Fix setting next multixid's offset at offset wraparound
authorHeikki Linnakangas <[email protected]>
Fri, 5 Dec 2025 09:32:38 +0000 (11:32 +0200)
committerHeikki Linnakangas <[email protected]>
Fri, 5 Dec 2025 09:32:38 +0000 (11:32 +0200)
In commit 789d65364c, we started updating the next multixid's offset
too when recording a multixid, so that it can always be used to
calculate the number of members. I got it wrong at offset wraparound:
we need to skip over offset 0. Fix that.

Discussion: https://www.postgresql.org/message-id/d9996478-389a-4340-8735-bfad456b313c@iki.fi
Backpatch-through: 14

src/backend/access/transam/multixact.c

index 27f02faec80f0cd651911ceb34c9cb169c900f52..8ed3fd9d071c44ec5772c5bb942145b1fe395647 100644 (file)
@@ -909,6 +909,7 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
        int64           next_pageno;
        int                     next_entryno;
        MultiXactOffset *next_offptr;
+       MultiXactOffset next_offset;
        LWLock     *lock;
        LWLock     *prevlock = NULL;
 
@@ -976,11 +977,15 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
                next_offptr += next_entryno;
        }
 
-       if (*next_offptr != offset + nmembers)
+       /* Like in GetNewMultiXactId(), skip over offset 0 */
+       next_offset = offset + nmembers;
+       if (next_offset == 0)
+               next_offset = 1;
+       if (*next_offptr != next_offset)
        {
                /* should already be set to the correct value, or not at all */
                Assert(*next_offptr == 0);
-               *next_offptr = offset + nmembers;
+               *next_offptr = next_offset;
                MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
        }