/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2008 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ #pragma ident "%Z%%M% %I% %E% SMI" /* * Server side RPC handler. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define SMB_CTXT_BUFSZ 65536 /* * Fragment size (5680: NT style). */ #define MLRPC_FRAG_SZ 5680 static unsigned long mlrpc_frag_size = MLRPC_FRAG_SZ; /* * Service context table. */ #define CTXT_TABLE_ENTRIES 128 static struct mlsvc_rpc_context context_table[CTXT_TABLE_ENTRIES]; static mutex_t mlrpc_context_lock; static int ndr_s_transact(struct mlsvc_rpc_context *); static struct mlsvc_rpc_context *ndr_s_lookup(int); static void ndr_s_release(struct mlsvc_rpc_context *); static struct mlsvc_rpc_context *ndr_s_allocate(int); static void ndr_s_deallocate(struct mlsvc_rpc_context *); static void ndr_s_rewind(struct mlsvc_rpc_context *); static void ndr_s_flush(struct mlsvc_rpc_context *); static int mlrpc_s_process(struct mlrpc_xaction *); static int mlrpc_s_bind(struct mlrpc_xaction *); static int mlrpc_s_request(struct mlrpc_xaction *); static void mlrpc_reply_prepare_hdr(struct mlrpc_xaction *); static int mlrpc_s_alter_context(struct mlrpc_xaction *); static void mlrpc_reply_bind_ack(struct mlrpc_xaction *); static void mlrpc_reply_fault(struct mlrpc_xaction *, unsigned long); static int mlrpc_build_reply(struct mlrpc_xaction *); static void mlrpc_build_frag(struct mlndr_stream *, uint8_t *, uint32_t); /* * Allocate and associate a service context with a fid. */ int ndr_s_open(int fid, uint8_t *data, uint32_t datalen) { struct mlsvc_rpc_context *svc; (void) mutex_lock(&mlrpc_context_lock); if ((svc = ndr_s_lookup(fid)) != NULL) { ndr_s_release(svc); (void) mutex_unlock(&mlrpc_context_lock); return (EEXIST); } if ((svc = ndr_s_allocate(fid)) == NULL) { (void) mutex_unlock(&mlrpc_context_lock); return (ENOMEM); } if (smb_opipe_context_decode(&svc->svc_ctx, data, datalen) == -1) { ndr_s_release(svc); (void) mutex_unlock(&mlrpc_context_lock); return (EINVAL); } mlrpc_binding_pool_initialize(&svc->binding, svc->binding_pool, CTXT_N_BINDING_POOL); (void) mutex_unlock(&mlrpc_context_lock); return (0); } /* * Release the context associated with a fid when an opipe is closed. */ int ndr_s_close(int fid) { struct mlsvc_rpc_context *svc; (void) mutex_lock(&mlrpc_context_lock); if ((svc = ndr_s_lookup(fid)) == NULL) { (void) mutex_unlock(&mlrpc_context_lock); return (ENOENT); } /* * Release twice: once for the lookup above * and again to close the fid. */ ndr_s_release(svc); ndr_s_release(svc); (void) mutex_unlock(&mlrpc_context_lock); return (0); } /* * Write RPC request data to the input stream. Input data is buffered * until the response is requested. */ int ndr_s_write(int fid, uint8_t *buf, uint32_t len) { struct mlsvc_rpc_context *svc; ssize_t nbytes; if (len == 0) return (0); (void) mutex_lock(&mlrpc_context_lock); if ((svc = ndr_s_lookup(fid)) == NULL) { (void) mutex_unlock(&mlrpc_context_lock); return (ENOENT); } nbytes = ndr_uiomove((caddr_t)buf, len, UIO_READ, &svc->in_uio); ndr_s_release(svc); (void) mutex_unlock(&mlrpc_context_lock); return ((nbytes == len) ? 0 : EIO); } /* * Read RPC response data. If the input stream contains an RPC request, * we need to process the RPC transaction, which will place the RPC * response in the output (frags) stream. Otherwise, read data from * the output stream. */ int ndr_s_read(int fid, uint8_t *buf, uint32_t *len, uint32_t *resid) { struct mlsvc_rpc_context *svc; ssize_t nbytes = *len; int rc; if (nbytes == 0) { *resid = 0; return (0); } (void) mutex_lock(&mlrpc_context_lock); if ((svc = ndr_s_lookup(fid)) == NULL) { (void) mutex_unlock(&mlrpc_context_lock); return (ENOENT); } (void) mutex_unlock(&mlrpc_context_lock); if (svc->in_uio.uio_offset) { if ((rc = ndr_s_transact(svc)) != 0) { ndr_s_flush(svc); (void) mutex_lock(&mlrpc_context_lock); ndr_s_release(svc); (void) mutex_unlock(&mlrpc_context_lock); return (rc); } } *len = ndr_uiomove((caddr_t)buf, nbytes, UIO_WRITE, &svc->frags.uio); *resid = svc->frags.uio.uio_resid; if (*resid == 0) { /* * Nothing left, cleanup the output stream. */ ndr_s_flush(svc); } (void) mutex_lock(&mlrpc_context_lock); ndr_s_release(svc); (void) mutex_unlock(&mlrpc_context_lock); return (0); } /* * Process a server-side RPC request. */ static int ndr_s_transact(struct mlsvc_rpc_context *svc) { ndr_xa_t *mxa; struct mlndr_stream *recv_mlnds; struct mlndr_stream *send_mlnds; char *data; int datalen; data = svc->in_buf; datalen = svc->in_uio.uio_offset; if ((mxa = (ndr_xa_t *)malloc(sizeof (ndr_xa_t))) == NULL) return (ENOMEM); bzero(mxa, sizeof (struct mlrpc_xaction)); mxa->fid = svc->fid; mxa->context = svc; mxa->binding_list = svc->binding; if ((mxa->heap = mlrpc_heap_create()) == NULL) { free(mxa); return (ENOMEM); } recv_mlnds = &mxa->recv_mlnds; mlnds_initialize(recv_mlnds, datalen, NDR_MODE_CALL_RECV, mxa->heap); /* * Copy the input data and reset the input stream. */ bcopy(data, recv_mlnds->pdu_base_addr, datalen); ndr_s_rewind(svc); send_mlnds = &mxa->send_mlnds; mlnds_initialize(send_mlnds, 0, NDR_MODE_RETURN_SEND, mxa->heap); (void) mlrpc_s_process(mxa); mlnds_finalize(send_mlnds, &svc->frags); mlnds_destruct(&mxa->recv_mlnds); mlnds_destruct(&mxa->send_mlnds); mlrpc_heap_destroy(mxa->heap); free(mxa); return (0); } /* * Must be called with mlrpc_context_lock held. */ static struct mlsvc_rpc_context * ndr_s_lookup(int fid) { struct mlsvc_rpc_context *svc; int i; for (i = 0; i < CTXT_TABLE_ENTRIES; ++i) { svc = &context_table[i]; if (svc->fid == fid) { if (svc->refcnt == 0) return (NULL); svc->refcnt++; return (svc); } } return (NULL); } /* * Must be called with mlrpc_context_lock held. */ static void ndr_s_release(struct mlsvc_rpc_context *svc) { svc->refcnt--; ndr_s_deallocate(svc); } /* * Must be called with mlrpc_context_lock held. */ static struct mlsvc_rpc_context * ndr_s_allocate(int fid) { struct mlsvc_rpc_context *svc = NULL; int i; for (i = 0; i < CTXT_TABLE_ENTRIES; ++i) { svc = &context_table[i]; if (svc->fid == 0) { bzero(svc, sizeof (struct mlsvc_rpc_context)); if ((svc->in_buf = malloc(SMB_CTXT_BUFSZ)) == NULL) return (NULL); ndr_s_rewind(svc); svc->fid = fid; svc->refcnt = 1; return (svc); } } return (NULL); } /* * Must be called with mlrpc_context_lock held. */ static void ndr_s_deallocate(struct mlsvc_rpc_context *svc) { if (svc->refcnt == 0) { /* * Ensure that there are no RPC service policy handles * (associated with this fid) left around. */ ndr_hdclose(svc->fid); ndr_s_rewind(svc); ndr_s_flush(svc); free(svc->in_buf); free(svc->svc_ctx.oc_domain); free(svc->svc_ctx.oc_account); free(svc->svc_ctx.oc_workstation); bzero(svc, sizeof (struct mlsvc_rpc_context)); } } /* * Rewind the input data stream, ready for the next write. */ static void ndr_s_rewind(struct mlsvc_rpc_context *svc) { svc->in_uio.uio_iov = &svc->in_iov; svc->in_uio.uio_iovcnt = 1; svc->in_uio.uio_offset = 0; svc->in_uio.uio_segflg = UIO_USERSPACE; svc->in_uio.uio_resid = SMB_CTXT_BUFSZ; svc->in_iov.iov_base = svc->in_buf; svc->in_iov.iov_len = SMB_CTXT_BUFSZ; } /* * Flush the output data stream. */ static void ndr_s_flush(struct mlsvc_rpc_context *svc) { ndr_frag_t *frag; while ((frag = svc->frags.head) != NULL) { svc->frags.head = frag->next; free(frag); } free(svc->frags.iov); bzero(&svc->frags, sizeof (ndr_fraglist_t)); } /* * Check whether or not the specified user has administrator privileges, * i.e. is a member of Domain Admins or Administrators. * Returns true if the user is an administrator, otherwise returns false. */ boolean_t ndr_is_admin(ndr_xa_t *xa) { smb_opipe_context_t *svc = &xa->context->svc_ctx; return (svc->oc_flags & SMB_ATF_ADMIN); } /* * Check whether or not the specified user has power-user privileges, * i.e. is a member of Domain Admins, Administrators or Power Users. * This is typically required for operations such as managing shares. * Returns true if the user is a power user, otherwise returns false. */ boolean_t ndr_is_poweruser(ndr_xa_t *xa) { smb_opipe_context_t *svc = &xa->context->svc_ctx; return ((svc->oc_flags & SMB_ATF_ADMIN) || (svc->oc_flags & SMB_ATF_POWERUSER)); } int32_t ndr_native_os(ndr_xa_t *xa) { smb_opipe_context_t *svc = &xa->context->svc_ctx; return (svc->oc_native_os); } /* * This is the entry point for all server-side RPC processing. * It is assumed that the PDU has already been received. */ static int mlrpc_s_process(struct mlrpc_xaction *mxa) { int rc; rc = mlrpc_decode_pdu_hdr(mxa); if (!MLRPC_DRC_IS_OK(rc)) return (-1); (void) mlrpc_reply_prepare_hdr(mxa); switch (mxa->ptype) { case MLRPC_PTYPE_BIND: rc = mlrpc_s_bind(mxa); break; case MLRPC_PTYPE_REQUEST: rc = mlrpc_s_request(mxa); break; case MLRPC_PTYPE_ALTER_CONTEXT: rc = mlrpc_s_alter_context(mxa); break; default: rc = MLRPC_DRC_FAULT_RPCHDR_PTYPE_INVALID; break; } if (MLRPC_DRC_IS_FAULT(rc)) mlrpc_reply_fault(mxa, rc); (void) mlrpc_build_reply(mxa); return (rc); } /* * Multiple p_cont_elem[]s, multiple transfer_syntaxes[] and multiple * p_results[] not supported. */ static int mlrpc_s_bind(struct mlrpc_xaction *mxa) { mlrpc_p_cont_list_t *cont_list; mlrpc_p_result_list_t *result_list; mlrpc_p_result_t *result; unsigned p_cont_id; struct mlrpc_binding *mbind; ndr_uuid_t *as_uuid; ndr_uuid_t *ts_uuid; char as_buf[64]; char ts_buf[64]; int as_vers; int ts_vers; struct mlndr_stream *send_mlnds; struct mlrpc_service *msvc; int rc; mlrpc_port_any_t *sec_addr; /* acquire targets */ cont_list = &mxa->recv_hdr.bind_hdr.p_context_elem; result_list = &mxa->send_hdr.bind_ack_hdr.p_result_list; result = &result_list->p_results[0]; /* * Set up temporary secondary address port. * We will correct this later (below). */ send_mlnds = &mxa->send_mlnds; sec_addr = &mxa->send_hdr.bind_ack_hdr.sec_addr; sec_addr->length = 13; (void) strcpy((char *)sec_addr->port_spec, "\\PIPE\\ntsvcs"); result_list->n_results = 1; result_list->reserved = 0; result_list->reserved2 = 0; result->result = MLRPC_PCDR_ACCEPTANCE; result->reason = 0; bzero(&result->transfer_syntax, sizeof (result->transfer_syntax)); /* sanity check */ if (cont_list->n_context_elem != 1 || cont_list->p_cont_elem[0].n_transfer_syn != 1) { mlndo_trace("mlrpc_s_bind: warning: multiple p_cont_elem"); } p_cont_id = cont_list->p_cont_elem[0].p_cont_id; if ((mbind = mlrpc_find_binding(mxa, p_cont_id)) != NULL) { /* * Duplicate p_cont_id. * Send a bind_ack with a better error. */ mlndo_trace("mlrpc_s_bind: duplicate binding"); return (MLRPC_DRC_FAULT_BIND_PCONT_BUSY); } if ((mbind = mlrpc_new_binding(mxa)) == NULL) { /* * No free binding slot */ result->result = MLRPC_PCDR_PROVIDER_REJECTION; result->reason = MLRPC_PPR_LOCAL_LIMIT_EXCEEDED; mlndo_trace("mlrpc_s_bind: no resources"); return (MLRPC_DRC_OK); } as_uuid = &cont_list->p_cont_elem[0].abstract_syntax.if_uuid; as_vers = cont_list->p_cont_elem[0].abstract_syntax.if_version; ts_uuid = &cont_list->p_cont_elem[0].transfer_syntaxes[0].if_uuid; ts_vers = cont_list->p_cont_elem[0].transfer_syntaxes[0].if_version; msvc = mlrpc_find_service_by_uuids(as_uuid, as_vers, ts_uuid, ts_vers); if (!msvc) { mlrpc_uuid_to_str(as_uuid, as_buf); mlrpc_uuid_to_str(ts_uuid, ts_buf); mlndo_printf(send_mlnds, 0, "mlrpc_s_bind: unknown service"); mlndo_printf(send_mlnds, 0, "abs=%s v%d, xfer=%s v%d", as_buf, as_vers, ts_buf, ts_vers); result->result = MLRPC_PCDR_PROVIDER_REJECTION; result->reason = MLRPC_PPR_ABSTRACT_SYNTAX_NOT_SUPPORTED; return (MLRPC_DRC_OK); } /* * We can now use the correct secondary address port. */ sec_addr = &mxa->send_hdr.bind_ack_hdr.sec_addr; sec_addr->length = strlen(msvc->sec_addr_port) + 1; (void) strlcpy((char *)sec_addr->port_spec, msvc->sec_addr_port, MLRPC_PORT_ANY_MAX_PORT_SPEC); mbind->p_cont_id = p_cont_id; mbind->which_side = MLRPC_BIND_SIDE_SERVER; /* mbind->context set by app */ mbind->service = msvc; mbind->instance_specific = 0; mxa->binding = mbind; if (msvc->bind_req) { /* * Call the service-specific bind() handler. If * this fails, we shouild send a specific error * on the bind ack. */ rc = (msvc->bind_req)(mxa); if (MLRPC_DRC_IS_FAULT(rc)) { mbind->service = 0; /* free binding slot */ mbind->which_side = 0; mbind->p_cont_id = 0; mbind->instance_specific = 0; return (rc); } } result->transfer_syntax = cont_list->p_cont_elem[0].transfer_syntaxes[0]; /* * Special rejection of Windows 2000 DSSETUP interface. * This interface was introduced in Windows 2000 but has * been subsequently deprecated due to problems. */ if (strcmp(msvc->name, "DSSETUP") == 0) { result->result = MLRPC_PCDR_PROVIDER_REJECTION; result->reason = MLRPC_PPR_ABSTRACT_SYNTAX_NOT_SUPPORTED; } return (MLRPC_DRC_BINDING_MADE); } /* * mlrpc_s_alter_context * * The alter context request is used to request additional presentation * context for another interface and/or version. It's very similar to a * bind request. * * We don't fully support multiple contexts so, for now, we reject this * request. Windows 2000 clients attempt to use an alternate LSA context * when ACLs are modified. */ static int mlrpc_s_alter_context(struct mlrpc_xaction *mxa) { mlrpc_p_result_list_t *result_list; mlrpc_p_result_t *result; mlrpc_p_cont_list_t *cont_list; struct mlrpc_binding *mbind; struct mlrpc_service *msvc; unsigned p_cont_id; ndr_uuid_t *as_uuid; ndr_uuid_t *ts_uuid; int as_vers; int ts_vers; mlrpc_port_any_t *sec_addr; result_list = &mxa->send_hdr.bind_ack_hdr.p_result_list; result_list->n_results = 1; result_list->reserved = 0; result_list->reserved2 = 0; result = &result_list->p_results[0]; result->result = MLRPC_PCDR_ACCEPTANCE; result->reason = 0; bzero(&result->transfer_syntax, sizeof (result->transfer_syntax)); if (mxa != NULL) { result->result = MLRPC_PCDR_PROVIDER_REJECTION; result->reason = MLRPC_PPR_ABSTRACT_SYNTAX_NOT_SUPPORTED; return (MLRPC_DRC_OK); } cont_list = &mxa->recv_hdr.bind_hdr.p_context_elem; p_cont_id = cont_list->p_cont_elem[0].p_cont_id; if ((mbind = mlrpc_find_binding(mxa, p_cont_id)) != NULL) return (MLRPC_DRC_FAULT_BIND_PCONT_BUSY); if ((mbind = mlrpc_new_binding(mxa)) == NULL) { result->result = MLRPC_PCDR_PROVIDER_REJECTION; result->reason = MLRPC_PPR_LOCAL_LIMIT_EXCEEDED; return (MLRPC_DRC_OK); } as_uuid = &cont_list->p_cont_elem[0].abstract_syntax.if_uuid; as_vers = cont_list->p_cont_elem[0].abstract_syntax.if_version; ts_uuid = &cont_list->p_cont_elem[0].transfer_syntaxes[0].if_uuid; ts_vers = cont_list->p_cont_elem[0].transfer_syntaxes[0].if_version; msvc = mlrpc_find_service_by_uuids(as_uuid, as_vers, ts_uuid, ts_vers); if (msvc == 0) { result->result = MLRPC_PCDR_PROVIDER_REJECTION; result->reason = MLRPC_PPR_ABSTRACT_SYNTAX_NOT_SUPPORTED; return (MLRPC_DRC_OK); } mbind->p_cont_id = p_cont_id; mbind->which_side = MLRPC_BIND_SIDE_SERVER; /* mbind->context set by app */ mbind->service = msvc; mbind->instance_specific = 0; mxa->binding = mbind; sec_addr = &mxa->send_hdr.bind_ack_hdr.sec_addr; sec_addr->length = 0; bzero(sec_addr->port_spec, MLRPC_PORT_ANY_MAX_PORT_SPEC); result->transfer_syntax = cont_list->p_cont_elem[0].transfer_syntaxes[0]; return (MLRPC_DRC_BINDING_MADE); } static int mlrpc_s_request(struct mlrpc_xaction *mxa) { struct mlrpc_binding *mbind; struct mlrpc_service *msvc; unsigned p_cont_id; int rc; mxa->opnum = mxa->recv_hdr.request_hdr.opnum; p_cont_id = mxa->recv_hdr.request_hdr.p_cont_id; if ((mbind = mlrpc_find_binding(mxa, p_cont_id)) == NULL) return (MLRPC_DRC_FAULT_REQUEST_PCONT_INVALID); mxa->binding = mbind; msvc = mbind->service; /* * Make room for the response hdr. */ mxa->send_mlnds.pdu_scan_offset = MLRPC_RSP_HDR_SIZE; if (msvc->call_stub) rc = (*msvc->call_stub)(mxa); else rc = mlrpc_generic_call_stub(mxa); if (MLRPC_DRC_IS_FAULT(rc)) { mlndo_printf(0, 0, "%s[0x%02x]: 0x%04x", msvc->name, mxa->opnum, rc); } return (rc); } /* * The transaction and the two mlnds streams use the same heap, which * should already exist at this point. The heap will also be available * to the stub. */ int mlrpc_generic_call_stub(struct mlrpc_xaction *mxa) { struct mlrpc_binding *mbind = mxa->binding; struct mlrpc_service *msvc = mbind->service; struct ndr_typeinfo *intf_ti = msvc->interface_ti; struct mlrpc_stub_table *ste; int opnum = mxa->opnum; unsigned p_len = intf_ti->c_size_fixed_part; char *param; int rc; if (mxa->heap == NULL) { mlndo_printf(0, 0, "%s[0x%02x]: no heap", msvc->name, opnum); return (MLRPC_DRC_FAULT_OUT_OF_MEMORY); } if ((ste = mlrpc_find_stub_in_svc(msvc, opnum)) == NULL) { mlndo_printf(0, 0, "%s[0x%02x]: invalid opnum", msvc->name, opnum); return (MLRPC_DRC_FAULT_REQUEST_OPNUM_INVALID); } if ((param = mlrpc_heap_malloc(mxa->heap, p_len)) == NULL) return (MLRPC_DRC_FAULT_OUT_OF_MEMORY); bzero(param, p_len); rc = mlrpc_decode_call(mxa, param); if (!MLRPC_DRC_IS_OK(rc)) return (rc); rc = (*ste->func)(param, mxa); if (rc == MLRPC_DRC_OK) rc = mlrpc_encode_return(mxa, param); return (rc); } /* * We can perform some initial setup of the response header here. * We also need to cache some of the information from the bind * negotiation for use during subsequent RPC calls. */ static void mlrpc_reply_prepare_hdr(struct mlrpc_xaction *mxa) { mlrpcconn_common_header_t *rhdr = &mxa->recv_hdr.common_hdr; mlrpcconn_common_header_t *hdr = &mxa->send_hdr.common_hdr; hdr->rpc_vers = 5; hdr->rpc_vers_minor = 0; hdr->pfc_flags = MLRPC_PFC_FIRST_FRAG + MLRPC_PFC_LAST_FRAG; hdr->packed_drep = rhdr->packed_drep; hdr->frag_length = 0; hdr->auth_length = 0; hdr->call_id = rhdr->call_id; #ifdef _BIG_ENDIAN hdr->packed_drep.intg_char_rep = MLRPC_REPLAB_CHAR_ASCII | MLRPC_REPLAB_INTG_BIG_ENDIAN; #else hdr->packed_drep.intg_char_rep = MLRPC_REPLAB_CHAR_ASCII | MLRPC_REPLAB_INTG_LITTLE_ENDIAN; #endif switch (mxa->ptype) { case MLRPC_PTYPE_BIND: hdr->ptype = MLRPC_PTYPE_BIND_ACK; mxa->send_hdr.bind_ack_hdr.max_xmit_frag = mxa->recv_hdr.bind_hdr.max_xmit_frag; mxa->send_hdr.bind_ack_hdr.max_recv_frag = mxa->recv_hdr.bind_hdr.max_recv_frag; mxa->send_hdr.bind_ack_hdr.assoc_group_id = mxa->recv_hdr.bind_hdr.assoc_group_id; if (mxa->send_hdr.bind_ack_hdr.assoc_group_id == 0) mxa->send_hdr.bind_ack_hdr.assoc_group_id = time(0); /* * Save the maximum fragment sizes * for use with subsequent requests. */ mxa->context->max_xmit_frag = mxa->recv_hdr.bind_hdr.max_xmit_frag; mxa->context->max_recv_frag = mxa->recv_hdr.bind_hdr.max_recv_frag; break; case MLRPC_PTYPE_REQUEST: hdr->ptype = MLRPC_PTYPE_RESPONSE; /* mxa->send_hdr.response_hdr.alloc_hint */ mxa->send_hdr.response_hdr.p_cont_id = mxa->recv_hdr.request_hdr.p_cont_id; mxa->send_hdr.response_hdr.cancel_count = 0; mxa->send_hdr.response_hdr.reserved = 0; break; case MLRPC_PTYPE_ALTER_CONTEXT: hdr->ptype = MLRPC_PTYPE_ALTER_CONTEXT_RESP; /* * The max_xmit_frag, max_recv_frag * and assoc_group_id are ignored. */ break; default: hdr->ptype = 0xFF; } } /* * Finish and encode the bind acknowledge (MLRPC_PTYPE_BIND_ACK) header. * The frag_length is different from a regular RPC response. */ static void mlrpc_reply_bind_ack(struct mlrpc_xaction *mxa) { mlrpcconn_common_header_t *hdr; mlrpcconn_bind_ack_hdr_t *bahdr; hdr = &mxa->send_hdr.common_hdr; bahdr = &mxa->send_hdr.bind_ack_hdr; hdr->frag_length = mlrpc_bind_ack_hdr_size(bahdr); } /* * Signal an RPC fault. The stream is reset and we overwrite whatever * was in the response header with the fault information. */ static void mlrpc_reply_fault(struct mlrpc_xaction *mxa, unsigned long drc) { mlrpcconn_common_header_t *rhdr = &mxa->recv_hdr.common_hdr; mlrpcconn_common_header_t *hdr = &mxa->send_hdr.common_hdr; struct mlndr_stream *mlnds = &mxa->send_mlnds; unsigned long fault_status; MLNDS_RESET(mlnds); hdr->rpc_vers = 5; hdr->rpc_vers_minor = 0; hdr->pfc_flags = MLRPC_PFC_FIRST_FRAG + MLRPC_PFC_LAST_FRAG; hdr->packed_drep = rhdr->packed_drep; hdr->frag_length = sizeof (mxa->send_hdr.fault_hdr); hdr->auth_length = 0; hdr->call_id = rhdr->call_id; #ifdef _BIG_ENDIAN hdr->packed_drep.intg_char_rep = MLRPC_REPLAB_CHAR_ASCII | MLRPC_REPLAB_INTG_BIG_ENDIAN; #else hdr->packed_drep.intg_char_rep = MLRPC_REPLAB_CHAR_ASCII | MLRPC_REPLAB_INTG_LITTLE_ENDIAN; #endif switch (drc & MLRPC_DRC_MASK_SPECIFIER) { case MLRPC_DRC_FAULT_OUT_OF_MEMORY: case MLRPC_DRC_FAULT_ENCODE_TOO_BIG: fault_status = MLRPC_FAULT_NCA_OUT_ARGS_TOO_BIG; break; case MLRPC_DRC_FAULT_REQUEST_PCONT_INVALID: fault_status = MLRPC_FAULT_NCA_INVALID_PRES_CONTEXT_ID; break; case MLRPC_DRC_FAULT_REQUEST_OPNUM_INVALID: fault_status = MLRPC_FAULT_NCA_OP_RNG_ERROR; break; case MLRPC_DRC_FAULT_DECODE_FAILED: case MLRPC_DRC_FAULT_ENCODE_FAILED: fault_status = MLRPC_FAULT_NCA_PROTO_ERROR; break; default: fault_status = MLRPC_FAULT_NCA_UNSPEC_REJECT; break; } mxa->send_hdr.fault_hdr.common_hdr.ptype = MLRPC_PTYPE_FAULT; mxa->send_hdr.fault_hdr.status = fault_status; mxa->send_hdr.response_hdr.alloc_hint = hdr->frag_length; } static int mlrpc_build_reply(struct mlrpc_xaction *mxa) { mlrpcconn_common_header_t *hdr = &mxa->send_hdr.common_hdr; struct mlndr_stream *mlnds = &mxa->send_mlnds; uint8_t *pdu_buf; unsigned long pdu_size; unsigned long frag_size; unsigned long pdu_data_size; unsigned long frag_data_size; frag_size = mlrpc_frag_size; pdu_size = mlnds->pdu_size; pdu_buf = mlnds->pdu_base_addr; if (pdu_size <= frag_size) { /* * Single fragment response. The PDU size may be zero * here (i.e. bind or fault response). So don't make * any assumptions about it until after the header is * encoded. */ switch (hdr->ptype) { case MLRPC_PTYPE_BIND_ACK: mlrpc_reply_bind_ack(mxa); break; case MLRPC_PTYPE_FAULT: /* already setup */ break; case MLRPC_PTYPE_RESPONSE: hdr->frag_length = pdu_size; mxa->send_hdr.response_hdr.alloc_hint = hdr->frag_length; break; default: hdr->frag_length = pdu_size; break; } mlnds->pdu_scan_offset = 0; (void) mlrpc_encode_pdu_hdr(mxa); pdu_size = mlnds->pdu_size; mlrpc_build_frag(mlnds, pdu_buf, pdu_size); return (0); } /* * Multiple fragment response. */ hdr->pfc_flags = MLRPC_PFC_FIRST_FRAG; hdr->frag_length = frag_size; mxa->send_hdr.response_hdr.alloc_hint = pdu_size - MLRPC_RSP_HDR_SIZE; mlnds->pdu_scan_offset = 0; (void) mlrpc_encode_pdu_hdr(mxa); mlrpc_build_frag(mlnds, pdu_buf, frag_size); /* * We need to update the 24-byte header in subsequent fragments. * * pdu_data_size: total data remaining to be handled * frag_size: total fragment size including header * frag_data_size: data in fragment * (i.e. frag_size - MLRPC_RSP_HDR_SIZE) */ pdu_data_size = pdu_size - MLRPC_RSP_HDR_SIZE; frag_data_size = frag_size - MLRPC_RSP_HDR_SIZE; while (pdu_data_size) { mxa->send_hdr.response_hdr.alloc_hint -= frag_data_size; pdu_data_size -= frag_data_size; pdu_buf += frag_data_size; if (pdu_data_size <= frag_data_size) { frag_data_size = pdu_data_size; frag_size = frag_data_size + MLRPC_RSP_HDR_SIZE; hdr->pfc_flags = MLRPC_PFC_LAST_FRAG; } else { hdr->pfc_flags = 0; } hdr->frag_length = frag_size; mlnds->pdu_scan_offset = 0; (void) mlrpc_encode_pdu_hdr(mxa); bcopy(mlnds->pdu_base_addr, pdu_buf, MLRPC_RSP_HDR_SIZE); mlrpc_build_frag(mlnds, pdu_buf, frag_size); if (hdr->pfc_flags & MLRPC_PFC_LAST_FRAG) break; } return (0); } /* * mlrpc_build_frag * * Build an RPC PDU fragment from the specified buffer. * If malloc fails, the client will see a header/pdu inconsistency * and report an error. */ static void mlrpc_build_frag(struct mlndr_stream *mlnds, uint8_t *buf, uint32_t len) { ndr_frag_t *frag; int size = sizeof (ndr_frag_t) + len; if ((frag = (ndr_frag_t *)malloc(size)) == NULL) return; frag->next = NULL; frag->buf = (uint8_t *)frag + sizeof (ndr_frag_t); frag->len = len; bcopy(buf, frag->buf, len); if (mlnds->frags.head == NULL) { mlnds->frags.head = frag; mlnds->frags.tail = frag; mlnds->frags.nfrag = 1; } else { mlnds->frags.tail->next = frag; mlnds->frags.tail = frag; ++mlnds->frags.nfrag; } }