/*
* REC_TOTAL -- Return the number of recno entries below a page.
*
* Parameters:
* h: page
*
* Returns:
* The number of recno entries below a page.
*
* XXX
* These values could be set by the bt_psplit routine. The problem is that the
* entry has to be popped off of the stack etc. or the values have to be passed
* all the way back to bt_split/bt_rroot and it's not very clean.
*/
static recno_t
rec_total(PAGE *h)
{
recno_t recs;
indx_t nxt, top;
for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt)
recs += GETRINTERNAL(h, nxt)->nrecs;
return (recs);
}
/*
* __REC_SEARCH -- Search a btree for a key.
*
* Parameters:
* t: tree to search
* recno: key to find
* op: search operation
*
* Returns:
* EPG for matching record, if any, or the EPG for the location of the
* key, if it were inserted into the tree.
*
* Returns:
* The EPG for matching record, if any, or the EPG for the location
* of the key, if it were inserted into the tree, is entered into
* the bt_cur field of the tree. A pointer to the field is returned.
*/
EPG *
__rec_search(BTREE *t, recno_t recno, enum SRCHOP op)
{
indx_t idx;
PAGE *h;
EPGNO *parent;
RINTERNAL *r;
pgno_t pg;
indx_t top;
recno_t total;
int sverrno;
BT_CLR(t);
for (pg = P_ROOT, total = 0;;) {
if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL)
goto err;
if (h->flags & P_RLEAF) {
t->bt_cur.page = h;
t->bt_cur.index = recno - total;
return (&t->bt_cur);
}
for (idx = 0, top = NEXTINDEX(h);;) {
r = GETRINTERNAL(h, idx);
if (++idx == top || total + r->nrecs > recno)
break;
total += r->nrecs;
}
BT_PUSH(t, pg, idx - 1);
pg = r->pgno;
switch (op) {
case SDELETE:
--GETRINTERNAL(h, (idx - 1))->nrecs;
mpool_put(t->bt_mp, h, MPOOL_DIRTY);
break;
case SINSERT:
++GETRINTERNAL(h, (idx - 1))->nrecs;
mpool_put(t->bt_mp, h, MPOOL_DIRTY);
break;
case SEARCH:
mpool_put(t->bt_mp, h, 0);
break;
}
}
/* Try and recover the tree. */
err: sverrno = errno;
if (op != SEARCH)
while ((parent = BT_POP(t)) != NULL) {
if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
break;
if (op == SINSERT)
--GETRINTERNAL(h, parent->index)->nrecs;
else
++GETRINTERNAL(h, parent->index)->nrecs;
mpool_put(t->bt_mp, h, MPOOL_DIRTY);
}
errno = sverrno;
return (NULL);
}
/*
* BT_STAT -- Gather/print the tree statistics
*
* Parameters:
* dbp: pointer to the DB
*/
void
__bt_stat(DB *dbp)
{
extern u_long bt_cache_hit, bt_cache_miss, bt_pfxsaved, bt_rootsplit;
extern u_long bt_sortsplit, bt_split;
BTREE *t;
PAGE *h;
pgno_t i, pcont, pinternal, pleaf;
u_long ifree, lfree, nkeys;
int levels;
t = dbp->internal;
pcont = pinternal = pleaf = 0;
nkeys = ifree = lfree = 0;
for (i = P_ROOT;
(h = mpool_get(t->bt_mp, i, MPOOL_IGNOREPIN)) != NULL; ++i)
switch (h->flags & P_TYPE) {
case P_BINTERNAL:
case P_RINTERNAL:
++pinternal;
ifree += h->upper - h->lower;
break;
case P_BLEAF:
case P_RLEAF:
++pleaf;
lfree += h->upper - h->lower;
nkeys += NEXTINDEX(h);
break;
case P_OVERFLOW:
++pcont;
break;
}
/* Count the levels of the tree. */
for (i = P_ROOT, levels = 0 ;; ++levels) {
h = mpool_get(t->bt_mp, i, MPOOL_IGNOREPIN);
if (h->flags & (P_BLEAF|P_RLEAF)) {
if (levels == 0)
levels = 1;
break;
}
i = F_ISSET(t, R_RECNO) ?
GETRINTERNAL(h, 0)->pgno :
GETBINTERNAL(h, 0)->pgno;
}
(void)fprintf(stderr, "%d level%s with %lu keys",
levels, levels == 1 ? "" : "s", nkeys);
if (F_ISSET(t, R_RECNO))
(void)fprintf(stderr, " (%u header count)", t->bt_nrecs);
(void)fprintf(stderr,
"\n%u pages (leaf %u, internal %u, overflow %u)\n",
pinternal + pleaf + pcont, pleaf, pinternal, pcont);
(void)fprintf(stderr, "%lu cache hits, %lu cache misses\n",
bt_cache_hit, bt_cache_miss);
(void)fprintf(stderr, "%lu splits (%lu root splits, %lu sort splits)\n",
bt_split, bt_rootsplit, bt_sortsplit);
pleaf *= t->bt_psize - BTDATAOFF;
if (pleaf)
(void)fprintf(stderr,
"%.0f%% leaf fill (%lu bytes used, %lu bytes free)\n",
((double)(pleaf - lfree) / pleaf) * 100,
pleaf - lfree, lfree);
pinternal *= t->bt_psize - BTDATAOFF;
if (pinternal)
(void)fprintf(stderr,
"%.0f%% internal fill (%lu bytes used, %lu bytes free\n",
((double)(pinternal - ifree) / pinternal) * 100,
pinternal - ifree, ifree);
if (bt_pfxsaved)
(void)fprintf(stderr, "prefix checking removed %lu bytes.\n",
bt_pfxsaved);
}
/*
* BT_PSPLIT -- Do the real work of splitting the page.
*
* Parameters:
* t: tree
* h: page to be split
* l: page to put lower half of data
* r: page to put upper half of data
* pskip: pointer to index to leave open
* ilen: insert length
*
* Returns:
* Pointer to page in which to insert.
*/
static PAGE *
bt_psplit(BTREE *t, PAGE *h, PAGE *l, PAGE *r, indx_t *pskip, size_t ilen)
{
BINTERNAL *bi;
BLEAF *bl;
CURSOR *c;
RLEAF *rl;
PAGE *rval;
void *src;
indx_t full, half, nxt, off, skip, top, used;
u_int32_t nbytes;
int bigkeycnt, isbigkey;
/*
* Split the data to the left and right pages. Leave the skip index
* open. Additionally, make some effort not to split on an overflow
* key. This makes internal page processing faster and can save
* space as overflow keys used by internal pages are never deleted.
*/
bigkeycnt = 0;
skip = *pskip;
full = t->bt_psize - BTDATAOFF;
half = full / 2;
used = 0;
for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) {
if (skip == off) {
nbytes = ilen;
isbigkey = 0; /* XXX: not really known. */
} else
switch (h->flags & P_TYPE) {
case P_BINTERNAL:
src = bi = GETBINTERNAL(h, nxt);
nbytes = NBINTERNAL(bi->ksize);
isbigkey = bi->flags & P_BIGKEY;
break;
case P_BLEAF:
src = bl = GETBLEAF(h, nxt);
nbytes = NBLEAF(bl);
isbigkey = bl->flags & P_BIGKEY;
break;
case P_RINTERNAL:
src = GETRINTERNAL(h, nxt);
nbytes = NRINTERNAL;
isbigkey = 0;
break;
case P_RLEAF:
src = rl = GETRLEAF(h, nxt);
nbytes = NRLEAF(rl);
isbigkey = 0;
break;
default:
abort();
}
/*
* If the key/data pairs are substantial fractions of the max
* possible size for the page, it's possible to get situations
* where we decide to try and copy too much onto the left page.
* Make sure that doesn't happen.
*/
if ((skip <= off && used + nbytes + sizeof(indx_t) >= full) ||
nxt == top - 1) {
--off;
break;
}
/* Copy the key/data pair, if not the skipped index. */
if (skip != off) {
++nxt;
l->linp[off] = l->upper -= nbytes;
memmove((char *)l + l->upper, src, nbytes);
}
used += nbytes + sizeof(indx_t);
if (used >= half) {
if (!isbigkey || bigkeycnt == 3)
break;
else
++bigkeycnt;
}
}
/*
* Off is the last offset that's valid for the left page.
* Nxt is the first offset to be placed on the right page.
*/
l->lower += (off + 1) * sizeof(indx_t);
/*
* If splitting the page that the cursor was on, the cursor has to be
* adjusted to point to the same record as before the split. If the
* cursor is at or past the skipped slot, the cursor is incremented by
* one. If the cursor is on the right page, it is decremented by the
* number of records split to the left page.
*/
c = &t->bt_cursor;
if (F_ISSET(c, CURS_INIT) && c->pg.pgno == h->pgno) {
if (c->pg.index >= skip)
++c->pg.index;
if (c->pg.index < nxt) /* Left page. */
c->pg.pgno = l->pgno;
else { /* Right page. */
c->pg.pgno = r->pgno;
c->pg.index -= nxt;
}
}
/*
* If the skipped index was on the left page, just return that page.
* Otherwise, adjust the skip index to reflect the new position on
* the right page.
*/
if (skip <= off) {
skip = MAX_PAGE_OFFSET;
rval = l;
} else {
rval = r;
*pskip -= nxt;
}
for (off = 0; nxt < top; ++off) {
if (skip == nxt) {
++off;
skip = MAX_PAGE_OFFSET;
}
switch (h->flags & P_TYPE) {
case P_BINTERNAL:
src = bi = GETBINTERNAL(h, nxt);
nbytes = NBINTERNAL(bi->ksize);
break;
case P_BLEAF:
src = bl = GETBLEAF(h, nxt);
nbytes = NBLEAF(bl);
break;
case P_RINTERNAL:
src = GETRINTERNAL(h, nxt);
nbytes = NRINTERNAL;
break;
case P_RLEAF:
src = rl = GETRLEAF(h, nxt);
nbytes = NRLEAF(rl);
break;
default:
abort();
}
++nxt;
r->linp[off] = r->upper -= nbytes;
memmove((char *)r + r->upper, src, nbytes);
}
r->lower += off * sizeof(indx_t);
/* If the key is being appended to the page, adjust the index. */
if (skip == top)
r->lower += sizeof(indx_t);
return (rval);
}
/*
* BT_DPAGE -- Dump the page
*
* Parameters:
* h: pointer to the PAGE
*/
void
__bt_dpage(PAGE *h)
{
BINTERNAL *bi;
BLEAF *bl;
RINTERNAL *ri;
RLEAF *rl;
indx_t cur, top;
char *sep;
(void)fprintf(stderr, " page %u: (", h->pgno);
#undef X
#define X(flag, name) \
if (h->flags & flag) { \
(void)fprintf(stderr, "%s%s", sep, name); \
sep = ", "; \
}
sep = "";
X(P_BINTERNAL, "BINTERNAL") /* types */
X(P_BLEAF, "BLEAF")
X(P_RINTERNAL, "RINTERNAL") /* types */
X(P_RLEAF, "RLEAF")
X(P_OVERFLOW, "OVERFLOW")
X(P_PRESERVE, "PRESERVE");
(void)fprintf(stderr, ")\n");
#undef X
(void)fprintf(stderr, "\tprev %2u next %2u", h->prevpg, h->nextpg);
if (h->flags & P_OVERFLOW)
return;
top = NEXTINDEX(h);
(void)fprintf(stderr, " lower %3d upper %3d nextind %d\n",
h->lower, h->upper, top);
for (cur = 0; cur < top; cur++) {
(void)fprintf(stderr, "\t[%03d] %4d ", cur, h->linp[cur]);
switch (h->flags & P_TYPE) {
case P_BINTERNAL:
bi = GETBINTERNAL(h, cur);
(void)fprintf(stderr,
"size %03d pgno %03d", bi->ksize, bi->pgno);
if (bi->flags & P_BIGKEY)
(void)fprintf(stderr, " (indirect)");
else if (bi->ksize)
(void)fprintf(stderr,
" {%.*s}", (int)bi->ksize, bi->bytes);
break;
case P_RINTERNAL:
ri = GETRINTERNAL(h, cur);
(void)fprintf(stderr, "entries %03d pgno %03d",
ri->nrecs, ri->pgno);
break;
case P_BLEAF:
bl = GETBLEAF(h, cur);
if (bl->flags & P_BIGKEY)
(void)fprintf(stderr,
"big key page %u size %u/",
*(pgno_t *)bl->bytes,
*(u_int32_t *)(bl->bytes + sizeof(pgno_t)));
else if (bl->ksize)
(void)fprintf(stderr, "%s/", bl->bytes);
if (bl->flags & P_BIGDATA)
(void)fprintf(stderr,
"big data page %u size %u",
*(pgno_t *)(bl->bytes + bl->ksize),
*(u_int32_t *)(bl->bytes + bl->ksize +
sizeof(pgno_t)));
else if (bl->dsize)
(void)fprintf(stderr, "%.*s",
(int)bl->dsize, bl->bytes + bl->ksize);
break;
case P_RLEAF:
rl = GETRLEAF(h, cur);
if (rl->flags & P_BIGDATA)
(void)fprintf(stderr,
"big data page %u size %u",
*(pgno_t *)rl->bytes,
*(u_int32_t *)(rl->bytes + sizeof(pgno_t)));
else if (rl->dsize)
(void)fprintf(stderr,
"%.*s", (int)rl->dsize, rl->bytes);
break;
}
(void)fprintf(stderr, "\n");
}
}
