1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #include "net/http/http_stream_parser.h"
8 #include "base/compiler_specific.h"
9 #include "base/logging.h"
10 #include "base/metrics/histogram_macros.h"
11 #include "base/profiler/scoped_tracker.h"
12 #include "base/strings/string_util.h"
13 #include "base/values.h"
14 #include "net/base/io_buffer.h"
15 #include "net/base/ip_endpoint.h"
16 #include "net/base/upload_data_stream.h"
17 #include "net/http/http_chunked_decoder.h"
18 #include "net/http/http_request_headers.h"
19 #include "net/http/http_request_info.h"
20 #include "net/http/http_response_headers.h"
21 #include "net/http/http_util.h"
22 #include "net/socket/client_socket_handle.h"
23 #include "net/socket/ssl_client_socket.h"
29 enum HttpHeaderParserEvent
{
30 HEADER_PARSER_INVOKED
= 0,
31 // Obsolete: HEADER_HTTP_09_RESPONSE = 1,
32 HEADER_ALLOWED_TRUNCATED_HEADERS
= 2,
33 HEADER_SKIPPED_WS_PREFIX
= 3,
34 HEADER_SKIPPED_NON_WS_PREFIX
= 4,
35 HEADER_HTTP_09_RESPONSE_OVER_HTTP
= 5,
36 HEADER_HTTP_09_RESPONSE_OVER_SSL
= 6,
37 HEADER_HTTP_09_ON_REUSED_SOCKET
= 7,
41 void RecordHeaderParserEvent(HttpHeaderParserEvent header_event
) {
42 UMA_HISTOGRAM_ENUMERATION("Net.HttpHeaderParserEvent", header_event
,
46 const uint64 kMaxMergedHeaderAndBodySize
= 1400;
47 const size_t kRequestBodyBufferSize
= 1 << 14; // 16KB
49 std::string
GetResponseHeaderLines(const HttpResponseHeaders
& headers
) {
50 std::string raw_headers
= headers
.raw_headers();
51 const char* null_separated_headers
= raw_headers
.c_str();
52 const char* header_line
= null_separated_headers
;
53 std::string cr_separated_headers
;
54 while (header_line
[0] != 0) {
55 cr_separated_headers
+= header_line
;
56 cr_separated_headers
+= "\n";
57 header_line
+= strlen(header_line
) + 1;
59 return cr_separated_headers
;
62 // Return true if |headers| contain multiple |field_name| fields with different
64 bool HeadersContainMultipleCopiesOfField(const HttpResponseHeaders
& headers
,
65 const std::string
& field_name
) {
67 std::string field_value
;
68 if (!headers
.EnumerateHeader(&it
, field_name
, &field_value
))
70 // There's at least one |field_name| header. Check if there are any more
71 // such headers, and if so, return true if they have different values.
72 std::string field_value2
;
73 while (headers
.EnumerateHeader(&it
, field_name
, &field_value2
)) {
74 if (field_value
!= field_value2
)
80 scoped_ptr
<base::Value
> NetLogSendRequestBodyCallback(
84 NetLogCaptureMode
/* capture_mode */) {
85 scoped_ptr
<base::DictionaryValue
> dict(new base::DictionaryValue());
86 dict
->SetInteger("length", static_cast<int>(length
));
87 dict
->SetBoolean("is_chunked", is_chunked
);
88 dict
->SetBoolean("did_merge", did_merge
);
92 // Returns true if |error_code| is an error for which we give the server a
93 // chance to send a body containing error information, if the error was received
94 // while trying to upload a request body.
95 bool ShouldTryReadingOnUploadError(int error_code
) {
96 return (error_code
== ERR_CONNECTION_RESET
);
101 // Similar to DrainableIOBuffer(), but this version comes with its own
102 // storage. The motivation is to avoid repeated allocations of
103 // DrainableIOBuffer.
107 // scoped_refptr<SeekableIOBuffer> buf = new SeekableIOBuffer(1024);
108 // // capacity() == 1024. size() == BytesRemaining() == BytesConsumed() == 0.
109 // // data() points to the beginning of the buffer.
111 // // Read() takes an IOBuffer.
112 // int bytes_read = some_reader->Read(buf, buf->capacity());
113 // buf->DidAppend(bytes_read);
114 // // size() == BytesRemaining() == bytes_read. data() is unaffected.
116 // while (buf->BytesRemaining() > 0) {
117 // // Write() takes an IOBuffer. If it takes const char*, we could
118 /// // simply use the regular IOBuffer like buf->data() + offset.
119 // int bytes_written = Write(buf, buf->BytesRemaining());
120 // buf->DidConsume(bytes_written);
122 // // BytesRemaining() == 0. BytesConsumed() == size().
123 // // data() points to the end of the consumed bytes (exclusive).
125 // // If you want to reuse the buffer, be sure to clear the buffer.
127 // // size() == BytesRemaining() == BytesConsumed() == 0.
128 // // data() points to the beginning of the buffer.
130 class HttpStreamParser::SeekableIOBuffer
: public IOBuffer
{
132 explicit SeekableIOBuffer(int capacity
)
133 : IOBuffer(capacity
),
140 // DidConsume() changes the |data_| pointer so that |data_| always points
141 // to the first unconsumed byte.
142 void DidConsume(int bytes
) {
143 SetOffset(used_
+ bytes
);
146 // Returns the number of unconsumed bytes.
147 int BytesRemaining() const {
148 return size_
- used_
;
151 // Seeks to an arbitrary point in the buffer. The notion of bytes consumed
152 // and remaining are updated appropriately.
153 void SetOffset(int bytes
) {
155 DCHECK_LE(bytes
, size_
);
157 data_
= real_data_
+ used_
;
160 // Called after data is added to the buffer. Adds |bytes| added to
161 // |size_|. data() is unaffected.
162 void DidAppend(int bytes
) {
164 DCHECK_GE(size_
+ bytes
, 0);
165 DCHECK_LE(size_
+ bytes
, capacity_
);
169 // Changes the logical size to 0, and the offset to 0.
175 // Returns the logical size of the buffer (i.e the number of bytes of data
177 int size() const { return size_
; }
179 // Returns the capacity of the buffer. The capacity is the size used when
180 // the object is created.
181 int capacity() const { return capacity_
; };
184 ~SeekableIOBuffer() override
{
185 // data_ will be deleted in IOBuffer::~IOBuffer().
195 // 2 CRLFs + max of 8 hex chars.
196 const size_t HttpStreamParser::kChunkHeaderFooterSize
= 12;
198 HttpStreamParser::HttpStreamParser(ClientSocketHandle
* connection
,
199 const HttpRequestInfo
* request
,
200 GrowableIOBuffer
* read_buffer
,
201 const BoundNetLog
& net_log
)
202 : io_state_(STATE_NONE
),
204 request_headers_(NULL
),
205 request_headers_length_(0),
206 read_buf_(read_buffer
),
207 read_buf_unused_offset_(0),
208 response_header_start_offset_(-1),
210 response_body_length_(-1),
211 response_body_read_(0),
212 user_read_buf_(NULL
),
213 user_read_buf_len_(0),
214 connection_(connection
),
216 sent_last_chunk_(false),
218 weak_ptr_factory_(this) {
219 io_callback_
= base::Bind(&HttpStreamParser::OnIOComplete
,
220 weak_ptr_factory_
.GetWeakPtr());
223 HttpStreamParser::~HttpStreamParser() {
226 int HttpStreamParser::SendRequest(const std::string
& request_line
,
227 const HttpRequestHeaders
& headers
,
228 HttpResponseInfo
* response
,
229 const CompletionCallback
& callback
) {
230 DCHECK_EQ(STATE_NONE
, io_state_
);
231 DCHECK(callback_
.is_null());
232 DCHECK(!callback
.is_null());
236 NetLog::TYPE_HTTP_TRANSACTION_SEND_REQUEST_HEADERS
,
237 base::Bind(&HttpRequestHeaders::NetLogCallback
,
238 base::Unretained(&headers
),
241 DVLOG(1) << __FUNCTION__
<< "()"
242 << " request_line = \"" << request_line
<< "\""
243 << " headers = \"" << headers
.ToString() << "\"";
244 response_
= response
;
246 // Put the peer's IP address and port into the response.
247 IPEndPoint ip_endpoint
;
248 int result
= connection_
->socket()->GetPeerAddress(&ip_endpoint
);
251 response_
->socket_address
= HostPortPair::FromIPEndPoint(ip_endpoint
);
253 std::string request
= request_line
+ headers
.ToString();
254 request_headers_length_
= request
.size();
256 if (request_
->upload_data_stream
!= NULL
) {
257 request_body_send_buf_
= new SeekableIOBuffer(kRequestBodyBufferSize
);
258 if (request_
->upload_data_stream
->is_chunked()) {
259 // Read buffer is adjusted to guarantee that |request_body_send_buf_| is
260 // large enough to hold the encoded chunk.
261 request_body_read_buf_
=
262 new SeekableIOBuffer(kRequestBodyBufferSize
- kChunkHeaderFooterSize
);
264 // No need to encode request body, just send the raw data.
265 request_body_read_buf_
= request_body_send_buf_
;
269 io_state_
= STATE_SEND_HEADERS
;
271 // If we have a small request body, then we'll merge with the headers into a
273 bool did_merge
= false;
274 if (ShouldMergeRequestHeadersAndBody(request
, request_
->upload_data_stream
)) {
275 int merged_size
= static_cast<int>(
276 request_headers_length_
+ request_
->upload_data_stream
->size());
277 scoped_refptr
<IOBuffer
> merged_request_headers_and_body(
278 new IOBuffer(merged_size
));
279 // We'll repurpose |request_headers_| to store the merged headers and
281 request_headers_
= new DrainableIOBuffer(
282 merged_request_headers_and_body
.get(), merged_size
);
284 memcpy(request_headers_
->data(), request
.data(), request_headers_length_
);
285 request_headers_
->DidConsume(request_headers_length_
);
287 uint64 todo
= request_
->upload_data_stream
->size();
289 int consumed
= request_
->upload_data_stream
->Read(
290 request_headers_
.get(), static_cast<int>(todo
), CompletionCallback());
291 DCHECK_GT(consumed
, 0); // Read() won't fail if not chunked.
292 request_headers_
->DidConsume(consumed
);
295 DCHECK(request_
->upload_data_stream
->IsEOF());
296 // Reset the offset, so the buffer can be read from the beginning.
297 request_headers_
->SetOffset(0);
301 NetLog::TYPE_HTTP_TRANSACTION_SEND_REQUEST_BODY
,
302 base::Bind(&NetLogSendRequestBodyCallback
,
303 request_
->upload_data_stream
->size(),
304 false, /* not chunked */
309 // If we didn't merge the body with the headers, then |request_headers_|
310 // contains just the HTTP headers.
311 scoped_refptr
<StringIOBuffer
> headers_io_buf(new StringIOBuffer(request
));
313 new DrainableIOBuffer(headers_io_buf
.get(), headers_io_buf
->size());
317 if (result
== ERR_IO_PENDING
)
318 callback_
= callback
;
320 return result
> 0 ? OK
: result
;
323 int HttpStreamParser::ReadResponseHeaders(const CompletionCallback
& callback
) {
324 DCHECK(io_state_
== STATE_NONE
|| io_state_
== STATE_DONE
);
325 DCHECK(callback_
.is_null());
326 DCHECK(!callback
.is_null());
327 DCHECK_EQ(0, read_buf_unused_offset_
);
329 // This function can be called with io_state_ == STATE_DONE if the
330 // connection is closed after seeing just a 1xx response code.
331 if (io_state_
== STATE_DONE
)
332 return ERR_CONNECTION_CLOSED
;
335 io_state_
= STATE_READ_HEADERS
;
337 if (read_buf_
->offset() > 0) {
338 // Simulate the state where the data was just read from the socket.
339 result
= read_buf_
->offset();
340 read_buf_
->set_offset(0);
343 io_state_
= STATE_READ_HEADERS_COMPLETE
;
345 result
= DoLoop(result
);
346 if (result
== ERR_IO_PENDING
)
347 callback_
= callback
;
349 return result
> 0 ? OK
: result
;
352 void HttpStreamParser::Close(bool not_reusable
) {
353 if (not_reusable
&& connection_
->socket())
354 connection_
->socket()->Disconnect();
355 connection_
->Reset();
358 int HttpStreamParser::ReadResponseBody(IOBuffer
* buf
, int buf_len
,
359 const CompletionCallback
& callback
) {
360 DCHECK(io_state_
== STATE_NONE
|| io_state_
== STATE_DONE
);
361 DCHECK(callback_
.is_null());
362 DCHECK(!callback
.is_null());
363 DCHECK_LE(buf_len
, kMaxBufSize
);
365 if (io_state_
== STATE_DONE
)
368 user_read_buf_
= buf
;
369 user_read_buf_len_
= buf_len
;
370 io_state_
= STATE_READ_BODY
;
372 int result
= DoLoop(OK
);
373 if (result
== ERR_IO_PENDING
)
374 callback_
= callback
;
379 void HttpStreamParser::OnIOComplete(int result
) {
380 result
= DoLoop(result
);
382 // The client callback can do anything, including destroying this class,
383 // so any pending callback must be issued after everything else is done.
384 if (result
!= ERR_IO_PENDING
&& !callback_
.is_null()) {
385 CompletionCallback c
= callback_
;
391 int HttpStreamParser::DoLoop(int result
) {
393 DCHECK_NE(ERR_IO_PENDING
, result
);
394 DCHECK_NE(STATE_DONE
, io_state_
);
395 DCHECK_NE(STATE_NONE
, io_state_
);
396 State state
= io_state_
;
397 io_state_
= STATE_NONE
;
399 case STATE_SEND_HEADERS
:
400 DCHECK_EQ(OK
, result
);
401 result
= DoSendHeaders();
403 case STATE_SEND_HEADERS_COMPLETE
:
404 result
= DoSendHeadersComplete(result
);
406 case STATE_SEND_BODY
:
407 DCHECK_EQ(OK
, result
);
408 result
= DoSendBody();
410 case STATE_SEND_BODY_COMPLETE
:
411 result
= DoSendBodyComplete(result
);
413 case STATE_SEND_REQUEST_READ_BODY_COMPLETE
:
414 result
= DoSendRequestReadBodyComplete(result
);
416 case STATE_READ_HEADERS
:
417 net_log_
.BeginEvent(NetLog::TYPE_HTTP_STREAM_PARSER_READ_HEADERS
);
418 DCHECK_GE(result
, 0);
419 result
= DoReadHeaders();
421 case STATE_READ_HEADERS_COMPLETE
:
422 result
= DoReadHeadersComplete(result
);
423 net_log_
.EndEventWithNetErrorCode(
424 NetLog::TYPE_HTTP_STREAM_PARSER_READ_HEADERS
, result
);
426 case STATE_READ_BODY
:
427 DCHECK_GE(result
, 0);
428 result
= DoReadBody();
430 case STATE_READ_BODY_COMPLETE
:
431 result
= DoReadBodyComplete(result
);
437 } while (result
!= ERR_IO_PENDING
&&
438 (io_state_
!= STATE_DONE
&& io_state_
!= STATE_NONE
));
443 int HttpStreamParser::DoSendHeaders() {
444 // TODO(mmenke): Remove ScopedTracker below once crbug.com/424359 is fixed.
445 tracked_objects::ScopedTracker
tracking_profile(
446 FROM_HERE_WITH_EXPLICIT_FUNCTION(
447 "424359 HttpStreamParser::DoSendHeaders"));
449 int bytes_remaining
= request_headers_
->BytesRemaining();
450 DCHECK_GT(bytes_remaining
, 0);
452 // Record our best estimate of the 'request time' as the time when we send
453 // out the first bytes of the request headers.
454 if (bytes_remaining
== request_headers_
->size())
455 response_
->request_time
= base::Time::Now();
457 io_state_
= STATE_SEND_HEADERS_COMPLETE
;
458 return connection_
->socket()
459 ->Write(request_headers_
.get(), bytes_remaining
, io_callback_
);
462 int HttpStreamParser::DoSendHeadersComplete(int result
) {
464 // In the unlikely case that the headers and body were merged, all the
465 // the headers were sent, but not all of the body way, and |result| is
466 // an error that this should try reading after, stash the error for now and
467 // act like the request was successfully sent.
468 if (request_headers_
->BytesConsumed() >= request_headers_length_
&&
469 ShouldTryReadingOnUploadError(result
)) {
470 upload_error_
= result
;
476 request_headers_
->DidConsume(result
);
477 if (request_headers_
->BytesRemaining() > 0) {
478 io_state_
= STATE_SEND_HEADERS
;
482 if (request_
->upload_data_stream
!= NULL
&&
483 (request_
->upload_data_stream
->is_chunked() ||
484 // !IsEOF() indicates that the body wasn't merged.
485 (request_
->upload_data_stream
->size() > 0 &&
486 !request_
->upload_data_stream
->IsEOF()))) {
488 NetLog::TYPE_HTTP_TRANSACTION_SEND_REQUEST_BODY
,
489 base::Bind(&NetLogSendRequestBodyCallback
,
490 request_
->upload_data_stream
->size(),
491 request_
->upload_data_stream
->is_chunked(),
492 false /* not merged */));
493 io_state_
= STATE_SEND_BODY
;
497 // Finished sending the request.
501 int HttpStreamParser::DoSendBody() {
502 if (request_body_send_buf_
->BytesRemaining() > 0) {
503 io_state_
= STATE_SEND_BODY_COMPLETE
;
504 return connection_
->socket()
505 ->Write(request_body_send_buf_
.get(),
506 request_body_send_buf_
->BytesRemaining(),
510 if (request_
->upload_data_stream
->is_chunked() && sent_last_chunk_
) {
511 // Finished sending the request.
515 request_body_read_buf_
->Clear();
516 io_state_
= STATE_SEND_REQUEST_READ_BODY_COMPLETE
;
517 return request_
->upload_data_stream
->Read(request_body_read_buf_
.get(),
518 request_body_read_buf_
->capacity(),
522 int HttpStreamParser::DoSendBodyComplete(int result
) {
524 // If |result| is an error that this should try reading after, stash the
525 // error for now and act like the request was successfully sent.
526 if (ShouldTryReadingOnUploadError(result
)) {
527 upload_error_
= result
;
533 request_body_send_buf_
->DidConsume(result
);
535 io_state_
= STATE_SEND_BODY
;
539 int HttpStreamParser::DoSendRequestReadBodyComplete(int result
) {
540 // |result| is the result of read from the request body from the last call to
542 DCHECK_GE(result
, 0); // There won't be errors.
544 // Chunked data needs to be encoded.
545 if (request_
->upload_data_stream
->is_chunked()) {
546 if (result
== 0) { // Reached the end.
547 DCHECK(request_
->upload_data_stream
->IsEOF());
548 sent_last_chunk_
= true;
550 // Encode the buffer as 1 chunk.
551 const base::StringPiece
payload(request_body_read_buf_
->data(), result
);
552 request_body_send_buf_
->Clear();
553 result
= EncodeChunk(payload
,
554 request_body_send_buf_
->data(),
555 request_body_send_buf_
->capacity());
558 if (result
== 0) { // Reached the end.
559 // Reaching EOF means we can finish sending request body unless the data is
560 // chunked. (i.e. No need to send the terminal chunk.)
561 DCHECK(request_
->upload_data_stream
->IsEOF());
562 DCHECK(!request_
->upload_data_stream
->is_chunked());
563 // Finished sending the request.
564 } else if (result
> 0) {
565 request_body_send_buf_
->DidAppend(result
);
567 io_state_
= STATE_SEND_BODY
;
572 int HttpStreamParser::DoReadHeaders() {
573 io_state_
= STATE_READ_HEADERS_COMPLETE
;
575 // Grow the read buffer if necessary.
576 if (read_buf_
->RemainingCapacity() == 0)
577 read_buf_
->SetCapacity(read_buf_
->capacity() + kHeaderBufInitialSize
);
579 // http://crbug.com/16371: We're seeing |user_buf_->data()| return NULL.
580 // See if the user is passing in an IOBuffer with a NULL |data_|.
581 CHECK(read_buf_
->data());
583 return connection_
->socket()
584 ->Read(read_buf_
.get(), read_buf_
->RemainingCapacity(), io_callback_
);
587 int HttpStreamParser::DoReadHeadersComplete(int result
) {
588 result
= HandleReadHeaderResult(result
);
590 // TODO(mmenke): The code below is ugly and hacky. A much better and more
591 // flexible long term solution would be to separate out the read and write
592 // loops, though this would involve significant changes, both here and
593 // elsewhere (WebSockets, for instance).
595 // If still reading the headers, or there was no error uploading the request
596 // body, just return the result.
597 if (io_state_
== STATE_READ_HEADERS
|| upload_error_
== OK
)
600 // If the result is ERR_IO_PENDING, |io_state_| should be STATE_READ_HEADERS.
601 DCHECK_NE(ERR_IO_PENDING
, result
);
603 // On errors, use the original error received when sending the request.
604 // The main cases where these are different is when there's a header-related
605 // error code, or when there's an ERR_CONNECTION_CLOSED, which can result in
606 // special handling of partial responses and HTTP/0.9 responses.
608 // Nothing else to do. In the HTTP/0.9 or only partial headers received
609 // cases, can normally go to other states after an error reading headers.
610 io_state_
= STATE_DONE
;
611 // Don't let caller see the headers.
612 response_
->headers
= NULL
;
613 return upload_error_
;
616 // Skip over 1xx responses as usual, and allow 4xx/5xx error responses to
617 // override the error received while uploading the body.
618 int response_code_class
= response_
->headers
->response_code() / 100;
619 if (response_code_class
== 1 || response_code_class
== 4 ||
620 response_code_class
== 5) {
624 // All other status codes are not allowed after an error during upload, to
625 // make sure the consumer has some indication there was an error.
627 // Nothing else to do.
628 io_state_
= STATE_DONE
;
629 // Don't let caller see the headers.
630 response_
->headers
= NULL
;
631 return upload_error_
;
634 int HttpStreamParser::DoReadBody() {
635 io_state_
= STATE_READ_BODY_COMPLETE
;
637 // There may be some data left over from reading the response headers.
638 if (read_buf_
->offset()) {
639 int available
= read_buf_
->offset() - read_buf_unused_offset_
;
641 CHECK_GT(available
, 0);
642 int bytes_from_buffer
= std::min(available
, user_read_buf_len_
);
643 memcpy(user_read_buf_
->data(),
644 read_buf_
->StartOfBuffer() + read_buf_unused_offset_
,
646 read_buf_unused_offset_
+= bytes_from_buffer
;
647 if (bytes_from_buffer
== available
) {
648 read_buf_
->SetCapacity(0);
649 read_buf_unused_offset_
= 0;
651 return bytes_from_buffer
;
653 read_buf_
->SetCapacity(0);
654 read_buf_unused_offset_
= 0;
658 // Check to see if we're done reading.
659 if (IsResponseBodyComplete())
662 DCHECK_EQ(0, read_buf_
->offset());
663 return connection_
->socket()
664 ->Read(user_read_buf_
.get(), user_read_buf_len_
, io_callback_
);
667 int HttpStreamParser::DoReadBodyComplete(int result
) {
668 // When the connection is closed, there are numerous ways to interpret it.
670 // - If a Content-Length header is present and the body contains exactly that
671 // number of bytes at connection close, the response is successful.
673 // - If a Content-Length header is present and the body contains fewer bytes
674 // than promised by the header at connection close, it may indicate that
675 // the connection was closed prematurely, or it may indicate that the
676 // server sent an invalid Content-Length header. Unfortunately, the invalid
677 // Content-Length header case does occur in practice and other browsers are
678 // tolerant of it. We choose to treat it as an error for now, but the
679 // download system treats it as a non-error, and URLRequestHttpJob also
680 // treats it as OK if the Content-Length is the post-decoded body content
683 // - If chunked encoding is used and the terminating chunk has been processed
684 // when the connection is closed, the response is successful.
686 // - If chunked encoding is used and the terminating chunk has not been
687 // processed when the connection is closed, it may indicate that the
688 // connection was closed prematurely or it may indicate that the server
689 // sent an invalid chunked encoding. We choose to treat it as
690 // an invalid chunked encoding.
692 // - If a Content-Length is not present and chunked encoding is not used,
693 // connection close is the only way to signal that the response is
694 // complete. Unfortunately, this also means that there is no way to detect
695 // early close of a connection. No error is returned.
696 if (result
== 0 && !IsResponseBodyComplete() && CanFindEndOfResponse()) {
697 if (chunked_decoder_
.get())
698 result
= ERR_INCOMPLETE_CHUNKED_ENCODING
;
700 result
= ERR_CONTENT_LENGTH_MISMATCH
;
704 received_bytes_
+= result
;
706 // Filter incoming data if appropriate. FilterBuf may return an error.
707 if (result
> 0 && chunked_decoder_
.get()) {
708 result
= chunked_decoder_
->FilterBuf(user_read_buf_
->data(), result
);
709 if (result
== 0 && !chunked_decoder_
->reached_eof()) {
710 // Don't signal completion of the Read call yet or else it'll look like
711 // we received end-of-file. Wait for more data.
712 io_state_
= STATE_READ_BODY
;
718 response_body_read_
+= result
;
720 if (result
<= 0 || IsResponseBodyComplete()) {
721 io_state_
= STATE_DONE
;
723 // Save the overflow data, which can be in two places. There may be
724 // some left over in |user_read_buf_|, plus there may be more
725 // in |read_buf_|. But the part left over in |user_read_buf_| must have
726 // come from the |read_buf_|, so there's room to put it back at the
728 int additional_save_amount
= read_buf_
->offset() - read_buf_unused_offset_
;
730 if (chunked_decoder_
.get()) {
731 save_amount
= chunked_decoder_
->bytes_after_eof();
732 } else if (response_body_length_
>= 0) {
733 int64 extra_data_read
= response_body_read_
- response_body_length_
;
734 if (extra_data_read
> 0) {
735 save_amount
= static_cast<int>(extra_data_read
);
737 result
-= save_amount
;
741 CHECK_LE(save_amount
+ additional_save_amount
, kMaxBufSize
);
742 if (read_buf_
->capacity() < save_amount
+ additional_save_amount
) {
743 read_buf_
->SetCapacity(save_amount
+ additional_save_amount
);
747 received_bytes_
-= save_amount
;
748 memcpy(read_buf_
->StartOfBuffer(), user_read_buf_
->data() + result
,
751 read_buf_
->set_offset(save_amount
);
752 if (additional_save_amount
) {
753 memmove(read_buf_
->data(),
754 read_buf_
->StartOfBuffer() + read_buf_unused_offset_
,
755 additional_save_amount
);
756 read_buf_
->set_offset(save_amount
+ additional_save_amount
);
758 read_buf_unused_offset_
= 0;
760 // Now waiting for more of the body to be read.
761 user_read_buf_
= NULL
;
762 user_read_buf_len_
= 0;
768 int HttpStreamParser::HandleReadHeaderResult(int result
) {
769 DCHECK_EQ(0, read_buf_unused_offset_
);
772 result
= ERR_CONNECTION_CLOSED
;
774 if (result
== ERR_CONNECTION_CLOSED
) {
775 // The connection closed without getting any more data.
776 if (read_buf_
->offset() == 0) {
777 io_state_
= STATE_DONE
;
778 // If the connection has not been reused, it may have been a 0-length
779 // HTTP/0.9 responses, but it was most likely an error, so just return
780 // ERR_EMPTY_RESPONSE instead. If the connection was reused, just pass
781 // on the original connection close error, as rather than being an
782 // empty HTTP/0.9 response it's much more likely the server closed the
783 // socket before it received the request.
784 if (!connection_
->is_reused())
785 return ERR_EMPTY_RESPONSE
;
789 // Accepting truncated headers over HTTPS is a potential security
790 // vulnerability, so just return an error in that case.
792 // If response_header_start_offset_ is -1, this may be a < 8 byte HTTP/0.9
793 // response. However, accepting such a response over HTTPS would allow a
794 // MITM to truncate an HTTP/1.x status line to look like a short HTTP/0.9
795 // response if the peer put a record boundary at the first 8 bytes. To
796 // ensure that all response headers received over HTTPS are pristine, treat
797 // such responses as errors.
799 // TODO(mmenke): Returning ERR_RESPONSE_HEADERS_TRUNCATED when a response
800 // looks like an HTTP/0.9 response is weird. Should either come up with
801 // another error code, or, better, disable HTTP/0.9 over HTTPS (and give
802 // that a new error code).
803 if (request_
->url
.SchemeIsCryptographic()) {
804 io_state_
= STATE_DONE
;
805 return ERR_RESPONSE_HEADERS_TRUNCATED
;
808 // Parse things as well as we can and let the caller decide what to do.
810 if (response_header_start_offset_
>= 0) {
811 // The response looks to be a truncated set of HTTP headers.
812 io_state_
= STATE_READ_BODY_COMPLETE
;
813 end_offset
= read_buf_
->offset();
814 RecordHeaderParserEvent(HEADER_ALLOWED_TRUNCATED_HEADERS
);
816 // The response is apparently using HTTP/0.9. Treat the entire response
820 int rv
= ParseResponseHeaders(end_offset
);
827 io_state_
= STATE_DONE
;
831 // Record our best estimate of the 'response time' as the time when we read
832 // the first bytes of the response headers.
833 if (read_buf_
->offset() == 0)
834 response_
->response_time
= base::Time::Now();
836 read_buf_
->set_offset(read_buf_
->offset() + result
);
837 DCHECK_LE(read_buf_
->offset(), read_buf_
->capacity());
838 DCHECK_GE(result
, 0);
840 int end_of_header_offset
= FindAndParseResponseHeaders();
842 // Note: -1 is special, it indicates we haven't found the end of headers.
843 // Anything less than -1 is a net::Error, so we bail out.
844 if (end_of_header_offset
< -1)
845 return end_of_header_offset
;
847 if (end_of_header_offset
== -1) {
848 io_state_
= STATE_READ_HEADERS
;
849 // Prevent growing the headers buffer indefinitely.
850 if (read_buf_
->offset() >= kMaxHeaderBufSize
) {
851 io_state_
= STATE_DONE
;
852 return ERR_RESPONSE_HEADERS_TOO_BIG
;
855 CalculateResponseBodySize();
856 // If the body is zero length, the caller may not call ReadResponseBody,
857 // which is where any extra data is copied to read_buf_, so we move the
859 if (response_body_length_
== 0) {
860 int extra_bytes
= read_buf_
->offset() - end_of_header_offset
;
862 CHECK_GT(extra_bytes
, 0);
863 memmove(read_buf_
->StartOfBuffer(),
864 read_buf_
->StartOfBuffer() + end_of_header_offset
,
867 read_buf_
->SetCapacity(extra_bytes
);
868 if (response_
->headers
->response_code() / 100 == 1) {
869 // After processing a 1xx response, the caller will ask for the next
870 // header, so reset state to support that. We don't completely ignore a
871 // 1xx response because it cannot be returned in reply to a CONNECT
872 // request so we return OK here, which lets the caller inspect the
873 // response and reject it in the event that we're setting up a CONNECT
875 response_header_start_offset_
= -1;
876 response_body_length_
= -1;
877 // Now waiting for the second set of headers to be read.
879 io_state_
= STATE_DONE
;
884 // Note where the headers stop.
885 read_buf_unused_offset_
= end_of_header_offset
;
886 // Now waiting for the body to be read.
891 int HttpStreamParser::FindAndParseResponseHeaders() {
893 DCHECK_EQ(0, read_buf_unused_offset_
);
895 // Look for the start of the status line, if it hasn't been found yet.
896 if (response_header_start_offset_
< 0) {
897 response_header_start_offset_
= HttpUtil::LocateStartOfStatusLine(
898 read_buf_
->StartOfBuffer(), read_buf_
->offset());
901 if (response_header_start_offset_
>= 0) {
902 end_offset
= HttpUtil::LocateEndOfHeaders(read_buf_
->StartOfBuffer(),
904 response_header_start_offset_
);
905 } else if (read_buf_
->offset() >= 8) {
906 // Enough data to decide that this is an HTTP/0.9 response.
907 // 8 bytes = (4 bytes of junk) + "http".length()
911 if (end_offset
== -1)
914 int rv
= ParseResponseHeaders(end_offset
);
920 int HttpStreamParser::ParseResponseHeaders(int end_offset
) {
921 scoped_refptr
<HttpResponseHeaders
> headers
;
922 DCHECK_EQ(0, read_buf_unused_offset_
);
924 RecordHeaderParserEvent(HEADER_PARSER_INVOKED
);
926 if (response_header_start_offset_
> 0) {
927 bool has_non_whitespace_in_prefix
= false;
928 for (int i
= 0; i
< response_header_start_offset_
; ++i
) {
929 if (!strchr(" \t\r\n", read_buf_
->StartOfBuffer()[i
])) {
930 has_non_whitespace_in_prefix
= true;
934 if (has_non_whitespace_in_prefix
) {
935 RecordHeaderParserEvent(HEADER_SKIPPED_NON_WS_PREFIX
);
937 RecordHeaderParserEvent(HEADER_SKIPPED_WS_PREFIX
);
941 if (response_header_start_offset_
>= 0) {
942 received_bytes_
+= end_offset
;
943 headers
= new HttpResponseHeaders(HttpUtil::AssembleRawHeaders(
944 read_buf_
->StartOfBuffer(), end_offset
));
946 // Enough data was read -- there is no status line.
947 headers
= new HttpResponseHeaders(std::string("HTTP/0.9 200 OK"));
949 if (request_
->url
.SchemeIsCryptographic()) {
950 RecordHeaderParserEvent(HEADER_HTTP_09_RESPONSE_OVER_SSL
);
952 RecordHeaderParserEvent(HEADER_HTTP_09_RESPONSE_OVER_HTTP
);
954 if (connection_
->is_reused())
955 RecordHeaderParserEvent(HEADER_HTTP_09_ON_REUSED_SOCKET
);
958 // Check for multiple Content-Length headers with no Transfer-Encoding header.
959 // If they exist, and have distinct values, it's a potential response
961 if (!headers
->HasHeader("Transfer-Encoding")) {
962 if (HeadersContainMultipleCopiesOfField(*headers
.get(), "Content-Length"))
963 return ERR_RESPONSE_HEADERS_MULTIPLE_CONTENT_LENGTH
;
966 // Check for multiple Content-Disposition or Location headers. If they exist,
967 // it's also a potential response smuggling attack.
968 if (HeadersContainMultipleCopiesOfField(*headers
.get(),
969 "Content-Disposition"))
970 return ERR_RESPONSE_HEADERS_MULTIPLE_CONTENT_DISPOSITION
;
971 if (HeadersContainMultipleCopiesOfField(*headers
.get(), "Location"))
972 return ERR_RESPONSE_HEADERS_MULTIPLE_LOCATION
;
974 response_
->headers
= headers
;
975 response_
->connection_info
= HttpResponseInfo::CONNECTION_INFO_HTTP1
;
976 response_
->vary_data
.Init(*request_
, *response_
->headers
.get());
977 DVLOG(1) << __FUNCTION__
<< "()"
978 << " content_length = \"" << response_
->headers
->GetContentLength()
981 << GetResponseHeaderLines(*response_
->headers
.get()) << "\"";
985 void HttpStreamParser::CalculateResponseBodySize() {
986 // Figure how to determine EOF:
988 // For certain responses, we know the content length is always 0. From
989 // RFC 7230 Section 3.3 Message Body:
991 // The presence of a message body in a response depends on both the
992 // request method to which it is responding and the response status code
993 // (Section 3.1.2). Responses to the HEAD request method (Section 4.3.2
994 // of [RFC7231]) never include a message body because the associated
995 // response header fields (e.g., Transfer-Encoding, Content-Length,
996 // etc.), if present, indicate only what their values would have been if
997 // the request method had been GET (Section 4.3.1 of [RFC7231]). 2xx
998 // (Successful) responses to a CONNECT request method (Section 4.3.6 of
999 // [RFC7231]) switch to tunnel mode instead of having a message body.
1000 // All 1xx (Informational), 204 (No Content), and 304 (Not Modified)
1001 // responses do not include a message body. All other responses do
1002 // include a message body, although the body might be of zero length.
1004 // From RFC 7231 Section 6.3.6 205 Reset Content:
1006 // Since the 205 status code implies that no additional content will be
1007 // provided, a server MUST NOT generate a payload in a 205 response.
1008 if (response_
->headers
->response_code() / 100 == 1) {
1009 response_body_length_
= 0;
1011 switch (response_
->headers
->response_code()) {
1012 case 204: // No Content
1013 case 205: // Reset Content
1014 case 304: // Not Modified
1015 response_body_length_
= 0;
1019 if (request_
->method
== "HEAD")
1020 response_body_length_
= 0;
1022 if (response_body_length_
== -1) {
1023 // "Transfer-Encoding: chunked" trumps "Content-Length: N"
1024 if (response_
->headers
->IsChunkEncoded()) {
1025 chunked_decoder_
.reset(new HttpChunkedDecoder());
1027 response_body_length_
= response_
->headers
->GetContentLength();
1028 // If response_body_length_ is still -1, then we have to wait
1029 // for the server to close the connection.
1034 UploadProgress
HttpStreamParser::GetUploadProgress() const {
1035 if (!request_
->upload_data_stream
)
1036 return UploadProgress();
1038 return UploadProgress(request_
->upload_data_stream
->position(),
1039 request_
->upload_data_stream
->size());
1042 bool HttpStreamParser::IsResponseBodyComplete() const {
1043 if (chunked_decoder_
.get())
1044 return chunked_decoder_
->reached_eof();
1045 if (response_body_length_
!= -1)
1046 return response_body_read_
>= response_body_length_
;
1048 return false; // Must read to EOF.
1051 bool HttpStreamParser::CanFindEndOfResponse() const {
1052 return chunked_decoder_
.get() || response_body_length_
>= 0;
1055 bool HttpStreamParser::IsMoreDataBuffered() const {
1056 return read_buf_
->offset() > read_buf_unused_offset_
;
1059 bool HttpStreamParser::IsConnectionReused() const {
1060 ClientSocketHandle::SocketReuseType reuse_type
= connection_
->reuse_type();
1061 return connection_
->is_reused() ||
1062 reuse_type
== ClientSocketHandle::UNUSED_IDLE
;
1065 void HttpStreamParser::SetConnectionReused() {
1066 connection_
->set_reuse_type(ClientSocketHandle::REUSED_IDLE
);
1069 bool HttpStreamParser::IsConnectionReusable() const {
1070 return connection_
->socket() && connection_
->socket()->IsConnectedAndIdle();
1073 void HttpStreamParser::GetSSLInfo(SSLInfo
* ssl_info
) {
1074 if (request_
->url
.SchemeIsCryptographic() && connection_
->socket()) {
1075 SSLClientSocket
* ssl_socket
=
1076 static_cast<SSLClientSocket
*>(connection_
->socket());
1077 ssl_socket
->GetSSLInfo(ssl_info
);
1081 void HttpStreamParser::GetSSLCertRequestInfo(
1082 SSLCertRequestInfo
* cert_request_info
) {
1083 if (request_
->url
.SchemeIsCryptographic() && connection_
->socket()) {
1084 SSLClientSocket
* ssl_socket
=
1085 static_cast<SSLClientSocket
*>(connection_
->socket());
1086 ssl_socket
->GetSSLCertRequestInfo(cert_request_info
);
1090 int HttpStreamParser::EncodeChunk(const base::StringPiece
& payload
,
1092 size_t output_size
) {
1093 if (output_size
< payload
.size() + kChunkHeaderFooterSize
)
1094 return ERR_INVALID_ARGUMENT
;
1096 char* cursor
= output
;
1098 const int num_chars
= base::snprintf(output
, output_size
,
1100 static_cast<int>(payload
.size()));
1101 cursor
+= num_chars
;
1102 // Add the payload if any.
1103 if (payload
.size() > 0) {
1104 memcpy(cursor
, payload
.data(), payload
.size());
1105 cursor
+= payload
.size();
1107 // Add the trailing CRLF.
1108 memcpy(cursor
, "\r\n", 2);
1111 return cursor
- output
;
1115 bool HttpStreamParser::ShouldMergeRequestHeadersAndBody(
1116 const std::string
& request_headers
,
1117 const UploadDataStream
* request_body
) {
1118 if (request_body
!= NULL
&&
1119 // IsInMemory() ensures that the request body is not chunked.
1120 request_body
->IsInMemory() &&
1121 request_body
->size() > 0) {
1122 uint64 merged_size
= request_headers
.size() + request_body
->size();
1123 if (merged_size
<= kMaxMergedHeaderAndBodySize
)