X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fbasic%2Fcgroup-util.c;h=315c449656aa073ce6aadad84e0506b82e246a6f;hp=b5de1cc143f55f9ec8c8ed719d22a873fb9eef75;hb=d93247127eb2e073a6d3b5bcc67bcc4048d674fe;hpb=eaca07ccfdf5d7dabc50afc7e539c2413dd69d3e diff --git a/src/basic/cgroup-util.c b/src/basic/cgroup-util.c index b5de1cc14..315c44965 100644 --- a/src/basic/cgroup-util.c +++ b/src/basic/cgroup-util.c @@ -1,5 +1,3 @@ -/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ - /*** This file is part of systemd. @@ -22,23 +20,30 @@ #include #include #include +//#include #include +//#include #include #include #include +//#include #include +#include #include #include "alloc-util.h" #include "cgroup-util.h" +//#include "def.h" #include "dirent-util.h" #include "extract-word.h" #include "fd-util.h" #include "fileio.h" -#include "formats-util.h" +#include "format-util.h" #include "fs-util.h" +//#include "log.h" #include "login-util.h" #include "macro.h" +//#include "missing.h" #include "mkdir.h" #include "parse-util.h" #include "path-util.h" @@ -47,11 +52,12 @@ #include "set.h" //#include "special.h" #include "stat-util.h" +#include "stdio-util.h" #include "string-table.h" #include "string-util.h" +#include "strv.h" #include "unit-name.h" #include "user-util.h" -#include "util.h" int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) { _cleanup_free_ char *fs = NULL; @@ -87,7 +93,7 @@ int cg_read_pid(FILE *f, pid_t *_pid) { if (feof(f)) return 0; - return errno ? -errno : -EIO; + return errno > 0 ? -errno : -EIO; } if (ul <= 0) @@ -97,6 +103,55 @@ int cg_read_pid(FILE *f, pid_t *_pid) { return 1; } +int cg_read_event(const char *controller, const char *path, const char *event, + char **val) +{ + _cleanup_free_ char *events = NULL, *content = NULL; + char *p, *line; + int r; + + r = cg_get_path(controller, path, "cgroup.events", &events); + if (r < 0) + return r; + + r = read_full_file(events, &content, NULL); + if (r < 0) + return r; + + p = content; + while ((line = strsep(&p, "\n"))) { + char *key; + + key = strsep(&line, " "); + if (!key || !line) + return -EINVAL; + + if (strcmp(key, event)) + continue; + + *val = strdup(line); + return 0; + } + + return -ENOENT; +} + +#if 0 /// UNNEEDED by elogind +bool cg_ns_supported(void) { + static thread_local int enabled = -1; + + if (enabled >= 0) + return enabled; + + if (access("/proc/self/ns/cgroup", F_OK) == 0) + enabled = 1; + else + enabled = 0; + + return enabled; +} +#endif //0 + int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) { _cleanup_free_ char *fs = NULL; int r; @@ -130,8 +185,7 @@ int cg_read_subgroup(DIR *d, char **fn) { if (de->d_type != DT_DIR) continue; - if (streq(de->d_name, ".") || - streq(de->d_name, "..")) + if (dot_or_dot_dot(de->d_name)) continue; b = strdup(de->d_name); @@ -157,10 +211,30 @@ int cg_rmdir(const char *controller, const char *path) { if (r < 0 && errno != ENOENT) return -errno; + r = cg_hybrid_unified(); + if (r < 0) + return r; + if (r == 0) + return 0; + + if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); + if (r < 0) + log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path); + } + return 0; } -int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) { +int cg_kill( + const char *controller, + const char *path, + int sig, + CGroupFlags flags, + Set *s, + cg_kill_log_func_t log_kill, + void *userdata) { + _cleanup_set_free_ Set *allocated_set = NULL; bool done = false; int r, ret = 0; @@ -168,6 +242,11 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo assert(sig >= 0); + /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send + * SIGCONT on SIGKILL. */ + if (IN_SET(sig, SIGCONT, SIGKILL)) + flags &= ~CGROUP_SIGCONT; + /* This goes through the tasks list and kills them all. This * is repeated until no further processes are added to the * tasks list, to properly handle forking processes */ @@ -195,19 +274,22 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo while ((r = cg_read_pid(f, &pid)) > 0) { - if (ignore_self && pid == my_pid) + if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) continue; if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) continue; + if (log_kill) + log_kill(pid, sig, userdata); + /* If we haven't killed this process yet, kill * it */ if (kill(pid, sig) < 0) { if (ret >= 0 && errno != ESRCH) ret = -errno; } else { - if (sigcont && sig != SIGKILL) + if (flags & CGROUP_SIGCONT) (void) kill(pid, SIGCONT); if (ret == 0) @@ -241,7 +323,15 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo return ret; } -int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) { +int cg_kill_recursive( + const char *controller, + const char *path, + int sig, + CGroupFlags flags, + Set *s, + cg_kill_log_func_t log_kill, + void *userdata) { + _cleanup_set_free_ Set *allocated_set = NULL; _cleanup_closedir_ DIR *d = NULL; int r, ret; @@ -256,7 +346,7 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si return -ENOMEM; } - ret = cg_kill(controller, path, sig, sigcont, ignore_self, s); + ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata); r = cg_enumerate_subgroups(controller, path, &d); if (r < 0) { @@ -269,20 +359,19 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si while ((r = cg_read_subgroup(d, &fn)) > 0) { _cleanup_free_ char *p = NULL; - p = strjoin(path, "/", fn, NULL); + p = strjoin(path, "/", fn); free(fn); if (!p) return -ENOMEM; - r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s); + r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata); if (r != 0 && ret >= 0) ret = r; } - if (ret >= 0 && r < 0) ret = r; - if (rem) { + if (flags & CGROUP_REMOVE) { r = cg_rmdir(controller, path); if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY) return r; @@ -291,7 +380,13 @@ int cg_kill_recursive(const char *controller, const char *path, int sig, bool si return ret; } -int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char *pto, bool ignore_self) { +int cg_migrate( + const char *cfrom, + const char *pfrom, + const char *cto, + const char *pto, + CGroupFlags flags) { + bool done = false; _cleanup_set_free_ Set *s = NULL; int r, ret = 0; @@ -310,8 +405,8 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char log_debug_elogind("Migrating \"%s\"/\"%s\" to \"%s\"/\"%s\" (%s)", cfrom, pfrom, cto, pto, - ignore_self ? "ignoring self" : "watching self"); - + (flags & CGROUP_IGNORE_SELF) + ? "ignoring self" : "watching self"); do { _cleanup_fclose_ FILE *f = NULL; pid_t pid = 0; @@ -330,7 +425,7 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char /* This might do weird stuff if we aren't a * single-threaded program. However, we * luckily know we are not */ - if (ignore_self && pid == my_pid) + if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) continue; if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) @@ -378,8 +473,7 @@ int cg_migrate_recursive( const char *pfrom, const char *cto, const char *pto, - bool ignore_self, - bool rem) { + CGroupFlags flags) { _cleanup_closedir_ DIR *d = NULL; int r, ret = 0; @@ -390,7 +484,7 @@ int cg_migrate_recursive( assert(cto); assert(pto); - ret = cg_migrate(cfrom, pfrom, cto, pto, ignore_self); + ret = cg_migrate(cfrom, pfrom, cto, pto, flags); r = cg_enumerate_subgroups(cfrom, pfrom, &d); if (r < 0) { @@ -403,12 +497,12 @@ int cg_migrate_recursive( while ((r = cg_read_subgroup(d, &fn)) > 0) { _cleanup_free_ char *p = NULL; - p = strjoin(pfrom, "/", fn, NULL); + p = strjoin(pfrom, "/", fn); free(fn); if (!p) - return -ENOMEM; + return -ENOMEM; - r = cg_migrate_recursive(cfrom, p, cto, pto, ignore_self, rem); + r = cg_migrate_recursive(cfrom, p, cto, pto, flags); if (r != 0 && ret >= 0) ret = r; } @@ -416,7 +510,7 @@ int cg_migrate_recursive( if (r < 0 && ret >= 0) ret = r; - if (rem) { + if (flags & CGROUP_REMOVE) { r = cg_rmdir(cfrom, pfrom); if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY) return r; @@ -430,8 +524,7 @@ int cg_migrate_recursive_fallback( const char *pfrom, const char *cto, const char *pto, - bool ignore_self, - bool rem) { + CGroupFlags flags) { int r; @@ -440,7 +533,7 @@ int cg_migrate_recursive_fallback( assert(cto); assert(pto); - r = cg_migrate_recursive(cfrom, pfrom, cto, pto, ignore_self, rem); + r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags); if (r < 0) { char prefix[strlen(pto) + 1]; @@ -449,7 +542,7 @@ int cg_migrate_recursive_fallback( PATH_FOREACH_PREFIX(prefix, pto) { int q; - q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, ignore_self, rem); + q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags); if (q >= 0) return q; } @@ -468,6 +561,13 @@ static const char *controller_to_dirname(const char *controller) { * just cuts off the name= prefixed used for named * hierarchies, if it is specified. */ + if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + if (cg_hybrid_unified() > 0) + controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID; + else + controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY; + } + e = startswith(controller, "name="); if (e) return e; @@ -487,17 +587,17 @@ static int join_path_legacy(const char *controller, const char *path, const char if (isempty(path) && isempty(suffix)) t = strappend("/sys/fs/cgroup/", dn); else if (isempty(path)) - t = strjoin("/sys/fs/cgroup/", dn, "/", suffix, NULL); + t = strjoin("/sys/fs/cgroup/", dn, "/", suffix); else if (isempty(suffix)) - t = strjoin("/sys/fs/cgroup/", dn, "/", path, NULL); - else - t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix, NULL); + t = strjoin("/sys/fs/cgroup/", dn, "/", path); + else + t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix); if (!t) return -ENOMEM; *fs = t; return 0; - } +} static int join_path_unified(const char *path, const char *suffix, char **fs) { char *t; @@ -511,7 +611,7 @@ static int join_path_unified(const char *path, const char *suffix, char **fs) { else if (isempty(suffix)) t = strappend("/sys/fs/cgroup/", path); else - t = strjoin("/sys/fs/cgroup/", path, "/", suffix, NULL); + t = strjoin("/sys/fs/cgroup/", path, "/", suffix); if (!t) return -ENOMEM; @@ -520,7 +620,7 @@ static int join_path_unified(const char *path, const char *suffix, char **fs) { } int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) { - int unified, r; + int r; assert(fs); @@ -538,7 +638,7 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch else if (!path) t = strdup(suffix); else - t = strjoin(path, "/", suffix, NULL); + t = strjoin(path, "/", suffix); if (!t) return -ENOMEM; @@ -549,15 +649,13 @@ 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(); - if (unified < 0) - return unified; - - if (unified > 0) + r = cg_all_unified(); + if (r < 0) + return r; + if (r > 0) r = join_path_unified(path, suffix, fs); else r = join_path_legacy(controller, path, suffix, fs); - if (r < 0) return r; @@ -566,7 +664,7 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch } static int controller_is_accessible(const char *controller) { - int unified; + int r; assert(controller); @@ -578,10 +676,10 @@ static int controller_is_accessible(const char *controller) { if (!cg_controller_is_valid(controller)) return -EINVAL; - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified > 0) { + r = cg_all_unified(); + if (r < 0) + return r; + if (r > 0) { /* We don't support named hierarchies if we are using * the unified hierarchy. */ @@ -635,7 +733,7 @@ static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct int cg_trim(const char *controller, const char *path, bool delete_root) { _cleanup_free_ char *fs = NULL; - int r = 0; + int r = 0, q; assert(path); @@ -647,7 +745,7 @@ int cg_trim(const char *controller, const char *path, bool delete_root) { if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) { if (errno == ENOENT) r = 0; - else if (errno != 0) + else if (errno > 0) r = -errno; else r = -EIO; @@ -658,6 +756,15 @@ int cg_trim(const char *controller, const char *path, bool delete_root) { return -errno; } + q = cg_hybrid_unified(); + if (q < 0) + return q; + if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root); + if (q < 0) + log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path); + } + return r; } @@ -681,6 +788,16 @@ int cg_create(const char *controller, const char *path) { return -errno; } + r = cg_hybrid_unified(); + if (r < 0) + return r; + + if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); + if (r < 0) + log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path); + } + return 1; } @@ -716,9 +833,23 @@ int cg_attach(const char *controller, const char *path, pid_t pid) { if (pid == 0) pid = getpid(); - snprintf(c, sizeof(c), PID_FMT"\n", pid); + xsprintf(c, PID_FMT "\n", pid); - return write_string_file(fs, c, 0); + r = write_string_file(fs, c, 0); + if (r < 0) + return r; + + r = cg_hybrid_unified(); + if (r < 0) + return r; + + if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid); + if (r < 0) + log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path); + } + + return 0; } int cg_attach_fallback(const char *controller, const char *path, pid_t pid) { @@ -747,8 +878,7 @@ int cg_attach_fallback(const char *controller, const char *path, pid_t pid) { return r; } -/// UNNEEDED by elogind -#if 0 +#if 0 /// UNNEEDED by elogind int cg_set_group_access( const char *controller, const char *path, @@ -769,7 +899,20 @@ int cg_set_group_access( if (r < 0) return r; - return chmod_and_chown(fs, mode, uid, gid); + r = chmod_and_chown(fs, mode, uid, gid); + if (r < 0) + return r; + + r = cg_hybrid_unified(); + if (r < 0) + return r; + if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid); + if (r < 0) + log_warning_errno(r, "Failed to set group access on compat systemd cgroup %s: %m", path); + } + + return 0; } int cg_set_task_access( @@ -780,7 +923,7 @@ int cg_set_task_access( gid_t gid) { _cleanup_free_ char *fs = NULL, *procs = NULL; - int r, unified; + int r; assert(path); @@ -798,42 +941,92 @@ int cg_set_task_access( if (r < 0) return r; - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified) - return 0; + r = cg_unified_controller(controller); + if (r < 0) + return r; + if (r == 0) { + /* Compatibility, Always keep values for "tasks" in sync with + * "cgroup.procs" */ + if (cg_get_path(controller, path, "tasks", &procs) >= 0) + (void) chmod_and_chown(procs, mode, uid, gid); + } - /* Compatibility, Always keep values for "tasks" in sync with - * "cgroup.procs" */ - if (cg_get_path(controller, path, "tasks", &procs) >= 0) - (void) chmod_and_chown(procs, mode, uid, gid); + r = cg_hybrid_unified(); + if (r < 0) + return r; + if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid); + if (r < 0) + log_warning_errno(r, "Failed to set task access on compat systemd cgroup %s: %m", path); + } return 0; } + +int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) { + _cleanup_free_ char *fs = NULL; + int r; + + assert(path); + assert(name); + assert(value || size <= 0); + + r = cg_get_path(controller, path, NULL, &fs); + if (r < 0) + return r; + + if (setxattr(fs, name, value, size, flags) < 0) + return -errno; + + return 0; +} + +int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) { + _cleanup_free_ char *fs = NULL; + ssize_t n; + int r; + + assert(path); + assert(name); + + r = cg_get_path(controller, path, NULL, &fs); + if (r < 0) + return r; + + n = getxattr(fs, name, value, size); + if (n < 0) + return -errno; + + return (int) n; +} #endif // 0 int cg_pid_get_path(const char *controller, pid_t pid, char **path) { _cleanup_fclose_ FILE *f = NULL; char line[LINE_MAX]; - const char *fs; + const char *fs, *controller_str = NULL; size_t cs = 0; int unified; 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(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 (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) + controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY; + else + controller_str = controller; - cs = strlen(controller); + cs = strlen(controller_str); } fs = procfs_file_alloca(pid, "cgroup"); @@ -873,7 +1066,7 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) { *e = 0; FOREACH_WORD_SEPARATOR(word, k, l, ",", state) { - if (k == cs && memcmp(word, controller, cs) == 0) { + if (k == cs && memcmp(word, controller_str, cs) == 0) { found = true; break; } @@ -898,14 +1091,14 @@ int cg_pid_get_path(const char *controller, pid_t pid, char **path) { int cg_install_release_agent(const char *controller, const char *agent) { _cleanup_free_ char *fs = NULL, *contents = NULL; const char *sc; - int r, unified; + int r; assert(agent); - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified) /* doesn't apply to unified hierarchy */ + r = cg_unified_controller(controller); + if (r < 0) + return r; + if (r > 0) /* doesn't apply to unified hierarchy */ return -EOPNOTSUPP; r = cg_get_path(controller, NULL, "release_agent", &fs); @@ -951,12 +1144,12 @@ int cg_install_release_agent(const char *controller, const char *agent) { int cg_uninstall_release_agent(const char *controller) { _cleanup_free_ char *fs = NULL; - int r, unified; + int r; - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified) /* Doesn't apply to unified hierarchy */ + r = cg_unified_controller(controller); + if (r < 0) + return r; + if (r > 0) /* Doesn't apply to unified hierarchy */ return -EOPNOTSUPP; r = cg_get_path(controller, NULL, "notify_on_release", &fs); @@ -1001,7 +1194,7 @@ int cg_is_empty(const char *controller, const char *path) { } int cg_is_empty_recursive(const char *controller, const char *path) { - int unified, r; + int r; assert(path); @@ -1009,23 +1202,16 @@ int cg_is_empty_recursive(const char *controller, const char *path) { if (controller && (isempty(path) || path_equal(path, "/"))) return false; - unified = cg_unified(); - if (unified < 0) - return unified; - - if (unified > 0) { - _cleanup_free_ char *populated = NULL, *t = NULL; + r = cg_unified_controller(controller); + if (r < 0) + return r; + if (r > 0) { + _cleanup_free_ char *t = NULL; /* On the unified hierarchy we can check empty state - * via the "cgroup.populated" attribute. */ + * via the "populated" attribute of "cgroup.events". */ - r = cg_get_path(controller, path, "cgroup.populated", &populated); - if (r < 0) - return r; - - r = read_one_line_file(populated, &t); - if (r == -ENOENT) - return 1; + r = cg_read_event(controller, path, "populated", &t); if (r < 0) return r; @@ -1039,15 +1225,15 @@ int cg_is_empty_recursive(const char *controller, const char *path) { return r; r = cg_enumerate_subgroups(controller, path, &d); - if (r == -ENOENT) - return 1; + if (r == -ENOENT) + return 1; if (r < 0) return r; while ((r = cg_read_subgroup(d, &fn)) > 0) { _cleanup_free_ char *p = NULL; - p = strjoin(path, "/", fn, NULL); + p = strjoin(path, "/", fn); free(fn); if (!p) return -ENOMEM; @@ -1174,8 +1360,7 @@ int cg_mangle_path(const char *path, char **result) { } int cg_get_root_path(char **path) { -/// elogind does not support systemd scopes and slices -#if 0 +#if 0 /// elogind does not support systemd scopes and slices char *p, *e; int r; @@ -1222,7 +1407,11 @@ int cg_shift_path(const char *cgroup, const char *root, const char **shifted) { } p = path_startswith(cgroup, root); +#if 0 /// With other controllers, elogind might end up in /elogind, and *p is 0 + if (p && p > cgroup) +#else if (p && p[0] && (p > cgroup)) +#endif // 0 *shifted = p - 1; else *shifted = cgroup; @@ -1265,9 +1454,8 @@ int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) { return 0; } -/// UNNEEDED by elogind -#if 0 -int cg_path_decode_unit(const char *cgroup, char **unit){ +#if 0 /// UNNEEDED by elogind +int cg_path_decode_unit(const char *cgroup, char **unit) { char *c, *s; size_t n; @@ -1518,9 +1706,7 @@ int cg_pid_get_machine_name(pid_t pid, char **machine) { #endif // 0 int cg_path_get_session(const char *path, char **session) { - /* Elogind uses a flat hierarchy, just "/SESSION". The only - wrinkle is that SESSION might be escaped. */ -#if 0 +#if 0 /// UNNEEDED by elogind _cleanup_free_ char *unit = NULL; char *start, *end; int r; @@ -1542,6 +1728,8 @@ int cg_path_get_session(const char *path, char **session) { if (!session_id_valid(start)) return -ENXIO; #else + /* Elogind uses a flat hierarchy, just "/SESSION". The only + wrinkle is that SESSION might be escaped. */ const char *e, *n, *start; assert(path); @@ -1585,8 +1773,7 @@ int cg_pid_get_session(pid_t pid, char **session) { return cg_path_get_session(cgroup, session); } -/// UNNEEDED by elogind -#if 0 +#if 0 /// UNNEEDED by elogind int cg_path_get_owner_uid(const char *path, uid_t *uid) { _cleanup_free_ char *slice = NULL; char *start, *end; @@ -1643,7 +1830,7 @@ int cg_path_get_slice(const char *p, char **slice) { if (!e) { char *s; - s = strdup("-.slice"); + s = strdup(SPECIAL_ROOT_SLICE); if (!s) return -ENOMEM; @@ -1774,6 +1961,9 @@ bool cg_controller_is_valid(const char *p) { if (!p) return false; + if (streq(p, SYSTEMD_CGROUP_CONTROLLER)) + return true; + s = startswith(p, "name="); if (s) p = s; @@ -1791,8 +1981,7 @@ bool cg_controller_is_valid(const char *p) { return true; } -/// UNNEEDED by elogind -#if 0 +#if 0 /// UNNEEDED by elogind int cg_slice_to_path(const char *unit, char **ret) { _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL; const char *dash; @@ -1801,7 +1990,7 @@ int cg_slice_to_path(const char *unit, char **ret) { assert(unit); assert(ret); - if (streq(unit, "-.slice")) { + if (streq(unit, SPECIAL_ROOT_SLICE)) { char *x; x = strdup(""); @@ -1874,8 +2063,6 @@ int cg_set_attribute(const char *controller, const char *path, const char *attri return write_string_file(p, value, 0); } -/// UNNEEDED by elogind -#if 0 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) { _cleanup_free_ char *p = NULL; int r; @@ -1887,9 +2074,53 @@ int cg_get_attribute(const char *controller, const char *path, const char *attri return read_one_line_file(p, ret); } +#if 0 /// UNNEEDED by elogind +int cg_get_keyed_attribute(const char *controller, const char *path, const char *attribute, const char **keys, char **values) { + _cleanup_free_ char *filename = NULL, *content = NULL; + char *line, *p; + int i, r; + + for (i = 0; keys[i]; i++) + values[i] = NULL; + + r = cg_get_path(controller, path, attribute, &filename); + if (r < 0) + return r; + + r = read_full_file(filename, &content, NULL); + if (r < 0) + return r; + + p = content; + while ((line = strsep(&p, "\n"))) { + char *key; + + key = strsep(&line, " "); + + for (i = 0; keys[i]; i++) { + if (streq(key, keys[i])) { + values[i] = strdup(line); + break; + } + } + } + + for (i = 0; keys[i]; i++) { + if (!values[i]) { + for (i = 0; keys[i]; i++) { + free(values[i]); + values[i] = NULL; + } + return -ENOENT; + } + } + + return 0; +} + int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) { CGroupController c; - int r, unified; + int r; /* This one will create a cgroup in our private tree, but also * duplicate it in the trees specified in mask, and remove it @@ -1901,10 +2132,10 @@ 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(); - if (unified < 0) - return unified; - if (unified > 0) + r = cg_all_unified(); + if (r < 0) + return r; + if (r > 0) return 0; /* Otherwise, do the same in the other hierarchies */ @@ -1925,30 +2156,30 @@ int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) { CGroupController c; - int r, unified; + int r; r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid); if (r < 0) return r; - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified > 0) + r = cg_all_unified(); + if (r < 0) + return r; + if (r > 0) return 0; for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); - const char *p = NULL; + const char *p = NULL; if (!(supported & bit)) continue; - if (path_callback) - p = path_callback(bit, userdata); + if (path_callback) + p = path_callback(bit, userdata); - if (!p) - p = path; + if (!p) + p = path; (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid); } @@ -1975,34 +2206,34 @@ int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) { CGroupController c; - int r = 0, unified; + int r = 0, q; if (!path_equal(from, to)) { - r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, false, true); + r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE); if (r < 0) return r; } - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified > 0) + q = cg_all_unified(); + if (q < 0) + return q; + if (q > 0) return r; for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); - const char *p = NULL; + const char *p = NULL; if (!(supported & bit)) continue; - if (to_callback) - p = to_callback(bit, userdata); + if (to_callback) + p = to_callback(bit, userdata); - if (!p) - p = to; + if (!p) + p = to; - (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, false, false); + (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0); } return 0; @@ -2010,16 +2241,16 @@ int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) { CGroupController c; - int r, unified; + int r, q; r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root); if (r < 0) return r; - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified > 0) + q = cg_all_unified(); + if (q < 0) + return q; + if (q > 0) return r; for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { @@ -2035,20 +2266,73 @@ int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) } #endif // 0 +int cg_mask_to_string(CGroupMask mask, char **ret) { + const char *controllers[_CGROUP_CONTROLLER_MAX + 1]; + CGroupController c; + int i = 0; + char *s; + + assert(ret); + + if (mask == 0) { + *ret = NULL; + return 0; + } + + for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { + + if (!(mask & CGROUP_CONTROLLER_TO_MASK(c))) + continue; + + controllers[i++] = cgroup_controller_to_string(c); + controllers[i] = NULL; + } + + s = strv_join((char **)controllers, NULL); + if (!s) + return -ENOMEM; + + *ret = s; + return 0; +} + +int cg_mask_from_string(const char *value, CGroupMask *mask) { + assert(mask); + assert(value); + + for (;;) { + _cleanup_free_ char *n = NULL; + CGroupController v; + int r; + + r = extract_first_word(&value, &n, NULL, 0); + if (r < 0) + return r; + if (r == 0) + break; + + v = cgroup_controller_from_string(n); + if (v < 0) + continue; + + *mask |= CGROUP_CONTROLLER_TO_MASK(v); + } + return 0; +} + int cg_mask_supported(CGroupMask *ret) { CGroupMask mask = 0; - int r, unified; + int r; /* Determines the mask of supported cgroup controllers. Only * includes controllers we can make sense of and that are * actually accessible. */ - unified = cg_unified(); - if (unified < 0) - return unified; - if (unified > 0) { + r = cg_all_unified(); + if (r < 0) + return r; + if (r > 0) { _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL; - const char *c; /* In the unified hierarchy we can read the supported * and accessible controllers from a the top-level @@ -2066,28 +2350,14 @@ int cg_mask_supported(CGroupMask *ret) { if (r < 0) return r; - c = controllers; - for (;;) { - _cleanup_free_ char *n = NULL; - CGroupController v; - - r = extract_first_word(&c, &n, NULL, 0); - if (r < 0) - return r; - if (r == 0) - break; - - v = cgroup_controller_from_string(n); - if (v < 0) - continue; - - mask |= CGROUP_CONTROLLER_TO_MASK(v); - } + r = cg_mask_from_string(controllers, &mask); + if (r < 0) + return r; - /* Currently, we only support the memory and pids + /* Currently, we support the cpu, memory, io and pids * controller in the unified hierarchy, mask * everything else off. */ - mask &= CGROUP_MASK_MEMORY | CGROUP_MASK_PIDS; + mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS; } else { CGroupController c; @@ -2108,8 +2378,7 @@ int cg_mask_supported(CGroupMask *ret) { return 0; } -/// UNNEEDED by elogind -#if 0 +#if 0 /// UNNEEDED by elogind int cg_kernel_controllers(Set *controllers) { _cleanup_fclose_ FILE *f = NULL; char buf[LINE_MAX]; @@ -2142,7 +2411,7 @@ int cg_kernel_controllers(Set *controllers) { if (feof(f)) break; - if (ferror(f) && errno != 0) + if (ferror(f) && errno > 0) return -errno; return -EBADMSG; @@ -2167,9 +2436,21 @@ int cg_kernel_controllers(Set *controllers) { } #endif // 0 -static thread_local int unified_cache = -1; +static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN; + +/* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd. This + * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on + * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains + * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools. + * + * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2 + * process management but disable the compat dual layout, we return %true on + * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified(). + */ +static thread_local bool unified_systemd_v232; + +static int cg_unified_update(void) { -int cg_unified(void) { struct statfs fs; /* Checks if we support the unified hierarchy. Returns an @@ -2177,51 +2458,102 @@ 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; -/// elogind can not support the unified hierarchy as a controller, -/// so always assume a classical hierarchy. -/// If, ond only *if*, someone really wants to substitute systemd-login -/// in an environment managed by systemd with elogin, we might have to -/// add such a support. -#if 0 - if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) - unified_cache = true; - else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) +#if 0 /// UNNEEDED by elogind + if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) + unified_cache = CGROUP_UNIFIED_ALL; + else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) { + if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 && + F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { + unified_cache = CGROUP_UNIFIED_SYSTEMD; + unified_systemd_v232 = false; + } else if (statfs("/sys/fs/cgroup/systemd/", &fs) == 0 && + F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { + unified_cache = CGROUP_UNIFIED_SYSTEMD; + unified_systemd_v232 = true; + } else { + if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0) + return -errno; + if (!F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) + return -ENOMEDIUM; + unified_cache = CGROUP_UNIFIED_NONE; + } + } else + return -ENOMEDIUM; #else - if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) + /* elogind can not support the unified hierarchy as a controller, + * so always assume a classical hierarchy. + * If, and only *if*, someone really wants to substitute systemd-login + * in an environment managed by systemd with elogind, we might have to + * add such a support. */ + unified_cache = CGROUP_UNIFIED_NONE; #endif // 0 - unified_cache = false; - else - return -ENOEXEC; - return unified_cache; + return 0; +} + +int cg_unified_controller(const char *controller) { + int r; + + r = cg_unified_update(); + if (r < 0) + return r; + + if (unified_cache == CGROUP_UNIFIED_NONE) + return false; + + if (unified_cache >= CGROUP_UNIFIED_ALL) + return true; + + return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER); +} + +int cg_all_unified(void) { + int r; + + r = cg_unified_update(); + if (r < 0) + return r; + + return unified_cache >= CGROUP_UNIFIED_ALL; } -/// UNNEEDED by elogind -#if 0 -void cg_unified_flush(void) { - unified_cache = -1; +int cg_hybrid_unified(void) { + int r; + + r = cg_unified_update(); + if (r < 0) + return r; + + return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232; } +int cg_unified_flush(void) { + unified_cache = CGROUP_UNIFIED_UNKNOWN; + + return cg_unified_update(); +} + +#if 0 /// UNNEEDED by elogind int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) { _cleanup_free_ char *fs = NULL; CGroupController c; - int r, unified; + int r; assert(p); if (supported == 0) return 0; - unified = cg_unified(); - if (unified < 0) - return unified; - if (!unified) /* on the legacy hiearchy there's no joining of controllers defined */ + r = cg_all_unified(); + if (r < 0) + return r; + if (r == 0) /* on the legacy hiearchy there's no joining of controllers defined */ return 0; r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs); @@ -2253,47 +2585,119 @@ int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) { bool cg_is_unified_wanted(void) { static thread_local int wanted = -1; - int r, unified; + int r; + bool b; + const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL; + + /* If we have a cached value, return that. */ + if (wanted >= 0) + return wanted; /* If the hierarchy is already mounted, then follow whatever * was chosen for it. */ - unified = cg_unified(); - if (unified >= 0) - return unified; + if (cg_unified_flush() >= 0) + return (wanted = unified_cache >= CGROUP_UNIFIED_ALL); + + /* Otherwise, let's see what the kernel command line has to say. + * Since checking is expensive, cache a non-error result. */ + r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b); - /* Otherwise, let's see what the kernel command line has to - * say. Since checking that is expensive, let's cache the - * result. */ + return (wanted = r > 0 ? b : is_default); +} + +bool cg_is_legacy_wanted(void) { + static thread_local int wanted = -1; + + /* If we have a cached value, return that. */ if (wanted >= 0) return wanted; - r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy", NULL); - if (r > 0) - return (wanted = true); - else { - _cleanup_free_ char *value = NULL; - - r = get_proc_cmdline_key("systemd.unified_cgroup_hierarchy=", &value); - if (r < 0) - return false; - if (r == 0) - return (wanted = false); + /* Check if we have cgroups2 already mounted. */ + if (cg_unified_flush() >= 0 && + unified_cache == CGROUP_UNIFIED_ALL) + return (wanted = false); - return (wanted = parse_boolean(value) > 0); - } + /* Otherwise, assume that at least partial legacy is wanted, + * since cgroups2 should already be mounted at this point. */ + return (wanted = true); } -bool cg_is_legacy_wanted(void) { - return !cg_is_unified_wanted(); +bool cg_is_hybrid_wanted(void) { + static thread_local int wanted = -1; + int r; + bool b; + const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD; + /* We default to true if the default is "hybrid", obviously, + * but also when the default is "unified", because if we get + * called, it means that unified hierarchy was not mounted. */ + + /* If we have a cached value, return that. */ + if (wanted >= 0) + return wanted; + + /* If the hierarchy is already mounted, then follow whatever + * was chosen for it. */ + if (cg_unified_flush() >= 0 && + unified_cache == CGROUP_UNIFIED_ALL) + return (wanted = false); + + /* Otherwise, let's see what the kernel command line has to say. + * Since checking is expensive, cache a non-error result. */ + r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b); + + /* The meaning of the kernel option is reversed wrt. to the return value + * of this function, hence the negation. */ + return (wanted = r > 0 ? !b : is_default); } #else +bool cg_is_unified_wanted(void) { + return false; +} bool cg_is_legacy_wanted(void) { return true; } +bool cg_is_hybrid_wanted(void) { + return false; +} #endif // 0 -/// UNNEEDED by elogind -#if 0 +#if 0 /// UNNEEDED by elogind +int cg_weight_parse(const char *s, uint64_t *ret) { + uint64_t u; + int r; + + if (isempty(s)) { + *ret = CGROUP_WEIGHT_INVALID; + return 0; + } + + r = safe_atou64(s, &u); + if (r < 0) + return r; + + if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX) + return -ERANGE; + + *ret = u; + return 0; +} + +const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = { + [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX, + [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX, + [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX, + [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX, +}; + +static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = { + [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax", + [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax", + [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax", + [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax", +}; + +DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType); + int cg_cpu_shares_parse(const char *s, uint64_t *ret) { uint64_t u; int r; @@ -2335,14 +2739,28 @@ int cg_blkio_weight_parse(const char *s, uint64_t *ret) { } #endif // 0 +bool is_cgroup_fs(const struct statfs *s) { + return is_fs_type(s, CGROUP_SUPER_MAGIC) || + is_fs_type(s, CGROUP2_SUPER_MAGIC); +} + +bool fd_is_cgroup_fs(int fd) { + struct statfs s; + + if (fstatfs(fd, &s) < 0) + return -errno; + + return is_cgroup_fs(&s); +} + static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = { [CGROUP_CONTROLLER_CPU] = "cpu", [CGROUP_CONTROLLER_CPUACCT] = "cpuacct", + [CGROUP_CONTROLLER_IO] = "io", [CGROUP_CONTROLLER_BLKIO] = "blkio", [CGROUP_CONTROLLER_MEMORY] = "memory", [CGROUP_CONTROLLER_DEVICES] = "devices", [CGROUP_CONTROLLER_PIDS] = "pids", - [CGROUP_CONTROLLER_NET_CLS] = "net_cls", }; DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);