cpulimit.c

   1 /**
   2  *
   3  * cpulimit - a cpu limiter for Linux
   4  *
   5  * Copyright (C) 2005-2008, by:  Angelo Marletta <marlonx80@hotmail.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; either version 2
  10  * of the License, or (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  20  *
  21  **********************************************************************
  22  *
  23  * This is a simple program to limit the cpu usage of a process
  24  * If you modify this code, send me a copy please
  25  *
  26  * Date:    15/2/2008
  27  * Version: 1.2 alpha
  28  * Get the latest version at: http://cpulimit.sourceforge.net
  29  *
  30  * Changelog:
  31  * - reorganization of the code, splitted in more source files
  32  * - control function process_monitor() optimized by eliminating an unnecessary loop
  33  * - experimental support for multiple control of children processes and threads
  34  *   children detection algorithm seems heavy because of the amount of code,
  35  *   but it's designed to be scalable when there are a lot of children processes
  36  * - cpu count detection, i.e. if you have 4 cpu, it is possible to limit up to 400%
  37  * - in order to avoid deadlock, cpulimit prevents to limit itself
  38  * - option --path eliminated, use --exe instead both for absolute path and file name
  39  * - deleted almost every setpriority(), just set it once at startup
  40  * - minor enhancements and bugfixes
  41  *
  42  */
  43
  44
  45 #include <getopt.h>
  46 #include <stdio.h>
  47 #include <fcntl.h>
  48 #include <stdlib.h>
  49 #include <time.h>
  50 #include <sys/time.h>
  51 #include <unistd.h>
  52 #include <sys/types.h>
  53 #include <signal.h>
  54 #include <sys/resource.h>
  55 #include <string.h>
  56 #include <dirent.h>
  57 #include <errno.h>
  58 #include <string.h>
  59
  60 #include "process.h"
  61 #include "procutils.h"
  62 #include "list.h"
  63
  64 //some useful macro
  65 #define MIN(a,b) (a<b?a:b)
  66 #define MAX(a,b) (a>b?a:b)
  67 #define print_caption() printf("\n%%CPU\twork quantum\tsleep quantum\tactive rate\n")
  68 //control time slot in microseconds
  69 //each slot is splitted in a working slice and a sleeping slice
  70 #define CONTROL_SLOT 100000
  71
  72 #define MAX_PRIORITY -10
  73
  74 //the "family"
  75 struct process_family pf;
  76 //pid of cpulimit
  77 int cpulimit_pid;
  78 //name of this program (maybe cpulimit...)
  79 char *program_name;
  80
  81 /* CONFIGURATION VARIABLES */
  82
  83 //verbose mode
  84 int verbose = 0;
  85 //lazy mode (exits if there is no process)
  86 int lazy = 0;
  87
  88 //how many cpu do we have?
  89 int get_cpu_count()
  90 {
  91         FILE *fd;
  92         int cpu_count = 0;
  93         char line[100];
  94         fd = fopen("/proc/stat", "r");
  95         if (fd < 0)
  96                 return 0; //are we running Linux??
  97         while (fgets(line,sizeof(line),fd)!=NULL) {
  98                 if (strncmp(line, "cpu", 3) != 0) break;
  99                 cpu_count++;
 100         }
 101         fclose(fd);
 102         return cpu_count - 1;
 103 }
 104
 105 //return t1-t2 in microseconds (no overflow checks, so better watch out!)
 106 inline unsigned long timediff(const struct timespec *t1,const struct timespec *t2)
 107 {
 108         return (t1->tv_sec - t2->tv_sec) * 1000000 + (t1->tv_nsec/1000 - t2->tv_nsec/1000);
 109 }
 110
 111 //returns t1-t2 in microseconds
 112 inline unsigned long long tv_diff(struct timeval *t1, struct timeval *t2)
 113 {
 114         return ((unsigned long long)(t1->tv_sec - t2->tv_sec)) * 1000000ULL + t1->tv_usec - t2->tv_usec;
 115 }
 116
 117 //SIGINT and SIGTERM signal handler
 118 void quit(int sig)
 119 {
 120         //let all the processes continue if stopped
 121         struct list_node *node = NULL;
 122         for (node=pf.members.first; node!= NULL; node=node->next) {
 123                 struct process *p = (struct process*)(node->data);
 124                 process_close(p->history);
 125                 kill(p->pid, SIGCONT);
 126         }
 127         //free all the memory
 128         cleanup_process_family(&pf);
 129         exit(0);
 130 }
 131
 132 void print_usage(FILE *stream, int exit_code)
 133 {
 134         fprintf(stream, "Usage: %s TARGET [OPTIONS...]\n",program_name);
 135         fprintf(stream, "   TARGET must be exactly one of these:\n");
 136         fprintf(stream, "      -p, --pid=N        pid of the process (implies -z)\n");
 137         fprintf(stream, "      -e, --exe=FILE     name of the executable program file or absolute path name\n");
 138         fprintf(stream, "   OPTIONS\n");
 139         fprintf(stream, "      -l, --limit=N      percentage of cpu allowed from 0 to 100 (required)\n");
 140         fprintf(stream, "      -v, --verbose      show control statistics\n");
 141         fprintf(stream, "      -z, --lazy         exit if there is no suitable target process, or if it dies\n");
 142         fprintf(stream, "      -h, --help         display this help and exit\n");
 143         exit(exit_code);
 144 }
 145
 146 void limit_process(int pid, float limit)
 147 {
 148         //slice of the slot in which the process is allowed to run
 149         struct timespec twork;
 150         //slice of the slot in which the process is stopped
 151         struct timespec tsleep;
 152         //when the last twork has started
 153         struct timespec startwork;
 154         //when the last twork has finished
 155         struct timespec endwork;
 156         //initialization
 157         memset(&twork, 0, sizeof(struct timespec));
 158         memset(&tsleep, 0, sizeof(struct timespec));
 159         memset(&startwork, 0, sizeof(struct timespec));
 160         memset(&endwork, 0, sizeof(struct timespec));
 161         //last working time in microseconds
 162         unsigned long workingtime = 0;
 163         int i = 0;
 164
 165         //build the family
 166         create_process_family(&pf, pid);
 167         struct list_node *node;
 168
 169         if (verbose) printf("Members in the family owned by %d: %d\n", pf.father, pf.members.count);
 170
 171         while(1) {
 172
 173                 if (i%100==0 && verbose) print_caption();
 174
 175                 if (i%10==0) {
 176                         //update the process family (checks only for new members)
 177                         int newborn = check_new_members(&pf);
 178                         if (newborn) {
 179                                 printf("%d new children processes detected (", newborn);
 180                                 int j;
 181                                 node = pf.members.last;
 182                                 for (j=0; j<newborn; j++) {
 183                                         printf("%d", ((struct process*)(node->data))->pid);
 184                                         if (j<newborn-1) printf(" ");
 185                                         node = node->previous;
 186                                 }
 187                                 printf(")\n");
 188                         }
 189                 }
 190
 191                 //total cpu actual usage (range 0-1)
 192                 //1 means that the processes are using 100% cpu
 193                 float pcpu = 0;
 194                 //rate at which we are keeping active the processes (range 0-1)
 195                 //1 means that the process are using all the twork slice
 196                 float workingrate = 0;
 197
 198                 //estimate how much the controlled processes are using the cpu in the working interval
 199                 for (node=pf.members.first; node!=NULL; node=node->next) {
 200                         struct process *proc = (struct process*)(node->data);
 201                         if (process_monitor(proc->history, workingtime, &(proc->history->usage))==-1) {
 202                                 //process is dead, remove it from family
 203                                 remove_process_from_family(&pf, proc->pid);
 204                                 fprintf(stderr,"Process %d dead!\n", proc->pid);
 205                         }
 206                         pcpu += proc->history->usage.pcpu;
 207                         workingrate += proc->history->usage.workingrate;
 208                 }
 209                 //average value
 210                 workingrate /= pf.members.count;
 211
 212                 //TODO: make workingtime customized for each process, now it's equal for all
 213
 214                 //adjust work and sleep time slices
 215                 if (pcpu>0) {
 216                         twork.tv_nsec = MIN(CONTROL_SLOT*limit*1000/pcpu*workingrate,CONTROL_SLOT*1000);
 217                 }
 218                 else if (pcpu==0) {
 219                         twork.tv_nsec = CONTROL_SLOT*1000;
 220                 }
 221                 else if (pcpu==-1) {
 222                         //not yet a valid idea of cpu usage
 223                         pcpu = limit;
 224                         workingrate = limit;
 225                         twork.tv_nsec = MIN(CONTROL_SLOT*limit*1000,CONTROL_SLOT*1000);
 226                 }
 227                 tsleep.tv_nsec = CONTROL_SLOT*1000-twork.tv_nsec;
 228
 229                 if (verbose && i%10==0 && i>0) {
 230                         printf("%0.2f%%\t%6ld us\t%6ld us\t%0.2f%%\n",pcpu*100,twork.tv_nsec/1000,tsleep.tv_nsec/1000,workingrate*100);
 231                 }
 232
 233                 //resume processes
 234                 for (node=pf.members.first; node!=NULL; node=node->next) {
 235                         struct process *proc = (struct process*)(node->data);
 236                         if (kill(proc->pid,SIGCONT)!=0) {
 237                                 //process is dead, remove it from family
 238                                 fprintf(stderr,"Process %d dead!\n", proc->pid);
 239                                 remove_process_from_family(&pf, proc->pid);
 240                         }
 241                 }
 242
 243                 //now processes are free to run (same working slice for all)
 244                 clock_gettime(CLOCK_REALTIME,&startwork);
 245                 nanosleep(&twork,NULL);
 246                 clock_gettime(CLOCK_REALTIME,&endwork);
 247                 workingtime = timediff(&endwork,&startwork);
 248
 249                 //stop processes, they have worked enough
 250                 //resume processes
 251                 for (node=pf.members.first; node!=NULL; node=node->next) {
 252                         struct process *proc = (struct process*)(node->data);
 253                         if (kill(proc->pid,SIGSTOP)!=0) {
 254                                 //process is dead, remove it from family
 255                                 fprintf(stderr,"Process %d dead!\n", proc->pid);
 256                                 remove_process_from_family(&pf, proc->pid);
 257                         }
 258                 }
 259                 //now the process is forced to sleep
 260                 nanosleep(&tsleep,NULL);
 261                 i++;
 262         }
 263         cleanup_process_family(&pf);
 264 }
 265
 266 int main(int argc, char **argv) {
 267
 268         //get program name
 269         char *p=(char*)memrchr(argv[0],(unsigned int)'/',strlen(argv[0]));
 270         program_name = p==NULL?argv[0]:(p+1);
 271         cpulimit_pid = getpid();
 272
 273         //argument variables
 274         const char *exe = NULL;
 275         int perclimit = 0;
 276         int pid_ok = 0;
 277         int process_ok = 0;
 278         int limit_ok = 0;
 279         int pid = 0;
 280
 281         //parse arguments
 282         int next_option;
 283     int option_index = 0;
 284         //A string listing valid short options letters
 285         const char* short_options = "p:e:l:vzh";
 286         //An array describing valid long options
 287         const struct option long_options[] = {
 288                 { "pid",        required_argument, NULL,     'p' },
 289                 { "exe",        required_argument, NULL,     'e' },
 290                 { "limit",      required_argument, NULL,     'l' },
 291                 { "verbose",    no_argument,       &verbose, 'v' },
 292                 { "lazy",       no_argument,       &lazy,    'z' },
 293                 { "help",       no_argument,       NULL,     'h' },
 294                 { 0,            0,                 0,         0  }
 295         };
 296
 297         do {
 298                 next_option = getopt_long(argc, argv, short_options,long_options, &option_index);
 299                 switch(next_option) {
 300                         case 'p':
 301                                 pid = atoi(optarg);
 302                                 //todo: verify pid is valid
 303                                 pid_ok = 1;
 304                                 process_ok = 1;
 305                                 break;
 306                         case 'e':
 307                                 exe = optarg;
 308                                 process_ok = 1;
 309                                 break;
 310                         case 'l':
 311                                 perclimit = atoi(optarg);
 312                                 limit_ok = 1;
 313                                 break;
 314                         case 'v':
 315                                 verbose = 1;
 316                                 break;
 317                         case 'z':
 318                                 lazy = 1;
 319                                 break;
 320                         case 'h':
 321                                 print_usage(stdout, 1);
 322                                 break;
 323                         case '?':
 324                                 print_usage(stderr, 1);
 325                                 break;
 326                         case -1:
 327                                 break;
 328                         default:
 329                                 abort();
 330                 }
 331         } while(next_option != -1);
 332
 333         if (pid!=0) {
 334                 lazy = 1;
 335         }
 336
 337         if (!process_ok) {
 338                 fprintf(stderr,"Error: You must specify a target process, by name or by PID\n");
 339                 print_usage(stderr, 1);
 340                 exit(1);
 341         }
 342         if (pid_ok && exe!=NULL) {
 343                 fprintf(stderr, "Error: You must specify exactly one process, by name or by PID\n");
 344                 print_usage(stderr, 1);
 345                 exit(1);
 346         }
 347         if (!limit_ok) {
 348                 fprintf(stderr,"Error: You must specify a cpu limit percentage\n");
 349                 print_usage(stderr, 1);
 350                 exit(1);
 351         }
 352         float limit = perclimit/100.0;
 353         int cpu_count = get_cpu_count();
 354         if (limit<0 || limit >cpu_count) {
 355                 fprintf(stderr,"Error: limit must be in the range 0-%d00\n", cpu_count);
 356                 print_usage(stderr, 1);
 357                 exit(1);
 358         }
 359         //parameters are all ok!
 360         signal(SIGINT, quit);
 361         signal(SIGTERM, quit);
 362
 363         //try to renice with the best value
 364         int old_priority = getpriority(PRIO_PROCESS, 0);
 365         int priority = old_priority;
 366         while (setpriority(PRIO_PROCESS, 0, priority-1) == 0 && priority>MAX_PRIORITY) {
 367                 priority--;
 368         }
 369         if (priority != old_priority) {
 370                 printf("Priority changed to %d\n", priority);
 371         }
 372         else {
 373                 printf("Cannot change priority\n");
 374         }
 375
 376         while(1) {
 377                 //look for the target process..or wait for it
 378                 int ret = 0;
 379                 if (pid_ok) {
 380                         //search by pid
 381                         ret = look_for_process_by_pid(pid);
 382                         if (ret == 0) {
 383                                 printf("No process found\n");
 384                         }
 385                         else if (ret < 0) {
 386                                 printf("Process found but you aren't allowed to control it\n");
 387                         }
 388                 }
 389                 else {
 390                         //search by file or path name
 391                         ret = look_for_process_by_name(exe);
 392                         if (ret == 0) {
 393                                 printf("No process found\n");
 394                         }
 395                         else if (ret < 0) {
 396                                 printf("Process found but you aren't allowed to control it\n");
 397                         }
 398                         else {
 399                                 pid = ret;
 400                         }
 401                 }
 402                 if (ret > 0) {
 403                         if (ret == cpulimit_pid) {
 404                                 printf("Process %d is cpulimit itself! Aborting to avoid deadlock\n", ret);
 405                                 exit(1);
 406                         }
 407                         printf("Process %d found\n", pid);
 408                         //control
 409                         limit_process(pid, limit);
 410                 }
 411                 if (lazy) {
 412                         printf("Giving up...\n");
 413                         break;
 414                 }
 415                 sleep(2);
 416         }
 417
 418         return 0;
 419 }
 420