X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcore%2Fcgroup.c;h=10fdcc9984e39f4a5171c77d68c750d7b7d3d490;hp=1c818ab300828989c54ea95981649ce351abe00f;hb=615938651d3a4fd9253b08da00db22d451a8cef8;hpb=0a1eb06d9aacc1f007be04c4133031e0acc91cdd diff --git a/src/core/cgroup.c b/src/core/cgroup.c index 1c818ab30..10fdcc998 100644 --- a/src/core/cgroup.c +++ b/src/core/cgroup.c @@ -20,28 +20,35 @@ ***/ #include +#include #include "path-util.h" #include "special.h" #include "cgroup-util.h" #include "cgroup.h" +#define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) + void cgroup_context_init(CGroupContext *c) { assert(c); /* Initialize everything to the kernel defaults, assuming the * structure is preinitialized to 0 */ - c->cpu_shares = 1024; - c->memory_limit = c->memory_soft_limit = (uint64_t) -1; - c->blockio_weight = 1000; + 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; + + c->cpu_quota_per_sec_usec = USEC_INFINITY; } void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { assert(c); assert(a); - LIST_REMOVE(CGroupDeviceAllow, device_allow, c->device_allow, a); + LIST_REMOVE(device_allow, c->device_allow, a); free(a->path); free(a); } @@ -50,7 +57,7 @@ void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODe assert(c); assert(w); - LIST_REMOVE(CGroupBlockIODeviceWeight, device_weights, c->blockio_device_weights, w); + LIST_REMOVE(device_weights, c->blockio_device_weights, w); free(w->path); free(w); } @@ -59,7 +66,7 @@ void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockI assert(c); assert(b); - LIST_REMOVE(CGroupBlockIODeviceBandwidth, device_bandwidths, c->blockio_device_bandwidths, b); + LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); free(b->path); free(b); } @@ -81,6 +88,7 @@ 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); @@ -92,18 +100,24 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { "%sBlockIOAccounting=%s\n" "%sMemoryAccounting=%s\n" "%sCPUShares=%lu\n" - "%sBlockIOWeight%lu\n" + "%sStartupCPUShares=%lu\n" + "%sCPUQuotaPerSecSec=%s\n" + "%sBlockIOWeight=%lu\n" + "%sStartupBlockIOWeight=%lu\n" "%sMemoryLimit=%" PRIu64 "\n" - "%sMemorySoftLimit=%" PRIu64 "\n" - "%sDevicePolicy=%s\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, c->memory_soft_limit, - prefix, cgroup_device_policy_to_string(c->device_policy)); + prefix, cgroup_device_policy_to_string(c->device_policy), + prefix, yes_no(c->delegate)); LIST_FOREACH(device_allow, a, c->device_allow) fprintf(f, @@ -114,7 +128,7 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { LIST_FOREACH(device_weights, w, c->blockio_device_weights) fprintf(f, - "%sBlockIOWeight=%s %lu", + "%sBlockIODeviceWeight=%s %lu", prefix, w->path, w->weight); @@ -139,10 +153,8 @@ static int lookup_blkio_device(const char *p, dev_t *dev) { assert(dev); r = stat(p, &st); - if (r < 0) { - log_warning("Couldn't stat device %s: %m", p); - return -errno; - } + 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; @@ -188,12 +200,90 @@ static int whitelist_device(const char *path, const char *node, const char *acc) r = cg_set_attribute("devices", path, "devices.allow", buf); if (r < 0) - log_warning("Failed to set devices.allow on %s: %s", path, strerror(-r)); + log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set devices.allow on %s: %m", path); return r; } -void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path) { +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; + + assert(path); + assert(acc); + assert(type == 'b' || type == 'c'); + + f = fopen("/proc/devices", "re"); + if (!f) + return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); + + FOREACH_LINE(line, f, goto fail) { + char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; + unsigned maj; + + truncate_nl(line); + + if (type == 'c' && streq(line, "Character devices:")) { + good = true; + continue; + } + + if (type == 'b' && streq(line, "Block devices:")) { + good = true; + continue; + } + + if (isempty(line)) { + good = false; + continue; + } + + if (!good) + continue; + + p = strstrip(line); + + w = strpbrk(p, WHITESPACE); + if (!w) + continue; + *w = 0; + + r = safe_atou(p, &maj); + if (r < 0) + continue; + if (maj <= 0) + continue; + + w++; + w += strspn(w, WHITESPACE); + + if (fnmatch(name, w, 0) != 0) + continue; + + sprintf(buf, + "%c %u:* %s", + type, + maj, + acc); + + r = cg_set_attribute("devices", path, "devices.allow", buf); + if (r < 0) + log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set devices.allow on %s: %m", path); + } + + return 0; + +fail: + log_warning_errno(errno, "Failed to read /proc/devices: %m"); + return -errno; +} + +void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path, ManagerState state) { + bool is_root; int r; assert(c); @@ -202,13 +292,42 @@ 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 supported on the root cgroup, + * hence silently ignore */ + is_root = isempty(path) || path_equal(path, "/"); + if (is_root) + /* Make sure we don't try to display messages with an empty path. */ + path = "/"; - sprintf(buf, "%lu\n", c->cpu_shares); + /* We generally ignore errors caused by read-only mounted + * cgroup trees (assuming we are running in a container then), + * and missing cgroups, i.e. EROFS and ENOENT. */ + + 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) - log_warning("Failed to set cpu.shares on %s: %s", path, strerror(-r)); + log_full_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.shares on %s: %m", path); + + 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_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.cfs_period_us on %s: %m", path); + + 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_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.cfs_quota_us on %s: %m", path); } if (mask & CGROUP_BLKIO) { @@ -218,23 +337,28 @@ 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)); - - /* 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 (!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_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r)); + log_full_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set blkio.weight on %s: %m", path); + + /* 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_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set blkio.weight_device on %s: %m", path); + } } /* FIXME: no way to reset this list */ @@ -251,47 +375,59 @@ void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const cha sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), b->bandwidth); r = cg_set_attribute("blkio", path, a, buf); if (r < 0) - log_error("Failed to set %s on %s: %s", a, path, strerror(-r)); + log_full_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set %s on %s: %m", a, path); } } - if (mask & CGROUP_MEMORY) { - char buf[DECIMAL_STR_MAX(uint64_t) + 1]; + if ((mask & CGROUP_MEMORY) && !is_root) { + if (c->memory_limit != (uint64_t) -1) { + char buf[DECIMAL_STR_MAX(uint64_t) + 1]; - sprintf(buf, "%" PRIu64 "\n", c->memory_limit); - r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); - if (r < 0) - log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r)); + 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"); - sprintf(buf, "%" PRIu64 "\n", c->memory_soft_limit); - cg_set_attribute("memory", path, "memory.soft_limit_in_bytes", buf); if (r < 0) - log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r)); + log_full_errno(IN_SET(r, -ENOENT, -EROFS) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set memory.limit_in_bytes on %s: %m", path); } - if (mask & CGROUP_DEVICE) { + if ((mask & CGROUP_DEVICE) && !is_root) { CGroupDeviceAllow *a; + /* Changing the devices list of a populated cgroup + * might result in EINVAL, hence ignore EINVAL + * here. */ + 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_error("Failed to reset devices.list on %s: %s", path, strerror(-r)); + log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to reset devices.list on %s: %m", path); 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) { @@ -309,7 +445,15 @@ void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const cha continue; acc[k++] = 0; - whitelist_device(path, a->path, acc); + + 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); } } } @@ -319,124 +463,277 @@ 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 != (unsigned long) -1 || + c->startup_cpu_shares != (unsigned long) -1 || + c->cpu_quota_per_sec_usec != USEC_INFINITY) mask |= CGROUP_CPUACCT | CGROUP_CPU; if (c->blockio_accounting || - c->blockio_weight != 1000 || + 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 (c->memory_accounting || - c->memory_limit != (uint64_t) -1 || - c->memory_soft_limit != (uint64_t) -1) + c->memory_limit != (uint64_t) -1) mask |= CGROUP_MEMORY; - if (c->device_allow || c->device_policy != CGROUP_AUTO) + if (c->device_allow || + c->device_policy != CGROUP_AUTO) mask |= CGROUP_DEVICE; return mask; } -static CGroupControllerMask unit_get_cgroup_mask(Unit *u) { +CGroupControllerMask unit_get_cgroup_mask(Unit *u) { CGroupContext *c; c = unit_get_cgroup_context(u); if (!c) return 0; + /* 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 (c->delegate) { + ExecContext *e; + + e = unit_get_exec_context(u); + if (!e || exec_context_maintains_privileges(e)) + return _CGROUP_CONTROLLER_MASK_ALL; + } + return cgroup_context_get_mask(c); } -static CGroupControllerMask unit_get_members_mask(Unit *u) { - CGroupControllerMask mask = 0; - Unit *m; - Iterator i; - +CGroupControllerMask unit_get_members_mask(Unit *u) { assert(u); - SET_FOREACH(m, u->dependencies[UNIT_BEFORE], i) { + if (u->cgroup_members_mask_valid) + return u->cgroup_members_mask; - if (UNIT_DEREF(m->slice) != u) - continue; + u->cgroup_members_mask = 0; + + if (u->type == UNIT_SLICE) { + Unit *member; + Iterator i; + + SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) { + + if (member == u) + continue; + + if (UNIT_DEREF(member->slice) != u) + continue; - mask |= unit_get_cgroup_mask(m) | unit_get_members_mask(m); + u->cgroup_members_mask |= + unit_get_cgroup_mask(member) | + unit_get_members_mask(member); + } } + u->cgroup_members_mask_valid = true; + return u->cgroup_members_mask; +} + +CGroupControllerMask unit_get_siblings_mask(Unit *u) { + assert(u); + + if (UNIT_ISSET(u->slice)) + return unit_get_members_mask(UNIT_DEREF(u->slice)); + + return unit_get_cgroup_mask(u) | unit_get_members_mask(u); +} + +CGroupControllerMask unit_get_target_mask(Unit *u) { + CGroupControllerMask mask; + + mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); + mask &= u->manager->cgroup_supported; + return mask; } -static CGroupControllerMask unit_get_siblings_mask(Unit *u) { +/* 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; + assert(u); - if (!UNIT_ISSET(u->slice)) - return 0; + /* Calculate subtree mask */ + m = unit_get_cgroup_mask(u) | unit_get_members_mask(u); + + /* 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; + + more = + u->cgroup_subtree_mask_valid && + ((m & ~u->cgroup_subtree_mask) != 0) && + ((~m & u->cgroup_subtree_mask) == 0); + + u->cgroup_subtree_mask = m; + u->cgroup_subtree_mask_valid = true; + + if (UNIT_ISSET(u->slice)) { + Unit *s = UNIT_DEREF(u->slice); + + 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). */ + + s->cgroup_members_mask |= m; + + 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 */ - /* Sibling propagation is only relevant for weight-based - * controllers, so let's mask out everything else */ - return unit_get_members_mask(UNIT_DEREF(u->slice)) & - (CGROUP_CPU|CGROUP_BLKIO|CGROUP_CPUACCT); + s->cgroup_members_mask_valid = false; + + /* And now make sure that this change also hits our + * grandparents */ + unit_update_cgroup_members_masks(s); + } +} + +static const char *migrate_callback(CGroupControllerMask mask, void *userdata) { + Unit *u = userdata; + + 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; } static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) { - char *path = NULL; + CGroupContext *c; int r; assert(u); - path = unit_default_cgroup_path(u); - if (!path) - return -ENOMEM; + c = unit_get_cgroup_context(u); + if (!c) + return 0; - r = hashmap_put(u->manager->cgroup_unit, path, u); - if (r < 0) - return r; + if (!u->cgroup_path) { + _cleanup_free_ char *path = NULL; + + path = unit_default_cgroup_path(u); + if (!path) + return log_oom(); + + 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; + } + } /* First, create our own group */ - r = cg_create_with_mask(mask, path); + r = cg_create_everywhere(u->manager->cgroup_supported, mask, u->cgroup_path); if (r < 0) - log_error("Failed to create cgroup %s: %s", path, strerror(-r)); + return log_error_errno(r, "Failed to create cgroup %s: %m", u->cgroup_path); + + /* Keep track that this is now realized */ + u->cgroup_realized = true; + u->cgroup_realized_mask = mask; + + if (u->type != UNIT_SLICE && !c->delegate) { - /* Then, possibly move things over */ - if (u->cgroup_path && !streq(path, u->cgroup_path)) { - r = cg_migrate_with_mask(mask, u->cgroup_path, path); + /* Then, possibly move things over, but not if + * subgroups may contain processes, which is the case + * for slice and delegation units. */ + r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); if (r < 0) - log_error("Failed to migrate cgroup %s: %s", path, strerror(-r)); + log_warning_errno(r, "Failed to migrate cgroup from to %s: %m", u->cgroup_path); } - /* And remember the new data */ - free(u->cgroup_path); - u->cgroup_path = path; - u->cgroup_realized = true; - u->cgroup_mask = mask; + return 0; +} + +int unit_attach_pids_to_cgroup(Unit *u) { + int r; + assert(u); + + r = unit_realize_cgroup(u); + if (r < 0) + return r; + + r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u); + if (r < 0) + return r; return 0; } -static int unit_realize_cgroup_now(Unit *u) { +static bool unit_has_mask_realized(Unit *u, CGroupControllerMask mask) { + assert(u); + + return u->cgroup_realized && u->cgroup_realized_mask == mask; +} + +/* 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(Unit, cgroup_queue, u->manager->cgroup_queue, u); + LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u); u->in_cgroup_queue = false; } - mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); - mask &= u->manager->cgroup_supported; + mask = unit_get_target_mask(u); - if (u->cgroup_realized && - u->cgroup_mask == mask) + if (unit_has_mask_realized(u, mask)) return 0; /* First, realize parents */ - if (UNIT_ISSET(u->slice)) - unit_realize_cgroup_now(UNIT_DEREF(u->slice)); + 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 */ - return unit_create_cgroups(u, mask); + 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; } static void unit_add_to_cgroup_queue(Unit *u) { @@ -444,19 +741,24 @@ static void unit_add_to_cgroup_queue(Unit *u) { if (u->in_cgroup_queue) return; - LIST_PREPEND(Unit, cgroup_queue, u->manager->cgroup_queue, u); + LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u); u->in_cgroup_queue = true; } unsigned manager_dispatch_cgroup_queue(Manager *m) { - Unit *i; + ManagerState state; unsigned n = 0; + Unit *i; + int r; + + state = manager_state(m); while ((i = m->cgroup_queue)) { assert(i->in_cgroup_queue); - if (unit_realize_cgroup_now(i) >= 0) - cgroup_context_apply(unit_get_cgroup_context(i), i->cgroup_mask, i->cgroup_path); + 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); n++; } @@ -479,9 +781,22 @@ static void unit_queue_siblings(Unit *u) { 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); } @@ -491,7 +806,6 @@ static void unit_queue_siblings(Unit *u) { int unit_realize_cgroup(Unit *u) { CGroupContext *c; - int r; assert(u); @@ -503,7 +817,7 @@ int unit_realize_cgroup(Unit *u) { * 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 cgroup too. Otherwise things + * 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 @@ -513,17 +827,11 @@ int unit_realize_cgroup(Unit *u) { /* Add all sibling slices to the cgroup queue. */ unit_queue_siblings(u); - /* And realize this one now */ - r = unit_realize_cgroup_now(u); - - /* And apply the values */ - if (r >= 0) - cgroup_context_apply(c, u->cgroup_mask, u->cgroup_path); - - return r; + /* And realize this one now (and apply the values) */ + return unit_realize_cgroup_now(u, manager_state(u->manager)); } -void unit_destroy_cgroup(Unit *u) { +void unit_destroy_cgroup_if_empty(Unit *u) { int r; assert(u); @@ -531,17 +839,18 @@ void unit_destroy_cgroup(Unit *u) { if (!u->cgroup_path) return; - r = cg_trim_with_mask(u->cgroup_mask, u->cgroup_path, true); - if (r < 0) - log_error("Failed to destroy cgroup %s: %s", u->cgroup_path, strerror(-r)); + 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); + return; + } hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); free(u->cgroup_path); u->cgroup_path = NULL; u->cgroup_realized = false; - u->cgroup_mask = 0; - + u->cgroup_realized_mask = 0; } pid_t unit_search_main_pid(Unit *u) { @@ -584,29 +893,26 @@ pid_t unit_search_main_pid(Unit *u) { int manager_setup_cgroup(Manager *m) { _cleanup_free_ char *path = NULL; int r; - char *e, *a; 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; r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); - if (r < 0) { - log_error("Cannot determine cgroup we are running in: %s", strerror(-r)); - return r; - } + if (r < 0) + return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); - /* Already in /system.slice? If so, let's cut this off again */ + /* 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; } @@ -619,43 +925,40 @@ int manager_setup_cgroup(Manager *m) { /* 2. Show data */ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); - if (r < 0) { - log_error("Cannot find cgroup mount point: %s", strerror(-r)); - return r; - } + 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); + /* 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) - 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."); - } + return log_error_errno(r, "Failed to create root cgroup hierarchy: %m"); - /* 4. Realize the system slice and put us in there */ - a = strappenda(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); - r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, a, 0); - if (r < 0) { - log_error("Failed to create root cgroup hierarchy: %s", strerror(-r)); - return r; - } + /* 5. And pin it, so that it cannot be unmounted */ + safe_close(m->pin_cgroupfs_fd); - /* 5. And pin it, so that it cannot be unmounted */ - if (m->pin_cgroupfs_fd >= 0) - close_nointr_nofail(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"); - 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"); - return -errno; + /* 6. Always enable hierarchical support if it exists... */ + cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); } - /* 6. Figure out which controllers are supported */ + /* 7. Figure out which controllers are supported */ m->cgroup_supported = cg_mask_supported(); return 0; @@ -669,10 +972,7 @@ 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; @@ -728,13 +1028,16 @@ int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { assert(m); assert(cgroup); - r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, cgroup, true); - if (r == 0) - return 0; - u = manager_get_unit_by_cgroup(m, cgroup); - if (u && UNIT_VTABLE(u)->notify_cgroup_empty) - UNIT_VTABLE(u)->notify_cgroup_empty(u); + 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; }