cgroup.c

   1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU General Public License as published by
  10   the Free Software Foundation; either version 2 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   General Public License for more details.
  17
  18   You should have received a copy of the GNU General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <errno.h>
  23 #include <assert.h>
  24 #include <unistd.h>
  25 #include <sys/types.h>
  26 #include <signal.h>
  27 #include <sys/mount.h>
  28
  29 #include "cgroup.h"
  30 #include "log.h"
  31
  32 static int translate_error(int error, int _errno) {
  33
  34         switch (error) {
  35
  36         case ECGROUPNOTCOMPILED:
  37         case ECGROUPNOTMOUNTED:
  38         case ECGROUPNOTEXIST:
  39         case ECGROUPNOTCREATED:
  40                 return -ENOENT;
  41
  42         case ECGINVAL:
  43                 return -EINVAL;
  44
  45         case ECGROUPNOTALLOWED:
  46                 return -EPERM;
  47
  48         case ECGOTHER:
  49                 return -_errno;
  50         }
  51
  52         return -EIO;
  53 }
  54
  55 int cgroup_bonding_realize(CGroupBonding *b) {
  56         int r;
  57
  58         assert(b);
  59         assert(b->path);
  60         assert(b->controller);
  61
  62         if (b->cgroup)
  63                 return 0;
  64
  65         if (!(b->cgroup = cgroup_new_cgroup(b->path)))
  66                 return -ENOMEM;
  67
  68         if (!cgroup_add_controller(b->cgroup, b->controller)) {
  69                 r = -ENOMEM;
  70                 goto fail;
  71         }
  72
  73         if (b->inherit)
  74                 r = cgroup_create_cgroup_from_parent(b->cgroup, true);
  75         else
  76                 r = cgroup_create_cgroup(b->cgroup, true);
  77
  78         if (r != 0) {
  79                 r = translate_error(r, errno);
  80                 goto fail;
  81         }
  82
  83         return 0;
  84
  85 fail:
  86         cgroup_free(&b->cgroup);
  87         b->cgroup = NULL;
  88         return r;
  89 }
  90
  91 int cgroup_bonding_realize_list(CGroupBonding *first) {
  92         CGroupBonding *b;
  93
  94         LIST_FOREACH(by_unit, b, first) {
  95                 int r;
  96
  97                 if ((r = cgroup_bonding_realize(b)) < 0)
  98                         return r;
  99         }
 100
 101         return 0;
 102 }
 103
 104 void cgroup_bonding_free(CGroupBonding *b) {
 105         assert(b);
 106
 107         if (b->unit) {
 108                 CGroupBonding *f;
 109
 110                 LIST_REMOVE(CGroupBonding, by_unit, b->unit->meta.cgroup_bondings, b);
 111
 112                 assert_se(f = hashmap_get(b->unit->meta.manager->cgroup_bondings, b->path));
 113                 LIST_REMOVE(CGroupBonding, by_path, f, b);
 114
 115                 if (f)
 116                         hashmap_replace(b->unit->meta.manager->cgroup_bondings, b->path, f);
 117                 else
 118                         hashmap_remove(b->unit->meta.manager->cgroup_bondings, b->path);
 119         }
 120
 121         free(b->controller);
 122         free(b->path);
 123
 124         if (b->cgroup) {
 125
 126                 if (b->only_us && b->clean_up)
 127                         cgroup_delete_cgroup(b->cgroup, true);
 128
 129                 cgroup_free(&b->cgroup);
 130         }
 131
 132         free(b);
 133 }
 134
 135 void cgroup_bonding_free_list(CGroupBonding *first) {
 136         CGroupBonding *b, *n;
 137
 138         LIST_FOREACH_SAFE(by_unit, b, n, first)
 139                 cgroup_bonding_free(b);
 140 }
 141
 142 int cgroup_bonding_install(CGroupBonding *b, pid_t pid) {
 143         int r;
 144
 145         assert(b);
 146         assert(pid >= 0);
 147
 148         if (pid == 0)
 149                 pid = getpid();
 150
 151         if (!b->cgroup)
 152                 return -ENOENT;
 153
 154         if ((r = cgroup_attach_task_pid(b->cgroup, pid)))
 155                 return translate_error(r, errno);
 156
 157         return 0;
 158 }
 159
 160 int cgroup_bonding_install_list(CGroupBonding *first, pid_t pid) {
 161         CGroupBonding *b;
 162
 163         LIST_FOREACH(by_unit, b, first) {
 164                 int r;
 165
 166                 if ((r = cgroup_bonding_install(b, pid)) < 0)
 167                         return r;
 168         }
 169
 170         return 0;
 171 }
 172
 173 int cgroup_bonding_kill(CGroupBonding *b, int sig) {
 174         int r;
 175         Set *s;
 176         bool done;
 177
 178         assert(b);
 179         assert(sig > 0);
 180
 181         if (!(s = set_new(trivial_hash_func, trivial_compare_func)))
 182                 return -ENOMEM;
 183
 184         do {
 185                 void *iterator;
 186                 pid_t pid;
 187
 188                 done = true;
 189
 190                 if ((r = cgroup_get_task_begin(b->path, b->controller, &iterator, &pid)) != 0) {
 191                         if (r == ECGEOF) {
 192                                 r = 0;
 193                                 goto kill_done;
 194                         } else {
 195                                 r = translate_error(r, errno);
 196                                 break;
 197                         }
 198                 }
 199
 200                 for (;;) {
 201                         if (set_get(s, INT_TO_PTR(pid)) != INT_TO_PTR(pid)) {
 202
 203                                 /* If we haven't killed this process
 204                                  * yet, kill it */
 205
 206                                 if (kill(pid, sig) < 0 && errno != ESRCH) {
 207                                         r = -errno;
 208                                         break;
 209                                 }
 210
 211                                 done = false;
 212
 213                                 if ((r = set_put(s, INT_TO_PTR(pid))) < 0)
 214                                     break;
 215                         }
 216
 217                         if ((r = cgroup_get_task_next(&iterator, &pid)) != 0) {
 218
 219                                 if (r == ECGEOF)
 220                                         r = 0;
 221                                 else
 222                                         r = translate_error(r, errno);
 223
 224                                 break;
 225                         }
 226                 }
 227
 228         kill_done:
 229                 assert_se(cgroup_get_task_end(&iterator) == 0);
 230
 231                 /* To avoid racing against processes which fork
 232                  * quicker than we can kill them we repeat this until
 233                  * no new pids need to be killed. */
 234
 235         } while (!done && r >= 0);
 236
 237         set_free(s);
 238         return r;
 239 }
 240
 241 int cgroup_bonding_kill_list(CGroupBonding *first, int sig) {
 242         CGroupBonding *b;
 243
 244         LIST_FOREACH(by_unit, b, first) {
 245                 int r;
 246
 247                 if ((r = cgroup_bonding_kill(b, sig)) < 0)
 248                         return r;
 249         }
 250
 251         return 0;
 252 }
 253
 254 /* Returns 1 if the group is empty, 0 if it is not, -EAGAIN if we
 255  * cannot know */
 256 int cgroup_bonding_is_empty(CGroupBonding *b) {
 257         void *iterator;
 258         pid_t pid;
 259         int r;
 260
 261         assert(b);
 262
 263         r = cgroup_get_task_begin(b->path, b->controller, &iterator, &pid);
 264
 265         if (r == 0 || r == ECGEOF)
 266                 cgroup_get_task_end(&iterator);
 267
 268         /* Hmm, no PID in this group? Then it is definitely empty */
 269         if (r == ECGEOF)
 270                 return 1;
 271
 272         /* Some error? Let's return it */
 273         if (r != 0)
 274                 return translate_error(r, errno);
 275
 276         /* It's not empty, and we are the only user, then it is
 277          * definitely not empty */
 278         if (b->only_us)
 279                 return 0;
 280
 281         /* There are PIDs in the group but we aren't the only users,
 282          * hence we cannot say */
 283         return -EAGAIN;
 284 }
 285
 286 int cgroup_bonding_is_empty_list(CGroupBonding *first) {
 287         CGroupBonding *b;
 288
 289         LIST_FOREACH(by_unit, b, first) {
 290                 int r;
 291
 292                 if ((r = cgroup_bonding_is_empty(b)) < 0) {
 293                         /* If this returned -EAGAIN, then we don't know if the
 294                          * group is empty, so let's see if another group can
 295                          * tell us */
 296
 297                         if (r != -EAGAIN)
 298                                 return r;
 299                 } else
 300                         return r;
 301         }
 302
 303         return -EAGAIN;
 304 }
 305
 306 static int install_release_agent(Manager *m, const char *mount_point) {
 307         char *p, *c, *sc;
 308         int r;
 309
 310         assert(m);
 311         assert(mount_point);
 312
 313         if (asprintf(&p, "%s/release_agent", mount_point) < 0)
 314                 return -ENOMEM;
 315
 316         if ((r = read_one_line_file(p, &c)) < 0) {
 317                 free(p);
 318                 return r;
 319         }
 320
 321         sc = strstrip(c);
 322
 323         if (sc[0] == 0) {
 324                 if ((r = write_one_line_file(p, CGROUP_AGENT_PATH "\n" )) < 0) {
 325                         free(p);
 326                         free(c);
 327                         return r;
 328                 }
 329         } else if (!streq(sc, CGROUP_AGENT_PATH)) {
 330                 free(p);
 331                 free(c);
 332                 return -EEXIST;
 333         }
 334
 335         free(c);
 336         free(p);
 337
 338         if (asprintf(&p, "%s/notify_on_release", mount_point) < 0)
 339                 return -ENOMEM;
 340
 341         if ((r = read_one_line_file(p, &c)) < 0) {
 342                 free(p);
 343                 return r;
 344         }
 345
 346         sc = strstrip(c);
 347
 348         if (streq(sc, "0")) {
 349                 if ((r = write_one_line_file(p, "1\n")) < 0) {
 350                         free(p);
 351                         free(c);
 352                         return r;
 353                 }
 354         } else if (!streq(sc, "1")) {
 355                 free(p);
 356                 free(c);
 357                 return -EIO;
 358         }
 359
 360         return 0;
 361 }
 362
 363 static int create_hierarchy_cgroup(Manager *m) {
 364         struct cgroup *cg;
 365         int r;
 366
 367         assert(m);
 368
 369         if (!(cg = cgroup_new_cgroup(m->cgroup_hierarchy)))
 370                 return -ENOMEM;
 371
 372         if (!(cgroup_add_controller(cg, m->cgroup_controller))) {
 373                 r = -ENOMEM;
 374                 goto finish;
 375         }
 376
 377         if ((r = cgroup_create_cgroup(cg, true)) != 0) {
 378                 log_error("Failed to create cgroup hierarchy group: %s", cgroup_strerror(r));
 379                 r = translate_error(r, errno);
 380                 goto finish;
 381         }
 382
 383         if ((r = cgroup_attach_task(cg)) != 0) {
 384                 log_error("Failed to add ourselves to hierarchy group: %s", cgroup_strerror(r));
 385                 r = translate_error(r, errno);
 386                 goto finish;
 387         }
 388
 389         r = 0;
 390
 391 finish:
 392         cgroup_free(&cg);
 393         return r;
 394 }
 395
 396 int manager_setup_cgroup(Manager *m) {
 397         char *mp, *cp;
 398         int r;
 399         pid_t pid;
 400
 401         assert(m);
 402
 403         if ((r = cgroup_init()) != 0) {
 404                 log_error("Failed to initialize libcg: %s", cgroup_strerror(r));
 405                 return translate_error(r, errno);
 406         }
 407
 408         free(m->cgroup_controller);
 409         if (!(m->cgroup_controller = strdup("debug")))
 410                 return -ENOMEM;
 411
 412         if ((r = cgroup_get_subsys_mount_point(m->cgroup_controller, &mp)))
 413                 return translate_error(r, errno);
 414
 415         pid = getpid();
 416
 417         if ((r = cgroup_get_current_controller_path(pid, m->cgroup_controller, &cp))) {
 418                 free(mp);
 419                 return translate_error(r, errno);
 420         }
 421
 422         free(m->cgroup_hierarchy);
 423         m->cgroup_hierarchy = NULL;
 424         if (asprintf(&m->cgroup_hierarchy, "%s/systemd-%llu", strcmp(cp, "/") == 0 ? "" : cp, (unsigned long long) pid) < 0) {
 425                 free(cp);
 426                 free(mp);
 427                 return -ENOMEM;
 428         }
 429
 430         log_info("Using cgroup controller <%s>, hierarchy mounted at <%s>, using root group <%s>.",
 431                  m->cgroup_controller,
 432                  mp,
 433                  m->cgroup_hierarchy);
 434
 435         if ((r = install_release_agent(m, mp)) < 0)
 436                 log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
 437         else
 438                 log_info("Installed release agent, or already installed.");
 439
 440         free(mp);
 441         free(cp);
 442
 443         if ((r = create_hierarchy_cgroup(m)) < 0)
 444                 log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
 445         else
 446                 log_info("Created root group.");
 447
 448         return r;
 449 }
 450
 451 int manager_shutdown_cgroup(Manager *m) {
 452         struct cgroup *cg;
 453         int r;
 454
 455         assert(m);
 456
 457         if (!m->cgroup_hierarchy)
 458                 return 0;
 459
 460         if (!(cg = cgroup_new_cgroup(m->cgroup_hierarchy)))
 461                 return -ENOMEM;
 462
 463         if (!(cgroup_add_controller(cg, m->cgroup_controller))) {
 464                 r = -ENOMEM;
 465                 goto finish;
 466         }
 467
 468         if ((r = cgroup_delete_cgroup_ext(cg, CGFLAG_DELETE_IGNORE_MIGRATION|CGFLAG_DELETE_RECURSIVE)) != 0) {
 469                 log_error("Failed to delete cgroup hierarchy group: %s", cgroup_strerror(r));
 470                 r = translate_error(r, errno);
 471                 goto finish;
 472         }
 473         r = 0;
 474
 475 finish:
 476         cgroup_free(&cg);
 477         return r;
 478
 479 }
 480
 481 int cgroup_notify_empty(Manager *m, const char *group) {
 482         CGroupBonding *l, *b;
 483
 484         assert(m);
 485         assert(group);
 486
 487         if (!(l = hashmap_get(m->cgroup_bondings, group)))
 488                 return 0;
 489
 490         LIST_FOREACH(by_path, b, l) {
 491                 int t;
 492
 493                 if (!b->unit)
 494                         continue;
 495
 496                 if ((t = cgroup_bonding_is_empty_list(b)) < 0) {
 497
 498                         /* If we don't know, we don't know */
 499                         if (t != -EAGAIN)
 500                                 log_warning("Failed to check whether cgroup is empty: %s", strerror(errno));
 501
 502                         continue;
 503                 }
 504
 505                 if (t > 0)
 506                         if (UNIT_VTABLE(b->unit)->cgroup_notify_empty)
 507                                 UNIT_VTABLE(b->unit)->cgroup_notify_empty(b->unit);
 508         }
 509
 510         return 0;
 511 }
 512
 513 CGroupBonding *cgroup_bonding_find_list(CGroupBonding *first, const char *controller) {
 514         CGroupBonding *b;
 515
 516         assert(controller);
 517
 518         LIST_FOREACH(by_unit, b, first)
 519                 if (streq(b->controller, controller))
 520                         return b;
 521
 522         return NULL;
 523 }