diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-27 11:11:40 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-27 11:11:40 +0000 |
commit | 7731832751ab9f3c6ddeb66f186d3d7fa1934a6d (patch) | |
tree | e91015872543a59be2aad26c2fea02e41b57005d /servers/slapd/backglue.c | |
parent | Initial commit. (diff) | |
download | openldap-7731832751ab9f3c6ddeb66f186d3d7fa1934a6d.tar.xz openldap-7731832751ab9f3c6ddeb66f186d3d7fa1934a6d.zip |
Adding upstream version 2.4.57+dfsg.upstream/2.4.57+dfsgupstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'servers/slapd/backglue.c')
-rw-r--r-- | servers/slapd/backglue.c | 1549 |
1 files changed, 1549 insertions, 0 deletions
diff --git a/servers/slapd/backglue.c b/servers/slapd/backglue.c new file mode 100644 index 0000000..1c3354e --- /dev/null +++ b/servers/slapd/backglue.c @@ -0,0 +1,1549 @@ +/* backglue.c - backend glue */ +/* $OpenLDAP$ */ +/* This work is part of OpenLDAP Software <http://www.openldap.org/>. + * + * Copyright 2001-2021 The OpenLDAP Foundation. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted only as authorized by the OpenLDAP + * Public License. + * + * A copy of this license is available in the file LICENSE in the + * top-level directory of the distribution or, alternatively, at + * <http://www.OpenLDAP.org/license.html>. + */ + +/* + * Functions to glue a bunch of other backends into a single tree. + * All of the glued backends must share a common suffix. E.g., you + * can glue o=foo and ou=bar,o=foo but you can't glue o=foo and o=bar. + * + * The purpose of these functions is to allow you to split a single database + * into pieces (for load balancing purposes, whatever) but still be able + * to treat it as a single database after it's been split. As such, each + * of the glued backends should have identical rootdn. + * -- Howard Chu + */ + +#include "portable.h" + +#include <stdio.h> + +#include <ac/string.h> +#include <ac/socket.h> + +#define SLAPD_TOOLS +#include "slap.h" +#include "lutil.h" +#include "config.h" + +typedef struct gluenode { + BackendDB *gn_be; + struct berval gn_pdn; +} gluenode; + +typedef struct glueinfo { + int gi_nodes; + struct berval gi_pdn; + gluenode gi_n[1]; +} glueinfo; + +static slap_overinst glue; + +static int glueMode; +static BackendDB *glueBack; +static BackendDB glueBackDone; +#define GLUEBACK_DONE (&glueBackDone) + +static slap_overinst * glue_tool_inst( BackendInfo *bi); + +static slap_response glue_op_response; + +/* Just like select_backend, but only for our backends */ +static BackendDB * +glue_back_select ( + BackendDB *be, + struct berval *dn +) +{ + slap_overinst *on = (slap_overinst *)be->bd_info; + glueinfo *gi = (glueinfo *)on->on_bi.bi_private; + int i; + + for (i = gi->gi_nodes-1; i >= 0; i--) { + assert( gi->gi_n[i].gn_be->be_nsuffix != NULL ); + + if (dnIsSuffix(dn, &gi->gi_n[i].gn_be->be_nsuffix[0])) { + return gi->gi_n[i].gn_be; + } + } + be->bd_info = on->on_info->oi_orig; + return be; +} + + +typedef struct glue_state { + char *matched; + BerVarray refs; + LDAPControl **ctrls; + int err; + int matchlen; + int nrefs; + int nctrls; +} glue_state; + +static int +glue_op_cleanup( Operation *op, SlapReply *rs ) +{ + /* This is not a final result */ + if (rs->sr_type == REP_RESULT ) + rs->sr_type = REP_GLUE_RESULT; + return SLAP_CB_CONTINUE; +} + +static int +glue_op_response ( Operation *op, SlapReply *rs ) +{ + glue_state *gs = op->o_callback->sc_private; + + switch(rs->sr_type) { + case REP_SEARCH: + case REP_SEARCHREF: + case REP_INTERMEDIATE: + return SLAP_CB_CONTINUE; + + default: + if (rs->sr_err == LDAP_SUCCESS || + rs->sr_err == LDAP_SIZELIMIT_EXCEEDED || + rs->sr_err == LDAP_TIMELIMIT_EXCEEDED || + rs->sr_err == LDAP_ADMINLIMIT_EXCEEDED || + rs->sr_err == LDAP_NO_SUCH_OBJECT || + gs->err != LDAP_SUCCESS) + gs->err = rs->sr_err; + if (gs->err == LDAP_SUCCESS && gs->matched) { + ch_free (gs->matched); + gs->matched = NULL; + gs->matchlen = 0; + } + if (gs->err != LDAP_SUCCESS && rs->sr_matched) { + int len; + len = strlen (rs->sr_matched); + if (len > gs->matchlen) { + if (gs->matched) + ch_free (gs->matched); + gs->matched = ch_strdup (rs->sr_matched); + gs->matchlen = len; + } + } + if (rs->sr_ref) { + int i, j, k; + BerVarray new; + + for (i=0; rs->sr_ref[i].bv_val; i++); + + j = gs->nrefs; + if (!j) { + new = ch_malloc ((i+1)*sizeof(struct berval)); + } else { + new = ch_realloc(gs->refs, + (j+i+1)*sizeof(struct berval)); + } + for (k=0; k<i; j++,k++) { + ber_dupbv( &new[j], &rs->sr_ref[k] ); + } + new[j].bv_val = NULL; + gs->nrefs = j; + gs->refs = new; + } + if (rs->sr_ctrls) { + int i, j, k; + LDAPControl **newctrls; + + for (i=0; rs->sr_ctrls[i]; i++); + + j = gs->nctrls; + if (!j) { + newctrls = op->o_tmpalloc((i+1)*sizeof(LDAPControl *), + op->o_tmpmemctx); + } else { + /* Forget old pagedResults response if we're sending + * a new one now + */ + if ( get_pagedresults( op ) > SLAP_CONTROL_IGNORED ) { + int newpage = 0; + for ( k=0; k<i; k++ ) { + if ( !strcmp(rs->sr_ctrls[k]->ldctl_oid, + LDAP_CONTROL_PAGEDRESULTS )) { + newpage = 1; + break; + } + } + if ( newpage ) { + for ( k=0; k<j; k++ ) { + if ( !strcmp(gs->ctrls[k]->ldctl_oid, + LDAP_CONTROL_PAGEDRESULTS )) + { + op->o_tmpfree(gs->ctrls[k], op->o_tmpmemctx); + gs->ctrls[k] = gs->ctrls[--j]; + gs->ctrls[j] = NULL; + break; + } + } + } + } + newctrls = op->o_tmprealloc(gs->ctrls, + (j+i+1)*sizeof(LDAPControl *), op->o_tmpmemctx); + } + for (k=0; k<i; j++,k++) { + ber_len_t oidlen = strlen( rs->sr_ctrls[k]->ldctl_oid ); + newctrls[j] = op->o_tmpalloc(sizeof(LDAPControl) + oidlen + 1 + rs->sr_ctrls[k]->ldctl_value.bv_len + 1, + op->o_tmpmemctx); + newctrls[j]->ldctl_iscritical = rs->sr_ctrls[k]->ldctl_iscritical; + newctrls[j]->ldctl_oid = (char *)&newctrls[j][1]; + lutil_strcopy( newctrls[j]->ldctl_oid, rs->sr_ctrls[k]->ldctl_oid ); + if ( !BER_BVISNULL( &rs->sr_ctrls[k]->ldctl_value ) ) { + newctrls[j]->ldctl_value.bv_val = &newctrls[j]->ldctl_oid[oidlen + 1]; + newctrls[j]->ldctl_value.bv_len = rs->sr_ctrls[k]->ldctl_value.bv_len; + lutil_memcopy( newctrls[j]->ldctl_value.bv_val, + rs->sr_ctrls[k]->ldctl_value.bv_val, + rs->sr_ctrls[k]->ldctl_value.bv_len + 1 ); + } else { + BER_BVZERO( &newctrls[j]->ldctl_value ); + } + } + newctrls[j] = NULL; + gs->nctrls = j; + gs->ctrls = newctrls; + } + } + return 0; +} + +static int +glue_op_func ( Operation *op, SlapReply *rs ) +{ + slap_overinst *on = (slap_overinst *)op->o_bd->bd_info; + BackendDB *b0 = op->o_bd; + BackendInfo *bi0 = op->o_bd->bd_info; + BI_op_modify **func; + slap_operation_t which = op_bind; + int rc; + + op->o_bd = glue_back_select (b0, &op->o_req_ndn); + + /* If we're on the primary backend, let overlay framework handle it */ + if ( op->o_bd == b0 ) + return SLAP_CB_CONTINUE; + + b0->bd_info = on->on_info->oi_orig; + + switch(op->o_tag) { + case LDAP_REQ_ADD: which = op_add; break; + case LDAP_REQ_DELETE: which = op_delete; break; + case LDAP_REQ_MODIFY: which = op_modify; break; + case LDAP_REQ_MODRDN: which = op_modrdn; break; + case LDAP_REQ_EXTENDED: which = op_extended; break; + default: assert( 0 ); break; + } + + func = &op->o_bd->bd_info->bi_op_bind; + if ( func[which] ) + rc = func[which]( op, rs ); + else + rc = SLAP_CB_BYPASS; + + op->o_bd = b0; + op->o_bd->bd_info = bi0; + return rc; +} + +static int +glue_op_abandon( Operation *op, SlapReply *rs ) +{ + slap_overinst *on = (slap_overinst *)op->o_bd->bd_info; + glueinfo *gi = (glueinfo *)on->on_bi.bi_private; + BackendDB *b0 = op->o_bd; + BackendInfo *bi0 = op->o_bd->bd_info; + int i; + + b0->bd_info = on->on_info->oi_orig; + + for (i = gi->gi_nodes-1; i >= 0; i--) { + assert( gi->gi_n[i].gn_be->be_nsuffix != NULL ); + op->o_bd = gi->gi_n[i].gn_be; + if ( op->o_bd == b0 ) + continue; + if ( op->o_bd->bd_info->bi_op_abandon ) + op->o_bd->bd_info->bi_op_abandon( op, rs ); + } + op->o_bd = b0; + op->o_bd->bd_info = bi0; + return SLAP_CB_CONTINUE; +} + +static int +glue_response ( Operation *op, SlapReply *rs ) +{ + BackendDB *be = op->o_bd; + be = glue_back_select (op->o_bd, &op->o_req_ndn); + + /* If we're on the primary backend, let overlay framework handle it. + * Otherwise, bail out. + */ + return ( op->o_bd == be ) ? SLAP_CB_CONTINUE : SLAP_CB_BYPASS; +} + +static int +glue_chk_referrals ( Operation *op, SlapReply *rs ) +{ + slap_overinst *on = (slap_overinst *)op->o_bd->bd_info; + BackendDB *b0 = op->o_bd; + BackendInfo *bi0 = op->o_bd->bd_info; + int rc; + + op->o_bd = glue_back_select (b0, &op->o_req_ndn); + if ( op->o_bd == b0 ) + return SLAP_CB_CONTINUE; + + b0->bd_info = on->on_info->oi_orig; + + if ( op->o_bd->bd_info->bi_chk_referrals ) + rc = ( *op->o_bd->bd_info->bi_chk_referrals )( op, rs ); + else + rc = SLAP_CB_CONTINUE; + + op->o_bd = b0; + op->o_bd->bd_info = bi0; + return rc; +} + +static int +glue_chk_controls ( Operation *op, SlapReply *rs ) +{ + slap_overinst *on = (slap_overinst *)op->o_bd->bd_info; + BackendDB *b0 = op->o_bd; + BackendInfo *bi0 = op->o_bd->bd_info; + int rc = SLAP_CB_CONTINUE; + + op->o_bd = glue_back_select (b0, &op->o_req_ndn); + if ( op->o_bd == b0 ) + return SLAP_CB_CONTINUE; + + b0->bd_info = on->on_info->oi_orig; + + /* if the subordinate database has overlays, the bi_chk_controls() + * hook is actually over_aux_chk_controls(); in case it actually + * wraps a missing hok, we need to mimic the behavior + * of the frontend applied to that database */ + if ( op->o_bd->bd_info->bi_chk_controls ) { + rc = ( *op->o_bd->bd_info->bi_chk_controls )( op, rs ); + } + + + if ( rc == SLAP_CB_CONTINUE ) { + rc = backend_check_controls( op, rs ); + } + + op->o_bd = b0; + op->o_bd->bd_info = bi0; + return rc; +} + +/* ITS#4615 - overlays configured above the glue overlay should be + * invoked for the entire glued tree. Overlays configured below the + * glue overlay should only be invoked on the primary backend. + * So, if we're searching on any subordinates, we need to force the + * current overlay chain to stop processing, without stopping the + * overall callback flow. + */ +static int +glue_sub_search( Operation *op, SlapReply *rs, BackendDB *b0, + slap_overinst *on ) +{ + /* Process any overlays on the primary backend */ + if ( op->o_bd == b0 && on->on_next ) { + BackendInfo *bi = op->o_bd->bd_info; + int rc = SLAP_CB_CONTINUE; + for ( on=on->on_next; on; on=on->on_next ) { + op->o_bd->bd_info = (BackendInfo *)on; + if ( on->on_bi.bi_op_search ) { + rc = on->on_bi.bi_op_search( op, rs ); + if ( rc != SLAP_CB_CONTINUE ) + break; + } + } + op->o_bd->bd_info = bi; + if ( rc != SLAP_CB_CONTINUE ) + return rc; + } + return op->o_bd->be_search( op, rs ); +} + +static const ID glueID = NOID; +static const struct berval gluecookie = { sizeof( glueID ), (char *)&glueID }; + +static int +glue_op_search ( Operation *op, SlapReply *rs ) +{ + slap_overinst *on = (slap_overinst *)op->o_bd->bd_info; + glueinfo *gi = (glueinfo *)on->on_bi.bi_private; + BackendDB *b0 = op->o_bd; + BackendDB *b1 = NULL, *btmp; + BackendInfo *bi0 = op->o_bd->bd_info; + int i; + long stoptime = 0, starttime; + glue_state gs = {NULL, NULL, NULL, 0, 0, 0, 0}; + slap_callback cb = { NULL, glue_op_response, glue_op_cleanup, NULL }; + int scope0, tlimit0; + struct berval dn, ndn, *pdn; + + cb.sc_private = &gs; + + cb.sc_next = op->o_callback; + + starttime = op->o_time; + stoptime = slap_get_time () + op->ors_tlimit; + + /* reset dummy cookie used to keep paged results going across databases */ + if ( get_pagedresults( op ) > SLAP_CONTROL_IGNORED + && bvmatch( &((PagedResultsState *)op->o_pagedresults_state)->ps_cookieval, &gluecookie ) ) + { + PagedResultsState *ps = op->o_pagedresults_state; + BerElementBuffer berbuf; + BerElement *ber = (BerElement *)&berbuf; + struct berval cookie = BER_BVC(""), value; + int c; + + for (c = 0; op->o_ctrls[c] != NULL; c++) { + if (strcmp(op->o_ctrls[c]->ldctl_oid, LDAP_CONTROL_PAGEDRESULTS) == 0) + break; + } + + assert( op->o_ctrls[c] != NULL ); + + ber_init2( ber, NULL, LBER_USE_DER ); + ber_printf( ber, "{iO}", ps->ps_size, &cookie ); + ber_flatten2( ber, &value, 0 ); + assert( op->o_ctrls[c]->ldctl_value.bv_len >= value.bv_len ); + op->o_ctrls[c]->ldctl_value.bv_len = value.bv_len; + lutil_memcopy( op->o_ctrls[c]->ldctl_value.bv_val, + value.bv_val, value.bv_len ); + ber_free_buf( ber ); + + ps->ps_cookie = (PagedResultsCookie)0; + BER_BVZERO( &ps->ps_cookieval ); + } + + op->o_bd = glue_back_select (b0, &op->o_req_ndn); + b0->bd_info = on->on_info->oi_orig; + + switch (op->ors_scope) { + case LDAP_SCOPE_BASE: + if ( op->o_bd == b0 ) + return SLAP_CB_CONTINUE; + + if (op->o_bd && op->o_bd->be_search) { + rs->sr_err = op->o_bd->be_search( op, rs ); + } else { + rs->sr_err = LDAP_UNWILLING_TO_PERFORM; + } + return rs->sr_err; + + case LDAP_SCOPE_ONELEVEL: + case LDAP_SCOPE_SUBTREE: + case LDAP_SCOPE_SUBORDINATE: /* FIXME */ + op->o_callback = &cb; + rs->sr_err = gs.err = LDAP_UNWILLING_TO_PERFORM; + scope0 = op->ors_scope; + tlimit0 = op->ors_tlimit; + dn = op->o_req_dn; + ndn = op->o_req_ndn; + b1 = op->o_bd; + + /* + * Execute in reverse order, most specific first + */ + for (i = gi->gi_nodes; i >= 0; i--) { + if ( i == gi->gi_nodes ) { + btmp = b0; + pdn = &gi->gi_pdn; + } else { + btmp = gi->gi_n[i].gn_be; + pdn = &gi->gi_n[i].gn_pdn; + } + if (!btmp || !btmp->be_search) + continue; + if (!dnIsSuffix(&btmp->be_nsuffix[0], &b1->be_nsuffix[0])) + continue; + if (get_no_subordinate_glue(op) && btmp != b1) + continue; + /* If we remembered which backend we were on before, + * skip down to it now + */ + if ( get_pagedresults( op ) > SLAP_CONTROL_IGNORED && + op->o_conn->c_pagedresults_state.ps_be && + op->o_conn->c_pagedresults_state.ps_be != btmp ) + continue; + + if (tlimit0 != SLAP_NO_LIMIT) { + op->o_time = slap_get_time(); + op->ors_tlimit = stoptime - op->o_time; + if (op->ors_tlimit <= 0) { + rs->sr_err = gs.err = LDAP_TIMELIMIT_EXCEEDED; + break; + } + } + rs->sr_err = 0; + /* + * check for abandon + */ + if (op->o_abandon) { + goto end_of_loop; + } + op->o_bd = btmp; + + assert( op->o_bd->be_suffix != NULL ); + assert( op->o_bd->be_nsuffix != NULL ); + + if (scope0 == LDAP_SCOPE_ONELEVEL && + dn_match(pdn, &ndn)) + { + struct berval mdn, mndn; + op->ors_scope = LDAP_SCOPE_BASE; + mdn = op->o_req_dn = op->o_bd->be_suffix[0]; + mndn = op->o_req_ndn = op->o_bd->be_nsuffix[0]; + rs->sr_err = op->o_bd->be_search(op, rs); + if ( rs->sr_err == LDAP_NO_SUCH_OBJECT ) { + gs.err = LDAP_SUCCESS; + } + op->ors_scope = LDAP_SCOPE_ONELEVEL; + if ( op->o_req_dn.bv_val == mdn.bv_val ) + op->o_req_dn = dn; + if ( op->o_req_ndn.bv_val == mndn.bv_val ) + op->o_req_ndn = ndn; + + } else if (scope0 == LDAP_SCOPE_SUBTREE && + dn_match(&op->o_bd->be_nsuffix[0], &ndn)) + { + rs->sr_err = glue_sub_search( op, rs, b0, on ); + + } else if (scope0 == LDAP_SCOPE_SUBTREE && + dnIsSuffix(&op->o_bd->be_nsuffix[0], &ndn)) + { + struct berval mdn, mndn; + mdn = op->o_req_dn = op->o_bd->be_suffix[0]; + mndn = op->o_req_ndn = op->o_bd->be_nsuffix[0]; + rs->sr_err = glue_sub_search( op, rs, b0, on ); + if ( rs->sr_err == LDAP_NO_SUCH_OBJECT ) { + gs.err = LDAP_SUCCESS; + } + if ( op->o_req_dn.bv_val == mdn.bv_val ) + op->o_req_dn = dn; + if ( op->o_req_ndn.bv_val == mndn.bv_val ) + op->o_req_ndn = ndn; + + } else if (dnIsSuffix(&ndn, &op->o_bd->be_nsuffix[0])) { + rs->sr_err = glue_sub_search( op, rs, b0, on ); + } + + switch ( gs.err ) { + + /* + * Add errors that should result in dropping + * the search + */ + case LDAP_SIZELIMIT_EXCEEDED: + case LDAP_TIMELIMIT_EXCEEDED: + case LDAP_ADMINLIMIT_EXCEEDED: + case LDAP_NO_SUCH_OBJECT: +#ifdef LDAP_CONTROL_X_CHAINING_BEHAVIOR + case LDAP_X_CANNOT_CHAIN: +#endif /* LDAP_CONTROL_X_CHAINING_BEHAVIOR */ + goto end_of_loop; + + case LDAP_SUCCESS: + if ( get_pagedresults( op ) > SLAP_CONTROL_IGNORED ) { + PagedResultsState *ps = op->o_pagedresults_state; + + /* Assume this backend can be forgotten now */ + op->o_conn->c_pagedresults_state.ps_be = NULL; + + /* If we have a full page, exit the loop. We may + * need to remember this backend so we can continue + * from here on a subsequent request. + */ + if ( rs->sr_nentries >= ps->ps_size ) { + PagedResultsState *cps = &op->o_conn->c_pagedresults_state; + + /* Don't bother to remember the first backend. + * Only remember the last one if there's more state left. + */ + if ( op->o_bd != b0 && + ( cps->ps_cookie != NOID + || !BER_BVISNULL( &cps->ps_cookieval ) + || op->o_bd != gi->gi_n[0].gn_be ) ) + { + op->o_conn->c_pagedresults_state.ps_be = op->o_bd; + } + + /* Check whether the cookie is empty, + * and give remaining databases a chance + */ + if ( op->o_bd != gi->gi_n[0].gn_be || cps->ps_cookie == NOID ) { + int c; + + for ( c = 0; gs.ctrls[c] != NULL; c++ ) { + if ( strcmp( gs.ctrls[c]->ldctl_oid, LDAP_CONTROL_PAGEDRESULTS ) == 0 ) { + break; + } + } + + if ( gs.ctrls[c] != NULL ) { + BerElementBuffer berbuf; + BerElement *ber = (BerElement *)&berbuf; + ber_tag_t tag; + ber_int_t size; + struct berval cookie, value; + + ber_init2( ber, &gs.ctrls[c]->ldctl_value, LBER_USE_DER ); + + tag = ber_scanf( ber, "{im}", &size, &cookie ); + assert( tag != LBER_ERROR ); + + if ( BER_BVISEMPTY( &cookie ) && op->o_bd != gi->gi_n[0].gn_be ) { + /* delete old, create new cookie with NOID */ + PagedResultsCookie respcookie = (PagedResultsCookie)NOID; + ber_len_t oidlen = strlen( gs.ctrls[c]->ldctl_oid ); + LDAPControl *newctrl; + + /* it's next database's turn */ + if ( btmp == b0 ) { + op->o_conn->c_pagedresults_state.ps_be = gi->gi_n[gi->gi_nodes - 1].gn_be; + + } else { + op->o_conn->c_pagedresults_state.ps_be = gi->gi_n[(i > 0 ? i - 1: 0)].gn_be; + } + + cookie.bv_val = (char *)&respcookie; + cookie.bv_len = sizeof( PagedResultsCookie ); + + ber_init2( ber, NULL, LBER_USE_DER ); + ber_printf( ber, "{iO}", 0, &cookie ); + ber_flatten2( ber, &value, 0 ); + + newctrl = op->o_tmprealloc( gs.ctrls[c], + sizeof(LDAPControl) + oidlen + 1 + value.bv_len + 1, + op->o_tmpmemctx); + newctrl->ldctl_iscritical = gs.ctrls[c]->ldctl_iscritical; + newctrl->ldctl_oid = (char *)&newctrl[1]; + lutil_strcopy( newctrl->ldctl_oid, gs.ctrls[c]->ldctl_oid ); + newctrl->ldctl_value.bv_len = value.bv_len; + lutil_memcopy( newctrl->ldctl_value.bv_val, + value.bv_val, value.bv_len ); + + gs.ctrls[c] = newctrl; + + ber_free_buf( ber ); + + } else if ( !BER_BVISEMPTY( &cookie ) && op->o_bd != b0 ) { + /* if cookie not empty, it's again this database's turn */ + op->o_conn->c_pagedresults_state.ps_be = op->o_bd; + } + } + } + + goto end_of_loop; + } + + /* This backend has run out of entries, but more responses + * can fit in the page. Fake a reset of the state so the + * next backend will start up properly. Only back-[bh]db + * and back-sql look at this state info. + */ + ps->ps_cookie = (PagedResultsCookie)0; + BER_BVZERO( &ps->ps_cookieval ); + + { + /* change the size of the page in the request + * that will be propagated, and reset the cookie */ + BerElementBuffer berbuf; + BerElement *ber = (BerElement *)&berbuf; + int size = ps->ps_size - rs->sr_nentries; + struct berval cookie = BER_BVC(""), value; + int c; + + for (c = 0; op->o_ctrls[c] != NULL; c++) { + if (strcmp(op->o_ctrls[c]->ldctl_oid, LDAP_CONTROL_PAGEDRESULTS) == 0) + break; + } + + assert( op->o_ctrls[c] != NULL ); + + ber_init2( ber, NULL, LBER_USE_DER ); + ber_printf( ber, "{iO}", size, &cookie ); + ber_flatten2( ber, &value, 0 ); + assert( op->o_ctrls[c]->ldctl_value.bv_len >= value.bv_len ); + op->o_ctrls[c]->ldctl_value.bv_len = value.bv_len; + lutil_memcopy( op->o_ctrls[c]->ldctl_value.bv_val, + value.bv_val, value.bv_len ); + ber_free_buf( ber ); + } + } + + default: + break; + } + } +end_of_loop:; + op->ors_scope = scope0; + op->ors_tlimit = tlimit0; + op->o_time = starttime; + + break; + } + + op->o_callback = cb.sc_next; + if ( op->o_abandon ) { + rs->sr_err = SLAPD_ABANDON; + } else { + rs->sr_err = gs.err; + rs->sr_matched = gs.matched; + rs->sr_ref = gs.refs; + } + rs->sr_ctrls = gs.ctrls; + + send_ldap_result( op, rs ); + + op->o_bd = b0; + op->o_bd->bd_info = bi0; + if (gs.matched) + free (gs.matched); + if (gs.refs) + ber_bvarray_free(gs.refs); + if (gs.ctrls) { + for (i = gs.nctrls; --i >= 0; ) { + op->o_tmpfree(gs.ctrls[i], op->o_tmpmemctx); + } + op->o_tmpfree(gs.ctrls, op->o_tmpmemctx); + } + return rs->sr_err; +} + +static BackendDB toolDB; + +static int +glue_tool_entry_open ( + BackendDB *b0, + int mode +) +{ + slap_overinfo *oi = (slap_overinfo *)b0->bd_info; + + /* We don't know which backend to talk to yet, so just + * remember the mode and move on... + */ + + glueMode = mode; + glueBack = NULL; + toolDB = *b0; + toolDB.bd_info = oi->oi_orig; + + /* Sanity checks */ + { + slap_overinst *on = glue_tool_inst( b0->bd_info ); + glueinfo *gi = on->on_bi.bi_private; + + int i; + for (i = 0; i < gi->gi_nodes; i++) { + BackendDB *bd; + struct berval pdn; + + dnParent( &gi->gi_n[i].gn_be->be_nsuffix[0], &pdn ); + bd = select_backend( &pdn, 0 ); + if ( bd ) { + ID id; + BackendDB db; + + if ( overlay_is_over( bd ) ) { + slap_overinfo *oi = (slap_overinfo *)bd->bd_info; + db = *bd; + db.bd_info = oi->oi_orig; + bd = &db; + } + + if ( !bd->bd_info->bi_tool_dn2id_get + || !bd->bd_info->bi_tool_entry_open + || !bd->bd_info->bi_tool_entry_close ) + { + continue; + } + + bd->bd_info->bi_tool_entry_open( bd, 0 ); + id = bd->bd_info->bi_tool_dn2id_get( bd, &gi->gi_n[i].gn_be->be_nsuffix[0] ); + bd->bd_info->bi_tool_entry_close( bd ); + if ( id != NOID ) { + Debug( LDAP_DEBUG_ANY, + "glue_tool_entry_open: subordinate database suffix entry DN=\"%s\" also present in superior database rooted at DN=\"%s\"\n", + gi->gi_n[i].gn_be->be_suffix[0].bv_val, bd->be_suffix[0].bv_val, 0 ); + return LDAP_OTHER; + } + } + } + } + + return 0; +} + +static int +glue_tool_entry_close ( + BackendDB *b0 +) +{ + int rc = 0; + + if (glueBack && glueBack != GLUEBACK_DONE) { + if (!glueBack->be_entry_close) + return 0; + rc = glueBack->be_entry_close (glueBack); + } + return rc; +} + +static slap_overinst * +glue_tool_inst( + BackendInfo *bi +) +{ + slap_overinfo *oi = (slap_overinfo *)bi; + slap_overinst *on; + + for ( on = oi->oi_list; on; on=on->on_next ) { + if ( !strcmp( on->on_bi.bi_type, glue.on_bi.bi_type )) + return on; + } + return NULL; +} + +/* This function will only be called in tool mode */ +static int +glue_open ( + BackendInfo *bi +) +{ + slap_overinst *on = glue_tool_inst( bi ); + glueinfo *gi = on->on_bi.bi_private; + static int glueOpened = 0; + int i, j, same, bsame = 0, rc = 0; + ConfigReply cr = {0}; + + if (glueOpened) return 0; + + glueOpened = 1; + + /* If we were invoked in tool mode, open all the underlying backends */ + if (slapMode & SLAP_TOOL_MODE) { + for (i = 0; i<gi->gi_nodes; i++) { + same = 0; + /* Same bi_open as our main backend? */ + if ( gi->gi_n[i].gn_be->bd_info->bi_open == + on->on_info->oi_orig->bi_open ) + bsame = 1; + + /* Loop thru the bd_info's and make sure we only + * invoke their bi_open functions once each. + */ + for ( j = 0; j<i; j++ ) { + if ( gi->gi_n[i].gn_be->bd_info->bi_open == + gi->gi_n[j].gn_be->bd_info->bi_open ) { + same = 1; + break; + } + } + /* OK, it's unique and non-NULL, call it. */ + if ( !same && gi->gi_n[i].gn_be->bd_info->bi_open ) + rc = gi->gi_n[i].gn_be->bd_info->bi_open( + gi->gi_n[i].gn_be->bd_info ); + /* Let backend.c take care of the rest of startup */ + if ( !rc ) + rc = backend_startup_one( gi->gi_n[i].gn_be, &cr ); + if ( rc ) break; + } + if ( !rc && !bsame && on->on_info->oi_orig->bi_open ) + rc = on->on_info->oi_orig->bi_open( on->on_info->oi_orig ); + + } /* other case is impossible */ + return rc; +} + +/* This function will only be called in tool mode */ +static int +glue_close ( + BackendInfo *bi +) +{ + static int glueClosed = 0; + int rc = 0; + + if (glueClosed) return 0; + + glueClosed = 1; + + if (slapMode & SLAP_TOOL_MODE) { + rc = backend_shutdown( NULL ); + } + return rc; +} + +static int +glue_entry_get_rw ( + Operation *op, + struct berval *dn, + ObjectClass *oc, + AttributeDescription *ad, + int rw, + Entry **e ) +{ + int rc; + BackendDB *b0 = op->o_bd; + op->o_bd = glue_back_select( b0, dn ); + + if ( op->o_bd->be_fetch ) { + rc = op->o_bd->be_fetch( op, dn, oc, ad, rw, e ); + } else { + rc = LDAP_UNWILLING_TO_PERFORM; + } + op->o_bd =b0; + return rc; +} + +static int +glue_entry_release_rw ( + Operation *op, + Entry *e, + int rw +) +{ + BackendDB *b0 = op->o_bd; + int rc = -1; + + op->o_bd = glue_back_select (b0, &e->e_nname); + + if ( op->o_bd->be_release ) { + rc = op->o_bd->be_release( op, e, rw ); + + } else { + /* FIXME: mimic be_entry_release_rw + * when no be_release() available */ + /* free entry */ + entry_free( e ); + rc = 0; + } + op->o_bd = b0; + return rc; +} + +static struct berval *glue_base; +static int glue_scope; +static Filter *glue_filter; + +static ID +glue_tool_entry_first ( + BackendDB *b0 +) +{ + slap_overinst *on = glue_tool_inst( b0->bd_info ); + glueinfo *gi = on->on_bi.bi_private; + int i; + ID rc; + + /* If we're starting from scratch, start at the most general */ + if (!glueBack) { + if ( toolDB.be_entry_open && toolDB.be_entry_first ) { + glueBack = &toolDB; + } else { + for (i = gi->gi_nodes-1; i >= 0; i--) { + if (gi->gi_n[i].gn_be->be_entry_open && + gi->gi_n[i].gn_be->be_entry_first) { + glueBack = gi->gi_n[i].gn_be; + break; + } + } + } + } + if (!glueBack || !glueBack->be_entry_open || !glueBack->be_entry_first || + glueBack->be_entry_open (glueBack, glueMode) != 0) + return NOID; + + rc = glueBack->be_entry_first (glueBack); + while ( rc == NOID ) { + if ( glueBack && glueBack->be_entry_close ) + glueBack->be_entry_close (glueBack); + for (i=0; i<gi->gi_nodes; i++) { + if (gi->gi_n[i].gn_be == glueBack) + break; + } + if (i == 0) { + glueBack = GLUEBACK_DONE; + break; + } else { + glueBack = gi->gi_n[i-1].gn_be; + rc = glue_tool_entry_first (b0); + if ( glueBack == GLUEBACK_DONE ) { + break; + } + } + } + return rc; +} + +static ID +glue_tool_entry_first_x ( + BackendDB *b0, + struct berval *base, + int scope, + Filter *f +) +{ + slap_overinst *on = glue_tool_inst( b0->bd_info ); + glueinfo *gi = on->on_bi.bi_private; + int i; + ID rc; + + glue_base = base; + glue_scope = scope; + glue_filter = f; + + /* If we're starting from scratch, start at the most general */ + if (!glueBack) { + if ( toolDB.be_entry_open && toolDB.be_entry_first_x ) { + glueBack = &toolDB; + } else { + for (i = gi->gi_nodes-1; i >= 0; i--) { + if (gi->gi_n[i].gn_be->be_entry_open && + gi->gi_n[i].gn_be->be_entry_first_x) + { + glueBack = gi->gi_n[i].gn_be; + break; + } + } + } + } + if (!glueBack || !glueBack->be_entry_open || !glueBack->be_entry_first_x || + glueBack->be_entry_open (glueBack, glueMode) != 0) + return NOID; + + rc = glueBack->be_entry_first_x (glueBack, + glue_base, glue_scope, glue_filter); + while ( rc == NOID ) { + if ( glueBack && glueBack->be_entry_close ) + glueBack->be_entry_close (glueBack); + for (i=0; i<gi->gi_nodes; i++) { + if (gi->gi_n[i].gn_be == glueBack) + break; + } + if (i == 0) { + glueBack = GLUEBACK_DONE; + break; + } else { + glueBack = gi->gi_n[i-1].gn_be; + rc = glue_tool_entry_first_x (b0, + glue_base, glue_scope, glue_filter); + if ( glueBack == GLUEBACK_DONE ) { + break; + } + } + } + return rc; +} + +static ID +glue_tool_entry_next ( + BackendDB *b0 +) +{ + slap_overinst *on = glue_tool_inst( b0->bd_info ); + glueinfo *gi = on->on_bi.bi_private; + int i; + ID rc; + + if (!glueBack || !glueBack->be_entry_next) + return NOID; + + rc = glueBack->be_entry_next (glueBack); + + /* If we ran out of entries in one database, move on to the next */ + while (rc == NOID) { + if ( glueBack && glueBack->be_entry_close ) + glueBack->be_entry_close (glueBack); + for (i=0; i<gi->gi_nodes; i++) { + if (gi->gi_n[i].gn_be == glueBack) + break; + } + if (i == 0) { + glueBack = GLUEBACK_DONE; + break; + } else { + glueBack = gi->gi_n[i-1].gn_be; + if ( glue_base || glue_filter ) { + /* using entry_first_x() */ + rc = glue_tool_entry_first_x (b0, + glue_base, glue_scope, glue_filter); + + } else { + /* using entry_first() */ + rc = glue_tool_entry_first (b0); + } + if ( glueBack == GLUEBACK_DONE ) { + break; + } + } + } + return rc; +} + +static ID +glue_tool_dn2id_get ( + BackendDB *b0, + struct berval *dn +) +{ + BackendDB *be, b2; + int rc = -1; + + b2 = *b0; + b2.bd_info = (BackendInfo *)glue_tool_inst( b0->bd_info ); + be = glue_back_select (&b2, dn); + if ( be == &b2 ) be = &toolDB; + + if (!be->be_dn2id_get) + return NOID; + + if (!glueBack) { + if ( be->be_entry_open ) { + rc = be->be_entry_open (be, glueMode); + } + if (rc != 0) { + return NOID; + } + } else if (be != glueBack) { + /* If this entry belongs in a different branch than the + * previous one, close the current database and open the + * new one. + */ + if ( glueBack->be_entry_close ) { + glueBack->be_entry_close (glueBack); + } + if ( be->be_entry_open ) { + rc = be->be_entry_open (be, glueMode); + } + if (rc != 0) { + return NOID; + } + } + glueBack = be; + return be->be_dn2id_get (be, dn); +} + +static Entry * +glue_tool_entry_get ( + BackendDB *b0, + ID id +) +{ + if (!glueBack || !glueBack->be_entry_get) + return NULL; + + return glueBack->be_entry_get (glueBack, id); +} + +static ID +glue_tool_entry_put ( + BackendDB *b0, + Entry *e, + struct berval *text +) +{ + BackendDB *be, b2; + int rc = -1; + + b2 = *b0; + b2.bd_info = (BackendInfo *)glue_tool_inst( b0->bd_info ); + be = glue_back_select (&b2, &e->e_nname); + if ( be == &b2 ) be = &toolDB; + + if (!be->be_entry_put) + return NOID; + + if (!glueBack) { + if ( be->be_entry_open ) { + rc = be->be_entry_open (be, glueMode); + } + if (rc != 0) { + return NOID; + } + } else if (be != glueBack) { + /* If this entry belongs in a different branch than the + * previous one, close the current database and open the + * new one. + */ + if ( glueBack->be_entry_close ) { + glueBack->be_entry_close (glueBack); + } + if ( be->be_entry_open ) { + rc = be->be_entry_open (be, glueMode); + } + if (rc != 0) { + return NOID; + } + } + glueBack = be; + return be->be_entry_put (be, e, text); +} + +static ID +glue_tool_entry_modify ( + BackendDB *b0, + Entry *e, + struct berval *text +) +{ + if (!glueBack || !glueBack->be_entry_modify) + return NOID; + + return glueBack->be_entry_modify (glueBack, e, text); +} + +static int +glue_tool_entry_reindex ( + BackendDB *b0, + ID id, + AttributeDescription **adv +) +{ + if (!glueBack || !glueBack->be_entry_reindex) + return -1; + + return glueBack->be_entry_reindex (glueBack, id, adv); +} + +static int +glue_tool_sync ( + BackendDB *b0 +) +{ + slap_overinst *on = glue_tool_inst( b0->bd_info ); + glueinfo *gi = on->on_bi.bi_private; + BackendInfo *bi = b0->bd_info; + int i; + + /* just sync everyone */ + for (i = 0; i<gi->gi_nodes; i++) + if (gi->gi_n[i].gn_be->be_sync) + gi->gi_n[i].gn_be->be_sync (gi->gi_n[i].gn_be); + b0->bd_info = on->on_info->oi_orig; + if ( b0->be_sync ) + b0->be_sync( b0 ); + b0->bd_info = bi; + return 0; +} + +typedef struct glue_Addrec { + struct glue_Addrec *ga_next; + BackendDB *ga_be; +} glue_Addrec; + +/* List of added subordinates */ +static glue_Addrec *ga_list; +static int ga_adding; + +static int +glue_db_init( + BackendDB *be, + ConfigReply *cr +) +{ + slap_overinst *on = (slap_overinst *)be->bd_info; + slap_overinfo *oi = on->on_info; + BackendInfo *bi = oi->oi_orig; + glueinfo *gi; + + if ( SLAP_GLUE_SUBORDINATE( be )) { + Debug( LDAP_DEBUG_ANY, "glue: backend %s is already subordinate, " + "cannot have glue overlay!\n", + be->be_suffix[0].bv_val, 0, 0 ); + return LDAP_OTHER; + } + + gi = ch_calloc( 1, sizeof(glueinfo)); + on->on_bi.bi_private = gi; + dnParent( be->be_nsuffix, &gi->gi_pdn ); + + /* Currently the overlay framework doesn't handle these entry points + * but we need them.... + */ + oi->oi_bi.bi_open = glue_open; + oi->oi_bi.bi_close = glue_close; + + /* Only advertise these if the root DB supports them */ + if ( bi->bi_tool_entry_open ) + oi->oi_bi.bi_tool_entry_open = glue_tool_entry_open; + if ( bi->bi_tool_entry_close ) + oi->oi_bi.bi_tool_entry_close = glue_tool_entry_close; + if ( bi->bi_tool_entry_first ) + oi->oi_bi.bi_tool_entry_first = glue_tool_entry_first; + /* FIXME: check whether all support bi_tool_entry_first_x() ? */ + if ( bi->bi_tool_entry_first_x ) + oi->oi_bi.bi_tool_entry_first_x = glue_tool_entry_first_x; + if ( bi->bi_tool_entry_next ) + oi->oi_bi.bi_tool_entry_next = glue_tool_entry_next; + if ( bi->bi_tool_entry_get ) + oi->oi_bi.bi_tool_entry_get = glue_tool_entry_get; + if ( bi->bi_tool_dn2id_get ) + oi->oi_bi.bi_tool_dn2id_get = glue_tool_dn2id_get; + if ( bi->bi_tool_entry_put ) + oi->oi_bi.bi_tool_entry_put = glue_tool_entry_put; + if ( bi->bi_tool_entry_reindex ) + oi->oi_bi.bi_tool_entry_reindex = glue_tool_entry_reindex; + if ( bi->bi_tool_entry_modify ) + oi->oi_bi.bi_tool_entry_modify = glue_tool_entry_modify; + if ( bi->bi_tool_sync ) + oi->oi_bi.bi_tool_sync = glue_tool_sync; + + SLAP_DBFLAGS( be ) |= SLAP_DBFLAG_GLUE_INSTANCE; + + if ( ga_list && ( slapMode & SLAP_SERVER_MODE ) ) { + be->bd_info = (BackendInfo *)oi; + glue_sub_attach( 1 ); + } + + return 0; +} + +static int +glue_db_destroy ( + BackendDB *be, + ConfigReply *cr +) +{ + slap_overinst *on = (slap_overinst *)be->bd_info; + glueinfo *gi = (glueinfo *)on->on_bi.bi_private; + + free (gi); + return SLAP_CB_CONTINUE; +} + +static int +glue_db_close( + BackendDB *be, + ConfigReply *cr +) +{ + slap_overinst *on = (slap_overinst *)be->bd_info; + + on->on_info->oi_bi.bi_db_close = 0; + return 0; +} + +int +glue_sub_del( BackendDB *b0 ) +{ + BackendDB *be; + int rc = 0; + + /* Find the top backend for this subordinate */ + be = b0; + while ( (be=LDAP_STAILQ_NEXT( be, be_next )) != NULL ) { + slap_overinfo *oi; + slap_overinst *on; + glueinfo *gi; + int i; + + if ( SLAP_GLUE_SUBORDINATE( be )) + continue; + if ( !SLAP_GLUE_INSTANCE( be )) + continue; + if ( !dnIsSuffix( &b0->be_nsuffix[0], &be->be_nsuffix[0] )) + continue; + + /* OK, got the right backend, find the overlay */ + oi = (slap_overinfo *)be->bd_info; + for ( on=oi->oi_list; on; on=on->on_next ) { + if ( on->on_bi.bi_type == glue.on_bi.bi_type ) + break; + } + assert( on != NULL ); + gi = on->on_bi.bi_private; + for ( i=0; i < gi->gi_nodes; i++ ) { + if ( gi->gi_n[i].gn_be == b0 ) { + int j; + + for (j=i+1; j < gi->gi_nodes; j++) + gi->gi_n[j-1] = gi->gi_n[j]; + + gi->gi_nodes--; + } + } + } + if ( be == NULL ) + rc = LDAP_NO_SUCH_OBJECT; + + return rc; +} + + +/* Attach all the subordinate backends to their superior */ +int +glue_sub_attach( int online ) +{ + glue_Addrec *ga, *gnext = NULL; + int rc = 0; + + if ( ga_adding ) + return 0; + + ga_adding = 1; + + /* For all the subordinate backends */ + for ( ga=ga_list; ga != NULL; ga = gnext ) { + BackendDB *be; + + gnext = ga->ga_next; + + /* Find the top backend for this subordinate */ + be = ga->ga_be; + while ( (be=LDAP_STAILQ_NEXT( be, be_next )) != NULL ) { + slap_overinfo *oi; + slap_overinst *on; + glueinfo *gi; + + if ( SLAP_GLUE_SUBORDINATE( be )) + continue; + if ( !dnIsSuffix( &ga->ga_be->be_nsuffix[0], &be->be_nsuffix[0] )) + continue; + + /* If it's not already configured, set up the overlay */ + if ( !SLAP_GLUE_INSTANCE( be )) { + rc = overlay_config( be, glue.on_bi.bi_type, -1, NULL, NULL); + if ( rc ) + break; + } + /* Find the overlay instance */ + oi = (slap_overinfo *)be->bd_info; + for ( on=oi->oi_list; on; on=on->on_next ) { + if ( on->on_bi.bi_type == glue.on_bi.bi_type ) + break; + } + assert( on != NULL ); + gi = on->on_bi.bi_private; + gi = (glueinfo *)ch_realloc( gi, sizeof(glueinfo) + + gi->gi_nodes * sizeof(gluenode)); + gi->gi_n[gi->gi_nodes].gn_be = ga->ga_be; + dnParent( &ga->ga_be->be_nsuffix[0], + &gi->gi_n[gi->gi_nodes].gn_pdn ); + gi->gi_nodes++; + on->on_bi.bi_private = gi; + ga->ga_be->be_flags |= SLAP_DBFLAG_GLUE_LINKED; + break; + } + if ( !be ) { + Debug( LDAP_DEBUG_ANY, "glue: no superior found for sub %s!\n", + ga->ga_be->be_suffix[0].bv_val, 0, 0 ); + /* allow this for now, assume a superior will + * be added later + */ + if ( online ) { + rc = 0; + gnext = ga_list; + break; + } + rc = LDAP_NO_SUCH_OBJECT; + } + ch_free( ga ); + if ( rc ) break; + } + + ga_list = gnext; + + ga_adding = 0; + + return rc; +} + +int +glue_sub_add( BackendDB *be, int advert, int online ) +{ + glue_Addrec *ga; + int rc = 0; + + if ( overlay_is_inst( be, "glue" )) { + Debug( LDAP_DEBUG_ANY, "glue: backend %s already has glue overlay, " + "cannot be a subordinate!\n", + be->be_suffix[0].bv_val, 0, 0 ); + return LDAP_OTHER; + } + SLAP_DBFLAGS( be ) |= SLAP_DBFLAG_GLUE_SUBORDINATE; + if ( advert ) + SLAP_DBFLAGS( be ) |= SLAP_DBFLAG_GLUE_ADVERTISE; + + ga = ch_malloc( sizeof( glue_Addrec )); + ga->ga_next = ga_list; + ga->ga_be = be; + ga_list = ga; + + if ( online ) + rc = glue_sub_attach( online ); + + return rc; +} + +static int +glue_access_allowed( + Operation *op, + Entry *e, + AttributeDescription *desc, + struct berval *val, + slap_access_t access, + AccessControlState *state, + slap_mask_t *maskp ) +{ + BackendDB *b0, *be = glue_back_select( op->o_bd, &e->e_nname ); + int rc; + + if ( be == NULL || be == op->o_bd || be->bd_info->bi_access_allowed == NULL ) + return SLAP_CB_CONTINUE; + + b0 = op->o_bd; + op->o_bd = be; + rc = be->bd_info->bi_access_allowed ( op, e, desc, val, access, state, maskp ); + op->o_bd = b0; + return rc; +} + +int +glue_sub_init() +{ + glue.on_bi.bi_type = "glue"; + + glue.on_bi.bi_db_init = glue_db_init; + glue.on_bi.bi_db_close = glue_db_close; + glue.on_bi.bi_db_destroy = glue_db_destroy; + + glue.on_bi.bi_op_search = glue_op_search; + glue.on_bi.bi_op_modify = glue_op_func; + glue.on_bi.bi_op_modrdn = glue_op_func; + glue.on_bi.bi_op_add = glue_op_func; + glue.on_bi.bi_op_delete = glue_op_func; + glue.on_bi.bi_op_abandon = glue_op_abandon; + glue.on_bi.bi_extended = glue_op_func; + + glue.on_bi.bi_chk_referrals = glue_chk_referrals; + glue.on_bi.bi_chk_controls = glue_chk_controls; + glue.on_bi.bi_entry_get_rw = glue_entry_get_rw; + glue.on_bi.bi_entry_release_rw = glue_entry_release_rw; + glue.on_bi.bi_access_allowed = glue_access_allowed; + + glue.on_response = glue_response; + + return overlay_register( &glue ); +} |