Closing libUV Handles Correctly - c

I'm trying to find out how to fix these memory leaks I'm getting while running this program with Valgrind. The leaks occur with the two allocations in nShell_client_main. But I'm not
sure how to properly free them.
I've tried freeing them at nShell_Connect, but it's causing libUV to abort the program. I've tried freeing them at the end of nShell_client_main, but then I get read/write errors when closing the loop. Does anyone know how I'm supposed to close these handles? I've read this, which got me started. But, it seams out-dated because uv_ip4_addr has a different prototype in the latest version.
(nShell_main is the "entry" point)
#include "nPort.h"
#include "nShell-main.h"
void nShell_Close(
uv_handle_t * term_handle
){
}
void nShell_Connect(uv_connect_t * term_handle, int status){
uv_close((uv_handle_t *) term_handle, 0);
}
nError * nShell_client_main(nShell * n_shell, uv_loop_t * n_shell_loop){
int uv_error = 0;
nError * n_error = 0;
uv_tcp_t * n_shell_socket = 0;
uv_connect_t * n_shell_connect = 0;
struct sockaddr_in dest_addr;
n_shell_socket = malloc(sizeof(uv_tcp_t));
if (!n_shell_socket){
// handle error
}
uv_error = uv_tcp_init(n_shell_loop, n_shell_socket);
if (uv_error){
// handle error
}
uv_error = uv_ip4_addr("127.0.0.1", NPORT, &dest_addr);
if (uv_error){
// handle error
}
n_shell_connect = malloc(sizeof(uv_connect_t));
if (!n_shell_connect){
// handle error
}
uv_error = uv_tcp_connect(n_shell_connect, n_shell_socket, (struct sockaddr *) &dest_addr, nShell_Connect);
if (uv_error){
// handle error
}
uv_error = uv_run(n_shell_loop, UV_RUN_DEFAULT);
if (uv_error){
// handle error
}
return 0;
}
nError * nShell_loop_main(nShell * n_shell){
int uv_error = 0;
nError * n_error = 0;
uv_loop_t * n_shell_loop = 0;
n_shell_loop = malloc(sizeof(uv_loop_t));
if (!n_shell_loop){
// handle error
}
uv_error = uv_loop_init(n_shell_loop);
if (uv_error){
// handle error
}
n_error = nShell_client_main(n_shell, n_shell_loop);
if (n_error){
// handle error
}
uv_loop_close(n_shell_loop);
free(n_shell_loop);
return 0;
}
The assertion is happening at the end of the switch statement in this excerpt of code (taken from Joyent's libUV page on Github):
void uv_close(uv_handle_t* handle, uv_close_cb close_cb) {
assert(!(handle->flags & (UV_CLOSING | UV_CLOSED)));
handle->flags |= UV_CLOSING;
handle->close_cb = close_cb;
switch (handle->type) {
case UV_NAMED_PIPE:
uv__pipe_close((uv_pipe_t*)handle);
break;
case UV_TTY:
uv__stream_close((uv_stream_t*)handle);
break;
case UV_TCP:
uv__tcp_close((uv_tcp_t*)handle);
break;
case UV_UDP:
uv__udp_close((uv_udp_t*)handle);
break;
case UV_PREPARE:
uv__prepare_close((uv_prepare_t*)handle);
break;
case UV_CHECK:
uv__check_close((uv_check_t*)handle);
break;
case UV_IDLE:
uv__idle_close((uv_idle_t*)handle);
break;
case UV_ASYNC:
uv__async_close((uv_async_t*)handle);
break;
case UV_TIMER:
uv__timer_close((uv_timer_t*)handle);
break;
case UV_PROCESS:
uv__process_close((uv_process_t*)handle);
break;
case UV_FS_EVENT:
uv__fs_event_close((uv_fs_event_t*)handle);
break;
case UV_POLL:
uv__poll_close((uv_poll_t*)handle);
break;
case UV_FS_POLL:
uv__fs_poll_close((uv_fs_poll_t*)handle);
break;
case UV_SIGNAL:
uv__signal_close((uv_signal_t*) handle);
/* Signal handles may not be closed immediately. The signal code will */
/* itself close uv__make_close_pending whenever appropriate. */
return;
default:
assert(0); // assertion is happening here
}
uv__make_close_pending(handle);
}
I could call uv__tcp_close manually, but it's not in the public headers (and probably not the right solution anyway).

libuv is not done with a handle until it's close callback is called. That is the exact moment when you can free the handle.
I see you call uv_loop_close, but you don't check for the return value. If there are still pending handles, it will return UV_EBUSY, so you should check for that.
If you want to close a loop and close all handles, you need to do the following:
Use uv_stop to stop the loop
Use uv_walk and call uv_close on all handles which are not closing
Run the loop again with uv_run so all close callbacks are called and you can free the memory in the callbacks
Call uv_loop_close, it should return 0 now

I finally figured out how to stop a loop and clean up all handles.
I created a bunch of handles and SIGINT signal handle:
uv_signal_t *sigint = new uv_signal_t;
uv_signal_init(uv_default_loop(), sigint);
uv_signal_start(sigint, on_sigint_received, SIGINT);
When SIGINT is received (Ctrl+C in console is pressed) the on_sigint_received callback is called.
The on_sigint_received looks like:
void on_sigint_received(uv_signal_t *handle, int signum)
{
int result = uv_loop_close(handle->loop);
if (result == UV_EBUSY)
{
uv_walk(handle->loop, on_uv_walk, NULL);
}
}
It triggers a call back function on_uv_walk:
void on_uv_walk(uv_handle_t* handle, void* arg)
{
uv_close(handle, on_uv_close);
}
It tries to close each opened libuv handle.
Note: that I do not call uv_stop before uv_walk, as mentioned saghul.
After on_sigint_received function is called libuv loop continuous the execution and on the next iteration calls on_uv_close for each opened handle. If you call the uv_stop function, then the on_uv_close callback will not be called.
void on_uv_close(uv_handle_t* handle)
{
if (handle != NULL)
{
delete handle;
}
}
After that libuv do not have opened handles and finishes the loop (exits from uv_run):
uv_run(uv_default_loop(), UV_RUN_DEFAULT);
int result = uv_loop_close(uv_default_loop());
if (result)
{
cerr << "failed to close libuv loop: " << uv_err_name(result) << endl;
}
else
{
cout << "libuv loop is closed successfully!\n";
}

I like the solution given by Konstantin Gindemit
I did run into a couple of problems however. His on_uv_close() function ends with a core dump. Also the uv_signal_t variable was causing valgrind to report a "definitely lost" memory leak.
I am using his code with fixes for these 2 situations.
void on_uv_walk(uv_handle_t* handle, void* arg) {
uv_close(handle, NULL);
}
void on_sigint_received(uv_signal_t *handle, int signum) {
int result = uv_loop_close(handle->loop);
if(result == UV_EBUSY) {
uv_walk(handle->loop, on_uv_walk, NULL);
}
}
int main(int argc, char *argv[]) {
uv_signal_t *sigint = new uv_signal_t;
uv_signal_init(uv_default_loop(), sigint);
uv_signal_start(sigint, on_sigint_received, SIGINT);
uv_loop_t* main_loop = uv_default_loop();
...
uv_run(main_loop, UV_RUN_DEFAULT));
uv_loop_close(uv_default_loop());
delete sigint;
return 0;
}

Related

accept call blocking thread termination

I'm having trouble terminating my server in my multithreaded program (one server, multiple clients).
When the variable global_var, which counts the number of currently connected clients, gets set to 0, the server should terminate, but it doesn't.
What I think is happening is since accept() is blocking , the code never reaches the break condition in main loop.
It's breaking correctly out of thread_func but then it blocks inside the while loop, just before the accept() call and after printing "Exiting thread_func".
volatile int finished = 0; // Gets set to 1 by catching SIGINT/SIGSTOP
int global_var = 0; // When it gets to 0, server should terminate
int server_fd;
void * thread_func(void* arg)
{
do_some_pre_stuff();
while(1)
{
if(!global_var)
{
close(server_fd);
finished = 1;
break;
}
if(recv(...) > 0)
{
do_more_stuff()
}
else
{
disconnect_client();
global_var--;
break;
}
}
free_more_ressources();
return NULL;
}
int main()
{
do_initial_stuff();
init_socket();
listen();
while (!finished)
{
if( (fd = accept(server_fd,...)) == -1)
exit(-1);
global_var++;
/* Some intermediate code */
if(!global_var)
break;
// Thread for the newly connected player
if(pthread_create(&thread_id[...], NULL, thread_func, (void*)some_arg)
exit(-1);
}
free_resources();
puts("Exiting thread_func");
}
I tried the advice listed here without success (except the pipe answer, not trying to mess with pipes).
I'm new to socket programming but what I tried so far looked correct but none of the solutions worked (including semaphores, pthread_cancel,etc)
PS: synchronization has been implemented, just omitted here for readability

libevent - event_base_loop() should it get events repeatly?

Here is a simple program using libevent on linux, it tracks the stdout fd, when it's writable, the callback will print some info to stdout.
Code
hello_libevent.c:
// libevent hello
#include <stdio.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
#include <event2/event.h>
#include <event2/thread.h>
void event_callback(evutil_socket_t fd, short events, void *arg) {
if(events | EV_WRITE) {
write(fd, "hello\n", 7);
}
sleep(1);
}
int libevent_test() {
int opr;
// enable pthread
if(evthread_use_pthreads() == -1) {
printf("error while evthread_use_pthreads(): %s\n", strerror(errno));
return -1;
}
// create event_base
struct event_base* eb;
if((eb = event_base_new()) == NULL) {
printf("error while event_base_new(): %s\n", strerror(errno));
return -1;
}
// create event
int fd_stdout = fileno(stdout);
struct event* event_stdout;
event_stdout = event_new(eb, fd_stdout, EV_WRITE, &event_callback, NULL);
// add event as pending
struct timeval timeout = {10, 0};
if(event_add(event_stdout, &timeout) == -1) {
printf("error while event_add(): %s\n", strerror(errno));
return -1;
}
// dispatch
if((opr = event_base_loop(eb, EVLOOP_NONBLOCK)) == -1) {
printf("error while event_base_dispatch(): %s\n", strerror(errno));
return -1;
} else if(opr == 1) {
printf("no more events\n");
} else {
printf("exit normally\n");
}
// free event
event_free(event_stdout);
return 0;
}
int main(int argc, char * argv[]) {
return libevent_test();
}
Compile:
gcc -Wall hello_libevent.c -levent -levent_pthreads
Execution result:
hello
no more events
Questions:
In the test, event only occur once, is that the expected behavior? Or it should loop to get more event until timeout?
How to make it get event continuously? Is it necessary to call event_base_loop within a loop, while it's already a loop?
I think the event flag EV_PERSIST mentioned in event.h File Reference might help.
Persistent event: won't get removed automatically when activated.
When a persistent event with a timeout becomes activated, its timeout is reset to 0.
Instead of
//...
event_stdout = event_new(eb, fd_stdout, EV_WRITE, &event_callback, NULL);
//...
you can pass this flag to event_new
//...
event_stdout = event_new(eb, fd_stdout, EV_WRITE|EV_PERSIST, &event_callback, NULL);
//...
and the other parts of code remain unchanged. At this time you create and add an event once and there's no need to call event_base_loop within a loop.
The compiled program just keeps printing lines of "hello" until it is terminated.
By the way, I noticed that changing
write(fd, "hello\n", 7);
into
write(fd, "hello\n", 6);
eliminates the leading character '\0' of each line.
From http://www.wangafu.net/~nickm/libevent-book/Ref3_eventloop.html it looks like you can call event_base_loop or event_base_dispatch from within a loop.
while (1) {
/* This schedules an exit ten seconds from now. */
event_base_loopexit(base, &ten_sec);``
event_base_dispatch(base);
puts("Tick");
}
The main purpose of events is to inform some busy thread about some event that happened elsewhere. So, this looks logical.

Does QEMU user mode emulation exit in a way that would prevent pthread_join from blocking?

I'm trying to run QEMU's user mode emulator as a thread in a larger program that I'm writing. I've modified the linux-user/main.c file so that the standard int main(int argc, char **argv, char **envp function is now called void *qemu_user_mode_func(void *arg). I've also added pthread_exit(NULL) to the end of that function, as is standard practice for pthreads (or so I've been told).
However, when I try to run a second thread that contains my own test function (shown below in void *test_func(void *arg)), the process exits before the second thread completes, even with a call to pthread_join(tid), which I've read blocks the calling thread until thread tid returns. Does QEMU's user mode emulation exit in such a way that would prevent pthread_join from exiting, or am I just using threads wrong?
Here's my code (not including the bulk of qemu_user_mode_func):
void *qemu_user_mode_func(void *arg)
{
thread_data_t *thread_data;
int argc;
char **argv;
char **envp;
/** QEMU's normal code **/
//return 0;
pthread_exit(NULL);
}
void *test_func(void *arg) {
struct timespec time;
time.tv_sec = 7;
time.tv_nsec = 0;
nanosleep(&time, NULL);
printf("hello, world - from a thread\n");
pthread_exit(NULL);
}
int main(int argc, char**argv, char **envp) {
//Initialize variables to create thread
int rc;
pthread_t threads[2];
thread_data_t main_args;
main_args.tid = 1;
main_args.argc = argc;
main_args.argv = argv;
main_args.envp = envp;
//Create thread
if ((rc = pthread_create(&(threads[0]), NULL, test_func, NULL))) {
fprintf(stderr, "error: pthread_create, rc: %d\n", rc);
return EXIT_FAILURE;
}
if ((rc = pthread_create(&(threads[1]), NULL, qemu_user_mode_func, (void *)&main_args))) {
fprintf(stderr, "error: pthread_create, rc: %d\n", rc);
return EXIT_FAILURE;
}
//Wait for thread to finish, then terminate process
for (rc = 0; rc < 2; rc++) {
pthread_join(threads[rc], NULL);
}
return 0;
}
EDIT: I've discovered in the void cpu_loop(CPUX86State *env) function that when the emulated program reaches its conclusion, QEMU calls syscall 231, which is sys_exit_group (as per 1). So I'm guessing this syscall is terminating the entire process that I'm running. I'd appreciate any tips on how to get around that!
If you turn a complicated preexisting application into thread there are going to be issues. One is that the application can call exit or its variants which will terminate your entire program. There are numerous other issues that could be causing a problem. I would suggest using gdb to determine what is making your program exit.
Problem was solved by editing the following section in void cpu_loop(CPUX86State *env). I capture either sys_exit_group and sys_exit system calls before they are executed, and just return from the function instead.
Original:
void cpu_loop(CPUX86State *env)
{
CPUState *cs = CPU(x86_env_get_cpu(env));
int trapnr;
abi_ulong pc;
target_siginfo_t info;
for(;;) {
cpu_exec_start(cs);
trapnr = cpu_x86_exec(env);
cpu_exec_end(cs);
switch(trapnr) {
case 0x80:
/* linux syscall from int $0x80 */
env->regs[R_EAX] = do_syscall(env,
env->regs[R_EAX],
env->regs[R_EBX],
env->regs[R_ECX],
env->regs[R_EDX],
env->regs[R_ESI],
env->regs[R_EDI],
env->regs[R_EBP],
0, 0);
break;
#ifndef TARGET_ABI32
case EXCP_SYSCALL:
/* linux syscall from syscall instruction */
env->regs[R_EAX] = do_syscall(env,
env->regs[R_EAX],
env->regs[R_EDI],
env->regs[R_ESI],
env->regs[R_EDX],
env->regs[10],
env->regs[8],
env->regs[9],
0, 0);
break;
#endif
Modified:
void cpu_loop(CPUX86State *env)
{
CPUState *cs = CPU(x86_env_get_cpu(env));
int trapnr;
abi_ulong pc;
target_siginfo_t info;
for(;;) {
cpu_exec_start(cs);
trapnr = cpu_x86_exec(env);
cpu_exec_end(cs);
switch(trapnr) {
case 0x80:
/* linux syscall from int $0x80 */
env->regs[R_EAX] = do_syscall(env,
env->regs[R_EAX],
env->regs[R_EBX],
env->regs[R_ECX],
env->regs[R_EDX],
env->regs[R_ESI],
env->regs[R_EDI],
env->regs[R_EBP],
0, 0);
break;
#ifndef TARGET_ABI32
case EXCP_SYSCALL:
/* linux syscall from syscall instruction */
----> if ((env->regs[R_EAX] == __NR_exit_group) || (env->regs[R_EAX] == __NR_exit)) {
return;
}
env->regs[R_EAX] = do_syscall(env,
env->regs[R_EAX],
env->regs[R_EDI],
env->regs[R_ESI],
env->regs[R_EDX],
env->regs[10],
env->regs[8],
env->regs[9],
0, 0);
break;
#endif

segfault handler for C only called once

I'm writing a program that scans the process memory and creates a memchunk structures to represent the accessibility of a block of memory. A side effect of this process is to learn how to handle signals, as there should be many segfaults over the course of this scan. So, I'm trying to cause many segfaults to learn about signals while also learning a bit about the virtual memory. This is compiled as a 32bit program on a linux platform.
I make use of setjmp and longjmp in order to re-enter my program.
My problem is that my segfault handler is only called once, and then the default segfault handler -- the core dumping one -- is called on the next segfault.
Here is my handler:
static void hdl (int sig, siginfo_t *siginfo, void *unused){
/* handler is called properly and can read curr_access */
printf("-------------------\n");
printf("handling segfault\n");
printf("curr_access = %d\n", curr_access);
printf("-----------------\n");
switch(curr_access){
case -1:
longjmp(buf, 1);
break;
case 0:
longjmp(buf, 2);
break;
default:
printf("error in hdl\n");
}
}
Here is where I register it
void set_hndlr(){
printf("------------------\n");
printf("setting handler\n");
printf("-----------------\n");
/* setting up signal handler */
struct sigaction sa;
sa.sa_flags = SA_SIGINFO;
sigemptyset(&sa.sa_mask);
sa.sa_sigaction = hdl;
if (sigaction(SIGSEGV, &sa, NULL) == -1)
printf("error setting segfault hanld");
}
here is where I "test" a memory location:
switch(jmp_code){
/* we haven't gone through testing yet */
case 0:
test_boolean = 1;
break;
/* we've gone through testing and we segfaulted at read */
case 1:
test_boolean = 0;
curr_access = -1;
break;
/* we've gone through testing and we segfaulted at write */
case 2:
test_boolean = 0;
curr_access = 0;
break;
/* if we reach here then there is an error in program logic */
default:
printf("error in programming logic regaurding jmp\n");
exit(-2);
}
if(1 == test_boolean){
/* not accessible */
curr_access = -1;
printf("testing read\n");
/*test read */
curr_cont = *curr_addr;
curr_access = 0;
printf("readable\n");
/*test write */
*curr_addr = curr_cont;
curr_access = 1;
printf("read/write\n");
}
Here is some example output:
setting handler
curr_addr = (nil)
jmp_code = 0
testing read
handling segfault
curr_access = -1
jmp_code = 1
base[0].RW = -1
base[0].length = 4096
base[0].start = (nil)
curr_addr = 0x1000
jmp_code = 0
testing read
Segmentation fault (core dumped)
It's strange that I can handle one segfault just fine, but then it wont handle again. I would really appreciate any help.
In some of the systems, it does not support the re-assignig of singal handler. we have to re assign the handle after the calling of that function_handler.
Unreliable Signals
One problem with these early versions is that the action for a signal was reset to its default each time the signal occurred.
int sig_int();
...
signal(SIGINT, sig_int);
...
sig_int()
{
signal(SIGINT, sig_int);
...
}

Run application continuously

Whats the smartest way to run an application continuously so that it doesn't exit after it hits the bottom? Instead it starts again from the top of main and only exits when commanded. (This is in C)
You should always have some way of exiting cleanly. I'd suggest moving the code off to another function that returns a flag to say whether to exit or not.
int main(int argc, char*argv[])
{
// param parsing, init code
while (DoStuff());
// cleanup code
return 0;
}
int DoStuff(void)
{
// code that you would have had in main
if (we_should_exit)
return 0;
return 1;
}
Most applications that don't fall through enter some kind of event processing loop that allows for event-driven programming.
Under Win32 development, for instance, you'd write your WinMain function to continually handle new messages until it receives the WM_QUIT message telling the application to finish. This code typically takes the following form:
// ...meanwhile, somewhere inside WinMain()
MSG msg;
while (GetMessage(&msg, NULL, 0, 0))
{
TranslateMessage(&msg);
DispatchMessage(&msg);
}
If you are writing a game using SDL, you would loop on SDL events until deciding to exit, such as when you detect that the user has hit the Esc key. Some code to do that might resemble the following:
bool done = false;
while (!done)
{
SDL_Event event;
while (SDL_PollEvent(&event))
{
switch (event.type)
{
case SDL_QUIT:
done = true;
break;
case SDL_KEYDOWN:
if (event.key.keysym.sym == SDLK_ESCAPE)
{
done = true;
}
break;
}
}
}
You may also want to read about Unix Daemons and Windows Services.
while (true)
{
....
}
To elaborate a bit more, you want to put something in that loop that allows you to let the user do repeated actions. Whether it's reading key strokes and performing actions based on the keys pressed, or reading data from the socket and sending back a response.
There are a number of ways to "command" your application to exit (such as a global exit flag or return codes). Some have already touched on using an exit code so I'll put forward an easy modification to make to an existing program using an exit flag.
Let's assume your program executes a system call to output a directory listing (full directory or a single file):
int main (int argCount, char *argValue[]) {
char *cmdLine;
if (argCount < 2) {
system ("ls");
} else {
cmdLine = malloc (strlen (argValue[1]) + 4);
sprintf (cmdLine, "ls %s", argValue[1]);
system (cmdLine);
}
}
How do we go about making that loop until an exit condition. The following steps are taken:
Change main() to oldMain().
Add new exitFlag.
Add new main() to continuously call oldMain() until exit flagged.
Change oldMain() to signal exit at some point.
This gives the following code:
static int exitFlag = 0;
int main (int argCount, char *argValue[]) {
int retVal = 0;
while (!exitFlag) {
retVal = oldMain (argCount, argValue);
}
return retVal;
}
static int oldMain (int argCount, char *argValue[]) {
char *cmdLine;
if (argCount < 2) {
system ("ls");
} else {
cmdLine = malloc (strlen (argValue[1]) + 4);
sprintf (cmdLine, "ls %s", argValue[1]);
system (cmdLine);
}
if (someCondition)
exitFlag = 1;
}

Resources