src/cgroup.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   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 #include <fcntl.h>
  29
  30 #include "cgroup.h"
  31 #include "cgroup-util.h"
  32 #include "log.h"
  33
  34 int cgroup_bonding_realize(CGroupBonding *b) {
  35         int r;
  36
  37         assert(b);
  38         assert(b->path);
  39         assert(b->controller);
  40
  41         r = cg_create(b->controller, b->path);
  42         if (r < 0) {
  43                 log_warning("Failed to create cgroup %s:%s: %s", b->controller, b->path, strerror(-r));
  44                 return r;
  45         }
  46
  47         b->realized = true;
  48
  49         return 0;
  50 }
  51
  52 int cgroup_bonding_realize_list(CGroupBonding *first) {
  53         CGroupBonding *b;
  54         int r;
  55
  56         LIST_FOREACH(by_unit, b, first)
  57                 if ((r = cgroup_bonding_realize(b)) < 0 && b->essential)
  58                         return r;
  59
  60         return 0;
  61 }
  62
  63 void cgroup_bonding_free(CGroupBonding *b, bool trim) {
  64         assert(b);
  65
  66         if (b->unit) {
  67                 CGroupBonding *f;
  68
  69                 LIST_REMOVE(CGroupBonding, by_unit, b->unit->cgroup_bondings, b);
  70
  71                 if (streq(b->controller, SYSTEMD_CGROUP_CONTROLLER)) {
  72                         assert_se(f = hashmap_get(b->unit->manager->cgroup_bondings, b->path));
  73                         LIST_REMOVE(CGroupBonding, by_path, f, b);
  74
  75                         if (f)
  76                                 hashmap_replace(b->unit->manager->cgroup_bondings, b->path, f);
  77                         else
  78                                 hashmap_remove(b->unit->manager->cgroup_bondings, b->path);
  79                 }
  80         }
  81
  82         if (b->realized && b->ours && trim)
  83                 cg_trim(b->controller, b->path, false);
  84
  85         free(b->controller);
  86         free(b->path);
  87         free(b);
  88 }
  89
  90 void cgroup_bonding_free_list(CGroupBonding *first, bool remove_or_trim) {
  91         CGroupBonding *b, *n;
  92
  93         LIST_FOREACH_SAFE(by_unit, b, n, first)
  94                 cgroup_bonding_free(b, remove_or_trim);
  95 }
  96
  97 void cgroup_bonding_trim(CGroupBonding *b, bool delete_root) {
  98         assert(b);
  99
 100         if (b->realized && b->ours)
 101                 cg_trim(b->controller, b->path, delete_root);
 102 }
 103
 104 void cgroup_bonding_trim_list(CGroupBonding *first, bool delete_root) {
 105         CGroupBonding *b;
 106
 107         LIST_FOREACH(by_unit, b, first)
 108                 cgroup_bonding_trim(b, delete_root);
 109 }
 110
 111 int cgroup_bonding_install(CGroupBonding *b, pid_t pid) {
 112         int r;
 113
 114         assert(b);
 115         assert(pid >= 0);
 116
 117         if ((r = cg_create_and_attach(b->controller, b->path, pid)) < 0)
 118                 return r;
 119
 120         b->realized = true;
 121         return 0;
 122 }
 123
 124 int cgroup_bonding_install_list(CGroupBonding *first, pid_t pid) {
 125         CGroupBonding *b;
 126         int r;
 127
 128         LIST_FOREACH(by_unit, b, first)
 129                 if ((r = cgroup_bonding_install(b, pid)) < 0 && b->essential)
 130                         return r;
 131
 132         return 0;
 133 }
 134
 135 int cgroup_bonding_set_group_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid) {
 136         assert(b);
 137
 138         if (!b->realized)
 139                 return -EINVAL;
 140
 141         return cg_set_group_access(b->controller, b->path, mode, uid, gid);
 142 }
 143
 144 int cgroup_bonding_set_group_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid) {
 145         CGroupBonding *b;
 146         int r;
 147
 148         LIST_FOREACH(by_unit, b, first) {
 149                 r = cgroup_bonding_set_group_access(b, mode, uid, gid);
 150                 if (r < 0)
 151                         return r;
 152         }
 153
 154         return 0;
 155 }
 156
 157 int cgroup_bonding_set_task_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid, int sticky) {
 158         assert(b);
 159
 160         if (!b->realized)
 161                 return -EINVAL;
 162
 163         return cg_set_task_access(b->controller, b->path, mode, uid, gid, sticky);
 164 }
 165
 166 int cgroup_bonding_set_task_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid, int sticky) {
 167         CGroupBonding *b;
 168         int r;
 169
 170         LIST_FOREACH(by_unit, b, first) {
 171                 r = cgroup_bonding_set_task_access(b, mode, uid, gid, sticky);
 172                 if (r < 0)
 173                         return r;
 174         }
 175
 176         return 0;
 177 }
 178
 179 int cgroup_bonding_kill(CGroupBonding *b, int sig, bool sigcont, Set *s) {
 180         assert(b);
 181         assert(sig >= 0);
 182
 183         /* Don't kill cgroups that aren't ours */
 184         if (!b->ours)
 185                 return 0;
 186
 187         return cg_kill_recursive(b->controller, b->path, sig, sigcont, true, false, s);
 188 }
 189
 190 int cgroup_bonding_kill_list(CGroupBonding *first, int sig, bool sigcont, Set *s) {
 191         CGroupBonding *b;
 192         Set *allocated_set = NULL;
 193         int ret = -EAGAIN, r;
 194
 195         if (!first)
 196                 return 0;
 197
 198         if (!s)
 199                 if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
 200                         return -ENOMEM;
 201
 202         LIST_FOREACH(by_unit, b, first) {
 203                 if ((r = cgroup_bonding_kill(b, sig, sigcont, s)) < 0) {
 204                         if (r == -EAGAIN || r == -ESRCH)
 205                                 continue;
 206
 207                         ret = r;
 208                         goto finish;
 209                 }
 210
 211                 if (ret < 0 || r > 0)
 212                         ret = r;
 213         }
 214
 215 finish:
 216         if (allocated_set)
 217                 set_free(allocated_set);
 218
 219         return ret;
 220 }
 221
 222 /* Returns 1 if the group is empty, 0 if it is not, -EAGAIN if we
 223  * cannot know */
 224 int cgroup_bonding_is_empty(CGroupBonding *b) {
 225         int r;
 226
 227         assert(b);
 228
 229         if ((r = cg_is_empty_recursive(b->controller, b->path, true)) < 0)
 230                 return r;
 231
 232         /* If it is empty it is empty */
 233         if (r > 0)
 234                 return 1;
 235
 236         /* It's not only us using this cgroup, so we just don't know */
 237         return b->ours ? 0 : -EAGAIN;
 238 }
 239
 240 int cgroup_bonding_is_empty_list(CGroupBonding *first) {
 241         CGroupBonding *b;
 242
 243         LIST_FOREACH(by_unit, b, first) {
 244                 int r;
 245
 246                 if ((r = cgroup_bonding_is_empty(b)) < 0) {
 247                         /* If this returned -EAGAIN, then we don't know if the
 248                          * group is empty, so let's see if another group can
 249                          * tell us */
 250
 251                         if (r != -EAGAIN)
 252                                 return r;
 253                 } else
 254                         return r;
 255         }
 256
 257         return -EAGAIN;
 258 }
 259
 260 int manager_setup_cgroup(Manager *m) {
 261         char *current = NULL, *path = NULL;
 262         int r;
 263         char suffix[32];
 264
 265         assert(m);
 266
 267         /* 0. Be nice to Ingo Molnar #628004 */
 268         if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) {
 269                 log_warning("No control group support available, not creating root group.");
 270                 return 0;
 271         }
 272
 273         /* 1. Determine hierarchy */
 274         if ((r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &current)) < 0) {
 275                 log_error("Cannot determine cgroup we are running in: %s", strerror(-r));
 276                 goto finish;
 277         }
 278
 279         if (m->running_as == MANAGER_SYSTEM)
 280                 strcpy(suffix, "/system");
 281         else {
 282                 snprintf(suffix, sizeof(suffix), "/systemd-%lu", (unsigned long) getpid());
 283                 char_array_0(suffix);
 284         }
 285
 286         free(m->cgroup_hierarchy);
 287         if (endswith(current, suffix)) {
 288                 /* We probably got reexecuted and can continue to use our root cgroup */
 289                 m->cgroup_hierarchy = current;
 290                 current = NULL;
 291
 292         } else {
 293                 /* We need a new root cgroup */
 294                 m->cgroup_hierarchy = NULL;
 295                 if (asprintf(&m->cgroup_hierarchy, "%s%s", streq(current, "/") ? "" : current, suffix) < 0) {
 296                         log_error("Out of memory");
 297                         r = -ENOMEM;
 298                         goto finish;
 299                 }
 300         }
 301
 302         /* 2. Show data */
 303         if ((r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, NULL, &path)) < 0) {
 304                 log_error("Cannot find cgroup mount point: %s", strerror(-r));
 305                 goto finish;
 306         }
 307
 308         log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
 309
 310         /* 3. Install agent */
 311         if ((r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH)) < 0)
 312                 log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
 313         else if (r > 0)
 314                 log_debug("Installed release agent.");
 315         else
 316                 log_debug("Release agent already installed.");
 317
 318         /* 4. Realize the group */
 319         if ((r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, 0)) < 0) {
 320                 log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
 321                 goto finish;
 322         }
 323
 324         /* 5. And pin it, so that it cannot be unmounted */
 325         if (m->pin_cgroupfs_fd >= 0)
 326                 close_nointr_nofail(m->pin_cgroupfs_fd);
 327
 328         if ((m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK)) < 0) {
 329                 log_error("Failed to open pin file: %m");
 330                 r = -errno;
 331                 goto finish;
 332         }
 333
 334         log_debug("Created root group.");
 335
 336 finish:
 337         free(current);
 338         free(path);
 339
 340         return r;
 341 }
 342
 343 void manager_shutdown_cgroup(Manager *m, bool delete) {
 344         assert(m);
 345
 346         if (delete && m->cgroup_hierarchy)
 347                 cg_delete(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy);
 348
 349         if (m->pin_cgroupfs_fd >= 0) {
 350                 close_nointr_nofail(m->pin_cgroupfs_fd);
 351                 m->pin_cgroupfs_fd = -1;
 352         }
 353
 354         free(m->cgroup_hierarchy);
 355         m->cgroup_hierarchy = NULL;
 356 }
 357
 358 int cgroup_notify_empty(Manager *m, const char *group) {
 359         CGroupBonding *l, *b;
 360
 361         assert(m);
 362         assert(group);
 363
 364         l = hashmap_get(m->cgroup_bondings, group);
 365         if (!l)
 366                 return 0;
 367
 368         LIST_FOREACH(by_path, b, l) {
 369                 int t;
 370
 371                 if (!b->unit)
 372                         continue;
 373
 374                 t = cgroup_bonding_is_empty_list(b);
 375                 if (t < 0) {
 376
 377                         /* If we don't know, we don't know */
 378                         if (t != -EAGAIN)
 379                                 log_warning("Failed to check whether cgroup is empty: %s", strerror(errno));
 380
 381                         continue;
 382                 }
 383
 384                 if (t > 0) {
 385                         /* If it is empty, let's delete it */
 386                         cgroup_bonding_trim_list(b->unit->cgroup_bondings, true);
 387
 388                         if (UNIT_VTABLE(b->unit)->cgroup_notify_empty)
 389                                 UNIT_VTABLE(b->unit)->cgroup_notify_empty(b->unit);
 390                 }
 391         }
 392
 393         return 0;
 394 }
 395
 396 Unit* cgroup_unit_by_pid(Manager *m, pid_t pid) {
 397         CGroupBonding *l, *b;
 398         char *group = NULL;
 399
 400         assert(m);
 401
 402         if (pid <= 1)
 403                 return NULL;
 404
 405         if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &group) < 0)
 406                 return NULL;
 407
 408         l = hashmap_get(m->cgroup_bondings, group);
 409
 410         if (!l) {
 411                 char *slash;
 412
 413                 while ((slash = strrchr(group, '/'))) {
 414                         if (slash == group)
 415                                 break;
 416
 417                         *slash = 0;
 418
 419                         if ((l = hashmap_get(m->cgroup_bondings, group)))
 420                                 break;
 421                 }
 422         }
 423
 424         free(group);
 425
 426         LIST_FOREACH(by_path, b, l) {
 427
 428                 if (!b->unit)
 429                         continue;
 430
 431                 if (b->ours)
 432                         return b->unit;
 433         }
 434
 435         return NULL;
 436 }
 437
 438 CGroupBonding *cgroup_bonding_find_list(CGroupBonding *first, const char *controller) {
 439         CGroupBonding *b;
 440
 441         assert(controller);
 442
 443         LIST_FOREACH(by_unit, b, first)
 444                 if (streq(b->controller, controller))
 445                         return b;
 446
 447         return NULL;
 448 }
 449
 450 char *cgroup_bonding_to_string(CGroupBonding *b) {
 451         char *r;
 452
 453         assert(b);
 454
 455         if (asprintf(&r, "%s:%s", b->controller, b->path) < 0)
 456                 return NULL;
 457
 458         return r;
 459 }
 460
 461 pid_t cgroup_bonding_search_main_pid(CGroupBonding *b) {
 462         FILE *f;
 463         pid_t pid = 0, npid, mypid;
 464
 465         assert(b);
 466
 467         if (!b->ours)
 468                 return 0;
 469
 470         if (cg_enumerate_processes(b->controller, b->path, &f) < 0)
 471                 return 0;
 472
 473         mypid = getpid();
 474
 475         while (cg_read_pid(f, &npid) > 0)  {
 476                 pid_t ppid;
 477
 478                 if (npid == pid)
 479                         continue;
 480
 481                 /* Ignore processes that aren't our kids */
 482                 if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid)
 483                         continue;
 484
 485                 if (pid != 0) {
 486                         /* Dang, there's more than one daemonized PID
 487                         in this group, so we don't know what process
 488                         is the main process. */
 489                         pid = 0;
 490                         break;
 491                 }
 492
 493                 pid = npid;
 494         }
 495
 496         fclose(f);
 497
 498         return pid;
 499 }
 500
 501 pid_t cgroup_bonding_search_main_pid_list(CGroupBonding *first) {
 502         CGroupBonding *b;
 503         pid_t pid;
 504
 505         /* Try to find a main pid from this cgroup, but checking if
 506          * there's only one PID in the cgroup and returning it. Later
 507          * on we might want to add additional, smarter heuristics
 508          * here. */
 509
 510         LIST_FOREACH(by_unit, b, first)
 511                 if ((pid = cgroup_bonding_search_main_pid(b)) != 0)
 512                         return pid;
 513
 514         return 0;
 515
 516 }