filesystem: prevent panic on mount points with non-UTF-8 names

[ricardbejarano]

Summary

On Linux, mount paths are byte strings and are usually — but not always — valid UTF-8. When a filesystem is mounted on a path containing non-UTF-8 bytes, the filesystem collector passes that value straight to prometheus.MustNewConstMetric as a label value, which panics (Prometheus label values must be valid UTF-8). This crashes the whole scrape:

panic: label value "/mnt/cass\xe9" is not valid UTF-8
github.com/prometheus/client_golang/prometheus.MustNewConstMetric(...)
github.com/prometheus/node_exporter/collector.(*filesystemCollector).Update(...)

Fix

Sanitize the device, mountPoint and fsType labels with strings.ToValidUTF8(..., "�") in parseFilesystemLabels. This mirrors the approach already used elsewhere in the tree (hwmon, dmi, netclass, powersupply collectors and collector/utils), so invalid bytes are replaced with the Unicode replacement character instead of taking the exporter down.

A unit test covering non-UTF-8 device/mountPoint/fsType input is included.

Fixes #3662

Reproduce

fallocate -l 32m /root/disk.dd
mkfs -t ext2 /root/disk.dd
mount --mkdir /root/disk.dd /mnt/cass$'\xe9'
./node_exporter &
curl http://localhost:9100/metrics

[rtreffer]

I was able to reproduce on an older node_exporter for mount point and device by doing a loopback mount of a ext4 disk image. Both the image and mount point had non-utf8 names.

panic: label value "/media/cass\xe9" is not valid UTF-8

goroutine 89 [running]:
github.com/prometheus/client_golang/prometheus.MustNewConstMetric(...)
        github.com/prometheus/client_golang/prometheus/value.go:129
github.com/prometheus/node_exporter/collector.(*filesystemCollector).Update(0xc000246c40, 0xc0002956c0)
        github.com/prometheus/node_exporter/collector/filesystem_common.go:198 +0xb49
github.com/prometheus/node_exporter/collector.execute({0xcddee3, 0xa}, {0xe15280, 0xc000246c40}, 0xc0002956c0, 0xc0000491f0)
        github.com/prometheus/node_exporter/collector/collector.go:160 +0x85
github.com/prometheus/node_exporter/collector.NodeCollector.Collect.func1({0xcddee3?, 0x0?}, {0xe15280?, 0xc000246c40?})
        github.com/prometheus/node_exporter/collector/collector.go:151 +0x33
created by github.com/prometheus/node_exporter/collector.NodeCollector.Collect in goroutine 44
        github.com/prometheus/node_exporter/collector/collector.go:150 +0xce

This change aligns labels with the output of mount

/media/cass�/out.img on /media/cass� type ext2 (rw,relatime)

listmount indicates that all values can be byte strings.
I doubt that this is relevant for fsType but it does not harm to normalize all values.