podman cgroup enhancement

currently, setting any sort of resource limit in a pod does nothing. With the newly refactored creation process in c/common, podman ca now set resources at a pod level
meaning that resource related flags can now be exposed to podman pod create.

cgroupfs and systemd are both supported with varying completion. cgroupfs is a much simpler process and one that is virtually complete for all resource types, the flags now just need to be added. systemd on the other hand
has to be handeled via the dbus api meaning that the limits need to be passed as recognized properties to systemd. The properties added so far are the ones that podman pod create supports as well as `cpuset-mems` as this will
be the next flag I work on.

Signed-off-by: Charlie Doern <cdoern@redhat.com>

2 files changed, 266 insertions, 0 deletions

diff --git a/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/rdma.go b/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/rdma.go
new file mode 100644
index 000000000..d463d15ee
--- /dev/null
+++ b/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/rdma.go
@@ -0,0 +1,121 @@
+package fscommon
+
+import (
+	"bufio"
+	"errors"
+	"math"
+	"os"
+	"strconv"
+	"strings"
+
+	"github.com/opencontainers/runc/libcontainer/cgroups"
+	"github.com/opencontainers/runc/libcontainer/configs"
+	"golang.org/x/sys/unix"
+)
+
+// parseRdmaKV parses raw string to RdmaEntry.
+func parseRdmaKV(raw string, entry *cgroups.RdmaEntry) error {
+	var value uint32
+
+	parts := strings.SplitN(raw, "=", 3)
+
+	if len(parts) != 2 {
+		return errors.New("Unable to parse RDMA entry")
+	}
+
+	k, v := parts[0], parts[1]
+
+	if v == "max" {
+		value = math.MaxUint32
+	} else {
+		val64, err := strconv.ParseUint(v, 10, 32)
+		if err != nil {
+			return err
+		}
+		value = uint32(val64)
+	}
+	if k == "hca_handle" {
+		entry.HcaHandles = value
+	} else if k == "hca_object" {
+		entry.HcaObjects = value
+	}
+
+	return nil
+}
+
+// readRdmaEntries reads and converts array of rawstrings to RdmaEntries from file.
+// example entry: mlx4_0 hca_handle=2 hca_object=2000
+func readRdmaEntries(dir, file string) ([]cgroups.RdmaEntry, error) {
+	rdmaEntries := make([]cgroups.RdmaEntry, 0)
+	fd, err := cgroups.OpenFile(dir, file, unix.O_RDONLY)
+	if err != nil {
+		return nil, err
+	}
+	defer fd.Close() //nolint:errorlint
+	scanner := bufio.NewScanner(fd)
+	for scanner.Scan() {
+		parts := strings.SplitN(scanner.Text(), " ", 4)
+		if len(parts) == 3 {
+			entry := new(cgroups.RdmaEntry)
+			entry.Device = parts[0]
+			err = parseRdmaKV(parts[1], entry)
+			if err != nil {
+				continue
+			}
+			err = parseRdmaKV(parts[2], entry)
+			if err != nil {
+				continue
+			}
+
+			rdmaEntries = append(rdmaEntries, *entry)
+		}
+	}
+	return rdmaEntries, scanner.Err()
+}
+
+// RdmaGetStats returns rdma stats such as totalLimit and current entries.
+func RdmaGetStats(path string, stats *cgroups.Stats) error {
+	currentEntries, err := readRdmaEntries(path, "rdma.current")
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			err = nil
+		}
+		return err
+	}
+	maxEntries, err := readRdmaEntries(path, "rdma.max")
+	if err != nil {
+		return err
+	}
+	// If device got removed between reading two files, ignore returning stats.
+	if len(currentEntries) != len(maxEntries) {
+		return nil
+	}
+
+	stats.RdmaStats = cgroups.RdmaStats{
+		RdmaLimit:   maxEntries,
+		RdmaCurrent: currentEntries,
+	}
+
+	return nil
+}
+
+func createCmdString(device string, limits configs.LinuxRdma) string {
+	cmdString := device
+	if limits.HcaHandles != nil {
+		cmdString += " hca_handle=" + strconv.FormatUint(uint64(*limits.HcaHandles), 10)
+	}
+	if limits.HcaObjects != nil {
+		cmdString += " hca_object=" + strconv.FormatUint(uint64(*limits.HcaObjects), 10)
+	}
+	return cmdString
+}
+
+// RdmaSet sets RDMA resources.
+func RdmaSet(path string, r *configs.Resources) error {
+	for device, limits := range r.Rdma {
+		if err := cgroups.WriteFile(path, "rdma.max", createCmdString(device, limits)); err != nil {
+			return err
+		}
+	}
+	return nil
+}
diff --git a/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/utils.go b/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/utils.go
new file mode 100644
index 000000000..f4a51c9e5
--- /dev/null
+++ b/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/utils.go
@@ -0,0 +1,145 @@
+package fscommon
+
+import (
+	"errors"
+	"fmt"
+	"math"
+	"path"
+	"strconv"
+	"strings"
+
+	"github.com/opencontainers/runc/libcontainer/cgroups"
+)
+
+var (
+	// Deprecated: use cgroups.OpenFile instead.
+	OpenFile = cgroups.OpenFile
+	// Deprecated: use cgroups.ReadFile instead.
+	ReadFile = cgroups.ReadFile
+	// Deprecated: use cgroups.WriteFile instead.
+	WriteFile = cgroups.WriteFile
+)
+
+// ParseError records a parse error details, including the file path.
+type ParseError struct {
+	Path string
+	File string
+	Err  error
+}
+
+func (e *ParseError) Error() string {
+	return "unable to parse " + path.Join(e.Path, e.File) + ": " + e.Err.Error()
+}
+
+func (e *ParseError) Unwrap() error { return e.Err }
+
+// ParseUint converts a string to an uint64 integer.
+// Negative values are returned at zero as, due to kernel bugs,
+// some of the memory cgroup stats can be negative.
+func ParseUint(s string, base, bitSize int) (uint64, error) {
+	value, err := strconv.ParseUint(s, base, bitSize)
+	if err != nil {
+		intValue, intErr := strconv.ParseInt(s, base, bitSize)
+		// 1. Handle negative values greater than MinInt64 (and)
+		// 2. Handle negative values lesser than MinInt64
+		if intErr == nil && intValue < 0 {
+			return 0, nil
+		} else if errors.Is(intErr, strconv.ErrRange) && intValue < 0 {
+			return 0, nil
+		}
+
+		return value, err
+	}
+
+	return value, nil
+}
+
+// ParseKeyValue parses a space-separated "name value" kind of cgroup
+// parameter and returns its key as a string, and its value as uint64
+// (ParseUint is used to convert the value). For example,
+// "io_service_bytes 1234" will be returned as "io_service_bytes", 1234.
+func ParseKeyValue(t string) (string, uint64, error) {
+	parts := strings.SplitN(t, " ", 3)
+	if len(parts) != 2 {
+		return "", 0, fmt.Errorf("line %q is not in key value format", t)
+	}
+
+	value, err := ParseUint(parts[1], 10, 64)
+	if err != nil {
+		return "", 0, err
+	}
+
+	return parts[0], value, nil
+}
+
+// GetValueByKey reads a key-value pairs from the specified cgroup file,
+// and returns a value of the specified key. ParseUint is used for value
+// conversion.
+func GetValueByKey(path, file, key string) (uint64, error) {
+	content, err := cgroups.ReadFile(path, file)
+	if err != nil {
+		return 0, err
+	}
+
+	lines := strings.Split(content, "\n")
+	for _, line := range lines {
+		arr := strings.Split(line, " ")
+		if len(arr) == 2 && arr[0] == key {
+			val, err := ParseUint(arr[1], 10, 64)
+			if err != nil {
+				err = &ParseError{Path: path, File: file, Err: err}
+			}
+			return val, err
+		}
+	}
+
+	return 0, nil
+}
+
+// GetCgroupParamUint reads a single uint64 value from the specified cgroup file.
+// If the value read is "max", the math.MaxUint64 is returned.
+func GetCgroupParamUint(path, file string) (uint64, error) {
+	contents, err := GetCgroupParamString(path, file)
+	if err != nil {
+		return 0, err
+	}
+	contents = strings.TrimSpace(contents)
+	if contents == "max" {
+		return math.MaxUint64, nil
+	}
+
+	res, err := ParseUint(contents, 10, 64)
+	if err != nil {
+		return res, &ParseError{Path: path, File: file, Err: err}
+	}
+	return res, nil
+}
+
+// GetCgroupParamInt reads a single int64 value from specified cgroup file.
+// If the value read is "max", the math.MaxInt64 is returned.
+func GetCgroupParamInt(path, file string) (int64, error) {
+	contents, err := cgroups.ReadFile(path, file)
+	if err != nil {
+		return 0, err
+	}
+	contents = strings.TrimSpace(contents)
+	if contents == "max" {
+		return math.MaxInt64, nil
+	}
+
+	res, err := strconv.ParseInt(contents, 10, 64)
+	if err != nil {
+		return res, &ParseError{Path: path, File: file, Err: err}
+	}
+	return res, nil
+}
+
+// GetCgroupParamString reads a string from the specified cgroup file.
+func GetCgroupParamString(path, file string) (string, error) {
+	contents, err := cgroups.ReadFile(path, file)
+	if err != nil {
+		return "", err
+	}
+
+	return strings.TrimSpace(contents), nil
+}