :
1446 58 : prev_pageno = pageno;
1447 :
1448 58 : pageno = MultiXactIdToOffsetPage(tmpMXact);
1449 58 : entryno = MultiXactIdToOffsetEntry(tmpMXact);
1450 :
1451 58 : if (pageno != prev_pageno)
1452 : {
1453 : LWLock *newlock;
1454 :
1455 : /*
1456 : * Since we're going to access a different SLRU page, if this page
1457 : * falls under a different bank, release the old bank's lock and
1458 : * acquire the lock of the new bank.
1459 : */
1460 0 : newlock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1461 0 : if (newlock != lock)
1462 : {
1463 0 : LWLockRelease(lock);
1464 0 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1465 0 : lock = newlock;
1466 : }
1467 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1468 : }
1469 :
1470 58 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1471 58 : offptr += entryno;
1472 58 : nextMXOffset = *offptr;
1473 :
1474 58 : if (nextMXOffset == 0)
1475 : {
1476 : /* Corner case 2: next multixact is still being filled in */
1477 2 : LWLockRelease(lock);
1478 2 : CHECK_FOR_INTERRUPTS();
1479 :
1480 2 : INJECTION_POINT("multixact-get-members-cv-sleep", NULL);
1481 :
1482 2 : ConditionVariableSleep(&MultiXactState->nextoff_cv,
1483 : WAIT_EVENT_MULTIXACT_CREATION);
1484 2 : slept = true;
1485 2 : goto retry;
1486 : }
1487 :
1488 56 : length = nextMXOffset - offset;
1489 : }
1490 :
1491 480 : LWLockRelease(lock);
1492 480 : lock = NULL;
1493 :
1494 : /*
1495 : * If we slept above, clean up state; it's no longer needed.
1496 : */
1497 480 : if (slept)
1498 2 : ConditionVariableCancelSleep();
1499 :
1500 480 : ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1501 :
1502 480 : truelength = 0;
1503 480 : prev_pageno = -1;
1504 1708 : for (int i = 0; i < length; i++, offset++)
1505 : {
1506 : TransactionId *xactptr;
1507 : uint32 *flagsptr;
1508 : int flagsoff;
1509 : int bshift;
1510 : int memberoff;
1511 :
1512 1228 : pageno = MXOffsetToMemberPage(offset);
1513 1228 : memberoff = MXOffsetToMemberOffset(offset);
1514 :
1515 1228 : if (pageno != prev_pageno)
1516 : {
1517 : LWLock *newlock;
1518 :
1519 : /*
1520 : * Since we're going to access a different SLRU page, if this page
1521 : * falls under a different bank, release the old bank's lock and
1522 : * acquire the lock of the new bank.
1523 : */
1524 480 : newlock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
1525 480 : if (newlock != lock)
1526 : {
1527 480 : if (lock)
1528 0 : LWLockRelease(lock);
1529 480 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1530 480 : lock = newlock;
1531 : }
1532 :
1533 480 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1534 480 : prev_pageno = pageno;
1535 : }
1536 :
1537 1228 : xactptr = (TransactionId *)
1538 1228 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1539 :
1540 1228 : if (!TransactionIdIsValid(*xactptr))
1541 : {
1542 : /* Corner case 3: we must be looking at unused slot zero */
1543 : Assert(offset == 0);
1544 0 : continue;
1545 : }
1546 :
1547 1228 : flagsoff = MXOffsetToFlagsOffset(offset);
1548 1228 : bshift = MXOffsetToFlagsBitShift(offset);
1549 1228 : flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1550 :
1551 1228 : ptr[truelength].xid = *xactptr;
1552 1228 : ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1553 1228 : truelength++;
1554 : }
1555 :
1556 480 : LWLockRelease(lock);
1557 :
1558 : /* A multixid with zero members should not happen */
1559 : Assert(truelength > 0);
1560 :
1561 : /*
1562 : * Copy the result into the local cache.
1563 : */
1564 480 : mXactCachePut(multi, truelength, ptr);
1565 :
1566 : debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1567 : mxid_to_string(multi, truelength, ptr));
1568 480 : *members = ptr;
1569 480 : return truelength;
1570 : }
1571 :
1572 : /*
1573 : * mxactMemberComparator
1574 : * qsort comparison function for MultiXactMember
1575 : *
1576 : * We can't use wraparound comparison for XIDs because that does not respect
1577 : * the triangle inequality! Any old sort order will do.
1578 : */
1579 : static int
1580 3862 : mxactMemberComparator(const void *arg1, const void *arg2)
1581 : {
1582 3862 : MultiXactMember member1 = *(const MultiXactMember *) arg1;
1583 3862 : MultiXactMember member2 = *(const MultiXactMember *) arg2;
1584 :
1585 3862 : if (member1.xid > member2.xid)
1586 48 : return 1;
1587 3814 : if (member1.xid < member2.xid)
1588 3468 : return -1;
1589 346 : if (member1.status > member2.status)
1590 4 : return 1;
1591 342 : if (member1.status < member2.status)
1592 342 : return -1;
1593 0 : return 0;
1594 : }
1595 :
1596 : /*
1597 : * mXactCacheGetBySet
1598 : * returns a MultiXactId from the cache based on the set of
1599 : * TransactionIds that compose it, or InvalidMultiXactId if
1600 : * none matches.
1601 : *
1602 : * This is helpful, for example, if two transactions want to lock a huge
1603 : * table. By using the cache, the second will use the same MultiXactId
1604 : * for the majority of tuples, thus keeping MultiXactId usage low (saving
1605 : * both I/O and wraparound issues).
1606 : *
1607 : * NB: the passed members array will be sorted in-place.
1608 : */
1609 : static MultiXactId
1610 2268 : mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1611 : {
1612 : dlist_iter iter;
1613 :
1614 : debug_elog3(DEBUG2, "CacheGet: looking for %s",
1615 : mxid_to_string(InvalidMultiXactId, nmembers, members));
1616 :
1617 : /* sort the array so comparison is easy */
1618 2268 : qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1619 :
1620 2748 : dclist_foreach(iter, &MXactCache)
1621 : {
1622 2180 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1623 : iter.cur);
1624 :
1625 2180 : if (entry->nmembers != nmembers)
1626 246 : continue;
1627 :
1628 : /*
1629 : * We assume the cache entries are sorted, and that the unused bits in
1630 : * "status" are zeroed.
1631 : */
1632 1934 : if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1633 : {
1634 : debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1635 1700 : dclist_move_head(&MXactCache, iter.cur);
1636 1700 : return entry->multi;
1637 : }
1638 : }
1639 :
1640 : debug_elog2(DEBUG2, "CacheGet: not found :-(");
1641 568 : return InvalidMultiXactId;
1642 : }
1643 :
1644 : /*
1645 : * mXactCacheGetById
1646 : * returns the composing MultiXactMember set from the cache for a
1647 : * given MultiXactId, if present.
1648 : *
1649 : * If successful, *xids is set to the address of a palloc'd copy of the
1650 : * MultiXactMember set. Return value is number of members, or -1 on failure.
1651 : */
1652 : static int
1653 6032 : mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
1654 : {
1655 : dlist_iter iter;
1656 :
1657 : debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1658 :
1659 6918 : dclist_foreach(iter, &MXactCache)
1660 : {
1661 5100 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1662 : iter.cur);
1663 :
1664 5100 : if (entry->multi == multi)
1665 : {
1666 : MultiXactMember *ptr;
1667 : Size size;
1668 :
1669 4214 : size = sizeof(MultiXactMember) * entry->nmembers;
1670 4214 : ptr = (MultiXactMember *) palloc(size);
1671 :
1672 4214 : memcpy(ptr, entry->members, size);
1673 :
1674 : debug_elog3(DEBUG2, "CacheGet: found %s",
1675 : mxid_to_string(multi,
1676 : entry->nmembers,
1677 : entry->members));
1678 :
1679 : /*
1680 : * Note we modify the list while not using a modifiable iterator.
1681 : * This is acceptable only because we exit the iteration
1682 : * immediately afterwards.
1683 : */
1684 4214 : dclist_move_head(&MXactCache, iter.cur);
1685 :
1686 4214 : *members = ptr;
1687 4214 : return entry->nmembers;
1688 : }
1689 : }
1690 :
1691 : debug_elog2(DEBUG2, "CacheGet: not found");
1692 1818 : return -1;
1693 : }
1694 :
1695 : /*
1696 : * mXactCachePut
1697 : * Add a new MultiXactId and its composing set into the local cache.
1698 : */
1699 : static void
1700 1048 : mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1701 : {
1702 : mXactCacheEnt *entry;
1703 :
1704 : debug_elog3(DEBUG2, "CachePut: storing %s",
1705 : mxid_to_string(multi, nmembers, members));
1706 :
1707 1048 : if (MXactContext == NULL)
1708 : {
1709 : /* The cache only lives as long as the current transaction */
1710 : debug_elog2(DEBUG2, "CachePut: initializing memory context");
1711 754 : MXactContext = AllocSetContextCreate(TopTransactionContext,
1712 : "MultiXact cache context",
1713 : ALLOCSET_SMALL_SIZES);
1714 : }
1715 :
1716 : entry = (mXactCacheEnt *)
1717 1048 : MemoryContextAlloc(MXactContext,
1718 1048 : offsetof(mXactCacheEnt, members) +
1719 : nmembers * sizeof(MultiXactMember));
1720 :
1721 1048 : entry->multi = multi;
1722 1048 : entry->nmembers = nmembers;
1723 1048 : memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1724 :
1725 : /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1726 1048 : qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1727 :
1728 1048 : dclist_push_head(&MXactCache, &entry->node);
1729 1048 : if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
1730 : {
1731 : dlist_node *node;
1732 :
1733 0 : node = dclist_tail_node(&MXactCache);
1734 0 : dclist_delete_from(&MXactCache, node);
1735 :
1736 0 : entry = dclist_container(mXactCacheEnt, node, node);
1737 : debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1738 : entry->multi);
1739 :
1740 0 : pfree(entry);
1741 : }
1742 1048 : }
1743 :
1744 : char *
1745 0 : mxstatus_to_string(MultiXactStatus status)
1746 : {
1747 0 : switch (status)
1748 : {
1749 0 : case MultiXactStatusForKeyShare:
1750 0 : return "keysh";
1751 0 : case MultiXactStatusForShare:
1752 0 : return "sh";
1753 0 : case MultiXactStatusForNoKeyUpdate:
1754 0 : return "fornokeyupd";
1755 0 : case MultiXactStatusForUpdate:
1756 0 : return "forupd";
1757 0 : case MultiXactStatusNoKeyUpdate:
1758 0 : return "nokeyupd";
1759 0 : case MultiXactStatusUpdate:
1760 0 : return "upd";
1761 0 : default:
1762 0 : elog(ERROR, "unrecognized multixact status %d", status);
1763 : return "";
1764 : }
1765 : }
1766 :
1767 : char *
1768 0 : mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1769 : {
1770 : static char *str = NULL;
1771 : StringInfoData buf;
1772 : int i;
1773 :
1774 0 : if (str != NULL)
1775 0 : pfree(str);
1776 :
1777 0 : initStringInfo(&buf);
1778 :
1779 0 : appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1780 : mxstatus_to_string(members[0].status));
1781 :
1782 0 : for (i = 1; i < nmembers; i++)
1783 0 : appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1784 0 : mxstatus_to_string(members[i].status));
1785 :
1786 0 : appendStringInfoChar(&buf, ']');
1787 0 : str = MemoryContextStrdup(TopMemoryContext, buf.data);
1788 0 : pfree(buf.data);
1789 0 : return str;
1790 : }
1791 :
1792 : /*
1793 : * AtEOXact_MultiXact
1794 : * Handle transaction end for MultiXact
1795 : *
1796 : * This is called at top transaction commit or abort (we don't care which).
1797 : */
1798 : void
1799 1093158 : AtEOXact_MultiXact(void)
1800 : {
1801 : /*
1802 : * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1803 : * which should only be valid while within a transaction.
1804 : *
1805 : * We assume that storing a MultiXactId is atomic and so we need not take
1806 : * MultiXactGenLock to do this.
1807 : */
1808 1093158 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1809 1093158 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1810 :
1811 : /*
1812 : * Discard the local MultiXactId cache. Since MXactContext was created as
1813 : * a child of TopTransactionContext, we needn't delete it explicitly.
1814 : */
1815 1093158 : MXactContext = NULL;
1816 1093158 : dclist_init(&MXactCache);
1817 1093158 : }
1818 :
1819 : /*
1820 : * AtPrepare_MultiXact
1821 : * Save multixact state at 2PC transaction prepare
1822 : *
1823 : * In this phase, we only store our OldestMemberMXactId value in the two-phase
1824 : * state file.
1825 : */
1826 : void
1827 566 : AtPrepare_MultiXact(void)
1828 : {
1829 566 : MultiXactId myOldestMember = OldestMemberMXactId[MyProcNumber];
1830 :
1831 566 : if (MultiXactIdIsValid(myOldestMember))
1832 100 : RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
1833 : &myOldestMember, sizeof(MultiXactId));
1834 566 : }
1835 :
1836 : /*
1837 : * PostPrepare_MultiXact
1838 : * Clean up after successful PREPARE TRANSACTION
1839 : */
1840 : void
1841 566 : PostPrepare_MultiXact(FullTransactionId fxid)
1842 : {
1843 : MultiXactId myOldestMember;
1844 :
1845 : /*
1846 : * Transfer our OldestMemberMXactId value to the slot reserved for the
1847 : * prepared transaction.
1848 : */
1849 566 : myOldestMember = OldestMemberMXactId[MyProcNumber];
1850 566 : if (MultiXactIdIsValid(myOldestMember))
1851 : {
1852 100 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1853 :
1854 : /*
1855 : * Even though storing MultiXactId is atomic, acquire lock to make
1856 : * sure others see both changes, not just the reset of the slot of the
1857 : * current backend. Using a volatile pointer might suffice, but this
1858 : * isn't a hot spot.
1859 : */
1860 100 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1861 :
1862 100 : OldestMemberMXactId[dummyProcNumber] = myOldestMember;
1863 100 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1864 :
1865 100 : LWLockRelease(MultiXactGenLock);
1866 : }
1867 :
1868 : /*
1869 : * We don't need to transfer OldestVisibleMXactId value, because the
1870 : * transaction is not going to be looking at any more multixacts once it's
1871 : * prepared.
1872 : *
1873 : * We assume that storing a MultiXactId is atomic and so we need not take
1874 : * MultiXactGenLock to do this.
1875 : */
1876 566 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1877 :
1878 : /*
1879 : * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1880 : */
1881 566 : MXactContext = NULL;
1882 566 : dclist_init(&MXactCache);
1883 566 : }
1884 :
1885 : /*
1886 : * multixact_twophase_recover
1887 : * Recover the state of a prepared transaction at startup
1888 : */
1889 : void
1890 16 : multixact_twophase_recover(FullTransactionId fxid, uint16 info,
1891 : void *recdata, uint32 len)
1892 : {
1893 16 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1894 : MultiXactId oldestMember;
1895 :
1896 : /*
1897 : * Get the oldest member XID from the state file record, and set it in the
1898 : * OldestMemberMXactId slot reserved for this prepared transaction.
1899 : */
1900 : Assert(len == sizeof(MultiXactId));
1901 16 : oldestMember = *((MultiXactId *) recdata);
1902 :
1903 16 : OldestMemberMXactId[dummyProcNumber] = oldestMember;
1904 16 : }
1905 :
1906 : /*
1907 : * multixact_twophase_postcommit
1908 : * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1909 : */
1910 : void
1911 108 : multixact_twophase_postcommit(FullTransactionId fxid, uint16 info,
1912 : void *recdata, uint32 len)
1913 : {
1914 108 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, true);
1915 :
1916 : Assert(len == sizeof(MultiXactId));
1917 :
1918 108 : OldestMemberMXactId[dummyProcNumber] = InvalidMultiXactId;
1919 108 : }
1920 :
1921 : /*
1922 : * multixact_twophase_postabort
1923 : * This is actually just the same as the COMMIT case.
1924 : */
1925 : void
1926 44 : multixact_twophase_postabort(FullTransactionId fxid, uint16 info,
1927 : void *recdata, uint32 len)
1928 : {
1929 44 : multixact_twophase_postcommit(fxid, info, recdata, len);
1930 44 : }
1931 :
1932 : /*
1933 : * Initialization of shared memory for MultiXact. We use two SLRU areas,
1934 : * thus double memory. Also, reserve space for the shared MultiXactState
1935 : * struct and the per-backend MultiXactId arrays (two of those, too).
1936 : */
1937 : Size
1938 4060 : MultiXactShmemSize(void)
1939 : {
1940 : Size size;
1941 :
1942 : /* We need 2*MaxOldestSlot perBackendXactIds[] entries */
1943 : #define SHARED_MULTIXACT_STATE_SIZE \
1944 : add_size(offsetof(MultiXactStateData, perBackendXactIds), \
1945 : mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1946 :
1947 4060 : size = SHARED_MULTIXACT_STATE_SIZE;
1948 4060 : size = add_size(size, SimpleLruShmemSize(multixact_offset_buffers, 0));
1949 4060 : size = add_size(size, SimpleLruShmemSize(multixact_member_buffers, 0));
1950 :
1951 4060 : return size;
1952 : }
1953 :
1954 : void
1955 2180 : MultiXactShmemInit(void)
1956 : {
1957 : bool found;
1958 :
1959 : debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1960 :
1961 2180 : MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
1962 2180 : MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
1963 :
1964 2180 : SimpleLruInit(MultiXactOffsetCtl,
1965 : "multixact_offset", multixact_offset_buffers, 0,
1966 : "pg_multixact/offsets", LWTRANCHE_MULTIXACTOFFSET_BUFFER,
1967 : LWTRANCHE_MULTIXACTOFFSET_SLRU,
1968 : SYNC_HANDLER_MULTIXACT_OFFSET,
1969 : false);
1970 : SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
1971 2180 : SimpleLruInit(MultiXactMemberCtl,
1972 : "multixact_member", multixact_member_buffers, 0,
1973 : "pg_multixact/members", LWTRANCHE_MULTIXACTMEMBER_BUFFER,
1974 : LWTRANCHE_MULTIXACTMEMBER_SLRU,
1975 : SYNC_HANDLER_MULTIXACT_MEMBER,
1976 : false);
1977 : /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1978 :
1979 : /* Initialize our shared state struct */
1980 2180 : MultiXactState = ShmemInitStruct("Shared MultiXact State",
1981 2180 : SHARED_MULTIXACT_STATE_SIZE,
1982 : &found);
1983 2180 : if (!IsUnderPostmaster)
1984 : {
1985 : Assert(!found);
1986 :
1987 : /* Make sure we zero out the per-backend state */
1988 2180 : MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
1989 2180 : ConditionVariableInit(&MultiXactState->nextoff_cv);
1990 : }
1991 : else
1992 : Assert(found);
1993 :
1994 : /*
1995 : * Set up array pointers.
1996 : */
1997 2180 : OldestMemberMXactId = MultiXactState->perBackendXactIds;
1998 2180 : OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
1999 2180 : }
2000 :
2001 : /*
2002 : * GUC check_hook for multixact_offset_buffers
2003 : */
2004 : bool
2005 2254 : check_multixact_offset_buffers(int *newval, void **extra, GucSource source)
2006 : {
2007 2254 : return check_slru_buffers("multixact_offset_buffers", newval);
2008 : }
2009 :
2010 : /*
2011 : * GUC check_hook for multixact_member_buffers
2012 : */
2013 : bool
2014 2254 : check_multixact_member_buffers(int *newval, void **extra, GucSource source)
2015 : {
2016 2254 : return check_slru_buffers("multixact_member_buffers", newval);
2017 : }
2018 :
2019 : /*
2020 : * This func must be called ONCE on system install. It creates the initial
2021 : * MultiXact segments. (The MultiXacts directories are assumed to have been
2022 : * created by initdb, and MultiXactShmemInit must have been called already.)
2023 : */
2024 : void
2025 100 : BootStrapMultiXact(void)
2026 : {
2027 : /* Zero the initial pages and flush them to disk */
2028 100 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, 0);
2029 100 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, 0);
2030 100 : }
2031 :
2032 : /*
2033 : * MaybeExtendOffsetSlru
2034 : * Extend the offsets SLRU area, if necessary
2035 : *
2036 : * After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
2037 : * contain files that are shorter than necessary; this would occur if the old
2038 : * installation had used multixacts beyond the first page (files cannot be
2039 : * copied, because the on-disk representation is different). pg_upgrade would
2040 : * update pg_control to set the next offset value to be at that position, so
2041 : * that tuples marked as locked by such MultiXacts would be seen as visible
2042 : * without having to consult multixact. However, trying to create and use a
2043 : * new MultiXactId would result in an error because the page on which the new
2044 : * value would reside does not exist. This routine is in charge of creating
2045 : * such pages.
2046 : */
2047 : static void
2048 94 : MaybeExtendOffsetSlru(void)
2049 : {
2050 : int64 pageno;
2051 : LWLock *lock;
2052 :
2053 94 : pageno = MultiXactIdToOffsetPage(MultiXactState->nextMXact);
2054 94 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2055 :
2056 94 : LWLockAcquire(lock, LW_EXCLUSIVE);
2057 :
2058 94 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2059 : {
2060 : int slotno;
2061 :
2062 : /*
2063 : * Fortunately for us, SimpleLruWritePage is already prepared to deal
2064 : * with creating a new segment file even if the page we're writing is
2065 : * not the first in it, so this is enough.
2066 : */
2067 0 : slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2068 0 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
2069 : }
2070 :
2071 94 : LWLockRelease(lock);
2072 94 : }
2073 :
2074 : /*
2075 : * This must be called ONCE during postmaster or standalone-backend startup.
2076 : *
2077 : * StartupXLOG has already established nextMXact/nextOffset by calling
2078 : * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
2079 : * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
2080 : * replayed WAL.
2081 : */
2082 : void
2083 1894 : StartupMultiXact(void)
2084 : {
2085 1894 : MultiXactId multi = MultiXactState->nextMXact;
2086 1894 : MultiXactOffset offset = MultiXactState->nextOffset;
2087 : int64 pageno;
2088 :
2089 : /*
2090 : * Initialize offset's idea of the latest page number.
2091 : */
2092 1894 : pageno = MultiXactIdToOffsetPage(multi);
2093 1894 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2094 : pageno);
2095 :
2096 : /*
2097 : * Initialize member's idea of the latest page number.
2098 : */
2099 1894 : pageno = MXOffsetToMemberPage(offset);
2100 1894 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2101 : pageno);
2102 1894 : }
2103 :
2104 : /*
2105 : * This must be called ONCE at the end of startup/recovery.
2106 : */
2107 : void
2108 1778 : TrimMultiXact(void)
2109 : {
2110 : MultiXactId nextMXact;
2111 : MultiXactOffset offset;
2112 : MultiXactId oldestMXact;
2113 : Oid oldestMXactDB;
2114 : int64 pageno;
2115 : int entryno;
2116 : int flagsoff;
2117 :
2118 1778 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2119 1778 : nextMXact = MultiXactState->nextMXact;
2120 1778 : offset = MultiXactState->nextOffset;
2121 1778 : oldestMXact = MultiXactState->oldestMultiXactId;
2122 1778 : oldestMXactDB = MultiXactState->oldestMultiXactDB;
2123 1778 : LWLockRelease(MultiXactGenLock);
2124 :
2125 : /* Clean up offsets state */
2126 :
2127 : /*
2128 : * (Re-)Initialize our idea of the latest page number for offsets.
2129 : */
2130 1778 : pageno = MultiXactIdToOffsetPage(nextMXact);
2131 1778 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2132 : pageno);
2133 :
2134 : /*
2135 : * Zero out the remainder of the current offsets page. See notes in
2136 : * TrimCLOG() for background. Unlike CLOG, some WAL record covers every
2137 : * pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
2138 : * rule "write xlog before data," nextMXact successors may carry obsolete,
2139 : * nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
2140 : * operates normally.
2141 : */
2142 1778 : entryno = MultiXactIdToOffsetEntry(nextMXact);
2143 1778 : if (entryno != 0)
2144 : {
2145 : int slotno;
2146 : MultiXactOffset *offptr;
2147 1776 : LWLock *lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2148 :
2149 1776 : LWLockAcquire(lock, LW_EXCLUSIVE);
2150 1776 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
2151 1776 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2152 1776 : offptr += entryno;
2153 :
2154 1776 : MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
2155 :
2156 1776 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
2157 1776 : LWLockRelease(lock);
2158 : }
2159 :
2160 : /*
2161 : * And the same for members.
2162 : *
2163 : * (Re-)Initialize our idea of the latest page number for members.
2164 : */
2165 1778 : pageno = MXOffsetToMemberPage(offset);
2166 1778 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2167 : pageno);
2168 :
2169 : /*
2170 : * Zero out the remainder of the current members page. See notes in
2171 : * TrimCLOG() for motivation.
2172 : */
2173 1778 : flagsoff = MXOffsetToFlagsOffset(offset);
2174 1778 : if (flagsoff != 0)
2175 : {
2176 : int slotno;
2177 : TransactionId *xidptr;
2178 : int memberoff;
2179 22 : LWLock *lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2180 :
2181 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2182 22 : memberoff = MXOffsetToMemberOffset(offset);
2183 22 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
2184 22 : xidptr = (TransactionId *)
2185 22 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
2186 :
2187 22 : MemSet(xidptr, 0, BLCKSZ - memberoff);
2188 :
2189 : /*
2190 : * Note: we don't need to zero out the flag bits in the remaining
2191 : * members of the current group, because they are always reset before
2192 : * writing.
2193 : */
2194 :
2195 22 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
2196 22 : LWLockRelease(lock);
2197 : }
2198 :
2199 : /* signal that we're officially up */
2200 1778 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2201 1778 : MultiXactState->finishedStartup = true;
2202 1778 : LWLockRelease(MultiXactGenLock);
2203 :
2204 : /* Now compute how far away the next members wraparound is. */
2205 1778 : SetMultiXactIdLimit(oldestMXact, oldestMXactDB, true);
2206 1778 : }
2207 :
2208 : /*
2209 : * Get the MultiXact data to save in a checkpoint record
2210 : */
2211 : void
2212 3054 : MultiXactGetCheckptMulti(bool is_shutdown,
2213 : MultiXactId *nextMulti,
2214 : MultiXactOffset *nextMultiOffset,
2215 : MultiXactId *oldestMulti,
2216 : Oid *oldestMultiDB)
2217 : {
2218 3054 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2219 3054 : *nextMulti = MultiXactState->nextMXact;
2220 3054 : *nextMultiOffset = MultiXactState->nextOffset;
2221 3054 : *oldestMulti = MultiXactState->oldestMultiXactId;
2222 3054 : *oldestMultiDB = MultiXactState->oldestMultiXactDB;
2223 3054 : LWLockRelease(MultiXactGenLock);
2224 :
2225 : debug_elog6(DEBUG2,
2226 : "MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
2227 : *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
2228 3054 : }
2229 :
2230 : /*
2231 : * Perform a checkpoint --- either during shutdown, or on-the-fly
2232 : */
2233 : void
2234 3434 : CheckPointMultiXact(void)
2235 : {
2236 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
2237 :
2238 : /*
2239 : * Write dirty MultiXact pages to disk. This may result in sync requests
2240 : * queued for later handling by ProcessSyncRequests(), as part of the
2241 : * checkpoint.
2242 : */
2243 3434 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2244 3434 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2245 :
2246 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
2247 3434 : }
2248 :
2249 : /*
2250 : * Set the next-to-be-assigned MultiXactId and offset
2251 : *
2252 : * This is used when we can determine the correct next ID/offset exactly
2253 : * from a checkpoint record. Although this is only called during bootstrap
2254 : * and XLog replay, we take the lock in case any hot-standby backends are
2255 : * examining the values.
2256 : */
2257 : void
2258 2064 : MultiXactSetNextMXact(MultiXactId nextMulti,
2259 : MultiXactOffset nextMultiOffset)
2260 : {
2261 : debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
2262 : nextMulti, nextMultiOffset);
2263 2064 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2264 2064 : MultiXactState->nextMXact = nextMulti;
2265 2064 : MultiXactState->nextOffset = nextMultiOffset;
2266 2064 : LWLockRelease(MultiXactGenLock);
2267 :
2268 : /*
2269 : * During a binary upgrade, make sure that the offsets SLRU is large
2270 : * enough to contain the next value that would be created.
2271 : *
2272 : * We need to do this pretty early during the first startup in binary
2273 : * upgrade mode: before StartupMultiXact() in fact, because this routine
2274 : * is called even before that by StartupXLOG(). And we can't do it
2275 : * earlier than at this point, because during that first call of this
2276 : * routine we determine the MultiXactState->nextMXact value that
2277 : * MaybeExtendOffsetSlru needs.
2278 : */
2279 2064 : if (IsBinaryUpgrade)
2280 94 : MaybeExtendOffsetSlru();
2281 2064 : }
2282 :
2283 : /*
2284 : * Determine the last safe MultiXactId to allocate given the currently oldest
2285 : * datminmxid (ie, the oldest MultiXactId that might exist in any database
2286 : * of our cluster), and the OID of the (or a) database with that value.
2287 : *
2288 : * is_startup is true when we are just starting the cluster, false when we
2289 : * are updating state in a running cluster. This only affects log messages.
2290 : */
2291 : void
2292 6104 : SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
2293 : bool is_startup)
2294 : {
2295 : MultiXactId multiVacLimit;
2296 : MultiXactId multiWarnLimit;
2297 : MultiXactId multiStopLimit;
2298 : MultiXactId multiWrapLimit;
2299 : MultiXactId curMulti;
2300 : bool needs_offset_vacuum;
2301 :
2302 : Assert(MultiXactIdIsValid(oldest_datminmxid));
2303 :
2304 : /*
2305 : * We pretend that a wrap will happen halfway through the multixact ID
2306 : * space, but that's not really true, because multixacts wrap differently
2307 : * from transaction IDs. Note that, separately from any concern about
2308 : * multixact IDs wrapping, we must ensure that multixact members do not
2309 : * wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
2310 : */
2311 6104 : multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2312 6104 : if (multiWrapLimit < FirstMultiXactId)
2313 0 : multiWrapLimit += FirstMultiXactId;
2314 :
2315 : /*
2316 : * We'll refuse to continue assigning MultiXactIds once we get within 3M
2317 : * multi of data loss. See SetTransactionIdLimit.
2318 : */
2319 6104 : multiStopLimit = multiWrapLimit - 3000000;
2320 6104 : if (multiStopLimit < FirstMultiXactId)
2321 0 : multiStopLimit -= FirstMultiXactId;
2322 :
2323 : /*
2324 : * We'll start complaining loudly when we get within 40M multis of data
2325 : * loss. This is kind of arbitrary, but if you let your gas gauge get
2326 : * down to 2% of full, would you be looking for the next gas station? We
2327 : * need to be fairly liberal about this number because there are lots of
2328 : * scenarios where most transactions are done by automatic clients that
2329 : * won't pay attention to warnings. (No, we're not gonna make this
2330 : * configurable. If you know enough to configure it, you know enough to
2331 : * not get in this kind of trouble in the first place.)
2332 : */
2333 6104 : multiWarnLimit = multiWrapLimit - 40000000;
2334 6104 : if (multiWarnLimit < FirstMultiXactId)
2335 0 : multiWarnLimit -= FirstMultiXactId;
2336 :
2337 : /*
2338 : * We'll start trying to force autovacuums when oldest_datminmxid gets to
2339 : * be more than autovacuum_multixact_freeze_max_age mxids old.
2340 : *
2341 : * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2342 : * so that we don't have to worry about dealing with on-the-fly changes in
2343 : * its value. See SetTransactionIdLimit.
2344 : */
2345 6104 : multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2346 6104 : if (multiVacLimit < FirstMultiXactId)
2347 0 : multiVacLimit += FirstMultiXactId;
2348 :
2349 : /* Grab lock for just long enough to set the new limit values */
2350 6104 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2351 6104 : MultiXactState->oldestMultiXactId = oldest_datminmxid;
2352 6104 : MultiXactState->oldestMultiXactDB = oldest_datoid;
2353 6104 : MultiXactState->multiVacLimit = multiVacLimit;
2354 6104 : MultiXactState->multiWarnLimit = multiWarnLimit;
2355 6104 : MultiXactState->multiStopLimit = multiStopLimit;
2356 6104 : MultiXactState->multiWrapLimit = multiWrapLimit;
2357 6104 : curMulti = MultiXactState->nextMXact;
2358 6104 : LWLockRelease(MultiXactGenLock);
2359 :
2360 : /* Log the info */
2361 6104 : ereport(DEBUG1,
2362 : (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2363 : multiWrapLimit, oldest_datoid)));
2364 :
2365 : /*
2366 : * Computing the actual limits is only possible once the data directory is
2367 : * in a consistent state. There's no need to compute the limits while
2368 : * still replaying WAL - no decisions about new multis are made even
2369 : * though multixact creations might be replayed. So we'll only do further
2370 : * checks after TrimMultiXact() has been called.
2371 : */
2372 6104 : if (!MultiXactState->finishedStartup)
2373 1996 : return;
2374 :
2375 : Assert(!InRecovery);
2376 :
2377 : /* Set limits for offset vacuum. */
2378 4108 : needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
2379 :
2380 : /*
2381 : * If past the autovacuum force point, immediately signal an autovac
2382 : * request. The reason for this is that autovac only processes one
2383 : * database per invocation. Once it's finished cleaning up the oldest
2384 : * database, it'll call here, and we'll signal the postmaster to start
2385 : * another iteration immediately if there are still any old databases.
2386 : */
2387 4108 : if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
2388 0 : needs_offset_vacuum) && IsUnderPostmaster)
2389 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
2390 :
2391 : /* Give an immediate warning if past the wrap warn point */
2392 4108 : if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2393 : {
2394 : char *oldest_datname;
2395 :
2396 : /*
2397 : * We can be called when not inside a transaction, for example during
2398 : * StartupXLOG(). In such a case we cannot do database access, so we
2399 : * must just report the oldest DB's OID.
2400 : *
2401 : * Note: it's also possible that get_database_name fails and returns
2402 : * NULL, for example because the database just got dropped. We'll
2403 : * still warn, even though the warning might now be unnecessary.
2404 : */
2405 0 : if (IsTransactionState())
2406 0 : oldest_datname = get_database_name(oldest_datoid);
2407 : else
2408 0 : oldest_datname = NULL;
2409 :
2410 0 : if (oldest_datname)
2411 0 : ereport(WARNING,
2412 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2413 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2414 : multiWrapLimit - curMulti,
2415 : oldest_datname,
2416 : multiWrapLimit - curMulti),
2417 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2418 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2419 : else
2420 0 : ereport(WARNING,
2421 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2422 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2423 : multiWrapLimit - curMulti,
2424 : oldest_datoid,
2425 : multiWrapLimit - curMulti),
2426 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2427 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2428 : }
2429 : }
2430 :
2431 : /*
2432 : * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2433 : * and similarly nextOffset is at least minMultiOffset.
2434 : *
2435 : * This is used when we can determine minimum safe values from an XLog
2436 : * record (either an on-line checkpoint or an mxact creation log entry).
2437 : * Although this is only called during XLog replay, we take the lock in case
2438 : * any hot-standby backends are examining the values.
2439 : */
2440 : void
2441 1354 : MultiXactAdvanceNextMXact(MultiXactId minMulti,
2442 : MultiXactOffset minMultiOffset)
2443 : {
2444 1354 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2445 1354 : if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
2446 : {
2447 : debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2448 4 : MultiXactState->nextMXact = minMulti;
2449 : }
2450 1354 : if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
2451 : {
2452 : debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
2453 : minMultiOffset);
2454 4 : MultiXactState->nextOffset = minMultiOffset;
2455 : }
2456 1354 : LWLockRelease(MultiXactGenLock);
2457 1354 : }
2458 :
2459 : /*
2460 : * Update our oldestMultiXactId value, but only if it's more recent than what
2461 : * we had.
2462 : *
2463 : * This may only be called during WAL replay.
2464 : */
2465 : void
2466 1418 : MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2467 : {
2468 : Assert(InRecovery);
2469 :
2470 1418 : if (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
2471 0 : SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
2472 1418 : }
2473 :
2474 : /*
2475 : * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2476 : *
2477 : * NB: this is called while holding MultiXactGenLock. We want it to be very
2478 : * fast most of the time; even when it's not so fast, no actual I/O need
2479 : * happen unless we're forced to write out a dirty log or xlog page to make
2480 : * room in shared memory.
2481 : */
2482 : static void
2483 568 : ExtendMultiXactOffset(MultiXactId multi)
2484 : {
2485 : int64 pageno;
2486 : LWLock *lock;
2487 :
2488 : /*
2489 : * No work except at first MultiXactId of a page. But beware: just after
2490 : * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2491 : */
2492 568 : if (MultiXactIdToOffsetEntry(multi) != 0 &&
2493 : multi != FirstMultiXactId)
2494 546 : return;
2495 :
2496 22 : pageno = MultiXactIdToOffsetPage(multi);
2497 22 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2498 :
2499 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2500 :
2501 : /* Zero the page and make a WAL entry about it */
2502 22 : SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2503 22 : XLogSimpleInsertInt64(RM_MULTIXACT_ID, XLOG_MULTIXACT_ZERO_OFF_PAGE,
2504 : pageno);
2505 :
2506 22 : LWLockRelease(lock);
2507 : }
2508 :
2509 : /*
2510 : * Make sure that MultiXactMember has room for the members of a newly-
2511 : * allocated MultiXactId.
2512 : *
2513 : * Like the above routine, this is called while holding MultiXactGenLock;
2514 : * same comments apply.
2515 : */
2516 : static void
2517 568 : ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
2518 : {
2519 : /*
2520 : * It's possible that the members span more than one page of the members
2521 : * file, so we loop to ensure we consider each page. The coding is not
2522 : * optimal if the members span several pages, but that seems unusual
2523 : * enough to not worry much about.
2524 : */
2525 1136 : while (nmembers > 0)
2526 : {
2527 : int flagsoff;
2528 : int flagsbit;
2529 : uint32 difference;
2530 :
2531 : /*
2532 : * Only zero when at first entry of a page.
2533 : */
2534 568 : flagsoff = MXOffsetToFlagsOffset(offset);
2535 568 : flagsbit = MXOffsetToFlagsBitShift(offset);
2536 568 : if (flagsoff == 0 && flagsbit == 0)
2537 : {
2538 : int64 pageno;
2539 : LWLock *lock;
2540 :
2541 22 : pageno = MXOffsetToMemberPage(offset);
2542 22 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2543 :
2544 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2545 :
2546 : /* Zero the page and make a WAL entry about it */
2547 22 : SimpleLruZeroPage(MultiXactMemberCtl, pageno);
2548 22 : XLogSimpleInsertInt64(RM_MULTIXACT_ID,
2549 : XLOG_MULTIXACT_ZERO_MEM_PAGE, pageno);
2550 :
2551 22 : LWLockRelease(lock);
2552 : }
2553 :
2554 : /*
2555 : * Compute the number of items till end of current page. Careful: if
2556 : * addition of unsigned ints wraps around, we're at the last page of
2557 : * the last segment; since that page holds a different number of items
2558 : * than other pages, we need to do it differently.
2559 : */
2560 568 : if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
2561 : {
2562 : /*
2563 : * This is the last page of the last segment; we can compute the
2564 : * number of items left to allocate in it without modulo
2565 : * arithmetic.
2566 : */
2567 0 : difference = MaxMultiXactOffset - offset + 1;
2568 : }
2569 : else
2570 568 : difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
2571 :
2572 : /*
2573 : * Advance to next page, taking care to properly handle the wraparound
2574 : * case. OK if nmembers goes negative.
2575 : */
2576 568 : nmembers -= difference;
2577 568 : offset += difference;
2578 : }
2579 568 : }
2580 :
2581 : /*
2582 : * GetOldestMultiXactId
2583 : *
2584 : * Return the oldest MultiXactId that's still possibly still seen as live by
2585 : * any running transaction. Older ones might still exist on disk, but they no
2586 : * longer have any running member transaction.
2587 : *
2588 : * It's not safe to truncate MultiXact SLRU segments on the value returned by
2589 : * this function; however, it can be set as the new relminmxid for any table
2590 : * that VACUUM knows has no remaining MXIDs < the same value. It is only safe
2591 : * to truncate SLRUs when no table can possibly still have a referencing MXID.
2592 : */
2593 : MultiXactId
2594 313586 : GetOldestMultiXactId(void)
2595 : {
2596 : MultiXactId oldestMXact;
2597 : MultiXactId nextMXact;
2598 : int i;
2599 :
2600 : /*
2601 : * This is the oldest valid value among all the OldestMemberMXactId[] and
2602 : * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2603 : */
2604 313586 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2605 :
2606 : /*
2607 : * We have to beware of the possibility that nextMXact is in the
2608 : * wrapped-around state. We don't fix the counter itself here, but we
2609 : * must be sure to use a valid value in our calculation.
2610 : */
2611 313586 : nextMXact = MultiXactState->nextMXact;
2612 313586 : if (nextMXact < FirstMultiXactId)
2613 0 : nextMXact = FirstMultiXactId;
2614 :
2615 313586 : oldestMXact = nextMXact;
2616 39137498 : for (i = 0; i < MaxOldestSlot; i++)
2617 : {
2618 : MultiXactId thisoldest;
2619 :
2620 38823912 : thisoldest = OldestMemberMXactId[i];
2621 38904250 : if (MultiXactIdIsValid(thisoldest) &&
2622 80338 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2623 10 : oldestMXact = thisoldest;
2624 38823912 : thisoldest = OldestVisibleMXactId[i];
2625 38824082 : if (MultiXactIdIsValid(thisoldest) &&
2626 170 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2627 4 : oldestMXact = thisoldest;
2628 : }
2629 :
2630 313586 : LWLockRelease(MultiXactGenLock);
2631 :
2632 313586 : return oldestMXact;
2633 : }
2634 :
2635 : /*
2636 : * Determine how aggressively we need to vacuum in order to prevent member
2637 : * wraparound.
2638 : *
2639 : * To do so determine what's the oldest member offset and install the limit
2640 : * info in MultiXactState, where it can be used to prevent overrun of old data
2641 : * in the members SLRU area.
2642 : *
2643 : * The return value is true if emergency autovacuum is required and false
2644 : * otherwise.
2645 : */
2646 : static bool
2647 4108 : SetOffsetVacuumLimit(bool is_startup)
2648 : {
2649 : MultiXactId oldestMultiXactId;
2650 : MultiXactId nextMXact;
2651 4108 : MultiXactOffset oldestOffset = 0; /* placate compiler */
2652 : MultiXactOffset prevOldestOffset;
2653 : MultiXactOffset nextOffset;
2654 4108 : bool oldestOffsetKnown = false;
2655 : bool prevOldestOffsetKnown;
2656 4108 : MultiXactOffset offsetStopLimit = 0;
2657 : MultiXactOffset prevOffsetStopLimit;
2658 :
2659 : /*
2660 : * NB: Have to prevent concurrent truncation, we might otherwise try to
2661 : * lookup an oldestMulti that's concurrently getting truncated away.
2662 : */
2663 4108 : LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2664 :
2665 : /* Read relevant fields from shared memory. */
2666 4108 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2667 4108 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2668 4108 : nextMXact = MultiXactState->nextMXact;
2669 4108 : nextOffset = MultiXactState->nextOffset;
2670 4108 : prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2671 4108 : prevOldestOffset = MultiXactState->oldestOffset;
2672 4108 : prevOffsetStopLimit = MultiXactState->offsetStopLimit;
2673 : Assert(MultiXactState->finishedStartup);
2674 4108 : LWLockRelease(MultiXactGenLock);
2675 :
2676 : /*
2677 : * Determine the offset of the oldest multixact. Normally, we can read
2678 : * the offset from the multixact itself, but there's an important special
2679 : * case: if there are no multixacts in existence at all, oldestMXact
2680 : * obviously can't point to one. It will instead point to the multixact
2681 : * ID that will be assigned the next time one is needed.
2682 : */
2683 4108 : if (oldestMultiXactId == nextMXact)
2684 : {
2685 : /*
2686 : * When the next multixact gets created, it will be stored at the next
2687 : * offset.
2688 : */
2689 4080 : oldestOffset = nextOffset;
2690 4080 : oldestOffsetKnown = true;
2691 : }
2692 : else
2693 : {
2694 : /*
2695 : * Figure out where the oldest existing multixact's offsets are
2696 : * stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
2697 : * the supposedly-earliest multixact might not really exist. We are
2698 : * careful not to fail in that case.
2699 : */
2700 : oldestOffsetKnown =
2701 28 : find_multixact_start(oldestMultiXactId, &oldestOffset);
2702 :
2703 28 : if (oldestOffsetKnown)
2704 28 : ereport(DEBUG1,
2705 : (errmsg_internal("oldest MultiXactId member is at offset %u",
2706 : oldestOffset)));
2707 : else
2708 0 : ereport(LOG,
2709 : (errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
2710 : oldestMultiXactId)));
2711 : }
2712 :
2713 4108 : LWLockRelease(MultiXactTruncationLock);
2714 :
2715 : /*
2716 : * If we can, compute limits (and install them MultiXactState) to prevent
2717 : * overrun of old data in the members SLRU area. We can only do so if the
2718 : * oldest offset is known though.
2719 : */
2720 4108 : if (oldestOffsetKnown)
2721 : {
2722 : /* move back to start of the corresponding segment */
2723 4108 : offsetStopLimit = oldestOffset - (oldestOffset %
2724 : (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT));
2725 :
2726 : /* always leave one segment before the wraparound point */
2727 4108 : offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
2728 :
2729 4108 : if (!prevOldestOffsetKnown && !is_startup)
2730 0 : ereport(LOG,
2731 : (errmsg("MultiXact member wraparound protections are now enabled")));
2732 :
2733 4108 : ereport(DEBUG1,
2734 : (errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
2735 : offsetStopLimit, oldestMultiXactId)));
2736 : }
2737 0 : else if (prevOldestOffsetKnown)
2738 : {
2739 : /*
2740 : * If we failed to get the oldest offset this time, but we have a
2741 : * value from a previous pass through this function, use the old
2742 : * values rather than automatically forcing an emergency autovacuum
2743 : * cycle again.
2744 : */
2745 0 : oldestOffset = prevOldestOffset;
2746 0 : oldestOffsetKnown = true;
2747 0 : offsetStopLimit = prevOffsetStopLimit;
2748 : }
2749 :
2750 : /* Install the computed values */
2751 4108 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2752 4108 : MultiXactState->oldestOffset = oldestOffset;
2753 4108 : MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
2754 4108 : MultiXactState->offsetStopLimit = offsetStopLimit;
2755 4108 : LWLockRelease(MultiXactGenLock);
2756 :
2757 : /*
2758 : * Do we need an emergency autovacuum? If we're not sure, assume yes.
2759 : */
2760 8216 : return !oldestOffsetKnown ||
2761 4108 : (nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
2762 : }
2763 :
2764 : /*
2765 : * Return whether adding "distance" to "start" would move past "boundary".
2766 : *
2767 : * We use this to determine whether the addition is "wrapping around" the
2768 : * boundary point, hence the name. The reason we don't want to use the regular
2769 : * 2^31-modulo arithmetic here is that we want to be able to use the whole of
2770 : * the 2^32-1 space here, allowing for more multixacts than would fit
2771 : * otherwise.
2772 : */
2773 : static bool
2774 1136 : MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start,
2775 : uint32 distance)
2776 : {
2777 : MultiXactOffset finish;
2778 :
2779 : /*
2780 : * Note that offset number 0 is not used (see GetMultiXactIdMembers), so
2781 : * if the addition wraps around the UINT_MAX boundary, skip that value.
2782 : */
2783 1136 : finish = start + distance;
2784 1136 : if (finish < start)
2785 0 : finish++;
2786 :
2787 : /*-----------------------------------------------------------------------
2788 : * When the boundary is numerically greater than the starting point, any
2789 : * value numerically between the two is not wrapped:
2790 : *
2791 : * <----S----B---->
2792 : * [---) = F wrapped past B (and UINT_MAX)
2793 : * [---) = F not wrapped
2794 : * [----] = F wrapped past B
2795 : *
2796 : * When the boundary is numerically less than the starting point (i.e. the
2797 : * UINT_MAX wraparound occurs somewhere in between) then all values in
2798 : * between are wrapped:
2799 : *
2800 : * <----B----S---->
2801 : * [---) = F not wrapped past B (but wrapped past UINT_MAX)
2802 : * [---) = F wrapped past B (and UINT_MAX)
2803 : * [----] = F not wrapped
2804 : *-----------------------------------------------------------------------
2805 : */
2806 1136 : if (start < boundary)
2807 1136 : return finish >= boundary || finish < start;
2808 : else
2809 0 : return finish >= boundary && finish < start;
2810 : }
2811 :
2812 : /*
2813 : * Find the starting offset of the given MultiXactId.
2814 : *
2815 : * Returns false if the file containing the multi does not exist on disk.
2816 : * Otherwise, returns true and sets *result to the starting member offset.
2817 : *
2818 : * This function does not prevent concurrent truncation, so if that's
2819 : * required, the caller has to protect against that.
2820 : */
2821 : static bool
2822 28 : find_multixact_start(MultiXactId multi, MultiXactOffset *result)
2823 : {
2824 : MultiXactOffset offset;
2825 : int64 pageno;
2826 : int entryno;
2827 : int slotno;
2828 : MultiXactOffset *offptr;
2829 :
2830 : Assert(MultiXactState->finishedStartup);
2831 :
2832 28 : pageno = MultiXactIdToOffsetPage(multi);
2833 28 : entryno = MultiXactIdToOffsetEntry(multi);
2834 :
2835 : /*
2836 : * Write out dirty data, so PhysicalPageExists can work correctly.
2837 : */
2838 28 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2839 28 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2840 :
2841 28 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2842 0 : return false;
2843 :
2844 : /* lock is acquired by SimpleLruReadPage_ReadOnly */
2845 28 : slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2846 28 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2847 28 : offptr += entryno;
2848 28 : offset = *offptr;
2849 28 : LWLockRelease(SimpleLruGetBankLock(MultiXactOffsetCtl, pageno));
2850 :
2851 28 : *result = offset;
2852 28 : return true;
2853 : }
2854 :
2855 : /*
2856 : * GetMultiXactInfo
2857 : *
2858 : * Returns information about the current MultiXact state, as of:
2859 : * multixacts: Number of MultiXacts (nextMultiXactId - oldestMultiXactId)
2860 : * members: Number of member entries (nextOffset - oldestOffset)
2861 : * oldestMultiXactId: Oldest MultiXact ID still in use
2862 : * oldestOffset: Oldest offset still in use
2863 : *
2864 : * Returns false if unable to determine, the oldest offset being unknown.
2865 : */
2866 : bool
2867 251246 : GetMultiXactInfo(uint32 *multixacts, MultiXactOffset *members,
2868 : MultiXactId *oldestMultiXactId, MultiXactOffset *oldestOffset)
2869 : {
2870 : MultiXactOffset nextOffset;
2871 : MultiXactId nextMultiXactId;
2872 : bool oldestOffsetKnown;
2873 :
2874 251246 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2875 251246 : nextOffset = MultiXactState->nextOffset;
2876 251246 : *oldestMultiXactId = MultiXactState->oldestMultiXactId;
2877 251246 : nextMultiXactId = MultiXactState->nextMXact;
2878 251246 : *oldestOffset = MultiXactState->oldestOffset;
2879 251246 : oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2880 251246 : LWLockRelease(MultiXactGenLock);
2881 :
2882 251246 : if (!oldestOffsetKnown)
2883 : {
2884 0 : *members = 0;
2885 0 : *multixacts = 0;
2886 0 : *oldestMultiXactId = InvalidMultiXactId;
2887 0 : *oldestOffset = 0;
2888 0 : return false;
2889 : }
2890 :
2891 251246 : *members = nextOffset - *oldestOffset;
2892 251246 : *multixacts = nextMultiXactId - *oldestMultiXactId;
2893 251246 : return true;
2894 : }
2895 :
2896 : /*
2897 : * Multixact members can be removed once the multixacts that refer to them
2898 : * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2899 : * vacuum_multixact_freeze_table_age work together to make sure we never have
2900 : * too many multixacts; we hope that, at least under normal circumstances,
2901 : * this will also be sufficient to keep us from using too many offsets.
2902 : * However, if the average multixact has many members, we might exhaust the
2903 : * members space while still using few enough members that these limits fail
2904 : * to trigger relminmxid advancement by VACUUM. At that point, we'd have no
2905 : * choice but to start failing multixact-creating operations with an error.
2906 : *
2907 : * To prevent that, if more than a threshold portion of the members space is
2908 : * used, we effectively reduce autovacuum_multixact_freeze_max_age and
2909 : * to a value just less than the number of multixacts in use. We hope that
2910 : * this will quickly trigger autovacuuming on the table or tables with the
2911 : * oldest relminmxid, thus allowing datminmxid values to advance and removing
2912 : * some members.
2913 : *
2914 : * As the fraction of the member space currently in use grows, we become
2915 : * more aggressive in clamping this value. That not only causes autovacuum
2916 : * to ramp up, but also makes any manual vacuums the user issues more
2917 : * aggressive. This happens because vacuum_get_cutoffs() will clamp the
2918 : * freeze table and the minimum freeze age cutoffs based on the effective
2919 : * autovacuum_multixact_freeze_max_age this function returns. In the worst
2920 : * case, we'll claim the freeze_max_age to zero, and every vacuum of any
2921 : * table will freeze every multixact.
2922 : */
2923 : int
2924 251246 : MultiXactMemberFreezeThreshold(void)
2925 : {
2926 : MultiXactOffset members;
2927 : uint32 multixacts;
2928 : uint32 victim_multixacts;
2929 : double fraction;
2930 : int result;
2931 : MultiXactId oldestMultiXactId;
2932 : MultiXactOffset oldestOffset;
2933 :
2934 : /* If we can't determine member space utilization, assume the worst. */
2935 251246 : if (!GetMultiXactInfo(&multixacts, &members, &oldestMultiXactId, &oldestOffset))
2936 0 : return 0;
2937 :
2938 : /* If member space utilization is low, no special action is required. */
2939 251246 : if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
2940 251246 : return autovacuum_multixact_freeze_max_age;
2941 :
2942 : /*
2943 : * Compute a target for relminmxid advancement. The number of multixacts
2944 : * we try to eliminate from the system is based on how far we are past
2945 : * MULTIXACT_MEMBER_SAFE_THRESHOLD.
2946 : */
2947 0 : fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
2948 : (MULTIXACT_MEMBER_DANGER_THRESHOLD - MULTIXACT_MEMBER_SAFE_THRESHOLD);
2949 0 : victim_multixacts = multixacts * fraction;
2950 :
2951 : /* fraction could be > 1.0, but lowest possible freeze age is zero */
2952 0 : if (victim_multixacts > multixacts)
2953 0 : return 0;
2954 0 : result = multixacts - victim_multixacts;
2955 :
2956 : /*
2957 : * Clamp to autovacuum_multixact_freeze_max_age, so that we never make
2958 : * autovacuum less aggressive than it would otherwise be.
2959 : */
2960 0 : return Min(result, autovacuum_multixact_freeze_max_age);
2961 : }
2962 :
2963 : typedef struct mxtruncinfo
2964 : {
2965 : int64 earliestExistingPage;
2966 : } mxtruncinfo;
2967 :
2968 : /*
2969 : * SlruScanDirectory callback
2970 : * This callback determines the earliest existing page number.
2971 : */
2972 : static bool
2973 0 : SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int64 segpage, void *data)
2974 : {
2975 0 : mxtruncinfo *trunc = (mxtruncinfo *) data;
2976 :
2977 0 : if (trunc->earliestExistingPage == -1 ||
2978 0 : ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
2979 : {
2980 0 : trunc->earliestExistingPage = segpage;
2981 : }
2982 :
2983 0 : return false; /* keep going */
2984 : }
2985 :
2986 :
2987 : /*
2988 : * Delete members segments [oldest, newOldest)
2989 : *
2990 : * The members SLRU can, in contrast to the offsets one, be filled to almost
2991 : * the full range at once. This means SimpleLruTruncate() can't trivially be
2992 : * used - instead the to-be-deleted range is computed using the offsets
2993 : * SLRU. C.f. TruncateMultiXact().
2994 : */
2995 : static void
2996 0 : PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
2997 : {
2998 0 : const int64 maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
2999 0 : int64 startsegment = MXOffsetToMemberSegment(oldestOffset);
3000 0 : int64 endsegment = MXOffsetToMemberSegment(newOldestOffset);
3001 0 : int64 segment = startsegment;
3002 :
3003 : /*
3004 : * Delete all the segments but the last one. The last segment can still
3005 : * contain, possibly partially, valid data.
3006 : */
3007 0 : while (segment != endsegment)
3008 : {
3009 0 : elog(DEBUG2, "truncating multixact members segment %" PRIx64,
3010 : segment);
3011 0 : SlruDeleteSegment(MultiXactMemberCtl, segment);
3012 :
3013 : /* move to next segment, handling wraparound correctly */
3014 0 : if (segment == maxsegment)
3015 0 : segment = 0;
3016 : else
3017 0 : segment += 1;
3018 : }
3019 0 : }
3020 :
3021 : /*
3022 : * Delete offsets segments [oldest, newOldest)
3023 : */
3024 : static void
3025 0 : PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
3026 : {
3027 : /*
3028 : * We step back one multixact to avoid passing a cutoff page that hasn't
3029 : * been created yet in the rare case that oldestMulti would be the first
3030 : * item on a page and oldestMulti == nextMulti. In that case, if we
3031 : * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
3032 : * detection.
3033 : */
3034 0 : SimpleLruTruncate(MultiXactOffsetCtl,
3035 : MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
3036 0 : }
3037 :
3038 : /*
3039 : * Remove all MultiXactOffset and MultiXactMember segments before the oldest
3040 : * ones still of interest.
3041 : *
3042 : * This is only called on a primary as part of vacuum (via
3043 : * vac_truncate_clog()). During recovery truncation is done by replaying
3044 : * truncation WAL records logged here.
3045 : *
3046 : * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
3047 : * is one of the databases preventing newOldestMulti from increasing.
3048 : */
3049 : void
3050 2330 : TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
3051 : {
3052 : MultiXactId oldestMulti;
3053 : MultiXactId nextMulti;
3054 : MultiXactOffset newOldestOffset;
3055 : MultiXactOffset oldestOffset;
3056 : MultiXactOffset nextOffset;
3057 : mxtruncinfo trunc;
3058 : MultiXactId earliest;
3059 :
3060 : Assert(!RecoveryInProgress());
3061 : Assert(MultiXactState->finishedStartup);
3062 :
3063 : /*
3064 : * We can only allow one truncation to happen at once. Otherwise parts of
3065 : * members might vanish while we're doing lookups or similar. There's no
3066 : * need to have an interlock with creating new multis or such, since those
3067 : * are constrained by the limits (which only grow, never shrink).
3068 : */
3069 2330 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3070 :
3071 2330 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
3072 2330 : nextMulti = MultiXactState->nextMXact;
3073 2330 : nextOffset = MultiXactState->nextOffset;
3074 2330 : oldestMulti = MultiXactState->oldestMultiXactId;
3075 2330 : LWLockRelease(MultiXactGenLock);
3076 : Assert(MultiXactIdIsValid(oldestMulti));
3077 :
3078 : /*
3079 : * Make sure to only attempt truncation if there's values to truncate
3080 : * away. In normal processing values shouldn't go backwards, but there's
3081 : * some corner cases (due to bugs) where that's possible.
3082 : */
3083 2330 : if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
3084 : {
3085 2330 : LWLockRelease(MultiXactTruncationLock);
3086 2330 : return;
3087 : }
3088 :
3089 : /*
3090 : * Note we can't just plow ahead with the truncation; it's possible that
3091 : * there are no segments to truncate, which is a problem because we are
3092 : * going to attempt to read the offsets page to determine where to
3093 : * truncate the members SLRU. So we first scan the directory to determine
3094 : * the earliest offsets page number that we can read without error.
3095 : *
3096 : * When nextMXact is less than one segment away from multiWrapLimit,
3097 : * SlruScanDirCbFindEarliest can find some early segment other than the
3098 : * actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
3099 : * returns false, because not all pairs of entries have the same answer.)
3100 : * That can also arise when an earlier truncation attempt failed unlink()
3101 : * or returned early from this function. The only consequence is
3102 : * returning early, which wastes space that we could have liberated.
3103 : *
3104 : * NB: It's also possible that the page that oldestMulti is on has already
3105 : * been truncated away, and we crashed before updating oldestMulti.
3106 : */
3107 0 : trunc.earliestExistingPage = -1;
3108 0 : SlruScanDirectory(MultiXactOffsetCtl, SlruScanDirCbFindEarliest, &trunc);
3109 0 : earliest = trunc.earliestExistingPage * MULTIXACT_OFFSETS_PER_PAGE;
3110 0 : if (earliest < FirstMultiXactId)
3111 0 : earliest = FirstMultiXactId;
3112 :
3113 : /* If there's nothing to remove, we can bail out early. */
3114 0 : if (MultiXactIdPrecedes(oldestMulti, earliest))
3115 : {
3116 0 : LWLockRelease(MultiXactTruncationLock);
3117 0 : return;
3118 : }
3119 :
3120 : /*
3121 : * First, compute the safe truncation point for MultiXactMember. This is
3122 : * the starting offset of the oldest multixact.
3123 : *
3124 : * Hopefully, find_multixact_start will always work here, because we've
3125 : * already checked that it doesn't precede the earliest MultiXact on disk.
3126 : * But if it fails, don't truncate anything, and log a message.
3127 : */
3128 0 : if (oldestMulti == nextMulti)
3129 : {
3130 : /* there are NO MultiXacts */
3131 0 : oldestOffset = nextOffset;
3132 : }
3133 0 : else if (!find_multixact_start(oldestMulti, &oldestOffset))
3134 : {
3135 0 : ereport(LOG,
3136 : (errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
3137 : oldestMulti, earliest)));
3138 0 : LWLockRelease(MultiXactTruncationLock);
3139 0 : return;
3140 : }
3141 :
3142 : /*
3143 : * Secondly compute up to where to truncate. Lookup the corresponding
3144 : * member offset for newOldestMulti for that.
3145 : */
3146 0 : if (newOldestMulti == nextMulti)
3147 : {
3148 : /* there are NO MultiXacts */
3149 0 : newOldestOffset = nextOffset;
3150 : }
3151 0 : else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
3152 : {
3153 0 : ereport(LOG,
3154 : (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
3155 : newOldestMulti)));
3156 0 : LWLockRelease(MultiXactTruncationLock);
3157 0 : return;
3158 : }
3159 :
3160 0 : elog(DEBUG1, "performing multixact truncation: "
3161 : "offsets [%u, %u), offsets segments [%" PRIx64 ", %" PRIx64 "), "
3162 : "members [%u, %u), members segments [%" PRIx64 ", %" PRIx64 ")",
3163 : oldestMulti, newOldestMulti,
3164 : MultiXactIdToOffsetSegment(oldestMulti),
3165 : MultiXactIdToOffsetSegment(newOldestMulti),
3166 : oldestOffset, newOldestOffset,
3167 : MXOffsetToMemberSegment(oldestOffset),
3168 : MXOffsetToMemberSegment(newOldestOffset));
3169 :
3170 : /*
3171 : * Do truncation, and the WAL logging of the truncation, in a critical
3172 : * section. That way offsets/members cannot get out of sync anymore, i.e.
3173 : * once consistent the newOldestMulti will always exist in members, even
3174 : * if we crashed in the wrong moment.
3175 : */
3176 0 : START_CRIT_SECTION();
3177 :
3178 : /*
3179 : * Prevent checkpoints from being scheduled concurrently. This is critical
3180 : * because otherwise a truncation record might not be replayed after a
3181 : * crash/basebackup, even though the state of the data directory would
3182 : * require it.
3183 : */
3184 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
3185 0 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
3186 :
3187 : /* WAL log truncation */
3188 0 : WriteMTruncateXlogRec(newOldestMultiDB,
3189 : oldestMulti, newOldestMulti,
3190 : oldestOffset, newOldestOffset);
3191 :
3192 : /*
3193 : * Update in-memory limits before performing the truncation, while inside
3194 : * the critical section: Have to do it before truncation, to prevent
3195 : * concurrent lookups of those values. Has to be inside the critical
3196 : * section as otherwise a future call to this function would error out,
3197 : * while looking up the oldest member in offsets, if our caller crashes
3198 : * before updating the limits.
3199 : */
3200 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
3201 0 : MultiXactState->oldestMultiXactId = newOldestMulti;
3202 0 : MultiXactState->oldestMultiXactDB = newOldestMultiDB;
3203 0 : LWLockRelease(MultiXactGenLock);
3204 :
3205 : /* First truncate members */
3206 0 : PerformMembersTruncation(oldestOffset, newOldestOffset);
3207 :
3208 : /* Then offsets */
3209 0 : PerformOffsetsTruncation(oldestMulti, newOldestMulti);
3210 :
3211 0 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
3212 :
3213 0 : END_CRIT_SECTION();
3214 0 : LWLockRelease(MultiXactTruncationLock);
3215 : }
3216 :
3217 : /*
3218 : * Decide whether a MultiXactOffset page number is "older" for truncation
3219 : * purposes. Analogous to CLOGPagePrecedes().
3220 : *
3221 : * Offsetting the values is optional, because MultiXactIdPrecedes() has
3222 : * translational symmetry.
3223 : */
3224 : static bool
3225 0 : MultiXactOffsetPagePrecedes(int64 page1, int64 page2)
3226 : {
3227 : MultiXactId multi1;
3228 : MultiXactId multi2;
3229 :
3230 0 : multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
3231 0 : multi1 += FirstMultiXactId + 1;
3232 0 : multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
3233 0 : multi2 += FirstMultiXactId + 1;
3234 :
3235 0 : return (MultiXactIdPrecedes(multi1, multi2) &&
3236 0 : MultiXactIdPrecedes(multi1,
3237 : multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
3238 : }
3239 :
3240 : /*
3241 : * Decide whether a MultiXactMember page number is "older" for truncation
3242 : * purposes. There is no "invalid offset number" so use the numbers verbatim.
3243 : */
3244 : static bool
3245 0 : MultiXactMemberPagePrecedes(int64 page1, int64 page2)
3246 : {
3247 : MultiXactOffset offset1;
3248 : MultiXactOffset offset2;
3249 :
3250 0 : offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
3251 0 : offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
3252 :
3253 0 : return (MultiXactOffsetPrecedes(offset1, offset2) &&
3254 0 : MultiXactOffsetPrecedes(offset1,
3255 : offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
3256 : }
3257 :
3258 : /*
3259 : * Decide which of two MultiXactIds is earlier.
3260 : *
3261 : * XXX do we need to do something special for InvalidMultiXactId?
3262 : * (Doesn't look like it.)
3263 : */
3264 : bool
3265 2803112 : MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
3266 : {
3267 2803112 : int32 diff = (int32) (multi1 - multi2);
3268 :
3269 2803112 : return (diff < 0);
3270 : }
3271 :
3272 : /*
3273 : * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
3274 : *
3275 : * XXX do we need to do something special for InvalidMultiXactId?
3276 : * (Doesn't look like it.)
3277 : */
3278 : bool
3279 13552 : MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
3280 : {
3281 13552 : int32 diff = (int32) (multi1 - multi2);
3282 :
3283 13552 : return (diff <= 0);
3284 : }
3285 :
3286 :
3287 : /*
3288 : * Decide which of two offsets is earlier.
3289 : */
3290 : static bool
3291 1354 : MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
3292 : {
3293 1354 : int32 diff = (int32) (offset1 - offset2);
3294 :
3295 1354 : return (diff < 0);
3296 : }
3297 :
3298 : /*
3299 : * Write a TRUNCATE xlog record
3300 : *
3301 : * We must flush the xlog record to disk before returning --- see notes in
3302 : * TruncateCLOG().
3303 : */
3304 : static void
3305 0 : WriteMTruncateXlogRec(Oid oldestMultiDB,
3306 : MultiXactId startTruncOff, MultiXactId endTruncOff,
3307 : MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
3308 : {
3309 : XLogRecPtr recptr;
3310 : xl_multixact_truncate xlrec;
3311 :
3312 0 : xlrec.oldestMultiDB = oldestMultiDB;
3313 :
3314 0 : xlrec.startTruncOff = startTruncOff;
3315 0 : xlrec.endTruncOff = endTruncOff;
3316 :
3317 0 : xlrec.startTruncMemb = startTruncMemb;
3318 0 : xlrec.endTruncMemb = endTruncMemb;
3319 :
3320 0 : XLogBeginInsert();
3321 0 : XLogRegisterData(&xlrec, SizeOfMultiXactTruncate);
3322 0 : recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
3323 0 : XLogFlush(recptr);
3324 0 : }
3325 :
3326 : /*
3327 : * MULTIXACT resource manager's routines
3328 : */
3329 : void
3330 8 : multixact_redo(XLogReaderState *record)
3331 : {
3332 8 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3333 :
3334 : /* Backup blocks are not used in multixact records */
3335 : Assert(!XLogRecHasAnyBlockRefs(record));
3336 :
3337 8 : if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
3338 : {
3339 : int64 pageno;
3340 :
3341 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3342 2 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, pageno);
3343 : }
3344 6 : else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
3345 : {
3346 : int64 pageno;
3347 :
3348 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3349 2 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, pageno);
3350 : }
3351 4 : else if (info == XLOG_MULTIXACT_CREATE_ID)
3352 : {
3353 4 : xl_multixact_create *xlrec =
3354 4 : (xl_multixact_create *) XLogRecGetData(record);
3355 : TransactionId max_xid;
3356 : int i;
3357 :
3358 : /* Store the data back into the SLRU files */
3359 4 : RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
3360 4 : xlrec->members);
3361 :
3362 : /* Make sure nextMXact/nextOffset are beyond what this record has */
3363 4 : MultiXactAdvanceNextMXact(xlrec->mid + 1,
3364 4 : xlrec->moff + xlrec->nmembers);
3365 :
3366 : /*
3367 : * Make sure nextXid is beyond any XID mentioned in the record. This
3368 : * should be unnecessary, since any XID found here ought to have other
3369 : * evidence in the XLOG, but let's be safe.
3370 : */
3371 4 : max_xid = XLogRecGetXid(record);
3372 12 : for (i = 0; i < xlrec->nmembers; i++)
3373 : {
3374 8 : if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
3375 0 : max_xid = xlrec->members[i].xid;
3376 : }
3377 :
3378 4 : AdvanceNextFullTransactionIdPastXid(max_xid);
3379 : }
3380 0 : else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
3381 : {
3382 : xl_multixact_truncate xlrec;
3383 : int64 pageno;
3384 :
3385 0 : memcpy(&xlrec, XLogRecGetData(record),
3386 : SizeOfMultiXactTruncate);
3387 :
3388 0 : elog(DEBUG1, "replaying multixact truncation: "
3389 : "offsets [%u, %u), offsets segments [%" PRIx64 ", %" PRIx64 "), "
3390 : "members [%u, %u), members segments [%" PRIx64 ", %" PRIx64 ")",
3391 : xlrec.startTruncOff, xlrec.endTruncOff,
3392 : MultiXactIdToOffsetSegment(xlrec.startTruncOff),
3393 : MultiXactIdToOffsetSegment(xlrec.endTruncOff),
3394 : xlrec.startTruncMemb, xlrec.endTruncMemb,
3395 : MXOffsetToMemberSegment(xlrec.startTruncMemb),
3396 : MXOffsetToMemberSegment(xlrec.endTruncMemb));
3397 :
3398 : /* should not be required, but more than cheap enough */
3399 0 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3400 :
3401 : /*
3402 : * Advance the horizon values, so they're current at the end of
3403 : * recovery.
3404 : */
3405 0 : SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
3406 :
3407 0 : PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
3408 :
3409 : /*
3410 : * During XLOG replay, latest_page_number isn't necessarily set up
3411 : * yet; insert a suitable value to bypass the sanity test in
3412 : * SimpleLruTruncate.
3413 : */
3414 0 : pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
3415 0 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
3416 : pageno);
3417 0 : PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
3418 :
3419 0 : LWLockRelease(MultiXactTruncationLock);
3420 : }
3421 : else
3422 0 : elog(PANIC, "multixact_redo: unknown op code %u", info);
3423 8 : }
3424 :
3425 : /*
3426 : * Entrypoint for sync.c to sync offsets files.
3427 : */
3428 : int
3429 0 : multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
3430 : {
3431 0 : return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
3432 : }
3433 :
3434 : /*
3435 : * Entrypoint for sync.c to sync members files.
3436 : */
3437 : int
3438 0 : multixactmemberssyncfiletag(const FileTag *ftag, char *path)
3439 : {
3440 0 : return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
3441 : }
|
