From 7ec6e537898e139cc33017e03465ef40a86dd433 Mon Sep 17 00:00:00 2001

From: Lennart Poettering <lennart@poettering.net>

Date: Tue, 19 Jul 2016 15:58:49 +0200

Subject: [PATCH] core: support percentage specifications on TasksMax=

This adds support for a TasksMax=40% syntax for specifying values relative to

the system's configured maximum number of processes. This is useful in order to

neatly subdivide the available room for tasks within containers.

Cherry-picked from: 83f8e80857090f63cf6a02c54d381dad3c0fad55

Related: #1337244

---

 man/systemd.resource-control.xml | 15 +++----

 src/core/dbus-cgroup.c           | 22 +++++++++++

 src/core/load-fragment.c         | 15 +++++--

 src/libsystemd/sd-bus/bus-util.c | 19 +++++++++

 src/shared/util.c                | 84 ++++++++++++++++++++++++++++++++++++++++

 src/shared/util.h                |  5 +++

 src/test/test-util.c             | 39 +++++++++++++++++++

 7 files changed, 187 insertions(+), 12 deletions(-)

diff --git a/man/systemd.resource-control.xml b/man/systemd.resource-control.xml

index 217105e..f507c67 100644

--- a/man/systemd.resource-control.xml

+++ b/man/systemd.resource-control.xml

@@ -221,15 +221,12 @@

         <term><varname>TasksMax=<replaceable>N</replaceable></varname></term>

         <listitem>

-          <para>Specify the maximum number of tasks that may be

-          created in the unit. This ensures that the number of tasks

-          accounted for the unit (see above) stays below a specific

-          limit. If assigned the special value

-          <literal>infinity</literal> no tasks limit is applied. This

-          controls the <literal>pids.max</literal> control group

-          attribute. For details about this control group attribute,

-          see 

-          url="https://www.kernel.org/doc/Documentation/cgroups/pids.txt">pids.txt</ulink>.</para>

+          <para>Specify the maximum number of tasks that may be created in the unit. This ensures that the number of

+          tasks accounted for the unit (see above) stays below a specific limit. This either takes an absolute number

+          of tasks or a percentage value that is taken relative to the configured maximum number of tasks on the

+          system.  If assigned the special value <literal>infinity</literal>, no tasks limit is applied. This controls

+          the <literal>pids.max</literal> control group attribute. For details about this control group attribute, see

+          <ulink url="https://www.kernel.org/doc/Documentation/cgroup-v1/pids.txt">pids.txt</ulink>.</para>

           <para>Implies <literal>TasksAccounting=true</literal>. The

           system default for this setting may be controlled with

diff --git a/src/core/dbus-cgroup.c b/src/core/dbus-cgroup.c

index a4465dc..fa76c60 100644

--- a/src/core/dbus-cgroup.c

+++ b/src/core/dbus-cgroup.c

@@ -694,6 +694,8 @@ int bus_cgroup_set_property(

                 r = sd_bus_message_read(message, "t", &limit);

                 if (r < 0)

                         return r;

+                if (limit <= 0)

+                        return sd_bus_error_set_errnof(error, EINVAL, "%s= is too small", name);

                 if (mode != UNIT_CHECK) {

                         c->tasks_max = limit;

@@ -706,6 +708,26 @@ int bus_cgroup_set_property(

                 }

                 return 1;

+        } else if (streq(name, "TasksMaxScale")) {

+                uint64_t limit;

+                uint32_t raw;

+

+                r = sd_bus_message_read(message, "u", &raw;;

+                if (r < 0)

+                        return r;

+

+                limit = system_tasks_max_scale(raw, UINT32_MAX);

+                if (limit <= 0 || limit >= UINT64_MAX)

+                        return sd_bus_error_set_errnof(error, EINVAL, "%s= is out of range", name);

+

+                if (mode != UNIT_CHECK) {

+                        c->tasks_max = limit;

+                        u->cgroup_realized_mask &= ~CGROUP_PIDS;

+                        unit_write_drop_in_private_format(u, mode, name, "TasksMax=%" PRIu32 "%%",

+                                                          (uint32_t) (DIV_ROUND_UP((uint64_t) raw * 100U, (uint64_t) UINT32_MAX)));

+                }

+

+                return 1;

         }

         if (u->transient && u->load_state == UNIT_STUB) {

diff --git a/src/core/load-fragment.c b/src/core/load-fragment.c

index c1ffee2..4114750 100644

--- a/src/core/load-fragment.c

+++ b/src/core/load-fragment.c

@@ -3089,9 +3089,18 @@ int config_parse_tasks_max(

                 return 0;

         }

-        r = safe_atou64(rvalue, &u);

-        if (r < 0 || u < 1) {

-                log_syntax(unit, LOG_ERR, filename, line, EINVAL, "Maximum tasks value '%s' invalid. Ignoring.", rvalue);

+        r = parse_percent(rvalue);

+        if (r < 0) {

+                r = safe_atou64(rvalue, &u);

+                if (r < 0) {

+                        log_syntax(unit, LOG_ERR, filename, line, r, "Maximum tasks value '%s' invalid. Ignoring.", rvalue);

+                        return 0;

+                }

+        } else

+                u = system_tasks_max_scale(r, 100U);

+

+        if (u <= 0 || u >= UINT64_MAX) {

+                log_syntax(unit, LOG_ERR, filename, line, 0, "Maximum tasks value '%s' out of range. Ignoring.", rvalue);

                 return 0;

         }

diff --git a/src/libsystemd/sd-bus/bus-util.c b/src/libsystemd/sd-bus/bus-util.c

index ed0849b..f46fa2b 100644

--- a/src/libsystemd/sd-bus/bus-util.c

+++ b/src/libsystemd/sd-bus/bus-util.c

@@ -1409,7 +1409,26 @@ int bus_append_unit_property_assignment(sd_bus_message *m, const char *assignmen

                 }

                 r = sd_bus_message_append(m, "v", "t", (uint64_t) bytes);

+        } else if (streq(field, "TasksMax")) {

+                uint64_t t;

+

+                if (isempty(eq) || streq(eq, "infinity"))

+                        t = (uint64_t) -1;

+                else {

+                        r = parse_percent(eq);

+                        if (r >= 0) {

+                                r = sd_bus_message_append(m, "sv", "TasksMaxScale", "u", (uint32_t) (((uint64_t) UINT32_MAX * r) / 100U));

+                                if (r < 0)

+                                        return bus_log_create_error(r);

+                        } else {

+                                r = safe_atou64(eq, &t);

+                                if (r < 0)

+                                        return log_error_errno(r, "Failed to parse maximum tasks specification %s", assignment);

+                        }

+

+                }

+                r = sd_bus_message_append(m, "sv", "TasksMax", "t", t);

         } else if (STR_IN_SET(field, "CPUShares", "BlockIOWeight")) {

                 uint64_t u;

diff --git a/src/shared/util.c b/src/shared/util.c

index cadadde..bbb4577 100644

--- a/src/shared/util.c

+++ b/src/shared/util.c

@@ -94,6 +94,7 @@

 #include "def.h"

 #include "sparse-endian.h"

 #include "conf-parser.h"

+#include "cgroup-util.h"

 int saved_argc = 0;

 char **saved_argv = NULL;

@@ -8707,3 +8708,86 @@ int extract_many_words(const char **p, const char *separators, ExtractFlags flag

         return c;

 }

+

+int parse_percent_unbounded(const char *p) {

+        const char *pc, *n;

+        unsigned v;

+        int r;

+

+        pc = endswith(p, "%");

+        if (!pc)

+                return -EINVAL;

+

+        n = strndupa(p, pc - p);

+        r = safe_atou(n, &v);

+        if (r < 0)

+                return r;

+

+        return (int) v;

+}

+

+int parse_percent(const char *p) {

+        int v;

+

+        v = parse_percent_unbounded(p);

+        if (v > 100)

+                return -ERANGE;

+

+        return v;

+}

+

+uint64_t system_tasks_max(void) {

+

+#if SIZEOF_PID_T == 4

+#define TASKS_MAX ((uint64_t) (INT32_MAX-1))

+#elif SIZEOF_PID_T == 2

+#define TASKS_MAX ((uint64_t) (INT16_MAX-1))

+#else

+#error "Unknown pid_t size"

+#endif

+

+        _cleanup_free_ char *value = NULL, *root = NULL;

+        uint64_t a = TASKS_MAX, b = TASKS_MAX;

+

+        /* Determine the maximum number of tasks that may run on this system. We check three sources to determine this

+         * limit:

+         *

+         * a) the maximum value for the pid_t type

+         * b) the cgroups pids_max attribute for the system

+         * c) the kernel's configure maximum PID value

+         *

+         * And then pick the smallest of the three */

+

+        if (read_one_line_file("/proc/sys/kernel/pid_max", &value) >= 0)

+                (void) safe_atou64(value, &a);

+

+        if (cg_get_root_path(&root) >= 0) {

+                free(value);

+                value = NULL;

+

+                if (cg_get_attribute("pids", root, "pids.max", &value) >= 0)

+                        (void) safe_atou64(value, &b);

+        }

+

+        return MIN3(TASKS_MAX,

+                    a <= 0 ? TASKS_MAX : a,

+                    b <= 0 ? TASKS_MAX : b);

+}

+

+uint64_t system_tasks_max_scale(uint64_t v, uint64_t max) {

+        uint64_t t, m;

+

+        assert(max > 0);

+

+        /* Multiply the system's task value by the fraction v/max. Hence, if max==100 this calculates percentages

+         * relative to the system's maximum number of tasks. Returns UINT64_MAX on overflow. */

+

+        t = system_tasks_max();

+        assert(t > 0);

+

+        m = t * v;

+        if (m / t != v) /* overflow? */

+                return UINT64_MAX;

+

+        return m / max;

+}

diff --git a/src/shared/util.h b/src/shared/util.h

index 12afcc3..f1b6c34 100644

--- a/src/shared/util.h

+++ b/src/shared/util.h

@@ -1098,3 +1098,8 @@ typedef enum ExtractFlags {

 int extract_first_word(const char **p, char **ret, const char *separators, ExtractFlags flags);

 int extract_first_word_and_warn(const char **p, char **ret, const char *separators, ExtractFlags flags, const char *unit, const char *filename, unsigned line, const char *rvalue);

 int extract_many_words(const char **p, const char *separators, ExtractFlags flags, ...) _sentinel_;

+int parse_percent_unbounded(const char *p);

+int parse_percent(const char *p);

+

+uint64_t system_tasks_max(void);

+uint64_t system_tasks_max_scale(uint64_t v, uint64_t max);

diff --git a/src/test/test-util.c b/src/test/test-util.c

index 9ae347b..971f97d 100644

--- a/src/test/test-util.c

+++ b/src/test/test-util.c

@@ -1530,6 +1530,43 @@ static void test_shell_maybe_quote(void) {

         test_shell_maybe_quote_one("foo$bar", "\"foo\\$bar\"");

 }

+static void test_system_tasks_max(void) {

+        uint64_t t;

+

+        t = system_tasks_max();

+        assert_se(t > 0);

+        assert_se(t < UINT64_MAX);

+

+        log_info("Max tasks: %" PRIu64, t);

+}

+

+static void test_system_tasks_max_scale(void) {

+        uint64_t t;

+

+        t = system_tasks_max();

+

+        assert_se(system_tasks_max_scale(0, 100) == 0);

+        assert_se(system_tasks_max_scale(100, 100) == t);

+

+        assert_se(system_tasks_max_scale(0, 1) == 0);

+        assert_se(system_tasks_max_scale(1, 1) == t);

+        assert_se(system_tasks_max_scale(2, 1) == 2*t);

+

+        assert_se(system_tasks_max_scale(0, 2) == 0);

+        assert_se(system_tasks_max_scale(1, 2) == t/2);

+        assert_se(system_tasks_max_scale(2, 2) == t);

+        assert_se(system_tasks_max_scale(3, 2) == (3*t)/2);

+        assert_se(system_tasks_max_scale(4, 2) == t*2);

+

+        assert_se(system_tasks_max_scale(0, UINT32_MAX) == 0);

+        assert_se(system_tasks_max_scale((UINT32_MAX-1)/2, UINT32_MAX-1) == t/2);

+        assert_se(system_tasks_max_scale(UINT32_MAX, UINT32_MAX) == t);

+

+        /* overflow */

+

+        assert_se(system_tasks_max_scale(UINT64_MAX/4, UINT64_MAX) == UINT64_MAX);

+}

+

 int main(int argc, char *argv[]) {

         log_parse_environment();

         log_open();

@@ -1608,6 +1645,8 @@ int main(int argc, char *argv[]) {

         test_uid_ptr();

         test_sparse_write();

         test_shell_maybe_quote();

+        test_system_tasks_max();

+        test_system_tasks_max_scale();

         return 0;

 }