Use lsblk to safeguard against outdated symlinks

Signed-off-by: Connor Catlett <[email protected]>

Makefile

@@ -35,8 +35,8 @@ OS?=linux
 ARCH?=amd64
 OSVERSION?=al2023
 
-ALL_OS?=linux windows
-ALL_ARCH_linux?=amd64 arm64
+ALL_OS?=linux
+ALL_ARCH_linux?=amd64
 ALL_OSVERSION_linux?=al2023
 ALL_OS_ARCH_OSVERSION_linux=$(foreach arch, $(ALL_ARCH_linux), $(foreach osversion, ${ALL_OSVERSION_linux}, linux-$(arch)-${osversion}))
 
@@ -86,7 +86,7 @@ create-manifest: all-image-registry
 .PHONY: all-image-docker
 all-image-docker: $(addprefix sub-image-docker-,$(ALL_OS_ARCH_OSVERSION_linux))
 .PHONY: all-image-registry
-all-image-registry: sub-image-registry-linux-arm64-al2 $(addprefix sub-image-registry-,$(ALL_OS_ARCH_OSVERSION))
+all-image-registry: $(addprefix sub-image-registry-,$(ALL_OS_ARCH_OSVERSION))
 
 sub-image-%:
 	$(MAKE) OUTPUT_TYPE=$(call word-hyphen,$*,1) OS=$(call word-hyphen,$*,2) ARCH=$(call word-hyphen,$*,3) OSVERSION=$(call word-hyphen,$*,4) image

pkg/driver/node_linux.go

@@ -22,7 +22,9 @@ package driver
 import (
 	"fmt"
 	"os"
+	"os/exec"
 	"path/filepath"
+	"regexp"
 	"strconv"
 	"strings"
 
@@ -87,7 +89,8 @@ func (d *nodeService) findDevicePath(devicePath, volumeID, partition string) (st
 	// which AWS presents NVME devices under /dev/disk/by-id/. For example,
 	// vol-0fab1d5e3f72a5e23 creates a symlink at
 	// /dev/disk/by-id/nvme-Amazon_Elastic_Block_Store_vol0fab1d5e3f72a5e23
-	nvmeName := "nvme-Amazon_Elastic_Block_Store_" + strings.Replace(volumeID, "-", "", -1)
+	strippedVolumeName := strings.Replace(volumeID, "-", "", -1)
+	nvmeName := "nvme-Amazon_Elastic_Block_Store_" + strippedVolumeName
 
 	nvmeDevicePath, err := findNvmeVolume(d.deviceIdentifier, nvmeName)
 
@@ -112,6 +115,26 @@ func (d *nodeService) findDevicePath(devicePath, volumeID, partition string) (st
 		return "", errNoDevicePathFound(devicePath, volumeID)
 	}
 
+	// In some rare cases, a race condition can lead to the /dev/disk/by-id/ symlink becoming out of date
+	// See https://github.com/kubernetes-sigs/aws-ebs-csi-driver/issues/1224 for more info
+	// Attempt to use lsblk to double check that the nvme device selected was the correct volume
+	output, err := exec.Command("lsblk", "--noheadings", "--ascii", "--nodeps", "--output", "SERIAL", canonicalDevicePath).CombinedOutput()
+
+	if err != nil {
+		// Look for an EBS volume ID in the output, compare all matches against what we expect
+		// (in some rare cases there may be multiple matches due to lsblk printing partitions)
+		// If no volume ID is in the output (non-Nitro instances, SBE devices, etc) silently proceed
+		volumeRegex := regexp.MustCompile(`vol[a-z0-9]+`)
+		for _, volume := range volumeRegex.FindAllString(string(output), -1) {
+			if volume != strippedVolumeName {
+				return "", fmt.Errorf("Refusing to mount %s because it claims to be %s but should be %s", canonicalDevicePath, volume, strippedVolumeName)
+			}
+		}
+	} else {
+		// If the command fails (for example, because lsblk is not available), silently ignore the error and proceed
+		klog.V(5).ErrorS(err, "Ignoring lsblk failure", "devicePath", devicePath, "volumeID", volumeID)
+	}
+
 	canonicalDevicePath = d.appendPartition(canonicalDevicePath, partition)
 	return canonicalDevicePath, nil
 }