core: add Condition[Memory/CPU/IO]Pressure

By default checks PSI on /proc/pressure, and causes a unit to be skipped
if the threshold is above the given configuration for the avg300
measurement.
Also allow to pass a custom timespan, and a particular slice unit to
check under.

Fixes #20139

man/systemd.unit.xml

@@ -1658,6 +1658,29 @@
           </listitem>
         </varlistentry>
 
+        <varlistentry>
+          <term><varname>ConditionMemoryPressure=</varname></term>
+          <term><varname>ConditionCPUPressure=</varname></term>
+          <term><varname>ConditionIOPressure=</varname></term>
+
+          <listitem><para>Verify that the overall system (memory, CPU or IO) pressure is below or equal to a threshold.
+          This setting takes a threshold value as argument. It can be specified as a simple percentage value,
+          suffixed with <literal>%</literal>, in which case the pressure will be measured as an average over the last
+          five minutes before the attempt to start the unit is performed.
+          Alternatively, the average timespan can also be specified using <literal>/</literal> as a separator, for
+          example: <literal>10%/1min</literal>. The supported timespans match what the kernel provides, and are
+          limited to <literal>10sec</literal>, <literal>1min</literal> and <literal>5min</literal>. The
+          <literal>full</literal> PSI will be checked first, and if not found <literal>some</literal> will be
+          checked. For more details, see the documentation on <ulink
+          url="https://www.kernel.org/doc/html/latest/accounting/psi.html">PSI (Pressure Stall Information)
+          </ulink>.</para>
+
+          <para>Optionally, the threshold value can be prefixed with the slice unit under which the pressure will be checked,
+          followed by a <literal>:</literal>. If the slice unit is not specified, the overall system pressure will be measured,
+          instead of a particular cgroup's.</para>
+          </listitem>
+        </varlistentry>
+
         <varlistentry>
           <term><varname>AssertArchitecture=</varname></term>
           <term><varname>AssertVirtualization=</varname></term>
@@ -1686,6 +1709,9 @@
           <term><varname>AssertMemory=</varname></term>
           <term><varname>AssertCPUs=</varname></term>
           <term><varname>AssertOSRelease=</varname></term>
+          <term><varname>AssertMemoryPressure=</varname></term>
+          <term><varname>AssertCPUPressure=</varname></term>
+          <term><varname>AssertIOPressure=</varname></term>
 
           <listitem><para>Similar to the <varname>ConditionArchitecture=</varname>,
           <varname>ConditionVirtualization=</varname>, …, condition settings described above, these settings

src/core/load-fragment-gperf.gperf.in

@@ -341,6 +341,9 @@ Unit.ConditionUser,                      config_parse_unit_condition_string,
 Unit.ConditionGroup,                     config_parse_unit_condition_string,          CONDITION_GROUP,                    offsetof(Unit, conditions)
 Unit.ConditionControlGroupController,    config_parse_unit_condition_string,          CONDITION_CONTROL_GROUP_CONTROLLER, offsetof(Unit, conditions)
 Unit.ConditionOSRelease,                 config_parse_unit_condition_string,          CONDITION_OS_RELEASE,               offsetof(Unit, conditions)
+Unit.ConditionMemoryPressure,            config_parse_unit_condition_string,          CONDITION_MEMORY_PRESSURE,          offsetof(Unit, conditions)
+Unit.ConditionCPUPressure,               config_parse_unit_condition_string,          CONDITION_CPU_PRESSURE,             offsetof(Unit, conditions)
+Unit.ConditionIOPressure,                config_parse_unit_condition_string,          CONDITION_IO_PRESSURE,              offsetof(Unit, conditions)
 Unit.AssertPathExists,                   config_parse_unit_condition_path,            CONDITION_PATH_EXISTS,              offsetof(Unit, asserts)
 Unit.AssertPathExistsGlob,               config_parse_unit_condition_path,            CONDITION_PATH_EXISTS_GLOB,         offsetof(Unit, asserts)
 Unit.AssertPathIsDirectory,              config_parse_unit_condition_path,            CONDITION_PATH_IS_DIRECTORY,        offsetof(Unit, asserts)
@@ -368,6 +371,9 @@ Unit.AssertUser,                         config_parse_unit_condition_string,
 Unit.AssertGroup,                        config_parse_unit_condition_string,          CONDITION_GROUP,                    offsetof(Unit, asserts)
 Unit.AssertControlGroupController,       config_parse_unit_condition_string,          CONDITION_CONTROL_GROUP_CONTROLLER, offsetof(Unit, asserts)
 Unit.AssertOSRelease,                    config_parse_unit_condition_string,          CONDITION_OS_RELEASE,               offsetof(Unit, asserts)
+Unit.AssertMemoryPressure,               config_parse_unit_condition_string,          CONDITION_MEMORY_PRESSURE,          offsetof(Unit, asserts)
+Unit.AssertCPUPressure,                  config_parse_unit_condition_string,          CONDITION_CPU_PRESSURE,             offsetof(Unit, asserts)
+Unit.AssertIOPressure,                   config_parse_unit_condition_string,          CONDITION_IO_PRESSURE,              offsetof(Unit, asserts)
 Unit.CollectMode,                        config_parse_collect_mode,                   0,                                  offsetof(Unit, collect_mode)
 Service.PIDFile,                         config_parse_pid_file,                       0,                                  offsetof(Service, pid_file)
 Service.ExecCondition,                   config_parse_exec,                           SERVICE_EXEC_CONDITION,             offsetof(Service, exec_command)

src/shared/condition.c

@@ -39,8 +39,10 @@
 #include "os-util.h"
 #include "parse-util.h"
 #include "path-util.h"
+#include "percent-util.h"
 #include "proc-cmdline.h"
 #include "process-util.h"
+#include "psi-util.h"
 #include "selinux-util.h"
 #include "smack-util.h"
 #include "stat-util.h"
@@ -962,6 +964,131 @@ static int condition_test_file_is_executable(Condition *c, char **env) {
                 (st.st_mode & 0111));
 }
 
+static int condition_test_psi(Condition *c, char **env) {
+        _cleanup_free_ char *first = NULL, *second = NULL, *third = NULL, *fourth = NULL, *pressure_path = NULL;
+        const char *p, *value, *pressure_type;
+        loadavg_t *current, limit;
+        ResourcePressure pressure;
+        int r;
+
+        assert(c);
+        assert(c->parameter);
+        assert(IN_SET(c->type, CONDITION_MEMORY_PRESSURE, CONDITION_CPU_PRESSURE, CONDITION_IO_PRESSURE));
+
+        if (!is_pressure_supported()) {
+                log_debug("Pressure Stall Information (PSI) is not supported, skipping.");
+                return 1;
+        }
+
+        pressure_type = c->type == CONDITION_MEMORY_PRESSURE ? "memory" :
+                        c->type == CONDITION_CPU_PRESSURE ? "cpu" :
+                        "io";
+
+        p = c->parameter;
+        r = extract_many_words(&p, ":", 0, &first, &second, NULL);
+        if (r <= 0)
+                return log_debug_errno(r < 0 ? r : SYNTHETIC_ERRNO(EINVAL), "Failed to parse condition parameter %s: %m", c->parameter);
+        /* If only one parameter is passed, then we look at the global system pressure rather than a specific cgroup. */
+        if (r == 1) {
+                pressure_path = path_join("/proc/pressure", pressure_type);
+                if (!pressure_path)
+                        return log_oom();
+
+                value = first;
+        } else {
+                const char *controller = strjoina(pressure_type, ".pressure");
+                _cleanup_free_ char *slice_path = NULL;
+                CGroupMask mask, required_mask;
+                char *slice;
+
+                required_mask = c->type == CONDITION_MEMORY_PRESSURE ? CGROUP_MASK_MEMORY :
+                                c->type == CONDITION_CPU_PRESSURE ? CGROUP_MASK_CPU :
+                                CGROUP_MASK_IO;
+
+                slice = strstrip(first);
+                if (!slice)
+                        return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse condition parameter %s: %m", c->parameter);
+
+                r = cg_all_unified();
+                if (r < 0)
+                        return log_debug_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
+                if (r == 0) {
+                        log_debug("PSI condition check requires the unified cgroups hierarchy, skipping.");
+                        return 1;
+                }
+
+                r = cg_mask_supported(&mask);
+                if (r < 0)
+                        return log_debug_errno(r, "Failed to get supported cgroup controllers: %m");
+
+                if (!FLAGS_SET(mask, required_mask)) {
+                        log_debug("Cgroup %s controller not available, skipping PSI condition check.", pressure_type);
+                        return 1;
+                }
+
+                r = cg_slice_to_path(slice, &slice_path);
+                if (r < 0)
+                        return log_debug_errno(r, "Cannot determine slice \"%s\" cgroup path: %m", slice);
+
+                r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, slice_path, controller, &pressure_path);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting cgroup pressure path from %s: %m", slice_path);
+
+                value = second;
+        }
+
+        /* If a value including a specific timespan (in the intervals allowed by the kernel),
+         * parse it, otherwise we assume just a plain percentage that will be checked if it is
+         * smaller or equal to the current pressure average over 5 minutes. */
+        r = extract_many_words(&value, "/", 0, &third, &fourth, NULL);
+        if (r <= 0)
+                return log_debug_errno(r < 0 ? r : SYNTHETIC_ERRNO(EINVAL), "Failed to parse condition parameter %s: %m", c->parameter);
+        if (r == 1)
+                current = &pressure.avg300;
+        else {
+                const char *timespan;
+
+                timespan = skip_leading_chars(fourth, NULL);
+                if (!timespan)
+                        return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse condition parameter %s: %m", c->parameter);
+
+                if (startswith(timespan, "10sec"))
+                        current = &pressure.avg10;
+                else if (startswith(timespan, "1min"))
+                        current = &pressure.avg60;
+                else if (startswith(timespan, "5min"))
+                        current = &pressure.avg300;
+                else
+                        return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse condition parameter %s: %m", c->parameter);
+        }
+
+        value = strstrip(third);
+        if (!value)
+                return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse condition parameter %s: %m", c->parameter);
+
+        r = parse_permyriad(value);
+        if (r < 0)
+                return log_debug_errno(r, "Failed to parse permyriad: %s", c->parameter);
+
+        r = store_loadavg_fixed_point(r / 100LU, r % 100LU, &limit);
+        if (r < 0)
+                return log_debug_errno(r, "Failed to parse loadavg: %s", c->parameter);
+
+        r = read_resource_pressure(pressure_path, PRESSURE_TYPE_FULL, &pressure);
+        if (r == -ENODATA) /* cpu.pressure 'full' was added recently, fall back to 'some'. */
+                r = read_resource_pressure(pressure_path, PRESSURE_TYPE_SOME, &pressure);
+        if (r == -ENOENT) {
+                /* We already checked that /proc/pressure exists, so this means we were given a cgroup
+                 * that doesn't exist or doesn't exist any longer. */
+                log_debug("\"%s\" not found, skipping PSI check.", pressure_path);
+                return 1;
+        }
+        if (r < 0)
+                return log_debug_errno(r, "Error parsing pressure from %s: %m", pressure_path);
+
+        return *current <= limit;
+}
+
 int condition_test(Condition *c, char **env) {
 
         static int (*const condition_tests[_CONDITION_TYPE_MAX])(Condition *c, char **env) = {
@@ -994,6 +1121,9 @@ int condition_test(Condition *c, char **env) {
                 [CONDITION_ENVIRONMENT]              = condition_test_environment,
                 [CONDITION_CPU_FEATURE]              = condition_test_cpufeature,
                 [CONDITION_OS_RELEASE]               = condition_test_osrelease,
+                [CONDITION_MEMORY_PRESSURE]          = condition_test_psi,
+                [CONDITION_CPU_PRESSURE]             = condition_test_psi,
+                [CONDITION_IO_PRESSURE]              = condition_test_psi,
         };
 
         int r, b;
@@ -1119,6 +1249,9 @@ static const char* const condition_type_table[_CONDITION_TYPE_MAX] = {
         [CONDITION_ENVIRONMENT] = "ConditionEnvironment",
         [CONDITION_CPU_FEATURE] = "ConditionCPUFeature",
         [CONDITION_OS_RELEASE] = "ConditionOSRelease",
+        [CONDITION_MEMORY_PRESSURE] = "ConditionMemoryPressure",
+        [CONDITION_CPU_PRESSURE] = "ConditionCPUPressure",
+        [CONDITION_IO_PRESSURE] = "ConditionIOPressure",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(condition_type, ConditionType);
@@ -1153,6 +1286,9 @@ static const char* const assert_type_table[_CONDITION_TYPE_MAX] = {
         [CONDITION_ENVIRONMENT] = "AssertEnvironment",
         [CONDITION_CPU_FEATURE] = "AssertCPUFeature",
         [CONDITION_OS_RELEASE] = "AssertOSRelease",
+        [CONDITION_MEMORY_PRESSURE] = "AssertMemoryPressure",
+        [CONDITION_CPU_PRESSURE] = "AssertCPUPressure",
+        [CONDITION_IO_PRESSURE] = "AssertIOPressure",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(assert_type, ConditionType);

src/shared/condition.h

@@ -22,6 +22,9 @@ typedef enum ConditionType {
         CONDITION_ENVIRONMENT,
         CONDITION_CPU_FEATURE,
         CONDITION_OS_RELEASE,
+        CONDITION_MEMORY_PRESSURE,
+        CONDITION_CPU_PRESSURE,
+        CONDITION_IO_PRESSURE,
 
         CONDITION_NEEDS_UPDATE,
         CONDITION_FIRST_BOOT,

src/test/test-condition.c

@@ -25,6 +25,7 @@
 #include "nulstr-util.h"
 #include "os-util.h"
 #include "process-util.h"
+#include "psi-util.h"
 #include "selinux-util.h"
 #include "set.h"
 #include "smack-util.h"
@@ -1031,4 +1032,158 @@ TEST(condition_test_os_release) {
         condition_free(condition);
 }
 
+TEST(condition_test_psi) {
+        Condition *condition;
+        CGroupMask mask;
+        int r;
+
+        if (!is_pressure_supported())
+                return (void) log_notice("Pressure Stall Information (PSI) is not supported, skipping %s", __func__);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "sbarabau", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "10%sbarabau", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "10% sbarabau", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "-10", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "10%/10min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "10min/10%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "10% 5min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "/5min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_IO_PRESSURE, "10s /   ", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "100%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "0%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "0.0%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "100%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "0%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "0.0%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "0.01%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "0.0%/10sec", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "100.0% / 1min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_IO_PRESSURE, "50.0% / 1min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        r = cg_all_unified();
+        if (r < 0)
+                return (void) log_notice("Failed to determine whether the unified cgroups hierarchy is used, skipping %s", __func__);
+        if (r == 0)
+                return (void) log_notice("Requires the unified cgroups hierarchy, skipping %s", __func__);
+
+        if (cg_mask_supported(&mask) < 0)
+                return (void) log_notice("Failed to get supported cgroup controllers, skipping %s", __func__);
+
+        if (!FLAGS_SET(mask, CGROUP_MASK_MEMORY))
+                return (void) log_notice("Requires the cgroup memory controller, skipping %s", __func__);
+
+        if (!FLAGS_SET(mask, CGROUP_MASK_CPU))
+                return (void) log_notice("Requires the cgroup CPU controller, skipping %s", __func__);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, " : / ", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) < 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "hopefullythisisnotarealone.slice:100% / 10sec", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) > 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_CPU_PRESSURE, "-.slice:100.0% / 1min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "-.slice:0.0%/5min", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_MEMORY_PRESSURE, "-.slice:100.0%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+
+        condition = condition_new(CONDITION_IO_PRESSURE, "-.slice:0.0%", false, false);
+        assert_se(condition);
+        assert_se(condition_test(condition, environ) >= 0);
+        condition_free(condition);
+}
+
 DEFINE_TEST_MAIN(LOG_DEBUG);

test/fuzz/fuzz-unit-file/directives-all.service

@@ -12,14 +12,17 @@ AssertACPower=
 AssertArchitecture=
 AssertCapability=
 AssertControlGroupController=
+AssertCPUPressure=
 AssertDirectoryNotEmpty=
 AssertFileIsExecutable=
 AssertFileNotEmpty=
 AssertFirstBoot=
 AssertGroup=
 AssertHost=
+AssertIOPressure=
 AssertKernelCommandLine=
 AssertKernelVersion=
+AssertMemoryPressure=
 AssertNeedsUpdate=
 AssertOSRelease=
 AssertPathExists=
@@ -56,14 +59,17 @@ ConditionACPower=
 ConditionArchitecture=
 ConditionCapability=
 ConditionControlGroupController=
+ConditionCPUPressure=
 ConditionDirectoryNotEmpty=
 ConditionFileIsExecutable=
 ConditionFileNotEmpty=
 ConditionFirstBoot=
 ConditionGroup=
 ConditionHost=
+ConditionIOPressure=
 ConditionKernelCommandLine=
 ConditionKernelVersion=
+ConditionMemoryPressure=
 ConditionNeedsUpdate=
 ConditionOSRelease=
 ConditionPathExists=

test/fuzz/fuzz-unit-file/directives.service

@@ -4,6 +4,7 @@ After=
 AllowIsolate=
 AssertACPower=
 AssertArchitecture=
+AssertCPUPressure=
 AssertCPUs=
 AssertCapability=
 AssertControlGroupController=
@@ -14,9 +15,11 @@ AssertFileNotEmpty=
 AssertFirstBoot=
 AssertGroup=
 AssertHost=
+AssertIOPressure=
 AssertKernelCommandLine=
 AssertKernelVersion=
 AssertMemory=
+AssertMemoryPressure=
 AssertNeedsUpdate=
 AssertOSRelease=
 AssertPathExists=
@@ -37,6 +40,7 @@ BindsTo=
 CollectMode=
 ConditionACPower=
 ConditionArchitecture=
+ConditionCPUPressure=
 ConditionCPUs=
 ConditionFirmware=
 ConditionCapability=
@@ -48,9 +52,11 @@ ConditionFileNotEmpty=
 ConditionFirstBoot=
 ConditionGroup=
 ConditionHost=
+ConditionIOPressure=
 ConditionKernelCommandLine=
 ConditionKernelVersion=
 ConditionMemory=
+ConditionMemoryPressure=
 ConditionNeedsUpdate=
 ConditionOSRelease=
 ConditionPathExists=