//
// SPDX-License-Identifier: AGPL-3.0
-package main
+package keepproxy
import (
+ "context"
"errors"
- "flag"
"fmt"
"io"
"io/ioutil"
"net"
"net/http"
- "os"
- "os/signal"
"regexp"
"strings"
- "sync"
- "syscall"
"time"
- "git.arvados.org/arvados.git/lib/config"
+ "git.arvados.org/arvados.git/lib/service"
"git.arvados.org/arvados.git/sdk/go/arvados"
"git.arvados.org/arvados.git/sdk/go/arvadosclient"
+ "git.arvados.org/arvados.git/sdk/go/ctxlog"
"git.arvados.org/arvados.git/sdk/go/health"
"git.arvados.org/arvados.git/sdk/go/httpserver"
"git.arvados.org/arvados.git/sdk/go/keepclient"
- "github.com/coreos/go-systemd/daemon"
- "github.com/ghodss/yaml"
+ "git.arvados.org/arvados.git/services/keepstore"
"github.com/gorilla/mux"
- log "github.com/sirupsen/logrus"
-)
-
-var version = "dev"
-
-var (
- listener net.Listener
- router http.Handler
+ lru "github.com/hashicorp/golang-lru"
+ "github.com/prometheus/client_golang/prometheus"
+ "github.com/sirupsen/logrus"
)
const rfc3339NanoFixed = "2006-01-02T15:04:05.000000000Z07:00"
-func configure(logger log.FieldLogger, args []string) (*arvados.Cluster, error) {
- flags := flag.NewFlagSet(args[0], flag.ExitOnError)
-
- dumpConfig := flags.Bool("dump-config", false, "write current configuration to stdout and exit")
- getVersion := flags.Bool("version", false, "Print version information and exit.")
-
- loader := config.NewLoader(os.Stdin, logger)
- loader.SetupFlags(flags)
-
- args = loader.MungeLegacyConfigArgs(logger, args[1:], "-legacy-keepproxy-config")
- flags.Parse(args)
-
- // Print version information if requested
- if *getVersion {
- fmt.Printf("keepproxy %s\n", version)
- return nil, nil
- }
-
- cfg, err := loader.Load()
- if err != nil {
- return nil, err
- }
- cluster, err := cfg.GetCluster("")
- if err != nil {
- return nil, err
- }
-
- if *dumpConfig {
- out, err := yaml.Marshal(cfg)
- if err != nil {
- return nil, err
- }
- if _, err := os.Stdout.Write(out); err != nil {
- return nil, err
- }
- return nil, nil
- }
- return cluster, nil
-}
-
-func main() {
- logger := log.New()
- logger.Formatter = &log.JSONFormatter{
- TimestampFormat: rfc3339NanoFixed,
- }
-
- cluster, err := configure(logger, os.Args)
- if err != nil {
- log.Fatal(err)
- }
- if cluster == nil {
- return
- }
+var Command = service.Command(arvados.ServiceNameKeepproxy, newHandlerOrErrorHandler)
- log.Printf("keepproxy %s started", version)
-
- if err := run(logger, cluster); err != nil {
- log.Fatal(err)
- }
-
- log.Println("shutting down")
-}
-
-func run(logger log.FieldLogger, cluster *arvados.Cluster) error {
+func newHandlerOrErrorHandler(ctx context.Context, cluster *arvados.Cluster, token string, reg *prometheus.Registry) service.Handler {
client, err := arvados.NewClientFromConfig(cluster)
if err != nil {
- return err
+ return service.ErrorHandler(ctx, cluster, fmt.Errorf("Error setting up arvados client: %w", err))
}
- client.AuthToken = cluster.SystemRootToken
-
arv, err := arvadosclient.New(client)
if err != nil {
- return fmt.Errorf("Error setting up arvados client %v", err)
- }
-
- // If a config file is available, use the keepstores defined there
- // instead of the legacy autodiscover mechanism via the API server
- for k := range cluster.Services.Keepstore.InternalURLs {
- arv.KeepServiceURIs = append(arv.KeepServiceURIs, strings.TrimRight(k.String(), "/"))
- }
-
- if cluster.SystemLogs.LogLevel == "debug" {
- keepclient.DebugPrintf = log.Printf
+ return service.ErrorHandler(ctx, cluster, fmt.Errorf("Error setting up arvados client: %w", err))
}
kc, err := keepclient.MakeKeepClient(arv)
if err != nil {
- return fmt.Errorf("Error setting up keep client %v", err)
+ return service.ErrorHandler(ctx, cluster, fmt.Errorf("Error setting up keep client: %w", err))
}
keepclient.RefreshServiceDiscoveryOnSIGHUP()
-
- if cluster.Collections.DefaultReplication > 0 {
- kc.Want_replicas = cluster.Collections.DefaultReplication
- }
-
- var listen arvados.URL
- for listen = range cluster.Services.Keepproxy.InternalURLs {
- break
- }
-
- var lErr error
- listener, lErr = net.Listen("tcp", listen.Host)
- if lErr != nil {
- return fmt.Errorf("listen(%s): %v", listen.Host, lErr)
- }
-
- if _, err := daemon.SdNotify(false, "READY=1"); err != nil {
- log.Printf("Error notifying init daemon: %v", err)
+ router, err := newHandler(ctx, kc, time.Duration(keepclient.DefaultProxyRequestTimeout), cluster)
+ if err != nil {
+ return service.ErrorHandler(ctx, cluster, err)
}
- log.Println("listening at", listener.Addr())
-
- // Shut down the server gracefully (by closing the listener)
- // if SIGTERM is received.
- term := make(chan os.Signal, 1)
- go func(sig <-chan os.Signal) {
- s := <-sig
- log.Println("caught signal:", s)
- listener.Close()
- }(term)
- signal.Notify(term, syscall.SIGTERM)
- signal.Notify(term, syscall.SIGINT)
+ return router
+}
- // Start serving requests.
- router = MakeRESTRouter(kc, time.Duration(keepclient.DefaultProxyRequestTimeout), cluster.ManagementToken)
- return http.Serve(listener, httpserver.AddRequestIDs(httpserver.LogRequests(router)))
+type tokenCacheEntry struct {
+ expire int64
+ user *arvados.User
}
-type APITokenCache struct {
- tokens map[string]int64
- lock sync.Mutex
+type apiTokenCache struct {
+ tokens *lru.TwoQueueCache
expireTime int64
}
-// RememberToken caches the token and set an expire time. If we already have
-// an expire time on the token, it is not updated.
-func (cache *APITokenCache) RememberToken(token string) {
- cache.lock.Lock()
- defer cache.lock.Unlock()
-
+// RememberToken caches the token and set an expire time. If the
+// token is already in the cache, it is not updated.
+func (cache *apiTokenCache) RememberToken(token string, user *arvados.User) {
now := time.Now().Unix()
- if cache.tokens[token] == 0 {
- cache.tokens[token] = now + cache.expireTime
+ _, ok := cache.tokens.Get(token)
+ if !ok {
+ cache.tokens.Add(token, tokenCacheEntry{
+ expire: now + cache.expireTime,
+ user: user,
+ })
}
}
// RecallToken checks if the cached token is known and still believed to be
// valid.
-func (cache *APITokenCache) RecallToken(token string) bool {
- cache.lock.Lock()
- defer cache.lock.Unlock()
+func (cache *apiTokenCache) RecallToken(token string) (bool, *arvados.User) {
+ val, ok := cache.tokens.Get(token)
+ if !ok {
+ return false, nil
+ }
+ cacheEntry := val.(tokenCacheEntry)
now := time.Now().Unix()
- if cache.tokens[token] == 0 {
- // Unknown token
- return false
- } else if now < cache.tokens[token] {
+ if now < cacheEntry.expire {
// Token is known and still valid
- return true
+ return true, cacheEntry.user
} else {
// Token is expired
- cache.tokens[token] = 0
- return false
+ cache.tokens.Remove(token)
+ return false, nil
}
}
-// GetRemoteAddress returns a string with the remote address for the request.
-// If the X-Forwarded-For header is set and has a non-zero length, it returns a
-// string made from a comma separated list of all the remote addresses,
-// starting with the one(s) from the X-Forwarded-For header.
-func GetRemoteAddress(req *http.Request) string {
- if xff := req.Header.Get("X-Forwarded-For"); xff != "" {
- return xff + "," + req.RemoteAddr
- }
- return req.RemoteAddr
+func (h *proxyHandler) Done() <-chan struct{} {
+ return nil
+}
+
+func (h *proxyHandler) CheckHealth() error {
+ return nil
}
-func CheckAuthorizationHeader(kc *keepclient.KeepClient, cache *APITokenCache, req *http.Request) (pass bool, tok string) {
+func (h *proxyHandler) checkAuthorizationHeader(req *http.Request) (pass bool, tok string, user *arvados.User) {
parts := strings.SplitN(req.Header.Get("Authorization"), " ", 2)
if len(parts) < 2 || !(parts[0] == "OAuth2" || parts[0] == "Bearer") || len(parts[1]) == 0 {
- return false, ""
+ return false, "", nil
}
tok = parts[1]
op = "write"
}
- if cache.RecallToken(op + ":" + tok) {
+ if ok, user := h.apiTokenCache.RecallToken(op + ":" + tok); ok {
// Valid in the cache, short circuit
- return true, tok
+ return true, tok, user
}
var err error
- arv := *kc.Arvados
+ arv := *h.KeepClient.Arvados
arv.ApiToken = tok
arv.RequestID = req.Header.Get("X-Request-Id")
- if op == "read" {
- err = arv.Call("HEAD", "keep_services", "", "accessible", nil, nil)
- } else {
- err = arv.Call("HEAD", "users", "", "current", nil, nil)
+ user = &arvados.User{}
+ userCurrentError := arv.Call("GET", "users", "", "current", nil, user)
+ err = userCurrentError
+ if err != nil && op == "read" {
+ apiError, ok := err.(arvadosclient.APIServerError)
+ if ok && apiError.HttpStatusCode == http.StatusForbidden {
+ // If it was a scoped "sharing" token it will
+ // return 403 instead of 401 for the current
+ // user check. If it is a download operation
+ // and they have permission to read the
+ // keep_services table, we can allow it.
+ err = arv.Call("HEAD", "keep_services", "", "accessible", nil, nil)
+ }
}
if err != nil {
- log.Printf("%s: CheckAuthorizationHeader error: %v", GetRemoteAddress(req), err)
- return false, ""
+ ctxlog.FromContext(req.Context()).WithError(err).Info("checkAuthorizationHeader error")
+ return false, "", nil
+ }
+
+ if userCurrentError == nil && user.IsAdmin {
+ // checking userCurrentError is probably redundant,
+ // IsAdmin would be false anyway. But can't hurt.
+ if op == "read" && !h.cluster.Collections.KeepproxyPermission.Admin.Download {
+ return false, "", nil
+ }
+ if op == "write" && !h.cluster.Collections.KeepproxyPermission.Admin.Upload {
+ return false, "", nil
+ }
+ } else {
+ if op == "read" && !h.cluster.Collections.KeepproxyPermission.User.Download {
+ return false, "", nil
+ }
+ if op == "write" && !h.cluster.Collections.KeepproxyPermission.User.Upload {
+ return false, "", nil
+ }
}
// Success! Update cache
- cache.RememberToken(op + ":" + tok)
+ h.apiTokenCache.RememberToken(op+":"+tok, user)
- return true, tok
+ return true, tok, user
}
-// We need to make a private copy of the default http transport early
-// in initialization, then make copies of our private copy later. It
-// won't be safe to copy http.DefaultTransport itself later, because
-// its private mutexes might have already been used. (Without this,
-// the test suite sometimes panics "concurrent map writes" in
-// net/http.(*Transport).removeIdleConnLocked().)
-var defaultTransport = *(http.DefaultTransport.(*http.Transport))
+// We can't copy the default http transport because http.Transport has
+// a mutex field, so we make our own using the values of the exported
+// fields.
+var defaultTransport = http.Transport{
+ Proxy: http.DefaultTransport.(*http.Transport).Proxy,
+ DialContext: http.DefaultTransport.(*http.Transport).DialContext,
+ ForceAttemptHTTP2: http.DefaultTransport.(*http.Transport).ForceAttemptHTTP2,
+ MaxIdleConns: http.DefaultTransport.(*http.Transport).MaxIdleConns,
+ IdleConnTimeout: http.DefaultTransport.(*http.Transport).IdleConnTimeout,
+ TLSHandshakeTimeout: http.DefaultTransport.(*http.Transport).TLSHandshakeTimeout,
+ ExpectContinueTimeout: http.DefaultTransport.(*http.Transport).ExpectContinueTimeout,
+}
type proxyHandler struct {
http.Handler
*keepclient.KeepClient
- *APITokenCache
+ *apiTokenCache
timeout time.Duration
transport *http.Transport
+ cluster *arvados.Cluster
}
-// MakeRESTRouter returns an http.Handler that passes GET and PUT
-// requests to the appropriate handlers.
-func MakeRESTRouter(kc *keepclient.KeepClient, timeout time.Duration, mgmtToken string) http.Handler {
+func newHandler(ctx context.Context, kc *keepclient.KeepClient, timeout time.Duration, cluster *arvados.Cluster) (service.Handler, error) {
rest := mux.NewRouter()
- transport := defaultTransport
- transport.DialContext = (&net.Dialer{
- Timeout: keepclient.DefaultConnectTimeout,
- KeepAlive: keepclient.DefaultKeepAlive,
- DualStack: true,
- }).DialContext
- transport.TLSClientConfig = arvadosclient.MakeTLSConfig(kc.Arvados.ApiInsecure)
- transport.TLSHandshakeTimeout = keepclient.DefaultTLSHandshakeTimeout
+ // We can't copy the default http transport because
+ // http.Transport has a mutex field, so we copy the fields
+ // that we know have non-zero values in http.DefaultTransport.
+ transport := &http.Transport{
+ Proxy: http.DefaultTransport.(*http.Transport).Proxy,
+ ForceAttemptHTTP2: http.DefaultTransport.(*http.Transport).ForceAttemptHTTP2,
+ MaxIdleConns: http.DefaultTransport.(*http.Transport).MaxIdleConns,
+ IdleConnTimeout: http.DefaultTransport.(*http.Transport).IdleConnTimeout,
+ ExpectContinueTimeout: http.DefaultTransport.(*http.Transport).ExpectContinueTimeout,
+ DialContext: (&net.Dialer{
+ Timeout: keepclient.DefaultConnectTimeout,
+ KeepAlive: keepclient.DefaultKeepAlive,
+ DualStack: true,
+ }).DialContext,
+ TLSClientConfig: arvadosclient.MakeTLSConfig(kc.Arvados.ApiInsecure),
+ TLSHandshakeTimeout: keepclient.DefaultTLSHandshakeTimeout,
+ }
+
+ cacheQ, err := lru.New2Q(500)
+ if err != nil {
+ return nil, fmt.Errorf("Error from lru.New2Q: %v", err)
+ }
h := &proxyHandler{
Handler: rest,
KeepClient: kc,
timeout: timeout,
- transport: &transport,
- APITokenCache: &APITokenCache{
- tokens: make(map[string]int64),
+ transport: transport,
+ apiTokenCache: &apiTokenCache{
+ tokens: cacheQ,
expireTime: 300,
},
+ cluster: cluster,
}
rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Get).Methods("GET", "HEAD")
rest.HandleFunc(`/`, h.Options).Methods("OPTIONS")
rest.Handle("/_health/{check}", &health.Handler{
- Token: mgmtToken,
+ Token: cluster.ManagementToken,
Prefix: "/_health/",
}).Methods("GET")
- rest.NotFoundHandler = InvalidPathHandler{}
- return h
+ rest.NotFoundHandler = invalidPathHandler{}
+ return h, nil
}
var errLoopDetected = errors.New("loop detected")
-func (*proxyHandler) checkLoop(resp http.ResponseWriter, req *http.Request) error {
+func (h *proxyHandler) checkLoop(resp http.ResponseWriter, req *http.Request) error {
if via := req.Header.Get("Via"); strings.Index(via, " "+viaAlias) >= 0 {
- log.Printf("proxy loop detected (request has Via: %q): perhaps keepproxy is misidentified by gateway config as an external client, or its keep_services record does not have service_type=proxy?", via)
+ ctxlog.FromContext(req.Context()).Printf("proxy loop detected (request has Via: %q): perhaps keepproxy is misidentified by gateway config as an external client, or its keep_services record does not have service_type=proxy?", via)
http.Error(resp, errLoopDetected.Error(), http.StatusInternalServerError)
return errLoopDetected
}
return nil
}
-func SetCorsHeaders(resp http.ResponseWriter) {
- resp.Header().Set("Access-Control-Allow-Methods", "GET, HEAD, POST, PUT, OPTIONS")
- resp.Header().Set("Access-Control-Allow-Origin", "*")
- resp.Header().Set("Access-Control-Allow-Headers", "Authorization, Content-Length, Content-Type, X-Keep-Desired-Replicas")
- resp.Header().Set("Access-Control-Max-Age", "86486400")
+func setCORSHeaders(resp http.ResponseWriter) {
+ keepstore.SetCORSHeaders(resp)
+ acam := "Access-Control-Allow-Methods"
+ resp.Header().Set(acam, resp.Header().Get(acam)+", POST")
}
-type InvalidPathHandler struct{}
+type invalidPathHandler struct{}
-func (InvalidPathHandler) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
- log.Printf("%s: %s %s unroutable", GetRemoteAddress(req), req.Method, req.URL.Path)
+func (invalidPathHandler) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
http.Error(resp, "Bad request", http.StatusBadRequest)
}
func (h *proxyHandler) Options(resp http.ResponseWriter, req *http.Request) {
- log.Printf("%s: %s %s", GetRemoteAddress(req), req.Method, req.URL.Path)
- SetCorsHeaders(resp)
+ setCORSHeaders(resp)
}
-var errBadAuthorizationHeader = errors.New("Missing or invalid Authorization header")
+var errBadAuthorizationHeader = errors.New("Missing or invalid Authorization header, or method not allowed")
var errContentLengthMismatch = errors.New("Actual length != expected content length")
var errMethodNotSupported = errors.New("Method not supported")
if err := h.checkLoop(resp, req); err != nil {
return
}
- SetCorsHeaders(resp)
+ setCORSHeaders(resp)
resp.Header().Set("Via", req.Proto+" "+viaAlias)
locator := mux.Vars(req)["locator"]
var err error
var status int
var expectLength, responseLength int64
- var proxiedURI = "-"
+ logger := ctxlog.FromContext(req.Context())
defer func() {
- log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, responseLength, proxiedURI, err)
+ httpserver.SetResponseLogFields(req.Context(), logrus.Fields{
+ "locator": locator,
+ "expectLength": expectLength,
+ "responseLength": responseLength,
+ "err": err,
+ })
if status != http.StatusOK {
http.Error(resp, err.Error(), status)
}
}()
kc := h.makeKeepClient(req)
+ kc.DiskCacheSize = keepclient.DiskCacheDisabled
var pass bool
var tok string
- if pass, tok = CheckAuthorizationHeader(kc, h.APITokenCache, req); !pass {
+ var user *arvados.User
+ if pass, tok, user = h.checkAuthorizationHeader(req); !pass {
status, err = http.StatusForbidden, errBadAuthorizationHeader
return
}
+ httpserver.SetResponseLogFields(req.Context(), logrus.Fields{
+ "userUUID": user.UUID,
+ "userFullName": user.FullName,
+ })
// Copy ArvadosClient struct and use the client's API token
arvclient := *kc.Arvados
switch req.Method {
case "HEAD":
- expectLength, proxiedURI, err = kc.Ask(locator)
+ expectLength, _, err = kc.Ask(locator)
case "GET":
- reader, expectLength, proxiedURI, err = kc.Get(locator)
+ reader, expectLength, _, err = kc.Get(locator)
if reader != nil {
defer reader.Close()
}
}
if expectLength == -1 {
- log.Println("Warning:", GetRemoteAddress(req), req.Method, proxiedURI, "Content-Length not provided")
+ logger.Warn("Content-Length not provided")
}
switch respErr := err.(type) {
if err := h.checkLoop(resp, req); err != nil {
return
}
- SetCorsHeaders(resp)
+ setCORSHeaders(resp)
resp.Header().Set("Via", "HTTP/1.1 "+viaAlias)
kc := h.makeKeepClient(req)
var locatorOut string = "-"
defer func() {
- log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, kc.Want_replicas, wroteReplicas, locatorOut, err)
+ httpserver.SetResponseLogFields(req.Context(), logrus.Fields{
+ "expectLength": expectLength,
+ "wantReplicas": kc.Want_replicas,
+ "wroteReplicas": wroteReplicas,
+ "locator": strings.SplitN(locatorOut, "+A", 2)[0],
+ "err": err,
+ })
if status != http.StatusOK {
http.Error(resp, err.Error(), status)
}
locatorIn := mux.Vars(req)["locator"]
// Check if the client specified storage classes
- if req.Header.Get("X-Keep-Storage-Classes") != "" {
+ if req.Header.Get(keepclient.XKeepStorageClasses) != "" {
var scl []string
- for _, sc := range strings.Split(req.Header.Get("X-Keep-Storage-Classes"), ",") {
+ for _, sc := range strings.Split(req.Header.Get(keepclient.XKeepStorageClasses), ",") {
scl = append(scl, strings.Trim(sc, " "))
}
- kc.StorageClasses = scl
+ kc.SetStorageClasses(scl)
}
_, err = fmt.Sscanf(req.Header.Get("Content-Length"), "%d", &expectLength)
var pass bool
var tok string
- if pass, tok = CheckAuthorizationHeader(kc, h.APITokenCache, req); !pass {
+ var user *arvados.User
+ if pass, tok, user = h.checkAuthorizationHeader(req); !pass {
err = errBadAuthorizationHeader
status = http.StatusForbidden
return
}
+ httpserver.SetResponseLogFields(req.Context(), logrus.Fields{
+ "userUUID": user.UUID,
+ "userFullName": user.FullName,
+ })
// Copy ArvadosClient struct and use the client's API token
arvclient := *kc.Arvados
// ServeHTTP implementation for IndexHandler
// Supports only GET requests for /index/{prefix:[0-9a-f]{0,32}}
// For each keep server found in LocalRoots:
-// Invokes GetIndex using keepclient
-// Expects "complete" response (terminating with blank new line)
-// Aborts on any errors
+// - Invokes GetIndex using keepclient
+// - Expects "complete" response (terminating with blank new line)
+// - Aborts on any errors
// Concatenates responses from all those keep servers and returns
func (h *proxyHandler) Index(resp http.ResponseWriter, req *http.Request) {
- SetCorsHeaders(resp)
+ setCORSHeaders(resp)
prefix := mux.Vars(req)["prefix"]
var err error
}()
kc := h.makeKeepClient(req)
- ok, token := CheckAuthorizationHeader(kc, h.APITokenCache, req)
+ ok, token, _ := h.checkAuthorizationHeader(req)
if !ok {
status, err = http.StatusForbidden, errBadAuthorizationHeader
return
}
func (h *proxyHandler) makeKeepClient(req *http.Request) *keepclient.KeepClient {
- kc := *h.KeepClient
+ kc := h.KeepClient.Clone()
kc.RequestID = req.Header.Get("X-Request-Id")
kc.HTTPClient = &proxyClient{
client: &http.Client{
},
proto: req.Proto,
}
- return &kc
+ return kc
}
projects / arvados.git / blobdiff