
Currently, the BSD socket API is implemented in libc, translating the API calls to character driver operations underneath. This approach has several issues: - it is inefficient, as most character driver operations are specific to the socket type, thus requiring that each operation start by bruteforcing the socket protocol family and type of the given file descriptor using several system calls; - it requires that libc itself be changed every time system support for a new protocol is added; - various parts of the libc implementations violate the asynchronous signal safety POSIX requirements. In order to resolve all these issues at once, the plan is to turn the BSD socket calls into system calls, thus making the BSD socket API the "native" ABI, removing the complexity from libc and instead letting VFS deal with the socket calls. The overall change is going to break all networking functionality. In order to smoothen the transition, this patch introduces the fifteen new BSD socket system calls, and makes libc try these first before falling back on the old behavior. For now, the VFS implementations of the new calls fail such that libc will always use the fallback cases. Later on, when we introduce the actual implementation of the native BSD socket calls, all statically linked programs will automatically use the new ABI, thus limiting actual application breakage. In other words: by itself, this patch does nothing, except add a bit of transitional overhead that will disappear in the future. The largest part of the patch is concerned with adding full support for the new BSD socket system calls to trace(1) - this early addition has the advantage of making system call tracing output of several socket calls much more readable already. Both the system call interfaces and the trace(1) support have already been tested using code that will be committed later on. Change-Id: I3460812be50c78be662d857f9d3d6840f3ca917f
202 lines
4.4 KiB
C
202 lines
4.4 KiB
C
#include <sys/cdefs.h>
|
|
#include "namespace.h"
|
|
#include <lib.h>
|
|
|
|
#include <errno.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <sys/ioctl.h>
|
|
#include <sys/ioc_net.h>
|
|
#include <sys/socket.h>
|
|
#include <sys/types.h>
|
|
#include <sys/un.h>
|
|
|
|
#define DEBUG 0
|
|
|
|
static ssize_t _uds_sendmsg_conn(int sock, const struct msghdr *msg,
|
|
int flags);
|
|
static ssize_t _uds_sendmsg_dgram(int sock, const struct msghdr *msg,
|
|
int flags);
|
|
|
|
/*
|
|
* Send a message on a socket using a message structure.
|
|
*/
|
|
static ssize_t
|
|
__sendmsg(int fd, const struct msghdr * msg, int flags)
|
|
{
|
|
struct iovec iov;
|
|
const struct msghdr *msgp;
|
|
struct msghdr msg2;
|
|
char *ptr;
|
|
message m;
|
|
ssize_t r;
|
|
|
|
/*
|
|
* Currently, MINIX3 does not support vector I/O operations. Like in
|
|
* the readv and writev implementations, we coalesce the data vector
|
|
* into a single buffer used for I/O. For future ABI compatibility, we
|
|
* then supply this buffer as a single vector element. This involves
|
|
* supplying a modified copy of the message header, as well as extra
|
|
* pre-checks. Once true vector I/O support has been added, the checks
|
|
* and vector I/O coalescing can be removed from here, leaving just the
|
|
* system call. Nothing will change at the system call ABI level.
|
|
*/
|
|
if (msg == NULL || (msg->msg_iovlen > 1 && msg->msg_iov == NULL)) {
|
|
errno = EFAULT;
|
|
return -1;
|
|
}
|
|
|
|
if (msg->msg_iovlen < 0 || msg->msg_iovlen > IOV_MAX) {
|
|
errno = EMSGSIZE; /* different from readv/writev */
|
|
return -1;
|
|
}
|
|
|
|
if (msg->msg_iovlen > 1) {
|
|
if ((r = _vectorio_setup(msg->msg_iov, msg->msg_iovlen, &ptr,
|
|
_VECTORIO_WRITE)) < 0)
|
|
return -1;
|
|
|
|
iov.iov_base = ptr;
|
|
iov.iov_len = r;
|
|
|
|
memcpy(&msg2, msg, sizeof(msg2));
|
|
msg2.msg_iov = &iov;
|
|
msg2.msg_iovlen = 1;
|
|
msgp = &msg2;
|
|
} else
|
|
msgp = msg;
|
|
|
|
memset(&m, 0, sizeof(m));
|
|
m.m_lc_vfs_sockmsg.fd = fd;
|
|
m.m_lc_vfs_sockmsg.msgbuf = (vir_bytes)msgp;
|
|
m.m_lc_vfs_sockmsg.flags = flags;
|
|
|
|
r = _syscall(VFS_PROC_NR, VFS_SENDMSG, &m);
|
|
|
|
/* If we coalesced the vector, clean up. */
|
|
if (msgp != msg) {
|
|
_vectorio_cleanup(msg->msg_iov, msg->msg_iovlen, ptr, r,
|
|
_VECTORIO_WRITE);
|
|
}
|
|
|
|
return r;
|
|
}
|
|
|
|
ssize_t sendmsg(int sock, const struct msghdr *msg, int flags)
|
|
{
|
|
int r;
|
|
int uds_sotype;
|
|
|
|
r = __sendmsg(sock, msg, flags);
|
|
if (r != -1 || errno != ENOTSOCK)
|
|
return r;
|
|
|
|
if (msg == NULL) {
|
|
errno= EFAULT;
|
|
return -1;
|
|
}
|
|
|
|
r= ioctl(sock, NWIOGUDSSOTYPE, &uds_sotype);
|
|
if (r != -1 || errno != ENOTTY) {
|
|
if (r == -1) {
|
|
return r;
|
|
}
|
|
|
|
if (uds_sotype == SOCK_DGRAM) {
|
|
return _uds_sendmsg_dgram(sock, msg, flags);
|
|
} else {
|
|
return _uds_sendmsg_conn(sock, msg, flags);
|
|
}
|
|
|
|
}
|
|
|
|
errno = ENOTSOCK;
|
|
return -1;
|
|
}
|
|
|
|
static ssize_t _uds_sendmsg_conn(int sock, const struct msghdr *msg,
|
|
int flags)
|
|
{
|
|
struct msg_control msg_ctrl;
|
|
int r;
|
|
|
|
if (flags != 0) {
|
|
#if DEBUG
|
|
fprintf(stderr, "sendmsg(uds): flags not implemented\n");
|
|
#endif
|
|
errno= ENOSYS;
|
|
return -1;
|
|
|
|
}
|
|
|
|
/* grab the control data */
|
|
memset(&msg_ctrl, '\0', sizeof(struct msg_control));
|
|
if (msg->msg_controllen > MSG_CONTROL_MAX) {
|
|
errno = ENOMEM;
|
|
return -1;
|
|
} else if (msg->msg_controllen > 0) {
|
|
memcpy(&msg_ctrl.msg_control, msg->msg_control,
|
|
msg->msg_controllen);
|
|
}
|
|
msg_ctrl.msg_controllen = msg->msg_controllen;
|
|
|
|
/* send the control data to PFS */
|
|
r= ioctl(sock, NWIOSUDSCTRL, (void *) &msg_ctrl);
|
|
if (r == -1) {
|
|
return r;
|
|
}
|
|
|
|
/* Silently ignore destination, if given. */
|
|
|
|
return writev(sock, msg->msg_iov, msg->msg_iovlen);
|
|
}
|
|
|
|
static ssize_t _uds_sendmsg_dgram(int sock, const struct msghdr *msg,
|
|
int flags)
|
|
{
|
|
struct msg_control msg_ctrl;
|
|
struct sockaddr_un *dest_addr;
|
|
int r;
|
|
|
|
if (flags != 0) {
|
|
#if DEBUG
|
|
fprintf(stderr, "sendmsg(uds): flags not implemented\n");
|
|
#endif
|
|
errno= ENOSYS;
|
|
return -1;
|
|
|
|
}
|
|
|
|
dest_addr = msg->msg_name;
|
|
if (dest_addr == NULL) {
|
|
errno= EFAULT;
|
|
return -1;
|
|
}
|
|
|
|
/* set the target address */
|
|
r= ioctl(sock, NWIOSUDSTADDR, (void *) dest_addr);
|
|
if (r == -1) {
|
|
return r;
|
|
}
|
|
|
|
/* grab the control data */
|
|
memset(&msg_ctrl, '\0', sizeof(struct msg_control));
|
|
if (msg->msg_controllen > MSG_CONTROL_MAX) {
|
|
errno = ENOMEM;
|
|
return -1;
|
|
} else if (msg->msg_controllen > 0) {
|
|
memcpy(&msg_ctrl.msg_control, msg->msg_control,
|
|
msg->msg_controllen);
|
|
}
|
|
msg_ctrl.msg_controllen = msg->msg_controllen;
|
|
|
|
/* send the control data to PFS */
|
|
r= ioctl(sock, NWIOSUDSCTRL, (void *) &msg_ctrl);
|
|
if (r == -1) {
|
|
return r;
|
|
}
|
|
|
|
/* do the send */
|
|
return writev(sock, msg->msg_iov, msg->msg_iovlen);
|
|
}
|