core: add MemoryAvailable unit property

Try to infer the unused memory that a unit can claim before the
memory.max limit is reached, including any limit set on any parent
slice above the unit itself.

man/org.freedesktop.systemd1.xml

@@ -2401,6 +2401,8 @@ node /org/freedesktop/systemd1/unit/avahi_2ddaemon_2eservice {
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t MemoryCurrent = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
+      readonly t MemoryAvailable = ...;
+      @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t CPUUsageNSec = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly ay EffectiveCPUs = [...];
@@ -3504,6 +3506,8 @@ node /org/freedesktop/systemd1/unit/avahi_2ddaemon_2eservice {
 
     <variablelist class="dbus-property" generated="True" extra-ref="MemoryCurrent"/>
 
+    <variablelist class="dbus-property" generated="True" extra-ref="MemoryAvailable"/>
+
     <variablelist class="dbus-property" generated="True" extra-ref="CPUUsageNSec"/>
 
     <variablelist class="dbus-property" generated="True" extra-ref="EffectiveCPUs"/>
@@ -4063,6 +4067,11 @@ node /org/freedesktop/systemd1/unit/avahi_2ddaemon_2eservice {
       <varname>MountImages</varname>
       <varname>ExtensionImages</varname>
       see systemd.exec(5) for their meaning.</para>
+
+      <para><varname>MemoryAvailable</varname> indicates how much unused memory is available to the unit before
+      the <literal>MemoryMax</literal> or <literal>MemoryHigh</literal> (whichever is lower) limit set by the cgroup
+      memory controller is reached. It will take into consideration limits on all parent slices, other than the
+      limits set on the unit itself.</para>
     </refsect2>
   </refsect1>
 
@@ -4196,6 +4205,8 @@ node /org/freedesktop/systemd1/unit/avahi_2ddaemon_2esocket {
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t MemoryCurrent = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
+      readonly t MemoryAvailable = ...;
+      @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t CPUUsageNSec = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly ay EffectiveCPUs = [...];
@@ -5321,6 +5332,8 @@ node /org/freedesktop/systemd1/unit/avahi_2ddaemon_2esocket {
 
     <variablelist class="dbus-property" generated="True" extra-ref="MemoryCurrent"/>
 
+    <variablelist class="dbus-property" generated="True" extra-ref="MemoryAvailable"/>
+
     <variablelist class="dbus-property" generated="True" extra-ref="CPUUsageNSec"/>
 
     <variablelist class="dbus-property" generated="True" extra-ref="EffectiveCPUs"/>
@@ -5915,6 +5928,8 @@ node /org/freedesktop/systemd1/unit/home_2emount {
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t MemoryCurrent = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
+      readonly t MemoryAvailable = ...;
+      @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t CPUUsageNSec = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly ay EffectiveCPUs = [...];
@@ -6886,6 +6901,8 @@ node /org/freedesktop/systemd1/unit/home_2emount {
 
     <variablelist class="dbus-property" generated="True" extra-ref="MemoryCurrent"/>
 
+    <variablelist class="dbus-property" generated="True" extra-ref="MemoryAvailable"/>
+
     <variablelist class="dbus-property" generated="True" extra-ref="CPUUsageNSec"/>
 
     <variablelist class="dbus-property" generated="True" extra-ref="EffectiveCPUs"/>
@@ -7601,6 +7618,8 @@ node /org/freedesktop/systemd1/unit/dev_2dsda3_2eswap {
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t MemoryCurrent = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
+      readonly t MemoryAvailable = ...;
+      @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t CPUUsageNSec = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly ay EffectiveCPUs = [...];
@@ -8544,6 +8563,8 @@ node /org/freedesktop/systemd1/unit/dev_2dsda3_2eswap {
 
     <variablelist class="dbus-property" generated="True" extra-ref="MemoryCurrent"/>
 
+    <variablelist class="dbus-property" generated="True" extra-ref="MemoryAvailable"/>
+
     <variablelist class="dbus-property" generated="True" extra-ref="CPUUsageNSec"/>
 
     <variablelist class="dbus-property" generated="True" extra-ref="EffectiveCPUs"/>
@@ -9112,6 +9133,8 @@ node /org/freedesktop/systemd1/unit/system_2eslice {
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t MemoryCurrent = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
+      readonly t MemoryAvailable = ...;
+      @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t CPUUsageNSec = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly ay EffectiveCPUs = [...];
@@ -9403,6 +9426,8 @@ node /org/freedesktop/systemd1/unit/system_2eslice {
 
     <variablelist class="dbus-property" generated="True" extra-ref="MemoryCurrent"/>
 
+    <variablelist class="dbus-property" generated="True" extra-ref="MemoryAvailable"/>
+
     <variablelist class="dbus-property" generated="True" extra-ref="CPUUsageNSec"/>
 
     <variablelist class="dbus-property" generated="True" extra-ref="EffectiveCPUs"/>
@@ -9571,6 +9596,8 @@ node /org/freedesktop/systemd1/unit/session_2d1_2escope {
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t MemoryCurrent = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
+      readonly t MemoryAvailable = ...;
+      @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly t CPUUsageNSec = ...;
       @org.freedesktop.DBus.Property.EmitsChangedSignal("false")
       readonly ay EffectiveCPUs = [...];
@@ -9904,6 +9931,8 @@ node /org/freedesktop/systemd1/unit/session_2d1_2escope {
 
     <variablelist class="dbus-property" generated="True" extra-ref="MemoryCurrent"/>
 
+    <variablelist class="dbus-property" generated="True" extra-ref="MemoryAvailable"/>
+
     <variablelist class="dbus-property" generated="True" extra-ref="CPUUsageNSec"/>
 
     <variablelist class="dbus-property" generated="True" extra-ref="EffectiveCPUs"/>

src/core/cgroup.c

@@ -3402,6 +3402,77 @@ int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
         return 1;
 }
 
+int unit_get_memory_available(Unit *u, uint64_t *ret) {
+        uint64_t unit_current, available = UINT64_MAX;
+        CGroupContext *unit_context;
+        const char *memory_file;
+        int r;
+
+        assert(u);
+        assert(ret);
+
+        /* If data from cgroups can be accessed, try to find out how much more memory a unit can
+         * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh
+         * and MemoryMax, and also any slice the unit might be nested below. */
+
+        if (!UNIT_CGROUP_BOOL(u, memory_accounting))
+                return -ENODATA;
+
+        if (!u->cgroup_path)
+                return -ENODATA;
+
+        /* The root cgroup doesn't expose this information */
+        if (unit_has_host_root_cgroup(u))
+                return -ENODATA;
+
+        if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
+                return -ENODATA;
+
+        r = cg_all_unified();
+        if (r < 0)
+                return r;
+        memory_file = r > 0 ? "memory.current" : "memory.usage_in_bytes";
+
+        r = cg_get_attribute_as_uint64("memory", u->cgroup_path, memory_file, &unit_current);
+        if (r < 0)
+                return r;
+
+        assert_se(unit_context = unit_get_cgroup_context(u));
+
+        if (unit_context->memory_max != UINT64_MAX || unit_context->memory_high != UINT64_MAX)
+                available = LESS_BY(MIN(unit_context->memory_max, unit_context->memory_high), unit_current);
+
+        for (Unit *slice = UNIT_GET_SLICE(u); slice; slice = UNIT_GET_SLICE(slice)) {
+                uint64_t slice_current, slice_available = UINT64_MAX;
+                CGroupContext *slice_context;
+
+                /* No point in continuing if we can't go any lower */
+                if (available == 0)
+                        break;
+
+                if (!slice->cgroup_path)
+                        continue;
+
+                slice_context = unit_get_cgroup_context(slice);
+                if (!slice_context)
+                        continue;
+
+                if (slice_context->memory_max == UINT64_MAX && slice_context->memory_high == UINT64_MAX)
+                        continue;
+
+                r = cg_get_attribute_as_uint64("memory", slice->cgroup_path, memory_file, &slice_current);
+                if (r < 0)
+                        continue;
+
+                slice_available = LESS_BY(MIN(slice_context->memory_max, slice_context->memory_high), slice_current);
+                available = MIN(slice_available, available);
+        }
+
+        *ret = available;
+
+        return 0;
+}
+
 int unit_get_memory_current(Unit *u, uint64_t *ret) {
         int r;
 

src/core/cgroup.h

@@ -282,6 +282,7 @@ int unit_watch_all_pids(Unit *u);
 int unit_synthesize_cgroup_empty_event(Unit *u);
 
 int unit_get_memory_current(Unit *u, uint64_t *ret);
+int unit_get_memory_available(Unit *u, uint64_t *ret);
 int unit_get_tasks_current(Unit *u, uint64_t *ret);
 int unit_get_cpu_usage(Unit *u, nsec_t *ret);
 int unit_get_io_accounting(Unit *u, CGroupIOAccountingMetric metric, bool allow_cache, uint64_t *ret);

src/core/dbus-unit.c

@@ -1097,6 +1097,30 @@ static int property_get_current_memory(
         return sd_bus_message_append(reply, "t", sz);
 }
 
+static int property_get_available_memory(
+                sd_bus *bus,
+                const char *path,
+                const char *interface,
+                const char *property,
+                sd_bus_message *reply,
+                void *userdata,
+                sd_bus_error *error) {
+
+        uint64_t sz = UINT64_MAX;
+        Unit *u = userdata;
+        int r;
+
+        assert(bus);
+        assert(reply);
+        assert(u);
+
+        r = unit_get_memory_available(u, &sz);
+        if (r < 0 && r != -ENODATA)
+                log_unit_warning_errno(u, r, "Failed to get total available memory from cgroup: %m");
+
+        return sd_bus_message_append(reply, "t", sz);
+}
+
 static int property_get_current_tasks(
                 sd_bus *bus,
                 const char *path,
@@ -1541,6 +1565,7 @@ const sd_bus_vtable bus_unit_cgroup_vtable[] = {
         SD_BUS_PROPERTY("Slice", "s", property_get_slice, 0, 0),
         SD_BUS_PROPERTY("ControlGroup", "s", property_get_cgroup, 0, 0),
         SD_BUS_PROPERTY("MemoryCurrent", "t", property_get_current_memory, 0, 0),
+        SD_BUS_PROPERTY("MemoryAvailable", "t", property_get_available_memory, 0, 0),
         SD_BUS_PROPERTY("CPUUsageNSec", "t", property_get_cpu_usage, 0, 0),
         SD_BUS_PROPERTY("EffectiveCPUs", "ay", property_get_cpuset_cpus, 0, 0),
         SD_BUS_PROPERTY("EffectiveMemoryNodes", "ay", property_get_cpuset_mems, 0, 0),

src/shared/bus-print-properties.c

@@ -165,7 +165,7 @@ static int bus_print_property(const char *name, const char *expected_value, sd_b
 
                         bus_print_property_value(name, expected_value, flags, "[not set]");
 
-                else if ((STR_IN_SET(name, "DefaultMemoryLow", "DefaultMemoryMin", "MemoryLow", "MemoryHigh", "MemoryMax", "MemorySwapMax", "MemoryLimit") && u == CGROUP_LIMIT_MAX) ||
+                else if ((STR_IN_SET(name, "DefaultMemoryLow", "DefaultMemoryMin", "MemoryLow", "MemoryHigh", "MemoryMax", "MemorySwapMax", "MemoryLimit", "MemoryAvailable") && u == CGROUP_LIMIT_MAX) ||
                          (STR_IN_SET(name, "TasksMax", "DefaultTasksMax") && u == UINT64_MAX) ||
                          (startswith(name, "Limit") && u == UINT64_MAX) ||
                          (startswith(name, "DefaultLimit") && u == UINT64_MAX))

src/systemctl/systemctl-show.c

@@ -247,6 +247,7 @@ typedef struct UnitStatusInfo {
         uint64_t memory_max;
         uint64_t memory_swap_max;
         uint64_t memory_limit;
+        uint64_t memory_available;
         uint64_t cpu_usage_nsec;
         uint64_t tasks_current;
         uint64_t tasks_max;
@@ -682,6 +683,7 @@ static void print_status_info(
                 if (i->memory_min > 0 || i->memory_low > 0 ||
                     i->memory_high != CGROUP_LIMIT_MAX || i->memory_max != CGROUP_LIMIT_MAX ||
                     i->memory_swap_max != CGROUP_LIMIT_MAX ||
+                    i->memory_available != CGROUP_LIMIT_MAX ||
                     i->memory_limit != CGROUP_LIMIT_MAX) {
                         const char *prefix = "";
 
@@ -710,6 +712,10 @@ static void print_status_info(
                                 printf("%slimit: %s", prefix, format_bytes(buf, sizeof(buf), i->memory_limit));
                                 prefix = " ";
                         }
+                        if (i->memory_available != CGROUP_LIMIT_MAX) {
+                                printf("%savailable: %s", prefix, format_bytes(buf, sizeof(buf), i->memory_available));
+                                prefix = " ";
+                        }
                         printf(")");
                 }
                 printf("\n");
@@ -1827,6 +1833,7 @@ static int show_one(
                 { "Where",                          "s",               NULL,           offsetof(UnitStatusInfo, where)                             },
                 { "What",                           "s",               NULL,           offsetof(UnitStatusInfo, what)                              },
                 { "MemoryCurrent",                  "t",               NULL,           offsetof(UnitStatusInfo, memory_current)                    },
+                { "MemoryAvailable",                "t",               NULL,           offsetof(UnitStatusInfo, memory_available)                  },
                 { "DefaultMemoryMin",               "t",               NULL,           offsetof(UnitStatusInfo, default_memory_min)                },
                 { "DefaultMemoryLow",               "t",               NULL,           offsetof(UnitStatusInfo, default_memory_low)                },
                 { "MemoryMin",                      "t",               NULL,           offsetof(UnitStatusInfo, memory_min)                        },
@@ -1869,6 +1876,7 @@ static int show_one(
                 .memory_max = CGROUP_LIMIT_MAX,
                 .memory_swap_max = CGROUP_LIMIT_MAX,
                 .memory_limit = UINT64_MAX,
+                .memory_available = CGROUP_LIMIT_MAX,
                 .cpu_usage_nsec = UINT64_MAX,
                 .tasks_current = UINT64_MAX,
                 .tasks_max = UINT64_MAX,