#define MDB_MINKEYS 4
#define MDB_MAGIC 0xBEEFC0DE
#define MDB_VERSION 1
-#define MAXKEYSIZE 255
+#define MAXKEYSIZE 511
#define P_INVALID (~0UL)
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 mdb_set_key(MDB_node *node, 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);
+ MDB_val *key, MDB_val *data, MDB_cursor_op op);
static int mdb_cursor_prev(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data);
+ MDB_val *key, MDB_val *data, MDB_cursor_op op);
static int mdb_cursor_set(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data, int *exactp);
+ MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
static int mdb_cursor_first(MDB_cursor *cursor,
MDB_val *key, MDB_val *data);
static int mdb_cursor_last(MDB_cursor *cursor,
MDB_val *key, MDB_val *data);
+static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
+static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_db *db);
+static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
+
static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
MDB_val *data);
static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
MDB_txn *txn;
int rc, toggle;
- if ((txn = calloc(1, sizeof(*txn))) == NULL) {
+ if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
DPRINTF("calloc: %s", strerror(errno));
return ENOMEM;
}
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;
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)) {
}
pthread_mutex_unlock(&env->me_txns->mt_wmutex);
- free(txn->mt_free_pgs);
free(txn->mt_u.dirty_queue);
free(txn);
txn = NULL;
if (m->mm_version != MDB_VERSION) {
DPRINTF("database is version %u, expected version %u",
m->mm_version, MDB_VERSION);
- return EINVAL;
+ return MDB_VERSION_MISMATCH;
}
memcpy(meta, m, sizeof(*m));
{
MDB_env *e;
- e = calloc(1, sizeof(*e));
+ e = calloc(1, sizeof(MDB_env));
if (!e) return ENOMEM;
e->me_maxreaders = DEFAULT_READERS;
} else {
if (env->me_txns->mt_magic != MDB_MAGIC) {
DPRINTF("lock region has invalid magic");
- errno = EINVAL;
+ rc = EINVAL;
+ goto fail;
}
if (env->me_txns->mt_version != MDB_VERSION) {
DPRINTF("lock region is version %u, expected version %u",
env->me_txns->mt_version, MDB_VERSION);
- errno = EINVAL;
+ rc = MDB_VERSION_MISMATCH;
+ goto fail;
}
if (errno != EACCES && errno != EAGAIN) {
rc = errno;
DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
- if ((ppage = calloc(1, sizeof(*ppage))) == NULL)
+ if ((ppage = calloc(1, sizeof(MDB_ppage))) == NULL)
return NULL;
ppage->mp_page = mp;
CURSOR_PUSH(cursor, ppage);
}
}
if (!found) {
+ if (pgno > txn->mt_env->me_meta->mm_last_pg)
+ return NULL;
p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
}
return p;
if (root == P_INVALID) { /* Tree is empty. */
DPRINTF("tree is empty");
- return ENOENT;
+ return MDB_NOTFOUND;
}
if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
return rc;
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
- if (leaf && exact)
+ if (leaf && exact) {
+ /* Return first duplicate data item */
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ MDB_xcursor mx;
+
+ mdb_xcursor_init0(txn, dbi, &mx);
+ mdb_xcursor_init1(txn, dbi, &mx, NODEDATA(leaf));
+ rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ leaf = NODEPTR(mpp.mp_page, 0);
+ }
rc = mdb_read_data(txn, leaf, data);
- else {
- rc = ENOENT;
+ } else {
+ rc = MDB_NOTFOUND;
}
return rc;
top = CURSOR_TOP(cursor);
if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
- return ENOENT; /* root has no siblings */
+ return MDB_NOTFOUND; /* root has no siblings */
}
DPRINTF("parent page is page %lu, index %u",
}
static int
-mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
MDB_ppage *top;
MDB_page *mp;
MDB_node *leaf;
+ int rc;
if (cursor->mc_eof) {
- return ENOENT;
+ return MDB_NOTFOUND;
}
assert(cursor->mc_initialized);
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
+ rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
+ if (op != MDB_NEXT || rc == MDB_SUCCESS)
+ return rc;
+ }
+ }
+
top = CURSOR_TOP(cursor);
mp = top->mp_page;
DPRINTF("=====> move to next sibling page");
if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
cursor->mc_eof = 1;
- return ENOENT;
+ return MDB_NOTFOUND;
}
top = CURSOR_TOP(cursor);
mp = top->mp_page;
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;
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
+ return rc;
+
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, NODEDATA(leaf));
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
static int
-mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
MDB_ppage *top;
MDB_page *mp;
MDB_node *leaf;
+ int rc;
assert(cursor->mc_initialized);
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ if (op == MDB_PREV || op == MDB_PREV_DUP) {
+ rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
+ if (op != MDB_PREV || rc == MDB_SUCCESS)
+ return rc;
+ }
+ }
+
top = CURSOR_TOP(cursor);
mp = top->mp_page;
if (top->mp_ki == 0) {
DPRINTF("=====> move to prev sibling page");
if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
- return ENOENT;
+ return MDB_NOTFOUND;
}
top = CURSOR_TOP(cursor);
mp = top->mp_page;
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;
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
+ return rc;
+
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, NODEDATA(leaf));
+ rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
static int
mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
- int *exactp)
+ MDB_cursor_op op, int *exactp)
{
int rc;
MDB_node *leaf;
assert(key);
assert(key->mv_size > 0);
+ while (CURSOR_TOP(cursor) != NULL)
+ cursor_pop_page(cursor);
+
rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
if (exactp != NULL && !*exactp) {
/* MDB_SET specified and not an exact match. */
- return ENOENT;
+ return MDB_NOTFOUND;
}
if (leaf == NULL) {
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, NODEDATA(leaf));
+ if (op == MDB_SET || op == MDB_SET_RANGE) {
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ } else {
+ int ex2, *ex2p;
+ MDB_cursor_op op2;
+ if (op == MDB_GET_BOTH) {
+ ex2p = &ex2;
+ op2 = MDB_SET;
+ } else {
+ ex2p = NULL;
+ op2 = MDB_SET_RANGE;
+ }
+ rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ }
+ }
rc = mdb_set_key(leaf, key);
if (rc == MDB_SUCCESS) {
MDB_pageparent mpp;
MDB_node *leaf;
+ while (CURSOR_TOP(cursor) != NULL)
+ cursor_pop_page(cursor);
+
rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, NODEDATA(leaf));
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
MDB_node *leaf;
MDB_val lkey;
+ while (CURSOR_TOP(cursor) != NULL)
+ cursor_pop_page(cursor);
+
lkey.mv_size = MAXKEYSIZE+1;
lkey.mv_data = NULL;
top = CURSOR_TOP(cursor);
top->mp_ki = NUMKEYS(top->mp_page) - 1;
- if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, NODEDATA(leaf));
+ rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
assert(cursor);
switch (op) {
+ case MDB_GET_BOTH:
+ case MDB_GET_BOTH_RANGE:
+ if (data == NULL) {
+ rc = EINVAL;
+ break;
+ }
+ /* FALLTHRU */
case MDB_SET:
case MDB_SET_RANGE:
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
rc = EINVAL;
- } else if (op == MDB_SET)
- rc = mdb_cursor_set(cursor, key, data, &exact);
+ } else if (op != MDB_SET_RANGE)
+ rc = mdb_cursor_set(cursor, key, data, op, NULL);
else
- rc = mdb_cursor_set(cursor, key, data, NULL);
+ rc = mdb_cursor_set(cursor, key, data, op, &exact);
break;
case MDB_NEXT:
+ case MDB_NEXT_DUP:
+ case MDB_NEXT_NODUP:
if (!cursor->mc_initialized)
rc = mdb_cursor_first(cursor, key, data);
else
- rc = mdb_cursor_next(cursor, key, data);
+ rc = mdb_cursor_next(cursor, key, data, op);
break;
case MDB_PREV:
- if (!cursor->mc_initialized || cursor->mc_eof) {
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
+ case MDB_PREV_DUP:
+ case MDB_PREV_NODUP:
+ if (!cursor->mc_initialized || cursor->mc_eof)
rc = mdb_cursor_last(cursor, key, data);
- } else
- rc = mdb_cursor_prev(cursor, key, data);
+ else
+ rc = mdb_cursor_prev(cursor, key, data, op);
break;
case MDB_FIRST:
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
rc = mdb_cursor_first(cursor, key, data);
break;
case MDB_LAST:
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
rc = mdb_cursor_last(cursor, key, data);
break;
default:
mp->mp_upper += sz;
}
+static void
+mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
+{
+ MDB_dbi dbn;
+
+ 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;
+
+ SLIST_INIT(&mx->mx_cursor.mc_stack);
+ mx->mx_cursor.mc_txn = &mx->mx_txn;
+ mx->mx_cursor.mc_dbi = dbn+1;
+}
+
+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[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
+ }
+}
+
int
mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
{
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
cursor->mc_xcursor = mx;
- 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];
- mx->mx_txn.mt_numdbs = 4;
- } else {
- mx->mx_txn.mt_numdbs = 3;
- }
+ mdb_xcursor_init0(txn, dbi, mx);
}
} else {
return ENOMEM;
mdb_cursor_close(MDB_cursor *cursor)
{
if (cursor != NULL) {
- while (!CURSOR_EMPTY(cursor))
+ while(!CURSOR_EMPTY(cursor))
cursor_pop_page(cursor);
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ mdb_xcursor_fini(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor);
+ while(!CURSOR_EMPTY(&cursor->mc_xcursor->mx_cursor))
+ cursor_pop_page(&cursor->mc_xcursor->mx_cursor);
+ }
-/* btree_close(cursor->bt); */
free(cursor);
}
}
int
mdb_del(MDB_txn *txn, MDB_dbi dbi,
- MDB_val *key, MDB_val *data)
+ MDB_val *key, MDB_val *data,
+ unsigned int flags)
{
int rc, exact;
unsigned int ki;
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf == NULL || !exact) {
- return ENOENT;
+ return MDB_NOTFOUND;
+ }
+
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ MDB_xcursor mx;
+ MDB_pageparent mp2;
+
+ mdb_xcursor_init0(txn, dbi, &mx);
+ mdb_xcursor_init1(txn, dbi, &mx, NODEDATA(leaf));
+ if (flags == MDB_DEL_DUP) {
+ rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
+ mdb_xcursor_fini(txn, dbi, &mx);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ /* If sub-DB still has entries, we're done */
+ if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
+ memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
+ sizeof(MDB_db));
+ return rc;
+ }
+ /* otherwise fall thru and delete the sub-DB */
+ } else {
+ /* add all the child DB's pages to the free list */
+ rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
+ NULL, &mx.mx_cursor, 0, &mp2);
+ if (rc == MDB_SUCCESS) {
+ MDB_ppage *top, *parent;
+ MDB_node *ni;
+ unsigned int i;
+
+ cursor_pop_page(&mx.mx_cursor);
+ top = CURSOR_TOP(&mx.mx_cursor);
+ if (top != NULL) {
+ parent = SLIST_NEXT(top, mp_entry);
+ while (parent != NULL) {
+ 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(&mx.mx_cursor);
+ top = CURSOR_TOP(&mx.mx_cursor);
+ parent = SLIST_NEXT(top, mp_entry);
+ } else {
+ ni = NODEPTR(parent->mp_page, parent->mp_ki);
+ top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
+ }
+ }
+ }
+ }
+ mdb_idl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
+ }
+ }
}
if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
unsigned int ki;
MDB_node *leaf;
MDB_pageparent mpp;
+ MDB_val xdata, *rdata;
+ MDB_db dummy;
assert(key != NULL);
assert(data != NULL);
return EINVAL;
}
- if (txn->mt_env->me_txn != txn) {
- return EINVAL;
- }
-
if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
return EINVAL;
}
if (rc == MDB_SUCCESS) {
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf && exact) {
- if (F_ISSET(flags, MDB_NOOVERWRITE)) {
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ goto put_sub;
+ }
+ if (flags == MDB_NOOVERWRITE) {
DPRINTF("duplicate key %.*s",
(int)key->mv_size, (char *)key->mv_data);
- return EEXIST;
+ return MDB_KEYEXIST;
}
/* same size, just replace it */
if (NODEDSZ(leaf) == data->mv_size) {
ki = NUMKEYS(mpp.mp_page);
DPRINTF("appending key at index %i", ki);
}
- } else if (rc == ENOENT) {
+ } else if (rc == MDB_NOTFOUND) {
MDB_dpage *dp;
/* new file, just write a root leaf page */
DPRINTF("allocating new root leaf page");
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 = "";
+ if (flags == MDB_NODUPDATA)
+ flags = MDB_NOOVERWRITE;
+ rc = mdb_put(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
+ mdb_xcursor_fini(txn, dbi, &mx);
+ memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
+ sizeof(MDB_db));
+ }
+ }
done:
return rc;
rc = mdb_get(txn, MAIN_DBI, &key, &data);
/* Create if requested */
- if (rc == ENOENT && (flags & MDB_CREATE)) {
+ if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
MDB_db dummy;
data.mv_size = sizeof(MDB_db);
data.mv_data = &dummy;
projects / openldap / blobdiff