rt-thread/components/lwp/lwp_syscall.c

4820 lines
108 KiB
C

/*
* Copyright (c) 2006-2023, RT-Thread Development Team
*
* SPDX-License-Identifier: Apache-2.0
*
* Change Logs:
* Date Author Notes
* 2018-06-10 Bernard first version
* 2021-02-03 lizhirui add limit condition for network syscall and add 64-bit arch support
* 2021-02-06 lizhirui fix some bugs
* 2021-02-12 lizhirui add 64-bit support for sys_brk
* 2021-02-20 lizhirui fix some warnings
* 2023-03-13 WangXiaoyao Format & fix syscall return value
*/
#define _GNU_SOURCE
/* RT-Thread System call */
#include <rtthread.h>
#include <rthw.h>
#include <board.h>
#include <mm_aspace.h>
#include <string.h>
#include <stdint.h>
#define DBG_TAG "SYSCALL"
#define DBG_LVL DBG_INFO
#include <rtdbg.h>
#include "syscall_generic.h"
#include <lwp.h>
#ifdef ARCH_MM_MMU
#include <lwp_user_mm.h>
#include <lwp_arch.h>
#endif
#include <fcntl.h>
#ifdef RT_USING_DFS
#include <poll.h>
#include <sys/select.h>
#include <dfs_file.h>
#include <unistd.h>
#include <stdio.h> /* rename() */
#include <sys/stat.h>
#include <sys/statfs.h> /* statfs() */
#endif
#include "mqueue.h"
#ifdef RT_USING_SAL
#include <netdev_ipaddr.h>
#include <netdev.h>
#include <sal_netdb.h>
#include <sal_socket.h>
#include <sys/socket.h>
#endif /* RT_USING_SAL */
#if (defined(RT_USING_SAL) && defined(SAL_USING_POSIX))
#include <sys/socket.h>
#define SYSCALL_NET(f) f
#else
#define SYSCALL_NET(f) SYSCALL_SIGN(sys_notimpl)
#endif /* (defined(RT_USING_SAL) && defined(SAL_USING_POSIX)) */
#if defined(RT_USING_DFS) && defined(ARCH_MM_MMU)
#define SYSCALL_USPACE(f) f
#else
#define SYSCALL_USPACE(f) SYSCALL_SIGN(sys_notimpl)
#endif /* defined(RT_USING_DFS) && defined(ARCH_MM_MMU) */
#include <tty.h>
#include "lwp_ipc_internal.h"
#include <sched.h>
#ifndef GRND_NONBLOCK
#define GRND_NONBLOCK 0x0001
#endif /* GRND_NONBLOCK */
#ifndef GRND_RANDOM
#define GRND_RANDOM 0x0002
#endif /*GRND_RANDOM */
#ifndef RT_USING_POSIX_TIMER
#error "No definition RT_USING_POSIX_TIMER"
#endif /* RT_USING_POSIX_TIMER */
#ifndef RT_USING_POSIX_CLOCK
#error "No definition RT_USING_POSIX_CLOCK"
#endif /* RT_USING_POSIX_CLOCK */
struct musl_sockaddr
{
uint16_t sa_family;
char sa_data[14];
};
void lwp_cleanup(struct rt_thread *tid);
#ifdef ARCH_MM_MMU
#define ALLOC_KERNEL_STACK_SIZE 5120
static void *kmem_get(size_t size)
{
return rt_malloc(size);
}
static void kmem_put(void *kptr)
{
rt_free(kptr);
}
#else /* ARCH_MM_MMU */
#define ALLOC_KERNEL_STACK_SIZE 1536
#define ALLOC_KERNEL_STACK_SIZE_MIN 1024
#define ALLOC_KERNEL_STACK_SIZE_MAX 4096
extern void set_user_context(void *stack);
#endif /* ARCH_MM_MMU */
#ifdef RT_USING_SAL
/* The same socket option is defined differently in the user interfaces and the
* implementation. The options should be converted in the kernel. */
#include "lwp_sys_socket.h"
static void convert_sockopt(int *level, int *optname)
{
if (*level == INTF_SOL_SOCKET)
{
*level = IMPL_SOL_SOCKET;
switch (*optname)
{
case INTF_SO_REUSEADDR:
*optname = IMPL_SO_REUSEADDR;
break;
case INTF_SO_KEEPALIVE:
*optname = IMPL_SO_KEEPALIVE;
break;
case INTF_SO_BROADCAST:
*optname = IMPL_SO_BROADCAST;
break;
case INTF_SO_ACCEPTCONN:
*optname = IMPL_SO_ACCEPTCONN;
break;
case INTF_SO_DONTROUTE:
*optname = IMPL_SO_DONTROUTE;
break;
case INTF_SO_LINGER:
*optname = IMPL_SO_LINGER;
break;
case INTF_SO_OOBINLINE:
*optname = IMPL_SO_OOBINLINE;
break;
case INTF_SO_REUSEPORT:
*optname = IMPL_SO_REUSEPORT;
break;
case INTF_SO_SNDBUF:
*optname = IMPL_SO_SNDBUF;
break;
case INTF_SO_RCVBUF:
*optname = IMPL_SO_RCVBUF;
break;
case INTF_SO_SNDLOWAT:
*optname = IMPL_SO_SNDLOWAT;
break;
case INTF_SO_RCVLOWAT:
*optname = IMPL_SO_RCVLOWAT;
break;
case INTF_SO_SNDTIMEO:
*optname = IMPL_SO_SNDTIMEO;
break;
case INTF_SO_RCVTIMEO:
*optname = IMPL_SO_RCVTIMEO;
break;
case INTF_SO_ERROR:
*optname = IMPL_SO_ERROR;
break;
case INTF_SO_TYPE:
*optname = IMPL_SO_TYPE;
break;
case INTF_SO_NO_CHECK:
*optname = IMPL_SO_NO_CHECK;
break;
/*
* SO_DONTLINGER (*level = ((int)(~SO_LINGER))),
* SO_USELOOPBACK (*level = 0x0040) and
* SO_CONTIMEO (*level = 0x1009) are not supported for now.
*/
default:
*optname = 0;
break;
}
return;
}
if (*level == INTF_IPPROTO_IP)
{
*level = IMPL_IPPROTO_IP;
switch (*optname)
{
case INTF_IP_TTL:
*optname = IMPL_IP_TTL;
break;
case INTF_IP_TOS:
*optname = IMPL_IP_TOS;
break;
case INTF_IP_MULTICAST_TTL:
*optname = IMPL_IP_MULTICAST_TTL;
break;
case INTF_IP_MULTICAST_IF:
*optname = IMPL_IP_MULTICAST_IF;
break;
case INTF_IP_MULTICAST_LOOP:
*optname = IMPL_IP_MULTICAST_LOOP;
break;
case INTF_IP_ADD_MEMBERSHIP:
*optname = IMPL_IP_ADD_MEMBERSHIP;
break;
case INTF_IP_DROP_MEMBERSHIP:
*optname = IMPL_IP_DROP_MEMBERSHIP;
break;
default:
break;
}
}
if (*level == INTF_IPPROTO_TCP)
{
*level = IMPL_IPPROTO_TCP;
switch (*optname)
{
case INTF_TCP_NODELAY:
*optname = IMPL_TCP_NODELAY;
break;
case INTF_TCP_KEEPALIVE:
*optname = IMPL_TCP_KEEPALIVE;
break;
case INTF_TCP_KEEPIDLE:
*optname = IMPL_TCP_KEEPIDLE;
break;
case INTF_TCP_KEEPINTVL:
*optname = IMPL_TCP_KEEPINTVL;
break;
case INTF_TCP_KEEPCNT:
*optname = IMPL_TCP_KEEPCNT;
break;
default:
break;
}
return;
}
if (*level == INTF_IPPROTO_IPV6)
{
*level = IMPL_IPPROTO_IPV6;
switch (*optname)
{
case INTF_IPV6_V6ONLY:
*optname = IMPL_IPV6_V6ONLY;
break;
default:
break;
}
return;
}
}
#endif /* RT_USING_SAL */
#if defined(RT_USING_LWIP) || defined(SAL_USING_UNET)
static void sockaddr_tolwip(const struct musl_sockaddr *std, struct sockaddr *lwip)
{
if (std && lwip)
{
lwip->sa_len = sizeof(*lwip);
lwip->sa_family = (sa_family_t) std->sa_family;
memcpy(lwip->sa_data, std->sa_data, sizeof(lwip->sa_data));
}
}
static void sockaddr_tomusl(const struct sockaddr *lwip, struct musl_sockaddr *std)
{
if (std && lwip)
{
std->sa_family = (uint16_t) lwip->sa_family;
memcpy(std->sa_data, lwip->sa_data, sizeof(std->sa_data));
}
}
#endif
static void _crt_thread_entry(void *parameter)
{
rt_thread_t tid;
rt_size_t user_stack;
tid = rt_thread_self();
user_stack = (rt_size_t)tid->user_stack + tid->user_stack_size;
user_stack &= ~7; //align 8
#ifdef ARCH_MM_MMU
arch_crt_start_umode(parameter, tid->user_entry, (void *)user_stack, tid->stack_addr + tid->stack_size);
#else
set_user_context((void*)user_stack);
arch_start_umode(parameter, tid->user_entry, ((struct rt_lwp *)tid->lwp)->data_entry, (void*)user_stack);
#endif /* ARCH_MM_MMU */
}
/* thread/process */
void sys_exit(int value)
{
rt_base_t level;
rt_thread_t tid, main_thread;
struct rt_lwp *lwp;
LOG_D("thread/process exit.");
tid = rt_thread_self();
lwp = (struct rt_lwp *)tid->lwp;
level = rt_hw_interrupt_disable();
#ifdef ARCH_MM_MMU
if (tid->clear_child_tid)
{
int t = 0;
int *clear_child_tid = tid->clear_child_tid;
tid->clear_child_tid = RT_NULL;
lwp_put_to_user(clear_child_tid, &t, sizeof t);
sys_futex(clear_child_tid, FUTEX_WAKE, 1, RT_NULL, RT_NULL, 0);
}
main_thread = rt_list_entry(lwp->t_grp.prev, struct rt_thread, sibling);
if (main_thread == tid)
{
lwp_terminate(lwp);
lwp_wait_subthread_exit();
lwp->lwp_ret = value;
}
#else
main_thread = rt_list_entry(lwp->t_grp.prev, struct rt_thread, sibling);
if (main_thread == tid)
{
rt_thread_t sub_thread;
rt_list_t *list;
lwp_terminate(lwp);
/* delete all subthread */
while ((list = tid->sibling.prev) != &lwp->t_grp)
{
sub_thread = rt_list_entry(list, struct rt_thread, sibling);
rt_list_remove(&sub_thread->sibling);
rt_thread_delete(sub_thread);
}
lwp->lwp_ret = value;
}
#endif /* ARCH_MM_MMU */
rt_thread_delete(tid);
rt_schedule();
rt_hw_interrupt_enable(level);
return;
}
/* exit group */
void sys_exit_group(int status)
{
return;
}
/* syscall: "read" ret: "ssize_t" args: "int" "void *" "size_t" */
ssize_t sys_read(int fd, void *buf, size_t nbyte)
{
#ifdef ARCH_MM_MMU
void *kmem = RT_NULL;
ssize_t ret = -1;
if (!nbyte)
{
return -EINVAL;
}
if (!lwp_user_accessable((void *)buf, nbyte))
{
return -EFAULT;
}
kmem = kmem_get(nbyte);
if (!kmem)
{
return -ENOMEM;
}
ret = read(fd, kmem, nbyte);
if (ret > 0)
{
lwp_put_to_user(buf, kmem, ret);
}
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kmem);
return ret;
#else
if (!lwp_user_accessable((void *)buf, nbyte))
{
return -EFAULT;
}
ssize_t ret = read(fd, buf, nbyte);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
/* syscall: "write" ret: "ssize_t" args: "int" "const void *" "size_t" */
ssize_t sys_write(int fd, const void *buf, size_t nbyte)
{
#ifdef ARCH_MM_MMU
void *kmem = RT_NULL;
ssize_t ret = -1;
if (!nbyte)
{
return -EINVAL;
}
if (!lwp_user_accessable((void *)buf, nbyte))
{
return -EFAULT;
}
kmem = kmem_get(nbyte);
if (!kmem)
{
return -ENOMEM;
}
lwp_get_from_user(kmem, (void *)buf, nbyte);
ret = write(fd, kmem, nbyte);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kmem);
return ret;
#else
if (!lwp_user_accessable((void *)buf, nbyte))
{
return -EFAULT;
}
ssize_t ret = write(fd, buf, nbyte);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
/* syscall: "lseek" ret: "off_t" args: "int" "off_t" "int" */
off_t sys_lseek(int fd, off_t offset, int whence)
{
off_t ret = lseek(fd, offset, whence);
return (ret < 0 ? GET_ERRNO() : ret);
}
/* syscall: "open" ret: "int" args: "const char *" "int" "..." */
sysret_t sys_open(const char *name, int flag, ...)
{
#ifdef ARCH_MM_MMU
int ret = -1;
rt_size_t len = 0;
char *kname = RT_NULL;
if (!lwp_user_accessable((void *)name, 1))
{
return -EFAULT;
}
len = rt_strlen(name);
if (!len)
{
return -EINVAL;
}
kname = (char *)kmem_get(len + 1);
if (!kname)
{
return -ENOMEM;
}
lwp_get_from_user(kname, (void *)name, len + 1);
ret = open(kname, flag, 0);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kname);
return ret;
#else
if (!lwp_user_accessable((void *)name, 1))
{
return -EFAULT;
}
int ret = open(name, flag, 0);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
/* syscall: "close" ret: "int" args: "int" */
sysret_t sys_close(int fd)
{
int ret = close(fd);
return (ret < 0 ? GET_ERRNO() : ret);
}
/* syscall: "ioctl" ret: "int" args: "int" "u_long" "..." */
sysret_t sys_ioctl(int fd, unsigned long cmd, void* data)
{
int ret = ioctl(fd, cmd, data);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_fstat(int file, struct stat *buf)
{
#ifdef ARCH_MM_MMU
int ret = -1;
struct stat statbuff = {0};
if (!lwp_user_accessable((void *)buf, sizeof(struct stat)))
{
return -EFAULT;
}
else
{
ret = fstat(file, &statbuff);
if (ret == 0)
{
lwp_put_to_user(buf, &statbuff, sizeof statbuff);
}
else
{
ret = GET_ERRNO();
}
return ret;
}
#else
if (!lwp_user_accessable((void *)buf, sizeof(struct stat)))
{
return -EFAULT;
}
int ret = fstat(file, buf);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
/* DFS and lwip definitions */
#define IMPL_POLLIN (0x01)
#define IMPL_POLLOUT (0x02)
#define IMPL_POLLERR (0x04)
#define IMPL_POLLHUP (0x08)
#define IMPL_POLLNVAL (0x10)
/* musl definitions */
#define INTF_POLLIN 0x001
#define INTF_POLLPRI 0x002
#define INTF_POLLOUT 0x004
#define INTF_POLLERR 0x008
#define INTF_POLLHUP 0x010
#define INTF_POLLNVAL 0x020
#define INTF_POLLRDNORM 0x040
#define INTF_POLLRDBAND 0x080
#define INTF_POLLWRNORM 0x100
#define INTF_POLLWRBAND 0x200
#define INTF_POLLMSG 0x400
#define INTF_POLLRDHUP 0x2000
#define INTF_POLLIN_MASK (INTF_POLLIN | INTF_POLLRDNORM | INTF_POLLRDBAND | INTF_POLLPRI)
#define INTF_POLLOUT_MASK (INTF_POLLOUT | INTF_POLLWRNORM | INTF_POLLWRBAND)
static void musl2dfs_events(short *events)
{
short origin_e = *events;
short result_e = 0;
if (origin_e & INTF_POLLIN_MASK)
{
result_e |= IMPL_POLLIN;
}
if (origin_e & INTF_POLLOUT_MASK)
{
result_e |= IMPL_POLLOUT;
}
if (origin_e & INTF_POLLERR)
{
result_e |= IMPL_POLLERR;
}
if (origin_e & INTF_POLLHUP)
{
result_e |= IMPL_POLLHUP;
}
if (origin_e & INTF_POLLNVAL)
{
result_e |= IMPL_POLLNVAL;
}
*events = result_e;
}
static void dfs2musl_events(short *events)
{
short origin_e = *events;
short result_e = 0;
if (origin_e & IMPL_POLLIN)
{
result_e |= INTF_POLLIN_MASK;
}
if (origin_e & IMPL_POLLOUT)
{
result_e |= INTF_POLLOUT_MASK;
}
if (origin_e & IMPL_POLLERR)
{
result_e |= INTF_POLLERR;
}
if (origin_e & IMPL_POLLHUP)
{
result_e |= INTF_POLLHUP;
}
if (origin_e & IMPL_POLLNVAL)
{
result_e |= INTF_POLLNVAL;
}
*events = result_e;
}
sysret_t sys_poll(struct pollfd *fds, nfds_t nfds, int timeout)
{
int ret = -1;
int i = 0;
#ifdef ARCH_MM_MMU
struct pollfd *kfds = RT_NULL;
if (!lwp_user_accessable((void *)fds, nfds * sizeof *fds))
{
return -EFAULT;
}
kfds = (struct pollfd *)kmem_get(nfds * sizeof *kfds);
if (!kfds)
{
return -ENOMEM;
}
lwp_get_from_user(kfds, fds, nfds * sizeof *kfds);
for (i = 0; i < nfds; i++)
{
musl2dfs_events(&kfds[i].events);
}
ret = poll(kfds, nfds, timeout);
if (ret > 0)
{
for (i = 0; i < nfds; i++)
{
dfs2musl_events(&kfds->revents);
}
lwp_put_to_user(fds, kfds, nfds * sizeof *kfds);
}
kmem_put(kfds);
return ret;
#else
#ifdef RT_USING_MUSL
for (i = 0; i < nfds; i++)
{
musl2dfs_events(&fds->events);
}
#endif /* RT_USING_MUSL */
if (!lwp_user_accessable((void *)fds, nfds * sizeof *fds))
{
return -EFAULT;
}
ret = poll(fds, nfds, timeout);
#ifdef RT_USING_MUSL
if (ret > 0)
{
for (i = 0; i < nfds; i++)
{
dfs2musl_events(&fds->revents);
}
}
#endif /* RT_USING_MUSL */
return ret;
#endif /* ARCH_MM_MMU */
}
sysret_t sys_select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout)
{
#ifdef ARCH_MM_MMU
int ret = -1;
fd_set *kreadfds = RT_NULL, *kwritefds = RT_NULL, *kexceptfds = RT_NULL;
if (readfds)
{
if (!lwp_user_accessable((void *)readfds, sizeof *readfds))
{
SET_ERRNO(EFAULT);
goto quit;
}
kreadfds = (fd_set *)kmem_get(sizeof *kreadfds);
if (!kreadfds)
{
SET_ERRNO(ENOMEM);
goto quit;
}
lwp_get_from_user(kreadfds, readfds, sizeof *kreadfds);
}
if (writefds)
{
if (!lwp_user_accessable((void *)writefds, sizeof *writefds))
{
SET_ERRNO(EFAULT);
goto quit;
}
kwritefds = (fd_set *)kmem_get(sizeof *kwritefds);
if (!kwritefds)
{
SET_ERRNO(ENOMEM);
goto quit;
}
lwp_get_from_user(kwritefds, writefds, sizeof *kwritefds);
}
if (exceptfds)
{
if (!lwp_user_accessable((void *)exceptfds, sizeof *exceptfds))
{
SET_ERRNO(EFAULT);
goto quit;
}
kexceptfds = (fd_set *)kmem_get(sizeof *kexceptfds);
if (!kexceptfds)
{
SET_ERRNO(EINVAL);
goto quit;
}
lwp_get_from_user(kexceptfds, exceptfds, sizeof *kexceptfds);
}
ret = select(nfds, kreadfds, kwritefds, kexceptfds, timeout);
if (kreadfds)
{
lwp_put_to_user(readfds, kreadfds, sizeof *kreadfds);
}
if (kwritefds)
{
lwp_put_to_user(writefds, kwritefds, sizeof *kwritefds);
}
if (kexceptfds)
{
lwp_put_to_user(exceptfds, kexceptfds, sizeof *kexceptfds);
}
quit:
if (ret < 0)
{
ret = GET_ERRNO();
}
if (kreadfds)
{
kmem_put(kreadfds);
}
if (kwritefds)
{
kmem_put(kwritefds);
}
if (kexceptfds)
{
kmem_put(kexceptfds);
}
return ret;
#else
int ret;
if (!lwp_user_accessable((void *)readfds, sizeof *readfds))
{
return -EFAULT;
}
if (!lwp_user_accessable((void *)writefds, sizeof *writefds))
{
return -EFAULT;
}
if (!lwp_user_accessable((void *)exceptfds, sizeof *exceptfds))
{
return -EFAULT;
}
ret = select(nfds, readfds, writefds, exceptfds, timeout);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_unlink(const char *pathname)
{
#ifdef ARCH_MM_MMU
int ret = -1;
rt_size_t len = 0;
char *kname = RT_NULL;
int a_err = 0;
lwp_user_strlen(pathname, &a_err);
if (a_err)
{
return -EFAULT;
}
len = rt_strlen(pathname);
if (!len)
{
return -EINVAL;
}
kname = (char *)kmem_get(len + 1);
if (!kname)
{
return -ENOMEM;
}
lwp_get_from_user(kname, (void *)pathname, len + 1);
ret = unlink(kname);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kname);
return ret;
#else
int ret = 0;
ret = unlink(pathname);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
/* syscall: "nanosleep" ret: "int" args: "const struct timespec *" "struct timespec *" */
sysret_t sys_nanosleep(const struct timespec *rqtp, struct timespec *rmtp)
{
int ret = 0;
dbg_log(DBG_LOG, "sys_nanosleep\n");
if (!lwp_user_accessable((void *)rqtp, sizeof *rqtp))
return -EFAULT;
#ifdef ARCH_MM_MMU
struct timespec rqtp_k;
struct timespec rmtp_k;
lwp_get_from_user(&rqtp_k, (void *)rqtp, sizeof rqtp_k);
ret = nanosleep(&rqtp_k, &rmtp_k);
if ((ret != -1 || rt_get_errno() == EINTR) && rmtp && lwp_user_accessable((void *)rmtp, sizeof *rmtp))
{
lwp_put_to_user(rmtp, (void *)&rmtp_k, sizeof rmtp_k);
if(ret != 0)
return -EINTR;
}
#else
if (rmtp)
{
if (!lwp_user_accessable((void *)rmtp, sizeof *rmtp))
return -EFAULT;
ret = nanosleep(rqtp, rmtp);
}
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
/* syscall: "gettimeofday" ret: "int" args: "struct timeval *" "struct timezone *" */
sysret_t sys_gettimeofday(struct timeval *tp, struct timezone *tzp)
{
#ifdef ARCH_MM_MMU
struct timeval t_k;
if (tp)
{
if (!lwp_user_accessable((void *)tp, sizeof *tp))
{
return -EFAULT;
}
t_k.tv_sec = rt_tick_get() / RT_TICK_PER_SECOND;
t_k.tv_usec = (rt_tick_get() % RT_TICK_PER_SECOND) * (1000000 / RT_TICK_PER_SECOND);
lwp_put_to_user(tp, (void *)&t_k, sizeof t_k);
}
#else
if (tp)
{
if (!lwp_user_accessable((void *)tp, sizeof *tp))
{
return -EFAULT;
}
tp->tv_sec = rt_tick_get() / RT_TICK_PER_SECOND;
tp->tv_usec = (rt_tick_get() % RT_TICK_PER_SECOND) * (1000000 / RT_TICK_PER_SECOND);
}
#endif
return 0;
}
sysret_t sys_settimeofday(const struct timeval *tv, const struct timezone *tzp)
{
return 0;
}
sysret_t sys_exec(char *filename, int argc, char **argv, char **envp)
{
return lwp_execve(filename, 0, argc, argv, envp);
}
sysret_t sys_kill(int pid, int sig)
{
int ret = 0;
ret = lwp_kill(pid, sig);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_getpid(void)
{
return lwp_getpid();
}
/* syscall: "getpriority" ret: "int" args: "int" "id_t" */
sysret_t sys_getpriority(int which, id_t who)
{
if (which == PRIO_PROCESS)
{
rt_thread_t tid;
tid = rt_thread_self();
if (who == (id_t)(rt_size_t)tid || who == 0xff)
{
return tid->current_priority;
}
}
return 0xff;
}
/* syscall: "setpriority" ret: "int" args: "int" "id_t" "int" */
sysret_t sys_setpriority(int which, id_t who, int prio)
{
if (which == PRIO_PROCESS)
{
rt_thread_t tid;
tid = rt_thread_self();
if ((who == (id_t)(rt_size_t)tid || who == 0xff) && (prio >= 0 && prio < RT_THREAD_PRIORITY_MAX))
{
rt_thread_control(tid, RT_THREAD_CTRL_CHANGE_PRIORITY, &prio);
return 0;
}
}
return -1;
}
rt_sem_t sys_sem_create(const char *name, rt_uint32_t value, rt_uint8_t flag)
{
rt_sem_t sem = rt_sem_create(name, value, flag);
if (lwp_user_object_add(lwp_self(), (rt_object_t)sem) != 0)
{
rt_sem_delete(sem);
sem = NULL;
}
return sem;
}
rt_err_t sys_sem_delete(rt_sem_t sem)
{
return lwp_user_object_delete(lwp_self(), (rt_object_t)sem);
}
rt_err_t sys_sem_take(rt_sem_t sem, rt_int32_t time)
{
return rt_sem_take_interruptible(sem, time);
}
rt_err_t sys_sem_release(rt_sem_t sem)
{
return rt_sem_release(sem);
}
rt_mutex_t sys_mutex_create(const char *name, rt_uint8_t flag)
{
rt_mutex_t mutex = rt_mutex_create(name, flag);
if (lwp_user_object_add(lwp_self(), (rt_object_t)mutex) != 0)
{
rt_mutex_delete(mutex);
mutex = NULL;
}
return mutex;
}
rt_err_t sys_mutex_delete(rt_mutex_t mutex)
{
return lwp_user_object_delete(lwp_self(), (rt_object_t)mutex);
}
rt_err_t sys_mutex_take(rt_mutex_t mutex, rt_int32_t time)
{
return rt_mutex_take_interruptible(mutex, time);
}
rt_err_t sys_mutex_release(rt_mutex_t mutex)
{
return rt_mutex_release(mutex);
}
#ifdef ARCH_MM_MMU
/* memory allocation */
rt_base_t sys_brk(void *addr)
{
return lwp_brk(addr);
}
void *sys_mmap2(void *addr, size_t length, int prot,
int flags, int fd, off_t pgoffset)
{
return lwp_mmap2(addr, length, prot, flags, fd, pgoffset);
}
sysret_t sys_munmap(void *addr, size_t length)
{
return lwp_munmap(addr);
}
void *sys_mremap(void *old_address, size_t old_size,
size_t new_size, int flags, void *new_address)
{
return (void *)-1;
}
sysret_t sys_madvise(void *addr, size_t len, int behav)
{
return -ENOSYS;
}
#endif
rt_event_t sys_event_create(const char *name, rt_uint8_t flag)
{
rt_event_t event = rt_event_create(name, flag);
if (lwp_user_object_add(lwp_self(), (rt_object_t)event) != 0)
{
rt_event_delete(event);
event = NULL;
}
return event;
}
rt_err_t sys_event_delete(rt_event_t event)
{
return lwp_user_object_delete(lwp_self(), (rt_object_t)event);
}
rt_err_t sys_event_send(rt_event_t event, rt_uint32_t set)
{
return rt_event_send(event, set);
}
rt_err_t sys_event_recv(rt_event_t event,
rt_uint32_t set,
rt_uint8_t opt,
rt_int32_t timeout,
rt_uint32_t *recved)
{
if ((recved != NULL) && !lwp_user_accessable((void *)recved, sizeof(rt_uint32_t *)))
{
return -EFAULT;
}
return rt_event_recv(event, set, opt, timeout, recved);
}
rt_mailbox_t sys_mb_create(const char *name, rt_size_t size, rt_uint8_t flag)
{
rt_mailbox_t mb = rt_mb_create(name, size, flag);
if (lwp_user_object_add(lwp_self(), (rt_object_t)mb) != 0)
{
rt_mb_delete(mb);
mb = NULL;
}
return mb;
}
rt_err_t sys_mb_delete(rt_mailbox_t mb)
{
return lwp_user_object_delete(lwp_self(), (rt_object_t)mb);
}
rt_err_t sys_mb_send(rt_mailbox_t mb, rt_ubase_t value)
{
return rt_mb_send(mb, value);
}
rt_err_t sys_mb_send_wait(rt_mailbox_t mb,
rt_ubase_t value,
rt_int32_t timeout)
{
return rt_mb_send_wait(mb, value, timeout);
}
rt_err_t sys_mb_recv(rt_mailbox_t mb, rt_ubase_t *value, rt_int32_t timeout)
{
if (!lwp_user_accessable((void *)value, sizeof(rt_ubase_t *)))
{
return -EFAULT;
}
return rt_mb_recv(mb, (rt_ubase_t *)value, timeout);
}
rt_mq_t sys_mq_create(const char *name,
rt_size_t msg_size,
rt_size_t max_msgs,
rt_uint8_t flag)
{
rt_mq_t mq = rt_mq_create(name, msg_size, max_msgs, flag);
if (lwp_user_object_add(lwp_self(), (rt_object_t)mq) != 0)
{
rt_mq_delete(mq);
mq = NULL;
}
return mq;
}
rt_err_t sys_mq_delete(rt_mq_t mq)
{
return lwp_user_object_delete(lwp_self(), (rt_object_t)mq);
}
rt_err_t sys_mq_send(rt_mq_t mq, void *buffer, rt_size_t size)
{
if (!lwp_user_accessable((void *)buffer, size))
{
return -EFAULT;
}
return rt_mq_send(mq, buffer, size);
}
rt_err_t sys_mq_urgent(rt_mq_t mq, void *buffer, rt_size_t size)
{
if (!lwp_user_accessable((void *)buffer, size))
{
return -EFAULT;
}
return rt_mq_urgent(mq, buffer, size);
}
rt_err_t sys_mq_recv(rt_mq_t mq,
void *buffer,
rt_size_t size,
rt_int32_t timeout)
{
if (!lwp_user_accessable((void *)buffer, size))
{
return -EFAULT;
}
return rt_mq_recv(mq, buffer, size, timeout);
}
static void timer_timeout_callback(void *parameter)
{
rt_sem_t sem = (rt_sem_t)parameter;
rt_sem_release(sem);
}
rt_timer_t sys_rt_timer_create(const char *name,
void *data,
rt_tick_t time,
rt_uint8_t flag)
{
rt_timer_t timer = rt_timer_create(name, timer_timeout_callback, (void *)data, time, flag);
if (lwp_user_object_add(lwp_self(), (rt_object_t)timer) != 0)
{
rt_timer_delete(timer);
timer = NULL;
}
return timer;
}
rt_err_t sys_rt_timer_delete(rt_timer_t timer)
{
return lwp_user_object_delete(lwp_self(), (rt_object_t)timer);
}
rt_err_t sys_rt_timer_start(rt_timer_t timer)
{
return rt_timer_start(timer);
}
rt_err_t sys_rt_timer_stop(rt_timer_t timer)
{
return rt_timer_stop(timer);
}
rt_err_t sys_rt_timer_control(rt_timer_t timer, int cmd, void *arg)
{
return rt_timer_control(timer, cmd, arg);
}
/* MUSL compatible */
struct ksigevent
{
union sigval sigev_value;
int sigev_signo;
int sigev_notify;
int sigev_tid;
};
/* to protect unsafe implementation in current rt-smart toolchain */
RT_CTASSERT(sigevent_compatible, offsetof(struct ksigevent, sigev_tid) == offsetof(struct sigevent, sigev_notify_function));
rt_err_t sys_timer_create(clockid_t clockid, struct sigevent *restrict sevp, timer_t *restrict timerid)
{
int ret = 0;
#ifdef ARCH_MM_MMU
struct sigevent sevp_k;
timer_t timerid_k;
int utimer;
if (sevp == NULL)
{
sevp_k.sigev_notify = SIGEV_SIGNAL;
sevp_k.sigev_signo = SIGALRM;
sevp = &sevp_k;
}
else
{
/* clear extra bytes if any */
if (sizeof(struct ksigevent) < sizeof(struct sigevent))
memset(&sevp_k, 0, sizeof(sevp_k));
/* musl passes `struct ksigevent` to kernel, we shoule only get size of that bytes */
if (!lwp_get_from_user(&sevp_k, (void *)sevp, sizeof(struct ksigevent)))
{
return -EINVAL;
}
}
ret = _SYS_WRAP(timer_create(clockid, &sevp_k, &timerid_k));
if (ret != -RT_ERROR)
{
utimer = (rt_ubase_t)timerid_k;
if (!lwp_put_to_user(sevp, (void *)&sevp_k, sizeof(struct ksigevent)) ||
!lwp_put_to_user(timerid, (void *)&utimer, sizeof(utimer)))
ret = -EINVAL;
}
#else
ret = _SYS_WRAP(timer_create(clockid, sevp, timerid));
#endif
return ret;
}
rt_err_t sys_timer_delete(timer_t timerid)
{
int ret = timer_delete(timerid);
return (ret < 0 ? GET_ERRNO() : ret);
}
rt_err_t sys_timer_settime(timer_t timerid, int flags,
const struct itimerspec *restrict new_value,
struct itimerspec *restrict old_value)
{
int ret = 0;
#ifdef ARCH_MM_MMU
struct itimerspec new_value_k;
struct itimerspec old_value_k;
if (!lwp_get_from_user(&new_value_k, (void *)new_value, sizeof(*new_value)) ||
(old_value && !lwp_get_from_user(&old_value_k, (void *)old_value, sizeof(*old_value))))
{
return -EFAULT;
}
ret = timer_settime(timerid, flags, &new_value_k, &old_value_k);
lwp_put_to_user(old_value, (void *)&old_value_k, sizeof old_value_k);
#else
ret = timer_settime(timerid, flags, new_value, old_value);
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
rt_err_t sys_timer_gettime(timer_t timerid, struct itimerspec *curr_value)
{
int ret = 0;
#ifdef ARCH_MM_MMU
struct itimerspec curr_value_k;
lwp_get_from_user(&curr_value_k, (void *)curr_value, sizeof curr_value_k);
ret = timer_gettime(timerid, &curr_value_k);
lwp_put_to_user(curr_value, (void *)&curr_value_k, sizeof curr_value_k);
#else
ret = timer_gettime(timerid, curr_value);
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
rt_err_t sys_timer_getoverrun(timer_t timerid)
{
int ret = 0;
ret = timer_getoverrun(timerid);
return (ret < 0 ? GET_ERRNO() : ret);
}
rt_thread_t sys_thread_create(void *arg[])
{
rt_base_t level = 0;
void *user_stack = 0;
struct rt_lwp *lwp = 0;
rt_thread_t thread = RT_NULL;
int tid = 0;
lwp = rt_thread_self()->lwp;
lwp_ref_inc(lwp);
#ifdef ARCH_MM_MMU
user_stack = lwp_map_user(lwp, 0, (size_t)arg[3], 0);
if (!user_stack)
{
goto fail;
}
if ((tid = lwp_tid_get()) == 0)
{
goto fail;
}
thread = rt_thread_create((const char *)arg[0],
_crt_thread_entry,
(void *)arg[2],
ALLOC_KERNEL_STACK_SIZE,
(rt_uint8_t)(size_t)arg[4],
(rt_uint32_t)(rt_size_t)arg[5]);
if (!thread)
{
goto fail;
}
#ifdef RT_USING_SMP
thread->bind_cpu = lwp->bind_cpu;
#endif
thread->cleanup = lwp_cleanup;
thread->user_entry = (void (*)(void *))arg[1];
thread->user_stack = (void *)user_stack;
thread->user_stack_size = (rt_size_t)arg[3];
#else
rt_uint32_t kstack_size = (rt_uint32_t)arg[7];
if (kstack_size < ALLOC_KERNEL_STACK_SIZE_MIN)
{
/* When kstack size is 0, the default size of the kernel stack is used */
kstack_size = kstack_size ? ALLOC_KERNEL_STACK_SIZE_MIN : ALLOC_KERNEL_STACK_SIZE;
}
else if (kstack_size > ALLOC_KERNEL_STACK_SIZE_MAX)
{
kstack_size = ALLOC_KERNEL_STACK_SIZE_MAX;
}
user_stack = (void *)arg[3];
if ((!user_stack) || ((rt_uint32_t)arg[6] == RT_NULL))
{
goto fail;
}
if ((tid = lwp_tid_get()) == 0)
{
goto fail;
}
thread = rt_thread_create((const char *)arg[0], _crt_thread_entry, (void *)arg[2], kstack_size, (rt_uint8_t)(size_t)arg[5], (rt_uint32_t)arg[6]);
if (!thread)
{
goto fail;
}
thread->cleanup = lwp_cleanup;
thread->user_entry = (void (*)(void *))arg[1];
thread->user_stack = (void *)user_stack;
thread->user_stack_size = (uint32_t)arg[4];
rt_memset(thread->user_stack, '#', thread->user_stack_size);
#endif /* ARCH_MM_MMU */
thread->lwp = (void*)lwp;
thread->tid = tid;
lwp_tid_set_thread(tid, thread);
if (lwp->debug)
{
rt_thread_control(thread, RT_THREAD_CTRL_BIND_CPU, (void*)0);
}
level = rt_hw_interrupt_disable();
rt_list_insert_after(&lwp->t_grp, &thread->sibling);
rt_hw_interrupt_enable(level);
return thread;
fail:
lwp_tid_put(tid);
if (lwp)
{
lwp_ref_dec(lwp);
}
return RT_NULL;
}
#ifdef ARCH_MM_MMU
#define CLONE_VM 0x00000100
#define CLONE_FS 0x00000200
#define CLONE_FILES 0x00000400
#define CLONE_SIGHAND 0x00000800
#define CLONE_PTRACE 0x00002000
#define CLONE_VFORK 0x00004000
#define CLONE_PARENT 0x00008000
#define CLONE_THREAD 0x00010000
#define CLONE_NEWNS 0x00020000
#define CLONE_SYSVSEM 0x00040000
#define CLONE_SETTLS 0x00080000
#define CLONE_PARENT_SETTID 0x00100000
#define CLONE_CHILD_CLEARTID 0x00200000
#define CLONE_DETACHED 0x00400000
#define CLONE_UNTRACED 0x00800000
#define CLONE_CHILD_SETTID 0x01000000
#define CLONE_NEWCGROUP 0x02000000
#define CLONE_NEWUTS 0x04000000
#define CLONE_NEWIPC 0x08000000
#define CLONE_NEWUSER 0x10000000
#define CLONE_NEWPID 0x20000000
#define CLONE_NEWNET 0x40000000
#define CLONE_IO 0x80000000
/* arg[] -> flags
* stack
* new_tid
* tls
* set_clear_tid_address
* quit_func
* start_args
* */
#define SYS_CLONE_ARGS_NR 7
long _sys_clone(void *arg[])
{
rt_base_t level = 0;
struct rt_lwp *lwp = 0;
rt_thread_t thread = RT_NULL;
rt_thread_t self = RT_NULL;
int tid = 0;
unsigned long flags = 0;
void *user_stack = RT_NULL;
int *new_tid = RT_NULL;
void *tls = RT_NULL;
/*
musl call flags (CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND
| CLONE_THREAD | CLONE_SYSVSEM | CLONE_SETTLS
| CLONE_PARENT_SETTID | CLONE_CHILD_CLEARTID | CLONE_DETACHED);
*/
/* check args */
if (!lwp_user_accessable(arg, sizeof(void *[SYS_CLONE_ARGS_NR])))
{
return -EFAULT;
}
flags = (unsigned long)(size_t)arg[0];
if ((flags & (CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_THREAD | CLONE_SYSVSEM))
!= (CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_THREAD | CLONE_SYSVSEM))
{
return -EINVAL;
}
user_stack = arg[1];
new_tid = (int *)arg[2];
tls = (void *)arg[3];
if ((flags & CLONE_PARENT_SETTID) == CLONE_PARENT_SETTID)
{
if (!lwp_user_accessable(new_tid, sizeof(int)))
{
return -EFAULT;
}
}
self = rt_thread_self();
lwp = self->lwp;
lwp_ref_inc(lwp);
if (!user_stack)
{
SET_ERRNO(EINVAL);
goto fail;
}
if ((tid = lwp_tid_get()) == 0)
{
SET_ERRNO(ENOMEM);
goto fail;
}
thread = rt_thread_create(self->name,
RT_NULL,
RT_NULL,
self->stack_size,
self->init_priority,
self->init_tick);
if (!thread)
{
goto fail;
}
#ifdef RT_USING_SMP
thread->bind_cpu = lwp->bind_cpu;
#endif
thread->cleanup = lwp_cleanup;
thread->user_entry = RT_NULL;
thread->user_stack = RT_NULL;
thread->user_stack_size = 0;
thread->lwp = (void *)lwp;
thread->tid = tid;
if ((flags & CLONE_SETTLS) == CLONE_SETTLS)
{
thread->thread_idr = tls;
}
if ((flags & CLONE_PARENT_SETTID) == CLONE_PARENT_SETTID)
{
*new_tid = (int)(tid);
}
if ((flags & CLONE_CHILD_CLEARTID) == CLONE_CHILD_CLEARTID)
{
thread->clear_child_tid = (int *)arg[4];
}
if (lwp->debug)
{
rt_thread_control(thread, RT_THREAD_CTRL_BIND_CPU, (void*)0);
}
level = rt_hw_interrupt_disable();
rt_list_insert_after(&lwp->t_grp, &thread->sibling);
rt_hw_interrupt_enable(level);
/* copy origin stack */
rt_memcpy(thread->stack_addr, self->stack_addr, thread->stack_size);
lwp_tid_set_thread(tid, thread);
arch_set_thread_context(arch_clone_exit,
(void *)((char *)thread->stack_addr + thread->stack_size),
user_stack, &thread->sp);
/* new thread never reach there */
rt_thread_startup(thread);
return (long)tid;
fail:
lwp_tid_put(tid);
if (lwp)
{
lwp_ref_dec(lwp);
}
return GET_ERRNO();
}
rt_weak long sys_clone(void *arg[])
{
return _sys_clone(arg);
}
int lwp_dup_user(rt_varea_t varea, void *arg);
static int _copy_process(struct rt_lwp *dest_lwp, struct rt_lwp *src_lwp)
{
int err;
dest_lwp->lwp_obj->source = src_lwp->aspace;
err = rt_aspace_traversal(src_lwp->aspace, lwp_dup_user, dest_lwp);
dest_lwp->lwp_obj->source = NULL;
return err;
}
static void lwp_struct_copy(struct rt_lwp *dst, struct rt_lwp *src)
{
#ifdef ARCH_MM_MMU
dst->end_heap = src->end_heap;
#endif
dst->lwp_type = src->lwp_type;
dst->text_entry = src->text_entry;
dst->text_size = src->text_size;
dst->data_entry = src->data_entry;
dst->data_size = src->data_size;
dst->args = src->args;
dst->leader = 0;
dst->session = src->session;
dst->tty_old_pgrp = 0;
dst->__pgrp = src->__pgrp;
dst->tty = src->tty;
rt_memcpy(dst->cmd, src->cmd, RT_NAME_MAX);
dst->sa_flags = src->sa_flags;
dst->signal_mask = src->signal_mask;
rt_memcpy(dst->signal_handler, src->signal_handler, sizeof dst->signal_handler);
}
static int lwp_copy_files(struct rt_lwp *dst, struct rt_lwp *src)
{
struct dfs_fdtable *dst_fdt;
struct dfs_fdtable *src_fdt;
src_fdt = &src->fdt;
dst_fdt = &dst->fdt;
/* init fds */
dst_fdt->fds = rt_calloc(src_fdt->maxfd, sizeof(void *));
if (dst_fdt->fds)
{
struct dfs_fd *d_s;
int i;
dst_fdt->maxfd = src_fdt->maxfd;
dfs_fd_lock();
/* dup files */
for (i = 0; i < src_fdt->maxfd; i++)
{
d_s = fdt_fd_get(src_fdt, i);
if (d_s)
{
dst_fdt->fds[i] = d_s;
d_s->ref_count++;
}
}
dfs_fd_unlock();
return 0;
}
return -RT_ERROR;
}
sysret_t _sys_fork(void)
{
rt_base_t level;
int tid = 0;
rt_err_t falival = 0;
struct rt_lwp *lwp = RT_NULL;
struct rt_lwp *self_lwp = RT_NULL;
rt_thread_t thread = RT_NULL;
rt_thread_t self_thread = RT_NULL;
void *user_stack = RT_NULL;
/* new lwp */
lwp = lwp_new();
if (!lwp)
{
SET_ERRNO(ENOMEM);
goto fail;
}
/* new tid */
if ((tid = lwp_tid_get()) == 0)
{
SET_ERRNO(ENOMEM);
goto fail;
}
/* user space init */
if (lwp_user_space_init(lwp, 1) != 0)
{
SET_ERRNO(ENOMEM);
goto fail;
}
self_lwp = lwp_self();
/* copy process */
if (_copy_process(lwp, self_lwp) != 0)
{
SET_ERRNO(ENOMEM);
goto fail;
}
/* copy lwp struct data */
lwp_struct_copy(lwp, self_lwp);
/* copy files */
if (lwp_copy_files(lwp, self_lwp) != 0)
{
SET_ERRNO(ENOMEM);
goto fail;
}
/* create thread */
self_thread = rt_thread_self();
thread = rt_thread_create(self_thread->name,
RT_NULL,
RT_NULL,
self_thread->stack_size,
self_thread->init_priority,
self_thread->init_tick);
if (!thread)
{
SET_ERRNO(ENOMEM);
goto fail;
}
thread->cleanup = self_thread->cleanup;
thread->user_entry = self_thread->user_entry;
thread->user_stack = self_thread->user_stack;
thread->user_stack_size = self_thread->user_stack_size;
thread->signal_mask = self_thread->signal_mask;
thread->thread_idr = self_thread->thread_idr;
thread->lwp = (void *)lwp;
thread->tid = tid;
level = rt_hw_interrupt_disable();
/* add thread to lwp process */
rt_list_insert_after(&lwp->t_grp, &thread->sibling);
/* lwp add to children link */
lwp->sibling = self_lwp->first_child;
self_lwp->first_child = lwp;
lwp->parent = self_lwp;
rt_hw_interrupt_enable(level);
/* copy origin stack */
rt_memcpy(thread->stack_addr, self_thread->stack_addr, self_thread->stack_size);
lwp_tid_set_thread(tid, thread);
/* duplicate user objects */
lwp_user_object_dup(lwp, self_lwp);
level = rt_hw_interrupt_disable();
user_stack = arch_get_user_sp();
rt_hw_interrupt_enable(level);
arch_set_thread_context(arch_fork_exit,
(void *)((char *)thread->stack_addr + thread->stack_size),
user_stack, &thread->sp);
/* new thread never reach there */
level = rt_hw_interrupt_disable();
if (lwp->tty != RT_NULL)
{
int ret;
struct rt_lwp *old_lwp;
old_lwp = lwp->tty->foreground;
rt_mutex_take(&lwp->tty->lock, RT_WAITING_FOREVER);
ret = tty_push(&lwp->tty->head, old_lwp);
rt_mutex_release(&lwp->tty->lock);
if (ret < 0)
{
LOG_E("malloc fail!\n");
SET_ERRNO(ENOMEM);
goto fail;
}
lwp->tty->foreground = lwp;
}
rt_hw_interrupt_enable(level);
rt_thread_startup(thread);
return lwp_to_pid(lwp);
fail:
falival = GET_ERRNO();
if (tid != 0)
{
lwp_tid_put(tid);
}
if (lwp)
{
lwp_ref_dec(lwp);
}
return falival;
}
size_t lwp_user_strlen(const char *s, int *err)
{
size_t len = 0;
while (1)
{
if (!lwp_user_accessable((void *)(s + len), sizeof(char)))
{
if (err)
{
*err = 1;
}
return 0;
}
if (s[len] == '\0')
{
if (err)
{
*err = 0;
}
return len;
}
len++;
}
}
/* arm needs to wrap fork/clone call to preserved lr & caller saved regs */
rt_weak sysret_t sys_fork(void)
{
return _sys_fork();
}
rt_weak sysret_t sys_vfork(void)
{
return sys_fork();
}
struct process_aux *lwp_argscopy(struct rt_lwp *lwp, int argc, char **argv, char **envp);
int lwp_load(const char *filename, struct rt_lwp *lwp, uint8_t *load_addr, size_t addr_size, struct process_aux *aux);
void lwp_user_obj_free(struct rt_lwp *lwp);
#define _swap_lwp_data(lwp_used, lwp_new, type, member) \
do {\
type tmp;\
tmp = lwp_used->member;\
lwp_used->member = lwp_new->member;\
lwp_new->member = tmp;\
} while (0)
static char *_insert_args(int new_argc, char *new_argv[], struct lwp_args_info *args)
{
void *page = NULL;
int err = 0;
char **nargv;
char **nenvp;
char *p;
int i, len;
int nsize;
if (new_argc == 0)
{
goto quit;
}
page = rt_pages_alloc(0); /* 1 page */
if (!page)
{
goto quit;
}
nsize = new_argc * sizeof(char *);
for (i = 0; i < new_argc; i++)
{
nsize += rt_strlen(new_argv[i]) + 1;
}
if (nsize + args->size > ARCH_PAGE_SIZE)
{
err = 1;
goto quit;
}
nargv = (char **)page;
nenvp = nargv + args->argc + new_argc + 1;
p = (char *)(nenvp + args->envc + 1);
/* insert argv */
for (i = 0; i < new_argc; i++)
{
nargv[i] = p;
len = rt_strlen(new_argv[i]) + 1;
rt_memcpy(p, new_argv[i], len);
p += len;
}
/* copy argv */
nargv += new_argc;
for (i = 0; i < args->argc; i++)
{
nargv[i] = p;
len = rt_strlen(args->argv[i]) + 1;
rt_memcpy(p, args->argv[i], len);
p += len;
}
nargv[i] = NULL;
/* copy envp */
for (i = 0; i < args->envc; i++)
{
nenvp[i] = p;
len = rt_strlen(args->envp[i]) + 1;
rt_memcpy(p, args->envp[i], len);
p += len;
}
nenvp[i] = NULL;
/* update args */
args->argv = (char **)page;
args->argc = args->argc + new_argc;
args->envp = args->argv + args->argc + 1;
/* args->envc no change */
args->size = args->size + nsize;
quit:
if (err && page)
{
rt_pages_free(page, 0);
page = NULL;
}
return page;
}
#define INTERP_BUF_SIZE 128
static char *_load_script(const char *filename, struct lwp_args_info *args)
{
void *page = NULL;
char *new_page;
int fd = -RT_ERROR;
int len;
char interp[INTERP_BUF_SIZE];
char *cp;
char *i_name;
char *i_arg;
fd = open(filename, O_BINARY | O_RDONLY, 0);
if (fd < 0)
{
goto quit;
}
len = read(fd, interp, INTERP_BUF_SIZE);
if (len < 2)
{
goto quit;
}
if ((interp[0] != '#') || (interp[1] != '!'))
{
goto quit;
}
if (len == INTERP_BUF_SIZE)
{
len--;
}
interp[len] = '\0';
if ((cp = strchr(interp, '\n')) == NULL)
{
cp = interp + INTERP_BUF_SIZE - 1;
}
*cp = '\0';
while (cp > interp)
{
cp--;
if ((*cp == ' ') || (*cp == '\t'))
{
*cp = '\0';
}
else
{
break;
}
}
for (cp = interp + 2; (*cp == ' ') || (*cp == '\t'); cp++)
{
/* nothing */
}
if (*cp == '\0')
{
goto quit; /* No interpreter name found */
}
i_name = cp;
i_arg = NULL;
for (; *cp && (*cp != ' ') && (*cp != '\t'); cp++)
{
/* nothing */
}
while ((*cp == ' ') || (*cp == '\t'))
{
*cp++ = '\0';
}
if (*cp)
{
i_arg = cp;
}
if (i_arg)
{
new_page = _insert_args(1, &i_arg, args);
rt_pages_free(page, 0);
page = new_page;
if (!page)
{
goto quit;
}
}
new_page = _insert_args(1, &i_name, args);
rt_pages_free(page, 0);
page = new_page;
quit:
if (fd >= 0)
{
close(fd);
}
return page;
}
int load_ldso(struct rt_lwp *lwp, char *exec_name, char *const argv[], char *const envp[])
{
int ret = -1;
int i;
void *page;
void *new_page;
int argc = 0;
int envc = 0;
int size;
char **kargv;
char **kenvp;
size_t len;
char *p;
char *i_arg;
struct lwp_args_info args_info;
struct process_aux *aux;
size = sizeof(char *);
if (argv)
{
while (1)
{
if (!argv[argc])
{
break;
}
len = rt_strlen((const char *)argv[argc]);
size += sizeof(char *) + len + 1;
argc++;
}
}
if (envp)
{
while (1)
{
if (!envp[envc])
{
break;
}
len = rt_strlen((const char *)envp[envc]);
size += sizeof(char *) + len + 1;
envc++;
}
}
page = rt_pages_alloc(0); /* 1 page */
if (!page)
{
SET_ERRNO(ENOMEM);
goto quit;
}
kargv = (char **)page;
kenvp = kargv + argc + 1;
p = (char *)(kenvp + envc + 1);
/* copy argv */
if (argv)
{
for (i = 0; i < argc; i++)
{
kargv[i] = p;
len = rt_strlen(argv[i]) + 1;
rt_memcpy(p, argv[i], len);
p += len;
}
kargv[i] = NULL;
}
/* copy envp */
if (envp)
{
for (i = 0; i < envc; i++)
{
kenvp[i] = p;
len = rt_strlen(envp[i]) + 1;
rt_memcpy(p, envp[i], len);
p += len;
}
kenvp[i] = NULL;
}
args_info.argc = argc;
args_info.argv = kargv;
args_info.envc = envc;
args_info.envp = kenvp;
args_info.size = size;
new_page = _insert_args(1, &exec_name, &args_info);
rt_pages_free(page, 0);
page = new_page;
if (!page)
{
SET_ERRNO(ENOMEM);
goto quit;
}
i_arg = "-e";
new_page = _insert_args(1, &i_arg, &args_info);
rt_pages_free(page, 0);
page = new_page;
if (!page)
{
SET_ERRNO(ENOMEM);
goto quit;
}
i_arg = "ld.so";
new_page = _insert_args(1, &i_arg, &args_info);
rt_pages_free(page, 0);
page = new_page;
if (!page)
{
SET_ERRNO(ENOMEM);
goto quit;
}
if ((aux = lwp_argscopy(lwp, args_info.argc, args_info.argv, args_info.envp)) == NULL)
{
SET_ERRNO(ENOMEM);
goto quit;
}
ret = lwp_load("/lib/ld.so", lwp, RT_NULL, 0, aux);
rt_strncpy(lwp->cmd, exec_name, RT_NAME_MAX);
quit:
if (page)
{
rt_pages_free(page, 0);
}
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_execve(const char *path, char *const argv[], char *const envp[])
{
int ret = -1;
int argc = 0;
int envc = 0;
void *page = NULL;
void *new_page;
int size = 0;
size_t len;
int access_err;
char **kargv;
char **kenvp;
char *p;
struct rt_lwp *new_lwp = NULL;
struct rt_lwp *lwp;
rt_base_t level;
int uni_thread;
rt_thread_t thread;
struct process_aux *aux;
int i;
struct lwp_args_info args_info;
lwp = lwp_self();
thread = rt_thread_self();
uni_thread = 1;
level = rt_hw_interrupt_disable();
if (lwp->t_grp.prev != &thread->sibling)
{
uni_thread = 0;
}
if (lwp->t_grp.next != &thread->sibling)
{
uni_thread = 0;
}
rt_hw_interrupt_enable(level);
if (!uni_thread)
{
SET_ERRNO(EINVAL);
goto quit;
}
len = lwp_user_strlen(path, &access_err);
if (access_err)
{
SET_ERRNO(EFAULT);
goto quit;
}
size += sizeof(char *);
if (argv)
{
while (1)
{
if (!lwp_user_accessable((void *)(argv + argc), sizeof(char *)))
{
SET_ERRNO(EFAULT);
goto quit;
}
if (!argv[argc])
{
break;
}
len = lwp_user_strlen((const char *)argv[argc], &access_err);
if (access_err)
{
SET_ERRNO(EFAULT);
goto quit;
}
size += sizeof(char *) + len + 1;
argc++;
}
}
size += sizeof(char *);
if (envp)
{
while (1)
{
if (!lwp_user_accessable((void *)(envp + envc), sizeof(char *)))
{
SET_ERRNO(EFAULT);
goto quit;
}
if (!envp[envc])
{
break;
}
len = lwp_user_strlen((const char *)envp[envc], &access_err);
if (access_err)
{
SET_ERRNO(EFAULT);
goto quit;
}
size += sizeof(char *) + len + 1;
envc++;
}
}
if (size > ARCH_PAGE_SIZE)
{
SET_ERRNO(EINVAL);
goto quit;
}
page = rt_pages_alloc(0); /* 1 page */
if (!page)
{
SET_ERRNO(ENOMEM);
goto quit;
}
kargv = (char **)page;
kenvp = kargv + argc + 1;
p = (char *)(kenvp + envc + 1);
/* copy argv */
if (argv)
{
for (i = 0; i < argc; i++)
{
kargv[i] = p;
len = rt_strlen(argv[i]) + 1;
rt_memcpy(p, argv[i], len);
p += len;
}
kargv[i] = NULL;
}
/* copy envp */
if (envp)
{
for (i = 0; i < envc; i++)
{
kenvp[i] = p;
len = rt_strlen(envp[i]) + 1;
rt_memcpy(p, envp[i], len);
p += len;
}
kenvp[i] = NULL;
}
/* alloc new lwp to operation */
new_lwp = (struct rt_lwp *)rt_malloc(sizeof(struct rt_lwp));
if (!new_lwp)
{
SET_ERRNO(ENOMEM);
goto quit;
}
rt_memset(new_lwp, 0, sizeof(struct rt_lwp));
new_lwp->ref = 1;
lwp_user_object_lock_init(new_lwp);
ret = lwp_user_space_init(new_lwp, 0);
if (ret != 0)
{
SET_ERRNO(ENOMEM);
goto quit;
}
/* file is a script ? */
args_info.argc = argc;
args_info.argv = kargv;
args_info.envc = envc;
args_info.envp = kenvp;
args_info.size = size;
while (1)
{
new_page = _load_script(path, &args_info);
if (!new_page)
{
break;
}
rt_pages_free(page, 0);
page = new_page;
path = args_info.argv[0];
}
/* now load elf */
if ((aux = lwp_argscopy(new_lwp, args_info.argc, args_info.argv, args_info.envp)) == NULL)
{
SET_ERRNO(ENOMEM);
goto quit;
}
ret = lwp_load(path, new_lwp, RT_NULL, 0, aux);
if (ret == 1)
{
/* dynamic */
lwp_unmap_user(new_lwp, (void *)(USER_VADDR_TOP - ARCH_PAGE_SIZE));
ret = load_ldso(new_lwp, (char *)path, args_info.argv, args_info.envp);
}
if (ret == RT_EOK)
{
int off = 0;
int last_backslash = 0;
char *run_name = args_info.argv[0];
/* clear all user objects */
lwp_user_object_clear(lwp);
/* find last \ or / */
while (1)
{
char c = run_name[off++];
if (c == '\0')
{
break;
}
if (c == '\\' || c == '/')
{
last_backslash = off;
}
}
/* load ok, now set thread name and swap the data of lwp and new_lwp */
level = rt_hw_interrupt_disable();
rt_strncpy(thread->name, run_name + last_backslash, RT_NAME_MAX);
rt_pages_free(page, 0);
#ifdef ARCH_MM_MMU
_swap_lwp_data(lwp, new_lwp, struct rt_aspace *, aspace);
_swap_lwp_data(lwp, new_lwp, struct rt_lwp_objs *, lwp_obj);
_swap_lwp_data(lwp, new_lwp, size_t, end_heap);
#endif
_swap_lwp_data(lwp, new_lwp, uint8_t, lwp_type);
_swap_lwp_data(lwp, new_lwp, void *, text_entry);
_swap_lwp_data(lwp, new_lwp, uint32_t, text_size);
_swap_lwp_data(lwp, new_lwp, void *, data_entry);
_swap_lwp_data(lwp, new_lwp, uint32_t, data_size);
_swap_lwp_data(lwp, new_lwp, void *, args);
rt_memset(&thread->signal_mask, 0, sizeof(thread->signal_mask));
rt_memset(&thread->signal_mask_bak, 0, sizeof(thread->signal_mask_bak));
lwp->sa_flags = 0;
rt_memset(&lwp->signal_mask, 0, sizeof(lwp->signal_mask));
rt_memset(&lwp->signal_mask_bak, 0, sizeof(lwp->signal_mask_bak));
rt_memset(lwp->signal_handler, 0, sizeof(lwp->signal_handler));
/* to do: clsoe files with flag CLOEXEC */
lwp_aspace_switch(thread);
rt_hw_interrupt_enable(level);
lwp_ref_dec(new_lwp);
arch_start_umode(lwp->args,
lwp->text_entry,
(void*)USER_STACK_VEND,
thread->stack_addr + thread->stack_size);
/* never reach here */
}
return -EINVAL;
quit:
if (page)
{
rt_pages_free(page, 0);
}
if (new_lwp)
{
lwp_ref_dec(new_lwp);
}
return (ret < 0 ? GET_ERRNO() : ret);
}
#endif /* ARCH_MM_MMU */
rt_err_t sys_thread_delete(rt_thread_t thread)
{
#ifdef ARCH_MM_MMU
return rt_thread_delete(thread);
#else
rt_err_t ret = 0;
if(thread->type != RT_Object_Class_Thread)
{
ret = -EINVAL;
goto __exit;
}
ret = rt_thread_delete(thread);
if (rt_thread_self() == thread)
{
rt_schedule();
}
__exit:
return ret;
#endif
}
rt_err_t sys_thread_startup(rt_thread_t thread)
{
return rt_thread_startup(thread);
}
rt_thread_t sys_thread_self(void)
{
return rt_thread_self();
}
/* sys channel */
sysret_t sys_channel_open(const char *name, int flags)
{
return lwp_channel_open(FDT_TYPE_LWP, name, flags);
}
rt_err_t sys_channel_close(int fd)
{
return lwp_channel_close(FDT_TYPE_LWP, fd);
}
rt_err_t sys_channel_send(int fd, rt_channel_msg_t data)
{
return lwp_channel_send(FDT_TYPE_LWP, fd, data);
}
rt_err_t sys_channel_send_recv_timeout(int fd, rt_channel_msg_t data, rt_channel_msg_t data_ret, rt_int32_t time)
{
return lwp_channel_send_recv_timeout(FDT_TYPE_LWP, fd, data, data_ret, time);
}
rt_err_t sys_channel_reply(int fd, rt_channel_msg_t data)
{
return lwp_channel_reply(FDT_TYPE_LWP, fd, data);
}
rt_err_t sys_channel_recv_timeout(int fd, rt_channel_msg_t data, rt_int32_t time)
{
return lwp_channel_recv_timeout(FDT_TYPE_LWP, fd, data, time);
}
static struct rt_semaphore critical_lock;
static int critical_init(void)
{
rt_sem_init(&critical_lock, "ct_lock", 1, RT_IPC_FLAG_FIFO);
return 0;
}
INIT_DEVICE_EXPORT(critical_init);
void sys_enter_critical(void)
{
rt_sem_take(&critical_lock, RT_WAITING_FOREVER);
}
void sys_exit_critical(void)
{
rt_sem_release(&critical_lock);
}
/* syscall: "sys_log" ret: "int" args: "const char*" "size" */
static int __sys_log_enable = 0;
static int sys_log_enable(int argc, char** argv)
{
if (argc == 1)
{
rt_kprintf("sys_log = %d\n", __sys_log_enable);
return 0;
}
else
{
__sys_log_enable = atoi(argv[1]);
}
return 0;
}
MSH_CMD_EXPORT_ALIAS(sys_log_enable, sys_log, sys_log 1(enable)/0(disable));
sysret_t sys_log(const char* log, int size)
{
rt_device_t console = rt_console_get_device();
if (console && __sys_log_enable)
{
rt_device_write(console, -1, log, size);
}
return 0;
}
sysret_t sys_stat(const char *file, struct stat *buf)
{
int ret = 0;
int err;
size_t len;
size_t copy_len;
char *copy_path;
struct stat statbuff = {0};
if (!lwp_user_accessable((void *)buf, sizeof(struct stat)))
{
return -EFAULT;
}
len = lwp_user_strlen(file, &err);
if (err)
{
return -EFAULT;
}
copy_path = (char*)rt_malloc(len + 1);
if (!copy_path)
{
return -ENOMEM;
}
copy_len = lwp_get_from_user(copy_path, (void*)file, len);
if (copy_len == 0)
{
rt_free(copy_path);
return -EFAULT;
}
copy_path[copy_len] = '\0';
ret = _SYS_WRAP(stat(copy_path, &statbuff));
rt_free(copy_path);
if (ret == 0)
{
lwp_put_to_user(buf, &statbuff, sizeof statbuff);
}
return ret;
}
sysret_t sys_notimpl(void)
{
return -ENOSYS;
}
uint32_t sys_hw_interrupt_disable(void)
{
return rt_hw_interrupt_disable();
}
void sys_hw_interrupt_enable(uint32_t level)
{
rt_hw_interrupt_enable(level);
}
#ifdef ARCH_MM_MMU
sysret_t sys_shmget(size_t key, size_t size, int create)
{
return lwp_shmget(key, size, create);
}
sysret_t sys_shmrm(int id)
{
return lwp_shmrm(id);
}
void* sys_shmat(int id, void* shm_vaddr)
{
return lwp_shmat(id, shm_vaddr);
}
sysret_t sys_shmdt(void* shm_vaddr)
{
return lwp_shmdt(shm_vaddr);
}
#elif defined RT_LWP_USING_SHM
void *sys_shm_alloc(int size)
{
if (size < 0)
{
return RT_NULL;
}
return lwp_shm_alloc((rt_size_t)size);
}
void *sys_shm_retain(void *mem)
{
if (!lwp_user_accessable(mem, sizeof (void *)))
{
return RT_NULL;
}
return lwp_shm_retain(mem);
}
sysret_t sys_shm_free(void *mem)
{
if (!lwp_user_accessable(mem, sizeof (void *)))
{
return -EFAULT;
}
lwp_shm_free(mem);
return 0;
}
#endif
/* device interfaces */
rt_err_t sys_device_init(rt_device_t dev)
{
return rt_device_init(dev);
}
rt_err_t sys_device_register(rt_device_t dev, const char *name, rt_uint16_t flags)
{
return rt_device_register(dev, name, flags);
}
rt_err_t sys_device_control(rt_device_t dev, int cmd, void *arg)
{
return rt_device_control(dev, cmd, arg);
}
rt_device_t sys_device_find(const char* name)
{
return rt_device_find(name);
}
rt_err_t sys_device_open(rt_device_t dev, rt_uint16_t oflag)
{
return rt_device_open(dev, oflag);
}
rt_err_t sys_device_close(rt_device_t dev)
{
return rt_device_close(dev);
}
rt_ssize_t sys_device_read(rt_device_t dev, rt_off_t pos, void *buffer, rt_size_t size)
{
return rt_device_read(dev, pos, buffer, size);
}
rt_ssize_t sys_device_write(rt_device_t dev, rt_off_t pos, const void *buffer, rt_size_t size)
{
return rt_device_write(dev, pos, buffer, size);
}
#ifdef RT_USING_SAL
/* network interfaces */
sysret_t sys_accept(int socket, struct musl_sockaddr *addr, socklen_t *addrlen)
{
int ret = -1;
struct sockaddr ksa;
struct musl_sockaddr kmusladdr;
socklen_t uaddrlen;
socklen_t kaddrlen;
if (addr)
{
if (!lwp_user_accessable(addrlen, sizeof (socklen_t)))
{
return -EFAULT;
}
lwp_get_from_user(&uaddrlen, addrlen, sizeof (socklen_t));
if (!uaddrlen)
{
return -EINVAL;
}
if (!lwp_user_accessable(addr, uaddrlen))
{
return -EFAULT;
}
}
kaddrlen = sizeof(struct sockaddr);
ret = accept(socket, &ksa, &kaddrlen);
if (ret >= 0)
{
if (addr)
{
sockaddr_tomusl(&ksa, &kmusladdr);
if (uaddrlen > sizeof(struct musl_sockaddr))
{
uaddrlen = sizeof(struct musl_sockaddr);
}
lwp_put_to_user(addr, &kmusladdr, uaddrlen);
lwp_put_to_user(addrlen, &uaddrlen, sizeof (socklen_t));
}
}
return ret;
}
sysret_t sys_bind(int socket, const struct musl_sockaddr *name, socklen_t namelen)
{
struct sockaddr sa;
struct musl_sockaddr kname;
if (!lwp_user_accessable((void *)name, namelen))
{
return -EFAULT;
}
#ifdef SAL_USING_AF_UNIX
if (name->sa_family == AF_UNIX)
{
namelen = sizeof(struct sockaddr);
}
#endif /* SAL_USING_AF_UNIX */
lwp_get_from_user(&kname, (void *)name, namelen);
sockaddr_tolwip(&kname, &sa);
return bind(socket, &sa, namelen);
}
sysret_t sys_shutdown(int socket, int how)
{
return shutdown(socket, how);
}
sysret_t sys_getpeername(int socket, struct musl_sockaddr *name, socklen_t *namelen)
{
int ret = -1;
struct sockaddr sa;
struct musl_sockaddr kname;
socklen_t unamelen;
socklen_t knamelen;
if (!lwp_user_accessable(namelen, sizeof (socklen_t *)))
{
return -EFAULT;
}
lwp_get_from_user(&unamelen, namelen, sizeof (socklen_t *));
if (!unamelen)
{
return -EINVAL;
}
if (!lwp_user_accessable(name, unamelen))
{
return -EFAULT;
}
knamelen = sizeof(struct sockaddr);
ret = getpeername(socket, &sa, &knamelen);
if (ret == 0)
{
sockaddr_tomusl(&sa, &kname);
if (unamelen > sizeof(struct musl_sockaddr))
{
unamelen = sizeof(struct musl_sockaddr);
}
lwp_put_to_user(name, &kname, unamelen);
lwp_put_to_user(namelen, &unamelen, sizeof (socklen_t *));
}
else
{
ret = GET_ERRNO();
}
return ret;
}
sysret_t sys_getsockname(int socket, struct musl_sockaddr *name, socklen_t *namelen)
{
int ret = -1;
struct sockaddr sa;
struct musl_sockaddr kname;
socklen_t unamelen;
socklen_t knamelen;
if (!lwp_user_accessable(namelen, sizeof (socklen_t *)))
{
return -EFAULT;
}
lwp_get_from_user(&unamelen, namelen, sizeof (socklen_t *));
if (!unamelen)
{
return -EINVAL;
}
if (!lwp_user_accessable(name, unamelen))
{
return -EFAULT;
}
knamelen = sizeof(struct sockaddr);
ret = getsockname(socket, &sa, &knamelen);
if (ret == 0)
{
sockaddr_tomusl(&sa, &kname);
if (unamelen > sizeof(struct musl_sockaddr))
{
unamelen = sizeof(struct musl_sockaddr);
}
lwp_put_to_user(name, &kname, unamelen);
lwp_put_to_user(namelen, &unamelen, sizeof(socklen_t *));
}
else
{
ret = GET_ERRNO();
}
return ret;
}
sysret_t sys_getsockopt(int socket, int level, int optname, void *optval, socklen_t *optlen)
{
int ret;
convert_sockopt(&level, &optname);
ret = getsockopt(socket, level, optname, optval, optlen);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_setsockopt(int socket, int level, int optname, const void *optval, socklen_t optlen)
{
int ret;
convert_sockopt(&level, &optname);
ret = setsockopt(socket, level, optname, optval, optlen);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_connect(int socket, const struct musl_sockaddr *name, socklen_t namelen)
{
int ret;
struct sockaddr sa;
struct musl_sockaddr kname;
if (!lwp_user_accessable((void *)name, namelen))
{
return -EFAULT;
}
#ifdef SAL_USING_AF_UNIX
if (name->sa_family == AF_UNIX)
{
namelen = sizeof(struct sockaddr);
}
#endif /* SAL_USING_AF_UNIX */
lwp_get_from_user(&kname, (void *)name, namelen);
sockaddr_tolwip(&kname, &sa);
ret = connect(socket, &sa, namelen);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_listen(int socket, int backlog)
{
return listen(socket, backlog);
}
#define MUSLC_MSG_OOB 0x0001
#define MUSLC_MSG_PEEK 0x0002
#define MUSLC_MSG_DONTWAIT 0x0040
#define MUSLC_MSG_WAITALL 0x0100
#define MUSLC_MSG_MORE 0x8000
static int netflags_muslc_2_lwip(int flags)
{
int flgs = 0;
if (flags & MUSLC_MSG_PEEK)
{
flgs |= MSG_PEEK;
}
if (flags & MUSLC_MSG_WAITALL)
{
flgs |= MSG_WAITALL;
}
if (flags & MUSLC_MSG_OOB)
{
flgs |= MSG_OOB;
}
if (flags & MUSLC_MSG_DONTWAIT)
{
flgs |= MSG_DONTWAIT;
}
if (flags & MUSLC_MSG_MORE)
{
flgs |= MSG_MORE;
}
return flgs;
}
sysret_t sys_recvfrom(int socket, void *mem, size_t len, int flags,
struct musl_sockaddr *from, socklen_t *fromlen)
{
int flgs = 0;
#ifdef ARCH_MM_MMU
int ret = -1;
void *kmem = RT_NULL;
#endif
flgs = netflags_muslc_2_lwip(flags);
#ifdef ARCH_MM_MMU
if (!len)
{
return -EINVAL;
}
if (!lwp_user_accessable((void *)mem, len))
{
return -EFAULT;
}
kmem = kmem_get(len);
if (!kmem)
{
return -ENOMEM;
}
if (flags == 0x2)
{
flags = 0x1;
}
if (from)
{
struct sockaddr sa;
ret = recvfrom(socket, kmem, len, flgs, &sa, fromlen);
sockaddr_tomusl(&sa, from);
}
else
{
ret = recvfrom(socket, kmem, len, flgs, NULL, NULL);
}
if (ret > 0)
{
lwp_put_to_user(mem, kmem, len);
}
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kmem);
return ret;
#else
int ret = -1;
if (from)
{
struct sockaddr sa = {0};
ret = recvfrom(socket, mem, len, flgs, &sa, fromlen);
sockaddr_tomusl(&sa, from);
}
else
{
ret = recvfrom(socket, mem, len, flags, NULL, NULL);
}
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_recv(int socket, void *mem, size_t len, int flags)
{
int flgs = 0;
int ret;
flgs = netflags_muslc_2_lwip(flags);
ret = recvfrom(socket, mem, len, flgs, NULL, NULL);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_sendto(int socket, const void *dataptr, size_t size, int flags,
const struct musl_sockaddr *to, socklen_t tolen)
{
int flgs = 0;
#ifdef ARCH_MM_MMU
int ret = -1;
void *kmem = RT_NULL;
#endif
flgs = netflags_muslc_2_lwip(flags);
#ifdef ARCH_MM_MMU
if (!size)
{
return -EINVAL;
}
if (!lwp_user_accessable((void *)dataptr, size))
{
return -EFAULT;
}
kmem = kmem_get(size);
if (!kmem)
{
return -ENOMEM;
}
lwp_get_from_user(kmem, (void *)dataptr, size);
if (to)
{
struct sockaddr sa;
sockaddr_tolwip(to, &sa);
ret = sendto(socket, kmem, size, flgs, &sa, tolen);
}
else
{
ret = sendto(socket, kmem, size, flgs, NULL, tolen);
}
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kmem);
return ret;
#else
int ret;
if (to)
{
struct sockaddr sa;
sockaddr_tolwip(to, &sa);
ret = sendto(socket, dataptr, size, flgs, &sa, tolen);
}
else
{
ret = sendto(socket, dataptr, size, flgs, NULL, tolen);
}
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_send(int socket, const void *dataptr, size_t size, int flags)
{
int ret;
int flgs = 0;
flgs = netflags_muslc_2_lwip(flags);
ret = sendto(socket, dataptr, size, flgs, NULL, 0);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_socket(int domain, int type, int protocol)
{
int fd = -1;
int nonblock = 0;
/* not support SOCK_CLOEXEC type */
if (type & SOCK_CLOEXEC)
{
type &= ~SOCK_CLOEXEC;
}
if (type & SOCK_NONBLOCK)
{
nonblock = 1;
type &= ~SOCK_NONBLOCK;
}
fd = socket(domain, type, protocol);
if (fd < 0)
{
goto out;
}
if (nonblock)
{
fcntl(fd, F_SETFL, O_NONBLOCK);
}
out:
return (fd < 0 ? GET_ERRNO() : fd);
}
sysret_t sys_closesocket(int socket)
{
return closesocket(socket);
}
#endif
rt_thread_t sys_thread_find(char *name)
{
return rt_thread_find(name);
}
rt_tick_t sys_tick_get(void)
{
return rt_tick_get();
}
rt_err_t sys_thread_mdelay(rt_int32_t ms)
{
return rt_thread_mdelay(ms);
}
struct k_sigaction {
void (*handler)(int);
unsigned long flags;
void (*restorer)(void);
unsigned mask[2];
};
sysret_t sys_sigaction(int sig, const struct k_sigaction *act,
struct k_sigaction *oact, size_t sigsetsize)
{
int ret = -RT_EINVAL;
struct lwp_sigaction kact, *pkact = RT_NULL;
struct lwp_sigaction koact, *pkoact = RT_NULL;
if (!sigsetsize)
{
SET_ERRNO(EINVAL);
goto out;
}
if (sigsetsize > sizeof(lwp_sigset_t))
{
sigsetsize = sizeof(lwp_sigset_t);
}
if (!act && !oact)
{
SET_ERRNO(EINVAL);
goto out;
}
if (oact)
{
if (!lwp_user_accessable((void *)oact, sizeof(*oact)))
{
SET_ERRNO(EFAULT);
goto out;
}
pkoact = &koact;
}
if (act)
{
if (!lwp_user_accessable((void *)act, sizeof(*act)))
{
SET_ERRNO(EFAULT);
goto out;
}
kact.sa_flags = act->flags;
kact.__sa_handler._sa_handler = act->handler;
memcpy(&kact.sa_mask, &act->mask, sigsetsize);
kact.sa_restorer = act->restorer;
pkact = &kact;
}
ret = lwp_sigaction(sig, pkact, pkoact, sigsetsize);
#ifdef ARCH_MM_MMU
if (ret == 0 && oact)
{
lwp_put_to_user(&oact->handler, &pkoact->__sa_handler._sa_handler, sizeof(void (*)(int)));
lwp_put_to_user(&oact->mask, &pkoact->sa_mask, sigsetsize);
lwp_put_to_user(&oact->flags, &pkoact->sa_flags, sizeof(int));
lwp_put_to_user(&oact->restorer, &pkoact->sa_restorer, sizeof(void (*)(void)));
}
#endif /* ARCH_MM_MMU */
out:
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_sigprocmask(int how, const sigset_t *sigset, sigset_t *oset, size_t size)
{
int ret = -1;
lwp_sigset_t *pnewset = RT_NULL, *poldset = RT_NULL;
#ifdef ARCH_MM_MMU
lwp_sigset_t newset, oldset;
#endif /* ARCH_MM_MMU*/
if (!size)
{
return -EINVAL;
}
if (!oset && !sigset)
{
return -EINVAL;
}
if (size > sizeof(lwp_sigset_t))
{
size = sizeof(lwp_sigset_t);
}
if (oset)
{
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)oset, size))
{
return -EFAULT;
}
poldset = &oldset;
#else
if (!lwp_user_accessable((void *)oset, size))
{
return -EFAULT;
}
poldset = (lwp_sigset_t *)oset;
#endif
}
if (sigset)
{
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)sigset, size))
{
return -EFAULT;
}
lwp_get_from_user(&newset, (void *)sigset, size);
pnewset = &newset;
#else
if (!lwp_user_accessable((void *)sigset, size))
{
return -EFAULT;
}
pnewset = (lwp_sigset_t *)sigset;
#endif /* ARCH_MM_MMU */
}
ret = lwp_sigprocmask(how, pnewset, poldset);
#ifdef ARCH_MM_MMU
if (ret < 0)
{
return ret;
}
if (oset)
{
lwp_put_to_user(oset, poldset, size);
}
#endif /* ARCH_MM_MMU */
return (ret < 0 ? -EFAULT: ret);
}
sysret_t sys_tkill(int tid, int sig)
{
#ifdef ARCH_MM_MMU
rt_base_t level;
rt_thread_t thread;
int ret;
level = rt_hw_interrupt_disable();
thread = lwp_tid_get_thread(tid);
ret = lwp_thread_kill(thread, sig);
rt_hw_interrupt_enable(level);
return ret;
#else
return lwp_thread_kill((rt_thread_t)tid, sig);
#endif
}
sysret_t sys_thread_sigprocmask(int how, const lwp_sigset_t *sigset, lwp_sigset_t *oset, size_t size)
{
int ret = -1;
lwp_sigset_t *pnewset = RT_NULL, *poldset = RT_NULL;
#ifdef ARCH_MM_MMU
lwp_sigset_t newset, oldset;
#endif /* ARCH_MM_MMU */
if (!size)
{
return -EINVAL;
}
if (!oset && !sigset)
{
return -EINVAL;
}
if (size != sizeof(lwp_sigset_t))
{
return -EINVAL;
}
if (oset)
{
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)oset, size))
{
return -EFAULT;
}
poldset = &oldset;
#else
if (!lwp_user_accessable((void *)oset, size))
{
return -EFAULT;
}
poldset = oset;
#endif
}
if (sigset)
{
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)sigset, size))
{
return -EFAULT;
}
lwp_get_from_user(&newset, (void *)sigset, sizeof(lwp_sigset_t));
pnewset = &newset;
#else
if (!lwp_user_accessable((void *)sigset, size))
{
return -EFAULT;
}
pnewset = (lwp_sigset_t *)sigset;
#endif
}
ret = lwp_thread_sigprocmask(how, pnewset, poldset);
if (ret < 0)
{
return ret;
}
#ifdef ARCH_MM_MMU
if (oset)
{
lwp_put_to_user(oset, poldset, sizeof(lwp_sigset_t));
}
#endif
return (ret < 0 ? -EFAULT: ret);
}
#ifndef ARCH_MM_MMU
sysret_t sys_lwp_sighandler_set(int sig, lwp_sighandler_t func)
{
if (!lwp_user_accessable((void *)func, sizeof(lwp_sighandler_t)))
{
return -EFAULT;
}
lwp_sighandler_set(sig, func);
return 0;
}
sysret_t sys_thread_sighandler_set(int sig, lwp_sighandler_t func)
{
if (!lwp_user_accessable((void *)func, sizeof(lwp_sighandler_t)))
{
return -EFAULT;
}
lwp_thread_sighandler_set(sig, func);
return 0;
}
#endif /* not defined ARCH_MM_MMU */
sysret_t sys_waitpid(int32_t pid, int *status, int options)
{
int ret = -1;
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)status, sizeof(int)))
{
return -EFAULT;
}
else
{
ret = waitpid(pid, status, options);
}
#else
if (!lwp_user_accessable((void *)status, sizeof(int)))
{
return -EFAULT;
}
ret = waitpid(pid, status, options);
#endif
return ret;
}
#if defined(RT_USING_SAL) && defined(SAL_USING_POSIX)
struct musl_addrinfo
{
int ai_flags;
int ai_family;
int ai_socktype;
int ai_protocol;
socklen_t ai_addrlen;
struct musl_sockaddr *ai_addr;
char *ai_canonname;
struct musl_addrinfo *ai_next;
};
sysret_t sys_getaddrinfo(const char *nodename,
const char *servname,
const struct musl_addrinfo *hints,
struct musl_addrinfo *res)
{
int ret = -1;
struct addrinfo *k_res = NULL;
char *k_nodename = NULL;
char *k_servname = NULL;
struct addrinfo *k_hints = NULL;
#ifdef ARCH_MM_MMU
int err;
#endif
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)res, sizeof(*res)))
{
SET_ERRNO(EFAULT);
goto exit;
}
#endif
if (nodename)
{
#ifdef ARCH_MM_MMU
lwp_user_strlen(nodename, &err);
if (err)
{
SET_ERRNO(EFAULT);
goto exit;
}
#endif
k_nodename = rt_strdup(nodename);
if (!k_nodename)
{
SET_ERRNO(ENOMEM);
goto exit;
}
}
if (servname)
{
#ifdef ARCH_MM_MMU
lwp_user_strlen(servname, &err);
if (err)
{
SET_ERRNO(EFAULT);
goto exit;
}
#endif
k_servname = rt_strdup(servname);
if (!k_servname)
{
SET_ERRNO(ENOMEM);
goto exit;
}
}
if (hints)
{
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)hints, sizeof(*hints)))
{
SET_ERRNO(EFAULT);
goto exit;
}
#endif
k_hints = (struct addrinfo *) rt_malloc(sizeof *hints);
if (!k_hints)
{
SET_ERRNO(ENOMEM);
goto exit;
}
rt_memset(k_hints, 0x0, sizeof(struct addrinfo));
k_hints->ai_flags = hints->ai_flags;
k_hints->ai_family = hints->ai_family;
k_hints->ai_socktype = hints->ai_socktype;
k_hints->ai_protocol = hints->ai_protocol;
k_hints->ai_addrlen = hints->ai_addrlen;
}
ret = sal_getaddrinfo(k_nodename, k_servname, k_hints, &k_res);
if (ret == 0)
{
/* set sockaddr */
sockaddr_tomusl(k_res->ai_addr, res->ai_addr);
res->ai_addrlen = k_res->ai_addrlen;
/* set up addrinfo */
res->ai_family = k_res->ai_family;
res->ai_flags = k_res->ai_flags;
res->ai_next = NULL;
if (hints != NULL)
{
/* copy socktype & protocol from hints if specified */
res->ai_socktype = hints->ai_socktype;
res->ai_protocol = hints->ai_protocol;
}
sal_freeaddrinfo(k_res);
k_res = NULL;
}
exit:
if (ret < 0)
{
ret = GET_ERRNO();
}
if (k_nodename)
{
rt_free(k_nodename);
}
if (k_servname)
{
rt_free(k_servname);
}
if (k_hints)
{
rt_free(k_hints);
}
return ret;
}
#define HOSTENT_BUFSZ 512
sysret_t sys_gethostbyname2_r(const char *name, int af, struct hostent *ret,
char *buf, size_t buflen,
struct hostent **result, int *err)
{
int ret_val = -1;
int sal_ret = -1 , sal_err = -1;
struct hostent sal_he;
struct hostent *sal_result = NULL;
char *sal_buf = NULL;
char *k_name = NULL;
int a_err = 0;
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)err, sizeof(*err)))
{
SET_ERRNO(EFAULT);
goto __exit;
}
if (!lwp_user_accessable((void *)result, sizeof(*result))
|| !lwp_user_accessable((void *)ret, sizeof(*ret))
|| !lwp_user_accessable((void *)buf, buflen))
{
/* not all arguments given */
*err = EFAULT;
SET_ERRNO(EFAULT);
goto __exit;
}
lwp_user_strlen(name, &a_err);
if (a_err)
{
*err = EFAULT;
SET_ERRNO(EFAULT);
goto __exit;
}
#endif
*result = ret;
sal_buf = (char *)malloc(HOSTENT_BUFSZ);
if (sal_buf == NULL)
{
SET_ERRNO(ENOMEM);
goto __exit;
}
k_name = rt_strdup(name);
if (k_name == NULL)
{
SET_ERRNO(ENOMEM);
goto __exit;
}
/* get host by name in SAL */
sal_ret = sal_gethostbyname_r(k_name, &sal_he, sal_buf, HOSTENT_BUFSZ, &sal_result, &sal_err);
if (sal_ret == 0)
{
int index = 0, cnt = 0;
char *ptr = buf;
/* get counter */
index = 0;
while (sal_he.h_addr_list[index] != NULL)
{
index++;
}
cnt = index + 1;
/* update user space hostent */
ret->h_addrtype = sal_he.h_addrtype;
ret->h_length = sal_he.h_length;
rt_strncpy(ptr, k_name, buflen - (ptr - buf));
ret->h_name = ptr;
ptr += rt_strlen(k_name);
ret->h_addr_list = (char**)ptr;
ptr += cnt * sizeof(char *);
index = 0;
while (sal_he.h_addr_list[index] != NULL)
{
ret->h_addr_list[index] = ptr;
rt_memcpy(ptr, sal_he.h_addr_list[index], sal_he.h_length);
ptr += sal_he.h_length;
index++;
}
ret->h_addr_list[index] = NULL;
}
ret_val = 0;
__exit:
if (ret_val < 0)
{
ret_val = GET_ERRNO();
}
/* release buffer */
if (sal_buf)
{
free(sal_buf);
}
if (k_name)
{
free(k_name);
}
return ret_val;
}
#endif
char *sys_getcwd(char *buf, size_t size)
{
if (!lwp_user_accessable((void *)buf, size))
{
return RT_NULL;
}
getcwd(buf, size);
return (char *)strlen(buf);
}
sysret_t sys_chdir(const char *path)
{
#ifdef ARCH_MM_MMU
int err = 0;
lwp_user_strlen(path, &err);
if (err)
{
return -EFAULT;
}
err = chdir(path);
return (err < 0 ? GET_ERRNO() : err);
#else
int ret = chdir(path);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_mkdir(const char *path, mode_t mode)
{
#ifdef ARCH_MM_MMU
int err = 0;
lwp_user_strlen(path, &err);
if (err)
{
return -EFAULT;
}
err = mkdir(path, mode);
return (err < 0 ? GET_ERRNO() : err);
#else
int ret = mkdir(path, mode);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_rmdir(const char *path)
{
#ifdef ARCH_MM_MMU
int err = 0;
lwp_user_strlen(path, &err);
if (err)
{
return -EFAULT;
}
err = unlink(path);
return (err < 0 ? GET_ERRNO() : err);
#else
int ret = unlink(path);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
#ifdef RT_USING_MUSL
typedef uint64_t ino_t;
#endif
struct libc_dirent {
ino_t d_ino;
off_t d_off;
unsigned short d_reclen;
unsigned char d_type;
char d_name[256];
};
sysret_t sys_getdents(int fd, struct libc_dirent *dirp, size_t nbytes)
{
int ret = -1;
struct dfs_fd *dfs_fd;
size_t cnt = (nbytes / sizeof(struct libc_dirent));
size_t rtt_nbytes = 0;
struct dirent *rtt_dirp;
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)dirp, sizeof(struct libc_dirent)))
{
return -EFAULT;
}
#endif
if (cnt == 0)
{
return -EINVAL;
}
rtt_nbytes = cnt * sizeof(struct dirent);
rtt_dirp = (struct dirent *)rt_malloc(rtt_nbytes);
if (!rtt_dirp)
{
return -ENOMEM;
}
dfs_fd = fd_get(fd);
ret = dfs_file_getdents(dfs_fd, rtt_dirp, rtt_nbytes);
if (ret > 0)
{
size_t i = 0;
cnt = ret / sizeof(struct dirent);
for (i = 0; i < cnt; i++)
{
dirp[i].d_ino = 0;
dirp[i].d_off = i*sizeof(struct libc_dirent);
dirp[i].d_type = rtt_dirp[i].d_type;
dirp[i].d_reclen = sizeof(struct libc_dirent);
strcpy(dirp[i].d_name, rtt_dirp[i].d_name);
}
ret = cnt * sizeof(struct libc_dirent);
}
if (ret < 0)
{
ret = GET_ERRNO();
}
rt_free(rtt_dirp);
return ret;
}
rt_err_t sys_get_errno(void)
{
return rt_get_errno();
}
#ifdef ARCH_MM_MMU
sysret_t sys_set_thread_area(void *p)
{
rt_thread_t thread;
thread = rt_thread_self();
thread->thread_idr = p;
arch_set_thread_area(p);
return 0;
}
sysret_t sys_set_tid_address(int *tidptr)
{
rt_thread_t thread;
#ifdef ARCH_MM_MMU
if (!lwp_user_accessable((void *)tidptr, sizeof(int)))
{
return -EFAULT;
}
#endif
thread = rt_thread_self();
thread->clear_child_tid = tidptr;
return thread->tid;
}
#endif /* ARCH_MM_MMU */
sysret_t sys_gettid(void)
{
return rt_thread_self()->tid;
}
sysret_t sys_access(const char *filename, int mode)
{
int ret = 0;
#ifdef ARCH_MM_MMU
rt_size_t len = 0;
char *kname = RT_NULL;
int a_err = 0;
lwp_user_strlen(filename, &a_err);
if (a_err)
{
return -EFAULT;
}
len = rt_strlen(filename);
if (!len)
{
return -EINVAL;
}
kname = (char *)kmem_get(len + 1);
if (!kname)
{
return -ENOMEM;
}
lwp_get_from_user(kname, (void *)filename, len + 1);
ret = access(kname, mode);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kname);
return ret;
#else
ret = access(filename, mode);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_pipe(int fd[2])
{
int ret;
if (!lwp_user_accessable((void *)fd, sizeof(int[2])))
{
return -EFAULT;
}
ret = pipe(fd);
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_clock_settime(clockid_t clk, const struct timespec *ts)
{
int ret = 0;
#ifdef ARCH_MM_MMU
size_t size = sizeof(struct timespec);
struct timespec *kts = NULL;
if (!lwp_user_accessable((void *)ts, size))
{
return -EFAULT;
}
kts = kmem_get(size);
if (!kts)
{
return -ENOMEM;
}
lwp_get_from_user(kts, (void *)ts, size);
ret = clock_settime(clk, kts);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kts);
return ret;
#else
if (!lwp_user_accessable((void *)ts, sizeof(struct timespec)))
{
return -EFAULT;
}
ret = clock_settime(clk, ts);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_clock_gettime(clockid_t clk, struct timespec *ts)
{
int ret = 0;
#ifdef ARCH_MM_MMU
size_t size = sizeof(struct timespec);
struct timespec *kts = NULL;
if (!lwp_user_accessable((void *)ts, size))
{
return -EFAULT;
}
kts = kmem_get(size);
if (!kts)
{
return -ENOMEM;
}
ret = clock_gettime(clk, kts);
if (ret != -1)
lwp_put_to_user(ts, kts, size);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kts);
return ret;
#else
if (!lwp_user_accessable((void *)ts, sizeof(struct timespec)))
{
return -EFAULT;
}
ret = clock_gettime(clk, ts);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_clock_nanosleep(clockid_t clk, int flags, const struct timespec *rqtp, struct timespec *rmtp)
{
int ret = 0;
dbg_log(DBG_LOG, "sys_nanosleep\n");
if (!lwp_user_accessable((void *)rqtp, sizeof *rqtp))
return -EFAULT;
#ifdef ARCH_MM_MMU
struct timespec rqtp_k;
struct timespec rmtp_k;
lwp_get_from_user(&rqtp_k, (void *)rqtp, sizeof rqtp_k);
ret = clock_nanosleep(clk, flags, &rqtp_k, &rmtp_k);
if ((ret != -1 || rt_get_errno() == EINTR) && rmtp && lwp_user_accessable((void *)rmtp, sizeof *rmtp))
{
lwp_put_to_user(rmtp, (void *)&rmtp_k, sizeof rmtp_k);
if(ret != 0)
return -EINTR;
}
#else
if (rmtp)
{
if (!lwp_user_accessable((void *)rmtp, sizeof *rmtp))
return -EFAULT;
ret = clock_nanosleep(clk, flags, rqtp, rmtp);
}
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_clock_getres(clockid_t clk, struct timespec *ts)
{
int ret = 0;
#ifdef ARCH_MM_MMU
struct timespec kts;
size_t size = sizeof(struct timespec);
if (!lwp_user_accessable((void *)ts, size))
{
return -EFAULT;
}
ret = clock_getres(clk, &kts);
if (ret != -1)
lwp_put_to_user(ts, &kts, size);
#else
if (!lwp_user_accessable((void *)ts, sizeof(struct timespec)))
{
return -EFAULT;
}
ret = clock_getres(clk, ts);
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_rename(const char *oldpath, const char *newpath)
{
int ret = -1;
#ifdef ARCH_MM_MMU
int err;
lwp_user_strlen(oldpath, &err);
if (err)
{
return -EFAULT;
}
lwp_user_strlen(newpath, &err);
if (err)
{
return -EFAULT;
}
#endif
ret = rename(oldpath, newpath);
return (ret < 0 ? GET_ERRNO() : ret);
}
typedef unsigned long long rlim_t;
struct rlimit {
rlim_t rlim_cur;
rlim_t rlim_max;
};
#define RLIMIT_CPU 0
#define RLIMIT_FSIZE 1
#define RLIMIT_DATA 2
#define RLIMIT_STACK 3
#define RLIMIT_CORE 4
#define RLIMIT_RSS 5
#define RLIMIT_NPROC 6
#define RLIMIT_NOFILE 7
#define RLIMIT_MEMLOCK 8
#define RLIMIT_AS 9
sysret_t sys_prlimit64(pid_t pid,
unsigned int resource,
const struct rlimit *new_rlim,
struct rlimit *old_rlim)
{
return -ENOSYS;
}
sysret_t sys_getrlimit(unsigned int resource, unsigned long rlim[2])
{
int ret = -1;
if (!lwp_user_accessable((void *)rlim, sizeof(unsigned long [2])))
{
return -EFAULT;
}
switch (resource)
{
case RLIMIT_NOFILE:
{
struct dfs_fdtable *fdt = dfs_fdtable_get();
dfs_fd_lock();
rlim[0] = fdt->maxfd;
dfs_fd_unlock();
rlim[1] = DFS_FD_MAX;
ret = 0;
}
break;
default:
return -EINVAL;
break;
}
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_setrlimit(unsigned int resource, struct rlimit *rlim)
{
return -ENOSYS;
}
sysret_t sys_setsid(void)
{
int ret = 0;
ret = setsid();
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_getrandom(void *buf, size_t buflen, unsigned int flags)
{
int ret = -1;
int count = 0;
void *kmem = RT_NULL;
rt_device_t rd_dev = RT_NULL;
if (flags & GRND_RANDOM)
rd_dev = rt_device_find("random");
else
rd_dev = rt_device_find("urandom");
if (rd_dev == RT_NULL)
{
return -EFAULT;
}
if (rt_device_open(rd_dev, RT_DEVICE_OFLAG_RDONLY) != RT_EOK)
{
return -EFAULT;
}
if (!lwp_user_accessable(buf, buflen))
{
rt_device_close(rd_dev);
return -EFAULT;
}
#ifdef ARCH_MM_MMU
kmem = kmem_get(buflen);
if (!kmem)
{
rt_device_close(rd_dev);
return -ENOMEM;
}
while (count < buflen)
{
ret = rt_device_read(rd_dev, count, (char *)kmem + count, buflen - count);
if (ret <= 0)
break;
count += ret;
}
rt_device_close(rd_dev);
ret = count;
if (count > 0)
{
ret = lwp_put_to_user(buf, kmem, count);
}
kmem_put(kmem);
#else
while (count < buflen)
{
ret = rt_device_read(rd_dev, count, (char *)kmem + count, buflen - count);
if (ret <= 0)
break;
count += ret;
}
rt_device_close(rd_dev);
ret = count;
#endif
return ret;
}
ssize_t sys_readlink(char* path, char *buf, size_t bufsz)
{
size_t len, copy_len;
int err;
int fd = -1;
struct dfs_fd *d;
char *copy_path;
len = lwp_user_strlen(path, &err);
if (err)
{
return -EFAULT;
}
if (!lwp_user_accessable(buf, bufsz))
{
return -EINVAL;
}
copy_path = (char*)rt_malloc(len + 1);
if (!copy_path)
{
return -ENOMEM;
}
copy_len = lwp_get_from_user(copy_path, path, len);
copy_path[copy_len] = '\0';
/* musl __procfdname */
err = sscanf(copy_path, "/proc/self/fd/%d", &fd);
rt_free(copy_path);
if (err != 1)
{
LOG_E("readlink: path not is /proc/self/fd/* , call by musl __procfdname()?");
return -EINVAL;
}
d = fd_get(fd);
if (!d)
{
return -EBADF;
}
if (!d->vnode)
{
return -EBADF;
}
copy_len = strlen(d->vnode->fullpath);
if (copy_len > bufsz)
{
copy_len = bufsz;
}
bufsz = lwp_put_to_user(buf, d->vnode->fullpath, copy_len);
return bufsz;
}
sysret_t sys_setaffinity(pid_t pid, size_t size, void *set)
{
if (!lwp_user_accessable(set, sizeof(cpu_set_t)))
{
return -EFAULT;
}
for (int i = 0;i < size * 8; i++)
{
if (CPU_ISSET(i, (cpu_set_t *)set))
{
return lwp_setaffinity(pid, i);
}
}
return -1;
}
sysret_t sys_sched_setparam(pid_t pid, void *param)
{
struct sched_param *sched_param = (struct sched_param *)param;
struct rt_lwp *lwp = NULL;
rt_thread_t main_thread;
int ret = -1;
if (!lwp_user_accessable(param, sizeof(struct sched_param)))
{
return -EFAULT;
}
if (pid > 0)
{
lwp = lwp_from_pid(pid);
}
else if (pid == 0)
{
lwp = lwp_self();
}
if (lwp)
{
main_thread = rt_list_entry(lwp->t_grp.prev, struct rt_thread, sibling);
return rt_thread_control(main_thread, RT_THREAD_CTRL_CHANGE_PRIORITY, (void *)&sched_param->sched_priority);
}
return ret;
}
sysret_t sys_sched_getparam(pid_t pid, void *param)
{
struct sched_param *sched_param = (struct sched_param *)param;
struct rt_lwp *lwp = NULL;
rt_thread_t main_thread;
int ret = -1;
if (!lwp_user_accessable(param, sizeof(struct sched_param)))
{
return -EFAULT;
}
if (pid > 0)
{
lwp = lwp_from_pid(pid);
}
else if (pid == 0)
{
lwp = lwp_self();
}
if (lwp)
{
main_thread = rt_list_entry(lwp->t_grp.prev, struct rt_thread, sibling);
sched_param->sched_priority = main_thread->current_priority;
ret = 0;
}
return ret;
}
sysret_t sys_sched_get_priority_max(int policy)
{
if(policy < 0)
{
SET_ERRNO(EINVAL);
return -rt_get_errno();
}
return RT_THREAD_PRIORITY_MAX;
}
sysret_t sys_sched_get_priority_min(int policy)
{
if(policy < 0)
{
SET_ERRNO(EINVAL);
return -rt_get_errno();
}
return 0;
}
sysret_t sys_sched_setscheduler(int tid, int policy, void *param)
{
struct sched_param *sched_param = (struct sched_param *)param;
rt_thread_t thread = lwp_tid_get_thread(tid);
if (!lwp_user_accessable(param, sizeof(struct sched_param)))
{
return -EFAULT;
}
return rt_thread_control(thread, RT_THREAD_CTRL_CHANGE_PRIORITY, (void *)&sched_param->sched_priority);
return 0;
}
sysret_t sys_sched_getscheduler(int tid, int *policy, void *param)
{
struct sched_param *sched_param = (struct sched_param *)param;
rt_thread_t thread = lwp_tid_get_thread(tid);
if (!lwp_user_accessable(sched_param, sizeof(struct sched_param)))
{
return -EFAULT;
}
sched_param->sched_priority = thread->current_priority;
*policy = 0;
return 0;
}
sysret_t sys_fsync(int fd)
{
int res = fsync(fd);
if (res < 0)
res = rt_get_errno();
return res;
}
mqd_t sys_mq_open(const char *name, int flags, mode_t mode, struct mq_attr *attr)
{
mqd_t mqdes;
rt_err_t ret = 0;
#ifdef ARCH_MM_MMU
char *kname = RT_NULL;
int a_err = 0;
rt_size_t len = 0;
struct mq_attr attr_k;
lwp_user_strlen(name, &a_err);
if (a_err)
return (mqd_t)-EFAULT;
len = rt_strlen(name);
if (!len)
return (mqd_t)-EINVAL;
kname = (char *)kmem_get(len + 1);
if (!kname)
return (mqd_t)-ENOMEM;
lwp_get_from_user(&attr_k, (void *)attr, sizeof(struct mq_attr));
lwp_get_from_user(kname, (void *)name, len + 1);
mqdes = mq_open(kname, flags, mode, &attr_k);
if (mqdes == RT_NULL)
{
ret = GET_ERRNO();
}
lwp_put_to_user(attr, &attr_k, sizeof(struct mq_attr));
kmem_put(kname);
#else
mqdes = mq_open(name, flags, mode, attr);
#endif
if (mqdes == RT_NULL)
return (mqd_t)ret;
else
return mqdes;
}
sysret_t sys_mq_unlink(const char *name)
{
int ret = 0;
#ifdef ARCH_MM_MMU
char *kname = RT_NULL;
int a_err = 0;
rt_size_t len = 0;
lwp_user_strlen(name, &a_err);
if (a_err)
return -EFAULT;
len = rt_strlen(name);
if (!len)
return -EINVAL;
kname = (char *)kmem_get(len + 1);
if (!kname)
return -ENOMEM;
lwp_get_from_user(kname, (void *)name, len + 1);
ret = mq_unlink(kname);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kname);
return ret;
#else
ret = mq_unlink(name);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_mq_timedsend(mqd_t mqd, const char *msg, size_t len, unsigned prio, const struct timespec *at)
{
int ret = 0;
#ifdef ARCH_MM_MMU
char *kmsg = RT_NULL;
int a_err = 0;
struct timespec at_k;
lwp_user_strlen(msg, &a_err);
if (a_err)
return -EFAULT;
kmsg = (char *)kmem_get(len + 1);
if (!kmsg)
return -ENOMEM;
lwp_get_from_user(&at_k, (void *)at, sizeof(struct timespec));
lwp_get_from_user(kmsg, (void *)msg, len + 1);
ret = mq_timedsend(mqd, kmsg, len, prio, &at_k);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kmsg);
return ret;
#else
ret = mq_timedsend(mqd, msg, len, prio, at);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_mq_timedreceive(mqd_t mqd, char *restrict msg, size_t len, unsigned *restrict prio, const struct timespec *restrict at)
{
int ret = 0;
#ifdef ARCH_MM_MMU
char *restrict kmsg = RT_NULL;
int a_err = 0;
struct timespec at_k;
lwp_user_strlen(msg, &a_err);
if (a_err)
return -EFAULT;
kmsg = (char *restrict)kmem_get(len + 1);
if (!kmsg)
return -ENOMEM;
lwp_get_from_user(&at_k, (void *)at, sizeof(struct timespec));
lwp_get_from_user(kmsg, (void *)msg, len + 1);
ret = mq_timedreceive(mqd, kmsg, len, prio, &at_k);
if (ret > 0)
lwp_put_to_user(msg, kmsg, len + 1);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kmsg);
return ret;
#else
ret = mq_timedreceive(mqd, msg, len, prio, at);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_mq_notify(mqd_t mqd, const struct sigevent *sev)
{
int ret = 0;
#ifdef ARCH_MM_MMU
struct sigevent sev_k;
lwp_get_from_user(&sev_k, (void *)sev, sizeof(struct timespec));
ret = mq_notify(mqd, &sev_k);
#else
ret = mq_notify(mqd, sev);
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
sysret_t sys_mq_getsetattr(mqd_t mqd, const struct mq_attr *restrict new, struct mq_attr *restrict old)
{
int ret = 0;
#ifdef ARCH_MM_MMU
size_t size = sizeof(struct mq_attr);
struct mq_attr *restrict knew = NULL;
struct mq_attr *restrict kold = NULL;
if (new != RT_NULL)
{
if (!lwp_user_accessable((void *)new, size))
return -EFAULT;
knew = kmem_get(size);
if (!knew)
return -ENOMEM;
lwp_get_from_user(knew, (void *)new, size);
}
if (!lwp_user_accessable((void *)old, size))
return -EFAULT;
kold = kmem_get(size);
if (!kold)
return -ENOMEM;
lwp_get_from_user(kold, (void *)old, size);
ret = mq_setattr(mqd, knew, kold);
if (ret != -1)
lwp_put_to_user(old, kold, size);
if (ret < 0)
{
ret = GET_ERRNO();
}
kmem_put(kold);
if (new != RT_NULL)
kmem_put(knew);
return ret;
#else
ret = mq_setattr(mqd, new, old);
return (ret < 0 ? GET_ERRNO() : ret);
#endif
}
sysret_t sys_mq_close(mqd_t mqd)
{
int ret = 0;
#ifdef ARCH_MM_MMU
ret = mq_close(mqd);
#else
ret = mq_close(mqd);
#endif
return (ret < 0 ? GET_ERRNO() : ret);
}
#define ICACHE (1<<0)
#define DCACHE (1<<1)
#define BCACHE (ICACHE|DCACHE)
rt_weak sysret_t sys_cacheflush(void *addr, int size, int cache)
{
if (addr < addr + size &&
(size_t)addr >= USER_VADDR_START &&
(size_t)addr + size < USER_VADDR_TOP)
{
if ((cache & DCACHE))
{
rt_hw_cpu_dcache_clean_and_invalidate(addr, size);
}
if ((cache & ICACHE))
{
rt_hw_cpu_icache_invalidate(addr, size);
}
return 0;
}
return -EFAULT;
}
const static struct rt_syscall_def func_table[] =
{
SYSCALL_SIGN(sys_exit), /* 01 */
SYSCALL_SIGN(sys_read),
SYSCALL_SIGN(sys_write),
SYSCALL_SIGN(sys_lseek),
SYSCALL_SIGN(sys_open), /* 05 */
SYSCALL_SIGN(sys_close),
SYSCALL_SIGN(sys_ioctl),
SYSCALL_SIGN(sys_fstat),
SYSCALL_SIGN(sys_poll),
SYSCALL_SIGN(sys_nanosleep), /* 10 */
SYSCALL_SIGN(sys_gettimeofday),
SYSCALL_SIGN(sys_settimeofday),
SYSCALL_SIGN(sys_exec),
SYSCALL_SIGN(sys_kill),
SYSCALL_SIGN(sys_getpid), /* 15 */
SYSCALL_SIGN(sys_getpriority),
SYSCALL_SIGN(sys_setpriority),
SYSCALL_SIGN(sys_sem_create),
SYSCALL_SIGN(sys_sem_delete),
SYSCALL_SIGN(sys_sem_take), /* 20 */
SYSCALL_SIGN(sys_sem_release),
SYSCALL_SIGN(sys_mutex_create),
SYSCALL_SIGN(sys_mutex_delete),
SYSCALL_SIGN(sys_mutex_take),
SYSCALL_SIGN(sys_mutex_release), /* 25 */
SYSCALL_SIGN(sys_event_create),
SYSCALL_SIGN(sys_event_delete),
SYSCALL_SIGN(sys_event_send),
SYSCALL_SIGN(sys_event_recv),
SYSCALL_SIGN(sys_mb_create), /* 30 */
SYSCALL_SIGN(sys_mb_delete),
SYSCALL_SIGN(sys_mb_send),
SYSCALL_SIGN(sys_mb_send_wait),
SYSCALL_SIGN(sys_mb_recv),
SYSCALL_SIGN(sys_mq_create), /* 35 */
SYSCALL_SIGN(sys_mq_delete),
SYSCALL_SIGN(sys_mq_send),
SYSCALL_SIGN(sys_mq_urgent),
SYSCALL_SIGN(sys_mq_recv),
SYSCALL_SIGN(sys_thread_create), /* 40 */
SYSCALL_SIGN(sys_thread_delete),
SYSCALL_SIGN(sys_thread_startup),
SYSCALL_SIGN(sys_thread_self),
SYSCALL_SIGN(sys_channel_open),
SYSCALL_SIGN(sys_channel_close), /* 45 */
SYSCALL_SIGN(sys_channel_send),
SYSCALL_SIGN(sys_channel_send_recv_timeout),
SYSCALL_SIGN(sys_channel_reply),
SYSCALL_SIGN(sys_channel_recv_timeout),
SYSCALL_SIGN(sys_enter_critical), /* 50 */
SYSCALL_SIGN(sys_exit_critical),
SYSCALL_USPACE(SYSCALL_SIGN(sys_brk)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_mmap2)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_munmap)),
#ifdef ARCH_MM_MMU
SYSCALL_USPACE(SYSCALL_SIGN(sys_shmget)), /* 55 */
SYSCALL_USPACE(SYSCALL_SIGN(sys_shmrm)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_shmat)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_shmdt)),
#else
#ifdef RT_LWP_USING_SHM
SYSCALL_SIGN(sys_shm_alloc), /* 55 */
SYSCALL_SIGN(sys_shm_free),
SYSCALL_SIGN(sys_shm_retain),
SYSCALL_SIGN(sys_notimpl),
#else
SYSCALL_SIGN(sys_notimpl), /* 55 */
SYSCALL_SIGN(sys_notimpl),
SYSCALL_SIGN(sys_notimpl),
SYSCALL_SIGN(sys_notimpl),
#endif /* RT_LWP_USING_SHM */
#endif /* ARCH_MM_MMU */
SYSCALL_SIGN(sys_device_init),
SYSCALL_SIGN(sys_device_register), /* 60 */
SYSCALL_SIGN(sys_device_control),
SYSCALL_SIGN(sys_device_find),
SYSCALL_SIGN(sys_device_open),
SYSCALL_SIGN(sys_device_close),
SYSCALL_SIGN(sys_device_read), /* 65 */
SYSCALL_SIGN(sys_device_write),
SYSCALL_SIGN(sys_stat),
SYSCALL_SIGN(sys_thread_find),
SYSCALL_NET(SYSCALL_SIGN(sys_accept)),
SYSCALL_NET(SYSCALL_SIGN(sys_bind)), /* 70 */
SYSCALL_NET(SYSCALL_SIGN(sys_shutdown)),
SYSCALL_NET(SYSCALL_SIGN(sys_getpeername)),
SYSCALL_NET(SYSCALL_SIGN(sys_getsockname)),
SYSCALL_NET(SYSCALL_SIGN(sys_getsockopt)),
SYSCALL_NET(SYSCALL_SIGN(sys_setsockopt)), /* 75 */
SYSCALL_NET(SYSCALL_SIGN(sys_connect)),
SYSCALL_NET(SYSCALL_SIGN(sys_listen)),
SYSCALL_NET(SYSCALL_SIGN(sys_recv)),
SYSCALL_NET(SYSCALL_SIGN(sys_recvfrom)),
SYSCALL_NET(SYSCALL_SIGN(sys_send)), /* 80 */
SYSCALL_NET(SYSCALL_SIGN(sys_sendto)),
SYSCALL_NET(SYSCALL_SIGN(sys_socket)),
SYSCALL_NET(SYSCALL_SIGN(sys_closesocket)),
SYSCALL_NET(SYSCALL_SIGN(sys_getaddrinfo)),
SYSCALL_NET(SYSCALL_SIGN(sys_gethostbyname2_r)), /* 85 */
SYSCALL_SIGN(sys_notimpl), //network,
SYSCALL_SIGN(sys_notimpl), //network,
SYSCALL_SIGN(sys_notimpl), //network,
SYSCALL_SIGN(sys_notimpl), //network,
SYSCALL_SIGN(sys_notimpl), //network, /* 90 */
SYSCALL_SIGN(sys_notimpl), //network,
SYSCALL_SIGN(sys_notimpl), //network,
SYSCALL_SIGN(sys_notimpl), //network,
#ifdef RT_USING_DFS
SYSCALL_SIGN(sys_select),
#else
SYSCALL_SIGN(sys_notimpl),
#endif
SYSCALL_SIGN(sys_notimpl), //SYSCALL_SIGN(sys_hw_interrupt_disable), /* 95 */
SYSCALL_SIGN(sys_notimpl), //SYSCALL_SIGN(sys_hw_interrupt_enable),
SYSCALL_SIGN(sys_tick_get),
SYSCALL_SIGN(sys_exit_group),
SYSCALL_SIGN(sys_notimpl), //rt_delayed_work_init,
SYSCALL_SIGN(sys_notimpl), //rt_work_submit, /* 100 */
SYSCALL_SIGN(sys_notimpl), //rt_wqueue_wakeup,
SYSCALL_SIGN(sys_thread_mdelay),
SYSCALL_SIGN(sys_sigaction),
SYSCALL_SIGN(sys_sigprocmask),
SYSCALL_SIGN(sys_tkill), /* 105 */
SYSCALL_SIGN(sys_thread_sigprocmask),
#ifdef ARCH_MM_MMU
SYSCALL_SIGN(sys_cacheflush),
SYSCALL_SIGN(sys_notimpl),
SYSCALL_SIGN(sys_notimpl),
#else
SYSCALL_SIGN(sys_notimpl),
SYSCALL_SIGN(sys_lwp_sighandler_set),
SYSCALL_SIGN(sys_thread_sighandler_set),
#endif
SYSCALL_SIGN(sys_waitpid), /* 110 */
SYSCALL_SIGN(sys_rt_timer_create),
SYSCALL_SIGN(sys_rt_timer_delete),
SYSCALL_SIGN(sys_rt_timer_start),
SYSCALL_SIGN(sys_rt_timer_stop),
SYSCALL_SIGN(sys_rt_timer_control), /* 115 */
SYSCALL_SIGN(sys_getcwd),
SYSCALL_SIGN(sys_chdir),
SYSCALL_SIGN(sys_unlink),
SYSCALL_SIGN(sys_mkdir),
SYSCALL_SIGN(sys_rmdir), /* 120 */
SYSCALL_SIGN(sys_getdents),
SYSCALL_SIGN(sys_get_errno),
#ifdef ARCH_MM_MMU
SYSCALL_SIGN(sys_set_thread_area),
SYSCALL_SIGN(sys_set_tid_address),
#else
SYSCALL_SIGN(sys_notimpl),
SYSCALL_SIGN(sys_notimpl),
#endif
SYSCALL_SIGN(sys_access), /* 125 */
SYSCALL_SIGN(sys_pipe),
SYSCALL_SIGN(sys_clock_settime),
SYSCALL_SIGN(sys_clock_gettime),
SYSCALL_SIGN(sys_clock_getres),
SYSCALL_USPACE(SYSCALL_SIGN(sys_clone)), /* 130 */
SYSCALL_USPACE(SYSCALL_SIGN(sys_futex)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_pmutex)),
SYSCALL_SIGN(sys_dup),
SYSCALL_SIGN(sys_dup2),
SYSCALL_SIGN(sys_rename), /* 135 */
SYSCALL_USPACE(SYSCALL_SIGN(sys_fork)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_execve)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_vfork)),
SYSCALL_SIGN(sys_gettid),
SYSCALL_SIGN(sys_prlimit64), /* 140 */
SYSCALL_SIGN(sys_getrlimit),
SYSCALL_SIGN(sys_setrlimit),
SYSCALL_SIGN(sys_setsid),
SYSCALL_SIGN(sys_getrandom),
SYSCALL_SIGN(sys_readlink), // SYSCALL_SIGN(sys_readlink) /* 145 */
SYSCALL_USPACE(SYSCALL_SIGN(sys_mremap)),
SYSCALL_USPACE(SYSCALL_SIGN(sys_madvise)),
SYSCALL_SIGN(sys_sched_setparam),
SYSCALL_SIGN(sys_sched_getparam),
SYSCALL_SIGN(sys_sched_get_priority_max), /* 150 */
SYSCALL_SIGN(sys_sched_get_priority_min),
SYSCALL_SIGN(sys_sched_setscheduler),
SYSCALL_SIGN(sys_sched_getscheduler),
SYSCALL_SIGN(sys_setaffinity),
SYSCALL_SIGN(sys_fsync), /* 155 */
SYSCALL_SIGN(sys_clock_nanosleep),
SYSCALL_SIGN(sys_timer_create),
SYSCALL_SIGN(sys_timer_delete),
SYSCALL_SIGN(sys_timer_settime),
SYSCALL_SIGN(sys_timer_gettime), /* 160 */
SYSCALL_SIGN(sys_timer_getoverrun),
SYSCALL_SIGN(sys_mq_open),
SYSCALL_SIGN(sys_mq_unlink),
SYSCALL_SIGN(sys_mq_timedsend),
SYSCALL_SIGN(sys_mq_timedreceive),
SYSCALL_SIGN(sys_mq_notify),
SYSCALL_SIGN(sys_mq_getsetattr),
SYSCALL_SIGN(sys_mq_close),
};
const void *lwp_get_sys_api(rt_uint32_t number)
{
const void *func = (const void *)sys_notimpl;
if (number == 0xff)
{
func = (void *)sys_log;
}
else
{
number -= 1;
if (number < sizeof(func_table) / sizeof(func_table[0]))
{
func = func_table[number].func;
}
}
return func;
}
const char *lwp_get_syscall_name(rt_uint32_t number)
{
const char *name = "sys_notimpl";
if (number == 0xff)
{
name = "sys_log";
}
else
{
number -= 1;
if (number < sizeof(func_table) / sizeof(func_table[0]))
{
name = (char*)func_table[number].name;
}
}
// skip sys_
return name;
}