From 5a01ef25da05a4812a6b63d03a223e73fa930f58 Mon Sep 17 00:00:00 2001 From: codex Date: Fri, 23 Oct 2026 22:10:01 +0800 Subject: [PATCH] fix(agent): boot rescue ISO before seed media --- agent/hypervisor/libvirt_domain_features.go | 110 ++++++++++++++++---- agent/hypervisor/libvirt_safety_test.go | 66 ++++++++++-- 2 files changed, 147 insertions(+), 29 deletions(-) diff --git a/agent/hypervisor/libvirt_domain_features.go b/agent/hypervisor/libvirt_domain_features.go index ce3dfda7..cf910efa 100644 --- a/agent/hypervisor/libvirt_domain_features.go +++ b/agent/hypervisor/libvirt_domain_features.go @@ -782,12 +782,9 @@ func (l *Libvirt) EnterRescue(ctx context.Context, uuid string, request RescueRe if definition.OS == nil || definition.Devices == nil { return NewError(ErrConflict, "invalid_domain", "managed domain XML is incomplete", nil) } - definition.OS.BootDevices = []libvirtxml.DomainBootDevice{{Dev: "cdrom"}, {Dev: "hd"}} - definition.Devices.Disks = append(definition.Devices.Disks, libvirtxml.DomainDisk{ - Device: "cdrom", Driver: &libvirtxml.DomainDiskDriver{Name: "qemu", Type: "raw"}, - Source: &libvirtxml.DomainDiskSource{File: &libvirtxml.DomainDiskSourceFile{File: path}}, - Target: &libvirtxml.DomainDiskTarget{Dev: "sdb", Bus: "sata"}, ReadOnly: &libvirtxml.DomainDiskReadOnly{}, - }) + if err := stageRescueBoot(definition, path); err != nil { + return err + } metadata.RescueTemplateID = request.TemplateID metadata.RescueWasRunning = wasRunning metadata.RescueAutostart = autostart @@ -951,15 +948,7 @@ func (l *Libvirt) ExitRescue(ctx context.Context, uuid string, request ExitRescu if definition.OS == nil || definition.Devices == nil { return NewError(ErrConflict, "invalid_domain", "managed domain XML is incomplete", nil) } - definition.OS.BootDevices = []libvirtxml.DomainBootDevice{{Dev: "hd"}} - disks := definition.Devices.Disks[:0] - for _, disk := range definition.Devices.Disks { - if disk.Device == "cdrom" && disk.Target != nil && disk.Target.Dev == "sdb" { - continue - } - disks = append(disks, disk) - } - definition.Devices.Disks = disks + restoreNormalBoot(definition) linkState := "up" if *request.NetworkSuspended { linkState = "down" @@ -1095,25 +1084,100 @@ func validateRescueDefinition(document, templateID, expectedPath string) error { if err != nil { return err } - if metadata.RescueTemplateID != templateID || definition.OS == nil || len(definition.OS.BootDevices) != 2 || definition.OS.BootDevices[0].Dev != "cdrom" || definition.OS.BootDevices[1].Dev != "hd" || definition.Devices == nil { + if metadata.RescueTemplateID != templateID || definition.OS == nil || definition.Devices == nil { + return NewError(ErrConflict, "rescue_definition_invalid", "rescue boot metadata or order is invalid", nil) + } + legacyBootOrder := len(definition.OS.BootDevices) == 2 && definition.OS.BootDevices[0].Dev == "cdrom" && definition.OS.BootDevices[1].Dev == "hd" + deviceBootOrder := len(definition.OS.BootDevices) == 0 + if !legacyBootOrder && !deviceBootOrder { return NewError(ErrConflict, "rescue_definition_invalid", "rescue boot metadata or order is invalid", nil) } - matches := 0 + rescueMatches := 0 + rootMatches := 0 + bootDevices := 0 for _, disk := range definition.Devices.Disks { - if disk.Device != "cdrom" || disk.Target == nil || disk.Target.Dev != "sdb" { + if disk.Boot != nil { + bootDevices++ + } + if disk.Device == "disk" && disk.Target != nil && disk.Target.Dev == "vda" { + if deviceBootOrder && (disk.Boot == nil || disk.Boot.Order != 2) { + return NewError(ErrConflict, "rescue_definition_invalid", "managed root disk is not the rescue fallback boot device", nil) + } + if legacyBootOrder && disk.Boot != nil { + return NewError(ErrConflict, "rescue_definition_invalid", "legacy rescue boot mixes global and per-device order", nil) + } + rootMatches++ + continue + } + if disk.Device == "cdrom" && disk.Target != nil && disk.Target.Dev == "sdb" { + if deviceBootOrder && (disk.Boot == nil || disk.Boot.Order != 1) { + return NewError(ErrConflict, "rescue_definition_invalid", "managed rescue disk is not the first boot device", nil) + } + if legacyBootOrder && disk.Boot != nil { + return NewError(ErrConflict, "rescue_definition_invalid", "legacy rescue boot mixes global and per-device order", nil) + } + if disk.Driver == nil || disk.Driver.Type != "raw" || disk.Source == nil || disk.Source.File == nil || filepath.Clean(disk.Source.File.File) != filepath.Clean(expectedPath) || disk.ReadOnly == nil { + return NewError(ErrConflict, "rescue_definition_invalid", "rescue boot disk does not match the managed template", nil) + } + rescueMatches++ continue } - if disk.Driver == nil || disk.Driver.Type != "raw" || disk.Source == nil || disk.Source.File == nil || filepath.Clean(disk.Source.File.File) != filepath.Clean(expectedPath) || disk.ReadOnly == nil { - return NewError(ErrConflict, "rescue_definition_invalid", "rescue boot disk does not match the managed template", nil) + if disk.Boot != nil { + return NewError(ErrConflict, "rescue_definition_invalid", "an unmanaged disk participates in rescue boot order", nil) } - matches++ } - if matches != 1 { - return NewError(ErrConflict, "rescue_definition_invalid", "rescue boot must contain exactly one managed rescue disk", nil) + expectedBootDevices := 2 + if legacyBootOrder { + expectedBootDevices = 0 + } + if rescueMatches != 1 || rootMatches != 1 || bootDevices != expectedBootDevices { + return NewError(ErrConflict, "rescue_definition_invalid", "rescue boot must contain exactly one rescue disk and one managed root fallback", nil) } return nil } +func stageRescueBoot(definition *libvirtxml.Domain, path string) error { + definition.OS.BootDevices = nil + rootIndex := -1 + for index := range definition.Devices.Disks { + disk := &definition.Devices.Disks[index] + disk.Boot = nil + if disk.Device == "disk" && disk.Target != nil && disk.Target.Dev == "vda" { + if rootIndex >= 0 { + return NewError(ErrConflict, "invalid_domain", "managed domain contains multiple root disks", nil) + } + rootIndex = index + } + if disk.Device == "cdrom" && disk.Target != nil && disk.Target.Dev == "sdb" { + return NewError(ErrConflict, "invalid_domain", "managed rescue disk target is already in use", nil) + } + } + if rootIndex < 0 { + return NewError(ErrConflict, "invalid_domain", "managed root disk is missing", nil) + } + definition.Devices.Disks[rootIndex].Boot = &libvirtxml.DomainDeviceBoot{Order: 2} + definition.Devices.Disks = append(definition.Devices.Disks, libvirtxml.DomainDisk{ + Device: "cdrom", Driver: &libvirtxml.DomainDiskDriver{Name: "qemu", Type: "raw"}, + Source: &libvirtxml.DomainDiskSource{File: &libvirtxml.DomainDiskSourceFile{File: path}}, + Target: &libvirtxml.DomainDiskTarget{Dev: "sdb", Bus: "sata"}, ReadOnly: &libvirtxml.DomainDiskReadOnly{}, + Boot: &libvirtxml.DomainDeviceBoot{Order: 1}, + }) + return nil +} + +func restoreNormalBoot(definition *libvirtxml.Domain) { + definition.OS.BootDevices = []libvirtxml.DomainBootDevice{{Dev: "hd"}} + disks := definition.Devices.Disks[:0] + for _, disk := range definition.Devices.Disks { + if disk.Device == "cdrom" && disk.Target != nil && disk.Target.Dev == "sdb" { + continue + } + disk.Boot = nil + disks = append(disks, disk) + } + definition.Devices.Disks = disks +} + func definitionNetworkSuspended(definition *libvirtxml.Domain) bool { if definition == nil || definition.Devices == nil || len(definition.Devices.Interfaces) == 0 { return false diff --git a/agent/hypervisor/libvirt_safety_test.go b/agent/hypervisor/libvirt_safety_test.go index f3ade69f..d57842c1 100644 --- a/agent/hypervisor/libvirt_safety_test.go +++ b/agent/hypervisor/libvirt_safety_test.go @@ -1633,12 +1633,9 @@ func TestValidateRescueDefinitionRequiresManagedReadOnlyBootDisk(t *testing.T) { templateID := "0123456789abcdef0123456789abcdef" rescuePath := "/var/lib/libvirt/images/rescue-0123456789abcdef0123456789abcdef.iso" staged, err := mutateDomainDocument(document, func(domain *libvirtxml.Domain, metadata *domainMetadataXML) error { - domain.OS.BootDevices = []libvirtxml.DomainBootDevice{{Dev: "cdrom"}, {Dev: "hd"}} - domain.Devices.Disks = append(domain.Devices.Disks, libvirtxml.DomainDisk{ - Device: "cdrom", Driver: &libvirtxml.DomainDiskDriver{Name: "qemu", Type: "raw"}, - Source: &libvirtxml.DomainDiskSource{File: &libvirtxml.DomainDiskSourceFile{File: rescuePath}}, - Target: &libvirtxml.DomainDiskTarget{Dev: "sdb", Bus: "sata"}, ReadOnly: &libvirtxml.DomainDiskReadOnly{}, - }) + if err := stageRescueBoot(domain, rescuePath); err != nil { + return err + } metadata.RescueTemplateID = templateID return nil }) @@ -1651,6 +1648,63 @@ func TestValidateRescueDefinitionRequiresManagedReadOnlyBootDisk(t *testing.T) { if err := validateRescueDefinition(staged, templateID, rescuePath+".foreign"); err == nil { t.Fatal("foreign rescue volume path was accepted") } + parsed, _, err := parseManagedDomain(staged) + if err != nil { + t.Fatal(err) + } + if len(parsed.OS.BootDevices) != 0 { + t.Fatalf("global boot order remains staged: %#v", parsed.OS.BootDevices) + } + for _, disk := range parsed.Devices.Disks { + if disk.Target == nil { + continue + } + switch disk.Target.Dev { + case "vda": + if disk.Boot == nil || disk.Boot.Order != 2 { + t.Fatalf("root disk boot order = %#v, want 2", disk.Boot) + } + case "sda": + if disk.Boot != nil { + t.Fatalf("seed ISO unexpectedly participates in rescue boot: %#v", disk.Boot) + } + case "sdb": + if disk.Boot == nil || disk.Boot.Order != 1 { + t.Fatalf("rescue ISO boot order = %#v, want 1", disk.Boot) + } + } + } + + legacy, err := mutateDomainDocument(staged, func(domain *libvirtxml.Domain, _ *domainMetadataXML) error { + domain.OS.BootDevices = []libvirtxml.DomainBootDevice{{Dev: "cdrom"}, {Dev: "hd"}} + for index := range domain.Devices.Disks { + domain.Devices.Disks[index].Boot = nil + } + return nil + }) + if err != nil { + t.Fatal(err) + } + if err := validateRescueDefinition(legacy, templateID, rescuePath); err != nil { + t.Fatalf("legacy rescue definition cannot be exited after upgrade: %v", err) + } + + restored, _, err := parseManagedDomain(staged) + if err != nil { + t.Fatal(err) + } + restoreNormalBoot(restored) + if len(restored.OS.BootDevices) != 1 || restored.OS.BootDevices[0].Dev != "hd" { + t.Fatalf("normal global boot order = %#v, want hd", restored.OS.BootDevices) + } + for _, disk := range restored.Devices.Disks { + if disk.Boot != nil { + t.Fatalf("per-device boot order remained after rescue exit: %#v", disk.Boot) + } + if disk.Target != nil && disk.Target.Dev == "sdb" { + t.Fatal("managed rescue disk remained after rescue exit") + } + } } func TestCollectVMManagedVolumeNamesIncludesTransitionsSnapshotsAndOrphans(t *testing.T) {