go/src/net/http/server.go

3697 lines
114 KiB
Go
Raw Normal View History

// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// HTTP server. See RFC 7230 through 7235.
package http
import (
"bufio"
"bytes"
"context"
"crypto/tls"
"errors"
"fmt"
"internal/godebug"
"io"
"log"
"math/rand"
"net"
"net/textproto"
"net/url"
urlpkg "net/url"
"path"
"runtime"
"strconv"
"strings"
"sync"
"sync/atomic"
"time"
"golang.org/x/net/http/httpguts"
)
// Errors used by the HTTP server.
var (
// ErrBodyNotAllowed is returned by ResponseWriter.Write calls
// when the HTTP method or response code does not permit a
// body.
ErrBodyNotAllowed = errors.New("http: request method or response status code does not allow body")
// ErrHijacked is returned by ResponseWriter.Write calls when
// the underlying connection has been hijacked using the
// Hijacker interface. A zero-byte write on a hijacked
// connection will return ErrHijacked without any other side
// effects.
ErrHijacked = errors.New("http: connection has been hijacked")
// ErrContentLength is returned by ResponseWriter.Write calls
// when a Handler set a Content-Length response header with a
// declared size and then attempted to write more bytes than
// declared.
ErrContentLength = errors.New("http: wrote more than the declared Content-Length")
// Deprecated: ErrWriteAfterFlush is no longer returned by
// anything in the net/http package. Callers should not
// compare errors against this variable.
ErrWriteAfterFlush = errors.New("unused")
)
// A Handler responds to an HTTP request.
//
// ServeHTTP should write reply headers and data to the [ResponseWriter]
// and then return. Returning signals that the request is finished; it
// is not valid to use the [ResponseWriter] or read from the
// [Request.Body] after or concurrently with the completion of the
// ServeHTTP call.
//
// Depending on the HTTP client software, HTTP protocol version, and
// any intermediaries between the client and the Go server, it may not
// be possible to read from the [Request.Body] after writing to the
// [ResponseWriter]. Cautious handlers should read the [Request.Body]
// first, and then reply.
//
// Except for reading the body, handlers should not modify the
// provided Request.
//
// If ServeHTTP panics, the server (the caller of ServeHTTP) assumes
// that the effect of the panic was isolated to the active request.
// It recovers the panic, logs a stack trace to the server error log,
// and either closes the network connection or sends an HTTP/2
// RST_STREAM, depending on the HTTP protocol. To abort a handler so
// the client sees an interrupted response but the server doesn't log
// an error, panic with the value [ErrAbortHandler].
type Handler interface {
ServeHTTP(ResponseWriter, *Request)
}
// A ResponseWriter interface is used by an HTTP handler to
// construct an HTTP response.
//
// A ResponseWriter may not be used after [Handler.ServeHTTP] has returned.
type ResponseWriter interface {
// Header returns the header map that will be sent by
// [ResponseWriter.WriteHeader]. The [Header] map also is the mechanism with which
// [Handler] implementations can set HTTP trailers.
//
// Changing the header map after a call to [ResponseWriter.WriteHeader] (or
// [ResponseWriter.Write]) has no effect unless the HTTP status code was of the
// 1xx class or the modified headers are trailers.
//
// There are two ways to set Trailers. The preferred way is to
// predeclare in the headers which trailers you will later
// send by setting the "Trailer" header to the names of the
// trailer keys which will come later. In this case, those
// keys of the Header map are treated as if they were
// trailers. See the example. The second way, for trailer
// keys not known to the [Handler] until after the first [ResponseWriter.Write],
// is to prefix the [Header] map keys with the [TrailerPrefix]
// constant value.
//
// To suppress automatic response headers (such as "Date"), set
// their value to nil.
Header() Header
// Write writes the data to the connection as part of an HTTP reply.
//
// If [ResponseWriter.WriteHeader] has not yet been called, Write calls
// WriteHeader(http.StatusOK) before writing the data. If the Header
// does not contain a Content-Type line, Write adds a Content-Type set
// to the result of passing the initial 512 bytes of written data to
// [DetectContentType]. Additionally, if the total size of all written
// data is under a few KB and there are no Flush calls, the
// Content-Length header is added automatically.
//
// Depending on the HTTP protocol version and the client, calling
// Write or WriteHeader may prevent future reads on the
// Request.Body. For HTTP/1.x requests, handlers should read any
// needed request body data before writing the response. Once the
// headers have been flushed (due to either an explicit Flusher.Flush
// call or writing enough data to trigger a flush), the request body
// may be unavailable. For HTTP/2 requests, the Go HTTP server permits
// handlers to continue to read the request body while concurrently
// writing the response. However, such behavior may not be supported
// by all HTTP/2 clients. Handlers should read before writing if
// possible to maximize compatibility.
Write([]byte) (int, error)
// WriteHeader sends an HTTP response header with the provided
// status code.
//
// If WriteHeader is not called explicitly, the first call to Write
// will trigger an implicit WriteHeader(http.StatusOK).
// Thus explicit calls to WriteHeader are mainly used to
// send error codes or 1xx informational responses.
//
// The provided code must be a valid HTTP 1xx-5xx status code.
// Any number of 1xx headers may be written, followed by at most
// one 2xx-5xx header. 1xx headers are sent immediately, but 2xx-5xx
// headers may be buffered. Use the Flusher interface to send
// buffered data. The header map is cleared when 2xx-5xx headers are
// sent, but not with 1xx headers.
//
// The server will automatically send a 100 (Continue) header
// on the first read from the request body if the request has
// an "Expect: 100-continue" header.
WriteHeader(statusCode int)
}
// The Flusher interface is implemented by ResponseWriters that allow
// an HTTP handler to flush buffered data to the client.
//
// The default HTTP/1.x and HTTP/2 ResponseWriter implementations
// support Flusher, but ResponseWriter wrappers may not. Handlers
// should always test for this ability at runtime.
//
// Note that even for ResponseWriters that support Flush,
// if the client is connected through an HTTP proxy,
// the buffered data may not reach the client until the response
// completes.
type Flusher interface {
// Flush sends any buffered data to the client.
Flush()
}
// The Hijacker interface is implemented by ResponseWriters that allow
// an HTTP handler to take over the connection.
//
// The default ResponseWriter for HTTP/1.x connections supports
// Hijacker, but HTTP/2 connections intentionally do not.
// ResponseWriter wrappers may also not support Hijacker. Handlers
// should always test for this ability at runtime.
type Hijacker interface {
// Hijack lets the caller take over the connection.
// After a call to Hijack the HTTP server library
// will not do anything else with the connection.
//
// It becomes the caller's responsibility to manage
// and close the connection.
//
// The returned net.Conn may have read or write deadlines
// already set, depending on the configuration of the
// Server. It is the caller's responsibility to set
// or clear those deadlines as needed.
//
// The returned bufio.Reader may contain unprocessed buffered
// data from the client.
//
// After a call to Hijack, the original Request.Body must not
// be used. The original Request's Context remains valid and
// is not canceled until the Request's ServeHTTP method
// returns.
Hijack() (net.Conn, *bufio.ReadWriter, error)
}
// The CloseNotifier interface is implemented by ResponseWriters which
// allow detecting when the underlying connection has gone away.
//
// This mechanism can be used to cancel long operations on the server
// if the client has disconnected before the response is ready.
//
// Deprecated: the CloseNotifier interface predates Go's context package.
// New code should use Request.Context instead.
type CloseNotifier interface {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// CloseNotify returns a channel that receives at most a
// single value (true) when the client connection has gone
// away.
//
// CloseNotify may wait to notify until Request.Body has been
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// fully read.
//
// After the Handler has returned, there is no guarantee
// that the channel receives a value.
//
// If the protocol is HTTP/1.1 and CloseNotify is called while
// processing an idempotent request (such a GET) while
// HTTP/1.1 pipelining is in use, the arrival of a subsequent
// pipelined request may cause a value to be sent on the
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// returned channel. In practice HTTP/1.1 pipelining is not
// enabled in browsers and not seen often in the wild. If this
// is a problem, use HTTP/2 or only use CloseNotify on methods
// such as POST.
CloseNotify() <-chan bool
}
var (
// ServerContextKey is a context key. It can be used in HTTP
// handlers with Context.Value to access the server that
// started the handler. The associated value will be of
// type *Server.
ServerContextKey = &contextKey{"http-server"}
// LocalAddrContextKey is a context key. It can be used in
// HTTP handlers with Context.Value to access the local
// address the connection arrived on.
// The associated value will be of type net.Addr.
LocalAddrContextKey = &contextKey{"local-addr"}
)
// A conn represents the server side of an HTTP connection.
type conn struct {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// server is the server on which the connection arrived.
// Immutable; never nil.
server *Server
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
// cancelCtx cancels the connection-level context.
cancelCtx context.CancelFunc
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// rwc is the underlying network connection.
// This is never wrapped by other types and is the value given out
// to CloseNotifier callers. It is usually of type *net.TCPConn or
// *tls.Conn.
rwc net.Conn
// remoteAddr is rwc.RemoteAddr().String(). It is not populated synchronously
// inside the Listener's Accept goroutine, as some implementations block.
// It is populated immediately inside the (*conn).serve goroutine.
// This is the value of a Handler's (*Request).RemoteAddr.
remoteAddr string
// tlsState is the TLS connection state when using TLS.
// nil means not TLS.
tlsState *tls.ConnectionState
// werr is set to the first write error to rwc.
// It is set via checkConnErrorWriter{w}, where bufw writes.
werr error
// r is bufr's read source. It's a wrapper around rwc that provides
// io.LimitedReader-style limiting (while reading request headers)
// and functionality to support CloseNotifier. See *connReader docs.
r *connReader
// bufr reads from r.
bufr *bufio.Reader
// bufw writes to checkConnErrorWriter{c}, which populates werr on error.
bufw *bufio.Writer
// lastMethod is the method of the most recent request
// on this connection, if any.
lastMethod string
curReq atomic.Pointer[response] // (which has a Request in it)
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
curState atomic.Uint64 // packed (unixtime<<8|uint8(ConnState))
// mu guards hijackedv
mu sync.Mutex
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// hijackedv is whether this connection has been hijacked
// by a Handler with the Hijacker interface.
// It is guarded by mu.
hijackedv bool
}
func (c *conn) hijacked() bool {
c.mu.Lock()
defer c.mu.Unlock()
return c.hijackedv
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// c.mu must be held.
func (c *conn) hijackLocked() (rwc net.Conn, buf *bufio.ReadWriter, err error) {
if c.hijackedv {
return nil, nil, ErrHijacked
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
c.r.abortPendingRead()
c.hijackedv = true
rwc = c.rwc
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
rwc.SetDeadline(time.Time{})
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
buf = bufio.NewReadWriter(c.bufr, bufio.NewWriter(rwc))
if c.r.hasByte {
if _, err := c.bufr.Peek(c.bufr.Buffered() + 1); err != nil {
return nil, nil, fmt.Errorf("unexpected Peek failure reading buffered byte: %v", err)
}
}
c.setState(rwc, StateHijacked, runHooks)
return
}
// This should be >= 512 bytes for DetectContentType,
// but otherwise it's somewhat arbitrary.
const bufferBeforeChunkingSize = 2048
// chunkWriter writes to a response's conn buffer, and is the writer
// wrapped by the response.w buffered writer.
//
// chunkWriter also is responsible for finalizing the Header, including
// conditionally setting the Content-Type and setting a Content-Length
// in cases where the handler's final output is smaller than the buffer
// size. It also conditionally adds chunk headers, when in chunking mode.
//
// See the comment above (*response).Write for the entire write flow.
type chunkWriter struct {
res *response
// header is either nil or a deep clone of res.handlerHeader
// at the time of res.writeHeader, if res.writeHeader is
// called and extra buffering is being done to calculate
// Content-Type and/or Content-Length.
header Header
// wroteHeader tells whether the header's been written to "the
// wire" (or rather: w.conn.buf). this is unlike
// (*response).wroteHeader, which tells only whether it was
// logically written.
wroteHeader bool
// set by the writeHeader method:
chunking bool // using chunked transfer encoding for reply body
}
var (
crlf = []byte("\r\n")
colonSpace = []byte(": ")
)
func (cw *chunkWriter) Write(p []byte) (n int, err error) {
if !cw.wroteHeader {
cw.writeHeader(p)
}
net/http: treat HEAD requests like GET requests A response to a HEAD request is supposed to look the same as a response to a GET request, just without a body. HEAD requests are incredibly rare in the wild. The Go net/http package has so far treated HEAD requests specially: a Write on our default ResponseWriter returned ErrBodyNotAllowed, telling handlers that something was wrong. This was to optimize the fast path for HEAD requests, but: 1) because HEAD requests are incredibly rare, they're not worth having a fast path for. 2) Letting the http.Handler handle but do nop Writes is still very fast. 3) this forces ugly error handling into the application. e.g. https://code.google.com/p/go/source/detail?r=6f596be7a31e and related. 4) The net/http package nowadays does Content-Type sniffing, but you don't get that for HEAD. 5) The net/http package nowadays does Content-Length counting for small (few KB) responses, but not for HEAD. 6) ErrBodyNotAllowed was useless. By the time you received it, you had probably already done all your heavy computation and I/O to calculate what to write. So, this change makes HEAD requests like GET requests. We now count content-length and sniff content-type for HEAD requests. If you Write, it doesn't return an error. If you want a fast-path in your code for HEAD, you have to do it early and set all the response headers yourself. Just like before. If you choose not to Write in HEAD requests, be sure to set Content-Length if you know it. We won't write "Content-Length: 0" because you might've just chosen to not write (or you don't know your Content-Length in advance). Fixes #5454 R=golang-dev, dsymonds CC=golang-dev https://golang.org/cl/12583043
2013-08-06 18:33:03 -07:00
if cw.res.req.Method == "HEAD" {
// Eat writes.
return len(p), nil
}
if cw.chunking {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
_, err = fmt.Fprintf(cw.res.conn.bufw, "%x\r\n", len(p))
if err != nil {
cw.res.conn.rwc.Close()
return
}
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
n, err = cw.res.conn.bufw.Write(p)
if cw.chunking && err == nil {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
_, err = cw.res.conn.bufw.Write(crlf)
}
if err != nil {
cw.res.conn.rwc.Close()
}
return
}
func (cw *chunkWriter) flush() error {
if !cw.wroteHeader {
cw.writeHeader(nil)
}
return cw.res.conn.bufw.Flush()
}
func (cw *chunkWriter) close() {
if !cw.wroteHeader {
cw.writeHeader(nil)
}
if cw.chunking {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
bw := cw.res.conn.bufw // conn's bufio writer
// zero chunk to mark EOF
bw.WriteString("0\r\n")
if trailers := cw.res.finalTrailers(); trailers != nil {
trailers.Write(bw) // the writer handles noting errors
}
// final blank line after the trailers (whether
// present or not)
bw.WriteString("\r\n")
}
}
// A response represents the server side of an HTTP response.
type response struct {
conn *conn
req *Request // request for this response
reqBody io.ReadCloser
cancelCtx context.CancelFunc // when ServeHTTP exits
wroteHeader bool // a non-1xx header has been (logically) written
wroteContinue bool // 100 Continue response was written
wants10KeepAlive bool // HTTP/1.0 w/ Connection "keep-alive"
wantsClose bool // HTTP request has Connection "close"
// canWriteContinue is an atomic boolean that says whether or
// not a 100 Continue header can be written to the
// connection.
// writeContinueMu must be held while writing the header.
// These two fields together synchronize the body reader (the
// expectContinueReader, which wants to write 100 Continue)
// against the main writer.
canWriteContinue atomic.Bool
writeContinueMu sync.Mutex
w *bufio.Writer // buffers output in chunks to chunkWriter
cw chunkWriter
// handlerHeader is the Header that Handlers get access to,
// which may be retained and mutated even after WriteHeader.
// handlerHeader is copied into cw.header at WriteHeader
// time, and privately mutated thereafter.
handlerHeader Header
calledHeader bool // handler accessed handlerHeader via Header
written int64 // number of bytes written in body
contentLength int64 // explicitly-declared Content-Length; or -1
status int // status code passed to WriteHeader
// close connection after this reply. set on request and
// updated after response from handler if there's a
// "Connection: keep-alive" response header and a
// Content-Length.
closeAfterReply bool
// When fullDuplex is false (the default), we consume any remaining
// request body before starting to write a response.
fullDuplex bool
// requestBodyLimitHit is set by requestTooLarge when
// maxBytesReader hits its max size. It is checked in
// WriteHeader, to make sure we don't consume the
// remaining request body to try to advance to the next HTTP
// request. Instead, when this is set, we stop reading
// subsequent requests on this connection and stop reading
// input from it.
requestBodyLimitHit bool
// trailers are the headers to be sent after the handler
// finishes writing the body. This field is initialized from
// the Trailer response header when the response header is
// written.
trailers []string
handlerDone atomic.Bool // set true when the handler exits
net/http: re-simplify HTTP/1.x status line writing It used to be simple, and then it got complicated for speed (to reduce allocations, mostly), but that involved a mutex and hurt multi-core performance, contending on the mutex. A change was sent to try to improve that mutex contention in https://go-review.googlesource.com/c/42110/2/src/net/http/server.go but that introduced its own allocations (the string->interface{} boxing for the sync.Map key), which runs counter to the whole point of that statusLine function: to remove allocations. Instead, make the code simple again and not have a mutex. It's a bit slower for the single-core case, but nobody with a single-user HTTP server cares about 50 nanoseconds: name old time/op new time/op delta ResponseStatusLine 37.5ns ± 2% 87.1ns ± 2% +132.42% (p=0.029 n=4+4) ResponseStatusLine-2 63.1ns ± 1% 43.1ns ±12% -31.67% (p=0.029 n=4+4) ResponseStatusLine-4 53.8ns ± 8% 40.2ns ± 2% -25.29% (p=0.029 n=4+4) name old alloc/op new alloc/op delta ResponseStatusLine 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) name old allocs/op new allocs/op delta ResponseStatusLine 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) (Note the code could be even simpler with fmt.Fprintf, but that is relatively slow and involves a bunch of allocations getting arguments into interface{} for the call) Change-Id: I1fa119132dbbf97a8e7204ce3e0707d433060da2 Reviewed-on: https://go-review.googlesource.com/42133 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan Mills <bcmills@google.com>
2017-04-28 16:46:18 +00:00
// Buffers for Date, Content-Length, and status code
dateBuf [len(TimeFormat)]byte
clenBuf [10]byte
statusBuf [3]byte
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
// closeNotifyCh is the channel returned by CloseNotify.
// TODO(bradfitz): this is currently (for Go 1.8) always
// non-nil. Make this lazily-created again as it used to be?
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
closeNotifyCh chan bool
didCloseNotify atomic.Bool // atomic (only false->true winner should send)
}
func (c *response) SetReadDeadline(deadline time.Time) error {
return c.conn.rwc.SetReadDeadline(deadline)
}
func (c *response) SetWriteDeadline(deadline time.Time) error {
return c.conn.rwc.SetWriteDeadline(deadline)
}
func (c *response) EnableFullDuplex() error {
c.fullDuplex = true
return nil
}
// TrailerPrefix is a magic prefix for ResponseWriter.Header map keys
// that, if present, signals that the map entry is actually for
// the response trailers, and not the response headers. The prefix
// is stripped after the ServeHTTP call finishes and the values are
// sent in the trailers.
//
// This mechanism is intended only for trailers that are not known
// prior to the headers being written. If the set of trailers is fixed
// or known before the header is written, the normal Go trailers mechanism
// is preferred:
//
// https://pkg.go.dev/net/http#ResponseWriter
// https://pkg.go.dev/net/http#example-ResponseWriter-Trailers
const TrailerPrefix = "Trailer:"
// finalTrailers is called after the Handler exits and returns a non-nil
// value if the Handler set any trailers.
func (w *response) finalTrailers() Header {
var t Header
for k, vv := range w.handlerHeader {
if kk, found := strings.CutPrefix(k, TrailerPrefix); found {
if t == nil {
t = make(Header)
}
t[kk] = vv
}
}
for _, k := range w.trailers {
if t == nil {
t = make(Header)
}
for _, v := range w.handlerHeader[k] {
t.Add(k, v)
}
}
return t
}
// declareTrailer is called for each Trailer header when the
// response header is written. It notes that a header will need to be
// written in the trailers at the end of the response.
func (w *response) declareTrailer(k string) {
k = CanonicalHeaderKey(k)
if !httpguts.ValidTrailerHeader(k) {
// Forbidden by RFC 7230, section 4.1.2
return
}
w.trailers = append(w.trailers, k)
}
// requestTooLarge is called by maxBytesReader when too much input has
// been read from the client.
func (w *response) requestTooLarge() {
w.closeAfterReply = true
w.requestBodyLimitHit = true
if !w.wroteHeader {
w.Header().Set("Connection", "close")
}
}
// writerOnly hides an io.Writer value's optional ReadFrom method
// from io.Copy.
type writerOnly struct {
io.Writer
}
// ReadFrom is here to optimize copying from an *os.File regular file
// to a *net.TCPConn with sendfile, or from a supported src type such
// as a *net.TCPConn on Linux with splice.
func (w *response) ReadFrom(src io.Reader) (n int64, err error) {
bufp := copyBufPool.Get().(*[]byte)
buf := *bufp
defer copyBufPool.Put(bufp)
// Our underlying w.conn.rwc is usually a *TCPConn (with its
// own ReadFrom method). If not, just fall back to the normal
// copy method.
rf, ok := w.conn.rwc.(io.ReaderFrom)
if !ok {
return io.CopyBuffer(writerOnly{w}, src, buf)
}
// Copy the first sniffLen bytes before switching to ReadFrom.
// This ensures we don't start writing the response before the
// source is available (see golang.org/issue/5660) and provides
// enough bytes to perform Content-Type sniffing when required.
if !w.cw.wroteHeader {
n0, err := io.CopyBuffer(writerOnly{w}, io.LimitReader(src, sniffLen), buf)
n += n0
if err != nil || n0 < sniffLen {
return n, err
}
}
w.w.Flush() // get rid of any previous writes
w.cw.flush() // make sure Header is written; flush data to rwc
// Now that cw has been flushed, its chunking field is guaranteed initialized.
if !w.cw.chunking && w.bodyAllowed() {
n0, err := rf.ReadFrom(src)
n += n0
w.written += n0
return n, err
}
n0, err := io.CopyBuffer(writerOnly{w}, src, buf)
n += n0
return n, err
}
// debugServerConnections controls whether all server connections are wrapped
// with a verbose logging wrapper.
const debugServerConnections = false
// Create new connection from rwc.
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
func (srv *Server) newConn(rwc net.Conn) *conn {
c := &conn{
server: srv,
rwc: rwc,
}
if debugServerConnections {
c.rwc = newLoggingConn("server", c.rwc)
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
return c
}
type readResult struct {
_ incomparable
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
n int
err error
b byte // byte read, if n == 1
}
// connReader is the io.Reader wrapper used by *conn. It combines a
// selectively-activated io.LimitedReader (to bound request header
// read sizes) with support for selectively keeping an io.Reader.Read
// call blocked in a background goroutine to wait for activity and
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// trigger a CloseNotifier channel.
type connReader struct {
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
conn *conn
mu sync.Mutex // guards following
hasByte bool
byteBuf [1]byte
cond *sync.Cond
inRead bool
aborted bool // set true before conn.rwc deadline is set to past
remain int64 // bytes remaining
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
func (cr *connReader) lock() {
cr.mu.Lock()
if cr.cond == nil {
cr.cond = sync.NewCond(&cr.mu)
}
}
func (cr *connReader) unlock() { cr.mu.Unlock() }
func (cr *connReader) startBackgroundRead() {
cr.lock()
defer cr.unlock()
if cr.inRead {
panic("invalid concurrent Body.Read call")
}
if cr.hasByte {
return
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.inRead = true
cr.conn.rwc.SetReadDeadline(time.Time{})
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
go cr.backgroundRead()
}
func (cr *connReader) backgroundRead() {
n, err := cr.conn.rwc.Read(cr.byteBuf[:])
cr.lock()
if n == 1 {
cr.hasByte = true
// We were past the end of the previous request's body already
// (since we wouldn't be in a background read otherwise), so
// this is a pipelined HTTP request. Prior to Go 1.11 we used to
// send on the CloseNotify channel and cancel the context here,
// but the behavior was documented as only "may", and we only
// did that because that's how CloseNotify accidentally behaved
// in very early Go releases prior to context support. Once we
// added context support, people used a Handler's
// Request.Context() and passed it along. Having that context
// cancel on pipelined HTTP requests caused problems.
// Fortunately, almost nothing uses HTTP/1.x pipelining.
// Unfortunately, apt-get does, or sometimes does.
// New Go 1.11 behavior: don't fire CloseNotify or cancel
// contexts on pipelined requests. Shouldn't affect people, but
// fixes cases like Issue 23921. This does mean that a client
// closing their TCP connection after sending a pipelined
// request won't cancel the context, but we'll catch that on any
// write failure (in checkConnErrorWriter.Write).
// If the server never writes, yes, there are still contrived
// server & client behaviors where this fails to ever cancel the
// context, but that's kinda why HTTP/1.x pipelining died
// anyway.
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
}
if ne, ok := err.(net.Error); ok && cr.aborted && ne.Timeout() {
// Ignore this error. It's the expected error from
// another goroutine calling abortPendingRead.
} else if err != nil {
cr.handleReadError(err)
}
cr.aborted = false
cr.inRead = false
cr.unlock()
cr.cond.Broadcast()
}
func (cr *connReader) abortPendingRead() {
cr.lock()
defer cr.unlock()
if !cr.inRead {
return
}
cr.aborted = true
cr.conn.rwc.SetReadDeadline(aLongTimeAgo)
for cr.inRead {
cr.cond.Wait()
}
cr.conn.rwc.SetReadDeadline(time.Time{})
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
}
func (cr *connReader) setReadLimit(remain int64) { cr.remain = remain }
func (cr *connReader) setInfiniteReadLimit() { cr.remain = maxInt64 }
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
func (cr *connReader) hitReadLimit() bool { return cr.remain <= 0 }
// handleReadError is called whenever a Read from the client returns a
// non-nil error.
//
// The provided non-nil err is almost always io.EOF or a "use of
// closed network connection". In any case, the error is not
// particularly interesting, except perhaps for debugging during
// development. Any error means the connection is dead and we should
// down its context.
//
// It may be called from multiple goroutines.
func (cr *connReader) handleReadError(_ error) {
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.conn.cancelCtx()
cr.closeNotify()
}
// may be called from multiple goroutines.
func (cr *connReader) closeNotify() {
res := cr.conn.curReq.Load()
if res != nil && !res.didCloseNotify.Swap(true) {
res.closeNotifyCh <- true
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
}
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
func (cr *connReader) Read(p []byte) (n int, err error) {
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.lock()
if cr.inRead {
cr.unlock()
if cr.conn.hijacked() {
panic("invalid Body.Read call. After hijacked, the original Request must not be used")
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
panic("invalid concurrent Body.Read call")
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
if cr.hitReadLimit() {
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.unlock()
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
return 0, io.EOF
}
if len(p) == 0 {
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.unlock()
return 0, nil
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
}
if int64(len(p)) > cr.remain {
p = p[:cr.remain]
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
if cr.hasByte {
p[0] = cr.byteBuf[0]
cr.hasByte = false
cr.unlock()
return 1, nil
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.inRead = true
cr.unlock()
n, err = cr.conn.rwc.Read(p)
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.lock()
cr.inRead = false
if err != nil {
cr.handleReadError(err)
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.remain -= int64(n)
cr.unlock()
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
cr.cond.Broadcast()
return n, err
}
var (
bufioReaderPool sync.Pool
bufioWriter2kPool sync.Pool
bufioWriter4kPool sync.Pool
)
var copyBufPool = sync.Pool{
New: func() any {
b := make([]byte, 32*1024)
return &b
},
}
func bufioWriterPool(size int) *sync.Pool {
switch size {
case 2 << 10:
return &bufioWriter2kPool
case 4 << 10:
return &bufioWriter4kPool
}
return nil
}
net/http: simplify server, use bufio Reader.Reset and Writer.Reset Update #5100 Update #6086 Remove switchReader, switchWriter, switchReaderPair, switchWriterPair, etc. Now it only maintains pools of bufio Readers and Writers, but uses Reset instead of working around all their previously-associated state. Compared to before the bufio Reset change, it's the same number of allocations, and also faster: benchmark old ns/op new ns/op delta BenchmarkClientServer 111218 109828 -1.25% BenchmarkClientServerParallel4 70580 70013 -0.80% BenchmarkClientServerParallel64 72636 68919 -5.12% BenchmarkServer 139858 137068 -1.99% BenchmarkServerFakeConnNoKeepAlive 14619 14314 -2.09% BenchmarkServerFakeConnWithKeepAlive 12390 11361 -8.31% BenchmarkServerFakeConnWithKeepAliveLite 7630 7306 -4.25% BenchmarkServerHandlerTypeLen 9688 9342 -3.57% BenchmarkServerHandlerNoLen 8700 8470 -2.64% BenchmarkServerHandlerNoType 9255 8949 -3.31% BenchmarkServerHandlerNoHeader 7058 6806 -3.57% benchmark old allocs new allocs delta BenchmarkClientServer 61 61 0.00% BenchmarkClientServerParallel4 61 61 0.00% BenchmarkClientServerParallel64 61 61 0.00% BenchmarkServer 16 16 0.00% BenchmarkServerFakeConnNoKeepAlive 24 24 0.00% BenchmarkServerFakeConnWithKeepAlive 19 19 0.00% BenchmarkServerFakeConnWithKeepAliveLite 9 9 0.00% BenchmarkServerHandlerTypeLen 17 17 0.00% BenchmarkServerHandlerNoLen 14 14 0.00% BenchmarkServerHandlerNoType 15 15 0.00% BenchmarkServerHandlerNoHeader 9 9 0.00% benchmark old bytes new bytes delta BenchmarkClientServer 6988 6985 -0.04% BenchmarkClientServerParallel4 6979 6985 0.09% BenchmarkClientServerParallel64 7002 7019 0.24% BenchmarkServer 1846 1848 0.11% BenchmarkServerFakeConnNoKeepAlive 2420 2412 -0.33% BenchmarkServerFakeConnWithKeepAlive 2126 2129 0.14% BenchmarkServerFakeConnWithKeepAliveLite 989 990 0.10% BenchmarkServerHandlerTypeLen 1818 1819 0.06% BenchmarkServerHandlerNoLen 1775 1777 0.11% BenchmarkServerHandlerNoType 1783 1785 0.11% BenchmarkServerHandlerNoHeader 989 990 0.10% R=golang-dev, r CC=golang-dev https://golang.org/cl/12708046
2013-08-10 19:22:44 -07:00
func newBufioReader(r io.Reader) *bufio.Reader {
if v := bufioReaderPool.Get(); v != nil {
br := v.(*bufio.Reader)
br.Reset(r)
return br
}
// Note: if this reader size is ever changed, update
// TestHandlerBodyClose's assumptions.
return bufio.NewReader(r)
}
net/http: simplify server, use bufio Reader.Reset and Writer.Reset Update #5100 Update #6086 Remove switchReader, switchWriter, switchReaderPair, switchWriterPair, etc. Now it only maintains pools of bufio Readers and Writers, but uses Reset instead of working around all their previously-associated state. Compared to before the bufio Reset change, it's the same number of allocations, and also faster: benchmark old ns/op new ns/op delta BenchmarkClientServer 111218 109828 -1.25% BenchmarkClientServerParallel4 70580 70013 -0.80% BenchmarkClientServerParallel64 72636 68919 -5.12% BenchmarkServer 139858 137068 -1.99% BenchmarkServerFakeConnNoKeepAlive 14619 14314 -2.09% BenchmarkServerFakeConnWithKeepAlive 12390 11361 -8.31% BenchmarkServerFakeConnWithKeepAliveLite 7630 7306 -4.25% BenchmarkServerHandlerTypeLen 9688 9342 -3.57% BenchmarkServerHandlerNoLen 8700 8470 -2.64% BenchmarkServerHandlerNoType 9255 8949 -3.31% BenchmarkServerHandlerNoHeader 7058 6806 -3.57% benchmark old allocs new allocs delta BenchmarkClientServer 61 61 0.00% BenchmarkClientServerParallel4 61 61 0.00% BenchmarkClientServerParallel64 61 61 0.00% BenchmarkServer 16 16 0.00% BenchmarkServerFakeConnNoKeepAlive 24 24 0.00% BenchmarkServerFakeConnWithKeepAlive 19 19 0.00% BenchmarkServerFakeConnWithKeepAliveLite 9 9 0.00% BenchmarkServerHandlerTypeLen 17 17 0.00% BenchmarkServerHandlerNoLen 14 14 0.00% BenchmarkServerHandlerNoType 15 15 0.00% BenchmarkServerHandlerNoHeader 9 9 0.00% benchmark old bytes new bytes delta BenchmarkClientServer 6988 6985 -0.04% BenchmarkClientServerParallel4 6979 6985 0.09% BenchmarkClientServerParallel64 7002 7019 0.24% BenchmarkServer 1846 1848 0.11% BenchmarkServerFakeConnNoKeepAlive 2420 2412 -0.33% BenchmarkServerFakeConnWithKeepAlive 2126 2129 0.14% BenchmarkServerFakeConnWithKeepAliveLite 989 990 0.10% BenchmarkServerHandlerTypeLen 1818 1819 0.06% BenchmarkServerHandlerNoLen 1775 1777 0.11% BenchmarkServerHandlerNoType 1783 1785 0.11% BenchmarkServerHandlerNoHeader 989 990 0.10% R=golang-dev, r CC=golang-dev https://golang.org/cl/12708046
2013-08-10 19:22:44 -07:00
func putBufioReader(br *bufio.Reader) {
br.Reset(nil)
bufioReaderPool.Put(br)
}
net/http: simplify server, use bufio Reader.Reset and Writer.Reset Update #5100 Update #6086 Remove switchReader, switchWriter, switchReaderPair, switchWriterPair, etc. Now it only maintains pools of bufio Readers and Writers, but uses Reset instead of working around all their previously-associated state. Compared to before the bufio Reset change, it's the same number of allocations, and also faster: benchmark old ns/op new ns/op delta BenchmarkClientServer 111218 109828 -1.25% BenchmarkClientServerParallel4 70580 70013 -0.80% BenchmarkClientServerParallel64 72636 68919 -5.12% BenchmarkServer 139858 137068 -1.99% BenchmarkServerFakeConnNoKeepAlive 14619 14314 -2.09% BenchmarkServerFakeConnWithKeepAlive 12390 11361 -8.31% BenchmarkServerFakeConnWithKeepAliveLite 7630 7306 -4.25% BenchmarkServerHandlerTypeLen 9688 9342 -3.57% BenchmarkServerHandlerNoLen 8700 8470 -2.64% BenchmarkServerHandlerNoType 9255 8949 -3.31% BenchmarkServerHandlerNoHeader 7058 6806 -3.57% benchmark old allocs new allocs delta BenchmarkClientServer 61 61 0.00% BenchmarkClientServerParallel4 61 61 0.00% BenchmarkClientServerParallel64 61 61 0.00% BenchmarkServer 16 16 0.00% BenchmarkServerFakeConnNoKeepAlive 24 24 0.00% BenchmarkServerFakeConnWithKeepAlive 19 19 0.00% BenchmarkServerFakeConnWithKeepAliveLite 9 9 0.00% BenchmarkServerHandlerTypeLen 17 17 0.00% BenchmarkServerHandlerNoLen 14 14 0.00% BenchmarkServerHandlerNoType 15 15 0.00% BenchmarkServerHandlerNoHeader 9 9 0.00% benchmark old bytes new bytes delta BenchmarkClientServer 6988 6985 -0.04% BenchmarkClientServerParallel4 6979 6985 0.09% BenchmarkClientServerParallel64 7002 7019 0.24% BenchmarkServer 1846 1848 0.11% BenchmarkServerFakeConnNoKeepAlive 2420 2412 -0.33% BenchmarkServerFakeConnWithKeepAlive 2126 2129 0.14% BenchmarkServerFakeConnWithKeepAliveLite 989 990 0.10% BenchmarkServerHandlerTypeLen 1818 1819 0.06% BenchmarkServerHandlerNoLen 1775 1777 0.11% BenchmarkServerHandlerNoType 1783 1785 0.11% BenchmarkServerHandlerNoHeader 989 990 0.10% R=golang-dev, r CC=golang-dev https://golang.org/cl/12708046
2013-08-10 19:22:44 -07:00
func newBufioWriterSize(w io.Writer, size int) *bufio.Writer {
pool := bufioWriterPool(size)
if pool != nil {
if v := pool.Get(); v != nil {
bw := v.(*bufio.Writer)
bw.Reset(w)
return bw
}
}
return bufio.NewWriterSize(w, size)
}
net/http: simplify server, use bufio Reader.Reset and Writer.Reset Update #5100 Update #6086 Remove switchReader, switchWriter, switchReaderPair, switchWriterPair, etc. Now it only maintains pools of bufio Readers and Writers, but uses Reset instead of working around all their previously-associated state. Compared to before the bufio Reset change, it's the same number of allocations, and also faster: benchmark old ns/op new ns/op delta BenchmarkClientServer 111218 109828 -1.25% BenchmarkClientServerParallel4 70580 70013 -0.80% BenchmarkClientServerParallel64 72636 68919 -5.12% BenchmarkServer 139858 137068 -1.99% BenchmarkServerFakeConnNoKeepAlive 14619 14314 -2.09% BenchmarkServerFakeConnWithKeepAlive 12390 11361 -8.31% BenchmarkServerFakeConnWithKeepAliveLite 7630 7306 -4.25% BenchmarkServerHandlerTypeLen 9688 9342 -3.57% BenchmarkServerHandlerNoLen 8700 8470 -2.64% BenchmarkServerHandlerNoType 9255 8949 -3.31% BenchmarkServerHandlerNoHeader 7058 6806 -3.57% benchmark old allocs new allocs delta BenchmarkClientServer 61 61 0.00% BenchmarkClientServerParallel4 61 61 0.00% BenchmarkClientServerParallel64 61 61 0.00% BenchmarkServer 16 16 0.00% BenchmarkServerFakeConnNoKeepAlive 24 24 0.00% BenchmarkServerFakeConnWithKeepAlive 19 19 0.00% BenchmarkServerFakeConnWithKeepAliveLite 9 9 0.00% BenchmarkServerHandlerTypeLen 17 17 0.00% BenchmarkServerHandlerNoLen 14 14 0.00% BenchmarkServerHandlerNoType 15 15 0.00% BenchmarkServerHandlerNoHeader 9 9 0.00% benchmark old bytes new bytes delta BenchmarkClientServer 6988 6985 -0.04% BenchmarkClientServerParallel4 6979 6985 0.09% BenchmarkClientServerParallel64 7002 7019 0.24% BenchmarkServer 1846 1848 0.11% BenchmarkServerFakeConnNoKeepAlive 2420 2412 -0.33% BenchmarkServerFakeConnWithKeepAlive 2126 2129 0.14% BenchmarkServerFakeConnWithKeepAliveLite 989 990 0.10% BenchmarkServerHandlerTypeLen 1818 1819 0.06% BenchmarkServerHandlerNoLen 1775 1777 0.11% BenchmarkServerHandlerNoType 1783 1785 0.11% BenchmarkServerHandlerNoHeader 989 990 0.10% R=golang-dev, r CC=golang-dev https://golang.org/cl/12708046
2013-08-10 19:22:44 -07:00
func putBufioWriter(bw *bufio.Writer) {
bw.Reset(nil)
if pool := bufioWriterPool(bw.Available()); pool != nil {
pool.Put(bw)
}
}
// DefaultMaxHeaderBytes is the maximum permitted size of the headers
// in an HTTP request.
// This can be overridden by setting Server.MaxHeaderBytes.
const DefaultMaxHeaderBytes = 1 << 20 // 1 MB
func (srv *Server) maxHeaderBytes() int {
if srv.MaxHeaderBytes > 0 {
return srv.MaxHeaderBytes
}
return DefaultMaxHeaderBytes
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
func (srv *Server) initialReadLimitSize() int64 {
return int64(srv.maxHeaderBytes()) + 4096 // bufio slop
}
// tlsHandshakeTimeout returns the time limit permitted for the TLS
// handshake, or zero for unlimited.
//
// It returns the minimum of any positive ReadHeaderTimeout,
// ReadTimeout, or WriteTimeout.
func (srv *Server) tlsHandshakeTimeout() time.Duration {
var ret time.Duration
for _, v := range [...]time.Duration{
srv.ReadHeaderTimeout,
srv.ReadTimeout,
srv.WriteTimeout,
} {
if v <= 0 {
continue
}
if ret == 0 || v < ret {
ret = v
}
}
return ret
}
// wrapper around io.ReadCloser which on first read, sends an
// HTTP/1.1 100 Continue header
type expectContinueReader struct {
resp *response
readCloser io.ReadCloser
closed atomic.Bool
sawEOF atomic.Bool
}
func (ecr *expectContinueReader) Read(p []byte) (n int, err error) {
if ecr.closed.Load() {
return 0, ErrBodyReadAfterClose
}
w := ecr.resp
if !w.wroteContinue && w.canWriteContinue.Load() && !w.conn.hijacked() {
w.wroteContinue = true
w.writeContinueMu.Lock()
if w.canWriteContinue.Load() {
w.conn.bufw.WriteString("HTTP/1.1 100 Continue\r\n\r\n")
w.conn.bufw.Flush()
w.canWriteContinue.Store(false)
}
w.writeContinueMu.Unlock()
}
n, err = ecr.readCloser.Read(p)
if err == io.EOF {
ecr.sawEOF.Store(true)
}
return
}
func (ecr *expectContinueReader) Close() error {
ecr.closed.Store(true)
return ecr.readCloser.Close()
}
// TimeFormat is the time format to use when generating times in HTTP
// headers. It is like time.RFC1123 but hard-codes GMT as the time
// zone. The time being formatted must be in UTC for Format to
// generate the correct format.
//
// For parsing this time format, see ParseTime.
const TimeFormat = "Mon, 02 Jan 2006 15:04:05 GMT"
// appendTime is a non-allocating version of []byte(t.UTC().Format(TimeFormat))
func appendTime(b []byte, t time.Time) []byte {
const days = "SunMonTueWedThuFriSat"
const months = "JanFebMarAprMayJunJulAugSepOctNovDec"
t = t.UTC()
yy, mm, dd := t.Date()
hh, mn, ss := t.Clock()
day := days[3*t.Weekday():]
mon := months[3*(mm-1):]
return append(b,
day[0], day[1], day[2], ',', ' ',
byte('0'+dd/10), byte('0'+dd%10), ' ',
mon[0], mon[1], mon[2], ' ',
byte('0'+yy/1000), byte('0'+(yy/100)%10), byte('0'+(yy/10)%10), byte('0'+yy%10), ' ',
byte('0'+hh/10), byte('0'+hh%10), ':',
byte('0'+mn/10), byte('0'+mn%10), ':',
byte('0'+ss/10), byte('0'+ss%10), ' ',
'G', 'M', 'T')
}
var errTooLarge = errors.New("http: request too large")
// Read next request from connection.
func (c *conn) readRequest(ctx context.Context) (w *response, err error) {
if c.hijacked() {
return nil, ErrHijacked
}
var (
wholeReqDeadline time.Time // or zero if none
hdrDeadline time.Time // or zero if none
)
t0 := time.Now()
if d := c.server.readHeaderTimeout(); d > 0 {
hdrDeadline = t0.Add(d)
}
if d := c.server.ReadTimeout; d > 0 {
wholeReqDeadline = t0.Add(d)
}
c.rwc.SetReadDeadline(hdrDeadline)
if d := c.server.WriteTimeout; d > 0 {
defer func() {
c.rwc.SetWriteDeadline(time.Now().Add(d))
}()
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
c.r.setReadLimit(c.server.initialReadLimitSize())
if c.lastMethod == "POST" {
// RFC 7230 section 3 tolerance for old buggy clients.
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
peek, _ := c.bufr.Peek(4) // ReadRequest will get err below
c.bufr.Discard(numLeadingCRorLF(peek))
}
req, err := readRequest(c.bufr)
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
if err != nil {
if c.r.hitReadLimit() {
return nil, errTooLarge
}
return nil, err
}
if !http1ServerSupportsRequest(req) {
return nil, statusError{StatusHTTPVersionNotSupported, "unsupported protocol version"}
}
c.lastMethod = req.Method
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
c.r.setInfiniteReadLimit()
hosts, haveHost := req.Header["Host"]
isH2Upgrade := req.isH2Upgrade()
if req.ProtoAtLeast(1, 1) && (!haveHost || len(hosts) == 0) && !isH2Upgrade && req.Method != "CONNECT" {
return nil, badRequestError("missing required Host header")
}
if len(hosts) == 1 && !httpguts.ValidHostHeader(hosts[0]) {
return nil, badRequestError("malformed Host header")
}
for k, vv := range req.Header {
if !httpguts.ValidHeaderFieldName(k) {
return nil, badRequestError("invalid header name")
}
for _, v := range vv {
if !httpguts.ValidHeaderFieldValue(v) {
return nil, badRequestError("invalid header value")
}
}
}
delete(req.Header, "Host")
ctx, cancelCtx := context.WithCancel(ctx)
req.ctx = ctx
req.RemoteAddr = c.remoteAddr
req.TLS = c.tlsState
if body, ok := req.Body.(*body); ok {
body.doEarlyClose = true
}
// Adjust the read deadline if necessary.
if !hdrDeadline.Equal(wholeReqDeadline) {
c.rwc.SetReadDeadline(wholeReqDeadline)
}
w = &response{
conn: c,
cancelCtx: cancelCtx,
req: req,
reqBody: req.Body,
handlerHeader: make(Header),
contentLength: -1,
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
closeNotifyCh: make(chan bool, 1),
// We populate these ahead of time so we're not
// reading from req.Header after their Handler starts
// and maybe mutates it (Issue 14940)
wants10KeepAlive: req.wantsHttp10KeepAlive(),
wantsClose: req.wantsClose(),
}
if isH2Upgrade {
w.closeAfterReply = true
}
w.cw.res = w
net/http: simplify server, use bufio Reader.Reset and Writer.Reset Update #5100 Update #6086 Remove switchReader, switchWriter, switchReaderPair, switchWriterPair, etc. Now it only maintains pools of bufio Readers and Writers, but uses Reset instead of working around all their previously-associated state. Compared to before the bufio Reset change, it's the same number of allocations, and also faster: benchmark old ns/op new ns/op delta BenchmarkClientServer 111218 109828 -1.25% BenchmarkClientServerParallel4 70580 70013 -0.80% BenchmarkClientServerParallel64 72636 68919 -5.12% BenchmarkServer 139858 137068 -1.99% BenchmarkServerFakeConnNoKeepAlive 14619 14314 -2.09% BenchmarkServerFakeConnWithKeepAlive 12390 11361 -8.31% BenchmarkServerFakeConnWithKeepAliveLite 7630 7306 -4.25% BenchmarkServerHandlerTypeLen 9688 9342 -3.57% BenchmarkServerHandlerNoLen 8700 8470 -2.64% BenchmarkServerHandlerNoType 9255 8949 -3.31% BenchmarkServerHandlerNoHeader 7058 6806 -3.57% benchmark old allocs new allocs delta BenchmarkClientServer 61 61 0.00% BenchmarkClientServerParallel4 61 61 0.00% BenchmarkClientServerParallel64 61 61 0.00% BenchmarkServer 16 16 0.00% BenchmarkServerFakeConnNoKeepAlive 24 24 0.00% BenchmarkServerFakeConnWithKeepAlive 19 19 0.00% BenchmarkServerFakeConnWithKeepAliveLite 9 9 0.00% BenchmarkServerHandlerTypeLen 17 17 0.00% BenchmarkServerHandlerNoLen 14 14 0.00% BenchmarkServerHandlerNoType 15 15 0.00% BenchmarkServerHandlerNoHeader 9 9 0.00% benchmark old bytes new bytes delta BenchmarkClientServer 6988 6985 -0.04% BenchmarkClientServerParallel4 6979 6985 0.09% BenchmarkClientServerParallel64 7002 7019 0.24% BenchmarkServer 1846 1848 0.11% BenchmarkServerFakeConnNoKeepAlive 2420 2412 -0.33% BenchmarkServerFakeConnWithKeepAlive 2126 2129 0.14% BenchmarkServerFakeConnWithKeepAliveLite 989 990 0.10% BenchmarkServerHandlerTypeLen 1818 1819 0.06% BenchmarkServerHandlerNoLen 1775 1777 0.11% BenchmarkServerHandlerNoType 1783 1785 0.11% BenchmarkServerHandlerNoHeader 989 990 0.10% R=golang-dev, r CC=golang-dev https://golang.org/cl/12708046
2013-08-10 19:22:44 -07:00
w.w = newBufioWriterSize(&w.cw, bufferBeforeChunkingSize)
return w, nil
}
// http1ServerSupportsRequest reports whether Go's HTTP/1.x server
// supports the given request.
func http1ServerSupportsRequest(req *Request) bool {
if req.ProtoMajor == 1 {
return true
}
// Accept "PRI * HTTP/2.0" upgrade requests, so Handlers can
// wire up their own HTTP/2 upgrades.
if req.ProtoMajor == 2 && req.ProtoMinor == 0 &&
req.Method == "PRI" && req.RequestURI == "*" {
return true
}
// Reject HTTP/0.x, and all other HTTP/2+ requests (which
// aren't encoded in ASCII anyway).
return false
}
func (w *response) Header() Header {
if w.cw.header == nil && w.wroteHeader && !w.cw.wroteHeader {
// Accessing the header between logically writing it
// and physically writing it means we need to allocate
// a clone to snapshot the logically written state.
w.cw.header = w.handlerHeader.Clone()
}
w.calledHeader = true
return w.handlerHeader
}
// maxPostHandlerReadBytes is the max number of Request.Body bytes not
// consumed by a handler that the server will read from the client
// in order to keep a connection alive. If there are more bytes than
// this then the server to be paranoid instead sends a "Connection:
// close" response.
//
// This number is approximately what a typical machine's TCP buffer
// size is anyway. (if we have the bytes on the machine, we might as
// well read them)
const maxPostHandlerReadBytes = 256 << 10
func checkWriteHeaderCode(code int) {
// Issue 22880: require valid WriteHeader status codes.
// For now we only enforce that it's three digits.
// In the future we might block things over 599 (600 and above aren't defined
// at https://httpwg.org/specs/rfc7231.html#status.codes).
// But for now any three digits.
//
// We used to send "HTTP/1.1 000 0" on the wire in responses but there's
// no equivalent bogus thing we can realistically send in HTTP/2,
// so we'll consistently panic instead and help people find their bugs
// early. (We can't return an error from WriteHeader even if we wanted to.)
if code < 100 || code > 999 {
panic(fmt.Sprintf("invalid WriteHeader code %v", code))
}
}
// relevantCaller searches the call stack for the first function outside of net/http.
// The purpose of this function is to provide more helpful error messages.
func relevantCaller() runtime.Frame {
pc := make([]uintptr, 16)
n := runtime.Callers(1, pc)
frames := runtime.CallersFrames(pc[:n])
var frame runtime.Frame
for {
frame, more := frames.Next()
if !strings.HasPrefix(frame.Function, "net/http.") {
return frame
}
if !more {
break
}
}
return frame
}
func (w *response) WriteHeader(code int) {
if w.conn.hijacked() {
caller := relevantCaller()
w.conn.server.logf("http: response.WriteHeader on hijacked connection from %s (%s:%d)", caller.Function, path.Base(caller.File), caller.Line)
return
}
if w.wroteHeader {
caller := relevantCaller()
w.conn.server.logf("http: superfluous response.WriteHeader call from %s (%s:%d)", caller.Function, path.Base(caller.File), caller.Line)
return
}
checkWriteHeaderCode(code)
// Handle informational headers.
//
// We shouldn't send any further headers after 101 Switching Protocols,
// so it takes the non-informational path.
if code >= 100 && code <= 199 && code != StatusSwitchingProtocols {
// Prevent a potential race with an automatically-sent 100 Continue triggered by Request.Body.Read()
if code == 100 && w.canWriteContinue.Load() {
w.writeContinueMu.Lock()
w.canWriteContinue.Store(false)
w.writeContinueMu.Unlock()
}
writeStatusLine(w.conn.bufw, w.req.ProtoAtLeast(1, 1), code, w.statusBuf[:])
// Per RFC 8297 we must not clear the current header map
w.handlerHeader.WriteSubset(w.conn.bufw, excludedHeadersNoBody)
w.conn.bufw.Write(crlf)
w.conn.bufw.Flush()
return
}
w.wroteHeader = true
w.status = code
if w.calledHeader && w.cw.header == nil {
w.cw.header = w.handlerHeader.Clone()
}
if cl := w.handlerHeader.get("Content-Length"); cl != "" {
v, err := strconv.ParseInt(cl, 10, 64)
if err == nil && v >= 0 {
w.contentLength = v
} else {
w.conn.server.logf("http: invalid Content-Length of %q", cl)
w.handlerHeader.Del("Content-Length")
}
}
}
// extraHeader is the set of headers sometimes added by chunkWriter.writeHeader.
// This type is used to avoid extra allocations from cloning and/or populating
// the response Header map and all its 1-element slices.
type extraHeader struct {
contentType string
connection string
transferEncoding string
date []byte // written if not nil
contentLength []byte // written if not nil
}
// Sorted the same as extraHeader.Write's loop.
var extraHeaderKeys = [][]byte{
[]byte("Content-Type"),
[]byte("Connection"),
[]byte("Transfer-Encoding"),
}
var (
headerContentLength = []byte("Content-Length: ")
headerDate = []byte("Date: ")
)
// Write writes the headers described in h to w.
//
// This method has a value receiver, despite the somewhat large size
// of h, because it prevents an allocation. The escape analysis isn't
// smart enough to realize this function doesn't mutate h.
func (h extraHeader) Write(w *bufio.Writer) {
if h.date != nil {
w.Write(headerDate)
w.Write(h.date)
w.Write(crlf)
}
if h.contentLength != nil {
w.Write(headerContentLength)
w.Write(h.contentLength)
w.Write(crlf)
}
for i, v := range []string{h.contentType, h.connection, h.transferEncoding} {
if v != "" {
w.Write(extraHeaderKeys[i])
w.Write(colonSpace)
w.WriteString(v)
w.Write(crlf)
}
}
}
// writeHeader finalizes the header sent to the client and writes it
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
// to cw.res.conn.bufw.
//
// p is not written by writeHeader, but is the first chunk of the body
// that will be written. It is sniffed for a Content-Type if none is
// set explicitly. It's also used to set the Content-Length, if the
// total body size was small and the handler has already finished
// running.
func (cw *chunkWriter) writeHeader(p []byte) {
if cw.wroteHeader {
return
}
cw.wroteHeader = true
w := cw.res
keepAlivesEnabled := w.conn.server.doKeepAlives()
net/http: treat HEAD requests like GET requests A response to a HEAD request is supposed to look the same as a response to a GET request, just without a body. HEAD requests are incredibly rare in the wild. The Go net/http package has so far treated HEAD requests specially: a Write on our default ResponseWriter returned ErrBodyNotAllowed, telling handlers that something was wrong. This was to optimize the fast path for HEAD requests, but: 1) because HEAD requests are incredibly rare, they're not worth having a fast path for. 2) Letting the http.Handler handle but do nop Writes is still very fast. 3) this forces ugly error handling into the application. e.g. https://code.google.com/p/go/source/detail?r=6f596be7a31e and related. 4) The net/http package nowadays does Content-Type sniffing, but you don't get that for HEAD. 5) The net/http package nowadays does Content-Length counting for small (few KB) responses, but not for HEAD. 6) ErrBodyNotAllowed was useless. By the time you received it, you had probably already done all your heavy computation and I/O to calculate what to write. So, this change makes HEAD requests like GET requests. We now count content-length and sniff content-type for HEAD requests. If you Write, it doesn't return an error. If you want a fast-path in your code for HEAD, you have to do it early and set all the response headers yourself. Just like before. If you choose not to Write in HEAD requests, be sure to set Content-Length if you know it. We won't write "Content-Length: 0" because you might've just chosen to not write (or you don't know your Content-Length in advance). Fixes #5454 R=golang-dev, dsymonds CC=golang-dev https://golang.org/cl/12583043
2013-08-06 18:33:03 -07:00
isHEAD := w.req.Method == "HEAD"
// header is written out to w.conn.buf below. Depending on the
// state of the handler, we either own the map or not. If we
// don't own it, the exclude map is created lazily for
// WriteSubset to remove headers. The setHeader struct holds
// headers we need to add.
header := cw.header
owned := header != nil
if !owned {
header = w.handlerHeader
}
var excludeHeader map[string]bool
delHeader := func(key string) {
if owned {
header.Del(key)
return
}
if _, ok := header[key]; !ok {
return
}
if excludeHeader == nil {
excludeHeader = make(map[string]bool)
}
excludeHeader[key] = true
}
var setHeader extraHeader
// Don't write out the fake "Trailer:foo" keys. See TrailerPrefix.
trailers := false
for k := range cw.header {
if strings.HasPrefix(k, TrailerPrefix) {
if excludeHeader == nil {
excludeHeader = make(map[string]bool)
}
excludeHeader[k] = true
trailers = true
}
}
for _, v := range cw.header["Trailer"] {
trailers = true
foreachHeaderElement(v, cw.res.declareTrailer)
}
te := header.get("Transfer-Encoding")
hasTE := te != ""
// If the handler is done but never sent a Content-Length
// response header and this is our first (and last) write, set
// it, even to zero. This helps HTTP/1.0 clients keep their
// "keep-alive" connections alive.
// Exceptions: 304/204/1xx responses never get Content-Length, and if
// it was a HEAD request, we don't know the difference between
// 0 actual bytes and 0 bytes because the handler noticed it
// was a HEAD request and chose not to write anything. So for
// HEAD, the handler should either write the Content-Length or
// write non-zero bytes. If it's actually 0 bytes and the
// handler never looked at the Request.Method, we just don't
// send a Content-Length header.
// Further, we don't send an automatic Content-Length if they
// set a Transfer-Encoding, because they're generally incompatible.
if w.handlerDone.Load() && !trailers && !hasTE && bodyAllowedForStatus(w.status) && !header.has("Content-Length") && (!isHEAD || len(p) > 0) {
w.contentLength = int64(len(p))
setHeader.contentLength = strconv.AppendInt(cw.res.clenBuf[:0], int64(len(p)), 10)
}
// If this was an HTTP/1.0 request with keep-alive and we sent a
// Content-Length back, we can make this a keep-alive response ...
if w.wants10KeepAlive && keepAlivesEnabled {
sentLength := header.get("Content-Length") != ""
if sentLength && header.get("Connection") == "keep-alive" {
w.closeAfterReply = false
}
}
// Check for an explicit (and valid) Content-Length header.
hasCL := w.contentLength != -1
if w.wants10KeepAlive && (isHEAD || hasCL || !bodyAllowedForStatus(w.status)) {
_, connectionHeaderSet := header["Connection"]
if !connectionHeaderSet {
setHeader.connection = "keep-alive"
}
} else if !w.req.ProtoAtLeast(1, 1) || w.wantsClose {
w.closeAfterReply = true
}
if header.get("Connection") == "close" || !keepAlivesEnabled {
w.closeAfterReply = true
}
// If the client wanted a 100-continue but we never sent it to
// them (or, more strictly: we never finished reading their
// request body), don't reuse this connection because it's now
// in an unknown state: we might be sending this response at
// the same time the client is now sending its request body
// after a timeout. (Some HTTP clients send Expect:
// 100-continue but knowing that some servers don't support
// it, the clients set a timer and send the body later anyway)
// If we haven't seen EOF, we can't skip over the unread body
// because we don't know if the next bytes on the wire will be
// the body-following-the-timer or the subsequent request.
// See Issue 11549.
if ecr, ok := w.req.Body.(*expectContinueReader); ok && !ecr.sawEOF.Load() {
w.closeAfterReply = true
}
// We do this by default because there are a number of clients that
// send a full request before starting to read the response, and they
// can deadlock if we start writing the response with unconsumed body
// remaining. See Issue 15527 for some history.
//
// If full duplex mode has been enabled with ResponseController.EnableFullDuplex,
// then leave the request body alone.
if w.req.ContentLength != 0 && !w.closeAfterReply && !w.fullDuplex {
var discard, tooBig bool
switch bdy := w.req.Body.(type) {
case *expectContinueReader:
if bdy.resp.wroteContinue {
discard = true
}
case *body:
bdy.mu.Lock()
switch {
case bdy.closed:
if !bdy.sawEOF {
// Body was closed in handler with non-EOF error.
w.closeAfterReply = true
}
case bdy.unreadDataSizeLocked() >= maxPostHandlerReadBytes:
tooBig = true
default:
discard = true
}
bdy.mu.Unlock()
default:
discard = true
}
if discard {
_, err := io.CopyN(io.Discard, w.reqBody, maxPostHandlerReadBytes+1)
switch err {
case nil:
// There must be even more data left over.
tooBig = true
case ErrBodyReadAfterClose:
// Body was already consumed and closed.
case io.EOF:
// The remaining body was just consumed, close it.
err = w.reqBody.Close()
if err != nil {
w.closeAfterReply = true
}
default:
// Some other kind of error occurred, like a read timeout, or
// corrupt chunked encoding. In any case, whatever remains
// on the wire must not be parsed as another HTTP request.
w.closeAfterReply = true
}
}
if tooBig {
w.requestTooLarge()
delHeader("Connection")
setHeader.connection = "close"
}
}
code := w.status
if bodyAllowedForStatus(code) {
// If no content type, apply sniffing algorithm to body.
_, haveType := header["Content-Type"]
// If the Content-Encoding was set and is non-blank,
// we shouldn't sniff the body. See Issue 31753.
ce := header.Get("Content-Encoding")
hasCE := len(ce) > 0
if !hasCE && !haveType && !hasTE && len(p) > 0 {
setHeader.contentType = DetectContentType(p)
}
} else {
for _, k := range suppressedHeaders(code) {
delHeader(k)
}
}
if !header.has("Date") {
setHeader.date = appendTime(cw.res.dateBuf[:0], time.Now())
}
if hasCL && hasTE && te != "identity" {
// TODO: return an error if WriteHeader gets a return parameter
// For now just ignore the Content-Length.
w.conn.server.logf("http: WriteHeader called with both Transfer-Encoding of %q and a Content-Length of %d",
te, w.contentLength)
delHeader("Content-Length")
hasCL = false
}
if w.req.Method == "HEAD" || !bodyAllowedForStatus(code) || code == StatusNoContent {
// Response has no body.
delHeader("Transfer-Encoding")
} else if hasCL {
// Content-Length has been provided, so no chunking is to be done.
delHeader("Transfer-Encoding")
} else if w.req.ProtoAtLeast(1, 1) {
// HTTP/1.1 or greater: Transfer-Encoding has been set to identity, and no
// content-length has been provided. The connection must be closed after the
// reply is written, and no chunking is to be done. This is the setup
// recommended in the Server-Sent Events candidate recommendation 11,
// section 8.
if hasTE && te == "identity" {
cw.chunking = false
w.closeAfterReply = true
delHeader("Transfer-Encoding")
} else {
// HTTP/1.1 or greater: use chunked transfer encoding
// to avoid closing the connection at EOF.
cw.chunking = true
setHeader.transferEncoding = "chunked"
if hasTE && te == "chunked" {
// We will send the chunked Transfer-Encoding header later.
delHeader("Transfer-Encoding")
}
}
} else {
// HTTP version < 1.1: cannot do chunked transfer
// encoding and we don't know the Content-Length so
// signal EOF by closing connection.
w.closeAfterReply = true
delHeader("Transfer-Encoding") // in case already set
}
// Cannot use Content-Length with non-identity Transfer-Encoding.
if cw.chunking {
delHeader("Content-Length")
}
if !w.req.ProtoAtLeast(1, 0) {
return
}
// Only override the Connection header if it is not a successful
// protocol switch response and if KeepAlives are not enabled.
// See https://golang.org/issue/36381.
delConnectionHeader := w.closeAfterReply &&
(!keepAlivesEnabled || !hasToken(cw.header.get("Connection"), "close")) &&
!isProtocolSwitchResponse(w.status, header)
if delConnectionHeader {
delHeader("Connection")
if w.req.ProtoAtLeast(1, 1) {
setHeader.connection = "close"
}
}
net/http: re-simplify HTTP/1.x status line writing It used to be simple, and then it got complicated for speed (to reduce allocations, mostly), but that involved a mutex and hurt multi-core performance, contending on the mutex. A change was sent to try to improve that mutex contention in https://go-review.googlesource.com/c/42110/2/src/net/http/server.go but that introduced its own allocations (the string->interface{} boxing for the sync.Map key), which runs counter to the whole point of that statusLine function: to remove allocations. Instead, make the code simple again and not have a mutex. It's a bit slower for the single-core case, but nobody with a single-user HTTP server cares about 50 nanoseconds: name old time/op new time/op delta ResponseStatusLine 37.5ns ± 2% 87.1ns ± 2% +132.42% (p=0.029 n=4+4) ResponseStatusLine-2 63.1ns ± 1% 43.1ns ±12% -31.67% (p=0.029 n=4+4) ResponseStatusLine-4 53.8ns ± 8% 40.2ns ± 2% -25.29% (p=0.029 n=4+4) name old alloc/op new alloc/op delta ResponseStatusLine 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) name old allocs/op new allocs/op delta ResponseStatusLine 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) (Note the code could be even simpler with fmt.Fprintf, but that is relatively slow and involves a bunch of allocations getting arguments into interface{} for the call) Change-Id: I1fa119132dbbf97a8e7204ce3e0707d433060da2 Reviewed-on: https://go-review.googlesource.com/42133 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan Mills <bcmills@google.com>
2017-04-28 16:46:18 +00:00
writeStatusLine(w.conn.bufw, w.req.ProtoAtLeast(1, 1), code, w.statusBuf[:])
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
cw.header.WriteSubset(w.conn.bufw, excludeHeader)
setHeader.Write(w.conn.bufw)
w.conn.bufw.Write(crlf)
}
// foreachHeaderElement splits v according to the "#rule" construction
// in RFC 7230 section 7 and calls fn for each non-empty element.
func foreachHeaderElement(v string, fn func(string)) {
v = textproto.TrimString(v)
if v == "" {
return
}
if !strings.Contains(v, ",") {
fn(v)
return
}
for _, f := range strings.Split(v, ",") {
if f = textproto.TrimString(f); f != "" {
fn(f)
}
}
}
// writeStatusLine writes an HTTP/1.x Status-Line (RFC 7230 Section 3.1.2)
net/http: re-simplify HTTP/1.x status line writing It used to be simple, and then it got complicated for speed (to reduce allocations, mostly), but that involved a mutex and hurt multi-core performance, contending on the mutex. A change was sent to try to improve that mutex contention in https://go-review.googlesource.com/c/42110/2/src/net/http/server.go but that introduced its own allocations (the string->interface{} boxing for the sync.Map key), which runs counter to the whole point of that statusLine function: to remove allocations. Instead, make the code simple again and not have a mutex. It's a bit slower for the single-core case, but nobody with a single-user HTTP server cares about 50 nanoseconds: name old time/op new time/op delta ResponseStatusLine 37.5ns ± 2% 87.1ns ± 2% +132.42% (p=0.029 n=4+4) ResponseStatusLine-2 63.1ns ± 1% 43.1ns ±12% -31.67% (p=0.029 n=4+4) ResponseStatusLine-4 53.8ns ± 8% 40.2ns ± 2% -25.29% (p=0.029 n=4+4) name old alloc/op new alloc/op delta ResponseStatusLine 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) name old allocs/op new allocs/op delta ResponseStatusLine 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) (Note the code could be even simpler with fmt.Fprintf, but that is relatively slow and involves a bunch of allocations getting arguments into interface{} for the call) Change-Id: I1fa119132dbbf97a8e7204ce3e0707d433060da2 Reviewed-on: https://go-review.googlesource.com/42133 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan Mills <bcmills@google.com>
2017-04-28 16:46:18 +00:00
// to bw. is11 is whether the HTTP request is HTTP/1.1. false means HTTP/1.0.
// code is the response status code.
// scratch is an optional scratch buffer. If it has at least capacity 3, it's used.
func writeStatusLine(bw *bufio.Writer, is11 bool, code int, scratch []byte) {
if is11 {
bw.WriteString("HTTP/1.1 ")
} else {
bw.WriteString("HTTP/1.0 ")
}
if text := StatusText(code); text != "" {
net/http: re-simplify HTTP/1.x status line writing It used to be simple, and then it got complicated for speed (to reduce allocations, mostly), but that involved a mutex and hurt multi-core performance, contending on the mutex. A change was sent to try to improve that mutex contention in https://go-review.googlesource.com/c/42110/2/src/net/http/server.go but that introduced its own allocations (the string->interface{} boxing for the sync.Map key), which runs counter to the whole point of that statusLine function: to remove allocations. Instead, make the code simple again and not have a mutex. It's a bit slower for the single-core case, but nobody with a single-user HTTP server cares about 50 nanoseconds: name old time/op new time/op delta ResponseStatusLine 37.5ns ± 2% 87.1ns ± 2% +132.42% (p=0.029 n=4+4) ResponseStatusLine-2 63.1ns ± 1% 43.1ns ±12% -31.67% (p=0.029 n=4+4) ResponseStatusLine-4 53.8ns ± 8% 40.2ns ± 2% -25.29% (p=0.029 n=4+4) name old alloc/op new alloc/op delta ResponseStatusLine 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00B ±NaN% 0.00B ±NaN% ~ (all samples are equal) name old allocs/op new allocs/op delta ResponseStatusLine 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) ResponseStatusLine-4 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal) (Note the code could be even simpler with fmt.Fprintf, but that is relatively slow and involves a bunch of allocations getting arguments into interface{} for the call) Change-Id: I1fa119132dbbf97a8e7204ce3e0707d433060da2 Reviewed-on: https://go-review.googlesource.com/42133 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan Mills <bcmills@google.com>
2017-04-28 16:46:18 +00:00
bw.Write(strconv.AppendInt(scratch[:0], int64(code), 10))
bw.WriteByte(' ')
bw.WriteString(text)
bw.WriteString("\r\n")
} else {
// don't worry about performance
fmt.Fprintf(bw, "%03d status code %d\r\n", code, code)
}
}
// bodyAllowed reports whether a Write is allowed for this response type.
// It's illegal to call this before the header has been flushed.
func (w *response) bodyAllowed() bool {
if !w.wroteHeader {
panic("")
}
return bodyAllowedForStatus(w.status)
}
// The Life Of A Write is like this:
//
// Handler starts. No header has been sent. The handler can either
// write a header, or just start writing. Writing before sending a header
// sends an implicitly empty 200 OK header.
//
// If the handler didn't declare a Content-Length up front, we either
// go into chunking mode or, if the handler finishes running before
// the chunking buffer size, we compute a Content-Length and send that
// in the header instead.
//
// Likewise, if the handler didn't set a Content-Type, we sniff that
// from the initial chunk of output.
//
// The Writers are wired together like:
//
// 1. *response (the ResponseWriter) ->
// 2. (*response).w, a *bufio.Writer of bufferBeforeChunkingSize bytes ->
// 3. chunkWriter.Writer (whose writeHeader finalizes Content-Length/Type)
// and which writes the chunk headers, if needed ->
// 4. conn.bufw, a *bufio.Writer of default (4kB) bytes, writing to ->
// 5. checkConnErrorWriter{c}, which notes any non-nil error on Write
// and populates c.werr with it if so, but otherwise writes to ->
// 6. the rwc, the net.Conn.
//
// TODO(bradfitz): short-circuit some of the buffering when the
// initial header contains both a Content-Type and Content-Length.
// Also short-circuit in (1) when the header's been sent and not in
// chunking mode, writing directly to (4) instead, if (2) has no
// buffered data. More generally, we could short-circuit from (1) to
// (3) even in chunking mode if the write size from (1) is over some
// threshold and nothing is in (2). The answer might be mostly making
// bufferBeforeChunkingSize smaller and having bufio's fast-paths deal
// with this instead.
func (w *response) Write(data []byte) (n int, err error) {
return w.write(len(data), data, "")
}
func (w *response) WriteString(data string) (n int, err error) {
return w.write(len(data), nil, data)
}
// either dataB or dataS is non-zero.
func (w *response) write(lenData int, dataB []byte, dataS string) (n int, err error) {
if w.conn.hijacked() {
if lenData > 0 {
caller := relevantCaller()
w.conn.server.logf("http: response.Write on hijacked connection from %s (%s:%d)", caller.Function, path.Base(caller.File), caller.Line)
}
return 0, ErrHijacked
}
if w.canWriteContinue.Load() {
// Body reader wants to write 100 Continue but hasn't yet.
// Tell it not to. The store must be done while holding the lock
// because the lock makes sure that there is not an active write
// this very moment.
w.writeContinueMu.Lock()
w.canWriteContinue.Store(false)
w.writeContinueMu.Unlock()
}
if !w.wroteHeader {
w.WriteHeader(StatusOK)
}
if lenData == 0 {
return 0, nil
}
if !w.bodyAllowed() {
return 0, ErrBodyNotAllowed
}
w.written += int64(lenData) // ignoring errors, for errorKludge
if w.contentLength != -1 && w.written > w.contentLength {
return 0, ErrContentLength
}
if dataB != nil {
return w.w.Write(dataB)
} else {
return w.w.WriteString(dataS)
}
}
func (w *response) finishRequest() {
w.handlerDone.Store(true)
if !w.wroteHeader {
w.WriteHeader(StatusOK)
}
w.w.Flush()
net/http: simplify server, use bufio Reader.Reset and Writer.Reset Update #5100 Update #6086 Remove switchReader, switchWriter, switchReaderPair, switchWriterPair, etc. Now it only maintains pools of bufio Readers and Writers, but uses Reset instead of working around all their previously-associated state. Compared to before the bufio Reset change, it's the same number of allocations, and also faster: benchmark old ns/op new ns/op delta BenchmarkClientServer 111218 109828 -1.25% BenchmarkClientServerParallel4 70580 70013 -0.80% BenchmarkClientServerParallel64 72636 68919 -5.12% BenchmarkServer 139858 137068 -1.99% BenchmarkServerFakeConnNoKeepAlive 14619 14314 -2.09% BenchmarkServerFakeConnWithKeepAlive 12390 11361 -8.31% BenchmarkServerFakeConnWithKeepAliveLite 7630 7306 -4.25% BenchmarkServerHandlerTypeLen 9688 9342 -3.57% BenchmarkServerHandlerNoLen 8700 8470 -2.64% BenchmarkServerHandlerNoType 9255 8949 -3.31% BenchmarkServerHandlerNoHeader 7058 6806 -3.57% benchmark old allocs new allocs delta BenchmarkClientServer 61 61 0.00% BenchmarkClientServerParallel4 61 61 0.00% BenchmarkClientServerParallel64 61 61 0.00% BenchmarkServer 16 16 0.00% BenchmarkServerFakeConnNoKeepAlive 24 24 0.00% BenchmarkServerFakeConnWithKeepAlive 19 19 0.00% BenchmarkServerFakeConnWithKeepAliveLite 9 9 0.00% BenchmarkServerHandlerTypeLen 17 17 0.00% BenchmarkServerHandlerNoLen 14 14 0.00% BenchmarkServerHandlerNoType 15 15 0.00% BenchmarkServerHandlerNoHeader 9 9 0.00% benchmark old bytes new bytes delta BenchmarkClientServer 6988 6985 -0.04% BenchmarkClientServerParallel4 6979 6985 0.09% BenchmarkClientServerParallel64 7002 7019 0.24% BenchmarkServer 1846 1848 0.11% BenchmarkServerFakeConnNoKeepAlive 2420 2412 -0.33% BenchmarkServerFakeConnWithKeepAlive 2126 2129 0.14% BenchmarkServerFakeConnWithKeepAliveLite 989 990 0.10% BenchmarkServerHandlerTypeLen 1818 1819 0.06% BenchmarkServerHandlerNoLen 1775 1777 0.11% BenchmarkServerHandlerNoType 1783 1785 0.11% BenchmarkServerHandlerNoHeader 989 990 0.10% R=golang-dev, r CC=golang-dev https://golang.org/cl/12708046
2013-08-10 19:22:44 -07:00
putBufioWriter(w.w)
w.cw.close()
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
w.conn.bufw.Flush()
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
w.conn.r.abortPendingRead()
// Close the body (regardless of w.closeAfterReply) so we can
// re-use its bufio.Reader later safely.
w.reqBody.Close()
if w.req.MultipartForm != nil {
w.req.MultipartForm.RemoveAll()
}
}
// shouldReuseConnection reports whether the underlying TCP connection can be reused.
// It must only be called after the handler is done executing.
func (w *response) shouldReuseConnection() bool {
if w.closeAfterReply {
// The request or something set while executing the
// handler indicated we shouldn't reuse this
// connection.
return false
}
net/http: treat HEAD requests like GET requests A response to a HEAD request is supposed to look the same as a response to a GET request, just without a body. HEAD requests are incredibly rare in the wild. The Go net/http package has so far treated HEAD requests specially: a Write on our default ResponseWriter returned ErrBodyNotAllowed, telling handlers that something was wrong. This was to optimize the fast path for HEAD requests, but: 1) because HEAD requests are incredibly rare, they're not worth having a fast path for. 2) Letting the http.Handler handle but do nop Writes is still very fast. 3) this forces ugly error handling into the application. e.g. https://code.google.com/p/go/source/detail?r=6f596be7a31e and related. 4) The net/http package nowadays does Content-Type sniffing, but you don't get that for HEAD. 5) The net/http package nowadays does Content-Length counting for small (few KB) responses, but not for HEAD. 6) ErrBodyNotAllowed was useless. By the time you received it, you had probably already done all your heavy computation and I/O to calculate what to write. So, this change makes HEAD requests like GET requests. We now count content-length and sniff content-type for HEAD requests. If you Write, it doesn't return an error. If you want a fast-path in your code for HEAD, you have to do it early and set all the response headers yourself. Just like before. If you choose not to Write in HEAD requests, be sure to set Content-Length if you know it. We won't write "Content-Length: 0" because you might've just chosen to not write (or you don't know your Content-Length in advance). Fixes #5454 R=golang-dev, dsymonds CC=golang-dev https://golang.org/cl/12583043
2013-08-06 18:33:03 -07:00
if w.req.Method != "HEAD" && w.contentLength != -1 && w.bodyAllowed() && w.contentLength != w.written {
// Did not write enough. Avoid getting out of sync.
return false
}
// There was some error writing to the underlying connection
// during the request, so don't re-use this conn.
if w.conn.werr != nil {
return false
}
if w.closedRequestBodyEarly() {
return false
}
return true
}
func (w *response) closedRequestBodyEarly() bool {
body, ok := w.req.Body.(*body)
return ok && body.didEarlyClose()
}
func (w *response) Flush() {
w.FlushError()
}
func (w *response) FlushError() error {
if !w.wroteHeader {
w.WriteHeader(StatusOK)
}
err := w.w.Flush()
e2 := w.cw.flush()
if err == nil {
err = e2
}
return err
}
func (c *conn) finalFlush() {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
if c.bufr != nil {
// Steal the bufio.Reader (~4KB worth of memory) and its associated
// reader for a future connection.
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
putBufioReader(c.bufr)
c.bufr = nil
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
if c.bufw != nil {
c.bufw.Flush()
// Steal the bufio.Writer (~4KB worth of memory) and its associated
// writer for a future connection.
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
putBufioWriter(c.bufw)
c.bufw = nil
}
}
// Close the connection.
func (c *conn) close() {
c.finalFlush()
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
c.rwc.Close()
}
// rstAvoidanceDelay is the amount of time we sleep after closing the
// write side of a TCP connection before closing the entire socket.
// By sleeping, we increase the chances that the client sees our FIN
// and processes its final data before they process the subsequent RST
// from closing a connection with known unread data.
// This RST seems to occur mostly on BSD systems. (And Windows?)
// This timeout is somewhat arbitrary (~latency around the planet),
// and may be modified by tests.
//
// TODO(bcmills): This should arguably be a server configuration parameter,
// not a hard-coded value.
var rstAvoidanceDelay = 500 * time.Millisecond
type closeWriter interface {
CloseWrite() error
}
var _ closeWriter = (*net.TCPConn)(nil)
// closeWriteAndWait flushes any outstanding data and sends a FIN packet (if
// client is connected via TCP), signaling that we're done. We then
// pause for a bit, hoping the client processes it before any
// subsequent RST.
//
// See https://golang.org/issue/3595
func (c *conn) closeWriteAndWait() {
c.finalFlush()
if tcp, ok := c.rwc.(closeWriter); ok {
tcp.CloseWrite()
}
// When we return from closeWriteAndWait, the caller will fully close the
// connection. If client is still writing to the connection, this will cause
// the write to fail with ECONNRESET or similar. Unfortunately, many TCP
// implementations will also drop unread packets from the client's read buffer
// when a write fails, causing our final response to be truncated away too.
//
// As a result, https://www.rfc-editor.org/rfc/rfc7230#section-6.6 recommends
// that “[t]he server … continues to read from the connection until it
// receives a corresponding close by the client, or until the server is
// reasonably certain that its own TCP stack has received the client's
// acknowledgement of the packet(s) containing the server's last response.”
//
// Unfortunately, we have no straightforward way to be “reasonably certain”
// that we have received the client's ACK, and at any rate we don't want to
// allow a misbehaving client to soak up server connections indefinitely by
// withholding an ACK, nor do we want to go through the complexity or overhead
// of using low-level APIs to figure out when a TCP round-trip has completed.
//
// Instead, we declare that we are “reasonably certain” that we received the
// ACK if maxRSTAvoidanceDelay has elapsed.
time.Sleep(rstAvoidanceDelay)
}
// validNextProto reports whether the proto is a valid ALPN protocol name.
// Everything is valid except the empty string and built-in protocol types,
// so that those can't be overridden with alternate implementations.
func validNextProto(proto string) bool {
switch proto {
case "", "http/1.1", "http/1.0":
return false
}
return true
}
const (
runHooks = true
skipHooks = false
)
func (c *conn) setState(nc net.Conn, state ConnState, runHook bool) {
srv := c.server
switch state {
case StateNew:
srv.trackConn(c, true)
case StateHijacked, StateClosed:
srv.trackConn(c, false)
}
if state > 0xff || state < 0 {
panic("internal error")
}
packedState := uint64(time.Now().Unix()<<8) | uint64(state)
c.curState.Store(packedState)
if !runHook {
return
}
if hook := srv.ConnState; hook != nil {
hook(nc, state)
}
}
func (c *conn) getState() (state ConnState, unixSec int64) {
packedState := c.curState.Load()
return ConnState(packedState & 0xff), int64(packedState >> 8)
}
// badRequestError is a literal string (used by in the server in HTML,
// unescaped) to tell the user why their request was bad. It should
// be plain text without user info or other embedded errors.
func badRequestError(e string) error { return statusError{StatusBadRequest, e} }
// statusError is an error used to respond to a request with an HTTP status.
// The text should be plain text without user info or other embedded errors.
type statusError struct {
code int
text string
}
func (e statusError) Error() string { return StatusText(e.code) + ": " + e.text }
// ErrAbortHandler is a sentinel panic value to abort a handler.
// While any panic from ServeHTTP aborts the response to the client,
// panicking with ErrAbortHandler also suppresses logging of a stack
// trace to the server's error log.
var ErrAbortHandler = errors.New("net/http: abort Handler")
// isCommonNetReadError reports whether err is a common error
// encountered during reading a request off the network when the
// client has gone away or had its read fail somehow. This is used to
// determine which logs are interesting enough to log about.
func isCommonNetReadError(err error) bool {
if err == io.EOF {
return true
}
if neterr, ok := err.(net.Error); ok && neterr.Timeout() {
return true
}
if oe, ok := err.(*net.OpError); ok && oe.Op == "read" {
return true
}
return false
}
// Serve a new connection.
func (c *conn) serve(ctx context.Context) {
if ra := c.rwc.RemoteAddr(); ra != nil {
c.remoteAddr = ra.String()
}
ctx = context.WithValue(ctx, LocalAddrContextKey, c.rwc.LocalAddr())
var inFlightResponse *response
defer func() {
if err := recover(); err != nil && err != ErrAbortHandler {
const size = 64 << 10
buf := make([]byte, size)
buf = buf[:runtime.Stack(buf, false)]
c.server.logf("http: panic serving %v: %v\n%s", c.remoteAddr, err, buf)
}
if inFlightResponse != nil {
inFlightResponse.cancelCtx()
}
if !c.hijacked() {
if inFlightResponse != nil {
inFlightResponse.conn.r.abortPendingRead()
inFlightResponse.reqBody.Close()
}
c.close()
c.setState(c.rwc, StateClosed, runHooks)
}
}()
if tlsConn, ok := c.rwc.(*tls.Conn); ok {
tlsTO := c.server.tlsHandshakeTimeout()
if tlsTO > 0 {
dl := time.Now().Add(tlsTO)
c.rwc.SetReadDeadline(dl)
c.rwc.SetWriteDeadline(dl)
}
if err := tlsConn.HandshakeContext(ctx); err != nil {
// If the handshake failed due to the client not speaking
// TLS, assume they're speaking plaintext HTTP and write a
// 400 response on the TLS conn's underlying net.Conn.
if re, ok := err.(tls.RecordHeaderError); ok && re.Conn != nil && tlsRecordHeaderLooksLikeHTTP(re.RecordHeader) {
io.WriteString(re.Conn, "HTTP/1.0 400 Bad Request\r\n\r\nClient sent an HTTP request to an HTTPS server.\n")
re.Conn.Close()
return
}
c.server.logf("http: TLS handshake error from %s: %v", c.rwc.RemoteAddr(), err)
return
}
// Restore Conn-level deadlines.
if tlsTO > 0 {
c.rwc.SetReadDeadline(time.Time{})
c.rwc.SetWriteDeadline(time.Time{})
}
c.tlsState = new(tls.ConnectionState)
*c.tlsState = tlsConn.ConnectionState()
if proto := c.tlsState.NegotiatedProtocol; validNextProto(proto) {
if fn := c.server.TLSNextProto[proto]; fn != nil {
h := initALPNRequest{ctx, tlsConn, serverHandler{c.server}}
// Mark freshly created HTTP/2 as active and prevent any server state hooks
// from being run on these connections. This prevents closeIdleConns from
// closing such connections. See issue https://golang.org/issue/39776.
c.setState(c.rwc, StateActive, skipHooks)
fn(c.server, tlsConn, h)
}
return
}
}
// HTTP/1.x from here on.
ctx, cancelCtx := context.WithCancel(ctx)
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
c.cancelCtx = cancelCtx
defer cancelCtx()
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
c.r = &connReader{conn: c}
c.bufr = newBufioReader(c.r)
c.bufw = newBufioWriterSize(checkConnErrorWriter{c}, 4<<10)
for {
w, err := c.readRequest(ctx)
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
if c.r.remain != c.server.initialReadLimitSize() {
// If we read any bytes off the wire, we're active.
c.setState(c.rwc, StateActive, runHooks)
}
if err != nil {
const errorHeaders = "\r\nContent-Type: text/plain; charset=utf-8\r\nConnection: close\r\n\r\n"
switch {
case err == errTooLarge:
// Their HTTP client may or may not be
// able to read this if we're
// responding to them and hanging up
// while they're still writing their
// request. Undefined behavior.
const publicErr = "431 Request Header Fields Too Large"
fmt.Fprintf(c.rwc, "HTTP/1.1 "+publicErr+errorHeaders+publicErr)
c.closeWriteAndWait()
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
return
case isUnsupportedTEError(err):
// Respond as per RFC 7230 Section 3.3.1 which says,
// A server that receives a request message with a
// transfer coding it does not understand SHOULD
// respond with 501 (Unimplemented).
code := StatusNotImplemented
// We purposefully aren't echoing back the transfer-encoding's value,
// so as to mitigate the risk of cross side scripting by an attacker.
fmt.Fprintf(c.rwc, "HTTP/1.1 %d %s%sUnsupported transfer encoding", code, StatusText(code), errorHeaders)
return
case isCommonNetReadError(err):
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
return // don't reply
default:
if v, ok := err.(statusError); ok {
fmt.Fprintf(c.rwc, "HTTP/1.1 %d %s: %s%s%d %s: %s", v.code, StatusText(v.code), v.text, errorHeaders, v.code, StatusText(v.code), v.text)
return
}
const publicErr = "400 Bad Request"
fmt.Fprintf(c.rwc, "HTTP/1.1 "+publicErr+errorHeaders+publicErr)
return
}
}
// Expect 100 Continue support
req := w.req
if req.expectsContinue() {
if req.ProtoAtLeast(1, 1) && req.ContentLength != 0 {
// Wrap the Body reader with one that replies on the connection
req.Body = &expectContinueReader{readCloser: req.Body, resp: w}
w.canWriteContinue.Store(true)
}
} else if req.Header.get("Expect") != "" {
w.sendExpectationFailed()
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
return
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
c.curReq.Store(w)
if requestBodyRemains(req.Body) {
registerOnHitEOF(req.Body, w.conn.r.startBackgroundRead)
} else {
w.conn.r.startBackgroundRead()
}
// HTTP cannot have multiple simultaneous active requests.[*]
// Until the server replies to this request, it can't read another,
// so we might as well run the handler in this goroutine.
// [*] Not strictly true: HTTP pipelining. We could let them all process
// in parallel even if their responses need to be serialized.
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
// But we're not going to implement HTTP pipelining because it
// was never deployed in the wild and the answer is HTTP/2.
inFlightResponse = w
serverHandler{c.server}.ServeHTTP(w, w.req)
inFlightResponse = nil
w.cancelCtx()
if c.hijacked() {
return
}
w.finishRequest()
c.rwc.SetWriteDeadline(time.Time{})
if !w.shouldReuseConnection() {
if w.requestBodyLimitHit || w.closedRequestBodyEarly() {
c.closeWriteAndWait()
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
return
}
c.setState(c.rwc, StateIdle, runHooks)
c.curReq.Store(nil)
if !w.conn.server.doKeepAlives() {
// We're in shutdown mode. We might've replied
// to the user without "Connection: close" and
// they might think they can send another
// request, but such is life with HTTP/1.1.
return
}
if d := c.server.idleTimeout(); d != 0 {
c.rwc.SetReadDeadline(time.Now().Add(d))
} else {
c.rwc.SetReadDeadline(time.Time{})
}
// Wait for the connection to become readable again before trying to
// read the next request. This prevents a ReadHeaderTimeout or
// ReadTimeout from starting until the first bytes of the next request
// have been received.
if _, err := c.bufr.Peek(4); err != nil {
return
}
c.rwc.SetReadDeadline(time.Time{})
}
}
func (w *response) sendExpectationFailed() {
// TODO(bradfitz): let ServeHTTP handlers handle
// requests with non-standard expectation[s]? Seems
// theoretical at best, and doesn't fit into the
// current ServeHTTP model anyway. We'd need to
// make the ResponseWriter an optional
// "ExpectReplier" interface or something.
//
// For now we'll just obey RFC 7231 5.1.1 which says
// "A server that receives an Expect field-value other
// than 100-continue MAY respond with a 417 (Expectation
// Failed) status code to indicate that the unexpected
// expectation cannot be met."
w.Header().Set("Connection", "close")
w.WriteHeader(StatusExpectationFailed)
w.finishRequest()
}
// Hijack implements the Hijacker.Hijack method. Our response is both a ResponseWriter
// and a Hijacker.
func (w *response) Hijack() (rwc net.Conn, buf *bufio.ReadWriter, err error) {
if w.handlerDone.Load() {
panic("net/http: Hijack called after ServeHTTP finished")
}
if w.wroteHeader {
w.cw.flush()
}
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
c := w.conn
c.mu.Lock()
defer c.mu.Unlock()
// Release the bufioWriter that writes to the chunk writer, it is not
// used after a connection has been hijacked.
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
rwc, buf, err = c.hijackLocked()
if err == nil {
putBufioWriter(w.w)
w.w = nil
}
return rwc, buf, err
}
func (w *response) CloseNotify() <-chan bool {
if w.handlerDone.Load() {
panic("net/http: CloseNotify called after ServeHTTP finished")
}
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
return w.closeNotifyCh
}
func registerOnHitEOF(rc io.ReadCloser, fn func()) {
switch v := rc.(type) {
case *expectContinueReader:
registerOnHitEOF(v.readCloser, fn)
case *body:
v.registerOnHitEOF(fn)
default:
panic("unexpected type " + fmt.Sprintf("%T", rc))
}
}
// requestBodyRemains reports whether future calls to Read
// on rc might yield more data.
func requestBodyRemains(rc io.ReadCloser) bool {
if rc == NoBody {
return false
}
switch v := rc.(type) {
case *expectContinueReader:
return requestBodyRemains(v.readCloser)
case *body:
return v.bodyRemains()
default:
panic("unexpected type " + fmt.Sprintf("%T", rc))
}
}
// The HandlerFunc type is an adapter to allow the use of
// ordinary functions as HTTP handlers. If f is a function
// with the appropriate signature, HandlerFunc(f) is a
// Handler that calls f.
type HandlerFunc func(ResponseWriter, *Request)
// ServeHTTP calls f(w, r).
func (f HandlerFunc) ServeHTTP(w ResponseWriter, r *Request) {
f(w, r)
}
// Helper handlers
// Error replies to the request with the specified error message and HTTP code.
// It does not otherwise end the request; the caller should ensure no further
// writes are done to w.
// The error message should be plain text.
func Error(w ResponseWriter, error string, code int) {
w.Header().Set("Content-Type", "text/plain; charset=utf-8")
w.Header().Set("X-Content-Type-Options", "nosniff")
w.WriteHeader(code)
fmt.Fprintln(w, error)
}
// NotFound replies to the request with an HTTP 404 not found error.
func NotFound(w ResponseWriter, r *Request) { Error(w, "404 page not found", StatusNotFound) }
// NotFoundHandler returns a simple request handler
// that replies to each request with a “404 page not found” reply.
func NotFoundHandler() Handler { return HandlerFunc(NotFound) }
// StripPrefix returns a handler that serves HTTP requests by removing the
// given prefix from the request URL's Path (and RawPath if set) and invoking
// the handler h. StripPrefix handles a request for a path that doesn't begin
// with prefix by replying with an HTTP 404 not found error. The prefix must
// match exactly: if the prefix in the request contains escaped characters
// the reply is also an HTTP 404 not found error.
func StripPrefix(prefix string, h Handler) Handler {
if prefix == "" {
return h
}
return HandlerFunc(func(w ResponseWriter, r *Request) {
p := strings.TrimPrefix(r.URL.Path, prefix)
rp := strings.TrimPrefix(r.URL.RawPath, prefix)
if len(p) < len(r.URL.Path) && (r.URL.RawPath == "" || len(rp) < len(r.URL.RawPath)) {
r2 := new(Request)
*r2 = *r
r2.URL = new(url.URL)
*r2.URL = *r.URL
r2.URL.Path = p
r2.URL.RawPath = rp
h.ServeHTTP(w, r2)
} else {
NotFound(w, r)
}
})
}
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
// Redirect replies to the request with a redirect to url,
// which may be a path relative to the request path.
//
// The provided code should be in the 3xx range and is usually
// StatusMovedPermanently, StatusFound or StatusSeeOther.
//
// If the Content-Type header has not been set, Redirect sets it
// to "text/html; charset=utf-8" and writes a small HTML body.
// Setting the Content-Type header to any value, including nil,
// disables that behavior.
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
func Redirect(w ResponseWriter, r *Request, url string, code int) {
if u, err := urlpkg.Parse(url); err == nil {
// If url was relative, make its path absolute by
// combining with request path.
// The client would probably do this for us,
// but doing it ourselves is more reliable.
// See RFC 7231, section 7.1.2
if u.Scheme == "" && u.Host == "" {
oldpath := r.URL.Path
if oldpath == "" { // should not happen, but avoid a crash if it does
oldpath = "/"
}
// no leading http://server
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
if url == "" || url[0] != '/' {
// make relative path absolute
olddir, _ := path.Split(oldpath)
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
url = olddir + url
}
var query string
if i := strings.Index(url, "?"); i != -1 {
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
url, query = url[:i], url[i:]
}
// clean up but preserve trailing slash
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
trailing := strings.HasSuffix(url, "/")
url = path.Clean(url)
if trailing && !strings.HasSuffix(url, "/") {
url += "/"
}
net/http: improve signature of Redirect, NewRequest In CL https://golang.org/cl/4893043 (6 years ago), a new package named "url" was created (it is currently known as "net/url"). During that change, some identifier name collisions were introduced, and two parameters in net/http were renamed to "urlStr". Since that time, Go has continued to put high emphasis on the quality and readability of the documentation. Sometimes, that means making small sacrifices in the implementation details of a package to ensure that the godoc reads better, since that's what the majority of users interact with. See https://golang.org/s/style#named-result-parameters: > Clarity of docs is always more important than saving a line or two > in your function. I think the "urlStr" parameter name is suboptimal for godoc purposes, and just "url" would be better. During the review of https://golang.org/cl/4893043, it was also noted by @rsc that having to rename parameters named "url" was suboptimal: > It's unfortunate that naming the package url means > you can't have a parameter or variable named url. However, at the time, the name of the url package was still being decided, and uri was an alternative name under consideration. The reason urlStr was chosen is because it was a lesser evil compared to naming the url package uri instead: > Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just > that the "i" in "uri" looks very similar to the "l" in "url" in many fonts. > Please let's go with urlStr instead of uri. Now that we have the Go 1 compatibility guarantee, the name of the net/url package is fixed. However, it's possible to improve the signature of Redirect, NewRequest functions in net/http package for godoc purposes by creating a package global alias to url.Parse, and renaming urlStr parameter to url in the exported funcs. This CL does so. Updates #21077. Change-Id: Ibcc10e3825863a663e6ad91b6eb47b1862a299a6 Reviewed-on: https://go-review.googlesource.com/49930 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-07-19 00:41:13 -04:00
url += query
}
}
h := w.Header()
// RFC 7231 notes that a short HTML body is usually included in
// the response because older user agents may not understand 301/307.
// Do it only if the request didn't already have a Content-Type header.
_, hadCT := h["Content-Type"]
h.Set("Location", hexEscapeNonASCII(url))
if !hadCT && (r.Method == "GET" || r.Method == "HEAD") {
h.Set("Content-Type", "text/html; charset=utf-8")
}
w.WriteHeader(code)
// Shouldn't send the body for POST or HEAD; that leaves GET.
if !hadCT && r.Method == "GET" {
body := "<a href=\"" + htmlEscape(url) + "\">" + StatusText(code) + "</a>.\n"
fmt.Fprintln(w, body)
}
}
var htmlReplacer = strings.NewReplacer(
"&", "&amp;",
"<", "&lt;",
">", "&gt;",
// "&#34;" is shorter than "&quot;".
`"`, "&#34;",
// "&#39;" is shorter than "&apos;" and apos was not in HTML until HTML5.
"'", "&#39;",
)
func htmlEscape(s string) string {
return htmlReplacer.Replace(s)
}
// Redirect to a fixed URL
type redirectHandler struct {
url string
code int
}
func (rh *redirectHandler) ServeHTTP(w ResponseWriter, r *Request) {
Redirect(w, r, rh.url, rh.code)
}
// RedirectHandler returns a request handler that redirects
// each request it receives to the given url using the given
// status code.
//
// The provided code should be in the 3xx range and is usually
// StatusMovedPermanently, StatusFound or StatusSeeOther.
func RedirectHandler(url string, code int) Handler {
return &redirectHandler{url, code}
}
// TODO(jba): rewrite the following doc for enhanced patterns (proposal
// https://go.dev/issue/61410).
// ServeMux is an HTTP request multiplexer.
// It matches the URL of each incoming request against a list of registered
// patterns and calls the handler for the pattern that
// most closely matches the URL.
//
// Patterns name fixed, rooted paths, like "/favicon.ico",
// or rooted subtrees, like "/images/" (note the trailing slash).
// Longer patterns take precedence over shorter ones, so that
// if there are handlers registered for both "/images/"
// and "/images/thumbnails/", the latter handler will be
// called for paths beginning with "/images/thumbnails/" and the
// former will receive requests for any other paths in the
// "/images/" subtree.
//
// Note that since a pattern ending in a slash names a rooted subtree,
// the pattern "/" matches all paths not matched by other registered
// patterns, not just the URL with Path == "/".
//
// If a subtree has been registered and a request is received naming the
// subtree root without its trailing slash, ServeMux redirects that
// request to the subtree root (adding the trailing slash). This behavior can
// be overridden with a separate registration for the path without
// the trailing slash. For example, registering "/images/" causes ServeMux
// to redirect a request for "/images" to "/images/", unless "/images" has
// been registered separately.
//
// Patterns may optionally begin with a host name, restricting matches to
// URLs on that host only. Host-specific patterns take precedence over
// general patterns, so that a handler might register for the two patterns
// "/codesearch" and "codesearch.google.com/" without also taking over
// requests for "http://www.google.com/".
//
// ServeMux also takes care of sanitizing the URL request path and the Host
// header, stripping the port number and redirecting any request containing . or
// .. elements or repeated slashes to an equivalent, cleaner URL.
type ServeMux struct {
mu sync.RWMutex
tree routingNode
patterns []*pattern
}
// NewServeMux allocates and returns a new ServeMux.
func NewServeMux() *ServeMux {
return &ServeMux{}
}
// DefaultServeMux is the default ServeMux used by Serve.
var DefaultServeMux = &defaultServeMux
var defaultServeMux ServeMux
// cleanPath returns the canonical path for p, eliminating . and .. elements.
func cleanPath(p string) string {
if p == "" {
return "/"
}
if p[0] != '/' {
p = "/" + p
}
np := path.Clean(p)
// path.Clean removes trailing slash except for root;
// put the trailing slash back if necessary.
if p[len(p)-1] == '/' && np != "/" {
// Fast path for common case of p being the string we want:
if len(p) == len(np)+1 && strings.HasPrefix(p, np) {
np = p
} else {
np += "/"
}
}
return np
}
// stripHostPort returns h without any trailing ":<port>".
func stripHostPort(h string) string {
// If no port on host, return unchanged
if !strings.Contains(h, ":") {
return h
}
host, _, err := net.SplitHostPort(h)
if err != nil {
return h // on error, return unchanged
}
return host
}
// Handler returns the handler to use for the given request,
// consulting r.Method, r.Host, and r.URL.Path. It always returns
// a non-nil handler. If the path is not in its canonical form, the
// handler will be an internally-generated handler that redirects
// to the canonical path. If the host contains a port, it is ignored
// when matching handlers.
//
// The path and host are used unchanged for CONNECT requests.
//
// Handler also returns the registered pattern that matches the
// request or, in the case of internally-generated redirects,
// the path that will match after following the redirect.
//
// If there is no registered handler that applies to the request,
// Handler returns a “page not found” handler and an empty pattern.
func (mux *ServeMux) Handler(r *Request) (h Handler, pattern string) {
return mux.findHandler(r)
}
// findHandler finds a handler for a request.
// If there is a matching handler, it returns it and the pattern that matched.
// Otherwise it returns a Redirect or NotFound handler with the path that would match
// after the redirect.
func (mux *ServeMux) findHandler(r *Request) (h Handler, patStr string) {
var n *routingNode
// TODO(jba): use escaped path. This is an independent change that is also part
// of proposal https://go.dev/issue/61410.
path := r.URL.Path
// CONNECT requests are not canonicalized.
if r.Method == "CONNECT" {
// If r.URL.Path is /tree and its handler is not registered,
// the /tree -> /tree/ redirect applies to CONNECT requests
// but the path canonicalization does not.
_, _, u := mux.matchOrRedirect(r.URL.Host, r.Method, path, r.URL)
if u != nil {
return RedirectHandler(u.String(), StatusMovedPermanently), u.Path
}
// Redo the match, this time with r.Host instead of r.URL.Host.
// Pass a nil URL to skip the trailing-slash redirect logic.
n, _, _ = mux.matchOrRedirect(r.Host, r.Method, path, nil)
} else {
// All other requests have any port stripped and path cleaned
// before passing to mux.handler.
host := stripHostPort(r.Host)
path = cleanPath(path)
// If the given path is /tree and its handler is not registered,
// redirect for /tree/.
var u *url.URL
n, _, u = mux.matchOrRedirect(host, r.Method, path, r.URL)
if u != nil {
return RedirectHandler(u.String(), StatusMovedPermanently), u.Path
}
if path != r.URL.Path {
// Redirect to cleaned path.
patStr := ""
if n != nil {
patStr = n.pattern.String()
}
u := &url.URL{Path: path, RawQuery: r.URL.RawQuery}
return RedirectHandler(u.String(), StatusMovedPermanently), patStr
}
}
if n == nil {
// TODO(jba): support 405 (MethodNotAllowed) by checking for patterns with different methods.
return NotFoundHandler(), ""
}
return n.handler, n.pattern.String()
}
// matchOrRedirect looks up a node in the tree that matches the host, method and path.
// The path is known to be in canonical form, except for CONNECT methods.
// If the url argument is non-nil, handler also deals with trailing-slash
// redirection: when a path doesn't match exactly, the match is tried again
// after appending "/" to the path. If that second match succeeds, the last
// return value is the URL to redirect to.
func (mux *ServeMux) matchOrRedirect(host, method, path string, u *url.URL) (_ *routingNode, matches []string, redirectTo *url.URL) {
mux.mu.RLock()
defer mux.mu.RUnlock()
n, matches := mux.tree.match(host, method, path)
// If we have an exact match, or we were asked not to try trailing-slash redirection,
// then we're done.
if !exactMatch(n, path) && u != nil {
// If there is an exact match with a trailing slash, then redirect.
path += "/"
n2, _ := mux.tree.match(host, method, path)
if exactMatch(n2, path) {
return nil, nil, &url.URL{Path: path, RawQuery: u.RawQuery}
}
}
return n, matches, nil
}
// exactMatch reports whether the node's pattern exactly matches the path.
// As a special case, if the node is nil, exactMatch return false.
//
// Before wildcards were introduced, it was clear that an exact match meant
// that the pattern and path were the same string. The only other possibility
// was that a trailing-slash pattern, like "/", matched a path longer than
// it, like "/a".
//
// With wildcards, we define an inexact match as any one where a multi wildcard
// matches a non-empty string. All other matches are exact.
// For example, these are all exact matches:
//
// pattern path
// /a /a
// /{x} /a
// /a/{$} /a/
// /a/ /a/
//
// The last case has a multi wildcard (implicitly), but the match is exact because
// the wildcard matches the empty string.
//
// Examples of matches that are not exact:
//
// pattern path
// / /a
// /a/{x...} /a/b
func exactMatch(n *routingNode, path string) bool {
if n == nil {
return false
}
// We can't directly implement the definition (empty match for multi
// wildcard) because we don't record a match for anonymous multis.
// If there is no multi, the match is exact.
if !n.pattern.lastSegment().multi {
return true
}
// If the path doesn't end in a trailing slash, then the multi match
// is non-empty.
if len(path) > 0 && path[len(path)-1] != '/' {
return false
}
// Only patterns ending in {$} or a multi wildcard can
// match a path with a trailing slash.
// For the match to be exact, the number of pattern
// segments should be the same as the number of slashes in the path.
// E.g. "/a/b/{$}" and "/a/b/{...}" exactly match "/a/b/", but "/a/" does not.
return len(n.pattern.segments) == strings.Count(path, "/")
}
// ServeHTTP dispatches the request to the handler whose
// pattern most closely matches the request URL.
func (mux *ServeMux) ServeHTTP(w ResponseWriter, r *Request) {
if r.RequestURI == "*" {
if r.ProtoAtLeast(1, 1) {
w.Header().Set("Connection", "close")
}
w.WriteHeader(StatusBadRequest)
return
}
h, _ := mux.findHandler(r)
// TODO(jba); save matches in Request.
h.ServeHTTP(w, r)
}
// The four functions below all call register so that callerLocation
// always refers to user code.
// Handle registers the handler for the given pattern.
// If a handler already exists for pattern, Handle panics.
func (mux *ServeMux) Handle(pattern string, handler Handler) {
mux.register(pattern, handler)
}
// HandleFunc registers the handler function for the given pattern.
func (mux *ServeMux) HandleFunc(pattern string, handler func(ResponseWriter, *Request)) {
mux.register(pattern, HandlerFunc(handler))
}
// Handle registers the handler for the given pattern in [DefaultServeMux].
// The documentation for [ServeMux] explains how patterns are matched.
func Handle(pattern string, handler Handler) {
DefaultServeMux.register(pattern, handler)
}
// HandleFunc registers the handler function for the given pattern in [DefaultServeMux].
// The documentation for [ServeMux] explains how patterns are matched.
func HandleFunc(pattern string, handler func(ResponseWriter, *Request)) {
DefaultServeMux.register(pattern, HandlerFunc(handler))
}
func (mux *ServeMux) register(pattern string, handler Handler) {
if err := mux.registerErr(pattern, handler); err != nil {
panic(err)
}
}
func (mux *ServeMux) registerErr(pattern string, handler Handler) error {
if pattern == "" {
return errors.New("http: invalid pattern")
}
if handler == nil {
return errors.New("http: nil handler")
}
if f, ok := handler.(HandlerFunc); ok && f == nil {
return errors.New("http: nil handler")
}
pat, err := parsePattern(pattern)
if err != nil {
return fmt.Errorf("parsing %q: %w", pattern, err)
}
// Get the caller's location, for better conflict error messages.
// Skip register and whatever calls it.
_, file, line, ok := runtime.Caller(3)
if !ok {
pat.loc = "unknown location"
} else {
pat.loc = fmt.Sprintf("%s:%d", file, line)
}
mux.mu.Lock()
defer mux.mu.Unlock()
// Check for conflict.
// This makes a quadratic number of calls to conflictsWith: we check
// each pattern against every other pattern.
// TODO(jba): add indexing to speed this up.
for _, pat2 := range mux.patterns {
if pat.conflictsWith(pat2) {
return fmt.Errorf("pattern %q (registered at %s) conflicts with pattern %q (registered at %s)",
pat, pat.loc, pat2, pat2.loc)
}
}
mux.tree.addPattern(pat, handler)
mux.patterns = append(mux.patterns, pat)
return nil
}
// Serve accepts incoming HTTP connections on the listener l,
// creating a new service goroutine for each. The service goroutines
// read requests and then call handler to reply to them.
//
// The handler is typically nil, in which case [DefaultServeMux] is used.
//
// HTTP/2 support is only enabled if the Listener returns *tls.Conn
// connections and they were configured with "h2" in the TLS
// Config.NextProtos.
//
// Serve always returns a non-nil error.
func Serve(l net.Listener, handler Handler) error {
srv := &Server{Handler: handler}
return srv.Serve(l)
}
// ServeTLS accepts incoming HTTPS connections on the listener l,
// creating a new service goroutine for each. The service goroutines
// read requests and then call handler to reply to them.
//
// The handler is typically nil, in which case [DefaultServeMux] is used.
//
// Additionally, files containing a certificate and matching private key
// for the server must be provided. If the certificate is signed by a
// certificate authority, the certFile should be the concatenation
// of the server's certificate, any intermediates, and the CA's certificate.
//
// ServeTLS always returns a non-nil error.
func ServeTLS(l net.Listener, handler Handler, certFile, keyFile string) error {
srv := &Server{Handler: handler}
return srv.ServeTLS(l, certFile, keyFile)
}
// A Server defines parameters for running an HTTP server.
// The zero value for Server is a valid configuration.
type Server struct {
// Addr optionally specifies the TCP address for the server to listen on,
// in the form "host:port". If empty, ":http" (port 80) is used.
// The service names are defined in RFC 6335 and assigned by IANA.
// See net.Dial for details of the address format.
Addr string
Handler Handler // handler to invoke, http.DefaultServeMux if nil
// DisableGeneralOptionsHandler, if true, passes "OPTIONS *" requests to the Handler,
// otherwise responds with 200 OK and Content-Length: 0.
DisableGeneralOptionsHandler bool
// TLSConfig optionally provides a TLS configuration for use
// by ServeTLS and ListenAndServeTLS. Note that this value is
// cloned by ServeTLS and ListenAndServeTLS, so it's not
// possible to modify the configuration with methods like
// tls.Config.SetSessionTicketKeys. To use
// SetSessionTicketKeys, use Server.Serve with a TLS Listener
// instead.
TLSConfig *tls.Config
// ReadTimeout is the maximum duration for reading the entire
// request, including the body. A zero or negative value means
// there will be no timeout.
//
// Because ReadTimeout does not let Handlers make per-request
// decisions on each request body's acceptable deadline or
// upload rate, most users will prefer to use
// ReadHeaderTimeout. It is valid to use them both.
ReadTimeout time.Duration
// ReadHeaderTimeout is the amount of time allowed to read
// request headers. The connection's read deadline is reset
// after reading the headers and the Handler can decide what
// is considered too slow for the body. If ReadHeaderTimeout
// is zero, the value of ReadTimeout is used. If both are
// zero, there is no timeout.
ReadHeaderTimeout time.Duration
// WriteTimeout is the maximum duration before timing out
// writes of the response. It is reset whenever a new
// request's header is read. Like ReadTimeout, it does not
// let Handlers make decisions on a per-request basis.
// A zero or negative value means there will be no timeout.
WriteTimeout time.Duration
// IdleTimeout is the maximum amount of time to wait for the
// next request when keep-alives are enabled. If IdleTimeout
// is zero, the value of ReadTimeout is used. If both are
// zero, there is no timeout.
IdleTimeout time.Duration
// MaxHeaderBytes controls the maximum number of bytes the
// server will read parsing the request header's keys and
// values, including the request line. It does not limit the
// size of the request body.
// If zero, DefaultMaxHeaderBytes is used.
MaxHeaderBytes int
// TLSNextProto optionally specifies a function to take over
// ownership of the provided TLS connection when an ALPN
// protocol upgrade has occurred. The map key is the protocol
// name negotiated. The Handler argument should be used to
// handle HTTP requests and will initialize the Request's TLS
// and RemoteAddr if not already set. The connection is
// automatically closed when the function returns.
// If TLSNextProto is not nil, HTTP/2 support is not enabled
// automatically.
TLSNextProto map[string]func(*Server, *tls.Conn, Handler)
// ConnState specifies an optional callback function that is
// called when a client connection changes state. See the
// ConnState type and associated constants for details.
ConnState func(net.Conn, ConnState)
// ErrorLog specifies an optional logger for errors accepting
// connections, unexpected behavior from handlers, and
// underlying FileSystem errors.
// If nil, logging is done via the log package's standard logger.
ErrorLog *log.Logger
// BaseContext optionally specifies a function that returns
// the base context for incoming requests on this server.
// The provided Listener is the specific Listener that's
// about to start accepting requests.
// If BaseContext is nil, the default is context.Background().
// If non-nil, it must return a non-nil context.
BaseContext func(net.Listener) context.Context
// ConnContext optionally specifies a function that modifies
// the context used for a new connection c. The provided ctx
// is derived from the base context and has a ServerContextKey
// value.
ConnContext func(ctx context.Context, c net.Conn) context.Context
inShutdown atomic.Bool // true when server is in shutdown
disableKeepAlives atomic.Bool
nextProtoOnce sync.Once // guards setupHTTP2_* init
nextProtoErr error // result of http2.ConfigureServer if used
mu sync.Mutex
listeners map[*net.Listener]struct{}
activeConn map[*conn]struct{}
onShutdown []func()
listenerGroup sync.WaitGroup
}
// Close immediately closes all active net.Listeners and any
// connections in state StateNew, StateActive, or StateIdle. For a
// graceful shutdown, use Shutdown.
//
// Close does not attempt to close (and does not even know about)
// any hijacked connections, such as WebSockets.
//
// Close returns any error returned from closing the Server's
// underlying Listener(s).
func (srv *Server) Close() error {
srv.inShutdown.Store(true)
srv.mu.Lock()
defer srv.mu.Unlock()
err := srv.closeListenersLocked()
// Unlock srv.mu while waiting for listenerGroup.
// The group Add and Done calls are made with srv.mu held,
// to avoid adding a new listener in the window between
// us setting inShutdown above and waiting here.
srv.mu.Unlock()
srv.listenerGroup.Wait()
srv.mu.Lock()
for c := range srv.activeConn {
c.rwc.Close()
delete(srv.activeConn, c)
}
return err
}
// shutdownPollIntervalMax is the max polling interval when checking
// quiescence during Server.Shutdown. Polling starts with a small
// interval and backs off to the max.
// Ideally we could find a solution that doesn't involve polling,
// but which also doesn't have a high runtime cost (and doesn't
// involve any contentious mutexes), but that is left as an
// exercise for the reader.
const shutdownPollIntervalMax = 500 * time.Millisecond
// Shutdown gracefully shuts down the server without interrupting any
// active connections. Shutdown works by first closing all open
// listeners, then closing all idle connections, and then waiting
// indefinitely for connections to return to idle and then shut down.
// If the provided context expires before the shutdown is complete,
// Shutdown returns the context's error, otherwise it returns any
// error returned from closing the Server's underlying Listener(s).
//
// When Shutdown is called, Serve, ListenAndServe, and
// ListenAndServeTLS immediately return ErrServerClosed. Make sure the
// program doesn't exit and waits instead for Shutdown to return.
//
// Shutdown does not attempt to close nor wait for hijacked
// connections such as WebSockets. The caller of Shutdown should
// separately notify such long-lived connections of shutdown and wait
// for them to close, if desired. See RegisterOnShutdown for a way to
// register shutdown notification functions.
//
// Once Shutdown has been called on a server, it may not be reused;
// future calls to methods such as Serve will return ErrServerClosed.
func (srv *Server) Shutdown(ctx context.Context) error {
srv.inShutdown.Store(true)
srv.mu.Lock()
lnerr := srv.closeListenersLocked()
for _, f := range srv.onShutdown {
go f()
}
srv.mu.Unlock()
srv.listenerGroup.Wait()
pollIntervalBase := time.Millisecond
nextPollInterval := func() time.Duration {
// Add 10% jitter.
interval := pollIntervalBase + time.Duration(rand.Intn(int(pollIntervalBase/10)))
// Double and clamp for next time.
pollIntervalBase *= 2
if pollIntervalBase > shutdownPollIntervalMax {
pollIntervalBase = shutdownPollIntervalMax
}
return interval
}
timer := time.NewTimer(nextPollInterval())
defer timer.Stop()
for {
if srv.closeIdleConns() {
return lnerr
}
select {
case <-ctx.Done():
return ctx.Err()
case <-timer.C:
timer.Reset(nextPollInterval())
}
}
}
// RegisterOnShutdown registers a function to call on Shutdown.
// This can be used to gracefully shutdown connections that have
// undergone ALPN protocol upgrade or that have been hijacked.
// This function should start protocol-specific graceful shutdown,
// but should not wait for shutdown to complete.
func (srv *Server) RegisterOnShutdown(f func()) {
srv.mu.Lock()
srv.onShutdown = append(srv.onShutdown, f)
srv.mu.Unlock()
}
// closeIdleConns closes all idle connections and reports whether the
// server is quiescent.
func (s *Server) closeIdleConns() bool {
s.mu.Lock()
defer s.mu.Unlock()
quiescent := true
for c := range s.activeConn {
st, unixSec := c.getState()
// Issue 22682: treat StateNew connections as if
// they're idle if we haven't read the first request's
// header in over 5 seconds.
if st == StateNew && unixSec < time.Now().Unix()-5 {
st = StateIdle
}
if st != StateIdle || unixSec == 0 {
// Assume unixSec == 0 means it's a very new
// connection, without state set yet.
quiescent = false
continue
}
c.rwc.Close()
delete(s.activeConn, c)
}
return quiescent
}
func (s *Server) closeListenersLocked() error {
var err error
for ln := range s.listeners {
if cerr := (*ln).Close(); cerr != nil && err == nil {
err = cerr
}
}
return err
}
// A ConnState represents the state of a client connection to a server.
// It's used by the optional Server.ConnState hook.
type ConnState int
const (
// StateNew represents a new connection that is expected to
// send a request immediately. Connections begin at this
// state and then transition to either StateActive or
// StateClosed.
StateNew ConnState = iota
// StateActive represents a connection that has read 1 or more
// bytes of a request. The Server.ConnState hook for
// StateActive fires before the request has entered a handler
// and doesn't fire again until the request has been
// handled. After the request is handled, the state
// transitions to StateClosed, StateHijacked, or StateIdle.
// For HTTP/2, StateActive fires on the transition from zero
// to one active request, and only transitions away once all
// active requests are complete. That means that ConnState
// cannot be used to do per-request work; ConnState only notes
// the overall state of the connection.
StateActive
// StateIdle represents a connection that has finished
// handling a request and is in the keep-alive state, waiting
// for a new request. Connections transition from StateIdle
// to either StateActive or StateClosed.
StateIdle
// StateHijacked represents a hijacked connection.
// This is a terminal state. It does not transition to StateClosed.
StateHijacked
// StateClosed represents a closed connection.
// This is a terminal state. Hijacked connections do not
// transition to StateClosed.
StateClosed
)
var stateName = map[ConnState]string{
StateNew: "new",
StateActive: "active",
StateIdle: "idle",
StateHijacked: "hijacked",
StateClosed: "closed",
}
func (c ConnState) String() string {
return stateName[c]
}
// serverHandler delegates to either the server's Handler or
// DefaultServeMux and also handles "OPTIONS *" requests.
type serverHandler struct {
srv *Server
}
func (sh serverHandler) ServeHTTP(rw ResponseWriter, req *Request) {
handler := sh.srv.Handler
if handler == nil {
handler = DefaultServeMux
}
if !sh.srv.DisableGeneralOptionsHandler && req.RequestURI == "*" && req.Method == "OPTIONS" {
handler = globalOptionsHandler{}
}
handler.ServeHTTP(rw, req)
}
// AllowQuerySemicolons returns a handler that serves requests by converting any
// unescaped semicolons in the URL query to ampersands, and invoking the handler h.
//
// This restores the pre-Go 1.17 behavior of splitting query parameters on both
// semicolons and ampersands. (See golang.org/issue/25192). Note that this
// behavior doesn't match that of many proxies, and the mismatch can lead to
// security issues.
//
// AllowQuerySemicolons should be invoked before Request.ParseForm is called.
func AllowQuerySemicolons(h Handler) Handler {
return HandlerFunc(func(w ResponseWriter, r *Request) {
if strings.Contains(r.URL.RawQuery, ";") {
r2 := new(Request)
*r2 = *r
r2.URL = new(url.URL)
*r2.URL = *r.URL
r2.URL.RawQuery = strings.ReplaceAll(r.URL.RawQuery, ";", "&")
h.ServeHTTP(w, r2)
} else {
h.ServeHTTP(w, r)
}
})
}
// ListenAndServe listens on the TCP network address srv.Addr and then
// calls Serve to handle requests on incoming connections.
// Accepted connections are configured to enable TCP keep-alives.
//
// If srv.Addr is blank, ":http" is used.
//
// ListenAndServe always returns a non-nil error. After Shutdown or Close,
// the returned error is ErrServerClosed.
func (srv *Server) ListenAndServe() error {
if srv.shuttingDown() {
return ErrServerClosed
}
addr := srv.Addr
if addr == "" {
addr = ":http"
}
ln, err := net.Listen("tcp", addr)
if err != nil {
return err
}
return srv.Serve(ln)
}
var testHookServerServe func(*Server, net.Listener) // used if non-nil
// shouldConfigureHTTP2ForServe reports whether Server.Serve should configure
// automatic HTTP/2. (which sets up the srv.TLSNextProto map)
func (srv *Server) shouldConfigureHTTP2ForServe() bool {
if srv.TLSConfig == nil {
// Compatibility with Go 1.6:
// If there's no TLSConfig, it's possible that the user just
// didn't set it on the http.Server, but did pass it to
// tls.NewListener and passed that listener to Serve.
// So we should configure HTTP/2 (to set up srv.TLSNextProto)
// in case the listener returns an "h2" *tls.Conn.
return true
}
// The user specified a TLSConfig on their http.Server.
// In this, case, only configure HTTP/2 if their tls.Config
// explicitly mentions "h2". Otherwise http2.ConfigureServer
// would modify the tls.Config to add it, but they probably already
// passed this tls.Config to tls.NewListener. And if they did,
// it's too late anyway to fix it. It would only be potentially racy.
// See Issue 15908.
return strSliceContains(srv.TLSConfig.NextProtos, http2NextProtoTLS)
}
// ErrServerClosed is returned by the Server's Serve, ServeTLS, ListenAndServe,
// and ListenAndServeTLS methods after a call to Shutdown or Close.
var ErrServerClosed = errors.New("http: Server closed")
// Serve accepts incoming connections on the Listener l, creating a
// new service goroutine for each. The service goroutines read requests and
// then call srv.Handler to reply to them.
//
// HTTP/2 support is only enabled if the Listener returns *tls.Conn
// connections and they were configured with "h2" in the TLS
// Config.NextProtos.
//
// Serve always returns a non-nil error and closes l.
// After Shutdown or Close, the returned error is ErrServerClosed.
func (srv *Server) Serve(l net.Listener) error {
if fn := testHookServerServe; fn != nil {
fn(srv, l) // call hook with unwrapped listener
}
origListener := l
l = &onceCloseListener{Listener: l}
defer l.Close()
if err := srv.setupHTTP2_Serve(); err != nil {
return err
}
if !srv.trackListener(&l, true) {
return ErrServerClosed
}
defer srv.trackListener(&l, false)
baseCtx := context.Background()
if srv.BaseContext != nil {
baseCtx = srv.BaseContext(origListener)
if baseCtx == nil {
panic("BaseContext returned a nil context")
}
}
var tempDelay time.Duration // how long to sleep on accept failure
ctx := context.WithValue(baseCtx, ServerContextKey, srv)
for {
rw, err := l.Accept()
if err != nil {
if srv.shuttingDown() {
return ErrServerClosed
}
if ne, ok := err.(net.Error); ok && ne.Temporary() {
if tempDelay == 0 {
tempDelay = 5 * time.Millisecond
} else {
tempDelay *= 2
}
if max := 1 * time.Second; tempDelay > max {
tempDelay = max
}
srv.logf("http: Accept error: %v; retrying in %v", err, tempDelay)
time.Sleep(tempDelay)
continue
}
return err
}
connCtx := ctx
if cc := srv.ConnContext; cc != nil {
connCtx = cc(connCtx, rw)
if connCtx == nil {
panic("ConnContext returned nil")
}
}
tempDelay = 0
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
c := srv.newConn(rw)
c.setState(c.rwc, StateNew, runHooks) // before Serve can return
go c.serve(connCtx)
}
}
// ServeTLS accepts incoming connections on the Listener l, creating a
// new service goroutine for each. The service goroutines perform TLS
// setup and then read requests, calling srv.Handler to reply to them.
//
// Files containing a certificate and matching private key for the
// server must be provided if neither the Server's
// TLSConfig.Certificates nor TLSConfig.GetCertificate are populated.
// If the certificate is signed by a certificate authority, the
// certFile should be the concatenation of the server's certificate,
// any intermediates, and the CA's certificate.
//
// ServeTLS always returns a non-nil error. After Shutdown or Close, the
// returned error is ErrServerClosed.
func (srv *Server) ServeTLS(l net.Listener, certFile, keyFile string) error {
// Setup HTTP/2 before srv.Serve, to initialize srv.TLSConfig
// before we clone it and create the TLS Listener.
if err := srv.setupHTTP2_ServeTLS(); err != nil {
return err
}
config := cloneTLSConfig(srv.TLSConfig)
if !strSliceContains(config.NextProtos, "http/1.1") {
config.NextProtos = append(config.NextProtos, "http/1.1")
}
configHasCert := len(config.Certificates) > 0 || config.GetCertificate != nil
if !configHasCert || certFile != "" || keyFile != "" {
var err error
config.Certificates = make([]tls.Certificate, 1)
config.Certificates[0], err = tls.LoadX509KeyPair(certFile, keyFile)
if err != nil {
return err
}
}
tlsListener := tls.NewListener(l, config)
return srv.Serve(tlsListener)
}
// trackListener adds or removes a net.Listener to the set of tracked
// listeners.
//
// We store a pointer to interface in the map set, in case the
// net.Listener is not comparable. This is safe because we only call
// trackListener via Serve and can track+defer untrack the same
// pointer to local variable there. We never need to compare a
// Listener from another caller.
//
// It reports whether the server is still up (not Shutdown or Closed).
func (s *Server) trackListener(ln *net.Listener, add bool) bool {
s.mu.Lock()
defer s.mu.Unlock()
if s.listeners == nil {
s.listeners = make(map[*net.Listener]struct{})
}
if add {
if s.shuttingDown() {
return false
}
s.listeners[ln] = struct{}{}
s.listenerGroup.Add(1)
} else {
delete(s.listeners, ln)
s.listenerGroup.Done()
}
return true
}
func (s *Server) trackConn(c *conn, add bool) {
s.mu.Lock()
defer s.mu.Unlock()
if s.activeConn == nil {
s.activeConn = make(map[*conn]struct{})
}
if add {
s.activeConn[c] = struct{}{}
} else {
delete(s.activeConn, c)
}
}
func (s *Server) idleTimeout() time.Duration {
if s.IdleTimeout != 0 {
return s.IdleTimeout
}
return s.ReadTimeout
}
func (s *Server) readHeaderTimeout() time.Duration {
if s.ReadHeaderTimeout != 0 {
return s.ReadHeaderTimeout
}
return s.ReadTimeout
}
func (s *Server) doKeepAlives() bool {
return !s.disableKeepAlives.Load() && !s.shuttingDown()
}
func (s *Server) shuttingDown() bool {
return s.inShutdown.Load()
}
// SetKeepAlivesEnabled controls whether HTTP keep-alives are enabled.
// By default, keep-alives are always enabled. Only very
// resource-constrained environments or servers in the process of
// shutting down should disable them.
func (srv *Server) SetKeepAlivesEnabled(v bool) {
if v {
srv.disableKeepAlives.Store(false)
return
}
srv.disableKeepAlives.Store(true)
// Close idle HTTP/1 conns:
srv.closeIdleConns()
// TODO: Issue 26303: close HTTP/2 conns as soon as they become idle.
}
func (s *Server) logf(format string, args ...any) {
if s.ErrorLog != nil {
s.ErrorLog.Printf(format, args...)
} else {
log.Printf(format, args...)
}
}
// logf prints to the ErrorLog of the *Server associated with request r
// via ServerContextKey. If there's no associated server, or if ErrorLog
// is nil, logging is done via the log package's standard logger.
func logf(r *Request, format string, args ...any) {
s, _ := r.Context().Value(ServerContextKey).(*Server)
if s != nil && s.ErrorLog != nil {
s.ErrorLog.Printf(format, args...)
} else {
log.Printf(format, args...)
}
}
// ListenAndServe listens on the TCP network address addr and then calls
// Serve with handler to handle requests on incoming connections.
// Accepted connections are configured to enable TCP keep-alives.
//
// The handler is typically nil, in which case [DefaultServeMux] is used.
//
// ListenAndServe always returns a non-nil error.
func ListenAndServe(addr string, handler Handler) error {
server := &Server{Addr: addr, Handler: handler}
return server.ListenAndServe()
}
// ListenAndServeTLS acts identically to [ListenAndServe], except that it
// expects HTTPS connections. Additionally, files containing a certificate and
// matching private key for the server must be provided. If the certificate
// is signed by a certificate authority, the certFile should be the concatenation
// of the server's certificate, any intermediates, and the CA's certificate.
func ListenAndServeTLS(addr, certFile, keyFile string, handler Handler) error {
server := &Server{Addr: addr, Handler: handler}
return server.ListenAndServeTLS(certFile, keyFile)
}
// ListenAndServeTLS listens on the TCP network address srv.Addr and
// then calls ServeTLS to handle requests on incoming TLS connections.
// Accepted connections are configured to enable TCP keep-alives.
//
// Filenames containing a certificate and matching private key for the
// server must be provided if neither the Server's TLSConfig.Certificates
// nor TLSConfig.GetCertificate are populated. If the certificate is
// signed by a certificate authority, the certFile should be the
// concatenation of the server's certificate, any intermediates, and
// the CA's certificate.
//
// If srv.Addr is blank, ":https" is used.
//
// ListenAndServeTLS always returns a non-nil error. After Shutdown or
// Close, the returned error is ErrServerClosed.
func (srv *Server) ListenAndServeTLS(certFile, keyFile string) error {
if srv.shuttingDown() {
return ErrServerClosed
}
addr := srv.Addr
if addr == "" {
addr = ":https"
}
ln, err := net.Listen("tcp", addr)
if err != nil {
return err
}
defer ln.Close()
return srv.ServeTLS(ln, certFile, keyFile)
}
// setupHTTP2_ServeTLS conditionally configures HTTP/2 on
// srv and reports whether there was an error setting it up. If it is
// not configured for policy reasons, nil is returned.
func (srv *Server) setupHTTP2_ServeTLS() error {
srv.nextProtoOnce.Do(srv.onceSetNextProtoDefaults)
return srv.nextProtoErr
}
// setupHTTP2_Serve is called from (*Server).Serve and conditionally
// configures HTTP/2 on srv using a more conservative policy than
// setupHTTP2_ServeTLS because Serve is called after tls.Listen,
// and may be called concurrently. See shouldConfigureHTTP2ForServe.
//
// The tests named TestTransportAutomaticHTTP2* and
// TestConcurrentServerServe in server_test.go demonstrate some
// of the supported use cases and motivations.
func (srv *Server) setupHTTP2_Serve() error {
srv.nextProtoOnce.Do(srv.onceSetNextProtoDefaults_Serve)
return srv.nextProtoErr
}
func (srv *Server) onceSetNextProtoDefaults_Serve() {
if srv.shouldConfigureHTTP2ForServe() {
srv.onceSetNextProtoDefaults()
}
}
internal/godebug: define more efficient API We have been expanding our use of GODEBUG for compatibility, and the current implementation forces a tradeoff between freshness and efficiency. It parses the environment variable in full each time it is called, which is expensive. But if clients cache the result, they won't respond to run-time GODEBUG changes, as happened with x509sha1 (#56436). This CL changes the GODEBUG API to provide efficient, up-to-date results. Instead of a single Get function, New returns a *godebug.Setting that itself has a Get method. Clients can save the result of New, which is no more expensive than errors.New, in a global variable, and then call that variable's Get method to get the value. Get costs only two atomic loads in the case where the variable hasn't changed since the last call. Unfortunately, these changes do require importing sync from godebug, which will mean that sync itself will never be able to use a GODEBUG setting. That doesn't seem like such a hardship. If it was really necessary, the runtime could pass a setting to package sync itself at startup, with the caveat that that setting, like the ones used by runtime itself, would not respond to run-time GODEBUG changes. Change-Id: I99a3acfa24fb2a692610af26a5d14bbc62c966ac Reviewed-on: https://go-review.googlesource.com/c/go/+/449504 Run-TryBot: Russ Cox <rsc@golang.org> Auto-Submit: Russ Cox <rsc@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Ian Lance Taylor <iant@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-11-11 12:36:31 -05:00
var http2server = godebug.New("http2server")
// onceSetNextProtoDefaults configures HTTP/2, if the user hasn't
// configured otherwise. (by setting srv.TLSNextProto non-nil)
// It must only be called via srv.nextProtoOnce (use srv.setupHTTP2_*).
func (srv *Server) onceSetNextProtoDefaults() {
if omitBundledHTTP2 {
return
}
if http2server.Value() == "0" {
http2server.IncNonDefault()
return
}
// Enable HTTP/2 by default if the user hasn't otherwise
// configured their TLSNextProto map.
if srv.TLSNextProto == nil {
conf := &http2Server{
NewWriteScheduler: func() http2WriteScheduler { return http2NewPriorityWriteScheduler(nil) },
}
srv.nextProtoErr = http2ConfigureServer(srv, conf)
}
}
// TimeoutHandler returns a Handler that runs h with the given time limit.
//
// The new Handler calls h.ServeHTTP to handle each request, but if a
// call runs for longer than its time limit, the handler responds with
// a 503 Service Unavailable error and the given message in its body.
// (If msg is empty, a suitable default message will be sent.)
// After such a timeout, writes by h to its ResponseWriter will return
// ErrHandlerTimeout.
//
// TimeoutHandler supports the Pusher interface but does not support
// the Hijacker or Flusher interfaces.
func TimeoutHandler(h Handler, dt time.Duration, msg string) Handler {
return &timeoutHandler{
handler: h,
body: msg,
dt: dt,
}
}
// ErrHandlerTimeout is returned on ResponseWriter Write calls
// in handlers which have timed out.
var ErrHandlerTimeout = errors.New("http: Handler timeout")
type timeoutHandler struct {
handler Handler
body string
dt time.Duration
// When set, no context will be created and this context will
// be used instead.
testContext context.Context
}
func (h *timeoutHandler) errorBody() string {
if h.body != "" {
return h.body
}
return "<html><head><title>Timeout</title></head><body><h1>Timeout</h1></body></html>"
}
func (h *timeoutHandler) ServeHTTP(w ResponseWriter, r *Request) {
ctx := h.testContext
if ctx == nil {
var cancelCtx context.CancelFunc
ctx, cancelCtx = context.WithTimeout(r.Context(), h.dt)
defer cancelCtx()
}
r = r.WithContext(ctx)
done := make(chan struct{})
tw := &timeoutWriter{
w: w,
h: make(Header),
req: r,
}
panicChan := make(chan any, 1)
go func() {
defer func() {
if p := recover(); p != nil {
panicChan <- p
}
}()
h.handler.ServeHTTP(tw, r)
close(done)
}()
select {
case p := <-panicChan:
panic(p)
case <-done:
tw.mu.Lock()
defer tw.mu.Unlock()
dst := w.Header()
for k, vv := range tw.h {
dst[k] = vv
}
if !tw.wroteHeader {
tw.code = StatusOK
}
w.WriteHeader(tw.code)
w.Write(tw.wbuf.Bytes())
case <-ctx.Done():
tw.mu.Lock()
defer tw.mu.Unlock()
switch err := ctx.Err(); err {
case context.DeadlineExceeded:
w.WriteHeader(StatusServiceUnavailable)
io.WriteString(w, h.errorBody())
tw.err = ErrHandlerTimeout
default:
w.WriteHeader(StatusServiceUnavailable)
tw.err = err
}
}
}
type timeoutWriter struct {
w ResponseWriter
h Header
wbuf bytes.Buffer
req *Request
mu sync.Mutex
err error
wroteHeader bool
code int
}
var _ Pusher = (*timeoutWriter)(nil)
// Push implements the Pusher interface.
func (tw *timeoutWriter) Push(target string, opts *PushOptions) error {
if pusher, ok := tw.w.(Pusher); ok {
return pusher.Push(target, opts)
}
return ErrNotSupported
}
func (tw *timeoutWriter) Header() Header { return tw.h }
func (tw *timeoutWriter) Write(p []byte) (int, error) {
tw.mu.Lock()
defer tw.mu.Unlock()
if tw.err != nil {
return 0, tw.err
}
if !tw.wroteHeader {
tw.writeHeaderLocked(StatusOK)
}
return tw.wbuf.Write(p)
}
func (tw *timeoutWriter) writeHeaderLocked(code int) {
checkWriteHeaderCode(code)
switch {
case tw.err != nil:
return
case tw.wroteHeader:
if tw.req != nil {
caller := relevantCaller()
logf(tw.req, "http: superfluous response.WriteHeader call from %s (%s:%d)", caller.Function, path.Base(caller.File), caller.Line)
}
default:
tw.wroteHeader = true
tw.code = code
}
}
func (tw *timeoutWriter) WriteHeader(code int) {
tw.mu.Lock()
defer tw.mu.Unlock()
tw.writeHeaderLocked(code)
}
// onceCloseListener wraps a net.Listener, protecting it from
// multiple Close calls.
type onceCloseListener struct {
net.Listener
once sync.Once
closeErr error
}
func (oc *onceCloseListener) Close() error {
oc.once.Do(oc.close)
return oc.closeErr
}
func (oc *onceCloseListener) close() { oc.closeErr = oc.Listener.Close() }
// globalOptionsHandler responds to "OPTIONS *" requests.
type globalOptionsHandler struct{}
func (globalOptionsHandler) ServeHTTP(w ResponseWriter, r *Request) {
w.Header().Set("Content-Length", "0")
if r.ContentLength != 0 {
// Read up to 4KB of OPTIONS body (as mentioned in the
// spec as being reserved for future use), but anything
// over that is considered a waste of server resources
// (or an attack) and we abort and close the connection,
// courtesy of MaxBytesReader's EOF behavior.
mb := MaxBytesReader(w, r.Body, 4<<10)
io.Copy(io.Discard, mb)
}
}
// initALPNRequest is an HTTP handler that initializes certain
// uninitialized fields in its *Request. Such partially-initialized
// Requests come from ALPN protocol handlers.
type initALPNRequest struct {
ctx context.Context
c *tls.Conn
h serverHandler
}
// BaseContext is an exported but unadvertised http.Handler method
// recognized by x/net/http2 to pass down a context; the TLSNextProto
// API predates context support so we shoehorn through the only
// interface we have available.
func (h initALPNRequest) BaseContext() context.Context { return h.ctx }
func (h initALPNRequest) ServeHTTP(rw ResponseWriter, req *Request) {
if req.TLS == nil {
req.TLS = &tls.ConnectionState{}
*req.TLS = h.c.ConnectionState()
}
if req.Body == nil {
req.Body = NoBody
}
if req.RemoteAddr == "" {
req.RemoteAddr = h.c.RemoteAddr().String()
}
h.h.ServeHTTP(rw, req)
}
// loggingConn is used for debugging.
type loggingConn struct {
name string
net.Conn
}
var (
uniqNameMu sync.Mutex
uniqNameNext = make(map[string]int)
)
func newLoggingConn(baseName string, c net.Conn) net.Conn {
uniqNameMu.Lock()
defer uniqNameMu.Unlock()
uniqNameNext[baseName]++
return &loggingConn{
name: fmt.Sprintf("%s-%d", baseName, uniqNameNext[baseName]),
Conn: c,
}
}
func (c *loggingConn) Write(p []byte) (n int, err error) {
log.Printf("%s.Write(%d) = ....", c.name, len(p))
n, err = c.Conn.Write(p)
log.Printf("%s.Write(%d) = %d, %v", c.name, len(p), n, err)
return
}
func (c *loggingConn) Read(p []byte) (n int, err error) {
log.Printf("%s.Read(%d) = ....", c.name, len(p))
n, err = c.Conn.Read(p)
log.Printf("%s.Read(%d) = %d, %v", c.name, len(p), n, err)
return
}
func (c *loggingConn) Close() (err error) {
log.Printf("%s.Close() = ...", c.name)
err = c.Conn.Close()
log.Printf("%s.Close() = %v", c.name, err)
return
}
// checkConnErrorWriter writes to c.rwc and records any write errors to c.werr.
// It only contains one field (and a pointer field at that), so it
// fits in an interface value without an extra allocation.
type checkConnErrorWriter struct {
c *conn
}
func (w checkConnErrorWriter) Write(p []byte) (n int, err error) {
net/http: rework CloseNotifier implementation, clarify expectations in docs CloseNotifier wasn't well specified previously. This CL simplifies its implementation, clarifies the public documentation on CloseNotifier, clarifies internal documentation on conn, and fixes two CloseNotifier bugs in the process. The main change, though, is tightening the rules and expectations for using CloseNotifier: * the caller must consume the Request.Body first (old rule, unwritten) * the received value is the "true" value (old rule, unwritten) * no promises for channel sends after Handler returns (old rule, unwritten) * a subsequent pipelined request fires the CloseNotifier (new behavior; previously it never fired and thus effectively deadlocked as in #13165) * advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2 or non-idempotent browsers). Not that browsers actually use pipelining. The main implementation change is that each Handler now gets its own CloseNotifier channel value, rather than sharing one between the whole conn. This means Handlers can't affect subsequent requests. This is how HTTP/2's Server works too. The old docs never clarified a behavior either way. The other side effect of each request getting its own CloseNotifier channel is that one handler can't "poison" the underlying conn preventing subsequent requests on the same connection from using CloseNotifier (this is #9763). In the old implementation, once any request on a connection used ClosedNotifier, the conn's underlying bufio.Reader source was switched from the TCPConn to the read side of the pipe being fed by a never-ending copy. Since it was impossible to abort that never-ending copy, we could never get back to a fresh state where it was possible to return the underlying TCPConn to callers of Hijack. Now, instead of a never-ending Copy, the background goroutine doing a Read from the TCPConn (or *tls.Conn) only reads a single byte. That single byte can be in the request body, a socket timeout error, io.EOF error, or the first byte of the second body. In any case, the new *connReader type stitches sync and async reads together like an io.MultiReader. To clarify the flow of Read data and combat the complexity of too many wrapper Reader types, the *connReader absorbs the io.LimitReader previously used for bounding request header reads. The liveSwitchReader type is removed. (an unused switchWriter type is also removed) Many fields on *conn are also documented more fully. Fixes #9763 (CloseNotify + Hijack together) Fixes #13165 (deadlock with CloseNotify + pipelined requests) Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65 Reviewed-on: https://go-review.googlesource.com/17750 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-01 19:07:41 +00:00
n, err = w.c.rwc.Write(p)
if err != nil && w.c.werr == nil {
w.c.werr = err
net/http: make Server Handler's Request.Context be done on conn errors This CL changes how the http1 Server reads from the client. The goal of this change is to make the Request.Context given to Server Handlers become done when the TCP connection dies (has seen any read or write error). I didn't finish that for Go 1.7 when Context was added to http package. We can't notice the peer disconnect unless we're blocked in a Read call, though, and previously we were only doing read calls as needed, when reading the body or the next request. One exception to that was the old pre-context CloseNotifier mechanism. The implementation of CloseNotifier has always been tricky. The past few releases have contained the complexity and moved the reading-from-TCP-conn logic into the "connReader" type. This CL extends connReader to make sure that it's always blocked in a Read call, at least once the request body has been fully consumed. In the process, this deletes all the old CloseNotify code and unifies it with the context cancelation code. The two notification mechanisms are nearly identical, except the CloseNotify path always notifies on the arrival of pipelined HTTP/1 requests. We might want to change that in a subsequent commit. I left a TODO for that. For now there's no change in behavior except that the context now cancels as it was supposed to. As a bonus that fell out for free, a Handler can now use CloseNotifier and Hijack together in the same request now. Fixes #15224 (make http1 Server always in a Read, like http2) Fixes #15927 (cancel context when underlying connection closes) Updates #9763 (CloseNotifier + Hijack) Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b Reviewed-on: https://go-review.googlesource.com/31173 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-10-14 11:45:59 +01:00
w.c.cancelCtx()
}
return
}
func numLeadingCRorLF(v []byte) (n int) {
for _, b := range v {
if b == '\r' || b == '\n' {
n++
continue
}
break
}
return
}
func strSliceContains(ss []string, s string) bool {
for _, v := range ss {
if v == s {
return true
}
}
return false
}
// tlsRecordHeaderLooksLikeHTTP reports whether a TLS record header
// looks like it might've been a misdirected plaintext HTTP request.
func tlsRecordHeaderLooksLikeHTTP(hdr [5]byte) bool {
switch string(hdr[:]) {
case "GET /", "HEAD ", "POST ", "PUT /", "OPTIO":
return true
}
return false
}
// MaxBytesHandler returns a Handler that runs h with its ResponseWriter and Request.Body wrapped by a MaxBytesReader.
func MaxBytesHandler(h Handler, n int64) Handler {
return HandlerFunc(func(w ResponseWriter, r *Request) {
r2 := *r
r2.Body = MaxBytesReader(w, r.Body, n)
h.ServeHTTP(w, &r2)
})
}