libct/cg: support hugetlb rsvd

This adds support for hugetlb.<pagesize>.rsvd limiting and accounting.

The previous non-rsvd max/limit_in_bytes does not account for reserved
huge page memory, making it possible for a processes to reserve all the
huge page memory, without being able to allocate it (due to cgroup
restrictions).

In practice this makes it possible to successfully mmap more huge page
memory than allowed via the cgroup settings, but when using the memory
the process will get a SIGBUS and crash. This is bad for applications
trying to mmap at startup (and it succeeds), but the program crashes
when starting to use the memory. eg. postgres is doing this by default.

This also keeps writing to the old max/limit_in_bytes, for backward
compatibility.

More info can be found here: https://lkml.org/lkml/2020/2/3/1153

(commit message mostly written by Odin Ugedal)

[1.1 backport: check for CGROUP_UNIFIED in integration test]

Co-authored-by: Odin Ugedal <odin@ugedal.com>
(cherry picked from commit 4a7d3ae)
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

libcontainer/cgroups/fs/hugetlb.go

@@ -1,6 +1,8 @@
 package fs
 
 import (
+	"errors"
+	"os"
 	"strconv"
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
@@ -19,8 +21,23 @@ func (s *HugetlbGroup) Apply(path string, _ *configs.Resources, pid int) error {
 }
 
 func (s *HugetlbGroup) Set(path string, r *configs.Resources) error {
+	const suffix = ".limit_in_bytes"
+	skipRsvd := false
+
 	for _, hugetlb := range r.HugetlbLimit {
-		if err := cgroups.WriteFile(path, "hugetlb."+hugetlb.Pagesize+".limit_in_bytes", strconv.FormatUint(hugetlb.Limit, 10)); err != nil {
+		prefix := "hugetlb." + hugetlb.Pagesize
+		val := strconv.FormatUint(hugetlb.Limit, 10)
+		if err := cgroups.WriteFile(path, prefix+suffix, val); err != nil {
+			return err
+		}
+		if skipRsvd {
+			continue
+		}
+		if err := cgroups.WriteFile(path, prefix+".rsvd"+suffix, val); err != nil {
+			if errors.Is(err, os.ErrNotExist) {
+				skipRsvd = true
+				continue
+			}
 			return err
 		}
 	}
@@ -32,24 +49,29 @@ func (s *HugetlbGroup) GetStats(path string, stats *cgroups.Stats) error {
 	if !cgroups.PathExists(path) {
 		return nil
 	}
+	rsvd := ".rsvd"
 	hugetlbStats := cgroups.HugetlbStats{}
 	for _, pageSize := range cgroups.HugePageSizes() {
-		usage := "hugetlb." + pageSize + ".usage_in_bytes"
-		value, err := fscommon.GetCgroupParamUint(path, usage)
+	again:
+		prefix := "hugetlb." + pageSize + rsvd
+
+		value, err := fscommon.GetCgroupParamUint(path, prefix+".usage_in_bytes")
 		if err != nil {
+			if rsvd != "" && errors.Is(err, os.ErrNotExist) {
+				rsvd = ""
+				goto again
+			}
 			return err
 		}
 		hugetlbStats.Usage = value
 
-		maxUsage := "hugetlb." + pageSize + ".max_usage_in_bytes"
-		value, err = fscommon.GetCgroupParamUint(path, maxUsage)
+		value, err = fscommon.GetCgroupParamUint(path, prefix+".max_usage_in_bytes")
 		if err != nil {
 			return err
 		}
 		hugetlbStats.MaxUsage = value
 
-		failcnt := "hugetlb." + pageSize + ".failcnt"
-		value, err = fscommon.GetCgroupParamUint(path, failcnt)
+		value, err = fscommon.GetCgroupParamUint(path, prefix+".failcnt")
 		if err != nil {
 			return err
 		}

libcontainer/cgroups/fs/hugetlb_test.go

@@ -21,6 +21,11 @@ const (
 	limit    = "hugetlb.%s.limit_in_bytes"
 	maxUsage = "hugetlb.%s.max_usage_in_bytes"
 	failcnt  = "hugetlb.%s.failcnt"
+
+	rsvdUsage    = "hugetlb.%s.rsvd.usage_in_bytes"
+	rsvdLimit    = "hugetlb.%s.rsvd.limit_in_bytes"
+	rsvdMaxUsage = "hugetlb.%s.rsvd.max_usage_in_bytes"
+	rsvdFailcnt  = "hugetlb.%s.rsvd.failcnt"
 )
 
 func TestHugetlbSetHugetlb(t *testing.T) {
@@ -52,13 +57,15 @@ func TestHugetlbSetHugetlb(t *testing.T) {
 	}
 
 	for _, pageSize := range cgroups.HugePageSizes() {
-		limit := fmt.Sprintf(limit, pageSize)
-		value, err := fscommon.GetCgroupParamUint(path, limit)
-		if err != nil {
-			t.Fatal(err)
-		}
-		if value != hugetlbAfter {
-			t.Fatalf("Set hugetlb.limit_in_bytes failed. Expected: %v, Got: %v", hugetlbAfter, value)
+		for _, f := range []string{limit, rsvdLimit} {
+			limit := fmt.Sprintf(f, pageSize)
+			value, err := fscommon.GetCgroupParamUint(path, limit)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if value != hugetlbAfter {
+				t.Fatalf("Set %s failed. Expected: %v, Got: %v", limit, hugetlbAfter, value)
+			}
 		}
 	}
 }
@@ -85,6 +92,28 @@ func TestHugetlbStats(t *testing.T) {
 	}
 }
 
+func TestHugetlbRStatsRsvd(t *testing.T) {
+	path := tempDir(t, "hugetlb")
+	for _, pageSize := range cgroups.HugePageSizes() {
+		writeFileContents(t, path, map[string]string{
+			fmt.Sprintf(rsvdUsage, pageSize):    hugetlbUsageContents,
+			fmt.Sprintf(rsvdMaxUsage, pageSize): hugetlbMaxUsageContents,
+			fmt.Sprintf(rsvdFailcnt, pageSize):  hugetlbFailcnt,
+		})
+	}
+
+	hugetlb := &HugetlbGroup{}
+	actualStats := *cgroups.NewStats()
+	err := hugetlb.GetStats(path, &actualStats)
+	if err != nil {
+		t.Fatal(err)
+	}
+	expectedStats := cgroups.HugetlbStats{Usage: 128, MaxUsage: 256, Failcnt: 100}
+	for _, pageSize := range cgroups.HugePageSizes() {
+		expectHugetlbStatEquals(t, expectedStats, actualStats.HugetlbStats[pageSize])
+	}
+}
+
 func TestHugetlbStatsNoUsageFile(t *testing.T) {
 	path := tempDir(t, "hugetlb")
 	writeFileContents(t, path, map[string]string{

libcontainer/cgroups/fs2/hugetlb.go

@@ -1,6 +1,8 @@
 package fs2
 
 import (
+	"errors"
+	"os"
 	"strconv"
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
@@ -16,8 +18,22 @@ func setHugeTlb(dirPath string, r *configs.Resources) error {
 	if !isHugeTlbSet(r) {
 		return nil
 	}
+	const suffix = ".max"
+	skipRsvd := false
 	for _, hugetlb := range r.HugetlbLimit {
-		if err := cgroups.WriteFile(dirPath, "hugetlb."+hugetlb.Pagesize+".max", strconv.FormatUint(hugetlb.Limit, 10)); err != nil {
+		prefix := "hugetlb." + hugetlb.Pagesize
+		val := strconv.FormatUint(hugetlb.Limit, 10)
+		if err := cgroups.WriteFile(dirPath, prefix+suffix, val); err != nil {
+			return err
+		}
+		if skipRsvd {
+			continue
+		}
+		if err := cgroups.WriteFile(dirPath, prefix+".rsvd"+suffix, val); err != nil {
+			if errors.Is(err, os.ErrNotExist) {
+				skipRsvd = true
+				continue
+			}
 			return err
 		}
 	}
@@ -27,15 +43,21 @@ func setHugeTlb(dirPath string, r *configs.Resources) error {
 
 func statHugeTlb(dirPath string, stats *cgroups.Stats) error {
 	hugetlbStats := cgroups.HugetlbStats{}
+	rsvd := ".rsvd"
 	for _, pagesize := range cgroups.HugePageSizes() {
-		value, err := fscommon.GetCgroupParamUint(dirPath, "hugetlb."+pagesize+".current")
+	again:
+		prefix := "hugetlb." + pagesize + rsvd
+		value, err := fscommon.GetCgroupParamUint(dirPath, prefix+".current")
 		if err != nil {
+			if rsvd != "" && errors.Is(err, os.ErrNotExist) {
+				rsvd = ""
+				goto again
+			}
 			return err
 		}
 		hugetlbStats.Usage = value
 
-		fileName := "hugetlb." + pagesize + ".events"
-		value, err = fscommon.GetValueByKey(dirPath, fileName, "max")
+		value, err = fscommon.GetValueByKey(dirPath, prefix+".events", "max")
 		if err != nil {
 			return err
 		}

tests/integration/cgroups.bats

@@ -187,6 +187,70 @@ function setup() {
 	[[ "$weights" == *"$major:$minor 444"* ]]
 }
 
+# Convert size in KB to hugetlb size suffix.
+convert_hugetlb_size() {
+	local size=$1
+	local units=("KB" "MB" "GB")
+	local idx=0
+
+	while ((size >= 1024)); do
+		((size /= 1024))
+		((idx++))
+	done
+
+	echo "$size${units[$idx]}"
+}
+
+@test "runc run (hugetlb limits)" {
+	requires cgroups_hugetlb
+	[ $EUID -ne 0 ] && requires rootless_cgroup
+	# shellcheck disable=SC2012 # ls is fine here.
+	mapfile -t sizes_kb < <(ls /sys/kernel/mm/hugepages/ | sed -e 's/.*hugepages-//' -e 's/kB$//') #
+	if [ "${#sizes_kb[@]}" -lt 1 ]; then
+		skip "requires hugetlb"
+	fi
+
+	# Create two arrays:
+	#  - sizes: hugetlb cgroup file suffixes;
+	#  - limits: limits for each size.
+	for size in "${sizes_kb[@]}"; do
+		sizes+=("$(convert_hugetlb_size "$size")")
+		# Limit to 1 page.
+		limits+=("$((size * 1024))")
+	done
+
+	# Set per-size limits.
+	for ((i = 0; i < ${#sizes[@]}; i++)); do
+		size="${sizes[$i]}"
+		limit="${limits[$i]}"
+		update_config '.linux.resources.hugepageLimits += [{ pagesize: "'"$size"'", limit: '"$limit"' }]'
+	done
+
+	set_cgroups_path
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_hugetlb
+	[ "$status" -eq 0 ]
+
+	lim="max"
+	[ "$CGROUP_UNIFIED" = "no" ] && lim="limit_in_bytes"
+
+	optional=("")
+	# Add rsvd, if available.
+	if test -f "$(get_cgroup_path hugetlb)/hugetlb.${sizes[0]}.rsvd.$lim"; then
+		optional+=(".rsvd")
+	fi
+
+	# Check if the limits are as expected.
+	for ((i = 0; i < ${#sizes[@]}; i++)); do
+		size="${sizes[$i]}"
+		limit="${limits[$i]}"
+		for rsvd in "${optional[@]}"; do
+			param="hugetlb.${size}${rsvd}.$lim"
+			echo "checking $param"
+			check_cgroup_value "$param" "$limit"
+		done
+	done
+}
+
 @test "runc run (cgroup v2 resources.unified only)" {
 	requires root cgroups_v2