dryang / rpms / systemd

Forked from rpms/systemd a year ago
Clone
Blob Blame History Raw
From 61e5aed87f1b82a51c6ea8ccde96805cb63e5b15 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Zbigniew=20J=C4=99drzejewski-Szmek?= <zbyszek@in.waw.pl>
Date: Tue, 21 May 2019 08:45:19 +0200
Subject: [PATCH] Rework cpu affinity parsing

The CPU_SET_S api is pretty bad. In particular, it has a parameter for the size
of the array, but operations which take two (CPU_EQUAL_S) or even three arrays
(CPU_{AND,OR,XOR}_S) still take just one size. This means that all arrays must
be of the same size, or buffer overruns will occur. This is exactly what our
code would do, if it received an array of unexpected size over the network.
("Unexpected" here means anything different from what cpu_set_malloc() detects
as the "right" size.)

Let's rework this, and store the size in bytes of the allocated storage area.

The code will now parse any number up to 8191, independently of what the current
kernel supports. This matches the kernel maximum setting for any architecture,
to make things more portable.

Fixes #12605.

(cherry picked from commit 0985c7c4e22c8dbbea4398cf3453da45ebf63800)

Related: #1734787
---
 src/basic/cpu-set-util.c     | 133 +++++++++++++++++++++-----
 src/basic/cpu-set-util.h     |  47 ++++++---
 src/core/dbus-execute.c      |  35 ++-----
 src/core/execute.c           |  12 +--
 src/core/execute.h           |   4 +-
 src/core/load-fragment.c     |  31 +-----
 src/core/main.c              |  14 +--
 src/nspawn/nspawn-settings.c |  33 +------
 src/nspawn/nspawn-settings.h |   4 +-
 src/nspawn/nspawn.c          |  29 +++---
 src/shared/bus-unit-util.c   |   4 +-
 src/test/test-cpu-set-util.c | 179 +++++++++++++++++++----------------
 src/test/test-sizeof.c       |   3 +
 13 files changed, 286 insertions(+), 242 deletions(-)

diff --git a/src/basic/cpu-set-util.c b/src/basic/cpu-set-util.c
index 8f24a2601a..fe440f6381 100644
--- a/src/basic/cpu-set-util.c
+++ b/src/basic/cpu-set-util.c
@@ -15,14 +15,15 @@
 #include "macro.h"
 #include "parse-util.h"
 #include "string-util.h"
+#include "util.h"
 
-char* cpu_set_to_string(const cpu_set_t *set, size_t setsize) {
+char* cpu_set_to_string(const CPUSet *a) {
         _cleanup_free_ char *str = NULL;
         size_t allocated = 0, len = 0;
         int i, r;
 
-        for (i = 0; (size_t) i < setsize * 8; i++) {
-                if (!CPU_ISSET_S(i, setsize, set))
+        for (i = 0; (size_t) i < a->allocated * 8; i++) {
+                if (!CPU_ISSET_S(i, a->allocated, a->set))
                         continue;
 
                 if (!GREEDY_REALLOC(str, allocated, len + 1 + DECIMAL_STR_MAX(int)))
@@ -65,24 +66,74 @@ cpu_set_t* cpu_set_malloc(unsigned *ncpus) {
         }
 }
 
-int parse_cpu_set_internal(
+static int cpu_set_realloc(CPUSet *cpu_set, unsigned ncpus) {
+        size_t need;
+
+        assert(cpu_set);
+
+        need = CPU_ALLOC_SIZE(ncpus);
+        if (need > cpu_set->allocated) {
+                cpu_set_t *t;
+
+                t = realloc(cpu_set->set, need);
+                if (!t)
+                        return -ENOMEM;
+
+                memzero((uint8_t*) t + cpu_set->allocated, need - cpu_set->allocated);
+
+                cpu_set->set = t;
+                cpu_set->allocated = need;
+        }
+
+        return 0;
+}
+
+static int cpu_set_add(CPUSet *cpu_set, unsigned cpu) {
+        int r;
+
+        if (cpu >= 8192)
+                /* As of kernel 5.1, CONFIG_NR_CPUS can be set to 8192 on PowerPC */
+                return -ERANGE;
+
+        r = cpu_set_realloc(cpu_set, cpu + 1);
+        if (r < 0)
+                return r;
+
+        CPU_SET_S(cpu, cpu_set->allocated, cpu_set->set);
+        return 0;
+}
+
+int cpu_set_add_all(CPUSet *a, const CPUSet *b) {
+        int r;
+
+        /* Do this backwards, so if we fail, we fail before changing anything. */
+        for (unsigned cpu_p1 = b->allocated * 8; cpu_p1 > 0; cpu_p1--)
+                if (CPU_ISSET_S(cpu_p1 - 1, b->allocated, b->set)) {
+                        r = cpu_set_add(a, cpu_p1 - 1);
+                        if (r < 0)
+                                return r;
+                }
+
+        return 0;
+}
+
+int parse_cpu_set_full(
                 const char *rvalue,
-                cpu_set_t **cpu_set,
+                CPUSet *cpu_set,
                 bool warn,
                 const char *unit,
                 const char *filename,
                 unsigned line,
                 const char *lvalue) {
 
-        _cleanup_cpu_free_ cpu_set_t *c = NULL;
+        _cleanup_(cpu_set_reset) CPUSet c = {};
         const char *p = rvalue;
-        unsigned ncpus = 0;
 
-        assert(rvalue);
+        assert(p);
 
         for (;;) {
                 _cleanup_free_ char *word = NULL;
-                unsigned cpu, cpu_lower, cpu_upper;
+                unsigned cpu_lower, cpu_upper;
                 int r;
 
                 r = extract_first_word(&p, &word, WHITESPACE ",", EXTRACT_QUOTES);
@@ -93,31 +144,63 @@ int parse_cpu_set_internal(
                 if (r == 0)
                         break;
 
-                if (!c) {
-                        c = cpu_set_malloc(&ncpus);
-                        if (!c)
-                                return warn ? log_oom() : -ENOMEM;
-                }
-
                 r = parse_range(word, &cpu_lower, &cpu_upper);
                 if (r < 0)
                         return warn ? log_syntax(unit, LOG_ERR, filename, line, r, "Failed to parse CPU affinity '%s'", word) : r;
-                if (cpu_lower >= ncpus || cpu_upper >= ncpus)
-                        return warn ? log_syntax(unit, LOG_ERR, filename, line, EINVAL, "CPU out of range '%s' ncpus is %u", word, ncpus) : -EINVAL;
 
                 if (cpu_lower > cpu_upper) {
                         if (warn)
-                                log_syntax(unit, LOG_WARNING, filename, line, 0, "Range '%s' is invalid, %u > %u, ignoring", word, cpu_lower, cpu_upper);
-                        continue;
+                                log_syntax(unit, LOG_WARNING, filename, line, 0, "Range '%s' is invalid, %u > %u, ignoring.",
+                                           word, cpu_lower, cpu_upper);
+
+                        /* Make sure something is allocated, to distinguish this from the empty case */
+                        r = cpu_set_realloc(&c, 1);
+                        if (r < 0)
+                                return r;
                 }
 
-                for (cpu = cpu_lower; cpu <= cpu_upper; cpu++)
-                        CPU_SET_S(cpu, CPU_ALLOC_SIZE(ncpus), c);
+                for (unsigned cpu_p1 = MIN(cpu_upper, UINT_MAX-1) + 1; cpu_p1 > cpu_lower; cpu_p1--) {
+                        r = cpu_set_add(&c, cpu_p1 - 1);
+                        if (r < 0)
+                                return warn ? log_syntax(unit, LOG_ERR, filename, line, r,
+                                                         "Cannot add CPU %u to set: %m", cpu_p1 - 1) : r;
+                }
         }
 
-        /* On success, sets *cpu_set and returns ncpus for the system. */
-        if (c)
-                *cpu_set = TAKE_PTR(c);
+        /* On success, transfer ownership to the output variable */
+        *cpu_set = c;
+        c = (CPUSet) {};
+
+        return 0;
+}
+
+int parse_cpu_set_extend(
+                const char *rvalue,
+                CPUSet *old,
+                bool warn,
+                const char *unit,
+                const char *filename,
+                unsigned line,
+                const char *lvalue) {
+
+        _cleanup_(cpu_set_reset) CPUSet cpuset = {};
+        int r;
+
+        r = parse_cpu_set_full(rvalue, &cpuset, true, unit, filename, line, lvalue);
+        if (r < 0)
+                return r;
+
+        if (!cpuset.set) {
+                /* An empty assignment resets the CPU list */
+                cpu_set_reset(old);
+                return 0;
+        }
+
+        if (!old->set) {
+                *old = cpuset;
+                cpuset = (CPUSet) {};
+                return 0;
+        }
 
-        return (int) ncpus;
+        return cpu_set_add_all(old, &cpuset);
 }
diff --git a/src/basic/cpu-set-util.h b/src/basic/cpu-set-util.h
index 20612a8876..eb31b362fe 100644
--- a/src/basic/cpu-set-util.h
+++ b/src/basic/cpu-set-util.h
@@ -12,23 +12,40 @@
 DEFINE_TRIVIAL_CLEANUP_FUNC(cpu_set_t*, CPU_FREE);
 #define _cleanup_cpu_free_ _cleanup_(CPU_FREEp)
 
-static inline cpu_set_t* cpu_set_mfree(cpu_set_t *p) {
-        if (p)
-                CPU_FREE(p);
-        return NULL;
-}
-
 cpu_set_t* cpu_set_malloc(unsigned *ncpus);
 
-char* cpu_set_to_string(const cpu_set_t *set, size_t setsize);
-int parse_cpu_set_internal(const char *rvalue, cpu_set_t **cpu_set, bool warn, const char *unit, const char *filename, unsigned line, const char *lvalue);
-
-static inline int parse_cpu_set_and_warn(const char *rvalue, cpu_set_t **cpu_set, const char *unit, const char *filename, unsigned line, const char *lvalue) {
-        assert(lvalue);
-
-        return parse_cpu_set_internal(rvalue, cpu_set, true, unit, filename, line, lvalue);
+/* This wraps the libc interface with a variable to keep the allocated size. */
+typedef struct CPUSet {
+        cpu_set_t *set;
+        size_t allocated; /* in bytes */
+} CPUSet;
+
+static inline void cpu_set_reset(CPUSet *a) {
+        assert((a->allocated > 0) == !!a->set);
+        if (a->set)
+                CPU_FREE(a->set);
+        *a = (CPUSet) {};
 }
 
-static inline int parse_cpu_set(const char *rvalue, cpu_set_t **cpu_set){
-        return parse_cpu_set_internal(rvalue, cpu_set, false, NULL, NULL, 0, NULL);
+int cpu_set_add_all(CPUSet *a, const CPUSet *b);
+
+char* cpu_set_to_string(const CPUSet *a);
+int parse_cpu_set_full(
+                const char *rvalue,
+                CPUSet *cpu_set,
+                bool warn,
+                const char *unit,
+                const char *filename, unsigned line,
+                const char *lvalue);
+int parse_cpu_set_extend(
+                const char *rvalue,
+                CPUSet *old,
+                bool warn,
+                const char *unit,
+                const char *filename,
+                unsigned line,
+                const char *lvalue);
+
+static inline int parse_cpu_set(const char *rvalue, CPUSet *cpu_set){
+        return parse_cpu_set_full(rvalue, cpu_set, false, NULL, NULL, 0, NULL);
 }
diff --git a/src/core/dbus-execute.c b/src/core/dbus-execute.c
index d9f4445745..08946627e3 100644
--- a/src/core/dbus-execute.c
+++ b/src/core/dbus-execute.c
@@ -220,7 +220,7 @@ static int property_get_cpu_affinity(
         assert(reply);
         assert(c);
 
-        return sd_bus_message_append_array(reply, 'y', c->cpuset, CPU_ALLOC_SIZE(c->cpuset_ncpus));
+        return sd_bus_message_append_array(reply, 'y', c->cpu_set.set, c->cpu_set.allocated);
 }
 
 static int property_get_timer_slack_nsec(
@@ -1560,37 +1560,22 @@ int bus_exec_context_set_transient_property(
 
                 if (!UNIT_WRITE_FLAGS_NOOP(flags)) {
                         if (n == 0) {
-                                c->cpuset = cpu_set_mfree(c->cpuset);
-                                c->cpuset_ncpus = 0;
+                                cpu_set_reset(&c->cpu_set);
                                 unit_write_settingf(u, flags, name, "%s=", name);
                         } else {
                                 _cleanup_free_ char *str = NULL;
-                                size_t ncpus;
+                                const CPUSet set = {(cpu_set_t*) a, n};
 
-                                str = cpu_set_to_string(a, n);
+                                str = cpu_set_to_string(&set);
                                 if (!str)
                                         return -ENOMEM;
 
-                                ncpus = CPU_SIZE_TO_NUM(n);
-
-                                if (!c->cpuset || c->cpuset_ncpus < ncpus) {
-                                        cpu_set_t *cpuset;
-
-                                        cpuset = CPU_ALLOC(ncpus);
-                                        if (!cpuset)
-                                                return -ENOMEM;
-
-                                        CPU_ZERO_S(n, cpuset);
-                                        if (c->cpuset) {
-                                                CPU_OR_S(CPU_ALLOC_SIZE(c->cpuset_ncpus), cpuset, c->cpuset, (cpu_set_t*) a);
-                                                CPU_FREE(c->cpuset);
-                                        } else
-                                                CPU_OR_S(n, cpuset, cpuset, (cpu_set_t*) a);
-
-                                        c->cpuset = cpuset;
-                                        c->cpuset_ncpus = ncpus;
-                                } else
-                                        CPU_OR_S(n, c->cpuset, c->cpuset, (cpu_set_t*) a);
+                                /* We forego any optimizations here, and always create the structure using
+                                 * cpu_set_add_all(), because we don't want to care if the existing size we
+                                 * got over dbus is appropriate. */
+                                r = cpu_set_add_all(&c->cpu_set, &set);
+                                if (r < 0)
+                                        return r;
 
                                 unit_write_settingf(u, flags, name, "%s=%s", name, str);
                         }
diff --git a/src/core/execute.c b/src/core/execute.c
index c42300a41e..22e5825905 100644
--- a/src/core/execute.c
+++ b/src/core/execute.c
@@ -2991,8 +2991,8 @@ static int exec_child(
                 }
         }
 
-        if (context->cpuset)
-                if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
+        if (context->cpu_set.set)
+                if (sched_setaffinity(0, context->cpu_set.allocated, context->cpu_set.set) < 0) {
                         *exit_status = EXIT_CPUAFFINITY;
                         return log_unit_error_errno(unit, errno, "Failed to set up CPU affinity: %m");
                 }
@@ -3694,7 +3694,7 @@ void exec_context_done(ExecContext *c) {
         c->temporary_filesystems = NULL;
         c->n_temporary_filesystems = 0;
 
-        c->cpuset = cpu_set_mfree(c->cpuset);
+        cpu_set_reset(&c->cpu_set);
 
         c->utmp_id = mfree(c->utmp_id);
         c->selinux_context = mfree(c->selinux_context);
@@ -4097,10 +4097,10 @@ void exec_context_dump(const ExecContext *c, FILE* f, const char *prefix) {
                         prefix, yes_no(c->cpu_sched_reset_on_fork));
         }
 
-        if (c->cpuset) {
+        if (c->cpu_set.set) {
                 fprintf(f, "%sCPUAffinity:", prefix);
-                for (i = 0; i < c->cpuset_ncpus; i++)
-                        if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
+                for (i = 0; i < c->cpu_set.allocated * 8; i++)
+                        if (CPU_ISSET_S(i, c->cpu_set.allocated, c->cpu_set.set))
                                 fprintf(f, " %u", i);
                 fputs("\n", f);
         }
diff --git a/src/core/execute.h b/src/core/execute.h
index 8c91636adc..e1e7a494cd 100644
--- a/src/core/execute.h
+++ b/src/core/execute.h
@@ -14,6 +14,7 @@ typedef struct Manager Manager;
 #include <sys/capability.h>
 
 #include "cgroup-util.h"
+#include "cpu-set-util.h"
 #include "fdset.h"
 #include "list.h"
 #include "missing.h"
@@ -148,8 +149,7 @@ struct ExecContext {
         int cpu_sched_policy;
         int cpu_sched_priority;
 
-        cpu_set_t *cpuset;
-        unsigned cpuset_ncpus;
+        CPUSet cpu_set;
 
         ExecInput std_input;
         ExecOutput std_output;
diff --git a/src/core/load-fragment.c b/src/core/load-fragment.c
index d9a5094aa0..34ae834188 100644
--- a/src/core/load-fragment.c
+++ b/src/core/load-fragment.c
@@ -1211,42 +1211,13 @@ int config_parse_exec_cpu_affinity(const char *unit,
                                    void *userdata) {
 
         ExecContext *c = data;
-        _cleanup_cpu_free_ cpu_set_t *cpuset = NULL;
-        int ncpus;
 
         assert(filename);
         assert(lvalue);
         assert(rvalue);
         assert(data);
 
-        ncpus = parse_cpu_set_and_warn(rvalue, &cpuset, unit, filename, line, lvalue);
-        if (ncpus < 0)
-                return ncpus;
-
-        if (ncpus == 0) {
-                /* An empty assignment resets the CPU list */
-                c->cpuset = cpu_set_mfree(c->cpuset);
-                c->cpuset_ncpus = 0;
-                return 0;
-        }
-
-        if (!c->cpuset) {
-                c->cpuset = TAKE_PTR(cpuset);
-                c->cpuset_ncpus = (unsigned) ncpus;
-                return 0;
-        }
-
-        if (c->cpuset_ncpus < (unsigned) ncpus) {
-                CPU_OR_S(CPU_ALLOC_SIZE(c->cpuset_ncpus), cpuset, c->cpuset, cpuset);
-                CPU_FREE(c->cpuset);
-                c->cpuset = TAKE_PTR(cpuset);
-                c->cpuset_ncpus = (unsigned) ncpus;
-                return 0;
-        }
-
-        CPU_OR_S(CPU_ALLOC_SIZE((unsigned) ncpus), c->cpuset, c->cpuset, cpuset);
-
-        return 0;
+        return parse_cpu_set_extend(rvalue, &c->cpu_set, true, unit, filename, line, lvalue);
 }
 
 int config_parse_capability_set(
diff --git a/src/core/main.c b/src/core/main.c
index af7b26d6f1..e62b2756ee 100644
--- a/src/core/main.c
+++ b/src/core/main.c
@@ -537,16 +537,18 @@ static int config_parse_cpu_affinity2(
                 void *data,
                 void *userdata) {
 
-        _cleanup_cpu_free_ cpu_set_t *c = NULL;
-        int ncpus;
+        _cleanup_(cpu_set_reset) CPUSet c = {};
+        int r;
 
-        ncpus = parse_cpu_set_and_warn(rvalue, &c, unit, filename, line, lvalue);
-        if (ncpus < 0)
-                return ncpus;
+        r = parse_cpu_set_full(rvalue, &c, true, unit, filename, line, lvalue);
+        if (r < 0)
+                return r;
 
-        if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
+        if (sched_setaffinity(0, c.allocated, c.set) < 0)
                 log_warning_errno(errno, "Failed to set CPU affinity: %m");
 
+        // FIXME: parsing and execution should be seperated.
+
         return 0;
 }
 
diff --git a/src/nspawn/nspawn-settings.c b/src/nspawn/nspawn-settings.c
index 62a3486952..21c24a1111 100644
--- a/src/nspawn/nspawn-settings.c
+++ b/src/nspawn/nspawn-settings.c
@@ -85,7 +85,7 @@ Settings* settings_free(Settings *s) {
         strv_free(s->syscall_blacklist);
         rlimit_free_all(s->rlimit);
         free(s->hostname);
-        s->cpuset = cpu_set_mfree(s->cpuset);
+        cpu_set_reset(&s->cpu_set);
 
         strv_free(s->network_interfaces);
         strv_free(s->network_macvlan);
@@ -687,41 +687,12 @@ int config_parse_cpu_affinity(
                 void *data,
                 void *userdata) {
 
-        _cleanup_cpu_free_ cpu_set_t *cpuset = NULL;
         Settings *settings = data;
-        int ncpus;
 
         assert(rvalue);
         assert(settings);
 
-        ncpus = parse_cpu_set_and_warn(rvalue, &cpuset, unit, filename, line, lvalue);
-        if (ncpus < 0)
-                return ncpus;
-
-        if (ncpus == 0) {
-                /* An empty assignment resets the CPU list */
-                settings->cpuset = cpu_set_mfree(settings->cpuset);
-                settings->cpuset_ncpus = 0;
-                return 0;
-        }
-
-        if (!settings->cpuset) {
-                settings->cpuset = TAKE_PTR(cpuset);
-                settings->cpuset_ncpus = (unsigned) ncpus;
-                return 0;
-        }
-
-        if (settings->cpuset_ncpus < (unsigned) ncpus) {
-                CPU_OR_S(CPU_ALLOC_SIZE(settings->cpuset_ncpus), cpuset, settings->cpuset, cpuset);
-                CPU_FREE(settings->cpuset);
-                settings->cpuset = TAKE_PTR(cpuset);
-                settings->cpuset_ncpus = (unsigned) ncpus;
-                return 0;
-        }
-
-        CPU_OR_S(CPU_ALLOC_SIZE((unsigned) ncpus), settings->cpuset, settings->cpuset, cpuset);
-
-        return 0;
+        return parse_cpu_set_extend(rvalue, &settings->cpu_set, true, unit, filename, line, lvalue);
 }
 
 DEFINE_CONFIG_PARSE_ENUM(config_parse_resolv_conf, resolv_conf_mode, ResolvConfMode, "Failed to parse resolv.conf mode");
diff --git a/src/nspawn/nspawn-settings.h b/src/nspawn/nspawn-settings.h
index d522f3cb36..da863ef11c 100644
--- a/src/nspawn/nspawn-settings.h
+++ b/src/nspawn/nspawn-settings.h
@@ -7,6 +7,7 @@
 #include "sd-id128.h"
 
 #include "conf-parser.h"
+#include "cpu-set-util.h"
 #include "macro.h"
 #include "nspawn-expose-ports.h"
 #include "nspawn-mount.h"
@@ -123,8 +124,7 @@ typedef struct Settings {
         int no_new_privileges;
         int oom_score_adjust;
         bool oom_score_adjust_set;
-        cpu_set_t *cpuset;
-        unsigned cpuset_ncpus;
+        CPUSet cpu_set;
         ResolvConfMode resolv_conf;
         LinkJournal link_journal;
         bool link_journal_try;
diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c
index b40411dcd0..08255b5724 100644
--- a/src/nspawn/nspawn.c
+++ b/src/nspawn/nspawn.c
@@ -199,8 +199,7 @@ static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
 static bool arg_no_new_privileges = false;
 static int arg_oom_score_adjust = 0;
 static bool arg_oom_score_adjust_set = false;
-static cpu_set_t *arg_cpuset = NULL;
-static unsigned arg_cpuset_ncpus = 0;
+static CPUSet arg_cpu_set = {};
 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
 
@@ -1186,17 +1185,14 @@ static int parse_argv(int argc, char *argv[]) {
                         break;
 
                 case ARG_CPU_AFFINITY: {
-                        _cleanup_cpu_free_ cpu_set_t *cpuset = NULL;
+                        CPUSet cpuset;
 
                         r = parse_cpu_set(optarg, &cpuset);
                         if (r < 0)
-                                return log_error_errno(r, "Failed to parse CPU affinity mask: %s", optarg);
+                                return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
 
-                        if (arg_cpuset)
-                                CPU_FREE(arg_cpuset);
-
-                        arg_cpuset = TAKE_PTR(cpuset);
-                        arg_cpuset_ncpus = r;
+                        cpu_set_reset(&arg_cpu_set);
+                        arg_cpu_set = cpuset;
                         arg_settings_mask |= SETTING_CPU_AFFINITY;
                         break;
                 }
@@ -2631,8 +2627,8 @@ static int inner_child(
                         return log_error_errno(r, "Failed to adjust OOM score: %m");
         }
 
-        if (arg_cpuset)
-                if (sched_setaffinity(0, CPU_ALLOC_SIZE(arg_cpuset_ncpus), arg_cpuset) < 0)
+        if (arg_cpu_set.set)
+                if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
                         return log_error_errno(errno, "Failed to set CPU affinity: %m");
 
         r = drop_capabilities();
@@ -3494,15 +3490,14 @@ static int merge_settings(Settings *settings, const char *path) {
         }
 
         if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
-            settings->cpuset) {
+            settings->cpu_set.set) {
 
                 if (!arg_settings_trusted)
                         log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
                 else {
-                        if (arg_cpuset)
-                                CPU_FREE(arg_cpuset);
-                        arg_cpuset = TAKE_PTR(settings->cpuset);
-                        arg_cpuset_ncpus = settings->cpuset_ncpus;
+                        cpu_set_reset(&arg_cpu_set);
+                        arg_cpu_set = settings->cpu_set;
+                        settings->cpu_set = (CPUSet) {};
                 }
         }
 
@@ -4600,7 +4595,7 @@ finish:
         rlimit_free_all(arg_rlimit);
         strv_free(arg_syscall_whitelist);
         strv_free(arg_syscall_blacklist);
-        arg_cpuset = cpu_set_mfree(arg_cpuset);
+        cpu_set_reset(&arg_cpu_set);
 
         return r < 0 ? EXIT_FAILURE : ret;
 }
diff --git a/src/shared/bus-unit-util.c b/src/shared/bus-unit-util.c
index 271cc054da..75b4aace84 100644
--- a/src/shared/bus-unit-util.c
+++ b/src/shared/bus-unit-util.c
@@ -932,13 +932,13 @@ static int bus_append_execute_property(sd_bus_message *m, const char *field, con
         }
 
         if (streq(field, "CPUAffinity")) {
-                _cleanup_cpu_free_ cpu_set_t *cpuset = NULL;
+                _cleanup_(cpu_set_reset) CPUSet cpuset = {};
 
                 r = parse_cpu_set(eq, &cpuset);
                 if (r < 0)
                         return log_error_errno(r, "Failed to parse %s value: %s", field, eq);
 
-                return bus_append_byte_array(m, field, cpuset, CPU_ALLOC_SIZE(r));
+                return bus_append_byte_array(m, field, cpuset.set, cpuset.allocated);
         }
 
         if (STR_IN_SET(field, "RestrictAddressFamilies", "SystemCallFilter")) {
diff --git a/src/test/test-cpu-set-util.c b/src/test/test-cpu-set-util.c
index ff5edb2a69..b9ec29af66 100644
--- a/src/test/test-cpu-set-util.c
+++ b/src/test/test-cpu-set-util.c
@@ -1,154 +1,171 @@
 /* SPDX-License-Identifier: LGPL-2.1+ */
 
+#include <errno.h>
+
 #include "alloc-util.h"
 #include "cpu-set-util.h"
 #include "macro.h"
 
 static void test_parse_cpu_set(void) {
-        cpu_set_t *c = NULL;
+        CPUSet c = {};
         _cleanup_free_ char *str = NULL;
-        int ncpus;
         int cpu;
 
         /* Simple range (from CPUAffinity example) */
-        ncpus = parse_cpu_set_and_warn("1 2", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_ISSET_S(1, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(CPU_ISSET_S(2, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 2);
-
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+        assert_se(parse_cpu_set_full("1 2", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.set);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_ISSET_S(1, c.allocated, c.set));
+        assert_se(CPU_ISSET_S(2, c.allocated, c.set));
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 2);
+
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* A more interesting range */
-        ncpus = parse_cpu_set_and_warn("0 1 2 3 8 9 10 11", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 8);
+        assert_se(parse_cpu_set_full("0 1 2 3 8 9 10 11", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 8);
         for (cpu = 0; cpu < 4; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
         for (cpu = 8; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Quoted strings */
-        ncpus = parse_cpu_set_and_warn("8 '9' 10 \"11\"", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 4);
+        assert_se(parse_cpu_set_full("8 '9' 10 \"11\"", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 4);
         for (cpu = 8; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Use commas as separators */
-        ncpus = parse_cpu_set_and_warn("0,1,2,3 8,9,10,11", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 8);
+        assert_se(parse_cpu_set_full("0,1,2,3 8,9,10,11", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 8);
         for (cpu = 0; cpu < 4; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
         for (cpu = 8; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Commas with spaces (and trailing comma, space) */
-        ncpus = parse_cpu_set_and_warn("0, 1, 2, 3, 4, 5, 6, 7, ", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 8);
+        assert_se(parse_cpu_set_full("0, 1, 2, 3, 4, 5, 6, 7, ", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 8);
         for (cpu = 0; cpu < 8; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Ranges */
-        ncpus = parse_cpu_set_and_warn("0-3,8-11", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 8);
+        assert_se(parse_cpu_set_full("0-3,8-11", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 8);
         for (cpu = 0; cpu < 4; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
         for (cpu = 8; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Ranges with trailing comma, space */
-        ncpus = parse_cpu_set_and_warn("0-3  8-11, ", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 8);
+        assert_se(parse_cpu_set_full("0-3  8-11, ", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 8);
         for (cpu = 0; cpu < 4; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
         for (cpu = 8; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Negative range (returns empty cpu_set) */
-        ncpus = parse_cpu_set_and_warn("3-0", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 0);
-        c = cpu_set_mfree(c);
+        assert_se(parse_cpu_set_full("3-0", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 0);
+        cpu_set_reset(&c);
 
         /* Overlapping ranges */
-        ncpus = parse_cpu_set_and_warn("0-7 4-11", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 12);
+        assert_se(parse_cpu_set_full("0-7 4-11", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 12);
         for (cpu = 0; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Mix ranges and individual CPUs */
-        ncpus = parse_cpu_set_and_warn("0,1 4-11", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus >= 1024);
-        assert_se(CPU_COUNT_S(CPU_ALLOC_SIZE(ncpus), c) == 10);
-        assert_se(CPU_ISSET_S(0, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(CPU_ISSET_S(1, CPU_ALLOC_SIZE(ncpus), c));
+        assert_se(parse_cpu_set_full("0,1 4-11", &c, true, NULL, "fake", 1, "CPUAffinity") >= 0);
+        assert_se(c.allocated >= sizeof(__cpu_mask) / 8);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 10);
+        assert_se(CPU_ISSET_S(0, c.allocated, c.set));
+        assert_se(CPU_ISSET_S(1, c.allocated, c.set));
         for (cpu = 4; cpu < 12; cpu++)
-                assert_se(CPU_ISSET_S(cpu, CPU_ALLOC_SIZE(ncpus), c));
-        assert_se(str = cpu_set_to_string(c, CPU_ALLOC_SIZE(ncpus)));
+                assert_se(CPU_ISSET_S(cpu, c.allocated, c.set));
+        assert_se(str = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", str);
         str = mfree(str);
-        c = cpu_set_mfree(c);
+        cpu_set_reset(&c);
 
         /* Garbage */
-        ncpus = parse_cpu_set_and_warn("0 1 2 3 garbage", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus < 0);
-        assert_se(!c);
+        assert_se(parse_cpu_set_full("0 1 2 3 garbage", &c, true, NULL, "fake", 1, "CPUAffinity") == -EINVAL);
+        assert_se(!c.set);
+        assert_se(c.allocated == 0);
 
         /* Range with garbage */
-        ncpus = parse_cpu_set_and_warn("0-3 8-garbage", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus < 0);
-        assert_se(!c);
+        assert_se(parse_cpu_set_full("0-3 8-garbage", &c, true, NULL, "fake", 1, "CPUAffinity") == -EINVAL);
+        assert_se(!c.set);
+        assert_se(c.allocated == 0);
 
         /* Empty string */
-        c = NULL;
-        ncpus = parse_cpu_set_and_warn("", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus == 0);  /* empty string returns 0 */
-        assert_se(!c);
+        assert_se(parse_cpu_set_full("", &c, true, NULL, "fake", 1, "CPUAffinity") == 0);
+        assert_se(!c.set);                /* empty string returns NULL */
+        assert_se(c.allocated == 0);
 
         /* Runaway quoted string */
-        ncpus = parse_cpu_set_and_warn("0 1 2 3 \"4 5 6 7 ", &c, NULL, "fake", 1, "CPUAffinity");
-        assert_se(ncpus < 0);
-        assert_se(!c);
+        assert_se(parse_cpu_set_full("0 1 2 3 \"4 5 6 7 ", &c, true, NULL, "fake", 1, "CPUAffinity") == -EINVAL);
+        assert_se(!c.set);
+        assert_se(c.allocated == 0);
+
+        /* Maximum allocation */
+        assert_se(parse_cpu_set_full("8000-8191", &c, true, NULL, "fake", 1, "CPUAffinity") == 0);
+        assert_se(CPU_COUNT_S(c.allocated, c.set) == 192);
+        assert_se(str = cpu_set_to_string(&c));
+        log_info("cpu_set_to_string: %s", str);
+        str = mfree(str);
+        cpu_set_reset(&c);
 }
 
 int main(int argc, char *argv[]) {
+        log_info("CPU_ALLOC_SIZE(1) = %zu", CPU_ALLOC_SIZE(1));
+        log_info("CPU_ALLOC_SIZE(9) = %zu", CPU_ALLOC_SIZE(9));
+        log_info("CPU_ALLOC_SIZE(64) = %zu", CPU_ALLOC_SIZE(64));
+        log_info("CPU_ALLOC_SIZE(65) = %zu", CPU_ALLOC_SIZE(65));
+        log_info("CPU_ALLOC_SIZE(1024) = %zu", CPU_ALLOC_SIZE(1024));
+        log_info("CPU_ALLOC_SIZE(1025) = %zu", CPU_ALLOC_SIZE(1025));
+        log_info("CPU_ALLOC_SIZE(8191) = %zu", CPU_ALLOC_SIZE(8191));
+
         test_parse_cpu_set();
 
         return 0;
diff --git a/src/test/test-sizeof.c b/src/test/test-sizeof.c
index 7a1e496ed2..396e68f35f 100644
--- a/src/test/test-sizeof.c
+++ b/src/test/test-sizeof.c
@@ -1,5 +1,6 @@
 /* SPDX-License-Identifier: LGPL-2.1+ */
 
+#include <sched.h>
 #include <stdio.h>
 #include <string.h>
 
@@ -64,6 +65,8 @@ int main(void) {
         info(uid_t);
         info(gid_t);
 
+        info(__cpu_mask);
+
         info(enum Enum);
         info(enum BigEnum);
         info(enum BigEnum2);