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/transam.h"
74 : #include "access/twophase.h"
75 : #include "access/twophase_rmgr.h"
76 : #include "access/xact.h"
77 : #include "access/xlog.h"
78 : #include "access/xloginsert.h"
79 : #include "access/xlogutils.h"
80 : #include "commands/dbcommands.h"
81 : #include "funcapi.h"
82 : #include "lib/ilist.h"
83 : #include "miscadmin.h"
84 : #include "pg_trace.h"
85 : #include "pgstat.h"
86 : #include "postmaster/autovacuum.h"
87 : #include "storage/pmsignal.h"
88 : #include "storage/proc.h"
89 : #include "storage/procarray.h"
90 : #include "utils/fmgrprotos.h"
91 : #include "utils/guc_hooks.h"
92 : #include "utils/injection_point.h"
93 : #include "utils/memutils.h"
94 :
95 :
96 : /*
97 : * Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
98 : * used everywhere else in Postgres.
99 : *
100 : * Note: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
101 : * MultiXact page numbering also wraps around at
102 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
103 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need
104 : * take no explicit notice of that fact in this module, except when comparing
105 : * segment and page numbers in TruncateMultiXact (see
106 : * MultiXactOffsetPagePrecedes).
107 : */
108 :
109 : /* We need four bytes per offset */
110 : #define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
111 :
112 : static inline int64
113 4834 : MultiXactIdToOffsetPage(MultiXactId multi)
114 : {
115 4834 : return multi / MULTIXACT_OFFSETS_PER_PAGE;
116 : }
117 :
118 : static inline int
119 3410 : MultiXactIdToOffsetEntry(MultiXactId multi)
120 : {
121 3410 : return multi % MULTIXACT_OFFSETS_PER_PAGE;
122 : }
123 :
124 : static inline int64
125 0 : MultiXactIdToOffsetSegment(MultiXactId multi)
126 : {
127 0 : return MultiXactIdToOffsetPage(multi) / SLRU_PAGES_PER_SEGMENT;
128 : }
129 :
130 : /*
131 : * The situation for members is a bit more complex: we store one byte of
132 : * additional flag bits for each TransactionId. To do this without getting
133 : * into alignment issues, we store four bytes of flags, and then the
134 : * corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
135 : * are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
136 : * per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
137 : * performance) trumps space efficiency here.
138 : *
139 : * Note that the "offset" macros work with byte offset, not array indexes, so
140 : * arithmetic must be done using "char *" pointers.
141 : */
142 : /* We need eight bits per xact, so one xact fits in a byte */
143 : #define MXACT_MEMBER_BITS_PER_XACT 8
144 : #define MXACT_MEMBER_FLAGS_PER_BYTE 1
145 : #define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
146 :
147 : /* how many full bytes of flags are there in a group? */
148 : #define MULTIXACT_FLAGBYTES_PER_GROUP 4
149 : #define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
150 : (MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
151 : /* size in bytes of a complete group */
152 : #define MULTIXACT_MEMBERGROUP_SIZE \
153 : (sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
154 : #define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
155 : #define MULTIXACT_MEMBERS_PER_PAGE \
156 : (MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
157 :
158 : /*
159 : * Because the number of items per page is not a divisor of the last item
160 : * number (member 0xFFFFFFFF), the last segment does not use the maximum number
161 : * of pages, and moreover the last used page therein does not use the same
162 : * number of items as previous pages. (Another way to say it is that the
163 : * 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
164 : * has some empty space after that item.)
165 : *
166 : * This constant is the number of members in the last page of the last segment.
167 : */
168 : #define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
169 : ((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
170 :
171 : /* page in which a member is to be found */
172 : static inline int64
173 6050 : MXOffsetToMemberPage(MultiXactOffset offset)
174 : {
175 6050 : return offset / MULTIXACT_MEMBERS_PER_PAGE;
176 : }
177 :
178 : static inline int64
179 0 : MXOffsetToMemberSegment(MultiXactOffset offset)
180 : {
181 0 : return MXOffsetToMemberPage(offset) / SLRU_PAGES_PER_SEGMENT;
182 : }
183 :
184 : /* Location (byte offset within page) of flag word for a given member */
185 : static inline int
186 7162 : MXOffsetToFlagsOffset(MultiXactOffset offset)
187 : {
188 7162 : MultiXactOffset group = offset / MULTIXACT_MEMBERS_PER_MEMBERGROUP;
189 7162 : int grouponpg = group % MULTIXACT_MEMBERGROUPS_PER_PAGE;
190 7162 : int byteoff = grouponpg * MULTIXACT_MEMBERGROUP_SIZE;
191 :
192 7162 : return byteoff;
193 : }
194 :
195 : static inline int
196 2974 : MXOffsetToFlagsBitShift(MultiXactOffset offset)
197 : {
198 2974 : int member_in_group = offset % MULTIXACT_MEMBERS_PER_MEMBERGROUP;
199 2974 : int bshift = member_in_group * MXACT_MEMBER_BITS_PER_XACT;
200 :
201 2974 : return bshift;
202 : }
203 :
204 : /* Location (byte offset within page) of TransactionId of given member */
205 : static inline int
206 2442 : MXOffsetToMemberOffset(MultiXactOffset offset)
207 : {
208 2442 : int member_in_group = offset % MULTIXACT_MEMBERS_PER_MEMBERGROUP;
209 :
210 2442 : return MXOffsetToFlagsOffset(offset) +
211 2442 : MULTIXACT_FLAGBYTES_PER_GROUP +
212 : member_in_group * sizeof(TransactionId);
213 : }
214 :
215 : /* Multixact members wraparound thresholds. */
216 : #define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
217 : #define MULTIXACT_MEMBER_DANGER_THRESHOLD \
218 : (MaxMultiXactOffset - MaxMultiXactOffset / 4)
219 :
220 : static inline MultiXactId
221 0 : PreviousMultiXactId(MultiXactId multi)
222 : {
223 0 : return multi == FirstMultiXactId ? MaxMultiXactId : multi - 1;
224 : }
225 :
226 : /*
227 : * Links to shared-memory data structures for MultiXact control
228 : */
229 : static SlruCtlData MultiXactOffsetCtlData;
230 : static SlruCtlData MultiXactMemberCtlData;
231 :
232 : #define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
233 : #define MultiXactMemberCtl (&MultiXactMemberCtlData)
234 :
235 : /*
236 : * MultiXact state shared across all backends. All this state is protected
237 : * by MultiXactGenLock. (We also use SLRU bank's lock of MultiXactOffset and
238 : * MultiXactMember to guard accesses to the two sets of SLRU buffers. For
239 : * concurrency's sake, we avoid holding more than one of these locks at a
240 : * time.)
241 : */
242 : typedef struct MultiXactStateData
243 : {
244 : /* next-to-be-assigned MultiXactId */
245 : MultiXactId nextMXact;
246 :
247 : /* next-to-be-assigned offset */
248 : MultiXactOffset nextOffset;
249 :
250 : /* Have we completed multixact startup? */
251 : bool finishedStartup;
252 :
253 : /*
254 : * Oldest multixact that is still potentially referenced by a relation.
255 : * Anything older than this should not be consulted. These values are
256 : * updated by vacuum.
257 : */
258 : MultiXactId oldestMultiXactId;
259 : Oid oldestMultiXactDB;
260 :
261 : /*
262 : * Oldest multixact offset that is potentially referenced by a multixact
263 : * referenced by a relation. We don't always know this value, so there's
264 : * a flag here to indicate whether or not we currently do.
265 : */
266 : MultiXactOffset oldestOffset;
267 : bool oldestOffsetKnown;
268 :
269 : /* support for anti-wraparound measures */
270 : MultiXactId multiVacLimit;
271 : MultiXactId multiWarnLimit;
272 : MultiXactId multiStopLimit;
273 : MultiXactId multiWrapLimit;
274 :
275 : /* support for members anti-wraparound measures */
276 : MultiXactOffset offsetStopLimit; /* known if oldestOffsetKnown */
277 :
278 : /*
279 : * This is used to sleep until a multixact offset is written when we want
280 : * to create the next one.
281 : */
282 : ConditionVariable nextoff_cv;
283 :
284 : /*
285 : * Per-backend data starts here. We have two arrays stored in the area
286 : * immediately following the MultiXactStateData struct. Each is indexed by
287 : * ProcNumber.
288 : *
289 : * In both arrays, there's a slot for all normal backends
290 : * (0..MaxBackends-1) followed by a slot for max_prepared_xacts prepared
291 : * transactions.
292 : *
293 : * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
294 : * transaction(s) could possibly be a member of, or InvalidMultiXactId
295 : * when the backend has no live transaction that could possibly be a
296 : * member of a MultiXact. Each backend sets its entry to the current
297 : * nextMXact counter just before first acquiring a shared lock in a given
298 : * transaction, and clears it at transaction end. (This works because only
299 : * during or after acquiring a shared lock could an XID possibly become a
300 : * member of a MultiXact, and that MultiXact would have to be created
301 : * during or after the lock acquisition.)
302 : *
303 : * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
304 : * current transaction(s) think is potentially live, or InvalidMultiXactId
305 : * when not in a transaction or not in a transaction that's paid any
306 : * attention to MultiXacts yet. This is computed when first needed in a
307 : * given transaction, and cleared at transaction end. We can compute it
308 : * as the minimum of the valid OldestMemberMXactId[] entries at the time
309 : * we compute it (using nextMXact if none are valid). Each backend is
310 : * required not to attempt to access any SLRU data for MultiXactIds older
311 : * than its own OldestVisibleMXactId[] setting; this is necessary because
312 : * the relevant SLRU data can be concurrently truncated away.
313 : *
314 : * The oldest valid value among all of the OldestMemberMXactId[] and
315 : * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
316 : * possible value still having any live member transaction -- OldestMxact.
317 : * Any value older than that is typically removed from tuple headers, or
318 : * "frozen" via being replaced with a new xmax. VACUUM can sometimes even
319 : * remove an individual MultiXact xmax whose value is >= its OldestMxact
320 : * cutoff, though typically only when no individual member XID is still
321 : * running. See FreezeMultiXactId for full details.
322 : *
323 : * Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
324 : * or the oldest extant Multi remaining in the table is used as the new
325 : * pg_class.relminmxid value (whichever is earlier). The minimum of all
326 : * relminmxid values in each database is stored in pg_database.datminmxid.
327 : * In turn, the minimum of all of those values is stored in pg_control.
328 : * This is used as the truncation point for pg_multixact when unneeded
329 : * segments get removed by vac_truncate_clog() during vacuuming.
330 : */
331 : MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
332 : } MultiXactStateData;
333 :
334 : /*
335 : * Size of OldestMemberMXactId and OldestVisibleMXactId arrays.
336 : */
337 : #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
338 :
339 : /* Pointers to the state data in shared memory */
340 : static MultiXactStateData *MultiXactState;
341 : static MultiXactId *OldestMemberMXactId;
342 : static MultiXactId *OldestVisibleMXactId;
343 :
344 :
345 : /*
346 : * Definitions for the backend-local MultiXactId cache.
347 : *
348 : * We use this cache to store known MultiXacts, so we don't need to go to
349 : * SLRU areas every time.
350 : *
351 : * The cache lasts for the duration of a single transaction, the rationale
352 : * for this being that most entries will contain our own TransactionId and
353 : * so they will be uninteresting by the time our next transaction starts.
354 : * (XXX not clear that this is correct --- other members of the MultiXact
355 : * could hang around longer than we did. However, it's not clear what a
356 : * better policy for flushing old cache entries would be.) FIXME actually
357 : * this is plain wrong now that multixact's may contain update Xids.
358 : *
359 : * We allocate the cache entries in a memory context that is deleted at
360 : * transaction end, so we don't need to do retail freeing of entries.
361 : */
362 : typedef struct mXactCacheEnt
363 : {
364 : MultiXactId multi;
365 : int nmembers;
366 : dlist_node node;
367 : MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
368 : } mXactCacheEnt;
369 :
370 : #define MAX_CACHE_ENTRIES 256
371 : static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
372 : static MemoryContext MXactContext = NULL;
373 :
374 : #ifdef MULTIXACT_DEBUG
375 : #define debug_elog2(a,b) elog(a,b)
376 : #define debug_elog3(a,b,c) elog(a,b,c)
377 : #define debug_elog4(a,b,c,d) elog(a,b,c,d)
378 : #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
379 : #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
380 : #else
381 : #define debug_elog2(a,b)
382 : #define debug_elog3(a,b,c)
383 : #define debug_elog4(a,b,c,d)
384 : #define debug_elog5(a,b,c,d,e)
385 : #define debug_elog6(a,b,c,d,e,f)
386 : #endif
387 :
388 : /* internal MultiXactId management */
389 : static void MultiXactIdSetOldestVisible(void);
390 : static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
391 : int nmembers, MultiXactMember *members);
392 : static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
393 :
394 : /* MultiXact cache management */
395 : static int mxactMemberComparator(const void *arg1, const void *arg2);
396 : static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
397 : static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
398 : static void mXactCachePut(MultiXactId multi, int nmembers,
399 : MultiXactMember *members);
400 :
401 : static char *mxstatus_to_string(MultiXactStatus status);
402 :
403 : /* management of SLRU infrastructure */
404 : static bool MultiXactOffsetPagePrecedes(int64 page1, int64 page2);
405 : static bool MultiXactMemberPagePrecedes(int64 page1, int64 page2);
406 : static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
407 : MultiXactOffset offset2);
408 : static void ExtendMultiXactOffset(MultiXactId multi);
409 : static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
410 : static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
411 : MultiXactOffset start, uint32 distance);
412 : static bool SetOffsetVacuumLimit(bool is_startup);
413 : static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
414 : static void WriteMTruncateXlogRec(Oid oldestMultiDB,
415 : MultiXactId startTruncOff,
416 : MultiXactId endTruncOff,
417 : MultiXactOffset startTruncMemb,
418 : MultiXactOffset endTruncMemb);
419 :
420 :
421 : /*
422 : * MultiXactIdCreate
423 : * Construct a MultiXactId representing two TransactionIds.
424 : *
425 : * The two XIDs must be different, or be requesting different statuses.
426 : *
427 : * NB - we don't worry about our local MultiXactId cache here, because that
428 : * is handled by the lower-level routines.
429 : */
430 : MultiXactId
431 2064 : MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
432 : TransactionId xid2, MultiXactStatus status2)
433 : {
434 : MultiXactId newMulti;
435 : MultiXactMember members[2];
436 :
437 : Assert(TransactionIdIsValid(xid1));
438 : Assert(TransactionIdIsValid(xid2));
439 :
440 : Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
441 :
442 : /* MultiXactIdSetOldestMember() must have been called already. */
443 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
444 :
445 : /*
446 : * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
447 : * are still running. In typical usage, xid2 will be our own XID and the
448 : * caller just did a check on xid1, so it'd be wasted effort.
449 : */
450 :
451 2064 : members[0].xid = xid1;
452 2064 : members[0].status = status1;
453 2064 : members[1].xid = xid2;
454 2064 : members[1].status = status2;
455 :
456 2064 : newMulti = MultiXactIdCreateFromMembers(2, members);
457 :
458 : debug_elog3(DEBUG2, "Create: %s",
459 : mxid_to_string(newMulti, 2, members));
460 :
461 2064 : return newMulti;
462 : }
463 :
464 : /*
465 : * MultiXactIdExpand
466 : * Add a TransactionId to a pre-existing MultiXactId.
467 : *
468 : * If the TransactionId is already a member of the passed MultiXactId with the
469 : * same status, just return it as-is.
470 : *
471 : * Note that we do NOT actually modify the membership of a pre-existing
472 : * MultiXactId; instead we create a new one. This is necessary to avoid
473 : * a race condition against code trying to wait for one MultiXactId to finish;
474 : * see notes in heapam.c.
475 : *
476 : * NB - we don't worry about our local MultiXactId cache here, because that
477 : * is handled by the lower-level routines.
478 : *
479 : * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
480 : * one upgraded by pg_upgrade from a cluster older than this feature) are not
481 : * passed in.
482 : */
483 : MultiXactId
484 192 : MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
485 : {
486 : MultiXactId newMulti;
487 : MultiXactMember *members;
488 : MultiXactMember *newMembers;
489 : int nmembers;
490 : int i;
491 : int j;
492 :
493 : Assert(MultiXactIdIsValid(multi));
494 : Assert(TransactionIdIsValid(xid));
495 :
496 : /* MultiXactIdSetOldestMember() must have been called already. */
497 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
498 :
499 : debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
500 : multi, xid, mxstatus_to_string(status));
501 :
502 : /*
503 : * Note: we don't allow for old multis here. The reason is that the only
504 : * caller of this function does a check that the multixact is no longer
505 : * running.
506 : */
507 192 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
508 :
509 192 : if (nmembers < 0)
510 : {
511 : MultiXactMember member;
512 :
513 : /*
514 : * The MultiXactId is obsolete. This can only happen if all the
515 : * MultiXactId members stop running between the caller checking and
516 : * passing it to us. It would be better to return that fact to the
517 : * caller, but it would complicate the API and it's unlikely to happen
518 : * too often, so just deal with it by creating a singleton MultiXact.
519 : */
520 0 : member.xid = xid;
521 0 : member.status = status;
522 0 : newMulti = MultiXactIdCreateFromMembers(1, &member);
523 :
524 : debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
525 : multi, newMulti);
526 0 : return newMulti;
527 : }
528 :
529 : /*
530 : * If the TransactionId is already a member of the MultiXactId with the
531 : * same status, just return the existing MultiXactId.
532 : */
533 592 : for (i = 0; i < nmembers; i++)
534 : {
535 400 : if (TransactionIdEquals(members[i].xid, xid) &&
536 108 : (members[i].status == status))
537 : {
538 : debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
539 : xid, multi);
540 0 : pfree(members);
541 0 : return multi;
542 : }
543 : }
544 :
545 : /*
546 : * Determine which of the members of the MultiXactId are still of
547 : * interest. This is any running transaction, and also any transaction
548 : * that grabbed something stronger than just a lock and was committed. (An
549 : * update that aborted is of no interest here; and having more than one
550 : * update Xid in a multixact would cause errors elsewhere.)
551 : *
552 : * Removing dead members is not just an optimization: freezing of tuples
553 : * whose Xmax are multis depends on this behavior.
554 : *
555 : * Note we have the same race condition here as above: j could be 0 at the
556 : * end of the loop.
557 : */
558 : newMembers = (MultiXactMember *)
559 192 : palloc(sizeof(MultiXactMember) * (nmembers + 1));
560 :
561 592 : for (i = 0, j = 0; i < nmembers; i++)
562 : {
563 400 : if (TransactionIdIsInProgress(members[i].xid) ||
564 92 : (ISUPDATE_from_mxstatus(members[i].status) &&
565 12 : TransactionIdDidCommit(members[i].xid)))
566 : {
567 320 : newMembers[j].xid = members[i].xid;
568 320 : newMembers[j++].status = members[i].status;
569 : }
570 : }
571 :
572 192 : newMembers[j].xid = xid;
573 192 : newMembers[j++].status = status;
574 192 : newMulti = MultiXactIdCreateFromMembers(j, newMembers);
575 :
576 192 : pfree(members);
577 192 : pfree(newMembers);
578 :
579 : debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
580 :
581 192 : return newMulti;
582 : }
583 :
584 : /*
585 : * MultiXactIdIsRunning
586 : * Returns whether a MultiXactId is "running".
587 : *
588 : * We return true if at least one member of the given MultiXactId is still
589 : * running. Note that a "false" result is certain not to change,
590 : * because it is not legal to add members to an existing MultiXactId.
591 : *
592 : * Caller is expected to have verified that the multixact does not come from
593 : * a pg_upgraded share-locked tuple.
594 : */
595 : bool
596 1968 : MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
597 : {
598 : MultiXactMember *members;
599 : int nmembers;
600 : int i;
601 :
602 : debug_elog3(DEBUG2, "IsRunning %u?", multi);
603 :
604 : /*
605 : * "false" here means we assume our callers have checked that the given
606 : * multi cannot possibly come from a pg_upgraded database.
607 : */
608 1968 : nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
609 :
610 1968 : if (nmembers <= 0)
611 : {
612 : debug_elog2(DEBUG2, "IsRunning: no members");
613 1338 : return false;
614 : }
615 :
616 : /*
617 : * Checking for myself is cheap compared to looking in shared memory;
618 : * return true if any live subtransaction of the current top-level
619 : * transaction is a member.
620 : *
621 : * This is not needed for correctness, it's just a fast path.
622 : */
623 1442 : for (i = 0; i < nmembers; i++)
624 : {
625 1124 : if (TransactionIdIsCurrentTransactionId(members[i].xid))
626 : {
627 : debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
628 312 : pfree(members);
629 312 : return true;
630 : }
631 : }
632 :
633 : /*
634 : * This could be made faster by having another entry point in procarray.c,
635 : * walking the PGPROC array only once for all the members. But in most
636 : * cases nmembers should be small enough that it doesn't much matter.
637 : */
638 570 : for (i = 0; i < nmembers; i++)
639 : {
640 478 : if (TransactionIdIsInProgress(members[i].xid))
641 : {
642 : debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
643 : i, members[i].xid);
644 226 : pfree(members);
645 226 : return true;
646 : }
647 : }
648 :
649 92 : pfree(members);
650 :
651 : debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
652 :
653 92 : return false;
654 : }
655 :
656 : /*
657 : * MultiXactIdSetOldestMember
658 : * Save the oldest MultiXactId this transaction could be a member of.
659 : *
660 : * We set the OldestMemberMXactId for a given transaction the first time it's
661 : * going to do some operation that might require a MultiXactId (tuple lock,
662 : * update or delete). We need to do this even if we end up using a
663 : * TransactionId instead of a MultiXactId, because there is a chance that
664 : * another transaction would add our XID to a MultiXactId.
665 : *
666 : * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
667 : * be called just before doing any such possibly-MultiXactId-able operation.
668 : */
669 : void
670 3766504 : MultiXactIdSetOldestMember(void)
671 : {
672 3766504 : if (!MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]))
673 : {
674 : MultiXactId nextMXact;
675 :
676 : /*
677 : * You might think we don't need to acquire a lock here, since
678 : * fetching and storing of TransactionIds is probably atomic, but in
679 : * fact we do: suppose we pick up nextMXact and then lose the CPU for
680 : * a long time. Someone else could advance nextMXact, and then
681 : * another someone else could compute an OldestVisibleMXactId that
682 : * would be after the value we are going to store when we get control
683 : * back. Which would be wrong.
684 : *
685 : * Note that a shared lock is sufficient, because it's enough to stop
686 : * someone from advancing nextMXact; and nobody else could be trying
687 : * to write to our OldestMember entry, only reading (and we assume
688 : * storing it is atomic.)
689 : */
690 141840 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
691 :
692 : /*
693 : * We have to beware of the possibility that nextMXact is in the
694 : * wrapped-around state. We don't fix the counter itself here, but we
695 : * must be sure to store a valid value in our array entry.
696 : */
697 141840 : nextMXact = MultiXactState->nextMXact;
698 141840 : if (nextMXact < FirstMultiXactId)
699 0 : nextMXact = FirstMultiXactId;
700 :
701 141840 : OldestMemberMXactId[MyProcNumber] = nextMXact;
702 :
703 141840 : LWLockRelease(MultiXactGenLock);
704 :
705 : debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
706 : MyProcNumber, nextMXact);
707 : }
708 3766504 : }
709 :
710 : /*
711 : * MultiXactIdSetOldestVisible
712 : * Save the oldest MultiXactId this transaction considers possibly live.
713 : *
714 : * We set the OldestVisibleMXactId for a given transaction the first time
715 : * it's going to inspect any MultiXactId. Once we have set this, we are
716 : * guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
717 : * won't be truncated away.
718 : *
719 : * The value to set is the oldest of nextMXact and all the valid per-backend
720 : * OldestMemberMXactId[] entries. Because of the locking we do, we can be
721 : * certain that no subsequent call to MultiXactIdSetOldestMember can set
722 : * an OldestMemberMXactId[] entry older than what we compute here. Therefore
723 : * there is no live transaction, now or later, that can be a member of any
724 : * MultiXactId older than the OldestVisibleMXactId we compute here.
725 : */
726 : static void
727 1804 : MultiXactIdSetOldestVisible(void)
728 : {
729 1804 : if (!MultiXactIdIsValid(OldestVisibleMXactId[MyProcNumber]))
730 : {
731 : MultiXactId oldestMXact;
732 : int i;
733 :
734 462 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
735 :
736 : /*
737 : * We have to beware of the possibility that nextMXact is in the
738 : * wrapped-around state. We don't fix the counter itself here, but we
739 : * must be sure to store a valid value in our array entry.
740 : */
741 462 : oldestMXact = MultiXactState->nextMXact;
742 462 : if (oldestMXact < FirstMultiXactId)
743 0 : oldestMXact = FirstMultiXactId;
744 :
745 62418 : for (i = 0; i < MaxOldestSlot; i++)
746 : {
747 61956 : MultiXactId thisoldest = OldestMemberMXactId[i];
748 :
749 62452 : if (MultiXactIdIsValid(thisoldest) &&
750 496 : MultiXactIdPrecedes(thisoldest, oldestMXact))
751 232 : oldestMXact = thisoldest;
752 : }
753 :
754 462 : OldestVisibleMXactId[MyProcNumber] = oldestMXact;
755 :
756 462 : LWLockRelease(MultiXactGenLock);
757 :
758 : debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
759 : MyProcNumber, oldestMXact);
760 : }
761 1804 : }
762 :
763 : /*
764 : * ReadNextMultiXactId
765 : * Return the next MultiXactId to be assigned, but don't allocate it
766 : */
767 : MultiXactId
768 393252 : ReadNextMultiXactId(void)
769 : {
770 : MultiXactId mxid;
771 :
772 : /* XXX we could presumably do this without a lock. */
773 393252 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
774 393252 : mxid = MultiXactState->nextMXact;
775 393252 : LWLockRelease(MultiXactGenLock);
776 :
777 393252 : if (mxid < FirstMultiXactId)
778 0 : mxid = FirstMultiXactId;
779 :
780 393252 : return mxid;
781 : }
782 :
783 : /*
784 : * ReadMultiXactIdRange
785 : * Get the range of IDs that may still be referenced by a relation.
786 : */
787 : void
788 2952 : ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
789 : {
790 2952 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
791 2952 : *oldest = MultiXactState->oldestMultiXactId;
792 2952 : *next = MultiXactState->nextMXact;
793 2952 : LWLockRelease(MultiXactGenLock);
794 :
795 2952 : if (*oldest < FirstMultiXactId)
796 0 : *oldest = FirstMultiXactId;
797 2952 : if (*next < FirstMultiXactId)
798 0 : *next = FirstMultiXactId;
799 2952 : }
800 :
801 :
802 : /*
803 : * MultiXactIdCreateFromMembers
804 : * Make a new MultiXactId from the specified set of members
805 : *
806 : * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
807 : * given TransactionIds as members. Returns the newly created MultiXactId.
808 : *
809 : * NB: the passed members[] array will be sorted in-place.
810 : */
811 : MultiXactId
812 2258 : MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
813 : {
814 : MultiXactId multi;
815 : MultiXactOffset offset;
816 : xl_multixact_create xlrec;
817 :
818 : debug_elog3(DEBUG2, "Create: %s",
819 : mxid_to_string(InvalidMultiXactId, nmembers, members));
820 :
821 : /*
822 : * See if the same set of members already exists in our cache; if so, just
823 : * re-use that MultiXactId. (Note: it might seem that looking in our
824 : * cache is insufficient, and we ought to search disk to see if a
825 : * duplicate definition already exists. But since we only ever create
826 : * MultiXacts containing our own XID, in most cases any such MultiXacts
827 : * were in fact created by us, and so will be in our cache. There are
828 : * corner cases where someone else added us to a MultiXact without our
829 : * knowledge, but it's not worth checking for.)
830 : */
831 2258 : multi = mXactCacheGetBySet(nmembers, members);
832 2258 : if (MultiXactIdIsValid(multi))
833 : {
834 : debug_elog2(DEBUG2, "Create: in cache!");
835 1704 : return multi;
836 : }
837 :
838 : /* Verify that there is a single update Xid among the given members. */
839 : {
840 : int i;
841 554 : bool has_update = false;
842 :
843 1778 : for (i = 0; i < nmembers; i++)
844 : {
845 1224 : if (ISUPDATE_from_mxstatus(members[i].status))
846 : {
847 264 : if (has_update)
848 0 : elog(ERROR, "new multixact has more than one updating member: %s",
849 : mxid_to_string(InvalidMultiXactId, nmembers, members));
850 264 : has_update = true;
851 : }
852 : }
853 : }
854 :
855 : /* Load the injection point before entering the critical section */
856 554 : INJECTION_POINT_LOAD("multixact-create-from-members");
857 :
858 : /*
859 : * Assign the MXID and offsets range to use, and make sure there is space
860 : * in the OFFSETs and MEMBERs files. NB: this routine does
861 : * START_CRIT_SECTION().
862 : *
863 : * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
864 : * that we've called MultiXactIdSetOldestMember here. This is because
865 : * this routine is used in some places to create new MultiXactIds of which
866 : * the current backend is not a member, notably during freezing of multis
867 : * in vacuum. During vacuum, in particular, it would be unacceptable to
868 : * keep OldestMulti set, in case it runs for long.
869 : */
870 554 : multi = GetNewMultiXactId(nmembers, &offset);
871 :
872 554 : INJECTION_POINT_CACHED("multixact-create-from-members", NULL);
873 :
874 : /* Make an XLOG entry describing the new MXID. */
875 554 : xlrec.mid = multi;
876 554 : xlrec.moff = offset;
877 554 : xlrec.nmembers = nmembers;
878 :
879 : /*
880 : * XXX Note: there's a lot of padding space in MultiXactMember. We could
881 : * find a more compact representation of this Xlog record -- perhaps all
882 : * the status flags in one XLogRecData, then all the xids in another one?
883 : * Not clear that it's worth the trouble though.
884 : */
885 554 : XLogBeginInsert();
886 554 : XLogRegisterData(&xlrec, SizeOfMultiXactCreate);
887 554 : XLogRegisterData(members, nmembers * sizeof(MultiXactMember));
888 :
889 554 : (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
890 :
891 : /* Now enter the information into the OFFSETs and MEMBERs logs */
892 554 : RecordNewMultiXact(multi, offset, nmembers, members);
893 :
894 : /* Done with critical section */
895 554 : END_CRIT_SECTION();
896 :
897 : /* Store the new MultiXactId in the local cache, too */
898 554 : mXactCachePut(multi, nmembers, members);
899 :
900 : debug_elog2(DEBUG2, "Create: all done");
901 :
902 554 : return multi;
903 : }
904 :
905 : /*
906 : * RecordNewMultiXact
907 : * Write info about a new multixact into the offsets and members files
908 : *
909 : * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
910 : * use it.
911 : */
912 : static void
913 558 : RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
914 : int nmembers, MultiXactMember *members)
915 : {
916 : int64 pageno;
917 : int64 prev_pageno;
918 : int entryno;
919 : int slotno;
920 : MultiXactOffset *offptr;
921 : int i;
922 : LWLock *lock;
923 558 : LWLock *prevlock = NULL;
924 :
925 558 : pageno = MultiXactIdToOffsetPage(multi);
926 558 : entryno = MultiXactIdToOffsetEntry(multi);
927 :
928 558 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
929 558 : LWLockAcquire(lock, LW_EXCLUSIVE);
930 :
931 : /*
932 : * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
933 : * to complain about if there's any I/O error. This is kinda bogus, but
934 : * since the errors will always give the full pathname, it should be clear
935 : * enough that a MultiXactId is really involved. Perhaps someday we'll
936 : * take the trouble to generalize the slru.c error reporting code.
937 : */
938 558 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
939 558 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
940 558 : offptr += entryno;
941 :
942 558 : *offptr = offset;
943 :
944 558 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
945 :
946 : /* Release MultiXactOffset SLRU lock. */
947 558 : LWLockRelease(lock);
948 :
949 : /*
950 : * If anybody was waiting to know the offset of this multixact ID we just
951 : * wrote, they can read it now, so wake them up.
952 : */
953 558 : ConditionVariableBroadcast(&MultiXactState->nextoff_cv);
954 :
955 558 : prev_pageno = -1;
956 :
957 1790 : for (i = 0; i < nmembers; i++, offset++)
958 : {
959 : TransactionId *memberptr;
960 : uint32 *flagsptr;
961 : uint32 flagsval;
962 : int bshift;
963 : int flagsoff;
964 : int memberoff;
965 :
966 : Assert(members[i].status <= MultiXactStatusUpdate);
967 :
968 1232 : pageno = MXOffsetToMemberPage(offset);
969 1232 : memberoff = MXOffsetToMemberOffset(offset);
970 1232 : flagsoff = MXOffsetToFlagsOffset(offset);
971 1232 : bshift = MXOffsetToFlagsBitShift(offset);
972 :
973 1232 : if (pageno != prev_pageno)
974 : {
975 : /*
976 : * MultiXactMember SLRU page is changed so check if this new page
977 : * fall into the different SLRU bank then release the old bank's
978 : * lock and acquire lock on the new bank.
979 : */
980 558 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
981 558 : if (lock != prevlock)
982 : {
983 558 : if (prevlock != NULL)
984 0 : LWLockRelease(prevlock);
985 :
986 558 : LWLockAcquire(lock, LW_EXCLUSIVE);
987 558 : prevlock = lock;
988 : }
989 558 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
990 558 : prev_pageno = pageno;
991 : }
992 :
993 1232 : memberptr = (TransactionId *)
994 1232 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
995 :
996 1232 : *memberptr = members[i].xid;
997 :
998 1232 : flagsptr = (uint32 *)
999 1232 : (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1000 :
1001 1232 : flagsval = *flagsptr;
1002 1232 : flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
1003 1232 : flagsval |= (members[i].status << bshift);
1004 1232 : *flagsptr = flagsval;
1005 :
1006 1232 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
1007 : }
1008 :
1009 558 : if (prevlock != NULL)
1010 558 : LWLockRelease(prevlock);
1011 558 : }
1012 :
1013 : /*
1014 : * GetNewMultiXactId
1015 : * Get the next MultiXactId.
1016 : *
1017 : * Also, reserve the needed amount of space in the "members" area. The
1018 : * starting offset of the reserved space is returned in *offset.
1019 : *
1020 : * This may generate XLOG records for expansion of the offsets and/or members
1021 : * files. Unfortunately, we have to do that while holding MultiXactGenLock
1022 : * to avoid race conditions --- the XLOG record for zeroing a page must appear
1023 : * before any backend can possibly try to store data in that page!
1024 : *
1025 : * We start a critical section before advancing the shared counters. The
1026 : * caller must end the critical section after writing SLRU data.
1027 : */
1028 : static MultiXactId
1029 554 : GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
1030 : {
1031 : MultiXactId result;
1032 : MultiXactOffset nextOffset;
1033 :
1034 : debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
1035 :
1036 : /* safety check, we should never get this far in a HS standby */
1037 554 : if (RecoveryInProgress())
1038 0 : elog(ERROR, "cannot assign MultiXactIds during recovery");
1039 :
1040 554 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1041 :
1042 : /* Handle wraparound of the nextMXact counter */
1043 554 : if (MultiXactState->nextMXact < FirstMultiXactId)
1044 0 : MultiXactState->nextMXact = FirstMultiXactId;
1045 :
1046 : /* Assign the MXID */
1047 554 : result = MultiXactState->nextMXact;
1048 :
1049 : /*----------
1050 : * Check to see if it's safe to assign another MultiXactId. This protects
1051 : * against catastrophic data loss due to multixact wraparound. The basic
1052 : * rules are:
1053 : *
1054 : * If we're past multiVacLimit or the safe threshold for member storage
1055 : * space, or we don't know what the safe threshold for member storage is,
1056 : * start trying to force autovacuum cycles.
1057 : * If we're past multiWarnLimit, start issuing warnings.
1058 : * If we're past multiStopLimit, refuse to create new MultiXactIds.
1059 : *
1060 : * Note these are pretty much the same protections in GetNewTransactionId.
1061 : *----------
1062 : */
1063 554 : if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
1064 : {
1065 : /*
1066 : * For safety's sake, we release MultiXactGenLock while sending
1067 : * signals, warnings, etc. This is not so much because we care about
1068 : * preserving concurrency in this situation, as to avoid any
1069 : * possibility of deadlock while doing get_database_name(). First,
1070 : * copy all the shared values we'll need in this path.
1071 : */
1072 0 : MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
1073 0 : MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
1074 0 : MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
1075 0 : Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
1076 :
1077 0 : LWLockRelease(MultiXactGenLock);
1078 :
1079 0 : if (IsUnderPostmaster &&
1080 0 : !MultiXactIdPrecedes(result, multiStopLimit))
1081 : {
1082 0 : char *oldest_datname = get_database_name(oldest_datoid);
1083 :
1084 : /*
1085 : * Immediately kick autovacuum into action as we're already in
1086 : * ERROR territory.
1087 : */
1088 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1089 :
1090 : /* complain even if that DB has disappeared */
1091 0 : if (oldest_datname)
1092 0 : ereport(ERROR,
1093 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1094 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database \"%s\"",
1095 : oldest_datname),
1096 : errhint("Execute a database-wide VACUUM in that database.\n"
1097 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1098 : else
1099 0 : ereport(ERROR,
1100 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1101 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database with OID %u",
1102 : oldest_datoid),
1103 : errhint("Execute a database-wide VACUUM in that database.\n"
1104 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1105 : }
1106 :
1107 : /*
1108 : * To avoid swamping the postmaster with signals, we issue the autovac
1109 : * request only once per 64K multis generated. This still gives
1110 : * plenty of chances before we get into real trouble.
1111 : */
1112 0 : if (IsUnderPostmaster && (result % 65536) == 0)
1113 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1114 :
1115 0 : if (!MultiXactIdPrecedes(result, multiWarnLimit))
1116 : {
1117 0 : char *oldest_datname = get_database_name(oldest_datoid);
1118 :
1119 : /* complain even if that DB has disappeared */
1120 0 : if (oldest_datname)
1121 0 : ereport(WARNING,
1122 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1123 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1124 : multiWrapLimit - result,
1125 : oldest_datname,
1126 : multiWrapLimit - result),
1127 : errhint("Execute a database-wide VACUUM in that database.\n"
1128 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1129 : else
1130 0 : ereport(WARNING,
1131 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1132 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1133 : multiWrapLimit - result,
1134 : oldest_datoid,
1135 : multiWrapLimit - result),
1136 : errhint("Execute a database-wide VACUUM in that database.\n"
1137 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1138 : }
1139 :
1140 : /* Re-acquire lock and start over */
1141 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1142 0 : result = MultiXactState->nextMXact;
1143 0 : if (result < FirstMultiXactId)
1144 0 : result = FirstMultiXactId;
1145 : }
1146 :
1147 : /* Make sure there is room for the MXID in the file. */
1148 554 : ExtendMultiXactOffset(result);
1149 :
1150 : /*
1151 : * Reserve the members space, similarly to above. Also, be careful not to
1152 : * return zero as the starting offset for any multixact. See
1153 : * GetMultiXactIdMembers() for motivation.
1154 : */
1155 554 : nextOffset = MultiXactState->nextOffset;
1156 554 : if (nextOffset == 0)
1157 : {
1158 22 : *offset = 1;
1159 22 : nmembers++; /* allocate member slot 0 too */
1160 : }
1161 : else
1162 532 : *offset = nextOffset;
1163 :
1164 : /*----------
1165 : * Protect against overrun of the members space as well, with the
1166 : * following rules:
1167 : *
1168 : * If we're past offsetStopLimit, refuse to generate more multis.
1169 : * If we're close to offsetStopLimit, emit a warning.
1170 : *
1171 : * Arbitrarily, we start emitting warnings when we're 20 segments or less
1172 : * from offsetStopLimit.
1173 : *
1174 : * Note we haven't updated the shared state yet, so if we fail at this
1175 : * point, the multixact ID we grabbed can still be used by the next guy.
1176 : *
1177 : * Note that there is no point in forcing autovacuum runs here: the
1178 : * multixact freeze settings would have to be reduced for that to have any
1179 : * effect.
1180 : *----------
1181 : */
1182 : #define OFFSET_WARN_SEGMENTS 20
1183 1108 : if (MultiXactState->oldestOffsetKnown &&
1184 554 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit, nextOffset,
1185 : nmembers))
1186 : {
1187 : /* see comment in the corresponding offsets wraparound case */
1188 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1189 :
1190 0 : ereport(ERROR,
1191 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1192 : errmsg("multixact \"members\" limit exceeded"),
1193 : errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
1194 : "This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
1195 : MultiXactState->offsetStopLimit - nextOffset - 1,
1196 : nmembers,
1197 : MultiXactState->offsetStopLimit - nextOffset - 1),
1198 : 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.",
1199 : MultiXactState->oldestMultiXactDB)));
1200 : }
1201 :
1202 : /*
1203 : * Check whether we should kick autovacuum into action, to prevent members
1204 : * wraparound. NB we use a much larger window to trigger autovacuum than
1205 : * just the warning limit. The warning is just a measure of last resort -
1206 : * this is in line with GetNewTransactionId's behaviour.
1207 : */
1208 554 : if (!MultiXactState->oldestOffsetKnown ||
1209 554 : (MultiXactState->nextOffset - MultiXactState->oldestOffset
1210 554 : > MULTIXACT_MEMBER_SAFE_THRESHOLD))
1211 : {
1212 : /*
1213 : * To avoid swamping the postmaster with signals, we issue the autovac
1214 : * request only when crossing a segment boundary. With default
1215 : * compilation settings that's roughly after 50k members. This still
1216 : * gives plenty of chances before we get into real trouble.
1217 : */
1218 0 : if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
1219 0 : (MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
1220 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1221 : }
1222 :
1223 1108 : if (MultiXactState->oldestOffsetKnown &&
1224 554 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit,
1225 : nextOffset,
1226 : nmembers + MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT * OFFSET_WARN_SEGMENTS))
1227 0 : ereport(WARNING,
1228 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1229 : errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
1230 : "database with OID %u must be vacuumed before %d more multixact members are used",
1231 : MultiXactState->offsetStopLimit - nextOffset + nmembers,
1232 : MultiXactState->oldestMultiXactDB,
1233 : MultiXactState->offsetStopLimit - nextOffset + nmembers),
1234 : errhint("Execute a database-wide VACUUM in that database with reduced \"vacuum_multixact_freeze_min_age\" and \"vacuum_multixact_freeze_table_age\" settings.")));
1235 :
1236 554 : ExtendMultiXactMember(nextOffset, nmembers);
1237 :
1238 : /*
1239 : * Critical section from here until caller has written the data into the
1240 : * just-reserved SLRU space; we don't want to error out with a partly
1241 : * written MultiXact structure. (In particular, failing to write our
1242 : * start offset after advancing nextMXact would effectively corrupt the
1243 : * previous MultiXact.)
1244 : */
1245 554 : START_CRIT_SECTION();
1246 :
1247 : /*
1248 : * Advance counters. As in GetNewTransactionId(), this must not happen
1249 : * until after file extension has succeeded!
1250 : *
1251 : * We don't care about MultiXactId wraparound here; it will be handled by
1252 : * the next iteration. But note that nextMXact may be InvalidMultiXactId
1253 : * or the first value on a segment-beginning page after this routine
1254 : * exits, so anyone else looking at the variable must be prepared to deal
1255 : * with either case. Similarly, nextOffset may be zero, but we won't use
1256 : * that as the actual start offset of the next multixact.
1257 : */
1258 554 : (MultiXactState->nextMXact)++;
1259 :
1260 554 : MultiXactState->nextOffset += nmembers;
1261 :
1262 554 : LWLockRelease(MultiXactGenLock);
1263 :
1264 : debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
1265 554 : return result;
1266 : }
1267 :
1268 : /*
1269 : * GetMultiXactIdMembers
1270 : * Return the set of MultiXactMembers that make up a MultiXactId
1271 : *
1272 : * Return value is the number of members found, or -1 if there are none,
1273 : * and *members is set to a newly palloc'ed array of members. It's the
1274 : * caller's responsibility to free it when done with it.
1275 : *
1276 : * from_pgupgrade must be passed as true if and only if only the multixact
1277 : * corresponds to a value from a tuple that was locked in a 9.2-or-older
1278 : * installation and later pg_upgrade'd (that is, the infomask is
1279 : * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1280 : * can still be running, so we return -1 just like for an empty multixact
1281 : * without any further checking. It would be wrong to try to resolve such a
1282 : * multixact: either the multixact is within the current valid multixact
1283 : * range, in which case the returned result would be bogus, or outside that
1284 : * range, in which case an error would be raised.
1285 : *
1286 : * In all other cases, the passed multixact must be within the known valid
1287 : * range, that is, greater than or equal to oldestMultiXactId, and less than
1288 : * nextMXact. Otherwise, an error is raised.
1289 : *
1290 : * isLockOnly must be set to true if caller is certain that the given multi
1291 : * is used only to lock tuples; can be false without loss of correctness,
1292 : * but passing a true means we can return quickly without checking for
1293 : * old updates.
1294 : */
1295 : int
1296 6010 : GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
1297 : bool from_pgupgrade, bool isLockOnly)
1298 : {
1299 : int64 pageno;
1300 : int64 prev_pageno;
1301 : int entryno;
1302 : int slotno;
1303 : MultiXactOffset *offptr;
1304 : MultiXactOffset offset;
1305 : int length;
1306 : int truelength;
1307 : MultiXactId oldestMXact;
1308 : MultiXactId nextMXact;
1309 : MultiXactId tmpMXact;
1310 : MultiXactOffset nextOffset;
1311 : MultiXactMember *ptr;
1312 : LWLock *lock;
1313 6010 : bool slept = false;
1314 :
1315 : debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1316 :
1317 6010 : if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1318 : {
1319 0 : *members = NULL;
1320 0 : return -1;
1321 : }
1322 :
1323 : /* See if the MultiXactId is in the local cache */
1324 6010 : length = mXactCacheGetById(multi, members);
1325 6010 : if (length >= 0)
1326 : {
1327 : debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1328 : mxid_to_string(multi, length, *members));
1329 4206 : return length;
1330 : }
1331 :
1332 : /* Set our OldestVisibleMXactId[] entry if we didn't already */
1333 1804 : MultiXactIdSetOldestVisible();
1334 :
1335 : /*
1336 : * If we know the multi is used only for locking and not for updates, then
1337 : * we can skip checking if the value is older than our oldest visible
1338 : * multi. It cannot possibly still be running.
1339 : */
1340 3304 : if (isLockOnly &&
1341 1500 : MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyProcNumber]))
1342 : {
1343 : debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1344 1340 : *members = NULL;
1345 1340 : return -1;
1346 : }
1347 :
1348 : /*
1349 : * We check known limits on MultiXact before resorting to the SLRU area.
1350 : *
1351 : * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1352 : * useful; it has already been removed, or will be removed shortly, by
1353 : * truncation. If one is passed, an error is raised.
1354 : *
1355 : * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1356 : * implies undetected ID wraparound has occurred. This raises a hard
1357 : * error.
1358 : *
1359 : * Shared lock is enough here since we aren't modifying any global state.
1360 : * Acquire it just long enough to grab the current counter values. We may
1361 : * need both nextMXact and nextOffset; see below.
1362 : */
1363 464 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1364 :
1365 464 : oldestMXact = MultiXactState->oldestMultiXactId;
1366 464 : nextMXact = MultiXactState->nextMXact;
1367 464 : nextOffset = MultiXactState->nextOffset;
1368 :
1369 464 : LWLockRelease(MultiXactGenLock);
1370 :
1371 464 : if (MultiXactIdPrecedes(multi, oldestMXact))
1372 0 : ereport(ERROR,
1373 : (errcode(ERRCODE_INTERNAL_ERROR),
1374 : errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1375 : multi)));
1376 :
1377 464 : if (!MultiXactIdPrecedes(multi, nextMXact))
1378 0 : ereport(ERROR,
1379 : (errcode(ERRCODE_INTERNAL_ERROR),
1380 : errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1381 : multi)));
1382 :
1383 : /*
1384 : * Find out the offset at which we need to start reading MultiXactMembers
1385 : * and the number of members in the multixact. We determine the latter as
1386 : * the difference between this multixact's starting offset and the next
1387 : * one's. However, there are some corner cases to worry about:
1388 : *
1389 : * 1. This multixact may be the latest one created, in which case there is
1390 : * no next one to look at. In this case the nextOffset value we just
1391 : * saved is the correct endpoint.
1392 : *
1393 : * 2. The next multixact may still be in process of being filled in: that
1394 : * is, another process may have done GetNewMultiXactId but not yet written
1395 : * the offset entry for that ID. In that scenario, it is guaranteed that
1396 : * the offset entry for that multixact exists (because GetNewMultiXactId
1397 : * won't release MultiXactGenLock until it does) but contains zero
1398 : * (because we are careful to pre-zero offset pages). Because
1399 : * GetNewMultiXactId will never return zero as the starting offset for a
1400 : * multixact, when we read zero as the next multixact's offset, we know we
1401 : * have this case. We handle this by sleeping on the condition variable
1402 : * we have just for this; the process in charge will signal the CV as soon
1403 : * as it has finished writing the multixact offset.
1404 : *
1405 : * 3. Because GetNewMultiXactId increments offset zero to offset one to
1406 : * handle case #2, there is an ambiguity near the point of offset
1407 : * wraparound. If we see next multixact's offset is one, is that our
1408 : * multixact's actual endpoint, or did it end at zero with a subsequent
1409 : * increment? We handle this using the knowledge that if the zero'th
1410 : * member slot wasn't filled, it'll contain zero, and zero isn't a valid
1411 : * transaction ID so it can't be a multixact member. Therefore, if we
1412 : * read a zero from the members array, just ignore it.
1413 : *
1414 : * This is all pretty messy, but the mess occurs only in infrequent corner
1415 : * cases, so it seems better than holding the MultiXactGenLock for a long
1416 : * time on every multixact creation.
1417 : */
1418 464 : retry:
1419 466 : pageno = MultiXactIdToOffsetPage(multi);
1420 466 : entryno = MultiXactIdToOffsetEntry(multi);
1421 :
1422 : /* Acquire the bank lock for the page we need. */
1423 466 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1424 466 : LWLockAcquire(lock, LW_EXCLUSIVE);
1425 :
1426 466 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1427 466 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1428 466 : offptr += entryno;
1429 466 : offset = *offptr;
1430 :
1431 : Assert(offset != 0);
1432 :
1433 : /*
1434 : * Use the same increment rule as GetNewMultiXactId(), that is, don't
1435 : * handle wraparound explicitly until needed.
1436 : */
1437 466 : tmpMXact = multi + 1;
1438 :
1439 466 : if (nextMXact == tmpMXact)
1440 : {
1441 : /* Corner case 1: there is no next multixact */
1442 408 : length = nextOffset - offset;
1443 : }
1444 : else
1445 : {
1446 : MultiXactOffset nextMXOffset;
1447 :
1448 : /* handle wraparound if needed */
1449 58 : if (tmpMXact < FirstMultiXactId)
1450 0 : tmpMXact = FirstMultiXactId;
1451 :
1452 58 : prev_pageno = pageno;
1453 :
1454 58 : pageno = MultiXactIdToOffsetPage(tmpMXact);
1455 58 : entryno = MultiXactIdToOffsetEntry(tmpMXact);
1456 :
1457 58 : if (pageno != prev_pageno)
1458 : {
1459 : LWLock *newlock;
1460 :
1461 : /*
1462 : * Since we're going to access a different SLRU page, if this page
1463 : * falls under a different bank, release the old bank's lock and
1464 : * acquire the lock of the new bank.
1465 : */
1466 0 : newlock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1467 0 : if (newlock != lock)
1468 : {
1469 0 : LWLockRelease(lock);
1470 0 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1471 0 : lock = newlock;
1472 : }
1473 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1474 : }
1475 :
1476 58 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1477 58 : offptr += entryno;
1478 58 : nextMXOffset = *offptr;
1479 :
1480 58 : if (nextMXOffset == 0)
1481 : {
1482 : /* Corner case 2: next multixact is still being filled in */
1483 2 : LWLockRelease(lock);
1484 2 : CHECK_FOR_INTERRUPTS();
1485 :
1486 2 : INJECTION_POINT("multixact-get-members-cv-sleep", NULL);
1487 :
1488 2 : ConditionVariableSleep(&MultiXactState->nextoff_cv,
1489 : WAIT_EVENT_MULTIXACT_CREATION);
1490 2 : slept = true;
1491 2 : goto retry;
1492 : }
1493 :
1494 56 : length = nextMXOffset - offset;
1495 : }
1496 :
1497 464 : LWLockRelease(lock);
1498 464 : lock = NULL;
1499 :
1500 : /*
1501 : * If we slept above, clean up state; it's no longer needed.
1502 : */
1503 464 : if (slept)
1504 2 : ConditionVariableCancelSleep();
1505 :
1506 464 : ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1507 :
1508 464 : truelength = 0;
1509 464 : prev_pageno = -1;
1510 1652 : for (int i = 0; i < length; i++, offset++)
1511 : {
1512 : TransactionId *xactptr;
1513 : uint32 *flagsptr;
1514 : int flagsoff;
1515 : int bshift;
1516 : int memberoff;
1517 :
1518 1188 : pageno = MXOffsetToMemberPage(offset);
1519 1188 : memberoff = MXOffsetToMemberOffset(offset);
1520 :
1521 1188 : if (pageno != prev_pageno)
1522 : {
1523 : LWLock *newlock;
1524 :
1525 : /*
1526 : * Since we're going to access a different SLRU page, if this page
1527 : * falls under a different bank, release the old bank's lock and
1528 : * acquire the lock of the new bank.
1529 : */
1530 464 : newlock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
1531 464 : if (newlock != lock)
1532 : {
1533 464 : if (lock)
1534 0 : LWLockRelease(lock);
1535 464 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1536 464 : lock = newlock;
1537 : }
1538 :
1539 464 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1540 464 : prev_pageno = pageno;
1541 : }
1542 :
1543 1188 : xactptr = (TransactionId *)
1544 1188 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1545 :
1546 1188 : if (!TransactionIdIsValid(*xactptr))
1547 : {
1548 : /* Corner case 3: we must be looking at unused slot zero */
1549 : Assert(offset == 0);
1550 0 : continue;
1551 : }
1552 :
1553 1188 : flagsoff = MXOffsetToFlagsOffset(offset);
1554 1188 : bshift = MXOffsetToFlagsBitShift(offset);
1555 1188 : flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1556 :
1557 1188 : ptr[truelength].xid = *xactptr;
1558 1188 : ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1559 1188 : truelength++;
1560 : }
1561 :
1562 464 : LWLockRelease(lock);
1563 :
1564 : /* A multixid with zero members should not happen */
1565 : Assert(truelength > 0);
1566 :
1567 : /*
1568 : * Copy the result into the local cache.
1569 : */
1570 464 : mXactCachePut(multi, truelength, ptr);
1571 :
1572 : debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1573 : mxid_to_string(multi, truelength, ptr));
1574 464 : *members = ptr;
1575 464 : return truelength;
1576 : }
1577 :
1578 : /*
1579 : * mxactMemberComparator
1580 : * qsort comparison function for MultiXactMember
1581 : *
1582 : * We can't use wraparound comparison for XIDs because that does not respect
1583 : * the triangle inequality! Any old sort order will do.
1584 : */
1585 : static int
1586 3800 : mxactMemberComparator(const void *arg1, const void *arg2)
1587 : {
1588 3800 : MultiXactMember member1 = *(const MultiXactMember *) arg1;
1589 3800 : MultiXactMember member2 = *(const MultiXactMember *) arg2;
1590 :
1591 3800 : if (member1.xid > member2.xid)
1592 54 : return 1;
1593 3746 : if (member1.xid < member2.xid)
1594 3400 : return -1;
1595 346 : if (member1.status > member2.status)
1596 4 : return 1;
1597 342 : if (member1.status < member2.status)
1598 342 : return -1;
1599 0 : return 0;
1600 : }
1601 :
1602 : /*
1603 : * mXactCacheGetBySet
1604 : * returns a MultiXactId from the cache based on the set of
1605 : * TransactionIds that compose it, or InvalidMultiXactId if
1606 : * none matches.
1607 : *
1608 : * This is helpful, for example, if two transactions want to lock a huge
1609 : * table. By using the cache, the second will use the same MultiXactId
1610 : * for the majority of tuples, thus keeping MultiXactId usage low (saving
1611 : * both I/O and wraparound issues).
1612 : *
1613 : * NB: the passed members array will be sorted in-place.
1614 : */
1615 : static MultiXactId
1616 2258 : mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1617 : {
1618 : dlist_iter iter;
1619 :
1620 : debug_elog3(DEBUG2, "CacheGet: looking for %s",
1621 : mxid_to_string(InvalidMultiXactId, nmembers, members));
1622 :
1623 : /* sort the array so comparison is easy */
1624 2258 : qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1625 :
1626 2734 : dclist_foreach(iter, &MXactCache)
1627 : {
1628 2180 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1629 : iter.cur);
1630 :
1631 2180 : if (entry->nmembers != nmembers)
1632 236 : continue;
1633 :
1634 : /*
1635 : * We assume the cache entries are sorted, and that the unused bits in
1636 : * "status" are zeroed.
1637 : */
1638 1944 : if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1639 : {
1640 : debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1641 1704 : dclist_move_head(&MXactCache, iter.cur);
1642 1704 : return entry->multi;
1643 : }
1644 : }
1645 :
1646 : debug_elog2(DEBUG2, "CacheGet: not found :-(");
1647 554 : return InvalidMultiXactId;
1648 : }
1649 :
1650 : /*
1651 : * mXactCacheGetById
1652 : * returns the composing MultiXactMember set from the cache for a
1653 : * given MultiXactId, if present.
1654 : *
1655 : * If successful, *xids is set to the address of a palloc'd copy of the
1656 : * MultiXactMember set. Return value is number of members, or -1 on failure.
1657 : */
1658 : static int
1659 6010 : mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
1660 : {
1661 : dlist_iter iter;
1662 :
1663 : debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1664 :
1665 6892 : dclist_foreach(iter, &MXactCache)
1666 : {
1667 5088 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1668 : iter.cur);
1669 :
1670 5088 : if (entry->multi == multi)
1671 : {
1672 : MultiXactMember *ptr;
1673 : Size size;
1674 :
1675 4206 : size = sizeof(MultiXactMember) * entry->nmembers;
1676 4206 : ptr = (MultiXactMember *) palloc(size);
1677 :
1678 4206 : memcpy(ptr, entry->members, size);
1679 :
1680 : debug_elog3(DEBUG2, "CacheGet: found %s",
1681 : mxid_to_string(multi,
1682 : entry->nmembers,
1683 : entry->members));
1684 :
1685 : /*
1686 : * Note we modify the list while not using a modifiable iterator.
1687 : * This is acceptable only because we exit the iteration
1688 : * immediately afterwards.
1689 : */
1690 4206 : dclist_move_head(&MXactCache, iter.cur);
1691 :
1692 4206 : *members = ptr;
1693 4206 : return entry->nmembers;
1694 : }
1695 : }
1696 :
1697 : debug_elog2(DEBUG2, "CacheGet: not found");
1698 1804 : return -1;
1699 : }
1700 :
1701 : /*
1702 : * mXactCachePut
1703 : * Add a new MultiXactId and its composing set into the local cache.
1704 : */
1705 : static void
1706 1018 : mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1707 : {
1708 : mXactCacheEnt *entry;
1709 :
1710 : debug_elog3(DEBUG2, "CachePut: storing %s",
1711 : mxid_to_string(multi, nmembers, members));
1712 :
1713 1018 : if (MXactContext == NULL)
1714 : {
1715 : /* The cache only lives as long as the current transaction */
1716 : debug_elog2(DEBUG2, "CachePut: initializing memory context");
1717 728 : MXactContext = AllocSetContextCreate(TopTransactionContext,
1718 : "MultiXact cache context",
1719 : ALLOCSET_SMALL_SIZES);
1720 : }
1721 :
1722 : entry = (mXactCacheEnt *)
1723 1018 : MemoryContextAlloc(MXactContext,
1724 1018 : offsetof(mXactCacheEnt, members) +
1725 : nmembers * sizeof(MultiXactMember));
1726 :
1727 1018 : entry->multi = multi;
1728 1018 : entry->nmembers = nmembers;
1729 1018 : memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1730 :
1731 : /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1732 1018 : qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1733 :
1734 1018 : dclist_push_head(&MXactCache, &entry->node);
1735 1018 : if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
1736 : {
1737 : dlist_node *node;
1738 :
1739 0 : node = dclist_tail_node(&MXactCache);
1740 0 : dclist_delete_from(&MXactCache, node);
1741 :
1742 0 : entry = dclist_container(mXactCacheEnt, node, node);
1743 : debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1744 : entry->multi);
1745 :
1746 0 : pfree(entry);
1747 : }
1748 1018 : }
1749 :
1750 : static char *
1751 0 : mxstatus_to_string(MultiXactStatus status)
1752 : {
1753 0 : switch (status)
1754 : {
1755 0 : case MultiXactStatusForKeyShare:
1756 0 : return "keysh";
1757 0 : case MultiXactStatusForShare:
1758 0 : return "sh";
1759 0 : case MultiXactStatusForNoKeyUpdate:
1760 0 : return "fornokeyupd";
1761 0 : case MultiXactStatusForUpdate:
1762 0 : return "forupd";
1763 0 : case MultiXactStatusNoKeyUpdate:
1764 0 : return "nokeyupd";
1765 0 : case MultiXactStatusUpdate:
1766 0 : return "upd";
1767 0 : default:
1768 0 : elog(ERROR, "unrecognized multixact status %d", status);
1769 : return "";
1770 : }
1771 : }
1772 :
1773 : char *
1774 0 : mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1775 : {
1776 : static char *str = NULL;
1777 : StringInfoData buf;
1778 : int i;
1779 :
1780 0 : if (str != NULL)
1781 0 : pfree(str);
1782 :
1783 0 : initStringInfo(&buf);
1784 :
1785 0 : appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1786 : mxstatus_to_string(members[0].status));
1787 :
1788 0 : for (i = 1; i < nmembers; i++)
1789 0 : appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1790 0 : mxstatus_to_string(members[i].status));
1791 :
1792 0 : appendStringInfoChar(&buf, ']');
1793 0 : str = MemoryContextStrdup(TopMemoryContext, buf.data);
1794 0 : pfree(buf.data);
1795 0 : return str;
1796 : }
1797 :
1798 : /*
1799 : * AtEOXact_MultiXact
1800 : * Handle transaction end for MultiXact
1801 : *
1802 : * This is called at top transaction commit or abort (we don't care which).
1803 : */
1804 : void
1805 1056212 : AtEOXact_MultiXact(void)
1806 : {
1807 : /*
1808 : * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1809 : * which should only be valid while within a transaction.
1810 : *
1811 : * We assume that storing a MultiXactId is atomic and so we need not take
1812 : * MultiXactGenLock to do this.
1813 : */
1814 1056212 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1815 1056212 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1816 :
1817 : /*
1818 : * Discard the local MultiXactId cache. Since MXactContext was created as
1819 : * a child of TopTransactionContext, we needn't delete it explicitly.
1820 : */
1821 1056212 : MXactContext = NULL;
1822 1056212 : dclist_init(&MXactCache);
1823 1056212 : }
1824 :
1825 : /*
1826 : * AtPrepare_MultiXact
1827 : * Save multixact state at 2PC transaction prepare
1828 : *
1829 : * In this phase, we only store our OldestMemberMXactId value in the two-phase
1830 : * state file.
1831 : */
1832 : void
1833 570 : AtPrepare_MultiXact(void)
1834 : {
1835 570 : MultiXactId myOldestMember = OldestMemberMXactId[MyProcNumber];
1836 :
1837 570 : if (MultiXactIdIsValid(myOldestMember))
1838 98 : RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
1839 : &myOldestMember, sizeof(MultiXactId));
1840 570 : }
1841 :
1842 : /*
1843 : * PostPrepare_MultiXact
1844 : * Clean up after successful PREPARE TRANSACTION
1845 : */
1846 : void
1847 570 : PostPrepare_MultiXact(FullTransactionId fxid)
1848 : {
1849 : MultiXactId myOldestMember;
1850 :
1851 : /*
1852 : * Transfer our OldestMemberMXactId value to the slot reserved for the
1853 : * prepared transaction.
1854 : */
1855 570 : myOldestMember = OldestMemberMXactId[MyProcNumber];
1856 570 : if (MultiXactIdIsValid(myOldestMember))
1857 : {
1858 98 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1859 :
1860 : /*
1861 : * Even though storing MultiXactId is atomic, acquire lock to make
1862 : * sure others see both changes, not just the reset of the slot of the
1863 : * current backend. Using a volatile pointer might suffice, but this
1864 : * isn't a hot spot.
1865 : */
1866 98 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1867 :
1868 98 : OldestMemberMXactId[dummyProcNumber] = myOldestMember;
1869 98 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1870 :
1871 98 : LWLockRelease(MultiXactGenLock);
1872 : }
1873 :
1874 : /*
1875 : * We don't need to transfer OldestVisibleMXactId value, because the
1876 : * transaction is not going to be looking at any more multixacts once it's
1877 : * prepared.
1878 : *
1879 : * We assume that storing a MultiXactId is atomic and so we need not take
1880 : * MultiXactGenLock to do this.
1881 : */
1882 570 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1883 :
1884 : /*
1885 : * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1886 : */
1887 570 : MXactContext = NULL;
1888 570 : dclist_init(&MXactCache);
1889 570 : }
1890 :
1891 : /*
1892 : * multixact_twophase_recover
1893 : * Recover the state of a prepared transaction at startup
1894 : */
1895 : void
1896 16 : multixact_twophase_recover(FullTransactionId fxid, uint16 info,
1897 : void *recdata, uint32 len)
1898 : {
1899 16 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1900 : MultiXactId oldestMember;
1901 :
1902 : /*
1903 : * Get the oldest member XID from the state file record, and set it in the
1904 : * OldestMemberMXactId slot reserved for this prepared transaction.
1905 : */
1906 : Assert(len == sizeof(MultiXactId));
1907 16 : oldestMember = *((MultiXactId *) recdata);
1908 :
1909 16 : OldestMemberMXactId[dummyProcNumber] = oldestMember;
1910 16 : }
1911 :
1912 : /*
1913 : * multixact_twophase_postcommit
1914 : * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1915 : */
1916 : void
1917 106 : multixact_twophase_postcommit(FullTransactionId fxid, uint16 info,
1918 : void *recdata, uint32 len)
1919 : {
1920 106 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, true);
1921 :
1922 : Assert(len == sizeof(MultiXactId));
1923 :
1924 106 : OldestMemberMXactId[dummyProcNumber] = InvalidMultiXactId;
1925 106 : }
1926 :
1927 : /*
1928 : * multixact_twophase_postabort
1929 : * This is actually just the same as the COMMIT case.
1930 : */
1931 : void
1932 44 : multixact_twophase_postabort(FullTransactionId fxid, uint16 info,
1933 : void *recdata, uint32 len)
1934 : {
1935 44 : multixact_twophase_postcommit(fxid, info, recdata, len);
1936 44 : }
1937 :
1938 : /*
1939 : * Initialization of shared memory for MultiXact. We use two SLRU areas,
1940 : * thus double memory. Also, reserve space for the shared MultiXactState
1941 : * struct and the per-backend MultiXactId arrays (two of those, too).
1942 : */
1943 : Size
1944 3998 : MultiXactShmemSize(void)
1945 : {
1946 : Size size;
1947 :
1948 : /* We need 2*MaxOldestSlot perBackendXactIds[] entries */
1949 : #define SHARED_MULTIXACT_STATE_SIZE \
1950 : add_size(offsetof(MultiXactStateData, perBackendXactIds), \
1951 : mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1952 :
1953 3998 : size = SHARED_MULTIXACT_STATE_SIZE;
1954 3998 : size = add_size(size, SimpleLruShmemSize(multixact_offset_buffers, 0));
1955 3998 : size = add_size(size, SimpleLruShmemSize(multixact_member_buffers, 0));
1956 :
1957 3998 : return size;
1958 : }
1959 :
1960 : void
1961 2152 : MultiXactShmemInit(void)
1962 : {
1963 : bool found;
1964 :
1965 : debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1966 :
1967 2152 : MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
1968 2152 : MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
1969 :
1970 2152 : SimpleLruInit(MultiXactOffsetCtl,
1971 : "multixact_offset", multixact_offset_buffers, 0,
1972 : "pg_multixact/offsets", LWTRANCHE_MULTIXACTOFFSET_BUFFER,
1973 : LWTRANCHE_MULTIXACTOFFSET_SLRU,
1974 : SYNC_HANDLER_MULTIXACT_OFFSET,
1975 : false);
1976 : SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
1977 2152 : SimpleLruInit(MultiXactMemberCtl,
1978 : "multixact_member", multixact_member_buffers, 0,
1979 : "pg_multixact/members", LWTRANCHE_MULTIXACTMEMBER_BUFFER,
1980 : LWTRANCHE_MULTIXACTMEMBER_SLRU,
1981 : SYNC_HANDLER_MULTIXACT_MEMBER,
1982 : false);
1983 : /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1984 :
1985 : /* Initialize our shared state struct */
1986 2152 : MultiXactState = ShmemInitStruct("Shared MultiXact State",
1987 2152 : SHARED_MULTIXACT_STATE_SIZE,
1988 : &found);
1989 2152 : if (!IsUnderPostmaster)
1990 : {
1991 : Assert(!found);
1992 :
1993 : /* Make sure we zero out the per-backend state */
1994 2152 : MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
1995 2152 : ConditionVariableInit(&MultiXactState->nextoff_cv);
1996 : }
1997 : else
1998 : Assert(found);
1999 :
2000 : /*
2001 : * Set up array pointers.
2002 : */
2003 2152 : OldestMemberMXactId = MultiXactState->perBackendXactIds;
2004 2152 : OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
2005 2152 : }
2006 :
2007 : /*
2008 : * GUC check_hook for multixact_offset_buffers
2009 : */
2010 : bool
2011 2228 : check_multixact_offset_buffers(int *newval, void **extra, GucSource source)
2012 : {
2013 2228 : return check_slru_buffers("multixact_offset_buffers", newval);
2014 : }
2015 :
2016 : /*
2017 : * GUC check_hook for multixact_member_buffers
2018 : */
2019 : bool
2020 2228 : check_multixact_member_buffers(int *newval, void **extra, GucSource source)
2021 : {
2022 2228 : return check_slru_buffers("multixact_member_buffers", newval);
2023 : }
2024 :
2025 : /*
2026 : * This func must be called ONCE on system install. It creates the initial
2027 : * MultiXact segments. (The MultiXacts directories are assumed to have been
2028 : * created by initdb, and MultiXactShmemInit must have been called already.)
2029 : */
2030 : void
2031 102 : BootStrapMultiXact(void)
2032 : {
2033 : /* Zero the initial pages and flush them to disk */
2034 102 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, 0);
2035 102 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, 0);
2036 102 : }
2037 :
2038 : /*
2039 : * MaybeExtendOffsetSlru
2040 : * Extend the offsets SLRU area, if necessary
2041 : *
2042 : * After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
2043 : * contain files that are shorter than necessary; this would occur if the old
2044 : * installation had used multixacts beyond the first page (files cannot be
2045 : * copied, because the on-disk representation is different). pg_upgrade would
2046 : * update pg_control to set the next offset value to be at that position, so
2047 : * that tuples marked as locked by such MultiXacts would be seen as visible
2048 : * without having to consult multixact. However, trying to create and use a
2049 : * new MultiXactId would result in an error because the page on which the new
2050 : * value would reside does not exist. This routine is in charge of creating
2051 : * such pages.
2052 : */
2053 : static void
2054 94 : MaybeExtendOffsetSlru(void)
2055 : {
2056 : int64 pageno;
2057 : LWLock *lock;
2058 :
2059 94 : pageno = MultiXactIdToOffsetPage(MultiXactState->nextMXact);
2060 94 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2061 :
2062 94 : LWLockAcquire(lock, LW_EXCLUSIVE);
2063 :
2064 94 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2065 : {
2066 : int slotno;
2067 :
2068 : /*
2069 : * Fortunately for us, SimpleLruWritePage is already prepared to deal
2070 : * with creating a new segment file even if the page we're writing is
2071 : * not the first in it, so this is enough.
2072 : */
2073 0 : slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2074 0 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
2075 : }
2076 :
2077 94 : LWLockRelease(lock);
2078 94 : }
2079 :
2080 : /*
2081 : * This must be called ONCE during postmaster or standalone-backend startup.
2082 : *
2083 : * StartupXLOG has already established nextMXact/nextOffset by calling
2084 : * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
2085 : * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
2086 : * replayed WAL.
2087 : */
2088 : void
2089 1862 : StartupMultiXact(void)
2090 : {
2091 1862 : MultiXactId multi = MultiXactState->nextMXact;
2092 1862 : MultiXactOffset offset = MultiXactState->nextOffset;
2093 : int64 pageno;
2094 :
2095 : /*
2096 : * Initialize offset's idea of the latest page number.
2097 : */
2098 1862 : pageno = MultiXactIdToOffsetPage(multi);
2099 1862 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2100 : pageno);
2101 :
2102 : /*
2103 : * Initialize member's idea of the latest page number.
2104 : */
2105 1862 : pageno = MXOffsetToMemberPage(offset);
2106 1862 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2107 : pageno);
2108 1862 : }
2109 :
2110 : /*
2111 : * This must be called ONCE at the end of startup/recovery.
2112 : */
2113 : void
2114 1746 : TrimMultiXact(void)
2115 : {
2116 : MultiXactId nextMXact;
2117 : MultiXactOffset offset;
2118 : MultiXactId oldestMXact;
2119 : Oid oldestMXactDB;
2120 : int64 pageno;
2121 : int entryno;
2122 : int flagsoff;
2123 :
2124 1746 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2125 1746 : nextMXact = MultiXactState->nextMXact;
2126 1746 : offset = MultiXactState->nextOffset;
2127 1746 : oldestMXact = MultiXactState->oldestMultiXactId;
2128 1746 : oldestMXactDB = MultiXactState->oldestMultiXactDB;
2129 1746 : LWLockRelease(MultiXactGenLock);
2130 :
2131 : /* Clean up offsets state */
2132 :
2133 : /*
2134 : * (Re-)Initialize our idea of the latest page number for offsets.
2135 : */
2136 1746 : pageno = MultiXactIdToOffsetPage(nextMXact);
2137 1746 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2138 : pageno);
2139 :
2140 : /*
2141 : * Zero out the remainder of the current offsets page. See notes in
2142 : * TrimCLOG() for background. Unlike CLOG, some WAL record covers every
2143 : * pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
2144 : * rule "write xlog before data," nextMXact successors may carry obsolete,
2145 : * nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
2146 : * operates normally.
2147 : */
2148 1746 : entryno = MultiXactIdToOffsetEntry(nextMXact);
2149 1746 : if (entryno != 0)
2150 : {
2151 : int slotno;
2152 : MultiXactOffset *offptr;
2153 1744 : LWLock *lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2154 :
2155 1744 : LWLockAcquire(lock, LW_EXCLUSIVE);
2156 1744 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
2157 1744 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2158 1744 : offptr += entryno;
2159 :
2160 1744 : MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
2161 :
2162 1744 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
2163 1744 : LWLockRelease(lock);
2164 : }
2165 :
2166 : /*
2167 : * And the same for members.
2168 : *
2169 : * (Re-)Initialize our idea of the latest page number for members.
2170 : */
2171 1746 : pageno = MXOffsetToMemberPage(offset);
2172 1746 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2173 : pageno);
2174 :
2175 : /*
2176 : * Zero out the remainder of the current members page. See notes in
2177 : * TrimCLOG() for motivation.
2178 : */
2179 1746 : flagsoff = MXOffsetToFlagsOffset(offset);
2180 1746 : if (flagsoff != 0)
2181 : {
2182 : int slotno;
2183 : TransactionId *xidptr;
2184 : int memberoff;
2185 22 : LWLock *lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2186 :
2187 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2188 22 : memberoff = MXOffsetToMemberOffset(offset);
2189 22 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
2190 22 : xidptr = (TransactionId *)
2191 22 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
2192 :
2193 22 : MemSet(xidptr, 0, BLCKSZ - memberoff);
2194 :
2195 : /*
2196 : * Note: we don't need to zero out the flag bits in the remaining
2197 : * members of the current group, because they are always reset before
2198 : * writing.
2199 : */
2200 :
2201 22 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
2202 22 : LWLockRelease(lock);
2203 : }
2204 :
2205 : /* signal that we're officially up */
2206 1746 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2207 1746 : MultiXactState->finishedStartup = true;
2208 1746 : LWLockRelease(MultiXactGenLock);
2209 :
2210 : /* Now compute how far away the next members wraparound is. */
2211 1746 : SetMultiXactIdLimit(oldestMXact, oldestMXactDB, true);
2212 1746 : }
2213 :
2214 : /*
2215 : * Get the MultiXact data to save in a checkpoint record
2216 : */
2217 : void
2218 3002 : MultiXactGetCheckptMulti(bool is_shutdown,
2219 : MultiXactId *nextMulti,
2220 : MultiXactOffset *nextMultiOffset,
2221 : MultiXactId *oldestMulti,
2222 : Oid *oldestMultiDB)
2223 : {
2224 3002 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2225 3002 : *nextMulti = MultiXactState->nextMXact;
2226 3002 : *nextMultiOffset = MultiXactState->nextOffset;
2227 3002 : *oldestMulti = MultiXactState->oldestMultiXactId;
2228 3002 : *oldestMultiDB = MultiXactState->oldestMultiXactDB;
2229 3002 : LWLockRelease(MultiXactGenLock);
2230 :
2231 : debug_elog6(DEBUG2,
2232 : "MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
2233 : *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
2234 3002 : }
2235 :
2236 : /*
2237 : * Perform a checkpoint --- either during shutdown, or on-the-fly
2238 : */
2239 : void
2240 3380 : CheckPointMultiXact(void)
2241 : {
2242 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
2243 :
2244 : /*
2245 : * Write dirty MultiXact pages to disk. This may result in sync requests
2246 : * queued for later handling by ProcessSyncRequests(), as part of the
2247 : * checkpoint.
2248 : */
2249 3380 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2250 3380 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2251 :
2252 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
2253 3380 : }
2254 :
2255 : /*
2256 : * Set the next-to-be-assigned MultiXactId and offset
2257 : *
2258 : * This is used when we can determine the correct next ID/offset exactly
2259 : * from a checkpoint record. Although this is only called during bootstrap
2260 : * and XLog replay, we take the lock in case any hot-standby backends are
2261 : * examining the values.
2262 : */
2263 : void
2264 2036 : MultiXactSetNextMXact(MultiXactId nextMulti,
2265 : MultiXactOffset nextMultiOffset)
2266 : {
2267 : debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
2268 : nextMulti, nextMultiOffset);
2269 2036 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2270 2036 : MultiXactState->nextMXact = nextMulti;
2271 2036 : MultiXactState->nextOffset = nextMultiOffset;
2272 2036 : LWLockRelease(MultiXactGenLock);
2273 :
2274 : /*
2275 : * During a binary upgrade, make sure that the offsets SLRU is large
2276 : * enough to contain the next value that would be created.
2277 : *
2278 : * We need to do this pretty early during the first startup in binary
2279 : * upgrade mode: before StartupMultiXact() in fact, because this routine
2280 : * is called even before that by StartupXLOG(). And we can't do it
2281 : * earlier than at this point, because during that first call of this
2282 : * routine we determine the MultiXactState->nextMXact value that
2283 : * MaybeExtendOffsetSlru needs.
2284 : */
2285 2036 : if (IsBinaryUpgrade)
2286 94 : MaybeExtendOffsetSlru();
2287 2036 : }
2288 :
2289 : /*
2290 : * Determine the last safe MultiXactId to allocate given the currently oldest
2291 : * datminmxid (ie, the oldest MultiXactId that might exist in any database
2292 : * of our cluster), and the OID of the (or a) database with that value.
2293 : *
2294 : * is_startup is true when we are just starting the cluster, false when we
2295 : * are updating state in a running cluster. This only affects log messages.
2296 : */
2297 : void
2298 5934 : SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
2299 : bool is_startup)
2300 : {
2301 : MultiXactId multiVacLimit;
2302 : MultiXactId multiWarnLimit;
2303 : MultiXactId multiStopLimit;
2304 : MultiXactId multiWrapLimit;
2305 : MultiXactId curMulti;
2306 : bool needs_offset_vacuum;
2307 :
2308 : Assert(MultiXactIdIsValid(oldest_datminmxid));
2309 :
2310 : /*
2311 : * We pretend that a wrap will happen halfway through the multixact ID
2312 : * space, but that's not really true, because multixacts wrap differently
2313 : * from transaction IDs. Note that, separately from any concern about
2314 : * multixact IDs wrapping, we must ensure that multixact members do not
2315 : * wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
2316 : */
2317 5934 : multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2318 5934 : if (multiWrapLimit < FirstMultiXactId)
2319 0 : multiWrapLimit += FirstMultiXactId;
2320 :
2321 : /*
2322 : * We'll refuse to continue assigning MultiXactIds once we get within 3M
2323 : * multi of data loss. See SetTransactionIdLimit.
2324 : */
2325 5934 : multiStopLimit = multiWrapLimit - 3000000;
2326 5934 : if (multiStopLimit < FirstMultiXactId)
2327 0 : multiStopLimit -= FirstMultiXactId;
2328 :
2329 : /*
2330 : * We'll start complaining loudly when we get within 40M multis of data
2331 : * loss. This is kind of arbitrary, but if you let your gas gauge get
2332 : * down to 2% of full, would you be looking for the next gas station? We
2333 : * need to be fairly liberal about this number because there are lots of
2334 : * scenarios where most transactions are done by automatic clients that
2335 : * won't pay attention to warnings. (No, we're not gonna make this
2336 : * configurable. If you know enough to configure it, you know enough to
2337 : * not get in this kind of trouble in the first place.)
2338 : */
2339 5934 : multiWarnLimit = multiWrapLimit - 40000000;
2340 5934 : if (multiWarnLimit < FirstMultiXactId)
2341 0 : multiWarnLimit -= FirstMultiXactId;
2342 :
2343 : /*
2344 : * We'll start trying to force autovacuums when oldest_datminmxid gets to
2345 : * be more than autovacuum_multixact_freeze_max_age mxids old.
2346 : *
2347 : * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2348 : * so that we don't have to worry about dealing with on-the-fly changes in
2349 : * its value. See SetTransactionIdLimit.
2350 : */
2351 5934 : multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2352 5934 : if (multiVacLimit < FirstMultiXactId)
2353 0 : multiVacLimit += FirstMultiXactId;
2354 :
2355 : /* Grab lock for just long enough to set the new limit values */
2356 5934 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2357 5934 : MultiXactState->oldestMultiXactId = oldest_datminmxid;
2358 5934 : MultiXactState->oldestMultiXactDB = oldest_datoid;
2359 5934 : MultiXactState->multiVacLimit = multiVacLimit;
2360 5934 : MultiXactState->multiWarnLimit = multiWarnLimit;
2361 5934 : MultiXactState->multiStopLimit = multiStopLimit;
2362 5934 : MultiXactState->multiWrapLimit = multiWrapLimit;
2363 5934 : curMulti = MultiXactState->nextMXact;
2364 5934 : LWLockRelease(MultiXactGenLock);
2365 :
2366 : /* Log the info */
2367 5934 : ereport(DEBUG1,
2368 : (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2369 : multiWrapLimit, oldest_datoid)));
2370 :
2371 : /*
2372 : * Computing the actual limits is only possible once the data directory is
2373 : * in a consistent state. There's no need to compute the limits while
2374 : * still replaying WAL - no decisions about new multis are made even
2375 : * though multixact creations might be replayed. So we'll only do further
2376 : * checks after TrimMultiXact() has been called.
2377 : */
2378 5934 : if (!MultiXactState->finishedStartup)
2379 1966 : return;
2380 :
2381 : Assert(!InRecovery);
2382 :
2383 : /* Set limits for offset vacuum. */
2384 3968 : needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
2385 :
2386 : /*
2387 : * If past the autovacuum force point, immediately signal an autovac
2388 : * request. The reason for this is that autovac only processes one
2389 : * database per invocation. Once it's finished cleaning up the oldest
2390 : * database, it'll call here, and we'll signal the postmaster to start
2391 : * another iteration immediately if there are still any old databases.
2392 : */
2393 3968 : if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
2394 0 : needs_offset_vacuum) && IsUnderPostmaster)
2395 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
2396 :
2397 : /* Give an immediate warning if past the wrap warn point */
2398 3968 : if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2399 : {
2400 : char *oldest_datname;
2401 :
2402 : /*
2403 : * We can be called when not inside a transaction, for example during
2404 : * StartupXLOG(). In such a case we cannot do database access, so we
2405 : * must just report the oldest DB's OID.
2406 : *
2407 : * Note: it's also possible that get_database_name fails and returns
2408 : * NULL, for example because the database just got dropped. We'll
2409 : * still warn, even though the warning might now be unnecessary.
2410 : */
2411 0 : if (IsTransactionState())
2412 0 : oldest_datname = get_database_name(oldest_datoid);
2413 : else
2414 0 : oldest_datname = NULL;
2415 :
2416 0 : if (oldest_datname)
2417 0 : ereport(WARNING,
2418 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2419 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2420 : multiWrapLimit - curMulti,
2421 : oldest_datname,
2422 : multiWrapLimit - curMulti),
2423 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2424 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2425 : else
2426 0 : ereport(WARNING,
2427 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2428 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2429 : multiWrapLimit - curMulti,
2430 : oldest_datoid,
2431 : multiWrapLimit - curMulti),
2432 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2433 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2434 : }
2435 : }
2436 :
2437 : /*
2438 : * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2439 : * and similarly nextOffset is at least minMultiOffset.
2440 : *
2441 : * This is used when we can determine minimum safe values from an XLog
2442 : * record (either an on-line checkpoint or an mxact creation log entry).
2443 : * Although this is only called during XLog replay, we take the lock in case
2444 : * any hot-standby backends are examining the values.
2445 : */
2446 : void
2447 1350 : MultiXactAdvanceNextMXact(MultiXactId minMulti,
2448 : MultiXactOffset minMultiOffset)
2449 : {
2450 1350 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2451 1350 : if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
2452 : {
2453 : debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2454 4 : MultiXactState->nextMXact = minMulti;
2455 : }
2456 1350 : if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
2457 : {
2458 : debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
2459 : minMultiOffset);
2460 4 : MultiXactState->nextOffset = minMultiOffset;
2461 : }
2462 1350 : LWLockRelease(MultiXactGenLock);
2463 1350 : }
2464 :
2465 : /*
2466 : * Update our oldestMultiXactId value, but only if it's more recent than what
2467 : * we had.
2468 : *
2469 : * This may only be called during WAL replay.
2470 : */
2471 : void
2472 1416 : MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2473 : {
2474 : Assert(InRecovery);
2475 :
2476 1416 : if (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
2477 0 : SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
2478 1416 : }
2479 :
2480 : /*
2481 : * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2482 : *
2483 : * NB: this is called while holding MultiXactGenLock. We want it to be very
2484 : * fast most of the time; even when it's not so fast, no actual I/O need
2485 : * happen unless we're forced to write out a dirty log or xlog page to make
2486 : * room in shared memory.
2487 : */
2488 : static void
2489 554 : ExtendMultiXactOffset(MultiXactId multi)
2490 : {
2491 : int64 pageno;
2492 : LWLock *lock;
2493 :
2494 : /*
2495 : * No work except at first MultiXactId of a page. But beware: just after
2496 : * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2497 : */
2498 554 : if (MultiXactIdToOffsetEntry(multi) != 0 &&
2499 : multi != FirstMultiXactId)
2500 532 : return;
2501 :
2502 22 : pageno = MultiXactIdToOffsetPage(multi);
2503 22 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2504 :
2505 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2506 :
2507 : /* Zero the page and make a WAL entry about it */
2508 22 : SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2509 22 : XLogSimpleInsertInt64(RM_MULTIXACT_ID, XLOG_MULTIXACT_ZERO_OFF_PAGE,
2510 : pageno);
2511 :
2512 22 : LWLockRelease(lock);
2513 : }
2514 :
2515 : /*
2516 : * Make sure that MultiXactMember has room for the members of a newly-
2517 : * allocated MultiXactId.
2518 : *
2519 : * Like the above routine, this is called while holding MultiXactGenLock;
2520 : * same comments apply.
2521 : */
2522 : static void
2523 554 : ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
2524 : {
2525 : /*
2526 : * It's possible that the members span more than one page of the members
2527 : * file, so we loop to ensure we consider each page. The coding is not
2528 : * optimal if the members span several pages, but that seems unusual
2529 : * enough to not worry much about.
2530 : */
2531 1108 : while (nmembers > 0)
2532 : {
2533 : int flagsoff;
2534 : int flagsbit;
2535 : uint32 difference;
2536 :
2537 : /*
2538 : * Only zero when at first entry of a page.
2539 : */
2540 554 : flagsoff = MXOffsetToFlagsOffset(offset);
2541 554 : flagsbit = MXOffsetToFlagsBitShift(offset);
2542 554 : if (flagsoff == 0 && flagsbit == 0)
2543 : {
2544 : int64 pageno;
2545 : LWLock *lock;
2546 :
2547 22 : pageno = MXOffsetToMemberPage(offset);
2548 22 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2549 :
2550 22 : LWLockAcquire(lock, LW_EXCLUSIVE);
2551 :
2552 : /* Zero the page and make a WAL entry about it */
2553 22 : SimpleLruZeroPage(MultiXactMemberCtl, pageno);
2554 22 : XLogSimpleInsertInt64(RM_MULTIXACT_ID,
2555 : XLOG_MULTIXACT_ZERO_MEM_PAGE, pageno);
2556 :
2557 22 : LWLockRelease(lock);
2558 : }
2559 :
2560 : /*
2561 : * Compute the number of items till end of current page. Careful: if
2562 : * addition of unsigned ints wraps around, we're at the last page of
2563 : * the last segment; since that page holds a different number of items
2564 : * than other pages, we need to do it differently.
2565 : */
2566 554 : if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
2567 : {
2568 : /*
2569 : * This is the last page of the last segment; we can compute the
2570 : * number of items left to allocate in it without modulo
2571 : * arithmetic.
2572 : */
2573 0 : difference = MaxMultiXactOffset - offset + 1;
2574 : }
2575 : else
2576 554 : difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
2577 :
2578 : /*
2579 : * Advance to next page, taking care to properly handle the wraparound
2580 : * case. OK if nmembers goes negative.
2581 : */
2582 554 : nmembers -= difference;
2583 554 : offset += difference;
2584 : }
2585 554 : }
2586 :
2587 : /*
2588 : * GetOldestMultiXactId
2589 : *
2590 : * Return the oldest MultiXactId that's still possibly still seen as live by
2591 : * any running transaction. Older ones might still exist on disk, but they no
2592 : * longer have any running member transaction.
2593 : *
2594 : * It's not safe to truncate MultiXact SLRU segments on the value returned by
2595 : * this function; however, it can be set as the new relminmxid for any table
2596 : * that VACUUM knows has no remaining MXIDs < the same value. It is only safe
2597 : * to truncate SLRUs when no table can possibly still have a referencing MXID.
2598 : */
2599 : MultiXactId
2600 296856 : GetOldestMultiXactId(void)
2601 : {
2602 : MultiXactId oldestMXact;
2603 : MultiXactId nextMXact;
2604 : int i;
2605 :
2606 : /*
2607 : * This is the oldest valid value among all the OldestMemberMXactId[] and
2608 : * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2609 : */
2610 296856 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2611 :
2612 : /*
2613 : * We have to beware of the possibility that nextMXact is in the
2614 : * wrapped-around state. We don't fix the counter itself here, but we
2615 : * must be sure to use a valid value in our calculation.
2616 : */
2617 296856 : nextMXact = MultiXactState->nextMXact;
2618 296856 : if (nextMXact < FirstMultiXactId)
2619 0 : nextMXact = FirstMultiXactId;
2620 :
2621 296856 : oldestMXact = nextMXact;
2622 37019458 : for (i = 0; i < MaxOldestSlot; i++)
2623 : {
2624 : MultiXactId thisoldest;
2625 :
2626 36722602 : thisoldest = OldestMemberMXactId[i];
2627 36789236 : if (MultiXactIdIsValid(thisoldest) &&
2628 66634 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2629 20 : oldestMXact = thisoldest;
2630 36722602 : thisoldest = OldestVisibleMXactId[i];
2631 36722794 : if (MultiXactIdIsValid(thisoldest) &&
2632 192 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2633 4 : oldestMXact = thisoldest;
2634 : }
2635 :
2636 296856 : LWLockRelease(MultiXactGenLock);
2637 :
2638 296856 : return oldestMXact;
2639 : }
2640 :
2641 : /*
2642 : * Determine how aggressively we need to vacuum in order to prevent member
2643 : * wraparound.
2644 : *
2645 : * To do so determine what's the oldest member offset and install the limit
2646 : * info in MultiXactState, where it can be used to prevent overrun of old data
2647 : * in the members SLRU area.
2648 : *
2649 : * The return value is true if emergency autovacuum is required and false
2650 : * otherwise.
2651 : */
2652 : static bool
2653 3968 : SetOffsetVacuumLimit(bool is_startup)
2654 : {
2655 : MultiXactId oldestMultiXactId;
2656 : MultiXactId nextMXact;
2657 3968 : MultiXactOffset oldestOffset = 0; /* placate compiler */
2658 : MultiXactOffset prevOldestOffset;
2659 : MultiXactOffset nextOffset;
2660 3968 : bool oldestOffsetKnown = false;
2661 : bool prevOldestOffsetKnown;
2662 3968 : MultiXactOffset offsetStopLimit = 0;
2663 : MultiXactOffset prevOffsetStopLimit;
2664 :
2665 : /*
2666 : * NB: Have to prevent concurrent truncation, we might otherwise try to
2667 : * lookup an oldestMulti that's concurrently getting truncated away.
2668 : */
2669 3968 : LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2670 :
2671 : /* Read relevant fields from shared memory. */
2672 3968 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2673 3968 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2674 3968 : nextMXact = MultiXactState->nextMXact;
2675 3968 : nextOffset = MultiXactState->nextOffset;
2676 3968 : prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2677 3968 : prevOldestOffset = MultiXactState->oldestOffset;
2678 3968 : prevOffsetStopLimit = MultiXactState->offsetStopLimit;
2679 : Assert(MultiXactState->finishedStartup);
2680 3968 : LWLockRelease(MultiXactGenLock);
2681 :
2682 : /*
2683 : * Determine the offset of the oldest multixact. Normally, we can read
2684 : * the offset from the multixact itself, but there's an important special
2685 : * case: if there are no multixacts in existence at all, oldestMXact
2686 : * obviously can't point to one. It will instead point to the multixact
2687 : * ID that will be assigned the next time one is needed.
2688 : */
2689 3968 : if (oldestMultiXactId == nextMXact)
2690 : {
2691 : /*
2692 : * When the next multixact gets created, it will be stored at the next
2693 : * offset.
2694 : */
2695 3940 : oldestOffset = nextOffset;
2696 3940 : oldestOffsetKnown = true;
2697 : }
2698 : else
2699 : {
2700 : /*
2701 : * Figure out where the oldest existing multixact's offsets are
2702 : * stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
2703 : * the supposedly-earliest multixact might not really exist. We are
2704 : * careful not to fail in that case.
2705 : */
2706 : oldestOffsetKnown =
2707 28 : find_multixact_start(oldestMultiXactId, &oldestOffset);
2708 :
2709 28 : if (oldestOffsetKnown)
2710 28 : ereport(DEBUG1,
2711 : (errmsg_internal("oldest MultiXactId member is at offset %u",
2712 : oldestOffset)));
2713 : else
2714 0 : ereport(LOG,
2715 : (errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
2716 : oldestMultiXactId)));
2717 : }
2718 :
2719 3968 : LWLockRelease(MultiXactTruncationLock);
2720 :
2721 : /*
2722 : * If we can, compute limits (and install them MultiXactState) to prevent
2723 : * overrun of old data in the members SLRU area. We can only do so if the
2724 : * oldest offset is known though.
2725 : */
2726 3968 : if (oldestOffsetKnown)
2727 : {
2728 : /* move back to start of the corresponding segment */
2729 3968 : offsetStopLimit = oldestOffset - (oldestOffset %
2730 : (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT));
2731 :
2732 : /* always leave one segment before the wraparound point */
2733 3968 : offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
2734 :
2735 3968 : if (!prevOldestOffsetKnown && !is_startup)
2736 0 : ereport(LOG,
2737 : (errmsg("MultiXact member wraparound protections are now enabled")));
2738 :
2739 3968 : ereport(DEBUG1,
2740 : (errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
2741 : offsetStopLimit, oldestMultiXactId)));
2742 : }
2743 0 : else if (prevOldestOffsetKnown)
2744 : {
2745 : /*
2746 : * If we failed to get the oldest offset this time, but we have a
2747 : * value from a previous pass through this function, use the old
2748 : * values rather than automatically forcing an emergency autovacuum
2749 : * cycle again.
2750 : */
2751 0 : oldestOffset = prevOldestOffset;
2752 0 : oldestOffsetKnown = true;
2753 0 : offsetStopLimit = prevOffsetStopLimit;
2754 : }
2755 :
2756 : /* Install the computed values */
2757 3968 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2758 3968 : MultiXactState->oldestOffset = oldestOffset;
2759 3968 : MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
2760 3968 : MultiXactState->offsetStopLimit = offsetStopLimit;
2761 3968 : LWLockRelease(MultiXactGenLock);
2762 :
2763 : /*
2764 : * Do we need an emergency autovacuum? If we're not sure, assume yes.
2765 : */
2766 7936 : return !oldestOffsetKnown ||
2767 3968 : (nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
2768 : }
2769 :
2770 : /*
2771 : * Return whether adding "distance" to "start" would move past "boundary".
2772 : *
2773 : * We use this to determine whether the addition is "wrapping around" the
2774 : * boundary point, hence the name. The reason we don't want to use the regular
2775 : * 2^31-modulo arithmetic here is that we want to be able to use the whole of
2776 : * the 2^32-1 space here, allowing for more multixacts than would fit
2777 : * otherwise.
2778 : */
2779 : static bool
2780 1108 : MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start,
2781 : uint32 distance)
2782 : {
2783 : MultiXactOffset finish;
2784 :
2785 : /*
2786 : * Note that offset number 0 is not used (see GetMultiXactIdMembers), so
2787 : * if the addition wraps around the UINT_MAX boundary, skip that value.
2788 : */
2789 1108 : finish = start + distance;
2790 1108 : if (finish < start)
2791 0 : finish++;
2792 :
2793 : /*-----------------------------------------------------------------------
2794 : * When the boundary is numerically greater than the starting point, any
2795 : * value numerically between the two is not wrapped:
2796 : *
2797 : * <----S----B---->
2798 : * [---) = F wrapped past B (and UINT_MAX)
2799 : * [---) = F not wrapped
2800 : * [----] = F wrapped past B
2801 : *
2802 : * When the boundary is numerically less than the starting point (i.e. the
2803 : * UINT_MAX wraparound occurs somewhere in between) then all values in
2804 : * between are wrapped:
2805 : *
2806 : * <----B----S---->
2807 : * [---) = F not wrapped past B (but wrapped past UINT_MAX)
2808 : * [---) = F wrapped past B (and UINT_MAX)
2809 : * [----] = F not wrapped
2810 : *-----------------------------------------------------------------------
2811 : */
2812 1108 : if (start < boundary)
2813 1108 : return finish >= boundary || finish < start;
2814 : else
2815 0 : return finish >= boundary && finish < start;
2816 : }
2817 :
2818 : /*
2819 : * Find the starting offset of the given MultiXactId.
2820 : *
2821 : * Returns false if the file containing the multi does not exist on disk.
2822 : * Otherwise, returns true and sets *result to the starting member offset.
2823 : *
2824 : * This function does not prevent concurrent truncation, so if that's
2825 : * required, the caller has to protect against that.
2826 : */
2827 : static bool
2828 28 : find_multixact_start(MultiXactId multi, MultiXactOffset *result)
2829 : {
2830 : MultiXactOffset offset;
2831 : int64 pageno;
2832 : int entryno;
2833 : int slotno;
2834 : MultiXactOffset *offptr;
2835 :
2836 : Assert(MultiXactState->finishedStartup);
2837 :
2838 28 : pageno = MultiXactIdToOffsetPage(multi);
2839 28 : entryno = MultiXactIdToOffsetEntry(multi);
2840 :
2841 : /*
2842 : * Write out dirty data, so PhysicalPageExists can work correctly.
2843 : */
2844 28 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2845 28 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2846 :
2847 28 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2848 0 : return false;
2849 :
2850 : /* lock is acquired by SimpleLruReadPage_ReadOnly */
2851 28 : slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2852 28 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2853 28 : offptr += entryno;
2854 28 : offset = *offptr;
2855 28 : LWLockRelease(SimpleLruGetBankLock(MultiXactOffsetCtl, pageno));
2856 :
2857 28 : *result = offset;
2858 28 : return true;
2859 : }
2860 :
2861 : /*
2862 : * Determine how many multixacts, and how many multixact members, currently
2863 : * exist. Return false if unable to determine.
2864 : */
2865 : static bool
2866 230730 : ReadMultiXactCounts(uint32 *multixacts, MultiXactOffset *members)
2867 : {
2868 : MultiXactOffset nextOffset;
2869 : MultiXactOffset oldestOffset;
2870 : MultiXactId oldestMultiXactId;
2871 : MultiXactId nextMultiXactId;
2872 : bool oldestOffsetKnown;
2873 :
2874 230730 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2875 230730 : nextOffset = MultiXactState->nextOffset;
2876 230730 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2877 230730 : nextMultiXactId = MultiXactState->nextMXact;
2878 230730 : oldestOffset = MultiXactState->oldestOffset;
2879 230730 : oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2880 230730 : LWLockRelease(MultiXactGenLock);
2881 :
2882 230730 : if (!oldestOffsetKnown)
2883 0 : return false;
2884 :
2885 230730 : *members = nextOffset - oldestOffset;
2886 230730 : *multixacts = nextMultiXactId - oldestMultiXactId;
2887 230730 : return true;
2888 : }
2889 :
2890 : /*
2891 : * Multixact members can be removed once the multixacts that refer to them
2892 : * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2893 : * vacuum_multixact_freeze_table_age work together to make sure we never have
2894 : * too many multixacts; we hope that, at least under normal circumstances,
2895 : * this will also be sufficient to keep us from using too many offsets.
2896 : * However, if the average multixact has many members, we might exhaust the
2897 : * members space while still using few enough members that these limits fail
2898 : * to trigger relminmxid advancement by VACUUM. At that point, we'd have no
2899 : * choice but to start failing multixact-creating operations with an error.
2900 : *
2901 : * To prevent that, if more than a threshold portion of the members space is
2902 : * used, we effectively reduce autovacuum_multixact_freeze_max_age and
2903 : * to a value just less than the number of multixacts in use. We hope that
2904 : * this will quickly trigger autovacuuming on the table or tables with the
2905 : * oldest relminmxid, thus allowing datminmxid values to advance and removing
2906 : * some members.
2907 : *
2908 : * As the fraction of the member space currently in use grows, we become
2909 : * more aggressive in clamping this value. That not only causes autovacuum
2910 : * to ramp up, but also makes any manual vacuums the user issues more
2911 : * aggressive. This happens because vacuum_get_cutoffs() will clamp the
2912 : * freeze table and the minimum freeze age cutoffs based on the effective
2913 : * autovacuum_multixact_freeze_max_age this function returns. In the worst
2914 : * case, we'll claim the freeze_max_age to zero, and every vacuum of any
2915 : * table will freeze every multixact.
2916 : */
2917 : int
2918 230730 : MultiXactMemberFreezeThreshold(void)
2919 : {
2920 : MultiXactOffset members;
2921 : uint32 multixacts;
2922 : uint32 victim_multixacts;
2923 : double fraction;
2924 : int result;
2925 :
2926 : /* If we can't determine member space utilization, assume the worst. */
2927 230730 : if (!ReadMultiXactCounts(&multixacts, &members))
2928 0 : return 0;
2929 :
2930 : /* If member space utilization is low, no special action is required. */
2931 230730 : if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
2932 230730 : return autovacuum_multixact_freeze_max_age;
2933 :
2934 : /*
2935 : * Compute a target for relminmxid advancement. The number of multixacts
2936 : * we try to eliminate from the system is based on how far we are past
2937 : * MULTIXACT_MEMBER_SAFE_THRESHOLD.
2938 : */
2939 0 : fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
2940 : (MULTIXACT_MEMBER_DANGER_THRESHOLD - MULTIXACT_MEMBER_SAFE_THRESHOLD);
2941 0 : victim_multixacts = multixacts * fraction;
2942 :
2943 : /* fraction could be > 1.0, but lowest possible freeze age is zero */
2944 0 : if (victim_multixacts > multixacts)
2945 0 : return 0;
2946 0 : result = multixacts - victim_multixacts;
2947 :
2948 : /*
2949 : * Clamp to autovacuum_multixact_freeze_max_age, so that we never make
2950 : * autovacuum less aggressive than it would otherwise be.
2951 : */
2952 0 : return Min(result, autovacuum_multixact_freeze_max_age);
2953 : }
2954 :
2955 : typedef struct mxtruncinfo
2956 : {
2957 : int64 earliestExistingPage;
2958 : } mxtruncinfo;
2959 :
2960 : /*
2961 : * SlruScanDirectory callback
2962 : * This callback determines the earliest existing page number.
2963 : */
2964 : static bool
2965 0 : SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int64 segpage, void *data)
2966 : {
2967 0 : mxtruncinfo *trunc = (mxtruncinfo *) data;
2968 :
2969 0 : if (trunc->earliestExistingPage == -1 ||
2970 0 : ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
2971 : {
2972 0 : trunc->earliestExistingPage = segpage;
2973 : }
2974 :
2975 0 : return false; /* keep going */
2976 : }
2977 :
2978 :
2979 : /*
2980 : * Delete members segments [oldest, newOldest)
2981 : *
2982 : * The members SLRU can, in contrast to the offsets one, be filled to almost
2983 : * the full range at once. This means SimpleLruTruncate() can't trivially be
2984 : * used - instead the to-be-deleted range is computed using the offsets
2985 : * SLRU. C.f. TruncateMultiXact().
2986 : */
2987 : static void
2988 0 : PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
2989 : {
2990 0 : const int64 maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
2991 0 : int64 startsegment = MXOffsetToMemberSegment(oldestOffset);
2992 0 : int64 endsegment = MXOffsetToMemberSegment(newOldestOffset);
2993 0 : int64 segment = startsegment;
2994 :
2995 : /*
2996 : * Delete all the segments but the last one. The last segment can still
2997 : * contain, possibly partially, valid data.
2998 : */
2999 0 : while (segment != endsegment)
3000 : {
3001 0 : elog(DEBUG2, "truncating multixact members segment %" PRIx64,
3002 : segment);
3003 0 : SlruDeleteSegment(MultiXactMemberCtl, segment);
3004 :
3005 : /* move to next segment, handling wraparound correctly */
3006 0 : if (segment == maxsegment)
3007 0 : segment = 0;
3008 : else
3009 0 : segment += 1;
3010 : }
3011 0 : }
3012 :
3013 : /*
3014 : * Delete offsets segments [oldest, newOldest)
3015 : */
3016 : static void
3017 0 : PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
3018 : {
3019 : /*
3020 : * We step back one multixact to avoid passing a cutoff page that hasn't
3021 : * been created yet in the rare case that oldestMulti would be the first
3022 : * item on a page and oldestMulti == nextMulti. In that case, if we
3023 : * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
3024 : * detection.
3025 : */
3026 0 : SimpleLruTruncate(MultiXactOffsetCtl,
3027 : MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
3028 0 : }
3029 :
3030 : /*
3031 : * Remove all MultiXactOffset and MultiXactMember segments before the oldest
3032 : * ones still of interest.
3033 : *
3034 : * This is only called on a primary as part of vacuum (via
3035 : * vac_truncate_clog()). During recovery truncation is done by replaying
3036 : * truncation WAL records logged here.
3037 : *
3038 : * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
3039 : * is one of the databases preventing newOldestMulti from increasing.
3040 : */
3041 : void
3042 2222 : TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
3043 : {
3044 : MultiXactId oldestMulti;
3045 : MultiXactId nextMulti;
3046 : MultiXactOffset newOldestOffset;
3047 : MultiXactOffset oldestOffset;
3048 : MultiXactOffset nextOffset;
3049 : mxtruncinfo trunc;
3050 : MultiXactId earliest;
3051 :
3052 : Assert(!RecoveryInProgress());
3053 : Assert(MultiXactState->finishedStartup);
3054 :
3055 : /*
3056 : * We can only allow one truncation to happen at once. Otherwise parts of
3057 : * members might vanish while we're doing lookups or similar. There's no
3058 : * need to have an interlock with creating new multis or such, since those
3059 : * are constrained by the limits (which only grow, never shrink).
3060 : */
3061 2222 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3062 :
3063 2222 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
3064 2222 : nextMulti = MultiXactState->nextMXact;
3065 2222 : nextOffset = MultiXactState->nextOffset;
3066 2222 : oldestMulti = MultiXactState->oldestMultiXactId;
3067 2222 : LWLockRelease(MultiXactGenLock);
3068 : Assert(MultiXactIdIsValid(oldestMulti));
3069 :
3070 : /*
3071 : * Make sure to only attempt truncation if there's values to truncate
3072 : * away. In normal processing values shouldn't go backwards, but there's
3073 : * some corner cases (due to bugs) where that's possible.
3074 : */
3075 2222 : if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
3076 : {
3077 2222 : LWLockRelease(MultiXactTruncationLock);
3078 2222 : return;
3079 : }
3080 :
3081 : /*
3082 : * Note we can't just plow ahead with the truncation; it's possible that
3083 : * there are no segments to truncate, which is a problem because we are
3084 : * going to attempt to read the offsets page to determine where to
3085 : * truncate the members SLRU. So we first scan the directory to determine
3086 : * the earliest offsets page number that we can read without error.
3087 : *
3088 : * When nextMXact is less than one segment away from multiWrapLimit,
3089 : * SlruScanDirCbFindEarliest can find some early segment other than the
3090 : * actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
3091 : * returns false, because not all pairs of entries have the same answer.)
3092 : * That can also arise when an earlier truncation attempt failed unlink()
3093 : * or returned early from this function. The only consequence is
3094 : * returning early, which wastes space that we could have liberated.
3095 : *
3096 : * NB: It's also possible that the page that oldestMulti is on has already
3097 : * been truncated away, and we crashed before updating oldestMulti.
3098 : */
3099 0 : trunc.earliestExistingPage = -1;
3100 0 : SlruScanDirectory(MultiXactOffsetCtl, SlruScanDirCbFindEarliest, &trunc);
3101 0 : earliest = trunc.earliestExistingPage * MULTIXACT_OFFSETS_PER_PAGE;
3102 0 : if (earliest < FirstMultiXactId)
3103 0 : earliest = FirstMultiXactId;
3104 :
3105 : /* If there's nothing to remove, we can bail out early. */
3106 0 : if (MultiXactIdPrecedes(oldestMulti, earliest))
3107 : {
3108 0 : LWLockRelease(MultiXactTruncationLock);
3109 0 : return;
3110 : }
3111 :
3112 : /*
3113 : * First, compute the safe truncation point for MultiXactMember. This is
3114 : * the starting offset of the oldest multixact.
3115 : *
3116 : * Hopefully, find_multixact_start will always work here, because we've
3117 : * already checked that it doesn't precede the earliest MultiXact on disk.
3118 : * But if it fails, don't truncate anything, and log a message.
3119 : */
3120 0 : if (oldestMulti == nextMulti)
3121 : {
3122 : /* there are NO MultiXacts */
3123 0 : oldestOffset = nextOffset;
3124 : }
3125 0 : else if (!find_multixact_start(oldestMulti, &oldestOffset))
3126 : {
3127 0 : ereport(LOG,
3128 : (errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
3129 : oldestMulti, earliest)));
3130 0 : LWLockRelease(MultiXactTruncationLock);
3131 0 : return;
3132 : }
3133 :
3134 : /*
3135 : * Secondly compute up to where to truncate. Lookup the corresponding
3136 : * member offset for newOldestMulti for that.
3137 : */
3138 0 : if (newOldestMulti == nextMulti)
3139 : {
3140 : /* there are NO MultiXacts */
3141 0 : newOldestOffset = nextOffset;
3142 : }
3143 0 : else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
3144 : {
3145 0 : ereport(LOG,
3146 : (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
3147 : newOldestMulti)));
3148 0 : LWLockRelease(MultiXactTruncationLock);
3149 0 : return;
3150 : }
3151 :
3152 0 : elog(DEBUG1, "performing multixact truncation: "
3153 : "offsets [%u, %u), offsets segments [%" PRIx64 ", %" PRIx64 "), "
3154 : "members [%u, %u), members segments [%" PRIx64 ", %" PRIx64 ")",
3155 : oldestMulti, newOldestMulti,
3156 : MultiXactIdToOffsetSegment(oldestMulti),
3157 : MultiXactIdToOffsetSegment(newOldestMulti),
3158 : oldestOffset, newOldestOffset,
3159 : MXOffsetToMemberSegment(oldestOffset),
3160 : MXOffsetToMemberSegment(newOldestOffset));
3161 :
3162 : /*
3163 : * Do truncation, and the WAL logging of the truncation, in a critical
3164 : * section. That way offsets/members cannot get out of sync anymore, i.e.
3165 : * once consistent the newOldestMulti will always exist in members, even
3166 : * if we crashed in the wrong moment.
3167 : */
3168 0 : START_CRIT_SECTION();
3169 :
3170 : /*
3171 : * Prevent checkpoints from being scheduled concurrently. This is critical
3172 : * because otherwise a truncation record might not be replayed after a
3173 : * crash/basebackup, even though the state of the data directory would
3174 : * require it.
3175 : */
3176 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
3177 0 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
3178 :
3179 : /* WAL log truncation */
3180 0 : WriteMTruncateXlogRec(newOldestMultiDB,
3181 : oldestMulti, newOldestMulti,
3182 : oldestOffset, newOldestOffset);
3183 :
3184 : /*
3185 : * Update in-memory limits before performing the truncation, while inside
3186 : * the critical section: Have to do it before truncation, to prevent
3187 : * concurrent lookups of those values. Has to be inside the critical
3188 : * section as otherwise a future call to this function would error out,
3189 : * while looking up the oldest member in offsets, if our caller crashes
3190 : * before updating the limits.
3191 : */
3192 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
3193 0 : MultiXactState->oldestMultiXactId = newOldestMulti;
3194 0 : MultiXactState->oldestMultiXactDB = newOldestMultiDB;
3195 0 : LWLockRelease(MultiXactGenLock);
3196 :
3197 : /* First truncate members */
3198 0 : PerformMembersTruncation(oldestOffset, newOldestOffset);
3199 :
3200 : /* Then offsets */
3201 0 : PerformOffsetsTruncation(oldestMulti, newOldestMulti);
3202 :
3203 0 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
3204 :
3205 0 : END_CRIT_SECTION();
3206 0 : LWLockRelease(MultiXactTruncationLock);
3207 : }
3208 :
3209 : /*
3210 : * Decide whether a MultiXactOffset page number is "older" for truncation
3211 : * purposes. Analogous to CLOGPagePrecedes().
3212 : *
3213 : * Offsetting the values is optional, because MultiXactIdPrecedes() has
3214 : * translational symmetry.
3215 : */
3216 : static bool
3217 0 : MultiXactOffsetPagePrecedes(int64 page1, int64 page2)
3218 : {
3219 : MultiXactId multi1;
3220 : MultiXactId multi2;
3221 :
3222 0 : multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
3223 0 : multi1 += FirstMultiXactId + 1;
3224 0 : multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
3225 0 : multi2 += FirstMultiXactId + 1;
3226 :
3227 0 : return (MultiXactIdPrecedes(multi1, multi2) &&
3228 0 : MultiXactIdPrecedes(multi1,
3229 : multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
3230 : }
3231 :
3232 : /*
3233 : * Decide whether a MultiXactMember page number is "older" for truncation
3234 : * purposes. There is no "invalid offset number" so use the numbers verbatim.
3235 : */
3236 : static bool
3237 0 : MultiXactMemberPagePrecedes(int64 page1, int64 page2)
3238 : {
3239 : MultiXactOffset offset1;
3240 : MultiXactOffset offset2;
3241 :
3242 0 : offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
3243 0 : offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
3244 :
3245 0 : return (MultiXactOffsetPrecedes(offset1, offset2) &&
3246 0 : MultiXactOffsetPrecedes(offset1,
3247 : offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
3248 : }
3249 :
3250 : /*
3251 : * Decide which of two MultiXactIds is earlier.
3252 : *
3253 : * XXX do we need to do something special for InvalidMultiXactId?
3254 : * (Doesn't look like it.)
3255 : */
3256 : bool
3257 2512762 : MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
3258 : {
3259 2512762 : int32 diff = (int32) (multi1 - multi2);
3260 :
3261 2512762 : return (diff < 0);
3262 : }
3263 :
3264 : /*
3265 : * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
3266 : *
3267 : * XXX do we need to do something special for InvalidMultiXactId?
3268 : * (Doesn't look like it.)
3269 : */
3270 : bool
3271 13338 : MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
3272 : {
3273 13338 : int32 diff = (int32) (multi1 - multi2);
3274 :
3275 13338 : return (diff <= 0);
3276 : }
3277 :
3278 :
3279 : /*
3280 : * Decide which of two offsets is earlier.
3281 : */
3282 : static bool
3283 1350 : MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
3284 : {
3285 1350 : int32 diff = (int32) (offset1 - offset2);
3286 :
3287 1350 : return (diff < 0);
3288 : }
3289 :
3290 : /*
3291 : * Write a TRUNCATE xlog record
3292 : *
3293 : * We must flush the xlog record to disk before returning --- see notes in
3294 : * TruncateCLOG().
3295 : */
3296 : static void
3297 0 : WriteMTruncateXlogRec(Oid oldestMultiDB,
3298 : MultiXactId startTruncOff, MultiXactId endTruncOff,
3299 : MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
3300 : {
3301 : XLogRecPtr recptr;
3302 : xl_multixact_truncate xlrec;
3303 :
3304 0 : xlrec.oldestMultiDB = oldestMultiDB;
3305 :
3306 0 : xlrec.startTruncOff = startTruncOff;
3307 0 : xlrec.endTruncOff = endTruncOff;
3308 :
3309 0 : xlrec.startTruncMemb = startTruncMemb;
3310 0 : xlrec.endTruncMemb = endTruncMemb;
3311 :
3312 0 : XLogBeginInsert();
3313 0 : XLogRegisterData(&xlrec, SizeOfMultiXactTruncate);
3314 0 : recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
3315 0 : XLogFlush(recptr);
3316 0 : }
3317 :
3318 : /*
3319 : * MULTIXACT resource manager's routines
3320 : */
3321 : void
3322 8 : multixact_redo(XLogReaderState *record)
3323 : {
3324 8 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3325 :
3326 : /* Backup blocks are not used in multixact records */
3327 : Assert(!XLogRecHasAnyBlockRefs(record));
3328 :
3329 8 : if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
3330 : {
3331 : int64 pageno;
3332 :
3333 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3334 2 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, pageno);
3335 : }
3336 6 : else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
3337 : {
3338 : int64 pageno;
3339 :
3340 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3341 2 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, pageno);
3342 : }
3343 4 : else if (info == XLOG_MULTIXACT_CREATE_ID)
3344 : {
3345 4 : xl_multixact_create *xlrec =
3346 4 : (xl_multixact_create *) XLogRecGetData(record);
3347 : TransactionId max_xid;
3348 : int i;
3349 :
3350 : /* Store the data back into the SLRU files */
3351 4 : RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
3352 4 : xlrec->members);
3353 :
3354 : /* Make sure nextMXact/nextOffset are beyond what this record has */
3355 4 : MultiXactAdvanceNextMXact(xlrec->mid + 1,
3356 4 : xlrec->moff + xlrec->nmembers);
3357 :
3358 : /*
3359 : * Make sure nextXid is beyond any XID mentioned in the record. This
3360 : * should be unnecessary, since any XID found here ought to have other
3361 : * evidence in the XLOG, but let's be safe.
3362 : */
3363 4 : max_xid = XLogRecGetXid(record);
3364 12 : for (i = 0; i < xlrec->nmembers; i++)
3365 : {
3366 8 : if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
3367 0 : max_xid = xlrec->members[i].xid;
3368 : }
3369 :
3370 4 : AdvanceNextFullTransactionIdPastXid(max_xid);
3371 : }
3372 0 : else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
3373 : {
3374 : xl_multixact_truncate xlrec;
3375 : int64 pageno;
3376 :
3377 0 : memcpy(&xlrec, XLogRecGetData(record),
3378 : SizeOfMultiXactTruncate);
3379 :
3380 0 : elog(DEBUG1, "replaying multixact truncation: "
3381 : "offsets [%u, %u), offsets segments [%" PRIx64 ", %" PRIx64 "), "
3382 : "members [%u, %u), members segments [%" PRIx64 ", %" PRIx64 ")",
3383 : xlrec.startTruncOff, xlrec.endTruncOff,
3384 : MultiXactIdToOffsetSegment(xlrec.startTruncOff),
3385 : MultiXactIdToOffsetSegment(xlrec.endTruncOff),
3386 : xlrec.startTruncMemb, xlrec.endTruncMemb,
3387 : MXOffsetToMemberSegment(xlrec.startTruncMemb),
3388 : MXOffsetToMemberSegment(xlrec.endTruncMemb));
3389 :
3390 : /* should not be required, but more than cheap enough */
3391 0 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3392 :
3393 : /*
3394 : * Advance the horizon values, so they're current at the end of
3395 : * recovery.
3396 : */
3397 0 : SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
3398 :
3399 0 : PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
3400 :
3401 : /*
3402 : * During XLOG replay, latest_page_number isn't necessarily set up
3403 : * yet; insert a suitable value to bypass the sanity test in
3404 : * SimpleLruTruncate.
3405 : */
3406 0 : pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
3407 0 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
3408 : pageno);
3409 0 : PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
3410 :
3411 0 : LWLockRelease(MultiXactTruncationLock);
3412 : }
3413 : else
3414 0 : elog(PANIC, "multixact_redo: unknown op code %u", info);
3415 8 : }
3416 :
3417 : Datum
3418 0 : pg_get_multixact_members(PG_FUNCTION_ARGS)
3419 : {
3420 : typedef struct
3421 : {
3422 : MultiXactMember *members;
3423 : int nmembers;
3424 : int iter;
3425 : } mxact;
3426 0 : MultiXactId mxid = PG_GETARG_TRANSACTIONID(0);
3427 : mxact *multi;
3428 : FuncCallContext *funccxt;
3429 :
3430 0 : if (mxid < FirstMultiXactId)
3431 0 : ereport(ERROR,
3432 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3433 : errmsg("invalid MultiXactId: %u", mxid)));
3434 :
3435 0 : if (SRF_IS_FIRSTCALL())
3436 : {
3437 : MemoryContext oldcxt;
3438 : TupleDesc tupdesc;
3439 :
3440 0 : funccxt = SRF_FIRSTCALL_INIT();
3441 0 : oldcxt = MemoryContextSwitchTo(funccxt->multi_call_memory_ctx);
3442 :
3443 0 : multi = palloc(sizeof(mxact));
3444 : /* no need to allow for old values here */
3445 0 : multi->nmembers = GetMultiXactIdMembers(mxid, &multi->members, false,
3446 : false);
3447 0 : multi->iter = 0;
3448 :
3449 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
3450 0 : elog(ERROR, "return type must be a row type");
3451 0 : funccxt->tuple_desc = tupdesc;
3452 0 : funccxt->attinmeta = TupleDescGetAttInMetadata(tupdesc);
3453 0 : funccxt->user_fctx = multi;
3454 :
3455 0 : MemoryContextSwitchTo(oldcxt);
3456 : }
3457 :
3458 0 : funccxt = SRF_PERCALL_SETUP();
3459 0 : multi = (mxact *) funccxt->user_fctx;
3460 :
3461 0 : while (multi->iter < multi->nmembers)
3462 : {
3463 : HeapTuple tuple;
3464 : char *values[2];
3465 :
3466 0 : values[0] = psprintf("%u", multi->members[multi->iter].xid);
3467 0 : values[1] = mxstatus_to_string(multi->members[multi->iter].status);
3468 :
3469 0 : tuple = BuildTupleFromCStrings(funccxt->attinmeta, values);
3470 :
3471 0 : multi->iter++;
3472 0 : pfree(values[0]);
3473 0 : SRF_RETURN_NEXT(funccxt, HeapTupleGetDatum(tuple));
3474 : }
3475 :
3476 0 : SRF_RETURN_DONE(funccxt);
3477 : }
3478 :
3479 : /*
3480 : * Entrypoint for sync.c to sync offsets files.
3481 : */
3482 : int
3483 0 : multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
3484 : {
3485 0 : return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
3486 : }
3487 :
3488 : /*
3489 : * Entrypoint for sync.c to sync members files.
3490 : */
3491 : int
3492 0 : multixactmemberssyncfiletag(const FileTag *ftag, char *path)
3493 : {
3494 0 : return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
3495 : }
|