src/slave.c

   1 #define _XOPEN_SOURCE 700
   2 #define _GNU_SOURCE
   3
   4
   5
   6 #include "slave.h"
   7 #include <assert.h>
   8
   9 #include "network.h"
  10 #include "daemon.h"
  11 #include "log.h"
  12 #include <errno.h>
  13 #include <pty.h>
  14 #include <fcntl.h>
  15 #include <string.h>
  16
  17 #include <pthread.h>
  18 #include <syslog.h>
  19 #include <signal.h>
  20 #include <unistd.h>
  21 #include <sys/types.h>
  22 #include <sys/wait.h>
  23 #include <sys/socket.h>
  24 #include <netinet/tcp.h>
  25
  26
  27
  28 Slave * slave;
  29
  30 char name[BUFSIZ];
  31
  32 void Slave_shell(int i, char * shell);
  33 void Slave_cleanup();
  34
  35
  36 void Slave_main(Options * o)
  37 {
  38         //fprintf(stderr, "%d\n", o->nCPU);
  39
  40
  41
  42         setbuf(stdin, NULL); setbuf(stdout, NULL); setbuf(stderr, NULL);
  43
  44         dup2(fileno(stdout), fileno(stderr)); // yes, this works, apparently
  45
  46         slave = (Slave*)(calloc(o->nCPU, sizeof(Slave)));
  47         atexit(Slave_cleanup);
  48
  49
  50         if (strcmp(o->master_addr, "-") != 0)
  51         {
  52                 if (fork() != 0)
  53                         exit(EXIT_SUCCESS);
  54
  55                 //log_print(2, "Slave_main", "Using unsecured networking; connect to %s:%d", o->master_addr, o->port);
  56                 //log_print(2, "Slave_main", "Connecting to %s:%d", o->master_addr, o->port);
  57                 int net_fd = Network_client(o->master_addr, o->port, 100);
  58                 dup2(net_fd, fileno(stdin));
  59                 dup2(net_fd, fileno(stdout));
  60                 dup2(net_fd, fileno(stderr));
  61
  62         }
  63         else
  64         {
  65                 o->master_addr = "localhost";
  66                 //log_print(2, "Slave_main", "Using port forwarding; connect to %s", o->master_addr);
  67         }
  68
  69         char buffer[BUFSIZ];
  70
  71         fgets(name, sizeof(name), stdin);
  72         name[strlen(name)-1] = '\0';
  73         //log_print(LOGINFO, "Slave_main", "Started remote swarm \"%s\"", name);
  74
  75
  76         fprintf(stdout, "%d\n", o->nCPU);
  77         //log_print(2, "Slave_main", "Wrote nCPU %d", o->nCPU);
  78
  79
  80         int port = 0;
  81         for (int i = 0; i < o->nCPU; ++i)
  82         {
  83                 //log_print(2, "Slave_main", "Waiting for port number...");
  84                 fgets(buffer, sizeof(buffer), stdin);
  85
  86                 buffer[strlen(buffer)-1] = '\0';
  87                 sscanf(buffer, "%d", &port);
  88                 //log_print(2, "Slave_main", "Port number %d", port);
  89                 slave[i].in = Network_client(o->master_addr, port,20);
  90                 //log_print(2, "Slave_main", "Connected to %s:%d\n", o->master_addr, port);
  91                 slave[i].out = slave[i].in;
  92
  93                 Slave_shell(i, o->shell);
  94         }
  95
  96
  97         Slave_loop(o);
  98
  99         exit(EXIT_SUCCESS);
 100 }
 101
 102 void Slave_shell(int i, char * shell)
 103 {
 104         slave[i].pid = fork();
 105
 106
 107
 108         if (slave[i].pid == 0)
 109         {
 110                 dup2(slave[i].in, fileno(stdin));
 111                 dup2(slave[i].out, fileno(stdout));
 112                 //dup2(error_socket[1], fileno(stderr));
 113
 114                 execlp(shell, shell, NULL);
 115         }
 116
 117         // if the input is a network socket, this message gets sent to the master
 118         // which will then echo it back to the socket and hence the shell
 119         FILE * f = fdopen(slave[i].in, "w"); setbuf(f, NULL);
 120         fprintf(f, "name=\"%s:%d\"\n", name,i);
 121 }
 122
 123 void Slave_loop(Options * o)
 124 {
 125         fd_set readSet;
 126         struct timeval tv;
 127         tv.tv_sec = 0;
 128         tv.tv_usec = 100000;
 129
 130         int p = -1; int s = 0;
 131         char buffer[BUFSIZ];
 132         while (true)
 133         {
 134                 FD_ZERO(&readSet);
 135                 FD_SET(fileno(stdin), &readSet);
 136                 p = waitpid(-1, &s, 0);
 137                 if (p == -1)
 138                 {
 139                         //log_print(0, "Slave_loop", "waitpid : %s", strerror(errno));
 140                         continue;
 141                 }
 142
 143                 //log_print(3, "Slave_loop", "Detected child %d exiting...", p);
 144
 145                 // check for an exit command from the master
 146                 select(fileno(stdin) + 1, &readSet, NULL, NULL, &tv);
 147
 148                 if (FD_ISSET(fileno(stdin), &readSet))
 149                 {
 150                         fgets(buffer, sizeof(buffer), stdin);
 151                         if (strcmp(buffer, "exit\n") == 0)
 152                         {
 153                                 log_print(2, "Slave_loop", "Received notification of exit.\n");
 154                                 exit(EXIT_SUCCESS);
 155                         }
 156                 }
 157
 158                 int i = 0;
 159                 for (i = 0; i < o->nCPU; ++i)
 160                 {
 161                         if (slave[i].pid == p) break;
 162                 }
 163                 if (i >= o->nCPU)
 164                         error("Slave_loop", "No child matches pid %d", p);
 165
 166                 sigchld_respond(s, name, i);
 167
 168
 169                 // cancel any tasks at the master for this slave
 170
 171                 write(slave[i].out, SHELL_OUTPUT_FINISHED, SHELL_OUTPUT_FINISHED_LENGTH);
 172
 173                 Slave_shell(i, o->shell);
 174
 175
 176         }
 177 }
 178
 179 void Slave_cleanup()
 180 {
 181         for (int i = 0; i < options.nCPU; ++i)
 182         {
 183                 kill(slave[i].pid, SIGTERM);
 184         }
 185         sleep(1);
 186         for (int i = 0; i < options.nCPU; ++i)
 187         {
 188                 kill(slave[i].pid, SIGKILL);
 189         }
 190         free(slave);
 191 }
 192
 193