multiple processes via socketpair hang sometimes - c

Question

I am trying to implement something that will give me a solution for:
| --> cmd3 --> cmd4 -->
cmd2-->|
| --> cmd5 --> cmd6 -->
and so on...
This is multiple executions of processes and pipe the results via chains of other's processes with threads, each commands chain should run in different thread.
I choose socketpair for the implementation of IPC, because pipe has a a bottleneck with the buffer size limit 64K.
When I test the program with single chain - it's work as expected, but when I am running master command and the output of it I send via socketpair to read end of multiple processes in each thread - the program stuck (look like a deadlock)
Whats I am doing wrong:
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <signal.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <fcntl.h>
#include <sys/socket.h>
typedef struct command {
char** argv;
int num_children;
struct command* master_cmd;
struct command** chains;
struct command* next;
int fd;
} command;
void be_child(command* cmd);
int execute_master_command_and_pipe_to_childs(command* cmd, int input);
int run_pipeline_sockets(command *cmd, int input);
void waitfor(int fd);
int main(int argc, char* argv[]) {
handle_segfault();
command* cmd1 = (command*) malloc(sizeof(command));
command* cmd2 = (command*) malloc(sizeof(command));
command* cmd3 = (command*) malloc(sizeof(command));
command* cmd4 = (command*) malloc(sizeof(command));
command* cmd5 = (command*) malloc(sizeof(command));
command* cmd6 = (command*) malloc(sizeof(command));
command* chains1[2];
chains1[0] = cmd3;
chains1[1] = cmd5;
char* args1[] = { "cat", "/tmp/test.log", NULL };
char* args3[] = { "sort", NULL, NULL };
char* args4[] = { "wc", "-l", NULL };
char* args5[] = { "wc", "-l", NULL };
char* args6[] = { "wc", "-l", NULL };
cmd1->argv = args1;
cmd2->argv = NULL;
cmd3->argv = args3;
cmd4->argv = args4;
cmd5->argv = args5;
cmd6->argv = args6;
cmd1->master_cmd = NULL;
cmd1->next = NULL;
cmd1->chains = NULL;
cmd1->num_children = -1;
cmd2->master_cmd = cmd1;
cmd2->chains = chains1;
cmd2->next = NULL;
cmd2->num_children = 2;
cmd3->master_cmd = NULL;
cmd3->next = cmd4;
cmd3->chains = NULL;
cmd3->num_children = -1;
cmd4->master_cmd = NULL;
cmd4->next = NULL;
cmd4->chains = NULL;
cmd4->num_children = -1;
cmd5->master_cmd = NULL;
cmd5->next = cmd6;
cmd5->chains = NULL;
cmd5->num_children = -1;
cmd6->master_cmd = NULL;
cmd6->next = NULL;
cmd6->chains = NULL;
cmd6->num_children = -1;
int rc = execute_master_command_and_pipe_to_childs(cmd2, -1);
return 0;
}
int execute_master_command_and_pipe_to_childs(command* cmd, int input) {
int num_children = cmd->num_children;
int write_pipes[num_children];
pthread_t threads[num_children];
command* master_cmd = cmd->master_cmd;
pid_t pid;
int i;
for (i = 0; i < num_children; i++) {
int new_pipe[2];
if (socketpair(AF_LOCAL, SOCK_STREAM, 0, new_pipe) < 0) {
int errnum = errno;
fprintf(STDERR_FILENO, "ERROR (%d: %s)\n", errnum,
strerror(errnum));
return EXIT_FAILURE;
}
if (cmd->chains[i] != NULL) {
cmd->chains[i]->fd = new_pipe[0];
if (pthread_create(&threads[i], NULL, (void *) be_child,
cmd->chains[i]) != 0) {
perror("pthread_create"), exit(1);
}
write_pipes[i] = new_pipe[1];
} else {
perror("ERROR\n");
}
}
if (input != -1) {
waitfor(input);
}
int pipefd = run_pipeline_sockets(master_cmd, input);
int buffer[1024];
int len = 0;
while ((len = read(pipefd, buffer, sizeof(buffer))) != 0) {
int j;
for (j = 0; j < num_children; j++) {
if (write(write_pipes[j], &buffer, len) != len) {
fprintf(STDERR_FILENO, "Write failed (child %d)\n", j);
exit(1);
}
}
}
close(pipefd);
for (i = 0; i < num_children; i++) {
close(write_pipes[i]);
}
for (i = 0; i < num_children; i++) {
if (pthread_join(threads[i], NULL) != 0) {
perror("pthread_join"), exit(1);
}
}
}
void waitfor(int fd) {
fd_set rfds;
struct timeval tv;
int retval;
FD_ZERO(&rfds);
FD_SET(fd, &rfds);
tv.tv_sec = 0;
tv.tv_usec = 500000;
retval = select(fd + 1, &rfds, NULL, NULL, &tv);
if (retval == -1)
perror("select()");
else if (retval) {
printf("Data is available now on: %d\n", fd);
} else {
printf("No data on: %d\n", fd);
///waitfor(fd);
}
}
void be_child(command* cmd) {
printf(
"fd = %d , argv = %s , args = %s , next = %d , master_cmd = %d , next_chain = %d\n",
cmd->fd, cmd->argv[0], cmd->argv[1], cmd->next, cmd->master_cmd,
cmd->chains);
waitfor(cmd->fd);
int fd = run_pipeline_sockets(cmd, cmd->fd);
waitfor(fd);
int buffer[1024];
int len = 0;
while ((len = read(fd, buffer, sizeof(buffer))) != 0) {
write(STDERR_FILENO, &buffer, len);
}
close(cmd->fd);
close(fd);
}
int run_pipeline_sockets(command *cmd, int input) {
int pfds[2] = { -1, -1 };
int pid = -1;
if (socketpair(AF_LOCAL, SOCK_STREAM, 0, pfds) < 0) {
int errnum = errno;
fprintf(STDERR_FILENO, "socketpair failed (%d: %s)\n", errnum,
strerror(errnum));
return EXIT_FAILURE;
}
if ((pid = fork()) == 0) { /* child */
if (input != -1) {
dup2(input, STDIN_FILENO);
close(input);
}
if (pfds[1] != -1) {
dup2(pfds[1], STDOUT_FILENO);
close(pfds[1]);
}
if (pfds[0] != -1) {
close(pfds[0]);
}
execvp(cmd->argv[0], cmd->argv);
exit(1);
} else { /* parent */
if (input != -1) {
close(input);
}
if (pfds[1] != -1) {
close(pfds[1]);
}
if (cmd->next != NULL) {
run_pipeline_sockets(cmd->next, pfds[0]);
} else {
return pfds[0];
}
}
}
void segfault_sigaction(int signal, siginfo_t *si, void *arg) {
printf("Caught segfault at address %p\n", si->si_addr);
printf("Caught segfault errno %p\n", si->si_errno);
exit(0);
}
void handle_segfault(void) {
struct sigaction sa;
memset(&sa, 0, sizeof(sigaction));
sigemptyset(&sa.sa_mask);
sa.sa_sigaction = segfault_sigaction;
sa.sa_flags = SA_SIGINFO;
sigaction(SIGSEGV, &sa, NULL);
}

Answer 1

I would come at this problem from a very different angle: rather than coming up with a large data structure to manage the pipe tree, and using threads (where an io blockage in a process may block in its threads) I would use only processes.
I also fail to see how a 64K buffer is your bottleneck when you're only using a 1K buffer.
2 simple functions should guide this: (error handling omitted for brevity, and using a pseudocodey parsecmd() function which turns a space separated string into an argument vector)
int mkproc(char *cmd, int outfd)
{
Command c = parsecmd(cmd);
int pipeleft[2];
pipe(pipeleft);
if(!fork()){
close(pipeleft[1]);
dup2(pipeleft[0], 0);
dup2(outfd, 1);
execvp(c.name, c.argv);
}
close(pipeleft[0]);
return pipeleft[1];
}
Mkproc takes the fd it will write to, and returns what it will read from. This way chains are really easy to initalize:
int chain_in = mkproc("cat foo.txt", mkproc("sort", mkproc("wc -l", 1)));
the next is:
int mktree(char *cmd, int ofd0, ...)
{
int piperight[2];
pipe(piperight);
int cmdin = mkproc(cmd, piperight[1]);
close(piperight[1]);
if(!fork()){
uchar buf[4096];
int n;
while((n=read(piperight[0], buf, sizeof buf))>0){
va_list ap;
int fd;
va_start(ap, ofd0);
for(fd=ofd0; fd!=-1; fd=va_arg(ap, int)){
write(fd, buf, n);
}
va_end(ap);
}
}
return cmdin;
}
Between the two of these, it is very easy to construct trees of arbitrary complexity, as so:
int tree_in = mktree("cat foo.txt",
mktree("rot13",
mkproc("uniq", mkproc("wc -l", 1)),
mkproc("wc -l", open("out.txt", O_WRONLY)), -1),
mkproc("sort", 2), -1);
This would output a sorted foo.txt to stderr, the number of lines in rot13'd foo.txt to out.txt, and the number of non-duplicate lines of rot13'd foo.txt to stdout.