Merge branch '16265-security-updates' into dependabot/bundler/apps/workbench/loofah...
[arvados.git] / lib / cloud / ec2 / ec2.go
index b8a360e19fabd36777b15112e8f47ccaf3a7bab3..2de82b1dcf9292f69959fb357839773b6a5df5e7 100644 (file)
@@ -5,14 +5,18 @@
 package ec2
 
 import (
+       "crypto/md5"
+       "crypto/rsa"
+       "crypto/sha1"
+       "crypto/x509"
        "encoding/base64"
        "encoding/json"
        "fmt"
-       "log"
-       "strings"
+       "math/big"
+       "sync"
 
-       "git.curoverse.com/arvados.git/lib/cloud"
-       "git.curoverse.com/arvados.git/sdk/go/arvados"
+       "git.arvados.org/arvados.git/lib/cloud"
+       "git.arvados.org/arvados.git/sdk/go/arvados"
        "github.com/aws/aws-sdk-go/aws"
        "github.com/aws/aws-sdk-go/aws/credentials"
        "github.com/aws/aws-sdk-go/aws/session"
@@ -21,9 +25,6 @@ import (
        "golang.org/x/crypto/ssh"
 )
 
-const arvadosDispatchID = "arvados-dispatch-id"
-const tagPrefix = "arvados-dispatch-tag-"
-
 // Driver is the ec2 implementation of the cloud.Driver interface.
 var Driver = cloud.DriverFunc(newEC2InstanceSet)
 
@@ -31,13 +32,14 @@ type ec2InstanceSetConfig struct {
        AccessKeyID      string
        SecretAccessKey  string
        Region           string
-       SecurityGroupIDs []string
+       SecurityGroupIDs arvados.StringSet
        SubnetID         string
        AdminUsername    string
-       KeyPairName      string
+       EBSVolumeType    string
 }
 
 type ec2Interface interface {
+       DescribeKeyPairs(input *ec2.DescribeKeyPairsInput) (*ec2.DescribeKeyPairsOutput, error)
        ImportKeyPair(input *ec2.ImportKeyPairInput) (*ec2.ImportKeyPairOutput, error)
        RunInstances(input *ec2.RunInstancesInput) (*ec2.Reservation, error)
        DescribeInstances(input *ec2.DescribeInstancesInput) (*ec2.DescribeInstancesOutput, error)
@@ -46,17 +48,18 @@ type ec2Interface interface {
 }
 
 type ec2InstanceSet struct {
-       ec2config    ec2InstanceSetConfig
-       dispatcherID cloud.InstanceSetID
-       logger       logrus.FieldLogger
-       client       ec2Interface
-       importedKey  bool
+       ec2config     ec2InstanceSetConfig
+       instanceSetID cloud.InstanceSetID
+       logger        logrus.FieldLogger
+       client        ec2Interface
+       keysMtx       sync.Mutex
+       keys          map[string]string
 }
 
-func newEC2InstanceSet(config json.RawMessage, dispatcherID cloud.InstanceSetID, logger logrus.FieldLogger) (prv cloud.InstanceSet, err error) {
+func newEC2InstanceSet(config json.RawMessage, instanceSetID cloud.InstanceSetID, _ cloud.SharedResourceTags, logger logrus.FieldLogger) (prv cloud.InstanceSet, err error) {
        instanceSet := &ec2InstanceSet{
-               dispatcherID: dispatcherID,
-               logger:       logger,
+               instanceSetID: instanceSetID,
+               logger:        logger,
        }
        err = json.Unmarshal(config, &instanceSet.ec2config)
        if err != nil {
@@ -69,9 +72,46 @@ func newEC2InstanceSet(config json.RawMessage, dispatcherID cloud.InstanceSetID,
                        "")).
                WithRegion(instanceSet.ec2config.Region)
        instanceSet.client = ec2.New(session.Must(session.NewSession(awsConfig)))
+       instanceSet.keys = make(map[string]string)
+       if instanceSet.ec2config.EBSVolumeType == "" {
+               instanceSet.ec2config.EBSVolumeType = "gp2"
+       }
        return instanceSet, nil
 }
 
+func awsKeyFingerprint(pk ssh.PublicKey) (md5fp string, sha1fp string, err error) {
+       // AWS key fingerprints don't use the usual key fingerprint
+       // you get from ssh-keygen or ssh.FingerprintLegacyMD5()
+       // (you can get that from md5.Sum(pk.Marshal())
+       //
+       // AWS uses the md5 or sha1 of the PKIX DER encoding of the
+       // public key, so calculate those fingerprints here.
+       var rsaPub struct {
+               Name string
+               E    *big.Int
+               N    *big.Int
+       }
+       if err := ssh.Unmarshal(pk.Marshal(), &rsaPub); err != nil {
+               return "", "", fmt.Errorf("agent: Unmarshal failed to parse public key: %v", err)
+       }
+       rsaPk := rsa.PublicKey{
+               E: int(rsaPub.E.Int64()),
+               N: rsaPub.N,
+       }
+       pkix, _ := x509.MarshalPKIXPublicKey(&rsaPk)
+       md5pkix := md5.Sum([]byte(pkix))
+       sha1pkix := sha1.Sum([]byte(pkix))
+       md5fp = ""
+       sha1fp = ""
+       for i := 0; i < len(md5pkix); i += 1 {
+               md5fp += fmt.Sprintf(":%02x", md5pkix[i])
+       }
+       for i := 0; i < len(sha1pkix); i += 1 {
+               sha1fp += fmt.Sprintf(":%02x", sha1pkix[i])
+       }
+       return md5fp[1:], sha1fp[1:], nil
+}
+
 func (instanceSet *ec2InstanceSet) Create(
        instanceType arvados.InstanceType,
        imageID cloud.ImageID,
@@ -79,63 +119,85 @@ func (instanceSet *ec2InstanceSet) Create(
        initCommand cloud.InitCommand,
        publicKey ssh.PublicKey) (cloud.Instance, error) {
 
-       if !instanceSet.importedKey {
-               instanceSet.client.ImportKeyPair(&ec2.ImportKeyPairInput{
-                       KeyName:           &instanceSet.ec2config.KeyPairName,
-                       PublicKeyMaterial: ssh.MarshalAuthorizedKey(publicKey),
-               })
-               instanceSet.importedKey = true
+       md5keyFingerprint, sha1keyFingerprint, err := awsKeyFingerprint(publicKey)
+       if err != nil {
+               return nil, fmt.Errorf("Could not make key fingerprint: %v", err)
        }
+       instanceSet.keysMtx.Lock()
+       var keyname string
+       var ok bool
+       if keyname, ok = instanceSet.keys[md5keyFingerprint]; !ok {
+               keyout, err := instanceSet.client.DescribeKeyPairs(&ec2.DescribeKeyPairsInput{
+                       Filters: []*ec2.Filter{&ec2.Filter{
+                               Name:   aws.String("fingerprint"),
+                               Values: []*string{&md5keyFingerprint, &sha1keyFingerprint},
+                       }},
+               })
+               if err != nil {
+                       return nil, fmt.Errorf("Could not search for keypair: %v", err)
+               }
 
-       ec2tags := []*ec2.Tag{
-               &ec2.Tag{
-                       Key:   aws.String(arvadosDispatchID),
-                       Value: aws.String(string(instanceSet.dispatcherID)),
-               },
-               &ec2.Tag{
-                       Key:   aws.String("arvados-class"),
-                       Value: aws.String("dynamic-compute"),
-               },
+               if len(keyout.KeyPairs) > 0 {
+                       keyname = *(keyout.KeyPairs[0].KeyName)
+               } else {
+                       keyname = "arvados-dispatch-keypair-" + md5keyFingerprint
+                       _, err := instanceSet.client.ImportKeyPair(&ec2.ImportKeyPairInput{
+                               KeyName:           &keyname,
+                               PublicKeyMaterial: ssh.MarshalAuthorizedKey(publicKey),
+                       })
+                       if err != nil {
+                               return nil, fmt.Errorf("Could not import keypair: %v", err)
+                       }
+               }
+               instanceSet.keys[md5keyFingerprint] = keyname
        }
+       instanceSet.keysMtx.Unlock()
+
+       ec2tags := []*ec2.Tag{}
        for k, v := range newTags {
                ec2tags = append(ec2tags, &ec2.Tag{
-                       Key:   aws.String(tagPrefix + k),
+                       Key:   aws.String(k),
                        Value: aws.String(v),
                })
        }
 
+       var groups []string
+       for sg := range instanceSet.ec2config.SecurityGroupIDs {
+               groups = append(groups, sg)
+       }
+
        rii := ec2.RunInstancesInput{
                ImageId:      aws.String(string(imageID)),
                InstanceType: &instanceType.ProviderType,
                MaxCount:     aws.Int64(1),
                MinCount:     aws.Int64(1),
-               KeyName:      &instanceSet.ec2config.KeyPairName,
+               KeyName:      &keyname,
 
                NetworkInterfaces: []*ec2.InstanceNetworkInterfaceSpecification{
                        &ec2.InstanceNetworkInterfaceSpecification{
                                AssociatePublicIpAddress: aws.Bool(false),
                                DeleteOnTermination:      aws.Bool(true),
                                DeviceIndex:              aws.Int64(0),
-                               Groups:                   aws.StringSlice(instanceSet.ec2config.SecurityGroupIDs),
+                               Groups:                   aws.StringSlice(groups),
                                SubnetId:                 &instanceSet.ec2config.SubnetID,
                        }},
                DisableApiTermination:             aws.Bool(false),
                InstanceInitiatedShutdownBehavior: aws.String("terminate"),
-               UserData: aws.String(base64.StdEncoding.EncodeToString([]byte("#!/bin/sh\n" + initCommand + "\n"))),
                TagSpecifications: []*ec2.TagSpecification{
                        &ec2.TagSpecification{
                                ResourceType: aws.String("instance"),
                                Tags:         ec2tags,
                        }},
+               UserData: aws.String(base64.StdEncoding.EncodeToString([]byte("#!/bin/sh\n" + initCommand + "\n"))),
        }
 
-       if instanceType.AttachScratch {
+       if instanceType.AddedScratch > 0 {
                rii.BlockDeviceMappings = []*ec2.BlockDeviceMapping{&ec2.BlockDeviceMapping{
                        DeviceName: aws.String("/dev/xvdt"),
                        Ebs: &ec2.EbsBlockDevice{
                                DeleteOnTermination: aws.Bool(true),
-                               VolumeSize:          aws.Int64((int64(instanceType.Scratch) / 1000000000) + 1),
-                               VolumeType:          aws.String("gp2"),
+                               VolumeSize:          aws.Int64((int64(instanceType.AddedScratch) + (1<<30 - 1)) >> 30),
+                               VolumeType:          &instanceSet.ec2config.EBSVolumeType,
                        }}}
        }
 
@@ -160,13 +222,15 @@ func (instanceSet *ec2InstanceSet) Create(
        }, nil
 }
 
-func (instanceSet *ec2InstanceSet) Instances(cloud.InstanceTags) (instances []cloud.Instance, err error) {
-       dii := &ec2.DescribeInstancesInput{
-               Filters: []*ec2.Filter{&ec2.Filter{
-                       Name:   aws.String("tag:" + arvadosDispatchID),
-                       Values: []*string{aws.String(string(instanceSet.dispatcherID))},
-               }}}
-
+func (instanceSet *ec2InstanceSet) Instances(tags cloud.InstanceTags) (instances []cloud.Instance, err error) {
+       var filters []*ec2.Filter
+       for k, v := range tags {
+               filters = append(filters, &ec2.Filter{
+                       Name:   aws.String("tag:" + k),
+                       Values: []*string{aws.String(v)},
+               })
+       }
+       dii := &ec2.DescribeInstancesInput{Filters: filters}
        for {
                dio, err := instanceSet.client.DescribeInstances(dii)
                if err != nil {
@@ -175,7 +239,9 @@ func (instanceSet *ec2InstanceSet) Instances(cloud.InstanceTags) (instances []cl
 
                for _, rsv := range dio.Reservations {
                        for _, inst := range rsv.Instances {
-                               instances = append(instances, &ec2Instance{instanceSet, inst})
+                               if *inst.State.Name != "shutting-down" && *inst.State.Name != "terminated" {
+                                       instances = append(instances, &ec2Instance{instanceSet, inst})
+                               }
                        }
                }
                if dio.NextToken == nil {
@@ -206,15 +272,10 @@ func (inst *ec2Instance) ProviderType() string {
 }
 
 func (inst *ec2Instance) SetTags(newTags cloud.InstanceTags) error {
-       ec2tags := []*ec2.Tag{
-               &ec2.Tag{
-                       Key:   aws.String(arvadosDispatchID),
-                       Value: aws.String(string(inst.provider.dispatcherID)),
-               },
-       }
+       var ec2tags []*ec2.Tag
        for k, v := range newTags {
                ec2tags = append(ec2tags, &ec2.Tag{
-                       Key:   aws.String(tagPrefix + k),
+                       Key:   aws.String(k),
                        Value: aws.String(v),
                })
        }
@@ -231,16 +292,13 @@ func (inst *ec2Instance) Tags() cloud.InstanceTags {
        tags := make(map[string]string)
 
        for _, t := range inst.instance.Tags {
-               if strings.HasPrefix(*t.Key, tagPrefix) {
-                       tags[(*t.Key)[len(tagPrefix):]] = *t.Value
-               }
+               tags[*t.Key] = *t.Value
        }
 
        return tags
 }
 
 func (inst *ec2Instance) Destroy() error {
-       log.Printf("terminating %v", *inst.instance.InstanceId)
        _, err := inst.provider.client.TerminateInstances(&ec2.TerminateInstancesInput{
                InstanceIds: []*string{inst.instance.InstanceId},
        })