1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2013 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
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 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
24 #include "path-util.h"
26 #include "cgroup-util.h"
29 void cgroup_context_init(CGroupContext *c) {
32 /* Initialize everything to the kernel defaults, assuming the
33 * structure is preinitialized to 0 */
36 c->memory_limit = c->memory_soft_limit = (uint64_t) -1;
37 c->blockio_weight = 1000;
40 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
44 LIST_REMOVE(CGroupDeviceAllow, device_allow, c->device_allow, a);
49 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
53 LIST_REMOVE(CGroupBlockIODeviceWeight, device_weights, c->blockio_device_weights, w);
58 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
62 LIST_REMOVE(CGroupBlockIODeviceBandwidth, device_bandwidths, c->blockio_device_bandwidths, b);
67 void cgroup_context_done(CGroupContext *c) {
70 while (c->blockio_device_weights)
71 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
73 while (c->blockio_device_bandwidths)
74 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
76 while (c->device_allow)
77 cgroup_context_free_device_allow(c, c->device_allow);
80 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
81 CGroupBlockIODeviceBandwidth *b;
82 CGroupBlockIODeviceWeight *w;
88 prefix = strempty(prefix);
91 "%sCPUAccounting=%s\n"
92 "%sBlockIOAccounting=%s\n"
93 "%sMemoryAccounting=%s\n"
95 "%sBlockIOWeight%lu\n"
96 "%sMemoryLimit=%" PRIu64 "\n"
97 "%sMemorySoftLimit=%" PRIu64 "\n"
98 "%sDevicePolicy=%s\n",
99 prefix, yes_no(c->cpu_accounting),
100 prefix, yes_no(c->blockio_accounting),
101 prefix, yes_no(c->memory_accounting),
102 prefix, c->cpu_shares,
103 prefix, c->blockio_weight,
104 prefix, c->memory_limit,
105 prefix, c->memory_soft_limit,
106 prefix, cgroup_device_policy_to_string(c->device_policy));
108 LIST_FOREACH(device_allow, a, c->device_allow)
110 "%sDeviceAllow=%s %s%s%s\n",
113 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
115 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
117 "%sBlockIODeviceWeight=%s %lu",
122 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
123 char buf[FORMAT_BYTES_MAX];
128 b->read ? "BlockIOReadBandwidth" : "BlockIOWriteBandwidth",
130 format_bytes(buf, sizeof(buf), b->bandwidth));
134 static int lookup_blkio_device(const char *p, dev_t *dev) {
143 log_warning("Couldn't stat device %s: %m", p);
147 if (S_ISBLK(st.st_mode))
149 else if (major(st.st_dev) != 0) {
150 /* If this is not a device node then find the block
151 * device this file is stored on */
154 /* If this is a partition, try to get the originating
156 block_get_whole_disk(*dev, dev);
158 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
165 static int whitelist_device(const char *path, const char *node, const char *acc) {
166 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
173 if (stat(node, &st) < 0) {
174 log_warning("Couldn't stat device %s", node);
178 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
179 log_warning("%s is not a device.", node);
185 S_ISCHR(st.st_mode) ? 'c' : 'b',
186 major(st.st_rdev), minor(st.st_rdev),
189 r = cg_set_attribute("devices", path, "devices.allow", buf);
191 log_warning("Failed to set devices.allow on %s: %s", path, strerror(-r));
196 void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path) {
205 if (mask & CGROUP_CPU) {
206 char buf[DECIMAL_STR_MAX(unsigned long) + 1];
208 sprintf(buf, "%lu\n", c->cpu_shares);
209 r = cg_set_attribute("cpu", path, "cpu.shares", buf);
211 log_warning("Failed to set cpu.shares on %s: %s", path, strerror(-r));
214 if (mask & CGROUP_BLKIO) {
215 char buf[MAX3(DECIMAL_STR_MAX(unsigned long)+1,
216 DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(unsigned long)*1,
217 DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)];
218 CGroupBlockIODeviceWeight *w;
219 CGroupBlockIODeviceBandwidth *b;
221 sprintf(buf, "%lu\n", c->blockio_weight);
222 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
224 log_warning("Failed to set blkio.weight on %s: %s", path, strerror(-r));
226 /* FIXME: no way to reset this list */
227 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
230 r = lookup_blkio_device(w->path, &dev);
234 sprintf(buf, "%u:%u %lu", major(dev), minor(dev), w->weight);
235 r = cg_set_attribute("blkio", path, "blkio.weight_device", buf);
237 log_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r));
240 /* FIXME: no way to reset this list */
241 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
245 r = lookup_blkio_device(b->path, &dev);
249 a = b->read ? "blkio.throttle.read_bps_device" : "blkio.throttle.write_bps_device";
251 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), b->bandwidth);
252 r = cg_set_attribute("blkio", path, a, buf);
254 log_error("Failed to set %s on %s: %s", a, path, strerror(-r));
258 if (mask & CGROUP_MEMORY) {
259 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
261 sprintf(buf, "%" PRIu64 "\n", c->memory_limit);
262 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
264 log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r));
266 sprintf(buf, "%" PRIu64 "\n", c->memory_soft_limit);
267 cg_set_attribute("memory", path, "memory.soft_limit_in_bytes", buf);
269 log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r));
272 if (mask & CGROUP_DEVICE) {
273 CGroupDeviceAllow *a;
275 if (c->device_allow || c->device_policy != CGROUP_AUTO)
276 r = cg_set_attribute("devices", path, "devices.deny", "a");
278 r = cg_set_attribute("devices", path, "devices.allow", "a");
280 log_error("Failed to reset devices.list on %s: %s", path, strerror(-r));
282 if (c->device_policy == CGROUP_CLOSED ||
283 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
284 static const char auto_devices[] =
288 "/dev/random\0" "rw\0"
289 "/dev/urandom\0" "rw\0";
293 NULSTR_FOREACH_PAIR(x, y, auto_devices)
294 whitelist_device(path, x, y);
297 LIST_FOREACH(device_allow, a, c->device_allow) {
312 whitelist_device(path, a->path, acc);
317 CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) {
318 CGroupControllerMask mask = 0;
320 /* Figure out which controllers we need */
322 if (c->cpu_accounting || c->cpu_shares != 1024)
323 mask |= CGROUP_CPUACCT | CGROUP_CPU;
325 if (c->blockio_accounting ||
326 c->blockio_weight != 1000 ||
327 c->blockio_device_weights ||
328 c->blockio_device_bandwidths)
329 mask |= CGROUP_BLKIO;
331 if (c->memory_accounting ||
332 c->memory_limit != (uint64_t) -1 ||
333 c->memory_soft_limit != (uint64_t) -1)
334 mask |= CGROUP_MEMORY;
336 if (c->device_allow || c->device_policy != CGROUP_AUTO)
337 mask |= CGROUP_DEVICE;
342 static CGroupControllerMask unit_get_cgroup_mask(Unit *u) {
345 c = unit_get_cgroup_context(u);
349 return cgroup_context_get_mask(c);
352 static CGroupControllerMask unit_get_members_mask(Unit *u) {
353 CGroupControllerMask mask = 0;
359 SET_FOREACH(m, u->dependencies[UNIT_BEFORE], i) {
361 if (UNIT_DEREF(m->slice) != u)
364 mask |= unit_get_cgroup_mask(m) | unit_get_members_mask(m);
370 static CGroupControllerMask unit_get_siblings_mask(Unit *u) {
373 if (!UNIT_ISSET(u->slice))
376 /* Sibling propagation is only relevant for weight-based
377 * controllers, so let's mask out everything else */
378 return unit_get_members_mask(UNIT_DEREF(u->slice)) &
379 (CGROUP_CPU|CGROUP_BLKIO|CGROUP_CPUACCT);
382 static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) {
388 path = unit_default_cgroup_path(u);
392 r = hashmap_put(u->manager->cgroup_unit, path, u);
396 /* First, create our own group */
397 r = cg_create_with_mask(mask, path);
399 log_error("Failed to create cgroup %s: %s", path, strerror(-r));
401 /* Then, possibly move things over */
402 if (u->cgroup_path && !streq(path, u->cgroup_path)) {
403 r = cg_migrate_with_mask(mask, u->cgroup_path, path);
405 log_error("Failed to migrate cgroup %s: %s", path, strerror(-r));
408 /* And remember the new data */
409 free(u->cgroup_path);
410 u->cgroup_path = path;
411 u->cgroup_realized = true;
412 u->cgroup_mask = mask;
417 static int unit_realize_cgroup_now(Unit *u) {
418 CGroupControllerMask mask;
422 if (u->in_cgroup_queue) {
423 LIST_REMOVE(Unit, cgroup_queue, u->manager->cgroup_queue, u);
424 u->in_cgroup_queue = false;
427 mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
428 mask &= u->manager->cgroup_supported;
430 if (u->cgroup_realized &&
431 u->cgroup_mask == mask)
434 /* First, realize parents */
435 if (UNIT_ISSET(u->slice))
436 unit_realize_cgroup_now(UNIT_DEREF(u->slice));
438 /* And then do the real work */
439 return unit_create_cgroups(u, mask);
442 static void unit_add_to_cgroup_queue(Unit *u) {
444 if (u->in_cgroup_queue)
447 LIST_PREPEND(Unit, cgroup_queue, u->manager->cgroup_queue, u);
448 u->in_cgroup_queue = true;
451 unsigned manager_dispatch_cgroup_queue(Manager *m) {
455 while ((i = m->cgroup_queue)) {
456 assert(i->in_cgroup_queue);
458 if (unit_realize_cgroup_now(i) >= 0)
459 cgroup_context_apply(unit_get_cgroup_context(i), i->cgroup_mask, i->cgroup_path);
467 static void unit_queue_siblings(Unit *u) {
470 /* This adds the siblings of the specified unit and the
471 * siblings of all parent units to the cgroup queue. (But
472 * neither the specified unit itself nor the parents.) */
474 while ((slice = UNIT_DEREF(u->slice))) {
478 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
482 if (UNIT_DEREF(m->slice) != slice)
485 unit_add_to_cgroup_queue(m);
492 int unit_realize_cgroup(Unit *u) {
498 c = unit_get_cgroup_context(u);
502 /* So, here's the deal: when realizing the cgroups for this
503 * unit, we need to first create all parents, but there's more
504 * actually: for the weight-based controllers we also need to
505 * make sure that all our siblings (i.e. units that are in the
506 * same slice as we are) have cgroup too. Otherwise things
507 * would become very uneven as each of their processes would
508 * get as much resources as all our group together. This call
509 * will synchronously create the parent cgroups, but will
510 * defer work on the siblings to the next event loop
513 /* Add all sibling slices to the cgroup queue. */
514 unit_queue_siblings(u);
516 /* And realize this one now */
517 r = unit_realize_cgroup_now(u);
519 /* And apply the values */
521 cgroup_context_apply(c, u->cgroup_mask, u->cgroup_path);
526 void unit_destroy_cgroup(Unit *u) {
534 r = cg_trim_with_mask(u->cgroup_mask, u->cgroup_path, !unit_has_name(u, SPECIAL_ROOT_SLICE));
536 log_debug("Failed to destroy cgroup %s: %s", u->cgroup_path, strerror(-r));
538 hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
540 free(u->cgroup_path);
541 u->cgroup_path = NULL;
542 u->cgroup_realized = false;
547 pid_t unit_search_main_pid(Unit *u) {
548 _cleanup_fclose_ FILE *f = NULL;
549 pid_t pid = 0, npid, mypid;
556 if (cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f) < 0)
560 while (cg_read_pid(f, &npid) > 0) {
566 /* Ignore processes that aren't our kids */
567 if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid)
571 /* Dang, there's more than one daemonized PID
572 in this group, so we don't know what process
573 is the main process. */
584 int manager_setup_cgroup(Manager *m) {
585 _cleanup_free_ char *path = NULL;
591 /* 0. Be nice to Ingo Molnar #628004 */
592 if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) {
593 log_warning("No control group support available, not creating root group.");
597 /* 1. Determine hierarchy */
598 free(m->cgroup_root);
599 m->cgroup_root = NULL;
601 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
603 log_error("Cannot determine cgroup we are running in: %s", strerror(-r));
607 /* Already in /system.slice? If so, let's cut this off again */
608 if (m->running_as == SYSTEMD_SYSTEM) {
609 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
614 /* And make sure to store away the root value without trailing
615 * slash, even for the root dir, so that we can easily prepend
617 if (streq(m->cgroup_root, "/"))
618 m->cgroup_root[0] = 0;
621 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
623 log_error("Cannot find cgroup mount point: %s", strerror(-r));
627 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
629 /* 3. Install agent */
630 if (m->running_as == SYSTEMD_SYSTEM) {
631 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
633 log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
635 log_debug("Installed release agent.");
637 log_debug("Release agent already installed.");
640 /* 4. Realize the system slice and put us in there */
641 if (m->running_as == SYSTEMD_SYSTEM) {
642 a = strappenda(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
643 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, a, 0);
645 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, 0);
647 log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
651 /* 5. And pin it, so that it cannot be unmounted */
652 if (m->pin_cgroupfs_fd >= 0)
653 close_nointr_nofail(m->pin_cgroupfs_fd);
655 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
657 log_error("Failed to open pin file: %m");
661 /* 6. Figure out which controllers are supported */
662 m->cgroup_supported = cg_mask_supported();
667 void manager_shutdown_cgroup(Manager *m, bool delete) {
670 /* We can't really delete the group, since we are in it. But
672 if (delete && m->cgroup_root)
673 cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
675 if (m->pin_cgroupfs_fd >= 0) {
676 close_nointr_nofail(m->pin_cgroupfs_fd);
677 m->pin_cgroupfs_fd = -1;
680 free(m->cgroup_root);
681 m->cgroup_root = NULL;
684 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
691 u = hashmap_get(m->cgroup_unit, cgroup);
705 u = hashmap_get(m->cgroup_unit, p);
711 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
712 _cleanup_free_ char *cgroup = NULL;
720 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
724 return manager_get_unit_by_cgroup(m, cgroup);
727 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
734 u = manager_get_unit_by_cgroup(m, cgroup);
736 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, true);
738 if (UNIT_VTABLE(u)->notify_cgroup_empty)
739 UNIT_VTABLE(u)->notify_cgroup_empty(u);
741 unit_add_to_gc_queue(u);
748 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
749 [CGROUP_AUTO] = "auto",
750 [CGROUP_CLOSED] = "closed",
751 [CGROUP_STRICT] = "strict",
754 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);