http2_client.go

/*
 *
 * Copyright 2014 gRPC authors.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 */

package transport

import (
	"context"
	"fmt"
	"io"
	"math"
	"net"
	"net/http"
	"path/filepath"
	"strconv"
	"strings"
	"sync"
	"sync/atomic"
	"time"

	"golang.org/x/net/http2"
	"golang.org/x/net/http2/hpack"
	"google.golang.org/grpc/codes"
	"google.golang.org/grpc/credentials"
	"google.golang.org/grpc/internal/channelz"
	icredentials "google.golang.org/grpc/internal/credentials"
	"google.golang.org/grpc/internal/grpcutil"
	imetadata "google.golang.org/grpc/internal/metadata"
	"google.golang.org/grpc/internal/syscall"
	"google.golang.org/grpc/internal/transport/networktype"
	"google.golang.org/grpc/keepalive"
	"google.golang.org/grpc/metadata"
	"google.golang.org/grpc/peer"
	"google.golang.org/grpc/resolver"
	"google.golang.org/grpc/stats"
	"google.golang.org/grpc/status"
)

// clientConnectionCounter counts the number of connections a client has
// initiated (equal to the number of http2Clients created). Must be accessed
// atomically.
var clientConnectionCounter uint64

// http2Client implements the ClientTransport interface with HTTP2.
type http2Client struct {
	lastRead   int64 // Keep this field 64-bit aligned. Accessed atomically.
	ctx        context.Context
	cancel     context.CancelFunc
	ctxDone    <-chan struct{} // Cache the ctx.Done() chan.
	userAgent  string
	md         metadata.MD
	conn       net.Conn // underlying communication channel
	loopy      *loopyWriter
	remoteAddr net.Addr
	localAddr  net.Addr
	authInfo   credentials.AuthInfo // auth info about the connection

	readerDone chan struct{} // sync point to enable testing.
	writerDone chan struct{} // sync point to enable testing.
	// goAway is closed to notify the upper layer (i.e., addrConn.transportMonitor)
	// that the server sent GoAway on this transport.
	goAway chan struct{}

	framer *framer
	// controlBuf delivers all the control related tasks (e.g., window
	// updates, reset streams, and various settings) to the controller.
	// Do not access controlBuf with mu held.
	controlBuf *controlBuffer
	fc         *trInFlow
	// The scheme used: https if TLS is on, http otherwise.
	scheme string

	isSecure bool

	perRPCCreds []credentials.PerRPCCredentials

	kp               keepalive.ClientParameters
	keepaliveEnabled bool

	statsHandlers []stats.Handler

	initialWindowSize int32

	// configured by peer through SETTINGS_MAX_HEADER_LIST_SIZE
	maxSendHeaderListSize *uint32

	bdpEst *bdpEstimator
	// onPrefaceReceipt is a callback that client transport calls upon
	// receiving server preface to signal that a succefull HTTP2
	// connection was established.
	onPrefaceReceipt func()

	maxConcurrentStreams  uint32
	streamQuota           int64
	streamsQuotaAvailable chan struct{}
	waitingStreams        uint32
	nextID                uint32

	// Do not access controlBuf with mu held.
	mu            sync.Mutex // guard the following variables
	state         transportState
	activeStreams map[uint32]*Stream
	// prevGoAway ID records the Last-Stream-ID in the previous GOAway frame.
	prevGoAwayID uint32
	// goAwayReason records the http2.ErrCode and debug data received with the
	// GoAway frame.
	goAwayReason GoAwayReason
	// goAwayDebugMessage contains a detailed human readable string about a
	// GoAway frame, useful for error messages.
	goAwayDebugMessage string
	// A condition variable used to signal when the keepalive goroutine should
	// go dormant. The condition for dormancy is based on the number of active
	// streams and the `PermitWithoutStream` keepalive client parameter. And
	// since the number of active streams is guarded by the above mutex, we use
	// the same for this condition variable as well.
	kpDormancyCond *sync.Cond
	// A boolean to track whether the keepalive goroutine is dormant or not.
	// This is checked before attempting to signal the above condition
	// variable.
	kpDormant bool

	// Fields below are for channelz metric collection.
	channelzID *channelz.Identifier
	czData     *channelzData

	onGoAway func(GoAwayReason)
	onClose  func()

	bufferPool *bufferPool

	connectionID uint64
}

func dial(ctx context.Context, fn func(context.Context, string) (net.Conn, error), addr resolver.Address, useProxy bool, grpcUA string) (net.Conn, error) {
	address := addr.Addr
	networkType, ok := networktype.Get(addr)
	if fn != nil {
		// Special handling for unix scheme with custom dialer. Back in the day,
		// we did not have a unix resolver and therefore targets with a unix
		// scheme would end up using the passthrough resolver. So, user's used a
		// custom dialer in this case and expected the original dial target to
		// be passed to the custom dialer. Now, we have a unix resolver. But if
		// a custom dialer is specified, we want to retain the old behavior in
		// terms of the address being passed to the custom dialer.
		if networkType == "unix" && !strings.HasPrefix(address, "\x00") {
			// Supported unix targets are either "unix://absolute-path" or
			// "unix:relative-path".
			if filepath.IsAbs(address) {
				return fn(ctx, "unix://"+address)
			}
			return fn(ctx, "unix:"+address)
		}
		return fn(ctx, address)
	}
	if !ok {
		networkType, address = parseDialTarget(address)
	}
	if networkType == "tcp" && useProxy {
		return proxyDial(ctx, address, grpcUA)
	}
	return (&net.Dialer{}).DialContext(ctx, networkType, address)
}

func isTemporary(err error) bool {
	switch err := err.(type) {
	case interface {
		Temporary() bool
	}:
		return err.Temporary()
	case interface {
		Timeout() bool
	}:
		// Timeouts may be resolved upon retry, and are thus treated as
		// temporary.
		return err.Timeout()
	}
	return true
}

// newHTTP2Client constructs a connected ClientTransport to addr based on HTTP2
// and starts to receive messages on it. Non-nil error returns if construction
// fails.
func newHTTP2Client(connectCtx, ctx context.Context, addr resolver.Address, opts ConnectOptions, onPrefaceReceipt func(), onGoAway func(GoAwayReason), onClose func()) (_ *http2Client, err error) {
	scheme := "http"
	ctx, cancel := context.WithCancel(ctx)
	defer func() {
		if err != nil {
			cancel()
		}
	}()

	// gRPC, resolver, balancer etc. can specify arbitrary data in the
	// Attributes field of resolver.Address, which is shoved into connectCtx
	// and passed to the dialer and credential handshaker. This makes it possible for
	// address specific arbitrary data to reach custom dialers and credential handshakers.
	connectCtx = icredentials.NewClientHandshakeInfoContext(connectCtx, credentials.ClientHandshakeInfo{Attributes: addr.Attributes})

	conn, err := dial(connectCtx, opts.Dialer, addr, opts.UseProxy, opts.UserAgent)
	if err != nil {
		if opts.FailOnNonTempDialError {
			return nil, connectionErrorf(isTemporary(err), err, "transport: error while dialing: %v", err)
		}
		return nil, connectionErrorf(true, err, "transport: Error while dialing %v", err)
	}
	// Any further errors will close the underlying connection
	defer func(conn net.Conn) {
		if err != nil {
			conn.Close()
		}
	}(conn)
	kp := opts.KeepaliveParams
	// Validate keepalive parameters.
	if kp.Time == 0 {
		kp.Time = defaultClientKeepaliveTime
	}
	if kp.Timeout == 0 {
		kp.Timeout = defaultClientKeepaliveTimeout
	}
	keepaliveEnabled := false
	if kp.Time != infinity {
		if err = syscall.SetTCPUserTimeout(conn, kp.Timeout); err != nil {
			return nil, connectionErrorf(false, err, "transport: failed to set TCP_USER_TIMEOUT: %v", err)
		}
		keepaliveEnabled = true
	}
	var (
		isSecure bool
		authInfo credentials.AuthInfo
	)
	transportCreds := opts.TransportCredentials
	perRPCCreds := opts.PerRPCCredentials

	if b := opts.CredsBundle; b != nil {
		if t := b.TransportCredentials(); t != nil {
			transportCreds = t
		}
		if t := b.PerRPCCredentials(); t != nil {
			perRPCCreds = append(perRPCCreds, t)
		}
	}
	if transportCreds != nil {
		rawConn := conn
		// Pull the deadline from the connectCtx, which will be used for
		// timeouts in the authentication protocol handshake. Can ignore the
		// boolean as the deadline will return the zero value, which will make
		// the conn not timeout on I/O operations.
		deadline, _ := connectCtx.Deadline()
		rawConn.SetDeadline(deadline)
		conn, authInfo, err = transportCreds.ClientHandshake(connectCtx, addr.ServerName, rawConn)
		rawConn.SetDeadline(time.Time{})
		if err != nil {
			return nil, connectionErrorf(isTemporary(err), err, "transport: authentication handshake failed: %v", err)
		}
		for _, cd := range perRPCCreds {
			if cd.RequireTransportSecurity() {
				if ci, ok := authInfo.(interface {
					GetCommonAuthInfo() credentials.CommonAuthInfo
				}); ok {
					secLevel := ci.GetCommonAuthInfo().SecurityLevel
					if secLevel != credentials.InvalidSecurityLevel && secLevel < credentials.PrivacyAndIntegrity {
						return nil, connectionErrorf(true, nil, "transport: cannot send secure credentials on an insecure connection")
					}
				}
			}
		}
		isSecure = true
		if transportCreds.Info().SecurityProtocol == "tls" {
			scheme = "https"
		}
	}
	dynamicWindow := true
	icwz := int32(initialWindowSize)
	if opts.InitialConnWindowSize >= defaultWindowSize {
		icwz = opts.InitialConnWindowSize
		dynamicWindow = false
	}
	writeBufSize := opts.WriteBufferSize
	readBufSize := opts.ReadBufferSize
	maxHeaderListSize := defaultClientMaxHeaderListSize
	if opts.MaxHeaderListSize != nil {
		maxHeaderListSize = *opts.MaxHeaderListSize
	}
	t := &http2Client{
		ctx:                   ctx,
		ctxDone:               ctx.Done(), // Cache Done chan.
		cancel:                cancel,
		userAgent:             opts.UserAgent,
		conn:                  conn,
		remoteAddr:            conn.RemoteAddr(),
		localAddr:             conn.LocalAddr(),
		authInfo:              authInfo,
		readerDone:            make(chan struct{}),
		writerDone:            make(chan struct{}),
		goAway:                make(chan struct{}),
		framer:                newFramer(conn, writeBufSize, readBufSize, maxHeaderListSize),
		fc:                    &trInFlow{limit: uint32(icwz)},
		scheme:                scheme,
		activeStreams:         make(map[uint32]*Stream),
		isSecure:              isSecure,
		perRPCCreds:           perRPCCreds,
		kp:                    kp,
		statsHandlers:         opts.StatsHandlers,
		initialWindowSize:     initialWindowSize,
		onPrefaceReceipt:      onPrefaceReceipt,
		nextID:                1,
		maxConcurrentStreams:  defaultMaxStreamsClient,
		streamQuota:           defaultMaxStreamsClient,
		streamsQuotaAvailable: make(chan struct{}, 1),
		czData:                new(channelzData),
		onGoAway:              onGoAway,
		onClose:               onClose,
		keepaliveEnabled:      keepaliveEnabled,
		bufferPool:            newBufferPool(),
	}
	// Add peer information to the http2client context.
	t.ctx = peer.NewContext(t.ctx, t.getPeer())

	if md, ok := addr.Metadata.(*metadata.MD); ok {
		t.md = *md
	} else if md := imetadata.Get(addr); md != nil {
		t.md = md
	}
	t.controlBuf = newControlBuffer(t.ctxDone)
	if opts.InitialWindowSize >= defaultWindowSize {
		t.initialWindowSize = opts.InitialWindowSize
		dynamicWindow = false
	}
	if dynamicWindow {
		t.bdpEst = &bdpEstimator{
			bdp:               initialWindowSize,
			updateFlowControl: t.updateFlowControl,
		}
	}
	for _, sh := range t.statsHandlers {
		t.ctx = sh.TagConn(t.ctx, &stats.ConnTagInfo{
			RemoteAddr: t.remoteAddr,
			LocalAddr:  t.localAddr,
		})
		connBegin := &stats.ConnBegin{
			Client: true,
		}
		sh.HandleConn(t.ctx, connBegin)
	}
	t.channelzID, err = channelz.RegisterNormalSocket(t, opts.ChannelzParentID, fmt.Sprintf("%s -> %s", t.localAddr, t.remoteAddr))
	if err != nil {
		return nil, err
	}
	if t.keepaliveEnabled {
		t.kpDormancyCond = sync.NewCond(&t.mu)
		go t.keepalive()
	}
	// Start the reader goroutine for incoming message. Each transport has
	// a dedicated goroutine which reads HTTP2 frame from network. Then it
	// dispatches the frame to the corresponding stream entity.
	go t.reader()

	// Send connection preface to server.
	n, err := t.conn.Write(clientPreface)
	if err != nil {
		err = connectionErrorf(true, err, "transport: failed to write client preface: %v", err)
		t.Close(err)
		return nil, err
	}
	if n != len(clientPreface) {
		err = connectionErrorf(true, nil, "transport: preface mismatch, wrote %d bytes; want %d", n, len(clientPreface))
		t.Close(err)
		return nil, err
	}
	var ss []http2.Setting

	if t.initialWindowSize != defaultWindowSize {
		ss = append(ss, http2.Setting{
			ID:  http2.SettingInitialWindowSize,
			Val: uint32(t.initialWindowSize),
		})
	}
	if opts.MaxHeaderListSize != nil {
		ss = append(ss, http2.Setting{
			ID:  http2.SettingMaxHeaderListSize,
			Val: *opts.MaxHeaderListSize,
		})
	}
	err = t.framer.fr.WriteSettings(ss...)
	if err != nil {
		err = connectionErrorf(true, err, "transport: failed to write initial settings frame: %v", err)
		t.Close(err)
		return nil, err
	}
	// Adjust the connection flow control window if needed.
	if delta := uint32(icwz - defaultWindowSize); delta > 0 {
		if err := t.framer.fr.WriteWindowUpdate(0, delta); err != nil {
			err = connectionErrorf(true, err, "transport: failed to write window update: %v", err)
			t.Close(err)
			return nil, err
		}
	}

	t.connectionID = atomic.AddUint64(&clientConnectionCounter, 1)

	if err := t.framer.writer.Flush(); err != nil {
		return nil, err
	}
	go func() {
		t.loopy = newLoopyWriter(clientSide, t.framer, t.controlBuf, t.bdpEst)
		err := t.loopy.run()
		if err != nil {
			if logger.V(logLevel) {
				logger.Errorf("transport: loopyWriter.run returning. Err: %v", err)
			}
		}
		// Do not close the transport.  Let reader goroutine handle it since
		// there might be data in the buffers.
		t.conn.Close()
		t.controlBuf.finish()
		close(t.writerDone)
	}()
	return t, nil
}

func (t *http2Client) newStream(ctx context.Context, callHdr *CallHdr) *Stream {
	// TODO(zhaoq): Handle uint32 overflow of Stream.id.
	s := &Stream{
		ct:             t,
		done:           make(chan struct{}),
		method:         callHdr.Method,
		sendCompress:   callHdr.SendCompress,
		buf:            newRecvBuffer(),
		headerChan:     make(chan struct{}),
		contentSubtype: callHdr.ContentSubtype,
		doneFunc:       callHdr.DoneFunc,
	}
	s.wq = newWriteQuota(defaultWriteQuota, s.done)
	s.requestRead = func(n int) {
		t.adjustWindow(s, uint32(n))
	}
	// The client side stream context should have exactly the same life cycle with the user provided context.
	// That means, s.ctx should be read-only. And s.ctx is done iff ctx is done.
	// So we use the original context here instead of creating a copy.
	s.ctx = ctx
	s.trReader = &transportReader{
		reader: &recvBufferReader{
			ctx:     s.ctx,
			ctxDone: s.ctx.Done(),
			recv:    s.buf,
			closeStream: func(err error) {
				t.CloseStream(s, err)
			},
			freeBuffer: t.bufferPool.put,
		},
		windowHandler: func(n int) {
			t.updateWindow(s, uint32(n))
		},
	}
	return s
}

func (t *http2Client) getPeer() *peer.Peer {
	return &peer.Peer{
		Addr:     t.remoteAddr,
		AuthInfo: t.authInfo, // Can be nil
	}
}

func (t *http2Client) createHeaderFields(ctx context.Context, callHdr *CallHdr) ([]hpack.HeaderField, error) {
	aud := t.createAudience(callHdr)
	ri := credentials.RequestInfo{
		Method:   callHdr.Method,
		AuthInfo: t.authInfo,
	}
	ctxWithRequestInfo := icredentials.NewRequestInfoContext(ctx, ri)
	authData, err := t.getTrAuthData(ctxWithRequestInfo, aud)
	if err != nil {
		return nil, err
	}
	callAuthData, err := t.getCallAuthData(ctxWithRequestInfo, aud, callHdr)
	if err != nil {
		return nil, err
	}
	// TODO(mmukhi): Benchmark if the performance gets better if count the metadata and other header fields
	// first and create a slice of that exact size.
	// Make the slice of certain predictable size to reduce allocations made by append.
	hfLen := 7 // :method, :scheme, :path, :authority, content-type, user-agent, te
	hfLen += len(authData) + len(callAuthData)
	headerFields := make([]hpack.HeaderField, 0, hfLen)
	headerFields = append(headerFields, hpack.HeaderField{Name: ":method", Value: "POST"})
	headerFields = append(headerFields, hpack.HeaderField{Name: ":scheme", Value: t.scheme})
	headerFields = append(headerFields, hpack.HeaderField{Name: ":path", Value: callHdr.Method})
	headerFields = append(headerFields, hpack.HeaderField{Name: ":authority", Value: callHdr.Host})
	headerFields = append(headerFields, hpack.HeaderField{Name: "content-type", Value: grpcutil.ContentType(callHdr.ContentSubtype)})
	headerFields = append(headerFields, hpack.HeaderField{Name: "user-agent", Value: t.userAgent})
	headerFields = append(headerFields, hpack.HeaderField{Name: "te", Value: "trailers"})
	if callHdr.PreviousAttempts > 0 {
		headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-previous-rpc-attempts", Value: strconv.Itoa(callHdr.PreviousAttempts)})
	}

	if callHdr.SendCompress != "" {
		headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-encoding", Value: callHdr.SendCompress})
		headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-accept-encoding", Value: callHdr.SendCompress})
	}
	if dl, ok := ctx.Deadline(); ok {
		// Send out timeout regardless its value. The server can detect timeout context by itself.
		// TODO(mmukhi): Perhaps this field should be updated when actually writing out to the wire.
		timeout := time.Until(dl)
		headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-timeout", Value: grpcutil.EncodeDuration(timeout)})
	}
	for k, v := range authData {
		headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)})
	}
	for k, v := range callAuthData {
		headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)})
	}
	if b := stats.OutgoingTags(ctx); b != nil {
		headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-tags-bin", Value: encodeBinHeader(b)})
	}
	if b := stats.OutgoingTrace(ctx); b != nil {
		headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-trace-bin", Value: encodeBinHeader(b)})
	}

	if md, added, ok := metadata.FromOutgoingContextRaw(ctx); ok {
		var k string
		for k, vv := range md {
			// HTTP doesn't allow you to set pseudoheaders after non pseudoheaders were set.
			if isReservedHeader(k) {
				continue
			}
			for _, v := range vv {
				headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)})
			}
		}
		for _, vv := range added {
			for i, v := range vv {
				if i%2 == 0 {
					k = strings.ToLower(v)
					continue
				}
				// HTTP doesn't allow you to set pseudoheaders after non pseudoheaders were set.
				if isReservedHeader(k) {
					continue
				}
				headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)})
			}
		}
	}
	for k, vv := range t.md {
		if isReservedHeader(k) {
			continue
		}
		for _, v := range vv {
			headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)})
		}
	}
	return headerFields, nil
}

func (t *http2Client) createAudience(callHdr *CallHdr) string {
	// Create an audience string only if needed.
	if len(t.perRPCCreds) == 0 && callHdr.Creds == nil {
		return ""
	}
	// Construct URI required to get auth request metadata.
	// Omit port if it is the default one.
	host := strings.TrimSuffix(callHdr.Host, ":443")
	pos := strings.LastIndex(callHdr.Method, "/")
	if pos == -1 {
		pos = len(callHdr.Method)
	}
	return "https://" + host + callHdr.Method[:pos]
}

func (t *http2Client) getTrAuthData(ctx context.Context, audience string) (map[string]string, error) {
	if len(t.perRPCCreds) == 0 {
		return nil, nil
	}
	authData := map[string]string{}
	for _, c := range t.perRPCCreds {
		data, err := c.GetRequestMetadata(ctx, audience)
		if err != nil {
			if _, ok := status.FromError(err); ok {
				return nil, err
			}

			return nil, status.Errorf(codes.Unauthenticated, "transport: per-RPC creds failed due to error: %v", err)
		}
		for k, v := range data {
			// Capital header names are illegal in HTTP/2.
			k = strings.ToLower(k)
			authData[k] = v
		}
	}
	return authData, nil
}

func (t *http2Client) getCallAuthData(ctx context.Context, audience string, callHdr *CallHdr) (map[string]string, error) {
	var callAuthData map[string]string
	// Check if credentials.PerRPCCredentials were provided via call options.
	// Note: if these credentials are provided both via dial options and call
	// options, then both sets of credentials will be applied.
	if callCreds := callHdr.Creds; callCreds != nil {
		if callCreds.RequireTransportSecurity() {
			ri, _ := credentials.RequestInfoFromContext(ctx)
			if !t.isSecure || credentials.CheckSecurityLevel(ri.AuthInfo, credentials.PrivacyAndIntegrity) != nil {
				return nil, status.Error(codes.Unauthenticated, "transport: cannot send secure credentials on an insecure connection")
			}
		}
		data, err := callCreds.GetRequestMetadata(ctx, audience)
		if err != nil {
			return nil, status.Errorf(codes.Internal, "transport: %v", err)
		}
		callAuthData = make(map[string]string, len(data))
		for k, v := range data {
			// Capital header names are illegal in HTTP/2
			k = strings.ToLower(k)
			callAuthData[k] = v
		}
	}
	return callAuthData, nil
}

// NewStreamError wraps an error and reports additional information.  Typically
// NewStream errors result in transparent retry, as they mean nothing went onto
// the wire.  However, there are two notable exceptions:
//
// 1. If the stream headers violate the max header list size allowed by the
//    server.  It's possible this could succeed on another transport, even if
//    it's unlikely, but do not transparently retry.
// 2. If the credentials errored when requesting their headers.  In this case,
//    it's possible a retry can fix the problem, but indefinitely transparently
//    retrying is not appropriate as it is likely the credentials, if they can
//    eventually succeed, would need I/O to do so.
type NewStreamError struct {
	Err error

	AllowTransparentRetry bool
}

func (e NewStreamError) Error() string {
	return e.Err.Error()
}

// NewStream creates a stream and registers it into the transport as "active"
// streams.  All non-nil errors returned will be *NewStreamError.
func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (*Stream, error) {
	ctx = peer.NewContext(ctx, t.getPeer())
	headerFields, err := t.createHeaderFields(ctx, callHdr)
	if err != nil {
		return nil, &NewStreamError{Err: err, AllowTransparentRetry: false}
	}
	s := t.newStream(ctx, callHdr)
	cleanup := func(err error) {
		if s.swapState(streamDone) == streamDone {
			// If it was already done, return.
			return
		}
		// The stream was unprocessed by the server.
		atomic.StoreUint32(&s.unprocessed, 1)
		s.write(recvMsg{err: err})
		close(s.done)
		// If headerChan isn't closed, then close it.
		if atomic.CompareAndSwapUint32(&s.headerChanClosed, 0, 1) {
			close(s.headerChan)
		}
	}
	hdr := &headerFrame{
		hf:        headerFields,
		endStream: false,
		initStream: func(id uint32) error {
			t.mu.Lock()
			if state := t.state; state != reachable {
				t.mu.Unlock()
				// Do a quick cleanup.
				err := error(errStreamDrain)
				if state == closing {
					err = ErrConnClosing
				}
				cleanup(err)
				return err
			}
			if channelz.IsOn() {
				atomic.AddInt64(&t.czData.streamsStarted, 1)
				atomic.StoreInt64(&t.czData.lastStreamCreatedTime, time.Now().UnixNano())
			}
			// If the keepalive goroutine has gone dormant, wake it up.
			if t.kpDormant {
				t.kpDormancyCond.Signal()
			}
			t.mu.Unlock()
			return nil
		},
		onOrphaned: cleanup,
		wq:         s.wq,
	}
	firstTry := true
	var ch chan struct{}
	checkForStreamQuota := func(it interface{}) bool {
		if t.streamQuota <= 0 { // Can go negative if server decreases it.
			if firstTry {
				t.waitingStreams++
			}
			ch = t.streamsQuotaAvailable
			return false
		}
		if !firstTry {
			t.waitingStreams--
		}
		t.streamQuota--
		h := it.(*headerFrame)
		h.streamID = t.nextID
		t.nextID += 2
		s.id = h.streamID
		s.fc = &inFlow{limit: uint32(t.initialWindowSize)}
		t.mu.Lock()
		if t.activeStreams == nil { // Can be niled from Close().
			t.mu.Unlock()
			return false // Don't create a stream if the transport is already closed.
		}
		t.activeStreams[s.id] = s
		t.mu.Unlock()
		if t.streamQuota > 0 && t.waitingStreams > 0 {
			select {
			case t.streamsQuotaAvailable <- struct{}{}:
			default:
			}
		}
		return true
	}
	var hdrListSizeErr error
	checkForHeaderListSize := func(it interface{}) bool {
		if t.maxSendHeaderListSize == nil {
			return true
		}
		hdrFrame := it.(*headerFrame)
		var sz int64
		for _, f := range hdrFrame.hf {
			if sz += int64(f.Size()); sz > int64(*t.maxSendHeaderListSize) {
				hdrListSizeErr = status.Errorf(codes.Internal, "header list size to send violates the maximum size (%d bytes) set by server", *t.maxSendHeaderListSize)
				return false
			}
		}
		return true
	}
	for {
		success, err := t.controlBuf.executeAndPut(func(it interface{}) bool {
			return checkForHeaderListSize(it) && checkForStreamQuota(it)
		}, hdr)
		if err != nil {
			// Connection closed.
			return nil, &NewStreamError{Err: err, AllowTransparentRetry: true}
		}
		if success {
			break
		}
		if hdrListSizeErr != nil {
			return nil, &NewStreamError{Err: hdrListSizeErr}
		}
		firstTry = false
		select {
		case <-ch:
		case <-ctx.Done():
			return nil, &NewStreamError{Err: ContextErr(ctx.Err())}
		case <-t.goAway:
			return nil, &NewStreamError{Err: errStreamDrain, AllowTransparentRetry: true}
		case <-t.ctx.Done():
			return nil, &NewStreamError{Err: ErrConnClosing, AllowTransparentRetry: true}
		}
	}
	if len(t.statsHandlers) != 0 {
		header, ok := metadata.FromOutgoingContext(ctx)
		if ok {
			header.Set("user-agent", t.userAgent)
		} else {
			header = metadata.Pairs("user-agent", t.userAgent)
		}
		for _, sh := range t.statsHandlers {
			// Note: The header fields are compressed with hpack after this call returns.
			// No WireLength field is set here.
			// Note: Creating a new stats object to prevent pollution.
			outHeader := &stats.OutHeader{
				Client:      true,
				FullMethod:  callHdr.Method,
				RemoteAddr:  t.remoteAddr,
				LocalAddr:   t.localAddr,
				Compression: callHdr.SendCompress,
				Header:      header,
			}
			sh.HandleRPC(s.ctx, outHeader)
		}
	}
	return s, nil
}

// CloseStream clears the footprint of a stream when the stream is not needed any more.
// This must not be executed in reader's goroutine.
func (t *http2Client) CloseStream(s *Stream, err error) {
	var (
		rst     bool
		rstCode http2.ErrCode
	)
	if err != nil {
		rst = true
		rstCode = http2.ErrCodeCancel
	}
	t.closeStream(s, err, rst, rstCode, status.Convert(err), nil, false)
}

func (t *http2Client) closeStream(s *Stream, err error, rst bool, rstCode http2.ErrCode, st *status.Status, mdata map[string][]string, eosReceived bool) {
	// Set stream status to done.
	if s.swapState(streamDone) == streamDone {
		// If it was already done, return.  If multiple closeStream calls
		// happen simultaneously, wait for the first to finish.
		<-s.done
		return
	}
	// status and trailers can be updated here without any synchronization because the stream goroutine will
	// only read it after it sees an io.EOF error from read or write and we'll write those errors
	// only after updating this.
	s.status = st
	if len(mdata) > 0 {
		s.trailer = mdata
	}
	if err != nil {
		// This will unblock reads eventually.
		s.write(recvMsg{err: err})
	}
	// If headerChan isn't closed, then close it.
	if atomic.CompareAndSwapUint32(&s.headerChanClosed, 0, 1) {
		s.noHeaders = true
		close(s.headerChan)
	}
	cleanup := &cleanupStream{
		streamID: s.id,
		onWrite: func() {
			t.mu.Lock()
			if t.activeStreams != nil {
				delete(t.activeStreams, s.id)
			}
			t.mu.Unlock()
			if channelz.IsOn() {
				if eosReceived {
					atomic.AddInt64(&t.czData.streamsSucceeded, 1)
				} else {
					atomic.AddInt64(&t.czData.streamsFailed, 1)
				}
			}
		},
		rst:     rst,
		rstCode: rstCode,
	}
	addBackStreamQuota := func(interface{}) bool {
		t.streamQuota++
		if t.streamQuota > 0 && t.waitingStreams > 0 {
			select {
			case t.streamsQuotaAvailable <- struct{}{}:
			default:
			}
		}
		return true
	}
	t.controlBuf.executeAndPut(addBackStreamQuota, cleanup)
	// This will unblock write.
	close(s.done)
	if s.doneFunc != nil {
		s.doneFunc()
	}
}

// Close kicks off the shutdown process of the transport. This should be called
// only once on a transport. Once it is called, the transport should not be
// accessed any more.
//
// This method blocks until the addrConn that initiated this transport is
// re-connected. This happens because t.onClose() begins reconnect logic at the
// addrConn level and blocks until the addrConn is successfully connected.
func (t *http2Client) Close(err error) {
	t.mu.Lock()
	// Make sure we only Close once.
	if t.state == closing {
		t.mu.Unlock()
		return
	}
	// Call t.onClose before setting the state to closing to prevent the client
	// from attempting to create new streams ASAP.
	t.onClose()
	t.state = closing
	streams := t.activeStreams
	t.activeStreams = nil
	if t.kpDormant {
		// If the keepalive goroutine is blocked on this condition variable, we
		// should unblock it so that the goroutine eventually exits.
		t.kpDormancyCond.Signal()
	}
	t.mu.Unlock()
	t.controlBuf.finish()
	t.cancel()
	t.conn.Close()
	channelz.RemoveEntry(t.channelzID)
	// Append info about previous goaways if there were any, since this may be important
	// for understanding the root cause for this connection to be closed.
	_, goAwayDebugMessage := t.GetGoAwayReason()

	var st *status.Status
	if len(goAwayDebugMessage) > 0 {
		st = status.Newf(codes.Unavailable, "closing transport due to: %v, received prior goaway: %v", err, goAwayDebugMessage)
		err = st.Err()
	} else {
		st = status.New(codes.Unavailable, err.Error())
	}

	// Notify all active streams.
	for _, s := range streams {
		t.closeStream(s, err, false, http2.ErrCodeNo, st, nil, false)
	}
	for _, sh := range t.statsHandlers {
		connEnd := &stats.ConnEnd{
			Client: true,
		}
		sh.HandleConn(t.ctx, connEnd)
	}
}

// GracefulClose sets the state to draining, which prevents new streams from
// being created and causes the transport to be closed when the last active
// stream is closed.  If there are no active streams, the transport is closed
// immediately.  This does nothing if the transport is already draining or
// closing.
func (t *http2Client) GracefulClose() {
	t.mu.Lock()
	// Make sure we move to draining only from active.
	if t.state == draining || t.state == closing {
		t.mu.Unlock()
		return
	}
	t.state = draining
	active := len(t.activeStreams)
	t.mu.Unlock()
	if active == 0 {
		t.Close(ErrConnClosing)
		return
	}
	t.controlBuf.put(&incomingGoAway{})
}

// Write formats the data into HTTP2 data frame(s) and sends it out. The caller
// should proceed only if Write returns nil.
func (t *http2Client) Write(s *Stream, hdr []byte, data []byte, opts *Options) error {
	if opts.Last {
		// If it's the last message, update stream state.
		if !s.compareAndSwapState(streamActive, streamWriteDone) {
			return errStreamDone
		}
	} else if s.getState() != streamActive {
		return errStreamDone
	}
	df := &dataFrame{
		streamID:  s.id,
		endStream: opts.Last,
		h:         hdr,
		d:         data,
	}
	if hdr != nil || data != nil { // If it's not an empty data frame, check quota.
		if err := s.wq.get(int32(len(hdr) + len(data))); err != nil {
			return err
		}
	}
	return t.controlBuf.put(df)
}

func (t *http2Client) getStream(f http2.Frame) *Stream {
	t.mu.Lock()
	s := t.activeStreams[f.Header().StreamID]
	t.mu.Unlock()
	return s
}

// adjustWindow sends out extra window update over the initial window size
// of stream if the application is requesting data larger in size than
// the window.
func (t *http2Client) adjustWindow(s *Stream, n uint32) {
	if w := s.fc.maybeAdjust(n); w > 0 {
		t.controlBuf.put(&outgoingWindowUpdate{streamID: s.id, increment: w})
	}
}

// updateWindow adjusts the inbound quota for the stream.
// Window updates will be sent out when the cumulative quota
// exceeds the corresponding threshold.
func (t *http2Client) updateWindow(s *Stream, n uint32) {
	if w := s.fc.onRead(n); w > 0 {
		t.controlBuf.put(&outgoingWindowUpdate{streamID: s.id, increment: w})
	}
}

// updateFlowControl updates the incoming flow control windows
// for the transport and the stream based on the current bdp
// estimation.
func (t *http2Client) updateFlowControl(n uint32) {
	updateIWS := func(interface{}) bool {
		t.initialWindowSize = int32(n)
		t.mu.Lock()
		for _, s := range t.activeStreams {
			s.fc.newLimit(n)
		}
		t.mu.Unlock()
		return true
	}
	t.controlBuf.executeAndPut(updateIWS, &outgoingWindowUpdate{streamID: 0, increment: t.fc.newLimit(n)})
	t.controlBuf.put(&outgoingSettings{
		ss: []http2.Setting{
			{
				ID:  http2.SettingInitialWindowSize,
				Val: n,
			},
		},
	})
}

func (t *http2Client) handleData(f *http2.DataFrame) {
	size := f.Header().Length
	var sendBDPPing bool
	if t.bdpEst != nil {
		sendBDPPing = t.bdpEst.add(size)
	}
	// Decouple connection's flow control from application's read.
	// An update on connection's flow control should not depend on
	// whether user application has read the data or not. Such a
	// restriction is already imposed on the stream's flow control,
	// and therefore the sender will be blocked anyways.
	// Decoupling the connection flow control will prevent other
	// active(fast) streams from starving in presence of slow or
	// inactive streams.
	//
	if w := t.fc.onData(size); w > 0 {
		t.controlBuf.put(&outgoingWindowUpdate{
			streamID:  0,
			increment: w,
		})
	}
	if sendBDPPing {
		// Avoid excessive ping detection (e.g. in an L7 proxy)
		// by sending a window update prior to the BDP ping.

		if w := t.fc.reset(); w > 0 {
			t.controlBuf.put(&outgoingWindowUpdate{
				streamID:  0,
				increment: w,
			})
		}

		t.controlBuf.put(bdpPing)
	}
	// Select the right stream to dispatch.
	s := t.getStream(f)
	if s == nil {
		return
	}
	if size > 0 {
		if err := s.fc.onData(size); err != nil {
			t.closeStream(s, io.EOF, true, http2.ErrCodeFlowControl, status.New(codes.Internal, err.Error()), nil, false)
			return
		}
		if f.Header().Flags.Has(http2.FlagDataPadded) {
			if w := s.fc.onRead(size - uint32(len(f.Data()))); w > 0 {
				t.controlBuf.put(&outgoingWindowUpdate{s.id, w})
			}
		}
		// TODO(bradfitz, zhaoq): A copy is required here because there is no
		// guarantee f.Data() is consumed before the arrival of next frame.
		// Can this copy be eliminated?
		if len(f.Data()) > 0 {
			buffer := t.bufferPool.get()
			buffer.Reset()
			buffer.Write(f.Data())
			s.write(recvMsg{buffer: buffer})
		}
	}
	// The server has closed the stream without sending trailers.  Record that
	// the read direction is closed, and set the status appropriately.
	if f.StreamEnded() {
		t.closeStream(s, io.EOF, false, http2.ErrCodeNo, status.New(codes.Internal, "server closed the stream without sending trailers"), nil, true)
	}
}

func (t *http2Client) handleRSTStream(f *http2.RSTStreamFrame) {
	s := t.getStream(f)
	if s == nil {
		return
	}
	if f.ErrCode == http2.ErrCodeRefusedStream {
		// The stream was unprocessed by the server.
		atomic.StoreUint32(&s.unprocessed, 1)
	}
	statusCode, ok := http2ErrConvTab[f.ErrCode]
	if !ok {
		if logger.V(logLevel) {
			logger.Warningf("transport: http2Client.handleRSTStream found no mapped gRPC status for the received http2 error %v", f.ErrCode)
		}
		statusCode = codes.Unknown
	}
	if statusCode == codes.Canceled {
		if d, ok := s.ctx.Deadline(); ok && !d.After(time.Now()) {
			// Our deadline was already exceeded, and that was likely the cause
			// of this cancelation.  Alter the status code accordingly.
			statusCode = codes.DeadlineExceeded
		}
	}
	t.closeStream(s, io.EOF, false, http2.ErrCodeNo, status.Newf(statusCode, "stream terminated by RST_STREAM with error code: %v", f.ErrCode), nil, false)
}

func (t *http2Client) handleSettings(f *http2.SettingsFrame, isFirst bool) {
	if f.IsAck() {
		return
	}
	var maxStreams *uint32
	var ss []http2.Setting
	var updateFuncs []func()
	f.ForeachSetting(func(s http2.Setting) error {
		switch s.ID {
		case http2.SettingMaxConcurrentStreams:
			maxStreams = new(uint32)
			*maxStreams = s.Val
		case http2.SettingMaxHeaderListSize:
			updateFuncs = append(updateFuncs, func() {
				t.maxSendHeaderListSize = new(uint32)
				*t.maxSendHeaderListSize = s.Val
			})
		default:
			ss = append(ss, s)
		}
		return nil
	})
	if isFirst && maxStreams == nil {
		maxStreams = new(uint32)
		*maxStreams = math.MaxUint32
	}
	sf := &incomingSettings{
		ss: ss,
	}
	if maxStreams != nil {
		updateStreamQuota := func() {
			delta := int64(*maxStreams) - int64(t.maxConcurrentStreams)
			t.maxConcurrentStreams = *maxStreams
			t.streamQuota += delta
			if delta > 0 && t.waitingStreams > 0 {
				close(t.streamsQuotaAvailable) // wake all of them up.
				t.streamsQuotaAvailable = make(chan struct{}, 1)
			}
		}
		updateFuncs = append(updateFuncs, updateStreamQuota)
	}
	t.controlBuf.executeAndPut(func(interface{}) bool {
		for _, f := range updateFuncs {
			f()
		}
		return true
	}, sf)
}

func (t *http2Client) handlePing(f *http2.PingFrame) {
	if f.IsAck() {
		// Maybe it's a BDP ping.
		if t.bdpEst != nil {
			t.bdpEst.calculate(f.Data)
		}
		return
	}
	pingAck := &ping{ack: true}
	copy(pingAck.data[:], f.Data[:])
	t.controlBuf.put(pingAck)
}

func (t *http2Client) handleGoAway(f *http2.GoAwayFrame) {
	t.mu.Lock()
	if t.state == closing {
		t.mu.Unlock()
		return
	}
	if f.ErrCode == http2.ErrCodeEnhanceYourCalm {
		if logger.V(logLevel) {
			logger.Infof("Client received GoAway with http2.ErrCodeEnhanceYourCalm.")
		}
	}
	id := f.LastStreamID
	if id > 0 && id%2 == 0 {
		t.mu.Unlock()
		t.Close(connectionErrorf(true, nil, "received goaway with non-zero even-numbered numbered stream id: %v", id))
		return
	}
	// A client can receive multiple GoAways from the server (see
	// https://github.com/grpc/grpc-go/issues/1387).  The idea is that the first
	// GoAway will be sent with an ID of MaxInt32 and the second GoAway will be
	// sent after an RTT delay with the ID of the last stream the server will
	// process.
	//
	// Therefore, when we get the first GoAway we don't necessarily close any
	// streams. While in case of second GoAway we close all streams created after
	// the GoAwayId. This way streams that were in-flight while the GoAway from
	// server was being sent don't get killed.
	select {
	case <-t.goAway: // t.goAway has been closed (i.e.,multiple GoAways).
		// If there are multiple GoAways the first one should always have an ID greater than the following ones.
		if id > t.prevGoAwayID {
			t.mu.Unlock()
			t.Close(connectionErrorf(true, nil, "received goaway with stream id: %v, which exceeds stream id of previous goaway: %v", id, t.prevGoAwayID))
			return
		}
	default:
		t.setGoAwayReason(f)
		close(t.goAway)
		defer t.controlBuf.put(&incomingGoAway{}) // Defer as t.mu is currently held.
		// Notify the clientconn about the GOAWAY before we set the state to
		// draining, to allow the client to stop attempting to create streams
		// before disallowing new streams on this connection.
		t.onGoAway(t.goAwayReason)
		t.state = draining
	}
	// All streams with IDs greater than the GoAwayId
	// and smaller than the previous GoAway ID should be killed.
	upperLimit := t.prevGoAwayID
	if upperLimit == 0 { // This is the first GoAway Frame.
		upperLimit = math.MaxUint32 // Kill all streams after the GoAway ID.
	}

	t.prevGoAwayID = id
	if len(t.activeStreams) == 0 {
		t.mu.Unlock()
		t.Close(connectionErrorf(true, nil, "received goaway and there are no active streams"))
		return
	}

	streamsToClose := make([]*Stream, 0)
	for streamID, stream := range t.activeStreams {
		if streamID > id && streamID <= upperLimit {
			// The stream was unprocessed by the server.
			if streamID > id && streamID <= upperLimit {
				atomic.StoreUint32(&stream.unprocessed, 1)
				streamsToClose = append(streamsToClose, stream)
			}
		}
	}
	t.mu.Unlock()
	// Called outside t.mu because closeStream can take controlBuf's mu, which
	// could induce deadlock and is not allowed.
	for _, stream := range streamsToClose {
		t.closeStream(stream, errStreamDrain, false, http2.ErrCodeNo, statusGoAway, nil, false)
	}
}

// setGoAwayReason sets the value of t.goAwayReason based
// on the GoAway frame received.
// It expects a lock on transport's mutext to be held by
// the caller.
func (t *http2Client) setGoAwayReason(f *http2.GoAwayFrame) {
	t.goAwayReason = GoAwayNoReason
	switch f.ErrCode {
	case http2.ErrCodeEnhanceYourCalm:
		if string(f.DebugData()) == "too_many_pings" {
			t.goAwayReason = GoAwayTooManyPings
		}
	}
	if len(f.DebugData()) == 0 {
		t.goAwayDebugMessage = fmt.Sprintf("code: %s", f.ErrCode)
	} else {
		t.goAwayDebugMessage = fmt.Sprintf("code: %s, debug data: %q", f.ErrCode, string(f.DebugData()))
	}
}

func (t *http2Client) GetGoAwayReason() (GoAwayReason, string) {
	t.mu.Lock()
	defer t.mu.Unlock()
	return t.goAwayReason, t.goAwayDebugMessage
}

func (t *http2Client) handleWindowUpdate(f *http2.WindowUpdateFrame) {
	t.controlBuf.put(&incomingWindowUpdate{
		streamID:  f.Header().StreamID,
		increment: f.Increment,
	})
}

// operateHeaders takes action on the decoded headers.
func (t *http2Client) operateHeaders(frame *http2.MetaHeadersFrame) {
	s := t.getStream(frame)
	if s == nil {
		return
	}
	endStream := frame.StreamEnded()
	atomic.StoreUint32(&s.bytesReceived, 1)
	initialHeader := atomic.LoadUint32(&s.headerChanClosed) == 0

	if !initialHeader && !endStream {
		// As specified by gRPC over HTTP2, a HEADERS frame (and associated CONTINUATION frames) can only appear at the start or end of a stream. Therefore, second HEADERS frame must have EOS bit set.
		st := status.New(codes.Internal, "a HEADERS frame cannot appear in the middle of a stream")
		t.closeStream(s, st.Err(), true, http2.ErrCodeProtocol, st, nil, false)
		return
	}

	// frame.Truncated is set to true when framer detects that the current header
	// list size hits MaxHeaderListSize limit.
	if frame.Truncated {
		se := status.New(codes.Internal, "peer header list size exceeded limit")
		t.closeStream(s, se.Err(), true, http2.ErrCodeFrameSize, se, nil, endStream)
		return
	}

	var (
		// If a gRPC Response-Headers has already been received, then it means
		// that the peer is speaking gRPC and we are in gRPC mode.
		isGRPC         = !initialHeader
		mdata          = make(map[string][]string)
		contentTypeErr = "malformed header: missing HTTP content-type"
		grpcMessage    string
		statusGen      *status.Status
		recvCompress   string
		httpStatusCode *int
		httpStatusErr  string
		rawStatusCode  = codes.Unknown
		// headerError is set if an error is encountered while parsing the headers
		headerError string
	)

	if initialHeader {
		httpStatusErr = "malformed header: missing HTTP status"
	}

	for _, hf := range frame.Fields {
		switch hf.Name {
		case "content-type":
			if _, validContentType := grpcutil.ContentSubtype(hf.Value); !validContentType {
				contentTypeErr = fmt.Sprintf("transport: received unexpected content-type %q", hf.Value)
				break
			}
			contentTypeErr = ""
			mdata[hf.Name] = append(mdata[hf.Name], hf.Value)
			isGRPC = true
		case "grpc-encoding":
			recvCompress = hf.Value
		case "grpc-status":
			code, err := strconv.ParseInt(hf.Value, 10, 32)
			if err != nil {
				se := status.New(codes.Internal, fmt.Sprintf("transport: malformed grpc-status: %v", err))
				t.closeStream(s, se.Err(), true, http2.ErrCodeProtocol, se, nil, endStream)
				return
			}
			rawStatusCode = codes.Code(uint32(code))
		case "grpc-message":
			grpcMessage = decodeGrpcMessage(hf.Value)
		case "grpc-status-details-bin":
			var err error
			statusGen, err = decodeGRPCStatusDetails(hf.Value)
			if err != nil {
				headerError = fmt.Sprintf("transport: malformed grpc-status-details-bin: %v", err)
			}
		case ":status":
			if hf.Value == "200" {
				httpStatusErr = ""
				statusCode := 200
				httpStatusCode = &statusCode
				break
			}

			c, err := strconv.ParseInt(hf.Value, 10, 32)
			if err != nil {
				se := status.New(codes.Internal, fmt.Sprintf("transport: malformed http-status: %v", err))
				t.closeStream(s, se.Err(), true, http2.ErrCodeProtocol, se, nil, endStream)
				return
			}
			statusCode := int(c)
			httpStatusCode = &statusCode

			httpStatusErr = fmt.Sprintf(
				"unexpected HTTP status code received from server: %d (%s)",
				statusCode,
				http.StatusText(statusCode),
			)
		default:
			if isReservedHeader(hf.Name) && !isWhitelistedHeader(hf.Name) {
				break
			}
			v, err := decodeMetadataHeader(hf.Name, hf.Value)
			if err != nil {
				headerError = fmt.Sprintf("transport: malformed %s: %v", hf.Name, err)
				logger.Warningf("Failed to decode metadata header (%q, %q): %v", hf.Name, hf.Value, err)
				break
			}
			mdata[hf.Name] = append(mdata[hf.Name], v)
		}
	}

	if !isGRPC || httpStatusErr != "" {
		var code = codes.Internal // when header does not include HTTP status, return INTERNAL

		if httpStatusCode != nil {
			var ok bool
			code, ok = HTTPStatusConvTab[*httpStatusCode]
			if !ok {
				code = codes.Unknown
			}
		}
		var errs []string
		if httpStatusErr != "" {
			errs = append(errs, httpStatusErr)
		}
		if contentTypeErr != "" {
			errs = append(errs, contentTypeErr)
		}
		// Verify the HTTP response is a 200.
		se := status.New(code, strings.Join(errs, "; "))
		t.closeStream(s, se.Err(), true, http2.ErrCodeProtocol, se, nil, endStream)
		return
	}

	if headerError != "" {
		se := status.New(codes.Internal, headerError)
		t.closeStream(s, se.Err(), true, http2.ErrCodeProtocol, se, nil, endStream)
		return
	}

	isHeader := false

	// If headerChan hasn't been closed yet
	if atomic.CompareAndSwapUint32(&s.headerChanClosed, 0, 1) {
		s.headerValid = true
		if !endStream {
			// HEADERS frame block carries a Response-Headers.
			isHeader = true
			// These values can be set without any synchronization because
			// stream goroutine will read it only after seeing a closed
			// headerChan which we'll close after setting this.
			s.recvCompress = recvCompress
			if len(mdata) > 0 {
				s.header = mdata
			}
		} else {
			// HEADERS frame block carries a Trailers-Only.
			s.noHeaders = true
		}
		close(s.headerChan)
	}

	for _, sh := range t.statsHandlers {
		if isHeader {
			inHeader := &stats.InHeader{
				Client:      true,
				WireLength:  int(frame.Header().Length),
				Header:      metadata.MD(mdata).Copy(),
				Compression: s.recvCompress,
			}
			sh.HandleRPC(s.ctx, inHeader)
		} else {
			inTrailer := &stats.InTrailer{
				Client:     true,
				WireLength: int(frame.Header().Length),
				Trailer:    metadata.MD(mdata).Copy(),
			}
			sh.HandleRPC(s.ctx, inTrailer)
		}
	}

	if !endStream {
		return
	}

	if statusGen == nil {
		statusGen = status.New(rawStatusCode, grpcMessage)
	}

	// if client received END_STREAM from server while stream was still active, send RST_STREAM
	rst := s.getState() == streamActive
	t.closeStream(s, io.EOF, rst, http2.ErrCodeNo, statusGen, mdata, true)
}

// reader runs as a separate goroutine in charge of reading data from network
// connection.
//
// TODO(zhaoq): currently one reader per transport. Investigate whether this is
// optimal.
// TODO(zhaoq): Check the validity of the incoming frame sequence.
func (t *http2Client) reader() {
	defer close(t.readerDone)
	// Check the validity of server preface.
	frame, err := t.framer.fr.ReadFrame()
	if err != nil {
		err = connectionErrorf(true, err, "error reading server preface: %v", err)
		t.Close(err) // this kicks off resetTransport, so must be last before return
		return
	}
	t.conn.SetReadDeadline(time.Time{}) // reset deadline once we get the settings frame (we didn't time out, yay!)
	if t.keepaliveEnabled {
		atomic.StoreInt64(&t.lastRead, time.Now().UnixNano())
	}
	sf, ok := frame.(*http2.SettingsFrame)
	if !ok {
		// this kicks off resetTransport, so must be last before return
		t.Close(connectionErrorf(true, nil, "initial http2 frame from server is not a settings frame: %T", frame))
		return
	}
	t.onPrefaceReceipt()
	t.handleSettings(sf, true)

	// loop to keep reading incoming messages on this transport.
	for {
		t.controlBuf.throttle()
		frame, err := t.framer.fr.ReadFrame()
		if t.keepaliveEnabled {
			atomic.StoreInt64(&t.lastRead, time.Now().UnixNano())
		}
		if err != nil {
			// Abort an active stream if the http2.Framer returns a
			// http2.StreamError. This can happen only if the server's response
			// is malformed http2.
			if se, ok := err.(http2.StreamError); ok {
				t.mu.Lock()
				s := t.activeStreams[se.StreamID]
				t.mu.Unlock()
				if s != nil {
					// use error detail to provide better err message
					code := http2ErrConvTab[se.Code]
					errorDetail := t.framer.fr.ErrorDetail()
					var msg string
					if errorDetail != nil {
						msg = errorDetail.Error()
					} else {
						msg = "received invalid frame"
					}
					t.closeStream(s, status.Error(code, msg), true, http2.ErrCodeProtocol, status.New(code, msg), nil, false)
				}
				continue
			} else {
				// Transport error.
				t.Close(connectionErrorf(true, err, "error reading from server: %v", err))
				return
			}
		}
		switch frame := frame.(type) {
		case *http2.MetaHeadersFrame:
			t.operateHeaders(frame)
		case *http2.DataFrame:
			t.handleData(frame)
		case *http2.RSTStreamFrame:
			t.handleRSTStream(frame)
		case *http2.SettingsFrame:
			t.handleSettings(frame, false)
		case *http2.PingFrame:
			t.handlePing(frame)
		case *http2.GoAwayFrame:
			t.handleGoAway(frame)
		case *http2.WindowUpdateFrame:
			t.handleWindowUpdate(frame)
		default:
			if logger.V(logLevel) {
				logger.Errorf("transport: http2Client.reader got unhandled frame type %v.", frame)
			}
		}
	}
}

func minTime(a, b time.Duration) time.Duration {
	if a < b {
		return a
	}
	return b
}

// keepalive running in a separate goroutine makes sure the connection is alive by sending pings.
func (t *http2Client) keepalive() {
	p := &ping{data: [8]byte{}}
	// True iff a ping has been sent, and no data has been received since then.
	outstandingPing := false
	// Amount of time remaining before which we should receive an ACK for the
	// last sent ping.
	timeoutLeft := time.Duration(0)
	// Records the last value of t.lastRead before we go block on the timer.
	// This is required to check for read activity since then.
	prevNano := time.Now().UnixNano()
	timer := time.NewTimer(t.kp.Time)
	for {
		select {
		case <-timer.C:
			lastRead := atomic.LoadInt64(&t.lastRead)
			if lastRead > prevNano {
				// There has been read activity since the last time we were here.
				outstandingPing = false
				// Next timer should fire at kp.Time seconds from lastRead time.
				timer.Reset(time.Duration(lastRead) + t.kp.Time - time.Duration(time.Now().UnixNano()))
				prevNano = lastRead
				continue
			}
			if outstandingPing && timeoutLeft <= 0 {
				t.Close(connectionErrorf(true, nil, "keepalive ping failed to receive ACK within timeout"))
				return
			}
			t.mu.Lock()
			if t.state == closing {
				// If the transport is closing, we should exit from the
				// keepalive goroutine here. If not, we could have a race
				// between the call to Signal() from Close() and the call to
				// Wait() here, whereby the keepalive goroutine ends up
				// blocking on the condition variable which will never be
				// signalled again.
				t.mu.Unlock()
				return
			}
			if len(t.activeStreams) < 1 && !t.kp.PermitWithoutStream {
				// If a ping was sent out previously (because there were active
				// streams at that point) which wasn't acked and its timeout
				// hadn't fired, but we got here and are about to go dormant,
				// we should make sure that we unconditionally send a ping once
				// we awaken.
				outstandingPing = false
				t.kpDormant = true
				t.kpDormancyCond.Wait()
			}
			t.kpDormant = false
			t.mu.Unlock()

			// We get here either because we were dormant and a new stream was
			// created which unblocked the Wait() call, or because the
			// keepalive timer expired. In both cases, we need to send a ping.
			if !outstandingPing {
				if channelz.IsOn() {
					atomic.AddInt64(&t.czData.kpCount, 1)
				}
				t.controlBuf.put(p)
				timeoutLeft = t.kp.Timeout
				outstandingPing = true
			}
			// The amount of time to sleep here is the minimum of kp.Time and
			// timeoutLeft. This will ensure that we wait only for kp.Time
			// before sending out the next ping (for cases where the ping is
			// acked).
			sleepDuration := minTime(t.kp.Time, timeoutLeft)
			timeoutLeft -= sleepDuration
			timer.Reset(sleepDuration)
		case <-t.ctx.Done():
			if !timer.Stop() {
				<-timer.C
			}
			return
		}
	}
}

func (t *http2Client) Error() <-chan struct{} {
	return t.ctx.Done()
}

func (t *http2Client) GoAway() <-chan struct{} {
	return t.goAway
}

func (t *http2Client) ChannelzMetric() *channelz.SocketInternalMetric {
	s := channelz.SocketInternalMetric{
		StreamsStarted:                  atomic.LoadInt64(&t.czData.streamsStarted),
		StreamsSucceeded:                atomic.LoadInt64(&t.czData.streamsSucceeded),
		StreamsFailed:                   atomic.LoadInt64(&t.czData.streamsFailed),
		MessagesSent:                    atomic.LoadInt64(&t.czData.msgSent),
		MessagesReceived:                atomic.LoadInt64(&t.czData.msgRecv),
		KeepAlivesSent:                  atomic.LoadInt64(&t.czData.kpCount),
		LastLocalStreamCreatedTimestamp: time.Unix(0, atomic.LoadInt64(&t.czData.lastStreamCreatedTime)),
		LastMessageSentTimestamp:        time.Unix(0, atomic.LoadInt64(&t.czData.lastMsgSentTime)),
		LastMessageReceivedTimestamp:    time.Unix(0, atomic.LoadInt64(&t.czData.lastMsgRecvTime)),
		LocalFlowControlWindow:          int64(t.fc.getSize()),
		SocketOptions:                   channelz.GetSocketOption(t.conn),
		LocalAddr:                       t.localAddr,
		RemoteAddr:                      t.remoteAddr,
		// RemoteName :
	}
	if au, ok := t.authInfo.(credentials.ChannelzSecurityInfo); ok {
		s.Security = au.GetSecurityValue()
	}
	s.RemoteFlowControlWindow = t.getOutFlowWindow()
	return &s
}

func (t *http2Client) RemoteAddr() net.Addr { return t.remoteAddr }

func (t *http2Client) IncrMsgSent() {
	atomic.AddInt64(&t.czData.msgSent, 1)
	atomic.StoreInt64(&t.czData.lastMsgSentTime, time.Now().UnixNano())
}

func (t *http2Client) IncrMsgRecv() {
	atomic.AddInt64(&t.czData.msgRecv, 1)
	atomic.StoreInt64(&t.czData.lastMsgRecvTime, time.Now().UnixNano())
}

func (t *http2Client) getOutFlowWindow() int64 {
	resp := make(chan uint32, 1)
	timer := time.NewTimer(time.Second)
	defer timer.Stop()
	t.controlBuf.put(&outFlowControlSizeRequest{resp})
	select {
	case sz := <-resp:
		return int64(sz)
	case <-t.ctxDone:
		return -1
	case <-timer.C:
		return -2
	}
}