1
0
mirror of https://github.com/samba-team/samba.git synced 2024-12-27 03:21:53 +03:00
samba-mirror/source3/rpc_server/srv_pipe_hnd.c
Volker Lendecke fe486d7b9f Add "queue" to writev_send
Unless higher levels queue themselves somehow, writev will *always* be queued.
So the queueing should be done at the right level.
2009-03-08 11:20:59 +01:00

1489 lines
40 KiB
C

/*
* Unix SMB/CIFS implementation.
* RPC Pipe client / server routines
* Copyright (C) Andrew Tridgell 1992-1998,
* Largely re-written : 2005
* Copyright (C) Jeremy Allison 1998 - 2005
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
#include "includes.h"
#include "librpc/gen_ndr/ndr_named_pipe_auth.h"
#undef DBGC_CLASS
#define DBGC_CLASS DBGC_RPC_SRV
static int pipes_open;
static pipes_struct *InternalPipes;
/* TODO
* the following prototypes are declared here to avoid
* code being moved about too much for a patch to be
* disrupted / less obvious.
*
* these functions, and associated functions that they
* call, should be moved behind a .so module-loading
* system _anyway_. so that's the next step...
*/
static int close_internal_rpc_pipe_hnd(struct pipes_struct *p);
/****************************************************************************
Internal Pipe iterator functions.
****************************************************************************/
pipes_struct *get_first_internal_pipe(void)
{
return InternalPipes;
}
pipes_struct *get_next_internal_pipe(pipes_struct *p)
{
return p->next;
}
/****************************************************************************
Initialise an outgoing packet.
****************************************************************************/
static bool pipe_init_outgoing_data(pipes_struct *p)
{
output_data *o_data = &p->out_data;
/* Reset the offset counters. */
o_data->data_sent_length = 0;
o_data->current_pdu_sent = 0;
prs_mem_free(&o_data->frag);
/* Free any memory in the current return data buffer. */
prs_mem_free(&o_data->rdata);
/*
* Initialize the outgoing RPC data buffer.
* we will use this as the raw data area for replying to rpc requests.
*/
if(!prs_init(&o_data->rdata, 128, p->mem_ctx, MARSHALL)) {
DEBUG(0,("pipe_init_outgoing_data: malloc fail.\n"));
return False;
}
return True;
}
/****************************************************************************
Make an internal namedpipes structure
****************************************************************************/
static struct pipes_struct *make_internal_rpc_pipe_p(TALLOC_CTX *mem_ctx,
const struct ndr_syntax_id *syntax,
const char *client_address,
struct auth_serversupplied_info *server_info)
{
pipes_struct *p;
DEBUG(4,("Create pipe requested %s\n",
get_pipe_name_from_iface(syntax)));
p = TALLOC_ZERO_P(mem_ctx, struct pipes_struct);
if (!p) {
DEBUG(0,("ERROR! no memory for pipes_struct!\n"));
return NULL;
}
if ((p->mem_ctx = talloc_init("pipe %s %p",
get_pipe_name_from_iface(syntax),
p)) == NULL) {
DEBUG(0,("open_rpc_pipe_p: talloc_init failed.\n"));
TALLOC_FREE(p);
return NULL;
}
if (!init_pipe_handle_list(p, syntax)) {
DEBUG(0,("open_rpc_pipe_p: init_pipe_handles failed.\n"));
talloc_destroy(p->mem_ctx);
TALLOC_FREE(p);
return NULL;
}
/*
* Initialize the incoming RPC data buffer with one PDU worth of memory.
* We cheat here and say we're marshalling, as we intend to add incoming
* data directly into the prs_struct and we want it to auto grow. We will
* change the type to UNMARSALLING before processing the stream.
*/
if(!prs_init(&p->in_data.data, 128, p->mem_ctx, MARSHALL)) {
DEBUG(0,("open_rpc_pipe_p: malloc fail for in_data struct.\n"));
talloc_destroy(p->mem_ctx);
close_policy_by_pipe(p);
TALLOC_FREE(p);
return NULL;
}
p->server_info = copy_serverinfo(p, server_info);
if (p->server_info == NULL) {
DEBUG(0, ("open_rpc_pipe_p: copy_serverinfo failed\n"));
talloc_destroy(p->mem_ctx);
close_policy_by_pipe(p);
TALLOC_FREE(p);
return NULL;
}
DLIST_ADD(InternalPipes, p);
memcpy(p->client_address, client_address, sizeof(p->client_address));
p->endian = RPC_LITTLE_ENDIAN;
/*
* Initialize the outgoing RPC data buffer with no memory.
*/
prs_init_empty(&p->out_data.rdata, p->mem_ctx, MARSHALL);
p->syntax = *syntax;
DEBUG(4,("Created internal pipe %s (pipes_open=%d)\n",
get_pipe_name_from_iface(syntax), pipes_open));
talloc_set_destructor(p, close_internal_rpc_pipe_hnd);
return p;
}
/****************************************************************************
Sets the fault state on incoming packets.
****************************************************************************/
static void set_incoming_fault(pipes_struct *p)
{
prs_mem_free(&p->in_data.data);
p->in_data.pdu_needed_len = 0;
p->in_data.pdu_received_len = 0;
p->fault_state = True;
DEBUG(10, ("set_incoming_fault: Setting fault state on pipe %s\n",
get_pipe_name_from_iface(&p->syntax)));
}
/****************************************************************************
Ensures we have at least RPC_HEADER_LEN amount of data in the incoming buffer.
****************************************************************************/
static ssize_t fill_rpc_header(pipes_struct *p, char *data, size_t data_to_copy)
{
size_t len_needed_to_complete_hdr = MIN(data_to_copy, RPC_HEADER_LEN - p->in_data.pdu_received_len);
DEBUG(10,("fill_rpc_header: data_to_copy = %u, len_needed_to_complete_hdr = %u, receive_len = %u\n",
(unsigned int)data_to_copy, (unsigned int)len_needed_to_complete_hdr,
(unsigned int)p->in_data.pdu_received_len ));
if (p->in_data.current_in_pdu == NULL) {
p->in_data.current_in_pdu = talloc_array(p, uint8_t,
RPC_HEADER_LEN);
}
if (p->in_data.current_in_pdu == NULL) {
DEBUG(0, ("talloc failed\n"));
return -1;
}
memcpy((char *)&p->in_data.current_in_pdu[p->in_data.pdu_received_len], data, len_needed_to_complete_hdr);
p->in_data.pdu_received_len += len_needed_to_complete_hdr;
return (ssize_t)len_needed_to_complete_hdr;
}
/****************************************************************************
Unmarshalls a new PDU header. Assumes the raw header data is in current_in_pdu.
****************************************************************************/
static ssize_t unmarshall_rpc_header(pipes_struct *p)
{
/*
* Unmarshall the header to determine the needed length.
*/
prs_struct rpc_in;
if(p->in_data.pdu_received_len != RPC_HEADER_LEN) {
DEBUG(0,("unmarshall_rpc_header: assert on rpc header length failed.\n"));
set_incoming_fault(p);
return -1;
}
prs_init_empty( &rpc_in, p->mem_ctx, UNMARSHALL);
prs_set_endian_data( &rpc_in, p->endian);
prs_give_memory( &rpc_in, (char *)&p->in_data.current_in_pdu[0],
p->in_data.pdu_received_len, False);
/*
* Unmarshall the header as this will tell us how much
* data we need to read to get the complete pdu.
* This also sets the endian flag in rpc_in.
*/
if(!smb_io_rpc_hdr("", &p->hdr, &rpc_in, 0)) {
DEBUG(0,("unmarshall_rpc_header: failed to unmarshall RPC_HDR.\n"));
set_incoming_fault(p);
prs_mem_free(&rpc_in);
return -1;
}
/*
* Validate the RPC header.
*/
if(p->hdr.major != 5 && p->hdr.minor != 0) {
DEBUG(0,("unmarshall_rpc_header: invalid major/minor numbers in RPC_HDR.\n"));
set_incoming_fault(p);
prs_mem_free(&rpc_in);
return -1;
}
/*
* If there's not data in the incoming buffer this should be the start of a new RPC.
*/
if(prs_offset(&p->in_data.data) == 0) {
/*
* AS/U doesn't set FIRST flag in a BIND packet it seems.
*/
if ((p->hdr.pkt_type == RPC_REQUEST) && !(p->hdr.flags & RPC_FLG_FIRST)) {
/*
* Ensure that the FIRST flag is set. If not then we have
* a stream missmatch.
*/
DEBUG(0,("unmarshall_rpc_header: FIRST flag not set in first PDU !\n"));
set_incoming_fault(p);
prs_mem_free(&rpc_in);
return -1;
}
/*
* If this is the first PDU then set the endianness
* flag in the pipe. We will need this when parsing all
* data in this RPC.
*/
p->endian = rpc_in.bigendian_data;
DEBUG(5,("unmarshall_rpc_header: using %sendian RPC\n",
p->endian == RPC_LITTLE_ENDIAN ? "little-" : "big-" ));
} else {
/*
* If this is *NOT* the first PDU then check the endianness
* flag in the pipe is the same as that in the PDU.
*/
if (p->endian != rpc_in.bigendian_data) {
DEBUG(0,("unmarshall_rpc_header: FIRST endianness flag (%d) different in next PDU !\n", (int)p->endian));
set_incoming_fault(p);
prs_mem_free(&rpc_in);
return -1;
}
}
/*
* Ensure that the pdu length is sane.
*/
if((p->hdr.frag_len < RPC_HEADER_LEN) || (p->hdr.frag_len > RPC_MAX_PDU_FRAG_LEN)) {
DEBUG(0,("unmarshall_rpc_header: assert on frag length failed.\n"));
set_incoming_fault(p);
prs_mem_free(&rpc_in);
return -1;
}
DEBUG(10,("unmarshall_rpc_header: type = %u, flags = %u\n", (unsigned int)p->hdr.pkt_type,
(unsigned int)p->hdr.flags ));
p->in_data.pdu_needed_len = (uint32)p->hdr.frag_len - RPC_HEADER_LEN;
prs_mem_free(&rpc_in);
p->in_data.current_in_pdu = TALLOC_REALLOC_ARRAY(
p, p->in_data.current_in_pdu, uint8_t, p->hdr.frag_len);
if (p->in_data.current_in_pdu == NULL) {
DEBUG(0, ("talloc failed\n"));
set_incoming_fault(p);
return -1;
}
return 0; /* No extra data processed. */
}
/****************************************************************************
Call this to free any talloc'ed memory. Do this before and after processing
a complete PDU.
****************************************************************************/
static void free_pipe_context(pipes_struct *p)
{
if (p->mem_ctx) {
DEBUG(3,("free_pipe_context: destroying talloc pool of size "
"%lu\n", (unsigned long)talloc_total_size(p->mem_ctx) ));
talloc_free_children(p->mem_ctx);
} else {
p->mem_ctx = talloc_init(
"pipe %s %p", get_pipe_name_from_iface(&p->syntax), p);
if (p->mem_ctx == NULL) {
p->fault_state = True;
}
}
}
/****************************************************************************
Processes a request pdu. This will do auth processing if needed, and
appends the data into the complete stream if the LAST flag is not set.
****************************************************************************/
static bool process_request_pdu(pipes_struct *p, prs_struct *rpc_in_p)
{
uint32 ss_padding_len = 0;
size_t data_len = p->hdr.frag_len - RPC_HEADER_LEN - RPC_HDR_REQ_LEN -
(p->hdr.auth_len ? RPC_HDR_AUTH_LEN : 0) - p->hdr.auth_len;
if(!p->pipe_bound) {
DEBUG(0,("process_request_pdu: rpc request with no bind.\n"));
set_incoming_fault(p);
return False;
}
/*
* Check if we need to do authentication processing.
* This is only done on requests, not binds.
*/
/*
* Read the RPC request header.
*/
if(!smb_io_rpc_hdr_req("req", &p->hdr_req, rpc_in_p, 0)) {
DEBUG(0,("process_request_pdu: failed to unmarshall RPC_HDR_REQ.\n"));
set_incoming_fault(p);
return False;
}
switch(p->auth.auth_type) {
case PIPE_AUTH_TYPE_NONE:
break;
case PIPE_AUTH_TYPE_SPNEGO_NTLMSSP:
case PIPE_AUTH_TYPE_NTLMSSP:
{
NTSTATUS status;
if(!api_pipe_ntlmssp_auth_process(p, rpc_in_p, &ss_padding_len, &status)) {
DEBUG(0,("process_request_pdu: failed to do auth processing.\n"));
DEBUG(0,("process_request_pdu: error was %s.\n", nt_errstr(status) ));
set_incoming_fault(p);
return False;
}
break;
}
case PIPE_AUTH_TYPE_SCHANNEL:
if (!api_pipe_schannel_process(p, rpc_in_p, &ss_padding_len)) {
DEBUG(3,("process_request_pdu: failed to do schannel processing.\n"));
set_incoming_fault(p);
return False;
}
break;
default:
DEBUG(0,("process_request_pdu: unknown auth type %u set.\n", (unsigned int)p->auth.auth_type ));
set_incoming_fault(p);
return False;
}
/* Now we've done the sign/seal we can remove any padding data. */
if (data_len > ss_padding_len) {
data_len -= ss_padding_len;
}
/*
* Check the data length doesn't go over the 15Mb limit.
* increased after observing a bug in the Windows NT 4.0 SP6a
* spoolsv.exe when the response to a GETPRINTERDRIVER2 RPC
* will not fit in the initial buffer of size 0x1068 --jerry 22/01/2002
*/
if(prs_offset(&p->in_data.data) + data_len > MAX_RPC_DATA_SIZE) {
DEBUG(0,("process_request_pdu: rpc data buffer too large (%u) + (%u)\n",
(unsigned int)prs_data_size(&p->in_data.data), (unsigned int)data_len ));
set_incoming_fault(p);
return False;
}
/*
* Append the data portion into the buffer and return.
*/
if(!prs_append_some_prs_data(&p->in_data.data, rpc_in_p, prs_offset(rpc_in_p), data_len)) {
DEBUG(0,("process_request_pdu: Unable to append data size %u to parse buffer of size %u.\n",
(unsigned int)data_len, (unsigned int)prs_data_size(&p->in_data.data) ));
set_incoming_fault(p);
return False;
}
if(p->hdr.flags & RPC_FLG_LAST) {
bool ret = False;
/*
* Ok - we finally have a complete RPC stream.
* Call the rpc command to process it.
*/
/*
* Ensure the internal prs buffer size is *exactly* the same
* size as the current offset.
*/
if(!prs_set_buffer_size(&p->in_data.data, prs_offset(&p->in_data.data))) {
DEBUG(0,("process_request_pdu: Call to prs_set_buffer_size failed!\n"));
set_incoming_fault(p);
return False;
}
/*
* Set the parse offset to the start of the data and set the
* prs_struct to UNMARSHALL.
*/
prs_set_offset(&p->in_data.data, 0);
prs_switch_type(&p->in_data.data, UNMARSHALL);
/*
* Process the complete data stream here.
*/
free_pipe_context(p);
if(pipe_init_outgoing_data(p)) {
ret = api_pipe_request(p);
}
free_pipe_context(p);
/*
* We have consumed the whole data stream. Set back to
* marshalling and set the offset back to the start of
* the buffer to re-use it (we could also do a prs_mem_free()
* and then re_init on the next start of PDU. Not sure which
* is best here.... JRA.
*/
prs_switch_type(&p->in_data.data, MARSHALL);
prs_set_offset(&p->in_data.data, 0);
return ret;
}
return True;
}
/****************************************************************************
Processes a finished PDU stored in current_in_pdu. The RPC_HEADER has
already been parsed and stored in p->hdr.
****************************************************************************/
static void process_complete_pdu(pipes_struct *p)
{
prs_struct rpc_in;
size_t data_len = p->in_data.pdu_received_len - RPC_HEADER_LEN;
char *data_p = (char *)&p->in_data.current_in_pdu[RPC_HEADER_LEN];
bool reply = False;
if(p->fault_state) {
DEBUG(10,("process_complete_pdu: pipe %s in fault state.\n",
get_pipe_name_from_iface(&p->syntax)));
set_incoming_fault(p);
setup_fault_pdu(p, NT_STATUS(DCERPC_FAULT_OP_RNG_ERROR));
return;
}
prs_init_empty( &rpc_in, p->mem_ctx, UNMARSHALL);
/*
* Ensure we're using the corrent endianness for both the
* RPC header flags and the raw data we will be reading from.
*/
prs_set_endian_data( &rpc_in, p->endian);
prs_set_endian_data( &p->in_data.data, p->endian);
prs_give_memory( &rpc_in, data_p, (uint32)data_len, False);
DEBUG(10,("process_complete_pdu: processing packet type %u\n",
(unsigned int)p->hdr.pkt_type ));
switch (p->hdr.pkt_type) {
case RPC_REQUEST:
reply = process_request_pdu(p, &rpc_in);
break;
case RPC_PING: /* CL request - ignore... */
DEBUG(0,("process_complete_pdu: Error. Connectionless packet type %u received on pipe %s.\n",
(unsigned int)p->hdr.pkt_type,
get_pipe_name_from_iface(&p->syntax)));
break;
case RPC_RESPONSE: /* No responses here. */
DEBUG(0,("process_complete_pdu: Error. RPC_RESPONSE received from client on pipe %s.\n",
get_pipe_name_from_iface(&p->syntax)));
break;
case RPC_FAULT:
case RPC_WORKING: /* CL request - reply to a ping when a call in process. */
case RPC_NOCALL: /* CL - server reply to a ping call. */
case RPC_REJECT:
case RPC_ACK:
case RPC_CL_CANCEL:
case RPC_FACK:
case RPC_CANCEL_ACK:
DEBUG(0,("process_complete_pdu: Error. Connectionless packet type %u received on pipe %s.\n",
(unsigned int)p->hdr.pkt_type,
get_pipe_name_from_iface(&p->syntax)));
break;
case RPC_BIND:
/*
* We assume that a pipe bind is only in one pdu.
*/
if(pipe_init_outgoing_data(p)) {
reply = api_pipe_bind_req(p, &rpc_in);
}
break;
case RPC_BINDACK:
case RPC_BINDNACK:
DEBUG(0,("process_complete_pdu: Error. RPC_BINDACK/RPC_BINDNACK packet type %u received on pipe %s.\n",
(unsigned int)p->hdr.pkt_type,
get_pipe_name_from_iface(&p->syntax)));
break;
case RPC_ALTCONT:
/*
* We assume that a pipe bind is only in one pdu.
*/
if(pipe_init_outgoing_data(p)) {
reply = api_pipe_alter_context(p, &rpc_in);
}
break;
case RPC_ALTCONTRESP:
DEBUG(0,("process_complete_pdu: Error. RPC_ALTCONTRESP on pipe %s: Should only be server -> client.\n",
get_pipe_name_from_iface(&p->syntax)));
break;
case RPC_AUTH3:
/*
* The third packet in an NTLMSSP auth exchange.
*/
if(pipe_init_outgoing_data(p)) {
reply = api_pipe_bind_auth3(p, &rpc_in);
}
break;
case RPC_SHUTDOWN:
DEBUG(0,("process_complete_pdu: Error. RPC_SHUTDOWN on pipe %s: Should only be server -> client.\n",
get_pipe_name_from_iface(&p->syntax)));
break;
case RPC_CO_CANCEL:
/* For now just free all client data and continue processing. */
DEBUG(3,("process_complete_pdu: RPC_ORPHANED. Abandoning rpc call.\n"));
/* As we never do asynchronous RPC serving, we can never cancel a
call (as far as I know). If we ever did we'd have to send a cancel_ack
reply. For now, just free all client data and continue processing. */
reply = True;
break;
#if 0
/* Enable this if we're doing async rpc. */
/* We must check the call-id matches the outstanding callid. */
if(pipe_init_outgoing_data(p)) {
/* Send a cancel_ack PDU reply. */
/* We should probably check the auth-verifier here. */
reply = setup_cancel_ack_reply(p, &rpc_in);
}
break;
#endif
case RPC_ORPHANED:
/* We should probably check the auth-verifier here.
For now just free all client data and continue processing. */
DEBUG(3,("process_complete_pdu: RPC_ORPHANED. Abandoning rpc call.\n"));
reply = True;
break;
default:
DEBUG(0,("process_complete_pdu: Unknown rpc type = %u received.\n", (unsigned int)p->hdr.pkt_type ));
break;
}
/* Reset to little endian. Probably don't need this but it won't hurt. */
prs_set_endian_data( &p->in_data.data, RPC_LITTLE_ENDIAN);
if (!reply) {
DEBUG(3,("process_complete_pdu: DCE/RPC fault sent on "
"pipe %s\n", get_pipe_name_from_iface(&p->syntax)));
set_incoming_fault(p);
setup_fault_pdu(p, NT_STATUS(DCERPC_FAULT_OP_RNG_ERROR));
prs_mem_free(&rpc_in);
} else {
/*
* Reset the lengths. We're ready for a new pdu.
*/
TALLOC_FREE(p->in_data.current_in_pdu);
p->in_data.pdu_needed_len = 0;
p->in_data.pdu_received_len = 0;
}
prs_mem_free(&rpc_in);
}
/****************************************************************************
Accepts incoming data on an rpc pipe. Processes the data in pdu sized units.
****************************************************************************/
static ssize_t process_incoming_data(pipes_struct *p, char *data, size_t n)
{
size_t data_to_copy = MIN(n, RPC_MAX_PDU_FRAG_LEN - p->in_data.pdu_received_len);
DEBUG(10,("process_incoming_data: Start: pdu_received_len = %u, pdu_needed_len = %u, incoming data = %u\n",
(unsigned int)p->in_data.pdu_received_len, (unsigned int)p->in_data.pdu_needed_len,
(unsigned int)n ));
if(data_to_copy == 0) {
/*
* This is an error - data is being received and there is no
* space in the PDU. Free the received data and go into the fault state.
*/
DEBUG(0,("process_incoming_data: No space in incoming pdu buffer. Current size = %u \
incoming data size = %u\n", (unsigned int)p->in_data.pdu_received_len, (unsigned int)n ));
set_incoming_fault(p);
return -1;
}
/*
* If we have no data already, wait until we get at least a RPC_HEADER_LEN
* number of bytes before we can do anything.
*/
if((p->in_data.pdu_needed_len == 0) && (p->in_data.pdu_received_len < RPC_HEADER_LEN)) {
/*
* Always return here. If we have more data then the RPC_HEADER
* will be processed the next time around the loop.
*/
return fill_rpc_header(p, data, data_to_copy);
}
/*
* At this point we know we have at least an RPC_HEADER_LEN amount of data
* stored in current_in_pdu.
*/
/*
* If pdu_needed_len is zero this is a new pdu.
* Unmarshall the header so we know how much more
* data we need, then loop again.
*/
if(p->in_data.pdu_needed_len == 0) {
ssize_t rret = unmarshall_rpc_header(p);
if (rret == -1 || p->in_data.pdu_needed_len > 0) {
return rret;
}
/* If rret == 0 and pdu_needed_len == 0 here we have a PDU that consists
of an RPC_HEADER only. This is a RPC_SHUTDOWN, RPC_CO_CANCEL or RPC_ORPHANED
pdu type. Deal with this in process_complete_pdu(). */
}
/*
* Ok - at this point we have a valid RPC_HEADER in p->hdr.
* Keep reading until we have a full pdu.
*/
data_to_copy = MIN(data_to_copy, p->in_data.pdu_needed_len);
/*
* Copy as much of the data as we need into the current_in_pdu buffer.
* pdu_needed_len becomes zero when we have a complete pdu.
*/
memcpy( (char *)&p->in_data.current_in_pdu[p->in_data.pdu_received_len], data, data_to_copy);
p->in_data.pdu_received_len += data_to_copy;
p->in_data.pdu_needed_len -= data_to_copy;
/*
* Do we have a complete PDU ?
* (return the number of bytes handled in the call)
*/
if(p->in_data.pdu_needed_len == 0) {
process_complete_pdu(p);
return data_to_copy;
}
DEBUG(10,("process_incoming_data: not a complete PDU yet. pdu_received_len = %u, pdu_needed_len = %u\n",
(unsigned int)p->in_data.pdu_received_len, (unsigned int)p->in_data.pdu_needed_len ));
return (ssize_t)data_to_copy;
}
/****************************************************************************
Accepts incoming data on an internal rpc pipe.
****************************************************************************/
static ssize_t write_to_internal_pipe(struct pipes_struct *p, char *data, size_t n)
{
size_t data_left = n;
while(data_left) {
ssize_t data_used;
DEBUG(10,("write_to_pipe: data_left = %u\n", (unsigned int)data_left ));
data_used = process_incoming_data(p, data, data_left);
DEBUG(10,("write_to_pipe: data_used = %d\n", (int)data_used ));
if(data_used < 0) {
return -1;
}
data_left -= data_used;
data += data_used;
}
return n;
}
/****************************************************************************
Replies to a request to read data from a pipe.
Headers are interspersed with the data at PDU intervals. By the time
this function is called, the start of the data could possibly have been
read by an SMBtrans (file_offset != 0).
Calling create_rpc_reply() here is a hack. The data should already
have been prepared into arrays of headers + data stream sections.
****************************************************************************/
static ssize_t read_from_internal_pipe(struct pipes_struct *p, char *data, size_t n,
bool *is_data_outstanding)
{
uint32 pdu_remaining = 0;
ssize_t data_returned = 0;
if (!p) {
DEBUG(0,("read_from_pipe: pipe not open\n"));
return -1;
}
DEBUG(6,(" name: %s len: %u\n", get_pipe_name_from_iface(&p->syntax),
(unsigned int)n));
/*
* We cannot return more than one PDU length per
* read request.
*/
/*
* This condition should result in the connection being closed.
* Netapp filers seem to set it to 0xffff which results in domain
* authentications failing. Just ignore it so things work.
*/
if(n > RPC_MAX_PDU_FRAG_LEN) {
DEBUG(5,("read_from_pipe: too large read (%u) requested on "
"pipe %s. We can only service %d sized reads.\n",
(unsigned int)n, get_pipe_name_from_iface(&p->syntax),
RPC_MAX_PDU_FRAG_LEN ));
n = RPC_MAX_PDU_FRAG_LEN;
}
/*
* Determine if there is still data to send in the
* pipe PDU buffer. Always send this first. Never
* send more than is left in the current PDU. The
* client should send a new read request for a new
* PDU.
*/
pdu_remaining = prs_offset(&p->out_data.frag)
- p->out_data.current_pdu_sent;
if (pdu_remaining > 0) {
data_returned = (ssize_t)MIN(n, pdu_remaining);
DEBUG(10,("read_from_pipe: %s: current_pdu_len = %u, "
"current_pdu_sent = %u returning %d bytes.\n",
get_pipe_name_from_iface(&p->syntax),
(unsigned int)prs_offset(&p->out_data.frag),
(unsigned int)p->out_data.current_pdu_sent,
(int)data_returned));
memcpy(data,
prs_data_p(&p->out_data.frag)
+ p->out_data.current_pdu_sent,
data_returned);
p->out_data.current_pdu_sent += (uint32)data_returned;
goto out;
}
/*
* At this point p->current_pdu_len == p->current_pdu_sent (which
* may of course be zero if this is the first return fragment.
*/
DEBUG(10,("read_from_pipe: %s: fault_state = %d : data_sent_length "
"= %u, prs_offset(&p->out_data.rdata) = %u.\n",
get_pipe_name_from_iface(&p->syntax), (int)p->fault_state,
(unsigned int)p->out_data.data_sent_length,
(unsigned int)prs_offset(&p->out_data.rdata) ));
if(p->out_data.data_sent_length >= prs_offset(&p->out_data.rdata)) {
/*
* We have sent all possible data, return 0.
*/
data_returned = 0;
goto out;
}
/*
* We need to create a new PDU from the data left in p->rdata.
* Create the header/data/footers. This also sets up the fields
* p->current_pdu_len, p->current_pdu_sent, p->data_sent_length
* and stores the outgoing PDU in p->current_pdu.
*/
if(!create_next_pdu(p)) {
DEBUG(0,("read_from_pipe: %s: create_next_pdu failed.\n",
get_pipe_name_from_iface(&p->syntax)));
return -1;
}
data_returned = MIN(n, prs_offset(&p->out_data.frag));
memcpy( data, prs_data_p(&p->out_data.frag), (size_t)data_returned);
p->out_data.current_pdu_sent += (uint32)data_returned;
out:
(*is_data_outstanding) = prs_offset(&p->out_data.frag) > n;
return data_returned;
}
/****************************************************************************
Close an rpc pipe.
****************************************************************************/
static int close_internal_rpc_pipe_hnd(struct pipes_struct *p)
{
if (!p) {
DEBUG(0,("Invalid pipe in close_internal_rpc_pipe_hnd\n"));
return False;
}
prs_mem_free(&p->out_data.frag);
prs_mem_free(&p->out_data.rdata);
prs_mem_free(&p->in_data.data);
if (p->auth.auth_data_free_func) {
(*p->auth.auth_data_free_func)(&p->auth);
}
TALLOC_FREE(p->mem_ctx);
free_pipe_rpc_context( p->contexts );
/* Free the handles database. */
close_policy_by_pipe(p);
DLIST_REMOVE(InternalPipes, p);
ZERO_STRUCTP(p);
TALLOC_FREE(p);
return True;
}
bool fsp_is_np(struct files_struct *fsp)
{
enum FAKE_FILE_TYPE type;
if ((fsp == NULL) || (fsp->fake_file_handle == NULL)) {
return false;
}
type = fsp->fake_file_handle->type;
return ((type == FAKE_FILE_TYPE_NAMED_PIPE)
|| (type == FAKE_FILE_TYPE_NAMED_PIPE_PROXY));
}
struct np_proxy_state {
struct async_req_queue *read_queue;
struct async_req_queue *write_queue;
int fd;
uint8_t *msg;
size_t sent;
};
static int np_proxy_state_destructor(struct np_proxy_state *state)
{
if (state->fd != -1) {
close(state->fd);
}
return 0;
}
static struct np_proxy_state *make_external_rpc_pipe_p(TALLOC_CTX *mem_ctx,
const char *pipe_name,
struct auth_serversupplied_info *server_info)
{
struct np_proxy_state *result;
struct sockaddr_un addr;
char *socket_path;
const char *socket_dir;
DATA_BLOB req_blob;
struct netr_SamInfo3 *info3;
struct named_pipe_auth_req req;
DATA_BLOB rep_blob;
uint8 rep_buf[20];
struct named_pipe_auth_rep rep;
enum ndr_err_code ndr_err;
NTSTATUS status;
ssize_t written;
result = talloc(mem_ctx, struct np_proxy_state);
if (result == NULL) {
DEBUG(0, ("talloc failed\n"));
return NULL;
}
result->fd = socket(AF_UNIX, SOCK_STREAM, 0);
if (result->fd == -1) {
DEBUG(10, ("socket(2) failed: %s\n", strerror(errno)));
goto fail;
}
talloc_set_destructor(result, np_proxy_state_destructor);
ZERO_STRUCT(addr);
addr.sun_family = AF_UNIX;
socket_dir = lp_parm_const_string(
GLOBAL_SECTION_SNUM, "external_rpc_pipe", "socket_dir",
get_dyn_NCALRPCDIR());
if (socket_dir == NULL) {
DEBUG(0, ("externan_rpc_pipe:socket_dir not set\n"));
goto fail;
}
socket_path = talloc_asprintf(talloc_tos(), "%s/np/%s",
socket_dir, pipe_name);
if (socket_path == NULL) {
DEBUG(0, ("talloc_asprintf failed\n"));
goto fail;
}
strncpy(addr.sun_path, socket_path, sizeof(addr.sun_path));
TALLOC_FREE(socket_path);
become_root();
if (sys_connect(result->fd, (struct sockaddr *)&addr) == -1) {
unbecome_root();
DEBUG(0, ("connect(%s) failed: %s\n", addr.sun_path,
strerror(errno)));
goto fail;
}
unbecome_root();
info3 = talloc(talloc_tos(), struct netr_SamInfo3);
if (info3 == NULL) {
DEBUG(0, ("talloc failed\n"));
goto fail;
}
status = serverinfo_to_SamInfo3(server_info, NULL, 0, info3);
if (!NT_STATUS_IS_OK(status)) {
TALLOC_FREE(info3);
DEBUG(0, ("serverinfo_to_SamInfo3 failed: %s\n",
nt_errstr(status)));
goto fail;
}
req.level = 1;
req.info.info1 = *info3;
ndr_err = ndr_push_struct_blob(
&req_blob, talloc_tos(), NULL, &req,
(ndr_push_flags_fn_t)ndr_push_named_pipe_auth_req);
if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
DEBUG(10, ("ndr_push_named_pipe_auth_req failed: %s\n",
ndr_errstr(ndr_err)));
goto fail;
}
DEBUG(10, ("named_pipe_auth_req(client)[%u]\n", (uint32_t)req_blob.length));
dump_data(10, req_blob.data, req_blob.length);
written = write_data(result->fd, (char *)req_blob.data,
req_blob.length);
if (written == -1) {
DEBUG(3, ("Could not write auth req data to RPC server\n"));
goto fail;
}
status = read_data(result->fd, (char *)rep_buf, sizeof(rep_buf));
if (!NT_STATUS_IS_OK(status)) {
DEBUG(3, ("Could not read auth result\n"));
goto fail;
}
rep_blob = data_blob_const(rep_buf, sizeof(rep_buf));
DEBUG(10,("name_pipe_auth_rep(client)[%u]\n", (uint32_t)rep_blob.length));
dump_data(10, rep_blob.data, rep_blob.length);
ndr_err = ndr_pull_struct_blob(
&rep_blob, talloc_tos(), NULL, &rep,
(ndr_pull_flags_fn_t)ndr_pull_named_pipe_auth_rep);
if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
DEBUG(0, ("ndr_pull_named_pipe_auth_rep failed: %s\n",
ndr_errstr(ndr_err)));
goto fail;
}
if (rep.length != 16) {
DEBUG(0, ("req invalid length: %u != 16\n",
rep.length));
goto fail;
}
if (strcmp(NAMED_PIPE_AUTH_MAGIC, rep.magic) != 0) {
DEBUG(0, ("req invalid magic: %s != %s\n",
rep.magic, NAMED_PIPE_AUTH_MAGIC));
goto fail;
}
if (!NT_STATUS_IS_OK(rep.status)) {
DEBUG(0, ("req failed: %s\n",
nt_errstr(rep.status)));
goto fail;
}
if (rep.level != 1) {
DEBUG(0, ("req invalid level: %u != 1\n",
rep.level));
goto fail;
}
result->msg = NULL;
result->read_queue = async_req_queue_init(result);
if (result->read_queue == NULL) {
goto fail;
}
result->write_queue = async_req_queue_init(result);
if (result->write_queue == NULL) {
goto fail;
}
return result;
fail:
TALLOC_FREE(result);
return NULL;
}
NTSTATUS np_open(TALLOC_CTX *mem_ctx, const char *name,
const char *client_address,
struct auth_serversupplied_info *server_info,
struct fake_file_handle **phandle)
{
const char **proxy_list;
struct fake_file_handle *handle;
proxy_list = lp_parm_string_list(-1, "np", "proxy", NULL);
handle = talloc(mem_ctx, struct fake_file_handle);
if (handle == NULL) {
return NT_STATUS_NO_MEMORY;
}
if ((proxy_list != NULL) && str_list_check_ci(proxy_list, name)) {
struct np_proxy_state *p;
p = make_external_rpc_pipe_p(handle, name, server_info);
handle->type = FAKE_FILE_TYPE_NAMED_PIPE_PROXY;
handle->private_data = p;
} else {
struct pipes_struct *p;
struct ndr_syntax_id syntax;
if (!is_known_pipename(name, &syntax)) {
TALLOC_FREE(handle);
return NT_STATUS_OBJECT_NAME_NOT_FOUND;
}
p = make_internal_rpc_pipe_p(handle, &syntax, client_address,
server_info);
handle->type = FAKE_FILE_TYPE_NAMED_PIPE;
handle->private_data = p;
}
if (handle->private_data == NULL) {
TALLOC_FREE(handle);
return NT_STATUS_PIPE_NOT_AVAILABLE;
}
*phandle = handle;
return NT_STATUS_OK;
}
struct np_write_state {
struct event_context *ev;
struct np_proxy_state *p;
struct iovec iov;
ssize_t nwritten;
};
static void np_write_trigger(struct async_req *req);
static void np_write_done(struct tevent_req *subreq);
struct async_req *np_write_send(TALLOC_CTX *mem_ctx, struct event_context *ev,
struct fake_file_handle *handle,
const uint8_t *data, size_t len)
{
struct async_req *result;
struct np_write_state *state;
NTSTATUS status;
DEBUG(6, ("np_write_send: len: %d\n", (int)len));
dump_data(50, data, len);
if (!async_req_setup(mem_ctx, &result, &state,
struct np_write_state)) {
return NULL;
}
if (len == 0) {
state->nwritten = 0;
status = NT_STATUS_OK;
goto post_status;
}
if (handle->type == FAKE_FILE_TYPE_NAMED_PIPE) {
struct pipes_struct *p = talloc_get_type_abort(
handle->private_data, struct pipes_struct);
state->nwritten = write_to_internal_pipe(p, (char *)data, len);
status = (state->nwritten >= 0)
? NT_STATUS_OK : NT_STATUS_UNEXPECTED_IO_ERROR;
goto post_status;
}
if (handle->type == FAKE_FILE_TYPE_NAMED_PIPE_PROXY) {
struct np_proxy_state *p = talloc_get_type_abort(
handle->private_data, struct np_proxy_state);
state->ev = ev;
state->p = p;
state->iov.iov_base = CONST_DISCARD(void *, data);
state->iov.iov_len = len;
if (!async_req_enqueue(p->write_queue, ev, result,
np_write_trigger)) {
goto fail;
}
return result;
}
status = NT_STATUS_INVALID_HANDLE;
post_status:
if (async_post_ntstatus(result, ev, status)) {
return result;
}
fail:
TALLOC_FREE(result);
return NULL;
}
static void np_write_trigger(struct async_req *req)
{
struct np_write_state *state = talloc_get_type_abort(
req->private_data, struct np_write_state);
struct tevent_req *subreq;
subreq = writev_send(state, state->ev, NULL, state->p->fd,
&state->iov, 1);
if (async_req_nomem(subreq, req)) {
return;
}
tevent_req_set_callback(subreq, np_write_done, req);
}
static void np_write_done(struct tevent_req *subreq)
{
struct async_req *req =
tevent_req_callback_data(subreq, struct async_req);
struct np_write_state *state = talloc_get_type_abort(
req->private_data, struct np_write_state);
ssize_t received;
int err;
received = writev_recv(subreq, &err);
if (received < 0) {
async_req_nterror(req, map_nt_error_from_unix(err));
return;
}
state->nwritten = received;
async_req_done(req);
}
NTSTATUS np_write_recv(struct async_req *req, ssize_t *pnwritten)
{
struct np_write_state *state = talloc_get_type_abort(
req->private_data, struct np_write_state);
NTSTATUS status;
if (async_req_is_nterror(req, &status)) {
return status;
}
*pnwritten = state->nwritten;
return NT_STATUS_OK;
}
static ssize_t rpc_frag_more_fn(uint8_t *buf, size_t buflen, void *priv)
{
prs_struct hdr_prs;
struct rpc_hdr_info hdr;
bool ret;
if (buflen > RPC_HEADER_LEN) {
return 0;
}
prs_init_empty(&hdr_prs, talloc_tos(), UNMARSHALL);
prs_give_memory(&hdr_prs, (char *)buf, RPC_HEADER_LEN, false);
ret = smb_io_rpc_hdr("", &hdr, &hdr_prs, 0);
prs_mem_free(&hdr_prs);
if (!ret) {
return -1;
}
return (hdr.frag_len - RPC_HEADER_LEN);
}
struct np_read_state {
struct event_context *ev;
struct np_proxy_state *p;
uint8_t *data;
size_t len;
size_t nread;
bool is_data_outstanding;
};
static void np_read_trigger(struct async_req *req);
static void np_read_done(struct tevent_req *subreq);
struct async_req *np_read_send(TALLOC_CTX *mem_ctx, struct event_context *ev,
struct fake_file_handle *handle,
uint8_t *data, size_t len)
{
struct async_req *result;
struct np_read_state *state;
NTSTATUS status;
if (!async_req_setup(mem_ctx, &result, &state,
struct np_read_state)) {
return NULL;
}
if (handle->type == FAKE_FILE_TYPE_NAMED_PIPE) {
struct pipes_struct *p = talloc_get_type_abort(
handle->private_data, struct pipes_struct);
state->nread = read_from_internal_pipe(
p, (char *)data, len, &state->is_data_outstanding);
status = (state->nread >= 0)
? NT_STATUS_OK : NT_STATUS_UNEXPECTED_IO_ERROR;
goto post_status;
}
if (handle->type == FAKE_FILE_TYPE_NAMED_PIPE_PROXY) {
struct np_proxy_state *p = talloc_get_type_abort(
handle->private_data, struct np_proxy_state);
if (p->msg != NULL) {
size_t thistime;
thistime = MIN(talloc_get_size(p->msg) - p->sent,
len);
memcpy(data, p->msg+p->sent, thistime);
state->nread = thistime;
p->sent += thistime;
if (p->sent < talloc_get_size(p->msg)) {
state->is_data_outstanding = true;
} else {
state->is_data_outstanding = false;
TALLOC_FREE(p->msg);
}
status = NT_STATUS_OK;
goto post_status;
}
state->ev = ev;
state->p = p;
state->data = data;
state->len = len;
if (!async_req_enqueue(p->read_queue, ev, result,
np_read_trigger)) {
goto fail;
}
return result;
}
status = NT_STATUS_INVALID_HANDLE;
post_status:
if (async_post_ntstatus(result, ev, status)) {
return result;
}
fail:
TALLOC_FREE(result);
return NULL;
}
static void np_read_trigger(struct async_req *req)
{
struct np_read_state *state = talloc_get_type_abort(
req->private_data, struct np_read_state);
struct tevent_req *subreq;
subreq = read_packet_send(state, state->ev, state->p->fd,
RPC_HEADER_LEN, rpc_frag_more_fn, NULL);
if (async_req_nomem(subreq, req)) {
return;
}
tevent_req_set_callback(subreq, np_read_done, req);
}
static void np_read_done(struct tevent_req *subreq)
{
struct async_req *req =
tevent_req_callback_data(subreq, struct async_req);
struct np_read_state *state = talloc_get_type_abort(
req->private_data, struct np_read_state);
ssize_t received;
size_t thistime;
int err;
received = read_packet_recv(subreq, state->p, &state->p->msg, &err);
TALLOC_FREE(subreq);
if (received == -1) {
async_req_nterror(req, map_nt_error_from_unix(err));
return;
}
thistime = MIN(received, state->len);
memcpy(state->data, state->p->msg, thistime);
state->p->sent = thistime;
state->nread = thistime;
if (state->p->sent < received) {
state->is_data_outstanding = true;
} else {
TALLOC_FREE(state->p->msg);
state->is_data_outstanding = false;
}
async_req_done(req);
return;
}
NTSTATUS np_read_recv(struct async_req *req, ssize_t *nread,
bool *is_data_outstanding)
{
struct np_read_state *state = talloc_get_type_abort(
req->private_data, struct np_read_state);
NTSTATUS status;
if (async_req_is_nterror(req, &status)) {
return status;
}
*nread = state->nread;
*is_data_outstanding = state->is_data_outstanding;
return NT_STATUS_OK;
}
/**
* Create a new RPC client context which uses a local dispatch function.
*/
NTSTATUS rpc_pipe_open_internal(TALLOC_CTX *mem_ctx,
const struct ndr_syntax_id *abstract_syntax,
NTSTATUS (*dispatch) (struct rpc_pipe_client *cli,
TALLOC_CTX *mem_ctx,
const struct ndr_interface_table *table,
uint32_t opnum, void *r),
struct auth_serversupplied_info *serversupplied_info,
struct rpc_pipe_client **presult)
{
struct rpc_pipe_client *result;
result = TALLOC_ZERO_P(mem_ctx, struct rpc_pipe_client);
if (result == NULL) {
return NT_STATUS_NO_MEMORY;
}
result->abstract_syntax = *abstract_syntax;
result->transfer_syntax = ndr_transfer_syntax;
result->dispatch = dispatch;
result->pipes_struct = make_internal_rpc_pipe_p(
result, abstract_syntax, "", serversupplied_info);
if (result->pipes_struct == NULL) {
TALLOC_FREE(result);
return NT_STATUS_NO_MEMORY;
}
result->max_xmit_frag = -1;
result->max_recv_frag = -1;
*presult = result;
return NT_STATUS_OK;
}