/*
* Determine size of a large object
*
* NOTE: LOs can contain gaps, just like Unix files. We actually return
* the offset of the last byte + 1.
*/
static uint32
inv_getsize(LargeObjectDesc *obj_desc)
{
bool found = false;
uint32 lastbyte = 0;
ScanKeyData skey[1];
IndexScanDesc sd;
HeapTuple tuple;
Assert(PointerIsValid(obj_desc));
open_lo_relation();
ScanKeyInit(&skey[0],
Anum_pg_largeobject_loid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(obj_desc->id));
sd = index_beginscan(lo_heap_r, lo_index_r,
obj_desc->snapshot, 1, skey);
/*
* Because the pg_largeobject index is on both loid and pageno, but we
* constrain only loid, a backwards scan should visit all pages of the
* large object in reverse pageno order. So, it's sufficient to examine
* the first valid tuple (== last valid page).
*/
while ((tuple = index_getnext(sd, BackwardScanDirection)) != NULL)
{
Form_pg_largeobject data;
bytea *datafield;
bool pfreeit;
found = true;
if (HeapTupleHasNulls(tuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
data = (Form_pg_largeobject) GETSTRUCT(tuple);
datafield = &(data->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;
}
lastbyte = data->pageno * LOBLKSIZE + getbytealen(datafield);
if (pfreeit)
pfree(datafield);
break;
}
index_endscan(sd);
if (!found)
ereport(ERROR,
(errcode(ERRCODE_UNDEFINED_OBJECT),
errmsg("large object %u does not exist", obj_desc->id)));
return lastbyte;
}
/*
* Determine size of a large object
*
* NOTE: LOs can contain gaps, just like Unix files. We actually return
* the offset of the last byte + 1.
*/
static uint64
inv_getsize(LargeObjectDesc *obj_desc)
{
uint64 lastbyte = 0;
ScanKeyData skey[1];
SysScanDesc sd;
HeapTuple tuple;
Assert(PointerIsValid(obj_desc));
open_lo_relation();
ScanKeyInit(&skey[0],
Anum_pg_largeobject_loid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(obj_desc->id));
sd = systable_beginscan_ordered(lo_heap_r, lo_index_r,
obj_desc->snapshot, 1, skey);
/*
* Because the pg_largeobject index is on both loid and pageno, but we
* constrain only loid, a backwards scan should visit all pages of the
* large object in reverse pageno order. So, it's sufficient to examine
* the first valid tuple (== last valid page).
*/
tuple = systable_getnext_ordered(sd, BackwardScanDirection);
if (HeapTupleIsValid(tuple))
{
Form_pg_largeobject data;
bytea *datafield;
bool pfreeit;
if (HeapTupleHasNulls(tuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
data = (Form_pg_largeobject) GETSTRUCT(tuple);
datafield = &(data->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;
}
lastbyte = (uint64) data->pageno * LOBLKSIZE + getbytealen(datafield);
if (pfreeit)
pfree(datafield);
}
systable_endscan_ordered(sd);
return lastbyte;
}
/*
* Determine size of a large object
*
* NOTE: LOs can contain gaps, just like Unix files. We actually return
* the offset of the last byte + 1.
*/
static uint32
inv_getsize(struct lobj *obj_desc)
{
uint32 lastbyte = 0;
struct scankey skey[1];
struct sys_scan* sd;
struct heap_tuple* tuple;
ASSERT(PTR_VALID(obj_desc));
open_lo_relation();
scankey_init(&skey[0], Anum_pg_largeobject_loid, BT_EQ_STRAT_NR, F_OIDEQ, OID_TO_D(obj_desc->id));
sd = systable_beginscan_ordered(lo_heap_r, lo_index_r, obj_desc->snapshot, 1, skey);
/*
* Because the pg_largeobject index is on both loid and pageno, but we
* constrain only loid, a backwards scan should visit all pages of the
* large object in reverse pageno order. So, it's sufficient to examine
* the first valid tuple (== last valid page).
*/
tuple = systable_getnext_ordered(sd, BACKWARD_SCANDIR);
if (HT_VALID(tuple)) {
Form_pg_largeobject data;
bytea* datafield;
bool pfreeit;
if (HT_HAS_NULLS(tuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
data = (Form_pg_largeobject) GET_STRUCT(tuple);
datafield = &(data->data); /* see note at top of file */
pfreeit = false;
if (VLA_EXTENDED(datafield)) {
datafield = (bytea *) heap_tuple_untoast_attr((struct vla *) datafield);
pfreeit = true;
}
lastbyte = data->pageno * LO_BLK_SIZE + getbytealen(datafield);
if (pfreeit)
pfree(datafield);
}
systable_endscan_ordered(sd);
return lastbyte;
}
void
inv_truncate(LargeObjectDesc *obj_desc, int len)
{
int32 pageno = (int32) (len / LOBLKSIZE);
int off;
ScanKeyData skey[2];
SysScanDesc 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)));
/* check existence of the target largeobject */
if (!LargeObjectExists(obj_desc->id))
ereport(ERROR,
(errcode(ERRCODE_UNDEFINED_OBJECT),
errmsg("large object %u was already dropped", obj_desc->id)));
open_lo_relation();
indstate = CatalogOpenIndexes(lo_heap_r);
/*
* Set up to find all pages with desired loid and pageno >= target
*/
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 = systable_beginscan_ordered(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 = systable_getnext_ordered(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
* because the loop below won't visit it again.
*/
if (olddata != NULL)
{
Assert(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. If the initial search
* didn't find a page, then of course there's nothing more to do.
*/
if (olddata != NULL)
{
while ((oldtuple = systable_getnext_ordered(sd, ForwardScanDirection)) != NULL)
{
simple_heap_delete(lo_heap_r, &oldtuple->t_self);
}
}
systable_endscan_ordered(sd);
CatalogCloseIndexes(indstate);
/*
* Advance command counter so that tuple updates will be seen by later
* large-object operations in this transaction.
*/
CommandCounterIncrement();
}
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];
SysScanDesc 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)));
/* check existence of the target largeobject */
if (!LargeObjectExists(obj_desc->id))
ereport(ERROR,
(errcode(ERRCODE_UNDEFINED_OBJECT),
errmsg("large object %u was already dropped", 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 = systable_beginscan_ordered(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 expect the
* indexscan will deliver these in order --- but there may be holes.
*/
if (neednextpage)
{
if ((oldtuple = systable_getnext_ordered(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++;
}
systable_endscan_ordered(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];
SysScanDesc 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 = systable_beginscan_ordered(lo_heap_r, lo_index_r,
obj_desc->snapshot, 2, skey);
while ((tuple = systable_getnext_ordered(sd, ForwardScanDirection)) != NULL)
{
Form_pg_largeobject data;
bytea *datafield;
bool pfreeit;
if (HeapTupleHasNulls(tuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
data = (Form_pg_largeobject) GETSTRUCT(tuple);
/*
* We expect 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); /* 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);
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;
}
systable_endscan_ordered(sd);
return nread;
}
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;
}
void
inv_truncate(struct lobj *obj_desc, int len)
{
int32 pageno = (int32) (len / LO_BLK_SIZE);
int off;
struct scankey skey[2];
struct sys_scan* sd;
struct heap_tuple* oldtuple;
Form_pg_largeobject olddata;
struct {
bytea hdr;
char data[LO_BLK_SIZE]; /* make struct big enough */
int32 align_it;/* ensure struct is aligned well enough */
} workbuf;
char* workb = VLA_DATA(&workbuf.hdr);
struct heap_tuple* newtup;
datum_t values[Natts_pg_largeobject];
bool nulls[Natts_pg_largeobject];
bool replace[Natts_pg_largeobject];
CatalogIndexState indstate;
ASSERT(PTR_VALID(obj_desc));
/* enforce writability because snapshot is probably wrong otherwise */
if ((obj_desc->flags & IFS_WRLOCK) == 0)
ereport(ERROR, (
errcode(E_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("large object %u was not opened for writing",
obj_desc->id)));
/* check existence of the target largeobject */
if (!large_obj_exists(obj_desc->id))
ereport(ERROR, (
errcode(E_UNDEFINED_OBJECT),
errmsg("large object %u was already dropped", obj_desc->id)));
open_lo_relation();
indstate = cat_open_indexes(lo_heap_r);
/*
* Set up to find all pages with desired loid and pageno >= target
*/
scankey_init(&skey[0], Anum_pg_largeobject_loid, BT_EQ_STRAT_NR, F_OIDEQ, OID_TO_D(obj_desc->id));
scankey_init(&skey[1], Anum_pg_largeobject_pageno, BT_GE_STRAT_NR, F_INT4GE, INT32_TO_D(pageno));
sd = systable_beginscan_ordered(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 = systable_getnext_ordered(sd, FORWARD_SCANDIR)) != NULL) {
if (HT_HAS_NULLS(oldtuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
olddata = (Form_pg_largeobject) GET_STRUCT(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 (VLA_EXTENDED(datafield)) {
datafield = (bytea *) heap_tuple_untoast_attr((struct vla *) datafield);
pfreeit = true;
}
pagelen = getbytealen(datafield);
ASSERT(pagelen <= LO_BLK_SIZE);
memcpy(workb, VLA_DATA(datafield), pagelen);
if (pfreeit)
pfree(datafield);
/*
* Fill any hole
*/
off = len % LO_BLK_SIZE;
if (off > pagelen)
pg_memset(workb + pagelen, 0, off - pagelen);
/* compute length of new page */
VLA_SET_SZ_STND(&workbuf.hdr, off + VAR_HDR_SZ);
/*
* 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] = PTR_TO_D(&workbuf);
replace[Anum_pg_largeobject_data - 1] = true;
newtup = heap_modify_tuple(oldtuple, REL_DESC(lo_heap_r), values, nulls, replace);
simple_heap_update(lo_heap_r, &newtup->t_self, newtup);
cat_index_insert(indstate, newtup);
heap_free_tuple(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
* because the loop below won't visit it again.
*/
if (olddata != NULL) {
ASSERT(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 % LO_BLK_SIZE;
if (off > 0)
pg_memset(workb, 0, off);
/* compute length of new page */
VLA_SET_SZ_STND(&workbuf.hdr, off + VAR_HDR_SZ);
/*
* Form and insert new tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, false, sizeof(nulls));
values[Anum_pg_largeobject_loid - 1] = OID_TO_D(obj_desc->id);
values[Anum_pg_largeobject_pageno - 1] = INT32_TO_D(pageno);
values[Anum_pg_largeobject_data - 1] = PTR_TO_D(&workbuf);
newtup = heap_form_tuple(lo_heap_r->rd_att, values, nulls);
simple_heap_insert(lo_heap_r, newtup);
cat_index_insert(indstate, newtup);
heap_free_tuple(newtup);
}
/*
* Delete any pages after the truncation point. If the initial search
* didn't find a page, then of course there's nothing more to do.
*/
if (olddata != NULL) {
while ((oldtuple = systable_getnext_ordered(sd, FORWARD_SCANDIR)) != NULL) {
simple_heap_delete(lo_heap_r, &oldtuple->t_self);
}
}
systable_endscan_ordered(sd);
cat_close_indexes(indstate);
/*
* Advance command counter so that tuple updates will be seen by later
* large-object operations in this transaction.
*/
cmd_count_incr();
}
int
inv_write(struct lobj *obj_desc, const char *buf, int nbytes)
{
int nwritten = 0;
int n;
int off;
int len;
int32 pageno = (int32) (obj_desc->offset / LO_BLK_SIZE);
struct scankey skey[2];
struct sys_scan * sd;
struct heap_tuple * oldtuple;
Form_pg_largeobject olddata;
bool neednextpage;
bytea* datafield;
bool pfreeit;
struct {
bytea hdr;
char data[LO_BLK_SIZE]; /* make struct big enough */
int32 align_it; /* ensure struct is aligned well enough */
} workbuf;
char* workb = VLA_DATA(&workbuf.hdr);
struct heap_tuple* newtup;
datum_t values[Natts_pg_largeobject];
bool nulls[Natts_pg_largeobject];
bool replace[Natts_pg_largeobject];
CatalogIndexState indstate;
ASSERT(PTR_VALID(obj_desc));
ASSERT(buf != NULL);
/* enforce writability because snapshot is probably wrong otherwise */
if ((obj_desc->flags & IFS_WRLOCK) == 0)
ereport(ERROR, (
errcode(E_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("large object %u was not opened for writing",
obj_desc->id)));
/* check existence of the target largeobject */
if (!large_obj_exists(obj_desc->id))
ereport(ERROR, (
errcode(E_UNDEFINED_OBJECT),
errmsg("large object %u was already dropped", obj_desc->id)));
if (nbytes <= 0)
return 0;
open_lo_relation();
indstate = cat_open_indexes(lo_heap_r);
scankey_init(&skey[0], Anum_pg_largeobject_loid, BT_EQ_STRAT_NR, F_OIDEQ, OID_TO_D(obj_desc->id));
scankey_init(&skey[1], Anum_pg_largeobject_pageno, BT_GE_STRAT_NR, F_INT4GE, INT32_TO_D(pageno));
sd = systable_beginscan_ordered(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 expect the
* indexscan will deliver these in order --- but there may be holes.
*/
if (neednextpage) {
if ((oldtuple = systable_getnext_ordered(sd, FORWARD_SCANDIR)) != NULL) {
if (HT_HAS_NULLS(oldtuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
olddata = (Form_pg_largeobject) GET_STRUCT(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 (VLA_EXTENDED(datafield)) {
datafield = (bytea *)
heap_tuple_untoast_attr((struct vla *) datafield);
pfreeit = true;
}
len = getbytealen(datafield);
ASSERT(len <= LO_BLK_SIZE);
memcpy(workb, VLA_DATA(datafield), len);
if (pfreeit)
pfree(datafield);
/*
* Fill any hole
*/
off = (int)(obj_desc->offset % LO_BLK_SIZE);
if (off > len)
pg_memset(workb + len, 0, off - len);
/*
* Insert appropriate portion of new data
*/
n = LO_BLK_SIZE - 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;
VLA_SET_SZ_STND(&workbuf.hdr, len + VAR_HDR_SZ);
/*
* 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] = PTR_TO_D(&workbuf);
replace[Anum_pg_largeobject_data - 1] = true;
newtup = heap_modify_tuple(oldtuple, REL_DESC(lo_heap_r), values, nulls, replace);
simple_heap_update(lo_heap_r, &newtup->t_self, newtup);
cat_index_insert(indstate, newtup);
heap_free_tuple(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 % LO_BLK_SIZE);
if (off > 0)
pg_memset(workb, 0, off);
/*
* Insert appropriate portion of new data
*/
n = LO_BLK_SIZE - 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;
VLA_SET_SZ_STND(&workbuf.hdr, len + VAR_HDR_SZ);
/*
* Form and insert updated tuple
*/
memset(values, 0, sizeof(values));
memset(nulls, false, sizeof(nulls));
values[Anum_pg_largeobject_loid - 1] = OID_TO_D(obj_desc->id);
values[Anum_pg_largeobject_pageno - 1] = INT32_TO_D(pageno);
values[Anum_pg_largeobject_data - 1] = PTR_TO_D(&workbuf);
newtup = heap_form_tuple(lo_heap_r->rd_att, values, nulls);
simple_heap_insert(lo_heap_r, newtup);
cat_index_insert(indstate, newtup);
heap_free_tuple(newtup);
}
pageno++;
}
systable_endscan_ordered(sd);
cat_close_indexes(indstate);
/*
* Advance command counter so that my tuple updates will be seen by
* later large-object operations in this transaction.
*/
cmd_count_incr();
return nwritten;
}
int
inv_read(struct lobj *obj_desc, char *buf, int nbytes)
{
int nread = 0;
int n;
int off;
int len;
int32 pageno = (int32) (obj_desc->offset / LO_BLK_SIZE);
uint32 pageoff;
struct scankey skey[2];
struct sys_scan* sd;
struct heap_tuple* tuple;
ASSERT(PTR_VALID(obj_desc));
ASSERT(buf != NULL);
if (nbytes <= 0)
return 0;
open_lo_relation();
scankey_init(&skey[0], Anum_pg_largeobject_loid, BT_EQ_STRAT_NR, F_OIDEQ, OID_TO_D(obj_desc->id));
scankey_init(&skey[1], Anum_pg_largeobject_pageno, BT_GE_STRAT_NR, F_INT4GE, INT32_TO_D(pageno));
sd = systable_beginscan_ordered(lo_heap_r, lo_index_r, obj_desc->snapshot, 2, skey);
while ((tuple = systable_getnext_ordered(sd, FORWARD_SCANDIR)) != NULL) {
Form_pg_largeobject data;
bytea* datafield;
bool pfreeit;
if (HT_HAS_NULLS(tuple)) /* paranoia */
elog(ERROR, "null field found in pg_largeobject");
data = (Form_pg_largeobject) GET_STRUCT(tuple);
/*
* We expect 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) * LO_BLK_SIZE;
if (pageoff > obj_desc->offset) {
n = pageoff - obj_desc->offset;
n = (n <= (nbytes - nread)) ? n : (nbytes - nread);
pg_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 < LO_BLK_SIZE);
datafield = &(data->data); /* see note at top of file */
pfreeit = false;
if (VLA_EXTENDED(datafield)) {
datafield = (bytea *) heap_tuple_untoast_attr((struct vla *) datafield);
pfreeit = true;
}
len = getbytealen(datafield);
if (len > off) {
n = len - off;
n = (n <= (nbytes - nread)) ? n : (nbytes - nread);
memcpy(buf + nread, VLA_DATA(datafield) + off, n);
nread += n;
obj_desc->offset += n;
}
if (pfreeit)
pfree(datafield);
}
if (nread >= nbytes)
break;
}
systable_endscan_ordered(sd);
return nread;
}