TLA 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) 1996-2023, PostgreSQL Global Development Group
63 : * Portions Copyright (c) 1994, Regents of the University of California
64 : *
65 : * src/backend/access/transam/multixact.c
66 : *
67 : *-------------------------------------------------------------------------
68 : */
69 : #include "postgres.h"
70 :
71 : #include "access/multixact.h"
72 : #include "access/slru.h"
73 : #include "access/transam.h"
74 : #include "access/twophase.h"
75 : #include "access/twophase_rmgr.h"
76 : #include "access/xact.h"
77 : #include "access/xloginsert.h"
78 : #include "access/xlogutils.h"
79 : #include "catalog/pg_type.h"
80 : #include "commands/dbcommands.h"
81 : #include "funcapi.h"
82 : #include "lib/ilist.h"
83 : #include "miscadmin.h"
84 : #include "pg_trace.h"
85 : #include "postmaster/autovacuum.h"
86 : #include "storage/lmgr.h"
87 : #include "storage/pmsignal.h"
88 : #include "storage/proc.h"
89 : #include "storage/procarray.h"
90 : #include "utils/builtins.h"
91 : #include "utils/memutils.h"
92 : #include "utils/snapmgr.h"
93 :
94 :
95 : /*
96 : * Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
97 : * used everywhere else in Postgres.
98 : *
99 : * Note: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
100 : * MultiXact page numbering also wraps around at
101 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
102 : * 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need
103 : * take no explicit notice of that fact in this module, except when comparing
104 : * segment and page numbers in TruncateMultiXact (see
105 : * MultiXactOffsetPagePrecedes).
106 : */
107 :
108 : /* We need four bytes per offset */
109 : #define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
110 :
111 : #define MultiXactIdToOffsetPage(xid) \
112 : ((xid) / (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
113 : #define MultiXactIdToOffsetEntry(xid) \
114 : ((xid) % (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
115 : #define MultiXactIdToOffsetSegment(xid) (MultiXactIdToOffsetPage(xid) / SLRU_PAGES_PER_SEGMENT)
116 :
117 : /*
118 : * The situation for members is a bit more complex: we store one byte of
119 : * additional flag bits for each TransactionId. To do this without getting
120 : * into alignment issues, we store four bytes of flags, and then the
121 : * corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
122 : * are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
123 : * per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
124 : * performance) trumps space efficiency here.
125 : *
126 : * Note that the "offset" macros work with byte offset, not array indexes, so
127 : * arithmetic must be done using "char *" pointers.
128 : */
129 : /* We need eight bits per xact, so one xact fits in a byte */
130 : #define MXACT_MEMBER_BITS_PER_XACT 8
131 : #define MXACT_MEMBER_FLAGS_PER_BYTE 1
132 : #define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
133 :
134 : /* how many full bytes of flags are there in a group? */
135 : #define MULTIXACT_FLAGBYTES_PER_GROUP 4
136 : #define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
137 : (MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
138 : /* size in bytes of a complete group */
139 : #define MULTIXACT_MEMBERGROUP_SIZE \
140 : (sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
141 : #define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
142 : #define MULTIXACT_MEMBERS_PER_PAGE \
143 : (MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
144 :
145 : /*
146 : * Because the number of items per page is not a divisor of the last item
147 : * number (member 0xFFFFFFFF), the last segment does not use the maximum number
148 : * of pages, and moreover the last used page therein does not use the same
149 : * number of items as previous pages. (Another way to say it is that the
150 : * 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
151 : * has some empty space after that item.)
152 : *
153 : * This constant is the number of members in the last page of the last segment.
154 : */
155 : #define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
156 : ((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
157 :
158 : /* page in which a member is to be found */
159 : #define MXOffsetToMemberPage(xid) ((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_PAGE)
160 : #define MXOffsetToMemberSegment(xid) (MXOffsetToMemberPage(xid) / SLRU_PAGES_PER_SEGMENT)
161 :
162 : /* Location (byte offset within page) of flag word for a given member */
163 : #define MXOffsetToFlagsOffset(xid) \
164 : ((((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) % \
165 : (TransactionId) MULTIXACT_MEMBERGROUPS_PER_PAGE) * \
166 : (TransactionId) MULTIXACT_MEMBERGROUP_SIZE)
167 : #define MXOffsetToFlagsBitShift(xid) \
168 : (((xid) % (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) * \
169 : MXACT_MEMBER_BITS_PER_XACT)
170 :
171 : /* Location (byte offset within page) of TransactionId of given member */
172 : #define MXOffsetToMemberOffset(xid) \
173 : (MXOffsetToFlagsOffset(xid) + MULTIXACT_FLAGBYTES_PER_GROUP + \
174 : ((xid) % MULTIXACT_MEMBERS_PER_MEMBERGROUP) * sizeof(TransactionId))
175 :
176 : /* Multixact members wraparound thresholds. */
177 : #define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
178 : #define MULTIXACT_MEMBER_DANGER_THRESHOLD \
179 : (MaxMultiXactOffset - MaxMultiXactOffset / 4)
180 :
181 : #define PreviousMultiXactId(xid) \
182 : ((xid) == FirstMultiXactId ? MaxMultiXactId : (xid) - 1)
183 :
184 : /*
185 : * Links to shared-memory data structures for MultiXact control
186 : */
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 MultiXactOffsetSLRULock and
196 : * MultiXactMemberSLRULock to guard accesses to the two sets of SLRU
197 : * buffers. For concurrency's sake, we avoid holding more than one of these
198 : * locks at a time.)
199 : */
200 : typedef struct MultiXactStateData
201 : {
202 : /* next-to-be-assigned MultiXactId */
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 : * Per-backend data starts here. We have two arrays stored in the area
238 : * immediately following the MultiXactStateData struct. Each is indexed by
239 : * BackendId.
240 : *
241 : * In both arrays, there's a slot for all normal backends (1..MaxBackends)
242 : * followed by a slot for max_prepared_xacts prepared transactions. Valid
243 : * BackendIds start from 1; element zero of each array is never used.
244 : *
245 : * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
246 : * transaction(s) could possibly be a member of, or InvalidMultiXactId
247 : * when the backend has no live transaction that could possibly be a
248 : * member of a MultiXact. Each backend sets its entry to the current
249 : * nextMXact counter just before first acquiring a shared lock in a given
250 : * transaction, and clears it at transaction end. (This works because only
251 : * during or after acquiring a shared lock could an XID possibly become a
252 : * member of a MultiXact, and that MultiXact would have to be created
253 : * during or after the lock acquisition.)
254 : *
255 : * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
256 : * current transaction(s) think is potentially live, or InvalidMultiXactId
257 : * when not in a transaction or not in a transaction that's paid any
258 : * attention to MultiXacts yet. This is computed when first needed in a
259 : * given transaction, and cleared at transaction end. We can compute it
260 : * as the minimum of the valid OldestMemberMXactId[] entries at the time
261 : * we compute it (using nextMXact if none are valid). Each backend is
262 : * required not to attempt to access any SLRU data for MultiXactIds older
263 : * than its own OldestVisibleMXactId[] setting; this is necessary because
264 : * the relevant SLRU data can be concurrently truncated away.
265 : *
266 : * The oldest valid value among all of the OldestMemberMXactId[] and
267 : * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
268 : * possible value still having any live member transaction -- OldestMxact.
269 : * Any value older than that is typically removed from tuple headers, or
270 : * "frozen" via being replaced with a new xmax. VACUUM can sometimes even
271 : * remove an individual MultiXact xmax whose value is >= its OldestMxact
272 : * cutoff, though typically only when no individual member XID is still
273 : * running. See FreezeMultiXactId for full details.
274 : *
275 : * Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
276 : * or the oldest extant Multi remaining in the table is used as the new
277 : * pg_class.relminmxid value (whichever is earlier). The minimum of all
278 : * relminmxid values in each database is stored in pg_database.datminmxid.
279 : * In turn, the minimum of all of those values is stored in pg_control.
280 : * This is used as the truncation point for pg_multixact when unneeded
281 : * segments get removed by vac_truncate_clog() during vacuuming.
282 : */
283 : MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
284 : } MultiXactStateData;
285 :
286 : /*
287 : * Last element of OldestMemberMXactId and OldestVisibleMXactId arrays.
288 : * Valid elements are (1..MaxOldestSlot); element 0 is never used.
289 : */
290 : #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
291 :
292 : /* Pointers to the state data in shared memory */
293 : static MultiXactStateData *MultiXactState;
294 : static MultiXactId *OldestMemberMXactId;
295 : static MultiXactId *OldestVisibleMXactId;
296 :
297 :
298 : /*
299 : * Definitions for the backend-local MultiXactId cache.
300 : *
301 : * We use this cache to store known MultiXacts, so we don't need to go to
302 : * SLRU areas every time.
303 : *
304 : * The cache lasts for the duration of a single transaction, the rationale
305 : * for this being that most entries will contain our own TransactionId and
306 : * so they will be uninteresting by the time our next transaction starts.
307 : * (XXX not clear that this is correct --- other members of the MultiXact
308 : * could hang around longer than we did. However, it's not clear what a
309 : * better policy for flushing old cache entries would be.) FIXME actually
310 : * this is plain wrong now that multixact's may contain update Xids.
311 : *
312 : * We allocate the cache entries in a memory context that is deleted at
313 : * transaction end, so we don't need to do retail freeing of entries.
314 : */
315 : typedef struct mXactCacheEnt
316 : {
317 : MultiXactId multi;
318 : int nmembers;
319 : dlist_node node;
320 : MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
321 : } mXactCacheEnt;
322 :
323 : #define MAX_CACHE_ENTRIES 256
324 : static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
325 : static MemoryContext MXactContext = NULL;
326 :
327 : #ifdef MULTIXACT_DEBUG
328 : #define debug_elog2(a,b) elog(a,b)
329 : #define debug_elog3(a,b,c) elog(a,b,c)
330 : #define debug_elog4(a,b,c,d) elog(a,b,c,d)
331 : #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
332 : #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
333 : #else
334 : #define debug_elog2(a,b)
335 : #define debug_elog3(a,b,c)
336 : #define debug_elog4(a,b,c,d)
337 : #define debug_elog5(a,b,c,d,e)
338 : #define debug_elog6(a,b,c,d,e,f)
339 : #endif
340 :
341 : /* internal MultiXactId management */
342 : static void MultiXactIdSetOldestVisible(void);
343 : static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
344 : int nmembers, MultiXactMember *members);
345 : static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
346 :
347 : /* MultiXact cache management */
348 : static int mxactMemberComparator(const void *arg1, const void *arg2);
349 : static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
350 : static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
351 : static void mXactCachePut(MultiXactId multi, int nmembers,
352 : MultiXactMember *members);
353 :
354 : static char *mxstatus_to_string(MultiXactStatus status);
355 :
356 : /* management of SLRU infrastructure */
357 : static int ZeroMultiXactOffsetPage(int pageno, bool writeXlog);
358 : static int ZeroMultiXactMemberPage(int pageno, bool writeXlog);
359 : static bool MultiXactOffsetPagePrecedes(int page1, int page2);
360 : static bool MultiXactMemberPagePrecedes(int page1, int page2);
361 : static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
362 : MultiXactOffset offset2);
363 : static void ExtendMultiXactOffset(MultiXactId multi);
364 : static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
365 : static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
366 : MultiXactOffset start, uint32 distance);
367 : static bool SetOffsetVacuumLimit(bool is_startup);
368 : static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
369 : static void WriteMZeroPageXlogRec(int pageno, uint8 info);
370 : static void WriteMTruncateXlogRec(Oid oldestMultiDB,
371 : MultiXactId startTruncOff,
372 : MultiXactId endTruncOff,
373 : MultiXactOffset startTruncMemb,
374 : MultiXactOffset endTruncMemb);
375 :
376 :
377 : /*
378 : * MultiXactIdCreate
379 : * Construct a MultiXactId representing two TransactionIds.
380 : *
381 : * The two XIDs must be different, or be requesting different statuses.
382 : *
383 : * NB - we don't worry about our local MultiXactId cache here, because that
384 : * is handled by the lower-level routines.
385 : */
386 : MultiXactId
387 GIC 1019 : MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
388 ECB : TransactionId xid2, MultiXactStatus status2)
389 : {
390 : MultiXactId newMulti;
391 : MultiXactMember members[2];
392 :
393 GNC 1019 : Assert(TransactionIdIsValid(xid1));
394 1019 : Assert(TransactionIdIsValid(xid2));
395 ECB :
396 GIC 1019 : Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
397 ECB :
398 : /* MultiXactIdSetOldestMember() must have been called already. */
399 GIC 1019 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyBackendId]));
400 ECB :
401 : /*
402 : * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
403 : * are still running. In typical usage, xid2 will be our own XID and the
404 : * caller just did a check on xid1, so it'd be wasted effort.
405 : */
406 :
407 GIC 1019 : members[0].xid = xid1;
408 CBC 1019 : members[0].status = status1;
409 1019 : members[1].xid = xid2;
410 1019 : members[1].status = status2;
411 ECB :
412 GIC 1019 : newMulti = MultiXactIdCreateFromMembers(2, members);
413 ECB :
414 : debug_elog3(DEBUG2, "Create: %s",
415 : mxid_to_string(newMulti, 2, members));
416 :
417 GIC 1019 : return newMulti;
418 ECB : }
419 :
420 : /*
421 : * MultiXactIdExpand
422 : * Add a TransactionId to a pre-existing MultiXactId.
423 : *
424 : * If the TransactionId is already a member of the passed MultiXactId with the
425 : * same status, just return it as-is.
426 : *
427 : * Note that we do NOT actually modify the membership of a pre-existing
428 : * MultiXactId; instead we create a new one. This is necessary to avoid
429 : * a race condition against code trying to wait for one MultiXactId to finish;
430 : * see notes in heapam.c.
431 : *
432 : * NB - we don't worry about our local MultiXactId cache here, because that
433 : * is handled by the lower-level routines.
434 : *
435 : * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
436 : * one upgraded by pg_upgrade from a cluster older than this feature) are not
437 : * passed in.
438 : */
439 : MultiXactId
440 GIC 97 : MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
441 ECB : {
442 : MultiXactId newMulti;
443 : MultiXactMember *members;
444 : MultiXactMember *newMembers;
445 : int nmembers;
446 : int i;
447 : int j;
448 :
449 GNC 97 : Assert(MultiXactIdIsValid(multi));
450 97 : Assert(TransactionIdIsValid(xid));
451 ECB :
452 : /* MultiXactIdSetOldestMember() must have been called already. */
453 GIC 97 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyBackendId]));
454 ECB :
455 : debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
456 : multi, xid, mxstatus_to_string(status));
457 :
458 : /*
459 : * Note: we don't allow for old multis here. The reason is that the only
460 : * caller of this function does a check that the multixact is no longer
461 : * running.
462 : */
463 GIC 97 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
464 ECB :
465 GIC 97 : if (nmembers < 0)
466 ECB : {
467 : MultiXactMember member;
468 :
469 : /*
470 : * The MultiXactId is obsolete. This can only happen if all the
471 : * MultiXactId members stop running between the caller checking and
472 : * passing it to us. It would be better to return that fact to the
473 : * caller, but it would complicate the API and it's unlikely to happen
474 : * too often, so just deal with it by creating a singleton MultiXact.
475 : */
476 UIC 0 : member.xid = xid;
477 UBC 0 : member.status = status;
478 0 : newMulti = MultiXactIdCreateFromMembers(1, &member);
479 EUB :
480 : debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
481 : multi, newMulti);
482 UIC 0 : return newMulti;
483 EUB : }
484 :
485 : /*
486 : * If the TransactionId is already a member of the MultiXactId with the
487 : * same status, just return the existing MultiXactId.
488 : */
489 GIC 301 : for (i = 0; i < nmembers; i++)
490 ECB : {
491 GIC 204 : if (TransactionIdEquals(members[i].xid, xid) &&
492 CBC 54 : (members[i].status == status))
493 ECB : {
494 : debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
495 : xid, multi);
496 UIC 0 : pfree(members);
497 UBC 0 : return multi;
498 EUB : }
499 : }
500 :
501 : /*
502 : * Determine which of the members of the MultiXactId are still of
503 : * interest. This is any running transaction, and also any transaction
504 : * that grabbed something stronger than just a lock and was committed. (An
505 : * update that aborted is of no interest here; and having more than one
506 : * update Xid in a multixact would cause errors elsewhere.)
507 : *
508 : * Removing dead members is not just an optimization: freezing of tuples
509 : * whose Xmax are multis depends on this behavior.
510 : *
511 : * Note we have the same race condition here as above: j could be 0 at the
512 : * end of the loop.
513 : */
514 : newMembers = (MultiXactMember *)
515 GIC 97 : palloc(sizeof(MultiXactMember) * (nmembers + 1));
516 ECB :
517 GIC 301 : for (i = 0, j = 0; i < nmembers; i++)
518 ECB : {
519 GIC 204 : if (TransactionIdIsInProgress(members[i].xid) ||
520 CBC 44 : (ISUPDATE_from_mxstatus(members[i].status) &&
521 6 : TransactionIdDidCommit(members[i].xid)))
522 ECB : {
523 GIC 166 : newMembers[j].xid = members[i].xid;
524 CBC 166 : newMembers[j++].status = members[i].status;
525 ECB : }
526 : }
527 :
528 GIC 97 : newMembers[j].xid = xid;
529 CBC 97 : newMembers[j++].status = status;
530 97 : newMulti = MultiXactIdCreateFromMembers(j, newMembers);
531 ECB :
532 GIC 97 : pfree(members);
533 CBC 97 : pfree(newMembers);
534 ECB :
535 : debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
536 :
537 GIC 97 : return newMulti;
538 ECB : }
539 :
540 : /*
541 : * MultiXactIdIsRunning
542 : * Returns whether a MultiXactId is "running".
543 : *
544 : * We return true if at least one member of the given MultiXactId is still
545 : * running. Note that a "false" result is certain not to change,
546 : * because it is not legal to add members to an existing MultiXactId.
547 : *
548 : * Caller is expected to have verified that the multixact does not come from
549 : * a pg_upgraded share-locked tuple.
550 : */
551 : bool
552 GIC 961 : MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
553 ECB : {
554 : MultiXactMember *members;
555 : int nmembers;
556 : int i;
557 :
558 : debug_elog3(DEBUG2, "IsRunning %u?", multi);
559 :
560 : /*
561 : * "false" here means we assume our callers have checked that the given
562 : * multi cannot possibly come from a pg_upgraded database.
563 : */
564 GIC 961 : nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
565 ECB :
566 GIC 961 : if (nmembers <= 0)
567 ECB : {
568 : debug_elog2(DEBUG2, "IsRunning: no members");
569 GIC 643 : return false;
570 ECB : }
571 :
572 : /*
573 : * Checking for myself is cheap compared to looking in shared memory;
574 : * return true if any live subtransaction of the current top-level
575 : * transaction is a member.
576 : *
577 : * This is not needed for correctness, it's just a fast path.
578 : */
579 GIC 736 : for (i = 0; i < nmembers; i++)
580 ECB : {
581 GIC 573 : if (TransactionIdIsCurrentTransactionId(members[i].xid))
582 ECB : {
583 : debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
584 GIC 155 : pfree(members);
585 CBC 155 : return true;
586 ECB : }
587 : }
588 :
589 : /*
590 : * This could be made faster by having another entry point in procarray.c,
591 : * walking the PGPROC array only once for all the members. But in most
592 : * cases nmembers should be small enough that it doesn't much matter.
593 : */
594 GIC 296 : for (i = 0; i < nmembers; i++)
595 ECB : {
596 GIC 248 : if (TransactionIdIsInProgress(members[i].xid))
597 ECB : {
598 : debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
599 : i, members[i].xid);
600 GIC 115 : pfree(members);
601 CBC 115 : return true;
602 ECB : }
603 : }
604 :
605 GIC 48 : pfree(members);
606 ECB :
607 : debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
608 :
609 GIC 48 : return false;
610 ECB : }
611 :
612 : /*
613 : * MultiXactIdSetOldestMember
614 : * Save the oldest MultiXactId this transaction could be a member of.
615 : *
616 : * We set the OldestMemberMXactId for a given transaction the first time it's
617 : * going to do some operation that might require a MultiXactId (tuple lock,
618 : * update or delete). We need to do this even if we end up using a
619 : * TransactionId instead of a MultiXactId, because there is a chance that
620 : * another transaction would add our XID to a MultiXactId.
621 : *
622 : * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
623 : * be called just before doing any such possibly-MultiXactId-able operation.
624 : */
625 : void
626 GIC 1917382 : MultiXactIdSetOldestMember(void)
627 ECB : {
628 GIC 1917382 : if (!MultiXactIdIsValid(OldestMemberMXactId[MyBackendId]))
629 ECB : {
630 : MultiXactId nextMXact;
631 :
632 : /*
633 : * You might think we don't need to acquire a lock here, since
634 : * fetching and storing of TransactionIds is probably atomic, but in
635 : * fact we do: suppose we pick up nextMXact and then lose the CPU for
636 : * a long time. Someone else could advance nextMXact, and then
637 : * another someone else could compute an OldestVisibleMXactId that
638 : * would be after the value we are going to store when we get control
639 : * back. Which would be wrong.
640 : *
641 : * Note that a shared lock is sufficient, because it's enough to stop
642 : * someone from advancing nextMXact; and nobody else could be trying
643 : * to write to our OldestMember entry, only reading (and we assume
644 : * storing it is atomic.)
645 : */
646 GIC 179372 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
647 ECB :
648 : /*
649 : * We have to beware of the possibility that nextMXact is in the
650 : * wrapped-around state. We don't fix the counter itself here, but we
651 : * must be sure to store a valid value in our array entry.
652 : */
653 GIC 179372 : nextMXact = MultiXactState->nextMXact;
654 CBC 179372 : if (nextMXact < FirstMultiXactId)
655 LBC 0 : nextMXact = FirstMultiXactId;
656 EUB :
657 GIC 179372 : OldestMemberMXactId[MyBackendId] = nextMXact;
658 ECB :
659 GIC 179372 : LWLockRelease(MultiXactGenLock);
660 ECB :
661 : debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
662 : MyBackendId, nextMXact);
663 : }
664 GIC 1917382 : }
665 ECB :
666 : /*
667 : * MultiXactIdSetOldestVisible
668 : * Save the oldest MultiXactId this transaction considers possibly live.
669 : *
670 : * We set the OldestVisibleMXactId for a given transaction the first time
671 : * it's going to inspect any MultiXactId. Once we have set this, we are
672 : * guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
673 : * won't be truncated away.
674 : *
675 : * The value to set is the oldest of nextMXact and all the valid per-backend
676 : * OldestMemberMXactId[] entries. Because of the locking we do, we can be
677 : * certain that no subsequent call to MultiXactIdSetOldestMember can set
678 : * an OldestMemberMXactId[] entry older than what we compute here. Therefore
679 : * there is no live transaction, now or later, that can be a member of any
680 : * MultiXactId older than the OldestVisibleMXactId we compute here.
681 : */
682 : static void
683 GIC 877 : MultiXactIdSetOldestVisible(void)
684 ECB : {
685 GIC 877 : if (!MultiXactIdIsValid(OldestVisibleMXactId[MyBackendId]))
686 ECB : {
687 : MultiXactId oldestMXact;
688 : int i;
689 :
690 GIC 222 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
691 ECB :
692 : /*
693 : * We have to beware of the possibility that nextMXact is in the
694 : * wrapped-around state. We don't fix the counter itself here, but we
695 : * must be sure to store a valid value in our array entry.
696 : */
697 GIC 222 : oldestMXact = MultiXactState->nextMXact;
698 CBC 222 : if (oldestMXact < FirstMultiXactId)
699 LBC 0 : oldestMXact = FirstMultiXactId;
700 EUB :
701 GIC 26738 : for (i = 1; i <= MaxOldestSlot; i++)
702 ECB : {
703 GIC 26516 : MultiXactId thisoldest = OldestMemberMXactId[i];
704 ECB :
705 GIC 26740 : if (MultiXactIdIsValid(thisoldest) &&
706 CBC 224 : MultiXactIdPrecedes(thisoldest, oldestMXact))
707 114 : oldestMXact = thisoldest;
708 ECB : }
709 :
710 GIC 222 : OldestVisibleMXactId[MyBackendId] = oldestMXact;
711 ECB :
712 GIC 222 : LWLockRelease(MultiXactGenLock);
713 ECB :
714 : debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
715 : MyBackendId, oldestMXact);
716 : }
717 GIC 877 : }
718 ECB :
719 : /*
720 : * ReadNextMultiXactId
721 : * Return the next MultiXactId to be assigned, but don't allocate it
722 : */
723 : MultiXactId
724 GIC 81007 : ReadNextMultiXactId(void)
725 ECB : {
726 : MultiXactId mxid;
727 :
728 : /* XXX we could presumably do this without a lock. */
729 GIC 81007 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
730 CBC 81007 : mxid = MultiXactState->nextMXact;
731 81007 : LWLockRelease(MultiXactGenLock);
732 ECB :
733 GIC 81007 : if (mxid < FirstMultiXactId)
734 LBC 0 : mxid = FirstMultiXactId;
735 EUB :
736 GIC 81007 : return mxid;
737 ECB : }
738 :
739 : /*
740 : * ReadMultiXactIdRange
741 : * Get the range of IDs that may still be referenced by a relation.
742 : */
743 : void
744 GIC 1202 : ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
745 ECB : {
746 GIC 1202 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
747 CBC 1202 : *oldest = MultiXactState->oldestMultiXactId;
748 1202 : *next = MultiXactState->nextMXact;
749 1202 : LWLockRelease(MultiXactGenLock);
750 ECB :
751 GIC 1202 : if (*oldest < FirstMultiXactId)
752 LBC 0 : *oldest = FirstMultiXactId;
753 GBC 1202 : if (*next < FirstMultiXactId)
754 LBC 0 : *next = FirstMultiXactId;
755 GBC 1202 : }
756 ECB :
757 :
758 : /*
759 : * MultiXactIdCreateFromMembers
760 : * Make a new MultiXactId from the specified set of members
761 : *
762 : * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
763 : * given TransactionIds as members. Returns the newly created MultiXactId.
764 : *
765 : * NB: the passed members[] array will be sorted in-place.
766 : */
767 : MultiXactId
768 GIC 1117 : MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
769 ECB : {
770 : MultiXactId multi;
771 : MultiXactOffset offset;
772 : xl_multixact_create xlrec;
773 :
774 : debug_elog3(DEBUG2, "Create: %s",
775 : mxid_to_string(InvalidMultiXactId, nmembers, members));
776 :
777 : /*
778 : * See if the same set of members already exists in our cache; if so, just
779 : * re-use that MultiXactId. (Note: it might seem that looking in our
780 : * cache is insufficient, and we ought to search disk to see if a
781 : * duplicate definition already exists. But since we only ever create
782 : * MultiXacts containing our own XID, in most cases any such MultiXacts
783 : * were in fact created by us, and so will be in our cache. There are
784 : * corner cases where someone else added us to a MultiXact without our
785 : * knowledge, but it's not worth checking for.)
786 : */
787 GIC 1117 : multi = mXactCacheGetBySet(nmembers, members);
788 CBC 1117 : if (MultiXactIdIsValid(multi))
789 ECB : {
790 : debug_elog2(DEBUG2, "Create: in cache!");
791 GIC 845 : return multi;
792 ECB : }
793 :
794 : /* Verify that there is a single update Xid among the given members. */
795 : {
796 : int i;
797 GIC 272 : bool has_update = false;
798 ECB :
799 GIC 879 : for (i = 0; i < nmembers; i++)
800 ECB : {
801 GIC 607 : if (ISUPDATE_from_mxstatus(members[i].status))
802 ECB : {
803 GIC 130 : if (has_update)
804 LBC 0 : elog(ERROR, "new multixact has more than one updating member: %s",
805 EUB : mxid_to_string(InvalidMultiXactId, nmembers, members));
806 GIC 130 : has_update = true;
807 ECB : }
808 : }
809 : }
810 :
811 : /*
812 : * Assign the MXID and offsets range to use, and make sure there is space
813 : * in the OFFSETs and MEMBERs files. NB: this routine does
814 : * START_CRIT_SECTION().
815 : *
816 : * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
817 : * that we've called MultiXactIdSetOldestMember here. This is because
818 : * this routine is used in some places to create new MultiXactIds of which
819 : * the current backend is not a member, notably during freezing of multis
820 : * in vacuum. During vacuum, in particular, it would be unacceptable to
821 : * keep OldestMulti set, in case it runs for long.
822 : */
823 GIC 272 : multi = GetNewMultiXactId(nmembers, &offset);
824 ECB :
825 : /* Make an XLOG entry describing the new MXID. */
826 GIC 272 : xlrec.mid = multi;
827 CBC 272 : xlrec.moff = offset;
828 272 : xlrec.nmembers = nmembers;
829 ECB :
830 : /*
831 : * XXX Note: there's a lot of padding space in MultiXactMember. We could
832 : * find a more compact representation of this Xlog record -- perhaps all
833 : * the status flags in one XLogRecData, then all the xids in another one?
834 : * Not clear that it's worth the trouble though.
835 : */
836 GIC 272 : XLogBeginInsert();
837 CBC 272 : XLogRegisterData((char *) (&xlrec), SizeOfMultiXactCreate);
838 272 : XLogRegisterData((char *) members, nmembers * sizeof(MultiXactMember));
839 ECB :
840 GIC 272 : (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
841 ECB :
842 : /* Now enter the information into the OFFSETs and MEMBERs logs */
843 GIC 272 : RecordNewMultiXact(multi, offset, nmembers, members);
844 ECB :
845 : /* Done with critical section */
846 GIC 272 : END_CRIT_SECTION();
847 ECB :
848 : /* Store the new MultiXactId in the local cache, too */
849 GIC 272 : mXactCachePut(multi, nmembers, members);
850 ECB :
851 : debug_elog2(DEBUG2, "Create: all done");
852 :
853 GIC 272 : return multi;
854 ECB : }
855 :
856 : /*
857 : * RecordNewMultiXact
858 : * Write info about a new multixact into the offsets and members files
859 : *
860 : * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
861 : * use it.
862 : */
863 : static void
864 GIC 274 : RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
865 ECB : int nmembers, MultiXactMember *members)
866 : {
867 : int pageno;
868 : int prev_pageno;
869 : int entryno;
870 : int slotno;
871 : MultiXactOffset *offptr;
872 : int i;
873 :
874 GIC 274 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
875 ECB :
876 GIC 274 : pageno = MultiXactIdToOffsetPage(multi);
877 CBC 274 : entryno = MultiXactIdToOffsetEntry(multi);
878 ECB :
879 : /*
880 : * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
881 : * to complain about if there's any I/O error. This is kinda bogus, but
882 : * since the errors will always give the full pathname, it should be clear
883 : * enough that a MultiXactId is really involved. Perhaps someday we'll
884 : * take the trouble to generalize the slru.c error reporting code.
885 : */
886 GIC 274 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
887 CBC 274 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
888 274 : offptr += entryno;
889 ECB :
890 GIC 274 : *offptr = offset;
891 ECB :
892 GIC 274 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
893 ECB :
894 : /* Exchange our lock */
895 GIC 274 : LWLockRelease(MultiXactOffsetSLRULock);
896 ECB :
897 GIC 274 : LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
898 ECB :
899 GIC 274 : prev_pageno = -1;
900 ECB :
901 GIC 885 : for (i = 0; i < nmembers; i++, offset++)
902 ECB : {
903 : TransactionId *memberptr;
904 : uint32 *flagsptr;
905 : uint32 flagsval;
906 : int bshift;
907 : int flagsoff;
908 : int memberoff;
909 :
910 GIC 611 : Assert(members[i].status <= MultiXactStatusUpdate);
911 ECB :
912 GIC 611 : pageno = MXOffsetToMemberPage(offset);
913 CBC 611 : memberoff = MXOffsetToMemberOffset(offset);
914 611 : flagsoff = MXOffsetToFlagsOffset(offset);
915 611 : bshift = MXOffsetToFlagsBitShift(offset);
916 ECB :
917 GIC 611 : if (pageno != prev_pageno)
918 ECB : {
919 GIC 274 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
920 CBC 274 : prev_pageno = pageno;
921 ECB : }
922 :
923 GIC 611 : memberptr = (TransactionId *)
924 CBC 611 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
925 ECB :
926 GIC 611 : *memberptr = members[i].xid;
927 ECB :
928 GIC 611 : flagsptr = (uint32 *)
929 CBC 611 : (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
930 ECB :
931 GIC 611 : flagsval = *flagsptr;
932 CBC 611 : flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
933 611 : flagsval |= (members[i].status << bshift);
934 611 : *flagsptr = flagsval;
935 ECB :
936 GIC 611 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
937 ECB : }
938 :
939 GIC 274 : LWLockRelease(MultiXactMemberSLRULock);
940 CBC 274 : }
941 ECB :
942 : /*
943 : * GetNewMultiXactId
944 : * Get the next MultiXactId.
945 : *
946 : * Also, reserve the needed amount of space in the "members" area. The
947 : * starting offset of the reserved space is returned in *offset.
948 : *
949 : * This may generate XLOG records for expansion of the offsets and/or members
950 : * files. Unfortunately, we have to do that while holding MultiXactGenLock
951 : * to avoid race conditions --- the XLOG record for zeroing a page must appear
952 : * before any backend can possibly try to store data in that page!
953 : *
954 : * We start a critical section before advancing the shared counters. The
955 : * caller must end the critical section after writing SLRU data.
956 : */
957 : static MultiXactId
958 GIC 272 : GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
959 ECB : {
960 : MultiXactId result;
961 : MultiXactOffset nextOffset;
962 :
963 : debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
964 :
965 : /* safety check, we should never get this far in a HS standby */
966 GIC 272 : if (RecoveryInProgress())
967 LBC 0 : elog(ERROR, "cannot assign MultiXactIds during recovery");
968 EUB :
969 GIC 272 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
970 ECB :
971 : /* Handle wraparound of the nextMXact counter */
972 GIC 272 : if (MultiXactState->nextMXact < FirstMultiXactId)
973 LBC 0 : MultiXactState->nextMXact = FirstMultiXactId;
974 EUB :
975 : /* Assign the MXID */
976 GIC 272 : result = MultiXactState->nextMXact;
977 ECB :
978 : /*----------
979 : * Check to see if it's safe to assign another MultiXactId. This protects
980 : * against catastrophic data loss due to multixact wraparound. The basic
981 : * rules are:
982 : *
983 : * If we're past multiVacLimit or the safe threshold for member storage
984 : * space, or we don't know what the safe threshold for member storage is,
985 : * start trying to force autovacuum cycles.
986 : * If we're past multiWarnLimit, start issuing warnings.
987 : * If we're past multiStopLimit, refuse to create new MultiXactIds.
988 : *
989 : * Note these are pretty much the same protections in GetNewTransactionId.
990 : *----------
991 : */
992 GIC 272 : if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
993 ECB : {
994 : /*
995 : * For safety's sake, we release MultiXactGenLock while sending
996 : * signals, warnings, etc. This is not so much because we care about
997 : * preserving concurrency in this situation, as to avoid any
998 : * possibility of deadlock while doing get_database_name(). First,
999 : * copy all the shared values we'll need in this path.
1000 : */
1001 UIC 0 : MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
1002 UBC 0 : MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
1003 0 : MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
1004 0 : Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
1005 EUB :
1006 UIC 0 : LWLockRelease(MultiXactGenLock);
1007 EUB :
1008 UIC 0 : if (IsUnderPostmaster &&
1009 UBC 0 : !MultiXactIdPrecedes(result, multiStopLimit))
1010 EUB : {
1011 UIC 0 : char *oldest_datname = get_database_name(oldest_datoid);
1012 EUB :
1013 : /*
1014 : * Immediately kick autovacuum into action as we're already in
1015 : * ERROR territory.
1016 : */
1017 UIC 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1018 EUB :
1019 : /* complain even if that DB has disappeared */
1020 UIC 0 : if (oldest_datname)
1021 UBC 0 : ereport(ERROR,
1022 EUB : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1023 : errmsg("database is not accepting commands that generate new MultiXactIds to avoid wraparound data loss in database \"%s\"",
1024 : oldest_datname),
1025 : errhint("Execute a database-wide VACUUM in that database.\n"
1026 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1027 : else
1028 UIC 0 : ereport(ERROR,
1029 EUB : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1030 : errmsg("database is not accepting commands that generate new MultiXactIds to avoid wraparound data loss in database with OID %u",
1031 : oldest_datoid),
1032 : errhint("Execute a database-wide VACUUM in that database.\n"
1033 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1034 : }
1035 :
1036 : /*
1037 : * To avoid swamping the postmaster with signals, we issue the autovac
1038 : * request only once per 64K multis generated. This still gives
1039 : * plenty of chances before we get into real trouble.
1040 : */
1041 UIC 0 : if (IsUnderPostmaster && (result % 65536) == 0)
1042 UBC 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1043 EUB :
1044 UIC 0 : if (!MultiXactIdPrecedes(result, multiWarnLimit))
1045 EUB : {
1046 UIC 0 : char *oldest_datname = get_database_name(oldest_datoid);
1047 EUB :
1048 : /* complain even if that DB has disappeared */
1049 UIC 0 : if (oldest_datname)
1050 UBC 0 : ereport(WARNING,
1051 EUB : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1052 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1053 : multiWrapLimit - result,
1054 : oldest_datname,
1055 : multiWrapLimit - result),
1056 : errhint("Execute a database-wide VACUUM in that database.\n"
1057 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1058 : else
1059 UIC 0 : ereport(WARNING,
1060 EUB : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1061 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1062 : multiWrapLimit - result,
1063 : oldest_datoid,
1064 : multiWrapLimit - result),
1065 : errhint("Execute a database-wide VACUUM in that database.\n"
1066 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1067 : }
1068 :
1069 : /* Re-acquire lock and start over */
1070 UIC 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1071 UBC 0 : result = MultiXactState->nextMXact;
1072 0 : if (result < FirstMultiXactId)
1073 0 : result = FirstMultiXactId;
1074 EUB : }
1075 :
1076 : /* Make sure there is room for the MXID in the file. */
1077 GIC 272 : ExtendMultiXactOffset(result);
1078 ECB :
1079 : /*
1080 : * Reserve the members space, similarly to above. Also, be careful not to
1081 : * return zero as the starting offset for any multixact. See
1082 : * GetMultiXactIdMembers() for motivation.
1083 : */
1084 GIC 272 : nextOffset = MultiXactState->nextOffset;
1085 CBC 272 : if (nextOffset == 0)
1086 ECB : {
1087 GIC 10 : *offset = 1;
1088 CBC 10 : nmembers++; /* allocate member slot 0 too */
1089 ECB : }
1090 : else
1091 GIC 262 : *offset = nextOffset;
1092 ECB :
1093 : /*----------
1094 : * Protect against overrun of the members space as well, with the
1095 : * following rules:
1096 : *
1097 : * If we're past offsetStopLimit, refuse to generate more multis.
1098 : * If we're close to offsetStopLimit, emit a warning.
1099 : *
1100 : * Arbitrarily, we start emitting warnings when we're 20 segments or less
1101 : * from offsetStopLimit.
1102 : *
1103 : * Note we haven't updated the shared state yet, so if we fail at this
1104 : * point, the multixact ID we grabbed can still be used by the next guy.
1105 : *
1106 : * Note that there is no point in forcing autovacuum runs here: the
1107 : * multixact freeze settings would have to be reduced for that to have any
1108 : * effect.
1109 : *----------
1110 : */
1111 : #define OFFSET_WARN_SEGMENTS 20
1112 GIC 544 : if (MultiXactState->oldestOffsetKnown &&
1113 CBC 272 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit, nextOffset,
1114 ECB : nmembers))
1115 : {
1116 : /* see comment in the corresponding offsets wraparound case */
1117 UIC 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1118 EUB :
1119 UIC 0 : ereport(ERROR,
1120 EUB : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1121 : errmsg("multixact \"members\" limit exceeded"),
1122 : errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
1123 : "This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
1124 : MultiXactState->offsetStopLimit - nextOffset - 1,
1125 : nmembers,
1126 : MultiXactState->offsetStopLimit - nextOffset - 1),
1127 : 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.",
1128 : MultiXactState->oldestMultiXactDB)));
1129 : }
1130 :
1131 : /*
1132 : * Check whether we should kick autovacuum into action, to prevent members
1133 : * wraparound. NB we use a much larger window to trigger autovacuum than
1134 : * just the warning limit. The warning is just a measure of last resort -
1135 : * this is in line with GetNewTransactionId's behaviour.
1136 : */
1137 GIC 272 : if (!MultiXactState->oldestOffsetKnown ||
1138 CBC 272 : (MultiXactState->nextOffset - MultiXactState->oldestOffset
1139 272 : > MULTIXACT_MEMBER_SAFE_THRESHOLD))
1140 ECB : {
1141 : /*
1142 : * To avoid swamping the postmaster with signals, we issue the autovac
1143 : * request only when crossing a segment boundary. With default
1144 : * compilation settings that's roughly after 50k members. This still
1145 : * gives plenty of chances before we get into real trouble.
1146 : */
1147 UIC 0 : if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
1148 UBC 0 : (MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
1149 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1150 EUB : }
1151 :
1152 GIC 544 : if (MultiXactState->oldestOffsetKnown &&
1153 CBC 272 : MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit,
1154 ECB : nextOffset,
1155 : nmembers + MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT * OFFSET_WARN_SEGMENTS))
1156 UIC 0 : ereport(WARNING,
1157 EUB : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1158 : errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
1159 : "database with OID %u must be vacuumed before %d more multixact members are used",
1160 : MultiXactState->offsetStopLimit - nextOffset + nmembers,
1161 : MultiXactState->oldestMultiXactDB,
1162 : MultiXactState->offsetStopLimit - nextOffset + nmembers),
1163 : errhint("Execute a database-wide VACUUM in that database with reduced vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age settings.")));
1164 :
1165 GIC 272 : ExtendMultiXactMember(nextOffset, nmembers);
1166 ECB :
1167 : /*
1168 : * Critical section from here until caller has written the data into the
1169 : * just-reserved SLRU space; we don't want to error out with a partly
1170 : * written MultiXact structure. (In particular, failing to write our
1171 : * start offset after advancing nextMXact would effectively corrupt the
1172 : * previous MultiXact.)
1173 : */
1174 GIC 272 : START_CRIT_SECTION();
1175 ECB :
1176 : /*
1177 : * Advance counters. As in GetNewTransactionId(), this must not happen
1178 : * until after file extension has succeeded!
1179 : *
1180 : * We don't care about MultiXactId wraparound here; it will be handled by
1181 : * the next iteration. But note that nextMXact may be InvalidMultiXactId
1182 : * or the first value on a segment-beginning page after this routine
1183 : * exits, so anyone else looking at the variable must be prepared to deal
1184 : * with either case. Similarly, nextOffset may be zero, but we won't use
1185 : * that as the actual start offset of the next multixact.
1186 : */
1187 GIC 272 : (MultiXactState->nextMXact)++;
1188 ECB :
1189 GIC 272 : MultiXactState->nextOffset += nmembers;
1190 ECB :
1191 GIC 272 : LWLockRelease(MultiXactGenLock);
1192 ECB :
1193 : debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
1194 GIC 272 : return result;
1195 ECB : }
1196 :
1197 : /*
1198 : * GetMultiXactIdMembers
1199 : * Return the set of MultiXactMembers that make up a MultiXactId
1200 : *
1201 : * Return value is the number of members found, or -1 if there are none,
1202 : * and *members is set to a newly palloc'ed array of members. It's the
1203 : * caller's responsibility to free it when done with it.
1204 : *
1205 : * from_pgupgrade must be passed as true if and only if only the multixact
1206 : * corresponds to a value from a tuple that was locked in a 9.2-or-older
1207 : * installation and later pg_upgrade'd (that is, the infomask is
1208 : * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1209 : * can still be running, so we return -1 just like for an empty multixact
1210 : * without any further checking. It would be wrong to try to resolve such a
1211 : * multixact: either the multixact is within the current valid multixact
1212 : * range, in which case the returned result would be bogus, or outside that
1213 : * range, in which case an error would be raised.
1214 : *
1215 : * In all other cases, the passed multixact must be within the known valid
1216 : * range, that is, greater to or equal than oldestMultiXactId, and less than
1217 : * nextMXact. Otherwise, an error is raised.
1218 : *
1219 : * isLockOnly must be set to true if caller is certain that the given multi
1220 : * is used only to lock tuples; can be false without loss of correctness,
1221 : * but passing a true means we can return quickly without checking for
1222 : * old updates.
1223 : */
1224 : int
1225 GIC 3001 : GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
1226 : bool from_pgupgrade, bool isLockOnly)
1227 : {
1228 : int pageno;
1229 : int prev_pageno;
1230 : int entryno;
1231 : int slotno;
1232 : MultiXactOffset *offptr;
1233 : MultiXactOffset offset;
1234 : int length;
1235 : int truelength;
1236 : int i;
1237 : MultiXactId oldestMXact;
1238 : MultiXactId nextMXact;
1239 : MultiXactId tmpMXact;
1240 : MultiXactOffset nextOffset;
1241 : MultiXactMember *ptr;
1242 :
1243 : debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1244 :
1245 3001 : if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1246 ECB : {
1247 UIC 0 : *members = NULL;
1248 UBC 0 : return -1;
1249 EUB : }
1250 :
1251 : /* See if the MultiXactId is in the local cache */
1252 GIC 3001 : length = mXactCacheGetById(multi, members);
1253 CBC 3001 : if (length >= 0)
1254 ECB : {
1255 : debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1256 : mxid_to_string(multi, length, *members));
1257 GIC 2124 : return length;
1258 ECB : }
1259 :
1260 : /* Set our OldestVisibleMXactId[] entry if we didn't already */
1261 GIC 877 : MultiXactIdSetOldestVisible();
1262 ECB :
1263 : /*
1264 : * If we know the multi is used only for locking and not for updates, then
1265 : * we can skip checking if the value is older than our oldest visible
1266 : * multi. It cannot possibly still be running.
1267 : */
1268 GNC 1601 : if (isLockOnly &&
1269 CBC 724 : MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyBackendId]))
1270 ECB : {
1271 : debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1272 GIC 644 : *members = NULL;
1273 CBC 644 : return -1;
1274 ECB : }
1275 :
1276 : /*
1277 : * We check known limits on MultiXact before resorting to the SLRU area.
1278 : *
1279 : * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1280 : * useful; it has already been removed, or will be removed shortly, by
1281 : * truncation. If one is passed, an error is raised.
1282 : *
1283 : * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1284 : * implies undetected ID wraparound has occurred. This raises a hard
1285 : * error.
1286 : *
1287 : * Shared lock is enough here since we aren't modifying any global state.
1288 : * Acquire it just long enough to grab the current counter values. We may
1289 : * need both nextMXact and nextOffset; see below.
1290 : */
1291 GIC 233 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1292 ECB :
1293 GIC 233 : oldestMXact = MultiXactState->oldestMultiXactId;
1294 CBC 233 : nextMXact = MultiXactState->nextMXact;
1295 233 : nextOffset = MultiXactState->nextOffset;
1296 ECB :
1297 GIC 233 : LWLockRelease(MultiXactGenLock);
1298 ECB :
1299 GIC 233 : if (MultiXactIdPrecedes(multi, oldestMXact))
1300 LBC 0 : ereport(ERROR,
1301 EUB : (errcode(ERRCODE_INTERNAL_ERROR),
1302 : errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1303 : multi)));
1304 :
1305 GIC 233 : if (!MultiXactIdPrecedes(multi, nextMXact))
1306 LBC 0 : ereport(ERROR,
1307 EUB : (errcode(ERRCODE_INTERNAL_ERROR),
1308 : errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1309 : multi)));
1310 :
1311 : /*
1312 : * Find out the offset at which we need to start reading MultiXactMembers
1313 : * and the number of members in the multixact. We determine the latter as
1314 : * the difference between this multixact's starting offset and the next
1315 : * one's. However, there are some corner cases to worry about:
1316 : *
1317 : * 1. This multixact may be the latest one created, in which case there is
1318 : * no next one to look at. In this case the nextOffset value we just
1319 : * saved is the correct endpoint.
1320 : *
1321 : * 2. The next multixact may still be in process of being filled in: that
1322 : * is, another process may have done GetNewMultiXactId but not yet written
1323 : * the offset entry for that ID. In that scenario, it is guaranteed that
1324 : * the offset entry for that multixact exists (because GetNewMultiXactId
1325 : * won't release MultiXactGenLock until it does) but contains zero
1326 : * (because we are careful to pre-zero offset pages). Because
1327 : * GetNewMultiXactId will never return zero as the starting offset for a
1328 : * multixact, when we read zero as the next multixact's offset, we know we
1329 : * have this case. We sleep for a bit and try again.
1330 : *
1331 : * 3. Because GetNewMultiXactId increments offset zero to offset one to
1332 : * handle case #2, there is an ambiguity near the point of offset
1333 : * wraparound. If we see next multixact's offset is one, is that our
1334 : * multixact's actual endpoint, or did it end at zero with a subsequent
1335 : * increment? We handle this using the knowledge that if the zero'th
1336 : * member slot wasn't filled, it'll contain zero, and zero isn't a valid
1337 : * transaction ID so it can't be a multixact member. Therefore, if we
1338 : * read a zero from the members array, just ignore it.
1339 : *
1340 : * This is all pretty messy, but the mess occurs only in infrequent corner
1341 : * cases, so it seems better than holding the MultiXactGenLock for a long
1342 : * time on every multixact creation.
1343 : */
1344 GIC 233 : retry:
1345 CBC 233 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
1346 ECB :
1347 GIC 233 : pageno = MultiXactIdToOffsetPage(multi);
1348 CBC 233 : entryno = MultiXactIdToOffsetEntry(multi);
1349 ECB :
1350 GIC 233 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1351 CBC 233 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1352 233 : offptr += entryno;
1353 233 : offset = *offptr;
1354 ECB :
1355 GIC 233 : Assert(offset != 0);
1356 ECB :
1357 : /*
1358 : * Use the same increment rule as GetNewMultiXactId(), that is, don't
1359 : * handle wraparound explicitly until needed.
1360 : */
1361 GIC 233 : tmpMXact = multi + 1;
1362 ECB :
1363 GIC 233 : if (nextMXact == tmpMXact)
1364 ECB : {
1365 : /* Corner case 1: there is no next multixact */
1366 GIC 206 : length = nextOffset - offset;
1367 ECB : }
1368 : else
1369 : {
1370 : MultiXactOffset nextMXOffset;
1371 :
1372 : /* handle wraparound if needed */
1373 GIC 27 : if (tmpMXact < FirstMultiXactId)
1374 LBC 0 : tmpMXact = FirstMultiXactId;
1375 EUB :
1376 GIC 27 : prev_pageno = pageno;
1377 ECB :
1378 GIC 27 : pageno = MultiXactIdToOffsetPage(tmpMXact);
1379 CBC 27 : entryno = MultiXactIdToOffsetEntry(tmpMXact);
1380 ECB :
1381 GIC 27 : if (pageno != prev_pageno)
1382 LBC 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1383 EUB :
1384 GIC 27 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1385 CBC 27 : offptr += entryno;
1386 27 : nextMXOffset = *offptr;
1387 ECB :
1388 GIC 27 : if (nextMXOffset == 0)
1389 ECB : {
1390 : /* Corner case 2: next multixact is still being filled in */
1391 UIC 0 : LWLockRelease(MultiXactOffsetSLRULock);
1392 UBC 0 : CHECK_FOR_INTERRUPTS();
1393 0 : pg_usleep(1000L);
1394 0 : goto retry;
1395 EUB : }
1396 :
1397 GIC 27 : length = nextMXOffset - offset;
1398 ECB : }
1399 :
1400 GIC 233 : LWLockRelease(MultiXactOffsetSLRULock);
1401 ECB :
1402 GIC 233 : ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1403 ECB :
1404 : /* Now get the members themselves. */
1405 GIC 233 : LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
1406 ECB :
1407 GIC 233 : truelength = 0;
1408 CBC 233 : prev_pageno = -1;
1409 835 : for (i = 0; i < length; i++, offset++)
1410 ECB : {
1411 : TransactionId *xactptr;
1412 : uint32 *flagsptr;
1413 : int flagsoff;
1414 : int bshift;
1415 : int memberoff;
1416 :
1417 GIC 602 : pageno = MXOffsetToMemberPage(offset);
1418 CBC 602 : memberoff = MXOffsetToMemberOffset(offset);
1419 ECB :
1420 GIC 602 : if (pageno != prev_pageno)
1421 ECB : {
1422 GIC 233 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1423 CBC 233 : prev_pageno = pageno;
1424 ECB : }
1425 :
1426 GIC 602 : xactptr = (TransactionId *)
1427 CBC 602 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1428 ECB :
1429 GIC 602 : if (!TransactionIdIsValid(*xactptr))
1430 ECB : {
1431 : /* Corner case 3: we must be looking at unused slot zero */
1432 UIC 0 : Assert(offset == 0);
1433 UBC 0 : continue;
1434 EUB : }
1435 :
1436 GIC 602 : flagsoff = MXOffsetToFlagsOffset(offset);
1437 CBC 602 : bshift = MXOffsetToFlagsBitShift(offset);
1438 602 : flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1439 ECB :
1440 GIC 602 : ptr[truelength].xid = *xactptr;
1441 CBC 602 : ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1442 602 : truelength++;
1443 ECB : }
1444 :
1445 GIC 233 : LWLockRelease(MultiXactMemberSLRULock);
1446 ECB :
1447 : /* A multixid with zero members should not happen */
1448 GIC 233 : Assert(truelength > 0);
1449 ECB :
1450 : /*
1451 : * Copy the result into the local cache.
1452 : */
1453 GIC 233 : mXactCachePut(multi, truelength, ptr);
1454 ECB :
1455 : debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1456 : mxid_to_string(multi, truelength, ptr));
1457 GIC 233 : *members = ptr;
1458 CBC 233 : return truelength;
1459 ECB : }
1460 :
1461 : /*
1462 : * mxactMemberComparator
1463 : * qsort comparison function for MultiXactMember
1464 : *
1465 : * We can't use wraparound comparison for XIDs because that does not respect
1466 : * the triangle inequality! Any old sort order will do.
1467 : */
1468 : static int
1469 GIC 1901 : mxactMemberComparator(const void *arg1, const void *arg2)
1470 ECB : {
1471 GIC 1901 : MultiXactMember member1 = *(const MultiXactMember *) arg1;
1472 CBC 1901 : MultiXactMember member2 = *(const MultiXactMember *) arg2;
1473 ECB :
1474 GIC 1901 : if (member1.xid > member2.xid)
1475 CBC 22 : return 1;
1476 1879 : if (member1.xid < member2.xid)
1477 1711 : return -1;
1478 168 : if (member1.status > member2.status)
1479 LBC 0 : return 1;
1480 GBC 168 : if (member1.status < member2.status)
1481 CBC 168 : return -1;
1482 LBC 0 : return 0;
1483 EUB : }
1484 :
1485 : /*
1486 : * mXactCacheGetBySet
1487 : * returns a MultiXactId from the cache based on the set of
1488 : * TransactionIds that compose it, or InvalidMultiXactId if
1489 : * none matches.
1490 : *
1491 : * This is helpful, for example, if two transactions want to lock a huge
1492 : * table. By using the cache, the second will use the same MultiXactId
1493 : * for the majority of tuples, thus keeping MultiXactId usage low (saving
1494 : * both I/O and wraparound issues).
1495 : *
1496 : * NB: the passed members array will be sorted in-place.
1497 : */
1498 : static MultiXactId
1499 GIC 1117 : mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1500 ECB : {
1501 : dlist_iter iter;
1502 :
1503 : debug_elog3(DEBUG2, "CacheGet: looking for %s",
1504 : mxid_to_string(InvalidMultiXactId, nmembers, members));
1505 :
1506 : /* sort the array so comparison is easy */
1507 GIC 1117 : qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1508 ECB :
1509 GNC 1356 : dclist_foreach(iter, &MXactCache)
1510 ECB : {
1511 GNC 1084 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1512 : iter.cur);
1513 ECB :
1514 GIC 1084 : if (entry->nmembers != nmembers)
1515 123 : continue;
1516 ECB :
1517 : /*
1518 : * We assume the cache entries are sorted, and that the unused bits in
1519 : * "status" are zeroed.
1520 : */
1521 GIC 961 : if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1522 : {
1523 ECB : debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1524 GNC 845 : dclist_move_head(&MXactCache, iter.cur);
1525 GIC 845 : return entry->multi;
1526 ECB : }
1527 : }
1528 :
1529 : debug_elog2(DEBUG2, "CacheGet: not found :-(");
1530 GIC 272 : return InvalidMultiXactId;
1531 : }
1532 ECB :
1533 : /*
1534 : * mXactCacheGetById
1535 : * returns the composing MultiXactMember set from the cache for a
1536 : * given MultiXactId, if present.
1537 : *
1538 : * If successful, *xids is set to the address of a palloc'd copy of the
1539 : * MultiXactMember set. Return value is number of members, or -1 on failure.
1540 : */
1541 : static int
1542 GIC 3001 : mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
1543 : {
1544 ECB : dlist_iter iter;
1545 :
1546 : debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1547 :
1548 GNC 3446 : dclist_foreach(iter, &MXactCache)
1549 : {
1550 2569 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1551 : iter.cur);
1552 :
1553 CBC 2569 : if (entry->multi == multi)
1554 : {
1555 : MultiXactMember *ptr;
1556 ECB : Size size;
1557 :
1558 GIC 2124 : size = sizeof(MultiXactMember) * entry->nmembers;
1559 2124 : ptr = (MultiXactMember *) palloc(size);
1560 :
1561 CBC 2124 : memcpy(ptr, entry->members, size);
1562 ECB :
1563 : debug_elog3(DEBUG2, "CacheGet: found %s",
1564 : mxid_to_string(multi,
1565 : entry->nmembers,
1566 : entry->members));
1567 :
1568 : /*
1569 : * Note we modify the list while not using a modifiable iterator.
1570 : * This is acceptable only because we exit the iteration
1571 : * immediately afterwards.
1572 : */
1573 GNC 2124 : dclist_move_head(&MXactCache, iter.cur);
1574 :
1575 GIC 2124 : *members = ptr;
1576 CBC 2124 : return entry->nmembers;
1577 : }
1578 ECB : }
1579 :
1580 : debug_elog2(DEBUG2, "CacheGet: not found");
1581 GIC 877 : return -1;
1582 : }
1583 :
1584 ECB : /*
1585 : * mXactCachePut
1586 : * Add a new MultiXactId and its composing set into the local cache.
1587 : */
1588 : static void
1589 GIC 505 : mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1590 : {
1591 : mXactCacheEnt *entry;
1592 ECB :
1593 : debug_elog3(DEBUG2, "CachePut: storing %s",
1594 : mxid_to_string(multi, nmembers, members));
1595 :
1596 GIC 505 : if (MXactContext == NULL)
1597 : {
1598 : /* The cache only lives as long as the current transaction */
1599 ECB : debug_elog2(DEBUG2, "CachePut: initializing memory context");
1600 GIC 358 : MXactContext = AllocSetContextCreate(TopTransactionContext,
1601 : "MultiXact cache context",
1602 : ALLOCSET_SMALL_SIZES);
1603 ECB : }
1604 :
1605 : entry = (mXactCacheEnt *)
1606 GIC 505 : MemoryContextAlloc(MXactContext,
1607 505 : offsetof(mXactCacheEnt, members) +
1608 : nmembers * sizeof(MultiXactMember));
1609 ECB :
1610 CBC 505 : entry->multi = multi;
1611 GIC 505 : entry->nmembers = nmembers;
1612 505 : memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1613 ECB :
1614 : /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1615 CBC 505 : qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1616 :
1617 GNC 505 : dclist_push_head(&MXactCache, &entry->node);
1618 505 : if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
1619 : {
1620 ECB : dlist_node *node;
1621 :
1622 UNC 0 : node = dclist_tail_node(&MXactCache);
1623 0 : dclist_delete_from(&MXactCache, node);
1624 EUB :
1625 UNC 0 : entry = dclist_container(mXactCacheEnt, node, node);
1626 EUB : debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1627 : entry->multi);
1628 :
1629 UIC 0 : pfree(entry);
1630 EUB : }
1631 GIC 505 : }
1632 ECB :
1633 : static char *
1634 UIC 0 : mxstatus_to_string(MultiXactStatus status)
1635 EUB : {
1636 UIC 0 : switch (status)
1637 EUB : {
1638 UIC 0 : case MultiXactStatusForKeyShare:
1639 UBC 0 : return "keysh";
1640 0 : case MultiXactStatusForShare:
1641 0 : return "sh";
1642 0 : case MultiXactStatusForNoKeyUpdate:
1643 0 : return "fornokeyupd";
1644 0 : case MultiXactStatusForUpdate:
1645 0 : return "forupd";
1646 0 : case MultiXactStatusNoKeyUpdate:
1647 0 : return "nokeyupd";
1648 0 : case MultiXactStatusUpdate:
1649 0 : return "upd";
1650 0 : default:
1651 0 : elog(ERROR, "unrecognized multixact status %d", status);
1652 EUB : return "";
1653 : }
1654 : }
1655 :
1656 : char *
1657 UIC 0 : mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1658 EUB : {
1659 : static char *str = NULL;
1660 : StringInfoData buf;
1661 : int i;
1662 :
1663 UIC 0 : if (str != NULL)
1664 UBC 0 : pfree(str);
1665 EUB :
1666 UIC 0 : initStringInfo(&buf);
1667 EUB :
1668 UIC 0 : appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1669 EUB : mxstatus_to_string(members[0].status));
1670 :
1671 UIC 0 : for (i = 1; i < nmembers; i++)
1672 UBC 0 : appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1673 0 : mxstatus_to_string(members[i].status));
1674 EUB :
1675 UIC 0 : appendStringInfoChar(&buf, ']');
1676 UBC 0 : str = MemoryContextStrdup(TopMemoryContext, buf.data);
1677 0 : pfree(buf.data);
1678 0 : return str;
1679 EUB : }
1680 :
1681 : /*
1682 : * AtEOXact_MultiXact
1683 : * Handle transaction end for MultiXact
1684 : *
1685 : * This is called at top transaction commit or abort (we don't care which).
1686 : */
1687 : void
1688 GIC 485037 : AtEOXact_MultiXact(void)
1689 ECB : {
1690 : /*
1691 : * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1692 : * which should only be valid while within a transaction.
1693 : *
1694 : * We assume that storing a MultiXactId is atomic and so we need not take
1695 : * MultiXactGenLock to do this.
1696 : */
1697 GIC 485037 : OldestMemberMXactId[MyBackendId] = InvalidMultiXactId;
1698 CBC 485037 : OldestVisibleMXactId[MyBackendId] = InvalidMultiXactId;
1699 ECB :
1700 : /*
1701 : * Discard the local MultiXactId cache. Since MXactContext was created as
1702 : * a child of TopTransactionContext, we needn't delete it explicitly.
1703 : */
1704 GIC 485037 : MXactContext = NULL;
1705 GNC 485037 : dclist_init(&MXactCache);
1706 CBC 485037 : }
1707 :
1708 : /*
1709 : * AtPrepare_MultiXact
1710 : * Save multixact state at 2PC transaction prepare
1711 : *
1712 : * In this phase, we only store our OldestMemberMXactId value in the two-phase
1713 : * state file.
1714 : */
1715 : void
1716 354 : AtPrepare_MultiXact(void)
1717 : {
1718 354 : MultiXactId myOldestMember = OldestMemberMXactId[MyBackendId];
1719 :
1720 354 : if (MultiXactIdIsValid(myOldestMember))
1721 47 : RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
1722 : &myOldestMember, sizeof(MultiXactId));
1723 354 : }
1724 :
1725 : /*
1726 : * PostPrepare_MultiXact
1727 : * Clean up after successful PREPARE TRANSACTION
1728 : */
1729 : void
1730 354 : PostPrepare_MultiXact(TransactionId xid)
1731 : {
1732 : MultiXactId myOldestMember;
1733 :
1734 : /*
1735 : * Transfer our OldestMemberMXactId value to the slot reserved for the
1736 : * prepared transaction.
1737 : */
1738 354 : myOldestMember = OldestMemberMXactId[MyBackendId];
1739 354 : if (MultiXactIdIsValid(myOldestMember))
1740 : {
1741 47 : BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, false);
1742 :
1743 : /*
1744 : * Even though storing MultiXactId is atomic, acquire lock to make
1745 : * sure others see both changes, not just the reset of the slot of the
1746 : * current backend. Using a volatile pointer might suffice, but this
1747 : * isn't a hot spot.
1748 : */
1749 47 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1750 :
1751 47 : OldestMemberMXactId[dummyBackendId] = myOldestMember;
1752 47 : OldestMemberMXactId[MyBackendId] = InvalidMultiXactId;
1753 :
1754 47 : LWLockRelease(MultiXactGenLock);
1755 : }
1756 :
1757 : /*
1758 : * We don't need to transfer OldestVisibleMXactId value, because the
1759 : * transaction is not going to be looking at any more multixacts once it's
1760 : * prepared.
1761 : *
1762 : * We assume that storing a MultiXactId is atomic and so we need not take
1763 : * MultiXactGenLock to do this.
1764 : */
1765 354 : OldestVisibleMXactId[MyBackendId] = InvalidMultiXactId;
1766 :
1767 : /*
1768 : * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1769 : */
1770 354 : MXactContext = NULL;
1771 GNC 354 : dclist_init(&MXactCache);
1772 GIC 354 : }
1773 :
1774 : /*
1775 : * multixact_twophase_recover
1776 : * Recover the state of a prepared transaction at startup
1777 : */
1778 ECB : void
1779 GIC 8 : multixact_twophase_recover(TransactionId xid, uint16 info,
1780 : void *recdata, uint32 len)
1781 ECB : {
1782 GIC 8 : BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, false);
1783 : MultiXactId oldestMember;
1784 :
1785 : /*
1786 : * Get the oldest member XID from the state file record, and set it in the
1787 : * OldestMemberMXactId slot reserved for this prepared transaction.
1788 ECB : */
1789 CBC 8 : Assert(len == sizeof(MultiXactId));
1790 GIC 8 : oldestMember = *((MultiXactId *) recdata);
1791 ECB :
1792 CBC 8 : OldestMemberMXactId[dummyBackendId] = oldestMember;
1793 GIC 8 : }
1794 :
1795 : /*
1796 : * multixact_twophase_postcommit
1797 : * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1798 : */
1799 ECB : void
1800 GIC 52 : multixact_twophase_postcommit(TransactionId xid, uint16 info,
1801 : void *recdata, uint32 len)
1802 ECB : {
1803 GIC 52 : BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, true);
1804 ECB :
1805 GIC 52 : Assert(len == sizeof(MultiXactId));
1806 ECB :
1807 CBC 52 : OldestMemberMXactId[dummyBackendId] = InvalidMultiXactId;
1808 GIC 52 : }
1809 :
1810 : /*
1811 : * multixact_twophase_postabort
1812 : * This is actually just the same as the COMMIT case.
1813 : */
1814 ECB : void
1815 GIC 22 : multixact_twophase_postabort(TransactionId xid, uint16 info,
1816 : void *recdata, uint32 len)
1817 ECB : {
1818 CBC 22 : multixact_twophase_postcommit(xid, info, recdata, len);
1819 GIC 22 : }
1820 :
1821 : /*
1822 : * Initialization of shared memory for MultiXact. We use two SLRU areas,
1823 : * thus double memory. Also, reserve space for the shared MultiXactState
1824 : * struct and the per-backend MultiXactId arrays (two of those, too).
1825 : */
1826 ECB : Size
1827 GIC 2738 : MultiXactShmemSize(void)
1828 : {
1829 : Size size;
1830 :
1831 : /* We need 2*MaxOldestSlot + 1 perBackendXactIds[] entries */
1832 : #define SHARED_MULTIXACT_STATE_SIZE \
1833 : add_size(offsetof(MultiXactStateData, perBackendXactIds) + sizeof(MultiXactId), \
1834 : mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1835 ECB :
1836 CBC 2738 : size = SHARED_MULTIXACT_STATE_SIZE;
1837 2738 : size = add_size(size, SimpleLruShmemSize(NUM_MULTIXACTOFFSET_BUFFERS, 0));
1838 GIC 2738 : size = add_size(size, SimpleLruShmemSize(NUM_MULTIXACTMEMBER_BUFFERS, 0));
1839 ECB :
1840 GIC 2738 : return size;
1841 : }
1842 :
1843 ECB : void
1844 GIC 1826 : MultiXactShmemInit(void)
1845 : {
1846 : bool found;
1847 :
1848 : debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1849 ECB :
1850 CBC 1826 : MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
1851 GIC 1826 : MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
1852 ECB :
1853 GIC 1826 : SimpleLruInit(MultiXactOffsetCtl,
1854 ECB : "MultiXactOffset", NUM_MULTIXACTOFFSET_BUFFERS, 0,
1855 GIC 1826 : MultiXactOffsetSLRULock, "pg_multixact/offsets",
1856 : LWTRANCHE_MULTIXACTOFFSET_BUFFER,
1857 ECB : SYNC_HANDLER_MULTIXACT_OFFSET);
1858 CBC 1826 : SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
1859 GIC 1826 : SimpleLruInit(MultiXactMemberCtl,
1860 ECB : "MultiXactMember", NUM_MULTIXACTMEMBER_BUFFERS, 0,
1861 GIC 1826 : MultiXactMemberSLRULock, "pg_multixact/members",
1862 : LWTRANCHE_MULTIXACTMEMBER_BUFFER,
1863 : SYNC_HANDLER_MULTIXACT_MEMBER);
1864 : /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1865 :
1866 ECB : /* Initialize our shared state struct */
1867 CBC 1826 : MultiXactState = ShmemInitStruct("Shared MultiXact State",
1868 GIC 1826 : SHARED_MULTIXACT_STATE_SIZE,
1869 ECB : &found);
1870 GIC 1826 : if (!IsUnderPostmaster)
1871 ECB : {
1872 GIC 1826 : Assert(!found);
1873 :
1874 ECB : /* Make sure we zero out the per-backend state */
1875 GIC 1826 : MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
1876 : }
1877 EUB : else
1878 UIC 0 : Assert(found);
1879 :
1880 : /*
1881 : * Set up array pointers. Note that perBackendXactIds[0] is wasted space
1882 : * since we only use indexes 1..MaxOldestSlot in each array.
1883 ECB : */
1884 CBC 1826 : OldestMemberMXactId = MultiXactState->perBackendXactIds;
1885 1826 : OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
1886 GIC 1826 : }
1887 :
1888 : /*
1889 : * This func must be called ONCE on system install. It creates the initial
1890 : * MultiXact segments. (The MultiXacts directories are assumed to have been
1891 : * created by initdb, and MultiXactShmemInit must have been called already.)
1892 : */
1893 ECB : void
1894 GIC 305 : BootStrapMultiXact(void)
1895 : {
1896 : int slotno;
1897 ECB :
1898 GIC 305 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
1899 :
1900 ECB : /* Create and zero the first page of the offsets log */
1901 GIC 305 : slotno = ZeroMultiXactOffsetPage(0, false);
1902 :
1903 ECB : /* Make sure it's written out */
1904 CBC 305 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
1905 GIC 305 : Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
1906 ECB :
1907 GIC 305 : LWLockRelease(MultiXactOffsetSLRULock);
1908 ECB :
1909 GIC 305 : LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
1910 :
1911 ECB : /* Create and zero the first page of the members log */
1912 GIC 305 : slotno = ZeroMultiXactMemberPage(0, false);
1913 :
1914 ECB : /* Make sure it's written out */
1915 CBC 305 : SimpleLruWritePage(MultiXactMemberCtl, slotno);
1916 GIC 305 : Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
1917 ECB :
1918 CBC 305 : LWLockRelease(MultiXactMemberSLRULock);
1919 GIC 305 : }
1920 :
1921 : /*
1922 : * Initialize (or reinitialize) a page of MultiXactOffset to zeroes.
1923 : * If writeXlog is true, also emit an XLOG record saying we did this.
1924 : *
1925 : * The page is not actually written, just set up in shared memory.
1926 : * The slot number of the new page is returned.
1927 : *
1928 : * Control lock must be held at entry, and will be held at exit.
1929 : */
1930 ECB : static int
1931 GIC 316 : ZeroMultiXactOffsetPage(int pageno, bool writeXlog)
1932 : {
1933 : int slotno;
1934 ECB :
1935 GIC 316 : slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
1936 ECB :
1937 CBC 316 : if (writeXlog)
1938 GIC 10 : WriteMZeroPageXlogRec(pageno, XLOG_MULTIXACT_ZERO_OFF_PAGE);
1939 ECB :
1940 GIC 316 : return slotno;
1941 : }
1942 :
1943 : /*
1944 : * Ditto, for MultiXactMember
1945 : */
1946 ECB : static int
1947 GIC 316 : ZeroMultiXactMemberPage(int pageno, bool writeXlog)
1948 : {
1949 : int slotno;
1950 ECB :
1951 GIC 316 : slotno = SimpleLruZeroPage(MultiXactMemberCtl, pageno);
1952 ECB :
1953 CBC 316 : if (writeXlog)
1954 GIC 10 : WriteMZeroPageXlogRec(pageno, XLOG_MULTIXACT_ZERO_MEM_PAGE);
1955 ECB :
1956 GIC 316 : return slotno;
1957 : }
1958 :
1959 : /*
1960 : * MaybeExtendOffsetSlru
1961 : * Extend the offsets SLRU area, if necessary
1962 : *
1963 : * After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
1964 : * contain files that are shorter than necessary; this would occur if the old
1965 : * installation had used multixacts beyond the first page (files cannot be
1966 : * copied, because the on-disk representation is different). pg_upgrade would
1967 : * update pg_control to set the next offset value to be at that position, so
1968 : * that tuples marked as locked by such MultiXacts would be seen as visible
1969 : * without having to consult multixact. However, trying to create and use a
1970 : * new MultiXactId would result in an error because the page on which the new
1971 : * value would reside does not exist. This routine is in charge of creating
1972 : * such pages.
1973 : */
1974 ECB : static void
1975 GIC 6 : MaybeExtendOffsetSlru(void)
1976 : {
1977 : int pageno;
1978 ECB :
1979 GIC 6 : pageno = MultiXactIdToOffsetPage(MultiXactState->nextMXact);
1980 ECB :
1981 GIC 6 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
1982 ECB :
1983 GIC 6 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
1984 : {
1985 : int slotno;
1986 :
1987 : /*
1988 : * Fortunately for us, SimpleLruWritePage is already prepared to deal
1989 : * with creating a new segment file even if the page we're writing is
1990 : * not the first in it, so this is enough.
1991 EUB : */
1992 UBC 0 : slotno = ZeroMultiXactOffsetPage(pageno, false);
1993 UIC 0 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
1994 : }
1995 ECB :
1996 CBC 6 : LWLockRelease(MultiXactOffsetSLRULock);
1997 GIC 6 : }
1998 :
1999 : /*
2000 : * This must be called ONCE during postmaster or standalone-backend startup.
2001 : *
2002 : * StartupXLOG has already established nextMXact/nextOffset by calling
2003 : * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
2004 : * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
2005 : * replayed WAL.
2006 : */
2007 ECB : void
2008 GIC 1176 : StartupMultiXact(void)
2009 ECB : {
2010 CBC 1176 : MultiXactId multi = MultiXactState->nextMXact;
2011 GIC 1176 : MultiXactOffset offset = MultiXactState->nextOffset;
2012 : int pageno;
2013 :
2014 : /*
2015 : * Initialize offset's idea of the latest page number.
2016 ECB : */
2017 CBC 1176 : pageno = MultiXactIdToOffsetPage(multi);
2018 GIC 1176 : MultiXactOffsetCtl->shared->latest_page_number = pageno;
2019 :
2020 : /*
2021 : * Initialize member's idea of the latest page number.
2022 ECB : */
2023 CBC 1176 : pageno = MXOffsetToMemberPage(offset);
2024 1176 : MultiXactMemberCtl->shared->latest_page_number = pageno;
2025 GIC 1176 : }
2026 :
2027 : /*
2028 : * This must be called ONCE at the end of startup/recovery.
2029 : */
2030 ECB : void
2031 GIC 1142 : TrimMultiXact(void)
2032 : {
2033 : MultiXactId nextMXact;
2034 : MultiXactOffset offset;
2035 : MultiXactId oldestMXact;
2036 : Oid oldestMXactDB;
2037 : int pageno;
2038 : int entryno;
2039 : int flagsoff;
2040 ECB :
2041 CBC 1142 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2042 1142 : nextMXact = MultiXactState->nextMXact;
2043 1142 : offset = MultiXactState->nextOffset;
2044 1142 : oldestMXact = MultiXactState->oldestMultiXactId;
2045 1142 : oldestMXactDB = MultiXactState->oldestMultiXactDB;
2046 GIC 1142 : LWLockRelease(MultiXactGenLock);
2047 :
2048 ECB : /* Clean up offsets state */
2049 GIC 1142 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
2050 :
2051 : /*
2052 : * (Re-)Initialize our idea of the latest page number for offsets.
2053 ECB : */
2054 CBC 1142 : pageno = MultiXactIdToOffsetPage(nextMXact);
2055 GIC 1142 : MultiXactOffsetCtl->shared->latest_page_number = pageno;
2056 :
2057 : /*
2058 : * Zero out the remainder of the current offsets page. See notes in
2059 : * TrimCLOG() for background. Unlike CLOG, some WAL record covers every
2060 : * pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
2061 : * rule "write xlog before data," nextMXact successors may carry obsolete,
2062 : * nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
2063 : * operates normally.
2064 ECB : */
2065 CBC 1142 : entryno = MultiXactIdToOffsetEntry(nextMXact);
2066 GIC 1142 : if (entryno != 0)
2067 : {
2068 : int slotno;
2069 : MultiXactOffset *offptr;
2070 ECB :
2071 CBC 1142 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
2072 1142 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2073 GIC 1142 : offptr += entryno;
2074 ECB :
2075 GIC 1142 : MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
2076 ECB :
2077 GIC 1142 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
2078 : }
2079 ECB :
2080 GIC 1142 : LWLockRelease(MultiXactOffsetSLRULock);
2081 :
2082 ECB : /* And the same for members */
2083 GIC 1142 : LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
2084 :
2085 : /*
2086 : * (Re-)Initialize our idea of the latest page number for members.
2087 ECB : */
2088 CBC 1142 : pageno = MXOffsetToMemberPage(offset);
2089 GIC 1142 : MultiXactMemberCtl->shared->latest_page_number = pageno;
2090 :
2091 : /*
2092 : * Zero out the remainder of the current members page. See notes in
2093 : * TrimCLOG() for motivation.
2094 ECB : */
2095 CBC 1142 : flagsoff = MXOffsetToFlagsOffset(offset);
2096 GIC 1142 : if (flagsoff != 0)
2097 : {
2098 : int slotno;
2099 : TransactionId *xidptr;
2100 : int memberoff;
2101 ECB :
2102 CBC 7 : memberoff = MXOffsetToMemberOffset(offset);
2103 7 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
2104 7 : xidptr = (TransactionId *)
2105 GIC 7 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
2106 ECB :
2107 GIC 7 : MemSet(xidptr, 0, BLCKSZ - memberoff);
2108 :
2109 : /*
2110 : * Note: we don't need to zero out the flag bits in the remaining
2111 : * members of the current group, because they are always reset before
2112 : * writing.
2113 : */
2114 ECB :
2115 GIC 7 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
2116 : }
2117 ECB :
2118 GIC 1142 : LWLockRelease(MultiXactMemberSLRULock);
2119 :
2120 ECB : /* signal that we're officially up */
2121 CBC 1142 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2122 1142 : MultiXactState->finishedStartup = true;
2123 GIC 1142 : LWLockRelease(MultiXactGenLock);
2124 :
2125 ECB : /* Now compute how far away the next members wraparound is. */
2126 CBC 1142 : SetMultiXactIdLimit(oldestMXact, oldestMXactDB, true);
2127 GIC 1142 : }
2128 :
2129 : /*
2130 : * Get the MultiXact data to save in a checkpoint record
2131 : */
2132 ECB : void
2133 GIC 2335 : MultiXactGetCheckptMulti(bool is_shutdown,
2134 : MultiXactId *nextMulti,
2135 : MultiXactOffset *nextMultiOffset,
2136 : MultiXactId *oldestMulti,
2137 : Oid *oldestMultiDB)
2138 ECB : {
2139 CBC 2335 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2140 2335 : *nextMulti = MultiXactState->nextMXact;
2141 2335 : *nextMultiOffset = MultiXactState->nextOffset;
2142 2335 : *oldestMulti = MultiXactState->oldestMultiXactId;
2143 2335 : *oldestMultiDB = MultiXactState->oldestMultiXactDB;
2144 GIC 2335 : LWLockRelease(MultiXactGenLock);
2145 :
2146 : debug_elog6(DEBUG2,
2147 : "MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
2148 ECB : *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
2149 GIC 2335 : }
2150 :
2151 : /*
2152 : * Perform a checkpoint --- either during shutdown, or on-the-fly
2153 : */
2154 ECB : void
2155 GIC 2363 : CheckPointMultiXact(void)
2156 : {
2157 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
2158 :
2159 : /*
2160 : * Write dirty MultiXact pages to disk. This may result in sync requests
2161 : * queued for later handling by ProcessSyncRequests(), as part of the
2162 : * checkpoint.
2163 ECB : */
2164 CBC 2363 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2165 GIC 2363 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2166 :
2167 ECB : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
2168 GIC 2363 : }
2169 :
2170 : /*
2171 : * Set the next-to-be-assigned MultiXactId and offset
2172 : *
2173 : * This is used when we can determine the correct next ID/offset exactly
2174 : * from a checkpoint record. Although this is only called during bootstrap
2175 : * and XLog replay, we take the lock in case any hot-standby backends are
2176 : * examining the values.
2177 : */
2178 ECB : void
2179 GIC 1506 : MultiXactSetNextMXact(MultiXactId nextMulti,
2180 : MultiXactOffset nextMultiOffset)
2181 : {
2182 : debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
2183 ECB : nextMulti, nextMultiOffset);
2184 CBC 1506 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2185 1506 : MultiXactState->nextMXact = nextMulti;
2186 1506 : MultiXactState->nextOffset = nextMultiOffset;
2187 GIC 1506 : LWLockRelease(MultiXactGenLock);
2188 :
2189 : /*
2190 : * During a binary upgrade, make sure that the offsets SLRU is large
2191 : * enough to contain the next value that would be created.
2192 : *
2193 : * We need to do this pretty early during the first startup in binary
2194 : * upgrade mode: before StartupMultiXact() in fact, because this routine
2195 : * is called even before that by StartupXLOG(). And we can't do it
2196 : * earlier than at this point, because during that first call of this
2197 : * routine we determine the MultiXactState->nextMXact value that
2198 : * MaybeExtendOffsetSlru needs.
2199 ECB : */
2200 CBC 1506 : if (IsBinaryUpgrade)
2201 6 : MaybeExtendOffsetSlru();
2202 GIC 1506 : }
2203 :
2204 : /*
2205 : * Determine the last safe MultiXactId to allocate given the currently oldest
2206 : * datminmxid (ie, the oldest MultiXactId that might exist in any database
2207 : * of our cluster), and the OID of the (or a) database with that value.
2208 : *
2209 : * is_startup is true when we are just starting the cluster, false when we
2210 : * are updating state in a running cluster. This only affects log messages.
2211 : */
2212 ECB : void
2213 GIC 2940 : SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
2214 : bool is_startup)
2215 : {
2216 : MultiXactId multiVacLimit;
2217 : MultiXactId multiWarnLimit;
2218 : MultiXactId multiStopLimit;
2219 : MultiXactId multiWrapLimit;
2220 : MultiXactId curMulti;
2221 : bool needs_offset_vacuum;
2222 ECB :
2223 GIC 2940 : Assert(MultiXactIdIsValid(oldest_datminmxid));
2224 :
2225 : /*
2226 : * We pretend that a wrap will happen halfway through the multixact ID
2227 : * space, but that's not really true, because multixacts wrap differently
2228 : * from transaction IDs. Note that, separately from any concern about
2229 : * multixact IDs wrapping, we must ensure that multixact members do not
2230 : * wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
2231 ECB : */
2232 CBC 2940 : multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2233 GBC 2940 : if (multiWrapLimit < FirstMultiXactId)
2234 UIC 0 : multiWrapLimit += FirstMultiXactId;
2235 :
2236 : /*
2237 : * We'll refuse to continue assigning MultiXactIds once we get within 3M
2238 : * multi of data loss. See SetTransactionIdLimit.
2239 ECB : */
2240 CBC 2940 : multiStopLimit = multiWrapLimit - 3000000;
2241 GBC 2940 : if (multiStopLimit < FirstMultiXactId)
2242 UIC 0 : multiStopLimit -= FirstMultiXactId;
2243 :
2244 : /*
2245 : * We'll start complaining loudly when we get within 40M multis of data
2246 : * loss. This is kind of arbitrary, but if you let your gas gauge get
2247 : * down to 2% of full, would you be looking for the next gas station? We
2248 : * need to be fairly liberal about this number because there are lots of
2249 : * scenarios where most transactions are done by automatic clients that
2250 : * won't pay attention to warnings. (No, we're not gonna make this
2251 : * configurable. If you know enough to configure it, you know enough to
2252 : * not get in this kind of trouble in the first place.)
2253 ECB : */
2254 CBC 2940 : multiWarnLimit = multiWrapLimit - 40000000;
2255 GBC 2940 : if (multiWarnLimit < FirstMultiXactId)
2256 UIC 0 : multiWarnLimit -= FirstMultiXactId;
2257 :
2258 : /*
2259 : * We'll start trying to force autovacuums when oldest_datminmxid gets to
2260 : * be more than autovacuum_multixact_freeze_max_age mxids old.
2261 : *
2262 : * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2263 : * so that we don't have to worry about dealing with on-the-fly changes in
2264 : * its value. See SetTransactionIdLimit.
2265 ECB : */
2266 CBC 2940 : multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2267 GBC 2940 : if (multiVacLimit < FirstMultiXactId)
2268 UIC 0 : multiVacLimit += FirstMultiXactId;
2269 :
2270 ECB : /* Grab lock for just long enough to set the new limit values */
2271 CBC 2940 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2272 2940 : MultiXactState->oldestMultiXactId = oldest_datminmxid;
2273 2940 : MultiXactState->oldestMultiXactDB = oldest_datoid;
2274 2940 : MultiXactState->multiVacLimit = multiVacLimit;
2275 2940 : MultiXactState->multiWarnLimit = multiWarnLimit;
2276 2940 : MultiXactState->multiStopLimit = multiStopLimit;
2277 2940 : MultiXactState->multiWrapLimit = multiWrapLimit;
2278 2940 : curMulti = MultiXactState->nextMXact;
2279 GIC 2940 : LWLockRelease(MultiXactGenLock);
2280 :
2281 ECB : /* Log the info */
2282 GIC 2940 : ereport(DEBUG1,
2283 : (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2284 : multiWrapLimit, oldest_datoid)));
2285 :
2286 : /*
2287 : * Computing the actual limits is only possible once the data directory is
2288 : * in a consistent state. There's no need to compute the limits while
2289 : * still replaying WAL - no decisions about new multis are made even
2290 : * though multixact creations might be replayed. So we'll only do further
2291 : * checks after TrimMultiXact() has been called.
2292 ECB : */
2293 CBC 2940 : if (!MultiXactState->finishedStartup)
2294 GIC 1481 : return;
2295 ECB :
2296 GIC 1459 : Assert(!InRecovery);
2297 :
2298 ECB : /* Set limits for offset vacuum. */
2299 GIC 1459 : needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
2300 :
2301 : /*
2302 : * If past the autovacuum force point, immediately signal an autovac
2303 : * request. The reason for this is that autovac only processes one
2304 : * database per invocation. Once it's finished cleaning up the oldest
2305 : * database, it'll call here, and we'll signal the postmaster to start
2306 : * another iteration immediately if there are still any old databases.
2307 ECB : */
2308 GBC 1459 : if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
2309 UBC 0 : needs_offset_vacuum) && IsUnderPostmaster)
2310 UIC 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
2311 :
2312 ECB : /* Give an immediate warning if past the wrap warn point */
2313 GIC 1459 : if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2314 : {
2315 : char *oldest_datname;
2316 :
2317 : /*
2318 : * We can be called when not inside a transaction, for example during
2319 : * StartupXLOG(). In such a case we cannot do database access, so we
2320 : * must just report the oldest DB's OID.
2321 : *
2322 : * Note: it's also possible that get_database_name fails and returns
2323 : * NULL, for example because the database just got dropped. We'll
2324 : * still warn, even though the warning might now be unnecessary.
2325 EUB : */
2326 UBC 0 : if (IsTransactionState())
2327 UIC 0 : oldest_datname = get_database_name(oldest_datoid);
2328 EUB : else
2329 UIC 0 : oldest_datname = NULL;
2330 EUB :
2331 UBC 0 : if (oldest_datname)
2332 UIC 0 : ereport(WARNING,
2333 : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2334 : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2335 : multiWrapLimit - curMulti,
2336 : oldest_datname,
2337 : multiWrapLimit - curMulti),
2338 : errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
2339 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2340 EUB : else
2341 UIC 0 : ereport(WARNING,
2342 : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2343 : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2344 : multiWrapLimit - curMulti,
2345 : oldest_datoid,
2346 : multiWrapLimit - curMulti),
2347 : errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
2348 : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2349 : }
2350 : }
2351 :
2352 : /*
2353 : * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2354 : * and similarly nextOffset is at least minMultiOffset.
2355 : *
2356 : * This is used when we can determine minimum safe values from an XLog
2357 : * record (either an on-line checkpoint or an mxact creation log entry).
2358 : * Although this is only called during XLog replay, we take the lock in case
2359 : * any hot-standby backends are examining the values.
2360 : */
2361 ECB : void
2362 GIC 143 : MultiXactAdvanceNextMXact(MultiXactId minMulti,
2363 : MultiXactOffset minMultiOffset)
2364 ECB : {
2365 CBC 143 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2366 GIC 143 : if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
2367 : {
2368 ECB : debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2369 GIC 2 : MultiXactState->nextMXact = minMulti;
2370 ECB : }
2371 GIC 143 : if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
2372 : {
2373 : debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
2374 ECB : minMultiOffset);
2375 GIC 2 : MultiXactState->nextOffset = minMultiOffset;
2376 ECB : }
2377 CBC 143 : LWLockRelease(MultiXactGenLock);
2378 GIC 143 : }
2379 :
2380 : /*
2381 : * Update our oldestMultiXactId value, but only if it's more recent than what
2382 : * we had.
2383 : *
2384 : * This may only be called during WAL replay.
2385 : */
2386 ECB : void
2387 GIC 166 : MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2388 ECB : {
2389 GIC 166 : Assert(InRecovery);
2390 ECB :
2391 GBC 166 : if (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
2392 LBC 0 : SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
2393 GIC 166 : }
2394 :
2395 : /*
2396 : * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2397 : *
2398 : * NB: this is called while holding MultiXactGenLock. We want it to be very
2399 : * fast most of the time; even when it's not so fast, no actual I/O need
2400 : * happen unless we're forced to write out a dirty log or xlog page to make
2401 : * room in shared memory.
2402 : */
2403 ECB : static void
2404 GIC 272 : ExtendMultiXactOffset(MultiXactId multi)
2405 : {
2406 : int pageno;
2407 :
2408 : /*
2409 : * No work except at first MultiXactId of a page. But beware: just after
2410 : * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2411 ECB : */
2412 GIC 272 : if (MultiXactIdToOffsetEntry(multi) != 0 &&
2413 ECB : multi != FirstMultiXactId)
2414 GIC 262 : return;
2415 ECB :
2416 GIC 10 : pageno = MultiXactIdToOffsetPage(multi);
2417 ECB :
2418 GIC 10 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
2419 :
2420 ECB : /* Zero the page and make an XLOG entry about it */
2421 GIC 10 : ZeroMultiXactOffsetPage(pageno, true);
2422 ECB :
2423 GIC 10 : LWLockRelease(MultiXactOffsetSLRULock);
2424 : }
2425 :
2426 : /*
2427 : * Make sure that MultiXactMember has room for the members of a newly-
2428 : * allocated MultiXactId.
2429 : *
2430 : * Like the above routine, this is called while holding MultiXactGenLock;
2431 : * same comments apply.
2432 : */
2433 ECB : static void
2434 GIC 272 : ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
2435 : {
2436 : /*
2437 : * It's possible that the members span more than one page of the members
2438 : * file, so we loop to ensure we consider each page. The coding is not
2439 : * optimal if the members span several pages, but that seems unusual
2440 : * enough to not worry much about.
2441 ECB : */
2442 GIC 544 : while (nmembers > 0)
2443 : {
2444 : int flagsoff;
2445 : int flagsbit;
2446 : uint32 difference;
2447 :
2448 : /*
2449 : * Only zero when at first entry of a page.
2450 ECB : */
2451 CBC 272 : flagsoff = MXOffsetToFlagsOffset(offset);
2452 272 : flagsbit = MXOffsetToFlagsBitShift(offset);
2453 GIC 272 : if (flagsoff == 0 && flagsbit == 0)
2454 : {
2455 : int pageno;
2456 ECB :
2457 GIC 10 : pageno = MXOffsetToMemberPage(offset);
2458 ECB :
2459 GIC 10 : LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
2460 :
2461 ECB : /* Zero the page and make an XLOG entry about it */
2462 GIC 10 : ZeroMultiXactMemberPage(pageno, true);
2463 ECB :
2464 GIC 10 : LWLockRelease(MultiXactMemberSLRULock);
2465 : }
2466 :
2467 : /*
2468 : * Compute the number of items till end of current page. Careful: if
2469 : * addition of unsigned ints wraps around, we're at the last page of
2470 : * the last segment; since that page holds a different number of items
2471 : * than other pages, we need to do it differently.
2472 ECB : */
2473 GIC 272 : if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
2474 : {
2475 : /*
2476 : * This is the last page of the last segment; we can compute the
2477 : * number of items left to allocate in it without modulo
2478 : * arithmetic.
2479 EUB : */
2480 UIC 0 : difference = MaxMultiXactOffset - offset + 1;
2481 : }
2482 ECB : else
2483 GIC 272 : difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
2484 :
2485 : /*
2486 : * Advance to next page, taking care to properly handle the wraparound
2487 : * case. OK if nmembers goes negative.
2488 ECB : */
2489 CBC 272 : nmembers -= difference;
2490 GIC 272 : offset += difference;
2491 ECB : }
2492 GIC 272 : }
2493 :
2494 : /*
2495 : * GetOldestMultiXactId
2496 : *
2497 : * Return the oldest MultiXactId that's still possibly still seen as live by
2498 : * any running transaction. Older ones might still exist on disk, but they no
2499 : * longer have any running member transaction.
2500 : *
2501 : * It's not safe to truncate MultiXact SLRU segments on the value returned by
2502 : * this function; however, it can be set as the new relminmxid for any table
2503 : * that VACUUM knows has no remaining MXIDs < the same value. It is only safe
2504 : * to truncate SLRUs when no table can possibly still have a referencing MXID.
2505 : */
2506 : MultiXactId
2507 CBC 93785 : GetOldestMultiXactId(void)
2508 : {
2509 : MultiXactId oldestMXact;
2510 : MultiXactId nextMXact;
2511 : int i;
2512 :
2513 : /*
2514 : * This is the oldest valid value among all the OldestMemberMXactId[] and
2515 : * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2516 : */
2517 93785 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2518 :
2519 : /*
2520 : * We have to beware of the possibility that nextMXact is in the
2521 : * wrapped-around state. We don't fix the counter itself here, but we
2522 : * must be sure to use a valid value in our calculation.
2523 : */
2524 93785 : nextMXact = MultiXactState->nextMXact;
2525 93785 : if (nextMXact < FirstMultiXactId)
2526 UBC 0 : nextMXact = FirstMultiXactId;
2527 :
2528 CBC 93785 : oldestMXact = nextMXact;
2529 11033905 : for (i = 1; i <= MaxOldestSlot; i++)
2530 : {
2531 : MultiXactId thisoldest;
2532 :
2533 10940120 : thisoldest = OldestMemberMXactId[i];
2534 10956348 : if (MultiXactIdIsValid(thisoldest) &&
2535 16228 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2536 8 : oldestMXact = thisoldest;
2537 10940120 : thisoldest = OldestVisibleMXactId[i];
2538 10940127 : if (MultiXactIdIsValid(thisoldest) &&
2539 7 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2540 2 : oldestMXact = thisoldest;
2541 : }
2542 :
2543 93785 : LWLockRelease(MultiXactGenLock);
2544 :
2545 93785 : return oldestMXact;
2546 : }
2547 :
2548 : /*
2549 : * Determine how aggressively we need to vacuum in order to prevent member
2550 : * wraparound.
2551 : *
2552 : * To do so determine what's the oldest member offset and install the limit
2553 : * info in MultiXactState, where it can be used to prevent overrun of old data
2554 : * in the members SLRU area.
2555 : *
2556 : * The return value is true if emergency autovacuum is required and false
2557 : * otherwise.
2558 : */
2559 : static bool
2560 1459 : SetOffsetVacuumLimit(bool is_startup)
2561 : {
2562 : MultiXactId oldestMultiXactId;
2563 : MultiXactId nextMXact;
2564 1459 : MultiXactOffset oldestOffset = 0; /* placate compiler */
2565 : MultiXactOffset prevOldestOffset;
2566 : MultiXactOffset nextOffset;
2567 1459 : bool oldestOffsetKnown = false;
2568 : bool prevOldestOffsetKnown;
2569 1459 : MultiXactOffset offsetStopLimit = 0;
2570 : MultiXactOffset prevOffsetStopLimit;
2571 :
2572 : /*
2573 : * NB: Have to prevent concurrent truncation, we might otherwise try to
2574 : * lookup an oldestMulti that's concurrently getting truncated away.
2575 : */
2576 1459 : LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2577 :
2578 : /* Read relevant fields from shared memory. */
2579 1459 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2580 1459 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2581 1459 : nextMXact = MultiXactState->nextMXact;
2582 1459 : nextOffset = MultiXactState->nextOffset;
2583 1459 : prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2584 1459 : prevOldestOffset = MultiXactState->oldestOffset;
2585 1459 : prevOffsetStopLimit = MultiXactState->offsetStopLimit;
2586 1459 : Assert(MultiXactState->finishedStartup);
2587 1459 : LWLockRelease(MultiXactGenLock);
2588 :
2589 : /*
2590 : * Determine the offset of the oldest multixact. Normally, we can read
2591 : * the offset from the multixact itself, but there's an important special
2592 : * case: if there are no multixacts in existence at all, oldestMXact
2593 : * obviously can't point to one. It will instead point to the multixact
2594 : * ID that will be assigned the next time one is needed.
2595 : */
2596 1459 : if (oldestMultiXactId == nextMXact)
2597 : {
2598 : /*
2599 : * When the next multixact gets created, it will be stored at the next
2600 : * offset.
2601 : */
2602 1452 : oldestOffset = nextOffset;
2603 1452 : oldestOffsetKnown = true;
2604 : }
2605 : else
2606 : {
2607 : /*
2608 : * Figure out where the oldest existing multixact's offsets are
2609 : * stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
2610 : * the supposedly-earliest multixact might not really exist. We are
2611 : * careful not to fail in that case.
2612 : */
2613 : oldestOffsetKnown =
2614 7 : find_multixact_start(oldestMultiXactId, &oldestOffset);
2615 :
2616 7 : if (oldestOffsetKnown)
2617 7 : ereport(DEBUG1,
2618 : (errmsg_internal("oldest MultiXactId member is at offset %u",
2619 : oldestOffset)));
2620 : else
2621 UBC 0 : ereport(LOG,
2622 : (errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
2623 : oldestMultiXactId)));
2624 : }
2625 :
2626 CBC 1459 : LWLockRelease(MultiXactTruncationLock);
2627 :
2628 : /*
2629 : * If we can, compute limits (and install them MultiXactState) to prevent
2630 : * overrun of old data in the members SLRU area. We can only do so if the
2631 : * oldest offset is known though.
2632 : */
2633 1459 : if (oldestOffsetKnown)
2634 : {
2635 : /* move back to start of the corresponding segment */
2636 1459 : offsetStopLimit = oldestOffset - (oldestOffset %
2637 : (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT));
2638 :
2639 : /* always leave one segment before the wraparound point */
2640 1459 : offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
2641 :
2642 1459 : if (!prevOldestOffsetKnown && !is_startup)
2643 UBC 0 : ereport(LOG,
2644 : (errmsg("MultiXact member wraparound protections are now enabled")));
2645 :
2646 CBC 1459 : ereport(DEBUG1,
2647 : (errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
2648 : offsetStopLimit, oldestMultiXactId)));
2649 : }
2650 UBC 0 : else if (prevOldestOffsetKnown)
2651 : {
2652 : /*
2653 : * If we failed to get the oldest offset this time, but we have a
2654 : * value from a previous pass through this function, use the old
2655 : * values rather than automatically forcing an emergency autovacuum
2656 : * cycle again.
2657 : */
2658 0 : oldestOffset = prevOldestOffset;
2659 0 : oldestOffsetKnown = true;
2660 0 : offsetStopLimit = prevOffsetStopLimit;
2661 : }
2662 :
2663 : /* Install the computed values */
2664 CBC 1459 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2665 1459 : MultiXactState->oldestOffset = oldestOffset;
2666 1459 : MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
2667 1459 : MultiXactState->offsetStopLimit = offsetStopLimit;
2668 1459 : LWLockRelease(MultiXactGenLock);
2669 :
2670 : /*
2671 : * Do we need an emergency autovacuum? If we're not sure, assume yes.
2672 : */
2673 2918 : return !oldestOffsetKnown ||
2674 1459 : (nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
2675 : }
2676 :
2677 : /*
2678 : * Return whether adding "distance" to "start" would move past "boundary".
2679 : *
2680 : * We use this to determine whether the addition is "wrapping around" the
2681 : * boundary point, hence the name. The reason we don't want to use the regular
2682 : * 2^31-modulo arithmetic here is that we want to be able to use the whole of
2683 : * the 2^32-1 space here, allowing for more multixacts than would fit
2684 : * otherwise.
2685 : */
2686 : static bool
2687 544 : MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start,
2688 : uint32 distance)
2689 : {
2690 : MultiXactOffset finish;
2691 :
2692 : /*
2693 : * Note that offset number 0 is not used (see GetMultiXactIdMembers), so
2694 : * if the addition wraps around the UINT_MAX boundary, skip that value.
2695 : */
2696 544 : finish = start + distance;
2697 544 : if (finish < start)
2698 UBC 0 : finish++;
2699 :
2700 : /*-----------------------------------------------------------------------
2701 : * When the boundary is numerically greater than the starting point, any
2702 : * value numerically between the two is not wrapped:
2703 : *
2704 : * <----S----B---->
2705 : * [---) = F wrapped past B (and UINT_MAX)
2706 : * [---) = F not wrapped
2707 : * [----] = F wrapped past B
2708 : *
2709 : * When the boundary is numerically less than the starting point (i.e. the
2710 : * UINT_MAX wraparound occurs somewhere in between) then all values in
2711 : * between are wrapped:
2712 : *
2713 : * <----B----S---->
2714 : * [---) = F not wrapped past B (but wrapped past UINT_MAX)
2715 : * [---) = F wrapped past B (and UINT_MAX)
2716 : * [----] = F not wrapped
2717 : *-----------------------------------------------------------------------
2718 : */
2719 CBC 544 : if (start < boundary)
2720 544 : return finish >= boundary || finish < start;
2721 : else
2722 UBC 0 : return finish >= boundary && finish < start;
2723 : }
2724 :
2725 : /*
2726 : * Find the starting offset of the given MultiXactId.
2727 : *
2728 : * Returns false if the file containing the multi does not exist on disk.
2729 : * Otherwise, returns true and sets *result to the starting member offset.
2730 : *
2731 : * This function does not prevent concurrent truncation, so if that's
2732 : * required, the caller has to protect against that.
2733 : */
2734 : static bool
2735 CBC 7 : find_multixact_start(MultiXactId multi, MultiXactOffset *result)
2736 : {
2737 : MultiXactOffset offset;
2738 : int pageno;
2739 : int entryno;
2740 : int slotno;
2741 : MultiXactOffset *offptr;
2742 :
2743 7 : Assert(MultiXactState->finishedStartup);
2744 :
2745 7 : pageno = MultiXactIdToOffsetPage(multi);
2746 7 : entryno = MultiXactIdToOffsetEntry(multi);
2747 :
2748 : /*
2749 : * Write out dirty data, so PhysicalPageExists can work correctly.
2750 : */
2751 7 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2752 7 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2753 :
2754 7 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2755 UBC 0 : return false;
2756 :
2757 : /* lock is acquired by SimpleLruReadPage_ReadOnly */
2758 CBC 7 : slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2759 7 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2760 7 : offptr += entryno;
2761 7 : offset = *offptr;
2762 7 : LWLockRelease(MultiXactOffsetSLRULock);
2763 :
2764 7 : *result = offset;
2765 7 : return true;
2766 : }
2767 :
2768 : /*
2769 : * Determine how many multixacts, and how many multixact members, currently
2770 : * exist. Return false if unable to determine.
2771 : */
2772 : static bool
2773 37057 : ReadMultiXactCounts(uint32 *multixacts, MultiXactOffset *members)
2774 : {
2775 : MultiXactOffset nextOffset;
2776 : MultiXactOffset oldestOffset;
2777 : MultiXactId oldestMultiXactId;
2778 : MultiXactId nextMultiXactId;
2779 : bool oldestOffsetKnown;
2780 :
2781 37057 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2782 37057 : nextOffset = MultiXactState->nextOffset;
2783 37057 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2784 37057 : nextMultiXactId = MultiXactState->nextMXact;
2785 37057 : oldestOffset = MultiXactState->oldestOffset;
2786 37057 : oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
2787 37057 : LWLockRelease(MultiXactGenLock);
2788 :
2789 37057 : if (!oldestOffsetKnown)
2790 UBC 0 : return false;
2791 :
2792 CBC 37057 : *members = nextOffset - oldestOffset;
2793 37057 : *multixacts = nextMultiXactId - oldestMultiXactId;
2794 37057 : return true;
2795 : }
2796 :
2797 : /*
2798 : * Multixact members can be removed once the multixacts that refer to them
2799 : * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2800 : * vacuum_multixact_freeze_table_age work together to make sure we never have
2801 : * too many multixacts; we hope that, at least under normal circumstances,
2802 : * this will also be sufficient to keep us from using too many offsets.
2803 : * However, if the average multixact has many members, we might exhaust the
2804 : * members space while still using few enough members that these limits fail
2805 : * to trigger relminmxid advancement by VACUUM. At that point, we'd have no
2806 : * choice but to start failing multixact-creating operations with an error.
2807 : *
2808 : * To prevent that, if more than a threshold portion of the members space is
2809 : * used, we effectively reduce autovacuum_multixact_freeze_max_age and
2810 : * to a value just less than the number of multixacts in use. We hope that
2811 : * this will quickly trigger autovacuuming on the table or tables with the
2812 : * oldest relminmxid, thus allowing datminmxid values to advance and removing
2813 : * some members.
2814 : *
2815 : * As the fraction of the member space currently in use grows, we become
2816 : * more aggressive in clamping this value. That not only causes autovacuum
2817 : * to ramp up, but also makes any manual vacuums the user issues more
2818 : * aggressive. This happens because vacuum_get_cutoffs() will clamp the
2819 : * freeze table and the minimum freeze age cutoffs based on the effective
2820 : * autovacuum_multixact_freeze_max_age this function returns. In the worst
2821 : * case, we'll claim the freeze_max_age to zero, and every vacuum of any
2822 : * table will freeze every multixact.
2823 : */
2824 : int
2825 GIC 37057 : MultiXactMemberFreezeThreshold(void)
2826 : {
2827 : MultiXactOffset members;
2828 : uint32 multixacts;
2829 ECB : uint32 victim_multixacts;
2830 EUB : double fraction;
2831 :
2832 : /* If we can't determine member space utilization, assume the worst. */
2833 CBC 37057 : if (!ReadMultiXactCounts(&multixacts, &members))
2834 LBC 0 : return 0;
2835 :
2836 : /* If member space utilization is low, no special action is required. */
2837 GIC 37057 : if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
2838 37057 : return autovacuum_multixact_freeze_max_age;
2839 :
2840 : /*
2841 EUB : * Compute a target for relminmxid advancement. The number of multixacts
2842 : * we try to eliminate from the system is based on how far we are past
2843 : * MULTIXACT_MEMBER_SAFE_THRESHOLD.
2844 : */
2845 UIC 0 : fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
2846 EUB : (MULTIXACT_MEMBER_DANGER_THRESHOLD - MULTIXACT_MEMBER_SAFE_THRESHOLD);
2847 UBC 0 : victim_multixacts = multixacts * fraction;
2848 EUB :
2849 : /* fraction could be > 1.0, but lowest possible freeze age is zero */
2850 UIC 0 : if (victim_multixacts > multixacts)
2851 0 : return 0;
2852 0 : return multixacts - victim_multixacts;
2853 : }
2854 :
2855 : typedef struct mxtruncinfo
2856 : {
2857 : int earliestExistingPage;
2858 : } mxtruncinfo;
2859 :
2860 : /*
2861 EUB : * SlruScanDirectory callback
2862 : * This callback determines the earliest existing page number.
2863 : */
2864 : static bool
2865 UBC 0 : SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int segpage, void *data)
2866 EUB : {
2867 UIC 0 : mxtruncinfo *trunc = (mxtruncinfo *) data;
2868 EUB :
2869 UIC 0 : if (trunc->earliestExistingPage == -1 ||
2870 0 : ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
2871 EUB : {
2872 UIC 0 : trunc->earliestExistingPage = segpage;
2873 : }
2874 :
2875 0 : return false; /* keep going */
2876 : }
2877 :
2878 :
2879 : /*
2880 : * Delete members segments [oldest, newOldest)
2881 : *
2882 : * The members SLRU can, in contrast to the offsets one, be filled to almost
2883 : * the full range at once. This means SimpleLruTruncate() can't trivially be
2884 EUB : * used - instead the to-be-deleted range is computed using the offsets
2885 : * SLRU. C.f. TruncateMultiXact().
2886 : */
2887 : static void
2888 UBC 0 : PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
2889 EUB : {
2890 UIC 0 : const int maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
2891 0 : int startsegment = MXOffsetToMemberSegment(oldestOffset);
2892 0 : int endsegment = MXOffsetToMemberSegment(newOldestOffset);
2893 0 : int segment = startsegment;
2894 :
2895 EUB : /*
2896 : * Delete all the segments but the last one. The last segment can still
2897 : * contain, possibly partially, valid data.
2898 : */
2899 UIC 0 : while (segment != endsegment)
2900 : {
2901 UBC 0 : elog(DEBUG2, "truncating multixact members segment %x", segment);
2902 0 : SlruDeleteSegment(MultiXactMemberCtl, segment);
2903 :
2904 EUB : /* move to next segment, handling wraparound correctly */
2905 UIC 0 : if (segment == maxsegment)
2906 UBC 0 : segment = 0;
2907 : else
2908 UIC 0 : segment += 1;
2909 : }
2910 0 : }
2911 :
2912 EUB : /*
2913 : * Delete offsets segments [oldest, newOldest)
2914 : */
2915 : static void
2916 UIC 0 : PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
2917 : {
2918 : /*
2919 : * We step back one multixact to avoid passing a cutoff page that hasn't
2920 : * been created yet in the rare case that oldestMulti would be the first
2921 EUB : * item on a page and oldestMulti == nextMulti. In that case, if we
2922 : * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
2923 : * detection.
2924 : */
2925 UIC 0 : SimpleLruTruncate(MultiXactOffsetCtl,
2926 0 : MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
2927 0 : }
2928 :
2929 : /*
2930 : * Remove all MultiXactOffset and MultiXactMember segments before the oldest
2931 : * ones still of interest.
2932 : *
2933 : * This is only called on a primary as part of vacuum (via
2934 : * vac_truncate_clog()). During recovery truncation is done by replaying
2935 : * truncation WAL records logged here.
2936 : *
2937 ECB : * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
2938 : * is one of the databases preventing newOldestMulti from increasing.
2939 : */
2940 : void
2941 GIC 317 : TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
2942 : {
2943 : MultiXactId oldestMulti;
2944 : MultiXactId nextMulti;
2945 : MultiXactOffset newOldestOffset;
2946 : MultiXactOffset oldestOffset;
2947 ECB : MultiXactOffset nextOffset;
2948 : mxtruncinfo trunc;
2949 : MultiXactId earliest;
2950 :
2951 GIC 317 : Assert(!RecoveryInProgress());
2952 317 : Assert(MultiXactState->finishedStartup);
2953 :
2954 : /*
2955 : * We can only allow one truncation to happen at once. Otherwise parts of
2956 ECB : * members might vanish while we're doing lookups or similar. There's no
2957 : * need to have an interlock with creating new multis or such, since those
2958 : * are constrained by the limits (which only grow, never shrink).
2959 : */
2960 CBC 317 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
2961 ECB :
2962 CBC 317 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2963 317 : nextMulti = MultiXactState->nextMXact;
2964 GIC 317 : nextOffset = MultiXactState->nextOffset;
2965 317 : oldestMulti = MultiXactState->oldestMultiXactId;
2966 317 : LWLockRelease(MultiXactGenLock);
2967 317 : Assert(MultiXactIdIsValid(oldestMulti));
2968 :
2969 : /*
2970 ECB : * Make sure to only attempt truncation if there's values to truncate
2971 : * away. In normal processing values shouldn't go backwards, but there's
2972 : * some corner cases (due to bugs) where that's possible.
2973 : */
2974 GIC 317 : if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
2975 : {
2976 317 : LWLockRelease(MultiXactTruncationLock);
2977 317 : return;
2978 : }
2979 :
2980 : /*
2981 : * Note we can't just plow ahead with the truncation; it's possible that
2982 : * there are no segments to truncate, which is a problem because we are
2983 : * going to attempt to read the offsets page to determine where to
2984 : * truncate the members SLRU. So we first scan the directory to determine
2985 : * the earliest offsets page number that we can read without error.
2986 : *
2987 : * When nextMXact is less than one segment away from multiWrapLimit,
2988 : * SlruScanDirCbFindEarliest can find some early segment other than the
2989 : * actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
2990 : * returns false, because not all pairs of entries have the same answer.)
2991 : * That can also arise when an earlier truncation attempt failed unlink()
2992 : * or returned early from this function. The only consequence is
2993 : * returning early, which wastes space that we could have liberated.
2994 EUB : *
2995 : * NB: It's also possible that the page that oldestMulti is on has already
2996 : * been truncated away, and we crashed before updating oldestMulti.
2997 : */
2998 UBC 0 : trunc.earliestExistingPage = -1;
2999 UIC 0 : SlruScanDirectory(MultiXactOffsetCtl, SlruScanDirCbFindEarliest, &trunc);
3000 0 : earliest = trunc.earliestExistingPage * MULTIXACT_OFFSETS_PER_PAGE;
3001 UBC 0 : if (earliest < FirstMultiXactId)
3002 UIC 0 : earliest = FirstMultiXactId;
3003 EUB :
3004 : /* If there's nothing to remove, we can bail out early. */
3005 UIC 0 : if (MultiXactIdPrecedes(oldestMulti, earliest))
3006 : {
3007 0 : LWLockRelease(MultiXactTruncationLock);
3008 0 : return;
3009 : }
3010 :
3011 : /*
3012 : * First, compute the safe truncation point for MultiXactMember. This is
3013 : * the starting offset of the oldest multixact.
3014 : *
3015 EUB : * Hopefully, find_multixact_start will always work here, because we've
3016 : * already checked that it doesn't precede the earliest MultiXact on disk.
3017 : * But if it fails, don't truncate anything, and log a message.
3018 : */
3019 UIC 0 : if (oldestMulti == nextMulti)
3020 EUB : {
3021 : /* there are NO MultiXacts */
3022 UBC 0 : oldestOffset = nextOffset;
3023 : }
3024 UIC 0 : else if (!find_multixact_start(oldestMulti, &oldestOffset))
3025 EUB : {
3026 UBC 0 : ereport(LOG,
3027 : (errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
3028 : oldestMulti, earliest)));
3029 UIC 0 : LWLockRelease(MultiXactTruncationLock);
3030 0 : return;
3031 : }
3032 :
3033 EUB : /*
3034 : * Secondly compute up to where to truncate. Lookup the corresponding
3035 : * member offset for newOldestMulti for that.
3036 : */
3037 UIC 0 : if (newOldestMulti == nextMulti)
3038 EUB : {
3039 : /* there are NO MultiXacts */
3040 UBC 0 : newOldestOffset = nextOffset;
3041 : }
3042 UIC 0 : else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
3043 EUB : {
3044 UBC 0 : ereport(LOG,
3045 : (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
3046 : newOldestMulti)));
3047 0 : LWLockRelease(MultiXactTruncationLock);
3048 UIC 0 : return;
3049 : }
3050 :
3051 0 : elog(DEBUG1, "performing multixact truncation: "
3052 : "offsets [%u, %u), offsets segments [%x, %x), "
3053 : "members [%u, %u), members segments [%x, %x)",
3054 : oldestMulti, newOldestMulti,
3055 : MultiXactIdToOffsetSegment(oldestMulti),
3056 : MultiXactIdToOffsetSegment(newOldestMulti),
3057 : oldestOffset, newOldestOffset,
3058 : MXOffsetToMemberSegment(oldestOffset),
3059 : MXOffsetToMemberSegment(newOldestOffset));
3060 :
3061 : /*
3062 : * Do truncation, and the WAL logging of the truncation, in a critical
3063 EUB : * section. That way offsets/members cannot get out of sync anymore, i.e.
3064 : * once consistent the newOldestMulti will always exist in members, even
3065 : * if we crashed in the wrong moment.
3066 : */
3067 UIC 0 : START_CRIT_SECTION();
3068 :
3069 : /*
3070 : * Prevent checkpoints from being scheduled concurrently. This is critical
3071 EUB : * because otherwise a truncation record might not be replayed after a
3072 : * crash/basebackup, even though the state of the data directory would
3073 : * require it.
3074 : */
3075 UBC 0 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
3076 UIC 0 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
3077 :
3078 : /* WAL log truncation */
3079 0 : WriteMTruncateXlogRec(newOldestMultiDB,
3080 : oldestMulti, newOldestMulti,
3081 : oldestOffset, newOldestOffset);
3082 :
3083 : /*
3084 : * Update in-memory limits before performing the truncation, while inside
3085 : * the critical section: Have to do it before truncation, to prevent
3086 : * concurrent lookups of those values. Has to be inside the critical
3087 EUB : * section as otherwise a future call to this function would error out,
3088 : * while looking up the oldest member in offsets, if our caller crashes
3089 : * before updating the limits.
3090 : */
3091 UIC 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
3092 0 : MultiXactState->oldestMultiXactId = newOldestMulti;
3093 UBC 0 : MultiXactState->oldestMultiXactDB = newOldestMultiDB;
3094 UIC 0 : LWLockRelease(MultiXactGenLock);
3095 :
3096 EUB : /* First truncate members */
3097 UIC 0 : PerformMembersTruncation(oldestOffset, newOldestOffset);
3098 EUB :
3099 : /* Then offsets */
3100 UBC 0 : PerformOffsetsTruncation(oldestMulti, newOldestMulti);
3101 EUB :
3102 UIC 0 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
3103 :
3104 0 : END_CRIT_SECTION();
3105 0 : LWLockRelease(MultiXactTruncationLock);
3106 : }
3107 :
3108 : /*
3109 : * Decide whether a MultiXactOffset page number is "older" for truncation
3110 : * purposes. Analogous to CLOGPagePrecedes().
3111 : *
3112 ECB : * Offsetting the values is optional, because MultiXactIdPrecedes() has
3113 : * translational symmetry.
3114 : */
3115 : static bool
3116 GIC 71214 : MultiXactOffsetPagePrecedes(int page1, int page2)
3117 ECB : {
3118 : MultiXactId multi1;
3119 : MultiXactId multi2;
3120 :
3121 GIC 71214 : multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
3122 CBC 71214 : multi1 += FirstMultiXactId + 1;
3123 71214 : multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
3124 GIC 71214 : multi2 += FirstMultiXactId + 1;
3125 :
3126 118690 : return (MultiXactIdPrecedes(multi1, multi2) &&
3127 47476 : MultiXactIdPrecedes(multi1,
3128 : multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
3129 : }
3130 :
3131 : /*
3132 EUB : * Decide whether a MultiXactMember page number is "older" for truncation
3133 : * purposes. There is no "invalid offset number" so use the numbers verbatim.
3134 : */
3135 : static bool
3136 UIC 0 : MultiXactMemberPagePrecedes(int page1, int page2)
3137 EUB : {
3138 : MultiXactOffset offset1;
3139 : MultiXactOffset offset2;
3140 :
3141 UBC 0 : offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
3142 UIC 0 : offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
3143 :
3144 0 : return (MultiXactOffsetPrecedes(offset1, offset2) &&
3145 0 : MultiXactOffsetPrecedes(offset1,
3146 : offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
3147 : }
3148 :
3149 : /*
3150 : * Decide which of two MultiXactIds is earlier.
3151 : *
3152 ECB : * XXX do we need to do something special for InvalidMultiXactId?
3153 : * (Doesn't look like it.)
3154 : */
3155 : bool
3156 CBC 714000 : MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
3157 : {
3158 GIC 714000 : int32 diff = (int32) (multi1 - multi2);
3159 :
3160 714000 : return (diff < 0);
3161 : }
3162 :
3163 : /*
3164 : * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
3165 : *
3166 ECB : * XXX do we need to do something special for InvalidMultiXactId?
3167 : * (Doesn't look like it.)
3168 : */
3169 : bool
3170 CBC 3880 : MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
3171 : {
3172 GIC 3880 : int32 diff = (int32) (multi1 - multi2);
3173 :
3174 3880 : return (diff <= 0);
3175 : }
3176 :
3177 :
3178 ECB : /*
3179 : * Decide which of two offsets is earlier.
3180 : */
3181 : static bool
3182 CBC 143 : MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
3183 : {
3184 GIC 143 : int32 diff = (int32) (offset1 - offset2);
3185 :
3186 143 : return (diff < 0);
3187 : }
3188 :
3189 : /*
3190 ECB : * Write an xlog record reflecting the zeroing of either a MEMBERs or
3191 : * OFFSETs page (info shows which)
3192 : */
3193 : static void
3194 CBC 20 : WriteMZeroPageXlogRec(int pageno, uint8 info)
3195 ECB : {
3196 GIC 20 : XLogBeginInsert();
3197 20 : XLogRegisterData((char *) (&pageno), sizeof(int));
3198 20 : (void) XLogInsert(RM_MULTIXACT_ID, info);
3199 20 : }
3200 :
3201 : /*
3202 : * Write a TRUNCATE xlog record
3203 : *
3204 EUB : * We must flush the xlog record to disk before returning --- see notes in
3205 : * TruncateCLOG().
3206 : */
3207 : static void
3208 UIC 0 : WriteMTruncateXlogRec(Oid oldestMultiDB,
3209 : MultiXactId startTruncOff, MultiXactId endTruncOff,
3210 : MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
3211 EUB : {
3212 : XLogRecPtr recptr;
3213 : xl_multixact_truncate xlrec;
3214 :
3215 UIC 0 : xlrec.oldestMultiDB = oldestMultiDB;
3216 EUB :
3217 UBC 0 : xlrec.startTruncOff = startTruncOff;
3218 UIC 0 : xlrec.endTruncOff = endTruncOff;
3219 EUB :
3220 UBC 0 : xlrec.startTruncMemb = startTruncMemb;
3221 0 : xlrec.endTruncMemb = endTruncMemb;
3222 EUB :
3223 UBC 0 : XLogBeginInsert();
3224 UIC 0 : XLogRegisterData((char *) (&xlrec), SizeOfMultiXactTruncate);
3225 0 : recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
3226 0 : XLogFlush(recptr);
3227 0 : }
3228 :
3229 ECB : /*
3230 : * MULTIXACT resource manager's routines
3231 : */
3232 : void
3233 GIC 4 : multixact_redo(XLogReaderState *record)
3234 ECB : {
3235 GIC 4 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3236 ECB :
3237 : /* Backup blocks are not used in multixact records */
3238 GIC 4 : Assert(!XLogRecHasAnyBlockRefs(record));
3239 :
3240 4 : if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
3241 ECB : {
3242 : int pageno;
3243 : int slotno;
3244 :
3245 CBC 1 : memcpy(&pageno, XLogRecGetData(record), sizeof(int));
3246 ECB :
3247 CBC 1 : LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
3248 :
3249 1 : slotno = ZeroMultiXactOffsetPage(pageno, false);
3250 GIC 1 : SimpleLruWritePage(MultiXactOffsetCtl, slotno);
3251 CBC 1 : Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
3252 :
3253 GIC 1 : LWLockRelease(MultiXactOffsetSLRULock);
3254 : }
3255 3 : else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
3256 ECB : {
3257 : int pageno;
3258 : int slotno;
3259 :
3260 CBC 1 : memcpy(&pageno, XLogRecGetData(record), sizeof(int));
3261 ECB :
3262 CBC 1 : LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
3263 :
3264 1 : slotno = ZeroMultiXactMemberPage(pageno, false);
3265 GIC 1 : SimpleLruWritePage(MultiXactMemberCtl, slotno);
3266 CBC 1 : Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
3267 :
3268 1 : LWLockRelease(MultiXactMemberSLRULock);
3269 ECB : }
3270 GIC 2 : else if (info == XLOG_MULTIXACT_CREATE_ID)
3271 : {
3272 2 : xl_multixact_create *xlrec =
3273 2 : (xl_multixact_create *) XLogRecGetData(record);
3274 ECB : TransactionId max_xid;
3275 : int i;
3276 :
3277 : /* Store the data back into the SLRU files */
3278 CBC 2 : RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
3279 2 : xlrec->members);
3280 :
3281 : /* Make sure nextMXact/nextOffset are beyond what this record has */
3282 GIC 2 : MultiXactAdvanceNextMXact(xlrec->mid + 1,
3283 2 : xlrec->moff + xlrec->nmembers);
3284 :
3285 : /*
3286 ECB : * Make sure nextXid is beyond any XID mentioned in the record. This
3287 : * should be unnecessary, since any XID found here ought to have other
3288 : * evidence in the XLOG, but let's be safe.
3289 : */
3290 GBC 2 : max_xid = XLogRecGetXid(record);
3291 GIC 6 : for (i = 0; i < xlrec->nmembers; i++)
3292 : {
3293 CBC 4 : if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
3294 UIC 0 : max_xid = xlrec->members[i].xid;
3295 EUB : }
3296 :
3297 GIC 2 : AdvanceNextFullTransactionIdPastXid(max_xid);
3298 : }
3299 UIC 0 : else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
3300 EUB : {
3301 : xl_multixact_truncate xlrec;
3302 : int pageno;
3303 :
3304 UIC 0 : memcpy(&xlrec, XLogRecGetData(record),
3305 : SizeOfMultiXactTruncate);
3306 :
3307 0 : elog(DEBUG1, "replaying multixact truncation: "
3308 : "offsets [%u, %u), offsets segments [%x, %x), "
3309 : "members [%u, %u), members segments [%x, %x)",
3310 : xlrec.startTruncOff, xlrec.endTruncOff,
3311 : MultiXactIdToOffsetSegment(xlrec.startTruncOff),
3312 : MultiXactIdToOffsetSegment(xlrec.endTruncOff),
3313 : xlrec.startTruncMemb, xlrec.endTruncMemb,
3314 EUB : MXOffsetToMemberSegment(xlrec.startTruncMemb),
3315 : MXOffsetToMemberSegment(xlrec.endTruncMemb));
3316 :
3317 : /* should not be required, but more than cheap enough */
3318 UIC 0 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
3319 :
3320 EUB : /*
3321 : * Advance the horizon values, so they're current at the end of
3322 : * recovery.
3323 : */
3324 UIC 0 : SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
3325 :
3326 0 : PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
3327 :
3328 : /*
3329 EUB : * During XLOG replay, latest_page_number isn't necessarily set up
3330 : * yet; insert a suitable value to bypass the sanity test in
3331 : * SimpleLruTruncate.
3332 : */
3333 UBC 0 : pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
3334 UIC 0 : MultiXactOffsetCtl->shared->latest_page_number = pageno;
3335 0 : PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
3336 EUB :
3337 LBC 0 : LWLockRelease(MultiXactTruncationLock);
3338 : }
3339 : else
3340 UBC 0 : elog(PANIC, "multixact_redo: unknown op code %u", info);
3341 GIC 4 : }
3342 :
3343 : Datum
3344 UIC 0 : pg_get_multixact_members(PG_FUNCTION_ARGS)
3345 : {
3346 : typedef struct
3347 : {
3348 EUB : MultiXactMember *members;
3349 : int nmembers;
3350 : int iter;
3351 : } mxact;
3352 UBC 0 : MultiXactId mxid = PG_GETARG_TRANSACTIONID(0);
3353 EUB : mxact *multi;
3354 : FuncCallContext *funccxt;
3355 :
3356 UIC 0 : if (mxid < FirstMultiXactId)
3357 UBC 0 : ereport(ERROR,
3358 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3359 : errmsg("invalid MultiXactId: %u", mxid)));
3360 :
3361 UIC 0 : if (SRF_IS_FIRSTCALL())
3362 EUB : {
3363 : MemoryContext oldcxt;
3364 : TupleDesc tupdesc;
3365 :
3366 UIC 0 : funccxt = SRF_FIRSTCALL_INIT();
3367 UBC 0 : oldcxt = MemoryContextSwitchTo(funccxt->multi_call_memory_ctx);
3368 :
3369 0 : multi = palloc(sizeof(mxact));
3370 : /* no need to allow for old values here */
3371 0 : multi->nmembers = GetMultiXactIdMembers(mxid, &multi->members, false,
3372 EUB : false);
3373 UBC 0 : multi->iter = 0;
3374 EUB :
3375 UNC 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
3376 0 : elog(ERROR, "return type must be a row type");
3377 0 : funccxt->tuple_desc = tupdesc;
3378 UBC 0 : funccxt->attinmeta = TupleDescGetAttInMetadata(tupdesc);
3379 UIC 0 : funccxt->user_fctx = multi;
3380 EUB :
3381 UIC 0 : MemoryContextSwitchTo(oldcxt);
3382 : }
3383 :
3384 0 : funccxt = SRF_PERCALL_SETUP();
3385 UBC 0 : multi = (mxact *) funccxt->user_fctx;
3386 EUB :
3387 UIC 0 : while (multi->iter < multi->nmembers)
3388 EUB : {
3389 : HeapTuple tuple;
3390 : char *values[2];
3391 :
3392 UBC 0 : values[0] = psprintf("%u", multi->members[multi->iter].xid);
3393 UIC 0 : values[1] = mxstatus_to_string(multi->members[multi->iter].status);
3394 :
3395 UBC 0 : tuple = BuildTupleFromCStrings(funccxt->attinmeta, values);
3396 :
3397 UIC 0 : multi->iter++;
3398 0 : pfree(values[0]);
3399 0 : SRF_RETURN_NEXT(funccxt, HeapTupleGetDatum(tuple));
3400 : }
3401 :
3402 UBC 0 : SRF_RETURN_DONE(funccxt);
3403 : }
3404 EUB :
3405 : /*
3406 : * Entrypoint for sync.c to sync offsets files.
3407 : */
3408 : int
3409 UIC 0 : multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
3410 : {
3411 UBC 0 : return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
3412 : }
3413 EUB :
3414 : /*
3415 : * Entrypoint for sync.c to sync members files.
3416 : */
3417 : int
3418 UIC 0 : multixactmemberssyncfiletag(const FileTag *ftag, char *path)
3419 : {
3420 0 : return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
3421 : }
|
