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) 1996-2024, 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/memutils.h"
93 :
94 :
95 : /*
96 : * Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
97 : * used everywhere else in Postgres.
98 : *
99 : * Note: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
100 : * MultiXact page numbering also wraps around at
101 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
102 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need
103 : * take no explicit notice of that fact in this module, except when comparing
104 : * segment and page numbers in TruncateMultiXact (see
105 : * MultiXactOffsetPagePrecedes).
106 : */
107 :
108 : /* We need four bytes per offset */
109 : #define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
110 :
111 : #define MultiXactIdToOffsetPage(xid) \
112 : ((xid) / (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
113 : #define MultiXactIdToOffsetEntry(xid) \
114 : ((xid) % (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
115 : #define MultiXactIdToOffsetSegment(xid) (MultiXactIdToOffsetPage(xid) / SLRU_PAGES_PER_SEGMENT)
116 :
117 : /*
118 : * The situation for members is a bit more complex: we store one byte of
119 : * additional flag bits for each TransactionId. To do this without getting
120 : * into alignment issues, we store four bytes of flags, and then the
121 : * corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
122 : * are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
123 : * per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
124 : * performance) trumps space efficiency here.
125 : *
126 : * Note that the "offset" macros work with byte offset, not array indexes, so
127 : * arithmetic must be done using "char *" pointers.
128 : */
129 : /* We need eight bits per xact, so one xact fits in a byte */
130 : #define MXACT_MEMBER_BITS_PER_XACT 8
131 : #define MXACT_MEMBER_FLAGS_PER_BYTE 1
132 : #define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
133 :
134 : /* how many full bytes of flags are there in a group? */
135 : #define MULTIXACT_FLAGBYTES_PER_GROUP 4
136 : #define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
137 : (MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
138 : /* size in bytes of a complete group */
139 : #define MULTIXACT_MEMBERGROUP_SIZE \
140 : (sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
141 : #define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
142 : #define MULTIXACT_MEMBERS_PER_PAGE \
143 : (MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
144 :
145 : /*
146 : * Because the number of items per page is not a divisor of the last item
147 : * number (member 0xFFFFFFFF), the last segment does not use the maximum number
148 : * of pages, and moreover the last used page therein does not use the same
149 : * number of items as previous pages. (Another way to say it is that the
150 : * 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
151 : * has some empty space after that item.)
152 : *
153 : * This constant is the number of members in the last page of the last segment.
154 : */
155 : #define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
156 : ((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
157 :
158 : /* page in which a member is to be found */
159 : #define MXOffsetToMemberPage(xid) ((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_PAGE)
160 : #define MXOffsetToMemberSegment(xid) (MXOffsetToMemberPage(xid) / SLRU_PAGES_PER_SEGMENT)
161 :
162 : /* Location (byte offset within page) of flag word for a given member */
163 : #define MXOffsetToFlagsOffset(xid) \
164 : ((((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) % \
165 : (TransactionId) MULTIXACT_MEMBERGROUPS_PER_PAGE) * \
166 : (TransactionId) MULTIXACT_MEMBERGROUP_SIZE)
167 : #define MXOffsetToFlagsBitShift(xid) \
168 : (((xid) % (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) * \
169 : MXACT_MEMBER_BITS_PER_XACT)
170 :
171 : /* Location (byte offset within page) of TransactionId of given member */
172 : #define MXOffsetToMemberOffset(xid) \
173 : (MXOffsetToFlagsOffset(xid) + MULTIXACT_FLAGBYTES_PER_GROUP + \
174 : ((xid) % MULTIXACT_MEMBERS_PER_MEMBERGROUP) * sizeof(TransactionId))
175 :
176 : /* Multixact members wraparound thresholds. */
177 : #define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
178 : #define MULTIXACT_MEMBER_DANGER_THRESHOLD \
179 : (MaxMultiXactOffset - MaxMultiXactOffset / 4)
180 :
181 : #define PreviousMultiXactId(xid) \
182 : ((xid) == FirstMultiXactId ? MaxMultiXactId : (xid) - 1)
183 :
184 : /*
185 : * Links to shared-memory data structures for MultiXact control
186 : */
187 : static SlruCtlData MultiXactOffsetCtlData;
188 : static SlruCtlData MultiXactMemberCtlData;
189 :
190 : #define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
191 : #define MultiXactMemberCtl (&MultiXactMemberCtlData)
192 :
193 : /*
194 : * MultiXact state shared across all backends. All this state is protected
195 : * by MultiXactGenLock. (We also use SLRU bank's lock of MultiXactOffset and
196 : * MultiXactMember to guard accesses to the two sets of SLRU buffers. For
197 : * concurrency's sake, we avoid holding more than one of these locks at a
198 : * time.)
199 : */
200 : typedef struct MultiXactStateData
201 : {
202 : /* next-to-be-assigned MultiXactId */
203 : MultiXactId nextMXact;
204 :
205 : /* next-to-be-assigned offset */
206 : MultiXactOffset nextOffset;
207 :
208 : /* Have we completed multixact startup? */
209 : bool finishedStartup;
210 :
211 : /*
212 : * Oldest multixact that is still potentially referenced by a relation.
213 : * Anything older than this should not be consulted. These values are
214 : * updated by vacuum.
215 : */
216 : MultiXactId oldestMultiXactId;
217 : Oid oldestMultiXactDB;
218 :
219 : /*
220 : * Oldest multixact offset that is potentially referenced by a multixact
221 : * referenced by a relation. We don't always know this value, so there's
222 : * a flag here to indicate whether or not we currently do.
223 : */
224 : MultiXactOffset oldestOffset;
225 : bool oldestOffsetKnown;
226 :
227 : /* support for anti-wraparound measures */
228 : MultiXactId multiVacLimit;
229 : MultiXactId multiWarnLimit;
230 : MultiXactId multiStopLimit;
231 : MultiXactId multiWrapLimit;
232 :
233 : /* support for members anti-wraparound measures */
234 : MultiXactOffset offsetStopLimit; /* known if oldestOffsetKnown */
235 :
236 : /*
237 : * This is used to sleep until a multixact offset is written when we want
238 : * to create the next one.
239 : */
240 : ConditionVariable nextoff_cv;
241 :
242 : /*
243 : * Per-backend data starts here. We have two arrays stored in the area
244 : * immediately following the MultiXactStateData struct. Each is indexed by
245 : * ProcNumber.
246 : *
247 : * In both arrays, there's a slot for all normal backends
248 : * (0..MaxBackends-1) followed by a slot for max_prepared_xacts prepared
249 : * transactions.
250 : *
251 : * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
252 : * transaction(s) could possibly be a member of, or InvalidMultiXactId
253 : * when the backend has no live transaction that could possibly be a
254 : * member of a MultiXact. Each backend sets its entry to the current
255 : * nextMXact counter just before first acquiring a shared lock in a given
256 : * transaction, and clears it at transaction end. (This works because only
257 : * during or after acquiring a shared lock could an XID possibly become a
258 : * member of a MultiXact, and that MultiXact would have to be created
259 : * during or after the lock acquisition.)
260 : *
261 : * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
262 : * current transaction(s) think is potentially live, or InvalidMultiXactId
263 : * when not in a transaction or not in a transaction that's paid any
264 : * attention to MultiXacts yet. This is computed when first needed in a
265 : * given transaction, and cleared at transaction end. We can compute it
266 : * as the minimum of the valid OldestMemberMXactId[] entries at the time
267 : * we compute it (using nextMXact if none are valid). Each backend is
268 : * required not to attempt to access any SLRU data for MultiXactIds older
269 : * than its own OldestVisibleMXactId[] setting; this is necessary because
270 : * the relevant SLRU data can be concurrently truncated away.
271 : *
272 : * The oldest valid value among all of the OldestMemberMXactId[] and
273 : * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
274 : * possible value still having any live member transaction -- OldestMxact.
275 : * Any value older than that is typically removed from tuple headers, or
276 : * "frozen" via being replaced with a new xmax. VACUUM can sometimes even
277 : * remove an individual MultiXact xmax whose value is >= its OldestMxact
278 : * cutoff, though typically only when no individual member XID is still
279 : * running. See FreezeMultiXactId for full details.
280 : *
281 : * Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
282 : * or the oldest extant Multi remaining in the table is used as the new
283 : * pg_class.relminmxid value (whichever is earlier). The minimum of all
284 : * relminmxid values in each database is stored in pg_database.datminmxid.
285 : * In turn, the minimum of all of those values is stored in pg_control.
286 : * This is used as the truncation point for pg_multixact when unneeded
287 : * segments get removed by vac_truncate_clog() during vacuuming.
288 : */
289 : MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
290 : } MultiXactStateData;
291 :
292 : /*
293 : * Size of OldestMemberMXactId and OldestVisibleMXactId arrays.
294 : */
295 : #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
296 :
297 : /* Pointers to the state data in shared memory */
298 : static MultiXactStateData *MultiXactState;
299 : static MultiXactId *OldestMemberMXactId;
300 : static MultiXactId *OldestVisibleMXactId;
301 :
302 :
303 : /*
304 : * Definitions for the backend-local MultiXactId cache.
305 : *
306 : * We use this cache to store known MultiXacts, so we don't need to go to
307 : * SLRU areas every time.
308 : *
309 : * The cache lasts for the duration of a single transaction, the rationale
310 : * for this being that most entries will contain our own TransactionId and
311 : * so they will be uninteresting by the time our next transaction starts.
312 : * (XXX not clear that this is correct --- other members of the MultiXact
313 : * could hang around longer than we did. However, it's not clear what a
314 : * better policy for flushing old cache entries would be.) FIXME actually
315 : * this is plain wrong now that multixact's may contain update Xids.
316 : *
317 : * We allocate the cache entries in a memory context that is deleted at
318 : * transaction end, so we don't need to do retail freeing of entries.
319 : */
320 : typedef struct mXactCacheEnt
321 : {
322 : MultiXactId multi;
323 : int nmembers;
324 : dlist_node node;
325 : MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
326 : } mXactCacheEnt;
327 :
328 : #define MAX_CACHE_ENTRIES 256
329 : static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
330 : static MemoryContext MXactContext = NULL;
331 :
332 : #ifdef MULTIXACT_DEBUG
333 : #define debug_elog2(a,b) elog(a,b)
334 : #define debug_elog3(a,b,c) elog(a,b,c)
335 : #define debug_elog4(a,b,c,d) elog(a,b,c,d)
336 : #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
337 : #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
338 : #else
339 : #define debug_elog2(a,b)
340 : #define debug_elog3(a,b,c)
341 : #define debug_elog4(a,b,c,d)
342 : #define debug_elog5(a,b,c,d,e)
343 : #define debug_elog6(a,b,c,d,e,f)
344 : #endif
345 :
346 : /* internal MultiXactId management */
347 : static void MultiXactIdSetOldestVisible(void);
348 : static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
349 : int nmembers, MultiXactMember *members);
350 : static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
351 :
352 : /* MultiXact cache management */
353 : static int mxactMemberComparator(const void *arg1, const void *arg2);
354 : static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
355 : static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
356 : static void mXactCachePut(MultiXactId multi, int nmembers,
357 : MultiXactMember *members);
358 :
359 : static char *mxstatus_to_string(MultiXactStatus status);
360 :
361 : /* management of SLRU infrastructure */
362 : static int ZeroMultiXactOffsetPage(int64 pageno, bool writeXlog);
363 : static int ZeroMultiXactMemberPage(int64 pageno, bool writeXlog);
364 : static bool MultiXactOffsetPagePrecedes(int64 page1, int64 page2);
365 : static bool MultiXactMemberPagePrecedes(int64 page1, int64 page2);
366 : static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
367 : MultiXactOffset offset2);
368 : static void ExtendMultiXactOffset(MultiXactId multi);
369 : static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
370 : static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
371 : MultiXactOffset start, uint32 distance);
372 : static bool SetOffsetVacuumLimit(bool is_startup);
373 : static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
374 : static void WriteMZeroPageXlogRec(int64 pageno, uint8 info);
375 : static void WriteMTruncateXlogRec(Oid oldestMultiDB,
376 : MultiXactId startTruncOff,
377 : MultiXactId endTruncOff,
378 : MultiXactOffset startTruncMemb,
379 : MultiXactOffset endTruncMemb);
380 :
381 :
382 : /*
383 : * MultiXactIdCreate
384 : * Construct a MultiXactId representing two TransactionIds.
385 : *
386 : * The two XIDs must be different, or be requesting different statuses.
387 : *
388 : * NB - we don't worry about our local MultiXactId cache here, because that
389 : * is handled by the lower-level routines.
390 : */
391 : MultiXactId
392 2046 : MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
393 : TransactionId xid2, MultiXactStatus status2)
394 : {
395 : MultiXactId newMulti;
396 : MultiXactMember members[2];
397 :
398 : Assert(TransactionIdIsValid(xid1));
399 : Assert(TransactionIdIsValid(xid2));
400 :
401 : Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
402 :
403 : /* MultiXactIdSetOldestMember() must have been called already. */
404 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
405 :
406 : /*
407 : * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
408 : * are still running. In typical usage, xid2 will be our own XID and the
409 : * caller just did a check on xid1, so it'd be wasted effort.
410 : */
411 :
412 2046 : members[0].xid = xid1;
413 2046 : members[0].status = status1;
414 2046 : members[1].xid = xid2;
415 2046 : members[1].status = status2;
416 :
417 2046 : newMulti = MultiXactIdCreateFromMembers(2, members);
418 :
419 : debug_elog3(DEBUG2, "Create: %s",
420 : mxid_to_string(newMulti, 2, members));
421 :
422 2046 : return newMulti;
423 : }
424 :
425 : /*
426 : * MultiXactIdExpand
427 : * Add a TransactionId to a pre-existing MultiXactId.
428 : *
429 : * If the TransactionId is already a member of the passed MultiXactId with the
430 : * same status, just return it as-is.
431 : *
432 : * Note that we do NOT actually modify the membership of a pre-existing
433 : * MultiXactId; instead we create a new one. This is necessary to avoid
434 : * a race condition against code trying to wait for one MultiXactId to finish;
435 : * see notes in heapam.c.
436 : *
437 : * NB - we don't worry about our local MultiXactId cache here, because that
438 : * is handled by the lower-level routines.
439 : *
440 : * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
441 : * one upgraded by pg_upgrade from a cluster older than this feature) are not
442 : * passed in.
443 : */
444 : MultiXactId
445 196 : MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
446 : {
447 : MultiXactId newMulti;
448 : MultiXactMember *members;
449 : MultiXactMember *newMembers;
450 : int nmembers;
451 : int i;
452 : int j;
453 :
454 : Assert(MultiXactIdIsValid(multi));
455 : Assert(TransactionIdIsValid(xid));
456 :
457 : /* MultiXactIdSetOldestMember() must have been called already. */
458 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
459 :
460 : debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
461 : multi, xid, mxstatus_to_string(status));
462 :
463 : /*
464 : * Note: we don't allow for old multis here. The reason is that the only
465 : * caller of this function does a check that the multixact is no longer
466 : * running.
467 : */
468 196 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
469 :
470 196 : if (nmembers < 0)
471 : {
472 : MultiXactMember member;
473 :
474 : /*
475 : * The MultiXactId is obsolete. This can only happen if all the
476 : * MultiXactId members stop running between the caller checking and
477 : * passing it to us. It would be better to return that fact to the
478 : * caller, but it would complicate the API and it's unlikely to happen
479 : * too often, so just deal with it by creating a singleton MultiXact.
480 : */
481 0 : member.xid = xid;
482 0 : member.status = status;
483 0 : newMulti = MultiXactIdCreateFromMembers(1, &member);
484 :
485 : debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
486 : multi, newMulti);
487 0 : return newMulti;
488 : }
489 :
490 : /*
491 : * If the TransactionId is already a member of the MultiXactId with the
492 : * same status, just return the existing MultiXactId.
493 : */
494 608 : for (i = 0; i < nmembers; i++)
495 : {
496 412 : if (TransactionIdEquals(members[i].xid, xid) &&
497 108 : (members[i].status == status))
498 : {
499 : debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
500 : xid, multi);
501 0 : pfree(members);
502 0 : return multi;
503 : }
504 : }
505 :
506 : /*
507 : * Determine which of the members of the MultiXactId are still of
508 : * interest. This is any running transaction, and also any transaction
509 : * that grabbed something stronger than just a lock and was committed. (An
510 : * update that aborted is of no interest here; and having more than one
511 : * update Xid in a multixact would cause errors elsewhere.)
512 : *
513 : * Removing dead members is not just an optimization: freezing of tuples
514 : * whose Xmax are multis depends on this behavior.
515 : *
516 : * Note we have the same race condition here as above: j could be 0 at the
517 : * end of the loop.
518 : */
519 : newMembers = (MultiXactMember *)
520 196 : palloc(sizeof(MultiXactMember) * (nmembers + 1));
521 :
522 608 : for (i = 0, j = 0; i < nmembers; i++)
523 : {
524 412 : if (TransactionIdIsInProgress(members[i].xid) ||
525 90 : (ISUPDATE_from_mxstatus(members[i].status) &&
526 12 : TransactionIdDidCommit(members[i].xid)))
527 : {
528 334 : newMembers[j].xid = members[i].xid;
529 334 : newMembers[j++].status = members[i].status;
530 : }
531 : }
532 :
533 196 : newMembers[j].xid = xid;
534 196 : newMembers[j++].status = status;
535 196 : newMulti = MultiXactIdCreateFromMembers(j, newMembers);
536 :
537 196 : pfree(members);
538 196 : pfree(newMembers);
539 :
540 : debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
541 :
542 196 : return newMulti;
543 : }
544 :
545 : /*
546 : * MultiXactIdIsRunning
547 : * Returns whether a MultiXactId is "running".
548 : *
549 : * We return true if at least one member of the given MultiXactId is still
550 : * running. Note that a "false" result is certain not to change,
551 : * because it is not legal to add members to an existing MultiXactId.
552 : *
553 : * Caller is expected to have verified that the multixact does not come from
554 : * a pg_upgraded share-locked tuple.
555 : */
556 : bool
557 1940 : MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
558 : {
559 : MultiXactMember *members;
560 : int nmembers;
561 : int i;
562 :
563 : debug_elog3(DEBUG2, "IsRunning %u?", multi);
564 :
565 : /*
566 : * "false" here means we assume our callers have checked that the given
567 : * multi cannot possibly come from a pg_upgraded database.
568 : */
569 1940 : nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
570 :
571 1940 : if (nmembers <= 0)
572 : {
573 : debug_elog2(DEBUG2, "IsRunning: no members");
574 1310 : return false;
575 : }
576 :
577 : /*
578 : * Checking for myself is cheap compared to looking in shared memory;
579 : * return true if any live subtransaction of the current top-level
580 : * transaction is a member.
581 : *
582 : * This is not needed for correctness, it's just a fast path.
583 : */
584 1454 : for (i = 0; i < nmembers; i++)
585 : {
586 1134 : if (TransactionIdIsCurrentTransactionId(members[i].xid))
587 : {
588 : debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
589 310 : pfree(members);
590 310 : return true;
591 : }
592 : }
593 :
594 : /*
595 : * This could be made faster by having another entry point in procarray.c,
596 : * walking the PGPROC array only once for all the members. But in most
597 : * cases nmembers should be small enough that it doesn't much matter.
598 : */
599 562 : for (i = 0; i < nmembers; i++)
600 : {
601 468 : if (TransactionIdIsInProgress(members[i].xid))
602 : {
603 : debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
604 : i, members[i].xid);
605 226 : pfree(members);
606 226 : return true;
607 : }
608 : }
609 :
610 94 : pfree(members);
611 :
612 : debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
613 :
614 94 : return false;
615 : }
616 :
617 : /*
618 : * MultiXactIdSetOldestMember
619 : * Save the oldest MultiXactId this transaction could be a member of.
620 : *
621 : * We set the OldestMemberMXactId for a given transaction the first time it's
622 : * going to do some operation that might require a MultiXactId (tuple lock,
623 : * update or delete). We need to do this even if we end up using a
624 : * TransactionId instead of a MultiXactId, because there is a chance that
625 : * another transaction would add our XID to a MultiXactId.
626 : *
627 : * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
628 : * be called just before doing any such possibly-MultiXactId-able operation.
629 : */
630 : void
631 3477404 : MultiXactIdSetOldestMember(void)
632 : {
633 3477404 : if (!MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]))
634 : {
635 : MultiXactId nextMXact;
636 :
637 : /*
638 : * You might think we don't need to acquire a lock here, since
639 : * fetching and storing of TransactionIds is probably atomic, but in
640 : * fact we do: suppose we pick up nextMXact and then lose the CPU for
641 : * a long time. Someone else could advance nextMXact, and then
642 : * another someone else could compute an OldestVisibleMXactId that
643 : * would be after the value we are going to store when we get control
644 : * back. Which would be wrong.
645 : *
646 : * Note that a shared lock is sufficient, because it's enough to stop
647 : * someone from advancing nextMXact; and nobody else could be trying
648 : * to write to our OldestMember entry, only reading (and we assume
649 : * storing it is atomic.)
650 : */
651 120466 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
652 :
653 : /*
654 : * We have to beware of the possibility that nextMXact is in the
655 : * wrapped-around state. We don't fix the counter itself here, but we
656 : * must be sure to store a valid value in our array entry.
657 : */
658 120466 : nextMXact = MultiXactState->nextMXact;
659 120466 : if (nextMXact < FirstMultiXactId)
660 0 : nextMXact = FirstMultiXactId;
661 :
662 120466 : OldestMemberMXactId[MyProcNumber] = nextMXact;
663 :
664 120466 : LWLockRelease(MultiXactGenLock);
665 :
666 : debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
667 : MyProcNumber, nextMXact);
668 : }
669 3477404 : }
670 :
671 : /*
672 : * MultiXactIdSetOldestVisible
673 : * Save the oldest MultiXactId this transaction considers possibly live.
674 : *
675 : * We set the OldestVisibleMXactId for a given transaction the first time
676 : * it's going to inspect any MultiXactId. Once we have set this, we are
677 : * guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
678 : * won't be truncated away.
679 : *
680 : * The value to set is the oldest of nextMXact and all the valid per-backend
681 : * OldestMemberMXactId[] entries. Because of the locking we do, we can be
682 : * certain that no subsequent call to MultiXactIdSetOldestMember can set
683 : * an OldestMemberMXactId[] entry older than what we compute here. Therefore
684 : * there is no live transaction, now or later, that can be a member of any
685 : * MultiXactId older than the OldestVisibleMXactId we compute here.
686 : */
687 : static void
688 1788 : MultiXactIdSetOldestVisible(void)
689 : {
690 1788 : if (!MultiXactIdIsValid(OldestVisibleMXactId[MyProcNumber]))
691 : {
692 : MultiXactId oldestMXact;
693 : int i;
694 :
695 462 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
696 :
697 : /*
698 : * We have to beware of the possibility that nextMXact is in the
699 : * wrapped-around state. We don't fix the counter itself here, but we
700 : * must be sure to store a valid value in our array entry.
701 : */
702 462 : oldestMXact = MultiXactState->nextMXact;
703 462 : if (oldestMXact < FirstMultiXactId)
704 0 : oldestMXact = FirstMultiXactId;
705 :
706 55974 : for (i = 0; i < MaxOldestSlot; i++)
707 : {
708 55512 : MultiXactId thisoldest = OldestMemberMXactId[i];
709 :
710 55990 : if (MultiXactIdIsValid(thisoldest) &&
711 478 : MultiXactIdPrecedes(thisoldest, oldestMXact))
712 228 : oldestMXact = thisoldest;
713 : }
714 :
715 462 : OldestVisibleMXactId[MyProcNumber] = oldestMXact;
716 :
717 462 : LWLockRelease(MultiXactGenLock);
718 :
719 : debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
720 : MyProcNumber, oldestMXact);
721 : }
722 1788 : }
723 :
724 : /*
725 : * ReadNextMultiXactId
726 : * Return the next MultiXactId to be assigned, but don't allocate it
727 : */
728 : MultiXactId
729 43932 : ReadNextMultiXactId(void)
730 : {
731 : MultiXactId mxid;
732 :
733 : /* XXX we could presumably do this without a lock. */
734 43932 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
735 43932 : mxid = MultiXactState->nextMXact;
736 43932 : LWLockRelease(MultiXactGenLock);
737 :
738 43932 : if (mxid < FirstMultiXactId)
739 0 : mxid = FirstMultiXactId;
740 :
741 43932 : return mxid;
742 : }
743 :
744 : /*
745 : * ReadMultiXactIdRange
746 : * Get the range of IDs that may still be referenced by a relation.
747 : */
748 : void
749 2956 : ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
750 : {
751 2956 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
752 2956 : *oldest = MultiXactState->oldestMultiXactId;
753 2956 : *next = MultiXactState->nextMXact;
754 2956 : LWLockRelease(MultiXactGenLock);
755 :
756 2956 : if (*oldest < FirstMultiXactId)
757 0 : *oldest = FirstMultiXactId;
758 2956 : if (*next < FirstMultiXactId)
759 0 : *next = FirstMultiXactId;
760 2956 : }
761 :
762 :
763 : /*
764 : * MultiXactIdCreateFromMembers
765 : * Make a new MultiXactId from the specified set of members
766 : *
767 : * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
768 : * given TransactionIds as members. Returns the newly created MultiXactId.
769 : *
770 : * NB: the passed members[] array will be sorted in-place.
771 : */
772 : MultiXactId
773 2244 : MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
774 : {
775 : MultiXactId multi;
776 : MultiXactOffset offset;
777 : xl_multixact_create xlrec;
778 :
779 : debug_elog3(DEBUG2, "Create: %s",
780 : mxid_to_string(InvalidMultiXactId, nmembers, members));
781 :
782 : /*
783 : * See if the same set of members already exists in our cache; if so, just
784 : * re-use that MultiXactId. (Note: it might seem that looking in our
785 : * cache is insufficient, and we ought to search disk to see if a
786 : * duplicate definition already exists. But since we only ever create
787 : * MultiXacts containing our own XID, in most cases any such MultiXacts
788 : * were in fact created by us, and so will be in our cache. There are
789 : * corner cases where someone else added us to a MultiXact without our
790 : * knowledge, but it's not worth checking for.)
791 : */
792 2244 : multi = mXactCacheGetBySet(nmembers, members);
793 2244 : if (MultiXactIdIsValid(multi))
794 : {
795 : debug_elog2(DEBUG2, "Create: in cache!");
796 1694 : return multi;
797 : }
798 :
799 : /* Verify that there is a single update Xid among the given members. */
800 : {
801 : int i;
802 550 : bool has_update = false;
803 :
804 1776 : for (i = 0; i < nmembers; i++)
805 : {
806 1226 : if (ISUPDATE_from_mxstatus(members[i].status))
807 : {
808 260 : if (has_update)
809 0 : elog(ERROR, "new multixact has more than one updating member: %s",
810 : mxid_to_string(InvalidMultiXactId, nmembers, members));
811 260 : has_update = true;
812 : }
813 : }
814 : }
815 :
816 : /*
817 : * Assign the MXID and offsets range to use, and make sure there is space
818 : * in the OFFSETs and MEMBERs files. NB: this routine does
819 : * START_CRIT_SECTION().
820 : *
821 : * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
822 : * that we've called MultiXactIdSetOldestMember here. This is because
823 : * this routine is used in some places to create new MultiXactIds of which
824 : * the current backend is not a member, notably during freezing of multis
825 : * in vacuum. During vacuum, in particular, it would be unacceptable to
826 : * keep OldestMulti set, in case it runs for long.
827 : */
828 550 : multi = GetNewMultiXactId(nmembers, &offset);
829 :
830 : /* Make an XLOG entry describing the new MXID. */
831 550 : xlrec.mid = multi;
832 550 : xlrec.moff = offset;
833 550 : xlrec.nmembers = nmembers;
834 :
835 : /*
836 : * XXX Note: there's a lot of padding space in MultiXactMember. We could
837 : * find a more compact representation of this Xlog record -- perhaps all
838 : * the status flags in one XLogRecData, then all the xids in another one?
839 : * Not clear that it's worth the trouble though.
840 : */
841 550 : XLogBeginInsert();
842 550 : XLogRegisterData((char *) (&xlrec), SizeOfMultiXactCreate);
843 550 : XLogRegisterData((char *) members, nmembers * sizeof(MultiXactMember));
844 :
845 550 : (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
846 :
847 : /* Now enter the information into the OFFSETs and MEMBERs logs */
848 550 : RecordNewMultiXact(multi, offset, nmembers, members);
849 :
850 : /* Done with critical section */
851 550 : END_CRIT_SECTION();
852 :
853 : /* Store the new MultiXactId in the local cache, too */
854 550 : mXactCachePut(multi, nmembers, members);
855 :
856 : debug_elog2(DEBUG2, "Create: all done");
857 :
858 550 : return multi;
859 : }
860 :
861 : /*
862 : * RecordNewMultiXact
863 : * Write info about a new multixact into the offsets and members files
864 : *
865 : * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
866 : * use it.
867 : */
868 : static void
869 554 : RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
870 : int nmembers, MultiXactMember *members)
871 : {
872 : int64 pageno;
873 : int64 prev_pageno;
874 : int entryno;
875 : int slotno;
876 : MultiXactOffset *offptr;
877 : int i;
878 : LWLock *lock;
879 554 : LWLock *prevlock = NULL;
880 :
881 554 : pageno = MultiXactIdToOffsetPage(multi);
882 554 : entryno = MultiXactIdToOffsetEntry(multi);
883 :
884 554 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
885 554 : LWLockAcquire(lock, LW_EXCLUSIVE);
886 :
887 : /*
888 : * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
889 : * to complain about if there's any I/O error. This is kinda bogus, but
890 : * since the errors will always give the full pathname, it should be clear
891 : * enough that a MultiXactId is really involved. Perhaps someday we'll
892 : * take the trouble to generalize the slru.c error reporting code.
893 : */
894 554 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
895 554 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
896 554 : offptr += entryno;
897 :
898 554 : *offptr = offset;
899 :
900 554 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
901 :
902 : /* Release MultiXactOffset SLRU lock. */
903 554 : LWLockRelease(lock);
904 :
905 : /*
906 : * If anybody was waiting to know the offset of this multixact ID we just
907 : * wrote, they can read it now, so wake them up.
908 : */
909 554 : ConditionVariableBroadcast(&MultiXactState->nextoff_cv);
910 :
911 554 : prev_pageno = -1;
912 :
913 1788 : for (i = 0; i < nmembers; i++, offset++)
914 : {
915 : TransactionId *memberptr;
916 : uint32 *flagsptr;
917 : uint32 flagsval;
918 : int bshift;
919 : int flagsoff;
920 : int memberoff;
921 :
922 : Assert(members[i].status <= MultiXactStatusUpdate);
923 :
924 1234 : pageno = MXOffsetToMemberPage(offset);
925 1234 : memberoff = MXOffsetToMemberOffset(offset);
926 1234 : flagsoff = MXOffsetToFlagsOffset(offset);
927 1234 : bshift = MXOffsetToFlagsBitShift(offset);
928 :
929 1234 : if (pageno != prev_pageno)
930 : {
931 : /*
932 : * MultiXactMember SLRU page is changed so check if this new page
933 : * fall into the different SLRU bank then release the old bank's
934 : * lock and acquire lock on the new bank.
935 : */
936 554 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
937 554 : if (lock != prevlock)
938 : {
939 554 : if (prevlock != NULL)
940 0 : LWLockRelease(prevlock);
941 :
942 554 : LWLockAcquire(lock, LW_EXCLUSIVE);
943 554 : prevlock = lock;
944 : }
945 554 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
946 554 : prev_pageno = pageno;
947 : }
948 :
949 1234 : memberptr = (TransactionId *)
950 1234 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
951 :
952 1234 : *memberptr = members[i].xid;
953 :
954 1234 : flagsptr = (uint32 *)
955 1234 : (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
956 :
957 1234 : flagsval = *flagsptr;
958 1234 : flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
959 1234 : flagsval |= (members[i].status << bshift);
960 1234 : *flagsptr = flagsval;
961 :
962 1234 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
963 : }
964 :
965 554 : if (prevlock != NULL)
966 554 : LWLockRelease(prevlock);
967 554 : }
968 :
969 : /*
970 : * GetNewMultiXactId
971 : * Get the next MultiXactId.
972 : *
973 : * Also, reserve the needed amount of space in the "members" area. The
974 : * starting offset of the reserved space is returned in *offset.
975 : *
976 : * This may generate XLOG records for expansion of the offsets and/or members
977 : * files. Unfortunately, we have to do that while holding MultiXactGenLock
978 : * to avoid race conditions --- the XLOG record for zeroing a page must appear
979 : * before any backend can possibly try to store data in that page!
980 : *
981 : * We start a critical section before advancing the shared counters. The
982 : * caller must end the critical section after writing SLRU data.
983 : */
984 : static MultiXactId
985 550 : GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
986 : {
987 : MultiXactId result;
988 : MultiXactOffset nextOffset;
989 :
990 : debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
991 :
992 : /* safety check, we should never get this far in a HS standby */
993 550 : if (RecoveryInProgress())
994 0 : elog(ERROR, "cannot assign MultiXactIds during recovery");
995 :
996 550 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
997 :
998 : /* Handle wraparound of the nextMXact counter */
999 550 : if (MultiXactState->nextMXact < FirstMultiXactId)
1000 0 : MultiXactState->nextMXact = FirstMultiXactId;
1001 :
1002 : /* Assign the MXID */
1003 550 : result = MultiXactState->nextMXact;
1004 :
1005 : /*----------
1006 : * Check to see if it's safe to assign another MultiXactId. This protects
1007 : * against catastrophic data loss due to multixact wraparound. The basic
1008 : * rules are:
1009 : *
1010 : * If we're past multiVacLimit or the safe threshold for member storage
1011 : * space, or we don't know what the safe threshold for member storage is,
1012 : * start trying to force autovacuum cycles.
1013 : * If we're past multiWarnLimit, start issuing warnings.
1014 : * If we're past multiStopLimit, refuse to create new MultiXactIds.
1015 : *
1016 : * Note these are pretty much the same protections in GetNewTransactionId.
1017 : *----------
1018 : */
1019 550 : if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
1020 : {
1021 : /*
1022 : * For safety's sake, we release MultiXactGenLock while sending
1023 : * signals, warnings, etc. This is not so much because we care about
1024 : * preserving concurrency in this situation, as to avoid any
1025 : * possibility of deadlock while doing get_database_name(). First,
1026 : * copy all the shared values we'll need in this path.
1027 : */
1028 0 : MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
1029 0 : MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
1030 0 : MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
1031 0 : Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
1032 :
1033 0 : LWLockRelease(MultiXactGenLock);
1034 :
1035 0 : if (IsUnderPostmaster &&
1036 0 : !MultiXactIdPrecedes(result, multiStopLimit))
1037 : {
1038 0 : char *oldest_datname = get_database_name(oldest_datoid);
1039 :
1040 : /*
1041 : * Immediately kick autovacuum into action as we're already in
1042 : * ERROR territory.
1043 : */
1044 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1045 :
1046 : /* complain even if that DB has disappeared */
1047 0 : if (oldest_datname)
1048 0 : ereport(ERROR,
1049 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1050 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database \"%s\"",
1051 : oldest_datname),
1052 : errhint("Execute a database-wide VACUUM in that database.\n"
1053 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1054 : else
1055 0 : ereport(ERROR,
1056 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1057 : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database with OID %u",
1058 : oldest_datoid),
1059 : errhint("Execute a database-wide VACUUM in that database.\n"
1060 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1061 : }
1062 :
1063 : /*
1064 : * To avoid swamping the postmaster with signals, we issue the autovac
1065 : * request only once per 64K multis generated. This still gives
1066 : * plenty of chances before we get into real trouble.
1067 : */
1068 0 : if (IsUnderPostmaster && (result % 65536) == 0)
1069 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1070 :
1071 0 : if (!MultiXactIdPrecedes(result, multiWarnLimit))
1072 : {
1073 0 : char *oldest_datname = get_database_name(oldest_datoid);
1074 :
1075 : /* complain even if that DB has disappeared */
1076 0 : if (oldest_datname)
1077 0 : ereport(WARNING,
1078 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1079 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1080 : multiWrapLimit - result,
1081 : oldest_datname,
1082 : multiWrapLimit - result),
1083 : errhint("Execute a database-wide VACUUM in that database.\n"
1084 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1085 : else
1086 0 : ereport(WARNING,
1087 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1088 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1089 : multiWrapLimit - result,
1090 : oldest_datoid,
1091 : multiWrapLimit - result),
1092 : errhint("Execute a database-wide VACUUM in that database.\n"
1093 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1094 : }
1095 :
1096 : /* Re-acquire lock and start over */
1097 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1098 0 : result = MultiXactState->nextMXact;
1099 0 : if (result < FirstMultiXactId)
1100 0 : result = FirstMultiXactId;
1101 : }
1102 :
1103 : /* Make sure there is room for the MXID in the file. */
1104 550 : ExtendMultiXactOffset(result);
1105 :
1106 : /*
1107 : * Reserve the members space, similarly to above. Also, be careful not to
1108 : * return zero as the starting offset for any multixact. See
1109 : * GetMultiXactIdMembers() for motivation.
1110 : */
1111 550 : nextOffset = MultiXactState->nextOffset;
1112 550 : if (nextOffset == 0)
1113 : {
1114 20 : *offset = 1;
1115 20 : nmembers++; /* allocate member slot 0 too */
1116 : }
1117 : else
1118 530 : *offset = nextOffset;
1119 :
1120 : /*----------
1121 : * Protect against overrun of the members space as well, with the
1122 : * following rules:
1123 : *
1124 : * If we're past offsetStopLimit, refuse to generate more multis.
1125 : * If we're close to offsetStopLimit, emit a warning.
1126 : *
1127 : * Arbitrarily, we start emitting warnings when we're 20 segments or less
1128 : * from offsetStopLimit.
1129 : *
1130 : * Note we haven't updated the shared state yet, so if we fail at this
1131 : * point, the multixact ID we grabbed can still be used by the next guy.
1132 : *
1133 : * Note that there is no point in forcing autovacuum runs here: the
1134 : * multixact freeze settings would have to be reduced for that to have any
1135 : * effect.
1136 : *----------
1137 : */
1138 : #define OFFSET_WARN_SEGMENTS 20
1139 1100 : if (MultiXactState->oldestOffsetKnown &&
1140 550 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit, nextOffset,
1141 : nmembers))
1142 : {
1143 : /* see comment in the corresponding offsets wraparound case */
1144 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1145 :
1146 0 : ereport(ERROR,
1147 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1148 : errmsg("multixact \"members\" limit exceeded"),
1149 : errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
1150 : "This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
1151 : MultiXactState->offsetStopLimit - nextOffset - 1,
1152 : nmembers,
1153 : MultiXactState->offsetStopLimit - nextOffset - 1),
1154 : 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.",
1155 : MultiXactState->oldestMultiXactDB)));
1156 : }
1157 :
1158 : /*
1159 : * Check whether we should kick autovacuum into action, to prevent members
1160 : * wraparound. NB we use a much larger window to trigger autovacuum than
1161 : * just the warning limit. The warning is just a measure of last resort -
1162 : * this is in line with GetNewTransactionId's behaviour.
1163 : */
1164 550 : if (!MultiXactState->oldestOffsetKnown ||
1165 550 : (MultiXactState->nextOffset - MultiXactState->oldestOffset
1166 550 : > MULTIXACT_MEMBER_SAFE_THRESHOLD))
1167 : {
1168 : /*
1169 : * To avoid swamping the postmaster with signals, we issue the autovac
1170 : * request only when crossing a segment boundary. With default
1171 : * compilation settings that's roughly after 50k members. This still
1172 : * gives plenty of chances before we get into real trouble.
1173 : */
1174 0 : if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
1175 0 : (MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
1176 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1177 : }
1178 :
1179 1100 : if (MultiXactState->oldestOffsetKnown &&
1180 550 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit,
1181 : nextOffset,
1182 : nmembers + MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT * OFFSET_WARN_SEGMENTS))
1183 0 : ereport(WARNING,
1184 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1185 : errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
1186 : "database with OID %u must be vacuumed before %d more multixact members are used",
1187 : MultiXactState->offsetStopLimit - nextOffset + nmembers,
1188 : MultiXactState->oldestMultiXactDB,
1189 : MultiXactState->offsetStopLimit - nextOffset + nmembers),
1190 : errhint("Execute a database-wide VACUUM in that database with reduced vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age settings.")));
1191 :
1192 550 : ExtendMultiXactMember(nextOffset, nmembers);
1193 :
1194 : /*
1195 : * Critical section from here until caller has written the data into the
1196 : * just-reserved SLRU space; we don't want to error out with a partly
1197 : * written MultiXact structure. (In particular, failing to write our
1198 : * start offset after advancing nextMXact would effectively corrupt the
1199 : * previous MultiXact.)
1200 : */
1201 550 : START_CRIT_SECTION();
1202 :
1203 : /*
1204 : * Advance counters. As in GetNewTransactionId(), this must not happen
1205 : * until after file extension has succeeded!
1206 : *
1207 : * We don't care about MultiXactId wraparound here; it will be handled by
1208 : * the next iteration. But note that nextMXact may be InvalidMultiXactId
1209 : * or the first value on a segment-beginning page after this routine
1210 : * exits, so anyone else looking at the variable must be prepared to deal
1211 : * with either case. Similarly, nextOffset may be zero, but we won't use
1212 : * that as the actual start offset of the next multixact.
1213 : */
1214 550 : (MultiXactState->nextMXact)++;
1215 :
1216 550 : MultiXactState->nextOffset += nmembers;
1217 :
1218 550 : LWLockRelease(MultiXactGenLock);
1219 :
1220 : debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
1221 550 : return result;
1222 : }
1223 :
1224 : /*
1225 : * GetMultiXactIdMembers
1226 : * Return the set of MultiXactMembers that make up a MultiXactId
1227 : *
1228 : * Return value is the number of members found, or -1 if there are none,
1229 : * and *members is set to a newly palloc'ed array of members. It's the
1230 : * caller's responsibility to free it when done with it.
1231 : *
1232 : * from_pgupgrade must be passed as true if and only if only the multixact
1233 : * corresponds to a value from a tuple that was locked in a 9.2-or-older
1234 : * installation and later pg_upgrade'd (that is, the infomask is
1235 : * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1236 : * can still be running, so we return -1 just like for an empty multixact
1237 : * without any further checking. It would be wrong to try to resolve such a
1238 : * multixact: either the multixact is within the current valid multixact
1239 : * range, in which case the returned result would be bogus, or outside that
1240 : * range, in which case an error would be raised.
1241 : *
1242 : * In all other cases, the passed multixact must be within the known valid
1243 : * range, that is, greater to or equal than oldestMultiXactId, and less than
1244 : * nextMXact. Otherwise, an error is raised.
1245 : *
1246 : * isLockOnly must be set to true if caller is certain that the given multi
1247 : * is used only to lock tuples; can be false without loss of correctness,
1248 : * but passing a true means we can return quickly without checking for
1249 : * old updates.
1250 : */
1251 : int
1252 6008 : GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
1253 : bool from_pgupgrade, bool isLockOnly)
1254 : {
1255 : int64 pageno;
1256 : int64 prev_pageno;
1257 : int entryno;
1258 : int slotno;
1259 : MultiXactOffset *offptr;
1260 : MultiXactOffset offset;
1261 : int length;
1262 : int truelength;
1263 : MultiXactId oldestMXact;
1264 : MultiXactId nextMXact;
1265 : MultiXactId tmpMXact;
1266 : MultiXactOffset nextOffset;
1267 : MultiXactMember *ptr;
1268 : LWLock *lock;
1269 6008 : bool slept = false;
1270 :
1271 : debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1272 :
1273 6008 : if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1274 : {
1275 0 : *members = NULL;
1276 0 : return -1;
1277 : }
1278 :
1279 : /* See if the MultiXactId is in the local cache */
1280 6008 : length = mXactCacheGetById(multi, members);
1281 6008 : if (length >= 0)
1282 : {
1283 : debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1284 : mxid_to_string(multi, length, *members));
1285 4220 : return length;
1286 : }
1287 :
1288 : /* Set our OldestVisibleMXactId[] entry if we didn't already */
1289 1788 : MultiXactIdSetOldestVisible();
1290 :
1291 : /*
1292 : * If we know the multi is used only for locking and not for updates, then
1293 : * we can skip checking if the value is older than our oldest visible
1294 : * multi. It cannot possibly still be running.
1295 : */
1296 3262 : if (isLockOnly &&
1297 1474 : MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyProcNumber]))
1298 : {
1299 : debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1300 1312 : *members = NULL;
1301 1312 : return -1;
1302 : }
1303 :
1304 : /*
1305 : * We check known limits on MultiXact before resorting to the SLRU area.
1306 : *
1307 : * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1308 : * useful; it has already been removed, or will be removed shortly, by
1309 : * truncation. If one is passed, an error is raised.
1310 : *
1311 : * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1312 : * implies undetected ID wraparound has occurred. This raises a hard
1313 : * error.
1314 : *
1315 : * Shared lock is enough here since we aren't modifying any global state.
1316 : * Acquire it just long enough to grab the current counter values. We may
1317 : * need both nextMXact and nextOffset; see below.
1318 : */
1319 476 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1320 :
1321 476 : oldestMXact = MultiXactState->oldestMultiXactId;
1322 476 : nextMXact = MultiXactState->nextMXact;
1323 476 : nextOffset = MultiXactState->nextOffset;
1324 :
1325 476 : LWLockRelease(MultiXactGenLock);
1326 :
1327 476 : if (MultiXactIdPrecedes(multi, oldestMXact))
1328 0 : ereport(ERROR,
1329 : (errcode(ERRCODE_INTERNAL_ERROR),
1330 : errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1331 : multi)));
1332 :
1333 476 : if (!MultiXactIdPrecedes(multi, nextMXact))
1334 0 : ereport(ERROR,
1335 : (errcode(ERRCODE_INTERNAL_ERROR),
1336 : errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1337 : multi)));
1338 :
1339 : /*
1340 : * Find out the offset at which we need to start reading MultiXactMembers
1341 : * and the number of members in the multixact. We determine the latter as
1342 : * the difference between this multixact's starting offset and the next
1343 : * one's. However, there are some corner cases to worry about:
1344 : *
1345 : * 1. This multixact may be the latest one created, in which case there is
1346 : * no next one to look at. In this case the nextOffset value we just
1347 : * saved is the correct endpoint.
1348 : *
1349 : * 2. The next multixact may still be in process of being filled in: that
1350 : * is, another process may have done GetNewMultiXactId but not yet written
1351 : * the offset entry for that ID. In that scenario, it is guaranteed that
1352 : * the offset entry for that multixact exists (because GetNewMultiXactId
1353 : * won't release MultiXactGenLock until it does) but contains zero
1354 : * (because we are careful to pre-zero offset pages). Because
1355 : * GetNewMultiXactId will never return zero as the starting offset for a
1356 : * multixact, when we read zero as the next multixact's offset, we know we
1357 : * have this case. We handle this by sleeping on the condition variable
1358 : * we have just for this; the process in charge will signal the CV as soon
1359 : * as it has finished writing the multixact offset.
1360 : *
1361 : * 3. Because GetNewMultiXactId increments offset zero to offset one to
1362 : * handle case #2, there is an ambiguity near the point of offset
1363 : * wraparound. If we see next multixact's offset is one, is that our
1364 : * multixact's actual endpoint, or did it end at zero with a subsequent
1365 : * increment? We handle this using the knowledge that if the zero'th
1366 : * member slot wasn't filled, it'll contain zero, and zero isn't a valid
1367 : * transaction ID so it can't be a multixact member. Therefore, if we
1368 : * read a zero from the members array, just ignore it.
1369 : *
1370 : * This is all pretty messy, but the mess occurs only in infrequent corner
1371 : * cases, so it seems better than holding the MultiXactGenLock for a long
1372 : * time on every multixact creation.
1373 : */
1374 476 : retry:
1375 476 : pageno = MultiXactIdToOffsetPage(multi);
1376 476 : entryno = MultiXactIdToOffsetEntry(multi);
1377 :
1378 : /* Acquire the bank lock for the page we need. */
1379 476 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1380 476 : LWLockAcquire(lock, LW_EXCLUSIVE);
1381 :
1382 476 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1383 476 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1384 476 : offptr += entryno;
1385 476 : offset = *offptr;
1386 :
1387 : Assert(offset != 0);
1388 :
1389 : /*
1390 : * Use the same increment rule as GetNewMultiXactId(), that is, don't
1391 : * handle wraparound explicitly until needed.
1392 : */
1393 476 : tmpMXact = multi + 1;
1394 :
1395 476 : if (nextMXact == tmpMXact)
1396 : {
1397 : /* Corner case 1: there is no next multixact */
1398 422 : length = nextOffset - offset;
1399 : }
1400 : else
1401 : {
1402 : MultiXactOffset nextMXOffset;
1403 :
1404 : /* handle wraparound if needed */
1405 54 : if (tmpMXact < FirstMultiXactId)
1406 0 : tmpMXact = FirstMultiXactId;
1407 :
1408 54 : prev_pageno = pageno;
1409 :
1410 54 : pageno = MultiXactIdToOffsetPage(tmpMXact);
1411 54 : entryno = MultiXactIdToOffsetEntry(tmpMXact);
1412 :
1413 54 : if (pageno != prev_pageno)
1414 : {
1415 : LWLock *newlock;
1416 :
1417 : /*
1418 : * Since we're going to access a different SLRU page, if this page
1419 : * falls under a different bank, release the old bank's lock and
1420 : * acquire the lock of the new bank.
1421 : */
1422 0 : newlock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1423 0 : if (newlock != lock)
1424 : {
1425 0 : LWLockRelease(lock);
1426 0 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1427 0 : lock = newlock;
1428 : }
1429 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1430 : }
1431 :
1432 54 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1433 54 : offptr += entryno;
1434 54 : nextMXOffset = *offptr;
1435 :
1436 54 : if (nextMXOffset == 0)
1437 : {
1438 : /* Corner case 2: next multixact is still being filled in */
1439 0 : LWLockRelease(lock);
1440 0 : CHECK_FOR_INTERRUPTS();
1441 :
1442 0 : ConditionVariableSleep(&MultiXactState->nextoff_cv,
1443 : WAIT_EVENT_MULTIXACT_CREATION);
1444 0 : slept = true;
1445 0 : goto retry;
1446 : }
1447 :
1448 54 : length = nextMXOffset - offset;
1449 : }
1450 :
1451 476 : LWLockRelease(lock);
1452 476 : lock = NULL;
1453 :
1454 : /*
1455 : * If we slept above, clean up state; it's no longer needed.
1456 : */
1457 476 : if (slept)
1458 0 : ConditionVariableCancelSleep();
1459 :
1460 476 : ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1461 :
1462 476 : truelength = 0;
1463 476 : prev_pageno = -1;
1464 1706 : for (int i = 0; i < length; i++, offset++)
1465 : {
1466 : TransactionId *xactptr;
1467 : uint32 *flagsptr;
1468 : int flagsoff;
1469 : int bshift;
1470 : int memberoff;
1471 :
1472 1230 : pageno = MXOffsetToMemberPage(offset);
1473 1230 : memberoff = MXOffsetToMemberOffset(offset);
1474 :
1475 1230 : if (pageno != prev_pageno)
1476 : {
1477 : LWLock *newlock;
1478 :
1479 : /*
1480 : * Since we're going to access a different SLRU page, if this page
1481 : * falls under a different bank, release the old bank's lock and
1482 : * acquire the lock of the new bank.
1483 : */
1484 476 : newlock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
1485 476 : if (newlock != lock)
1486 : {
1487 476 : if (lock)
1488 0 : LWLockRelease(lock);
1489 476 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1490 476 : lock = newlock;
1491 : }
1492 :
1493 476 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1494 476 : prev_pageno = pageno;
1495 : }
1496 :
1497 1230 : xactptr = (TransactionId *)
1498 1230 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1499 :
1500 1230 : if (!TransactionIdIsValid(*xactptr))
1501 : {
1502 : /* Corner case 3: we must be looking at unused slot zero */
1503 : Assert(offset == 0);
1504 0 : continue;
1505 : }
1506 :
1507 1230 : flagsoff = MXOffsetToFlagsOffset(offset);
1508 1230 : bshift = MXOffsetToFlagsBitShift(offset);
1509 1230 : flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1510 :
1511 1230 : ptr[truelength].xid = *xactptr;
1512 1230 : ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1513 1230 : truelength++;
1514 : }
1515 :
1516 476 : LWLockRelease(lock);
1517 :
1518 : /* A multixid with zero members should not happen */
1519 : Assert(truelength > 0);
1520 :
1521 : /*
1522 : * Copy the result into the local cache.
1523 : */
1524 476 : mXactCachePut(multi, truelength, ptr);
1525 :
1526 : debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1527 : mxid_to_string(multi, truelength, ptr));
1528 476 : *members = ptr;
1529 476 : return truelength;
1530 : }
1531 :
1532 : /*
1533 : * mxactMemberComparator
1534 : * qsort comparison function for MultiXactMember
1535 : *
1536 : * We can't use wraparound comparison for XIDs because that does not respect
1537 : * the triangle inequality! Any old sort order will do.
1538 : */
1539 : static int
1540 3838 : mxactMemberComparator(const void *arg1, const void *arg2)
1541 : {
1542 3838 : MultiXactMember member1 = *(const MultiXactMember *) arg1;
1543 3838 : MultiXactMember member2 = *(const MultiXactMember *) arg2;
1544 :
1545 3838 : if (member1.xid > member2.xid)
1546 52 : return 1;
1547 3786 : if (member1.xid < member2.xid)
1548 3450 : return -1;
1549 336 : if (member1.status > member2.status)
1550 0 : return 1;
1551 336 : if (member1.status < member2.status)
1552 336 : return -1;
1553 0 : return 0;
1554 : }
1555 :
1556 : /*
1557 : * mXactCacheGetBySet
1558 : * returns a MultiXactId from the cache based on the set of
1559 : * TransactionIds that compose it, or InvalidMultiXactId if
1560 : * none matches.
1561 : *
1562 : * This is helpful, for example, if two transactions want to lock a huge
1563 : * table. By using the cache, the second will use the same MultiXactId
1564 : * for the majority of tuples, thus keeping MultiXactId usage low (saving
1565 : * both I/O and wraparound issues).
1566 : *
1567 : * NB: the passed members array will be sorted in-place.
1568 : */
1569 : static MultiXactId
1570 2244 : mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1571 : {
1572 : dlist_iter iter;
1573 :
1574 : debug_elog3(DEBUG2, "CacheGet: looking for %s",
1575 : mxid_to_string(InvalidMultiXactId, nmembers, members));
1576 :
1577 : /* sort the array so comparison is easy */
1578 2244 : qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1579 :
1580 2728 : dclist_foreach(iter, &MXactCache)
1581 : {
1582 2178 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1583 : iter.cur);
1584 :
1585 2178 : if (entry->nmembers != nmembers)
1586 250 : continue;
1587 :
1588 : /*
1589 : * We assume the cache entries are sorted, and that the unused bits in
1590 : * "status" are zeroed.
1591 : */
1592 1928 : if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1593 : {
1594 : debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1595 1694 : dclist_move_head(&MXactCache, iter.cur);
1596 1694 : return entry->multi;
1597 : }
1598 : }
1599 :
1600 : debug_elog2(DEBUG2, "CacheGet: not found :-(");
1601 550 : return InvalidMultiXactId;
1602 : }
1603 :
1604 : /*
1605 : * mXactCacheGetById
1606 : * returns the composing MultiXactMember set from the cache for a
1607 : * given MultiXactId, if present.
1608 : *
1609 : * If successful, *xids is set to the address of a palloc'd copy of the
1610 : * MultiXactMember set. Return value is number of members, or -1 on failure.
1611 : */
1612 : static int
1613 6008 : mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
1614 : {
1615 : dlist_iter iter;
1616 :
1617 : debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1618 :
1619 6828 : dclist_foreach(iter, &MXactCache)
1620 : {
1621 5040 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1622 : iter.cur);
1623 :
1624 5040 : if (entry->multi == multi)
1625 : {
1626 : MultiXactMember *ptr;
1627 : Size size;
1628 :
1629 4220 : size = sizeof(MultiXactMember) * entry->nmembers;
1630 4220 : ptr = (MultiXactMember *) palloc(size);
1631 :
1632 4220 : memcpy(ptr, entry->members, size);
1633 :
1634 : debug_elog3(DEBUG2, "CacheGet: found %s",
1635 : mxid_to_string(multi,
1636 : entry->nmembers,
1637 : entry->members));
1638 :
1639 : /*
1640 : * Note we modify the list while not using a modifiable iterator.
1641 : * This is acceptable only because we exit the iteration
1642 : * immediately afterwards.
1643 : */
1644 4220 : dclist_move_head(&MXactCache, iter.cur);
1645 :
1646 4220 : *members = ptr;
1647 4220 : return entry->nmembers;
1648 : }
1649 : }
1650 :
1651 : debug_elog2(DEBUG2, "CacheGet: not found");
1652 1788 : return -1;
1653 : }
1654 :
1655 : /*
1656 : * mXactCachePut
1657 : * Add a new MultiXactId and its composing set into the local cache.
1658 : */
1659 : static void
1660 1026 : mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1661 : {
1662 : mXactCacheEnt *entry;
1663 :
1664 : debug_elog3(DEBUG2, "CachePut: storing %s",
1665 : mxid_to_string(multi, nmembers, members));
1666 :
1667 1026 : if (MXactContext == NULL)
1668 : {
1669 : /* The cache only lives as long as the current transaction */
1670 : debug_elog2(DEBUG2, "CachePut: initializing memory context");
1671 728 : MXactContext = AllocSetContextCreate(TopTransactionContext,
1672 : "MultiXact cache context",
1673 : ALLOCSET_SMALL_SIZES);
1674 : }
1675 :
1676 : entry = (mXactCacheEnt *)
1677 1026 : MemoryContextAlloc(MXactContext,
1678 1026 : offsetof(mXactCacheEnt, members) +
1679 : nmembers * sizeof(MultiXactMember));
1680 :
1681 1026 : entry->multi = multi;
1682 1026 : entry->nmembers = nmembers;
1683 1026 : memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1684 :
1685 : /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1686 1026 : qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1687 :
1688 1026 : dclist_push_head(&MXactCache, &entry->node);
1689 1026 : if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
1690 : {
1691 : dlist_node *node;
1692 :
1693 0 : node = dclist_tail_node(&MXactCache);
1694 0 : dclist_delete_from(&MXactCache, node);
1695 :
1696 0 : entry = dclist_container(mXactCacheEnt, node, node);
1697 : debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1698 : entry->multi);
1699 :
1700 0 : pfree(entry);
1701 : }
1702 1026 : }
1703 :
1704 : static char *
1705 0 : mxstatus_to_string(MultiXactStatus status)
1706 : {
1707 0 : switch (status)
1708 : {
1709 0 : case MultiXactStatusForKeyShare:
1710 0 : return "keysh";
1711 0 : case MultiXactStatusForShare:
1712 0 : return "sh";
1713 0 : case MultiXactStatusForNoKeyUpdate:
1714 0 : return "fornokeyupd";
1715 0 : case MultiXactStatusForUpdate:
1716 0 : return "forupd";
1717 0 : case MultiXactStatusNoKeyUpdate:
1718 0 : return "nokeyupd";
1719 0 : case MultiXactStatusUpdate:
1720 0 : return "upd";
1721 0 : default:
1722 0 : elog(ERROR, "unrecognized multixact status %d", status);
1723 : return "";
1724 : }
1725 : }
1726 :
1727 : char *
1728 0 : mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1729 : {
1730 : static char *str = NULL;
1731 : StringInfoData buf;
1732 : int i;
1733 :
1734 0 : if (str != NULL)
1735 0 : pfree(str);
1736 :
1737 0 : initStringInfo(&buf);
1738 :
1739 0 : appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1740 : mxstatus_to_string(members[0].status));
1741 :
1742 0 : for (i = 1; i < nmembers; i++)
1743 0 : appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1744 0 : mxstatus_to_string(members[i].status));
1745 :
1746 0 : appendStringInfoChar(&buf, ']');
1747 0 : str = MemoryContextStrdup(TopMemoryContext, buf.data);
1748 0 : pfree(buf.data);
1749 0 : return str;
1750 : }
1751 :
1752 : /*
1753 : * AtEOXact_MultiXact
1754 : * Handle transaction end for MultiXact
1755 : *
1756 : * This is called at top transaction commit or abort (we don't care which).
1757 : */
1758 : void
1759 564086 : AtEOXact_MultiXact(void)
1760 : {
1761 : /*
1762 : * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1763 : * which should only be valid while within a transaction.
1764 : *
1765 : * We assume that storing a MultiXactId is atomic and so we need not take
1766 : * MultiXactGenLock to do this.
1767 : */
1768 564086 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1769 564086 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1770 :
1771 : /*
1772 : * Discard the local MultiXactId cache. Since MXactContext was created as
1773 : * a child of TopTransactionContext, we needn't delete it explicitly.
1774 : */
1775 564086 : MXactContext = NULL;
1776 564086 : dclist_init(&MXactCache);
1777 564086 : }
1778 :
1779 : /*
1780 : * AtPrepare_MultiXact
1781 : * Save multixact state at 2PC transaction prepare
1782 : *
1783 : * In this phase, we only store our OldestMemberMXactId value in the two-phase
1784 : * state file.
1785 : */
1786 : void
1787 694 : AtPrepare_MultiXact(void)
1788 : {
1789 694 : MultiXactId myOldestMember = OldestMemberMXactId[MyProcNumber];
1790 :
1791 694 : if (MultiXactIdIsValid(myOldestMember))
1792 96 : RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
1793 : &myOldestMember, sizeof(MultiXactId));
1794 694 : }
1795 :
1796 : /*
1797 : * PostPrepare_MultiXact
1798 : * Clean up after successful PREPARE TRANSACTION
1799 : */
1800 : void
1801 694 : PostPrepare_MultiXact(TransactionId xid)
1802 : {
1803 : MultiXactId myOldestMember;
1804 :
1805 : /*
1806 : * Transfer our OldestMemberMXactId value to the slot reserved for the
1807 : * prepared transaction.
1808 : */
1809 694 : myOldestMember = OldestMemberMXactId[MyProcNumber];
1810 694 : if (MultiXactIdIsValid(myOldestMember))
1811 : {
1812 96 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(xid, false);
1813 :
1814 : /*
1815 : * Even though storing MultiXactId is atomic, acquire lock to make
1816 : * sure others see both changes, not just the reset of the slot of the
1817 : * current backend. Using a volatile pointer might suffice, but this
1818 : * isn't a hot spot.
1819 : */
1820 96 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1821 :
1822 96 : OldestMemberMXactId[dummyProcNumber] = myOldestMember;
1823 96 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1824 :
1825 96 : LWLockRelease(MultiXactGenLock);
1826 : }
1827 :
1828 : /*
1829 : * We don't need to transfer OldestVisibleMXactId value, because the
1830 : * transaction is not going to be looking at any more multixacts once it's
1831 : * prepared.
1832 : *
1833 : * We assume that storing a MultiXactId is atomic and so we need not take
1834 : * MultiXactGenLock to do this.
1835 : */
1836 694 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1837 :
1838 : /*
1839 : * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1840 : */
1841 694 : MXactContext = NULL;
1842 694 : dclist_init(&MXactCache);
1843 694 : }
1844 :
1845 : /*
1846 : * multixact_twophase_recover
1847 : * Recover the state of a prepared transaction at startup
1848 : */
1849 : void
1850 16 : multixact_twophase_recover(TransactionId xid, uint16 info,
1851 : void *recdata, uint32 len)
1852 : {
1853 16 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(xid, false);
1854 : MultiXactId oldestMember;
1855 :
1856 : /*
1857 : * Get the oldest member XID from the state file record, and set it in the
1858 : * OldestMemberMXactId slot reserved for this prepared transaction.
1859 : */
1860 : Assert(len == sizeof(MultiXactId));
1861 16 : oldestMember = *((MultiXactId *) recdata);
1862 :
1863 16 : OldestMemberMXactId[dummyProcNumber] = oldestMember;
1864 16 : }
1865 :
1866 : /*
1867 : * multixact_twophase_postcommit
1868 : * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1869 : */
1870 : void
1871 104 : multixact_twophase_postcommit(TransactionId xid, uint16 info,
1872 : void *recdata, uint32 len)
1873 : {
1874 104 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(xid, true);
1875 :
1876 : Assert(len == sizeof(MultiXactId));
1877 :
1878 104 : OldestMemberMXactId[dummyProcNumber] = InvalidMultiXactId;
1879 104 : }
1880 :
1881 : /*
1882 : * multixact_twophase_postabort
1883 : * This is actually just the same as the COMMIT case.
1884 : */
1885 : void
1886 44 : multixact_twophase_postabort(TransactionId xid, uint16 info,
1887 : void *recdata, uint32 len)
1888 : {
1889 44 : multixact_twophase_postcommit(xid, info, recdata, len);
1890 44 : }
1891 :
1892 : /*
1893 : * Initialization of shared memory for MultiXact. We use two SLRU areas,
1894 : * thus double memory. Also, reserve space for the shared MultiXactState
1895 : * struct and the per-backend MultiXactId arrays (two of those, too).
1896 : */
1897 : Size
1898 3298 : MultiXactShmemSize(void)
1899 : {
1900 : Size size;
1901 :
1902 : /* We need 2*MaxOldestSlot perBackendXactIds[] entries */
1903 : #define SHARED_MULTIXACT_STATE_SIZE \
1904 : add_size(offsetof(MultiXactStateData, perBackendXactIds), \
1905 : mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1906 :
1907 3298 : size = SHARED_MULTIXACT_STATE_SIZE;
1908 3298 : size = add_size(size, SimpleLruShmemSize(multixact_offset_buffers, 0));
1909 3298 : size = add_size(size, SimpleLruShmemSize(multixact_member_buffers, 0));
1910 :
1911 3298 : return size;
1912 : }
1913 :
1914 : void
1915 1768 : MultiXactShmemInit(void)
1916 : {
1917 : bool found;
1918 :
1919 : debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1920 :
1921 1768 : MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
1922 1768 : MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
1923 :
1924 1768 : SimpleLruInit(MultiXactOffsetCtl,
1925 : "multixact_offset", multixact_offset_buffers, 0,
1926 : "pg_multixact/offsets", LWTRANCHE_MULTIXACTOFFSET_BUFFER,
1927 : LWTRANCHE_MULTIXACTOFFSET_SLRU,
1928 : SYNC_HANDLER_MULTIXACT_OFFSET,
1929 : false);
1930 : SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
1931 1768 : SimpleLruInit(MultiXactMemberCtl,
1932 : "multixact_member", multixact_member_buffers, 0,
1933 : "pg_multixact/members", LWTRANCHE_MULTIXACTMEMBER_BUFFER,
1934 : LWTRANCHE_MULTIXACTMEMBER_SLRU,
1935 : SYNC_HANDLER_MULTIXACT_MEMBER,
1936 : false);
1937 : /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1938 :
1939 : /* Initialize our shared state struct */
1940 1768 : MultiXactState = ShmemInitStruct("Shared MultiXact State",
1941 1768 : SHARED_MULTIXACT_STATE_SIZE,
1942 : &found);
1943 1768 : if (!IsUnderPostmaster)
1944 : {
1945 : Assert(!found);
1946 :
1947 : /* Make sure we zero out the per-backend state */
1948 1768 : MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
1949 1768 : ConditionVariableInit(&MultiXactState->nextoff_cv);
1950 : }
1951 : else
1952 : Assert(found);
1953 :
1954 : /*
1955 : * Set up array pointers.
1956 : */
1957 1768 : OldestMemberMXactId = MultiXactState->perBackendXactIds;
1958 1768 : OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
1959 1768 : }
1960 :
1961 : /*
1962 : * GUC check_hook for multixact_offset_buffers
1963 : */
1964 : bool
1965 1830 : check_multixact_offset_buffers(int *newval, void **extra, GucSource source)
1966 : {
1967 1830 : return check_slru_buffers("multixact_offset_buffers", newval);
1968 : }
1969 :
1970 : /*
1971 : * GUC check_hook for multixact_member_buffer
1972 : */
1973 : bool
1974 1830 : check_multixact_member_buffers(int *newval, void **extra, GucSource source)
1975 : {
1976 1830 : return check_slru_buffers("multixact_member_buffers", newval);
1977 : }
1978 :
1979 : /*
1980 : * This func must be called ONCE on system install. It creates the initial
1981 : * MultiXact segments. (The MultiXacts directories are assumed to have been
1982 : * created by initdb, and MultiXactShmemInit must have been called already.)
1983 : */
1984 : void
1985 80 : BootStrapMultiXact(void)
1986 : {
1987 : int slotno;
1988 : LWLock *lock;
1989 :
1990 80 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, 0);
1991 80 : LWLockAcquire(lock, LW_EXCLUSIVE);
1992 :
1993 : /* Create and zero the first page of the offsets log */
1994 80 : slotno = ZeroMultiXactOffsetPage(0, false);
1995 :
1996 : /* Make sure it's written out */
1997 80 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
1998 : Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
1999 :
2000 80 : LWLockRelease(lock);
2001 :
2002 80 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, 0);
2003 80 : LWLockAcquire(lock, LW_EXCLUSIVE);
2004 :
2005 : /* Create and zero the first page of the members log */
2006 80 : slotno = ZeroMultiXactMemberPage(0, false);
2007 :
2008 : /* Make sure it's written out */
2009 80 : SimpleLruWritePage(MultiXactMemberCtl, slotno);
2010 : Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
2011 :
2012 80 : LWLockRelease(lock);
2013 80 : }
2014 :
2015 : /*
2016 : * Initialize (or reinitialize) a page of MultiXactOffset to zeroes.
2017 : * If writeXlog is true, also emit an XLOG record saying we did this.
2018 : *
2019 : * The page is not actually written, just set up in shared memory.
2020 : * The slot number of the new page is returned.
2021 : *
2022 : * Control lock must be held at entry, and will be held at exit.
2023 : */
2024 : static int
2025 102 : ZeroMultiXactOffsetPage(int64 pageno, bool writeXlog)
2026 : {
2027 : int slotno;
2028 :
2029 102 : slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2030 :
2031 102 : if (writeXlog)
2032 20 : WriteMZeroPageXlogRec(pageno, XLOG_MULTIXACT_ZERO_OFF_PAGE);
2033 :
2034 102 : return slotno;
2035 : }
2036 :
2037 : /*
2038 : * Ditto, for MultiXactMember
2039 : */
2040 : static int
2041 102 : ZeroMultiXactMemberPage(int64 pageno, bool writeXlog)
2042 : {
2043 : int slotno;
2044 :
2045 102 : slotno = SimpleLruZeroPage(MultiXactMemberCtl, pageno);
2046 :
2047 102 : if (writeXlog)
2048 20 : WriteMZeroPageXlogRec(pageno, XLOG_MULTIXACT_ZERO_MEM_PAGE);
2049 :
2050 102 : return slotno;
2051 : }
2052 :
2053 : /*
2054 : * MaybeExtendOffsetSlru
2055 : * Extend the offsets SLRU area, if necessary
2056 : *
2057 : * After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
2058 : * contain files that are shorter than necessary; this would occur if the old
2059 : * installation had used multixacts beyond the first page (files cannot be
2060 : * copied, because the on-disk representation is different). pg_upgrade would
2061 : * update pg_control to set the next offset value to be at that position, so
2062 : * that tuples marked as locked by such MultiXacts would be seen as visible
2063 : * without having to consult multixact. However, trying to create and use a
2064 : * new MultiXactId would result in an error because the page on which the new
2065 : * value would reside does not exist. This routine is in charge of creating
2066 : * such pages.
2067 : */
2068 : static void
2069 44 : MaybeExtendOffsetSlru(void)
2070 : {
2071 : int64 pageno;
2072 : LWLock *lock;
2073 :
2074 44 : pageno = MultiXactIdToOffsetPage(MultiXactState->nextMXact);
2075 44 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2076 :
2077 44 : LWLockAcquire(lock, LW_EXCLUSIVE);
2078 :
2079 44 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2080 : {
2081 : int slotno;
2082 :
2083 : /*
2084 : * Fortunately for us, SimpleLruWritePage is already prepared to deal
2085 : * with creating a new segment file even if the page we're writing is
2086 : * not the first in it, so this is enough.
2087 : */
2088 0 : slotno = ZeroMultiXactOffsetPage(pageno, false);
2089 0 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
2090 : }
2091 :
2092 44 : LWLockRelease(lock);
2093 44 : }
2094 :
2095 : /*
2096 : * This must be called ONCE during postmaster or standalone-backend startup.
2097 : *
2098 : * StartupXLOG has already established nextMXact/nextOffset by calling
2099 : * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
2100 : * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
2101 : * replayed WAL.
2102 : */
2103 : void
2104 1520 : StartupMultiXact(void)
2105 : {
2106 1520 : MultiXactId multi = MultiXactState->nextMXact;
2107 1520 : MultiXactOffset offset = MultiXactState->nextOffset;
2108 : int64 pageno;
2109 :
2110 : /*
2111 : * Initialize offset's idea of the latest page number.
2112 : */
2113 1520 : pageno = MultiXactIdToOffsetPage(multi);
2114 1520 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2115 : pageno);
2116 :
2117 : /*
2118 : * Initialize member's idea of the latest page number.
2119 : */
2120 1520 : pageno = MXOffsetToMemberPage(offset);
2121 1520 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2122 : pageno);
2123 1520 : }
2124 :
2125 : /*
2126 : * This must be called ONCE at the end of startup/recovery.
2127 : */
2128 : void
2129 1420 : TrimMultiXact(void)
2130 : {
2131 : MultiXactId nextMXact;
2132 : MultiXactOffset offset;
2133 : MultiXactId oldestMXact;
2134 : Oid oldestMXactDB;
2135 : int64 pageno;
2136 : int entryno;
2137 : int flagsoff;
2138 :
2139 1420 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2140 1420 : nextMXact = MultiXactState->nextMXact;
2141 1420 : offset = MultiXactState->nextOffset;
2142 1420 : oldestMXact = MultiXactState->oldestMultiXactId;
2143 1420 : oldestMXactDB = MultiXactState->oldestMultiXactDB;
2144 1420 : LWLockRelease(MultiXactGenLock);
2145 :
2146 : /* Clean up offsets state */
2147 :
2148 : /*
2149 : * (Re-)Initialize our idea of the latest page number for offsets.
2150 : */
2151 1420 : pageno = MultiXactIdToOffsetPage(nextMXact);
2152 1420 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2153 : pageno);
2154 :
2155 : /*
2156 : * Zero out the remainder of the current offsets page. See notes in
2157 : * TrimCLOG() for background. Unlike CLOG, some WAL record covers every
2158 : * pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
2159 : * rule "write xlog before data," nextMXact successors may carry obsolete,
2160 : * nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
2161 : * operates normally.
2162 : */
2163 1420 : entryno = MultiXactIdToOffsetEntry(nextMXact);
2164 1420 : if (entryno != 0)
2165 : {
2166 : int slotno;
2167 : MultiXactOffset *offptr;
2168 1418 : LWLock *lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2169 :
2170 1418 : LWLockAcquire(lock, LW_EXCLUSIVE);
2171 1418 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
2172 1418 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2173 1418 : offptr += entryno;
2174 :
2175 1418 : MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
2176 :
2177 1418 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
2178 1418 : LWLockRelease(lock);
2179 : }
2180 :
2181 : /*
2182 : * And the same for members.
2183 : *
2184 : * (Re-)Initialize our idea of the latest page number for members.
2185 : */
2186 1420 : pageno = MXOffsetToMemberPage(offset);
2187 1420 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
2188 : pageno);
2189 :
2190 : /*
2191 : * Zero out the remainder of the current members page. See notes in
2192 : * TrimCLOG() for motivation.
2193 : */
2194 1420 : flagsoff = MXOffsetToFlagsOffset(offset);
2195 1420 : if (flagsoff != 0)
2196 : {
2197 : int slotno;
2198 : TransactionId *xidptr;
2199 : int memberoff;
2200 18 : LWLock *lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2201 :
2202 18 : LWLockAcquire(lock, LW_EXCLUSIVE);
2203 18 : memberoff = MXOffsetToMemberOffset(offset);
2204 18 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
2205 18 : xidptr = (TransactionId *)
2206 18 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
2207 :
2208 18 : MemSet(xidptr, 0, BLCKSZ - memberoff);
2209 :
2210 : /*
2211 : * Note: we don't need to zero out the flag bits in the remaining
2212 : * members of the current group, because they are always reset before
2213 : * writing.
2214 : */
2215 :
2216 18 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
2217 18 : LWLockRelease(lock);
2218 : }
2219 :
2220 : /* signal that we're officially up */
2221 1420 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2222 1420 : MultiXactState->finishedStartup = true;
2223 1420 : LWLockRelease(MultiXactGenLock);
2224 :
2225 : /* Now compute how far away the next members wraparound is. */
2226 1420 : SetMultiXactIdLimit(oldestMXact, oldestMXactDB, true);
2227 1420 : }
2228 :
2229 : /*
2230 : * Get the MultiXact data to save in a checkpoint record
2231 : */
2232 : void
2233 1628 : MultiXactGetCheckptMulti(bool is_shutdown,
2234 : MultiXactId *nextMulti,
2235 : MultiXactOffset *nextMultiOffset,
2236 : MultiXactId *oldestMulti,
2237 : Oid *oldestMultiDB)
2238 : {
2239 1628 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2240 1628 : *nextMulti = MultiXactState->nextMXact;
2241 1628 : *nextMultiOffset = MultiXactState->nextOffset;
2242 1628 : *oldestMulti = MultiXactState->oldestMultiXactId;
2243 1628 : *oldestMultiDB = MultiXactState->oldestMultiXactDB;
2244 1628 : LWLockRelease(MultiXactGenLock);
2245 :
2246 : debug_elog6(DEBUG2,
2247 : "MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
2248 : *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
2249 1628 : }
2250 :
2251 : /*
2252 : * Perform a checkpoint --- either during shutdown, or on-the-fly
2253 : */
2254 : void
2255 1704 : CheckPointMultiXact(void)
2256 : {
2257 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
2258 :
2259 : /*
2260 : * Write dirty MultiXact pages to disk. This may result in sync requests
2261 : * queued for later handling by ProcessSyncRequests(), as part of the
2262 : * checkpoint.
2263 : */
2264 1704 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2265 1704 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2266 :
2267 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
2268 1704 : }
2269 :
2270 : /*
2271 : * Set the next-to-be-assigned MultiXactId and offset
2272 : *
2273 : * This is used when we can determine the correct next ID/offset exactly
2274 : * from a checkpoint record. Although this is only called during bootstrap
2275 : * and XLog replay, we take the lock in case any hot-standby backends are
2276 : * examining the values.
2277 : */
2278 : void
2279 1660 : MultiXactSetNextMXact(MultiXactId nextMulti,
2280 : MultiXactOffset nextMultiOffset)
2281 : {
2282 : debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
2283 : nextMulti, nextMultiOffset);
2284 1660 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2285 1660 : MultiXactState->nextMXact = nextMulti;
2286 1660 : MultiXactState->nextOffset = nextMultiOffset;
2287 1660 : LWLockRelease(MultiXactGenLock);
2288 :
2289 : /*
2290 : * During a binary upgrade, make sure that the offsets SLRU is large
2291 : * enough to contain the next value that would be created.
2292 : *
2293 : * We need to do this pretty early during the first startup in binary
2294 : * upgrade mode: before StartupMultiXact() in fact, because this routine
2295 : * is called even before that by StartupXLOG(). And we can't do it
2296 : * earlier than at this point, because during that first call of this
2297 : * routine we determine the MultiXactState->nextMXact value that
2298 : * MaybeExtendOffsetSlru needs.
2299 : */
2300 1660 : if (IsBinaryUpgrade)
2301 44 : MaybeExtendOffsetSlru();
2302 1660 : }
2303 :
2304 : /*
2305 : * Determine the last safe MultiXactId to allocate given the currently oldest
2306 : * datminmxid (ie, the oldest MultiXactId that might exist in any database
2307 : * of our cluster), and the OID of the (or a) database with that value.
2308 : *
2309 : * is_startup is true when we are just starting the cluster, false when we
2310 : * are updating state in a running cluster. This only affects log messages.
2311 : */
2312 : void
2313 3152 : SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
2314 : bool is_startup)
2315 : {
2316 : MultiXactId multiVacLimit;
2317 : MultiXactId multiWarnLimit;
2318 : MultiXactId multiStopLimit;
2319 : MultiXactId multiWrapLimit;
2320 : MultiXactId curMulti;
2321 : bool needs_offset_vacuum;
2322 :
2323 : Assert(MultiXactIdIsValid(oldest_datminmxid));
2324 :
2325 : /*
2326 : * We pretend that a wrap will happen halfway through the multixact ID
2327 : * space, but that's not really true, because multixacts wrap differently
2328 : * from transaction IDs. Note that, separately from any concern about
2329 : * multixact IDs wrapping, we must ensure that multixact members do not
2330 : * wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
2331 : */
2332 3152 : multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2333 3152 : if (multiWrapLimit < FirstMultiXactId)
2334 0 : multiWrapLimit += FirstMultiXactId;
2335 :
2336 : /*
2337 : * We'll refuse to continue assigning MultiXactIds once we get within 3M
2338 : * multi of data loss. See SetTransactionIdLimit.
2339 : */
2340 3152 : multiStopLimit = multiWrapLimit - 3000000;
2341 3152 : if (multiStopLimit < FirstMultiXactId)
2342 0 : multiStopLimit -= FirstMultiXactId;
2343 :
2344 : /*
2345 : * We'll start complaining loudly when we get within 40M multis of data
2346 : * loss. This is kind of arbitrary, but if you let your gas gauge get
2347 : * down to 2% of full, would you be looking for the next gas station? We
2348 : * need to be fairly liberal about this number because there are lots of
2349 : * scenarios where most transactions are done by automatic clients that
2350 : * won't pay attention to warnings. (No, we're not gonna make this
2351 : * configurable. If you know enough to configure it, you know enough to
2352 : * not get in this kind of trouble in the first place.)
2353 : */
2354 3152 : multiWarnLimit = multiWrapLimit - 40000000;
2355 3152 : if (multiWarnLimit < FirstMultiXactId)
2356 0 : multiWarnLimit -= FirstMultiXactId;
2357 :
2358 : /*
2359 : * We'll start trying to force autovacuums when oldest_datminmxid gets to
2360 : * be more than autovacuum_multixact_freeze_max_age mxids old.
2361 : *
2362 : * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2363 : * so that we don't have to worry about dealing with on-the-fly changes in
2364 : * its value. See SetTransactionIdLimit.
2365 : */
2366 3152 : multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2367 3152 : if (multiVacLimit < FirstMultiXactId)
2368 0 : multiVacLimit += FirstMultiXactId;
2369 :
2370 : /* Grab lock for just long enough to set the new limit values */
2371 3152 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2372 3152 : MultiXactState->oldestMultiXactId = oldest_datminmxid;
2373 3152 : MultiXactState->oldestMultiXactDB = oldest_datoid;
2374 3152 : MultiXactState->multiVacLimit = multiVacLimit;
2375 3152 : MultiXactState->multiWarnLimit = multiWarnLimit;
2376 3152 : MultiXactState->multiStopLimit = multiStopLimit;
2377 3152 : MultiXactState->multiWrapLimit = multiWrapLimit;
2378 3152 : curMulti = MultiXactState->nextMXact;
2379 3152 : LWLockRelease(MultiXactGenLock);
2380 :
2381 : /* Log the info */
2382 3152 : ereport(DEBUG1,
2383 : (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2384 : multiWrapLimit, oldest_datoid)));
2385 :
2386 : /*
2387 : * Computing the actual limits is only possible once the data directory is
2388 : * in a consistent state. There's no need to compute the limits while
2389 : * still replaying WAL - no decisions about new multis are made even
2390 : * though multixact creations might be replayed. So we'll only do further
2391 : * checks after TrimMultiXact() has been called.
2392 : */
2393 3152 : if (!MultiXactState->finishedStartup)
2394 1600 : return;
2395 :
2396 : Assert(!InRecovery);
2397 :
2398 : /* Set limits for offset vacuum. */
2399 1552 : needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
2400 :
2401 : /*
2402 : * If past the autovacuum force point, immediately signal an autovac
2403 : * request. The reason for this is that autovac only processes one
2404 : * database per invocation. Once it's finished cleaning up the oldest
2405 : * database, it'll call here, and we'll signal the postmaster to start
2406 : * another iteration immediately if there are still any old databases.
2407 : */
2408 1552 : if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
2409 0 : needs_offset_vacuum) && IsUnderPostmaster)
2410 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
2411 :
2412 : /* Give an immediate warning if past the wrap warn point */
2413 1552 : if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2414 : {
2415 : char *oldest_datname;
2416 :
2417 : /*
2418 : * We can be called when not inside a transaction, for example during
2419 : * StartupXLOG(). In such a case we cannot do database access, so we
2420 : * must just report the oldest DB's OID.
2421 : *
2422 : * Note: it's also possible that get_database_name fails and returns
2423 : * NULL, for example because the database just got dropped. We'll
2424 : * still warn, even though the warning might now be unnecessary.
2425 : */
2426 0 : if (IsTransactionState())
2427 0 : oldest_datname = get_database_name(oldest_datoid);
2428 : else
2429 0 : oldest_datname = NULL;
2430 :
2431 0 : if (oldest_datname)
2432 0 : ereport(WARNING,
2433 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2434 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2435 : multiWrapLimit - curMulti,
2436 : oldest_datname,
2437 : multiWrapLimit - curMulti),
2438 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2439 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2440 : else
2441 0 : ereport(WARNING,
2442 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2443 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2444 : multiWrapLimit - curMulti,
2445 : oldest_datoid,
2446 : multiWrapLimit - curMulti),
2447 : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2448 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2449 : }
2450 : }
2451 :
2452 : /*
2453 : * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2454 : * and similarly nextOffset is at least minMultiOffset.
2455 : *
2456 : * This is used when we can determine minimum safe values from an XLog
2457 : * record (either an on-line checkpoint or an mxact creation log entry).
2458 : * Although this is only called during XLog replay, we take the lock in case
2459 : * any hot-standby backends are examining the values.
2460 : */
2461 : void
2462 354 : MultiXactAdvanceNextMXact(MultiXactId minMulti,
2463 : MultiXactOffset minMultiOffset)
2464 : {
2465 354 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2466 354 : if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
2467 : {
2468 : debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2469 4 : MultiXactState->nextMXact = minMulti;
2470 : }
2471 354 : if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
2472 : {
2473 : debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
2474 : minMultiOffset);
2475 4 : MultiXactState->nextOffset = minMultiOffset;
2476 : }
2477 354 : LWLockRelease(MultiXactGenLock);
2478 354 : }
2479 :
2480 : /*
2481 : * Update our oldestMultiXactId value, but only if it's more recent than what
2482 : * we had.
2483 : *
2484 : * This may only be called during WAL replay.
2485 : */
2486 : void
2487 410 : MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2488 : {
2489 : Assert(InRecovery);
2490 :
2491 410 : if (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
2492 0 : SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
2493 410 : }
2494 :
2495 : /*
2496 : * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2497 : *
2498 : * NB: this is called while holding MultiXactGenLock. We want it to be very
2499 : * fast most of the time; even when it's not so fast, no actual I/O need
2500 : * happen unless we're forced to write out a dirty log or xlog page to make
2501 : * room in shared memory.
2502 : */
2503 : static void
2504 550 : ExtendMultiXactOffset(MultiXactId multi)
2505 : {
2506 : int64 pageno;
2507 : LWLock *lock;
2508 :
2509 : /*
2510 : * No work except at first MultiXactId of a page. But beware: just after
2511 : * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2512 : */
2513 550 : if (MultiXactIdToOffsetEntry(multi) != 0 &&
2514 : multi != FirstMultiXactId)
2515 530 : return;
2516 :
2517 20 : pageno = MultiXactIdToOffsetPage(multi);
2518 20 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2519 :
2520 20 : LWLockAcquire(lock, LW_EXCLUSIVE);
2521 :
2522 : /* Zero the page and make an XLOG entry about it */
2523 20 : ZeroMultiXactOffsetPage(pageno, true);
2524 :
2525 20 : LWLockRelease(lock);
2526 : }
2527 :
2528 : /*
2529 : * Make sure that MultiXactMember has room for the members of a newly-
2530 : * allocated MultiXactId.
2531 : *
2532 : * Like the above routine, this is called while holding MultiXactGenLock;
2533 : * same comments apply.
2534 : */
2535 : static void
2536 550 : ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
2537 : {
2538 : /*
2539 : * It's possible that the members span more than one page of the members
2540 : * file, so we loop to ensure we consider each page. The coding is not
2541 : * optimal if the members span several pages, but that seems unusual
2542 : * enough to not worry much about.
2543 : */
2544 1100 : while (nmembers > 0)
2545 : {
2546 : int flagsoff;
2547 : int flagsbit;
2548 : uint32 difference;
2549 :
2550 : /*
2551 : * Only zero when at first entry of a page.
2552 : */
2553 550 : flagsoff = MXOffsetToFlagsOffset(offset);
2554 550 : flagsbit = MXOffsetToFlagsBitShift(offset);
2555 550 : if (flagsoff == 0 && flagsbit == 0)
2556 : {
2557 : int64 pageno;
2558 : LWLock *lock;
2559 :
2560 20 : pageno = MXOffsetToMemberPage(offset);
2561 20 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2562 :
2563 20 : LWLockAcquire(lock, LW_EXCLUSIVE);
2564 :
2565 : /* Zero the page and make an XLOG entry about it */
2566 20 : ZeroMultiXactMemberPage(pageno, true);
2567 :
2568 20 : LWLockRelease(lock);
2569 : }
2570 :
2571 : /*
2572 : * Compute the number of items till end of current page. Careful: if
2573 : * addition of unsigned ints wraps around, we're at the last page of
2574 : * the last segment; since that page holds a different number of items
2575 : * than other pages, we need to do it differently.
2576 : */
2577 550 : if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
2578 : {
2579 : /*
2580 : * This is the last page of the last segment; we can compute the
2581 : * number of items left to allocate in it without modulo
2582 : * arithmetic.
2583 : */
2584 0 : difference = MaxMultiXactOffset - offset + 1;
2585 : }
2586 : else
2587 550 : difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
2588 :
2589 : /*
2590 : * Advance to next page, taking care to properly handle the wraparound
2591 : * case. OK if nmembers goes negative.
2592 : */
2593 550 : nmembers -= difference;
2594 550 : offset += difference;
2595 : }
2596 550 : }
2597 :
2598 : /*
2599 : * GetOldestMultiXactId
2600 : *
2601 : * Return the oldest MultiXactId that's still possibly still seen as live by
2602 : * any running transaction. Older ones might still exist on disk, but they no
2603 : * longer have any running member transaction.
2604 : *
2605 : * It's not safe to truncate MultiXact SLRU segments on the value returned by
2606 : * this function; however, it can be set as the new relminmxid for any table
2607 : * that VACUUM knows has no remaining MXIDs < the same value. It is only safe
2608 : * to truncate SLRUs when no table can possibly still have a referencing MXID.
2609 : */
2610 : MultiXactId
2611 79804 : GetOldestMultiXactId(void)
2612 : {
2613 : MultiXactId oldestMXact;
2614 : MultiXactId nextMXact;
2615 : int i;
2616 :
2617 : /*
2618 : * This is the oldest valid value among all the OldestMemberMXactId[] and
2619 : * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2620 : */
2621 79804 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2622 :
2623 : /*
2624 : * We have to beware of the possibility that nextMXact is in the
2625 : * wrapped-around state. We don't fix the counter itself here, but we
2626 : * must be sure to use a valid value in our calculation.
2627 : */
2628 79804 : nextMXact = MultiXactState->nextMXact;
2629 79804 : if (nextMXact < FirstMultiXactId)
2630 0 : nextMXact = FirstMultiXactId;
2631 :
2632 79804 : oldestMXact = nextMXact;
2633 8630420 : for (i = 0; i < MaxOldestSlot; i++)
2634 : {
2635 : MultiXactId thisoldest;
2636 :
2637 8550616 : thisoldest = OldestMemberMXactId[i];
2638 8585810 : if (MultiXactIdIsValid(thisoldest) &&
2639 35194 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2640 8 : oldestMXact = thisoldest;
2641 8550616 : thisoldest = OldestVisibleMXactId[i];
2642 8550800 : if (MultiXactIdIsValid(thisoldest) &&
2643 184 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2644 4 : oldestMXact = thisoldest;
2645 : }
2646 :
2647 79804 : LWLockRelease(MultiXactGenLock);
2648 :
2649 79804 : return oldestMXact;
2650 : }
2651 :
2652 : /*
2653 : * Determine how aggressively we need to vacuum in order to prevent member
2654 : * wraparound.
2655 : *
2656 : * To do so determine what's the oldest member offset and install the limit
2657 : * info in MultiXactState, where it can be used to prevent overrun of old data
2658 : * in the members SLRU area.
2659 : *
2660 : * The return value is true if emergency autovacuum is required and false
2661 : * otherwise.
2662 : */
2663 : static bool
2664 1552 : SetOffsetVacuumLimit(bool is_startup)
2665 : {
2666 : MultiXactId oldestMultiXactId;
2667 : MultiXactId nextMXact;
2668 1552 : MultiXactOffset oldestOffset = 0; /* placate compiler */
2669 : MultiXactOffset prevOldestOffset;
2670 : MultiXactOffset nextOffset;
2671 1552 : bool oldestOffsetKnown = false;
2672 : bool prevOldestOffsetKnown;
2673 1552 : MultiXactOffset offsetStopLimit = 0;
2674 : MultiXactOffset prevOffsetStopLimit;
2675 :
2676 : /*
2677 : * NB: Have to prevent concurrent truncation, we might otherwise try to
2678 : * lookup an oldestMulti that's concurrently getting truncated away.
2679 : */
2680 1552 : LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2681 :
2682 : /* Read relevant fields from shared memory. */
2683 1552 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2684 1552 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2685 1552 : nextMXact = MultiXactState->nextMXact;
2686 1552 : nextOffset = MultiXactState->nextOffset;
2687 1552 : prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2688 1552 : prevOldestOffset = MultiXactState->oldestOffset;
2689 1552 : prevOffsetStopLimit = MultiXactState->offsetStopLimit;
2690 : Assert(MultiXactState->finishedStartup);
2691 1552 : LWLockRelease(MultiXactGenLock);
2692 :
2693 : /*
2694 : * Determine the offset of the oldest multixact. Normally, we can read
2695 : * the offset from the multixact itself, but there's an important special
2696 : * case: if there are no multixacts in existence at all, oldestMXact
2697 : * obviously can't point to one. It will instead point to the multixact
2698 : * ID that will be assigned the next time one is needed.
2699 : */
2700 1552 : if (oldestMultiXactId == nextMXact)
2701 : {
2702 : /*
2703 : * When the next multixact gets created, it will be stored at the next
2704 : * offset.
2705 : */
2706 1532 : oldestOffset = nextOffset;
2707 1532 : oldestOffsetKnown = true;
2708 : }
2709 : else
2710 : {
2711 : /*
2712 : * Figure out where the oldest existing multixact's offsets are
2713 : * stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
2714 : * the supposedly-earliest multixact might not really exist. We are
2715 : * careful not to fail in that case.
2716 : */
2717 : oldestOffsetKnown =
2718 20 : find_multixact_start(oldestMultiXactId, &oldestOffset);
2719 :
2720 20 : if (oldestOffsetKnown)
2721 20 : ereport(DEBUG1,
2722 : (errmsg_internal("oldest MultiXactId member is at offset %u",
2723 : oldestOffset)));
2724 : else
2725 0 : ereport(LOG,
2726 : (errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
2727 : oldestMultiXactId)));
2728 : }
2729 :
2730 1552 : LWLockRelease(MultiXactTruncationLock);
2731 :
2732 : /*
2733 : * If we can, compute limits (and install them MultiXactState) to prevent
2734 : * overrun of old data in the members SLRU area. We can only do so if the
2735 : * oldest offset is known though.
2736 : */
2737 1552 : if (oldestOffsetKnown)
2738 : {
2739 : /* move back to start of the corresponding segment */
2740 1552 : offsetStopLimit = oldestOffset - (oldestOffset %
2741 : (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT));
2742 :
2743 : /* always leave one segment before the wraparound point */
2744 1552 : offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
2745 :
2746 1552 : if (!prevOldestOffsetKnown && !is_startup)
2747 0 : ereport(LOG,
2748 : (errmsg("MultiXact member wraparound protections are now enabled")));
2749 :
2750 1552 : ereport(DEBUG1,
2751 : (errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
2752 : offsetStopLimit, oldestMultiXactId)));
2753 : }
2754 0 : else if (prevOldestOffsetKnown)
2755 : {
2756 : /*
2757 : * If we failed to get the oldest offset this time, but we have a
2758 : * value from a previous pass through this function, use the old
2759 : * values rather than automatically forcing an emergency autovacuum
2760 : * cycle again.
2761 : */
2762 0 : oldestOffset = prevOldestOffset;
2763 0 : oldestOffsetKnown = true;
2764 0 : offsetStopLimit = prevOffsetStopLimit;
2765 : }
2766 :
2767 : /* Install the computed values */
2768 1552 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2769 1552 : MultiXactState->oldestOffset = oldestOffset;
2770 1552 : MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
2771 1552 : MultiXactState->offsetStopLimit = offsetStopLimit;
2772 1552 : LWLockRelease(MultiXactGenLock);
2773 :
2774 : /*
2775 : * Do we need an emergency autovacuum? If we're not sure, assume yes.
2776 : */
2777 3104 : return !oldestOffsetKnown ||
2778 1552 : (nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
2779 : }
2780 :
2781 : /*
2782 : * Return whether adding "distance" to "start" would move past "boundary".
2783 : *
2784 : * We use this to determine whether the addition is "wrapping around" the
2785 : * boundary point, hence the name. The reason we don't want to use the regular
2786 : * 2^31-modulo arithmetic here is that we want to be able to use the whole of
2787 : * the 2^32-1 space here, allowing for more multixacts than would fit
2788 : * otherwise.
2789 : */
2790 : static bool
2791 1100 : MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start,
2792 : uint32 distance)
2793 : {
2794 : MultiXactOffset finish;
2795 :
2796 : /*
2797 : * Note that offset number 0 is not used (see GetMultiXactIdMembers), so
2798 : * if the addition wraps around the UINT_MAX boundary, skip that value.
2799 : */
2800 1100 : finish = start + distance;
2801 1100 : if (finish < start)
2802 0 : finish++;
2803 :
2804 : /*-----------------------------------------------------------------------
2805 : * When the boundary is numerically greater than the starting point, any
2806 : * value numerically between the two is not wrapped:
2807 : *
2808 : * <----S----B---->
2809 : * [---) = F wrapped past B (and UINT_MAX)
2810 : * [---) = F not wrapped
2811 : * [----] = F wrapped past B
2812 : *
2813 : * When the boundary is numerically less than the starting point (i.e. the
2814 : * UINT_MAX wraparound occurs somewhere in between) then all values in
2815 : * between are wrapped:
2816 : *
2817 : * <----B----S---->
2818 : * [---) = F not wrapped past B (but wrapped past UINT_MAX)
2819 : * [---) = F wrapped past B (and UINT_MAX)
2820 : * [----] = F not wrapped
2821 : *-----------------------------------------------------------------------
2822 : */
2823 1100 : if (start < boundary)
2824 1100 : return finish >= boundary || finish < start;
2825 : else
2826 0 : return finish >= boundary && finish < start;
2827 : }
2828 :
2829 : /*
2830 : * Find the starting offset of the given MultiXactId.
2831 : *
2832 : * Returns false if the file containing the multi does not exist on disk.
2833 : * Otherwise, returns true and sets *result to the starting member offset.
2834 : *
2835 : * This function does not prevent concurrent truncation, so if that's
2836 : * required, the caller has to protect against that.
2837 : */
2838 : static bool
2839 20 : find_multixact_start(MultiXactId multi, MultiXactOffset *result)
2840 : {
2841 : MultiXactOffset offset;
2842 : int64 pageno;
2843 : int entryno;
2844 : int slotno;
2845 : MultiXactOffset *offptr;
2846 :
2847 : Assert(MultiXactState->finishedStartup);
2848 :
2849 20 : pageno = MultiXactIdToOffsetPage(multi);
2850 20 : entryno = MultiXactIdToOffsetEntry(multi);
2851 :
2852 : /*
2853 : * Write out dirty data, so PhysicalPageExists can work correctly.
2854 : */
2855 20 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2856 20 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2857 :
2858 20 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2859 0 : return false;
2860 :
2861 : /* lock is acquired by SimpleLruReadPage_ReadOnly */
2862 20 : slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2863 20 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2864 20 : offptr += entryno;
2865 20 : offset = *offptr;
2866 20 : LWLockRelease(SimpleLruGetBankLock(MultiXactOffsetCtl, pageno));
2867 :
2868 20 : *result = offset;
2869 20 : return true;
2870 : }
2871 :
2872 : /*
2873 : * Determine how many multixacts, and how many multixact members, currently
2874 : * exist. Return false if unable to determine.
2875 : */
2876 : static bool
2877 19736 : ReadMultiXactCounts(uint32 *multixacts, MultiXactOffset *members)
2878 : {
2879 : MultiXactOffset nextOffset;
2880 : MultiXactOffset oldestOffset;
2881 : MultiXactId oldestMultiXactId;
2882 : MultiXactId nextMultiXactId;
2883 : bool oldestOffsetKnown;
2884 :
2885 19736 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2886 19736 : nextOffset = MultiXactState->nextOffset;
2887 19736 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2888 19736 : nextMultiXactId = MultiXactState->nextMXact;
2889 19736 : oldestOffset = MultiXactState->oldestOffset;
2890 19736 : oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2891 19736 : LWLockRelease(MultiXactGenLock);
2892 :
2893 19736 : if (!oldestOffsetKnown)
2894 0 : return false;
2895 :
2896 19736 : *members = nextOffset - oldestOffset;
2897 19736 : *multixacts = nextMultiXactId - oldestMultiXactId;
2898 19736 : return true;
2899 : }
2900 :
2901 : /*
2902 : * Multixact members can be removed once the multixacts that refer to them
2903 : * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2904 : * vacuum_multixact_freeze_table_age work together to make sure we never have
2905 : * too many multixacts; we hope that, at least under normal circumstances,
2906 : * this will also be sufficient to keep us from using too many offsets.
2907 : * However, if the average multixact has many members, we might exhaust the
2908 : * members space while still using few enough members that these limits fail
2909 : * to trigger relminmxid advancement by VACUUM. At that point, we'd have no
2910 : * choice but to start failing multixact-creating operations with an error.
2911 : *
2912 : * To prevent that, if more than a threshold portion of the members space is
2913 : * used, we effectively reduce autovacuum_multixact_freeze_max_age and
2914 : * to a value just less than the number of multixacts in use. We hope that
2915 : * this will quickly trigger autovacuuming on the table or tables with the
2916 : * oldest relminmxid, thus allowing datminmxid values to advance and removing
2917 : * some members.
2918 : *
2919 : * As the fraction of the member space currently in use grows, we become
2920 : * more aggressive in clamping this value. That not only causes autovacuum
2921 : * to ramp up, but also makes any manual vacuums the user issues more
2922 : * aggressive. This happens because vacuum_get_cutoffs() will clamp the
2923 : * freeze table and the minimum freeze age cutoffs based on the effective
2924 : * autovacuum_multixact_freeze_max_age this function returns. In the worst
2925 : * case, we'll claim the freeze_max_age to zero, and every vacuum of any
2926 : * table will freeze every multixact.
2927 : */
2928 : int
2929 19736 : MultiXactMemberFreezeThreshold(void)
2930 : {
2931 : MultiXactOffset members;
2932 : uint32 multixacts;
2933 : uint32 victim_multixacts;
2934 : double fraction;
2935 :
2936 : /* If we can't determine member space utilization, assume the worst. */
2937 19736 : if (!ReadMultiXactCounts(&multixacts, &members))
2938 0 : return 0;
2939 :
2940 : /* If member space utilization is low, no special action is required. */
2941 19736 : if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
2942 19736 : return autovacuum_multixact_freeze_max_age;
2943 :
2944 : /*
2945 : * Compute a target for relminmxid advancement. The number of multixacts
2946 : * we try to eliminate from the system is based on how far we are past
2947 : * MULTIXACT_MEMBER_SAFE_THRESHOLD.
2948 : */
2949 0 : fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
2950 : (MULTIXACT_MEMBER_DANGER_THRESHOLD - MULTIXACT_MEMBER_SAFE_THRESHOLD);
2951 0 : victim_multixacts = multixacts * fraction;
2952 :
2953 : /* fraction could be > 1.0, but lowest possible freeze age is zero */
2954 0 : if (victim_multixacts > multixacts)
2955 0 : return 0;
2956 0 : return multixacts - victim_multixacts;
2957 : }
2958 :
2959 : typedef struct mxtruncinfo
2960 : {
2961 : int64 earliestExistingPage;
2962 : } mxtruncinfo;
2963 :
2964 : /*
2965 : * SlruScanDirectory callback
2966 : * This callback determines the earliest existing page number.
2967 : */
2968 : static bool
2969 0 : SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int64 segpage, void *data)
2970 : {
2971 0 : mxtruncinfo *trunc = (mxtruncinfo *) data;
2972 :
2973 0 : if (trunc->earliestExistingPage == -1 ||
2974 0 : ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
2975 : {
2976 0 : trunc->earliestExistingPage = segpage;
2977 : }
2978 :
2979 0 : return false; /* keep going */
2980 : }
2981 :
2982 :
2983 : /*
2984 : * Delete members segments [oldest, newOldest)
2985 : *
2986 : * The members SLRU can, in contrast to the offsets one, be filled to almost
2987 : * the full range at once. This means SimpleLruTruncate() can't trivially be
2988 : * used - instead the to-be-deleted range is computed using the offsets
2989 : * SLRU. C.f. TruncateMultiXact().
2990 : */
2991 : static void
2992 0 : PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
2993 : {
2994 0 : const int maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
2995 0 : int startsegment = MXOffsetToMemberSegment(oldestOffset);
2996 0 : int endsegment = MXOffsetToMemberSegment(newOldestOffset);
2997 0 : int segment = startsegment;
2998 :
2999 : /*
3000 : * Delete all the segments but the last one. The last segment can still
3001 : * contain, possibly partially, valid data.
3002 : */
3003 0 : while (segment != endsegment)
3004 : {
3005 0 : elog(DEBUG2, "truncating multixact members segment %x", segment);
3006 0 : SlruDeleteSegment(MultiXactMemberCtl, segment);
3007 :
3008 : /* move to next segment, handling wraparound correctly */
3009 0 : if (segment == maxsegment)
3010 0 : segment = 0;
3011 : else
3012 0 : segment += 1;
3013 : }
3014 0 : }
3015 :
3016 : /*
3017 : * Delete offsets segments [oldest, newOldest)
3018 : */
3019 : static void
3020 0 : PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
3021 : {
3022 : /*
3023 : * We step back one multixact to avoid passing a cutoff page that hasn't
3024 : * been created yet in the rare case that oldestMulti would be the first
3025 : * item on a page and oldestMulti == nextMulti. In that case, if we
3026 : * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
3027 : * detection.
3028 : */
3029 0 : SimpleLruTruncate(MultiXactOffsetCtl,
3030 0 : MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
3031 0 : }
3032 :
3033 : /*
3034 : * Remove all MultiXactOffset and MultiXactMember segments before the oldest
3035 : * ones still of interest.
3036 : *
3037 : * This is only called on a primary as part of vacuum (via
3038 : * vac_truncate_clog()). During recovery truncation is done by replaying
3039 : * truncation WAL records logged here.
3040 : *
3041 : * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
3042 : * is one of the databases preventing newOldestMulti from increasing.
3043 : */
3044 : void
3045 132 : TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
3046 : {
3047 : MultiXactId oldestMulti;
3048 : MultiXactId nextMulti;
3049 : MultiXactOffset newOldestOffset;
3050 : MultiXactOffset oldestOffset;
3051 : MultiXactOffset nextOffset;
3052 : mxtruncinfo trunc;
3053 : MultiXactId earliest;
3054 :
3055 : Assert(!RecoveryInProgress());
3056 : Assert(MultiXactState->finishedStartup);
3057 :
3058 : /*
3059 : * We can only allow one truncation to happen at once. Otherwise parts of
3060 : * members might vanish while we're doing lookups or similar. There's no
3061 : * need to have an interlock with creating new multis or such, since those
3062 : * are constrained by the limits (which only grow, never shrink).
3063 : */
3064 132 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3065 :
3066 132 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
3067 132 : nextMulti = MultiXactState->nextMXact;
3068 132 : nextOffset = MultiXactState->nextOffset;
3069 132 : oldestMulti = MultiXactState->oldestMultiXactId;
3070 132 : LWLockRelease(MultiXactGenLock);
3071 : Assert(MultiXactIdIsValid(oldestMulti));
3072 :
3073 : /*
3074 : * Make sure to only attempt truncation if there's values to truncate
3075 : * away. In normal processing values shouldn't go backwards, but there's
3076 : * some corner cases (due to bugs) where that's possible.
3077 : */
3078 132 : if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
3079 : {
3080 132 : LWLockRelease(MultiXactTruncationLock);
3081 132 : return;
3082 : }
3083 :
3084 : /*
3085 : * Note we can't just plow ahead with the truncation; it's possible that
3086 : * there are no segments to truncate, which is a problem because we are
3087 : * going to attempt to read the offsets page to determine where to
3088 : * truncate the members SLRU. So we first scan the directory to determine
3089 : * the earliest offsets page number that we can read without error.
3090 : *
3091 : * When nextMXact is less than one segment away from multiWrapLimit,
3092 : * SlruScanDirCbFindEarliest can find some early segment other than the
3093 : * actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
3094 : * returns false, because not all pairs of entries have the same answer.)
3095 : * That can also arise when an earlier truncation attempt failed unlink()
3096 : * or returned early from this function. The only consequence is
3097 : * returning early, which wastes space that we could have liberated.
3098 : *
3099 : * NB: It's also possible that the page that oldestMulti is on has already
3100 : * been truncated away, and we crashed before updating oldestMulti.
3101 : */
3102 0 : trunc.earliestExistingPage = -1;
3103 0 : SlruScanDirectory(MultiXactOffsetCtl, SlruScanDirCbFindEarliest, &trunc);
3104 0 : earliest = trunc.earliestExistingPage * MULTIXACT_OFFSETS_PER_PAGE;
3105 0 : if (earliest < FirstMultiXactId)
3106 0 : earliest = FirstMultiXactId;
3107 :
3108 : /* If there's nothing to remove, we can bail out early. */
3109 0 : if (MultiXactIdPrecedes(oldestMulti, earliest))
3110 : {
3111 0 : LWLockRelease(MultiXactTruncationLock);
3112 0 : return;
3113 : }
3114 :
3115 : /*
3116 : * First, compute the safe truncation point for MultiXactMember. This is
3117 : * the starting offset of the oldest multixact.
3118 : *
3119 : * Hopefully, find_multixact_start will always work here, because we've
3120 : * already checked that it doesn't precede the earliest MultiXact on disk.
3121 : * But if it fails, don't truncate anything, and log a message.
3122 : */
3123 0 : if (oldestMulti == nextMulti)
3124 : {
3125 : /* there are NO MultiXacts */
3126 0 : oldestOffset = nextOffset;
3127 : }
3128 0 : else if (!find_multixact_start(oldestMulti, &oldestOffset))
3129 : {
3130 0 : ereport(LOG,
3131 : (errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
3132 : oldestMulti, earliest)));
3133 0 : LWLockRelease(MultiXactTruncationLock);
3134 0 : return;
3135 : }
3136 :
3137 : /*
3138 : * Secondly compute up to where to truncate. Lookup the corresponding
3139 : * member offset for newOldestMulti for that.
3140 : */
3141 0 : if (newOldestMulti == nextMulti)
3142 : {
3143 : /* there are NO MultiXacts */
3144 0 : newOldestOffset = nextOffset;
3145 : }
3146 0 : else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
3147 : {
3148 0 : ereport(LOG,
3149 : (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
3150 : newOldestMulti)));
3151 0 : LWLockRelease(MultiXactTruncationLock);
3152 0 : return;
3153 : }
3154 :
3155 0 : elog(DEBUG1, "performing multixact truncation: "
3156 : "offsets [%u, %u), offsets segments [%x, %x), "
3157 : "members [%u, %u), members segments [%x, %x)",
3158 : oldestMulti, newOldestMulti,
3159 : MultiXactIdToOffsetSegment(oldestMulti),
3160 : MultiXactIdToOffsetSegment(newOldestMulti),
3161 : oldestOffset, newOldestOffset,
3162 : MXOffsetToMemberSegment(oldestOffset),
3163 : MXOffsetToMemberSegment(newOldestOffset));
3164 :
3165 : /*
3166 : * Do truncation, and the WAL logging of the truncation, in a critical
3167 : * section. That way offsets/members cannot get out of sync anymore, i.e.
3168 : * once consistent the newOldestMulti will always exist in members, even
3169 : * if we crashed in the wrong moment.
3170 : */
3171 0 : START_CRIT_SECTION();
3172 :
3173 : /*
3174 : * Prevent checkpoints from being scheduled concurrently. This is critical
3175 : * because otherwise a truncation record might not be replayed after a
3176 : * crash/basebackup, even though the state of the data directory would
3177 : * require it.
3178 : */
3179 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
3180 0 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
3181 :
3182 : /* WAL log truncation */
3183 0 : WriteMTruncateXlogRec(newOldestMultiDB,
3184 : oldestMulti, newOldestMulti,
3185 : oldestOffset, newOldestOffset);
3186 :
3187 : /*
3188 : * Update in-memory limits before performing the truncation, while inside
3189 : * the critical section: Have to do it before truncation, to prevent
3190 : * concurrent lookups of those values. Has to be inside the critical
3191 : * section as otherwise a future call to this function would error out,
3192 : * while looking up the oldest member in offsets, if our caller crashes
3193 : * before updating the limits.
3194 : */
3195 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
3196 0 : MultiXactState->oldestMultiXactId = newOldestMulti;
3197 0 : MultiXactState->oldestMultiXactDB = newOldestMultiDB;
3198 0 : LWLockRelease(MultiXactGenLock);
3199 :
3200 : /* First truncate members */
3201 0 : PerformMembersTruncation(oldestOffset, newOldestOffset);
3202 :
3203 : /* Then offsets */
3204 0 : PerformOffsetsTruncation(oldestMulti, newOldestMulti);
3205 :
3206 0 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
3207 :
3208 0 : END_CRIT_SECTION();
3209 0 : LWLockRelease(MultiXactTruncationLock);
3210 : }
3211 :
3212 : /*
3213 : * Decide whether a MultiXactOffset page number is "older" for truncation
3214 : * purposes. Analogous to CLOGPagePrecedes().
3215 : *
3216 : * Offsetting the values is optional, because MultiXactIdPrecedes() has
3217 : * translational symmetry.
3218 : */
3219 : static bool
3220 0 : MultiXactOffsetPagePrecedes(int64 page1, int64 page2)
3221 : {
3222 : MultiXactId multi1;
3223 : MultiXactId multi2;
3224 :
3225 0 : multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
3226 0 : multi1 += FirstMultiXactId + 1;
3227 0 : multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
3228 0 : multi2 += FirstMultiXactId + 1;
3229 :
3230 0 : return (MultiXactIdPrecedes(multi1, multi2) &&
3231 0 : MultiXactIdPrecedes(multi1,
3232 : multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
3233 : }
3234 :
3235 : /*
3236 : * Decide whether a MultiXactMember page number is "older" for truncation
3237 : * purposes. There is no "invalid offset number" so use the numbers verbatim.
3238 : */
3239 : static bool
3240 0 : MultiXactMemberPagePrecedes(int64 page1, int64 page2)
3241 : {
3242 : MultiXactOffset offset1;
3243 : MultiXactOffset offset2;
3244 :
3245 0 : offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
3246 0 : offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
3247 :
3248 0 : return (MultiXactOffsetPrecedes(offset1, offset2) &&
3249 0 : MultiXactOffsetPrecedes(offset1,
3250 : offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
3251 : }
3252 :
3253 : /*
3254 : * Decide which of two MultiXactIds is earlier.
3255 : *
3256 : * XXX do we need to do something special for InvalidMultiXactId?
3257 : * (Doesn't look like it.)
3258 : */
3259 : bool
3260 844790 : MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
3261 : {
3262 844790 : int32 diff = (int32) (multi1 - multi2);
3263 :
3264 844790 : return (diff < 0);
3265 : }
3266 :
3267 : /*
3268 : * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
3269 : *
3270 : * XXX do we need to do something special for InvalidMultiXactId?
3271 : * (Doesn't look like it.)
3272 : */
3273 : bool
3274 8824 : MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
3275 : {
3276 8824 : int32 diff = (int32) (multi1 - multi2);
3277 :
3278 8824 : return (diff <= 0);
3279 : }
3280 :
3281 :
3282 : /*
3283 : * Decide which of two offsets is earlier.
3284 : */
3285 : static bool
3286 354 : MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
3287 : {
3288 354 : int32 diff = (int32) (offset1 - offset2);
3289 :
3290 354 : return (diff < 0);
3291 : }
3292 :
3293 : /*
3294 : * Write an xlog record reflecting the zeroing of either a MEMBERs or
3295 : * OFFSETs page (info shows which)
3296 : */
3297 : static void
3298 40 : WriteMZeroPageXlogRec(int64 pageno, uint8 info)
3299 : {
3300 40 : XLogBeginInsert();
3301 40 : XLogRegisterData((char *) (&pageno), sizeof(pageno));
3302 40 : (void) XLogInsert(RM_MULTIXACT_ID, info);
3303 40 : }
3304 :
3305 : /*
3306 : * Write a TRUNCATE xlog record
3307 : *
3308 : * We must flush the xlog record to disk before returning --- see notes in
3309 : * TruncateCLOG().
3310 : */
3311 : static void
3312 0 : WriteMTruncateXlogRec(Oid oldestMultiDB,
3313 : MultiXactId startTruncOff, MultiXactId endTruncOff,
3314 : MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
3315 : {
3316 : XLogRecPtr recptr;
3317 : xl_multixact_truncate xlrec;
3318 :
3319 0 : xlrec.oldestMultiDB = oldestMultiDB;
3320 :
3321 0 : xlrec.startTruncOff = startTruncOff;
3322 0 : xlrec.endTruncOff = endTruncOff;
3323 :
3324 0 : xlrec.startTruncMemb = startTruncMemb;
3325 0 : xlrec.endTruncMemb = endTruncMemb;
3326 :
3327 0 : XLogBeginInsert();
3328 0 : XLogRegisterData((char *) (&xlrec), SizeOfMultiXactTruncate);
3329 0 : recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
3330 0 : XLogFlush(recptr);
3331 0 : }
3332 :
3333 : /*
3334 : * MULTIXACT resource manager's routines
3335 : */
3336 : void
3337 8 : multixact_redo(XLogReaderState *record)
3338 : {
3339 8 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3340 :
3341 : /* Backup blocks are not used in multixact records */
3342 : Assert(!XLogRecHasAnyBlockRefs(record));
3343 :
3344 8 : if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
3345 : {
3346 : int64 pageno;
3347 : int slotno;
3348 : LWLock *lock;
3349 :
3350 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3351 :
3352 2 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
3353 2 : LWLockAcquire(lock, LW_EXCLUSIVE);
3354 :
3355 2 : slotno = ZeroMultiXactOffsetPage(pageno, false);
3356 2 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
3357 : Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
3358 :
3359 2 : LWLockRelease(lock);
3360 : }
3361 6 : else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
3362 : {
3363 : int64 pageno;
3364 : int slotno;
3365 : LWLock *lock;
3366 :
3367 2 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
3368 :
3369 2 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
3370 2 : LWLockAcquire(lock, LW_EXCLUSIVE);
3371 :
3372 2 : slotno = ZeroMultiXactMemberPage(pageno, false);
3373 2 : SimpleLruWritePage(MultiXactMemberCtl, slotno);
3374 : Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
3375 :
3376 2 : LWLockRelease(lock);
3377 : }
3378 4 : else if (info == XLOG_MULTIXACT_CREATE_ID)
3379 : {
3380 4 : xl_multixact_create *xlrec =
3381 4 : (xl_multixact_create *) XLogRecGetData(record);
3382 : TransactionId max_xid;
3383 : int i;
3384 :
3385 : /* Store the data back into the SLRU files */
3386 4 : RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
3387 4 : xlrec->members);
3388 :
3389 : /* Make sure nextMXact/nextOffset are beyond what this record has */
3390 4 : MultiXactAdvanceNextMXact(xlrec->mid + 1,
3391 4 : xlrec->moff + xlrec->nmembers);
3392 :
3393 : /*
3394 : * Make sure nextXid is beyond any XID mentioned in the record. This
3395 : * should be unnecessary, since any XID found here ought to have other
3396 : * evidence in the XLOG, but let's be safe.
3397 : */
3398 4 : max_xid = XLogRecGetXid(record);
3399 12 : for (i = 0; i < xlrec->nmembers; i++)
3400 : {
3401 8 : if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
3402 0 : max_xid = xlrec->members[i].xid;
3403 : }
3404 :
3405 4 : AdvanceNextFullTransactionIdPastXid(max_xid);
3406 : }
3407 0 : else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
3408 : {
3409 : xl_multixact_truncate xlrec;
3410 : int64 pageno;
3411 :
3412 0 : memcpy(&xlrec, XLogRecGetData(record),
3413 : SizeOfMultiXactTruncate);
3414 :
3415 0 : elog(DEBUG1, "replaying multixact truncation: "
3416 : "offsets [%u, %u), offsets segments [%x, %x), "
3417 : "members [%u, %u), members segments [%x, %x)",
3418 : xlrec.startTruncOff, xlrec.endTruncOff,
3419 : MultiXactIdToOffsetSegment(xlrec.startTruncOff),
3420 : MultiXactIdToOffsetSegment(xlrec.endTruncOff),
3421 : xlrec.startTruncMemb, xlrec.endTruncMemb,
3422 : MXOffsetToMemberSegment(xlrec.startTruncMemb),
3423 : MXOffsetToMemberSegment(xlrec.endTruncMemb));
3424 :
3425 : /* should not be required, but more than cheap enough */
3426 0 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3427 :
3428 : /*
3429 : * Advance the horizon values, so they're current at the end of
3430 : * recovery.
3431 : */
3432 0 : SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
3433 :
3434 0 : PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
3435 :
3436 : /*
3437 : * During XLOG replay, latest_page_number isn't necessarily set up
3438 : * yet; insert a suitable value to bypass the sanity test in
3439 : * SimpleLruTruncate.
3440 : */
3441 0 : pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
3442 0 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
3443 : pageno);
3444 0 : PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
3445 :
3446 0 : LWLockRelease(MultiXactTruncationLock);
3447 : }
3448 : else
3449 0 : elog(PANIC, "multixact_redo: unknown op code %u", info);
3450 8 : }
3451 :
3452 : Datum
3453 0 : pg_get_multixact_members(PG_FUNCTION_ARGS)
3454 : {
3455 : typedef struct
3456 : {
3457 : MultiXactMember *members;
3458 : int nmembers;
3459 : int iter;
3460 : } mxact;
3461 0 : MultiXactId mxid = PG_GETARG_TRANSACTIONID(0);
3462 : mxact *multi;
3463 : FuncCallContext *funccxt;
3464 :
3465 0 : if (mxid < FirstMultiXactId)
3466 0 : ereport(ERROR,
3467 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3468 : errmsg("invalid MultiXactId: %u", mxid)));
3469 :
3470 0 : if (SRF_IS_FIRSTCALL())
3471 : {
3472 : MemoryContext oldcxt;
3473 : TupleDesc tupdesc;
3474 :
3475 0 : funccxt = SRF_FIRSTCALL_INIT();
3476 0 : oldcxt = MemoryContextSwitchTo(funccxt->multi_call_memory_ctx);
3477 :
3478 0 : multi = palloc(sizeof(mxact));
3479 : /* no need to allow for old values here */
3480 0 : multi->nmembers = GetMultiXactIdMembers(mxid, &multi->members, false,
3481 : false);
3482 0 : multi->iter = 0;
3483 :
3484 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
3485 0 : elog(ERROR, "return type must be a row type");
3486 0 : funccxt->tuple_desc = tupdesc;
3487 0 : funccxt->attinmeta = TupleDescGetAttInMetadata(tupdesc);
3488 0 : funccxt->user_fctx = multi;
3489 :
3490 0 : MemoryContextSwitchTo(oldcxt);
3491 : }
3492 :
3493 0 : funccxt = SRF_PERCALL_SETUP();
3494 0 : multi = (mxact *) funccxt->user_fctx;
3495 :
3496 0 : while (multi->iter < multi->nmembers)
3497 : {
3498 : HeapTuple tuple;
3499 : char *values[2];
3500 :
3501 0 : values[0] = psprintf("%u", multi->members[multi->iter].xid);
3502 0 : values[1] = mxstatus_to_string(multi->members[multi->iter].status);
3503 :
3504 0 : tuple = BuildTupleFromCStrings(funccxt->attinmeta, values);
3505 :
3506 0 : multi->iter++;
3507 0 : pfree(values[0]);
3508 0 : SRF_RETURN_NEXT(funccxt, HeapTupleGetDatum(tuple));
3509 : }
3510 :
3511 0 : SRF_RETURN_DONE(funccxt);
3512 : }
3513 :
3514 : /*
3515 : * Entrypoint for sync.c to sync offsets files.
3516 : */
3517 : int
3518 0 : multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
3519 : {
3520 0 : return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
3521 : }
3522 :
3523 : /*
3524 : * Entrypoint for sync.c to sync members files.
3525 : */
3526 : int
3527 0 : multixactmemberssyncfiletag(const FileTag *ftag, char *path)
3528 : {
3529 0 : return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
3530 : }
|