#define MDB_MINKEYS 4
#define MDB_MAGIC 0xBEEFC0DE
#define MDB_VERSION 1
-#define MAXKEYSIZE 255
+#define MAXKEYSIZE 511
#define P_INVALID (~0UL)
#define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
#define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
+struct MDB_xcursor;
+
struct MDB_cursor {
MDB_txn *mc_txn;
struct page_stack mc_stack; /* stack of parent pages */
MDB_dbi mc_dbi;
short mc_initialized; /* 1 if initialized */
short mc_eof; /* 1 if end is reached */
+ struct MDB_xcursor *mc_xcursor;
};
#define METAHASHLEN offsetof(MDB_meta, mm_hash)
MDB_cmp_func *md_cmp; /* user compare function */
MDB_cmp_func *md_dcmp; /* user dupsort function */
MDB_rel_func *md_rel; /* user relocate function */
+ MDB_dbi md_parent;
+ unsigned int md_dirty;
} MDB_dbx;
struct MDB_txn {
#define MDB_TXN_RDONLY 0x01 /* read-only transaction */
#define MDB_TXN_ERROR 0x02 /* an error has occurred */
#define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
- unsigned int mt_flags;
+ unsigned int mt_flags;
};
+/* Context for sorted-dup records */
+typedef struct MDB_xcursor {
+ MDB_cursor mx_cursor;
+ MDB_txn mx_txn;
+ MDB_dbx mx_dbxs[4];
+ MDB_db mx_dbs[4];
+} MDB_xcursor;
+
struct MDB_env {
int me_fd;
int me_lfd;
MDB_db *me_dbs[2];
MDB_oldpages *me_pghead;
pthread_key_t me_txkey; /* thread-key for readers */
+ pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
};
#define NODESIZE offsetof(MDB_node, mn_data)
static int mdbenv_read_header(MDB_env *env, MDB_meta *meta);
static int mdbenv_read_meta(MDB_env *env, int *which);
static int mdbenv_write_meta(MDB_txn *txn);
-static MDB_page *mdbenv_get_page(MDB_env *env, pgno_t pgno);
+static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
MDB_val *key, int *exactp, unsigned int *kip);
static void mdb_del_node(MDB_page *mp, indx_t indx);
static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
MDB_pageparent *mpp, MDB_node *leaf);
-static int mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data);
+static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
static int mdb_sibling(MDB_cursor *cursor, int move_right);
static int mdb_cursor_next(MDB_cursor *cursor,
MDB_val *key, MDB_val *data);
+static int mdb_cursor_prev(MDB_cursor *cursor,
+ MDB_val *key, MDB_val *data);
static int mdb_cursor_set(MDB_cursor *cursor,
MDB_val *key, MDB_val *data, int *exactp);
static int mdb_cursor_first(MDB_cursor *cursor,
MDB_val data;
pgno_t *idl;
- mdb_read_data(txn->mt_env, leaf, &data);
+ mdb_read_data(txn, leaf, &data);
idl = (ULONG *)data.mv_data;
mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
mop->mo_next = txn->mt_env->me_pghead;
pthread_mutex_lock(&env->me_txns->mt_wmutex);
env->me_txns->mt_txnid++;
- txn->mt_free_pgs = malloc(MDB_IDL_UM_SIZEOF);
- if (txn->mt_free_pgs == NULL) {
- free(txn->mt_u.dirty_queue);
- free(txn);
- return ENOMEM;
- }
+ txn->mt_free_pgs = env->me_free_pgs;
txn->mt_free_pgs[0] = 0;
}
if (!r) {
unsigned int i;
pthread_mutex_lock(&env->me_txns->mt_mutex);
- for (i=0; i<env->me_maxreaders; i++) {
- if (env->me_txns->mt_readers[i].mr_pid == 0) {
- env->me_txns->mt_readers[i].mr_pid = getpid();
- env->me_txns->mt_readers[i].mr_tid = pthread_self();
- r = &env->me_txns->mt_readers[i];
- pthread_setspecific(env->me_txkey, r);
- if (i >= env->me_txns->mt_numreaders)
- env->me_txns->mt_numreaders = i+1;
+ for (i=0; i<env->me_txns->mt_numreaders; i++)
+ if (env->me_txns->mt_readers[i].mr_pid == 0)
break;
- }
- }
- pthread_mutex_unlock(&env->me_txns->mt_mutex);
if (i == env->me_maxreaders) {
+ pthread_mutex_unlock(&env->me_txns->mti_mutex);
return ENOSPC;
}
+ env->me_txns->mt_readers[i].mr_pid = getpid();
+ env->me_txns->mt_readers[i].mr_tid = pthread_self();
+ r = &env->me_txns->mt_readers[i];
+ pthread_setspecific(env->me_txkey, r);
+ if (i >= env->me_txns->mt_numreaders)
+ env->me_txns->mt_numreaders = i+1;
+ pthread_mutex_unlock(&env->me_txns->mt_mutex);
}
r->mr_txnid = txn->mt_txnid;
txn->mt_u.reader = r;
txn->mt_u.reader->mr_txnid = 0;
} else {
MDB_oldpages *mop;
+ unsigned int i;
+
/* Discard all dirty pages. */
while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
free(dp);
}
- free(txn->mt_free_pgs);
free(txn->mt_u.dirty_queue);
while ((mop = txn->mt_env->me_pghead)) {
env->me_txn = NULL;
env->me_txns->mt_txnid--;
+ for (i=2; i<env->me_numdbs; i++)
+ env->me_dbxs[i].md_dirty = 0;
pthread_mutex_unlock(&env->me_txns->mt_wmutex);
}
MDB_val data;
data.mv_size = sizeof(MDB_db);
- for (i = 2; i < env->me_numdbs; i++) {
- if (env->me_dbs[env->me_db_toggle][i].md_root != txn->mt_dbs[i].md_root) {
+ for (i = 2; i < txn->mt_numdbs; i++) {
+ if (txn->mt_dbxs[i].md_dirty) {
data.mv_data = &txn->mt_dbs[i];
mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
}
}
- for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
- data.mv_data = &txn->mt_dbs[i];
- mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
- }
}
/* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
{
int toggle = !env->me_db_toggle;
- for (i = 0; i < env->me_numdbs; i++) {
- if (env->me_dbs[toggle][i].md_root != txn->mt_dbs[i].md_root)
+ for (i = 2; i < env->me_numdbs; i++) {
+ if (txn->mt_dbxs[i].md_dirty) {
env->me_dbs[toggle][i] = txn->mt_dbs[i];
+ txn->mt_dbxs[i].md_dirty = 0;
+ }
}
for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
+ txn->mt_dbxs[i].md_dirty = 0;
env->me_dbxs[i] = txn->mt_dbxs[i];
env->me_dbs[toggle][i] = txn->mt_dbs[i];
}
}
pthread_mutex_unlock(&env->me_txns->mt_wmutex);
- free(txn->mt_free_pgs);
free(txn->mt_u.dirty_queue);
free(txn);
txn = NULL;
}
static MDB_page *
-mdbenv_get_page(MDB_env *env, pgno_t pgno)
+mdb_get_page(MDB_txn *txn, pgno_t pgno)
{
MDB_page *p = NULL;
- MDB_txn *txn = env->me_txn;
int found = 0;
- if (txn && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
+ if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
MDB_dpage *dp;
STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
if (dp->p.mp_pgno == pgno) {
}
}
if (!found) {
- p = (MDB_page *)(env->me_map + env->me_psize * pgno);
+ p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
}
return p;
}
CURSOR_TOP(cursor)->mp_ki = i;
mpp->mp_parent = mp;
- if ((mp = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
+ if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
return MDB_FAIL;
mpp->mp_pi = i;
mpp->mp_page = mp;
return ENOENT;
}
- if ((mpp->mp_page = mdbenv_get_page(txn->mt_env, root)) == NULL)
+ if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
return MDB_FAIL;
DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
if (modify) {
/* For sub-databases, update main root first */
- if (dbi > MAIN_DBI && txn->mt_dbs[dbi].md_root ==
- txn->mt_env->me_dbs[txn->mt_env->me_db_toggle][dbi].md_root) {
+ if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
MDB_pageparent mp2;
rc = mdb_search_page(txn, 0, &txn->mt_dbxs[dbi].md_name,
NULL, 1, &mp2);
if (rc)
return rc;
+ txn->mt_dbxs[dbi].md_dirty = 1;
}
if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
mpp->mp_parent = NULL;
}
static int
-mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data)
+mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
{
MDB_page *omp; /* overflow mpage */
pgno_t pgno;
*/
data->mv_size = leaf->mn_dsize;
memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
- if ((omp = mdbenv_get_page(env, pgno)) == NULL) {
+ if ((omp = mdb_get_page(txn, pgno)) == NULL) {
DPRINTF("read overflow page %lu failed", pgno);
return MDB_FAIL;
}
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
if (leaf && exact)
- rc = mdb_read_data(txn->mt_env, leaf, data);
+ rc = mdb_read_data(txn, leaf, data);
else {
rc = ENOENT;
}
assert(IS_BRANCH(parent->mp_page));
indx = NODEPTR(parent->mp_page, parent->mp_ki);
- if ((mp = mdbenv_get_page(cursor->mc_txn->mt_env, indx->mn_pgno)) == NULL)
+ if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
return MDB_FAIL;
#if 0
mp->parent = parent->mp_page;
assert(IS_LEAF(mp));
leaf = NODEPTR(mp, top->mp_ki);
- if (data && mdb_read_data(cursor->mc_txn->mt_env, leaf, data) != MDB_SUCCESS)
+ if (data && mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS)
+ return MDB_FAIL;
+
+ return mdb_set_key(leaf, key);
+}
+
+static int
+mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+{
+ MDB_ppage *top;
+ MDB_page *mp;
+ MDB_node *leaf;
+
+ assert(cursor->mc_initialized);
+
+ top = CURSOR_TOP(cursor);
+ mp = top->mp_page;
+
+ DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
+
+ if (top->mp_ki == 0) {
+ DPRINTF("=====> move to prev sibling page");
+ if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
+ return ENOENT;
+ }
+ top = CURSOR_TOP(cursor);
+ mp = top->mp_page;
+ top->mp_ki = NUMKEYS(mp) - 1;
+ DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
+ } else
+ top->mp_ki--;
+
+ cursor->mc_eof = 0;
+
+ DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
+ mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
+
+ assert(IS_LEAF(mp));
+ leaf = NODEPTR(mp, top->mp_ki);
+
+ if (data && mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS)
return MDB_FAIL;
return mdb_set_key(leaf, key);
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
rc = mdb_set_key(leaf, key);
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
return mdb_set_key(leaf, key);
mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
{
int rc;
+ MDB_ppage *top;
MDB_pageparent mpp;
MDB_node *leaf;
MDB_val lkey;
leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
cursor->mc_initialized = 1;
- cursor->mc_eof = 1;
+ cursor->mc_eof = 0;
+
+ top = CURSOR_TOP(cursor);
+ top->mp_ki = NUMKEYS(top->mp_page) - 1;
- if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
return mdb_set_key(leaf, key);
else
rc = mdb_cursor_next(cursor, key, data);
break;
+ case MDB_PREV:
+ if (!cursor->mc_initialized || cursor->mc_eof) {
+ while (CURSOR_TOP(cursor) != NULL)
+ cursor_pop_page(cursor);
+ rc = mdb_cursor_last(cursor, key, data);
+ } else
+ rc = mdb_cursor_prev(cursor, key, data);
+ break;
case MDB_FIRST:
while (CURSOR_TOP(cursor) != NULL)
cursor_pop_page(cursor);
mp->mp_upper += sz;
}
+static void
+mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
+{
+ MDB_dbi dbn;
+
+ mx->mx_cursor.mc_txn = &mx->mx_txn;
+ mx->mx_txn = *txn;
+ mx->mx_txn.mt_dbxs = mx->mx_dbxs;
+ mx->mx_txn.mt_dbs = mx->mx_dbs;
+ mx->mx_dbxs[0] = txn->mt_dbxs[0];
+ mx->mx_dbxs[1] = txn->mt_dbxs[1];
+ if (dbi > 1) {
+ mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
+ dbn = 2;
+ } else {
+ dbn = 1;
+ }
+ mx->mx_dbxs[dbn+1].md_parent = dbn;
+ mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
+ mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
+ mx->mx_dbxs[dbn+1].md_dirty = 0;
+ mx->mx_txn.mt_numdbs = dbn+2;
+}
+
+static void
+mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_db *db)
+{
+ mx->mx_dbs[0] = txn->mt_dbs[0];
+ mx->mx_dbs[1] = txn->mt_dbs[1];
+ if (dbi > 1) {
+ mx->mx_dbs[2] = txn->mt_dbs[dbi];
+ mx->mx_dbs[3] = *db;
+ } else {
+ mx->mx_dbs[2] = *db;
+ }
+}
+
+static void
+mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
+{
+ txn->mt_dbs[0] = mx->mx_dbs[0];
+ txn->mt_dbs[1] = mx->mx_dbs[1];
+ txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
+ txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
+ if (dbi > 1) {
+ txn->mt_dbs[dbi] = mx->mx_dbs[2];
+ txn->mt_dbxs[2].md_dirty = mx->mx_dbxs[2].md_dirty;
+ }
+}
+
int
mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
{
MDB_cursor *cursor;
+ size_t size = sizeof(MDB_cursor);
- if (txn == NULL || ret == NULL)
+ if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
return EINVAL;
- if ((cursor = calloc(1, sizeof(*cursor))) != NULL) {
+ if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
+ size += sizeof(MDB_xcursor);
+
+ if ((cursor = calloc(1, size)) != NULL) {
SLIST_INIT(&cursor->mc_stack);
cursor->mc_dbi = dbi;
cursor->mc_txn = txn;
+ if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
+ MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
+ cursor->mc_xcursor = mx;
+ mdb_xcursor_init0(txn, dbi, mx);
+ }
+ } else {
+ return ENOMEM;
}
*ret = cursor;
} else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
DPRINTF("collapsing root page!");
txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
- if ((root = mdbenv_get_page(txn->mt_env, txn->mt_dbs[dbi].md_root)) == NULL)
+ if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
return MDB_FAIL;
txn->mt_dbs[dbi].md_depth--;
txn->mt_dbs[dbi].md_branch_pages--;
*/
DPRINTF("reading right neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
- if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
+ if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
return MDB_FAIL;
npp.mp_pi = mpp->mp_pi + 1;
si = 0;
*/
DPRINTF("reading left neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
- if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
+ if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
return MDB_FAIL;
npp.mp_pi = mpp->mp_pi - 1;
si = NUMKEYS(npp.mp_page) - 1;
return ENOENT;
}
- if (data && (rc = mdb_read_data(txn->mt_env, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
return rc;
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ /* add all the child DB's pages to the free list */
+ MDB_cursor mc;
+ MDB_xcursor mx;
+ MDB_pageparent mp2;
+
+ mdb_xcursor_init0(txn, dbi, &mx);
+ mdb_xcursor_init1(txn, dbi, &mx, NODEDATA(leaf));
+ SLIST_INIT(&mc.mc_stack);
+ mc.mc_dbi = mx.mx_txn.mt_numdbs-1;
+ mc.mc_txn = &mx.mx_txn;
+ rc = mdb_search_page(&mx.mx_txn, mx.mx_txn.mt_numdbs - 1, NULL, &mc, 0, &mp2);
+ if (rc == MDB_SUCCESS) {
+ MDB_ppage *top, *parent;
+ MDB_node *ni;
+ unsigned int i;
+
+ cursor_pop_page(&mc);
+ top = CURSOR_TOP(&mc);
+ parent = SLIST_NEXT(top, mp_entry);
+ do {
+ for (i=0; i<NUMKEYS(top->mp_page); i++) {
+ ni = NODEPTR(top->mp_page, i);
+ mdb_idl_insert(txn->mt_free_pgs, ni->mn_pgno);
+ }
+ if (parent) {
+ parent->mp_ki++;
+ if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
+ cursor_pop_page(&mc);
+ top = CURSOR_TOP(&mc);
+ parent = SLIST_NEXT(top, mp_entry);
+ } else {
+ ni = NODEPTR(parent->mp_page, parent->mp_ki);
+ top = mdb_get_page(mc.mc_txn, ni->mn_pgno);
+ }
+ }
+ }
+ mdb_idl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mc.mc_dbi].md_root);
+ }
+ }
+
return mdb_del0(txn, dbi, ki, &mpp, leaf);
}
unsigned int ki;
MDB_node *leaf;
MDB_pageparent mpp;
+ MDB_val xdata, *rdata;
+ MDB_db dummy;
assert(key != NULL);
assert(data != NULL);
if (rc == MDB_SUCCESS) {
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf && exact) {
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ goto put_sub;
+ }
if (F_ISSET(flags, MDB_NOOVERWRITE)) {
DPRINTF("duplicate key %.*s",
(int)key->mv_size, (char *)key->mv_data);
DPRINTF("there are %u keys, should insert new key at index %i",
NUMKEYS(mpp.mp_page), ki);
+ /* For sorted dups, the data item at this level is a DB record
+ * for a child DB; the actual data elements are stored as keys
+ * in the child DB.
+ */
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ rdata = &xdata;
+ xdata.mv_size = sizeof(MDB_db);
+ xdata.mv_data = &dummy;
+ memset(&dummy, 0, sizeof(dummy));
+ dummy.md_root = P_INVALID;
+ } else {
+ rdata = data;
+ }
+
if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, data)) {
rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, data, P_INVALID);
} else {
if (rc != MDB_SUCCESS)
txn->mt_flags |= MDB_TXN_ERROR;
- else
+ else {
txn->mt_dbs[dbi].md_entries++;
+ /* Now store the actual data in the child DB. Note that we're
+ * storing the user data in the keys field, so there are strict
+ * size limits on dupdata. The actual data fields of the child
+ * DB are all zero size.
+ */
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ MDB_xcursor mx;
+
+ leaf = NODEPTR(mpp.mp_page, ki);
+put_sub:
+ mdb_xcursor_init0(txn, dbi, &mx);
+ mdb_xcursor_init1(txn, dbi, &mx, NODEDATA(leaf));
+ xdata.mv_size = 0;
+ xdata.mv_data = "";
+ rc = mdb_put(&mx.mx_txn, mx.mx_txn.mt_numdbs-1, data, &xdata, flags);
+ mdb_xcursor_fini(txn, dbi, &mx);
+ }
+ }
done:
return rc;
txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
+ txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
+ txn->mt_dbxs[txn->mt_numdbs].md_dirty = 0;
memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
*dbi = txn->mt_numdbs;
txn->mt_numdbs++;