X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fshared%2Fcgroup-util.c;h=5816b7d4d6fd991a3c4d9cb8ce17ba978a8ab502;hp=c17e1d4d1bde18080cd82a7ebb2b1bd6778a384f;hb=4ad490007b70e6ac18d3cb04fa2ed92eba1451fa;hpb=2d76d14ea3682c1c144001449654eb0c6bbda08d diff --git a/src/shared/cgroup-util.c b/src/shared/cgroup-util.c index c17e1d4d1..5816b7d4d 100644 --- a/src/shared/cgroup-util.c +++ b/src/shared/cgroup-util.c @@ -38,44 +38,20 @@ #include "strv.h" #include "unit-name.h" #include "fileio.h" +#include "special.h" int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) { - char *fs; - int r; + _cleanup_free_ char *fs = NULL; FILE *f; - - assert(path); - assert(_f); - - r = cg_get_path(controller, path, "cgroup.procs", &fs); - if (r < 0) - return r; - - f = fopen(fs, "re"); - free(fs); - - if (!f) - return -errno; - - *_f = f; - return 0; -} - -int cg_enumerate_tasks(const char *controller, const char *path, FILE **_f) { - char *fs; int r; - FILE *f; - assert(path); assert(_f); - r = cg_get_path(controller, path, "tasks", &fs); + r = cg_get_path(controller, path, "cgroup.procs", &fs); if (r < 0) return r; f = fopen(fs, "re"); - free(fs); - if (!f) return -errno; @@ -89,6 +65,9 @@ int cg_read_pid(FILE *f, pid_t *_pid) { /* Note that the cgroup.procs might contain duplicates! See * cgroups.txt for details. */ + assert(f); + assert(_pid); + errno = 0; if (fscanf(f, "%lu", &ul) != 1) { @@ -106,11 +85,10 @@ int cg_read_pid(FILE *f, pid_t *_pid) { } int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) { - char *fs; + _cleanup_free_ char *fs = NULL; int r; DIR *d; - assert(path); assert(_d); /* This is not recursive! */ @@ -120,8 +98,6 @@ int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) { return r; d = opendir(fs); - free(fs); - if (!d) return -errno; @@ -133,9 +109,9 @@ int cg_read_subgroup(DIR *d, char **fn) { struct dirent *de; assert(d); + assert(fn); - errno = 0; - while ((de = readdir(d))) { + FOREACH_DIRENT(de, d, return -errno) { char *b; if (de->d_type != DT_DIR) @@ -145,88 +121,68 @@ int cg_read_subgroup(DIR *d, char **fn) { streq(de->d_name, "..")) continue; - if (!(b = strdup(de->d_name))) + b = strdup(de->d_name); + if (!b) return -ENOMEM; *fn = b; return 1; } - if (errno) - return -errno; - return 0; } -int cg_rmdir(const char *controller, const char *path, bool honour_sticky) { - char *p; +int cg_rmdir(const char *controller, const char *path) { + _cleanup_free_ char *p = NULL; int r; r = cg_get_path(controller, path, NULL, &p); if (r < 0) return r; - if (honour_sticky) { - char *tasks; - - /* If the sticky bit is set don't remove the directory */ - - tasks = strappend(p, "/tasks"); - if (!tasks) { - free(p); - return -ENOMEM; - } - - r = file_is_priv_sticky(tasks); - free(tasks); - - if (r > 0) { - free(p); - return 0; - } - } - r = rmdir(p); - free(p); + if (r < 0 && errno != ENOENT) + return -errno; - return (r < 0 && errno != ENOENT) ? -errno : 0; + return 0; } int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) { + _cleanup_set_free_ Set *allocated_set = NULL; bool done = false; int r, ret = 0; pid_t my_pid; - FILE *f = NULL; - Set *allocated_set = NULL; - assert(controller); - assert(path); assert(sig >= 0); /* 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 */ - if (!s) - if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func))) + if (!s) { + s = allocated_set = set_new(trivial_hash_func, trivial_compare_func); + if (!s) return -ENOMEM; + } my_pid = getpid(); do { + _cleanup_fclose_ FILE *f = NULL; pid_t pid = 0; done = true; - if ((r = cg_enumerate_processes(controller, path, &f)) < 0) { + r = cg_enumerate_processes(controller, path, &f); + if (r < 0) { if (ret >= 0 && r != -ENOENT) - ret = r; + return r; - goto finish; + return ret; } while ((r = cg_read_pid(f, &pid)) > 0) { - if (pid == my_pid && ignore_self) + if (ignore_self && pid == my_pid) continue; if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid)) @@ -247,100 +203,77 @@ int cg_kill(const char *controller, const char *path, int sig, bool sigcont, boo done = false; - if ((r = set_put(s, LONG_TO_PTR(pid))) < 0) { + r = set_put(s, LONG_TO_PTR(pid)); + if (r < 0) { if (ret >= 0) - ret = r; + return r; - goto finish; + return ret; } } if (r < 0) { if (ret >= 0) - ret = r; + return r; - goto finish; + return ret; } - fclose(f); - f = NULL; - /* To avoid racing against processes which fork * quicker than we can kill them we repeat this until * no new pids need to be killed. */ } while (!done); -finish: - if (allocated_set) - set_free(allocated_set); - - if (f) - fclose(f); - return ret; } int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) { + _cleanup_set_free_ Set *allocated_set = NULL; + _cleanup_closedir_ DIR *d = NULL; int r, ret = 0; - DIR *d = NULL; char *fn; - Set *allocated_set = NULL; assert(path); - assert(controller); assert(sig >= 0); - if (!s) - if (!(s = allocated_set = set_new(trivial_hash_func, trivial_compare_func))) + if (!s) { + s = allocated_set = set_new(trivial_hash_func, trivial_compare_func); + if (!s) return -ENOMEM; + } ret = cg_kill(controller, path, sig, sigcont, ignore_self, s); - if ((r = cg_enumerate_subgroups(controller, path, &d)) < 0) { + r = cg_enumerate_subgroups(controller, path, &d); + if (r < 0) { if (ret >= 0 && r != -ENOENT) - ret = r; + return r; - goto finish; + return ret; } while ((r = cg_read_subgroup(d, &fn)) > 0) { - char *p = NULL; + _cleanup_free_ char *p = NULL; - r = asprintf(&p, "%s/%s", path, fn); + p = strjoin(path, "/", fn, NULL); free(fn); - - if (r < 0) { - if (ret >= 0) - ret = -ENOMEM; - - goto finish; - } + if (!p) + return -ENOMEM; r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s); - free(p); - - if (r != 0 && ret >= 0) + if (ret >= 0 && r != 0) ret = r; } - if (r < 0 && ret >= 0) + if (ret >= 0 && r < 0) ret = r; - if (rem) - if ((r = cg_rmdir(controller, path, true)) < 0) { - if (ret >= 0 && - r != -ENOENT && - r != -EBUSY) - ret = r; - } - -finish: - if (d) - closedir(d); - - if (allocated_set) - set_free(allocated_set); + if (rem) { + r = cg_rmdir(controller, path); + if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY) + return r; + } return ret; } @@ -349,7 +282,6 @@ int cg_kill_recursive_and_wait(const char *controller, const char *path, bool re unsigned i; assert(path); - assert(controller); /* This safely kills all processes; first it sends a SIGTERM, * then checks 8 times after 200ms whether the group is now @@ -367,7 +299,8 @@ int cg_kill_recursive_and_wait(const char *controller, const char *path, bool re else sig = 0; - if ((r = cg_kill_recursive(controller, path, sig, true, true, rem, NULL)) <= 0) + r = cg_kill_recursive(controller, path, sig, true, true, rem, NULL); + if (r <= 0) return r; usleep(200 * USEC_PER_MSEC); @@ -381,7 +314,6 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char _cleanup_set_free_ Set *s = NULL; int r, ret = 0; pid_t my_pid; - _cleanup_fclose_ FILE *f = NULL; assert(cfrom); assert(pfrom); @@ -395,13 +327,14 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char my_pid = getpid(); do { + _cleanup_fclose_ FILE *f = NULL; pid_t pid = 0; done = true; - r = cg_enumerate_tasks(cfrom, pfrom, &f); + r = cg_enumerate_processes(cfrom, pfrom, &f); if (r < 0) { if (ret >= 0 && r != -ENOENT) - ret = r; + return r; return ret; } @@ -411,7 +344,7 @@ 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 (pid == my_pid && ignore_self) + if (ignore_self && pid == my_pid) continue; if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid)) @@ -429,7 +362,7 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char r = set_put(s, LONG_TO_PTR(pid)); if (r < 0) { if (ret >= 0) - ret = r; + return r; return ret; } @@ -437,21 +370,25 @@ int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char if (r < 0) { if (ret >= 0) - ret = r; + return r; return ret; } - - fclose(f); - f = NULL; } while (!done); return ret; } -int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, const char *pto, bool ignore_self, bool rem) { - int r, ret = 0; +int cg_migrate_recursive( + const char *cfrom, + const char *pfrom, + const char *cto, + const char *pto, + bool ignore_self, + bool rem) { + _cleanup_closedir_ DIR *d = NULL; + int r, ret = 0; char *fn; assert(cfrom); @@ -464,7 +401,8 @@ int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, r = cg_enumerate_subgroups(cfrom, pfrom, &d); if (r < 0) { if (ret >= 0 && r != -ENOENT) - ret = r; + return r; + return ret; } @@ -475,7 +413,7 @@ int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, free(fn); if (!p) { if (ret >= 0) - ret = -ENOMEM; + return -ENOMEM; return ret; } @@ -489,7 +427,7 @@ int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, ret = r; if (rem) { - r = cg_rmdir(cfrom, pfrom, true); + r = cg_rmdir(cfrom, pfrom); if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY) return r; } @@ -499,6 +437,8 @@ int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, static const char *normalize_controller(const char *controller) { + assert(controller); + if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) return "systemd"; else if (startswith(controller, "name=")) @@ -510,23 +450,22 @@ static const char *normalize_controller(const char *controller) { static int join_path(const char *controller, const char *path, const char *suffix, char **fs) { char *t = NULL; - if (!(controller || path)) - return -EINVAL; - - if (controller) { - if (path && suffix) + if (!isempty(controller)) { + if (!isempty(path) && !isempty(suffix)) t = strjoin("/sys/fs/cgroup/", controller, "/", path, "/", suffix, NULL); - else if (path) + else if (!isempty(path)) t = strjoin("/sys/fs/cgroup/", controller, "/", path, NULL); - else if (suffix) + else if (!isempty(suffix)) t = strjoin("/sys/fs/cgroup/", controller, "/", suffix, NULL); else - t = strjoin("/sys/fs/cgroup/", controller, NULL); + t = strappend("/sys/fs/cgroup/", controller); } else { - if (path && suffix) + if (!isempty(path) && !isempty(suffix)) t = strjoin(path, "/", suffix, NULL); - else if (path) + else if (!isempty(path)) t = strdup(path); + else + return -EINVAL; } if (!t) @@ -544,6 +483,9 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch assert(fs); + if (controller && !cg_controller_is_valid(controller, true)) + return -EINVAL; + if (_unlikely_(!good)) { int r; @@ -556,10 +498,11 @@ int cg_get_path(const char *controller, const char *path, const char *suffix, ch } p = controller ? normalize_controller(controller) : NULL; + return join_path(p, path, suffix, fs); } -static int check(const char *p) { +static int check_hierarchy(const char *p) { char *cc; assert(p); @@ -577,17 +520,16 @@ int cg_get_path_and_check(const char *controller, const char *path, const char * const char *p; int r; - assert(controller); assert(fs); - if (isempty(controller)) + if (!cg_controller_is_valid(controller, true)) return -EINVAL; /* Normalize the controller syntax */ p = normalize_controller(controller); /* Check if this controller actually really exists */ - r = check(p); + r = check_hierarchy(p); if (r < 0) return r; @@ -595,8 +537,9 @@ int cg_get_path_and_check(const char *controller, const char *path, const char * } static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) { - char *p; - bool is_sticky; + assert(path); + assert(sb); + assert(ftwbuf); if (typeflag != FTW_DP) return 0; @@ -604,27 +547,14 @@ static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct if (ftwbuf->level < 1) return 0; - p = strappend(path, "/tasks"); - if (!p) { - errno = ENOMEM; - return 1; - } - - is_sticky = file_is_priv_sticky(p) > 0; - free(p); - - if (is_sticky) - return 0; - rmdir(path); return 0; } int cg_trim(const char *controller, const char *path, bool delete_root) { - char *fs; + _cleanup_free_ char *fs = NULL; int r = 0; - assert(controller); assert(path); r = cg_get_path(controller, path, NULL, &fs); @@ -632,47 +562,28 @@ int cg_trim(const char *controller, const char *path, bool delete_root) { return r; errno = 0; - if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) < 0) + if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) r = errno ? -errno : -EIO; if (delete_root) { - bool is_sticky; - char *p; - - p = strappend(fs, "/tasks"); - if (!p) { - free(fs); - return -ENOMEM; - } - - is_sticky = file_is_priv_sticky(p) > 0; - free(p); - - if (!is_sticky) - if (rmdir(fs) < 0 && errno != ENOENT) { - if (r == 0) - r = -errno; - } + if (rmdir(fs) < 0 && errno != ENOENT) + return -errno; } - free(fs); - return r; } int cg_delete(const char *controller, const char *path) { - char *parent; + _cleanup_free_ char *parent = NULL; int r; - assert(controller); assert(path); - if ((r = path_get_parent(path, &parent)) < 0) + r = path_get_parent(path, &parent); + if (r < 0) return r; r = cg_migrate_recursive(controller, path, controller, parent, false, true); - free(parent); - return r == -ENOENT ? 0 : r; } @@ -681,11 +592,10 @@ int cg_attach(const char *controller, const char *path, pid_t pid) { char c[DECIMAL_STR_MAX(pid_t) + 2]; int r; - assert(controller); assert(path); assert(pid >= 0); - r = cg_get_path_and_check(controller, path, "tasks", &fs); + r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs); if (r < 0) return r; @@ -707,7 +617,6 @@ int cg_set_group_access( _cleanup_free_ char *fs = NULL; int r; - assert(controller); assert(path); if (mode != (mode_t) -1) @@ -725,197 +634,171 @@ int cg_set_task_access( const char *path, mode_t mode, uid_t uid, - gid_t gid, - int sticky) { + gid_t gid) { _cleanup_free_ char *fs = NULL, *procs = NULL; int r; - assert(controller); assert(path); - if (mode == (mode_t) -1 && uid == (uid_t) -1 && gid == (gid_t) -1 && sticky < 0) + if (mode == (mode_t) -1 && uid == (uid_t) -1 && gid == (gid_t) -1) return 0; if (mode != (mode_t) -1) mode &= 0666; - r = cg_get_path(controller, path, "tasks", &fs); + r = cg_get_path(controller, path, "cgroup.procs", &fs); if (r < 0) return r; - if (sticky >= 0 && mode != (mode_t) -1) - /* Both mode and sticky param are passed */ - mode |= (sticky ? S_ISVTX : 0); - else if ((sticky >= 0 && mode == (mode_t) -1) || - (mode != (mode_t) -1 && sticky < 0)) { - struct stat st; - - /* Only one param is passed, hence read the current - * mode from the file itself */ - - r = lstat(fs, &st); - if (r < 0) - return -errno; - - if (mode == (mode_t) -1) - /* No mode set, we just shall set the sticky bit */ - mode = (st.st_mode & ~S_ISVTX) | (sticky ? S_ISVTX : 0); - else - /* Only mode set, leave sticky bit untouched */ - mode = (st.st_mode & ~0777) | mode; - } - r = chmod_and_chown(fs, mode, uid, gid); if (r < 0) return r; - /* Always keep values for "cgroup.procs" in sync with "tasks" */ - r = cg_get_path(controller, path, "cgroup.procs", &procs); + /* Compatibility, Always keep values for "tasks" in sync with + * "cgroup.procs" */ + r = cg_get_path(controller, path, "tasks", &procs); if (r < 0) return r; return chmod_and_chown(procs, mode, uid, gid); } -int cg_get_by_pid(const char *controller, pid_t pid, char **path) { - int r; - char *p = NULL; - FILE *f; - char *fs; +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; size_t cs; - assert(controller); assert(path); assert(pid >= 0); - if (pid == 0) - pid = getpid(); + if (controller) { + if (!cg_controller_is_valid(controller, true)) + return -EINVAL; - if (asprintf(&fs, "/proc/%lu/cgroup", (unsigned long) pid) < 0) - return -ENOMEM; + controller = normalize_controller(controller); + } else + controller = SYSTEMD_CGROUP_CONTROLLER; - f = fopen(fs, "re"); - free(fs); + if (pid == 0) + fs = "/proc/self/cgroup"; + else + fs = procfs_file_alloca(pid, "cgroup"); + f = fopen(fs, "re"); if (!f) return errno == ENOENT ? -ESRCH : -errno; cs = strlen(controller); - while (!feof(f)) { - char line[LINE_MAX]; - char *l; - - errno = 0; - if (!(fgets(line, sizeof(line), f))) { - if (feof(f)) - break; - - r = errno ? -errno : -EIO; - goto finish; - } + FOREACH_LINE(line, f, return -errno) { + char *l, *p, *w, *e; + size_t k; + char *state; + bool found = false; truncate_nl(line); - if (!(l = strchr(line, ':'))) + l = strchr(line, ':'); + if (!l) continue; l++; - if (!strneq(l, controller, cs)) + e = strchr(l, ':'); + if (!e) continue; - if (l[cs] != ':') - continue; + *e = 0; + + FOREACH_WORD_SEPARATOR(w, k, l, ",", state) { - if (!(p = strdup(l + cs + 1))) { - r = -ENOMEM; - goto finish; + if (k == cs && memcmp(w, controller, cs) == 0) { + found = true; + break; + } + + if (k == 5 + cs && + memcmp(w, "name=", 5) == 0 && + memcmp(w+5, controller, cs) == 0) { + found = true; + break; + } } - *path = p; - r = 0; - goto finish; - } + if (!found) + continue; - r = -ENOENT; + p = strdup(e + 1); + if (!p) + return -ENOMEM; -finish: - fclose(f); + *path = p; + return 0; + } - return r; + return -ENOENT; } int cg_install_release_agent(const char *controller, const char *agent) { - char *fs = NULL, *contents = NULL, *line = NULL, *sc; + _cleanup_free_ char *fs = NULL, *contents = NULL; + char *sc; int r; - assert(controller); assert(agent); - if ((r = cg_get_path(controller, NULL, "release_agent", &fs)) < 0) + r = cg_get_path(controller, NULL, "release_agent", &fs); + if (r < 0) return r; - if ((r = read_one_line_file(fs, &contents)) < 0) - goto finish; + r = read_one_line_file(fs, &contents); + if (r < 0) + return r; sc = strstrip(contents); if (sc[0] == 0) { - - if (asprintf(&line, "%s\n", agent) < 0) { - r = -ENOMEM; - goto finish; - } - - r = write_string_file(fs, line); + r = write_string_file(fs, agent); if (r < 0) - goto finish; - - } else if (!streq(sc, agent)) { - r = -EEXIST; - goto finish; - } + return r; + } else if (!streq(sc, agent)) + return -EEXIST; free(fs); fs = NULL; - if ((r = cg_get_path(controller, NULL, "notify_on_release", &fs)) < 0) - goto finish; + r = cg_get_path(controller, NULL, "notify_on_release", &fs); + if (r < 0) + return r; free(contents); contents = NULL; - if ((r = read_one_line_file(fs, &contents)) < 0) - goto finish; + r = read_one_line_file(fs, &contents); + if (r < 0) + return r; sc = strstrip(contents); - if (streq(sc, "0")) { - if ((r = write_string_file(fs, "1\n")) < 0) - goto finish; + r = write_string_file(fs, "1"); + if (r < 0) + return r; - r = 1; - } else if (!streq(sc, "1")) { - r = -EIO; - goto finish; - } else - r = 0; + return 1; + } -finish: - free(fs); - free(contents); - free(line); + if (!streq(sc, "1")) + return -EIO; - return r; + return 0; } int cg_is_empty(const char *controller, const char *path, bool ignore_self) { + _cleanup_fclose_ FILE *f = NULL; pid_t pid = 0, self_pid; - int r; - FILE *f = NULL; bool found = false; + int r; assert(path); - r = cg_enumerate_tasks(controller, path, &f); + r = cg_enumerate_processes(controller, path, &f); if (r < 0) return r == -ENOENT ? 1 : r; @@ -930,8 +813,6 @@ int cg_is_empty(const char *controller, const char *path, bool ignore_self) { break; } - fclose(f); - if (r < 0) return r; @@ -939,8 +820,8 @@ int cg_is_empty(const char *controller, const char *path, bool ignore_self) { } int cg_is_empty_by_spec(const char *spec, bool ignore_self) { - int r; _cleanup_free_ char *controller = NULL, *path = NULL; + int r; assert(spec); @@ -952,9 +833,9 @@ int cg_is_empty_by_spec(const char *spec, bool ignore_self) { } int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) { - int r; - DIR *d = NULL; + _cleanup_closedir_ DIR *d = NULL; char *fn; + int r; assert(path); @@ -967,37 +848,28 @@ int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_ return r == -ENOENT ? 1 : r; while ((r = cg_read_subgroup(d, &fn)) > 0) { - char *p = NULL; + _cleanup_free_ char *p = NULL; - r = asprintf(&p, "%s/%s", path, fn); + p = strjoin(path, "/", fn, NULL); free(fn); - - if (r < 0) { - r = -ENOMEM; - goto finish; - } + if (!p) + return -ENOMEM; r = cg_is_empty_recursive(controller, p, ignore_self); - free(p); - if (r <= 0) - goto finish; + return r; } - if (r >= 0) - r = 1; - -finish: - - if (d) - closedir(d); + if (r < 0) + return r; - return r; + return 1; } int cg_split_spec(const char *spec, char **controller, char **path) { const char *e; char *t = NULL, *u = NULL; + _cleanup_free_ char *v = NULL; assert(spec); @@ -1010,6 +882,7 @@ int cg_split_spec(const char *spec, char **controller, char **path) { if (!t) return -ENOMEM; + path_kill_slashes(t); *path = t; } @@ -1021,11 +894,11 @@ int cg_split_spec(const char *spec, char **controller, char **path) { e = strchr(spec, ':'); if (!e) { - if (!filename_is_safe(spec)) + if (!cg_controller_is_valid(spec, true)) return -EINVAL; if (controller) { - t = strdup(spec); + t = strdup(normalize_controller(spec)); if (!t) return -ENOMEM; @@ -1038,10 +911,13 @@ int cg_split_spec(const char *spec, char **controller, char **path) { return 0; } - t = strndup(spec, e-spec); + v = strndup(spec, e-spec); + if (!v) + return -ENOMEM; + t = strdup(normalize_controller(v)); if (!t) return -ENOMEM; - if (!filename_is_safe(t)) { + if (!cg_controller_is_valid(t, true)) { free(t); return -EINVAL; } @@ -1051,12 +927,15 @@ int cg_split_spec(const char *spec, char **controller, char **path) { free(t); return -ENOMEM; } - if (!path_is_safe(u)) { + if (!path_is_safe(u) || + !path_is_absolute(u)) { free(t); free(u); return -EINVAL; } + path_kill_slashes(u); + if (controller) *controller = t; else @@ -1071,36 +950,48 @@ int cg_split_spec(const char *spec, char **controller, char **path) { } int cg_join_spec(const char *controller, const char *path, char **spec) { - assert(controller); + char *s; + assert(path); - if (!path_is_absolute(path) || - controller[0] == 0 || - strchr(controller, ':') || - strchr(controller, '/')) + if (!controller) + controller = "systemd"; + else { + if (!cg_controller_is_valid(controller, true)) + return -EINVAL; + + controller = normalize_controller(controller); + } + + if (!path_is_absolute(path)) return -EINVAL; - if (asprintf(spec, "%s:%s", controller, path) < 0) + s = strjoin(controller, ":", path, NULL); + if (!s) return -ENOMEM; + path_kill_slashes(s + strlen(controller) + 1); + + *spec = s; return 0; } -int cg_fix_path(const char *path, char **result) { - char *t, *c, *p; +int cg_mangle_path(const char *path, char **result) { + _cleanup_free_ char *c = NULL, *p = NULL; + char *t; int r; assert(path); assert(result); /* First check if it already is a filesystem path */ - if (path_startswith(path, "/sys/fs/cgroup") && - access(path, F_OK) >= 0) { + if (path_startswith(path, "/sys/fs/cgroup")) { t = strdup(path); if (!t) return -ENOMEM; + path_kill_slashes(t); *result = t; return 0; } @@ -1110,36 +1001,22 @@ int cg_fix_path(const char *path, char **result) { if (r < 0) return r; - r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result); - free(c); - free(p); - - return r; + return cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result); } -int cg_get_user_path(char **path) { - char *root, *p; +int cg_get_root_path(char **path) { + char *p, *e; + int r; assert(path); - /* Figure out the place to put user cgroups below. We use the - * same as PID 1 has but with the "/system" suffix replaced by - * "/user" */ - - if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &root) < 0) - p = strdup("/user"); - else { - if (endswith(root, "/system")) - root[strlen(root) - 7] = 0; - else if (streq(root, "/")) - root[0] = 0; - - p = strappend(root, "/user"); - free(root); - } + r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p); + if (r < 0) + return r; - if (!p) - return -ENOMEM; + e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); + if (e) + *e = 0; *path = p; return 0; @@ -1148,25 +1025,29 @@ int cg_get_user_path(char **path) { char **cg_shorten_controllers(char **controllers) { char **f, **t; - controllers = strv_uniq(controllers); - if (!controllers) return controllers; for (f = controllers, t = controllers; *f; f++) { - int r; const char *p; + int r; + + p = normalize_controller(*f); - if (streq(*f, "systemd") || streq(*f, SYSTEMD_CGROUP_CONTROLLER)) { + if (streq(p, "systemd")) { free(*f); continue; } - p = normalize_controller(*f); + if (!cg_controller_is_valid(p, true)) { + log_warning("Controller %s is not valid, removing from controllers list.", p); + free(*f); + continue; + } - r = check(p); + r = check_hierarchy(p); if (r < 0) { - log_debug("Controller %s is not available, removing from controllers list.", *f); + log_debug("Controller %s is not available, removing from controllers list.", p); free(*f); continue; } @@ -1175,40 +1056,28 @@ char **cg_shorten_controllers(char **controllers) { } *t = NULL; - return controllers; + return strv_uniq(controllers); } -int cg_pid_get_cgroup(pid_t pid, char **root, char **cgroup) { - char *cg_process, *cg_init, *p; +int cg_pid_get_path_shifted(pid_t pid, char **root, char **cgroup) { + _cleanup_free_ char *cg_root = NULL; + char *cg_process, *p; int r; - assert(pid >= 0); - - if (pid == 0) - pid = getpid(); - - r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &cg_process); + r = cg_get_root_path(&cg_root); if (r < 0) return r; - r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &cg_init); - if (r < 0) { - free(cg_process); + r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cg_process); + if (r < 0) return r; - } - if (endswith(cg_init, "/system")) - cg_init[strlen(cg_init)-7] = 0; - else if (streq(cg_init, "/")) - cg_init[0] = 0; - - if (startswith(cg_process, cg_init)) - p = cg_process + strlen(cg_init); + p = path_startswith(cg_process, cg_root); + if (p) + p--; else p = cg_process; - free(cg_init); - if (cgroup) { char* c; @@ -1230,84 +1099,339 @@ int cg_pid_get_cgroup(pid_t pid, char **root, char **cgroup) { return 0; } -static int instance_unit_from_cgroup(char *cgroup){ - char *at; +int cg_path_decode_unit(const char *cgroup, char **unit){ + char *p, *e, *c, *s, *k; assert(cgroup); + assert(unit); - at = strstr(cgroup, "@."); - if (at) { - /* This is a templated service */ + e = strchrnul(cgroup, '/'); + c = strndupa(cgroup, e - cgroup); + c = cg_unescape(c); - char *i; - char _cleanup_free_ *i2 = NULL, *s = NULL; + /* Could this be a valid unit name? */ + if (!unit_name_is_valid(c, true)) + return -EINVAL; - i = strchr(at, '/'); - if (!i || !i[1]) /* disallow empty instances */ + if (!unit_name_is_template(c)) + s = strdup(c); + else { + if (*e != '/') return -EINVAL; - s = strndup(at + 1, i - at - 1); - i2 = strdup(i + 1); - if (!s || !i2) - return -ENOMEM; + e += strspn(e, "/"); + + p = strchrnul(e, '/'); + k = strndupa(e, p - e); + k = cg_unescape(k); + + if (!unit_name_is_valid(k, false)) + return -EINVAL; - strcpy(at + 1, i2); - strcat(at + 1, s); + s = strdup(k); } + if (!s) + return -ENOMEM; + + *unit = s; return 0; } -/* non-static only for testing purposes */ -int cgroup_to_unit(char *cgroup, char **unit){ +static const char *skip_slices(const char *p) { + /* Skips over all slice assignments */ + + for (;;) { + size_t n; + + p += strspn(p, "/"); + + n = strcspn(p, "/"); + if (n <= 6 || memcmp(p + n - 6, ".slice", 6) != 0) + return p; + + p += n; + } +} + +int cg_path_get_unit(const char *path, char **unit) { + const char *e; + + assert(path); + assert(unit); + + e = skip_slices(path); + + return cg_path_decode_unit(e, unit); +} + +int cg_pid_get_unit(pid_t pid, char **unit) { + _cleanup_free_ char *cgroup = NULL; int r; - char *p; - assert(cgroup); assert(unit); - r = instance_unit_from_cgroup(cgroup); + r = cg_pid_get_path_shifted(pid, NULL, &cgroup); if (r < 0) return r; - p = strrchr(cgroup, '/'); + return cg_path_get_unit(cgroup, unit); +} + +static const char *skip_user(const char *p) { + size_t n; + + assert(p); + + p += strspn(p, "/"); + + n = strcspn(p, "/"); + if (n <= 5 || memcmp(p + n - 5, ".user", 5) != 0) + return p; + + p += n; + p += strspn(p, "/"); + + return p; +} + +static const char *skip_session(const char *p) { + size_t n; + assert(p); - r = unit_name_is_valid(p + 1, true); + p += strspn(p, "/"); + + n = strcspn(p, "/"); + if (n <= 8 || memcmp(p + n - 8, ".session", 8) != 0) + return NULL; + + p += n; + p += strspn(p, "/"); + + return p; +} + +static const char *skip_systemd_label(const char *p) { + size_t n; + + assert(p); + + p += strspn(p, "/"); + + n = strcspn(p, "/"); + if (n < 8 || memcmp(p, "systemd-", 8) != 0) + return p; + + p += n; + p += strspn(p, "/"); + + return p; +} + +int cg_path_get_user_unit(const char *path, char **unit) { + const char *e; + + assert(path); + assert(unit); + + /* We always have to parse the path from the beginning as unit + * cgroups might have arbitrary child cgroups and we shouldn't get + * confused by those */ + + /* Skip slices, if there are any */ + e = skip_slices(path); + + /* Skip the user name, if there is one */ + e = skip_user(e); + + /* Skip the session ID, require that there is one */ + e = skip_session(e); + if (!e) + return -ENOENT; + + /* Skip the systemd cgroup, if there is one */ + e = skip_systemd_label(e); + + return cg_path_decode_unit(e, unit); +} + +int cg_pid_get_user_unit(pid_t pid, char **unit) { + _cleanup_free_ char *cgroup = NULL; + int r; + + assert(unit); + + r = cg_pid_get_path_shifted(pid, NULL, &cgroup); + if (r < 0) + return r; + + return cg_path_get_user_unit(cgroup, unit); +} + +int cg_path_get_machine_name(const char *path, char **machine) { + const char *e, *n, *x; + char *s, *r; + + assert(path); + assert(machine); + + /* Skip slices, if there are any */ + e = skip_slices(path); + + n = strchrnul(e, '/'); + if (e == n) + return -ENOENT; + + s = strndupa(e, n - e); + s = cg_unescape(s); + + x = endswith(s, ".machine"); + if (!x) + return -ENOENT; + + r = strndup(s, x - s); if (!r) - return -EINVAL; + return -ENOMEM; + + *machine = r; + return 0; +} + +int cg_pid_get_machine_name(pid_t pid, char **machine) { + _cleanup_free_ char *cgroup = NULL; + int r; + + assert(machine); + + r = cg_pid_get_path_shifted(pid, NULL, &cgroup); + if (r < 0) + return r; - *unit = strdup(p + 1); - if (!*unit) + return cg_path_get_machine_name(cgroup, machine); +} + +int cg_path_get_session(const char *path, char **session) { + const char *e, *n; + char *s; + + assert(path); + assert(session); + + /* Skip slices, if there are any */ + e = skip_slices(path); + + /* Skip the user name, if there is one */ + e = skip_user(e); + + n = strchrnul(e, '/'); + if (n - e < 8) + return -ENOENT; + if (memcmp(n - 8, ".session", 8) != 0) + return -ENOENT; + + s = strndup(e, n - e - 8); + if (!s) return -ENOMEM; + *session = s; return 0; } -static int cg_pid_get(const char *prefix, pid_t pid, char **unit) { +int cg_pid_get_session(pid_t pid, char **session) { + _cleanup_free_ char *cgroup = NULL; int r; - char _cleanup_free_ *cgroup = NULL; - assert(pid >= 0); - assert(unit); + assert(session); - r = cg_pid_get_cgroup(pid, NULL, &cgroup); + r = cg_pid_get_path_shifted(pid, NULL, &cgroup); if (r < 0) return r; - if (!startswith(cgroup, prefix)) + return cg_path_get_session(cgroup, session); +} + +int cg_path_get_owner_uid(const char *path, uid_t *uid) { + const char *e, *n; + char *s; + + assert(path); + assert(uid); + + /* Skip slices, if there are any */ + e = skip_slices(path); + + n = strchrnul(e, '/'); + if (n - e < 5) + return -ENOENT; + if (memcmp(n - 5, ".user", 5) != 0) return -ENOENT; - r = cgroup_to_unit(cgroup, unit); - return r; + s = strndupa(e, n - e - 5); + if (!s) + return -ENOMEM; + + return parse_uid(s, uid); } -int cg_pid_get_unit(pid_t pid, char **unit) { - return cg_pid_get("/system/", pid, unit); +int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) { + _cleanup_free_ char *cgroup = NULL; + int r; + + assert(uid); + + r = cg_pid_get_path_shifted(pid, NULL, &cgroup); + if (r < 0) + return r; + + return cg_path_get_owner_uid(cgroup, uid); } -int cg_pid_get_user_unit(pid_t pid, char **unit) { - return cg_pid_get("/user/", pid, unit); +int cg_path_get_slice(const char *p, char **slice) { + const char *e = NULL; + size_t m = 0; + + assert(p); + assert(slice); + + for (;;) { + size_t n; + + p += strspn(p, "/"); + + n = strcspn(p, "/"); + if (n <= 6 || memcmp(p + n - 6, ".slice", 6) != 0) { + char *s; + + if (!e) + return -ENOENT; + + s = strndup(e, m); + if (!s) + return -ENOMEM; + + *slice = s; + return 0; + } + + e = p; + m = n; + + p += n; + } +} + +int cg_pid_get_slice(pid_t pid, char **slice) { + _cleanup_free_ char *cgroup = NULL; + int r; + + assert(slice); + + r = cg_pid_get_path_shifted(pid, NULL, &cgroup); + if (r < 0) + return r; + + return cg_path_get_slice(cgroup, slice); } int cg_controller_from_attr(const char *attr, char **controller) { @@ -1330,7 +1454,7 @@ int cg_controller_from_attr(const char *attr, char **controller) { if (!c) return -ENOMEM; - if (!filename_is_safe(c)) { + if (!cg_controller_is_valid(c, false)) { free(c); return -EINVAL; } @@ -1338,3 +1462,286 @@ int cg_controller_from_attr(const char *attr, char **controller) { *controller = c; return 1; } + +char *cg_escape(const char *p) { + bool need_prefix = false; + + /* This implements very minimal escaping for names to be used + * as file names in the cgroup tree: any name which might + * conflict with a kernel name or is prefixed with '_' is + * prefixed with a '_'. That way, when reading cgroup names it + * is sufficient to remove a single prefixing underscore if + * there is one. */ + + /* The return value of this function (unlike cg_unescape()) + * needs free()! */ + + if (p[0] == 0 || + p[0] == '_' || + p[0] == '.' || + streq(p, "notify_on_release") || + streq(p, "release_agent") || + streq(p, "tasks")) + need_prefix = true; + else { + const char *dot; + + dot = strrchr(p, '.'); + if (dot) { + + if (dot - p == 6 && memcmp(p, "cgroup", 6) == 0) + need_prefix = true; + else { + char *n; + + n = strndupa(p, dot - p); + + if (check_hierarchy(n) >= 0) + need_prefix = true; + } + } + } + + if (need_prefix) + return strappend("_", p); + else + return strdup(p); +} + +char *cg_unescape(const char *p) { + assert(p); + + /* The return value of this function (unlike cg_escape()) + * doesn't need free()! */ + + if (p[0] == '_') + return (char*) p+1; + + return (char*) p; +} + +#define CONTROLLER_VALID \ + "0123456789" \ + "abcdefghijklmnopqrstuvwxyz" \ + "ABCDEFGHIJKLMNOPQRSTUVWXYZ" \ + "_" + +bool cg_controller_is_valid(const char *p, bool allow_named) { + const char *t, *s; + + if (!p) + return false; + + if (allow_named) { + s = startswith(p, "name="); + if (s) + p = s; + } + + if (*p == 0 || *p == '_') + return false; + + for (t = p; *t; t++) + if (!strchr(CONTROLLER_VALID, *t)) + return false; + + if (t - p > FILENAME_MAX) + return false; + + return true; +} + +int cg_slice_to_path(const char *unit, char **ret) { + _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL; + const char *dash; + + assert(unit); + assert(ret); + + if (!unit_name_is_valid(unit, false)) + return -EINVAL; + + if (!endswith(unit, ".slice")) + return -EINVAL; + + p = unit_name_to_prefix(unit); + if (!p) + return -ENOMEM; + + dash = strchr(p, '-'); + while (dash) { + _cleanup_free_ char *escaped = NULL; + char n[dash - p + sizeof(".slice")]; + + strcpy(stpncpy(n, p, dash - p), ".slice"); + + if (!unit_name_is_valid(n, false)) + return -EINVAL; + + escaped = cg_escape(n); + if (!escaped) + return -ENOMEM; + + if (!strextend(&s, escaped, "/", NULL)) + return -ENOMEM; + + dash = strchr(dash+1, '-'); + } + + e = cg_escape(unit); + if (!e) + return -ENOMEM; + + if (!strextend(&s, e, NULL)) + return -ENOMEM; + + *ret = s; + s = NULL; + + return 0; +} + +int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) { + _cleanup_free_ char *p = NULL; + int r; + + r = cg_get_path(controller, path, attribute, &p); + if (r < 0) + return r; + + return write_string_file(p, value); +} + +static const char mask_names[] = + "cpu\0" + "cpuacct\0" + "blkio\0" + "memory\0" + "devices\0"; + +int cg_create_with_mask(CGroupControllerMask mask, const char *path) { + CGroupControllerMask bit = 1; + const char *n; + 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 + * in all others */ + + /* First create the cgroup in our own hierarchy. */ + r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path); + if (r < 0) + return r; + + /* Then, do the same in the other hierarchies */ + NULSTR_FOREACH(n, mask_names) { + if (bit & mask) + cg_create(n, path); + else + cg_trim(n, path, true); + + bit <<= 1; + } + + return r; +} + +int cg_attach_with_mask(CGroupControllerMask mask, const char *path, pid_t pid) { + CGroupControllerMask bit = 1; + const char *n; + int r; + + r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid); + + NULSTR_FOREACH(n, mask_names) { + if (bit & mask) + cg_attach(n, path, pid); + else { + char prefix[strlen(path) + 1], *slash; + + /* OK, this one is a bit harder... Now we need + * to add to the closest parent cgroup we + * can find */ + strcpy(prefix, path); + while ((slash = strrchr(prefix, '/'))) { + int q; + *slash = 0; + + q = cg_attach(n, prefix, pid); + if (q >= 0) + break; + } + } + + bit <<= 1; + } + + return r; +} + +int cg_migrate_with_mask(CGroupControllerMask mask, const char *from, const char *to) { + CGroupControllerMask bit = 1; + const char *n; + int r; + + if (path_equal(from, to)) + return 0; + + r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, false, true); + + NULSTR_FOREACH(n, mask_names) { + if (bit & mask) + cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, to, n, to, false, false); + else { + char prefix[strlen(to) + 1], *slash; + + strcpy(prefix, to); + while ((slash = strrchr(prefix, '/'))) { + int q; + + *slash = 0; + + q = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, to, n, prefix, false, false); + if (q >= 0) + break; + } + } + + bit <<= 1; + } + + return r; +} + +int cg_trim_with_mask(CGroupControllerMask mask, const char *path, bool delete_root) { + CGroupControllerMask bit = 1; + const char *n; + int r; + + r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root); + if (r < 0) + return r; + + NULSTR_FOREACH(n, mask_names) { + if (bit & mask) + cg_trim(n, path, delete_root); + + bit <<= 1; + } + + return r; +} + +CGroupControllerMask cg_mask_supported(void) { + CGroupControllerMask bit = 1, mask = 0; + const char *n; + + NULSTR_FOREACH(n, mask_names) { + if (check_hierarchy(n) >= 0) + mask |= bit; + + bit <<= 1; + } + + return mask; +}