/*
* Replay a revmap page extension
*/
static void
brin_xlog_revmap_extend(XLogReaderState *record)
{
XLogRecPtr lsn = record->EndRecPtr;
xl_brin_revmap_extend *xlrec;
Buffer metabuf;
Buffer buf;
Page page;
BlockNumber targetBlk;
XLogRedoAction action;
xlrec = (xl_brin_revmap_extend *) XLogRecGetData(record);
XLogRecGetBlockTag(record, 1, NULL, NULL, &targetBlk);
Assert(xlrec->targetBlk == targetBlk);
/* Update the metapage */
action = XLogReadBufferForRedo(record, 0, &metabuf);
if (action == BLK_NEEDS_REDO)
{
Page metapg;
BrinMetaPageData *metadata;
metapg = BufferGetPage(metabuf);
metadata = (BrinMetaPageData *) PageGetContents(metapg);
Assert(metadata->lastRevmapPage == xlrec->targetBlk - 1);
metadata->lastRevmapPage = xlrec->targetBlk;
PageSetLSN(metapg, lsn);
/*
* Set pd_lower just past the end of the metadata. This is essential,
* because without doing so, metadata will be lost if xlog.c
* compresses the page. (We must do this here because pre-v11
* versions of PG did not set the metapage's pd_lower correctly, so a
* pg_upgraded index might contain the wrong value.)
*/
((PageHeader) metapg)->pd_lower =
((char *) metadata + sizeof(BrinMetaPageData)) - (char *) metapg;
MarkBufferDirty(metabuf);
}
/*
* Re-init the target block as a revmap page. There's never a full- page
* image here.
*/
buf = XLogInitBufferForRedo(record, 1);
page = (Page) BufferGetPage(buf);
brin_page_init(page, BRIN_PAGETYPE_REVMAP);
PageSetLSN(page, lsn);
MarkBufferDirty(buf);
UnlockReleaseBuffer(buf);
if (BufferIsValid(metabuf))
UnlockReleaseBuffer(metabuf);
}
/*
* Initialize a new BRIN index' metapage.
*/
void
brin_metapage_init(Page page, BlockNumber pagesPerRange, uint16 version)
{
BrinMetaPageData *metadata;
brin_page_init(page, BRIN_PAGETYPE_META);
metadata = (BrinMetaPageData *) PageGetContents(page);
metadata->brinMagic = BRIN_META_MAGIC;
metadata->brinVersion = version;
metadata->pagesPerRange = pagesPerRange;
/*
* Note we cheat here a little. 0 is not a valid revmap block number
* (because it's the metapage buffer), but doing this enables the first
* revmap page to be created when the index is.
*/
metadata->lastRevmapPage = 0;
}
/*
* Initialize a page as an empty regular BRIN page, WAL-log this, and record
* the page in FSM.
*
* There are several corner situations in which we extend the relation to
* obtain a new page and later find that we cannot use it immediately. When
* that happens, we don't want to leave the page go unrecorded in FSM, because
* there is no mechanism to get the space back and the index would bloat.
* Also, because we would not WAL-log the action that would initialize the
* page, the page would go uninitialized in a standby (or after recovery).
*/
static void
brin_initialize_empty_new_buffer(Relation idxrel, Buffer buffer)
{
Page page;
BRIN_elog((DEBUG2,
"brin_initialize_empty_new_buffer: initializing blank page %u",
BufferGetBlockNumber(buffer)));
START_CRIT_SECTION();
page = BufferGetPage(buffer);
brin_page_init(page, BRIN_PAGETYPE_REGULAR);
MarkBufferDirty(buffer);
log_newpage_buffer(buffer, true);
END_CRIT_SECTION();
/*
* We update the FSM for this page, but this is not WAL-logged. This is
* acceptable because VACUUM will scan the index and update the FSM with
* pages whose FSM records were forgotten in a crash.
*/
RecordPageWithFreeSpace(idxrel, BufferGetBlockNumber(buffer),
br_page_get_freespace(page));
}
/*
* Try to extend the revmap by one page. This might not happen for a number of
* reasons; caller is expected to retry until the expected outcome is obtained.
*/
static void
revmap_physical_extend(BrinRevmap *revmap)
{
Buffer buf;
Page page;
Page metapage;
BrinMetaPageData *metadata;
BlockNumber mapBlk;
BlockNumber nblocks;
Relation irel = revmap->rm_irel;
bool needLock = !RELATION_IS_LOCAL(irel);
/*
* Lock the metapage. This locks out concurrent extensions of the revmap,
* but note that we still need to grab the relation extension lock because
* another backend can extend the index with regular BRIN pages.
*/
LockBuffer(revmap->rm_metaBuf, BUFFER_LOCK_EXCLUSIVE);
metapage = BufferGetPage(revmap->rm_metaBuf);
metadata = (BrinMetaPageData *) PageGetContents(metapage);
/*
* Check that our cached lastRevmapPage value was up-to-date; if it
* wasn't, update the cached copy and have caller start over.
*/
if (metadata->lastRevmapPage != revmap->rm_lastRevmapPage)
{
revmap->rm_lastRevmapPage = metadata->lastRevmapPage;
LockBuffer(revmap->rm_metaBuf, BUFFER_LOCK_UNLOCK);
return;
}
mapBlk = metadata->lastRevmapPage + 1;
nblocks = RelationGetNumberOfBlocks(irel);
if (mapBlk < nblocks)
{
buf = ReadBuffer(irel, mapBlk);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
page = BufferGetPage(buf);
}
else
{
if (needLock)
LockRelationForExtension(irel, ExclusiveLock);
buf = ReadBuffer(irel, P_NEW);
if (BufferGetBlockNumber(buf) != mapBlk)
{
/*
* Very rare corner case: somebody extended the relation
* concurrently after we read its length. If this happens, give
* up and have caller start over. We will have to evacuate that
* page from under whoever is using it.
*/
if (needLock)
UnlockRelationForExtension(irel, ExclusiveLock);
LockBuffer(revmap->rm_metaBuf, BUFFER_LOCK_UNLOCK);
return;
}
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
page = BufferGetPage(buf);
if (needLock)
UnlockRelationForExtension(irel, ExclusiveLock);
}
/* Check that it's a regular block (or an empty page) */
if (!PageIsNew(page) && !BRIN_IS_REGULAR_PAGE(page))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg("unexpected page type 0x%04X in BRIN index \"%s\" block %u",
BRIN_PAGE_TYPE(page),
RelationGetRelationName(irel),
BufferGetBlockNumber(buf))));
/* If the page is in use, evacuate it and restart */
if (brin_start_evacuating_page(irel, buf))
{
LockBuffer(revmap->rm_metaBuf, BUFFER_LOCK_UNLOCK);
brin_evacuate_page(irel, revmap->rm_pagesPerRange, revmap, buf);
/* have caller start over */
return;
}
/*
* Ok, we have now locked the metapage and the target block. Re-initialize
* it as a revmap page.
*/
START_CRIT_SECTION();
/* the rm_tids array is initialized to all invalid by PageInit */
brin_page_init(page, BRIN_PAGETYPE_REVMAP);
MarkBufferDirty(buf);
metadata->lastRevmapPage = mapBlk;
MarkBufferDirty(revmap->rm_metaBuf);
if (RelationNeedsWAL(revmap->rm_irel))
{
xl_brin_revmap_extend xlrec;
XLogRecPtr recptr;
XLogRecData rdata[2];
xlrec.node = revmap->rm_irel->rd_node;
xlrec.targetBlk = mapBlk;
rdata[0].data = (char *) &xlrec;
rdata[0].len = SizeOfBrinRevmapExtend;
rdata[0].buffer = InvalidBuffer;
rdata[0].buffer_std = false;
rdata[0].next = &(rdata[1]);
rdata[1].data = (char *) NULL;
rdata[1].len = 0;
rdata[1].buffer = revmap->rm_metaBuf;
rdata[1].buffer_std = false;
rdata[1].next = NULL;
recptr = XLogInsert(RM_BRIN_ID, XLOG_BRIN_REVMAP_EXTEND, rdata);
PageSetLSN(metapage, recptr);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
LockBuffer(revmap->rm_metaBuf, BUFFER_LOCK_UNLOCK);
UnlockReleaseBuffer(buf);
}
/*
* Update tuple origtup (size origsz), located in offset oldoff of buffer
* oldbuf, to newtup (size newsz) as summary tuple for the page range starting
* at heapBlk. oldbuf must not be locked on entry, and is not locked at exit.
*
* If samepage is true, attempt to put the new tuple in the same page, but if
* there's no room, use some other one.
*
* If the update is successful, return true; the revmap is updated to point to
* the new tuple. If the update is not done for whatever reason, return false.
* Caller may retry the update if this happens.
*/
bool
brin_doupdate(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, BlockNumber heapBlk,
Buffer oldbuf, OffsetNumber oldoff,
const BrinTuple *origtup, Size origsz,
const BrinTuple *newtup, Size newsz,
bool samepage)
{
Page oldpage;
ItemId oldlp;
BrinTuple *oldtup;
Size oldsz;
Buffer newbuf;
bool extended;
Assert(newsz == MAXALIGN(newsz));
/* If the item is oversized, don't bother. */
if (newsz > BrinMaxItemSize)
{
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) newsz,
(unsigned long) BrinMaxItemSize,
RelationGetRelationName(idxrel))));
return false; /* keep compiler quiet */
}
/* make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
if (!samepage)
{
/* need a page on which to put the item */
newbuf = brin_getinsertbuffer(idxrel, oldbuf, newsz, &extended);
if (!BufferIsValid(newbuf))
{
Assert(!extended);
return false;
}
/*
* Note: it's possible (though unlikely) that the returned newbuf is
* the same as oldbuf, if brin_getinsertbuffer determined that the old
* buffer does in fact have enough space.
*/
if (newbuf == oldbuf)
{
Assert(!extended);
newbuf = InvalidBuffer;
}
}
else
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
newbuf = InvalidBuffer;
extended = false;
}
oldpage = BufferGetPage(oldbuf);
oldlp = PageGetItemId(oldpage, oldoff);
/*
* Check that the old tuple wasn't updated concurrently: it might have
* moved someplace else entirely ...
*/
if (!ItemIdIsNormal(oldlp))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
/*
* If this happens, and the new buffer was obtained by extending the
* relation, then we need to ensure we don't leave it uninitialized or
* forget about it.
*/
if (BufferIsValid(newbuf))
{
if (extended)
brin_initialize_empty_new_buffer(idxrel, newbuf);
UnlockReleaseBuffer(newbuf);
if (extended)
FreeSpaceMapVacuum(idxrel);
}
return false;
}
oldsz = ItemIdGetLength(oldlp);
oldtup = (BrinTuple *) PageGetItem(oldpage, oldlp);
/*
* ... or it might have been updated in place to different contents.
*/
if (!brin_tuples_equal(oldtup, oldsz, origtup, origsz))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
if (BufferIsValid(newbuf))
{
if (extended)
brin_initialize_empty_new_buffer(idxrel, newbuf);
UnlockReleaseBuffer(newbuf);
if (extended)
FreeSpaceMapVacuum(idxrel);
}
return false;
}
/*
* Great, the old tuple is intact. We can proceed with the update.
*
* If there's enough room in the old page for the new tuple, replace it.
*
* Note that there might now be enough space on the page even though the
* caller told us there isn't, if a concurrent update moved another tuple
* elsewhere or replaced a tuple with a smaller one.
*/
if (((BrinPageFlags(oldpage) & BRIN_EVACUATE_PAGE) == 0) &&
brin_can_do_samepage_update(oldbuf, origsz, newsz))
{
if (BufferIsValid(newbuf))
{
/* as above */
if (extended)
brin_initialize_empty_new_buffer(idxrel, newbuf);
UnlockReleaseBuffer(newbuf);
}
START_CRIT_SECTION();
if (!PageIndexTupleOverwrite(oldpage, oldoff, (Item) newtup, newsz))
elog(ERROR, "failed to replace BRIN tuple");
MarkBufferDirty(oldbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_samepage_update xlrec;
XLogRecPtr recptr;
uint8 info = XLOG_BRIN_SAMEPAGE_UPDATE;
xlrec.offnum = oldoff;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinSamepageUpdate);
XLogRegisterBuffer(0, oldbuf, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) newtup, newsz);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(oldpage, recptr);
}
END_CRIT_SECTION();
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return true;
}
else if (newbuf == InvalidBuffer)
{
/*
* Not enough space, but caller said that there was. Tell them to
* start over.
*/
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
return false;
}
else
{
/*
* Not enough free space on the oldpage. Put the new tuple on the new
* page, and update the revmap.
*/
Page newpage = BufferGetPage(newbuf);
Buffer revmapbuf;
ItemPointerData newtid;
OffsetNumber newoff;
BlockNumber newblk = InvalidBlockNumber;
Size freespace = 0;
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
START_CRIT_SECTION();
/*
* We need to initialize the page if it's newly obtained. Note we
* will WAL-log the initialization as part of the update, so we don't
* need to do that here.
*/
if (extended)
brin_page_init(BufferGetPage(newbuf), BRIN_PAGETYPE_REGULAR);
PageIndexTupleDeleteNoCompact(oldpage, oldoff);
newoff = PageAddItem(newpage, (Item) newtup, newsz,
InvalidOffsetNumber, false, false);
if (newoff == InvalidOffsetNumber)
elog(ERROR, "failed to add BRIN tuple to new page");
MarkBufferDirty(oldbuf);
MarkBufferDirty(newbuf);
/* needed to update FSM below */
if (extended)
{
newblk = BufferGetBlockNumber(newbuf);
freespace = br_page_get_freespace(newpage);
}
ItemPointerSet(&newtid, BufferGetBlockNumber(newbuf), newoff);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, newtid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_update xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_UPDATE | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.insert.offnum = newoff;
xlrec.insert.heapBlk = heapBlk;
xlrec.insert.pagesPerRange = pagesPerRange;
xlrec.oldOffnum = oldoff;
XLogBeginInsert();
/* new page */
XLogRegisterData((char *) &xlrec, SizeOfBrinUpdate);
XLogRegisterBuffer(0, newbuf, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) newtup, newsz);
/* revmap page */
XLogRegisterBuffer(1, revmapbuf, 0);
/* old page */
XLogRegisterBuffer(2, oldbuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(oldpage, recptr);
PageSetLSN(newpage, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
UnlockReleaseBuffer(newbuf);
if (extended)
{
Assert(BlockNumberIsValid(newblk));
RecordPageWithFreeSpace(idxrel, newblk, freespace);
FreeSpaceMapVacuum(idxrel);
}
return true;
}
}
/*
* Insert an index tuple into the index relation. The revmap is updated to
* mark the range containing the given page as pointing to the inserted entry.
* A WAL record is written.
*
* The buffer, if valid, is first checked for free space to insert the new
* entry; if there isn't enough, a new buffer is obtained and pinned. No
* buffer lock must be held on entry, no buffer lock is held on exit.
*
* Return value is the offset number where the tuple was inserted.
*/
OffsetNumber
brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, Buffer *buffer, BlockNumber heapBlk,
BrinTuple *tup, Size itemsz)
{
Page page;
BlockNumber blk;
OffsetNumber off;
Buffer revmapbuf;
ItemPointerData tid;
bool extended;
Assert(itemsz == MAXALIGN(itemsz));
/* If the item is oversized, don't even bother. */
if (itemsz > BrinMaxItemSize)
{
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) itemsz,
(unsigned long) BrinMaxItemSize,
RelationGetRelationName(idxrel))));
return InvalidOffsetNumber; /* keep compiler quiet */
}
/* Make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
/*
* Acquire lock on buffer supplied by caller, if any. If it doesn't have
* enough space, unpin it to obtain a new one below.
*/
if (BufferIsValid(*buffer))
{
/*
* It's possible that another backend (or ourselves!) extended the
* revmap over the page we held a pin on, so we cannot assume that
* it's still a regular page.
*/
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (br_page_get_freespace(BufferGetPage(*buffer)) < itemsz)
{
UnlockReleaseBuffer(*buffer);
*buffer = InvalidBuffer;
}
}
/*
* If we still don't have a usable buffer, have brin_getinsertbuffer
* obtain one for us.
*/
if (!BufferIsValid(*buffer))
{
do
*buffer = brin_getinsertbuffer(idxrel, InvalidBuffer, itemsz, &extended);
while (!BufferIsValid(*buffer));
}
else
extended = false;
/* Now obtain lock on revmap buffer */
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
page = BufferGetPage(*buffer);
blk = BufferGetBlockNumber(*buffer);
/* Execute the actual insertion */
START_CRIT_SECTION();
if (extended)
brin_page_init(BufferGetPage(*buffer), BRIN_PAGETYPE_REGULAR);
off = PageAddItem(page, (Item) tup, itemsz, InvalidOffsetNumber,
false, false);
if (off == InvalidOffsetNumber)
elog(ERROR, "could not insert new index tuple to page");
MarkBufferDirty(*buffer);
BRIN_elog((DEBUG2, "inserted tuple (%u,%u) for range starting at %u",
blk, off, heapBlk));
ItemPointerSet(&tid, blk, off);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, tid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_insert xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_INSERT | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.heapBlk = heapBlk;
xlrec.pagesPerRange = pagesPerRange;
xlrec.offnum = off;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinInsert);
XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) tup, itemsz);
XLogRegisterBuffer(1, revmapbuf, 0);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(page, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
/* Tuple is firmly on buffer; we can release our locks */
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return off;
}
/*
* Common part of an insert or update. Inserts the new tuple and updates the
* revmap.
*/
static void
brin_xlog_insert_update(XLogReaderState *record,
xl_brin_insert *xlrec)
{
XLogRecPtr lsn = record->EndRecPtr;
Buffer buffer;
BlockNumber regpgno;
Page page;
XLogRedoAction action;
/*
* If we inserted the first and only tuple on the page, re-initialize the
* page from scratch.
*/
if (XLogRecGetInfo(record) & XLOG_BRIN_INIT_PAGE)
{
buffer = XLogInitBufferForRedo(record, 0);
page = BufferGetPage(buffer);
brin_page_init(page, BRIN_PAGETYPE_REGULAR);
action = BLK_NEEDS_REDO;
}
else
{
action = XLogReadBufferForRedo(record, 0, &buffer);
}
/* need this page's blkno to store in revmap */
regpgno = BufferGetBlockNumber(buffer);
/* insert the index item into the page */
if (action == BLK_NEEDS_REDO)
{
OffsetNumber offnum;
BrinTuple *tuple;
Size tuplen;
tuple = (BrinTuple *) XLogRecGetBlockData(record, 0, &tuplen);
Assert(tuple->bt_blkno == xlrec->heapBlk);
page = (Page) BufferGetPage(buffer);
offnum = xlrec->offnum;
if (PageGetMaxOffsetNumber(page) + 1 < offnum)
elog(PANIC, "brin_xlog_insert_update: invalid max offset number");
offnum = PageAddItem(page, (Item) tuple, tuplen, offnum, true, false);
if (offnum == InvalidOffsetNumber)
elog(PANIC, "brin_xlog_insert_update: failed to add tuple");
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
/* update the revmap */
action = XLogReadBufferForRedo(record, 1, &buffer);
if (action == BLK_NEEDS_REDO)
{
ItemPointerData tid;
ItemPointerSet(&tid, regpgno, xlrec->offnum);
page = (Page) BufferGetPage(buffer);
brinSetHeapBlockItemptr(buffer, xlrec->pagesPerRange, xlrec->heapBlk,
tid);
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
/* XXX no FSM updates here ... */
}
/*
* Return a pinned and exclusively locked buffer which can be used to insert an
* index item of size itemsz. If oldbuf is a valid buffer, it is also locked
* (in a order determined to avoid deadlocks.)
*
* If there's no existing page with enough free space to accomodate the new
* item, the relation is extended. If this happens, *extended is set to true.
*
* If we find that the old page is no longer a regular index page (because
* of a revmap extension), the old buffer is unlocked and we return
* InvalidBuffer.
*/
static Buffer
brin_getinsertbuffer(Relation irel, Buffer oldbuf, Size itemsz,
bool *was_extended)
{
BlockNumber oldblk;
BlockNumber newblk;
Page page;
int freespace;
if (BufferIsValid(oldbuf))
oldblk = BufferGetBlockNumber(oldbuf);
else
oldblk = InvalidBlockNumber;
/*
* Loop until we find a page with sufficient free space. By the time we
* return to caller out of this loop, both buffers are valid and locked;
* if we have to restart here, neither buffer is locked and buf is not a
* pinned buffer.
*/
newblk = RelationGetTargetBlock(irel);
if (newblk == InvalidBlockNumber)
newblk = GetPageWithFreeSpace(irel, itemsz);
for (;;)
{
Buffer buf;
bool extensionLockHeld = false;
bool extended = false;
CHECK_FOR_INTERRUPTS();
if (newblk == InvalidBlockNumber)
{
/*
* There's not enough free space in any existing index page,
* according to the FSM: extend the relation to obtain a shiny new
* page.
*/
if (!RELATION_IS_LOCAL(irel))
{
LockRelationForExtension(irel, ExclusiveLock);
extensionLockHeld = true;
}
buf = ReadBuffer(irel, P_NEW);
newblk = BufferGetBlockNumber(buf);
*was_extended = extended = true;
BRIN_elog((DEBUG2, "brin_getinsertbuffer: extending to page %u",
BufferGetBlockNumber(buf)));
}
else if (newblk == oldblk)
{
/*
* There's an odd corner-case here where the FSM is out-of-date,
* and gave us the old page.
*/
buf = oldbuf;
}
else
{
buf = ReadBuffer(irel, newblk);
}
/*
* We lock the old buffer first, if it's earlier than the new one; but
* before we do, we need to check that it hasn't been turned into a
* revmap page concurrently; if we detect that it happened, give up
* and tell caller to start over.
*/
if (BufferIsValid(oldbuf) && oldblk < newblk)
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
if (!BRIN_IS_REGULAR_PAGE(BufferGetPage(oldbuf)))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
ReleaseBuffer(buf);
return InvalidBuffer;
}
}
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (extensionLockHeld)
UnlockRelationForExtension(irel, ExclusiveLock);
page = BufferGetPage(buf);
if (extended)
brin_page_init(page, BRIN_PAGETYPE_REGULAR);
/*
* We have a new buffer to insert into. Check that the new page has
* enough free space, and return it if it does; otherwise start over.
* Note that we allow for the FSM to be out of date here, and in that
* case we update it and move on.
*
* (br_page_get_freespace also checks that the FSM didn't hand us a
* page that has since been repurposed for the revmap.)
*/
freespace = br_page_get_freespace(page);
if (freespace >= itemsz)
{
RelationSetTargetBlock(irel, BufferGetBlockNumber(buf));
/*
* Since the target block specification can get lost on cache
* invalidations, make sure we update the more permanent FSM with
* data about it before going away.
*/
if (extended)
RecordPageWithFreeSpace(irel, BufferGetBlockNumber(buf),
freespace);
/*
* Lock the old buffer if not locked already. Note that in this
* case we know for sure it's a regular page: it's later than the
* new page we just got, which is not a revmap page, and revmap
* pages are always consecutive.
*/
if (BufferIsValid(oldbuf) && oldblk > newblk)
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
Assert(BRIN_IS_REGULAR_PAGE(BufferGetPage(oldbuf)));
}
return buf;
}
/* This page is no good. */
/*
* If an entirely new page does not contain enough free space for the
* new item, then surely that item is oversized. Complain loudly; but
* first make sure we record the page as free, for next time.
*/
if (extended)
{
RecordPageWithFreeSpace(irel, BufferGetBlockNumber(buf),
freespace);
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) itemsz,
(unsigned long) freespace,
RelationGetRelationName(irel))));
return InvalidBuffer; /* keep compiler quiet */
}
if (newblk != oldblk)
UnlockReleaseBuffer(buf);
if (BufferIsValid(oldbuf) && oldblk <= newblk)
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
newblk = RecordAndGetPageWithFreeSpace(irel, newblk, freespace, itemsz);
}
}