/*
* Fetches the next entry from a visimap store index scan.
*
* It is the responsibility of the caller to decode the return value
* correctly.
*
*/
static HeapTuple
AppendOnlyVisimapStore_GetNextTuple(
AppendOnlyVisimapStore *visiMapStore,
IndexScanDesc indexScan,
ScanDirection scanDirection)
{
Assert(visiMapStore);
Assert(RelationIsValid(visiMapStore->visimapRelation));
Assert(RelationIsValid(visiMapStore->visimapIndex));
Assert(indexScan);
return index_getnext(indexScan, scanDirection);
}
/*
* systable_getnext_ordered --- get next tuple in an ordered catalog scan
*/
HeapTuple
systable_getnext_ordered(SysScanDesc sysscan, ScanDirection direction)
{
HeapTuple htup;
Assert(sysscan->irel);
htup = index_getnext(sysscan->iscan, direction);
/* See notes in systable_getnext */
if (htup && sysscan->iscan->xs_recheck)
elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
return htup;
}
/* ----------
* toast_delete_datum -
*
* Delete a single external stored value.
* ----------
*/
static void
toast_delete_datum(Relation rel __attribute__((unused)), Datum value)
{
varattrib *attr = (varattrib *) DatumGetPointer(value);
Relation toastrel;
Relation toastidx;
ScanKeyData toastkey;
IndexScanDesc toastscan;
HeapTuple toasttup;
if (!VARATT_IS_EXTERNAL(attr))
return;
/*
* Open the toast relation and its index
*/
toastrel = heap_open(attr->va_external.va_toastrelid,
RowExclusiveLock);
toastidx = index_open(toastrel->rd_rel->reltoastidxid, RowExclusiveLock);
/*
* Setup a scan key to fetch from the index by va_valueid (we don't
* particularly care whether we see them in sequence or not)
*/
ScanKeyInit(&toastkey,
(AttrNumber) 1,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(attr->va_external.va_valueid));
/*
* Find all the chunks. (We don't actually care whether we see them in
* sequence or not, but since we've already locked the index we might as
* well use systable_beginscan_ordered.)
*/
toastscan = index_beginscan(toastrel, toastidx,
SnapshotToast, 1, &toastkey);
while ((toasttup = index_getnext(toastscan, ForwardScanDirection)) != NULL)
{
/*
* Have a chunk, delete it
*/
simple_heap_delete(toastrel, &toasttup->t_self);
}
/*
* End scan and close relations
*/
index_endscan(toastscan);
index_close(toastidx, RowExclusiveLock);
heap_close(toastrel, RowExclusiveLock);
}
/* ----------
* toast_delete_datum -
*
* Delete a single external stored value.
* ----------
*/
static void
toast_delete_datum(Relation rel, Datum value)
{
varattrib *attr = (varattrib *) DatumGetPointer(value);
Relation toastrel;
Relation toastidx;
ScanKeyData toastkey;
IndexScanDesc toastscan;
HeapTuple toasttup;
if (!VARATT_IS_EXTERNAL(attr))
return;
/*
* Open the toast relation and it's index
*/
toastrel = heap_open(attr->va_content.va_external.va_toastrelid,
RowExclusiveLock);
toastidx = index_open(toastrel->rd_rel->reltoastidxid);
/*
* Setup a scan key to fetch from the index by va_valueid (we don't
* particularly care whether we see them in sequence or not)
*/
ScanKeyEntryInitialize(&toastkey,
(bits16) 0,
(AttrNumber) 1,
(RegProcedure) F_OIDEQ,
ObjectIdGetDatum(attr->va_content.va_external.va_valueid));
/*
* Find the chunks by index
*/
toastscan = index_beginscan(toastrel, toastidx, SnapshotToast,
1, &toastkey);
while ((toasttup = index_getnext(toastscan, ForwardScanDirection)) != NULL)
{
/*
* Have a chunk, delete it
*/
simple_heap_delete(toastrel, &toasttup->t_self);
}
/*
* End scan and close relations
*/
index_endscan(toastscan);
index_close(toastidx);
heap_close(toastrel, RowExclusiveLock);
}
/*
* GetNewOidWithIndex
* Guts of GetNewOid: use the supplied index
*
* This is exported separately because there are cases where we want to use
* an index that will not be recognized by RelationGetOidIndex: TOAST tables
* and pg_largeobject have indexes that are usable, but have multiple columns
* and are on ordinary columns rather than a true OID column. This code
* will work anyway, so long as the OID is the index's first column.
*
* Caller must have a suitable lock on the relation.
*/
Oid
GetNewOidWithIndex(Relation relation, Relation indexrel)
{
Oid newOid;
IndexScanDesc scan;
ScanKeyData key;
bool collides;
/* Generate new OIDs until we find one not in the table */
do
{
CHECK_FOR_INTERRUPTS();
newOid = GetNewObjectId();
ScanKeyInit(&key,
(AttrNumber) 1,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(newOid));
/* see notes above about using SnapshotDirty */
scan = index_beginscan(relation, indexrel,
SnapshotDirty, 1, &key);
collides = HeapTupleIsValid(index_getnext(scan, ForwardScanDirection));
index_endscan(scan);
} while (collides);
if (IsSystemNamespace(RelationGetNamespace(relation)))
{
if (Gp_role == GP_ROLE_EXECUTE)
{
if (relation->rd_id != 2604 /* pg_attrdef */ && relation->rd_id != 2606 /* pg_constraint */ && relation->rd_id != 2615 /* pg_namespace */)
elog(DEBUG1,"Allocating Oid %u with index on relid %u %s in EXECUTE mode",newOid,relation->rd_id, RelationGetRelationName(relation));
else
elog(DEBUG4,"Allocating Oid %u with index on relid %u %s in EXECUTE mode",newOid,relation->rd_id, RelationGetRelationName(relation));
}
if (Gp_role == GP_ROLE_DISPATCH)
{
elog(DEBUG5,"Allocating Oid %u with index on relid %u %s in DISPATCH mode",newOid,relation->rd_id, RelationGetRelationName(relation));
}
}
return newOid;
}
/*
* AppendOnlyBlockDirectory_DeleteSegmentFile
*
* Deletes all block directory entries for given segment file of an
* append-only relation.
*/
void
AppendOnlyBlockDirectory_DeleteSegmentFile(
AppendOnlyEntry *aoEntry,
Snapshot snapshot,
int segno,
int columnGroupNo)
{
Assert(aoEntry);
Assert(OidIsValid(aoEntry->blkdirrelid));
Assert(OidIsValid(aoEntry->blkdiridxid));
Relation blkdirRel = heap_open(aoEntry->blkdirrelid, RowExclusiveLock);
Relation blkdirIdx = index_open(aoEntry->blkdiridxid, RowExclusiveLock);
ScanKeyData scanKey;
ScanKeyInit(&scanKey,
1, /* segno */
BTEqualStrategyNumber,
F_INT4EQ,
Int32GetDatum(segno));
IndexScanDesc indexScan = index_beginscan(
blkdirRel,
blkdirIdx,
snapshot,
1,
&scanKey);
HeapTuple tuple = NULL;
while ((tuple = index_getnext(indexScan, ForwardScanDirection)) != NULL)
{
simple_heap_delete(blkdirRel,
&tuple->t_self);
}
index_endscan(indexScan);
index_close(blkdirIdx, RowExclusiveLock);
heap_close(blkdirRel, RowExclusiveLock);
}
void
rebuildheap(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex)
{
Relation LocalNewHeap, LocalOldHeap, LocalOldIndex;
IndexScanDesc ScanDesc;
RetrieveIndexResult ScanResult;
ItemPointer HeapTid;
HeapTuple LocalHeapTuple;
Buffer LocalBuffer;
Oid OIDNewHeapInsert;
/*
* Open the relations I need. Scan through the OldHeap on the OldIndex and
* insert each tuple into the NewHeap.
*/
LocalNewHeap=(Relation)heap_open(OIDNewHeap);
LocalOldHeap=(Relation)heap_open(OIDOldHeap);
LocalOldIndex=(Relation)index_open(OIDOldIndex);
ScanDesc=index_beginscan(LocalOldIndex, false, 0, (ScanKey) NULL);
while ((ScanResult =
index_getnext(ScanDesc, ForwardScanDirection)) != NULL) {
HeapTid = &ScanResult->heap_iptr;
LocalHeapTuple = heap_fetch(LocalOldHeap, 0, HeapTid, &LocalBuffer);
OIDNewHeapInsert =
heap_insert(LocalNewHeap, LocalHeapTuple);
pfree(ScanResult);
ReleaseBuffer(LocalBuffer);
}
index_close(LocalOldIndex);
heap_close(LocalOldHeap);
heap_close(LocalNewHeap);
}
/*
* systable_getnext --- get next tuple in a heap-or-index scan
*
* Returns NULL if no more tuples available.
*
* Note that returned tuple is a reference to data in a disk buffer;
* it must not be modified, and should be presumed inaccessible after
* next getnext() or endscan() call.
*/
HeapTuple
systable_getnext(SysScanDesc sysscan)
{
HeapTuple htup;
if (sysscan->irel)
{
htup = index_getnext(sysscan->iscan, ForwardScanDirection);
/*
* We currently don't need to support lossy index operators for any
* system catalog scan. It could be done here, using the scan keys to
* drive the operator calls, if we arranged to save the heap attnums
* during systable_beginscan(); this is practical because we still
* wouldn't need to support indexes on expressions.
*/
if (htup && sysscan->iscan->xs_recheck)
elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
}
else
htup = heap_getnext(sysscan->scan, ForwardScanDirection);
return htup;
}
/* ----------
* toast_fetch_datum_slice -
*
* Reconstruct a segment of a varattrib from the chunks saved
* in the toast relation
* ----------
*/
static varattrib *
toast_fetch_datum_slice(varattrib *attr, int32 sliceoffset, int32 length)
{
Relation toastrel;
Relation toastidx;
ScanKeyData toastkey[3];
int nscankeys;
IndexScanDesc toastscan;
HeapTuple ttup;
TupleDesc toasttupDesc;
varattrib *result;
int32 attrsize;
int32 residx;
int32 nextidx;
int numchunks;
int startchunk;
int endchunk;
int32 startoffset;
int32 endoffset;
int totalchunks;
Pointer chunk;
bool isnull;
int32 chunksize;
int32 chcpystrt;
int32 chcpyend;
attrsize = attr->va_content.va_external.va_extsize;
totalchunks = ((attrsize - 1) / TOAST_MAX_CHUNK_SIZE) + 1;
if (sliceoffset >= attrsize)
{
sliceoffset = 0;
length = 0;
}
if (((sliceoffset + length) > attrsize) || length < 0)
length = attrsize - sliceoffset;
result = (varattrib *) palloc(length + VARHDRSZ);
VARATT_SIZEP(result) = length + VARHDRSZ;
if (VARATT_IS_COMPRESSED(attr))
VARATT_SIZEP(result) |= VARATT_FLAG_COMPRESSED;
if (length == 0)
return (result); /* Can save a lot of work at this point! */
startchunk = sliceoffset / TOAST_MAX_CHUNK_SIZE;
endchunk = (sliceoffset + length - 1) / TOAST_MAX_CHUNK_SIZE;
numchunks = (endchunk - startchunk) + 1;
startoffset = sliceoffset % TOAST_MAX_CHUNK_SIZE;
endoffset = (sliceoffset + length - 1) % TOAST_MAX_CHUNK_SIZE;
/*
* Open the toast relation and it's index
*/
toastrel = heap_open(attr->va_content.va_external.va_toastrelid,
AccessShareLock);
toasttupDesc = toastrel->rd_att;
toastidx = index_open(toastrel->rd_rel->reltoastidxid);
/*
* Setup a scan key to fetch from the index. This is either two keys
* or three depending on the number of chunks.
*/
ScanKeyEntryInitialize(&toastkey[0],
(bits16) 0,
(AttrNumber) 1,
(RegProcedure) F_OIDEQ,
ObjectIdGetDatum(attr->va_content.va_external.va_valueid));
/*
* Now dependent on number of chunks:
*/
if (numchunks == 1)
{
ScanKeyEntryInitialize(&toastkey[1],
(bits16) 0,
(AttrNumber) 2,
(RegProcedure) F_INT4EQ,
Int32GetDatum(startchunk));
nscankeys = 2;
}
else
{
ScanKeyEntryInitialize(&toastkey[1],
(bits16) 0,
(AttrNumber) 2,
(RegProcedure) F_INT4GE,
Int32GetDatum(startchunk));
ScanKeyEntryInitialize(&toastkey[2],
(bits16) 0,
(AttrNumber) 2,
(RegProcedure) F_INT4LE,
Int32GetDatum(endchunk));
nscankeys = 3;
}
/*
* Read the chunks by index
*
* The index is on (valueid, chunkidx) so they will come in order
*/
nextidx = startchunk;
toastscan = index_beginscan(toastrel, toastidx, SnapshotToast,
nscankeys, toastkey);
while ((ttup = index_getnext(toastscan, ForwardScanDirection)) != NULL)
{
/*
* Have a chunk, extract the sequence number and the data
*/
residx = DatumGetInt32(heap_getattr(ttup, 2, toasttupDesc, &isnull));
Assert(!isnull);
chunk = DatumGetPointer(heap_getattr(ttup, 3, toasttupDesc, &isnull));
Assert(!isnull);
chunksize = VARATT_SIZE(chunk) - VARHDRSZ;
/*
* Some checks on the data we've found
*/
if ((residx != nextidx) || (residx > endchunk) || (residx < startchunk))
elog(ERROR, "unexpected chunk number %d (expected %d) for toast value %u",
residx, nextidx,
attr->va_content.va_external.va_valueid);
if (residx < totalchunks - 1)
{
if (chunksize != TOAST_MAX_CHUNK_SIZE)
elog(ERROR, "unexpected chunk size %d in chunk %d for toast value %u",
chunksize, residx,
attr->va_content.va_external.va_valueid);
}
else
{
if ((residx * TOAST_MAX_CHUNK_SIZE + chunksize) != attrsize)
elog(ERROR, "unexpected chunk size %d in chunk %d for toast value %u",
chunksize, residx,
attr->va_content.va_external.va_valueid);
}
/*
* Copy the data into proper place in our result
*/
chcpystrt = 0;
chcpyend = chunksize - 1;
if (residx == startchunk)
chcpystrt = startoffset;
if (residx == endchunk)
chcpyend = endoffset;
memcpy(((char *) VARATT_DATA(result)) +
(residx * TOAST_MAX_CHUNK_SIZE - sliceoffset) + chcpystrt,
VARATT_DATA(chunk) + chcpystrt,
(chcpyend - chcpystrt) + 1);
nextidx++;
}
/*
* Final checks that we successfully fetched the datum
*/
if (nextidx != (endchunk + 1))
elog(ERROR, "missing chunk number %d for toast value %u",
nextidx,
attr->va_content.va_external.va_valueid);
/*
* End scan and close relations
*/
index_endscan(toastscan);
index_close(toastidx);
heap_close(toastrel, AccessShareLock);
return result;
}
/* ----------
* toast_fetch_datum -
*
* Reconstruct an in memory varattrib from the chunks saved
* in the toast relation
* ----------
*/
static varattrib *
toast_fetch_datum(varattrib *attr)
{
Relation toastrel;
Relation toastidx;
ScanKeyData toastkey;
IndexScanDesc toastscan;
HeapTuple ttup;
TupleDesc toasttupDesc;
varattrib *result;
int32 ressize;
int32 residx,
nextidx;
int32 numchunks;
Pointer chunk;
bool isnull;
int32 chunksize;
ressize = attr->va_content.va_external.va_extsize;
numchunks = ((ressize - 1) / TOAST_MAX_CHUNK_SIZE) + 1;
result = (varattrib *) palloc(ressize + VARHDRSZ);
VARATT_SIZEP(result) = ressize + VARHDRSZ;
if (VARATT_IS_COMPRESSED(attr))
VARATT_SIZEP(result) |= VARATT_FLAG_COMPRESSED;
/*
* Open the toast relation and its index
*/
toastrel = heap_open(attr->va_content.va_external.va_toastrelid,
AccessShareLock);
toasttupDesc = toastrel->rd_att;
toastidx = index_open(toastrel->rd_rel->reltoastidxid);
/*
* Setup a scan key to fetch from the index by va_valueid
*/
ScanKeyEntryInitialize(&toastkey,
(bits16) 0,
(AttrNumber) 1,
(RegProcedure) F_OIDEQ,
ObjectIdGetDatum(attr->va_content.va_external.va_valueid));
/*
* Read the chunks by index
*
* Note that because the index is actually on (valueid, chunkidx) we will
* see the chunks in chunkidx order, even though we didn't explicitly
* ask for it.
*/
nextidx = 0;
toastscan = index_beginscan(toastrel, toastidx, SnapshotToast,
1, &toastkey);
while ((ttup = index_getnext(toastscan, ForwardScanDirection)) != NULL)
{
/*
* Have a chunk, extract the sequence number and the data
*/
residx = DatumGetInt32(heap_getattr(ttup, 2, toasttupDesc, &isnull));
Assert(!isnull);
chunk = DatumGetPointer(heap_getattr(ttup, 3, toasttupDesc, &isnull));
Assert(!isnull);
chunksize = VARATT_SIZE(chunk) - VARHDRSZ;
/*
* Some checks on the data we've found
*/
if (residx != nextidx)
elog(ERROR, "unexpected chunk number %d (expected %d) for toast value %u",
residx, nextidx,
attr->va_content.va_external.va_valueid);
if (residx < numchunks - 1)
{
if (chunksize != TOAST_MAX_CHUNK_SIZE)
elog(ERROR, "unexpected chunk size %d in chunk %d for toast value %u",
chunksize, residx,
attr->va_content.va_external.va_valueid);
}
else if (residx < numchunks)
{
if ((residx * TOAST_MAX_CHUNK_SIZE + chunksize) != ressize)
elog(ERROR, "unexpected chunk size %d in chunk %d for toast value %u",
chunksize, residx,
attr->va_content.va_external.va_valueid);
}
else
elog(ERROR, "unexpected chunk number %d for toast value %u",
residx,
attr->va_content.va_external.va_valueid);
/*
* Copy the data into proper place in our result
*/
memcpy(((char *) VARATT_DATA(result)) + residx * TOAST_MAX_CHUNK_SIZE,
VARATT_DATA(chunk),
chunksize);
nextidx++;
}
/*
* Final checks that we successfully fetched the datum
*/
if (nextidx != numchunks)
elog(ERROR, "missing chunk number %d for toast value %u",
nextidx,
attr->va_content.va_external.va_valueid);
/*
* End scan and close relations
*/
index_endscan(toastscan);
index_close(toastidx);
heap_close(toastrel, AccessShareLock);
return result;
}
/*
* load_last_minipage
*
* Search through the block directory btree to find the last row that
* contains the last minipage.
*/
static void
load_last_minipage(AppendOnlyBlockDirectory *blockDirectory,
int64 lastSequence,
int columnGroupNo)
{
Relation blkdirRel = blockDirectory->blkdirRel;
Relation blkdirIdx = blockDirectory->blkdirIdx;
TupleDesc heapTupleDesc;
IndexScanDesc idxScanDesc;
HeapTuple tuple = NULL;
MemoryContext oldcxt;
int numScanKeys = blockDirectory->numScanKeys;
ScanKey scanKeys = blockDirectory->scanKeys;
#ifdef USE_ASSERT_CHECKING
StrategyNumber *strategyNumbers = blockDirectory->strategyNumbers;
#endif /* USE_ASSERT_CHECKING */
Assert(blockDirectory->aoRel != NULL);
Assert(blockDirectory->blkdirRel != NULL);
Assert(blockDirectory->blkdirIdx != NULL);
oldcxt = MemoryContextSwitchTo(blockDirectory->memoryContext);
heapTupleDesc = RelationGetDescr(blkdirRel);
Assert(numScanKeys == 3);
Assert(blockDirectory->currentSegmentFileInfo != NULL);
/* Setup the scan keys for the scan. */
Assert(scanKeys != NULL);
Assert(strategyNumbers != NULL);
if (lastSequence == 0)
lastSequence = 1;
scanKeys[0].sk_argument =
Int32GetDatum(blockDirectory->currentSegmentFileNum);
scanKeys[1].sk_argument = Int32GetDatum(columnGroupNo);
scanKeys[2].sk_argument = Int64GetDatum(lastSequence);
/*
* Search the btree to find the entry in the block directory
* that contains the last minipage.
*/
idxScanDesc = index_beginscan(blkdirRel, blkdirIdx,
blockDirectory->appendOnlyMetaDataSnapshot,
numScanKeys, scanKeys);
tuple = index_getnext(idxScanDesc, BackwardScanDirection);
if (tuple != NULL)
{
extract_minipage(blockDirectory,
tuple,
heapTupleDesc,
columnGroupNo);
}
index_endscan(idxScanDesc);
MemoryContextSwitchTo(oldcxt);
ereportif(Debug_appendonly_print_blockdirectory, LOG,
(errmsg("Append-only block directory load last minipage: "
"(columnGroupNo, lastSequence, nEntries) = (%d, " INT64_FORMAT ", %u)",
columnGroupNo, lastSequence,
blockDirectory->minipages[columnGroupNo].numMinipageEntries)));
}
bool
CheckNewRelFileNodeIsOk(Oid newOid, Oid reltablespace, bool relisshared,
Relation pg_class)
{
RelFileNode rnode;
char *rpath;
int fd;
bool collides;
if (pg_class)
{
Oid oidIndex;
Relation indexrel;
IndexScanDesc scan;
ScanKeyData key;
Assert(!IsBootstrapProcessingMode());
Assert(pg_class->rd_rel->relhasoids);
/* The relcache will cache the identity of the OID index for us */
oidIndex = RelationGetOidIndex(pg_class);
Assert(OidIsValid(oidIndex));
indexrel = index_open(oidIndex, AccessShareLock);
ScanKeyInit(&key,
(AttrNumber) 1,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(newOid));
scan = index_beginscan(pg_class, indexrel, SnapshotDirty, 1, &key);
collides = HeapTupleIsValid(index_getnext(scan, ForwardScanDirection));
index_endscan(scan);
index_close(indexrel, AccessShareLock);
if (collides)
elog(ERROR, "relfilenode %d already in use in \"pg_class\"",
newOid);
}
/* This should match RelationInitPhysicalAddr */
rnode.spcNode = reltablespace ? reltablespace : MyDatabaseTableSpace;
rnode.dbNode = relisshared ? InvalidOid : MyDatabaseId;
rnode.relNode = newOid;
/* Check for existing file of same name */
rpath = relpath(rnode);
fd = BasicOpenFile(rpath, O_RDONLY | PG_BINARY, 0);
if (fd >= 0)
{
/* definite collision */
gp_retry_close(fd);
collides = true;
}
else
collides = false;
pfree(rpath);
if (collides && !relisshared)
elog(ERROR, "oid %d already in use", newOid);
while(GetNewObjectId() < newOid);
return !collides;
}
/*
* AppendOnlyBlockDirectory_GetEntry
*
* Find a directory entry for the given AOTupleId in the block directory.
* If such an entry is found, return true. Otherwise, return false.
*
* The range for directoryEntry is assigned accordingly in this function.
*
* The block directory for the appendonly table should exist before calling
* this function.
*/
bool
AppendOnlyBlockDirectory_GetEntry(
AppendOnlyBlockDirectory *blockDirectory,
AOTupleId *aoTupleId,
int columnGroupNo,
AppendOnlyBlockDirectoryEntry *directoryEntry)
{
int segmentFileNum = AOTupleIdGet_segmentFileNum(aoTupleId);
int64 rowNum = AOTupleIdGet_rowNum(aoTupleId);
int i;
Relation blkdirRel = blockDirectory->blkdirRel;
Relation blkdirIdx = blockDirectory->blkdirIdx;
int numScanKeys = blockDirectory->numScanKeys;
ScanKey scanKeys = blockDirectory->scanKeys;
TupleDesc heapTupleDesc;
FileSegInfo *fsInfo = NULL;
IndexScanDesc idxScanDesc;
HeapTuple tuple = NULL;
MinipagePerColumnGroup *minipageInfo =
&blockDirectory->minipages[columnGroupNo];
int entry_no = -1;
int tmpGroupNo;
if (blkdirRel == NULL || blkdirIdx == NULL)
{
Assert(RelationIsValid(blockDirectory->aoRel));
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR),
errmsg("Block directory for append-only relation '%s' does not exist",
RelationGetRelationName(blockDirectory->aoRel))));
return false;
}
ereportif(Debug_appendonly_print_blockdirectory, LOG,
(errmsg("Append-only block directory get entry: "
"(columnGroupNo, segmentFileNum, rowNum) = "
"(%d, %d, " INT64_FORMAT ")",
columnGroupNo, segmentFileNum, rowNum)));
/*
* If the segment file number is the same as
* blockDirectory->currentSegmentFileNum, the in-memory minipage
* may contain such an entry. We search the in-memory minipage
* first. If such an entry can not be found, we search for the
* appropriate minipage by using the block directory btree index.
*/
if (segmentFileNum == blockDirectory->currentSegmentFileNum &&
minipageInfo->numMinipageEntries > 0)
{
Assert(blockDirectory->currentSegmentFileInfo != NULL);
MinipageEntry *firstentry =
&minipageInfo->minipage->entry[0];
if (rowNum >= firstentry->firstRowNum)
{
/*
* Check if the existing minipage contains the requested
* rowNum. If so, just get it.
*/
entry_no = find_minipage_entry(minipageInfo->minipage,
minipageInfo->numMinipageEntries,
rowNum);
if (entry_no != -1)
{
return set_directoryentry_range(blockDirectory,
columnGroupNo,
entry_no,
directoryEntry);
}
/*
* The given rowNum may point to a tuple that does not exist
* in the AO table any more, either because of cancellation of
* an insert, or due to crashes during an insert. If this is
* the case, rowNum is smaller than the highest entry in
* the in-memory minipage entry.
*/
else
{
MinipageEntry *entry =
&minipageInfo->minipage->entry[minipageInfo->numMinipageEntries - 1];
if (rowNum < entry->firstRowNum + entry->rowCount - 1)
return false;
}
}
}
for (i = 0; i < blockDirectory->totalSegfiles; i++)
{
fsInfo = blockDirectory->segmentFileInfo[i];
if (!blockDirectory->isAOCol && segmentFileNum == fsInfo->segno)
break;
else if (blockDirectory->isAOCol && segmentFileNum ==
((AOCSFileSegInfo*)fsInfo)->segno)
break;
}
Assert(fsInfo != NULL);
/*
* Search the btree index to find the minipage that contains
* the rowNum. We find the minipages for all column groups, since
* currently we will need to access all columns at the same time.
*/
heapTupleDesc = RelationGetDescr(blkdirRel);
Assert(numScanKeys == 3);
for (tmpGroupNo = 0; tmpGroupNo < blockDirectory->numColumnGroups; tmpGroupNo++)
{
if (blockDirectory->proj && !blockDirectory->proj[tmpGroupNo])
{
/* Ignore columns that are not projected. */
continue;
}
/* Setup the scan keys for the scan. */
Assert(scanKeys != NULL);
scanKeys[0].sk_argument = Int32GetDatum(segmentFileNum);
scanKeys[1].sk_argument = Int32GetDatum(tmpGroupNo);
scanKeys[2].sk_argument = Int64GetDatum(rowNum);
idxScanDesc = index_beginscan(blkdirRel, blkdirIdx,
blockDirectory->appendOnlyMetaDataSnapshot,
numScanKeys, scanKeys);
tuple = index_getnext(idxScanDesc, BackwardScanDirection);
if (tuple != NULL)
{
/*
* MPP-17061: we need to update currentSegmentFileNum
* & currentSegmentFileInfo at the same time when we
* load the minipage for the block directory entry we
* found, otherwise we would risk having inconsistency
* between currentSegmentFileNum/currentSegmentFileInfo
* and minipage contents, which would cause wrong block
* header offset being returned in following block
* directory entry look up.
*/
blockDirectory->currentSegmentFileNum = segmentFileNum;
blockDirectory->currentSegmentFileInfo = fsInfo;
extract_minipage(blockDirectory,
tuple,
heapTupleDesc,
tmpGroupNo);
}
else
{
/* MPP-17061: index look up failed, row is invisible */
index_endscan(idxScanDesc);
return false;
}
index_endscan(idxScanDesc);
}
{
MinipagePerColumnGroup *minipageInfo;
minipageInfo = &blockDirectory->minipages[columnGroupNo];
/*
* Perform a binary search over the minipage to find
* the entry about the AO block.
*/
entry_no = find_minipage_entry(minipageInfo->minipage,
minipageInfo->numMinipageEntries,
rowNum);
/* If there are no entries, return false. */
if (entry_no == -1 && minipageInfo->numMinipageEntries == 0)
return false;
if (entry_no == -1)
{
/*
* Since the last few blocks may not be logged in the block
* directory, we always use the last entry.
*/
entry_no = minipageInfo->numMinipageEntries - 1;
}
return set_directoryentry_range(blockDirectory,
columnGroupNo,
entry_no,
directoryEntry);
}
return false;
}
/* ----------
* toast_fetch_datum_slice -
*
* Reconstruct a segment of a Datum from the chunks saved
* in the toast relation
* ----------
*/
static struct varlena *
toast_fetch_datum_slice(struct varlena *attr, int32 sliceoffset, int32 length)
{
Relation toastrel;
Relation toastidx;
ScanKeyData toastkey[3];
int nscankeys;
IndexScanDesc toastscan;
HeapTuple ttup;
TupleDesc toasttupDesc;
varattrib *result;
int32 attrsize;
int32 residx;
int32 nextidx;
int numchunks;
int startchunk;
int endchunk;
int32 startoffset;
int32 endoffset;
int totalchunks;
Pointer chunk;
bool isnull;
int32 chunksize;
int32 chcpystrt;
int32 chcpyend;
attrsize = ((varattrib *)attr)->va_external.va_extsize;
totalchunks = ((attrsize - 1) / TOAST_MAX_CHUNK_SIZE) + 1;
if (sliceoffset >= attrsize)
{
sliceoffset = 0;
length = 0;
}
if (((sliceoffset + length) > attrsize) || length < 0)
length = attrsize - sliceoffset;
result = (varattrib *) palloc(length + VARHDRSZ);
SET_VARSIZE(result, length + VARHDRSZ);
if (VARATT_EXTERNAL_IS_COMPRESSED(attr))
VARATT_SET_COMPRESSED(result);
if (length == 0)
return (struct varlena *)result; /* Can save a lot of work at this point! */
startchunk = sliceoffset / TOAST_MAX_CHUNK_SIZE;
endchunk = (sliceoffset + length - 1) / TOAST_MAX_CHUNK_SIZE;
numchunks = (endchunk - startchunk) + 1;
startoffset = sliceoffset % TOAST_MAX_CHUNK_SIZE;
endoffset = (sliceoffset + length - 1) % TOAST_MAX_CHUNK_SIZE;
/*
* Open the toast relation and its index
*/
toastrel = heap_open(((varattrib *)attr)->va_external.va_toastrelid, AccessShareLock);
toasttupDesc = toastrel->rd_att;
toastidx = index_open(toastrel->rd_rel->reltoastidxid, AccessShareLock);
/*
* Setup a scan key to fetch from the index. This is either two keys or
* three depending on the number of chunks.
*/
ScanKeyInit(&toastkey[0],
(AttrNumber) 1,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(((varattrib *)attr)->va_external.va_valueid));
/*
* Use equality condition for one chunk, a range condition otherwise:
*/
if (numchunks == 1)
{
ScanKeyInit(&toastkey[1],
(AttrNumber) 2,
BTEqualStrategyNumber, F_INT4EQ,
Int32GetDatum(startchunk));
nscankeys = 2;
}
else
{
ScanKeyInit(&toastkey[1],
(AttrNumber) 2,
BTGreaterEqualStrategyNumber, F_INT4GE,
Int32GetDatum(startchunk));
ScanKeyInit(&toastkey[2],
(AttrNumber) 2,
BTLessEqualStrategyNumber, F_INT4LE,
Int32GetDatum(endchunk));
nscankeys = 3;
}
/*
* Read the chunks by index
*
* The index is on (valueid, chunkidx) so they will come in order
*/
nextidx = startchunk;
toastscan = index_beginscan(toastrel, toastidx,
SnapshotToast, nscankeys, toastkey);
while ((ttup = index_getnext(toastscan, ForwardScanDirection)) != NULL)
{
/*
* Have a chunk, extract the sequence number and the data
*/
residx = DatumGetInt32(fastgetattr(ttup, 2, toasttupDesc, &isnull));
Assert(!isnull);
chunk = DatumGetPointer(fastgetattr(ttup, 3, toasttupDesc, &isnull));
Assert(!isnull);
if (VARATT_IS_SHORT((varattrib *)chunk))
chunksize = VARSIZE_SHORT((varattrib *)chunk) - VARHDRSZ_SHORT;
else if (!VARATT_IS_EXTENDED((varattrib *)chunk))
chunksize = VARSIZE((varattrib *)chunk) - VARHDRSZ;
else {
elog(ERROR, "found toasted toast chunk?");
chunksize = 0; /* shut compiler up */
}
/*
* Some checks on the data we've found
*/
if ((residx != nextidx) || (residx > endchunk) || (residx < startchunk))
elog(ERROR, "unexpected chunk number %d (expected %d) for toast value %u",
residx, nextidx,
((varattrib *)attr)->va_external.va_valueid);
if (residx < totalchunks - 1)
{
if (chunksize != TOAST_MAX_CHUNK_SIZE)
elog(ERROR, "unexpected chunk size %d in chunk %d for toast value %u of %d when fetching slice (expected %d)",
chunksize, residx,
((varattrib *)attr)->va_external.va_valueid, totalchunks-1,
(int)TOAST_MAX_CHUNK_SIZE);
}
else if (residx == totalchunks-1)
{
if ((residx * TOAST_MAX_CHUNK_SIZE + chunksize) != attrsize)
elog(ERROR, "unexpected chunk size %d in chunk %d for final toast value %u when fetching slice (expected %d)",
chunksize, residx,
((varattrib *)attr)->va_external.va_valueid,
attrsize - residx * (int)TOAST_MAX_CHUNK_SIZE);
}
else
{
elog(ERROR, "unexpected chunk");
}
/*
* Copy the data into proper place in our result
*/
chcpystrt = 0;
chcpyend = chunksize - 1;
if (residx == startchunk)
chcpystrt = startoffset;
if (residx == endchunk)
chcpyend = endoffset;
memcpy(((char *) VARDATA(result)) +
(residx * TOAST_MAX_CHUNK_SIZE - sliceoffset) + chcpystrt,
VARDATA((varattrib *)chunk) + chcpystrt,
(chcpyend - chcpystrt) + 1);
nextidx++;
}
/*
* Final checks that we successfully fetched the datum
*/
if (nextidx != (endchunk + 1))
elog(ERROR, "missing chunk number %d for toast value %u",
nextidx,
((varattrib *)attr)->va_external.va_valueid);
/*
* End scan and close relations
*/
index_endscan(toastscan);
index_close(toastidx, AccessShareLock);
heap_close(toastrel, AccessShareLock);
return (struct varlena *)result;
}
/* ----------------------------------------------------------------
* IndexNext
*
* Retrieve a tuple from the IndexScan node's currentRelation
* using the index specified in the IndexScanState information.
* ----------------------------------------------------------------
*/
TupleTableSlot *
IndexNext(IndexScanState *node)
{
EState *estate;
ExprContext *econtext;
ScanDirection direction;
IndexScanDesc scandesc;
Index scanrelid;
HeapTuple tuple;
TupleTableSlot *slot;
/*
* extract necessary information from index scan node
*/
estate = node->ss.ps.state;
direction = estate->es_direction;
initScanDesc(node);
/* flip direction if this is an overall backward scan */
if (ScanDirectionIsBackward(((IndexScan *) node->ss.ps.plan)->indexorderdir))
{
if (ScanDirectionIsForward(direction))
direction = BackwardScanDirection;
else if (ScanDirectionIsBackward(direction))
direction = ForwardScanDirection;
}
scandesc = node->iss_ScanDesc;
econtext = node->ss.ps.ps_ExprContext;
slot = node->ss.ss_ScanTupleSlot;
scanrelid = ((IndexScan *) node->ss.ps.plan)->scan.scanrelid;
/*
* Check if we are evaluating PlanQual for tuple of this relation.
* Additional checking is not good, but no other way for now. We could
* introduce new nodes for this case and handle IndexScan --> NewNode
* switching in Init/ReScan plan...
*/
if (estate->es_evTuple != NULL &&
estate->es_evTuple[scanrelid - 1] != NULL)
{
if (estate->es_evTupleNull[scanrelid - 1])
{
if (!node->ss.ps.delayEagerFree)
{
ExecEagerFreeIndexScan(node);
}
return ExecClearTuple(slot);
}
ExecStoreGenericTuple(estate->es_evTuple[scanrelid - 1], slot, false);
/* Does the tuple meet the indexqual condition? */
econtext->ecxt_scantuple = slot;
ResetExprContext(econtext);
if (!ExecQual(node->indexqualorig, econtext, false))
{
if (!node->ss.ps.delayEagerFree)
{
ExecEagerFreeIndexScan(node);
}
ExecClearTuple(slot); /* would not be returned by scan */
}
/* Flag for the next call that no more tuples */
estate->es_evTupleNull[scanrelid - 1] = true;
Gpmon_M_Incr_Rows_Out(GpmonPktFromIndexScanState(node));
CheckSendPlanStateGpmonPkt(&node->ss.ps);
return slot;
}
/*
* ok, now that we have what we need, fetch the next tuple.
*/
if ((tuple = index_getnext(scandesc, direction)) != NULL)
{
/*
* Store the scanned tuple in the scan tuple slot of the scan state.
* Note: we pass 'false' because tuples returned by amgetnext are
* pointers onto disk pages and must not be pfree()'d.
*/
ExecStoreHeapTuple(tuple, /* tuple to store */
slot, /* slot to store in */
scandesc->xs_cbuf, /* buffer containing tuple */
false); /* don't pfree */
Gpmon_M_Incr_Rows_Out(GpmonPktFromIndexScanState(node));
CheckSendPlanStateGpmonPkt(&node->ss.ps);
return slot;
}
if (!node->ss.ps.delayEagerFree)
{
ExecEagerFreeIndexScan(node);
}
/*
* if we get here it means the index scan failed so we are at the end of
* the scan..
*/
return ExecClearTuple(slot);
}
int
inv_write(LargeObjectDesc *obj_desc, const char *buf, int nbytes)
{
int nwritten = 0;
int n;
int off;
int len;
int32 pageno = (int32) (obj_desc->offset / LOBLKSIZE);
ScanKeyData skey[2];
IndexScanDesc sd;
HeapTuple oldtuple;
Form_pg_largeobject olddata;
bool neednextpage;
bytea *datafield;
bool pfreeit;
struct
{
bytea hdr;
char data[LOBLKSIZE]; /* make struct big enough */
int32 align_it; /* ensure struct is aligned well enough */
} workbuf;
char *workb = VARDATA(&workbuf.hdr);
HeapTuple newtup;
Datum values[Natts_pg_largeobject];
bool nulls[Natts_pg_largeobject];
bool replace[Natts_pg_largeobject];
CatalogIndexState indstate;
Assert(PointerIsValid(obj_desc));
Assert(buf != NULL);
/* enforce writability because snapshot is probably wrong otherwise */
if ((obj_desc->flags & IFS_WRLOCK) == 0)
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("large object %u was not opened for writing",
obj_desc->id)));
if (nbytes <= 0)
return 0;
open_lo_relation();
indstate = CatalogOpenIndexes(lo_heap_r);
ScanKeyInit(&skey[0],
Anum_pg_largeobject_loid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(obj_desc->id));
ScanKeyInit(&skey[1],
Anum_pg_largeobject_pageno,
BTGreaterEqualStrategyNumber, F_INT4GE,
Int32GetDatum(pageno));
sd = index_beginscan(lo_heap_r, lo_index_r,
obj_desc->snapshot, 2, skey);
oldtuple = NULL;
olddata = NULL;
neednextpage = true;
while (nwritten < nbytes)
{
/*
* If possible, get next pre-existing page of the LO. We assume the
* indexscan will deliver these in order --- but there may be holes.
*/
if (neednextpage)
{
if ((oldtuple = index_getnext(sd, ForwardScanDirection)) != NULL)
{
if (HeapTupleHasNulls(oldtuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
olddata = (Form_pg_largeobject) GETSTRUCT(oldtuple);
Assert(olddata->pageno >= pageno);
}
neednextpage = false;
}
/*
* If we have a pre-existing page, see if it is the page we want to
* write, or a later one.
*/
if (olddata != NULL && olddata->pageno == pageno)
{
/*
* Update an existing page with fresh data.
*
* First, load old data into workbuf
*/
datafield = &(olddata->data); /* see note at top of file */
pfreeit = false;
if (VARATT_IS_EXTENDED(datafield))
{
datafield = (bytea *)
heap_tuple_untoast_attr((struct varlena *) datafield);
pfreeit = true;
}
len = getbytealen(datafield);
Assert(len <= LOBLKSIZE);
memcpy(workb, VARDATA(datafield), len);
if (pfreeit)
pfree(datafield);
/*
* Fill any hole
*/
off = (int) (obj_desc->offset % LOBLKSIZE);
if (off > len)
MemSet(workb + len, 0, off - len);
/*
* Insert appropriate portion of new data
*/
n = LOBLKSIZE - off;
n = (n <= (nbytes - nwritten)) ? n : (nbytes - nwritten);
memcpy(workb + off, buf + nwritten, n);
nwritten += n;
obj_desc->offset += n;
off += n;
/* compute valid length of new page */
len = (len >= off) ? len : off;
SET_VARSIZE(&workbuf.hdr, len + VARHDRSZ);
/*
* Form and insert updated tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, false, sizeof(nulls));
memset(replace, false, sizeof(replace));
values[Anum_pg_largeobject_data - 1] = PointerGetDatum(&workbuf);
replace[Anum_pg_largeobject_data - 1] = true;
newtup = heap_modify_tuple(oldtuple, RelationGetDescr(lo_heap_r),
values, nulls, replace);
simple_heap_update(lo_heap_r, &newtup->t_self, newtup);
CatalogIndexInsert(indstate, newtup);
heap_freetuple(newtup);
/*
* We're done with this old page.
*/
oldtuple = NULL;
olddata = NULL;
neednextpage = true;
}
else
{
/*
* Write a brand new page.
*
* First, fill any hole
*/
off = (int) (obj_desc->offset % LOBLKSIZE);
if (off > 0)
MemSet(workb, 0, off);
/*
* Insert appropriate portion of new data
*/
n = LOBLKSIZE - off;
n = (n <= (nbytes - nwritten)) ? n : (nbytes - nwritten);
memcpy(workb + off, buf + nwritten, n);
nwritten += n;
obj_desc->offset += n;
/* compute valid length of new page */
len = off + n;
SET_VARSIZE(&workbuf.hdr, len + VARHDRSZ);
/*
* Form and insert updated tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, false, sizeof(nulls));
values[Anum_pg_largeobject_loid - 1] = ObjectIdGetDatum(obj_desc->id);
values[Anum_pg_largeobject_pageno - 1] = Int32GetDatum(pageno);
values[Anum_pg_largeobject_data - 1] = PointerGetDatum(&workbuf);
newtup = heap_form_tuple(lo_heap_r->rd_att, values, nulls);
simple_heap_insert(lo_heap_r, newtup);
CatalogIndexInsert(indstate, newtup);
heap_freetuple(newtup);
}
pageno++;
}
index_endscan(sd);
CatalogCloseIndexes(indstate);
/*
* Advance command counter so that my tuple updates will be seen by later
* large-object operations in this transaction.
*/
CommandCounterIncrement();
return nwritten;
}
/* ----------
* toast_fetch_datum -
*
* Reconstruct an in memory Datum from the chunks saved
* in the toast relation
* ----------
*/
static struct varlena *
toast_fetch_datum(struct varlena *attr)
{
Relation toastrel;
Relation toastidx;
ScanKeyData toastkey;
IndexScanDesc toastscan;
HeapTuple ttup;
TupleDesc toasttupDesc;
varattrib *result;
int32 ressize;
int32 residx,
nextidx;
int32 numchunks;
Pointer chunk;
bool isnull;
int32 chunksize;
void *chunkdata;
ressize = ((varattrib *)attr)->va_external.va_extsize;
numchunks = ((ressize - 1) / TOAST_MAX_CHUNK_SIZE) + 1;
result = (varattrib *) palloc(ressize + VARHDRSZ);
SET_VARSIZE(result, ressize + VARHDRSZ);
if (VARATT_EXTERNAL_IS_COMPRESSED(attr))
VARATT_SET_COMPRESSED(result);
/*
* Open the toast relation and its index
*/
toastrel = heap_open(((varattrib *)attr)->va_external.va_toastrelid, AccessShareLock);
toasttupDesc = toastrel->rd_att;
toastidx = index_open(toastrel->rd_rel->reltoastidxid, AccessShareLock);
/*
* Setup a scan key to fetch from the index by va_valueid
*/
ScanKeyInit(&toastkey,
(AttrNumber) 1,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(((varattrib *)attr)->va_external.va_valueid));
/*
* Read the chunks by index
*
* Note that because the index is actually on (valueid, chunkidx) we will
* see the chunks in chunkidx order, even though we didn't explicitly ask
* for it.
*/
nextidx = 0;
toastscan = index_beginscan(toastrel, toastidx,
SnapshotToast, 1, &toastkey);
while ((ttup = index_getnext(toastscan, ForwardScanDirection)) != NULL)
{
/*
* Have a chunk, extract the sequence number and the data
*/
residx = DatumGetInt32(fastgetattr(ttup, 2, toasttupDesc, &isnull));
Assert(!isnull);
chunk = DatumGetPointer(fastgetattr(ttup, 3, toasttupDesc, &isnull));
Assert(!isnull);
if (VARATT_IS_SHORT(chunk))
{
chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
chunkdata = VARDATA_SHORT(chunk);
}
else if (!VARATT_IS_EXTENDED(chunk))
{
chunksize = VARSIZE(chunk) - VARHDRSZ;
chunkdata = VARDATA(chunk);
}
else
{
elog(ERROR, "found toasted toast chunk?");
chunksize = 0; /* shut compiler up */
chunkdata = NULL;
}
/*
* Some checks on the data we've found
*/
if (residx != nextidx)
elog(ERROR, "unexpected chunk number %d (expected %d) for toast value %u",
residx, nextidx,
((varattrib *)attr)->va_external.va_valueid);
if (residx < numchunks - 1)
{
if (chunksize != TOAST_MAX_CHUNK_SIZE)
elog(ERROR, "unexpected chunk size %d in chunk %d of %d for toast value %u (expected %d)",
chunksize, residx,
((varattrib *)attr)->va_external.va_valueid, numchunks-1,
(int)TOAST_MAX_CHUNK_SIZE);
}
else if (residx == numchunks-1)
{
if ((residx * TOAST_MAX_CHUNK_SIZE + chunksize) != ressize)
elog(ERROR, "unexpected chunk size %d in final chunk %d for toast value %u (expected %d)",
chunksize, residx,
((varattrib *)attr)->va_external.va_valueid,
ressize - residx*(int)TOAST_MAX_CHUNK_SIZE);
}
else
elog(ERROR, "unexpected chunk number %d for toast value %u (expected in %d..%d)",
residx,
((varattrib *)attr)->va_external.va_valueid,
0, numchunks-1);
/*
* Copy the data into proper place in our result
*/
memcpy(((char *) VARDATA(result)) + residx * TOAST_MAX_CHUNK_SIZE,
chunkdata,
chunksize);
nextidx++;
}
/*
* Final checks that we successfully fetched the datum
*/
if (nextidx != numchunks)
elog(ERROR, "missing chunk number %d for toast value %u",
nextidx,
((varattrib *)attr)->va_external.va_valueid);
/*
* End scan and close relations
*/
index_endscan(toastscan);
index_close(toastidx, AccessShareLock);
heap_close(toastrel, AccessShareLock);
return (struct varlena *)result;
}
void
inv_truncate(LargeObjectDesc *obj_desc, int len)
{
int32 pageno = (int32) (len / LOBLKSIZE);
int off;
ScanKeyData skey[2];
IndexScanDesc sd;
HeapTuple oldtuple;
Form_pg_largeobject olddata;
struct
{
bytea hdr;
char data[LOBLKSIZE]; /* make struct big enough */
int32 align_it; /* ensure struct is aligned well enough */
} workbuf;
char *workb = VARDATA(&workbuf.hdr);
HeapTuple newtup;
Datum values[Natts_pg_largeobject];
bool nulls[Natts_pg_largeobject];
bool replace[Natts_pg_largeobject];
CatalogIndexState indstate;
Assert(PointerIsValid(obj_desc));
/* enforce writability because snapshot is probably wrong otherwise */
if ((obj_desc->flags & IFS_WRLOCK) == 0)
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("large object %u was not opened for writing",
obj_desc->id)));
open_lo_relation();
indstate = CatalogOpenIndexes(lo_heap_r);
ScanKeyInit(&skey[0],
Anum_pg_largeobject_loid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(obj_desc->id));
ScanKeyInit(&skey[1],
Anum_pg_largeobject_pageno,
BTGreaterEqualStrategyNumber, F_INT4GE,
Int32GetDatum(pageno));
sd = index_beginscan(lo_heap_r, lo_index_r,
obj_desc->snapshot, 2, skey);
/*
* If possible, get the page the truncation point is in. The truncation
* point may be beyond the end of the LO or in a hole.
*/
olddata = NULL;
if ((oldtuple = index_getnext(sd, ForwardScanDirection)) != NULL)
{
if (HeapTupleHasNulls(oldtuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
olddata = (Form_pg_largeobject) GETSTRUCT(oldtuple);
Assert(olddata->pageno >= pageno);
}
/*
* If we found the page of the truncation point we need to truncate the
* data in it. Otherwise if we're in a hole, we need to create a page to
* mark the end of data.
*/
if (olddata != NULL && olddata->pageno == pageno)
{
/* First, load old data into workbuf */
bytea *datafield = &(olddata->data); /* see note at top of
* file */
bool pfreeit = false;
int pagelen;
if (VARATT_IS_EXTENDED(datafield))
{
datafield = (bytea *)
heap_tuple_untoast_attr((struct varlena *) datafield);
pfreeit = true;
}
pagelen = getbytealen(datafield);
Assert(pagelen <= LOBLKSIZE);
memcpy(workb, VARDATA(datafield), pagelen);
if (pfreeit)
pfree(datafield);
/*
* Fill any hole
*/
off = len % LOBLKSIZE;
if (off > pagelen)
MemSet(workb + pagelen, 0, off - pagelen);
/* compute length of new page */
SET_VARSIZE(&workbuf.hdr, off + VARHDRSZ);
/*
* Form and insert updated tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, false, sizeof(nulls));
memset(replace, false, sizeof(replace));
values[Anum_pg_largeobject_data - 1] = PointerGetDatum(&workbuf);
replace[Anum_pg_largeobject_data - 1] = true;
newtup = heap_modify_tuple(oldtuple, RelationGetDescr(lo_heap_r),
values, nulls, replace);
simple_heap_update(lo_heap_r, &newtup->t_self, newtup);
CatalogIndexInsert(indstate, newtup);
heap_freetuple(newtup);
}
else
{
/*
* If the first page we found was after the truncation point, we're in
* a hole that we'll fill, but we need to delete the later page.
*/
if (olddata != NULL && olddata->pageno > pageno)
simple_heap_delete(lo_heap_r, &oldtuple->t_self);
/*
* Write a brand new page.
*
* Fill the hole up to the truncation point
*/
off = len % LOBLKSIZE;
if (off > 0)
MemSet(workb, 0, off);
/* compute length of new page */
SET_VARSIZE(&workbuf.hdr, off + VARHDRSZ);
/*
* Form and insert new tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, false, sizeof(nulls));
values[Anum_pg_largeobject_loid - 1] = ObjectIdGetDatum(obj_desc->id);
values[Anum_pg_largeobject_pageno - 1] = Int32GetDatum(pageno);
values[Anum_pg_largeobject_data - 1] = PointerGetDatum(&workbuf);
newtup = heap_form_tuple(lo_heap_r->rd_att, values, nulls);
simple_heap_insert(lo_heap_r, newtup);
CatalogIndexInsert(indstate, newtup);
heap_freetuple(newtup);
}
/*
* Delete any pages after the truncation point
*/
while ((oldtuple = index_getnext(sd, ForwardScanDirection)) != NULL)
{
simple_heap_delete(lo_heap_r, &oldtuple->t_self);
}
index_endscan(sd);
CatalogCloseIndexes(indstate);
/*
* Advance command counter so that tuple updates will be seen by later
* large-object operations in this transaction.
*/
CommandCounterIncrement();
}
/*
* Search the relation 'rel' for tuple using the index.
*
* If a matching tuple is found, lock it with lockmode, fill the slot with its
* contents, and return true. Return false otherwise.
*/
bool
RelationFindReplTupleByIndex(Relation rel, Oid idxoid,
LockTupleMode lockmode,
TupleTableSlot *searchslot,
TupleTableSlot *outslot)
{
HeapTuple scantuple;
ScanKeyData skey[INDEX_MAX_KEYS];
IndexScanDesc scan;
SnapshotData snap;
TransactionId xwait;
Relation idxrel;
bool found;
/* Open the index. */
idxrel = index_open(idxoid, RowExclusiveLock);
/* Start an index scan. */
InitDirtySnapshot(snap);
scan = index_beginscan(rel, idxrel, &snap,
RelationGetNumberOfAttributes(idxrel),
0);
/* Build scan key. */
build_replindex_scan_key(skey, rel, idxrel, searchslot);
retry:
found = false;
index_rescan(scan, skey, RelationGetNumberOfAttributes(idxrel), NULL, 0);
/* Try to find the tuple */
if ((scantuple = index_getnext(scan, ForwardScanDirection)) != NULL)
{
found = true;
ExecStoreTuple(scantuple, outslot, InvalidBuffer, false);
ExecMaterializeSlot(outslot);
xwait = TransactionIdIsValid(snap.xmin) ?
snap.xmin : snap.xmax;
/*
* If the tuple is locked, wait for locking transaction to finish and
* retry.
*/
if (TransactionIdIsValid(xwait))
{
XactLockTableWait(xwait, NULL, NULL, XLTW_None);
goto retry;
}
}
/* Found tuple, try to lock it in the lockmode. */
if (found)
{
Buffer buf;
HeapUpdateFailureData hufd;
HTSU_Result res;
HeapTupleData locktup;
ItemPointerCopy(&outslot->tts_tuple->t_self, &locktup.t_self);
PushActiveSnapshot(GetLatestSnapshot());
res = heap_lock_tuple(rel, &locktup, GetCurrentCommandId(false),
lockmode,
LockWaitBlock,
false /* don't follow updates */ ,
&buf, &hufd);
/* the tuple slot already has the buffer pinned */
ReleaseBuffer(buf);
PopActiveSnapshot();
switch (res)
{
case HeapTupleMayBeUpdated:
break;
case HeapTupleUpdated:
/* XXX: Improve handling here */
ereport(LOG,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("concurrent update, retrying")));
goto retry;
case HeapTupleInvisible:
elog(ERROR, "attempted to lock invisible tuple");
default:
elog(ERROR, "unexpected heap_lock_tuple status: %u", res);
break;
}
}
index_endscan(scan);
/* Don't release lock until commit. */
index_close(idxrel, NoLock);
return found;
}
/*
* GetNewSequenceRelationOid
* Get a sequence relation Oid and verify it is valid against
* the pg_class relation by doing an index lookup. The caller
* should have a suitable lock on pg_class.
*/
Oid
GetNewSequenceRelationOid(Relation relation)
{
Oid newOid;
Oid oidIndex;
Relation indexrel;
SnapshotData SnapshotDirty;
IndexScanDesc scan;
ScanKeyData key;
bool collides;
RelFileNode rnode;
char *rpath;
int fd;
/* This should match RelationInitPhysicalAddr */
rnode.spcNode = relation->rd_rel->reltablespace ? relation->rd_rel->reltablespace : MyDatabaseTableSpace;
rnode.dbNode = relation->rd_rel->relisshared ? InvalidOid : MyDatabaseId;
/* We should only be using pg_class */
Assert(RelationGetRelid(relation) == RelationRelationId);
/* The relcache will cache the identity of the OID index for us */
oidIndex = RelationGetOidIndex(relation);
/* Otherwise, use the index to find a nonconflicting OID */
indexrel = index_open(oidIndex, AccessShareLock);
InitDirtySnapshot(SnapshotDirty);
/* Generate new sequence relation OIDs until we find one not in the table */
do
{
CHECK_FOR_INTERRUPTS();
newOid = GetNewSequenceRelationObjectId();
ScanKeyInit(&key,
(AttrNumber) 1,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(newOid));
/* see notes above about using SnapshotDirty */
scan = index_beginscan(relation, indexrel,
&SnapshotDirty, 1, &key);
collides = HeapTupleIsValid(index_getnext(scan, ForwardScanDirection));
index_endscan(scan);
if (!collides)
{
/* Check for existing file of same name */
rpath = relpath(rnode);
fd = BasicOpenFile(rpath, O_RDONLY | PG_BINARY, 0);
if (fd >= 0)
{
/* definite collision */
gp_retry_close(fd);
collides = true;
}
else
{
/*
* Here we have a little bit of a dilemma: if errno is something
* other than ENOENT, should we declare a collision and loop? In
* particular one might think this advisable for, say, EPERM.
* However there really shouldn't be any unreadable files in a
* tablespace directory, and if the EPERM is actually complaining
* that we can't read the directory itself, we'd be in an infinite
* loop. In practice it seems best to go ahead regardless of the
* errno. If there is a colliding file we will get an smgr
* failure when we attempt to create the new relation file.
*/
collides = false;
}
}
/*
* Also check that the OID hasn't been pre-assigned for a different
* relation.
*
* We're a bit sloppy between OIDs and relfilenodes here; it would be
* OK to use a value that's been reserved for use as a type or
* relation OID here, as long as the relfilenode is free. But there's
* no harm in skipping over those too, so we don't bother to
* distinguish them.
*/
if (!collides && !IsOidAcceptable(newOid))
collides = true;
} while (collides);
index_close(indexrel, AccessShareLock);
return newOid;
}
int
inv_write(LargeObjectDesc *obj_desc, char *buf, int nbytes)
{
int nwritten = 0;
int n;
int off;
int len;
int32 pageno = (int32) (obj_desc->offset / LOBLKSIZE);
ScanKeyData skey[2];
IndexScanDesc sd;
HeapTuple oldtuple;
Form_pg_largeobject olddata;
bool neednextpage;
bytea *datafield;
bool pfreeit;
struct
{
bytea hdr;
char data[LOBLKSIZE];
} workbuf;
char *workb = VARATT_DATA(&workbuf.hdr);
HeapTuple newtup;
Datum values[Natts_pg_largeobject];
char nulls[Natts_pg_largeobject];
char replace[Natts_pg_largeobject];
CatalogIndexState indstate;
Assert(PointerIsValid(obj_desc));
Assert(buf != NULL);
if (nbytes <= 0)
return 0;
open_lo_relation();
indstate = CatalogOpenIndexes(lo_heap_r);
ScanKeyInit(&skey[0],
Anum_pg_largeobject_loid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(obj_desc->id));
ScanKeyInit(&skey[1],
Anum_pg_largeobject_pageno,
BTGreaterEqualStrategyNumber, F_INT4GE,
Int32GetDatum(pageno));
sd = index_beginscan(lo_heap_r, lo_index_r,
SnapshotNow, 2, skey);
oldtuple = NULL;
olddata = NULL;
neednextpage = true;
while (nwritten < nbytes)
{
/*
* If possible, get next pre-existing page of the LO. We assume
* the indexscan will deliver these in order --- but there may be
* holes.
*/
if (neednextpage)
{
if ((oldtuple = index_getnext(sd, ForwardScanDirection)) != NULL)
{
olddata = (Form_pg_largeobject) GETSTRUCT(oldtuple);
Assert(olddata->pageno >= pageno);
}
neednextpage = false;
}
/*
* If we have a pre-existing page, see if it is the page we want
* to write, or a later one.
*/
if (olddata != NULL && olddata->pageno == pageno)
{
/*
* Update an existing page with fresh data.
*
* First, load old data into workbuf
*/
datafield = &(olddata->data);
pfreeit = false;
if (VARATT_IS_EXTENDED(datafield))
{
datafield = (bytea *)
heap_tuple_untoast_attr((varattrib *) datafield);
pfreeit = true;
}
len = getbytealen(datafield);
Assert(len <= LOBLKSIZE);
memcpy(workb, VARDATA(datafield), len);
if (pfreeit)
pfree(datafield);
/*
* Fill any hole
*/
off = (int) (obj_desc->offset % LOBLKSIZE);
if (off > len)
MemSet(workb + len, 0, off - len);
/*
* Insert appropriate portion of new data
*/
n = LOBLKSIZE - off;
n = (n <= (nbytes - nwritten)) ? n : (nbytes - nwritten);
memcpy(workb + off, buf + nwritten, n);
nwritten += n;
obj_desc->offset += n;
off += n;
/* compute valid length of new page */
len = (len >= off) ? len : off;
VARATT_SIZEP(&workbuf.hdr) = len + VARHDRSZ;
/*
* Form and insert updated tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, ' ', sizeof(nulls));
memset(replace, ' ', sizeof(replace));
values[Anum_pg_largeobject_data - 1] = PointerGetDatum(&workbuf);
replace[Anum_pg_largeobject_data - 1] = 'r';
newtup = heap_modifytuple(oldtuple, lo_heap_r,
values, nulls, replace);
simple_heap_update(lo_heap_r, &newtup->t_self, newtup);
CatalogIndexInsert(indstate, newtup);
heap_freetuple(newtup);
/*
* We're done with this old page.
*/
oldtuple = NULL;
olddata = NULL;
neednextpage = true;
}
else
{
/*
* Write a brand new page.
*
* First, fill any hole
*/
off = (int) (obj_desc->offset % LOBLKSIZE);
if (off > 0)
MemSet(workb, 0, off);
/*
* Insert appropriate portion of new data
*/
n = LOBLKSIZE - off;
n = (n <= (nbytes - nwritten)) ? n : (nbytes - nwritten);
memcpy(workb + off, buf + nwritten, n);
nwritten += n;
obj_desc->offset += n;
/* compute valid length of new page */
len = off + n;
VARATT_SIZEP(&workbuf.hdr) = len + VARHDRSZ;
/*
* Form and insert updated tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, ' ', sizeof(nulls));
values[Anum_pg_largeobject_loid - 1] = ObjectIdGetDatum(obj_desc->id);
values[Anum_pg_largeobject_pageno - 1] = Int32GetDatum(pageno);
values[Anum_pg_largeobject_data - 1] = PointerGetDatum(&workbuf);
newtup = heap_formtuple(lo_heap_r->rd_att, values, nulls);
simple_heap_insert(lo_heap_r, newtup);
CatalogIndexInsert(indstate, newtup);
heap_freetuple(newtup);
}
pageno++;
}
index_endscan(sd);
CatalogCloseIndexes(indstate);
/*
* Advance command counter so that my tuple updates will be seen by
* later large-object operations in this transaction.
*/
CommandCounterIncrement();
return nwritten;
}
int
inv_read(LargeObjectDesc *obj_desc, char *buf, int nbytes)
{
int nread = 0;
int n;
int off;
int len;
int32 pageno = (int32) (obj_desc->offset / LOBLKSIZE);
uint32 pageoff;
ScanKeyData skey[2];
IndexScanDesc sd;
HeapTuple tuple;
Assert(PointerIsValid(obj_desc));
Assert(buf != NULL);
if (nbytes <= 0)
return 0;
open_lo_relation();
ScanKeyInit(&skey[0],
Anum_pg_largeobject_loid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(obj_desc->id));
ScanKeyInit(&skey[1],
Anum_pg_largeobject_pageno,
BTGreaterEqualStrategyNumber, F_INT4GE,
Int32GetDatum(pageno));
sd = index_beginscan(lo_heap_r, lo_index_r,
SnapshotNow, 2, skey);
while ((tuple = index_getnext(sd, ForwardScanDirection)) != NULL)
{
Form_pg_largeobject data;
bytea *datafield;
bool pfreeit;
data = (Form_pg_largeobject) GETSTRUCT(tuple);
/*
* We assume the indexscan will deliver pages in order. However,
* there may be missing pages if the LO contains unwritten
* "holes". We want missing sections to read out as zeroes.
*/
pageoff = ((uint32) data->pageno) * LOBLKSIZE;
if (pageoff > obj_desc->offset)
{
n = pageoff - obj_desc->offset;
n = (n <= (nbytes - nread)) ? n : (nbytes - nread);
MemSet(buf + nread, 0, n);
nread += n;
obj_desc->offset += n;
}
if (nread < nbytes)
{
Assert(obj_desc->offset >= pageoff);
off = (int) (obj_desc->offset - pageoff);
Assert(off >= 0 && off < LOBLKSIZE);
datafield = &(data->data);
pfreeit = false;
if (VARATT_IS_EXTENDED(datafield))
{
datafield = (bytea *)
heap_tuple_untoast_attr((varattrib *) datafield);
pfreeit = true;
}
len = getbytealen(datafield);
if (len > off)
{
n = len - off;
n = (n <= (nbytes - nread)) ? n : (nbytes - nread);
memcpy(buf + nread, VARDATA(datafield) + off, n);
nread += n;
obj_desc->offset += n;
}
if (pfreeit)
pfree(datafield);
}
if (nread >= nbytes)
break;
}
index_endscan(sd);
return nread;
}
/* ----------------------------------------------------------------
* IndexNext
*
* Retrieve a tuple from the IndexScan node's currentRelation
* using the index specified in the IndexScanState information.
* ----------------------------------------------------------------
*/
static TupleTableSlot *
IndexNext(IndexScanState *node)
{
EState *estate;
ExprContext *econtext;
ScanDirection direction;
IndexScanDesc scandesc;
HeapTuple tuple;
TupleTableSlot *slot;
/*
* extract necessary information from index scan node
*/
estate = node->ss.ps.state;
direction = estate->es_direction;
/* flip direction if this is an overall backward scan */
if (ScanDirectionIsBackward(((IndexScan *) node->ss.ps.plan)->indexorderdir))
{
if (ScanDirectionIsForward(direction))
direction = BackwardScanDirection;
else if (ScanDirectionIsBackward(direction))
direction = ForwardScanDirection;
}
scandesc = node->iss_ScanDesc;
econtext = node->ss.ps.ps_ExprContext;
slot = node->ss.ss_ScanTupleSlot;
/*
* ok, now that we have what we need, fetch the next tuple.
*/
while ((tuple = index_getnext(scandesc, direction)) != NULL)
{
/*
* Store the scanned tuple in the scan tuple slot of the scan state.
* Note: we pass 'false' because tuples returned by amgetnext are
* pointers onto disk pages and must not be pfree()'d.
*/
ExecStoreTuple(tuple, /* tuple to store */
slot, /* slot to store in */
scandesc->xs_cbuf, /* buffer containing tuple */
false); /* don't pfree */
/*
* If the index was lossy, we have to recheck the index quals using
* the real tuple.
*/
if (scandesc->xs_recheck)
{
econtext->ecxt_scantuple = slot;
ResetExprContext(econtext);
if (!ExecQual(node->indexqualorig, econtext, false))
continue; /* nope, so ask index for another one */
}
return slot;
}
/*
* if we get here it means the index scan failed so we are at the end of
* the scan..
*/
return ExecClearTuple(slot);
}
/* ----------------------------------------------------------------
* IndexNext
*
* Retrieve a tuple from the index_scan_sc node's currentRelation
* using the index specified in the index_ss information.
* ----------------------------------------------------------------
*/
static struct tupslot*
IndexNext(index_ss *node)
{
exec_state_n* estate;
expr_ctx_n* econtext;
enum scandir direction;
struct index_scan* scandesc;
struct heap_tuple* tuple;
struct tupslot* slot;
/*
* extract necessary information from index scan node
*/
estate = node->ss.ps.state;
direction = estate->es_direction;
/* flip direction if this is an overall backward scan */
if (SCANDIR_BACKWARD(((index_scan_sc *) node->ss.ps.plan)->indexorderdir)) {
if (SCANDIR_FORWARD(direction))
direction = BACKWARD_SCANDIR;
else if (SCANDIR_BACKWARD(direction))
direction = FORWARD_SCANDIR;
}
scandesc = node->iss_ScanDesc;
econtext = node->ss.ps.ps_ExprContext;
slot = node->ss.ss_ScanTupleSlot;
/*
* ok, now that we have what we need, fetch the next tuple.
*/
while ((tuple = index_getnext(scandesc, direction)) != NULL) {
/*
* Store the scanned tuple in the scan tuple slot of the scan state.
* Note: we pass 'false' because tuples returned by amgetnext are
* pointers onto disk pages and must not be pfree()'d.
*/
exec_store_tuple(tuple, /* tuple to store */
slot, /* slot to store in */
scandesc->xs_cbuf, /* buffer containing tuple */
false); /* don't pfree */
/*
* If the index was lossy, we have to recheck the index quals using
* the real tuple.
*/
if (scandesc->xs_recheck) {
econtext->ecxt_scantuple = slot;
ResetExprContext(econtext);
if (!exec_qual(node->indexqualorig, econtext, false))
continue; /* nope, so ask index for another one */
}
return slot;
}
/*
* if we get here it means the index scan failed so we are at the end of
* the scan..
*/
return exec_clear_tuple(slot);
}
