Skip to content

[btrfs] support dynamic_reclaim #2164

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
k8s-ci-robot merged 5 commits into from
Sep 9, 2025
Merged

[btrfs] support dynamic_reclaim #2164

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
dbaa4ae
[btrfs] support dynamic_reclaim
motiejus Sep 4, 2025
57f1758
convert kernel settings to a table, patch up a few things.
motiejus Sep 9, 2025
f5dd219
add metadata/dynamic_reclaim
motiejus Sep 9, 2025
47f3fd6
heading wording
motiejus Sep 9, 2025
427e119
more heading wording
motiejus Sep 9, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 7 additions & 3 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,9 +100,13 @@ As noted in [GCP PD documentation](https://cloud.google.com/kubernetes-engine/do

`btrfs` filesystem accepts the following "special" mount options and the sysfs paths they target:

- `btrfs-data-bg_reclaim_threshold`: `/sys/fs/btrfs/FS-UUID/allocation/data/bg_reclaim_threshold`.
- `btrfs-metadata-bg_reclaim_thresho: `/sys/fs/btrfs/FS-UUID/allocation/metadata/bg_reclaim_threshold`.
- `btrfs-bdi-read_ahead_kb`: `/sys/fs/btrfs/FS-UUID/bdi/read_ahead_kb`.
| Setting | Sysfs path | Value | Default | Supported on Linux versions | Notes |
|--------------------------------------------------|-------------------------------------------------------------------|----------------|-------------------------|-----------------------------|-------|
| `btrfs-allocation-data-bg_reclaim_threshold` | `/sys/fs/btrfs/FS-UUID/allocation/data/bg_reclaim_threshold` | 0–99 (percent) | 0 (off) | v5.19+ | Triggers background reclaim for DATA block groups when usage drops to the threshold. |
| `btrfs-allocation-metadata-bg_reclaim_threshold` | `/sys/fs/btrfs/FS-UUID/allocation/metadata/bg_reclaim_threshold` | 0–99 (percent) | 0 (off) | v5.19+ | Same as above, for METADATA block groups. |
| `btrfs-allocation-data-dynamic_reclaim` | `/sys/fs/btrfs/FS-UUID/allocation/data/dynamic_reclaim` | `0` or `1` | 0 (off) | v6.11+ | Heuristic reclaim that addresses [some concerns](https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f5ff64ccf7bb7274ed66b0d835b2f6ae10af5d7a) of `bg_reclaim_threshold`. |
| `btrfs-allocation-metadata-dynamic_reclaim` | `/sys/fs/btrfs/FS-UUID/allocation/metadata/dynamic_reclaim` | `0` or `1` | 0 (off) | v6.11+ | Same as above, for METADATA block groups. |
| `btrfs-bdi-read_ahead_kb` | `/sys/fs/btrfs/FS-UUID/bdi/read_ahead_kb` | integer kB ≥ 0 | kernel/device dependent | v5.9+ | Per-BDI readahead. Powers of two are commonly used. |

See more in the [in btrfs docs](https://btrfs.readthedocs.io/en/latest/ch-sysfs.html#uuid-allocations-data-metadata-system).

Expand Down
66 changes: 44 additions & 22 deletions pkg/gce-pd-csi-driver/node.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,19 +121,31 @@ const (
fsTypeExt3 = "ext3"
fsTypeBtrfs = "btrfs"

readAheadKBMountFlagRegexPattern = "^read_ahead_kb=(.+)$"
btrfsReclaimDataRegexPattern = "^btrfs-allocation-data-bg_reclaim_threshold=(\\d{1,2})$" // 0-99 are valid, incl. 00
btrfsReclaimMetadataRegexPattern = "^btrfs-allocation-metadata-bg_reclaim_threshold=(\\d{1,2})$" // ditto ^
btrfsReadAheadKBRegexPattern = "^btrfs-bdi-read_ahead_kb=(\\d+)$"
readAheadKBMountFlagRegexPattern = "^read_ahead_kb=(.+)$"
btrfsReclaimDataRegexPattern = "^btrfs-allocation-data-bg_reclaim_threshold=(\\d{1,2})$" // 0-99 are valid, incl. 00
btrfsReclaimMetadataRegexPattern = "^btrfs-allocation-metadata-bg_reclaim_threshold=(\\d{1,2})$" // ditto ^
btrfsDynamicReclaimDataRegexPattern = "^btrfs-allocation-data-dynamic_reclaim=(0|1)$" // boolean in kernel, so accepting 0 or 1
btrfsDynamicReclaimMetadataRegexPattern = "^btrfs-allocation-metadata-dynamic_reclaim=(0|1)$" // ditto ^
btrfsReadAheadKBRegexPattern = "^btrfs-bdi-read_ahead_kb=(\\d+)$"
)

var (
readAheadKBMountFlagRegex = regexp.MustCompile(readAheadKBMountFlagRegexPattern)
btrfsReclaimDataRegex = regexp.MustCompile(btrfsReclaimDataRegexPattern)
btrfsReclaimMetadataRegex = regexp.MustCompile(btrfsReclaimMetadataRegexPattern)
btrfsReadAheadKBRegex = regexp.MustCompile(btrfsReadAheadKBRegexPattern)
readAheadKBMountFlagRegex = regexp.MustCompile(readAheadKBMountFlagRegexPattern)
btrfsReclaimDataRegex = regexp.MustCompile(btrfsReclaimDataRegexPattern)
btrfsReclaimMetadataRegex = regexp.MustCompile(btrfsReclaimMetadataRegexPattern)
btrfsDynamicReclaimDataRegex = regexp.MustCompile(btrfsDynamicReclaimDataRegexPattern)
btrfsDynamicReclaimMetadataRegex = regexp.MustCompile(btrfsDynamicReclaimMetadataRegexPattern)
btrfsReadAheadKBRegex = regexp.MustCompile(btrfsReadAheadKBRegexPattern)
)

type btrfsFlags struct {
reclaimData,
reclaimMetadata,
dynamicReclaimData,
dynamicReclaimMetadata,
readAheadKb string
}

func getDefaultFsType() string {
if runtime.GOOS == "windows" {
return defaultWindowsFsType
Expand Down Expand Up @@ -404,7 +416,7 @@ func (ns *GCENodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStage
// Part 3: Mount device to stagingTargetPath
fstype := getDefaultFsType()

var btrfsReclaimData, btrfsReclaimMetadata, btrfsReadAheadKb string
var btrfsFlags btrfsFlags
shouldUpdateReadAhead := false
var readAheadKB int64
options := []string{}
Expand All @@ -420,7 +432,7 @@ func (ns *GCENodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStage
}

if mnt.FsType == fsTypeBtrfs {
btrfsReclaimData, btrfsReclaimMetadata, btrfsReadAheadKb = extractBtrfsFlags(mnt.MountFlags)
btrfsFlags = extractBtrfsFlags(mnt.MountFlags)
}
} else if blk := volumeCapability.GetBlock(); blk != nil {
// Noop for Block NodeStageVolume
Expand Down Expand Up @@ -476,16 +488,22 @@ func (ns *GCENodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStage

btrfsSysfs := map[string]string{}

if btrfsReadAheadKb != "" {
btrfsSysfs["bdi/read_ahead_kb"] = btrfsReadAheadKb
if btrfsFlags.readAheadKb != "" {
btrfsSysfs["bdi/read_ahead_kb"] = btrfsFlags.readAheadKb
}

if !readonly {
if btrfsReclaimData != "" {
btrfsSysfs["allocation/data/bg_reclaim_threshold"] = btrfsReclaimData
if btrfsFlags.reclaimData != "" {
btrfsSysfs["allocation/data/bg_reclaim_threshold"] = btrfsFlags.reclaimData
}
if btrfsFlags.reclaimMetadata != "" {
btrfsSysfs["allocation/metadata/bg_reclaim_threshold"] = btrfsFlags.reclaimMetadata
}
if btrfsFlags.dynamicReclaimData != "" {
btrfsSysfs["allocation/data/dynamic_reclaim"] = btrfsFlags.dynamicReclaimData
}
if btrfsReclaimMetadata != "" {
btrfsSysfs["allocation/metadata/bg_reclaim_threshold"] = btrfsReclaimMetadata
if btrfsFlags.dynamicReclaimMetadata != "" {
btrfsSysfs["allocation/metadata/dynamic_reclaim"] = btrfsFlags.dynamicReclaimMetadata
}
}

Expand Down Expand Up @@ -552,18 +570,22 @@ func (ns *GCENodeServer) updateReadAhead(devicePath string, readAheadKB int64) e
return nil
}

func extractBtrfsFlags(mountFlags []string) (string, string, string) {
var reclaimData, reclaimMetadata, readAheadKb string
func extractBtrfsFlags(mountFlags []string) btrfsFlags {
var flags btrfsFlags
for _, mountFlag := range mountFlags {
if got := btrfsReclaimDataRegex.FindStringSubmatch(mountFlag); len(got) == 2 {
reclaimData = got[1]
flags.reclaimData = got[1]
} else if got := btrfsReclaimMetadataRegex.FindStringSubmatch(mountFlag); len(got) == 2 {
reclaimMetadata = got[1]
flags.reclaimMetadata = got[1]
} else if got := btrfsReadAheadKBRegex.FindStringSubmatch(mountFlag); len(got) == 2 {
readAheadKb = got[1]
flags.readAheadKb = got[1]
} else if got := btrfsDynamicReclaimDataRegex.FindStringSubmatch(mountFlag); len(got) == 2 {
flags.dynamicReclaimData = got[1]
} else if got := btrfsDynamicReclaimMetadataRegex.FindStringSubmatch(mountFlag); len(got) == 2 {
flags.dynamicReclaimMetadata = got[1]
}
}
return reclaimData, reclaimMetadata, readAheadKb
return flags
}

func extractReadAheadKBMountFlag(mountFlags []string) (int64, bool, error) {
Expand Down
58 changes: 35 additions & 23 deletions pkg/gce-pd-csi-driver/node_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -631,7 +631,9 @@ func TestNodeStageVolume(t *testing.T) {
btrfsPrefix = fmt.Sprintf("%s/sys/fs/btrfs/%s", tempDir, btrfsUUID)
btrfsFixtures = map[string]string{
"allocation/data/bg_reclaim_threshold": "0\n",
"allocation/data/dynamic_reclaim": "0\n",
"allocation/metadata/bg_reclaim_threshold": "0\n",
"allocation/metadata/dynamic_reclaim": "0\n",
"bdi/read_ahead_kb": "4096\n",
}
)
Expand All @@ -648,22 +650,24 @@ func TestNodeStageVolume(t *testing.T) {
}

testCases := []struct {
name string
req *csi.NodeStageVolumeRequest
deviceSize int
blockExtSize int
readonlyBit string
expResize bool
expReadAheadUpdate bool
expReadAheadKB string
expReadOnlyRemount bool
expCommandList []fakeCmd
readAheadSectors string
btrfsReclaimData string
btrfsReclaimMetadata string
btrfsReadAheadKb string
sectorSizeInBytes int
expErrCode codes.Code
name string
req *csi.NodeStageVolumeRequest
deviceSize int
blockExtSize int
readonlyBit string
expResize bool
expReadAheadUpdate bool
expReadAheadKB string
expReadOnlyRemount bool
expCommandList []fakeCmd
readAheadSectors string
btrfsReclaimData string
btrfsReclaimMetadata string
btrfsDynamicReclaimData string
btrfsDynamicReclaimMetadata string
btrfsReadAheadKb string
sectorSizeInBytes int
expErrCode codes.Code
}{
{
name: "Valid request, resize even though block and filesystem sizes match",
Expand Down Expand Up @@ -927,6 +931,8 @@ func TestNodeStageVolume(t *testing.T) {
MountFlags: []string{
"btrfs-allocation-data-bg_reclaim_threshold=90",
"btrfs-allocation-metadata-bg_reclaim_threshold=91",
"btrfs-allocation-data-dynamic_reclaim=1",
"btrfs-allocation-metadata-dynamic_reclaim=1",
"btrfs-bdi-read_ahead_kb=128",
},
},
Expand All @@ -936,12 +942,14 @@ func TestNodeStageVolume(t *testing.T) {
},
},
},
deviceSize: 1,
blockExtSize: 1,
readonlyBit: "0",
btrfsReclaimData: "90",
btrfsReclaimMetadata: "91",
btrfsReadAheadKb: "128",
deviceSize: 1,
blockExtSize: 1,
readonlyBit: "0",
btrfsReclaimData: "90",
btrfsReclaimMetadata: "91",
btrfsDynamicReclaimData: "1",
btrfsDynamicReclaimMetadata: "1",
btrfsReadAheadKb: "128",
expCommandList: []fakeCmd{
{
cmd: "blkid",
Expand Down Expand Up @@ -1267,13 +1275,17 @@ func TestNodeStageVolume(t *testing.T) {
if tc.expReadAheadUpdate == false && readAheadUpdateCalled == true {
t.Fatalf("Test updated read ahead, but it was not expected.")
}
if tc.btrfsReclaimData == "" && tc.btrfsReclaimMetadata == "" && tc.btrfsReadAheadKb == "" && blkidCalled {
if tc.btrfsReclaimData == "" && tc.btrfsReclaimMetadata == "" &&
tc.btrfsDynamicReclaimData == "" && tc.btrfsDynamicReclaimMetadata == "" &&
tc.btrfsReadAheadKb == "" && blkidCalled {
t.Fatalf("blkid was called, but was not expected.")
}

btrfsProps := map[string]string{
"/allocation/data/bg_reclaim_threshold": tc.btrfsReclaimData,
"/allocation/metadata/bg_reclaim_threshold": tc.btrfsReclaimMetadata,
"/allocation/data/dynamic_reclaim": tc.btrfsDynamicReclaimData,
"/allocation/metadata/dynamic_reclaim": tc.btrfsDynamicReclaimMetadata,
"/bdi/read_ahead_kb": tc.btrfsReadAheadKb,
}

Expand Down
6 changes: 6 additions & 0 deletions pkg/gce-pd-csi-driver/utils.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -317,6 +317,12 @@ func collectMountOptions(fsType string, mntFlags []string) []string {
if btrfsReclaimMetadataRegex.FindString(opt) != "" {
continue
}
if btrfsDynamicReclaimDataRegex.FindString(opt) != "" {
continue
}
if btrfsDynamicReclaimMetadataRegex.FindString(opt) != "" {
continue
}
if btrfsReadAheadKBRegex.FindString(opt) != "" {
continue
}
Expand Down