1 /*-------------------------------------------------------------------------
4 * PostgreSQL subtransaction-log manager
6 * The pg_subtrans manager is a pg_xact-like manager that stores the parent
7 * transaction Id for each transaction. It is a fundamental part of the
8 * nested transactions implementation. A main transaction has a parent
9 * of InvalidTransactionId, and each subtransaction has its immediate parent.
10 * The tree can easily be walked from child to parent, but not in the
13 * This code is based on xact.c, but the robustness requirements
14 * are completely different from pg_xact, because we only need to remember
15 * pg_subtrans information for currently-open transactions. Thus, there is
16 * no need to preserve data over a crash and restart.
18 * There are no XLOG interactions since we do not care about preserving
19 * data across crashes. During database startup, we simply force the
20 * currently-active page of SUBTRANS to zeroes.
22 * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
23 * Portions Copyright (c) 1994, Regents of the University of California
25 * src/backend/access/transam/subtrans.c
27 *-------------------------------------------------------------------------
31 #include "access/slru.h"
32 #include "access/subtrans.h"
33 #include "access/transam.h"
35 #include "utils/snapmgr.h"
39 * Defines for SubTrans page sizes. A page is the same BLCKSZ as is used
40 * everywhere else in Postgres.
42 * Note: because TransactionIds are 32 bits and wrap around at 0xFFFFFFFF,
43 * SubTrans page numbering also wraps around at
44 * 0xFFFFFFFF/SUBTRANS_XACTS_PER_PAGE, and segment numbering at
45 * 0xFFFFFFFF/SUBTRANS_XACTS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need take no
46 * explicit notice of that fact in this module, except when comparing segment
47 * and page numbers in TruncateSUBTRANS (see SubTransPagePrecedes) and zeroing
48 * them in StartupSUBTRANS.
51 /* We need four bytes per xact */
52 #define SUBTRANS_XACTS_PER_PAGE (BLCKSZ / sizeof(TransactionId))
54 #define TransactionIdToPage(xid) ((xid) / (TransactionId) SUBTRANS_XACTS_PER_PAGE)
55 #define TransactionIdToEntry(xid) ((xid) % (TransactionId) SUBTRANS_XACTS_PER_PAGE)
59 * Link to shared-memory data structures for SUBTRANS control
61 static SlruCtlData SubTransCtlData
;
63 #define SubTransCtl (&SubTransCtlData)
66 static int ZeroSUBTRANSPage(int pageno
);
67 static bool SubTransPagePrecedes(int page1
, int page2
);
71 * Record the parent of a subtransaction in the subtrans log.
74 SubTransSetParent(TransactionId xid
, TransactionId parent
)
76 int pageno
= TransactionIdToPage(xid
);
77 int entryno
= TransactionIdToEntry(xid
);
81 Assert(TransactionIdIsValid(parent
));
82 Assert(TransactionIdFollows(xid
, parent
));
84 LWLockAcquire(SubtransSLRULock
, LW_EXCLUSIVE
);
86 slotno
= SimpleLruReadPage(SubTransCtl
, pageno
, true, xid
);
87 ptr
= (TransactionId
*) SubTransCtl
->shared
->page_buffer
[slotno
];
91 * It's possible we'll try to set the parent xid multiple times but we
92 * shouldn't ever be changing the xid from one valid xid to another valid
93 * xid, which would corrupt the data structure.
97 Assert(*ptr
== InvalidTransactionId
);
99 SubTransCtl
->shared
->page_dirty
[slotno
] = true;
102 LWLockRelease(SubtransSLRULock
);
106 * Interrogate the parent of a transaction in the subtrans log.
109 SubTransGetParent(TransactionId xid
)
111 int pageno
= TransactionIdToPage(xid
);
112 int entryno
= TransactionIdToEntry(xid
);
115 TransactionId parent
;
117 /* Can't ask about stuff that might not be around anymore */
118 Assert(TransactionIdFollowsOrEquals(xid
, TransactionXmin
));
120 /* Bootstrap and frozen XIDs have no parent */
121 if (!TransactionIdIsNormal(xid
))
122 return InvalidTransactionId
;
124 /* lock is acquired by SimpleLruReadPage_ReadOnly */
126 slotno
= SimpleLruReadPage_ReadOnly(SubTransCtl
, pageno
, xid
);
127 ptr
= (TransactionId
*) SubTransCtl
->shared
->page_buffer
[slotno
];
132 LWLockRelease(SubtransSLRULock
);
138 * SubTransGetTopmostTransaction
140 * Returns the topmost transaction of the given transaction id.
142 * Because we cannot look back further than TransactionXmin, it is possible
143 * that this function will lie and return an intermediate subtransaction ID
144 * instead of the true topmost parent ID. This is OK, because in practice
145 * we only care about detecting whether the topmost parent is still running
146 * or is part of a current snapshot's list of still-running transactions.
147 * Therefore, any XID before TransactionXmin is as good as any other.
150 SubTransGetTopmostTransaction(TransactionId xid
)
152 TransactionId parentXid
= xid
,
155 /* Can't ask about stuff that might not be around anymore */
156 Assert(TransactionIdFollowsOrEquals(xid
, TransactionXmin
));
158 while (TransactionIdIsValid(parentXid
))
160 previousXid
= parentXid
;
161 if (TransactionIdPrecedes(parentXid
, TransactionXmin
))
163 parentXid
= SubTransGetParent(parentXid
);
166 * By convention the parent xid gets allocated first, so should always
167 * precede the child xid. Anything else points to a corrupted data
168 * structure that could lead to an infinite loop, so exit.
170 if (!TransactionIdPrecedes(parentXid
, previousXid
))
171 elog(ERROR
, "pg_subtrans contains invalid entry: xid %u points to parent xid %u",
172 previousXid
, parentXid
);
175 Assert(TransactionIdIsValid(previousXid
));
182 * Initialization of shared memory for SUBTRANS
185 SUBTRANSShmemSize(void)
187 return SimpleLruShmemSize(NUM_SUBTRANS_BUFFERS
, 0);
191 SUBTRANSShmemInit(void)
193 SubTransCtl
->PagePrecedes
= SubTransPagePrecedes
;
194 SimpleLruInit(SubTransCtl
, "Subtrans", NUM_SUBTRANS_BUFFERS
, 0,
195 SubtransSLRULock
, "pg_subtrans",
196 LWTRANCHE_SUBTRANS_BUFFER
, SYNC_HANDLER_NONE
);
197 SlruPagePrecedesUnitTests(SubTransCtl
, SUBTRANS_XACTS_PER_PAGE
);
201 * This func must be called ONCE on system install. It creates
202 * the initial SUBTRANS segment. (The SUBTRANS directory is assumed to
203 * have been created by the initdb shell script, and SUBTRANSShmemInit
204 * must have been called already.)
206 * Note: it's not really necessary to create the initial segment now,
207 * since slru.c would create it on first write anyway. But we may as well
208 * do it to be sure the directory is set up correctly.
211 BootStrapSUBTRANS(void)
215 LWLockAcquire(SubtransSLRULock
, LW_EXCLUSIVE
);
217 /* Create and zero the first page of the subtrans log */
218 slotno
= ZeroSUBTRANSPage(0);
220 /* Make sure it's written out */
221 SimpleLruWritePage(SubTransCtl
, slotno
);
222 Assert(!SubTransCtl
->shared
->page_dirty
[slotno
]);
224 LWLockRelease(SubtransSLRULock
);
228 * Initialize (or reinitialize) a page of SUBTRANS to zeroes.
230 * The page is not actually written, just set up in shared memory.
231 * The slot number of the new page is returned.
233 * Control lock must be held at entry, and will be held at exit.
236 ZeroSUBTRANSPage(int pageno
)
238 return SimpleLruZeroPage(SubTransCtl
, pageno
);
242 * This must be called ONCE during postmaster or standalone-backend startup,
243 * after StartupXLOG has initialized ShmemVariableCache->nextXid.
245 * oldestActiveXID is the oldest XID of any prepared transaction, or nextXid
249 StartupSUBTRANS(TransactionId oldestActiveXID
)
251 FullTransactionId nextXid
;
256 * Since we don't expect pg_subtrans to be valid across crashes, we
257 * initialize the currently-active page(s) to zeroes during startup.
258 * Whenever we advance into a new page, ExtendSUBTRANS will likewise zero
259 * the new page without regard to whatever was previously on disk.
261 LWLockAcquire(SubtransSLRULock
, LW_EXCLUSIVE
);
263 startPage
= TransactionIdToPage(oldestActiveXID
);
264 nextXid
= ShmemVariableCache
->nextXid
;
265 endPage
= TransactionIdToPage(XidFromFullTransactionId(nextXid
));
267 while (startPage
!= endPage
)
269 (void) ZeroSUBTRANSPage(startPage
);
271 /* must account for wraparound */
272 if (startPage
> TransactionIdToPage(MaxTransactionId
))
275 (void) ZeroSUBTRANSPage(startPage
);
277 LWLockRelease(SubtransSLRULock
);
281 * Perform a checkpoint --- either during shutdown, or on-the-fly
284 CheckPointSUBTRANS(void)
287 * Write dirty SUBTRANS pages to disk
289 * This is not actually necessary from a correctness point of view. We do
290 * it merely to improve the odds that writing of dirty pages is done by
291 * the checkpoint process and not by backends.
293 TRACE_POSTGRESQL_SUBTRANS_CHECKPOINT_START(true);
294 SimpleLruWriteAll(SubTransCtl
, true);
295 TRACE_POSTGRESQL_SUBTRANS_CHECKPOINT_DONE(true);
300 * Make sure that SUBTRANS has room for a newly-allocated XID.
302 * NB: this is called while holding XidGenLock. We want it to be very fast
303 * most of the time; even when it's not so fast, no actual I/O need happen
304 * unless we're forced to write out a dirty subtrans page to make room
308 ExtendSUBTRANS(TransactionId newestXact
)
313 * No work except at first XID of a page. But beware: just after
314 * wraparound, the first XID of page zero is FirstNormalTransactionId.
316 if (TransactionIdToEntry(newestXact
) != 0 &&
317 !TransactionIdEquals(newestXact
, FirstNormalTransactionId
))
320 pageno
= TransactionIdToPage(newestXact
);
322 LWLockAcquire(SubtransSLRULock
, LW_EXCLUSIVE
);
325 ZeroSUBTRANSPage(pageno
);
327 LWLockRelease(SubtransSLRULock
);
332 * Remove all SUBTRANS segments before the one holding the passed transaction ID
334 * oldestXact is the oldest TransactionXmin of any running transaction. This
335 * is called only during checkpoint.
338 TruncateSUBTRANS(TransactionId oldestXact
)
343 * The cutoff point is the start of the segment containing oldestXact. We
344 * pass the *page* containing oldestXact to SimpleLruTruncate. We step
345 * back one transaction to avoid passing a cutoff page that hasn't been
346 * created yet in the rare case that oldestXact would be the first item on
347 * a page and oldestXact == next XID. In that case, if we didn't subtract
348 * one, we'd trigger SimpleLruTruncate's wraparound detection.
350 TransactionIdRetreat(oldestXact
);
351 cutoffPage
= TransactionIdToPage(oldestXact
);
353 SimpleLruTruncate(SubTransCtl
, cutoffPage
);
358 * Decide whether a SUBTRANS page number is "older" for truncation purposes.
359 * Analogous to CLOGPagePrecedes().
362 SubTransPagePrecedes(int page1
, int page2
)
367 xid1
= ((TransactionId
) page1
) * SUBTRANS_XACTS_PER_PAGE
;
368 xid1
+= FirstNormalTransactionId
+ 1;
369 xid2
= ((TransactionId
) page2
) * SUBTRANS_XACTS_PER_PAGE
;
370 xid2
+= FirstNormalTransactionId
+ 1;
372 return (TransactionIdPrecedes(xid1
, xid2
) &&
373 TransactionIdPrecedes(xid1
, xid2
+ SUBTRANS_XACTS_PER_PAGE
- 1));