0
0
mirror of https://gitlab.nic.cz/labs/bird.git synced 2024-12-23 02:01:55 +00:00
bird/proto/snmp/subagent.c
2023-10-25 16:58:47 +02:00

1594 lines
41 KiB
C

/*
* BIRD -- Simple Network Management Protocol (SNMP)
*
* (c) 2022 Vojtech Vilimek <vojtech.vilimek@nic.cz>
* (c) 2022 CZ.NIC z.s.p.o
*
* Can be freely distributed and used under the terms of the GNU GPL.
*
*/
#include "lib/unaligned.h"
#include "subagent.h"
#include "snmp_utils.h"
#include "bgp_mib.h"
/* =============================================================
* Problems
* ------------------------------------------------------------
*
* change of remote ip -> no notification, no update
* same ip, different ports
* distinct VRF (two interfaces with overlapping private addrs)
* posible link-local addresses in LOCAL_IP
*
* context is allocated as copied, is it approach really needed? wouldn't it
* sufficient just use the context in rx-buffer?
*
*/
static void snmp_mib_fill2(struct snmp_proto *p, struct oid *oid, struct snmp_pdu *c);
static uint parse_response(struct snmp_proto *p, byte *buf, uint size);
static void do_response(struct snmp_proto *p, byte *buf, uint size);
static uint parse_gets2_pdu(struct snmp_proto *p, byte *buf, uint size, uint *skip);
static struct agentx_response *prepare_response(struct snmp_proto *p, struct snmp_pdu *c);
static void response_err_ind(struct agentx_response *res, uint err, uint ind);
static uint update_packet_size(struct snmp_proto *p, const byte *start, byte *end);
static struct oid *search_mib(struct snmp_proto *p, const struct oid *o_start, const struct oid *o_end, struct oid *o_curr, struct snmp_pdu *c, enum snmp_search_res *result);
static void snmp_tx(sock *sk);
u32 snmp_internet[] = { SNMP_ISO, SNMP_ORG, SNMP_DOD, SNMP_INTERNET };
static const char * const snmp_errs[] UNUSED = {
#define SNMP_ERR_SHIFT 256
[AGENTX_RES_OPEN_FAILED - SNMP_ERR_SHIFT] = "Open failed",
[AGENTX_RES_NOT_OPEN - SNMP_ERR_SHIFT] = "Not open",
[AGENTX_RES_INDEX_WRONG_TYPE - SNMP_ERR_SHIFT] = "Index wrong type",
[AGENTX_RES_INDEX_ALREADY_ALLOC - SNMP_ERR_SHIFT] = "Index already allocated",
[AGENTX_RES_INDEX_NONE_AVAIL - SNMP_ERR_SHIFT] = "Index none availlable",
[AGENTX_RES_NOT_ALLOCATED - SNMP_ERR_SHIFT] = "Not allocated",
[AGENTX_RES_UNSUPPORTED_CONTEXT - SNMP_ERR_SHIFT] = "Unsupported contex",
[AGENTX_RES_DUPLICATE_REGISTER - SNMP_ERR_SHIFT] = "Duplicate registration",
[AGENTX_RES_UNKNOWN_REGISTER - SNMP_ERR_SHIFT] = "Unknown registration",
[AGENTX_RES_UNKNOWN_AGENT_CAPS - SNMP_ERR_SHIFT] = "Unknown agent caps",
[AGENTX_RES_PARSE_ERROR - SNMP_ERR_SHIFT] = "Parse error",
[AGENTX_RES_REQUEST_DENIED - SNMP_ERR_SHIFT] = "Request denied",
[AGENTX_RES_PROCESSING_ERR - SNMP_ERR_SHIFT] = "Processing error",
};
static const char * const snmp_pkt_type[] UNUSED = {
[AGENTX_OPEN_PDU] = "Open-PDU",
[AGENTX_CLOSE_PDU] = "Close-PDU",
[AGENTX_REGISTER_PDU] = "Register-PDU",
[AGENTX_UNREGISTER_PDU] = "Unregister-PDU",
[AGENTX_GET_PDU] = "Get-PDU",
[AGENTX_GET_NEXT_PDU] = "GetNext-PDU",
[AGENTX_GET_BULK_PDU] = "GetBulk-PDU",
[AGENTX_TEST_SET_PDU] = "TestSet-PDU",
[AGENTX_COMMIT_SET_PDU] = "CommitSet-PDU",
[AGENTX_UNDO_SET_PDU] = "UndoSet-PDU",
[AGENTX_CLEANUP_SET_PDU] = "CleanupSet-PDU",
[AGENTX_NOTIFY_PDU] = "Notify-PDU",
[AGENTX_PING_PDU] = "Ping-PDU",
[AGENTX_INDEX_ALLOCATE_PDU] = "IndexAllocate-PDU",
[AGENTX_INDEX_DEALLOCATE_PDU] = "IndexDeallocate-PDU",
[AGENTX_ADD_AGENT_CAPS_PDU] = "AddAgentCaps-PDU",
[AGENTX_REMOVE_AGENT_CAPS_PDU] = "RemoveAgentCaps-PDU",
[AGENTX_RESPONSE_PDU] = "Response-PDU",
};
static void
snmp_register_ok(struct snmp_proto *p, struct agentx_response *res, struct oid *oid, u8 UNUSED class)
{
// TODO switch based on oid type
snmp_bgp_reg_ok(p, res, oid);
}
static void
snmp_register_failed(struct snmp_proto *p, struct agentx_response *res, struct oid *oid, u8 UNUSED class)
{
// TODO switch based on oid type
snmp_bgp_reg_failed(p, res, oid);
}
void
snmp_register_ack(struct snmp_proto *p, struct agentx_response *res, u8 class)
{
struct snmp_register *reg;
WALK_LIST(reg, p->register_queue)
{
// TODO add support for more mib trees (other than BGP)
if (snmp_register_same(reg, &res->h, class))
{
struct snmp_registered_oid *ro = \
mb_alloc(p->p.pool, sizeof(struct snmp_registered_oid));
ro->n.prev = ro->n.next = NULL;
ro->oid = reg->oid;
rem_node(&reg->n);
mb_free(reg);
p->register_to_ack--;
add_tail(&p->bgp_registered, &ro->n);
if (res->error == AGENTX_RES_NO_ERROR)
snmp_register_ok(p, res, ro->oid, class);
else
snmp_register_failed(p, res, ro->oid, class);
return;
}
}
}
static int
snmp_rx_skip(sock UNUSED *sk, uint UNUSED size)
{
return 1;
}
static inline void
snmp_error(struct snmp_proto *p)
{
snmp_log("changing state to RESET");
p->state = SNMP_RESET;
p->sock->rx_hook = snmp_rx_skip;
p->sock->tx_hook = snmp_tx;
}
static void
snmp_simple_response(struct snmp_proto *p, enum agentx_response_err error, u16 index)
{
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
if (c.size < sizeof(struct agentx_response))
snmp_manage_tbuf(p, &c);
struct agentx_response *res = prepare_response(p, &c);
response_err_ind(res, error, index);
sk_send(sk, sizeof(struct agentx_response));
}
static void
open_pdu(struct snmp_proto *p, struct oid *oid)
{
const struct snmp_config *cf = SKIP_BACK(struct snmp_config, cf, p->p.cf);
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
#define TIMEOUT_SIZE 4 /* 1B timeout, 3B zero padding */
if (c.size < AGENTX_HEADER_SIZE + TIMEOUT_SIZE + snmp_oid_size(oid) +
+ snmp_str_size(cf->description))
snmp_manage_tbuf(p, &c);
struct agentx_header *h = (struct agentx_header *) c.buffer;
ADVANCE(c.buffer, c.size, AGENTX_HEADER_SIZE);
SNMP_BLANK_HEADER(h, AGENTX_OPEN_PDU);
c.byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
STORE_U32(h->session_id, 1);
STORE_U32(h->transaction_id, 1);
STORE_U32(h->packet_id, 1);
c.size -= (4 + snmp_oid_size(oid) + snmp_str_size(cf->description));
c.buffer = snmp_put_fbyte(c.buffer, p->timeout);
c.buffer = snmp_put_oid(c.buffer, oid);
c.buffer = snmp_put_str(c.buffer, cf->description);
uint s = update_packet_size(p, sk->tpos, c.buffer);
sk_send(sk, s);
#undef TIMEOUT_SIZE
}
void
snmp_notify_pdu(struct snmp_proto *p, struct oid *oid, void *data, uint size, int include_uptime)
{
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
#define UPTIME_SIZE \
(6 * sizeof(u32)) /* sizeof( { u32 vb_type, u32 oid_hdr, u32 ids[4] } )*/
#define TRAP0_HEADER_SIZE \
(7 * sizeof(u32)) /* sizeof( { u32 vb_type, u32 oid_hdr, u32 ids[6] } ) */
uint sz = AGENTX_HEADER_SIZE + TRAP0_HEADER_SIZE + snmp_oid_size(oid) \
+ size;
if (include_uptime)
sz += UPTIME_SIZE;
if (c.size < sz)
snmp_manage_tbuf(p, &c);
struct agentx_header *h = (struct agentx_header *) c.buffer;
ADVANCE(c.buffer, c.size, AGENTX_HEADER_SIZE);
SNMP_BLANK_HEADER(h, AGENTX_NOTIFY_PDU);
p->packet_id++;
SNMP_SESSION(h, p);
c.byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
if (include_uptime)
{
/* sysUpTime.0 oid */
struct oid uptime = {
.n_subid = 4,
.prefix = SNMP_MGMT,
.include = 0,
.pad = 0,
};
u32 uptime_ids[] = { 1, 1, 3, 0 };
struct agentx_varbind *vb = snmp_create_varbind(c.buffer, &uptime);
for (uint i = 0; i < uptime.n_subid; i++)
STORE_U32(vb->name.ids[i], uptime_ids[i]);
snmp_varbind_ticks(vb, c.size, (current_time() TO_S) / 100);
ADVANCE(c.buffer, c.size, snmp_varbind_size(vb, c.byte_ord));
}
/* snmpTrapOID.0 oid */
struct oid trap0 = {
.n_subid = 6,
.prefix = 6,
.include = 0,
.pad = 0,
};
u32 trap0_ids[] = { 3, 1, 1, 4, 1, 0 };
struct agentx_varbind *trap_vb = snmp_create_varbind(c.buffer, &trap0);
for (uint i = 0; i < trap0.n_subid; i++)
STORE_U32(trap_vb->name.ids[i], trap0_ids[i]);
trap_vb->type = AGENTX_OBJECT_ID;
snmp_put_oid(SNMP_VB_DATA(trap_vb), oid);
ADVANCE(c.buffer, c.size, snmp_varbind_size(trap_vb, c.byte_ord));
memcpy(c.buffer, data, size);
ADVANCE(c.buffer, c.size, size);
uint s = update_packet_size(p, sk->tbuf, c.buffer);
sk_send(sk, s);
#undef TRAP0_HEADER_SIZE
#undef UPTIME_SIZE
}
/* index allocate / deallocate pdu * /
static void
de_allocate_pdu(struct snmp_proto *p, struct oid *oid, u8 type)
{
sock *sk = p->sock;
byte *buf, *pkt;
buf = pkt = sk->tbuf;
uint size = sk->tbsize;
if (size > AGENTX_HEADER_SIZE + )
{
struct agentx_header *h;
SNMP_CREATE(pkt, struct agentx_header, h);
SNMP_BLANK_HEADER(h, type);
SNMP_SESSION(h,p);
struct agentx_varbind *vb = (struct agentx_varbind *) pkt;
STORE_16(vb->type, AGENTX_OBJECT_ID);
STORE(vb->oid,
}
}
*/
/* Register-PDU / Unregister-PDU */
static void
un_register_pdu(struct snmp_proto *p, struct oid *oid, uint len, uint index, u8 type, u8 is_instance, uint UNUSED contid)
{
const struct snmp_config *cf = SKIP_BACK(struct snmp_config, cf, p->p.cf);
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
/* conditional +4 for upper-bound (optinal field) */
uint sz = AGENTX_HEADER_SIZE + snmp_oid_size(oid) + ((len > 1) ? 4 : 0);
if (c.size < sz)
snmp_manage_tbuf(p, &c);
struct agentx_header *h = (struct agentx_header *) c.buffer;
ADVANCE(c.buffer, c.size, AGENTX_HEADER_SIZE);
SNMP_HEADER(h, type, is_instance ? AGENTX_FLAG_INSTANCE_REGISTRATION : 0);
p->packet_id++;
SNMP_SESSION(h, p);
c.byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
struct agentx_un_register_hdr *ur = (struct agentx_un_register_hdr *) c.buffer;
/* do not override timeout */
STORE_U8(ur->timeout, p->timeout);
/* default priority */
STORE_U8(ur->priority, cf->priority);
STORE_U8(ur->range_subid, (len > 1) ? index : 0);
STORE_U8(ur->pad, 0);
ADVANCE(c.buffer, c.size, sizeof(struct agentx_un_register_hdr));
snmp_put_oid(c.buffer, oid);
ADVANCE(c.buffer, c.size, snmp_oid_size(oid));
/* place upper-bound if needed */
if (len > 1)
{
STORE_PTR(c.buffer, len);
ADVANCE(c.buffer, c.size, 4);
}
uint s = update_packet_size(p, sk->tpos, c.buffer);
sk_send(sk, s);
}
/* Register-PDU */
void
snmp_register(struct snmp_proto *p, struct oid *oid, uint len, uint index, u8 is_instance, uint contid)
{
un_register_pdu(p, oid, len, index, AGENTX_REGISTER_PDU, is_instance, contid);
}
/* Unregister-PDU */
void UNUSED
snmp_unregister(struct snmp_proto *p, struct oid *oid, uint len, uint index, uint contid)
{
un_register_pdu(p, oid, len, index, AGENTX_UNREGISTER_PDU, 0, contid);
}
static void
close_pdu(struct snmp_proto *p, u8 reason)
{
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
#define REASON_SIZE 4
if (c.size < AGENTX_HEADER_SIZE + REASON_SIZE)
snmp_manage_tbuf(p, &c);
struct agentx_header *h = (struct agentx_header *) c.buffer;
ADVANCE(c.buffer, c.size, AGENTX_HEADER_SIZE);
SNMP_BLANK_HEADER(h, AGENTX_CLOSE_PDU);
p->packet_id++;
SNMP_SESSION(h, p);
c.byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
snmp_put_fbyte(c.buffer, reason);
ADVANCE(c.buffer, c.size, 4);
uint s = update_packet_size(p, sk->tpos, c.buffer);
sk_send(sk, s);
#undef REASON_SIZE
}
static uint
parse_close_pdu(struct snmp_proto *p, byte * const pkt_start, uint size)
{
byte *pkt = pkt_start;
struct agentx_header *h = (void *) pkt;
ADVANCE(pkt, size, AGENTX_HEADER_SIZE);
int byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
uint pkt_size = LOAD_U32(h->payload, byte_ord);
if (pkt_size != 0)
{
snmp_simple_response(p, AGENTX_RES_GEN_ERROR, 0);
// TODO: best solution for possibly malicious pkt_size
//return AGENTX_HEADER_SIZE + MIN(size, pkt_size);
return AGENTX_HEADER_SIZE;
}
snmp_simple_response(p, AGENTX_RES_NO_ERROR, 0);
snmp_sock_disconnect(p, 1); // TODO: should we try to reconnect (2nd arg) ??
return AGENTX_HEADER_SIZE;
}
/* MUCH better signature would be
static int snmp_testset(struct snmp_proto *p, const struct agentx_varbind *vb, uint pkt_size);
*/
/* return 1 if the value could be set */
static int UNUSED
snmp_testset(struct snmp_proto *p, const struct agentx_varbind *vb, struct oid *oid, uint pkt_size)
{
/* Hard-coded no support for writing */
(void)p;(void)vb;(void)oid;(void)pkt_size;
return 0;
#if 0
// TODO better logic
if (!oid)
return 0;
switch (oid->ids[1])
{
case SNMP_BGP4_MIB:
return snmp_bgp_testset(p, vb, oid, pkt_size);
default:
return 0;
}
#endif
}
#if 0
static void UNUSED
addagentcaps_pdu(struct snmp_proto *p, struct oid *cap, char *descr,
uint descr_len)
{
ASSUME(descr != NULL && descr_len > 0);
sock *sk = p->sock;
//byte *buf = sk->tbuf;
//uint size = sk->tbsize;
// TO-DO rename to pkt and add pkt_start
byte *buf = sk->tpos;
uint size = sk->tbuf + sk->tbsize - sk->tpos;
if (size < AGENTX_HEADER_SIZE + snmp_oid_size(cap) + snmp_str_size_from_len(descr_len))
{
/* TO-DO need more mem */
return;
}
struct agentx_header *h;
SNMP_CREATE(buf, struct agentx_header, h);
SNMP_BLANK_HEADER(h, AGENTX_ADD_AGENT_CAPS_PDU);
SNMP_SESSION(h, p);
ADVANCE(buf, size, AGENTX_HEADER_SIZE);
uint in_pkt;
if (c && c->length)
{
SNMP_HAS_CONTEXT(h);
ADVANCE(buf, size, in_pkt);
}
// memcpy(buf, cap, snmp_oid_size(cap));
ADVANCE(buf, size, snmp_oid_size(cap));
in_pkt = snmp_put_nstr(buf, descr, descr_len) - buf;
ADVANCE(buf, size, in_pkt);
// make a note in the snmp_proto structure
//int ret = sk_send(sk, buf - sk->tbuf);
sk_send(sk, buf - sk->tpos);
}
static void UNUSED
removeagentcaps_pdu(struct snmp_proto *p, struct oid *cap)
{
sock *sk = p->sock;
//byte *buf = sk->tbuf;
//uint size = sk->tbsize;
// TO-DO rename to pkt and add pkt_start
byte *buf = sk->tpos;
uint size = sk->tbuf + sk->tbsize - sk->tpos;
if (size < AGENTX_HEADER_SIZE + snmp_oid_size(cap))
{
/* TO-DO need more mem */
return;
}
struct agentx_header *h;
SNMP_CREATE(buf, struct agentx_header, h);
SNMP_SESSION(h, p);
ADVANCE(buf, size, AGENTX_HEADER_SIZE);
uint in_pkt;
if (c && c->length)
{
SNMP_HAS_CONTEXT(h);
ADVANCE(buf, size, in_pkt);
}
memcpy(buf, cap, snmp_oid_size(cap));
ADVANCE(buf, size, snmp_oid_size(cap));
// update state in snmp_proto structure
sk_send(sk, buf - sk->tpos);
}
#endif
static inline void
refresh_ids(struct snmp_proto *p, struct agentx_header *h)
{
int byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
p->transaction_id = LOAD_U32(h->transaction_id, byte_ord);
p->packet_id = LOAD_U32(h->packet_id, byte_ord);
}
static uint
parse_test_set_pdu(struct snmp_proto *p, byte * const pkt_start, uint size)
{
byte *pkt = pkt_start; /* pointer to agentx-TestSet-PDU in RX-buffer */
uint s; /* final packat size */
struct agentx_response *res; /* pointer to reponse in TX-buffer */
struct agentx_header *h = (void *) pkt;
ADVANCE(pkt, size, AGENTX_HEADER_SIZE);
uint pkt_size = LOAD_U32(h->payload, h->flags & AGENTX_NETWORK_BYTE_ORDER);
if (pkt_size < size)
return 0;
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
c.byte_ord = 0; /* use little-endian */
if (c.size < AGENTX_HEADER_SIZE)
snmp_manage_tbuf(p, &c);
res = prepare_response(p, &c);
/* 0 if there is piece, that we cannot set */
int all_possible = 0;
/* the all_possible is currently hard-coded with no support for writing to mib
* variables, when implementing the mentioned support, change the initializer
* to 1
*/
#if 0
// TODO think about future value setting data structure
//struct agentx_transaction *tr = mb_alloc(...);
void *tr = mb_alloc(p->pool, 16);
struct agentx_varbind *vb;
uint sz;
while (size > 0 && all_possible)
{
vb = (void *) pkt;
sz = snmp_varbind_size(vb, 0);
if (sz > size)
/* wait for more data to arive */
return 0;
if (sz > pkt_size)
{
c.error = AGENTX_RES_PARSE_ERROR;
goto error;
}
/* Unknown VarBind type check */
if (!snmp_test_varbind(vb))
{
c.error = AGENTX_RES_PARSE_ERROR;
goto error;
}
ADVANCE(pkt, size, snmp_varbind_size(vb, 0));
// TODO remove the mb_alloc() in prefixize()
struct oid *work = snmp_prefixize(p, &vb->name, c.byte_ord);
(void)work;
all_possible = snmp_testset(p, vb, tr, work, pkt_size);
mb_free(work);
}
mb_free(tr);
#endif
goto error;
error:
s = update_packet_size(p, sk->tpos, c.buffer);
if (c.error != AGENTX_RES_NO_ERROR)
{
response_err_ind(res, c.error, c.index + 1);
snmp_error(p);
}
else if (all_possible)
response_err_ind(res, AGENTX_RES_NO_ERROR, 0);
else
/* This is an recoverable error, we do not need to reset the connection */
//response_err_ind(res, AGENTX_RES_RESOURCE_UNAVAILABLE, c.index + 1);
response_err_ind(res, AGENTX_RES_NOT_WRITABLE, c.index + 1);
sk_send(sk, s);
return pkt - pkt_start;
}
/*
* Common code for packets:
* agentx-CommitSet-PDU
* agentx-UndoSet-PDU
*/
static uint
parse_set_pdu(struct snmp_proto *p, byte * const pkt_start, uint size, uint err)
{
byte *pkt = pkt_start;
struct agentx_header *h = (void *) pkt;
ADVANCE(pkt, size, AGENTX_HEADER_SIZE);
uint pkt_size = LOAD_U32(h->payload, h->flags & AGENTX_NETWORK_BYTE_ORDER);
if (pkt_size != 0)
{
snmp_simple_response(p, AGENTX_RES_PARSE_ERROR, 0);
// TODO: best solution for possibly malicious pkt_size
return MIN(size, pkt_size + AGENTX_HEADER_SIZE);
// use varbind_list_size()??
}
struct snmp_pdu c = SNMP_PDU_CONTEXT(p->sock);
if (c.size < sizeof(struct agentx_response))
snmp_manage_tbuf(p, &c);
struct agentx_response *r = prepare_response(p, &c);
if (size < pkt_size)
{
c.error = AGENTX_RES_PARSE_ERROR;
goto error;
}
// TODO: free resource allocated by parse_test_set_pdu()
// TODO: do something meaningful
//mb_free(tr);
c.error = err;
error:;
response_err_ind(r, c.error, 0);
sk_send(p->sock, AGENTX_HEADER_SIZE);
/* Reset the connection on unrecoverable error */
if (c.error != AGENTX_RES_NO_ERROR && c.error != err)
snmp_error(p);
return pkt - pkt_start;
}
/* agentx-CommitSet-PDU */
static uint
parse_commit_set_pdu(struct snmp_proto *p, byte *pkt, uint size)
{
// don't forget to free resoures allocated by parse_test_set_pdu()
//mb_free(tr);
return parse_set_pdu(p, pkt, size, AGENTX_RES_COMMIT_FAILED);
}
/* agentx-UndoSet-PDU */
static uint
parse_undo_set_pdu(struct snmp_proto *p, byte *pkt, uint size)
{
// don't forget to free resources allocated by parse_test_set_pdu()
//mb_free(tr);
return parse_set_pdu(p, pkt, size, AGENTX_RES_UNDO_FAILED);
}
/* agentx-CleanupSet-PDU */
static uint
parse_cleanup_set_pdu(struct snmp_proto *p, byte * const pkt_start, uint size)
{
(void)p;
//TODO:
// don't forget to free resources allocated by parse_test_set_pdu()
//mb_free(p->tr);
byte *pkt = pkt_start;
struct agentx_header *h = (void *) pkt;
uint pkt_size = LOAD_U32(h->payload, h->flags & AGENTX_NETWORK_BYTE_ORDER);
/* errors are dropped silently, we must not send any agentx-Response-PDU */
if (pkt_size != 0)
{
// TODO should we free even for malformed packets ??
return MIN(size, pkt_size + AGENTX_HEADER_SIZE);
}
/* No agentx-Response-PDU is sent in response to agentx-CleanupSet-PDU */
return pkt_size;
}
/**
* parse_pkt - parse recieved AgentX packet
* @p:
* @pkt: packet buffer
* @size: number of packet bytes in buffer
* retval number of byte parsed
*
* Returns number of bytes parsed from RX-buffer.
*/
static uint
parse_pkt(struct snmp_proto *p, byte *pkt, uint size, uint *skip)
{
if (size < AGENTX_HEADER_SIZE)
return 0;
struct agentx_header *h = (void *) pkt;
int byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
uint pkt_size = LOAD_U32(h->payload, byte_ord);
/* We need to see the responses for PDU such as
* agentx-Open-PDU, agentx-Register-PDU, ...
* even when we are outside the SNMP_CONNECTED state
*/
if (h->type == AGENTX_RESPONSE_PDU)
return parse_response(p, pkt, size);
if (p->state != SNMP_CONN ||
p->session_id != LOAD_U32(h->session_id, byte_ord))
{
struct agentx_header copy = {
.session_id = p->session_id,
.transaction_id = p->transaction_id,
.packet_id = p->packet_id,
};
snmp_simple_response(p, AGENTX_RES_NOT_OPEN, 0);
p->session_id = copy.session_id;
p->transaction_id = copy.transaction_id;
p->packet_id = copy.packet_id;
// TODO: fix the parsed size for possibly malitious packets
// this could be broken when sender sends the packet in two parts
// -> size < pkt_size;
// maliciously large pkt_size -> breaks the size >= pkt_size
// maybe move this test down to parse_*() functions
return MIN(size, pkt_size + AGENTX_HEADER_SIZE);
}
if (h->flags & AGENTX_NON_DEFAULT_CONTEXT)
{
// TODO: shouldn't we return a parseError or genError for
// packets that mustn't have a non-default context ??
snmp_simple_response(p, AGENTX_RES_UNSUPPORTED_CONTEXT, 0);
// TODO: fix the parsed size (as above)
return MIN(size, pkt_size + AGENTX_HEADER_SIZE);
}
refresh_ids(p, h);
switch (h->type)
{
case AGENTX_GET_PDU:
case AGENTX_GET_NEXT_PDU:
case AGENTX_GET_BULK_PDU:
return parse_gets2_pdu(p, pkt, size, skip);
case AGENTX_CLOSE_PDU:
return parse_close_pdu(p, pkt, size);
case AGENTX_TEST_SET_PDU:
return parse_test_set_pdu(p, pkt, size);
case AGENTX_COMMIT_SET_PDU:
return parse_commit_set_pdu(p, pkt, size);
case AGENTX_UNDO_SET_PDU:
return parse_undo_set_pdu(p, pkt, size);
case AGENTX_CLEANUP_SET_PDU:
return parse_cleanup_set_pdu(p, pkt, size);
default:
/* drop the packet with unknown type silently */
return AGENTX_HEADER_SIZE;
}
}
static uint
parse_response(struct snmp_proto *p, byte *res, uint size)
{
if (size < sizeof(struct agentx_response))
return 0;
struct agentx_response *r = (void *) res;
struct agentx_header *h = &r->h;
int byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
uint pkt_size = LOAD_U32(h->payload, byte_ord);
if (size < pkt_size + AGENTX_HEADER_SIZE)
return 0;
switch (r->error)
{
case AGENTX_RES_NO_ERROR:
do_response(p, res, size);
break;
case AGENTX_RES_NOT_OPEN:
if (p->state == SNMP_LOCKED || p->state == SNMP_INIT)
snmp_startup(p);
else
snmp_connected(p->sock);
break;
case AGENTX_RES_OPEN_FAILED:
if (p->state == SNMP_LOCKED || p->state == SNMP_INIT)
{
ASSUME(p->startup_timer);
p->startup_timer->hook = snmp_startup_timeout;
// TODO: better timeout
tm_set(p->startup_timer, current_time() + p->timeout S);
}
else
{
ASSUME(p->startup_timer);
p->startup_timer->hook = snmp_reconnect;
// TODO: better timeout
tm_set(p->startup_timer, current_time() + p->timeout S);
}
break;
/* Registration errors */
case AGENTX_RES_DUPLICATE_REGISTER:
case AGENTX_RES_REQUEST_DENIED:
// TODO: more direct path to mib-specifiec code
snmp_register_ack(p, r, size);
break;
/* We are trying to unregister a MIB, the unknownRegistration has same
* effect as success */
case AGENTX_RES_UNKNOWN_REGISTER:
case AGENTX_RES_UNKNOWN_AGENT_CAPS:
case AGENTX_RES_UNSUPPORTED_CONTEXT:
case AGENTX_RES_PARSE_ERROR:
case AGENTX_RES_PROCESSING_ERR:
default:
/* erronous packet should be dropped quietly */
// TODO correct error?
snmp_log("recieved response with error '%s'", snmp_errs[get_u16(&r->error) - SNMP_ERR_SHIFT]);
break;
}
return pkt_size + AGENTX_HEADER_SIZE;
}
static void
snmp_register_mibs(struct snmp_proto *p)
{
snmp_bgp_register(p);
/* snmp_ospf_regsiter(p); ... */
}
static void
do_response(struct snmp_proto *p, byte *buf, uint size)
{
struct agentx_response *r = (void *) buf;
struct agentx_header *h = &r->h;
int byte_ord = h->flags & AGENTX_NETWORK_BYTE_ORDER;
/* TODO make it asynchronous for better speed */
switch (p->state)
{
case SNMP_INIT:
case SNMP_LOCKED:
/* silent drop of recieved packet */
break;
case SNMP_OPEN:
/* copy session info from recieved packet */
p->session_id = LOAD_U32(h->session_id, byte_ord);
refresh_ids(p, h);
/* the state needs to be changed before sending registering PDUs to
* use correct do_response action on them
*/
snmp_log("changing state to REGISTER");
p->state = SNMP_REGISTER;
snmp_register_mibs(p);
break;
case SNMP_REGISTER:;
byte *pkt = buf;
ADVANCE(pkt, size, AGENTX_HEADER_SIZE);
const struct oid *oid = (void *) pkt;
snmp_register_ack(p, r, snmp_get_mib_class(oid));
if (p->register_to_ack == 0)
{
snmp_log("changing state to CONNECTED");
p->state = SNMP_CONN;
proto_notify_state(&p->p, PS_UP);
}
break;
case SNMP_CONN:
break;
case SNMP_STOP:
break;
default:
die("unkonwn SNMP state");
}
}
u8
snmp_get_mib_class(const struct oid *oid)
{
// TODO check code paths for oid->n_subid < 3
if (oid->prefix != SNMP_MGMT && oid->ids[0] != SNMP_MIB_2)
return SNMP_CLASS_INVALID;
switch (oid->ids[1])
{
case SNMP_BGP4_MIB:
return SNMP_CLASS_BGP;
default:
return SNMP_CLASS_END;
}
}
/* return 0 if the created varbind type is END_OF_MIB_VIEW, 1 otherwise */
static int
snmp_get_next2(struct snmp_proto *p, struct oid *o_start, struct oid *o_end,
struct snmp_pdu *c)
{
enum snmp_search_res r;
struct oid *o_copy = search_mib(p, o_start, o_end, NULL, c, &r);
struct agentx_varbind *vb = NULL;
switch (r)
{
case SNMP_SEARCH_NO_OBJECT:
case SNMP_SEARCH_NO_INSTANCE:
case SNMP_SEARCH_END_OF_VIEW:;
uint sz = snmp_varbind_hdr_size_from_oid(o_start);
if (c->size < sz)
{
/* TODO create NULL varbind */
c->error = AGENTX_RES_GEN_ERROR;
return 0;
}
vb = snmp_create_varbind(c->buffer, o_start);
vb->type = AGENTX_END_OF_MIB_VIEW;
ADVANCE(c->buffer, c->size, snmp_varbind_header_size(vb));
return 0;
case SNMP_SEARCH_OK:
default:
break;
}
if (o_copy)
{
/* basicaly snmp_create_varbind(c->buffer, o_copy), but without any copying */
vb = (void *) c->buffer;
snmp_mib_fill2(p, o_copy, c);
/* override the error for GetNext-PDU object not find */
switch (vb->type)
{
case AGENTX_NO_SUCH_OBJECT:
case AGENTX_NO_SUCH_INSTANCE:
case AGENTX_END_OF_MIB_VIEW:
vb->type = AGENTX_END_OF_MIB_VIEW;
return 0;
default:
return 1;
}
}
if (c->size < snmp_varbind_hdr_size_from_oid(o_start))
{
// TODO FIXME this is a bit tricky as we need to renew all TX buffer pointers
snmp_manage_tbuf(p, c);
}
vb = snmp_create_varbind(c->buffer, o_start);
vb->type = AGENTX_END_OF_MIB_VIEW;
ADVANCE(c->buffer, c->size, snmp_varbind_header_size(vb));
return 0;
}
/* returns 0 if the created varbind has type EndOfMibView, 1 otherwise */
static int
snmp_get_bulk2(struct snmp_proto *p, struct oid *o_start, struct oid *o_end,
struct agentx_bulk_state *state, struct snmp_pdu *c)
{
struct oid *o_curr = NULL;
struct oid *o_predecessor = NULL;
enum snmp_search_res r;
uint i = 0;
do
{
o_predecessor = o_curr;
o_curr = search_mib(p, o_start, o_end, o_curr, c, &r);
i++;
} while (o_curr && i < state->repetition);
// TODO check if the approach below works
// it need to generate varbinds that will be only of type EndOfMibView
/* Object Identifier fall-backs */
if (!o_curr)
o_curr = o_predecessor;
if (!o_curr)
o_curr = o_start;
uint sz = snmp_varbind_hdr_size_from_oid(o_curr);
if (c->size < sz)
{
c->error = AGENTX_RES_GEN_ERROR;
return 0;
}
/* we need the varbind handle to be able to override it's type */
struct agentx_varbind *vb = (void *) c->buffer;
vb->type = AGENTX_END_OF_MIB_VIEW;
if (r == SNMP_SEARCH_OK)
/* the varbind will be recreated inside the snmp_mib_fill2() */
snmp_mib_fill2(p, o_curr, c);
else
ADVANCE(c->buffer, c->size, snmp_varbind_header_size(vb));
/* override the error for GetBulk-PDU object not found */
switch (vb->type)
{
case AGENTX_NO_SUCH_OBJECT:
case AGENTX_NO_SUCH_INSTANCE:
case AGENTX_END_OF_MIB_VIEW:
vb->type = AGENTX_END_OF_MIB_VIEW;
return 0;
default:
return 1;
}
}
static inline uint
update_packet_size(struct snmp_proto *p, const byte *start, byte *end)
{
struct agentx_header *h = (void *) p->sock->tpos;
size_t s = snmp_pkt_len(start, end);
STORE_U32(h->payload, s);
return AGENTX_HEADER_SIZE + s;
}
static inline void
response_err_ind(struct agentx_response *res, uint err, uint ind)
{
STORE_U32(res->error, err);
if (err != AGENTX_RES_NO_ERROR && err != AGENTX_RES_PARSE_ERROR)
STORE_U32(res->index, ind);
else
STORE_U32(res->index, 0);
}
static uint
parse_gets2_pdu(struct snmp_proto *p, byte * const pkt_start, uint size, uint *skip)
{
// TODO checks for c.size underflow
uint ret = 0;
struct oid *o_start = NULL, *o_end = NULL;
byte *pkt = pkt_start;
struct agentx_header *h = (void *) pkt;
ADVANCE(pkt, size, AGENTX_HEADER_SIZE);
uint pkt_size = LOAD_U32(h->payload, h->flags & AGENTX_NETWORK_BYTE_ORDER);
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
// TODO better handling of endianness
c.byte_ord = 0; /* use little-endian */
/*
* Get-Bulk processing stops if all the varbind have type END_OF_MIB_VIEW
* has_any is true if some varbind has type other than END_OF_MIB_VIEW
*/
int has_any = 0;
struct agentx_bulk_state bulk_state = { };
if (h->type == AGENTX_GET_BULK_PDU)
{
if (size < sizeof(struct agentx_getbulk))
goto wait;
if (pkt_size < sizeof(struct agentx_getbulk))
{
c.error = AGENTX_RES_PARSE_ERROR;
c.index = 0;
ret = MIN(size, pkt_size + AGENTX_HEADER_SIZE);
goto error;
}
struct agentx_getbulk *bulk_info = (void *) pkt;
ADVANCE(pkt, pkt_size, sizeof(struct agentx_getbulk));
bulk_state = (struct agentx_bulk_state) {
.getbulk = {
.non_repeaters = LOAD_U32(bulk_info->non_repeaters, c.byte_ord),
.max_repetitions = LOAD_U32(bulk_info->max_repetitions, c.byte_ord),
},
/* In contrast to the RFC, we use 0-based indices. */
.index = 0,
.repetition = 0,
};
}
if (c.size < sizeof(struct agentx_response))
{
snmp_manage_tbuf(p, &c);
// TODO renew pkt, pkt_start pointers
}
struct agentx_response *response_header = prepare_response(p, &c);
while (c.error == AGENTX_RES_NO_ERROR && size > 0 && pkt_size > 0)
{
if (size < snmp_oid_sizeof(0))
goto partial;
/* We load search range start OID */
const struct oid *o_start_b = (void *) pkt;
uint sz;
if ((sz = snmp_oid_size(o_start_b)) > pkt_size)
{
c.error = AGENTX_RES_PARSE_ERROR;
ret = MIN(size, pkt_size + AGENTX_HEADER_SIZE);
goto error;
}
/*
* If we already have written same relevant data to the TX buffer, then
* we send processed part, otherwise we don't have anything to send and
* need to wait for more data to be recieved.
*/
if (sz > size && c.index > 0)
{
goto partial; /* send already processed part */
}
else if (sz > size)
{
goto wait;
}
/* Update buffer pointer and remaining size counters. */
ADVANCE(pkt, pkt_size, sz);
size -= sz;
/*
* We load search range end OID
* The exactly same process of sanity checking is preformed while loading
* the SearchRange's end OID
*/
const struct oid *o_end_b = (void *) pkt;
if ((sz = snmp_oid_size(o_end_b)) > pkt_size)
{
c.error = AGENTX_RES_PARSE_ERROR;
ret = MIN(size, pkt_size + AGENTX_HEADER_SIZE);
goto error;
}
if (sz > size && c.index > 0)
{
size += snmp_oid_size(o_start_b);
goto partial;
}
else if (sz > size)
{
goto wait;
}
ADVANCE(pkt, pkt_size, sz);
size -= sz;
// TODO check for oversized OIDs before any allocation (in prefixize())
/* We create copy of OIDs outside of rx-buffer and also prefixize them */
o_start = snmp_prefixize(p, o_start_b, c.byte_ord);
o_end = snmp_prefixize(p, o_end_b, c.byte_ord);
if (!snmp_is_oid_empty(o_end) && snmp_oid_compare(o_start, o_end) > 0)
{
c.error = AGENTX_RES_GEN_ERROR;
ret = MIN(size, pkt_size + AGENTX_HEADER_SIZE);
goto error;
}
/* TODO find mib_class, check if type is GET of GET_NEXT, act acordingly */
switch (h->type)
{
case AGENTX_GET_PDU:
snmp_mib_fill2(p, o_start, &c);
break;
case AGENTX_GET_NEXT_PDU:
snmp_get_next2(p, o_start, o_end, &c);
break;
case AGENTX_GET_BULK_PDU:
if (c.index >= bulk_state.getbulk.non_repeaters)
bulk_state.repeaters++;
// store the o_start, o_end
/* The behavior of GetBulk pdu in the first iteration is
* identical to GetNext pdu. */
has_any = snmp_get_next2(p, o_start, o_end, &c) | has_any;
break;
default:
die("incorrect usage");
}
mb_free(o_start);
o_start = NULL;
mb_free(o_end);
o_end = NULL;
c.index++;
} /* while (c.error == AGENTX_RES_NO_ERROR && size > 0) */
if (h->type == AGENTX_GET_BULK_PDU)
{
for (bulk_state.repetition++;
has_any && bulk_state.repetition < bulk_state.getbulk.max_repetitions;
bulk_state.repetition++)
{
// TODO find propper start and end
struct oid *start = NULL;
struct oid *end = NULL;
has_any = 0;
for (bulk_state.index = 0; bulk_state.index < bulk_state.repeaters;
bulk_state.repeaters++)
has_any = has_any || snmp_get_bulk2(p, start, end, &bulk_state, &c);
}
}
/* send the constructed packet */
struct agentx_response *res = (void *) sk->tbuf;
/* We update the error, index pair on the beginning of the packet. */
response_err_ind(res, c.error, c.index + 1);
uint s = update_packet_size(p, (byte *) response_header, c.buffer);
/* We send the message in TX-buffer. */
sk_send(sk, s);
// TODO think through the error state
/* number of bytes parsed from RX-buffer */
ret = pkt - pkt_start;
goto free;
partial:
/* need to tweak RX buffer packet size */
(c.byte_ord) ? put_u32(&h->payload, pkt_size) : (h->payload = pkt_size);
*skip = AGENTX_HEADER_SIZE;
/* number of bytes parsed from RX-buffer */
ret = pkt - pkt_start;
goto free;
wait:
p->packet_id--; /* we did not use the packetID */
ret = 0;
goto free;
error:
if (c.index > UINT16_MAX)
snmp_simple_response(p, AGENTX_RES_GEN_ERROR, UINT16_MAX);
else
snmp_simple_response(p, c.error, c.index);
free:
mb_free(o_start);
mb_free(o_end);
return ret;
}
void
snmp_start_subagent(struct snmp_proto *p)
{
/* blank oid means unsupported */
struct oid *blank = snmp_oid_blank(p);
open_pdu(p, blank);
p->state = SNMP_OPEN;
mb_free(blank);
}
void
snmp_stop_subagent(struct snmp_proto *p)
{
if (p->state == SNMP_STOP)
close_pdu(p, AGENTX_CLOSE_SHUTDOWN);
}
static inline int
oid_prefix(struct oid *o, u32 *prefix, uint len)
{
for (uint i = 0; i < len; i++)
if (o->ids[i] != prefix[i])
return 0;
return 1;
}
int
snmp_rx(sock *sk, uint size)
{
struct snmp_proto *p = sk->data;
byte *pkt_start = sk->rbuf;
byte *end = pkt_start + size;
/*
* In some cases we want to save the header for future parsing, skip is number
* of bytes that should not be overriden by memmove()
*/
uint skip = 0;
while (end >= pkt_start + AGENTX_HEADER_SIZE && skip == 0)
{
uint parsed_len = parse_pkt(p, pkt_start, size, &skip);
if (parsed_len == 0)
break;
pkt_start += parsed_len;
size -= parsed_len;
}
/* Incomplete packets */
if (skip != 0 || pkt_start != end)
{
memmove(sk->rbuf + skip, pkt_start, size);
sk->rpos = sk->rbuf + size + skip;
return 0;
}
return 1;
}
/* snmp_tx - used to reset the connection when the
* agentx-Response-PDU was sent
*/
static void
snmp_tx(sock *sk)
{
struct snmp_proto *p = sk->data;
snmp_sock_disconnect(p, 1);
}
/* Ping-PDU */
void
snmp_ping(struct snmp_proto *p)
{
sock *sk = p->sock;
struct snmp_pdu c = SNMP_PDU_CONTEXT(sk);
if (c.size < AGENTX_HEADER_SIZE)
snmp_manage_tbuf(p, &c);
struct agentx_header *h = (struct agentx_header *) c.buffer;
ADVANCE(c.buffer, c.size, AGENTX_HEADER_SIZE);
SNMP_BLANK_HEADER(h, AGENTX_PING_PDU);
p->packet_id++;
SNMP_SESSION(h, p);
c.byte_ord = AGENTX_NETWORK_BYTE_ORDER;
/* sending only header -> pkt - buf */
uint s = update_packet_size(p, sk->tpos, c.buffer);
sk_send(sk, s);
}
static inline int
is_bgp4_mib_prefix(struct oid *o)
{
if (o->prefix == SNMP_MGMT && o->ids[0] == SNMP_MIB_2 &&
o->ids[1] == SNMP_BGP4_MIB)
return 1;
else
return 0;
}
static inline int
has_inet_prefix(struct oid *o)
{
return (o->n_subid > 4 && o->ids[0] == 1 &&
o->ids[1] == 3 && o->ids[2] == 6 &&
o->ids[3] == 1);
}
/**
* snmp_search_check_end_oid - check if oid is before SearchRange end
*
* @found: best oid found in MIB tree
* @bound: upper bound specified in SearchRange
*
* check if found oid meet the SearchRange upper bound condition in
* lexicographical order, returns boolean value
*/
int snmp_search_check_end_oid(const struct oid *found, const struct oid *bound)
{
if (snmp_is_oid_empty(bound))
return 1;
return (snmp_oid_compare(found, bound) < 0);
}
/* tree is tree with "internet" prefix .1.3.6.1
working only with o_start, o_end allocated in heap (not from buffer)*/
static struct oid *
search_mib(struct snmp_proto *p, const struct oid *o_start, const struct oid *o_end,
struct oid *o_curr, struct snmp_pdu UNUSED *c,
enum snmp_search_res *result)
{
ASSUME(o_start != NULL);
if (o_curr && (o_curr->n_subid < 2 || o_curr->ids[0] != 1))
return NULL;
if (!o_curr && (o_start->n_subid < 2 || o_start->ids[0] != 1))
return NULL;
if (!o_curr)
{
o_curr = snmp_oid_duplicate(p->pool, o_start);
// XXX is it right time to free o_start right now (here) ?
// not for use in snmp_get_next2() the o_start comes and ends in _gets2_()
}
const struct oid *blank = NULL;
if (!snmp_is_oid_empty(o_end) &&
snmp_get_mib_class(o_curr) < snmp_get_mib_class(o_end))
{
o_end = blank = snmp_oid_blank(p);
}
enum snmp_search_res r;
switch (o_curr->ids[1])
{
case SNMP_BGP4_MIB:
r = snmp_bgp_search2(p, &o_curr, o_end, 0);
if (r == SNMP_SEARCH_OK)
{
*result = r;
break;
return o_curr;
}
// TODO add early break for o_end less then thinkable maximum in each tree
/* fall through */
default:
if (o_curr) mb_free(o_curr);
o_curr = snmp_oid_duplicate(p->pool, o_start);
*result = SNMP_SEARCH_END_OF_VIEW;
break;
}
if (o_end == blank)
mb_free((void *) blank);
return o_curr;
}
/**
* snmp_prefixize - return prefixed oid copy if possible
* @proto: allocation pool holder
* @oid: from packet loaded object identifier
* @byte_ord: byte order of @oid
*
* Returns prefixed (meaning with nonzero prefix field) oid copy of @oid if
* possible, NULL otherwise. Returned pointer is always allocated from @proto's
* pool not a pointer to recieve buffer (from which is most likely @oid).
*/
struct oid *
snmp_prefixize(struct snmp_proto *proto, const struct oid *oid, int byte_ord)
{
ASSERT(oid != NULL);
if (snmp_is_oid_empty(oid))
{
/* allocate new zeroed oid */
return snmp_oid_blank(proto);
}
/* already in prefixed form */
else if (oid->prefix != 0) {
struct oid *new = snmp_oid_duplicate(proto->pool, oid);
return new;
}
if (oid->n_subid < 5)
return NULL;
for (int i = 0; i < 4; i++)
if (LOAD_U32(oid->ids[i], byte_ord) != snmp_internet[i])
return NULL;
/* validity check here */
if (oid->ids[4] >= 256)
return NULL;
struct oid *new = mb_alloc(proto->pool,
sizeof(struct oid) + MAX((oid->n_subid - 5) * sizeof(u32), 0));
memcpy(new, oid, sizeof(struct oid));
new->n_subid = oid->n_subid - 5;
/* validity check before allocation => ids[4] < 256
and can be copied to one byte new->prefix */
new->prefix = oid->ids[4];
memcpy(&new->ids, &oid->ids[5], new->n_subid * sizeof(u32));
return new;
}
static void
snmp_mib_fill2(struct snmp_proto *p, struct oid *oid,
struct snmp_pdu *c)
{
ASSUME(oid != NULL);
if (c->size < snmp_varbind_hdr_size_from_oid(oid))
snmp_manage_tbuf(p, c);
struct agentx_varbind *vb = snmp_create_varbind(c->buffer, oid);
if (oid->n_subid < 2 || (oid->prefix != SNMP_MGMT && oid->ids[0] != SNMP_MIB_2))
{
vb->type = AGENTX_NO_SUCH_OBJECT;
ADVANCE(c->buffer, c->size, snmp_varbind_header_size(vb));
return;
}
u8 mib_class = snmp_get_mib_class(oid);
switch (mib_class)
{
case SNMP_CLASS_BGP:
snmp_bgp_fill(p, vb, c);
break;
case SNMP_CLASS_INVALID:
case SNMP_CLASS_END:
default:
break;
vb->type = AGENTX_NO_SUCH_OBJECT;
ADVANCE(c->buffer, c->size, snmp_varbind_header_size(vb));
}
}
/**
*
* Important note: After managing insufficient buffer size all in buffer pointers
* are invalidated!
*/
void
snmp_manage_tbuf(struct snmp_proto UNUSED *p, struct snmp_pdu *c)
{
sock *sk = p->sock;
sk_set_tbsize(sk, sk->tbsize + 2048);
c->size += 2048;
}
static struct agentx_response *
prepare_response(struct snmp_proto *p, struct snmp_pdu *c)
{
struct agentx_response *r = (void *) c->buffer;
struct agentx_header *h = &r->h;
SNMP_BLANK_HEADER(h, AGENTX_RESPONSE_PDU);
SNMP_SESSION(h, p);
/* protocol doesn't care about subagent upTime */
STORE_U32(r->uptime, 0);
STORE_U16(r->error, AGENTX_RES_NO_ERROR);
STORE_U16(r->index, 0);
ADVANCE(c->buffer, c->size, sizeof(struct agentx_response));
return r;
}
#undef SNMP_ERR_SHIFT