/* ----------
* heap_tuple_untoast_attr -
*
* Public entry point to get back a toasted value from compression
* or external storage.
* ----------
*/
varattrib *
heap_tuple_untoast_attr(varattrib *attr)
{
varattrib *result;
if (VARATT_IS_EXTERNAL(attr))
{
if (VARATT_IS_COMPRESSED(attr))
{
/* ----------
* This is an external stored compressed value
* Fetch it from the toast heap and decompress.
* ----------
*/
varattrib *tmp;
tmp = toast_fetch_datum(attr);
result = (varattrib *) palloc(attr->va_content.va_external.va_rawsize
+ VARHDRSZ);
VARATT_SIZEP(result) = attr->va_content.va_external.va_rawsize
+ VARHDRSZ;
pglz_decompress((PGLZ_Header *) tmp, VARATT_DATA(result));
pfree(tmp);
}
else
{
/*
* This is an external stored plain value
*/
result = toast_fetch_datum(attr);
}
}
else if (VARATT_IS_COMPRESSED(attr))
{
/*
* This is a compressed value inside of the main tuple
*/
result = (varattrib *) palloc(attr->va_content.va_compressed.va_rawsize
+ VARHDRSZ);
VARATT_SIZEP(result) = attr->va_content.va_compressed.va_rawsize
+ VARHDRSZ;
pglz_decompress((PGLZ_Header *) attr, VARATT_DATA(result));
}
else
/*
* This is a plain value inside of the main tuple - why am I
* called?
*/
return attr;
return result;
}
/* ----------
* heap_tuple_untoast_attr_slice -
*
* Public entry point to get back part of a toasted value
* from compression or external storage.
* ----------
*/
varattrib *
heap_tuple_untoast_attr_slice(varattrib *attr, int32 sliceoffset, int32 slicelength)
{
varattrib *preslice;
varattrib *result;
int32 attrsize;
if (VARATT_IS_COMPRESSED(attr))
{
varattrib *tmp;
if (VARATT_IS_EXTERNAL(attr))
tmp = toast_fetch_datum(attr);
else
{
tmp = attr; /* compressed in main tuple */
}
preslice = (varattrib *) palloc(attr->va_content.va_external.va_rawsize
+ VARHDRSZ);
VARATT_SIZEP(preslice) = attr->va_content.va_external.va_rawsize + VARHDRSZ;
pglz_decompress((PGLZ_Header *) tmp, VARATT_DATA(preslice));
if (tmp != attr)
pfree(tmp);
}
else
{
/* Plain value */
if (VARATT_IS_EXTERNAL(attr))
{
/* fast path */
return (toast_fetch_datum_slice(attr, sliceoffset, slicelength));
}
else
preslice = attr;
}
/* slicing of datum for compressed cases and plain value */
attrsize = VARSIZE(preslice) - VARHDRSZ;
if (sliceoffset >= attrsize)
{
sliceoffset = 0;
slicelength = 0;
}
if (((sliceoffset + slicelength) > attrsize) || slicelength < 0)
slicelength = attrsize - sliceoffset;
result = (varattrib *) palloc(slicelength + VARHDRSZ);
VARATT_SIZEP(result) = slicelength + VARHDRSZ;
memcpy(VARDATA(result), VARDATA(preslice) + sliceoffset, slicelength);
if (preslice != attr)
pfree(preslice);
return result;
}
/* ----------
* toast_compress_datum -
*
* Create a compressed version of a varlena datum
*
* If we fail (ie, compressed result is actually bigger than original)
* then return NULL. We must not use compressed data if it'd expand
* the tuple!
* ----------
*/
Datum
toast_compress_datum(Datum value)
{
varattrib *tmp;
tmp = (varattrib *) palloc(sizeof(PGLZ_Header) + VARATT_SIZE(value));
pglz_compress(VARATT_DATA(value), VARATT_SIZE(value) - VARHDRSZ,
(PGLZ_Header *) tmp,
PGLZ_strategy_default);
if (VARATT_SIZE(tmp) < VARATT_SIZE(value))
{
/* successful compression */
VARATT_SIZEP(tmp) |= VARATT_FLAG_COMPRESSED;
return PointerGetDatum(tmp);
}
else
{
/* incompressible data */
pfree(tmp);
return PointerGetDatum(NULL);
}
}
/* ----------
* toast_save_datum -
*
* Save one single datum into the secondary relation and return
* a varattrib reference for it.
* ----------
*/
static Datum
toast_save_datum(Relation rel, Datum value)
{
Relation toastrel;
Relation toastidx;
HeapTuple toasttup;
InsertIndexResult idxres;
TupleDesc toasttupDesc;
Datum t_values[3];
char t_nulls[3];
varattrib *result;
struct
{
struct varlena hdr;
char data[TOAST_MAX_CHUNK_SIZE];
} chunk_data;
int32 chunk_size;
int32 chunk_seq = 0;
char *data_p;
int32 data_todo;
/*
* Create the varattrib reference
*/
result = (varattrib *) palloc(sizeof(varattrib));
result->va_header = sizeof(varattrib) | VARATT_FLAG_EXTERNAL;
if (VARATT_IS_COMPRESSED(value))
{
result->va_header |= VARATT_FLAG_COMPRESSED;
result->va_content.va_external.va_rawsize =
((varattrib *) value)->va_content.va_compressed.va_rawsize;
}
else
result->va_content.va_external.va_rawsize = VARATT_SIZE(value);
result->va_content.va_external.va_extsize =
VARATT_SIZE(value) - VARHDRSZ;
result->va_content.va_external.va_valueid = newoid();
result->va_content.va_external.va_toastrelid =
rel->rd_rel->reltoastrelid;
/*
* Initialize constant parts of the tuple data
*/
t_values[0] = ObjectIdGetDatum(result->va_content.va_external.va_valueid);
t_values[2] = PointerGetDatum(&chunk_data);
t_nulls[0] = ' ';
t_nulls[1] = ' ';
t_nulls[2] = ' ';
/*
* Get the data to process
*/
data_p = VARATT_DATA(value);
data_todo = VARATT_SIZE(value) - VARHDRSZ;
/*
* Open the toast relation
*/
toastrel = heap_open(rel->rd_rel->reltoastrelid, RowExclusiveLock);
toasttupDesc = toastrel->rd_att;
toastidx = index_open(toastrel->rd_rel->reltoastidxid);
/*
* Split up the item into chunks
*/
while (data_todo > 0)
{
/*
* Calculate the size of this chunk
*/
chunk_size = Min(TOAST_MAX_CHUNK_SIZE, data_todo);
/*
* Build a tuple and store it
*/
t_values[1] = Int32GetDatum(chunk_seq++);
VARATT_SIZEP(&chunk_data) = chunk_size + VARHDRSZ;
memcpy(VARATT_DATA(&chunk_data), data_p, chunk_size);
toasttup = heap_formtuple(toasttupDesc, t_values, t_nulls);
if (!HeapTupleIsValid(toasttup))
elog(ERROR, "failed to build TOAST tuple");
simple_heap_insert(toastrel, toasttup);
/*
* Create the index entry. We cheat a little here by not using
* FormIndexDatum: this relies on the knowledge that the index
* columns are the same as the initial columns of the table.
*
* Note also that there had better not be any user-created index on
* the TOAST table, since we don't bother to update anything else.
*/
idxres = index_insert(toastidx, t_values, t_nulls,
&(toasttup->t_self),
toastrel, toastidx->rd_index->indisunique);
if (idxres == NULL)
elog(ERROR, "failed to insert index entry for TOAST tuple");
/*
* Free memory
*/
pfree(idxres);
heap_freetuple(toasttup);
/*
* Move on to next chunk
*/
data_todo -= chunk_size;
data_p += chunk_size;
}
/*
* Done - close toast relation and return the reference
*/
index_close(toastidx);
heap_close(toastrel, RowExclusiveLock);
return PointerGetDatum(result);
}
/* ----------
* 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;
}
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;
}