diff options
author | Zbigniew Jędrzejewski-Szmek <zbyszek@in.waw.pl> | 2016-08-19 19:00:25 -0400 |
---|---|---|
committer | Zbigniew Jędrzejewski-Szmek <zbyszek@in.waw.pl> | 2016-08-19 19:58:01 -0400 |
commit | 2056ec192742d45aa72a851dbd22ad1fe0bc91a2 (patch) | |
tree | 5f134b9c2d466b1726771df416f812d49ad1e60d | |
parent | e6c9fa74a5b1dc1c29797cab43a078b512438885 (diff) | |
parent | 5da38d0768bf53f611ccdd47d7d941e1c560e44e (diff) |
Merge pull request #3965 from htejun/systemd-controller-on-unified
-rw-r--r-- | src/basic/cgroup-util.c | 131 | ||||
-rw-r--r-- | src/basic/cgroup-util.h | 12 | ||||
-rw-r--r-- | src/cgls/cgls.c | 2 | ||||
-rw-r--r-- | src/cgtop/cgtop.c | 10 | ||||
-rw-r--r-- | src/core/cgroup.c | 34 | ||||
-rw-r--r-- | src/core/manager.c | 2 | ||||
-rw-r--r-- | src/core/mount-setup.c | 6 | ||||
-rw-r--r-- | src/core/scope.c | 2 | ||||
-rw-r--r-- | src/core/service.c | 2 | ||||
-rw-r--r-- | src/core/unit.c | 2 | ||||
-rw-r--r-- | src/libsystemd/sd-bus/test-bus-creds.c | 2 | ||||
-rw-r--r-- | src/nspawn/nspawn-cgroup.c | 13 | ||||
-rw-r--r-- | src/nspawn/nspawn-cgroup.h | 6 | ||||
-rw-r--r-- | src/nspawn/nspawn-mount.c | 46 | ||||
-rw-r--r-- | src/nspawn/nspawn-mount.h | 6 | ||||
-rw-r--r-- | src/nspawn/nspawn.c | 26 |
16 files changed, 205 insertions, 97 deletions
diff --git a/src/basic/cgroup-util.c b/src/basic/cgroup-util.c index 25ef8a5c76..9b44c5a7a5 100644 --- a/src/basic/cgroup-util.c +++ b/src/basic/cgroup-util.c @@ -623,7 +623,7 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch if (!cg_controller_is_valid(controller)) return -EINVAL; - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; @@ -651,7 +651,7 @@ static int controller_is_accessible(const char *controller) { if (!cg_controller_is_valid(controller)) return -EINVAL; - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (unified > 0) { @@ -869,7 +869,7 @@ int cg_set_task_access( if (r < 0) return r; - unified = cg_unified(); + unified = cg_unified(controller); if (unified < 0) return unified; if (unified) @@ -893,18 +893,17 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) { assert(path); assert(pid >= 0); - unified = cg_unified(); + if (controller) { + if (!cg_controller_is_valid(controller)) + return -EINVAL; + } else + controller = SYSTEMD_CGROUP_CONTROLLER; + + unified = cg_unified(controller); if (unified < 0) return unified; - if (unified == 0) { - if (controller) { - if (!cg_controller_is_valid(controller)) - return -EINVAL; - } else - controller = SYSTEMD_CGROUP_CONTROLLER; - + if (unified == 0) cs = strlen(controller); - } fs = procfs_file_alloca(pid, "cgroup"); f = fopen(fs, "re"); @@ -969,7 +968,7 @@ int cg_install_release_agent(const char *controller, const char *agent) { assert(agent); - unified = cg_unified(); + unified = cg_unified(controller); if (unified < 0) return unified; if (unified) /* doesn't apply to unified hierarchy */ @@ -1020,7 +1019,7 @@ int cg_uninstall_release_agent(const char *controller) { _cleanup_free_ char *fs = NULL; int r, unified; - unified = cg_unified(); + unified = cg_unified(controller); if (unified < 0) return unified; if (unified) /* Doesn't apply to unified hierarchy */ @@ -1076,7 +1075,7 @@ int cg_is_empty_recursive(const char *controller, const char *path) { if (controller && (isempty(path) || path_equal(path, "/"))) return false; - unified = cg_unified(); + unified = cg_unified(controller); if (unified < 0) return unified; @@ -1962,7 +1961,7 @@ int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path return r; /* If we are in the unified hierarchy, we are done now */ - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (unified > 0) @@ -1992,7 +1991,7 @@ int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_m if (r < 0) return r; - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (unified > 0) @@ -2044,7 +2043,7 @@ int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to return r; } - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (unified > 0) @@ -2077,7 +2076,7 @@ int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) if (r < 0) return r; - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (unified > 0) @@ -2103,7 +2102,7 @@ int cg_mask_supported(CGroupMask *ret) { * includes controllers we can make sense of and that are * actually accessible. */ - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (unified > 0) { @@ -2224,9 +2223,10 @@ int cg_kernel_controllers(Set *controllers) { return 0; } -static thread_local int unified_cache = -1; +static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN; + +static int cg_update_unified(void) { -int cg_unified(void) { struct statfs fs; /* Checks if we support the unified hierarchy. Returns an @@ -2234,24 +2234,47 @@ int cg_unified(void) { * have any other trouble determining if the unified hierarchy * is supported. */ - if (unified_cache >= 0) - return unified_cache; + if (unified_cache >= CGROUP_UNIFIED_NONE) + return 0; if (statfs("/sys/fs/cgroup/", &fs) < 0) return -errno; if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) - unified_cache = true; - else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) - unified_cache = false; - else + unified_cache = CGROUP_UNIFIED_ALL; + else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) { + if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0) + return -errno; + + unified_cache = F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC) ? + CGROUP_UNIFIED_SYSTEMD : CGROUP_UNIFIED_NONE; + } else return -ENOMEDIUM; - return unified_cache; + return 0; +} + +int cg_unified(const char *controller) { + + int r; + + r = cg_update_unified(); + if (r < 0) + return r; + + if (streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER)) + return unified_cache >= CGROUP_UNIFIED_SYSTEMD; + else + return unified_cache >= CGROUP_UNIFIED_ALL; +} + +int cg_all_unified(void) { + + return cg_unified(NULL); } void cg_unified_flush(void) { - unified_cache = -1; + unified_cache = CGROUP_UNIFIED_UNKNOWN; } int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) { @@ -2264,7 +2287,7 @@ int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) { if (supported == 0) return 0; - unified = cg_unified(); + unified = cg_all_unified(); if (unified < 0) return unified; if (!unified) /* on the legacy hiearchy there's no joining of controllers defined */ @@ -2303,7 +2326,7 @@ bool cg_is_unified_wanted(void) { /* If the hierarchy is already mounted, then follow whatever * was chosen for it. */ - unified = cg_unified(); + unified = cg_all_unified(); if (unified >= 0) return unified; @@ -2333,6 +2356,50 @@ bool cg_is_legacy_wanted(void) { return !cg_is_unified_wanted(); } +bool cg_is_unified_systemd_controller_wanted(void) { + static thread_local int wanted = -1; + int r, unified; + + /* If the unified hierarchy is requested in full, no need to + * bother with this. */ + if (cg_is_unified_wanted()) + return 0; + + /* If the hierarchy is already mounted, then follow whatever + * was chosen for it. */ + unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER); + if (unified >= 0) + return unified; + + /* Otherwise, let's see what the kernel command line has to + * say. Since checking that is expensive, let's cache the + * result. */ + if (wanted >= 0) + return wanted; + + r = get_proc_cmdline_key("systemd.legacy_systemd_cgroup_controller", NULL); + if (r > 0) + wanted = false; + else { + _cleanup_free_ char *value = NULL; + + r = get_proc_cmdline_key("systemd.legacy_systemd_cgroup_controller=", &value); + if (r < 0) + return true; + + if (r == 0) + wanted = true; + else + wanted = parse_boolean(value) <= 0; + } + + return wanted; +} + +bool cg_is_legacy_systemd_controller_wanted(void) { + return cg_is_legacy_wanted() && !cg_is_unified_systemd_controller_wanted(); +} + int cg_weight_parse(const char *s, uint64_t *ret) { uint64_t u; int r; diff --git a/src/basic/cgroup-util.h b/src/basic/cgroup-util.h index f1617a16be..1a61c7ad22 100644 --- a/src/basic/cgroup-util.h +++ b/src/basic/cgroup-util.h @@ -116,6 +116,13 @@ static inline bool CGROUP_BLKIO_WEIGHT_IS_OK(uint64_t x) { #define DEFAULT_TASKS_MAX_PERCENTAGE 15U /* 15% of PIDs, 4915 on default settings */ #define DEFAULT_USER_TASKS_MAX_PERCENTAGE 33U /* 33% of PIDs, 10813 on default settings */ +typedef enum CGroupUnified { + CGROUP_UNIFIED_UNKNOWN = -1, + CGROUP_UNIFIED_NONE = 0, /* Both systemd and controllers on legacy */ + CGROUP_UNIFIED_SYSTEMD = 1, /* Only systemd on unified */ + CGROUP_UNIFIED_ALL = 2, /* Both systemd and controllers on unified */ +} CGroupUnified; + /* * General rules: * @@ -229,11 +236,14 @@ int cg_kernel_controllers(Set *controllers); bool cg_ns_supported(void); -int cg_unified(void); +int cg_all_unified(void); +int cg_unified(const char *controller); void cg_unified_flush(void); bool cg_is_unified_wanted(void); bool cg_is_legacy_wanted(void); +bool cg_is_unified_systemd_controller_wanted(void); +bool cg_is_legacy_systemd_controller_wanted(void); const char* cgroup_controller_to_string(CGroupController c) _const_; CGroupController cgroup_controller_from_string(const char *s) _pure_; diff --git a/src/cgls/cgls.c b/src/cgls/cgls.c index dcb5912b83..adf488e8e1 100644 --- a/src/cgls/cgls.c +++ b/src/cgls/cgls.c @@ -166,7 +166,7 @@ static int get_cgroup_root(char **ret) { static void show_cg_info(const char *controller, const char *path) { - if (cg_unified() <= 0 && controller && !streq(controller, SYSTEMD_CGROUP_CONTROLLER)) + if (cg_all_unified() <= 0 && controller && !streq(controller, SYSTEMD_CGROUP_CONTROLLER)) printf("Controller %s; ", controller); printf("Control group %s:\n", isempty(path) ? "/" : path); diff --git a/src/cgtop/cgtop.c b/src/cgtop/cgtop.c index 6045ae0437..aba17c9829 100644 --- a/src/cgtop/cgtop.c +++ b/src/cgtop/cgtop.c @@ -213,7 +213,7 @@ static int process( uint64_t new_usage; nsec_t timestamp; - if (cg_unified() > 0) { + if (cg_all_unified() > 0) { const char *keys[] = { "usage_usec", NULL }; _cleanup_free_ char *val = NULL; @@ -273,7 +273,7 @@ static int process( } else if (streq(controller, "memory")) { _cleanup_free_ char *p = NULL, *v = NULL; - if (cg_unified() <= 0) + if (cg_all_unified() <= 0) r = cg_get_path(controller, path, "memory.usage_in_bytes", &p); else r = cg_get_path(controller, path, "memory.current", &p); @@ -293,11 +293,11 @@ static int process( if (g->memory > 0) g->memory_valid = true; - } else if ((streq(controller, "io") && cg_unified() > 0) || - (streq(controller, "blkio") && cg_unified() <= 0)) { + } else if ((streq(controller, "io") && cg_all_unified() > 0) || + (streq(controller, "blkio") && cg_all_unified() <= 0)) { _cleanup_fclose_ FILE *f = NULL; _cleanup_free_ char *p = NULL; - bool unified = cg_unified() > 0; + bool unified = cg_all_unified() > 0; uint64_t wr = 0, rd = 0; nsec_t timestamp; diff --git a/src/core/cgroup.c b/src/core/cgroup.c index ca3c3366f3..086cd7a789 100644 --- a/src/core/cgroup.c +++ b/src/core/cgroup.c @@ -665,7 +665,7 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { bool has_weight = cgroup_context_has_cpu_weight(c); bool has_shares = cgroup_context_has_cpu_shares(c); - if (cg_unified() > 0) { + if (cg_all_unified() > 0) { uint64_t weight; if (has_weight) @@ -846,7 +846,7 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { } if ((mask & CGROUP_MASK_MEMORY) && !is_root) { - if (cg_unified() > 0) { + if (cg_all_unified() > 0) { uint64_t max = c->memory_max; if (cgroup_context_has_unified_memory_config(c)) @@ -1019,7 +1019,7 @@ CGroupMask unit_get_own_mask(Unit *u) { e = unit_get_exec_context(u); if (!e || exec_context_maintains_privileges(e) || - cg_unified() > 0) + cg_all_unified() > 0) return _CGROUP_MASK_ALL; } @@ -1245,7 +1245,7 @@ int unit_watch_cgroup(Unit *u) { return 0; /* Only applies to the unified hierarchy */ - r = cg_unified(); + r = cg_unified(SYSTEMD_CGROUP_CONTROLLER); if (r < 0) return log_unit_error_errno(u, r, "Failed detect whether the unified hierarchy is used: %m"); if (r == 0) @@ -1645,7 +1645,7 @@ int unit_watch_all_pids(Unit *u) { if (!u->cgroup_path) return -ENOENT; - if (cg_unified() > 0) /* On unified we can use proper notifications */ + if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* On unified we can use proper notifications */ return 0; return unit_watch_pids_in_path(u, u->cgroup_path); @@ -1718,7 +1718,7 @@ static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, int manager_setup_cgroup(Manager *m) { _cleanup_free_ char *path = NULL; CGroupController c; - int r, unified; + int r, all_unified, systemd_unified; char *e; assert(m); @@ -1755,11 +1755,17 @@ int manager_setup_cgroup(Manager *m) { if (r < 0) return log_error_errno(r, "Cannot find cgroup mount point: %m"); - unified = cg_unified(); - if (unified < 0) - return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); - if (unified > 0) + all_unified = cg_all_unified(); + systemd_unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER); + + if (all_unified < 0 || systemd_unified < 0) + return log_error_errno(all_unified < 0 ? all_unified : systemd_unified, + "Couldn't determine if we are running in the unified hierarchy: %m"); + + if (all_unified > 0) log_debug("Unified cgroup hierarchy is located at %s.", path); + else if (systemd_unified > 0) + log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); else log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path); @@ -1767,7 +1773,7 @@ int manager_setup_cgroup(Manager *m) { const char *scope_path; /* 3. Install agent */ - if (unified) { + if (systemd_unified) { /* In the unified hierarchy we can get * cgroup empty notifications via inotify. */ @@ -1827,7 +1833,7 @@ int manager_setup_cgroup(Manager *m) { return log_error_errno(errno, "Failed to open pin file: %m"); /* 6. Always enable hierarchical support if it exists... */ - if (!unified) + if (!all_unified) (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); } @@ -1953,7 +1959,7 @@ int unit_get_memory_current(Unit *u, uint64_t *ret) { if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) return -ENODATA; - if (cg_unified() <= 0) + if (cg_all_unified() <= 0) r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); else r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); @@ -1998,7 +2004,7 @@ static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { if (!u->cgroup_path) return -ENODATA; - if (cg_unified() > 0) { + if (cg_all_unified() > 0) { const char *keys[] = { "usage_usec", NULL }; _cleanup_free_ char *val = NULL; uint64_t us; diff --git a/src/core/manager.c b/src/core/manager.c index 6f2477eef4..7576d038a2 100644 --- a/src/core/manager.c +++ b/src/core/manager.c @@ -771,7 +771,7 @@ static int manager_setup_cgroups_agent(Manager *m) { if (!MANAGER_IS_SYSTEM(m)) return 0; - if (cg_unified() > 0) /* We don't need this anymore on the unified hierarchy */ + if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* We don't need this anymore on the unified hierarchy */ return 0; if (m->cgroups_agent_fd < 0) { diff --git a/src/core/mount-setup.c b/src/core/mount-setup.c index 5d8ab0ec70..ca63a93e8b 100644 --- a/src/core/mount-setup.c +++ b/src/core/mount-setup.c @@ -99,10 +99,12 @@ static const MountPoint mount_table[] = { cg_is_unified_wanted, MNT_FATAL|MNT_IN_CONTAINER }, { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER }, + { "cgroup", "/sys/fs/cgroup/systemd", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, + cg_is_unified_systemd_controller_wanted, MNT_IN_CONTAINER }, { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV, - cg_is_legacy_wanted, MNT_IN_CONTAINER }, + cg_is_legacy_systemd_controller_wanted, MNT_IN_CONTAINER }, { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV, - cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER }, + cg_is_legacy_systemd_controller_wanted, MNT_FATAL|MNT_IN_CONTAINER }, { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL, MNT_NONE }, #ifdef ENABLE_EFI diff --git a/src/core/scope.c b/src/core/scope.c index b278aed3d6..65fa65493b 100644 --- a/src/core/scope.c +++ b/src/core/scope.c @@ -441,7 +441,7 @@ static void scope_sigchld_event(Unit *u, pid_t pid, int code, int status) { /* If the PID set is empty now, then let's finish this off (On unified we use proper notifications) */ - if (cg_unified() <= 0 && set_isempty(u->pids)) + if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) <= 0 && set_isempty(u->pids)) scope_notify_cgroup_empty_event(u); } diff --git a/src/core/service.c b/src/core/service.c index 1951ba9222..969c62bd83 100644 --- a/src/core/service.c +++ b/src/core/service.c @@ -2869,7 +2869,7 @@ static void service_sigchld_event(Unit *u, pid_t pid, int code, int status) { /* If the PID set is empty now, then let's finish this off (On unified we use proper notifications) */ - if (cg_unified() <= 0 && set_isempty(u->pids)) + if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) <= 0 && set_isempty(u->pids)) service_notify_cgroup_empty_event(u); } diff --git a/src/core/unit.c b/src/core/unit.c index 4b8d81c3f1..c58c501bc3 100644 --- a/src/core/unit.c +++ b/src/core/unit.c @@ -3726,7 +3726,7 @@ int unit_kill_context( * there we get proper events. Hence rely on * them.*/ - if (cg_unified() > 0 || + if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0 || (detect_container() == 0 && !unit_cgroup_delegate(u))) wait_for_exit = true; diff --git a/src/libsystemd/sd-bus/test-bus-creds.c b/src/libsystemd/sd-bus/test-bus-creds.c index e9ef483bdd..82237af115 100644 --- a/src/libsystemd/sd-bus/test-bus-creds.c +++ b/src/libsystemd/sd-bus/test-bus-creds.c @@ -27,7 +27,7 @@ int main(int argc, char *argv[]) { _cleanup_(sd_bus_creds_unrefp) sd_bus_creds *creds = NULL; int r; - if (cg_unified() == -ENOMEDIUM) { + if (cg_all_unified() == -ENOMEDIUM) { puts("Skipping test: /sys/fs/cgroup/ not available"); return EXIT_TEST_SKIP; } diff --git a/src/nspawn/nspawn-cgroup.c b/src/nspawn/nspawn-cgroup.c index b1580236c9..aa0da04955 100644 --- a/src/nspawn/nspawn-cgroup.c +++ b/src/nspawn/nspawn-cgroup.c @@ -20,7 +20,6 @@ #include <sys/mount.h> #include "alloc-util.h" -#include "cgroup-util.h" #include "fd-util.h" #include "fileio.h" #include "mkdir.h" @@ -63,18 +62,18 @@ int chown_cgroup(pid_t pid, uid_t uid_shift) { return 0; } -int sync_cgroup(pid_t pid, bool unified_requested) { +int sync_cgroup(pid_t pid, CGroupUnified unified_requested) { _cleanup_free_ char *cgroup = NULL; char tree[] = "/tmp/unifiedXXXXXX", pid_string[DECIMAL_STR_MAX(pid) + 1]; bool undo_mount = false; const char *fn; int unified, r; - unified = cg_unified(); + unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER); if (unified < 0) return log_error_errno(unified, "Failed to determine whether the unified hierarchy is used: %m"); - if ((unified > 0) == unified_requested) + if ((unified > 0) == (unified_requested >= CGROUP_UNIFIED_SYSTEMD)) return 0; /* When the host uses the legacy cgroup setup, but the @@ -117,7 +116,7 @@ finish: return r; } -int create_subcgroup(pid_t pid, bool unified_requested) { +int create_subcgroup(pid_t pid, CGroupUnified unified_requested) { _cleanup_free_ char *cgroup = NULL; const char *child; int unified, r; @@ -129,10 +128,10 @@ int create_subcgroup(pid_t pid, bool unified_requested) { * did not create a scope unit for the container move us and * the container into two separate subcgroups. */ - if (!unified_requested) + if (unified_requested == CGROUP_UNIFIED_NONE) return 0; - unified = cg_unified(); + unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER); if (unified < 0) return log_error_errno(unified, "Failed to determine whether the unified hierarchy is used: %m"); if (unified == 0) diff --git a/src/nspawn/nspawn-cgroup.h b/src/nspawn/nspawn-cgroup.h index 1ff35a299a..dc33da8abe 100644 --- a/src/nspawn/nspawn-cgroup.h +++ b/src/nspawn/nspawn-cgroup.h @@ -22,6 +22,8 @@ #include <stdbool.h> #include <sys/types.h> +#include "cgroup-util.h" + int chown_cgroup(pid_t pid, uid_t uid_shift); -int sync_cgroup(pid_t pid, bool unified_requested); -int create_subcgroup(pid_t pid, bool unified_requested); +int sync_cgroup(pid_t pid, CGroupUnified unified_requested); +int create_subcgroup(pid_t pid, CGroupUnified unified_requested); diff --git a/src/nspawn/nspawn-mount.c b/src/nspawn/nspawn-mount.c index 803caef3dd..295b75341f 100644 --- a/src/nspawn/nspawn-mount.c +++ b/src/nspawn/nspawn-mount.c @@ -21,7 +21,6 @@ #include <linux/magic.h> #include "alloc-util.h" -#include "cgroup-util.h" #include "escape.h" #include "fd-util.h" #include "fileio.h" @@ -661,7 +660,8 @@ static int get_controllers(Set *subsystems) { return 0; } -static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, bool read_only) { +static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, + CGroupUnified unified_requested, bool read_only) { char *to; int r; @@ -677,7 +677,15 @@ static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controlle /* The superblock mount options of the mount point need to be * identical to the hosts', and hence writable... */ - if (mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller) < 0) + if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) + r = mount("cgroup", to, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL); + else + r = mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr"); + } else + r = mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller); + + if (r < 0) return log_error_errno(errno, "Failed to mount to %s: %m", to); /* ... hence let's only make the bind mount read-only, not the @@ -691,8 +699,8 @@ static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controlle /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */ static int mount_legacy_cgns_supported( - bool userns, uid_t uid_shift, uid_t uid_range, - const char *selinux_apifs_context) { + CGroupUnified unified_requested, bool userns, uid_t uid_shift, + uid_t uid_range, const char *selinux_apifs_context) { _cleanup_set_free_free_ Set *controllers = NULL; const char *cgroup_root = "/sys/fs/cgroup", *c; int r; @@ -721,7 +729,7 @@ static int mount_legacy_cgns_supported( return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m"); } - if (cg_unified() > 0) + if (cg_all_unified() > 0) goto skip_controllers; controllers = set_new(&string_hash_ops); @@ -739,7 +747,7 @@ static int mount_legacy_cgns_supported( if (!controller) break; - r = mount_legacy_cgroup_hierarchy("", controller, controller, !userns); + r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns); if (r < 0) return r; @@ -773,7 +781,7 @@ static int mount_legacy_cgns_supported( } skip_controllers: - r = mount_legacy_cgroup_hierarchy("", "none,name=systemd,xattr", "systemd", false); + r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false); if (r < 0) return r; @@ -788,7 +796,7 @@ skip_controllers: /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */ static int mount_legacy_cgns_unsupported( const char *dest, - bool userns, uid_t uid_shift, uid_t uid_range, + CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context) { _cleanup_set_free_free_ Set *controllers = NULL; const char *cgroup_root; @@ -813,7 +821,7 @@ static int mount_legacy_cgns_unsupported( return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m"); } - if (cg_unified() > 0) + if (cg_all_unified() > 0) goto skip_controllers; controllers = set_new(&string_hash_ops); @@ -839,7 +847,7 @@ static int mount_legacy_cgns_unsupported( if (r == -EINVAL) { /* Not a symbolic link, but directly a single cgroup hierarchy */ - r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true); + r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true); if (r < 0) return r; @@ -859,7 +867,7 @@ static int mount_legacy_cgns_unsupported( continue; } - r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true); + r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true); if (r < 0) return r; @@ -872,7 +880,7 @@ static int mount_legacy_cgns_unsupported( } skip_controllers: - r = mount_legacy_cgroup_hierarchy(dest, "none,name=systemd,xattr", "systemd", false); + r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false); if (r < 0) return r; @@ -914,22 +922,22 @@ static int mount_unified_cgroups(const char *dest) { int mount_cgroups( const char *dest, - bool unified_requested, + CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context, bool use_cgns) { - if (unified_requested) + if (unified_requested >= CGROUP_UNIFIED_ALL) return mount_unified_cgroups(dest); else if (use_cgns && cg_ns_supported()) - return mount_legacy_cgns_supported(userns, uid_shift, uid_range, selinux_apifs_context); + return mount_legacy_cgns_supported(unified_requested, userns, uid_shift, uid_range, selinux_apifs_context); - return mount_legacy_cgns_unsupported(dest, userns, uid_shift, uid_range, selinux_apifs_context); + return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context); } int mount_systemd_cgroup_writable( const char *dest, - bool unified_requested) { + CGroupUnified unified_requested) { _cleanup_free_ char *own_cgroup_path = NULL; const char *systemd_root, *systemd_own; @@ -945,7 +953,7 @@ int mount_systemd_cgroup_writable( if (path_equal(own_cgroup_path, "/")) return 0; - if (unified_requested) { + if (unified_requested >= CGROUP_UNIFIED_ALL) { systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path); systemd_root = prefix_roota(dest, "/sys/fs/cgroup"); } else { diff --git a/src/nspawn/nspawn-mount.h b/src/nspawn/nspawn-mount.h index 0eff8e1006..7307a838a5 100644 --- a/src/nspawn/nspawn-mount.h +++ b/src/nspawn/nspawn-mount.h @@ -21,6 +21,8 @@ #include <stdbool.h> +#include "cgroup-util.h" + typedef enum VolatileMode { VOLATILE_NO, VOLATILE_YES, @@ -58,8 +60,8 @@ int custom_mount_compare(const void *a, const void *b); int mount_all(const char *dest, bool use_userns, bool in_userns, bool use_netns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context); int mount_sysfs(const char *dest); -int mount_cgroups(const char *dest, bool unified_requested, bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context, bool use_cgns); -int mount_systemd_cgroup_writable(const char *dest, bool unified_requested); +int mount_cgroups(const char *dest, CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context, bool use_cgns); +int mount_systemd_cgroup_writable(const char *dest, CGroupUnified unified_requested); int mount_custom(const char *dest, CustomMount *mounts, unsigned n, bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context); diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c index fcf14bba4c..24d243109a 100644 --- a/src/nspawn/nspawn.c +++ b/src/nspawn/nspawn.c @@ -188,7 +188,7 @@ static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO; static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U; static bool arg_userns_chown = false; static int arg_kill_signal = 0; -static bool arg_unified_cgroup_hierarchy = false; +static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN; static SettingsMask arg_settings_mask = 0; static int arg_settings_trusted = -1; static char **arg_parameters = NULL; @@ -318,7 +318,14 @@ static int custom_mounts_prepare(void) { static int detect_unified_cgroup_hierarchy(void) { const char *e; - int r; + int r, all_unified, systemd_unified; + + all_unified = cg_all_unified(); + systemd_unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER); + + if (all_unified < 0 || systemd_unified < 0) + return log_error_errno(all_unified < 0 ? all_unified : systemd_unified, + "Failed to determine whether the unified cgroups hierarchy is used: %m"); /* Allow the user to control whether the unified hierarchy is used */ e = getenv("UNIFIED_CGROUP_HIERARCHY"); @@ -326,17 +333,22 @@ static int detect_unified_cgroup_hierarchy(void) { r = parse_boolean(e); if (r < 0) return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY."); + if (r > 0) + arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL; + else + arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE; - arg_unified_cgroup_hierarchy = r; return 0; } /* Otherwise inherit the default from the host system */ - r = cg_unified(); - if (r < 0) - return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m"); + if (all_unified > 0) + arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL; + else if (systemd_unified > 0) + arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD; + else + arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE; - arg_unified_cgroup_hierarchy = r; return 0; } |