/*
* __curjoin_endpoint_init_key --
* Set the key in the reference endpoint.
*/
static int
__curjoin_endpoint_init_key(WT_SESSION_IMPL *session,
WT_CURSOR_JOIN_ENTRY *entry, WT_CURSOR_JOIN_ENDPOINT *endpoint)
{
WT_CURSOR *cursor;
WT_CURSOR_INDEX *cindex;
WT_ITEM *k;
uint64_t r;
if ((cursor = endpoint->cursor) != NULL) {
if (entry->index != NULL) {
/* Extract and save the index's logical key. */
cindex = (WT_CURSOR_INDEX *)endpoint->cursor;
WT_RET(__wt_struct_repack(session,
cindex->child->key_format,
(entry->repack_format != NULL ?
entry->repack_format : cindex->iface.key_format),
&cindex->child->key, &endpoint->key));
} else {
k = &((WT_CURSOR_TABLE *)cursor)->cg_cursors[0]->key;
if (WT_CURSOR_RECNO(cursor)) {
r = *(uint64_t *)k->data;
WT_RET(__curjoin_pack_recno(session, r,
endpoint->recno_buf,
sizeof(endpoint->recno_buf),
&endpoint->key));
} else
endpoint->key = *k;
}
}
return (0);
}
/*
* __curdump_get_key --
* WT_CURSOR->get_key for dump cursors.
*/
static int
__curdump_get_key(WT_CURSOR *cursor, ...)
{
WT_CURSOR *child;
WT_CURSOR_DUMP *cdump;
WT_DECL_RET;
WT_ITEM item, *itemp;
WT_SESSION_IMPL *session;
uint64_t recno;
va_list ap;
cdump = (WT_CURSOR_DUMP *)cursor;
child = cdump->child;
CURSOR_API_CALL(cursor, session, get_key, NULL);
if (WT_CURSOR_RECNO(cursor) && !F_ISSET(cursor, WT_CURSTD_RAW)) {
WT_ERR(child->get_key(child, &recno));
WT_ERR(__wt_buf_fmt(session, &cursor->key, "%" PRIu64, recno));
} else {
WT_ERR(child->get_key(child, &item));
WT_ERR(__raw_to_dump(session, &item,
&cursor->key, F_ISSET(cursor, WT_CURSTD_DUMP_HEX) ? 1 : 0));
}
va_start(ap, cursor);
if (F_ISSET(cursor, WT_CURSTD_RAW)) {
itemp = va_arg(ap, WT_ITEM *);
itemp->data = cursor->key.data;
itemp->size = cursor->key.size;
} else
/*
* __curjoin_split_key --
* Copy the primary key from a cursor (either main table or index)
* to another cursor. When copying from an index file, the index
* key is also returned.
*
*/
static int
__curjoin_split_key(WT_SESSION_IMPL *session, WT_CURSOR_JOIN *cjoin,
WT_ITEM *idxkey, WT_CURSOR *tocur, WT_CURSOR *fromcur,
const char *repack_fmt, bool isindex)
{
WT_CURSOR *firstcg_cur;
WT_CURSOR_INDEX *cindex;
WT_ITEM *keyp;
const uint8_t *p;
if (isindex) {
cindex = ((WT_CURSOR_INDEX *)fromcur);
/*
* Repack tells us where the index key ends; advance past
* that to get where the raw primary key starts.
*/
WT_RET(__wt_struct_repack(session, cindex->child->key_format,
repack_fmt != NULL ? repack_fmt : cindex->iface.key_format,
&cindex->child->key, idxkey));
WT_ASSERT(session, cindex->child->key.size > idxkey->size);
tocur->key.data = (uint8_t *)idxkey->data + idxkey->size;
tocur->key.size = cindex->child->key.size - idxkey->size;
if (WT_CURSOR_RECNO(tocur)) {
p = (const uint8_t *)tocur->key.data;
WT_RET(__wt_vunpack_uint(&p, tocur->key.size,
&tocur->recno));
} else
tocur->recno = 0;
} else {
firstcg_cur = ((WT_CURSOR_TABLE *)fromcur)->cg_cursors[0];
keyp = &firstcg_cur->key;
if (WT_CURSOR_RECNO(tocur)) {
WT_ASSERT(session, keyp->size == sizeof(uint64_t));
tocur->recno = *(uint64_t *)keyp->data;
WT_RET(__curjoin_pack_recno(session, tocur->recno,
cjoin->recno_buf, sizeof(cjoin->recno_buf),
&tocur->key));
} else {
WT_ITEM_SET(tocur->key, *keyp);
tocur->recno = 0;
}
idxkey->data = NULL;
idxkey->size = 0;
}
return (0);
}
/*
* __async_set_key --
* WT_ASYNC_OP->set_key implementation for op handles.
*/
static void
__async_set_key(WT_ASYNC_OP *asyncop, ...)
{
WT_CURSOR *c;
va_list ap;
c = &asyncop->c;
va_start(ap, asyncop);
__wt_cursor_set_keyv(c, c->flags, ap);
if (!WT_DATA_IN_ITEM(&c->key) && !WT_CURSOR_RECNO(c))
c->saved_err = __wt_buf_set(
O2S((WT_ASYNC_OP_IMPL *)asyncop),
&c->key, c->key.data, c->key.size);
va_end(ap);
}
/*
* __curjoin_endpoint_init_key --
* Set the key in the reference endpoint.
*/
static int
__curjoin_endpoint_init_key(WT_SESSION_IMPL *session,
WT_CURSOR_JOIN_ENTRY *entry, WT_CURSOR_JOIN_ENDPOINT *endpoint)
{
WT_CURSOR *cursor;
WT_CURSOR_INDEX *cindex;
WT_DECL_RET;
WT_ITEM *k;
uint64_t r;
void *allocbuf;
allocbuf = NULL;
if ((cursor = endpoint->cursor) != NULL) {
if (entry->index != NULL) {
cindex = (WT_CURSOR_INDEX *)endpoint->cursor;
if (cindex->index->extractor == NULL) {
WT_ERR(__wt_struct_repack(session,
cindex->child->key_format,
entry->main->value_format,
&cindex->child->key, &endpoint->key,
&allocbuf));
if (allocbuf != NULL)
F_SET(endpoint, WT_CURJOIN_END_OWN_KEY);
} else
endpoint->key = cindex->child->key;
} else {
k = &((WT_CURSOR_TABLE *)cursor)->cg_cursors[0]->key;
if (WT_CURSOR_RECNO(cursor)) {
r = *(uint64_t *)k->data;
WT_ERR(__curjoin_pack_recno(session, r,
endpoint->recno_buf,
sizeof(endpoint->recno_buf),
&endpoint->key));
}
else
endpoint->key = *k;
}
}
if (0) {
err: __wt_free(session, allocbuf);
}
return (ret);
}
/*
* __curjoin_entry_iter_next --
* Get the next item in an iteration.
*
*/
static int
__curjoin_entry_iter_next(WT_CURSOR_JOIN_ITER *iter, WT_ITEM *primkey,
uint64_t *rp)
{
WT_CURSOR *firstcg_cur;
WT_CURSOR_JOIN *cjoin;
WT_DECL_RET;
WT_SESSION_IMPL *session;
uint64_t r;
if (iter->advance)
WT_ERR(iter->cursor->next(iter->cursor));
else
iter->advance = true;
session = iter->session;
cjoin = iter->cjoin;
/*
* Set our key to the primary key, we'll also need this
* to check membership.
*/
if (iter->entry->index != NULL)
firstcg_cur = ((WT_CURSOR_INDEX *)iter->cursor)->cg_cursors[0];
else
firstcg_cur = ((WT_CURSOR_TABLE *)iter->cursor)->cg_cursors[0];
if (WT_CURSOR_RECNO(&cjoin->iface)) {
r = *(uint64_t *)firstcg_cur->key.data;
WT_ERR(__curjoin_pack_recno(session, r, cjoin->recno_buf,
sizeof(cjoin->recno_buf), primkey));
*rp = r;
} else {
WT_ITEM_SET(*primkey, firstcg_cur->key);
*rp = 0;
}
iter->curkey = primkey;
iter->entry->stats.actual_count++;
iter->entry->stats.accesses++;
err: return (ret);
}
/*
* __curds_compare --
* WT_CURSOR.compare method for the data-source cursor type.
*/
static int
__curds_compare(WT_CURSOR *a, WT_CURSOR *b, int *cmpp)
{
WT_COLLATOR *collator;
WT_DECL_RET;
WT_SESSION_IMPL *session;
CURSOR_API_CALL(a, session, compare, NULL);
/*
* Confirm both cursors refer to the same source and have keys, then
* compare them.
*/
if (strcmp(a->internal_uri, b->internal_uri) != 0)
WT_ERR_MSG(session, EINVAL,
"Cursors must reference the same object");
WT_ERR(__cursor_needkey(a));
WT_ERR(__cursor_needkey(b));
if (WT_CURSOR_RECNO(a)) {
if (a->recno < b->recno)
*cmpp = -1;
else if (a->recno == b->recno)
*cmpp = 0;
else
*cmpp = 1;
} else {
/*
* The assumption is data-sources don't provide WiredTiger with
* WT_CURSOR.compare methods, instead, we'll copy the key/value
* out of the underlying data-source cursor and any comparison
* to be done can be done at this level.
*/
collator = ((WT_CURSOR_DATA_SOURCE *)a)->collator;
WT_ERR(__wt_compare(
session, collator, &a->key, &b->key, cmpp));
}
err: API_END_RET(session, ret);
}
/*
* __wt_schema_project_slice --
* Given list of cursors and a projection, read columns from the
* a raw buffer.
*/
int
__wt_schema_project_slice(WT_SESSION_IMPL *session, WT_CURSOR **cp,
const char *proj_arg, int key_only, const char *vformat, WT_ITEM *value)
{
WT_CURSOR *c;
WT_ITEM *buf;
WT_PACK pack, vpack;
WT_PACK_VALUE pv, vpv;
char *proj;
uint8_t *end, *p;
const uint8_t *next, *vp, *vend;
size_t len, offset, old_len;
uint32_t arg;
int skip;
WT_CLEAR(pack); /* -Wuninitialized */
WT_CLEAR(vpv); /* -Wuninitialized */
buf = NULL; /* -Wuninitialized */
p = end = NULL; /* -Wuninitialized */
WT_RET(__pack_init(session, &vpack, vformat));
vp = (uint8_t *)value->data;
vend = vp + value->size;
/* Reset any of the buffers we will be setting. */
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = (uint32_t)strtoul(proj, &proj, 10);
if (*proj == WT_PROJ_KEY) {
c = cp[arg];
WT_RET(__wt_buf_init(session, &c->key, 0));
} else if (*proj == WT_PROJ_VALUE && !key_only) {
c = cp[arg];
WT_RET(__wt_buf_init(session, &c->value, 0));
}
}
skip = key_only;
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = (uint32_t)strtoul(proj, &proj, 10);
switch (*proj) {
case WT_PROJ_KEY:
skip = 0;
c = cp[arg];
if (WT_CURSOR_RECNO(c)) {
c->key.data = &c->recno;
c->key.size = sizeof(c->recno);
WT_RET(__pack_init(session, &pack, "R"));
} else
WT_RET(__pack_init(
session, &pack, c->key_format));
buf = &c->key;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
case WT_PROJ_VALUE:
if ((skip = key_only) != 0)
continue;
c = cp[arg];
WT_RET(__pack_init(session, &pack, c->value_format));
buf = &c->value;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
}
/*
* Otherwise, the argument is a count, where a missing
* count means a count of 1.
*/
for (arg = (arg == 0) ? 1 : arg; arg > 0; arg--) {
switch (*proj) {
case WT_PROJ_NEXT:
case WT_PROJ_SKIP:
if (!skip) {
WT_RET(__pack_next(&pack, &pv));
/*
* A nasty case: if we are inserting
* out-of-order, append a zero value
* to keep the buffer in the correct
* format.
*/
if (*proj == WT_PROJ_SKIP &&
p == end) {
/* Set up an empty value. */
WT_CLEAR(pv.u);
if (pv.type == 'S' ||
pv.type == 's')
pv.u.s = "";
len = __pack_size(session, &pv);
WT_RET(__wt_buf_grow(session,
buf, buf->size + len));
p = (uint8_t *)buf->data +
buf->size;
WT_RET(__pack_write(
session, &pv, &p, len));
end = p;
buf->size += WT_STORE_SIZE(len);
} else if (*proj == WT_PROJ_SKIP)
WT_RET(__unpack_read(session,
&pv, (const uint8_t **)&p,
(size_t)(end - p)));
}
if (*proj == WT_PROJ_SKIP)
break;
WT_RET(__pack_next(&vpack, &vpv));
WT_RET(__unpack_read(session, &vpv,
&vp, (size_t)(vend - vp)));
/* FALLTHROUGH */
case WT_PROJ_REUSE:
if (skip)
break;
/* Read the item we're about to overwrite. */
next = p;
if (p < end)
WT_RET(__unpack_read(session, &pv,
&next, (size_t)(end - p)));
old_len = (size_t)(next - p);
/*
* There is subtlety here: the value format
* may not exactly match the cursor's format.
* In particular, we need lengths with raw
* columns in the middle of a packed struct,
* but not if they are at the end of a column.
*/
pv.u = vpv.u;
len = __pack_size(session, &pv);
offset = WT_PTRDIFF(p, buf->data);
WT_RET(__wt_buf_grow(session,
buf, buf->size + len - old_len));
p = (uint8_t *)buf->data + offset;
/* Make room if we're inserting out-of-order. */
if (offset + old_len < buf->size)
memmove(p + len, p + old_len,
buf->size - (offset + old_len));
WT_RET(__pack_write(session, &pv, &p, len));
buf->size += WT_STORE_SIZE(len - old_len);
end = (uint8_t *)buf->data + buf->size;
break;
default:
WT_RET_MSG(session, EINVAL,
"unexpected projection plan: %c",
(int)*proj);
}
}
}
return (0);
}
/*
* __wt_curfile_create --
* Open a cursor for a given btree handle.
*/
int
__wt_curfile_create(WT_SESSION_IMPL *session,
WT_CURSOR *owner, const char *cfg[], bool bulk, bool bitmap,
WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__wt_cursor_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__wt_cursor_set_value, /* set-value */
__curfile_compare, /* compare */
__curfile_equals, /* equals */
__curfile_next, /* next */
__curfile_prev, /* prev */
__curfile_reset, /* reset */
__curfile_search, /* search */
__curfile_search_near, /* search-near */
__curfile_insert, /* insert */
__curfile_update, /* update */
__curfile_remove, /* remove */
__wt_cursor_reconfigure, /* reconfigure */
__curfile_close); /* close */
WT_BTREE *btree;
WT_CONFIG_ITEM cval;
WT_CURSOR *cursor;
WT_CURSOR_BTREE *cbt;
WT_CURSOR_BULK *cbulk;
WT_DECL_RET;
size_t csize;
WT_STATIC_ASSERT(offsetof(WT_CURSOR_BTREE, iface) == 0);
cbt = NULL;
btree = S2BT(session);
WT_ASSERT(session, btree != NULL);
csize = bulk ? sizeof(WT_CURSOR_BULK) : sizeof(WT_CURSOR_BTREE);
WT_RET(__wt_calloc(session, 1, csize, &cbt));
cursor = &cbt->iface;
*cursor = iface;
cursor->session = &session->iface;
cursor->internal_uri = btree->dhandle->name;
cursor->key_format = btree->key_format;
cursor->value_format = btree->value_format;
cbt->btree = btree;
if (session->dhandle->checkpoint != NULL)
F_SET(cbt, WT_CBT_NO_TXN);
if (bulk) {
F_SET(cursor, WT_CURSTD_BULK);
cbulk = (WT_CURSOR_BULK *)cbt;
/* Optionally skip the validation of each bulk-loaded key. */
WT_ERR(__wt_config_gets_def(
session, cfg, "skip_sort_check", 0, &cval));
WT_ERR(__wt_curbulk_init(
session, cbulk, bitmap, cval.val == 0 ? 0 : 1));
}
/*
* Random retrieval, row-store only.
* Random retrieval cursors support a limited set of methods.
*/
WT_ERR(__wt_config_gets_def(session, cfg, "next_random", 0, &cval));
if (cval.val != 0) {
if (WT_CURSOR_RECNO(cursor))
WT_ERR_MSG(session, ENOTSUP,
"next_random configuration not supported for "
"column-store objects");
__wt_cursor_set_notsup(cursor);
cursor->next = __curfile_next_random;
cursor->reset = __curfile_reset;
WT_ERR(__wt_config_gets_def(
session, cfg, "next_random_sample_size", 0, &cval));
if (cval.val != 0)
cbt->next_random_sample_size = (u_int)cval.val;
}
/* Underlying btree initialization. */
__wt_btcur_open(cbt);
/* __wt_cursor_init is last so we don't have to clean up on error. */
WT_ERR(__wt_cursor_init(
cursor, cursor->internal_uri, owner, cfg, cursorp));
WT_STAT_FAST_CONN_INCR(session, cursor_create);
WT_STAT_FAST_DATA_INCR(session, cursor_create);
if (0) {
err: __wt_free(session, cbt);
}
return (ret);
}
/*
* __curdump_get_key --
* WT_CURSOR->get_key for dump cursors.
*/
static int
__curdump_get_key(WT_CURSOR *cursor, ...)
{
WT_CURSOR *child;
WT_CURSOR_DUMP *cdump;
WT_CURSOR_JSON *json;
WT_DECL_RET;
WT_ITEM item, *itemp;
WT_SESSION_IMPL *session;
size_t size;
uint64_t recno;
const char *fmt;
const void *buffer;
va_list ap;
cdump = (WT_CURSOR_DUMP *)cursor;
child = cdump->child;
va_start(ap, cursor);
CURSOR_API_CALL(cursor, session, get_key, NULL);
if (F_ISSET(cursor, WT_CURSTD_DUMP_JSON)) {
json = (WT_CURSOR_JSON *)cursor->json_private;
WT_ASSERT(session, json != NULL);
if (WT_CURSOR_RECNO(cursor)) {
WT_ERR(child->get_key(child, &recno));
buffer = &recno;
size = sizeof(recno);
fmt = "R";
} else {
WT_ERR(__wt_cursor_get_raw_key(child, &item));
buffer = item.data;
size = item.size;
if (F_ISSET(cursor, WT_CURSTD_RAW))
fmt = "u";
else
fmt = cursor->key_format;
}
ret = __wt_json_alloc_unpack(
session, buffer, size, fmt, json, true, ap);
} else {
if (WT_CURSOR_RECNO(cursor) &&
!F_ISSET(cursor, WT_CURSTD_RAW)) {
WT_ERR(child->get_key(child, &recno));
WT_ERR(__wt_buf_fmt(session, &cursor->key, "%"
PRIu64, recno));
} else {
WT_ERR(child->get_key(child, &item));
WT_ERR(__raw_to_dump(session, &item, &cursor->key,
F_ISSET(cursor, WT_CURSTD_DUMP_HEX)));
}
if (F_ISSET(cursor, WT_CURSTD_RAW)) {
itemp = va_arg(ap, WT_ITEM *);
itemp->data = cursor->key.data;
itemp->size = cursor->key.size;
} else
*va_arg(ap, const char **) = cursor->key.data;
}
/*
* __wt_schema_project_slice --
* Given list of cursors and a projection, read columns from the
* a raw buffer.
*/
int
__wt_schema_project_slice(WT_SESSION_IMPL *session, WT_CURSOR **cp,
const char *proj_arg, bool key_only, const char *vformat, WT_ITEM *value)
{
WT_CURSOR *c;
WT_DECL_ITEM(buf);
WT_DECL_PACK(pack);
WT_DECL_PACK_VALUE(pv);
WT_DECL_PACK_VALUE(vpv);
WT_PACK vpack;
u_long arg;
char *proj;
uint8_t *end, *p;
const uint8_t *next, *vp, *vend;
size_t len, offset, old_len;
bool skip;
p = end = NULL; /* -Wuninitialized */
WT_RET(__pack_init(session, &vpack, vformat));
vp = value->data;
vend = vp + value->size;
/* Reset any of the buffers we will be setting. */
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = strtoul(proj, &proj, 10);
if (*proj == WT_PROJ_KEY) {
c = cp[arg];
WT_RET(__wt_buf_init(session, &c->key, 0));
} else if (*proj == WT_PROJ_VALUE && !key_only) {
c = cp[arg];
WT_RET(__wt_buf_init(session, &c->value, 0));
}
}
skip = key_only;
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = strtoul(proj, &proj, 10);
switch (*proj) {
case WT_PROJ_KEY:
skip = false;
c = cp[arg];
if (WT_CURSOR_RECNO(c)) {
c->key.data = &c->recno;
c->key.size = sizeof(c->recno);
WT_RET(__pack_init(session, &pack, "R"));
} else
WT_RET(__pack_init(
session, &pack, c->key_format));
buf = &c->key;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
case WT_PROJ_VALUE:
skip = key_only;
if (skip)
continue;
c = cp[arg];
WT_RET(__pack_init(session, &pack, c->value_format));
buf = &c->value;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
}
/* We have to get a key or value before any operations. */
WT_ASSERT(session, skip || buf != NULL);
/*
* Otherwise, the argument is a count, where a missing
* count means a count of 1.
*/
for (arg = (arg == 0) ? 1 : arg; arg > 0; arg--) {
switch (*proj) {
case WT_PROJ_SKIP:
if (skip)
break;
WT_RET(__pack_next(&pack, &pv));
/*
* A nasty case: if we are inserting
* out-of-order, append a zero value to keep
* the buffer in the correct format.
*/
if (p == end) {
/* Set up an empty value. */
WT_CLEAR(pv.u);
if (pv.type == 'S' || pv.type == 's')
pv.u.s = "";
WT_RET(__pack_size(session, &pv, &len));
WT_RET(__wt_buf_grow(session,
buf, buf->size + len));
p = (uint8_t *)buf->data + buf->size;
WT_RET(__pack_write(
session, &pv, &p, len));
end = p;
buf->size += len;
} else
WT_RET(__unpack_read(session,
&pv, (const uint8_t **)&p,
(size_t)(end - p)));
break;
case WT_PROJ_NEXT:
WT_RET(__pack_next(&vpack, &vpv));
WT_RET(__unpack_read(session, &vpv,
&vp, (size_t)(vend - vp)));
/* FALLTHROUGH */
case WT_PROJ_REUSE:
if (skip)
break;
/*
* Read the item we're about to overwrite.
*
* There is subtlety here: the value format
* may not exactly match the cursor's format.
* In particular, we need lengths with raw
* columns in the middle of a packed struct,
* but not if they are at the end of a struct.
*/
WT_RET(__pack_next(&pack, &pv));
next = p;
if (p < end)
WT_RET(__unpack_read(session, &pv,
&next, (size_t)(end - p)));
old_len = (size_t)(next - p);
/* Make sure the types are compatible. */
WT_ASSERT(session,
__wt_tolower((u_char)pv.type) ==
__wt_tolower((u_char)vpv.type));
pv.u = vpv.u;
WT_RET(__pack_size(session, &pv, &len));
offset = WT_PTRDIFF(p, buf->data);
/*
* Avoid growing the buffer if the value fits.
* This is not just a performance issue: it
* covers the case of record number keys, which
* have to be written to cursor->recno.
*/
if (len > old_len)
WT_RET(__wt_buf_grow(session,
buf, buf->size + len - old_len));
p = (uint8_t *)buf->data + offset;
/* Make room if we're inserting out-of-order. */
if (offset + old_len < buf->size)
memmove(p + len, p + old_len,
buf->size - (offset + old_len));
WT_RET(__pack_write(session, &pv, &p, len));
buf->size += len - old_len;
end = (uint8_t *)buf->data + buf->size;
break;
default:
WT_RET_MSG(session, EINVAL,
"unexpected projection plan: %c",
(int)*proj);
}
}
}
return (0);
}
/*
* __wt_schema_project_merge --
* Given list of cursors and a projection, build a buffer containing the
* column values read from the cursors.
*/
int
__wt_schema_project_merge(WT_SESSION_IMPL *session,
WT_CURSOR **cp, const char *proj_arg, const char *vformat, WT_ITEM *value)
{
WT_CURSOR *c;
WT_DECL_PACK(pack);
WT_DECL_PACK_VALUE(pv);
WT_DECL_PACK_VALUE(vpv);
WT_ITEM *buf;
WT_PACK vpack;
u_long arg;
char *proj;
const uint8_t *p, *end;
uint8_t *vp;
size_t len;
p = end = NULL; /* -Wuninitialized */
WT_RET(__wt_buf_init(session, value, 0));
WT_RET(__pack_init(session, &vpack, vformat));
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = strtoul(proj, &proj, 10);
switch (*proj) {
case WT_PROJ_KEY:
c = cp[arg];
if (WT_CURSOR_RECNO(c)) {
c->key.data = &c->recno;
c->key.size = sizeof(c->recno);
WT_RET(__pack_init(session, &pack, "R"));
} else
WT_RET(__pack_init(
session, &pack, c->key_format));
buf = &c->key;
p = buf->data;
end = p + buf->size;
continue;
case WT_PROJ_VALUE:
c = cp[arg];
WT_RET(__pack_init(session, &pack, c->value_format));
buf = &c->value;
p = buf->data;
end = p + buf->size;
continue;
}
/*
* Otherwise, the argument is a count, where a missing
* count means a count of 1.
*/
for (arg = (arg == 0) ? 1 : arg; arg > 0; arg--) {
switch (*proj) {
case WT_PROJ_NEXT:
case WT_PROJ_SKIP:
case WT_PROJ_REUSE:
WT_RET(__pack_next(&pack, &pv));
WT_RET(__unpack_read(session, &pv,
&p, (size_t)(end - p)));
/* Only copy the value out once. */
if (*proj != WT_PROJ_NEXT)
break;
WT_RET(__pack_next(&vpack, &vpv));
/* Make sure the types are compatible. */
WT_ASSERT(session,
__wt_tolower((u_char)pv.type) ==
__wt_tolower((u_char)vpv.type));
vpv.u = pv.u;
WT_RET(__pack_size(session, &vpv, &len));
WT_RET(__wt_buf_grow(session,
value, value->size + len));
vp = (uint8_t *)value->mem + value->size;
WT_RET(__pack_write(session, &vpv, &vp, len));
value->size += len;
break;
}
}
}
return (0);
}
/*
* __wt_schema_project_out --
* Given list of cursors and a projection, read columns from the
* dependent cursors and return them to the application.
*/
int
__wt_schema_project_out(WT_SESSION_IMPL *session,
WT_CURSOR **cp, const char *proj_arg, va_list ap)
{
WT_CURSOR *c;
WT_DECL_PACK(pack);
WT_DECL_PACK_VALUE(pv);
u_long arg;
char *proj;
uint8_t *p, *end;
p = end = NULL; /* -Wuninitialized */
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = strtoul(proj, &proj, 10);
switch (*proj) {
case WT_PROJ_KEY:
c = cp[arg];
if (WT_CURSOR_RECNO(c)) {
c->key.data = &c->recno;
c->key.size = sizeof(c->recno);
WT_RET(__pack_init(session, &pack, "R"));
} else
WT_RET(__pack_init(
session, &pack, c->key_format));
p = (uint8_t *)c->key.data;
end = p + c->key.size;
continue;
case WT_PROJ_VALUE:
c = cp[arg];
WT_RET(__pack_init(session, &pack, c->value_format));
p = (uint8_t *)c->value.data;
end = p + c->value.size;
continue;
}
/*
* Otherwise, the argument is a count, where a missing
* count means a count of 1.
*/
for (arg = (arg == 0) ? 1 : arg; arg > 0; arg--) {
switch (*proj) {
case WT_PROJ_NEXT:
case WT_PROJ_SKIP:
case WT_PROJ_REUSE:
WT_RET(__pack_next(&pack, &pv));
WT_RET(__unpack_read(session, &pv,
(const uint8_t **)&p, (size_t)(end - p)));
/* Only copy the value out once. */
if (*proj != WT_PROJ_NEXT)
break;
WT_UNPACK_PUT(session, pv, ap);
break;
}
}
}
return (0);
}
/*
* __wt_schema_project_in --
* Given list of cursors and a projection, read columns from the
* application into the dependent cursors.
*/
int
__wt_schema_project_in(WT_SESSION_IMPL *session,
WT_CURSOR **cp, const char *proj_arg, va_list ap)
{
WT_CURSOR *c;
WT_DECL_ITEM(buf);
WT_DECL_PACK(pack);
WT_DECL_PACK_VALUE(pv);
WT_PACK_VALUE old_pv;
size_t len, offset, old_len;
u_long arg;
char *proj;
uint8_t *p, *end;
const uint8_t *next;
p = end = NULL; /* -Wuninitialized */
/* Reset any of the buffers we will be setting. */
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = strtoul(proj, &proj, 10);
if (*proj == WT_PROJ_KEY) {
c = cp[arg];
WT_RET(__wt_buf_init(session, &c->key, 0));
} else if (*proj == WT_PROJ_VALUE) {
c = cp[arg];
WT_RET(__wt_buf_init(session, &c->value, 0));
}
}
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = strtoul(proj, &proj, 10);
switch (*proj) {
case WT_PROJ_KEY:
c = cp[arg];
if (WT_CURSOR_RECNO(c)) {
c->key.data = &c->recno;
c->key.size = sizeof(c->recno);
WT_RET(__pack_init(session, &pack, "R"));
} else
WT_RET(__pack_init(
session, &pack, c->key_format));
buf = &c->key;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
case WT_PROJ_VALUE:
c = cp[arg];
WT_RET(__pack_init(session, &pack, c->value_format));
buf = &c->value;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
}
/* We have to get a key or value before any operations. */
WT_ASSERT(session, buf != NULL);
/*
* Otherwise, the argument is a count, where a missing
* count means a count of 1.
*/
for (arg = (arg == 0) ? 1 : arg; arg > 0; arg--) {
switch (*proj) {
case WT_PROJ_SKIP:
WT_RET(__pack_next(&pack, &pv));
/*
* A nasty case: if we are inserting
* out-of-order, we may reach the end of the
* data. That's okay: we want to append in
* that case, and we're positioned to do that.
*/
if (p == end) {
/* Set up an empty value. */
WT_CLEAR(pv.u);
if (pv.type == 'S' || pv.type == 's')
pv.u.s = "";
WT_RET(__pack_size(session, &pv, &len));
WT_RET(__wt_buf_grow(session,
buf, buf->size + len));
p = (uint8_t *)buf->mem + buf->size;
WT_RET(__pack_write(
session, &pv, &p, len));
buf->size += len;
end = (uint8_t *)buf->mem + buf->size;
} else if (*proj == WT_PROJ_SKIP)
WT_RET(__unpack_read(session,
&pv, (const uint8_t **)&p,
(size_t)(end - p)));
break;
case WT_PROJ_NEXT:
WT_RET(__pack_next(&pack, &pv));
WT_PACK_GET(session, pv, ap);
/* FALLTHROUGH */
case WT_PROJ_REUSE:
/* Read the item we're about to overwrite. */
next = p;
if (p < end) {
old_pv = pv;
WT_RET(__unpack_read(session, &old_pv,
&next, (size_t)(end - p)));
}
old_len = (size_t)(next - p);
WT_RET(__pack_size(session, &pv, &len));
offset = WT_PTRDIFF(p, buf->mem);
WT_RET(__wt_buf_grow(session,
buf, buf->size + len));
p = (uint8_t *)buf->mem + offset;
end = (uint8_t *)buf->mem + buf->size + len;
/* Make room if we're inserting out-of-order. */
if (offset + old_len < buf->size)
memmove(p + len, p + old_len,
buf->size - (offset + old_len));
WT_RET(__pack_write(session, &pv, &p, len));
buf->size += len;
break;
default:
WT_RET_MSG(session, EINVAL,
"unexpected projection plan: %c",
(int)*proj);
}
}
}
return (0);
}
/*
* __wt_curindex_open --
* WT_SESSION->open_cursor method for index cursors.
*/
int
__wt_curindex_open(WT_SESSION_IMPL *session,
const char *uri, WT_CURSOR *owner, const char *cfg[], WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__curindex_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__curindex_set_value, /* set-value */
__curindex_compare, /* compare */
__wt_cursor_equals, /* equals */
__curindex_next, /* next */
__curindex_prev, /* prev */
__curindex_reset, /* reset */
__curindex_search, /* search */
__curindex_search_near, /* search-near */
__wt_cursor_notsup, /* insert */
__wt_cursor_notsup, /* update */
__wt_cursor_notsup, /* remove */
__wt_cursor_reconfigure_notsup, /* reconfigure */
__curindex_close); /* close */
WT_CURSOR_INDEX *cindex;
WT_CURSOR *cursor;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
WT_INDEX *idx;
WT_TABLE *table;
const char *columns, *idxname, *tablename;
size_t namesize;
tablename = uri;
if (!WT_PREFIX_SKIP(tablename, "index:") ||
(idxname = strchr(tablename, ':')) == NULL)
WT_RET_MSG(session, EINVAL, "Invalid cursor URI: '%s'", uri);
namesize = (size_t)(idxname - tablename);
++idxname;
if ((ret = __wt_schema_get_table(session,
tablename, namesize, false, &table)) != 0) {
if (ret == WT_NOTFOUND)
WT_RET_MSG(session, EINVAL,
"Cannot open cursor '%s' on unknown table", uri);
return (ret);
}
columns = strchr(idxname, '(');
if (columns == NULL)
namesize = strlen(idxname);
else
namesize = (size_t)(columns - idxname);
if ((ret = __wt_schema_open_index(
session, table, idxname, namesize, &idx)) != 0) {
__wt_schema_release_table(session, table);
return (ret);
}
WT_RET(__wt_calloc_one(session, &cindex));
cursor = &cindex->iface;
*cursor = iface;
cursor->session = &session->iface;
cindex->table = table;
cindex->index = idx;
cindex->key_plan = idx->key_plan;
cindex->value_plan = idx->value_plan;
cursor->internal_uri = idx->name;
cursor->key_format = idx->idxkey_format;
cursor->value_format = table->value_format;
/*
* XXX
* A very odd corner case is an index with a recno key.
* The only way to get here is by creating an index on a column store
* using only the primary's recno as the index key. Disallow that for
* now.
*/
if (WT_CURSOR_RECNO(cursor))
WT_ERR_MSG(session, WT_ERROR,
"Column store indexes based on a record number primary "
"key are not supported");
/* Handle projections. */
if (columns != NULL) {
WT_ERR(__wt_scr_alloc(session, 0, &tmp));
WT_ERR(__wt_struct_reformat(session, table,
columns, strlen(columns), NULL, false, tmp));
WT_ERR(__wt_strndup(
session, tmp->data, tmp->size, &cursor->value_format));
WT_ERR(__wt_buf_init(session, tmp, 0));
WT_ERR(__wt_struct_plan(session, table,
columns, strlen(columns), false, tmp));
WT_ERR(__wt_strndup(
session, tmp->data, tmp->size, &cindex->value_plan));
}
WT_ERR(__wt_cursor_init(
cursor, cursor->internal_uri, owner, cfg, cursorp));
WT_ERR(__wt_open_cursor(
session, idx->source, cursor, cfg, &cindex->child));
/* Open the column groups needed for this index cursor. */
WT_ERR(__curindex_open_colgroups(session, cindex, cfg));
if (F_ISSET(cursor, WT_CURSTD_DUMP_JSON))
__wt_json_column_init(
cursor, table->key_format, &idx->colconf, &table->colconf);
if (0) {
err: WT_TRET(__curindex_close(cursor));
*cursorp = NULL;
}
__wt_scr_free(session, &tmp);
return (ret);
}
/*
* __wt_schema_project_merge --
* Given list of cursors and a projection, build a buffer containing the
* column values read from the cursors.
*/
int
__wt_schema_project_merge(WT_SESSION_IMPL *session,
WT_CURSOR **cp, const char *proj_arg, const char *vformat, WT_ITEM *value)
{
WT_CURSOR *c;
WT_ITEM *buf;
WT_PACK pack, vpack;
WT_PACK_VALUE pv, vpv;
char *proj;
uint8_t *p, *end, *vp;
size_t len;
uint32_t arg;
WT_CLEAR(pack); /* -Wuninitialized */
WT_CLEAR(pv); /* -Wuninitialized */
WT_CLEAR(vpv); /* -Wuninitialized */
p = end = NULL; /* -Wuninitialized */
WT_RET(__wt_buf_init(session, value, 0));
WT_RET(__pack_init(session, &vpack, vformat));
for (proj = (char *)proj_arg; *proj != '\0'; proj++) {
arg = (uint32_t)strtoul(proj, &proj, 10);
switch (*proj) {
case WT_PROJ_KEY:
c = cp[arg];
if (WT_CURSOR_RECNO(c)) {
c->key.data = &c->recno;
c->key.size = sizeof(c->recno);
WT_RET(__pack_init(session, &pack, "R"));
} else
WT_RET(__pack_init(
session, &pack, c->key_format));
buf = &c->key;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
case WT_PROJ_VALUE:
c = cp[arg];
WT_RET(__pack_init(session, &pack, c->value_format));
buf = &c->value;
p = (uint8_t *)buf->data;
end = p + buf->size;
continue;
}
/*
* Otherwise, the argument is a count, where a missing
* count means a count of 1.
*/
for (arg = (arg == 0) ? 1 : arg; arg > 0; arg--) {
switch (*proj) {
case WT_PROJ_NEXT:
case WT_PROJ_SKIP:
WT_RET(__pack_next(&pack, &pv));
WT_RET(__unpack_read(session, &pv,
(const uint8_t **)&p,
(size_t)(end - p)));
if (*proj == WT_PROJ_SKIP)
break;
WT_RET(__pack_next(&vpack, &vpv));
vpv.u = pv.u;
len = __pack_size(session, &vpv);
WT_RET(__wt_buf_grow(session,
value, value->size + len));
vp = (uint8_t *)value->data + value->size;
WT_RET(__pack_write(session, &vpv, &vp, len));
value->size += WT_STORE_SIZE(len);
/* FALLTHROUGH */
case WT_PROJ_REUSE:
/* Don't copy the same value twice. */
break;
}
}
}
return (0);
}
/*
* __curfile_create --
* Open a cursor for a given btree handle.
*/
static int
__curfile_create(WT_SESSION_IMPL *session,
WT_CURSOR *owner, const char *cfg[], bool bulk, bool bitmap,
WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__wt_cursor_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__wt_cursor_set_value, /* set-value */
__curfile_compare, /* compare */
__curfile_equals, /* equals */
__curfile_next, /* next */
__curfile_prev, /* prev */
__curfile_reset, /* reset */
__curfile_search, /* search */
__curfile_search_near, /* search-near */
__curfile_insert, /* insert */
__wt_cursor_modify_notsup, /* modify */
__curfile_update, /* update */
__curfile_remove, /* remove */
__curfile_reserve, /* reserve */
__wt_cursor_reconfigure, /* reconfigure */
__curfile_close); /* close */
WT_BTREE *btree;
WT_CONFIG_ITEM cval;
WT_CURSOR *cursor;
WT_CURSOR_BTREE *cbt;
WT_CURSOR_BULK *cbulk;
WT_DECL_RET;
size_t csize;
WT_STATIC_ASSERT(offsetof(WT_CURSOR_BTREE, iface) == 0);
cbt = NULL;
btree = S2BT(session);
WT_ASSERT(session, btree != NULL);
csize = bulk ? sizeof(WT_CURSOR_BULK) : sizeof(WT_CURSOR_BTREE);
WT_RET(__wt_calloc(session, 1, csize, &cbt));
cursor = &cbt->iface;
*cursor = iface;
cursor->session = &session->iface;
cursor->internal_uri = btree->dhandle->name;
cursor->key_format = btree->key_format;
cursor->value_format = btree->value_format;
cbt->btree = btree;
/*
* Increment the data-source's in-use counter; done now because closing
* the cursor will decrement it, and all failure paths from here close
* the cursor.
*/
__wt_cursor_dhandle_incr_use(session);
if (session->dhandle->checkpoint != NULL)
F_SET(cbt, WT_CBT_NO_TXN);
if (bulk) {
F_SET(cursor, WT_CURSTD_BULK);
cbulk = (WT_CURSOR_BULK *)cbt;
/* Optionally skip the validation of each bulk-loaded key. */
WT_ERR(__wt_config_gets_def(
session, cfg, "skip_sort_check", 0, &cval));
WT_ERR(__wt_curbulk_init(
session, cbulk, bitmap, cval.val == 0 ? 0 : 1));
}
/*
* Random retrieval, row-store only.
* Random retrieval cursors support a limited set of methods.
*/
WT_ERR(__wt_config_gets_def(session, cfg, "next_random", 0, &cval));
if (cval.val != 0) {
if (WT_CURSOR_RECNO(cursor))
WT_ERR_MSG(session, ENOTSUP,
"next_random configuration not supported for "
"column-store objects");
__wt_cursor_set_notsup(cursor);
cursor->next = __wt_curfile_next_random;
cursor->reset = __curfile_reset;
WT_ERR(__wt_config_gets_def(
session, cfg, "next_random_sample_size", 0, &cval));
if (cval.val != 0)
cbt->next_random_sample_size = (u_int)cval.val;
}
/* Underlying btree initialization. */
__wt_btcur_open(cbt);
/*
* WT_CURSOR.modify supported on 'u' value formats, but the fast-path
* through the btree code requires log file format changes, it's not
* available in all versions.
*/
if (WT_STREQ(cursor->value_format, "u") &&
S2C(session)->compat_major >= WT_LOG_V2)
cursor->modify = __curfile_modify;
WT_ERR(__wt_cursor_init(
cursor, cursor->internal_uri, owner, cfg, cursorp));
WT_STAT_CONN_INCR(session, cursor_create);
WT_STAT_DATA_INCR(session, cursor_create);
if (0) {
err: /*
* Our caller expects to release the data handle if we fail.
* Disconnect it from the cursor before closing.
*/
if (session->dhandle != NULL)
__wt_cursor_dhandle_decr_use(session);
cbt->btree = NULL;
WT_TRET(__curfile_close(cursor));
*cursorp = NULL;
}
return (ret);
}