]> git.ipfire.org Git - thirdparty/squid.git/blob - src/http.cc
Bug 7: Update cached entries on 304 responses.
[thirdparty/squid.git] / src / http.cc
1 /*
2 * Copyright (C) 1996-2016 The Squid Software Foundation and contributors
3 *
4 * Squid software is distributed under GPLv2+ license and includes
5 * contributions from numerous individuals and organizations.
6 * Please see the COPYING and CONTRIBUTORS files for details.
7 */
8
9 /* DEBUG: section 11 Hypertext Transfer Protocol (HTTP) */
10
11 /*
12 * Anonymizing patch by lutz@as-node.jena.thur.de
13 * have a look into http-anon.c to get more informations.
14 */
15
16 #include "squid.h"
17 #include "acl/FilledChecklist.h"
18 #include "base/AsyncJobCalls.h"
19 #include "base/TextException.h"
20 #include "base64.h"
21 #include "CachePeer.h"
22 #include "client_side.h"
23 #include "comm/Connection.h"
24 #include "comm/Read.h"
25 #include "comm/Write.h"
26 #include "CommRead.h"
27 #include "err_detail_type.h"
28 #include "errorpage.h"
29 #include "fd.h"
30 #include "fde.h"
31 #include "globals.h"
32 #include "http.h"
33 #include "http/one/ResponseParser.h"
34 #include "http/one/TeChunkedParser.h"
35 #include "http/Stream.h"
36 #include "HttpControlMsg.h"
37 #include "HttpHdrCc.h"
38 #include "HttpHdrContRange.h"
39 #include "HttpHdrSc.h"
40 #include "HttpHdrScTarget.h"
41 #include "HttpHeaderTools.h"
42 #include "HttpReply.h"
43 #include "HttpRequest.h"
44 #include "HttpStateFlags.h"
45 #include "log/access_log.h"
46 #include "MemBuf.h"
47 #include "MemObject.h"
48 #include "neighbors.h"
49 #include "peer_proxy_negotiate_auth.h"
50 #include "profiler/Profiler.h"
51 #include "refresh.h"
52 #include "RefreshPattern.h"
53 #include "rfc1738.h"
54 #include "SquidConfig.h"
55 #include "SquidTime.h"
56 #include "StatCounters.h"
57 #include "Store.h"
58 #include "StrList.h"
59 #include "tools.h"
60 #include "URL.h"
61 #include "util.h"
62
63 #if USE_AUTH
64 #include "auth/UserRequest.h"
65 #endif
66 #if USE_DELAY_POOLS
67 #include "DelayPools.h"
68 #endif
69
70 #define SQUID_ENTER_THROWING_CODE() try {
71 #define SQUID_EXIT_THROWING_CODE(status) \
72 status = true; \
73 } \
74 catch (const std::exception &e) { \
75 debugs (11, 1, "Exception error:" << e.what()); \
76 status = false; \
77 }
78
79 CBDATA_CLASS_INIT(HttpStateData);
80
81 static const char *const crlf = "\r\n";
82
83 static void httpMaybeRemovePublic(StoreEntry *, Http::StatusCode);
84 static void copyOneHeaderFromClientsideRequestToUpstreamRequest(const HttpHeaderEntry *e, const String strConnection, const HttpRequest * request,
85 HttpHeader * hdr_out, const int we_do_ranges, const HttpStateFlags &);
86 //Declared in HttpHeaderTools.cc
87 void httpHdrAdd(HttpHeader *heads, HttpRequest *request, const AccessLogEntryPointer &al, HeaderWithAclList &headers_add);
88
89 HttpStateData::HttpStateData(FwdState *theFwdState) :
90 AsyncJob("HttpStateData"),
91 Client(theFwdState),
92 lastChunk(0),
93 httpChunkDecoder(NULL),
94 payloadSeen(0),
95 payloadTruncated(0)
96 {
97 debugs(11,5,HERE << "HttpStateData " << this << " created");
98 ignoreCacheControl = false;
99 surrogateNoStore = false;
100 serverConnection = fwd->serverConnection();
101
102 // reset peer response time stats for %<pt
103 request->hier.peer_http_request_sent.tv_sec = 0;
104 request->hier.peer_http_request_sent.tv_usec = 0;
105
106 if (fwd->serverConnection() != NULL)
107 _peer = cbdataReference(fwd->serverConnection()->getPeer()); /* might be NULL */
108
109 if (_peer) {
110 request->flags.proxying = true;
111 /*
112 * This NEIGHBOR_PROXY_ONLY check probably shouldn't be here.
113 * We might end up getting the object from somewhere else if,
114 * for example, the request to this neighbor fails.
115 */
116 if (_peer->options.proxy_only)
117 entry->releaseRequest();
118
119 #if USE_DELAY_POOLS
120 entry->setNoDelay(_peer->options.no_delay);
121 #endif
122 }
123
124 /*
125 * register the handler to free HTTP state data when the FD closes
126 */
127 typedef CommCbMemFunT<HttpStateData, CommCloseCbParams> Dialer;
128 closeHandler = JobCallback(9, 5, Dialer, this, HttpStateData::httpStateConnClosed);
129 comm_add_close_handler(serverConnection->fd, closeHandler);
130 }
131
132 HttpStateData::~HttpStateData()
133 {
134 /*
135 * don't forget that ~Client() gets called automatically
136 */
137
138 if (httpChunkDecoder)
139 delete httpChunkDecoder;
140
141 cbdataReferenceDone(_peer);
142
143 debugs(11,5, HERE << "HttpStateData " << this << " destroyed; " << serverConnection);
144 }
145
146 const Comm::ConnectionPointer &
147 HttpStateData::dataConnection() const
148 {
149 return serverConnection;
150 }
151
152 void
153 HttpStateData::httpStateConnClosed(const CommCloseCbParams &params)
154 {
155 debugs(11, 5, "httpStateFree: FD " << params.fd << ", httpState=" << params.data);
156 doneWithFwd = "httpStateConnClosed()"; // assume FwdState is monitoring too
157 mustStop("HttpStateData::httpStateConnClosed");
158 }
159
160 void
161 HttpStateData::httpTimeout(const CommTimeoutCbParams &)
162 {
163 debugs(11, 4, serverConnection << ": '" << entry->url() << "'");
164
165 if (entry->store_status == STORE_PENDING) {
166 fwd->fail(new ErrorState(ERR_READ_TIMEOUT, Http::scGatewayTimeout, fwd->request));
167 }
168
169 closeServer();
170 mustStop("HttpStateData::httpTimeout");
171 }
172
173 /// Remove an existing public store entry if the incoming response (to be
174 /// stored in a currently private entry) is going to invalidate it.
175 static void
176 httpMaybeRemovePublic(StoreEntry * e, Http::StatusCode status)
177 {
178 int remove = 0;
179 int forbidden = 0;
180 StoreEntry *pe;
181
182 // If the incoming response already goes into a public entry, then there is
183 // nothing to remove. This protects ready-for-collapsing entries as well.
184 if (!EBIT_TEST(e->flags, KEY_PRIVATE))
185 return;
186
187 switch (status) {
188
189 case Http::scOkay:
190
191 case Http::scNonAuthoritativeInformation:
192
193 case Http::scMultipleChoices:
194
195 case Http::scMovedPermanently:
196
197 case Http::scFound:
198
199 case Http::scGone:
200
201 case Http::scNotFound:
202 remove = 1;
203
204 break;
205
206 case Http::scForbidden:
207
208 case Http::scMethodNotAllowed:
209 forbidden = 1;
210
211 break;
212
213 #if WORK_IN_PROGRESS
214
215 case Http::scUnauthorized:
216 forbidden = 1;
217
218 break;
219
220 #endif
221
222 default:
223 #if QUESTIONABLE
224 /*
225 * Any 2xx response should eject previously cached entities...
226 */
227
228 if (status >= 200 && status < 300)
229 remove = 1;
230
231 #endif
232
233 break;
234 }
235
236 if (!remove && !forbidden)
237 return;
238
239 assert(e->mem_obj);
240
241 if (e->mem_obj->request)
242 pe = storeGetPublicByRequest(e->mem_obj->request);
243 else
244 pe = storeGetPublic(e->mem_obj->storeId(), e->mem_obj->method);
245
246 if (pe != NULL) {
247 assert(e != pe);
248 #if USE_HTCP
249 neighborsHtcpClear(e, NULL, e->mem_obj->request, e->mem_obj->method, HTCP_CLR_INVALIDATION);
250 #endif
251 pe->release();
252 }
253
254 /** \par
255 * Also remove any cached HEAD response in case the object has
256 * changed.
257 */
258 if (e->mem_obj->request)
259 pe = storeGetPublicByRequestMethod(e->mem_obj->request, Http::METHOD_HEAD);
260 else
261 pe = storeGetPublic(e->mem_obj->storeId(), Http::METHOD_HEAD);
262
263 if (pe != NULL) {
264 assert(e != pe);
265 #if USE_HTCP
266 neighborsHtcpClear(e, NULL, e->mem_obj->request, HttpRequestMethod(Http::METHOD_HEAD), HTCP_CLR_INVALIDATION);
267 #endif
268 pe->release();
269 }
270 }
271
272 void
273 HttpStateData::processSurrogateControl(HttpReply *reply)
274 {
275 if (request->flags.accelerated && reply->surrogate_control) {
276 HttpHdrScTarget *sctusable = reply->surrogate_control->getMergedTarget(Config.Accel.surrogate_id);
277
278 if (sctusable) {
279 if (sctusable->noStore() ||
280 (Config.onoff.surrogate_is_remote
281 && sctusable->noStoreRemote())) {
282 surrogateNoStore = true;
283 entry->makePrivate();
284 }
285
286 /* The HttpHeader logic cannot tell if the header it's parsing is a reply to an
287 * accelerated request or not...
288 * Still, this is an abstraction breach. - RC
289 */
290 if (sctusable->hasMaxAge()) {
291 if (sctusable->maxAge() < sctusable->maxStale())
292 reply->expires = reply->date + sctusable->maxAge();
293 else
294 reply->expires = reply->date + sctusable->maxStale();
295
296 /* And update the timestamps */
297 entry->timestampsSet();
298 }
299
300 /* We ignore cache-control directives as per the Surrogate specification */
301 ignoreCacheControl = true;
302
303 delete sctusable;
304 }
305 }
306 }
307
308 int
309 HttpStateData::cacheableReply()
310 {
311 HttpReply const *rep = finalReply();
312 HttpHeader const *hdr = &rep->header;
313 const char *v;
314 #if USE_HTTP_VIOLATIONS
315
316 const RefreshPattern *R = NULL;
317
318 /* This strange looking define first looks up the refresh pattern
319 * and then checks if the specified flag is set. The main purpose
320 * of this is to simplify the refresh pattern lookup and USE_HTTP_VIOLATIONS
321 * condition
322 */
323 #define REFRESH_OVERRIDE(flag) \
324 ((R = (R ? R : refreshLimits(entry->mem_obj->storeId()))) , \
325 (R && R->flags.flag))
326 #else
327 #define REFRESH_OVERRIDE(flag) 0
328 #endif
329
330 if (EBIT_TEST(entry->flags, RELEASE_REQUEST)) {
331 debugs(22, 3, "NO because " << *entry << " has been released.");
332 return 0;
333 }
334
335 // Check for Surrogate/1.0 protocol conditions
336 // NP: reverse-proxy traffic our parent server has instructed us never to cache
337 if (surrogateNoStore) {
338 debugs(22, 3, HERE << "NO because Surrogate-Control:no-store");
339 return 0;
340 }
341
342 // RFC 2616: HTTP/1.1 Cache-Control conditions
343 if (!ignoreCacheControl) {
344 // XXX: check to see if the request headers alone were enough to prevent caching earlier
345 // (ie no-store request header) no need to check those all again here if so.
346 // for now we are not reliably doing that so we waste CPU re-checking request CC
347
348 // RFC 2616 section 14.9.2 - MUST NOT cache any response with request CC:no-store
349 if (request && request->cache_control && request->cache_control->noStore() &&
350 !REFRESH_OVERRIDE(ignore_no_store)) {
351 debugs(22, 3, HERE << "NO because client request Cache-Control:no-store");
352 return 0;
353 }
354
355 // NP: request CC:no-cache only means cache READ is forbidden. STORE is permitted.
356 if (rep->cache_control && rep->cache_control->hasNoCache() && rep->cache_control->noCache().size() > 0) {
357 /* TODO: we are allowed to cache when no-cache= has parameters.
358 * Provided we strip away any of the listed headers unless they are revalidated
359 * successfully (ie, must revalidate AND these headers are prohibited on stale replies).
360 * That is a bit tricky for squid right now so we avoid caching entirely.
361 */
362 debugs(22, 3, HERE << "NO because server reply Cache-Control:no-cache has parameters");
363 return 0;
364 }
365
366 // NP: request CC:private is undefined. We ignore.
367 // NP: other request CC flags are limiters on HIT/MISS. We don't care about here.
368
369 // RFC 2616 section 14.9.2 - MUST NOT cache any response with CC:no-store
370 if (rep->cache_control && rep->cache_control->noStore() &&
371 !REFRESH_OVERRIDE(ignore_no_store)) {
372 debugs(22, 3, HERE << "NO because server reply Cache-Control:no-store");
373 return 0;
374 }
375
376 // RFC 2616 section 14.9.1 - MUST NOT cache any response with CC:private in a shared cache like Squid.
377 // CC:private overrides CC:public when both are present in a response.
378 // TODO: add a shared/private cache configuration possibility.
379 if (rep->cache_control &&
380 rep->cache_control->hasPrivate() &&
381 !REFRESH_OVERRIDE(ignore_private)) {
382 /* TODO: we are allowed to cache when private= has parameters.
383 * Provided we strip away any of the listed headers unless they are revalidated
384 * successfully (ie, must revalidate AND these headers are prohibited on stale replies).
385 * That is a bit tricky for squid right now so we avoid caching entirely.
386 */
387 debugs(22, 3, HERE << "NO because server reply Cache-Control:private");
388 return 0;
389 }
390 }
391
392 // RFC 2068, sec 14.9.4 - MUST NOT cache any response with Authentication UNLESS certain CC controls are present
393 // allow HTTP violations to IGNORE those controls (ie re-block caching Auth)
394 if (request && (request->flags.auth || request->flags.authSent)) {
395 if (!rep->cache_control) {
396 debugs(22, 3, HERE << "NO because Authenticated and server reply missing Cache-Control");
397 return 0;
398 }
399
400 if (ignoreCacheControl) {
401 debugs(22, 3, HERE << "NO because Authenticated and ignoring Cache-Control");
402 return 0;
403 }
404
405 bool mayStore = false;
406 // HTTPbis pt6 section 3.2: a response CC:public is present
407 if (rep->cache_control->Public()) {
408 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:public");
409 mayStore = true;
410
411 // HTTPbis pt6 section 3.2: a response CC:must-revalidate is present
412 } else if (rep->cache_control->mustRevalidate()) {
413 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:must-revalidate");
414 mayStore = true;
415
416 #if USE_HTTP_VIOLATIONS
417 // NP: given the must-revalidate exception we should also be able to exempt no-cache.
418 // HTTPbis WG verdict on this is that it is omitted from the spec due to being 'unexpected' by
419 // some. The caching+revalidate is not exactly unsafe though with Squids interpretation of no-cache
420 // (without parameters) as equivalent to must-revalidate in the reply.
421 } else if (rep->cache_control->hasNoCache() && rep->cache_control->noCache().size() == 0) {
422 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:no-cache (equivalent to must-revalidate)");
423 mayStore = true;
424 #endif
425
426 // HTTPbis pt6 section 3.2: a response CC:s-maxage is present
427 } else if (rep->cache_control->sMaxAge()) {
428 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:s-maxage");
429 mayStore = true;
430 }
431
432 if (!mayStore) {
433 debugs(22, 3, HERE << "NO because Authenticated transaction");
434 return 0;
435 }
436
437 // NP: response CC:no-cache is equivalent to CC:must-revalidate,max-age=0. We MAY cache, and do so.
438 // NP: other request CC flags are limiters on HIT/MISS/REFRESH. We don't care about here.
439 }
440
441 /* HACK: The "multipart/x-mixed-replace" content type is used for
442 * continuous push replies. These are generally dynamic and
443 * probably should not be cachable
444 */
445 if ((v = hdr->getStr(Http::HdrType::CONTENT_TYPE)))
446 if (!strncasecmp(v, "multipart/x-mixed-replace", 25)) {
447 debugs(22, 3, HERE << "NO because Content-Type:multipart/x-mixed-replace");
448 return 0;
449 }
450
451 switch (rep->sline.status()) {
452 /* Responses that are cacheable */
453
454 case Http::scOkay:
455
456 case Http::scNonAuthoritativeInformation:
457
458 case Http::scMultipleChoices:
459
460 case Http::scMovedPermanently:
461 case Http::scPermanentRedirect:
462
463 case Http::scGone:
464 /*
465 * Don't cache objects that need to be refreshed on next request,
466 * unless we know how to refresh it.
467 */
468
469 if (!refreshIsCachable(entry) && !REFRESH_OVERRIDE(store_stale)) {
470 debugs(22, 3, "NO because refreshIsCachable() returned non-cacheable..");
471 return 0;
472 } else {
473 debugs(22, 3, HERE << "YES because HTTP status " << rep->sline.status());
474 return 1;
475 }
476 /* NOTREACHED */
477 break;
478
479 /* Responses that only are cacheable if the server says so */
480
481 case Http::scFound:
482 case Http::scTemporaryRedirect:
483 if (rep->date <= 0) {
484 debugs(22, 3, HERE << "NO because HTTP status " << rep->sline.status() << " and Date missing/invalid");
485 return 0;
486 }
487 if (rep->expires > rep->date) {
488 debugs(22, 3, HERE << "YES because HTTP status " << rep->sline.status() << " and Expires > Date");
489 return 1;
490 } else {
491 debugs(22, 3, HERE << "NO because HTTP status " << rep->sline.status() << " and Expires <= Date");
492 return 0;
493 }
494 /* NOTREACHED */
495 break;
496
497 /* Errors can be negatively cached */
498
499 case Http::scNoContent:
500
501 case Http::scUseProxy:
502
503 case Http::scBadRequest:
504
505 case Http::scForbidden:
506
507 case Http::scNotFound:
508
509 case Http::scMethodNotAllowed:
510
511 case Http::scUriTooLong:
512
513 case Http::scInternalServerError:
514
515 case Http::scNotImplemented:
516
517 case Http::scBadGateway:
518
519 case Http::scServiceUnavailable:
520
521 case Http::scGatewayTimeout:
522 case Http::scMisdirectedRequest:
523
524 debugs(22, 3, "MAYBE because HTTP status " << rep->sline.status());
525 return -1;
526
527 /* NOTREACHED */
528 break;
529
530 /* Some responses can never be cached */
531
532 case Http::scPartialContent: /* Not yet supported */
533
534 case Http::scSeeOther:
535
536 case Http::scNotModified:
537
538 case Http::scUnauthorized:
539
540 case Http::scProxyAuthenticationRequired:
541
542 case Http::scInvalidHeader: /* Squid header parsing error */
543
544 case Http::scHeaderTooLarge:
545
546 case Http::scPaymentRequired:
547 case Http::scNotAcceptable:
548 case Http::scRequestTimeout:
549 case Http::scConflict:
550 case Http::scLengthRequired:
551 case Http::scPreconditionFailed:
552 case Http::scPayloadTooLarge:
553 case Http::scUnsupportedMediaType:
554 case Http::scUnprocessableEntity:
555 case Http::scLocked:
556 case Http::scFailedDependency:
557 case Http::scInsufficientStorage:
558 case Http::scRequestedRangeNotSatisfied:
559 case Http::scExpectationFailed:
560
561 debugs(22, 3, HERE << "NO because HTTP status " << rep->sline.status());
562 return 0;
563
564 default:
565 /* RFC 2616 section 6.1.1: an unrecognized response MUST NOT be cached. */
566 debugs (11, 3, HERE << "NO because unknown HTTP status code " << rep->sline.status());
567 return 0;
568
569 /* NOTREACHED */
570 break;
571 }
572
573 /* NOTREACHED */
574 }
575
576 /*
577 * For Vary, store the relevant request headers as
578 * virtual headers in the reply
579 * Returns false if the variance cannot be stored
580 */
581 const char *
582 httpMakeVaryMark(HttpRequest * request, HttpReply const * reply)
583 {
584 String vary, hdr;
585 const char *pos = NULL;
586 const char *item;
587 const char *value;
588 int ilen;
589 static String vstr;
590
591 vstr.clean();
592 vary = reply->header.getList(Http::HdrType::VARY);
593
594 while (strListGetItem(&vary, ',', &item, &ilen, &pos)) {
595 static const SBuf asterisk("*");
596 SBuf name(item, ilen);
597 if (name == asterisk) {
598 vstr.clean();
599 break;
600 }
601 name.toLower();
602 strListAdd(&vstr, name.c_str(), ',');
603 hdr = request->header.getByName(name);
604 value = hdr.termedBuf();
605 if (value) {
606 value = rfc1738_escape_part(value);
607 vstr.append("=\"", 2);
608 vstr.append(value);
609 vstr.append("\"", 1);
610 }
611
612 hdr.clean();
613 }
614
615 vary.clean();
616 #if X_ACCELERATOR_VARY
617
618 pos = NULL;
619 vary = reply->header.getList(Http::HdrType::HDR_X_ACCELERATOR_VARY);
620
621 while (strListGetItem(&vary, ',', &item, &ilen, &pos)) {
622 char *name = (char *)xmalloc(ilen + 1);
623 xstrncpy(name, item, ilen + 1);
624 Tolower(name);
625 strListAdd(&vstr, name, ',');
626 hdr = request->header.getByName(name);
627 safe_free(name);
628 value = hdr.termedBuf();
629
630 if (value) {
631 value = rfc1738_escape_part(value);
632 vstr.append("=\"", 2);
633 vstr.append(value);
634 vstr.append("\"", 1);
635 }
636
637 hdr.clean();
638 }
639
640 vary.clean();
641 #endif
642
643 debugs(11, 3, "httpMakeVaryMark: " << vstr);
644 return vstr.termedBuf();
645 }
646
647 void
648 HttpStateData::keepaliveAccounting(HttpReply *reply)
649 {
650 if (flags.keepalive)
651 if (_peer)
652 ++ _peer->stats.n_keepalives_sent;
653
654 if (reply->keep_alive) {
655 if (_peer)
656 ++ _peer->stats.n_keepalives_recv;
657
658 if (Config.onoff.detect_broken_server_pconns
659 && reply->bodySize(request->method) == -1 && !flags.chunked) {
660 debugs(11, DBG_IMPORTANT, "keepaliveAccounting: Impossible keep-alive header from '" << entry->url() << "'" );
661 // debugs(11, 2, "GOT HTTP REPLY HDR:\n---------\n" << readBuf->content() << "\n----------" );
662 flags.keepalive_broken = true;
663 }
664 }
665 }
666
667 void
668 HttpStateData::checkDateSkew(HttpReply *reply)
669 {
670 if (reply->date > -1 && !_peer) {
671 int skew = abs((int)(reply->date - squid_curtime));
672
673 if (skew > 86400)
674 debugs(11, 3, "" << request->url.host() << "'s clock is skewed by " << skew << " seconds!");
675 }
676 }
677
678 /**
679 * This creates the error page itself.. its likely
680 * that the forward ported reply header max size patch
681 * generates non http conformant error pages - in which
682 * case the errors where should be 'BAD_GATEWAY' etc
683 */
684 void
685 HttpStateData::processReplyHeader()
686 {
687 /** Creates a blank header. If this routine is made incremental, this will not do */
688
689 /* NP: all exit points to this function MUST call ctx_exit(ctx) */
690 Ctx ctx = ctx_enter(entry->mem_obj->urlXXX());
691
692 debugs(11, 3, "processReplyHeader: key '" << entry->getMD5Text() << "'");
693
694 assert(!flags.headers_parsed);
695
696 if (!inBuf.length()) {
697 ctx_exit(ctx);
698 return;
699 }
700
701 /* Attempt to parse the first line; this will define where the protocol, status, reason-phrase and header begin */
702 {
703 if (hp == NULL)
704 hp = new Http1::ResponseParser;
705
706 bool parsedOk = hp->parse(inBuf);
707
708 // sync the buffers after parsing.
709 inBuf = hp->remaining();
710
711 if (hp->needsMoreData()) {
712 if (eof) { // no more data coming
713 /* Bug 2879: Replies may terminate with \r\n then EOF instead of \r\n\r\n.
714 * We also may receive truncated responses.
715 * Ensure here that we have at minimum two \r\n when EOF is seen.
716 */
717 inBuf.append("\r\n\r\n", 4);
718 // retry the parse
719 parsedOk = hp->parse(inBuf);
720 // sync the buffers after parsing.
721 inBuf = hp->remaining();
722 } else {
723 debugs(33, 5, "Incomplete response, waiting for end of response headers");
724 ctx_exit(ctx);
725 return;
726 }
727 }
728
729 if (!parsedOk) {
730 // unrecoverable parsing error
731 // TODO: Use Raw! XXX: inBuf no longer has the [beginning of the] malformed header.
732 debugs(11, 3, "Non-HTTP-compliant header:\n---------\n" << inBuf << "\n----------");
733 flags.headers_parsed = true;
734 HttpReply *newrep = new HttpReply;
735 newrep->sline.set(Http::ProtocolVersion(), hp->parseStatusCode);
736 setVirginReply(newrep);
737 ctx_exit(ctx);
738 return;
739 }
740 }
741
742 /* We know the whole response is in parser now */
743 debugs(11, 2, "HTTP Server " << serverConnection);
744 debugs(11, 2, "HTTP Server RESPONSE:\n---------\n" <<
745 hp->messageProtocol() << " " << hp->messageStatus() << " " << hp->reasonPhrase() << "\n" <<
746 hp->mimeHeader() <<
747 "----------");
748
749 // reset payload tracking to begin after message headers
750 payloadSeen = inBuf.length();
751
752 HttpReply *newrep = new HttpReply;
753 // XXX: RFC 7230 indicates we MAY ignore the reason phrase,
754 // and use an empty string on unknown status.
755 // We do that now to avoid performance regression from using SBuf::c_str()
756 newrep->sline.set(Http::ProtocolVersion(1,1), hp->messageStatus() /* , hp->reasonPhrase() */);
757 newrep->sline.protocol = newrep->sline.version.protocol = hp->messageProtocol().protocol;
758 newrep->sline.version.major = hp->messageProtocol().major;
759 newrep->sline.version.minor = hp->messageProtocol().minor;
760
761 // parse headers
762 if (!newrep->parseHeader(*hp)) {
763 // XXX: when Http::ProtocolVersion is a function, remove this hack. just set with messageProtocol()
764 newrep->sline.set(Http::ProtocolVersion(), Http::scInvalidHeader);
765 newrep->sline.version.protocol = hp->messageProtocol().protocol;
766 newrep->sline.version.major = hp->messageProtocol().major;
767 newrep->sline.version.minor = hp->messageProtocol().minor;
768 debugs(11, 2, "error parsing response headers mime block");
769 }
770
771 // done with Parser, now process using the HttpReply
772 hp = NULL;
773
774 newrep->sources |= request->url.getScheme() == AnyP::PROTO_HTTPS ? HttpMsg::srcHttps : HttpMsg::srcHttp;
775
776 newrep->removeStaleWarnings();
777
778 if (newrep->sline.protocol == AnyP::PROTO_HTTP && newrep->sline.status() >= 100 && newrep->sline.status() < 200) {
779 handle1xx(newrep);
780 ctx_exit(ctx);
781 return;
782 }
783
784 flags.chunked = false;
785 if (newrep->sline.protocol == AnyP::PROTO_HTTP && newrep->header.chunked()) {
786 flags.chunked = true;
787 httpChunkDecoder = new Http1::TeChunkedParser;
788 }
789
790 if (!peerSupportsConnectionPinning())
791 request->flags.connectionAuthDisabled = true;
792
793 HttpReply *vrep = setVirginReply(newrep);
794 flags.headers_parsed = true;
795
796 keepaliveAccounting(vrep);
797
798 checkDateSkew(vrep);
799
800 processSurrogateControl (vrep);
801
802 request->hier.peer_reply_status = newrep->sline.status();
803
804 ctx_exit(ctx);
805 }
806
807 /// ignore or start forwarding the 1xx response (a.k.a., control message)
808 void
809 HttpStateData::handle1xx(HttpReply *reply)
810 {
811 HttpReply::Pointer msg(reply); // will destroy reply if unused
812
813 // one 1xx at a time: we must not be called while waiting for previous 1xx
814 Must(!flags.handling1xx);
815 flags.handling1xx = true;
816
817 if (!request->canHandle1xx() || request->forcedBodyContinuation) {
818 debugs(11, 2, "ignoring 1xx because it is " << (request->forcedBodyContinuation ? "already sent" : "not supported by client"));
819 proceedAfter1xx();
820 return;
821 }
822
823 #if USE_HTTP_VIOLATIONS
824 // check whether the 1xx response forwarding is allowed by squid.conf
825 if (Config.accessList.reply) {
826 ACLFilledChecklist ch(Config.accessList.reply, originalRequest(), NULL);
827 ch.reply = reply;
828 HTTPMSGLOCK(ch.reply);
829 if (ch.fastCheck() != ACCESS_ALLOWED) { // TODO: support slow lookups?
830 debugs(11, 3, HERE << "ignoring denied 1xx");
831 proceedAfter1xx();
832 return;
833 }
834 }
835 #endif // USE_HTTP_VIOLATIONS
836
837 debugs(11, 2, HERE << "forwarding 1xx to client");
838
839 // the Sink will use this to call us back after writing 1xx to the client
840 typedef NullaryMemFunT<HttpStateData> CbDialer;
841 const AsyncCall::Pointer cb = JobCallback(11, 3, CbDialer, this,
842 HttpStateData::proceedAfter1xx);
843 CallJobHere1(11, 4, request->clientConnectionManager, ConnStateData,
844 ConnStateData::sendControlMsg, HttpControlMsg(msg, cb));
845 // If the call is not fired, then the Sink is gone, and HttpStateData
846 // will terminate due to an aborted store entry or another similar error.
847 // If we get stuck, it is not handle1xx fault if we could get stuck
848 // for similar reasons without a 1xx response.
849 }
850
851 /// restores state and resumes processing after 1xx is ignored or forwarded
852 void
853 HttpStateData::proceedAfter1xx()
854 {
855 Must(flags.handling1xx);
856 debugs(11, 2, "continuing with " << payloadSeen << " bytes in buffer after 1xx");
857 CallJobHere(11, 3, this, HttpStateData, HttpStateData::processReply);
858 }
859
860 /**
861 * returns true if the peer can support connection pinning
862 */
863 bool HttpStateData::peerSupportsConnectionPinning() const
864 {
865 const HttpReply *rep = entry->mem_obj->getReply();
866 const HttpHeader *hdr = &rep->header;
867 bool rc;
868 String header;
869
870 if (!_peer)
871 return true;
872
873 /*If this peer does not support connection pinning (authenticated
874 connections) return false
875 */
876 if (!_peer->connection_auth)
877 return false;
878
879 /*The peer supports connection pinning and the http reply status
880 is not unauthorized, so the related connection can be pinned
881 */
882 if (rep->sline.status() != Http::scUnauthorized)
883 return true;
884
885 /*The server respond with Http::scUnauthorized and the peer configured
886 with "connection-auth=on" we know that the peer supports pinned
887 connections
888 */
889 if (_peer->connection_auth == 1)
890 return true;
891
892 /*At this point peer has configured with "connection-auth=auto"
893 parameter so we need some extra checks to decide if we are going
894 to allow pinned connections or not
895 */
896
897 /*if the peer configured with originserver just allow connection
898 pinning (squid 2.6 behaviour)
899 */
900 if (_peer->options.originserver)
901 return true;
902
903 /*if the connections it is already pinned it is OK*/
904 if (request->flags.pinned)
905 return true;
906
907 /*Allow pinned connections only if the Proxy-support header exists in
908 reply and has in its list the "Session-Based-Authentication"
909 which means that the peer supports connection pinning.
910 */
911 if (!hdr->has(Http::HdrType::PROXY_SUPPORT))
912 return false;
913
914 header = hdr->getStrOrList(Http::HdrType::PROXY_SUPPORT);
915 /* XXX This ought to be done in a case-insensitive manner */
916 rc = (strstr(header.termedBuf(), "Session-Based-Authentication") != NULL);
917
918 return rc;
919 }
920
921 // Called when we parsed (and possibly adapted) the headers but
922 // had not starting storing (a.k.a., sending) the body yet.
923 void
924 HttpStateData::haveParsedReplyHeaders()
925 {
926 Client::haveParsedReplyHeaders();
927
928 Ctx ctx = ctx_enter(entry->mem_obj->urlXXX());
929 HttpReply *rep = finalReply();
930
931 entry->timestampsSet();
932
933 /* Check if object is cacheable or not based on reply code */
934 debugs(11, 3, "HTTP CODE: " << rep->sline.status());
935
936 if (neighbors_do_private_keys)
937 httpMaybeRemovePublic(entry, rep->sline.status());
938
939 bool varyFailure = false;
940 if (rep->header.has(Http::HdrType::VARY)
941 #if X_ACCELERATOR_VARY
942 || rep->header.has(Http::HdrType::HDR_X_ACCELERATOR_VARY)
943 #endif
944 ) {
945 const char *vary = httpMakeVaryMark(request, rep);
946
947 if (!vary) {
948 entry->makePrivate();
949 if (!fwd->reforwardableStatus(rep->sline.status()))
950 EBIT_CLR(entry->flags, ENTRY_FWD_HDR_WAIT);
951 varyFailure = true;
952 } else {
953 entry->mem_obj->vary_headers = xstrdup(vary);
954 }
955 }
956
957 if (!varyFailure) {
958 /*
959 * If its not a reply that we will re-forward, then
960 * allow the client to get it.
961 */
962 if (!fwd->reforwardableStatus(rep->sline.status()))
963 EBIT_CLR(entry->flags, ENTRY_FWD_HDR_WAIT);
964
965 switch (cacheableReply()) {
966
967 case 1:
968 entry->makePublic();
969 break;
970
971 case 0:
972 entry->makePrivate();
973 break;
974
975 case -1:
976
977 #if USE_HTTP_VIOLATIONS
978 if (Config.negativeTtl > 0)
979 entry->cacheNegatively();
980 else
981 #endif
982 entry->makePrivate();
983 break;
984
985 default:
986 assert(0);
987 break;
988 }
989 }
990
991 if (!ignoreCacheControl) {
992 if (rep->cache_control) {
993 // We are required to revalidate on many conditions.
994 // For security reasons we do so even if storage was caused by refresh_pattern ignore-* option
995
996 // CC:must-revalidate or CC:proxy-revalidate
997 const bool ccMustRevalidate = (rep->cache_control->proxyRevalidate() || rep->cache_control->mustRevalidate());
998
999 // CC:no-cache (only if there are no parameters)
1000 const bool ccNoCacheNoParams = (rep->cache_control->hasNoCache() && rep->cache_control->noCache().size()==0);
1001
1002 // CC:s-maxage=N
1003 const bool ccSMaxAge = rep->cache_control->hasSMaxAge();
1004
1005 // CC:private (yes, these can sometimes be stored)
1006 const bool ccPrivate = rep->cache_control->hasPrivate();
1007
1008 if (ccMustRevalidate || ccNoCacheNoParams || ccSMaxAge || ccPrivate)
1009 EBIT_SET(entry->flags, ENTRY_REVALIDATE);
1010 }
1011 #if USE_HTTP_VIOLATIONS // response header Pragma::no-cache is undefined in HTTP
1012 else {
1013 // Expensive calculation. So only do it IF the CC: header is not present.
1014
1015 /* HACK: Pragma: no-cache in _replies_ is not documented in HTTP,
1016 * but servers like "Active Imaging Webcast/2.0" sure do use it */
1017 if (rep->header.has(Http::HdrType::PRAGMA) &&
1018 rep->header.hasListMember(Http::HdrType::PRAGMA,"no-cache",','))
1019 EBIT_SET(entry->flags, ENTRY_REVALIDATE);
1020 }
1021 #endif
1022 }
1023
1024 #if HEADERS_LOG
1025 headersLog(1, 0, request->method, rep);
1026
1027 #endif
1028
1029 ctx_exit(ctx);
1030 }
1031
1032 HttpStateData::ConnectionStatus
1033 HttpStateData::statusIfComplete() const
1034 {
1035 const HttpReply *rep = virginReply();
1036 /** \par
1037 * If the reply wants to close the connection, it takes precedence */
1038
1039 if (httpHeaderHasConnDir(&rep->header, "close"))
1040 return COMPLETE_NONPERSISTENT_MSG;
1041
1042 /** \par
1043 * If we didn't send a keep-alive request header, then this
1044 * can not be a persistent connection.
1045 */
1046 if (!flags.keepalive)
1047 return COMPLETE_NONPERSISTENT_MSG;
1048
1049 /** \par
1050 * If we haven't sent the whole request then this can not be a persistent
1051 * connection.
1052 */
1053 if (!flags.request_sent) {
1054 debugs(11, 2, "Request not yet fully sent " << request->method << ' ' << entry->url());
1055 return COMPLETE_NONPERSISTENT_MSG;
1056 }
1057
1058 /** \par
1059 * What does the reply have to say about keep-alive?
1060 */
1061 /**
1062 \bug XXX BUG?
1063 * If the origin server (HTTP/1.0) does not send a keep-alive
1064 * header, but keeps the connection open anyway, what happens?
1065 * We'll return here and http.c waits for an EOF before changing
1066 * store_status to STORE_OK. Combine this with ENTRY_FWD_HDR_WAIT
1067 * and an error status code, and we might have to wait until
1068 * the server times out the socket.
1069 */
1070 if (!rep->keep_alive)
1071 return COMPLETE_NONPERSISTENT_MSG;
1072
1073 return COMPLETE_PERSISTENT_MSG;
1074 }
1075
1076 HttpStateData::ConnectionStatus
1077 HttpStateData::persistentConnStatus() const
1078 {
1079 debugs(11, 3, HERE << serverConnection << " eof=" << eof);
1080 if (eof) // already reached EOF
1081 return COMPLETE_NONPERSISTENT_MSG;
1082
1083 /* If server fd is closing (but we have not been notified yet), stop Comm
1084 I/O to avoid assertions. TODO: Change Comm API to handle callers that
1085 want more I/O after async closing (usually initiated by others). */
1086 // XXX: add canReceive or s/canSend/canTalkToServer/
1087 if (!Comm::IsConnOpen(serverConnection))
1088 return COMPLETE_NONPERSISTENT_MSG;
1089
1090 /** \par
1091 * In chunked response we do not know the content length but we are absolutely
1092 * sure about the end of response, so we are calling the statusIfComplete to
1093 * decide if we can be persistant
1094 */
1095 if (lastChunk && flags.chunked)
1096 return statusIfComplete();
1097
1098 const HttpReply *vrep = virginReply();
1099 debugs(11, 5, "persistentConnStatus: content_length=" << vrep->content_length);
1100
1101 const int64_t clen = vrep->bodySize(request->method);
1102
1103 debugs(11, 5, "persistentConnStatus: clen=" << clen);
1104
1105 /* If the body size is unknown we must wait for EOF */
1106 if (clen < 0)
1107 return INCOMPLETE_MSG;
1108
1109 /** \par
1110 * If the body size is known, we must wait until we've gotten all of it. */
1111 if (clen > 0) {
1112 debugs(11,5, "payloadSeen=" << payloadSeen << " content_length=" << vrep->content_length);
1113
1114 if (payloadSeen < vrep->content_length)
1115 return INCOMPLETE_MSG;
1116
1117 if (payloadTruncated > 0) // already read more than needed
1118 return COMPLETE_NONPERSISTENT_MSG; // disable pconns
1119 }
1120
1121 /** \par
1122 * If there is no message body or we got it all, we can be persistent */
1123 return statusIfComplete();
1124 }
1125
1126 #if USE_DELAY_POOLS
1127 static void
1128 readDelayed(void *context, CommRead const &)
1129 {
1130 HttpStateData *state = static_cast<HttpStateData*>(context);
1131 state->flags.do_next_read = true;
1132 state->maybeReadVirginBody();
1133 }
1134 #endif
1135
1136 void
1137 HttpStateData::readReply(const CommIoCbParams &io)
1138 {
1139 Must(!flags.do_next_read); // XXX: should have been set false by mayReadVirginBody()
1140 flags.do_next_read = false;
1141
1142 debugs(11, 5, io.conn);
1143
1144 // Bail out early on Comm::ERR_CLOSING - close handlers will tidy up for us
1145 if (io.flag == Comm::ERR_CLOSING) {
1146 debugs(11, 3, "http socket closing");
1147 return;
1148 }
1149
1150 if (EBIT_TEST(entry->flags, ENTRY_ABORTED)) {
1151 abortTransaction("store entry aborted while reading reply");
1152 return;
1153 }
1154
1155 Must(Comm::IsConnOpen(serverConnection));
1156 Must(io.conn->fd == serverConnection->fd);
1157
1158 /*
1159 * Don't reset the timeout value here. The value should be
1160 * counting Config.Timeout.request and applies to the request
1161 * as a whole, not individual read() calls.
1162 * Plus, it breaks our lame *HalfClosed() detection
1163 */
1164
1165 Must(maybeMakeSpaceAvailable(true));
1166 CommIoCbParams rd(this); // will be expanded with ReadNow results
1167 rd.conn = io.conn;
1168 rd.size = entry->bytesWanted(Range<size_t>(0, inBuf.spaceSize()));
1169 #if USE_DELAY_POOLS
1170 if (rd.size < 1) {
1171 assert(entry->mem_obj);
1172
1173 /* read ahead limit */
1174 /* Perhaps these two calls should both live in MemObject */
1175 AsyncCall::Pointer nilCall;
1176 if (!entry->mem_obj->readAheadPolicyCanRead()) {
1177 entry->mem_obj->delayRead(DeferredRead(readDelayed, this, CommRead(io.conn, NULL, 0, nilCall)));
1178 return;
1179 }
1180
1181 /* delay id limit */
1182 entry->mem_obj->mostBytesAllowed().delayRead(DeferredRead(readDelayed, this, CommRead(io.conn, NULL, 0, nilCall)));
1183 return;
1184 }
1185 #endif
1186
1187 switch (Comm::ReadNow(rd, inBuf)) {
1188 case Comm::INPROGRESS:
1189 if (inBuf.isEmpty())
1190 debugs(33, 2, io.conn << ": no data to process, " << xstrerr(rd.xerrno));
1191 flags.do_next_read = true;
1192 maybeReadVirginBody();
1193 return;
1194
1195 case Comm::OK:
1196 {
1197 payloadSeen += rd.size;
1198 #if USE_DELAY_POOLS
1199 DelayId delayId = entry->mem_obj->mostBytesAllowed();
1200 delayId.bytesIn(rd.size);
1201 #endif
1202
1203 statCounter.server.all.kbytes_in += rd.size;
1204 statCounter.server.http.kbytes_in += rd.size;
1205 ++ IOStats.Http.reads;
1206
1207 int bin = 0;
1208 for (int clen = rd.size - 1; clen; ++bin)
1209 clen >>= 1;
1210
1211 ++ IOStats.Http.read_hist[bin];
1212
1213 // update peer response time stats (%<pt)
1214 const timeval &sent = request->hier.peer_http_request_sent;
1215 if (sent.tv_sec)
1216 tvSub(request->hier.peer_response_time, sent, current_time);
1217 else
1218 request->hier.peer_response_time.tv_sec = -1;
1219 }
1220
1221 /* Continue to process previously read data */
1222 break;
1223
1224 case Comm::ENDFILE: // close detected by 0-byte read
1225 eof = 1;
1226 flags.do_next_read = false;
1227
1228 /* Continue to process previously read data */
1229 break;
1230
1231 // case Comm::COMM_ERROR:
1232 default: // no other flags should ever occur
1233 debugs(11, 2, io.conn << ": read failure: " << xstrerr(rd.xerrno));
1234 ErrorState *err = new ErrorState(ERR_READ_ERROR, Http::scBadGateway, fwd->request);
1235 err->xerrno = rd.xerrno;
1236 fwd->fail(err);
1237 flags.do_next_read = false;
1238 closeServer();
1239 mustStop("HttpStateData::readReply");
1240 return;
1241 }
1242
1243 /* Process next response from buffer */
1244 processReply();
1245 }
1246
1247 /// processes the already read and buffered response data, possibly after
1248 /// waiting for asynchronous 1xx control message processing
1249 void
1250 HttpStateData::processReply()
1251 {
1252
1253 if (flags.handling1xx) { // we came back after handling a 1xx response
1254 debugs(11, 5, HERE << "done with 1xx handling");
1255 flags.handling1xx = false;
1256 Must(!flags.headers_parsed);
1257 }
1258
1259 if (!flags.headers_parsed) { // have not parsed headers yet?
1260 PROF_start(HttpStateData_processReplyHeader);
1261 processReplyHeader();
1262 PROF_stop(HttpStateData_processReplyHeader);
1263
1264 if (!continueAfterParsingHeader()) // parsing error or need more data
1265 return; // TODO: send errors to ICAP
1266
1267 adaptOrFinalizeReply(); // may write to, abort, or "close" the entry
1268 }
1269
1270 // kick more reads if needed and/or process the response body, if any
1271 PROF_start(HttpStateData_processReplyBody);
1272 processReplyBody(); // may call serverComplete()
1273 PROF_stop(HttpStateData_processReplyBody);
1274 }
1275
1276 /**
1277 \retval true if we can continue with processing the body or doing ICAP.
1278 */
1279 bool
1280 HttpStateData::continueAfterParsingHeader()
1281 {
1282 if (flags.handling1xx) {
1283 debugs(11, 5, HERE << "wait for 1xx handling");
1284 Must(!flags.headers_parsed);
1285 return false;
1286 }
1287
1288 if (!flags.headers_parsed && !eof) {
1289 debugs(11, 9, "needs more at " << inBuf.length());
1290 flags.do_next_read = true;
1291 /** \retval false If we have not finished parsing the headers and may get more data.
1292 * Schedules more reads to retrieve the missing data.
1293 */
1294 maybeReadVirginBody(); // schedules all kinds of reads; TODO: rename
1295 return false;
1296 }
1297
1298 /** If we are done with parsing, check for errors */
1299
1300 err_type error = ERR_NONE;
1301
1302 if (flags.headers_parsed) { // parsed headers, possibly with errors
1303 // check for header parsing errors
1304 if (HttpReply *vrep = virginReply()) {
1305 const Http::StatusCode s = vrep->sline.status();
1306 const AnyP::ProtocolVersion &v = vrep->sline.version;
1307 if (s == Http::scInvalidHeader && v != Http::ProtocolVersion(0,9)) {
1308 debugs(11, DBG_IMPORTANT, "WARNING: HTTP: Invalid Response: Bad header encountered from " << entry->url() << " AKA " << request->url);
1309 error = ERR_INVALID_RESP;
1310 } else if (s == Http::scHeaderTooLarge) {
1311 fwd->dontRetry(true);
1312 error = ERR_TOO_BIG;
1313 } else if (vrep->header.conflictingContentLength()) {
1314 fwd->dontRetry(true);
1315 error = ERR_INVALID_RESP;
1316 } else {
1317 return true; // done parsing, got reply, and no error
1318 }
1319 } else {
1320 // parsed headers but got no reply
1321 debugs(11, DBG_IMPORTANT, "WARNING: HTTP: Invalid Response: No reply at all for " << entry->url() << " AKA " << request->url);
1322 error = ERR_INVALID_RESP;
1323 }
1324 } else {
1325 assert(eof);
1326 if (inBuf.length()) {
1327 error = ERR_INVALID_RESP;
1328 debugs(11, DBG_IMPORTANT, "WARNING: HTTP: Invalid Response: Headers did not parse at all for " << entry->url() << " AKA " << request->url);
1329 } else {
1330 error = ERR_ZERO_SIZE_OBJECT;
1331 debugs(11, (request->flags.accelerated?DBG_IMPORTANT:2), "WARNING: HTTP: Invalid Response: No object data received for " << entry->url() << " AKA " << request->url);
1332 }
1333 }
1334
1335 assert(error != ERR_NONE);
1336 entry->reset();
1337 fwd->fail(new ErrorState(error, Http::scBadGateway, fwd->request));
1338 flags.do_next_read = false;
1339 closeServer();
1340 mustStop("HttpStateData::continueAfterParsingHeader");
1341 return false; // quit on error
1342 }
1343
1344 /** truncate what we read if we read too much so that writeReplyBody()
1345 writes no more than what we should have read */
1346 void
1347 HttpStateData::truncateVirginBody()
1348 {
1349 assert(flags.headers_parsed);
1350
1351 HttpReply *vrep = virginReply();
1352 int64_t clen = -1;
1353 if (!vrep->expectingBody(request->method, clen) || clen < 0)
1354 return; // no body or a body of unknown size, including chunked
1355
1356 if (payloadSeen - payloadTruncated <= clen)
1357 return; // we did not read too much or already took care of the extras
1358
1359 if (const int64_t extras = payloadSeen - payloadTruncated - clen) {
1360 // server sent more that the advertised content length
1361 debugs(11, 5, "payloadSeen=" << payloadSeen <<
1362 " clen=" << clen << '/' << vrep->content_length <<
1363 " trucated=" << payloadTruncated << '+' << extras);
1364
1365 inBuf.chop(0, inBuf.length() - extras);
1366 payloadTruncated += extras;
1367 }
1368 }
1369
1370 /**
1371 * Call this when there is data from the origin server
1372 * which should be sent to either StoreEntry, or to ICAP...
1373 */
1374 void
1375 HttpStateData::writeReplyBody()
1376 {
1377 truncateVirginBody(); // if needed
1378 const char *data = inBuf.rawContent();
1379 int len = inBuf.length();
1380 addVirginReplyBody(data, len);
1381 inBuf.consume(len);
1382 }
1383
1384 bool
1385 HttpStateData::decodeAndWriteReplyBody()
1386 {
1387 const char *data = NULL;
1388 int len;
1389 bool wasThereAnException = false;
1390 assert(flags.chunked);
1391 assert(httpChunkDecoder);
1392 SQUID_ENTER_THROWING_CODE();
1393 MemBuf decodedData;
1394 decodedData.init();
1395 httpChunkDecoder->setPayloadBuffer(&decodedData);
1396 const bool doneParsing = httpChunkDecoder->parse(inBuf);
1397 inBuf = httpChunkDecoder->remaining(); // sync buffers after parse
1398 len = decodedData.contentSize();
1399 data=decodedData.content();
1400 addVirginReplyBody(data, len);
1401 if (doneParsing) {
1402 lastChunk = 1;
1403 flags.do_next_read = false;
1404 }
1405 SQUID_EXIT_THROWING_CODE(wasThereAnException);
1406 return wasThereAnException;
1407 }
1408
1409 /**
1410 * processReplyBody has two purposes:
1411 * 1 - take the reply body data, if any, and put it into either
1412 * the StoreEntry, or give it over to ICAP.
1413 * 2 - see if we made it to the end of the response (persistent
1414 * connections and such)
1415 */
1416 void
1417 HttpStateData::processReplyBody()
1418 {
1419 Ip::Address client_addr;
1420 bool ispinned = false;
1421
1422 if (!flags.headers_parsed) {
1423 flags.do_next_read = true;
1424 maybeReadVirginBody();
1425 return;
1426 }
1427
1428 #if USE_ADAPTATION
1429 debugs(11,5, HERE << "adaptationAccessCheckPending=" << adaptationAccessCheckPending);
1430 if (adaptationAccessCheckPending)
1431 return;
1432
1433 #endif
1434
1435 /*
1436 * At this point the reply headers have been parsed and consumed.
1437 * That means header content has been removed from readBuf and
1438 * it contains only body data.
1439 */
1440 if (entry->isAccepting()) {
1441 if (flags.chunked) {
1442 if (!decodeAndWriteReplyBody()) {
1443 flags.do_next_read = false;
1444 serverComplete();
1445 return;
1446 }
1447 } else
1448 writeReplyBody();
1449 }
1450
1451 // storing/sending methods like earlier adaptOrFinalizeReply() or
1452 // above writeReplyBody() may release/abort the store entry.
1453 if (EBIT_TEST(entry->flags, ENTRY_ABORTED)) {
1454 // TODO: In some cases (e.g., 304), we should keep persistent conn open.
1455 // Detect end-of-reply (and, hence, pool our idle pconn) earlier (ASAP).
1456 abortTransaction("store entry aborted while storing reply");
1457 return;
1458 } else
1459 switch (persistentConnStatus()) {
1460 case INCOMPLETE_MSG: {
1461 debugs(11, 5, "processReplyBody: INCOMPLETE_MSG from " << serverConnection);
1462 /* Wait for more data or EOF condition */
1463 AsyncCall::Pointer nil;
1464 if (flags.keepalive_broken) {
1465 commSetConnTimeout(serverConnection, 10, nil);
1466 } else {
1467 commSetConnTimeout(serverConnection, Config.Timeout.read, nil);
1468 }
1469
1470 flags.do_next_read = true;
1471 }
1472 break;
1473
1474 case COMPLETE_PERSISTENT_MSG:
1475 debugs(11, 5, "processReplyBody: COMPLETE_PERSISTENT_MSG from " << serverConnection);
1476 /* yes we have to clear all these! */
1477 commUnsetConnTimeout(serverConnection);
1478 flags.do_next_read = false;
1479
1480 comm_remove_close_handler(serverConnection->fd, closeHandler);
1481 closeHandler = NULL;
1482 fwd->unregister(serverConnection);
1483
1484 if (request->flags.spoofClientIp)
1485 client_addr = request->client_addr;
1486
1487 if (request->flags.pinned) {
1488 ispinned = true;
1489 } else if (request->flags.connectionAuth && request->flags.authSent) {
1490 ispinned = true;
1491 }
1492
1493 if (ispinned && request->clientConnectionManager.valid()) {
1494 request->clientConnectionManager->pinConnection(serverConnection, request, _peer,
1495 (request->flags.connectionAuth));
1496 } else {
1497 fwd->pconnPush(serverConnection, request->url.host());
1498 }
1499
1500 serverConnection = NULL;
1501 serverComplete();
1502 return;
1503
1504 case COMPLETE_NONPERSISTENT_MSG:
1505 debugs(11, 5, "processReplyBody: COMPLETE_NONPERSISTENT_MSG from " << serverConnection);
1506 serverComplete();
1507 return;
1508 }
1509
1510 maybeReadVirginBody();
1511 }
1512
1513 bool
1514 HttpStateData::mayReadVirginReplyBody() const
1515 {
1516 // TODO: Be more precise here. For example, if/when reading trailer, we may
1517 // not be doneWithServer() yet, but we should return false. Similarly, we
1518 // could still be writing the request body after receiving the whole reply.
1519 return !doneWithServer();
1520 }
1521
1522 void
1523 HttpStateData::maybeReadVirginBody()
1524 {
1525 // too late to read
1526 if (!Comm::IsConnOpen(serverConnection) || fd_table[serverConnection->fd].closing())
1527 return;
1528
1529 if (!maybeMakeSpaceAvailable(false))
1530 return;
1531
1532 // XXX: get rid of the do_next_read flag
1533 // check for the proper reasons preventing read(2)
1534 if (!flags.do_next_read)
1535 return;
1536
1537 flags.do_next_read = false;
1538
1539 // must not already be waiting for read(2) ...
1540 assert(!Comm::MonitorsRead(serverConnection->fd));
1541
1542 // wait for read(2) to be possible.
1543 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
1544 AsyncCall::Pointer call = JobCallback(11, 5, Dialer, this, HttpStateData::readReply);
1545 Comm::Read(serverConnection, call);
1546 }
1547
1548 bool
1549 HttpStateData::maybeMakeSpaceAvailable(bool doGrow)
1550 {
1551 // how much we are allowed to buffer
1552 const int limitBuffer = (flags.headers_parsed ? Config.readAheadGap : Config.maxReplyHeaderSize);
1553
1554 if (limitBuffer < 0 || inBuf.length() >= (SBuf::size_type)limitBuffer) {
1555 // when buffer is at or over limit already
1556 debugs(11, 7, "wont read up to " << limitBuffer << ". buffer has (" << inBuf.length() << "/" << inBuf.spaceSize() << ") from " << serverConnection);
1557 debugs(11, DBG_DATA, "buffer has {" << inBuf << "}");
1558 // Process next response from buffer
1559 processReply();
1560 return false;
1561 }
1562
1563 // how much we want to read
1564 const size_t read_size = calcBufferSpaceToReserve(inBuf.spaceSize(), (limitBuffer - inBuf.length()));
1565
1566 if (!read_size) {
1567 debugs(11, 7, "wont read up to " << read_size << " into buffer (" << inBuf.length() << "/" << inBuf.spaceSize() << ") from " << serverConnection);
1568 return false;
1569 }
1570
1571 // just report whether we could grow or not, dont actually do it
1572 if (doGrow)
1573 return (read_size >= 2);
1574
1575 // we may need to grow the buffer
1576 inBuf.reserveSpace(read_size);
1577 debugs(11, 8, (!flags.do_next_read ? "wont" : "may") <<
1578 " read up to " << read_size << " bytes info buf(" << inBuf.length() << "/" << inBuf.spaceSize() <<
1579 ") from " << serverConnection);
1580
1581 return (inBuf.spaceSize() >= 2); // only read if there is 1+ bytes of space available
1582 }
1583
1584 /// called after writing the very last request byte (body, last-chunk, etc)
1585 void
1586 HttpStateData::wroteLast(const CommIoCbParams &io)
1587 {
1588 debugs(11, 5, HERE << serverConnection << ": size " << io.size << ": errflag " << io.flag << ".");
1589 #if URL_CHECKSUM_DEBUG
1590
1591 entry->mem_obj->checkUrlChecksum();
1592 #endif
1593
1594 if (io.size > 0) {
1595 fd_bytes(io.fd, io.size, FD_WRITE);
1596 statCounter.server.all.kbytes_out += io.size;
1597 statCounter.server.http.kbytes_out += io.size;
1598 }
1599
1600 if (io.flag == Comm::ERR_CLOSING)
1601 return;
1602
1603 if (io.flag) {
1604 ErrorState *err = new ErrorState(ERR_WRITE_ERROR, Http::scBadGateway, fwd->request);
1605 err->xerrno = io.xerrno;
1606 fwd->fail(err);
1607 closeServer();
1608 mustStop("HttpStateData::wroteLast");
1609 return;
1610 }
1611
1612 sendComplete();
1613 }
1614
1615 /// successfully wrote the entire request (including body, last-chunk, etc.)
1616 void
1617 HttpStateData::sendComplete()
1618 {
1619 /*
1620 * Set the read timeout here because it hasn't been set yet.
1621 * We only set the read timeout after the request has been
1622 * fully written to the peer. If we start the timeout
1623 * after connection establishment, then we are likely to hit
1624 * the timeout for POST/PUT requests that have very large
1625 * request bodies.
1626 */
1627 typedef CommCbMemFunT<HttpStateData, CommTimeoutCbParams> TimeoutDialer;
1628 AsyncCall::Pointer timeoutCall = JobCallback(11, 5,
1629 TimeoutDialer, this, HttpStateData::httpTimeout);
1630
1631 commSetConnTimeout(serverConnection, Config.Timeout.read, timeoutCall);
1632 flags.request_sent = true;
1633 request->hier.peer_http_request_sent = current_time;
1634 }
1635
1636 void
1637 HttpStateData::closeServer()
1638 {
1639 debugs(11,5, HERE << "closing HTTP server " << serverConnection << " this " << this);
1640
1641 if (Comm::IsConnOpen(serverConnection)) {
1642 fwd->unregister(serverConnection);
1643 comm_remove_close_handler(serverConnection->fd, closeHandler);
1644 closeHandler = NULL;
1645 serverConnection->close();
1646 }
1647 }
1648
1649 bool
1650 HttpStateData::doneWithServer() const
1651 {
1652 return !Comm::IsConnOpen(serverConnection);
1653 }
1654
1655 /*
1656 * Fixup authentication request headers for special cases
1657 */
1658 static void
1659 httpFixupAuthentication(HttpRequest * request, const HttpHeader * hdr_in, HttpHeader * hdr_out, const HttpStateFlags &flags)
1660 {
1661 Http::HdrType header = flags.originpeer ? Http::HdrType::AUTHORIZATION : Http::HdrType::PROXY_AUTHORIZATION;
1662
1663 /* Nothing to do unless we are forwarding to a peer */
1664 if (!request->flags.proxying)
1665 return;
1666
1667 /* Needs to be explicitly enabled */
1668 if (!request->peer_login)
1669 return;
1670
1671 /* Maybe already dealt with? */
1672 if (hdr_out->has(header))
1673 return;
1674
1675 /* Nothing to do here for PASSTHRU */
1676 if (strcmp(request->peer_login, "PASSTHRU") == 0)
1677 return;
1678
1679 /* PROXYPASS is a special case, single-signon to servers with the proxy password (basic only) */
1680 if (flags.originpeer && strcmp(request->peer_login, "PROXYPASS") == 0 && hdr_in->has(Http::HdrType::PROXY_AUTHORIZATION)) {
1681 const char *auth = hdr_in->getStr(Http::HdrType::PROXY_AUTHORIZATION);
1682
1683 if (auth && strncasecmp(auth, "basic ", 6) == 0) {
1684 hdr_out->putStr(header, auth);
1685 return;
1686 }
1687 }
1688
1689 uint8_t loginbuf[base64_encode_len(MAX_LOGIN_SZ)];
1690 size_t blen;
1691 struct base64_encode_ctx ctx;
1692 base64_encode_init(&ctx);
1693
1694 /* Special mode to pass the username to the upstream cache */
1695 if (*request->peer_login == '*') {
1696 const char *username = "-";
1697
1698 if (request->extacl_user.size())
1699 username = request->extacl_user.termedBuf();
1700 #if USE_AUTH
1701 else if (request->auth_user_request != NULL)
1702 username = request->auth_user_request->username();
1703 #endif
1704
1705 blen = base64_encode_update(&ctx, loginbuf, strlen(username), reinterpret_cast<const uint8_t*>(username));
1706 blen += base64_encode_update(&ctx, loginbuf+blen, strlen(request->peer_login +1), reinterpret_cast<const uint8_t*>(request->peer_login +1));
1707 blen += base64_encode_final(&ctx, loginbuf+blen);
1708 httpHeaderPutStrf(hdr_out, header, "Basic %.*s", (int)blen, loginbuf);
1709 return;
1710 }
1711
1712 /* external_acl provided credentials */
1713 if (request->extacl_user.size() && request->extacl_passwd.size() &&
1714 (strcmp(request->peer_login, "PASS") == 0 ||
1715 strcmp(request->peer_login, "PROXYPASS") == 0)) {
1716
1717 blen = base64_encode_update(&ctx, loginbuf, request->extacl_user.size(), reinterpret_cast<const uint8_t*>(request->extacl_user.rawBuf()));
1718 blen += base64_encode_update(&ctx, loginbuf+blen, 1, reinterpret_cast<const uint8_t*>(":"));
1719 blen += base64_encode_update(&ctx, loginbuf+blen, request->extacl_passwd.size(), reinterpret_cast<const uint8_t*>(request->extacl_passwd.rawBuf()));
1720 blen += base64_encode_final(&ctx, loginbuf+blen);
1721 httpHeaderPutStrf(hdr_out, header, "Basic %.*s", (int)blen, loginbuf);
1722 return;
1723 }
1724 // if no external user credentials are available to fake authentication with PASS acts like PASSTHRU
1725 if (strcmp(request->peer_login, "PASS") == 0)
1726 return;
1727
1728 /* Kerberos login to peer */
1729 #if HAVE_AUTH_MODULE_NEGOTIATE && HAVE_KRB5 && HAVE_GSSAPI
1730 if (strncmp(request->peer_login, "NEGOTIATE",strlen("NEGOTIATE")) == 0) {
1731 char *Token=NULL;
1732 char *PrincipalName=NULL,*p;
1733 int negotiate_flags = 0;
1734
1735 if ((p=strchr(request->peer_login,':')) != NULL ) {
1736 PrincipalName=++p;
1737 }
1738 if (request->flags.auth_no_keytab) {
1739 negotiate_flags |= PEER_PROXY_NEGOTIATE_NOKEYTAB;
1740 }
1741 Token = peer_proxy_negotiate_auth(PrincipalName, request->peer_host, negotiate_flags);
1742 if (Token) {
1743 httpHeaderPutStrf(hdr_out, header, "Negotiate %s",Token);
1744 }
1745 return;
1746 }
1747 #endif /* HAVE_KRB5 && HAVE_GSSAPI */
1748
1749 blen = base64_encode_update(&ctx, loginbuf, strlen(request->peer_login), reinterpret_cast<const uint8_t*>(request->peer_login));
1750 blen += base64_encode_final(&ctx, loginbuf+blen);
1751 httpHeaderPutStrf(hdr_out, header, "Basic %.*s", (int)blen, loginbuf);
1752 return;
1753 }
1754
1755 /*
1756 * build request headers and append them to a given MemBuf
1757 * used by buildRequestPrefix()
1758 * note: initialised the HttpHeader, the caller is responsible for Clean()-ing
1759 */
1760 void
1761 HttpStateData::httpBuildRequestHeader(HttpRequest * request,
1762 StoreEntry * entry,
1763 const AccessLogEntryPointer &al,
1764 HttpHeader * hdr_out,
1765 const HttpStateFlags &flags)
1766 {
1767 /* building buffer for complex strings */
1768 #define BBUF_SZ (MAX_URL+32)
1769 LOCAL_ARRAY(char, bbuf, BBUF_SZ);
1770 LOCAL_ARRAY(char, ntoabuf, MAX_IPSTRLEN);
1771 const HttpHeader *hdr_in = &request->header;
1772 const HttpHeaderEntry *e = NULL;
1773 HttpHeaderPos pos = HttpHeaderInitPos;
1774 assert (hdr_out->owner == hoRequest);
1775
1776 /* use our IMS header if the cached entry has Last-Modified time */
1777 if (request->lastmod > -1)
1778 hdr_out->putTime(Http::HdrType::IF_MODIFIED_SINCE, request->lastmod);
1779
1780 // Add our own If-None-Match field if the cached entry has a strong ETag.
1781 // copyOneHeaderFromClientsideRequestToUpstreamRequest() adds client ones.
1782 if (request->etag.size() > 0) {
1783 hdr_out->addEntry(new HttpHeaderEntry(Http::HdrType::IF_NONE_MATCH, NULL,
1784 request->etag.termedBuf()));
1785 }
1786
1787 bool we_do_ranges = decideIfWeDoRanges (request);
1788
1789 String strConnection (hdr_in->getList(Http::HdrType::CONNECTION));
1790
1791 while ((e = hdr_in->getEntry(&pos)))
1792 copyOneHeaderFromClientsideRequestToUpstreamRequest(e, strConnection, request, hdr_out, we_do_ranges, flags);
1793
1794 /* Abstraction break: We should interpret multipart/byterange responses
1795 * into offset-length data, and this works around our inability to do so.
1796 */
1797 if (!we_do_ranges && request->multipartRangeRequest()) {
1798 /* don't cache the result */
1799 request->flags.cachable = false;
1800 /* pretend it's not a range request */
1801 request->ignoreRange("want to request the whole object");
1802 request->flags.isRanged = false;
1803 }
1804
1805 /* append Via */
1806 if (Config.onoff.via) {
1807 String strVia;
1808 strVia = hdr_in->getList(Http::HdrType::VIA);
1809 snprintf(bbuf, BBUF_SZ, "%d.%d %s",
1810 request->http_ver.major,
1811 request->http_ver.minor, ThisCache);
1812 strListAdd(&strVia, bbuf, ',');
1813 hdr_out->putStr(Http::HdrType::VIA, strVia.termedBuf());
1814 strVia.clean();
1815 }
1816
1817 if (request->flags.accelerated) {
1818 /* Append Surrogate-Capabilities */
1819 String strSurrogate(hdr_in->getList(Http::HdrType::SURROGATE_CAPABILITY));
1820 #if USE_SQUID_ESI
1821 snprintf(bbuf, BBUF_SZ, "%s=\"Surrogate/1.0 ESI/1.0\"", Config.Accel.surrogate_id);
1822 #else
1823 snprintf(bbuf, BBUF_SZ, "%s=\"Surrogate/1.0\"", Config.Accel.surrogate_id);
1824 #endif
1825 strListAdd(&strSurrogate, bbuf, ',');
1826 hdr_out->putStr(Http::HdrType::SURROGATE_CAPABILITY, strSurrogate.termedBuf());
1827 }
1828
1829 /** \pre Handle X-Forwarded-For */
1830 if (strcmp(opt_forwarded_for, "delete") != 0) {
1831
1832 String strFwd = hdr_in->getList(Http::HdrType::X_FORWARDED_FOR);
1833
1834 // if we cannot double strFwd size, then it grew past 50% of the limit
1835 if (!strFwd.canGrowBy(strFwd.size())) {
1836 // There is probably a forwarding loop with Via detection disabled.
1837 // If we do nothing, String will assert on overflow soon.
1838 // TODO: Terminate all transactions with huge XFF?
1839 strFwd = "error";
1840
1841 static int warnedCount = 0;
1842 if (warnedCount++ < 100) {
1843 const SBuf url(entry ? SBuf(entry->url()) : request->effectiveRequestUri());
1844 debugs(11, DBG_IMPORTANT, "Warning: likely forwarding loop with " << url);
1845 }
1846 }
1847
1848 if (strcmp(opt_forwarded_for, "on") == 0) {
1849 /** If set to ON - append client IP or 'unknown'. */
1850 if ( request->client_addr.isNoAddr() )
1851 strListAdd(&strFwd, "unknown", ',');
1852 else
1853 strListAdd(&strFwd, request->client_addr.toStr(ntoabuf, MAX_IPSTRLEN), ',');
1854 } else if (strcmp(opt_forwarded_for, "off") == 0) {
1855 /** If set to OFF - append 'unknown'. */
1856 strListAdd(&strFwd, "unknown", ',');
1857 } else if (strcmp(opt_forwarded_for, "transparent") == 0) {
1858 /** If set to TRANSPARENT - pass through unchanged. */
1859 } else if (strcmp(opt_forwarded_for, "truncate") == 0) {
1860 /** If set to TRUNCATE - drop existing list and replace with client IP or 'unknown'. */
1861 if ( request->client_addr.isNoAddr() )
1862 strFwd = "unknown";
1863 else
1864 strFwd = request->client_addr.toStr(ntoabuf, MAX_IPSTRLEN);
1865 }
1866 if (strFwd.size() > 0)
1867 hdr_out->putStr(Http::HdrType::X_FORWARDED_FOR, strFwd.termedBuf());
1868 }
1869 /** If set to DELETE - do not copy through. */
1870
1871 /* append Host if not there already */
1872 if (!hdr_out->has(Http::HdrType::HOST)) {
1873 if (request->peer_domain) {
1874 hdr_out->putStr(Http::HdrType::HOST, request->peer_domain);
1875 } else {
1876 SBuf authority = request->url.authority();
1877 hdr_out->putStr(Http::HdrType::HOST, authority.c_str());
1878 }
1879 }
1880
1881 /* append Authorization if known in URL, not in header and going direct */
1882 if (!hdr_out->has(Http::HdrType::AUTHORIZATION)) {
1883 if (!request->flags.proxying && !request->url.userInfo().isEmpty()) {
1884 static uint8_t result[base64_encode_len(MAX_URL*2)]; // should be big enough for a single URI segment
1885 struct base64_encode_ctx ctx;
1886 base64_encode_init(&ctx);
1887 size_t blen = base64_encode_update(&ctx, result, request->url.userInfo().length(), reinterpret_cast<const uint8_t*>(request->url.userInfo().rawContent()));
1888 blen += base64_encode_final(&ctx, result+blen);
1889 result[blen] = '\0';
1890 if (blen)
1891 httpHeaderPutStrf(hdr_out, Http::HdrType::AUTHORIZATION, "Basic %.*s", (int)blen, result);
1892 }
1893 }
1894
1895 /* Fixup (Proxy-)Authorization special cases. Plain relaying dealt with above */
1896 httpFixupAuthentication(request, hdr_in, hdr_out, flags);
1897
1898 /* append Cache-Control, add max-age if not there already */
1899 {
1900 HttpHdrCc *cc = hdr_in->getCc();
1901
1902 if (!cc)
1903 cc = new HttpHdrCc();
1904
1905 #if 0 /* see bug 2330 */
1906 /* Set no-cache if determined needed but not found */
1907 if (request->flags.nocache)
1908 EBIT_SET(cc->mask, HttpHdrCcType::CC_NO_CACHE);
1909 #endif
1910
1911 /* Add max-age only without no-cache */
1912 if (!cc->hasMaxAge() && !cc->hasNoCache()) {
1913 // XXX: performance regression. c_str() reallocates
1914 SBuf tmp(request->effectiveRequestUri());
1915 cc->maxAge(getMaxAge(entry ? entry->url() : tmp.c_str()));
1916 }
1917
1918 /* Enforce sibling relations */
1919 if (flags.only_if_cached)
1920 cc->onlyIfCached(true);
1921
1922 hdr_out->putCc(cc);
1923
1924 delete cc;
1925 }
1926
1927 /* maybe append Connection: keep-alive */
1928 if (flags.keepalive) {
1929 hdr_out->putStr(Http::HdrType::CONNECTION, "keep-alive");
1930 }
1931
1932 /* append Front-End-Https */
1933 if (flags.front_end_https) {
1934 if (flags.front_end_https == 1 || request->url.getScheme() == AnyP::PROTO_HTTPS)
1935 hdr_out->putStr(Http::HdrType::FRONT_END_HTTPS, "On");
1936 }
1937
1938 if (flags.chunked_request) {
1939 // Do not just copy the original value so that if the client-side
1940 // starts decode other encodings, this code may remain valid.
1941 hdr_out->putStr(Http::HdrType::TRANSFER_ENCODING, "chunked");
1942 }
1943
1944 /* Now mangle the headers. */
1945 if (Config2.onoff.mangle_request_headers)
1946 httpHdrMangleList(hdr_out, request, ROR_REQUEST);
1947
1948 if (Config.request_header_add && !Config.request_header_add->empty())
1949 httpHdrAdd(hdr_out, request, al, *Config.request_header_add);
1950
1951 strConnection.clean();
1952 }
1953
1954 /**
1955 * Decides whether a particular header may be cloned from the received Clients request
1956 * to our outgoing fetch request.
1957 */
1958 void
1959 copyOneHeaderFromClientsideRequestToUpstreamRequest(const HttpHeaderEntry *e, const String strConnection, const HttpRequest * request, HttpHeader * hdr_out, const int we_do_ranges, const HttpStateFlags &flags)
1960 {
1961 debugs(11, 5, "httpBuildRequestHeader: " << e->name << ": " << e->value );
1962
1963 switch (e->id) {
1964
1965 /** \par RFC 2616 sect 13.5.1 - Hop-by-Hop headers which Squid should not pass on. */
1966
1967 case Http::HdrType::PROXY_AUTHORIZATION:
1968 /** \par Proxy-Authorization:
1969 * Only pass on proxy authentication to peers for which
1970 * authentication forwarding is explicitly enabled
1971 */
1972 if (!flags.originpeer && flags.proxying && request->peer_login &&
1973 (strcmp(request->peer_login, "PASS") == 0 ||
1974 strcmp(request->peer_login, "PROXYPASS") == 0 ||
1975 strcmp(request->peer_login, "PASSTHRU") == 0)) {
1976 hdr_out->addEntry(e->clone());
1977 }
1978 break;
1979
1980 /** \par RFC 2616 sect 13.5.1 - Hop-by-Hop headers which Squid does not pass on. */
1981
1982 case Http::HdrType::CONNECTION: /** \par Connection: */
1983 case Http::HdrType::TE: /** \par TE: */
1984 case Http::HdrType::KEEP_ALIVE: /** \par Keep-Alive: */
1985 case Http::HdrType::PROXY_AUTHENTICATE: /** \par Proxy-Authenticate: */
1986 case Http::HdrType::TRAILER: /** \par Trailer: */
1987 case Http::HdrType::UPGRADE: /** \par Upgrade: */
1988 case Http::HdrType::TRANSFER_ENCODING: /** \par Transfer-Encoding: */
1989 break;
1990
1991 /** \par OTHER headers I haven't bothered to track down yet. */
1992
1993 case Http::HdrType::AUTHORIZATION:
1994 /** \par WWW-Authorization:
1995 * Pass on WWW authentication */
1996
1997 if (!flags.originpeer) {
1998 hdr_out->addEntry(e->clone());
1999 } else {
2000 /** \note In accelerators, only forward authentication if enabled
2001 * (see also httpFixupAuthentication for special cases)
2002 */
2003 if (request->peer_login &&
2004 (strcmp(request->peer_login, "PASS") == 0 ||
2005 strcmp(request->peer_login, "PASSTHRU") == 0 ||
2006 strcmp(request->peer_login, "PROXYPASS") == 0)) {
2007 hdr_out->addEntry(e->clone());
2008 }
2009 }
2010
2011 break;
2012
2013 case Http::HdrType::HOST:
2014 /** \par Host:
2015 * Normally Squid rewrites the Host: header.
2016 * However, there is one case when we don't: If the URL
2017 * went through our redirector and the admin configured
2018 * 'redir_rewrites_host' to be off.
2019 */
2020 if (request->peer_domain)
2021 hdr_out->putStr(Http::HdrType::HOST, request->peer_domain);
2022 else if (request->flags.redirected && !Config.onoff.redir_rewrites_host)
2023 hdr_out->addEntry(e->clone());
2024 else {
2025 SBuf authority = request->url.authority();
2026 hdr_out->putStr(Http::HdrType::HOST, authority.c_str());
2027 }
2028
2029 break;
2030
2031 case Http::HdrType::IF_MODIFIED_SINCE:
2032 /** \par If-Modified-Since:
2033 * append unless we added our own,
2034 * but only if cache_miss_revalidate is enabled, or
2035 * the request is not cacheable, or
2036 * the request contains authentication credentials.
2037 * \note at most one client's If-Modified-Since header can pass through
2038 */
2039 // XXX: need to check and cleanup the auth case so cacheable auth requests get cached.
2040 if (hdr_out->has(Http::HdrType::IF_MODIFIED_SINCE))
2041 break;
2042 else if (Config.onoff.cache_miss_revalidate || !request->flags.cachable || request->flags.auth)
2043 hdr_out->addEntry(e->clone());
2044 break;
2045
2046 case Http::HdrType::IF_NONE_MATCH:
2047 /** \par If-None-Match:
2048 * append if the wildcard '*' special case value is present, or
2049 * cache_miss_revalidate is disabled, or
2050 * the request is not cacheable in this proxy, or
2051 * the request contains authentication credentials.
2052 * \note this header lists a set of responses for the server to elide sending. Squid added values are extending that set.
2053 */
2054 // XXX: need to check and cleanup the auth case so cacheable auth requests get cached.
2055 if (hdr_out->hasListMember(Http::HdrType::IF_MATCH, "*", ',') || Config.onoff.cache_miss_revalidate || !request->flags.cachable || request->flags.auth)
2056 hdr_out->addEntry(e->clone());
2057 break;
2058
2059 case Http::HdrType::MAX_FORWARDS:
2060 /** \par Max-Forwards:
2061 * pass only on TRACE or OPTIONS requests */
2062 if (request->method == Http::METHOD_TRACE || request->method == Http::METHOD_OPTIONS) {
2063 const int64_t hops = e->getInt64();
2064
2065 if (hops > 0)
2066 hdr_out->putInt64(Http::HdrType::MAX_FORWARDS, hops - 1);
2067 }
2068
2069 break;
2070
2071 case Http::HdrType::VIA:
2072 /** \par Via:
2073 * If Via is disabled then forward any received header as-is.
2074 * Otherwise leave for explicit updated addition later. */
2075
2076 if (!Config.onoff.via)
2077 hdr_out->addEntry(e->clone());
2078
2079 break;
2080
2081 case Http::HdrType::RANGE:
2082
2083 case Http::HdrType::IF_RANGE:
2084
2085 case Http::HdrType::REQUEST_RANGE:
2086 /** \par Range:, If-Range:, Request-Range:
2087 * Only pass if we accept ranges */
2088 if (!we_do_ranges)
2089 hdr_out->addEntry(e->clone());
2090
2091 break;
2092
2093 case Http::HdrType::PROXY_CONNECTION: // SHOULD ignore. But doing so breaks things.
2094 break;
2095
2096 case Http::HdrType::CONTENT_LENGTH:
2097 // pass through unless we chunk; also, keeping this away from default
2098 // prevents request smuggling via Connection: Content-Length tricks
2099 if (!flags.chunked_request)
2100 hdr_out->addEntry(e->clone());
2101 break;
2102
2103 case Http::HdrType::X_FORWARDED_FOR:
2104
2105 case Http::HdrType::CACHE_CONTROL:
2106 /** \par X-Forwarded-For:, Cache-Control:
2107 * handled specially by Squid, so leave off for now.
2108 * append these after the loop if needed */
2109 break;
2110
2111 case Http::HdrType::FRONT_END_HTTPS:
2112 /** \par Front-End-Https:
2113 * Pass thru only if peer is configured with front-end-https */
2114 if (!flags.front_end_https)
2115 hdr_out->addEntry(e->clone());
2116
2117 break;
2118
2119 default:
2120 /** \par default.
2121 * pass on all other header fields
2122 * which are NOT listed by the special Connection: header. */
2123
2124 if (strConnection.size()>0 && strListIsMember(&strConnection, e->name.termedBuf(), ',')) {
2125 debugs(11, 2, "'" << e->name << "' header cropped by Connection: definition");
2126 return;
2127 }
2128
2129 hdr_out->addEntry(e->clone());
2130 }
2131 }
2132
2133 bool
2134 HttpStateData::decideIfWeDoRanges (HttpRequest * request)
2135 {
2136 bool result = true;
2137 /* decide if we want to do Ranges ourselves
2138 * and fetch the whole object now)
2139 * We want to handle Ranges ourselves iff
2140 * - we can actually parse client Range specs
2141 * - the specs are expected to be simple enough (e.g. no out-of-order ranges)
2142 * - reply will be cachable
2143 * (If the reply will be uncachable we have to throw it away after
2144 * serving this request, so it is better to forward ranges to
2145 * the server and fetch only the requested content)
2146 */
2147
2148 int64_t roffLimit = request->getRangeOffsetLimit();
2149
2150 if (NULL == request->range || !request->flags.cachable
2151 || request->range->offsetLimitExceeded(roffLimit) || request->flags.connectionAuth)
2152 result = false;
2153
2154 debugs(11, 8, "decideIfWeDoRanges: range specs: " <<
2155 request->range << ", cachable: " <<
2156 request->flags.cachable << "; we_do_ranges: " << result);
2157
2158 return result;
2159 }
2160
2161 /* build request prefix and append it to a given MemBuf;
2162 * return the length of the prefix */
2163 mb_size_t
2164 HttpStateData::buildRequestPrefix(MemBuf * mb)
2165 {
2166 const int offset = mb->size;
2167 /* Uses a local httpver variable to print the HTTP label
2168 * since the HttpRequest may have an older version label.
2169 * XXX: This could create protocol bugs as the headers sent and
2170 * flow control should all be based on the HttpRequest version
2171 * not the one we are sending. Needs checking.
2172 */
2173 const AnyP::ProtocolVersion httpver = Http::ProtocolVersion();
2174 const SBuf url(_peer && !_peer->options.originserver ? request->effectiveRequestUri() : request->url.path());
2175 mb->appendf(SQUIDSBUFPH " " SQUIDSBUFPH " %s/%d.%d\r\n",
2176 SQUIDSBUFPRINT(request->method.image()),
2177 SQUIDSBUFPRINT(url),
2178 AnyP::ProtocolType_str[httpver.protocol],
2179 httpver.major,httpver.minor);
2180 /* build and pack headers */
2181 {
2182 HttpHeader hdr(hoRequest);
2183 httpBuildRequestHeader(request, entry, fwd->al, &hdr, flags);
2184
2185 if (request->flags.pinned && request->flags.connectionAuth)
2186 request->flags.authSent = true;
2187 else if (hdr.has(Http::HdrType::AUTHORIZATION))
2188 request->flags.authSent = true;
2189
2190 hdr.packInto(mb);
2191 hdr.clean();
2192 }
2193 /* append header terminator */
2194 mb->append(crlf, 2);
2195 return mb->size - offset;
2196 }
2197
2198 /* This will be called when connect completes. Write request. */
2199 bool
2200 HttpStateData::sendRequest()
2201 {
2202 MemBuf mb;
2203
2204 debugs(11, 5, HERE << serverConnection << ", request " << request << ", this " << this << ".");
2205
2206 if (!Comm::IsConnOpen(serverConnection)) {
2207 debugs(11,3, HERE << "cannot send request to closing " << serverConnection);
2208 assert(closeHandler != NULL);
2209 return false;
2210 }
2211
2212 typedef CommCbMemFunT<HttpStateData, CommTimeoutCbParams> TimeoutDialer;
2213 AsyncCall::Pointer timeoutCall = JobCallback(11, 5,
2214 TimeoutDialer, this, HttpStateData::httpTimeout);
2215 commSetConnTimeout(serverConnection, Config.Timeout.lifetime, timeoutCall);
2216 flags.do_next_read = true;
2217 maybeReadVirginBody();
2218
2219 if (request->body_pipe != NULL) {
2220 if (!startRequestBodyFlow()) // register to receive body data
2221 return false;
2222 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2223 requestSender = JobCallback(11,5,
2224 Dialer, this, HttpStateData::sentRequestBody);
2225
2226 Must(!flags.chunked_request);
2227 // use chunked encoding if we do not know the length
2228 if (request->content_length < 0)
2229 flags.chunked_request = true;
2230 } else {
2231 assert(!requestBodySource);
2232 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2233 requestSender = JobCallback(11,5,
2234 Dialer, this, HttpStateData::wroteLast);
2235 }
2236
2237 flags.originpeer = (_peer != NULL && _peer->options.originserver);
2238 flags.proxying = (_peer != NULL && !flags.originpeer);
2239
2240 /*
2241 * Is keep-alive okay for all request methods?
2242 */
2243 if (request->flags.mustKeepalive)
2244 flags.keepalive = true;
2245 else if (request->flags.pinned)
2246 flags.keepalive = request->persistent();
2247 else if (!Config.onoff.server_pconns)
2248 flags.keepalive = false;
2249 else if (_peer == NULL)
2250 flags.keepalive = true;
2251 else if (_peer->stats.n_keepalives_sent < 10)
2252 flags.keepalive = true;
2253 else if ((double) _peer->stats.n_keepalives_recv /
2254 (double) _peer->stats.n_keepalives_sent > 0.50)
2255 flags.keepalive = true;
2256
2257 if (_peer) {
2258 /*The old code here was
2259 if (neighborType(_peer, request->url) == PEER_SIBLING && ...
2260 which is equivalent to:
2261 if (neighborType(_peer, URL()) == PEER_SIBLING && ...
2262 or better:
2263 if (((_peer->type == PEER_MULTICAST && p->options.mcast_siblings) ||
2264 _peer->type == PEER_SIBLINGS ) && _peer->options.allow_miss)
2265 flags.only_if_cached = 1;
2266
2267 But I suppose it was a bug
2268 */
2269 if (neighborType(_peer, request->url) == PEER_SIBLING && !_peer->options.allow_miss)
2270 flags.only_if_cached = true;
2271
2272 flags.front_end_https = _peer->front_end_https;
2273 }
2274
2275 mb.init();
2276 request->peer_host=_peer?_peer->host:NULL;
2277 buildRequestPrefix(&mb);
2278
2279 debugs(11, 2, "HTTP Server " << serverConnection);
2280 debugs(11, 2, "HTTP Server REQUEST:\n---------\n" << mb.buf << "\n----------");
2281
2282 Comm::Write(serverConnection, &mb, requestSender);
2283 return true;
2284 }
2285
2286 bool
2287 HttpStateData::getMoreRequestBody(MemBuf &buf)
2288 {
2289 // parent's implementation can handle the no-encoding case
2290 if (!flags.chunked_request)
2291 return Client::getMoreRequestBody(buf);
2292
2293 MemBuf raw;
2294
2295 Must(requestBodySource != NULL);
2296 if (!requestBodySource->getMoreData(raw))
2297 return false; // no request body bytes to chunk yet
2298
2299 // optimization: pre-allocate buffer size that should be enough
2300 const mb_size_t rawDataSize = raw.contentSize();
2301 // we may need to send: hex-chunk-size CRLF raw-data CRLF last-chunk
2302 buf.init(16 + 2 + rawDataSize + 2 + 5, raw.max_capacity);
2303
2304 buf.appendf("%x\r\n", static_cast<unsigned int>(rawDataSize));
2305 buf.append(raw.content(), rawDataSize);
2306 buf.append("\r\n", 2);
2307
2308 Must(rawDataSize > 0); // we did not accidently created last-chunk above
2309
2310 // Do not send last-chunk unless we successfully received everything
2311 if (receivedWholeRequestBody) {
2312 Must(!flags.sentLastChunk);
2313 flags.sentLastChunk = true;
2314 buf.append("0\r\n\r\n", 5);
2315 }
2316
2317 return true;
2318 }
2319
2320 void
2321 httpStart(FwdState *fwd)
2322 {
2323 debugs(11, 3, fwd->request->method << ' ' << fwd->entry->url());
2324 AsyncJob::Start(new HttpStateData(fwd));
2325 }
2326
2327 void
2328 HttpStateData::start()
2329 {
2330 if (!sendRequest()) {
2331 debugs(11, 3, "httpStart: aborted");
2332 mustStop("HttpStateData::start failed");
2333 return;
2334 }
2335
2336 ++ statCounter.server.all.requests;
2337 ++ statCounter.server.http.requests;
2338
2339 /*
2340 * We used to set the read timeout here, but not any more.
2341 * Now its set in httpSendComplete() after the full request,
2342 * including request body, has been written to the server.
2343 */
2344 }
2345
2346 /// if broken posts are enabled for the request, try to fix and return true
2347 bool
2348 HttpStateData::finishingBrokenPost()
2349 {
2350 #if USE_HTTP_VIOLATIONS
2351 if (!Config.accessList.brokenPosts) {
2352 debugs(11, 5, HERE << "No brokenPosts list");
2353 return false;
2354 }
2355
2356 ACLFilledChecklist ch(Config.accessList.brokenPosts, originalRequest(), NULL);
2357 if (ch.fastCheck() != ACCESS_ALLOWED) {
2358 debugs(11, 5, HERE << "didn't match brokenPosts");
2359 return false;
2360 }
2361
2362 if (!Comm::IsConnOpen(serverConnection)) {
2363 debugs(11, 3, HERE << "ignoring broken POST for closed " << serverConnection);
2364 assert(closeHandler != NULL);
2365 return true; // prevent caller from proceeding as if nothing happened
2366 }
2367
2368 debugs(11, 3, "finishingBrokenPost: fixing broken POST");
2369 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2370 requestSender = JobCallback(11,5,
2371 Dialer, this, HttpStateData::wroteLast);
2372 Comm::Write(serverConnection, "\r\n", 2, requestSender, NULL);
2373 return true;
2374 #else
2375 return false;
2376 #endif /* USE_HTTP_VIOLATIONS */
2377 }
2378
2379 /// if needed, write last-chunk to end the request body and return true
2380 bool
2381 HttpStateData::finishingChunkedRequest()
2382 {
2383 if (flags.sentLastChunk) {
2384 debugs(11, 5, HERE << "already sent last-chunk");
2385 return false;
2386 }
2387
2388 Must(receivedWholeRequestBody); // or we should not be sending last-chunk
2389 flags.sentLastChunk = true;
2390
2391 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2392 requestSender = JobCallback(11,5, Dialer, this, HttpStateData::wroteLast);
2393 Comm::Write(serverConnection, "0\r\n\r\n", 5, requestSender, NULL);
2394 return true;
2395 }
2396
2397 void
2398 HttpStateData::doneSendingRequestBody()
2399 {
2400 Client::doneSendingRequestBody();
2401 debugs(11,5, HERE << serverConnection);
2402
2403 // do we need to write something after the last body byte?
2404 if (flags.chunked_request && finishingChunkedRequest())
2405 return;
2406 if (!flags.chunked_request && finishingBrokenPost())
2407 return;
2408
2409 sendComplete();
2410 }
2411
2412 // more origin request body data is available
2413 void
2414 HttpStateData::handleMoreRequestBodyAvailable()
2415 {
2416 if (eof || !Comm::IsConnOpen(serverConnection)) {
2417 // XXX: we should check this condition in other callbacks then!
2418 // TODO: Check whether this can actually happen: We should unsubscribe
2419 // as a body consumer when the above condition(s) are detected.
2420 debugs(11, DBG_IMPORTANT, HERE << "Transaction aborted while reading HTTP body");
2421 return;
2422 }
2423
2424 assert(requestBodySource != NULL);
2425
2426 if (requestBodySource->buf().hasContent()) {
2427 // XXX: why does not this trigger a debug message on every request?
2428
2429 if (flags.headers_parsed && !flags.abuse_detected) {
2430 flags.abuse_detected = true;
2431 debugs(11, DBG_IMPORTANT, "http handleMoreRequestBodyAvailable: Likely proxy abuse detected '" << request->client_addr << "' -> '" << entry->url() << "'" );
2432
2433 if (virginReply()->sline.status() == Http::scInvalidHeader) {
2434 closeServer();
2435 mustStop("HttpStateData::handleMoreRequestBodyAvailable");
2436 return;
2437 }
2438 }
2439 }
2440
2441 HttpStateData::handleMoreRequestBodyAvailable();
2442 }
2443
2444 // premature end of the request body
2445 void
2446 HttpStateData::handleRequestBodyProducerAborted()
2447 {
2448 Client::handleRequestBodyProducerAborted();
2449 if (entry->isEmpty()) {
2450 debugs(11, 3, "request body aborted: " << serverConnection);
2451 // We usually get here when ICAP REQMOD aborts during body processing.
2452 // We might also get here if client-side aborts, but then our response
2453 // should not matter because either client-side will provide its own or
2454 // there will be no response at all (e.g., if the the client has left).
2455 ErrorState *err = new ErrorState(ERR_ICAP_FAILURE, Http::scInternalServerError, fwd->request);
2456 err->detailError(ERR_DETAIL_SRV_REQMOD_REQ_BODY);
2457 fwd->fail(err);
2458 }
2459
2460 abortTransaction("request body producer aborted");
2461 }
2462
2463 // called when we wrote request headers(!) or a part of the body
2464 void
2465 HttpStateData::sentRequestBody(const CommIoCbParams &io)
2466 {
2467 if (io.size > 0)
2468 statCounter.server.http.kbytes_out += io.size;
2469
2470 Client::sentRequestBody(io);
2471 }
2472
2473 void
2474 HttpStateData::abortAll(const char *reason)
2475 {
2476 debugs(11,5, HERE << "aborting transaction for " << reason <<
2477 "; " << serverConnection << ", this " << this);
2478 mustStop(reason);
2479 }
2480