/*! Perform system bind on the OS backing socket.
* \param ep Endpoint context
* \param fd Callers FD
* \param ip_addr_be32 Local address to which to bind
* \param port_be16 [in] requested port [out] assigned port
* \return 0 - success & [port_be16] updated
* CI_SOCKET_HANDOVER, Pass to OS, OS bound ok, (no error)
* CI_SOCKET_ERROR & errno set
*/
ci_inline int __ci_bind(ci_netif *ni, ci_sock_cmn *s,
ci_uint32 ip_addr_be32, ci_uint16* port_be16 )
{
int rc;
ci_uint16 user_port; /* Port number specified by user, not by OS.
* See bug 4015 for details */
union ci_sockaddr_u sa_u;
ci_assert(s->domain == AF_INET || s->domain == AF_INET6);
ci_assert( port_be16 );
user_port = *port_be16;
#if CI_CFG_FAKE_IPV6
ci_assert(s->domain == AF_INET || s->domain == AF_INET6);
if( s->domain == AF_INET )
ci_make_sockaddr(&sa_u.sin, s->domain, user_port, ip_addr_be32);
else
ci_make_sockaddr6(&sa_u.sin6, s->domain, user_port, ip_addr_be32);
#else
ci_assert(s->domain == AF_INET);
ci_make_sockaddr(&sa_u.sin, s->domain, user_port, ip_addr_be32);
#endif
#ifdef __ci_driver__
rc = efab_tcp_helper_bind_os_sock(netif2tcp_helper_resource(ni),
SC_SP(s),
&sa_u.sa, sizeof(sa_u), port_be16);
#else
rc = ci_tcp_helper_bind_os_sock(ni, SC_SP(s), &sa_u.sa,
sizeof(sa_u), port_be16);
#endif
/* bug1781: only do this if the earlier bind succeeded.
* check if we can handle this socket */
if( rc != 0 )
return rc;
if( user_port != 0 )
s->s_flags |= CI_SOCK_FLAG_PORT_BOUND;
if( ip_addr_be32 != INADDR_ANY )
s->s_flags |= CI_SOCK_FLAG_ADDR_BOUND;
s->s_flags &= ~CI_SOCK_FLAG_CONNECT_MUST_BIND;
#ifndef __ci_driver__
/* We do not call bind() to alien address from in-kernel code */
if( ip_addr_be32 != INADDR_ANY &&
!cicp_user_addr_is_local_efab(CICP_HANDLE(ni), &ip_addr_be32) )
s->s_flags |= CI_SOCK_FLAG_BOUND_ALIEN;
#endif
return rc;
}
static int __ci_udp_recvmsg_try_os(ci_netif *ni, ci_udp_state *us,
struct msghdr* msg, int flags, int* prc)
{
int rc;
rc = oo_os_sock_recvmsg(ni, SC_SP(&us->s), msg, flags | MSG_DONTWAIT);
if( rc >= 0 ) {
++us->stats.n_rx_os;
us->udpflags &= ~CI_UDPF_LAST_RECV_ON;
if( ! (flags & MSG_PEEK) )
us->udpflags &=~ CI_UDPF_PEEK_FROM_OS;
else
us->udpflags |= CI_UDPF_PEEK_FROM_OS;
}
else {
if( rc == -EAGAIN )
return 0;
ci_assert(-rc == errno);
rc = -1;
++us->stats.n_rx_os_error;
}
*prc = rc;
return 1;
}
void sig_handler_common_tt(int sig, void *sc_ptr)
{
struct sigcontext *sc = sc_ptr;
struct tt_regs save_regs, *r;
struct signal_info *info;
int save_errno = errno, is_user;
unprotect_kernel_mem();
r = &TASK_REGS(get_current())->tt;
save_regs = *r;
is_user = user_context(SC_SP(sc));
r->sc = sc;
if(sig != SIGUSR2)
r->syscall = -1;
change_sig(SIGUSR1, 1);
info = &sig_info[sig];
if(!info->is_irq) unblock_signals();
(*info->handler)(sig, (union uml_pt_regs *) r);
if(is_user){
interrupt_end();
block_signals();
change_sig(SIGUSR1, 0);
set_user_mode(NULL);
}
*r = save_regs;
errno = save_errno;
if(is_user) protect_kernel_mem();
}
int copy_thread_tt(int nr, unsigned long clone_flags, unsigned long sp,
unsigned long stack_top, struct task_struct * p,
struct pt_regs *regs)
{
int (*tramp)(void *);
int new_pid, err;
unsigned long stack;
if(current->thread.forking)
tramp = fork_tramp;
else {
tramp = new_thread_proc;
p->thread.request.u.thread = current->thread.request.u.thread;
}
err = os_pipe(p->thread.mode.tt.switch_pipe, 1, 1);
if(err < 0){
printk("copy_thread : pipe failed, err = %d\n", -err);
return(err);
}
stack = alloc_stack(0, 0);
if(stack == 0){
printk(KERN_ERR "copy_thread : failed to allocate "
"temporary stack\n");
return(-ENOMEM);
}
clone_flags &= CLONE_VM;
p->thread.temp_stack = stack;
new_pid = start_fork_tramp(task_stack_page(p), stack, clone_flags, tramp);
if(new_pid < 0){
printk(KERN_ERR "copy_thread : clone failed - errno = %d\n",
-new_pid);
return(new_pid);
}
if(current->thread.forking){
sc_to_sc(UPT_SC(&p->thread.regs.regs), UPT_SC(®s->regs));
SC_SET_SYSCALL_RETURN(UPT_SC(&p->thread.regs.regs), 0);
if(sp != 0)
SC_SP(UPT_SC(&p->thread.regs.regs)) = sp;
}
p->thread.mode.tt.extern_pid = new_pid;
current->thread.request.op = OP_FORK;
current->thread.request.u.fork.pid = new_pid;
os_usr1_process(os_getpid());
/* Enable the signal and then disable it to ensure that it is handled
* here, and nowhere else.
*/
change_sig(SIGUSR1, 1);
change_sig(SIGUSR1, 0);
err = 0;
return(err);
}
ci_fd_t ci_udp_ep_ctor(citp_socket* ep, ci_netif* netif, int domain, int type)
{
ci_udp_state* us;
ci_fd_t fd;
VERB( log(LPFIN "ctor( )" ) );
ci_assert(ep);
ci_assert(netif);
ci_netif_lock(netif);
us = ci_udp_get_state_buf(netif);
if (!us) {
ci_netif_unlock(netif);
LOG_E(ci_log("%s: [%d] out of socket buffers", __FUNCTION__,NI_ID(netif)));
return -ENOMEM;
}
/* It's required to set protocol before ci_tcp_helper_sock_attach()
* since it's used to determine if TCP or UDP file operations should be
* attached to the file descriptor in kernel. */
sock_protocol(&us->s) = IPPROTO_UDP;
/* NB: this attach will close the os_sock_fd */
fd = ci_tcp_helper_sock_attach(ci_netif_get_driver_handle(netif),
SC_SP(&us->s), domain, type);
if( fd < 0 ) {
if( fd == -EAFNOSUPPORT )
LOG_U(ci_log("%s: ci_tcp_helper_sock_attach (domain=%d, type=%d) "
"failed %d", __FUNCTION__, domain, type, fd));
else
LOG_E(ci_log("%s: ci_tcp_helper_sock_attach (domain=%d, type=%d) "
"failed %d", __FUNCTION__, domain, type, fd));
ci_netif_unlock(netif);
return fd;
}
ci_assert(~us->s.b.sb_aflags & CI_SB_AFLAG_ORPHAN);
us->s.rx_errno = 0;
us->s.tx_errno = 0;
us->s.so_error = 0;
us->s.cp.sock_cp_flags |= OO_SCP_UDP_WILD;
ep->s = &us->s;
ep->netif = netif;
CHECK_UEP(ep);
ci_netif_unlock(netif);
return fd;
}
void ci_sock_cmn_init(ci_netif* ni, ci_sock_cmn* s)
{
oo_p sp;
/* Poison. */
CI_DEBUG(memset(&s->b + 1, 0xf0, (char*) (s + 1) - (char*) (&s->b + 1)));
citp_waitable_reinit(ni, &s->b);
oo_sock_cplane_init(&s->cp);
s->local_peer = OO_SP_NULL;
s->s_flags = CI_SOCK_FLAG_CONNECT_MUST_BIND | CI_SOCK_FLAG_PMTU_DO;
s->s_aflags = 0u;
ci_assert_equal( 0, CI_IP_DFLT_TOS );
s->so_priority = 0;
/* SO_SNDBUF & SO_RCVBUF. See also ci_tcp_set_established_state() which
* may modify these values.
*/
memset(&s->so, 0, sizeof(s->so));
s->so.sndbuf = NI_OPTS(ni).tcp_sndbuf_def;
s->so.rcvbuf = NI_OPTS(ni).tcp_rcvbuf_def;
s->rx_bind2dev_ifindex = CI_IFID_BAD;
/* These don't really need to be initialised, as only significant when
* rx_bind2dev_ifindex != CI_IFID_BAD. But makes stackdump output
* cleaner this way...
*/
s->rx_bind2dev_base_ifindex = 0;
s->rx_bind2dev_vlan = 0;
s->cmsg_flags = 0u;
s->timestamping_flags = 0u;
s->os_sock_status = OO_OS_STATUS_TX;
ci_ip_queue_init(&s->timestamp_q);
s->timestamp_q_extract = OO_PP_NULL;
ci_sock_cmn_reinit(ni, s);
sp = oo_sockp_to_statep(ni, SC_SP(s));
OO_P_ADD(sp, CI_MEMBER_OFFSET(ci_sock_cmn, reap_link));
ci_ni_dllist_link_init(ni, &s->reap_link, sp, "reap");
ci_ni_dllist_self_link(ni, &s->reap_link);
}
static int __ci_udp_recvmsg_try_os(ci_netif *ni, ci_udp_state *us,
ci_msghdr* msg, int flags, int* prc)
{
int rc, total_bytes, i;
tcp_helper_endpoint_t *ep;
struct socket *sock;
struct file *os_sock;
struct msghdr kmsg;
total_bytes = 0;
for( i = 0; i < msg->msg_iovlen; ++i )
total_bytes += msg->msg_iov[i].iov_len;
rc = -EMSGSIZE;
if( total_bytes < 0 )
return -EINVAL;
ep = ci_netif_ep_get(ni, SC_SP(&us->s));
os_sock = ep->os_socket->file;
ci_assert(S_ISSOCK(os_sock->f_dentry->d_inode->i_mode));
sock = SOCKET_I(os_sock->f_dentry->d_inode);
oo_msg_iov_init(&kmsg, READ, msg->msg_iov, msg->msg_iovlen, total_bytes);
kmsg.msg_namelen = 0;
kmsg.msg_controllen = 0;
rc = sock_recvmsg(sock, &kmsg, total_bytes, flags | MSG_DONTWAIT);
/* Clear OS RX flag if we've got everything */
oo_os_sock_status_bit_clear(&us->s, OO_OS_STATUS_RX,
os_sock->f_op->poll(os_sock, NULL) & POLLIN);
if( rc >= 0 ) {
++us->stats.n_rx_os;
}
else {
if( rc == -EAGAIN )
return 0;
++us->stats.n_rx_os_error;
}
if( rc >= 0 ) {
us->udpflags &= ~CI_UDPF_LAST_RECV_ON;
if( ! (flags & MSG_PEEK) )
us->udpflags &=~ CI_UDPF_PEEK_FROM_OS;
else
us->udpflags |= CI_UDPF_PEEK_FROM_OS;
}
*prc = rc;
return 1;
}
void sig_handler_common_tt(int sig, void *sc_ptr)
{
struct sigcontext *sc = sc_ptr;
struct tt_regs save_regs, *r;
struct signal_info *info;
int save_errno = errno, is_user;
unprotect_kernel_mem();
/* This is done because to allow SIGSEGV to be delivered inside a SEGV
* handler. This can happen in copy_user, and if SEGV is disabled,
* the process will die.
*/
if(sig == SIGSEGV)
change_sig(SIGSEGV, 1);
/* This is done because to allow SIGSEGV to be delivered inside a SEGV
* handler. This can happen in copy_user, and if SEGV is disabled,
* the process will die.
*/
if(sig == SIGSEGV)
change_sig(SIGSEGV, 1);
r = &TASK_REGS(get_current())->tt;
save_regs = *r;
is_user = user_context(SC_SP(sc));
r->sc = sc;
if(sig != SIGUSR2)
r->syscall = -1;
info = &sig_info[sig];
if(!info->is_irq) unblock_signals();
(*info->handler)(sig, (union uml_pt_regs *) r);
if(is_user){
interrupt_end();
block_signals();
set_user_mode(NULL);
}
*r = save_regs;
errno = save_errno;
if(is_user) protect_kernel_mem();
}
int copy_sc_to_user_tt(struct sigcontext __user *to, struct _fpstate __user *fp,
struct sigcontext *from, int fpsize, unsigned long sp)
{
struct _fpstate __user *to_fp;
struct _fpstate *from_fp;
int err;
to_fp = (fp ? fp : (struct _fpstate __user *) (to + 1));
from_fp = from->fpstate;
err = copy_to_user(to, from, sizeof(*to));
/* The SP in the sigcontext is the updated one for the signal
* delivery. The sp passed in is the original, and this needs
* to be restored, so we stick it in separately.
*/
err |= copy_to_user(&SC_SP(to), &sp, sizeof(sp));
if(from_fp != NULL){
err |= copy_to_user(&to->fpstate, &to_fp, sizeof(to->fpstate));
err |= copy_to_user(to_fp, from_fp, fpsize);
}
return err;
}
void
sigtrap_handler(HANDLER_ARGS)
{
unsigned int trap;
os_context_t* os_context = (os_context_t *) context;
#if 0
fprintf(stderr, "x86sigtrap: %8x %x\n",
SC_PC(os_os_context), *(unsigned char *) (SC_PC(os_context) - 1));
fprintf(stderr, "sigtrap(%d %d %x)\n", signal, CODE(code), os_context);
#endif
if (single_stepping && (signal == SIGTRAP)) {
#if 0
fprintf(stderr, "* Single step trap %p\n", single_stepping);
#endif
#ifdef SC_EFLAGS
/* Disable single-stepping */
SC_EFLAGS(os_context) ^= 0x100;
#else
/* Un-install single step helper instructions. */
*(single_stepping - 3) = single_step_save1;
*(single_stepping - 2) = single_step_save2;
*(single_stepping - 1) = single_step_save3;
DPRINTF(0, (stderr, "Uninstalling helper instructions\n"));
#endif
/*
* Re-install the breakpoint if possible.
*/
if ((int) SC_PC(os_context) == (int) single_stepping + 1)
fprintf(stderr, "* Breakpoint not re-install\n");
else {
char *ptr = (char *) single_stepping;
ptr[0] = BREAKPOINT_INST; /* x86 INT3 */
ptr[1] = trap_Breakpoint;
}
single_stepping = NULL;
return;
}
/* This is just for info in case monitor wants to print an approx */
current_control_stack_pointer = (unsigned long *) SC_SP(os_context);
RESTORE_FPU(os_context);
/*
* On entry %eip points just after the INT3 byte and aims at the
* 'kind' value (eg trap_Cerror). For error-trap and Cerror-trap a
* number of bytes will follow, the first is the length of the byte
* arguments to follow.
*/
trap = *(unsigned char *) SC_PC(os_context);
switch (trap) {
case trap_PendingInterrupt:
DPRINTF(0, (stderr, "<trap Pending Interrupt.>\n"));
arch_skip_instruction(os_context);
interrupt_handle_pending(os_context);
break;
case trap_Halt:
{
FPU_STATE(fpu_state);
save_fpu_state(fpu_state);
fake_foreign_function_call(os_context);
lose("%%primitive halt called; the party is over.\n");
undo_fake_foreign_function_call(os_context);
restore_fpu_state(fpu_state);
arch_skip_instruction(os_context);
break;
}
case trap_Error:
case trap_Cerror:
DPRINTF(0, (stderr, "<trap Error %x>\n", CODE(code)));
interrupt_internal_error(signal, code, os_context, CODE(code) == trap_Cerror);
break;
case trap_Breakpoint:
#if 0
fprintf(stderr, "*C break\n");
#endif
SC_PC(os_context) -= 1;
handle_breakpoint(signal, CODE(code), os_context);
#if 0
fprintf(stderr, "*C break return\n");
#endif
break;
case trap_FunctionEndBreakpoint:
SC_PC(os_context) -= 1;
SC_PC(os_context) =
(int) handle_function_end_breakpoint(signal, CODE(code), os_context);
break;
#ifdef trap_DynamicSpaceOverflowWarning
case trap_DynamicSpaceOverflowWarning:
interrupt_handle_space_overflow(SymbolFunction
(DYNAMIC_SPACE_OVERFLOW_WARNING_HIT),
os_context);
break;
#endif
#ifdef trap_DynamicSpaceOverflowError
case trap_DynamicSpaceOverflowError:
interrupt_handle_space_overflow(SymbolFunction
(DYNAMIC_SPACE_OVERFLOW_ERROR_HIT),
os_context);
break;
#endif
default:
DPRINTF(0,
(stderr, "[C--trap default %d %d %p]\n", signal, CODE(code),
os_context));
interrupt_handle_now(signal, code, os_context);
break;
}
}
static int ci_udp_recvmsg_socklocked_slowpath(ci_udp_iomsg_args* a,
ci_msghdr* msg,
ci_iovec_ptr *piov, int flags)
{
int rc = 0;
ci_netif* ni = a->ni;
ci_udp_state* us = a->us;
if(CI_UNLIKELY( ni->state->rxq_low ))
ci_netif_rxq_low_on_recv(ni, &us->s,
1 /* assume at least one pkt freed */);
/* In the kernel recv() with flags is not called.
* only read(). So flags may only contain MSG_DONTWAIT */
#ifdef __KERNEL__
ci_assert_equal(flags, 0);
#endif
#ifndef __KERNEL__
if( flags & MSG_ERRQUEUE_CHK ) {
if( OO_PP_NOT_NULL(us->timestamp_q.extract) ) {
ci_ip_pkt_fmt* pkt;
struct timespec ts[3];
struct cmsg_state cmsg_state;
ci_udp_hdr* udp;
int paylen;
/* TODO is this necessary? - mirroring ci_udp_recvmsg_get() */
ci_rmb();
pkt = PKT_CHK_NNL(ni, us->timestamp_q.extract);
if( pkt->tx_hw_stamp.tv_sec == CI_PKT_TX_HW_STAMP_CONSUMED ) {
if( OO_PP_IS_NULL(pkt->tsq_next) )
goto errqueue_empty;
us->timestamp_q.extract = pkt->tsq_next;
pkt = PKT_CHK_NNL(ni, us->timestamp_q.extract);
ci_assert(pkt->tx_hw_stamp.tv_sec != CI_PKT_TX_HW_STAMP_CONSUMED);
}
udp = oo_ip_data(pkt);
paylen = CI_BSWAP_BE16(oo_ip_hdr(pkt)->ip_tot_len_be16) -
sizeof(ci_ip4_hdr) - sizeof(udp);
msg->msg_flags = 0;
cmsg_state.msg = msg;
cmsg_state.cm = msg->msg_control;
cmsg_state.cmsg_bytes_used = 0;
ci_iovec_ptr_init_nz(piov, msg->msg_iov, msg->msg_iovlen);
memset(ts, 0, sizeof(ts));
if( us->s.timestamping_flags & ONLOAD_SOF_TIMESTAMPING_RAW_HARDWARE ) {
ts[2].tv_sec = pkt->tx_hw_stamp.tv_sec;
ts[2].tv_nsec = pkt->tx_hw_stamp.tv_nsec;
}
if( (us->s.timestamping_flags & ONLOAD_SOF_TIMESTAMPING_SYS_HARDWARE) &&
(pkt->tx_hw_stamp.tv_nsec & CI_IP_PKT_HW_STAMP_FLAG_IN_SYNC) ) {
ts[1].tv_sec = pkt->tx_hw_stamp.tv_sec;
ts[1].tv_nsec = pkt->tx_hw_stamp.tv_nsec;
}
ci_put_cmsg(&cmsg_state, SOL_SOCKET, ONLOAD_SCM_TIMESTAMPING,
sizeof(ts), &ts);
oo_offbuf_set_start(&pkt->buf, udp + 1);
oo_offbuf_set_len(&pkt->buf, paylen);
rc = oo_copy_pkt_to_iovec_no_adv(ni, pkt, piov, paylen);
/* Mark this packet/timestamp as consumed */
pkt->tx_hw_stamp.tv_sec = CI_PKT_TX_HW_STAMP_CONSUMED;
ci_ip_cmsg_finish(&cmsg_state);
msg->msg_flags |= MSG_ERRQUEUE_CHK;
return rc;
}
errqueue_empty:
/* ICMP is handled via OS, so get OS error */
rc = oo_os_sock_recvmsg(ni, SC_SP(&us->s), msg, flags);
if( rc < 0 ) {
ci_assert(-rc == errno);
return -1;
}
else
return rc;
}
#endif
if( (rc = ci_get_so_error(&us->s)) != 0 ) {
CI_SET_ERROR(rc, rc);
return rc;
}
if( msg->msg_iovlen > 0 && msg->msg_iov == NULL ) {
CI_SET_ERROR(rc, EFAULT);
return rc;
}
#if MSG_OOB_CHK
if( flags & MSG_OOB_CHK ) {
CI_SET_ERROR(rc, EOPNOTSUPP);
return rc;
}
#endif
#if CI_CFG_POSIX_RECV
if( ! udp_lport_be16(us)) {
LOG_UV(log("%s: -1 (ENOTCONN)", __FUNCTION__));
CI_SET_ERROR(rc, ENOTCONN);
return rc;
}
#endif
if( msg->msg_iovlen == 0 ) {
/* We have a difference in behaviour from the Linux stack here. When
** msg_iovlen is 0 Linux 2.4.21-15.EL does not set MSG_TRUNC when a
** datagram has non-zero length. We do. */
CI_IOVEC_LEN(&piov->io) = piov->iovlen = 0;
return IOVLEN_WORKAROUND_RC_VALUE;
}
return 0;
}
s->rx_bind2dev_ifindex = CI_IFID_BAD;
/* These don't really need to be initialised, as only significant when
* rx_bind2dev_ifindex != CI_IFID_BAD. But makes stackdump output
* cleaner this way...
*/
s->rx_bind2dev_base_ifindex = 0;
s->rx_bind2dev_vlan = 0;
s->cmsg_flags = 0u;
s->timestamping_flags = 0u;
s->os_sock_status = OO_OS_STATUS_TX;
ci_sock_cmn_reinit(ni, s);
sp = oo_sockp_to_statep(ni, SC_SP(s));
OO_P_ADD(sp, CI_MEMBER_OFFSET(ci_sock_cmn, reap_link));
ci_ni_dllist_link_init(ni, &s->reap_link, sp, "reap");
ci_ni_dllist_self_link(ni, &s->reap_link);
}
void ci_sock_cmn_dump(ci_netif* ni, ci_sock_cmn* s, const char* pf,
oo_dump_log_fn_t logger, void* log_arg)
{
logger(log_arg, "%s uid=%d"CI_DEBUG(" pid=%d")
" s_flags: "CI_SOCK_FLAGS_FMT, pf,
(int) s->uid CI_DEBUG_ARG((int)s->pid),
CI_SOCK_FLAGS_PRI_ARG(s));