sd-bus: sd_bus_set/get_method_call_timeout docs + cleanups.

core: add support for setting CPUAffinity= to special "numa" value

man/rules/meson.build

@@ -340,6 +340,10 @@ manpages = [
    'sd_bus_set_anonymous',
    'sd_bus_set_trusted'],
   ''],
+ ['sd_bus_set_method_call_timeout',
+  '3',
+  ['sd_bus_get_method_call_timeout'],
+  ''],
  ['sd_bus_set_sender', '3', ['sd_bus_get_sender'], ''],
  ['sd_bus_set_watch_bind', '3', ['sd_bus_get_watch_bind'], ''],
  ['sd_bus_slot_ref',

man/sd-bus.xml

@@ -55,6 +55,7 @@
 <citerefentry><refentrytitle>sd_bus_error</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_error_add_map</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_get_fd</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+<citerefentry><refentrytitle>sd_bus_get_method_call_timeout</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_get_n_queued_read</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_message_append</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_message_append_array</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
@@ -87,6 +88,7 @@
 <citerefentry><refentrytitle>sd_bus_request_name</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_set_connected_signal</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_set_description</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+<citerefentry><refentrytitle>sd_bus_set_method_call_timeout</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_set_sender</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
 <citerefentry><refentrytitle>sd_bus_set_watch_bind</refentrytitle><manvolnum>3</manvolnum></citerefentry>
 <citerefentry><refentrytitle>sd_bus_set_close_on_exit</refentrytitle><manvolnum>3</manvolnum></citerefentry>

man/sd_bus_call.xml

@@ -81,15 +81,8 @@
     on the reply object returned by <function>sd_bus_call()</function> or passed to the callback of
     <function>sd_bus_call_async()</function>.</para>
 
-    <para>If <parameter>usec</parameter> is zero, the timeout set using
-    <citerefentry><refentrytitle>sd_bus_set_method_call_timeout</refentrytitle><manvolnum>3</manvolnum></citerefentry>
-    is used. If no method call timeout was set, the timeout is read from the
-    <varname>$SYSTEMD_BUS_TIMEOUT</varname> environment variable. If this environment variable is
-    unset or does not contain a valid timeout, the implementation falls back to a predefined method
-    call timeout of 25 seconds. Note that <varname>$SYSTEMD_BUS_TIMEOUT</varname> is read once and
-    cached so callers should not rely on being able to change the default method call timeout at
-    runtime by changing the value of <varname>$SYSTEMD_BUS_TIMEOUT</varname>. Instead, call
-    <function>sd_bus_set_method_call_timeout()</function> to change the default method call timeout.
+    <para>If <parameter>usec</parameter> is zero, the default D-Bus method call timeout is used. See
+    <citerefentry><refentrytitle>sd_bus_get_method_call_timeout</refentrytitle><manvolnum>3</manvolnum></citerefentry>.
     </para>
 
   </refsect1>

man/sd_bus_set_description.xml

@@ -116,8 +116,8 @@
   <refsect1>
     <title>Return Value</title>
 
-    <para>On success, these functions return 0 or a positive integer. On failure, they return a negative
-    errno-style error code.</para>
+    <para>On success, these functions return a non-negative integer. On failure, they return a
+    negative errno-style error code.</para>
 
     <refsect2>
       <title>Errors</title>

man/sd_bus_set_method_call_timeout.xml

@@ -0,0 +1,104 @@
+<?xml version='1.0'?>
+<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.5//EN"
+  "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
+<!-- SPDX-License-Identifier: LGPL-2.1+ -->
+
+<refentry id="sd_bus_set_method_call_timeout" xmlns:xi="http://www.w3.org/2001/XInclude">
+
+  <refentryinfo>
+    <title>sd_bus_set_method_call_timeout</title>
+    <productname>systemd</productname>
+  </refentryinfo>
+
+  <refmeta>
+    <refentrytitle>sd_bus_set_method_call_timeout</refentrytitle>
+    <manvolnum>3</manvolnum>
+  </refmeta>
+
+  <refnamediv>
+    <refname>sd_bus_set_method_call_timeout</refname>
+    <refname>sd_bus_get_method_call_timeout</refname>
+
+    <refpurpose>Set or query the default D-Bus method call timeout of a bus object</refpurpose>
+  </refnamediv>
+
+  <refsynopsisdiv>
+    <funcsynopsis>
+      <funcsynopsisinfo>#include &lt;systemd/sd-bus.h&gt;</funcsynopsisinfo>
+
+      <funcprototype>
+        <funcdef>int <function>sd_bus_set_method_call_timeout</function></funcdef>
+        <paramdef>sd_bus *<parameter>bus</parameter></paramdef>
+        <paramdef>uint64_t <parameter>usec</parameter></paramdef>
+      </funcprototype>
+
+      <funcprototype>
+        <funcdef>int <function>sd_bus_get_method_call_timeout</function></funcdef>
+        <paramdef>sd_bus *<parameter>bus</parameter></paramdef>
+        <paramdef>uint64_t *<parameter>ret</parameter></paramdef>
+      </funcprototype>
+    </funcsynopsis>
+  </refsynopsisdiv>
+
+  <refsect1>
+    <title>Description</title>
+
+    <para><function>sd_bus_set_method_call_timeout()</function> sets the default D-Bus method call
+    timeout of <parameter>bus</parameter> to <parameter>usec</parameter> microseconds.</para>
+
+    <para><function>sd_bus_get_method_call_timeout()</function> queries the default D-Bus method
+    call timeout of <parameter>bus</parameter>. If no method call timeout was set using
+    <function>sd_bus_set_method_call_timeout()</function>, the timeout is read from the
+    <varname>$SYSTEMD_BUS_TIMEOUT</varname> environment variable. If this environment variable is
+    unset or does not contain a valid timeout, the implementation falls back to a predefined method
+    call timeout of 25 seconds. Note that <varname>$SYSTEMD_BUS_TIMEOUT</varname> is read once and
+    cached so callers should not rely on being able to change the default method call timeout at
+    runtime by changing the value of <varname>$SYSTEMD_BUS_TIMEOUT</varname>. Instead, call
+    <function>sd_bus_set_method_call_timeout()</function> to change the default method call timeout.
+    </para>
+  </refsect1>
+
+  <refsect1>
+    <title>Return Value</title>
+
+    <para>On success, these functions return a non-negative integer. On failure, they return a
+    negative errno-style error code.</para>
+
+    <refsect2>
+      <title>Errors</title>
+
+      <para>Returned errors may indicate the following problems:</para>
+
+      <variablelist>
+        <varlistentry>
+          <term><constant>-EINVAL</constant></term>
+
+          <listitem><para>The parameters <parameter>bus</parameter> or <parameter>ret</parameter>
+          are <constant>NULL</constant>.</para></listitem>
+        </varlistentry>
+
+        <varlistentry>
+          <term><constant>-ENOPKG</constant></term>
+
+          <listitem><para>Bus object <parameter>bus</parameter> could not be resolved.</para>
+          </listitem>
+        </varlistentry>
+      </variablelist>
+
+    </refsect2>
+  </refsect1>
+
+  <xi:include href="libsystemd-pkgconfig.xml" />
+
+  <refsect1>
+    <title>See Also</title>
+
+    <para>
+      <citerefentry><refentrytitle>systemd</refentrytitle><manvolnum>1</manvolnum></citerefentry>,
+      <citerefentry><refentrytitle>sd-bus</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+      <citerefentry><refentrytitle>sd_bus_call</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+      <citerefentry><refentrytitle>sd_bus_message_seal</refentrytitle><manvolnum>3</manvolnum></citerefentry>
+    </para>
+  </refsect1>
+
+</refentry>

man/systemd.exec.xml

@@ -774,10 +774,11 @@ CapabilityBoundingSet=~CAP_B CAP_C</programlisting>
         <term><varname>CPUAffinity=</varname></term>
 
         <listitem><para>Controls the CPU affinity of the executed processes. Takes a list of CPU indices or ranges
-        separated by either whitespace or commas. CPU ranges are specified by the lower and upper CPU indices separated
-        by a dash. This option may be specified more than once, in which case the specified CPU affinity masks are
-        merged. If the empty string is assigned, the mask is reset, all assignments prior to this will have no
-        effect. See
+        separated by either whitespace or commas. Alternatively, takes a special "numa" value in which case systemd
+        automatically derives allowed CPU range based on the value of <varname>NUMAMask=</varname> option. CPU ranges
+        are specified by the lower and upper CPU indices separated by a dash. This option may be specified more than
+        once, in which case the specified CPU affinity masks are merged. If the empty string is assigned, the mask
+        is reset, all assignments prior to this will have no effect. See
         <citerefentry><refentrytitle>sched_setaffinity</refentrytitle><manvolnum>2</manvolnum></citerefentry> for
         details.</para></listitem>
       </varlistentry>

src/core/dbus-execute.c

@@ -56,6 +56,8 @@ static BUS_DEFINE_PROPERTY_GET2(property_get_ioprio_priority, "i", ExecContext,
 static BUS_DEFINE_PROPERTY_GET_GLOBAL(property_get_empty_string, "s", NULL);
 static BUS_DEFINE_PROPERTY_GET_REF(property_get_syslog_level, "i", int, LOG_PRI);
 static BUS_DEFINE_PROPERTY_GET_REF(property_get_syslog_facility, "i", int, LOG_FAC);
+static BUS_DEFINE_PROPERTY_GET(property_get_cpu_affinity_from_numa, "b", ExecContext, exec_context_get_cpu_affinity_from_numa);
+
 
 static int property_get_environment_files(
                 sd_bus *bus,
@@ -213,6 +215,7 @@ static int property_get_cpu_affinity(
                 sd_bus_error *error) {
 
         ExecContext *c = userdata;
+        _cleanup_(cpu_set_reset) CPUSet s = {};
         _cleanup_free_ uint8_t *array = NULL;
         size_t allocated;
 
@@ -220,7 +223,16 @@ static int property_get_cpu_affinity(
         assert(reply);
         assert(c);
 
-        (void) cpu_set_to_dbus(&c->cpu_set, &array, &allocated);
+        if (c->cpu_affinity_from_numa) {
+                int r;
+
+                r = numa_to_cpu_set(&c->numa_policy, &s);
+                if (r < 0)
+                        return r;
+        }
+
+        (void) cpu_set_to_dbus(c->cpu_affinity_from_numa ? &s : &c->cpu_set,  &array, &allocated);
+
         return sd_bus_message_append_array(reply, 'y', array, allocated);
 }
 
@@ -741,6 +753,7 @@ const sd_bus_vtable bus_exec_vtable[] = {
         SD_BUS_PROPERTY("CPUSchedulingPolicy", "i", property_get_cpu_sched_policy, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("CPUSchedulingPriority", "i", property_get_cpu_sched_priority, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("CPUAffinity", "ay", property_get_cpu_affinity, 0, SD_BUS_VTABLE_PROPERTY_CONST),
+        SD_BUS_PROPERTY("CPUAffinityFromNUMA", "b", property_get_cpu_affinity_from_numa, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("NUMAPolicy", "i", property_get_numa_policy, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("NUMAMask", "ay", property_get_numa_mask, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("TimerSlackNSec", "t", property_get_timer_slack_nsec, 0, SD_BUS_VTABLE_PROPERTY_CONST),
@@ -1770,6 +1783,20 @@ int bus_exec_context_set_transient_property(
 
                 return 1;
 
+        } else if (streq(name, "CPUAffinityFromNUMA")) {
+                int q;
+
+                r = sd_bus_message_read_basic(message, 'b', &q);
+                if (r < 0)
+                        return r;
+
+                if (!UNIT_WRITE_FLAGS_NOOP(flags)) {
+                        c->cpu_affinity_from_numa = q;
+                        unit_write_settingf(u, flags, name, "%s=%s", "CPUAffinity", "numa");
+                }
+
+                return 1;
+
         } else if (streq(name, "NUMAPolicy")) {
                 int32_t type;
 
@@ -1784,6 +1811,7 @@ int bus_exec_context_set_transient_property(
                         c->numa_policy.type = type;
 
                 return 1;
+
         } else if (streq(name, "Nice")) {
                 int32_t q;
 

src/core/execute.c

@@ -3021,6 +3021,33 @@ static int exec_parameters_get_cgroup_path(const ExecParameters *params, char **
         return using_subcgroup;
 }
 
+static int exec_context_cpu_affinity_from_numa(const ExecContext *c, CPUSet *ret) {
+        _cleanup_(cpu_set_reset) CPUSet s = {};
+        int r;
+
+        assert(c);
+        assert(ret);
+
+        if (!c->numa_policy.nodes.set) {
+                log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
+                return 0;
+        }
+
+        r = numa_to_cpu_set(&c->numa_policy, &s);
+        if (r < 0)
+                return r;
+
+        cpu_set_reset(ret);
+
+        return cpu_set_add_all(ret, &s);
+}
+
+bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c) {
+        assert(c);
+
+        return c->cpu_affinity_from_numa;
+}
+
 static int exec_child(
                 Unit *unit,
                 const ExecCommand *command,
@@ -3318,11 +3345,26 @@ static int exec_child(
                 }
         }
 
-        if (context->cpu_set.set)
-                if (sched_setaffinity(0, context->cpu_set.allocated, context->cpu_set.set) < 0) {
+        if (context->cpu_affinity_from_numa || context->cpu_set.set) {
+                _cleanup_(cpu_set_reset) CPUSet converted_cpu_set = {};
+                const CPUSet *cpu_set;
+
+                if (context->cpu_affinity_from_numa) {
+                        r = exec_context_cpu_affinity_from_numa(context, &converted_cpu_set);
+                        if (r < 0) {
+                                *exit_status = EXIT_CPUAFFINITY;
+                                return log_unit_error_errno(unit, r, "Failed to derive CPU affinity mask from NUMA mask: %m");
+                        }
+
+                        cpu_set = &converted_cpu_set;
+                } else
+                        cpu_set = &context->cpu_set;
+
+                if (sched_setaffinity(0, cpu_set->allocated, cpu_set->set) < 0) {
                         *exit_status = EXIT_CPUAFFINITY;
                         return log_unit_error_errno(unit, errno, "Failed to set up CPU affinity: %m");
                 }
+        }
 
         if (mpol_is_valid(numa_policy_get_type(&context->numa_policy))) {
                 r = apply_numa_policy(&context->numa_policy);

src/core/execute.h

@@ -21,6 +21,7 @@ typedef struct Manager Manager;
 #include "missing_resource.h"
 #include "namespace.h"
 #include "nsflags.h"
+#include "numa-util.h"
 #include "time-util.h"
 
 #define EXEC_STDIN_DATA_MAX (64U*1024U*1024U)
@@ -181,6 +182,7 @@ struct ExecContext {
 
         CPUSet cpu_set;
         NUMAPolicy numa_policy;
+        bool cpu_affinity_from_numa;
 
         ExecInput std_input;
         ExecOutput std_output;
@@ -405,6 +407,8 @@ void exec_runtime_vacuum(Manager *m);
 
 void exec_params_clear(ExecParameters *p);
 
+bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c);
+
 const char* exec_output_to_string(ExecOutput i) _const_;
 ExecOutput exec_output_from_string(const char *s) _pure_;
 

src/core/load-fragment.c

@@ -1330,13 +1330,25 @@ int config_parse_exec_cpu_affinity(const char *unit,
                                    void *userdata) {
 
         ExecContext *c = data;
+        int r;
 
         assert(filename);
         assert(lvalue);
         assert(rvalue);
         assert(data);
 
-        return parse_cpu_set_extend(rvalue, &c->cpu_set, true, unit, filename, line, lvalue);
+        if (streq(rvalue, "numa")) {
+                c->cpu_affinity_from_numa = true;
+                cpu_set_reset(&c->cpu_set);
+
+                return 0;
+        }
+
+        r = parse_cpu_set_extend(rvalue, &c->cpu_set, true, unit, filename, line, lvalue);
+        if (r >= 0)
+                c->cpu_affinity_from_numa = false;
+
+        return r;
 }
 
 int config_parse_capability_set(

src/network/networkctl.c

@@ -161,6 +161,12 @@ typedef struct LinkInfo {
         union in_addr_union local;
         union in_addr_union remote;
 
+        /* bonding info */
+        uint8_t mode;
+        uint32_t miimon;
+        uint32_t updelay;
+        uint32_t downdelay;
+
         /* ethtool info */
         int autonegotiation;
         uint64_t speed;
@@ -229,7 +235,11 @@ static int decode_netdev(sd_netlink_message *m, LinkInfo *info) {
                 (void) sd_netlink_message_read_u32(m, IFLA_BR_STP_STATE, &info->stp_state);
                 (void) sd_netlink_message_read_u16(m, IFLA_BR_PRIORITY, &info->priority);
                 (void) sd_netlink_message_read_u8(m, IFLA_BR_MCAST_IGMP_VERSION, &info->mcast_igmp_version);
-
+        } if (streq(received_kind, "bond")) {
+                (void) sd_netlink_message_read_u8(m, IFLA_BOND_MODE, &info->mode);
+                (void) sd_netlink_message_read_u32(m, IFLA_BOND_MIIMON, &info->miimon);
+                (void) sd_netlink_message_read_u32(m, IFLA_BOND_DOWNDELAY, &info->downdelay);
+                (void) sd_netlink_message_read_u32(m, IFLA_BOND_UPDELAY, &info->updelay);
         } else if (streq(received_kind, "vxlan")) {
                 (void) sd_netlink_message_read_u32(m, IFLA_VXLAN_ID, &info->vxlan_info.vni);
 
@@ -1434,6 +1444,35 @@ static int link_status_one(
                 if (r < 0)
                         return table_log_add_error(r);
 
+        } else if (streq_ptr(info->netdev_kind, "bond")) {
+                static const struct {
+                        const char *mode;
+                } mode_table[] = {
+                        { "balance-rr" },
+                        { "active-backup" },
+                        { "balance-xor" },
+                        { "broadcast" },
+                        { "802.3ad" },
+                        { "balance-tlb" },
+                        { "balance-alb" },
+                };
+
+                r = table_add_many(table,
+                                   TABLE_EMPTY,
+                                   TABLE_STRING, "Mode:",
+                                   TABLE_STRING, mode_table[info->mode],
+                                   TABLE_EMPTY,
+                                   TABLE_STRING, "Miimon:",
+                                   TABLE_TIMESPAN_MSEC, jiffies_to_usec(info->miimon),
+                                   TABLE_EMPTY,
+                                   TABLE_STRING, "Updelay:",
+                                   TABLE_TIMESPAN_MSEC, jiffies_to_usec(info->updelay),
+                                   TABLE_EMPTY,
+                                   TABLE_STRING, "Downdelay:",
+                                   TABLE_TIMESPAN_MSEC, jiffies_to_usec(info->downdelay));
+                if (r < 0)
+                        return table_log_add_error(r);
+
         } else if (streq_ptr(info->netdev_kind, "vxlan")) {
                 if (info->vxlan_info.vni > 0) {
                         r = table_add_many(table,

src/shared/bus-unit-util.c

@@ -21,6 +21,7 @@
 #include "missing_fs.h"
 #include "mountpoint-util.h"
 #include "nsflags.h"
+#include "numa-util.h"
 #include "parse-util.h"
 #include "process-util.h"
 #include "rlimit-util.h"
@@ -28,6 +29,7 @@
 #include "signal-util.h"
 #include "socket-util.h"
 #include "sort-util.h"
+#include "stdio-util.h"
 #include "string-util.h"
 #include "syslog-util.h"
 #include "terminal-util.h"
@@ -1102,6 +1104,13 @@ static int bus_append_execute_property(sd_bus_message *m, const char *field, con
                 _cleanup_free_ uint8_t *array = NULL;
                 size_t allocated;
 
+                if (eq && streq(eq, "numa")) {
+                        r = sd_bus_message_append(m, "(sv)", "CPUAffinityFromNUMA", "b", true);
+                        if (r < 0)
+                                return bus_log_create_error(r);
+                        return r;
+                }
+
                 r = parse_cpu_set(eq, &cpuset);
                 if (r < 0)
                         return log_error_errno(r, "Failed to parse %s value: %s", field, eq);

src/shared/cpu-set-util.c

@@ -14,11 +14,9 @@
 #include "log.h"
 #include "macro.h"
 #include "memory-util.h"
-#include "missing_syscall.h"
 #include "parse-util.h"
 #include "stat-util.h"
 #include "string-util.h"
-#include "string-table.h"
 #include "strv.h"
 #include "util.h"
 
@@ -133,7 +131,7 @@ int cpu_set_add_all(CPUSet *a, const CPUSet *b) {
                                 return r;
                 }
 
-        return 0;
+        return 1;
 }
 
 int parse_cpu_set_full(
@@ -218,7 +216,7 @@ int parse_cpu_set_extend(
         if (!old->set) {
                 *old = cpuset;
                 cpuset = (CPUSet) {};
-                return 0;
+                return 1;
         }
 
         return cpu_set_add_all(old, &cpuset);
@@ -295,88 +293,3 @@ int cpu_set_from_dbus(const uint8_t *bits, size_t size, CPUSet *set) {
         s = (CPUSet) {};
         return 0;
 }
-
-bool numa_policy_is_valid(const NUMAPolicy *policy) {
-        assert(policy);
-
-        if (!mpol_is_valid(numa_policy_get_type(policy)))
-                return false;
-
-        if (!policy->nodes.set &&
-            !IN_SET(numa_policy_get_type(policy), MPOL_DEFAULT, MPOL_LOCAL, MPOL_PREFERRED))
-                return false;
-
-        if (policy->nodes.set &&
-            numa_policy_get_type(policy) == MPOL_PREFERRED &&
-            CPU_COUNT_S(policy->nodes.allocated, policy->nodes.set) != 1)
-                return false;
-
-        return true;
-}
-
-static int numa_policy_to_mempolicy(const NUMAPolicy *policy, unsigned long *ret_maxnode, unsigned long **ret_nodes) {
-        unsigned node, bits = 0, ulong_bits;
-        _cleanup_free_ unsigned long *out = NULL;
-
-        assert(policy);
-        assert(ret_maxnode);
-        assert(ret_nodes);
-
-        if (IN_SET(numa_policy_get_type(policy), MPOL_DEFAULT, MPOL_LOCAL) ||
-            (numa_policy_get_type(policy) == MPOL_PREFERRED && !policy->nodes.set)) {
-                *ret_nodes = NULL;
-                *ret_maxnode = 0;
-                return 0;
-        }
-
-        bits = policy->nodes.allocated * 8;
-        ulong_bits = sizeof(unsigned long) * 8;
-
-        out = new0(unsigned long, DIV_ROUND_UP(policy->nodes.allocated, sizeof(unsigned long)));
-        if (!out)
-                return -ENOMEM;
-
-        /* We don't make any assumptions about internal type libc is using to store NUMA node mask.
-           Hence we need to convert the node mask to the representation expected by set_mempolicy() */
-        for (node = 0; node < bits; node++)
-                if (CPU_ISSET_S(node, policy->nodes.allocated, policy->nodes.set))
-                        out[node / ulong_bits] |= 1ul << (node % ulong_bits);
-
-        *ret_nodes = TAKE_PTR(out);
-        *ret_maxnode = bits + 1;
-        return 0;
-}
-
-int apply_numa_policy(const NUMAPolicy *policy) {
-        int r;
-        _cleanup_free_ unsigned long *nodes = NULL;
-        unsigned long maxnode;
-
-        assert(policy);
-
-        if (get_mempolicy(NULL, NULL, 0, 0, 0) < 0 && errno == ENOSYS)
-                return -EOPNOTSUPP;
-
-        if (!numa_policy_is_valid(policy))
-                return -EINVAL;
-
-        r = numa_policy_to_mempolicy(policy, &maxnode, &nodes);
-        if (r < 0)
-                return r;
-
-        r = set_mempolicy(numa_policy_get_type(policy), nodes, maxnode);
-        if (r < 0)
-                return -errno;
-
-        return 0;
-}
-
-static const char* const mpol_table[] = {
-        [MPOL_DEFAULT]    = "default",
-        [MPOL_PREFERRED]  = "preferred",
-        [MPOL_BIND]       = "bind",
-        [MPOL_INTERLEAVE] = "interleave",
-        [MPOL_LOCAL]      = "local",
-};
-
-DEFINE_STRING_TABLE_LOOKUP(mpol, int);

src/shared/cpu-set-util.h

@@ -49,30 +49,3 @@ int cpu_set_to_dbus(const CPUSet *set, uint8_t **ret, size_t *allocated);
 int cpu_set_from_dbus(const uint8_t *bits, size_t size, CPUSet *set);
 
 int cpus_in_affinity_mask(void);
-
-static inline bool mpol_is_valid(int t) {
-        return t >= MPOL_DEFAULT && t <= MPOL_LOCAL;
-}
-
-typedef struct NUMAPolicy {
-        /* Always use numa_policy_get_type() to read the value */
-        int type;
-        CPUSet nodes;
-} NUMAPolicy;
-
-bool numa_policy_is_valid(const NUMAPolicy *p);
-
-static inline int numa_policy_get_type(const NUMAPolicy *p) {
-        return p->type < 0 ? (p->nodes.set ? MPOL_PREFERRED : -1) : p->type;
-}
-
-static inline void numa_policy_reset(NUMAPolicy *p) {
-        assert(p);
-        cpu_set_reset(&p->nodes);
-        p->type = -1;
-}
-
-int apply_numa_policy(const NUMAPolicy *policy);
-
-const char* mpol_to_string(int i) _const_;
-int mpol_from_string(const char *s) _pure_;

src/shared/meson.build

@@ -147,6 +147,8 @@ shared_sources = files('''
         nscd-flush.h
         nsflags.c
         nsflags.h
+        numa-util.c
+        numa-util.h
         openssl-util.h
         os-util.c
         os-util.h

src/shared/numa-util.c

@@ -0,0 +1,135 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+
+#include <errno.h>
+#include <sched.h>
+
+#include "alloc-util.h"
+#include "cpu-set-util.h"
+#include "fileio.h"
+#include "macro.h"
+#include "missing_syscall.h"
+#include "numa-util.h"
+#include "stdio-util.h"
+#include "string-table.h"
+
+bool numa_policy_is_valid(const NUMAPolicy *policy) {
+        assert(policy);
+
+        if (!mpol_is_valid(numa_policy_get_type(policy)))
+                return false;
+
+        if (!policy->nodes.set &&
+            !IN_SET(numa_policy_get_type(policy), MPOL_DEFAULT, MPOL_LOCAL, MPOL_PREFERRED))
+                return false;
+
+        if (policy->nodes.set &&
+            numa_policy_get_type(policy) == MPOL_PREFERRED &&
+            CPU_COUNT_S(policy->nodes.allocated, policy->nodes.set) != 1)
+                return false;
+
+        return true;
+}
+
+static int numa_policy_to_mempolicy(const NUMAPolicy *policy, unsigned long *ret_maxnode, unsigned long **ret_nodes) {
+        unsigned node, bits = 0, ulong_bits;
+        _cleanup_free_ unsigned long *out = NULL;
+
+        assert(policy);
+        assert(ret_maxnode);
+        assert(ret_nodes);
+
+        if (IN_SET(numa_policy_get_type(policy), MPOL_DEFAULT, MPOL_LOCAL) ||
+            (numa_policy_get_type(policy) == MPOL_PREFERRED && !policy->nodes.set)) {
+                *ret_nodes = NULL;
+                *ret_maxnode = 0;
+                return 0;
+        }
+
+        bits = policy->nodes.allocated * 8;
+        ulong_bits = sizeof(unsigned long) * 8;
+
+        out = new0(unsigned long, DIV_ROUND_UP(policy->nodes.allocated, sizeof(unsigned long)));
+        if (!out)
+                return -ENOMEM;
+
+        /* We don't make any assumptions about internal type libc is using to store NUMA node mask.
+           Hence we need to convert the node mask to the representation expected by set_mempolicy() */
+        for (node = 0; node < bits; node++)
+                if (CPU_ISSET_S(node, policy->nodes.allocated, policy->nodes.set))
+                        out[node / ulong_bits] |= 1ul << (node % ulong_bits);
+
+        *ret_nodes = TAKE_PTR(out);
+        *ret_maxnode = bits + 1;
+        return 0;
+}
+
+int apply_numa_policy(const NUMAPolicy *policy) {
+        int r;
+        _cleanup_free_ unsigned long *nodes = NULL;
+        unsigned long maxnode;
+
+        assert(policy);
+
+        if (get_mempolicy(NULL, NULL, 0, 0, 0) < 0 && errno == ENOSYS)
+                return -EOPNOTSUPP;
+
+        if (!numa_policy_is_valid(policy))
+                return -EINVAL;
+
+        r = numa_policy_to_mempolicy(policy, &maxnode, &nodes);
+        if (r < 0)
+                return r;
+
+        r = set_mempolicy(numa_policy_get_type(policy), nodes, maxnode);
+        if (r < 0)
+                return -errno;
+
+        return 0;
+}
+
+int numa_to_cpu_set(const NUMAPolicy *policy, CPUSet *ret) {
+        int r;
+        size_t i;
+        _cleanup_(cpu_set_reset) CPUSet s = {};
+
+        assert(policy);
+        assert(ret);
+
+        for (i = 0; i < policy->nodes.allocated * 8; i++) {
+                _cleanup_free_ char *l = NULL;
+                char p[STRLEN("/sys/devices/system/node/node//cpulist") + DECIMAL_STR_MAX(size_t) + 1];
+                _cleanup_(cpu_set_reset) CPUSet part = {};
+
+                if (!CPU_ISSET_S(i, policy->nodes.allocated, policy->nodes.set))
+                        continue;
+
+                xsprintf(p, "/sys/devices/system/node/node%zu/cpulist", i);
+
+                r = read_one_line_file(p, &l);
+                if (r < 0)
+                        return r;
+
+                r = parse_cpu_set(l, &part);
+                if (r < 0)
+                        return r;
+
+                r = cpu_set_add_all(&s, &part);
+                if (r < 0)
+                        return r;
+        }
+
+        *ret = s;
+        s = (CPUSet) {};
+
+        return 0;
+}
+
+static const char* const mpol_table[] = {
+        [MPOL_DEFAULT]    = "default",
+        [MPOL_PREFERRED]  = "preferred",
+        [MPOL_BIND]       = "bind",
+        [MPOL_INTERLEAVE] = "interleave",
+        [MPOL_LOCAL]      = "local",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(mpol, int);

src/shared/numa-util.h

@@ -0,0 +1,33 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+#pragma once
+
+#include "cpu-set-util.h"
+#include "missing_syscall.h"
+
+static inline bool mpol_is_valid(int t) {
+        return t >= MPOL_DEFAULT && t <= MPOL_LOCAL;
+}
+
+typedef struct NUMAPolicy {
+        /* Always use numa_policy_get_type() to read the value */
+        int type;
+        CPUSet nodes;
+} NUMAPolicy;
+
+bool numa_policy_is_valid(const NUMAPolicy *p);
+
+static inline int numa_policy_get_type(const NUMAPolicy *p) {
+        return p->type < 0 ? (p->nodes.set ? MPOL_PREFERRED : -1) : p->type;
+}
+
+static inline void numa_policy_reset(NUMAPolicy *p) {
+        assert(p);
+        cpu_set_reset(&p->nodes);
+        p->type = -1;
+}
+
+int apply_numa_policy(const NUMAPolicy *policy);
+int numa_to_cpu_set(const NUMAPolicy *policy, CPUSet *set);
+
+const char* mpol_to_string(int i) _const_;
+int mpol_from_string(const char *s) _pure_;

src/systemctl/systemctl.c

@@ -58,6 +58,7 @@
 #include "main-func.h"
 #include "memory-util.h"
 #include "mkdir.h"
+#include "numa-util.h"
 #include "pager.h"
 #include "parse-util.h"
 #include "path-lookup.h"

src/test/test-cpu-set-util.c

@@ -216,12 +216,12 @@ static void test_parse_cpu_set_extend(void) {
 
         log_info("/* %s */", __func__);
 
-        assert_se(parse_cpu_set_extend("1 3", &c, true, NULL, "fake", 1, "CPUAffinity") == 0);
+        assert_se(parse_cpu_set_extend("1 3", &c, true, NULL, "fake", 1, "CPUAffinity") == 1);
         assert_se(CPU_COUNT_S(c.allocated, c.set) == 2);
         assert_se(s1 = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", s1);
 
-        assert_se(parse_cpu_set_extend("4", &c, true, NULL, "fake", 1, "CPUAffinity") == 0);
+        assert_se(parse_cpu_set_extend("4", &c, true, NULL, "fake", 1, "CPUAffinity") == 1);
         assert_se(CPU_COUNT_S(c.allocated, c.set) == 3);
         assert_se(s2 = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", s2);
@@ -238,7 +238,7 @@ static void test_cpu_set_to_from_dbus(void) {
 
         log_info("/* %s */", __func__);
 
-        assert_se(parse_cpu_set_extend("1 3 8 100-200", &c, true, NULL, "fake", 1, "CPUAffinity") == 0);
+        assert_se(parse_cpu_set_extend("1 3 8 100-200", &c, true, NULL, "fake", 1, "CPUAffinity") == 1);
         assert_se(s = cpu_set_to_string(&c));
         log_info("cpu_set_to_string: %s", s);
         assert_se(CPU_COUNT_S(c.allocated, c.set) == 104);

test/TEST-36-NUMAPOLICY/testsuite.sh

@@ -279,6 +279,18 @@ else
     # Maks must be ignored
     grep -E "set_mempolicy\((MPOL_LOCAL|0x4 [^,]*), NULL" $straceLog
 
+    echo "Unit file CPUAffinity=NUMA support"
+    writeTestUnitNUMAPolicy "bind" "0"
+    echo "CPUAffinity=numa" >> $testUnitNUMAConf
+    systemctl daemon-reload
+    systemctl start $testUnit
+    systemctlCheckNUMAProperties $testUnit "bind" "0"
+    pid=$(systemctl show --value -p MainPID $testUnit)
+    cpulist=$(cat /sys/devices/system/node/node0/cpulist)
+    affinity_systemd=$(systemctl show --value -p CPUAffinity $testUnit)
+    [ $cpulist = $affinity_systemd ]
+    pid1StopUnit $testUnit
+
     echo "systemd-run NUMAPolicy support"
     runUnit='numa-systemd-run-test.service'
 
@@ -309,6 +321,12 @@ else
     systemd-run -p NUMAPolicy=local -p NUMAMask=0 --unit $runUnit sleep 1000
     systemctlCheckNUMAProperties $runUnit "local" ""
     pid1StopUnit $runUnit
+
+    systemd-run -p NUMAPolicy=local -p NUMAMask=0 -p CPUAffinity=numa --unit $runUnit sleep 1000
+    systemctlCheckNUMAProperties $runUnit "local" ""
+    systemctl cat $runUnit | grep -q 'CPUAffinity=numa'
+    pid1StopUnit $runUnit
+
 fi
 
 # Cleanup