/***
This file is part of systemd.
- Copyright 2010 Lennart Poettering
+ Copyright 2013 Lennart Poettering
systemd is free software; you can redistribute it and/or modify it
under the terms of the GNU Lesser General Public License as published by
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
-#include <errno.h>
-#include <assert.h>
-#include <unistd.h>
-#include <sys/types.h>
-#include <signal.h>
-#include <sys/mount.h>
#include <fcntl.h>
+#include <fnmatch.h>
-#include "cgroup.h"
-#include "cgroup-util.h"
-#include "log.h"
-#include "strv.h"
#include "path-util.h"
+#include "special.h"
+#include "cgroup-util.h"
+#include "cgroup.h"
-int cgroup_bonding_realize(CGroupBonding *b) {
- int r;
+#define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
- assert(b);
- assert(b->path);
- assert(b->controller);
+void cgroup_context_init(CGroupContext *c) {
+ assert(c);
- r = cg_create(b->controller, b->path);
- if (r < 0) {
- log_warning("Failed to create cgroup %s:%s: %s", b->controller, b->path, strerror(-r));
- return r;
- }
+ /* Initialize everything to the kernel defaults, assuming the
+ * structure is preinitialized to 0 */
- b->realized = true;
+ c->cpu_shares = (unsigned long) -1;
+ c->startup_cpu_shares = (unsigned long) -1;
+ c->memory_limit = (uint64_t) -1;
+ c->blockio_weight = (unsigned long) -1;
+ c->startup_blockio_weight = (unsigned long) -1;
- return 0;
+ c->cpu_quota_per_sec_usec = USEC_INFINITY;
}
-int cgroup_bonding_realize_list(CGroupBonding *first) {
- CGroupBonding *b;
- int r;
+void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
+ assert(c);
+ assert(a);
- LIST_FOREACH(by_unit, b, first)
- if ((r = cgroup_bonding_realize(b)) < 0 && b->essential)
- return r;
+ LIST_REMOVE(device_allow, c->device_allow, a);
+ free(a->path);
+ free(a);
+}
- return 0;
+void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
+ assert(c);
+ assert(w);
+
+ LIST_REMOVE(device_weights, c->blockio_device_weights, w);
+ free(w->path);
+ free(w);
}
-void cgroup_bonding_free(CGroupBonding *b, bool trim) {
+void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
+ assert(c);
assert(b);
- if (b->unit) {
- CGroupBonding *f;
+ LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
+ free(b->path);
+ free(b);
+}
- LIST_REMOVE(CGroupBonding, by_unit, b->unit->cgroup_bondings, b);
+void cgroup_context_done(CGroupContext *c) {
+ assert(c);
- if (streq(b->controller, SYSTEMD_CGROUP_CONTROLLER)) {
- assert_se(f = hashmap_get(b->unit->manager->cgroup_bondings, b->path));
- LIST_REMOVE(CGroupBonding, by_path, f, b);
+ while (c->blockio_device_weights)
+ cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
- if (f)
- hashmap_replace(b->unit->manager->cgroup_bondings, b->path, f);
- else
- hashmap_remove(b->unit->manager->cgroup_bondings, b->path);
- }
- }
-
- if (b->realized && b->ours && trim)
- cg_trim(b->controller, b->path, false);
+ while (c->blockio_device_bandwidths)
+ cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
- free(b->controller);
- free(b->path);
- free(b);
+ while (c->device_allow)
+ cgroup_context_free_device_allow(c, c->device_allow);
}
-void cgroup_bonding_free_list(CGroupBonding *first, bool remove_or_trim) {
- CGroupBonding *b, *n;
-
- LIST_FOREACH_SAFE(by_unit, b, n, first)
- cgroup_bonding_free(b, remove_or_trim);
+void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
+ CGroupBlockIODeviceBandwidth *b;
+ CGroupBlockIODeviceWeight *w;
+ CGroupDeviceAllow *a;
+ char u[FORMAT_TIMESPAN_MAX];
+
+ assert(c);
+ assert(f);
+
+ prefix = strempty(prefix);
+
+ fprintf(f,
+ "%sCPUAccounting=%s\n"
+ "%sBlockIOAccounting=%s\n"
+ "%sMemoryAccounting=%s\n"
+ "%sCPUShares=%lu\n"
+ "%sStartupCPUShares=%lu\n"
+ "%sCPUQuotaPerSecSec=%s\n"
+ "%sBlockIOWeight=%lu\n"
+ "%sStartupBlockIOWeight=%lu\n"
+ "%sMemoryLimit=%" PRIu64 "\n"
+ "%sDevicePolicy=%s\n"
+ "%sDelegate=%s\n",
+ prefix, yes_no(c->cpu_accounting),
+ prefix, yes_no(c->blockio_accounting),
+ prefix, yes_no(c->memory_accounting),
+ prefix, c->cpu_shares,
+ prefix, c->startup_cpu_shares,
+ prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
+ prefix, c->blockio_weight,
+ prefix, c->startup_blockio_weight,
+ prefix, c->memory_limit,
+ prefix, cgroup_device_policy_to_string(c->device_policy),
+ prefix, yes_no(c->delegate));
+
+ LIST_FOREACH(device_allow, a, c->device_allow)
+ fprintf(f,
+ "%sDeviceAllow=%s %s%s%s\n",
+ prefix,
+ a->path,
+ a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
+
+ LIST_FOREACH(device_weights, w, c->blockio_device_weights)
+ fprintf(f,
+ "%sBlockIODeviceWeight=%s %lu",
+ prefix,
+ w->path,
+ w->weight);
+
+ LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
+ char buf[FORMAT_BYTES_MAX];
+
+ fprintf(f,
+ "%s%s=%s %s\n",
+ prefix,
+ b->read ? "BlockIOReadBandwidth" : "BlockIOWriteBandwidth",
+ b->path,
+ format_bytes(buf, sizeof(buf), b->bandwidth));
+ }
}
-void cgroup_bonding_trim(CGroupBonding *b, bool delete_root) {
- assert(b);
+static int lookup_blkio_device(const char *p, dev_t *dev) {
+ struct stat st;
+ int r;
- if (b->realized && b->ours)
- cg_trim(b->controller, b->path, delete_root);
-}
+ assert(p);
+ assert(dev);
-void cgroup_bonding_trim_list(CGroupBonding *first, bool delete_root) {
- CGroupBonding *b;
+ r = stat(p, &st);
+ if (r < 0)
+ return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
+
+ if (S_ISBLK(st.st_mode))
+ *dev = st.st_rdev;
+ else if (major(st.st_dev) != 0) {
+ /* If this is not a device node then find the block
+ * device this file is stored on */
+ *dev = st.st_dev;
+
+ /* If this is a partition, try to get the originating
+ * block device */
+ block_get_whole_disk(*dev, dev);
+ } else {
+ log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
+ return -ENODEV;
+ }
- LIST_FOREACH(by_unit, b, first)
- cgroup_bonding_trim(b, delete_root);
+ return 0;
}
-int cgroup_bonding_install(CGroupBonding *b, pid_t pid, const char *cgroup_suffix) {
- _cleanup_free_ char *p = NULL;
- const char *path;
+static int whitelist_device(const char *path, const char *node, const char *acc) {
+ char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
+ struct stat st;
int r;
- assert(b);
- assert(pid >= 0);
+ assert(path);
+ assert(acc);
+
+ if (stat(node, &st) < 0) {
+ log_warning("Couldn't stat device %s", node);
+ return -errno;
+ }
- if (cgroup_suffix) {
- p = strjoin(b->path, "/", cgroup_suffix, NULL);
- if (!p)
- return -ENOMEM;
+ if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
+ log_warning("%s is not a device.", node);
+ return -ENODEV;
+ }
- path = p;
- } else
- path = b->path;
+ sprintf(buf,
+ "%c %u:%u %s",
+ S_ISCHR(st.st_mode) ? 'c' : 'b',
+ major(st.st_rdev), minor(st.st_rdev),
+ acc);
- r = cg_create_and_attach(b->controller, path, pid);
+ r = cg_set_attribute("devices", path, "devices.allow", buf);
if (r < 0)
- return r;
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set devices.allow on %s: %s", path, strerror(-r));
- b->realized = true;
- return 0;
+ return r;
}
-int cgroup_bonding_install_list(CGroupBonding *first, pid_t pid, const char *cgroup_suffix) {
- CGroupBonding *b;
+static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
+ _cleanup_fclose_ FILE *f = NULL;
+ char line[LINE_MAX];
+ bool good = false;
int r;
- LIST_FOREACH(by_unit, b, first) {
- r = cgroup_bonding_install(b, pid, cgroup_suffix);
- if (r < 0 && b->essential)
- return r;
- }
+ assert(path);
+ assert(acc);
+ assert(type == 'b' || type == 'c');
- return 0;
-}
+ f = fopen("/proc/devices", "re");
+ if (!f)
+ return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
-int cgroup_bonding_migrate(CGroupBonding *b, CGroupBonding *list) {
- CGroupBonding *q;
- int ret = 0;
+ FOREACH_LINE(line, f, goto fail) {
+ char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
+ unsigned maj;
- LIST_FOREACH(by_unit, q, list) {
- int r;
+ truncate_nl(line);
- if (q == b)
+ if (type == 'c' && streq(line, "Character devices:")) {
+ good = true;
continue;
+ }
- if (!q->ours)
+ if (type == 'b' && streq(line, "Block devices:")) {
+ good = true;
continue;
+ }
- r = cg_migrate_recursive(q->controller, q->path, b->controller, b->path, true, false);
- if (r < 0 && ret == 0)
- ret = r;
- }
+ if (isempty(line)) {
+ good = false;
+ continue;
+ }
- return ret;
-}
+ if (!good)
+ continue;
-int cgroup_bonding_migrate_to(CGroupBonding *b, const char *target, bool rem) {
- assert(b);
- assert(target);
+ p = strstrip(line);
- return cg_migrate_recursive(b->controller, b->path, b->controller, target, true, rem);
-}
+ w = strpbrk(p, WHITESPACE);
+ if (!w)
+ continue;
+ *w = 0;
-int cgroup_bonding_set_group_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid) {
- assert(b);
+ r = safe_atou(p, &maj);
+ if (r < 0)
+ continue;
+ if (maj <= 0)
+ continue;
- if (!b->realized)
- return -EINVAL;
+ w++;
+ w += strspn(w, WHITESPACE);
- return cg_set_group_access(b->controller, b->path, mode, uid, gid);
-}
+ if (fnmatch(name, w, 0) != 0)
+ continue;
-int cgroup_bonding_set_group_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid) {
- CGroupBonding *b;
- int r;
+ sprintf(buf,
+ "%c %u:* %s",
+ type,
+ maj,
+ acc);
- LIST_FOREACH(by_unit, b, first) {
- r = cgroup_bonding_set_group_access(b, mode, uid, gid);
+ r = cg_set_attribute("devices", path, "devices.allow", buf);
if (r < 0)
- return r;
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set devices.allow on %s: %s", path, strerror(-r));
}
return 0;
+
+fail:
+ log_warning_errno(errno, "Failed to read /proc/devices: %m");
+ return -errno;
}
-int cgroup_bonding_set_task_access(CGroupBonding *b, mode_t mode, uid_t uid, gid_t gid, int sticky) {
- assert(b);
+void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path, ManagerState state) {
+ bool is_root;
+ int r;
- if (!b->realized)
- return -EINVAL;
+ assert(c);
+ assert(path);
- return cg_set_task_access(b->controller, b->path, mode, uid, gid, sticky);
-}
+ if (mask == 0)
+ return;
-int cgroup_bonding_set_task_access_list(CGroupBonding *first, mode_t mode, uid_t uid, gid_t gid, int sticky) {
- CGroupBonding *b;
- int r;
+ /* Some cgroup attributes are not support on the root cgroup,
+ * hence silently ignore */
+ is_root = isempty(path) || path_equal(path, "/");
- LIST_FOREACH(by_unit, b, first) {
- r = cgroup_bonding_set_task_access(b, mode, uid, gid, sticky);
+ if ((mask & CGROUP_CPU) && !is_root) {
+ char buf[MAX(DECIMAL_STR_MAX(unsigned long), DECIMAL_STR_MAX(usec_t)) + 1];
+
+ sprintf(buf, "%lu\n",
+ IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && c->startup_cpu_shares != (unsigned long) -1 ? c->startup_cpu_shares :
+ c->cpu_shares != (unsigned long) -1 ? c->cpu_shares : 1024);
+ r = cg_set_attribute("cpu", path, "cpu.shares", buf);
if (r < 0)
- return r;
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set cpu.shares on %s: %s", path, strerror(-r));
+
+ sprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
+ r = cg_set_attribute("cpu", path, "cpu.cfs_period_us", buf);
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set cpu.cfs_period_us on %s: %s", path, strerror(-r));
+
+ if (c->cpu_quota_per_sec_usec != USEC_INFINITY) {
+ sprintf(buf, USEC_FMT "\n", c->cpu_quota_per_sec_usec * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
+ r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", buf);
+ } else
+ r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", "-1");
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set cpu.cfs_quota_us on %s: %s", path, strerror(-r));
}
- return 0;
-}
+ if (mask & CGROUP_BLKIO) {
+ char buf[MAX3(DECIMAL_STR_MAX(unsigned long)+1,
+ DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(unsigned long)*1,
+ DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)];
+ CGroupBlockIODeviceWeight *w;
+ CGroupBlockIODeviceBandwidth *b;
+
+ if (!is_root) {
+ sprintf(buf, "%lu\n", IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && c->startup_blockio_weight != (unsigned long) -1 ? c->startup_blockio_weight :
+ c->blockio_weight != (unsigned long) -1 ? c->blockio_weight : 1000);
+ r = cg_set_attribute("blkio", path, "blkio.weight", buf);
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set blkio.weight on %s: %s", path, strerror(-r));
+
+ /* FIXME: no way to reset this list */
+ LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
+ dev_t dev;
+
+ r = lookup_blkio_device(w->path, &dev);
+ if (r < 0)
+ continue;
+
+ sprintf(buf, "%u:%u %lu", major(dev), minor(dev), w->weight);
+ r = cg_set_attribute("blkio", path, "blkio.weight_device", buf);
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set blkio.weight_device on %s: %s", path, strerror(-r));
+ }
+ }
-int cgroup_bonding_kill(CGroupBonding *b, int sig, bool sigcont, bool rem, Set *s, const char *cgroup_suffix) {
- char *p = NULL;
- const char *path;
- int r;
+ /* FIXME: no way to reset this list */
+ LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
+ const char *a;
+ dev_t dev;
- assert(b);
- assert(sig >= 0);
+ r = lookup_blkio_device(b->path, &dev);
+ if (r < 0)
+ continue;
- /* Don't kill cgroups that aren't ours */
- if (!b->ours)
- return 0;
+ a = b->read ? "blkio.throttle.read_bps_device" : "blkio.throttle.write_bps_device";
- if (cgroup_suffix) {
- p = strjoin(b->path, "/", cgroup_suffix, NULL);
- if (!p)
- return -ENOMEM;
+ sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), b->bandwidth);
+ r = cg_set_attribute("blkio", path, a, buf);
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set %s on %s: %s", a, path, strerror(-r));
+ }
+ }
- path = p;
- } else
- path = b->path;
+ if (mask & CGROUP_MEMORY) {
+ if (c->memory_limit != (uint64_t) -1) {
+ char buf[DECIMAL_STR_MAX(uint64_t) + 1];
- r = cg_kill_recursive(b->controller, path, sig, sigcont, true, rem, s);
- free(p);
+ sprintf(buf, "%" PRIu64 "\n", c->memory_limit);
+ r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
+ } else
+ r = cg_set_attribute("memory", path, "memory.limit_in_bytes", "-1");
- return r;
-}
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r));
+ }
-int cgroup_bonding_kill_list(CGroupBonding *first, int sig, bool sigcont, bool rem, Set *s, const char *cgroup_suffix) {
- CGroupBonding *b;
- Set *allocated_set = NULL;
- int ret = -EAGAIN, r;
+ if ((mask & CGROUP_DEVICE) && !is_root) {
+ CGroupDeviceAllow *a;
- if (!first)
- return 0;
+ if (c->device_allow || c->device_policy != CGROUP_AUTO)
+ r = cg_set_attribute("devices", path, "devices.deny", "a");
+ else
+ r = cg_set_attribute("devices", path, "devices.allow", "a");
+ if (r < 0)
+ log_full(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, "Failed to reset devices.list on %s: %s", path, strerror(-r));
+
+ if (c->device_policy == CGROUP_CLOSED ||
+ (c->device_policy == CGROUP_AUTO && c->device_allow)) {
+ static const char auto_devices[] =
+ "/dev/null\0" "rwm\0"
+ "/dev/zero\0" "rwm\0"
+ "/dev/full\0" "rwm\0"
+ "/dev/random\0" "rwm\0"
+ "/dev/urandom\0" "rwm\0"
+ "/dev/tty\0" "rwm\0"
+ "/dev/pts/ptmx\0" "rw\0"; /* /dev/pts/ptmx may not be duplicated, but accessed */
+
+ const char *x, *y;
+
+ NULSTR_FOREACH_PAIR(x, y, auto_devices)
+ whitelist_device(path, x, y);
+
+ whitelist_major(path, "pts", 'c', "rw");
+ whitelist_major(path, "kdbus", 'c', "rw");
+ whitelist_major(path, "kdbus/*", 'c', "rw");
+ }
+
+ LIST_FOREACH(device_allow, a, c->device_allow) {
+ char acc[4];
+ unsigned k = 0;
- if (!s)
- if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func)))
- return -ENOMEM;
+ if (a->r)
+ acc[k++] = 'r';
+ if (a->w)
+ acc[k++] = 'w';
+ if (a->m)
+ acc[k++] = 'm';
- LIST_FOREACH(by_unit, b, first) {
- r = cgroup_bonding_kill(b, sig, sigcont, rem, s, cgroup_suffix);
- if (r < 0) {
- if (r == -EAGAIN || r == -ESRCH)
+ if (k == 0)
continue;
- ret = r;
- goto finish;
- }
+ acc[k++] = 0;
- if (ret < 0 || r > 0)
- ret = r;
+ if (startswith(a->path, "/dev/"))
+ whitelist_device(path, a->path, acc);
+ else if (startswith(a->path, "block-"))
+ whitelist_major(path, a->path + 6, 'b', acc);
+ else if (startswith(a->path, "char-"))
+ whitelist_major(path, a->path + 5, 'c', acc);
+ else
+ log_debug("Ignoring device %s while writing cgroup attribute.", a->path);
+ }
}
+}
-finish:
- if (allocated_set)
- set_free(allocated_set);
+CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) {
+ CGroupControllerMask mask = 0;
- return ret;
-}
+ /* Figure out which controllers we need */
-/* Returns 1 if the group is empty, 0 if it is not, -EAGAIN if we
- * cannot know */
-int cgroup_bonding_is_empty(CGroupBonding *b) {
- int r;
+ if (c->cpu_accounting ||
+ c->cpu_shares != (unsigned long) -1 ||
+ c->startup_cpu_shares != (unsigned long) -1 ||
+ c->cpu_quota_per_sec_usec != USEC_INFINITY)
+ mask |= CGROUP_CPUACCT | CGROUP_CPU;
- assert(b);
+ if (c->blockio_accounting ||
+ c->blockio_weight != (unsigned long) -1 ||
+ c->startup_blockio_weight != (unsigned long) -1 ||
+ c->blockio_device_weights ||
+ c->blockio_device_bandwidths)
+ mask |= CGROUP_BLKIO;
- if ((r = cg_is_empty_recursive(b->controller, b->path, true)) < 0)
- return r;
+ if (c->memory_accounting ||
+ c->memory_limit != (uint64_t) -1)
+ mask |= CGROUP_MEMORY;
- /* If it is empty it is empty */
- if (r > 0)
- return 1;
+ if (c->device_allow ||
+ c->device_policy != CGROUP_AUTO)
+ mask |= CGROUP_DEVICE;
- /* It's not only us using this cgroup, so we just don't know */
- return b->ours ? 0 : -EAGAIN;
+ return mask;
}
-int cgroup_bonding_is_empty_list(CGroupBonding *first) {
- CGroupBonding *b;
+CGroupControllerMask unit_get_cgroup_mask(Unit *u) {
+ CGroupContext *c;
+
+ c = unit_get_cgroup_context(u);
+ if (!c)
+ return 0;
- LIST_FOREACH(by_unit, b, first) {
- int r;
+ /* If delegation is turned on, then turn on all cgroups,
+ * unless the process we fork into it is known to drop
+ * privileges anyway, and shouldn't get access to the
+ * controllers anyway. */
- if ((r = cgroup_bonding_is_empty(b)) < 0) {
- /* If this returned -EAGAIN, then we don't know if the
- * group is empty, so let's see if another group can
- * tell us */
+ if (c->delegate) {
+ ExecContext *e;
- if (r != -EAGAIN)
- return r;
- } else
- return r;
+ e = unit_get_exec_context(u);
+ if (!e || exec_context_maintains_privileges(e))
+ return _CGROUP_CONTROLLER_MASK_ALL;
}
- return -EAGAIN;
+ return cgroup_context_get_mask(c);
}
-int manager_setup_cgroup(Manager *m) {
- char *current = NULL, *path = NULL;
- int r;
- char suffix[32];
+CGroupControllerMask unit_get_members_mask(Unit *u) {
+ assert(u);
- assert(m);
+ if (u->cgroup_members_mask_valid)
+ return u->cgroup_members_mask;
- /* 0. Be nice to Ingo Molnar #628004 */
- if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) {
- log_warning("No control group support available, not creating root group.");
- return 0;
- }
+ u->cgroup_members_mask = 0;
- /* 1. Determine hierarchy */
- r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, ¤t);
- if (r < 0) {
- log_error("Cannot determine cgroup we are running in: %s", strerror(-r));
- goto finish;
- }
+ if (u->type == UNIT_SLICE) {
+ Unit *member;
+ Iterator i;
- if (m->running_as == SYSTEMD_SYSTEM)
- strcpy(suffix, "/system");
- else {
- snprintf(suffix, sizeof(suffix), "/systemd-%lu", (unsigned long) getpid());
- char_array_0(suffix);
- }
+ SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
- free(m->cgroup_hierarchy);
- if (endswith(current, suffix)) {
- /* We probably got reexecuted and can continue to use our root cgroup */
- m->cgroup_hierarchy = current;
- current = NULL;
+ if (member == u)
+ continue;
- } else {
- /* We need a new root cgroup */
- m->cgroup_hierarchy = NULL;
- if (asprintf(&m->cgroup_hierarchy, "%s%s", streq(current, "/") ? "" : current, suffix) < 0) {
- r = log_oom();
- goto finish;
+ if (UNIT_DEREF(member->slice) != u)
+ continue;
+
+ u->cgroup_members_mask |=
+ unit_get_cgroup_mask(member) |
+ unit_get_members_mask(member);
}
}
- /* 2. Show data */
- r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, NULL, &path);
- if (r < 0) {
- log_error("Cannot find cgroup mount point: %s", strerror(-r));
- goto finish;
- }
+ u->cgroup_members_mask_valid = true;
+ return u->cgroup_members_mask;
+}
- log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
+CGroupControllerMask unit_get_siblings_mask(Unit *u) {
+ assert(u);
- /* 3. Install agent */
- r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
- if (r < 0)
- log_warning("Failed to install release agent, ignoring: %s", strerror(-r));
- else if (r > 0)
- log_debug("Installed release agent.");
- else
- log_debug("Release agent already installed.");
-
- /* 4. Realize the group */
- r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy, 0);
- if (r < 0) {
- log_error("Failed to create root cgroup hierarchy: %s", strerror(-r));
- goto finish;
- }
+ if (UNIT_ISSET(u->slice))
+ return unit_get_members_mask(UNIT_DEREF(u->slice));
- /* 5. And pin it, so that it cannot be unmounted */
- if (m->pin_cgroupfs_fd >= 0)
- close_nointr_nofail(m->pin_cgroupfs_fd);
+ return unit_get_cgroup_mask(u) | unit_get_members_mask(u);
+}
- m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
- if (r < 0) {
- log_error("Failed to open pin file: %m");
- r = -errno;
- goto finish;
- }
+CGroupControllerMask unit_get_target_mask(Unit *u) {
+ CGroupControllerMask mask;
- log_debug("Created root group.");
+ mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
+ mask &= u->manager->cgroup_supported;
- cg_shorten_controllers(m->default_controllers);
+ return mask;
+}
-finish:
- free(current);
- free(path);
+/* Recurse from a unit up through its containing slices, propagating
+ * mask bits upward. A unit is also member of itself. */
+void unit_update_cgroup_members_masks(Unit *u) {
+ CGroupControllerMask m;
+ bool more;
- return r;
-}
+ assert(u);
-void manager_shutdown_cgroup(Manager *m, bool delete) {
- assert(m);
+ /* Calculate subtree mask */
+ m = unit_get_cgroup_mask(u) | unit_get_members_mask(u);
- if (delete && m->cgroup_hierarchy)
- cg_delete(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_hierarchy);
+ /* See if anything changed from the previous invocation. If
+ * not, we're done. */
+ if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
+ return;
- if (m->pin_cgroupfs_fd >= 0) {
- close_nointr_nofail(m->pin_cgroupfs_fd);
- m->pin_cgroupfs_fd = -1;
- }
+ more =
+ u->cgroup_subtree_mask_valid &&
+ ((m & ~u->cgroup_subtree_mask) != 0) &&
+ ((~m & u->cgroup_subtree_mask) == 0);
- free(m->cgroup_hierarchy);
- m->cgroup_hierarchy = NULL;
-}
+ u->cgroup_subtree_mask = m;
+ u->cgroup_subtree_mask_valid = true;
-int cgroup_bonding_get(Manager *m, const char *cgroup, CGroupBonding **bonding) {
- CGroupBonding *b;
- char *p;
+ if (UNIT_ISSET(u->slice)) {
+ Unit *s = UNIT_DEREF(u->slice);
- assert(m);
- assert(cgroup);
- assert(bonding);
+ if (more)
+ /* There's more set now than before. We
+ * propagate the new mask to the parent's mask
+ * (not caring if it actually was valid or
+ * not). */
- b = hashmap_get(m->cgroup_bondings, cgroup);
- if (b) {
- *bonding = b;
- return 1;
- }
+ s->cgroup_members_mask |= m;
- p = strdupa(cgroup);
- if (!p)
- return -ENOMEM;
+ else
+ /* There's less set now than before (or we
+ * don't know), we need to recalculate
+ * everything, so let's invalidate the
+ * parent's members mask */
- for (;;) {
- char *e;
+ s->cgroup_members_mask_valid = false;
- e = strrchr(p, '/');
- if (e == p || !e) {
- *bonding = NULL;
- return 0;
- }
+ /* And now make sure that this change also hits our
+ * grandparents */
+ unit_update_cgroup_members_masks(s);
+ }
+}
- *e = 0;
+static const char *migrate_callback(CGroupControllerMask mask, void *userdata) {
+ Unit *u = userdata;
- b = hashmap_get(m->cgroup_bondings, p);
- if (b) {
- *bonding = b;
- return 1;
- }
+ assert(mask != 0);
+ assert(u);
+
+ while (u) {
+ if (u->cgroup_path &&
+ u->cgroup_realized &&
+ (u->cgroup_realized_mask & mask) == mask)
+ return u->cgroup_path;
+
+ u = UNIT_DEREF(u->slice);
}
+
+ return NULL;
}
-int cgroup_notify_empty(Manager *m, const char *group) {
- CGroupBonding *l, *b;
+static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) {
+ _cleanup_free_ char *path = NULL;
int r;
- assert(m);
- assert(group);
+ assert(u);
+
+ path = unit_default_cgroup_path(u);
+ if (!path)
+ return log_oom();
- r = cgroup_bonding_get(m, group, &l);
- if (r <= 0)
+ r = hashmap_put(u->manager->cgroup_unit, path, u);
+ if (r < 0) {
+ log_error(r == -EEXIST ? "cgroup %s exists already: %s" : "hashmap_put failed for %s: %s", path, strerror(-r));
return r;
+ }
+ if (r > 0) {
+ u->cgroup_path = path;
+ path = NULL;
+ }
- LIST_FOREACH(by_path, b, l) {
- int t;
+ /* First, create our own group */
+ r = cg_create_everywhere(u->manager->cgroup_supported, mask, u->cgroup_path);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create cgroup %s: %m", u->cgroup_path);
- if (!b->unit)
- continue;
+ /* Keep track that this is now realized */
+ u->cgroup_realized = true;
+ u->cgroup_realized_mask = mask;
- t = cgroup_bonding_is_empty_list(b);
- if (t < 0) {
+ /* Then, possibly move things over */
+ r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
+ if (r < 0)
+ log_warning_errno(r, "Failed to migrate cgroup from to %s: %m", u->cgroup_path);
- /* If we don't know, we don't know */
- if (t != -EAGAIN)
- log_warning("Failed to check whether cgroup is empty: %s", strerror(errno));
+ return 0;
+}
- continue;
- }
+static bool unit_has_mask_realized(Unit *u, CGroupControllerMask mask) {
+ assert(u);
- if (t > 0) {
- /* If it is empty, let's delete it */
- cgroup_bonding_trim_list(b->unit->cgroup_bondings, true);
+ return u->cgroup_realized && u->cgroup_realized_mask == mask;
+}
- if (UNIT_VTABLE(b->unit)->cgroup_notify_empty)
- UNIT_VTABLE(b->unit)->cgroup_notify_empty(b->unit);
- }
+/* Check if necessary controllers and attributes for a unit are in place.
+ *
+ * If so, do nothing.
+ * If not, create paths, move processes over, and set attributes.
+ *
+ * Returns 0 on success and < 0 on failure. */
+static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
+ CGroupControllerMask mask;
+ int r;
+
+ assert(u);
+
+ if (u->in_cgroup_queue) {
+ LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u);
+ u->in_cgroup_queue = false;
+ }
+
+ mask = unit_get_target_mask(u);
+
+ if (unit_has_mask_realized(u, mask))
+ return 0;
+
+ /* First, realize parents */
+ if (UNIT_ISSET(u->slice)) {
+ r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
+ if (r < 0)
+ return r;
}
+ /* And then do the real work */
+ r = unit_create_cgroups(u, mask);
+ if (r < 0)
+ return r;
+
+ /* Finally, apply the necessary attributes. */
+ cgroup_context_apply(unit_get_cgroup_context(u), mask, u->cgroup_path, state);
+
return 0;
}
-Unit* cgroup_unit_by_pid(Manager *m, pid_t pid) {
- CGroupBonding *l, *b;
- char *group = NULL;
+static void unit_add_to_cgroup_queue(Unit *u) {
- assert(m);
+ if (u->in_cgroup_queue)
+ return;
- if (pid <= 1)
- return NULL;
-
- if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &group) < 0)
- return NULL;
+ LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u);
+ u->in_cgroup_queue = true;
+}
- l = hashmap_get(m->cgroup_bondings, group);
+unsigned manager_dispatch_cgroup_queue(Manager *m) {
+ ManagerState state;
+ unsigned n = 0;
+ Unit *i;
+ int r;
- if (!l) {
- char *slash;
+ state = manager_state(m);
- while ((slash = strrchr(group, '/'))) {
- if (slash == group)
- break;
+ while ((i = m->cgroup_queue)) {
+ assert(i->in_cgroup_queue);
- *slash = 0;
+ r = unit_realize_cgroup_now(i, state);
+ if (r < 0)
+ log_warning_errno(r, "Failed to realize cgroups for queued unit %s: %m", i->id);
- if ((l = hashmap_get(m->cgroup_bondings, group)))
- break;
- }
+ n++;
}
- free(group);
+ return n;
+}
- LIST_FOREACH(by_path, b, l) {
+static void unit_queue_siblings(Unit *u) {
+ Unit *slice;
- if (!b->unit)
- continue;
+ /* This adds the siblings of the specified unit and the
+ * siblings of all parent units to the cgroup queue. (But
+ * neither the specified unit itself nor the parents.) */
- if (b->ours)
- return b->unit;
- }
+ while ((slice = UNIT_DEREF(u->slice))) {
+ Iterator i;
+ Unit *m;
- return NULL;
+ SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
+ if (m == u)
+ continue;
+
+ /* Skip units that have a dependency on the slice
+ * but aren't actually in it. */
+ if (UNIT_DEREF(m->slice) != slice)
+ continue;
+
+ /* No point in doing cgroup application for units
+ * without active processes. */
+ if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
+ continue;
+
+ /* If the unit doesn't need any new controllers
+ * and has current ones realized, it doesn't need
+ * any changes. */
+ if (unit_has_mask_realized(m, unit_get_target_mask(m)))
+ continue;
+
+ unit_add_to_cgroup_queue(m);
+ }
+
+ u = slice;
+ }
}
-CGroupBonding *cgroup_bonding_find_list(CGroupBonding *first, const char *controller) {
- CGroupBonding *b;
+int unit_realize_cgroup(Unit *u) {
+ CGroupContext *c;
- if (!controller)
- controller = SYSTEMD_CGROUP_CONTROLLER;
+ assert(u);
- LIST_FOREACH(by_unit, b, first)
- if (streq(b->controller, controller))
- return b;
+ c = unit_get_cgroup_context(u);
+ if (!c)
+ return 0;
- return NULL;
+ /* So, here's the deal: when realizing the cgroups for this
+ * unit, we need to first create all parents, but there's more
+ * actually: for the weight-based controllers we also need to
+ * make sure that all our siblings (i.e. units that are in the
+ * same slice as we are) have cgroups, too. Otherwise, things
+ * would become very uneven as each of their processes would
+ * get as much resources as all our group together. This call
+ * will synchronously create the parent cgroups, but will
+ * defer work on the siblings to the next event loop
+ * iteration. */
+
+ /* Add all sibling slices to the cgroup queue. */
+ unit_queue_siblings(u);
+
+ /* And realize this one now (and apply the values) */
+ return unit_realize_cgroup_now(u, manager_state(u->manager));
}
-char *cgroup_bonding_to_string(CGroupBonding *b) {
- char *r;
+void unit_destroy_cgroup(Unit *u) {
+ int r;
- assert(b);
+ assert(u);
- if (asprintf(&r, "%s:%s", b->controller, b->path) < 0)
- return NULL;
+ if (!u->cgroup_path)
+ return;
+
+ r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !unit_has_name(u, SPECIAL_ROOT_SLICE));
+ if (r < 0)
+ log_debug_errno(r, "Failed to destroy cgroup %s: %m", u->cgroup_path);
+
+ hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
+
+ free(u->cgroup_path);
+ u->cgroup_path = NULL;
+ u->cgroup_realized = false;
+ u->cgroup_realized_mask = 0;
- return r;
}
-pid_t cgroup_bonding_search_main_pid(CGroupBonding *b) {
- FILE *f;
+pid_t unit_search_main_pid(Unit *u) {
+ _cleanup_fclose_ FILE *f = NULL;
pid_t pid = 0, npid, mypid;
- assert(b);
+ assert(u);
- if (!b->ours)
+ if (!u->cgroup_path)
return 0;
- if (cg_enumerate_processes(b->controller, b->path, &f) < 0)
+ if (cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f) < 0)
return 0;
mypid = getpid();
-
while (cg_read_pid(f, &npid) > 0) {
pid_t ppid;
pid = npid;
}
- fclose(f);
-
return pid;
}
-pid_t cgroup_bonding_search_main_pid_list(CGroupBonding *first) {
- CGroupBonding *b;
- pid_t pid;
+int manager_setup_cgroup(Manager *m) {
+ _cleanup_free_ char *path = NULL;
+ int r;
- /* Try to find a main pid from this cgroup, but checking if
- * there's only one PID in the cgroup and returning it. Later
- * on we might want to add additional, smarter heuristics
- * here. */
+ assert(m);
+
+ /* 1. Determine hierarchy */
+ free(m->cgroup_root);
+ m->cgroup_root = NULL;
+
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
+ if (r < 0)
+ return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
+
+ /* LEGACY: Already in /system.slice? If so, let's cut this
+ * off. This is to support live upgrades from older systemd
+ * versions where PID 1 was moved there. */
+ if (m->running_as == SYSTEMD_SYSTEM) {
+ char *e;
+
+ e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
+ if (!e)
+ e = endswith(m->cgroup_root, "/system");
+ if (e)
+ *e = 0;
+ }
+
+ /* And make sure to store away the root value without trailing
+ * slash, even for the root dir, so that we can easily prepend
+ * it everywhere. */
+ if (streq(m->cgroup_root, "/"))
+ m->cgroup_root[0] = 0;
+
+ /* 2. Show data */
+ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
+ if (r < 0)
+ return log_error_errno(r, "Cannot find cgroup mount point: %m");
+
+ log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
+ if (!m->test_run) {
+
+ /* 3. Install agent */
+ if (m->running_as == SYSTEMD_SYSTEM) {
+ r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
+ if (r < 0)
+ log_warning_errno(r, "Failed to install release agent, ignoring: %m");
+ else if (r > 0)
+ log_debug("Installed release agent.");
+ else
+ log_debug("Release agent already installed.");
+ }
+
+ /* 4. Make sure we are in the root cgroup */
+ r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create root cgroup hierarchy: %m");
+
+ /* 5. And pin it, so that it cannot be unmounted */
+ safe_close(m->pin_cgroupfs_fd);
+
+ m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
+ if (m->pin_cgroupfs_fd < 0)
+ return log_error_errno(errno, "Failed to open pin file: %m");
+
+ /* 6. Always enable hierarchial support if it exists... */
+ cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
+ }
- LIST_FOREACH(by_unit, b, first)
- if ((pid = cgroup_bonding_search_main_pid(b)) != 0)
- return pid;
+ /* 7. Figure out which controllers are supported */
+ m->cgroup_supported = cg_mask_supported();
return 0;
+}
+
+void manager_shutdown_cgroup(Manager *m, bool delete) {
+ assert(m);
+
+ /* We can't really delete the group, since we are in it. But
+ * let's trim it. */
+ if (delete && m->cgroup_root)
+ cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
+
+ m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
+
+ free(m->cgroup_root);
+ m->cgroup_root = NULL;
+}
+
+Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
+ char *p;
+ Unit *u;
+
+ assert(m);
+ assert(cgroup);
+ u = hashmap_get(m->cgroup_unit, cgroup);
+ if (u)
+ return u;
+
+ p = strdupa(cgroup);
+ for (;;) {
+ char *e;
+
+ e = strrchr(p, '/');
+ if (e == p || !e)
+ return NULL;
+
+ *e = 0;
+
+ u = hashmap_get(m->cgroup_unit, p);
+ if (u)
+ return u;
+ }
+}
+
+Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
+ _cleanup_free_ char *cgroup = NULL;
+ int r;
+
+ assert(m);
+
+ if (pid <= 1)
+ return NULL;
+
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
+ if (r < 0)
+ return NULL;
+
+ return manager_get_unit_by_cgroup(m, cgroup);
+}
+
+int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
+ Unit *u;
+ int r;
+
+ assert(m);
+ assert(cgroup);
+
+ u = manager_get_unit_by_cgroup(m, cgroup);
+ if (u) {
+ r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, true);
+ if (r > 0) {
+ if (UNIT_VTABLE(u)->notify_cgroup_empty)
+ UNIT_VTABLE(u)->notify_cgroup_empty(u);
+
+ unit_add_to_gc_queue(u);
+ }
+ }
+
+ return 0;
}
+
+static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
+ [CGROUP_AUTO] = "auto",
+ [CGROUP_CLOSED] = "closed",
+ [CGROUP_STRICT] = "strict",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);