chiark / gitweb /
core: require cgroups filesystem to be available
[elogind.git] / src / core / cgroup.c
index 50de02d0ce041a52ade7ca9cc64706c74007a59e..3aeaf5665622dc41cd85d47337af59308ccf8f21 100644 (file)
@@ -20,6 +20,7 @@
 ***/
 
 #include <fcntl.h>
+#include <fnmatch.h>
 
 #include "path-util.h"
 #include "special.h"
@@ -35,6 +36,10 @@ void cgroup_context_init(CGroupContext *c) {
         c->cpu_shares = 1024;
         c->memory_limit = (uint64_t) -1;
         c->blockio_weight = 1000;
+
+        c->cpu_quota_per_sec_usec = (usec_t) -1;
+        c->cpu_quota_usec = (usec_t) -1;
+        c->cpu_quota_period_usec = 100*USEC_PER_MSEC;
 }
 
 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
@@ -77,10 +82,37 @@ void cgroup_context_done(CGroupContext *c) {
                 cgroup_context_free_device_allow(c, c->device_allow);
 }
 
+usec_t cgroup_context_get_cpu_quota_usec(CGroupContext *c) {
+        assert(c);
+
+        /* Returns the absolute CPU quota */
+
+        if (c->cpu_quota_usec != (usec_t) -1)
+                return c->cpu_quota_usec;
+        else if (c->cpu_quota_per_sec_usec != (usec_t) -1)
+                return c->cpu_quota_per_sec_usec*c->cpu_quota_period_usec/USEC_PER_SEC;
+        else
+                return (usec_t) -1;
+}
+
+usec_t cgroup_context_get_cpu_quota_per_sec_usec(CGroupContext *c) {
+        assert(c);
+
+        /* Returns the CPU quota relative to 1s */
+
+        if (c->cpu_quota_usec != (usec_t) -1)
+                return c->cpu_quota_usec*USEC_PER_SEC/c->cpu_quota_period_usec;
+        else if (c->cpu_quota_per_sec_usec != (usec_t) -1)
+                return c->cpu_quota_per_sec_usec;
+        else
+                return (usec_t) -1;
+}
+
 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
         CGroupBlockIODeviceBandwidth *b;
         CGroupBlockIODeviceWeight *w;
         CGroupDeviceAllow *a;
+        char t[FORMAT_TIMESPAN_MAX], s[FORMAT_TIMESPAN_MAX], u[FORMAT_TIMESPAN_MAX];
 
         assert(c);
         assert(f);
@@ -92,6 +124,9 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
                 "%sBlockIOAccounting=%s\n"
                 "%sMemoryAccounting=%s\n"
                 "%sCPUShares=%lu\n"
+                "%sCPUQuota=%s\n"
+                "%sCPUQuotaPerSecSec=%s\n"
+                "%sCPUQuotaPeriodSec=%s\n"
                 "%sBlockIOWeight=%lu\n"
                 "%sMemoryLimit=%" PRIu64 "\n"
                 "%sDevicePolicy=%s\n",
@@ -99,6 +134,9 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
                 prefix, yes_no(c->blockio_accounting),
                 prefix, yes_no(c->memory_accounting),
                 prefix, c->cpu_shares,
+                prefix, strna(format_timespan(u, sizeof(u), cgroup_context_get_cpu_quota_usec(c), 1)),
+                prefix, strna(format_timespan(t, sizeof(t), cgroup_context_get_cpu_quota_per_sec_usec(c), 1)),
+                prefix, strna(format_timespan(s, sizeof(s), c->cpu_quota_period_usec, 1)),
                 prefix, c->blockio_weight,
                 prefix, c->memory_limit,
                 prefix, cgroup_device_policy_to_string(c->device_policy));
@@ -246,7 +284,8 @@ static int whitelist_major(const char *path, const char *name, char type, const
 
                 w++;
                 w += strspn(w, WHITESPACE);
-                if (!streq(w, name))
+
+                if (fnmatch(name, w, 0) != 0)
                         continue;
 
                 sprintf(buf,
@@ -268,6 +307,7 @@ fail:
 }
 
 void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path) {
+        bool is_root;
         int r;
 
         assert(c);
@@ -276,13 +316,32 @@ void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const cha
         if (mask == 0)
                 return;
 
-        if (mask & CGROUP_CPU) {
-                char buf[DECIMAL_STR_MAX(unsigned long) + 1];
+        /* Some cgroup attributes are not support on the root cgroup,
+         * hence silently ignore */
+        is_root = isempty(path) || path_equal(path, "/");
+
+        if ((mask & CGROUP_CPU) && !is_root) {
+                char buf[MAX(DECIMAL_STR_MAX(unsigned long), DECIMAL_STR_MAX(usec_t)) + 1];
+                usec_t q;
 
                 sprintf(buf, "%lu\n", c->cpu_shares);
                 r = cg_set_attribute("cpu", path, "cpu.shares", buf);
                 if (r < 0)
                         log_warning("Failed to set cpu.shares on %s: %s", path, strerror(-r));
+
+                sprintf(buf, USEC_FMT "\n", c->cpu_quota_period_usec);
+                r = cg_set_attribute("cpu", path, "cpu.cfs_period_us", buf);
+                if (r < 0)
+                        log_warning("Failed to set cpu.cfs_period_us on %s: %s", path, strerror(-r));
+
+                q = cgroup_context_get_cpu_quota_usec(c);
+                if (q != (usec_t) -1) {
+                        sprintf(buf, USEC_FMT "\n", q);
+                        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_warning("Failed to set cpu.cfs_quota_us on %s: %s", path, strerror(-r));
         }
 
         if (mask & CGROUP_BLKIO) {
@@ -292,23 +351,25 @@ void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const cha
                 CGroupBlockIODeviceWeight *w;
                 CGroupBlockIODeviceBandwidth *b;
 
-                sprintf(buf, "%lu\n", c->blockio_weight);
-                r = cg_set_attribute("blkio", path, "blkio.weight", buf);
-                if (r < 0)
-                        log_warning("Failed to set blkio.weight on %s: %s", path, strerror(-r));
+                if (!is_root) {
+                        sprintf(buf, "%lu\n", c->blockio_weight);
+                        r = cg_set_attribute("blkio", path, "blkio.weight", buf);
+                        if (r < 0)
+                                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;
+                        /* 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;
+                                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_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r));
+                                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_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r));
+                        }
                 }
 
                 /* FIXME: no way to reset this list */
@@ -342,7 +403,7 @@ void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const cha
                         log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r));
         }
 
-        if (mask & CGROUP_DEVICE) {
+        if ((mask & CGROUP_DEVICE) && !is_root) {
                 CGroupDeviceAllow *a;
 
                 if (c->device_allow || c->device_policy != CGROUP_AUTO)
@@ -350,21 +411,27 @@ void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const cha
                 else
                         r = cg_set_attribute("devices", path, "devices.allow", "a");
                 if (r < 0)
-                        log_error("Failed to reset devices.list on %s: %s", path, strerror(-r));
+                        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" "rw\0"
-                                "/dev/zero\0" "rw\0"
-                                "/dev/full\0" "rw\0"
-                                "/dev/random\0" "rw\0"
-                                "/dev/urandom\0" "rw\0";
+                                "/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) {
@@ -400,7 +467,10 @@ CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) {
 
         /* Figure out which controllers we need */
 
-        if (c->cpu_accounting || c->cpu_shares != 1024)
+        if (c->cpu_accounting ||
+            c->cpu_shares != 1024 ||
+            c->cpu_quota_usec != (usec_t) -1 ||
+            c->cpu_quota_per_sec_usec != (usec_t) -1)
                 mask |= CGROUP_CPUACCT | CGROUP_CPU;
 
         if (c->blockio_accounting ||
@@ -792,12 +862,6 @@ int manager_setup_cgroup(Manager *m) {
 
         assert(m);
 
-        /* 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;
-        }
-
         /* 1. Determine hierarchy */
         free(m->cgroup_root);
         m->cgroup_root = NULL;
@@ -853,8 +917,7 @@ int manager_setup_cgroup(Manager *m) {
         }
 
         /* 5. And pin it, so that it cannot be unmounted */
-        if (m->pin_cgroupfs_fd >= 0)
-                close_nointr_nofail(m->pin_cgroupfs_fd);
+        safe_close(m->pin_cgroupfs_fd);
 
         m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
         if (r < 0) {
@@ -879,10 +942,7 @@ void manager_shutdown_cgroup(Manager *m, bool delete) {
         if (delete && m->cgroup_root)
                 cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
 
-        if (m->pin_cgroupfs_fd >= 0) {
-                close_nointr_nofail(m->pin_cgroupfs_fd);
-                m->pin_cgroupfs_fd = -1;
-        }
+        m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
 
         free(m->cgroup_root);
         m->cgroup_root = NULL;