minix/servers/lwip/udp.c

419 lines
9.3 KiB
C
Raw Normal View History

#include <stdlib.h>
#include <minix/sysutil.h>
#include <net/ioctl.h>
#include <net/gen/in.h>
#include <net/gen/udp.h>
#include <net/gen/udp_io.h>
#include <net/gen/udp_io_hdr.h>
#include <lwip/udp.h>
#include <lwip/ip_addr.h>
#include "socket.h"
#include "proto.h"
#define UDP_BUF_SIZE (4 << 10)
#define sock_alloc_buf(s) debug_malloc(s)
#define sock_free_buf(x) debug_free(x)
#if 0
#define debug_udp_print(str, ...) printf("LWIP %s:%d : " str "\n", \
__func__, __LINE__, ##__VA_ARGS__)
#else
#define debug_udp_print(...) debug_print(__VA_ARGS__)
#endif
struct udp_recv_data {
ip_addr_t ip;
u16_t port;
struct pbuf * pbuf;
};
#define udp_recv_alloc() debug_malloc(sizeof(struct udp_recv_data))
static void udp_recv_free(void * data)
{
if (((struct udp_recv_data *)data)->pbuf)
pbuf_free(((struct udp_recv_data *)data)->pbuf);
debug_free(data);
}
static int udp_op_open(struct socket * sock, __unused message * m)
{
struct udp_pcb * pcb;
debug_udp_print("socket num %ld", get_sock_num(sock));
if (!(pcb = udp_new()))
return ENOMEM;
sock->buf = NULL;
sock->buf_size = 0;
sock->pcb = pcb;
return OK;
}
static void udp_op_close(struct socket * sock, __unused message * m)
{
debug_udp_print("socket num %ld", get_sock_num(sock));
/* deque and free all enqueued data before closing */
sock_dequeue_data_all(sock, udp_recv_free);
if (sock->pcb)
udp_remove(sock->pcb);
assert(sock->buf == NULL);
/* mark it as unused */
sock->ops = NULL;
sock_reply(sock, OK);
}
static int udp_do_receive(struct socket * sock,
message * m,
struct udp_pcb *pcb,
struct pbuf *pbuf,
ip_addr_t *addr,
u16_t port)
{
struct pbuf * p;
unsigned rem_len = m->COUNT;
unsigned written = 0, hdr_sz = 0;
int err;
debug_udp_print("user buffer size : %d", rem_len);
/* FIXME make it both a single copy */
if (!(sock->usr_flags & NWUO_RWDATONLY)) {
udp_io_hdr_t hdr;
hdr.uih_src_addr = addr->addr;
hdr.uih_src_port = htons(port);
hdr.uih_dst_addr = pcb->local_ip.addr;
hdr.uih_dst_port = htons(pcb->local_port);
hdr.uih_data_len = 0;
hdr.uih_ip_opt_len = 0;
err = copy_to_user(m->m_source,
&hdr, sizeof(hdr),
(cp_grant_id_t) m->IO_GRANT,
0);
if (err != OK)
return err;
rem_len -= (hdr_sz = sizeof(hdr));
}
for (p = pbuf; p && rem_len; p = p->next) {
size_t cp_len;
cp_len = (rem_len < p->len) ? rem_len : p->len;
err = copy_to_user(m->m_source, p->payload, cp_len,
(cp_grant_id_t) m->IO_GRANT,
hdr_sz + written);
if (err != OK)
return err;
written += cp_len;
rem_len -= cp_len;
}
debug_udp_print("copied %d bytes", written + hdr_sz);
return written + hdr_sz;
}
static void udp_recv_callback(void *arg,
struct udp_pcb *pcb,
struct pbuf *pbuf,
ip_addr_t *addr,
u16_t port)
{
struct socket * sock = (struct socket *) arg;
struct udp_recv_data * data;
debug_udp_print("socket num : %ld addr : %x port : %d\n",
get_sock_num(sock), (unsigned int) addr->addr, port);
if (sock->flags & SOCK_FLG_OP_PENDING) {
/* we are resuming a suspended operation */
int ret;
ret = udp_do_receive(sock, &sock->mess, pcb, pbuf, addr, port);
if (ret > 0) {
pbuf_free(pbuf);
sock_revive(sock, ret);
sock->flags &= ~SOCK_FLG_OP_PENDING;
return;
} else {
sock_revive(sock, ret);
sock->flags &= ~SOCK_FLG_OP_PENDING;
}
}
/* Do not enqueue more data than allowed */
if (sock->recv_data_size > UDP_BUF_SIZE) {
pbuf_free(pbuf);
return;
}
/*
* nobody is waiting for the data or an error occured above, we enqueue
* the packet
*/
if (!(data = udp_recv_alloc())) {
pbuf_free(pbuf);
return;
}
data->ip = *addr;
data->port = port;
data->pbuf = pbuf;
if (sock_enqueue_data(sock, data, data->pbuf->tot_len) != OK) {
udp_recv_free(data);
return;
}
/*
* We don't need to notify when somebody is already waiting, reviving
* read operation will do the trick for us. But we must announce new
* data available here.
*/
if (sock_select_read_set(sock))
sock_select_notify(sock);
}
static void udp_op_read(struct socket * sock, message * m)
{
debug_udp_print("socket num %ld", get_sock_num(sock));
if (sock->recv_head) {
/* data available receive immeditely */
struct udp_recv_data * data;
int ret;
data = (struct udp_recv_data *) sock->recv_head->data;
ret = udp_do_receive(sock, m, (struct udp_pcb *) sock->pcb,
data->pbuf, &data->ip, data->port);
if (ret > 0) {
sock_dequeue_data(sock);
sock->recv_data_size -= data->pbuf->tot_len;
udp_recv_free(data);
}
sock_reply(sock, ret);
} else {
/* store the message so we know how to reply */
sock->mess = *m;
/* operation is being processes */
sock->flags |= SOCK_FLG_OP_PENDING;
debug_udp_print("no data to read, suspending\n");
sock_reply(sock, SUSPEND);
}
}
static int udp_op_send(__unused struct socket * sock,
__unused struct pbuf * pbuf,
__unused message * m)
{
int err;
debug_udp_print("pbuf len %d\n", pbuf->len);
if ((err = udp_send(sock->pcb, pbuf)) == ERR_OK)
return m->COUNT;
else {
debug_udp_print("udp_send failed %d", err);
return EIO;
}
}
static int udp_op_sendto(struct socket * sock, struct pbuf * pbuf, message * m)
{
int err;
udp_io_hdr_t hdr;
hdr = *(udp_io_hdr_t *) pbuf->payload;
pbuf_header(pbuf, -(s16_t)sizeof(udp_io_hdr_t));
debug_udp_print("data len %d pbuf len %d\n",
hdr.uih_data_len, pbuf->len);
if ((err = udp_sendto(sock->pcb, pbuf, (ip_addr_t *) &hdr.uih_dst_addr,
ntohs(hdr.uih_dst_port))) == ERR_OK)
return m->COUNT;
else {
debug_udp_print("udp_sendto failed %d", err);
return EIO;
}
}
static void udp_op_write(struct socket * sock, message * m)
{
int ret;
struct pbuf * pbuf;
debug_udp_print("socket num %ld data size %d",
get_sock_num(sock), m->COUNT);
pbuf = pbuf_alloc(PBUF_TRANSPORT, m->COUNT, PBUF_POOL);
if (!pbuf) {
ret = ENOMEM;
goto write_err;
}
if ((ret = copy_from_user(m->m_source, pbuf->payload, m->COUNT,
(cp_grant_id_t) m->IO_GRANT, 0)) != OK) {
pbuf_free(pbuf);
goto write_err;
}
if (sock->usr_flags & NWUO_RWDATONLY)
ret = udp_op_send(sock, pbuf, m);
else
ret = udp_op_sendto(sock, pbuf, m);
if (pbuf_free(pbuf) == 0) {
panic("We cannot buffer udp packets yet!");
}
write_err:
sock_reply(sock, ret);
}
static void udp_set_opt(struct socket * sock, message * m)
{
int err;
nwio_udpopt_t udpopt;
struct udp_pcb * pcb = (struct udp_pcb *) sock->pcb;
ip_addr_t loc_ip = ip_addr_any;
assert(pcb);
err = copy_from_user(m->m_source, &udpopt, sizeof(udpopt),
(cp_grant_id_t) m->IO_GRANT, 0);
if (err != OK)
sock_reply(sock, err);
debug_udp_print("udpopt.nwuo_flags = 0x%lx", udpopt.nwuo_flags);
debug_udp_print("udpopt.nwuo_remaddr = 0x%x",
(unsigned int) udpopt.nwuo_remaddr);
debug_udp_print("udpopt.nwuo_remport = 0x%x",
ntohs(udpopt.nwuo_remport));
debug_udp_print("udpopt.nwuo_locaddr = 0x%x",
(unsigned int) udpopt.nwuo_locaddr);
debug_udp_print("udpopt.nwuo_locport = 0x%x",
ntohs(udpopt.nwuo_locport));
sock->usr_flags = udpopt.nwuo_flags;
/*
* We will only get data from userspace and the remote address
* and port are being set which means that from now on we must
* know where to send data. Thus we should interpret this as
* connect() call
*/
if (sock->usr_flags & NWUO_RWDATONLY &&
sock->usr_flags & NWUO_RP_SET &&
sock->usr_flags & NWUO_RA_SET)
udp_connect(pcb, (ip_addr_t *) &udpopt.nwuo_remaddr,
ntohs(udpopt.nwuo_remport));
/* Setting local address means binding */
if (sock->usr_flags & NWUO_LP_SET)
udp_bind(pcb, &loc_ip, ntohs(udpopt.nwuo_locport));
/* We can only bind to random local port */
if (sock->usr_flags & NWUO_LP_SEL)
udp_bind(pcb, &loc_ip, 0);
/* register a receive hook */
udp_recv((struct udp_pcb *) sock->pcb, udp_recv_callback, sock);
sock_reply(sock, OK);
}
static void udp_get_opt(struct socket * sock, message * m)
{
int err;
nwio_udpopt_t udpopt;
struct udp_pcb * pcb = (struct udp_pcb *) sock->pcb;
assert(pcb);
udpopt.nwuo_locaddr = pcb->local_ip.addr;
udpopt.nwuo_locport = htons(pcb->local_port);
udpopt.nwuo_remaddr = pcb->remote_ip.addr;
udpopt.nwuo_remport = htons(pcb->remote_port);
udpopt.nwuo_flags = sock->usr_flags;
debug_udp_print("udpopt.nwuo_flags = 0x%lx", udpopt.nwuo_flags);
debug_udp_print("udpopt.nwuo_remaddr = 0x%x",
(unsigned int) udpopt.nwuo_remaddr);
debug_udp_print("udpopt.nwuo_remport = 0x%x",
ntohs(udpopt.nwuo_remport));
debug_udp_print("udpopt.nwuo_locaddr = 0x%x",
(unsigned int) udpopt.nwuo_locaddr);
debug_udp_print("udpopt.nwuo_locport = 0x%x",
ntohs(udpopt.nwuo_locport));
if ((unsigned) m->COUNT < sizeof(udpopt)) {
sock_reply(sock, EINVAL);
return;
}
err = copy_to_user(m->m_source, &udpopt, sizeof(udpopt),
(cp_grant_id_t) m->IO_GRANT, 0);
if (err != OK)
sock_reply(sock, err);
sock_reply(sock, OK);
}
static void udp_op_ioctl(struct socket * sock, message * m)
{
debug_udp_print("socket num %ld req %c %d %d",
get_sock_num(sock),
(m->REQUEST >> 8) & 0xff,
m->REQUEST & 0xff,
(m->REQUEST >> 16) & _IOCPARM_MASK);
switch (m->REQUEST) {
case NWIOSUDPOPT:
udp_set_opt(sock, m);
break;
case NWIOGUDPOPT:
udp_get_opt(sock, m);
break;
default:
sock_reply(sock, EBADIOCTL);
return;
}
}
struct sock_ops sock_udp_ops = {
.open = udp_op_open,
.close = udp_op_close,
.read = udp_op_read,
.write = udp_op_write,
.ioctl = udp_op_ioctl,
.select = generic_op_select,
.select_reply = generic_op_select_reply
};