]> git.ipfire.org Git - thirdparty/squid.git/blob - src/http.cc
Collapse internal revalidation requests (SMP-unaware caches), again.
[thirdparty/squid.git] / src / http.cc
1 /*
2 * Copyright (C) 1996-2017 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 "log/access_log.h"
45 #include "MemBuf.h"
46 #include "MemObject.h"
47 #include "neighbors.h"
48 #include "peer_proxy_negotiate_auth.h"
49 #include "profiler/Profiler.h"
50 #include "refresh.h"
51 #include "RefreshPattern.h"
52 #include "rfc1738.h"
53 #include "SquidConfig.h"
54 #include "SquidTime.h"
55 #include "StatCounters.h"
56 #include "Store.h"
57 #include "StrList.h"
58 #include "tools.h"
59 #include "URL.h"
60 #include "util.h"
61
62 #if USE_AUTH
63 #include "auth/UserRequest.h"
64 #endif
65 #if USE_DELAY_POOLS
66 #include "DelayPools.h"
67 #endif
68
69 #define SQUID_ENTER_THROWING_CODE() try {
70 #define SQUID_EXIT_THROWING_CODE(status) \
71 status = true; \
72 } \
73 catch (const std::exception &e) { \
74 debugs (11, 1, "Exception error:" << e.what()); \
75 status = false; \
76 }
77
78 CBDATA_CLASS_INIT(HttpStateData);
79
80 static const char *const crlf = "\r\n";
81
82 static void httpMaybeRemovePublic(StoreEntry *, Http::StatusCode);
83 static void copyOneHeaderFromClientsideRequestToUpstreamRequest(const HttpHeaderEntry *e, const String strConnection, const HttpRequest * request,
84 HttpHeader * hdr_out, const int we_do_ranges, const Http::StateFlags &);
85
86 HttpStateData::HttpStateData(FwdState *theFwdState) :
87 AsyncJob("HttpStateData"),
88 Client(theFwdState),
89 lastChunk(0),
90 httpChunkDecoder(NULL),
91 payloadSeen(0),
92 payloadTruncated(0),
93 sawDateGoBack(false)
94 {
95 debugs(11,5,HERE << "HttpStateData " << this << " created");
96 ignoreCacheControl = false;
97 surrogateNoStore = false;
98 serverConnection = fwd->serverConnection();
99
100 // reset peer response time stats for %<pt
101 request->hier.peer_http_request_sent.tv_sec = 0;
102 request->hier.peer_http_request_sent.tv_usec = 0;
103
104 if (fwd->serverConnection() != NULL)
105 _peer = cbdataReference(fwd->serverConnection()->getPeer()); /* might be NULL */
106
107 if (_peer) {
108 request->flags.proxying = true;
109 /*
110 * This NEIGHBOR_PROXY_ONLY check probably shouldn't be here.
111 * We might end up getting the object from somewhere else if,
112 * for example, the request to this neighbor fails.
113 */
114 if (_peer->options.proxy_only)
115 entry->releaseRequest();
116
117 #if USE_DELAY_POOLS
118 entry->setNoDelay(_peer->options.no_delay);
119 #endif
120 }
121
122 /*
123 * register the handler to free HTTP state data when the FD closes
124 */
125 typedef CommCbMemFunT<HttpStateData, CommCloseCbParams> Dialer;
126 closeHandler = JobCallback(9, 5, Dialer, this, HttpStateData::httpStateConnClosed);
127 comm_add_close_handler(serverConnection->fd, closeHandler);
128 }
129
130 HttpStateData::~HttpStateData()
131 {
132 /*
133 * don't forget that ~Client() gets called automatically
134 */
135
136 if (httpChunkDecoder)
137 delete httpChunkDecoder;
138
139 cbdataReferenceDone(_peer);
140
141 debugs(11,5, HERE << "HttpStateData " << this << " destroyed; " << serverConnection);
142 }
143
144 const Comm::ConnectionPointer &
145 HttpStateData::dataConnection() const
146 {
147 return serverConnection;
148 }
149
150 void
151 HttpStateData::httpStateConnClosed(const CommCloseCbParams &params)
152 {
153 debugs(11, 5, "httpStateFree: FD " << params.fd << ", httpState=" << params.data);
154 doneWithFwd = "httpStateConnClosed()"; // assume FwdState is monitoring too
155 mustStop("HttpStateData::httpStateConnClosed");
156 }
157
158 void
159 HttpStateData::httpTimeout(const CommTimeoutCbParams &)
160 {
161 debugs(11, 4, serverConnection << ": '" << entry->url() << "'");
162
163 if (entry->store_status == STORE_PENDING) {
164 fwd->fail(new ErrorState(ERR_READ_TIMEOUT, Http::scGatewayTimeout, fwd->request));
165 }
166
167 closeServer();
168 mustStop("HttpStateData::httpTimeout");
169 }
170
171 static StoreEntry *
172 findPreviouslyCachedEntry(StoreEntry *newEntry) {
173 assert(newEntry->mem_obj);
174 return newEntry->mem_obj->request ?
175 storeGetPublicByRequest(newEntry->mem_obj->request.getRaw()) :
176 storeGetPublic(newEntry->mem_obj->storeId(), newEntry->mem_obj->method);
177 }
178
179 /// Remove an existing public store entry if the incoming response (to be
180 /// stored in a currently private entry) is going to invalidate it.
181 static void
182 httpMaybeRemovePublic(StoreEntry * e, Http::StatusCode status)
183 {
184 int remove = 0;
185 int forbidden = 0;
186
187 // If the incoming response already goes into a public entry, then there is
188 // nothing to remove. This protects ready-for-collapsing entries as well.
189 if (!EBIT_TEST(e->flags, KEY_PRIVATE))
190 return;
191
192 // If the new/incoming response cannot be stored, then it does not
193 // compete with the old stored response for the public key, and the
194 // old stored response should be left as is.
195 if (e->mem_obj->request && !e->mem_obj->request->flags.cachable)
196 return;
197
198 switch (status) {
199
200 case Http::scOkay:
201
202 case Http::scNonAuthoritativeInformation:
203
204 case Http::scMultipleChoices:
205
206 case Http::scMovedPermanently:
207
208 case Http::scFound:
209
210 case Http::scSeeOther:
211
212 case Http::scGone:
213
214 case Http::scNotFound:
215 remove = 1;
216
217 break;
218
219 case Http::scForbidden:
220
221 case Http::scMethodNotAllowed:
222 forbidden = 1;
223
224 break;
225
226 #if WORK_IN_PROGRESS
227
228 case Http::scUnauthorized:
229 forbidden = 1;
230
231 break;
232
233 #endif
234
235 default:
236 #if QUESTIONABLE
237 /*
238 * Any 2xx response should eject previously cached entities...
239 */
240
241 if (status >= 200 && status < 300)
242 remove = 1;
243
244 #endif
245
246 break;
247 }
248
249 if (!remove && !forbidden)
250 return;
251
252 StoreEntry *pe = findPreviouslyCachedEntry(e);
253
254 if (pe != NULL) {
255 assert(e != pe);
256 #if USE_HTCP
257 neighborsHtcpClear(e, nullptr, e->mem_obj->request.getRaw(), e->mem_obj->method, HTCP_CLR_INVALIDATION);
258 #endif
259 pe->release();
260 }
261
262 /** \par
263 * Also remove any cached HEAD response in case the object has
264 * changed.
265 */
266 if (e->mem_obj->request)
267 pe = storeGetPublicByRequestMethod(e->mem_obj->request.getRaw(), Http::METHOD_HEAD);
268 else
269 pe = storeGetPublic(e->mem_obj->storeId(), Http::METHOD_HEAD);
270
271 if (pe != NULL) {
272 assert(e != pe);
273 #if USE_HTCP
274 neighborsHtcpClear(e, nullptr, e->mem_obj->request.getRaw(), HttpRequestMethod(Http::METHOD_HEAD), HTCP_CLR_INVALIDATION);
275 #endif
276 pe->release();
277 }
278 }
279
280 void
281 HttpStateData::processSurrogateControl(HttpReply *reply)
282 {
283 if (request->flags.accelerated && reply->surrogate_control) {
284 HttpHdrScTarget *sctusable = reply->surrogate_control->getMergedTarget(Config.Accel.surrogate_id);
285
286 if (sctusable) {
287 if (sctusable->hasNoStore() ||
288 (Config.onoff.surrogate_is_remote
289 && sctusable->noStoreRemote())) {
290 surrogateNoStore = true;
291 // Be conservative for now and make it non-shareable because
292 // there is no enough information here to make the decision.
293 entry->makePrivate(false);
294 }
295
296 /* The HttpHeader logic cannot tell if the header it's parsing is a reply to an
297 * accelerated request or not...
298 * Still, this is an abstraction breach. - RC
299 */
300 if (sctusable->hasMaxAge()) {
301 if (sctusable->maxAge() < sctusable->maxStale())
302 reply->expires = reply->date + sctusable->maxAge();
303 else
304 reply->expires = reply->date + sctusable->maxStale();
305
306 /* And update the timestamps */
307 entry->timestampsSet();
308 }
309
310 /* We ignore cache-control directives as per the Surrogate specification */
311 ignoreCacheControl = true;
312
313 delete sctusable;
314 }
315 }
316 }
317
318 HttpStateData::ReuseDecision::Answers
319 HttpStateData::reusableReply(HttpStateData::ReuseDecision &decision)
320 {
321 HttpReply const *rep = finalReply();
322 HttpHeader const *hdr = &rep->header;
323 const char *v;
324 #if USE_HTTP_VIOLATIONS
325
326 const RefreshPattern *R = NULL;
327
328 /* This strange looking define first looks up the refresh pattern
329 * and then checks if the specified flag is set. The main purpose
330 * of this is to simplify the refresh pattern lookup and USE_HTTP_VIOLATIONS
331 * condition
332 */
333 #define REFRESH_OVERRIDE(flag) \
334 ((R = (R ? R : refreshLimits(entry->mem_obj->storeId()))) , \
335 (R && R->flags.flag))
336 #else
337 #define REFRESH_OVERRIDE(flag) 0
338 #endif
339
340 if (EBIT_TEST(entry->flags, RELEASE_REQUEST))
341 return decision.make(ReuseDecision::reuseNot, "the entry has been released");
342
343 // RFC 7234 section 4: a cache MUST use the most recent response
344 // (as determined by the Date header field)
345 // TODO: whether such responses could be shareable?
346 if (sawDateGoBack)
347 return decision.make(ReuseDecision::reuseNot, "the response has an older date header");
348
349 // Check for Surrogate/1.0 protocol conditions
350 // NP: reverse-proxy traffic our parent server has instructed us never to cache
351 if (surrogateNoStore)
352 return decision.make(ReuseDecision::reuseNot, "Surrogate-Control:no-store");
353
354 // RFC 2616: HTTP/1.1 Cache-Control conditions
355 if (!ignoreCacheControl) {
356 // XXX: check to see if the request headers alone were enough to prevent caching earlier
357 // (ie no-store request header) no need to check those all again here if so.
358 // for now we are not reliably doing that so we waste CPU re-checking request CC
359
360 // RFC 2616 section 14.9.2 - MUST NOT cache any response with request CC:no-store
361 if (request && request->cache_control && request->cache_control->hasNoStore() &&
362 !REFRESH_OVERRIDE(ignore_no_store))
363 return decision.make(ReuseDecision::reuseNot,
364 "client request Cache-Control:no-store");
365
366 // NP: request CC:no-cache only means cache READ is forbidden. STORE is permitted.
367 if (rep->cache_control && rep->cache_control->hasNoCacheWithParameters()) {
368 /* TODO: we are allowed to cache when no-cache= has parameters.
369 * Provided we strip away any of the listed headers unless they are revalidated
370 * successfully (ie, must revalidate AND these headers are prohibited on stale replies).
371 * That is a bit tricky for squid right now so we avoid caching entirely.
372 */
373 return decision.make(ReuseDecision::reuseNot,
374 "server reply Cache-Control:no-cache has parameters");
375 }
376
377 // NP: request CC:private is undefined. We ignore.
378 // NP: other request CC flags are limiters on HIT/MISS. We don't care about here.
379
380 // RFC 2616 section 14.9.2 - MUST NOT cache any response with CC:no-store
381 if (rep->cache_control && rep->cache_control->hasNoStore() &&
382 !REFRESH_OVERRIDE(ignore_no_store))
383 return decision.make(ReuseDecision::reuseNot,
384 "server reply Cache-Control:no-store");
385
386 // RFC 2616 section 14.9.1 - MUST NOT cache any response with CC:private in a shared cache like Squid.
387 // CC:private overrides CC:public when both are present in a response.
388 // TODO: add a shared/private cache configuration possibility.
389 if (rep->cache_control &&
390 rep->cache_control->hasPrivate() &&
391 !REFRESH_OVERRIDE(ignore_private)) {
392 /* TODO: we are allowed to cache when private= has parameters.
393 * Provided we strip away any of the listed headers unless they are revalidated
394 * successfully (ie, must revalidate AND these headers are prohibited on stale replies).
395 * That is a bit tricky for squid right now so we avoid caching entirely.
396 */
397 return decision.make(ReuseDecision::reuseNot,
398 "server reply Cache-Control:private");
399 }
400 }
401
402 // RFC 2068, sec 14.9.4 - MUST NOT cache any response with Authentication UNLESS certain CC controls are present
403 // allow HTTP violations to IGNORE those controls (ie re-block caching Auth)
404 if (request && (request->flags.auth || request->flags.authSent)) {
405 if (!rep->cache_control)
406 return decision.make(ReuseDecision::reuseNot,
407 "authenticated and server reply missing Cache-Control");
408
409 if (ignoreCacheControl)
410 return decision.make(ReuseDecision::reuseNot,
411 "authenticated and ignoring Cache-Control");
412
413 bool mayStore = false;
414 // HTTPbis pt6 section 3.2: a response CC:public is present
415 if (rep->cache_control->hasPublic()) {
416 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:public");
417 mayStore = true;
418
419 // HTTPbis pt6 section 3.2: a response CC:must-revalidate is present
420 } else if (rep->cache_control->hasMustRevalidate()) {
421 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:must-revalidate");
422 mayStore = true;
423
424 #if USE_HTTP_VIOLATIONS
425 // NP: given the must-revalidate exception we should also be able to exempt no-cache.
426 // HTTPbis WG verdict on this is that it is omitted from the spec due to being 'unexpected' by
427 // some. The caching+revalidate is not exactly unsafe though with Squids interpretation of no-cache
428 // (without parameters) as equivalent to must-revalidate in the reply.
429 } else if (rep->cache_control->hasNoCacheWithoutParameters()) {
430 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:no-cache (equivalent to must-revalidate)");
431 mayStore = true;
432 #endif
433
434 // HTTPbis pt6 section 3.2: a response CC:s-maxage is present
435 } else if (rep->cache_control->hasSMaxAge()) {
436 debugs(22, 3, HERE << "Authenticated but server reply Cache-Control:s-maxage");
437 mayStore = true;
438 }
439
440 if (!mayStore)
441 return decision.make(ReuseDecision::reuseNot, "authenticated transaction");
442
443 // NP: response CC:no-cache is equivalent to CC:must-revalidate,max-age=0. We MAY cache, and do so.
444 // NP: other request CC flags are limiters on HIT/MISS/REFRESH. We don't care about here.
445 }
446
447 /* HACK: The "multipart/x-mixed-replace" content type is used for
448 * continuous push replies. These are generally dynamic and
449 * probably should not be cachable
450 */
451 if ((v = hdr->getStr(Http::HdrType::CONTENT_TYPE)))
452 if (!strncasecmp(v, "multipart/x-mixed-replace", 25))
453 return decision.make(ReuseDecision::reuseNot, "Content-Type:multipart/x-mixed-replace");
454
455 // TODO: if possible, provide more specific message for each status code
456 static const char *shareableError = "shareable error status code";
457 static const char *nonShareableError = "non-shareable error status code";
458 ReuseDecision::Answers statusAnswer = ReuseDecision::reuseNot;
459 const char *statusReason = nonShareableError;
460
461 switch (rep->sline.status()) {
462
463 /* There are several situations when a non-cacheable response may be
464 * still shareable (e.g., among collapsed clients). We assume that these
465 * are 3xx and 5xx responses, indicating server problems and some of
466 * 4xx responses, common for all clients with a given cache key (e.g.,
467 * 404 Not Found or 414 URI Too Long). On the other hand, we should not
468 * share non-cacheable client-specific errors, such as 400 Bad Request
469 * or 406 Not Acceptable.
470 */
471
472 /* Responses that are cacheable */
473
474 case Http::scOkay:
475
476 case Http::scNonAuthoritativeInformation:
477
478 case Http::scMultipleChoices:
479
480 case Http::scMovedPermanently:
481 case Http::scPermanentRedirect:
482
483 case Http::scGone:
484 /*
485 * Don't cache objects that need to be refreshed on next request,
486 * unless we know how to refresh it.
487 */
488
489 if (refreshIsCachable(entry) || REFRESH_OVERRIDE(store_stale))
490 decision.make(ReuseDecision::cachePositively, "refresh check returned cacheable");
491 else
492 decision.make(ReuseDecision::doNotCacheButShare, "refresh check returned non-cacheable");
493 break;
494
495 /* Responses that only are cacheable if the server says so */
496
497 case Http::scFound:
498 case Http::scTemporaryRedirect:
499 if (rep->date <= 0)
500 decision.make(ReuseDecision::doNotCacheButShare, "Date is missing/invalid");
501 if (rep->expires > rep->date)
502 decision.make(ReuseDecision::cachePositively, "Expires > Date");
503 else
504 decision.make(ReuseDecision::doNotCacheButShare, "Expires <= Date");
505 break;
506
507 /* Some responses can be negatively cached */
508 case Http::scNoContent:
509 case Http::scUseProxy:
510 case Http::scForbidden:
511 case Http::scNotFound:
512 case Http::scMethodNotAllowed:
513 case Http::scUriTooLong:
514 case Http::scInternalServerError:
515 case Http::scNotImplemented:
516 case Http::scBadGateway:
517 case Http::scServiceUnavailable:
518 case Http::scGatewayTimeout:
519 case Http::scMisdirectedRequest:
520
521 statusAnswer = ReuseDecision::doNotCacheButShare;
522 statusReason = shareableError;
523 case Http::scBadRequest:
524
525 #if USE_HTTP_VIOLATIONS
526 if (Config.negativeTtl > 0)
527 decision.make(ReuseDecision::cacheNegatively, "Config.negativeTtl > 0");
528 else
529 #endif
530 decision.make(statusAnswer, statusReason);
531 break;
532
533 /* these responses can never be cached, some
534 of them can be shared though */
535 case Http::scSeeOther:
536 case Http::scNotModified:
537 case Http::scUnauthorized:
538 case Http::scProxyAuthenticationRequired:
539 case Http::scPaymentRequired:
540 case Http::scInsufficientStorage:
541 // TODO: use more specific reason for non-error status codes
542 decision.make(ReuseDecision::doNotCacheButShare, shareableError);
543 break;
544
545 case Http::scPartialContent: /* Not yet supported. TODO: make shareable for suitable ranges */
546 case Http::scNotAcceptable:
547 case Http::scRequestTimeout: // TODO: is this shareable?
548 case Http::scConflict: // TODO: is this shareable?
549 case Http::scLengthRequired:
550 case Http::scPreconditionFailed:
551 case Http::scPayloadTooLarge:
552 case Http::scUnsupportedMediaType:
553 case Http::scUnprocessableEntity:
554 case Http::scLocked: // TODO: is this shareable?
555 case Http::scFailedDependency:
556 case Http::scRequestedRangeNotSatisfied:
557 case Http::scExpectationFailed:
558 case Http::scInvalidHeader: /* Squid header parsing error */
559 case Http::scHeaderTooLarge:
560 decision.make(ReuseDecision::reuseNot, nonShareableError);
561 break;
562
563 default:
564 /* RFC 2616 section 6.1.1: an unrecognized response MUST NOT be cached. */
565 decision.make(ReuseDecision::reuseNot, "unknown status code");
566 break;
567 }
568
569 return decision.answer;
570 }
571
572 /// assemble a variant key (vary-mark) from the given Vary header and HTTP request
573 static void
574 assembleVaryKey(String &vary, SBuf &vstr, const HttpRequest &request)
575 {
576 static const SBuf asterisk("*");
577 const char *pos = nullptr;
578 const char *item = nullptr;
579 int ilen = 0;
580
581 while (strListGetItem(&vary, ',', &item, &ilen, &pos)) {
582 SBuf name(item, ilen);
583 if (name == asterisk) {
584 vstr = asterisk;
585 break;
586 }
587 name.toLower();
588 if (!vstr.isEmpty())
589 vstr.append(", ", 2);
590 vstr.append(name);
591 String hdr(request.header.getByName(name));
592 const char *value = hdr.termedBuf();
593 if (value) {
594 value = rfc1738_escape_part(value);
595 vstr.append("=\"", 2);
596 vstr.append(value);
597 vstr.append("\"", 1);
598 }
599
600 hdr.clean();
601 }
602 }
603
604 /*
605 * For Vary, store the relevant request headers as
606 * virtual headers in the reply
607 * Returns an empty SBuf if the variance cannot be stored
608 */
609 SBuf
610 httpMakeVaryMark(HttpRequest * request, HttpReply const * reply)
611 {
612 SBuf vstr;
613 String vary;
614
615 vary = reply->header.getList(Http::HdrType::VARY);
616 assembleVaryKey(vary, vstr, *request);
617
618 #if X_ACCELERATOR_VARY
619 vary.clean();
620 vary = reply->header.getList(Http::HdrType::HDR_X_ACCELERATOR_VARY);
621 assembleVaryKey(vary, vstr, *request);
622 #endif
623
624 debugs(11, 3, vstr);
625 return vstr;
626 }
627
628 void
629 HttpStateData::keepaliveAccounting(HttpReply *reply)
630 {
631 if (flags.keepalive)
632 if (_peer)
633 ++ _peer->stats.n_keepalives_sent;
634
635 if (reply->keep_alive) {
636 if (_peer)
637 ++ _peer->stats.n_keepalives_recv;
638
639 if (Config.onoff.detect_broken_server_pconns
640 && reply->bodySize(request->method) == -1 && !flags.chunked) {
641 debugs(11, DBG_IMPORTANT, "keepaliveAccounting: Impossible keep-alive header from '" << entry->url() << "'" );
642 // debugs(11, 2, "GOT HTTP REPLY HDR:\n---------\n" << readBuf->content() << "\n----------" );
643 flags.keepalive_broken = true;
644 }
645 }
646 }
647
648 void
649 HttpStateData::checkDateSkew(HttpReply *reply)
650 {
651 if (reply->date > -1 && !_peer) {
652 int skew = abs((int)(reply->date - squid_curtime));
653
654 if (skew > 86400)
655 debugs(11, 3, "" << request->url.host() << "'s clock is skewed by " << skew << " seconds!");
656 }
657 }
658
659 /**
660 * This creates the error page itself.. its likely
661 * that the forward ported reply header max size patch
662 * generates non http conformant error pages - in which
663 * case the errors where should be 'BAD_GATEWAY' etc
664 */
665 void
666 HttpStateData::processReplyHeader()
667 {
668 /** Creates a blank header. If this routine is made incremental, this will not do */
669
670 /* NP: all exit points to this function MUST call ctx_exit(ctx) */
671 Ctx ctx = ctx_enter(entry->mem_obj->urlXXX());
672
673 debugs(11, 3, "processReplyHeader: key '" << entry->getMD5Text() << "'");
674
675 assert(!flags.headers_parsed);
676
677 if (!inBuf.length()) {
678 ctx_exit(ctx);
679 return;
680 }
681
682 /* Attempt to parse the first line; this will define where the protocol, status, reason-phrase and header begin */
683 {
684 if (hp == NULL)
685 hp = new Http1::ResponseParser;
686
687 bool parsedOk = hp->parse(inBuf);
688
689 // sync the buffers after parsing.
690 inBuf = hp->remaining();
691
692 if (hp->needsMoreData()) {
693 if (eof) { // no more data coming
694 /* Bug 2879: Replies may terminate with \r\n then EOF instead of \r\n\r\n.
695 * We also may receive truncated responses.
696 * Ensure here that we have at minimum two \r\n when EOF is seen.
697 */
698 inBuf.append("\r\n\r\n", 4);
699 // retry the parse
700 parsedOk = hp->parse(inBuf);
701 // sync the buffers after parsing.
702 inBuf = hp->remaining();
703 } else {
704 debugs(33, 5, "Incomplete response, waiting for end of response headers");
705 ctx_exit(ctx);
706 return;
707 }
708 }
709
710 if (!parsedOk) {
711 // unrecoverable parsing error
712 // TODO: Use Raw! XXX: inBuf no longer has the [beginning of the] malformed header.
713 debugs(11, 3, "Non-HTTP-compliant header:\n---------\n" << inBuf << "\n----------");
714 flags.headers_parsed = true;
715 HttpReply *newrep = new HttpReply;
716 newrep->sline.set(Http::ProtocolVersion(), hp->parseStatusCode);
717 setVirginReply(newrep);
718 ctx_exit(ctx);
719 return;
720 }
721 }
722
723 /* We know the whole response is in parser now */
724 debugs(11, 2, "HTTP Server " << serverConnection);
725 debugs(11, 2, "HTTP Server RESPONSE:\n---------\n" <<
726 hp->messageProtocol() << " " << hp->messageStatus() << " " << hp->reasonPhrase() << "\n" <<
727 hp->mimeHeader() <<
728 "----------");
729
730 // reset payload tracking to begin after message headers
731 payloadSeen = inBuf.length();
732
733 HttpReply *newrep = new HttpReply;
734 // XXX: RFC 7230 indicates we MAY ignore the reason phrase,
735 // and use an empty string on unknown status.
736 // We do that now to avoid performance regression from using SBuf::c_str()
737 newrep->sline.set(Http::ProtocolVersion(1,1), hp->messageStatus() /* , hp->reasonPhrase() */);
738 newrep->sline.protocol = newrep->sline.version.protocol = hp->messageProtocol().protocol;
739 newrep->sline.version.major = hp->messageProtocol().major;
740 newrep->sline.version.minor = hp->messageProtocol().minor;
741
742 // parse headers
743 if (!newrep->parseHeader(*hp)) {
744 // XXX: when Http::ProtocolVersion is a function, remove this hack. just set with messageProtocol()
745 newrep->sline.set(Http::ProtocolVersion(), Http::scInvalidHeader);
746 newrep->sline.version.protocol = hp->messageProtocol().protocol;
747 newrep->sline.version.major = hp->messageProtocol().major;
748 newrep->sline.version.minor = hp->messageProtocol().minor;
749 debugs(11, 2, "error parsing response headers mime block");
750 }
751
752 // done with Parser, now process using the HttpReply
753 hp = NULL;
754
755 newrep->sources |= request->url.getScheme() == AnyP::PROTO_HTTPS ? Http::Message::srcHttps : Http::Message::srcHttp;
756
757 newrep->removeStaleWarnings();
758
759 if (newrep->sline.protocol == AnyP::PROTO_HTTP && newrep->sline.status() >= 100 && newrep->sline.status() < 200) {
760 handle1xx(newrep);
761 ctx_exit(ctx);
762 return;
763 }
764
765 flags.chunked = false;
766 if (newrep->sline.protocol == AnyP::PROTO_HTTP && newrep->header.chunked()) {
767 flags.chunked = true;
768 httpChunkDecoder = new Http1::TeChunkedParser;
769 }
770
771 if (!peerSupportsConnectionPinning())
772 request->flags.connectionAuthDisabled = true;
773
774 HttpReply *vrep = setVirginReply(newrep);
775 flags.headers_parsed = true;
776
777 keepaliveAccounting(vrep);
778
779 checkDateSkew(vrep);
780
781 processSurrogateControl (vrep);
782
783 request->hier.peer_reply_status = newrep->sline.status();
784
785 ctx_exit(ctx);
786 }
787
788 /// ignore or start forwarding the 1xx response (a.k.a., control message)
789 void
790 HttpStateData::handle1xx(HttpReply *reply)
791 {
792 HttpReply::Pointer msg(reply); // will destroy reply if unused
793
794 // one 1xx at a time: we must not be called while waiting for previous 1xx
795 Must(!flags.handling1xx);
796 flags.handling1xx = true;
797
798 if (!request->canHandle1xx() || request->forcedBodyContinuation) {
799 debugs(11, 2, "ignoring 1xx because it is " << (request->forcedBodyContinuation ? "already sent" : "not supported by client"));
800 proceedAfter1xx();
801 return;
802 }
803
804 #if USE_HTTP_VIOLATIONS
805 // check whether the 1xx response forwarding is allowed by squid.conf
806 if (Config.accessList.reply) {
807 ACLFilledChecklist ch(Config.accessList.reply, originalRequest().getRaw());
808 ch.reply = reply;
809 HTTPMSGLOCK(ch.reply);
810 if (ch.fastCheck() != ACCESS_ALLOWED) { // TODO: support slow lookups?
811 debugs(11, 3, HERE << "ignoring denied 1xx");
812 proceedAfter1xx();
813 return;
814 }
815 }
816 #endif // USE_HTTP_VIOLATIONS
817
818 debugs(11, 2, HERE << "forwarding 1xx to client");
819
820 // the Sink will use this to call us back after writing 1xx to the client
821 typedef NullaryMemFunT<HttpStateData> CbDialer;
822 const AsyncCall::Pointer cb = JobCallback(11, 3, CbDialer, this,
823 HttpStateData::proceedAfter1xx);
824 CallJobHere1(11, 4, request->clientConnectionManager, ConnStateData,
825 ConnStateData::sendControlMsg, HttpControlMsg(msg, cb));
826 // If the call is not fired, then the Sink is gone, and HttpStateData
827 // will terminate due to an aborted store entry or another similar error.
828 // If we get stuck, it is not handle1xx fault if we could get stuck
829 // for similar reasons without a 1xx response.
830 }
831
832 /// restores state and resumes processing after 1xx is ignored or forwarded
833 void
834 HttpStateData::proceedAfter1xx()
835 {
836 Must(flags.handling1xx);
837 debugs(11, 2, "continuing with " << payloadSeen << " bytes in buffer after 1xx");
838 CallJobHere(11, 3, this, HttpStateData, HttpStateData::processReply);
839 }
840
841 /**
842 * returns true if the peer can support connection pinning
843 */
844 bool
845 HttpStateData::peerSupportsConnectionPinning() const
846 {
847 if (!_peer)
848 return true;
849
850 /*If this peer does not support connection pinning (authenticated
851 connections) return false
852 */
853 if (!_peer->connection_auth)
854 return false;
855
856 const HttpReplyPointer rep(entry->mem_obj->getReply());
857
858 /*The peer supports connection pinning and the http reply status
859 is not unauthorized, so the related connection can be pinned
860 */
861 if (rep->sline.status() != Http::scUnauthorized)
862 return true;
863
864 /*The server respond with Http::scUnauthorized and the peer configured
865 with "connection-auth=on" we know that the peer supports pinned
866 connections
867 */
868 if (_peer->connection_auth == 1)
869 return true;
870
871 /*At this point peer has configured with "connection-auth=auto"
872 parameter so we need some extra checks to decide if we are going
873 to allow pinned connections or not
874 */
875
876 /*if the peer configured with originserver just allow connection
877 pinning (squid 2.6 behaviour)
878 */
879 if (_peer->options.originserver)
880 return true;
881
882 /*if the connections it is already pinned it is OK*/
883 if (request->flags.pinned)
884 return true;
885
886 /*Allow pinned connections only if the Proxy-support header exists in
887 reply and has in its list the "Session-Based-Authentication"
888 which means that the peer supports connection pinning.
889 */
890 if (rep->header.hasListMember(Http::HdrType::PROXY_SUPPORT, "Session-Based-Authentication", ','))
891 return true;
892
893 return false;
894 }
895
896 // Called when we parsed (and possibly adapted) the headers but
897 // had not starting storing (a.k.a., sending) the body yet.
898 void
899 HttpStateData::haveParsedReplyHeaders()
900 {
901 Client::haveParsedReplyHeaders();
902
903 Ctx ctx = ctx_enter(entry->mem_obj->urlXXX());
904 HttpReply *rep = finalReply();
905 const Http::StatusCode statusCode = rep->sline.status();
906
907 entry->timestampsSet();
908
909 /* Check if object is cacheable or not based on reply code */
910 debugs(11, 3, "HTTP CODE: " << statusCode);
911
912 if (const StoreEntry *oldEntry = findPreviouslyCachedEntry(entry))
913 sawDateGoBack = rep->olderThan(oldEntry->getReply());
914
915 if (neighbors_do_private_keys && !sawDateGoBack)
916 httpMaybeRemovePublic(entry, rep->sline.status());
917
918 bool varyFailure = false;
919 if (rep->header.has(Http::HdrType::VARY)
920 #if X_ACCELERATOR_VARY
921 || rep->header.has(Http::HdrType::HDR_X_ACCELERATOR_VARY)
922 #endif
923 ) {
924 const SBuf vary(httpMakeVaryMark(request.getRaw(), rep));
925
926 if (vary.isEmpty()) {
927 // TODO: check whether such responses are shareable.
928 // Do not share for now.
929 entry->makePrivate(false);
930 if (!fwd->reforwardableStatus(rep->sline.status()))
931 EBIT_CLR(entry->flags, ENTRY_FWD_HDR_WAIT);
932 varyFailure = true;
933 } else {
934 entry->mem_obj->vary_headers = vary;
935
936 // RFC 7231 section 7.1.4
937 // Vary:* can be cached, but has mandatory revalidation
938 static const SBuf asterisk("*");
939 if (vary == asterisk)
940 EBIT_SET(entry->flags, ENTRY_REVALIDATE_ALWAYS);
941 }
942 }
943
944 if (!varyFailure) {
945 /*
946 * If its not a reply that we will re-forward, then
947 * allow the client to get it.
948 */
949 if (!fwd->reforwardableStatus(rep->sline.status()))
950 EBIT_CLR(entry->flags, ENTRY_FWD_HDR_WAIT);
951
952 ReuseDecision decision(entry, statusCode);
953
954 switch (reusableReply(decision)) {
955
956 case ReuseDecision::reuseNot:
957 entry->makePrivate(false);
958 break;
959
960 case ReuseDecision::cachePositively:
961 entry->makePublic();
962 break;
963
964 case ReuseDecision::cacheNegatively:
965 entry->cacheNegatively();
966 break;
967
968 case ReuseDecision::doNotCacheButShare:
969 entry->makePrivate(true);
970 break;
971
972 default:
973 assert(0);
974 break;
975 }
976 debugs(11, 3, "decided: " << decision);
977 }
978
979 if (!ignoreCacheControl) {
980 if (rep->cache_control) {
981 // We are required to revalidate on many conditions.
982 // For security reasons we do so even if storage was caused by refresh_pattern ignore-* option
983
984 // CC:must-revalidate or CC:proxy-revalidate
985 const bool ccMustRevalidate = (rep->cache_control->hasProxyRevalidate() || rep->cache_control->hasMustRevalidate());
986
987 // CC:no-cache (only if there are no parameters)
988 const bool ccNoCacheNoParams = rep->cache_control->hasNoCacheWithoutParameters();
989
990 // CC:s-maxage=N
991 const bool ccSMaxAge = rep->cache_control->hasSMaxAge();
992
993 // CC:private (yes, these can sometimes be stored)
994 const bool ccPrivate = rep->cache_control->hasPrivate();
995
996 if (ccNoCacheNoParams || ccPrivate)
997 EBIT_SET(entry->flags, ENTRY_REVALIDATE_ALWAYS);
998 else if (ccMustRevalidate || ccSMaxAge)
999 EBIT_SET(entry->flags, ENTRY_REVALIDATE_STALE);
1000 }
1001 #if USE_HTTP_VIOLATIONS // response header Pragma::no-cache is undefined in HTTP
1002 else {
1003 // Expensive calculation. So only do it IF the CC: header is not present.
1004
1005 /* HACK: Pragma: no-cache in _replies_ is not documented in HTTP,
1006 * but servers like "Active Imaging Webcast/2.0" sure do use it */
1007 if (rep->header.has(Http::HdrType::PRAGMA) &&
1008 rep->header.hasListMember(Http::HdrType::PRAGMA,"no-cache",','))
1009 EBIT_SET(entry->flags, ENTRY_REVALIDATE_ALWAYS);
1010 }
1011 #endif
1012 }
1013
1014 #if HEADERS_LOG
1015 headersLog(1, 0, request->method, rep);
1016
1017 #endif
1018
1019 ctx_exit(ctx);
1020 }
1021
1022 HttpStateData::ConnectionStatus
1023 HttpStateData::statusIfComplete() const
1024 {
1025 const HttpReply *rep = virginReply();
1026 /** \par
1027 * If the reply wants to close the connection, it takes precedence */
1028
1029 static SBuf close("close", 5);
1030 if (httpHeaderHasConnDir(&rep->header, close))
1031 return COMPLETE_NONPERSISTENT_MSG;
1032
1033 /** \par
1034 * If we didn't send a keep-alive request header, then this
1035 * can not be a persistent connection.
1036 */
1037 if (!flags.keepalive)
1038 return COMPLETE_NONPERSISTENT_MSG;
1039
1040 /** \par
1041 * If we haven't sent the whole request then this can not be a persistent
1042 * connection.
1043 */
1044 if (!flags.request_sent) {
1045 debugs(11, 2, "Request not yet fully sent " << request->method << ' ' << entry->url());
1046 return COMPLETE_NONPERSISTENT_MSG;
1047 }
1048
1049 /** \par
1050 * What does the reply have to say about keep-alive?
1051 */
1052 /**
1053 \bug XXX BUG?
1054 * If the origin server (HTTP/1.0) does not send a keep-alive
1055 * header, but keeps the connection open anyway, what happens?
1056 * We'll return here and http.c waits for an EOF before changing
1057 * store_status to STORE_OK. Combine this with ENTRY_FWD_HDR_WAIT
1058 * and an error status code, and we might have to wait until
1059 * the server times out the socket.
1060 */
1061 if (!rep->keep_alive)
1062 return COMPLETE_NONPERSISTENT_MSG;
1063
1064 return COMPLETE_PERSISTENT_MSG;
1065 }
1066
1067 HttpStateData::ConnectionStatus
1068 HttpStateData::persistentConnStatus() const
1069 {
1070 debugs(11, 3, HERE << serverConnection << " eof=" << eof);
1071 if (eof) // already reached EOF
1072 return COMPLETE_NONPERSISTENT_MSG;
1073
1074 /* If server fd is closing (but we have not been notified yet), stop Comm
1075 I/O to avoid assertions. TODO: Change Comm API to handle callers that
1076 want more I/O after async closing (usually initiated by others). */
1077 // XXX: add canReceive or s/canSend/canTalkToServer/
1078 if (!Comm::IsConnOpen(serverConnection))
1079 return COMPLETE_NONPERSISTENT_MSG;
1080
1081 /** \par
1082 * In chunked response we do not know the content length but we are absolutely
1083 * sure about the end of response, so we are calling the statusIfComplete to
1084 * decide if we can be persistant
1085 */
1086 if (lastChunk && flags.chunked)
1087 return statusIfComplete();
1088
1089 const HttpReply *vrep = virginReply();
1090 debugs(11, 5, "persistentConnStatus: content_length=" << vrep->content_length);
1091
1092 const int64_t clen = vrep->bodySize(request->method);
1093
1094 debugs(11, 5, "persistentConnStatus: clen=" << clen);
1095
1096 /* If the body size is unknown we must wait for EOF */
1097 if (clen < 0)
1098 return INCOMPLETE_MSG;
1099
1100 /** \par
1101 * If the body size is known, we must wait until we've gotten all of it. */
1102 if (clen > 0) {
1103 debugs(11,5, "payloadSeen=" << payloadSeen << " content_length=" << vrep->content_length);
1104
1105 if (payloadSeen < vrep->content_length)
1106 return INCOMPLETE_MSG;
1107
1108 if (payloadTruncated > 0) // already read more than needed
1109 return COMPLETE_NONPERSISTENT_MSG; // disable pconns
1110 }
1111
1112 /** \par
1113 * If there is no message body or we got it all, we can be persistent */
1114 return statusIfComplete();
1115 }
1116
1117 static void
1118 readDelayed(void *context, CommRead const &)
1119 {
1120 HttpStateData *state = static_cast<HttpStateData*>(context);
1121 state->flags.do_next_read = true;
1122 state->maybeReadVirginBody();
1123 }
1124
1125 void
1126 HttpStateData::readReply(const CommIoCbParams &io)
1127 {
1128 Must(!flags.do_next_read); // XXX: should have been set false by mayReadVirginBody()
1129 flags.do_next_read = false;
1130
1131 debugs(11, 5, io.conn);
1132
1133 // Bail out early on Comm::ERR_CLOSING - close handlers will tidy up for us
1134 if (io.flag == Comm::ERR_CLOSING) {
1135 debugs(11, 3, "http socket closing");
1136 return;
1137 }
1138
1139 if (EBIT_TEST(entry->flags, ENTRY_ABORTED)) {
1140 abortTransaction("store entry aborted while reading reply");
1141 return;
1142 }
1143
1144 Must(Comm::IsConnOpen(serverConnection));
1145 Must(io.conn->fd == serverConnection->fd);
1146
1147 /*
1148 * Don't reset the timeout value here. The value should be
1149 * counting Config.Timeout.request and applies to the request
1150 * as a whole, not individual read() calls.
1151 * Plus, it breaks our lame *HalfClosed() detection
1152 */
1153
1154 Must(maybeMakeSpaceAvailable(true));
1155 CommIoCbParams rd(this); // will be expanded with ReadNow results
1156 rd.conn = io.conn;
1157 rd.size = entry->bytesWanted(Range<size_t>(0, inBuf.spaceSize()));
1158
1159 if (rd.size <= 0) {
1160 assert(entry->mem_obj);
1161 AsyncCall::Pointer nilCall;
1162 entry->mem_obj->delayRead(DeferredRead(readDelayed, this, CommRead(io.conn, NULL, 0, nilCall)));
1163 return;
1164 }
1165
1166 switch (Comm::ReadNow(rd, inBuf)) {
1167 case Comm::INPROGRESS:
1168 if (inBuf.isEmpty())
1169 debugs(33, 2, io.conn << ": no data to process, " << xstrerr(rd.xerrno));
1170 flags.do_next_read = true;
1171 maybeReadVirginBody();
1172 return;
1173
1174 case Comm::OK:
1175 {
1176 payloadSeen += rd.size;
1177 #if USE_DELAY_POOLS
1178 DelayId delayId = entry->mem_obj->mostBytesAllowed();
1179 delayId.bytesIn(rd.size);
1180 #endif
1181
1182 statCounter.server.all.kbytes_in += rd.size;
1183 statCounter.server.http.kbytes_in += rd.size;
1184 ++ IOStats.Http.reads;
1185
1186 int bin = 0;
1187 for (int clen = rd.size - 1; clen; ++bin)
1188 clen >>= 1;
1189
1190 ++ IOStats.Http.read_hist[bin];
1191
1192 // update peer response time stats (%<pt)
1193 const timeval &sent = request->hier.peer_http_request_sent;
1194 if (sent.tv_sec)
1195 tvSub(request->hier.peer_response_time, sent, current_time);
1196 else
1197 request->hier.peer_response_time.tv_sec = -1;
1198 }
1199
1200 /* Continue to process previously read data */
1201 break;
1202
1203 case Comm::ENDFILE: // close detected by 0-byte read
1204 eof = 1;
1205 flags.do_next_read = false;
1206
1207 /* Continue to process previously read data */
1208 break;
1209
1210 // case Comm::COMM_ERROR:
1211 default: // no other flags should ever occur
1212 debugs(11, 2, io.conn << ": read failure: " << xstrerr(rd.xerrno));
1213 ErrorState *err = new ErrorState(ERR_READ_ERROR, Http::scBadGateway, fwd->request);
1214 err->xerrno = rd.xerrno;
1215 fwd->fail(err);
1216 flags.do_next_read = false;
1217 closeServer();
1218 mustStop("HttpStateData::readReply");
1219 return;
1220 }
1221
1222 /* Process next response from buffer */
1223 processReply();
1224 }
1225
1226 /// processes the already read and buffered response data, possibly after
1227 /// waiting for asynchronous 1xx control message processing
1228 void
1229 HttpStateData::processReply()
1230 {
1231
1232 if (flags.handling1xx) { // we came back after handling a 1xx response
1233 debugs(11, 5, HERE << "done with 1xx handling");
1234 flags.handling1xx = false;
1235 Must(!flags.headers_parsed);
1236 }
1237
1238 if (!flags.headers_parsed) { // have not parsed headers yet?
1239 PROF_start(HttpStateData_processReplyHeader);
1240 processReplyHeader();
1241 PROF_stop(HttpStateData_processReplyHeader);
1242
1243 if (!continueAfterParsingHeader()) // parsing error or need more data
1244 return; // TODO: send errors to ICAP
1245
1246 adaptOrFinalizeReply(); // may write to, abort, or "close" the entry
1247 }
1248
1249 // kick more reads if needed and/or process the response body, if any
1250 PROF_start(HttpStateData_processReplyBody);
1251 processReplyBody(); // may call serverComplete()
1252 PROF_stop(HttpStateData_processReplyBody);
1253 }
1254
1255 /**
1256 \retval true if we can continue with processing the body or doing ICAP.
1257 */
1258 bool
1259 HttpStateData::continueAfterParsingHeader()
1260 {
1261 if (flags.handling1xx) {
1262 debugs(11, 5, HERE << "wait for 1xx handling");
1263 Must(!flags.headers_parsed);
1264 return false;
1265 }
1266
1267 if (!flags.headers_parsed && !eof) {
1268 debugs(11, 9, "needs more at " << inBuf.length());
1269 flags.do_next_read = true;
1270 /** \retval false If we have not finished parsing the headers and may get more data.
1271 * Schedules more reads to retrieve the missing data.
1272 */
1273 maybeReadVirginBody(); // schedules all kinds of reads; TODO: rename
1274 return false;
1275 }
1276
1277 /** If we are done with parsing, check for errors */
1278
1279 err_type error = ERR_NONE;
1280
1281 if (flags.headers_parsed) { // parsed headers, possibly with errors
1282 // check for header parsing errors
1283 if (HttpReply *vrep = virginReply()) {
1284 const Http::StatusCode s = vrep->sline.status();
1285 const AnyP::ProtocolVersion &v = vrep->sline.version;
1286 if (s == Http::scInvalidHeader && v != Http::ProtocolVersion(0,9)) {
1287 debugs(11, DBG_IMPORTANT, "WARNING: HTTP: Invalid Response: Bad header encountered from " << entry->url() << " AKA " << request->url);
1288 error = ERR_INVALID_RESP;
1289 } else if (s == Http::scHeaderTooLarge) {
1290 fwd->dontRetry(true);
1291 error = ERR_TOO_BIG;
1292 } else if (vrep->header.conflictingContentLength()) {
1293 fwd->dontRetry(true);
1294 error = ERR_INVALID_RESP;
1295 } else {
1296 return true; // done parsing, got reply, and no error
1297 }
1298 } else {
1299 // parsed headers but got no reply
1300 debugs(11, DBG_IMPORTANT, "WARNING: HTTP: Invalid Response: No reply at all for " << entry->url() << " AKA " << request->url);
1301 error = ERR_INVALID_RESP;
1302 }
1303 } else {
1304 assert(eof);
1305 if (inBuf.length()) {
1306 error = ERR_INVALID_RESP;
1307 debugs(11, DBG_IMPORTANT, "WARNING: HTTP: Invalid Response: Headers did not parse at all for " << entry->url() << " AKA " << request->url);
1308 } else {
1309 error = ERR_ZERO_SIZE_OBJECT;
1310 debugs(11, (request->flags.accelerated?DBG_IMPORTANT:2), "WARNING: HTTP: Invalid Response: No object data received for " << entry->url() << " AKA " << request->url);
1311 }
1312 }
1313
1314 assert(error != ERR_NONE);
1315 entry->reset();
1316 fwd->fail(new ErrorState(error, Http::scBadGateway, fwd->request));
1317 flags.do_next_read = false;
1318 closeServer();
1319 mustStop("HttpStateData::continueAfterParsingHeader");
1320 return false; // quit on error
1321 }
1322
1323 /** truncate what we read if we read too much so that writeReplyBody()
1324 writes no more than what we should have read */
1325 void
1326 HttpStateData::truncateVirginBody()
1327 {
1328 assert(flags.headers_parsed);
1329
1330 HttpReply *vrep = virginReply();
1331 int64_t clen = -1;
1332 if (!vrep->expectingBody(request->method, clen) || clen < 0)
1333 return; // no body or a body of unknown size, including chunked
1334
1335 if (payloadSeen - payloadTruncated <= clen)
1336 return; // we did not read too much or already took care of the extras
1337
1338 if (const int64_t extras = payloadSeen - payloadTruncated - clen) {
1339 // server sent more that the advertised content length
1340 debugs(11, 5, "payloadSeen=" << payloadSeen <<
1341 " clen=" << clen << '/' << vrep->content_length <<
1342 " trucated=" << payloadTruncated << '+' << extras);
1343
1344 inBuf.chop(0, inBuf.length() - extras);
1345 payloadTruncated += extras;
1346 }
1347 }
1348
1349 /**
1350 * Call this when there is data from the origin server
1351 * which should be sent to either StoreEntry, or to ICAP...
1352 */
1353 void
1354 HttpStateData::writeReplyBody()
1355 {
1356 truncateVirginBody(); // if needed
1357 const char *data = inBuf.rawContent();
1358 int len = inBuf.length();
1359 addVirginReplyBody(data, len);
1360 inBuf.consume(len);
1361 }
1362
1363 bool
1364 HttpStateData::decodeAndWriteReplyBody()
1365 {
1366 const char *data = NULL;
1367 int len;
1368 bool wasThereAnException = false;
1369 assert(flags.chunked);
1370 assert(httpChunkDecoder);
1371 SQUID_ENTER_THROWING_CODE();
1372 MemBuf decodedData;
1373 decodedData.init();
1374 httpChunkDecoder->setPayloadBuffer(&decodedData);
1375 const bool doneParsing = httpChunkDecoder->parse(inBuf);
1376 inBuf = httpChunkDecoder->remaining(); // sync buffers after parse
1377 len = decodedData.contentSize();
1378 data=decodedData.content();
1379 addVirginReplyBody(data, len);
1380 if (doneParsing) {
1381 lastChunk = 1;
1382 flags.do_next_read = false;
1383 }
1384 SQUID_EXIT_THROWING_CODE(wasThereAnException);
1385 return wasThereAnException;
1386 }
1387
1388 /**
1389 * processReplyBody has two purposes:
1390 * 1 - take the reply body data, if any, and put it into either
1391 * the StoreEntry, or give it over to ICAP.
1392 * 2 - see if we made it to the end of the response (persistent
1393 * connections and such)
1394 */
1395 void
1396 HttpStateData::processReplyBody()
1397 {
1398 Ip::Address client_addr;
1399 bool ispinned = false;
1400
1401 if (!flags.headers_parsed) {
1402 flags.do_next_read = true;
1403 maybeReadVirginBody();
1404 return;
1405 }
1406
1407 #if USE_ADAPTATION
1408 debugs(11,5, HERE << "adaptationAccessCheckPending=" << adaptationAccessCheckPending);
1409 if (adaptationAccessCheckPending)
1410 return;
1411
1412 #endif
1413
1414 /*
1415 * At this point the reply headers have been parsed and consumed.
1416 * That means header content has been removed from readBuf and
1417 * it contains only body data.
1418 */
1419 if (entry->isAccepting()) {
1420 if (flags.chunked) {
1421 if (!decodeAndWriteReplyBody()) {
1422 flags.do_next_read = false;
1423 serverComplete();
1424 return;
1425 }
1426 } else
1427 writeReplyBody();
1428 }
1429
1430 // storing/sending methods like earlier adaptOrFinalizeReply() or
1431 // above writeReplyBody() may release/abort the store entry.
1432 if (EBIT_TEST(entry->flags, ENTRY_ABORTED)) {
1433 // TODO: In some cases (e.g., 304), we should keep persistent conn open.
1434 // Detect end-of-reply (and, hence, pool our idle pconn) earlier (ASAP).
1435 abortTransaction("store entry aborted while storing reply");
1436 return;
1437 } else
1438 switch (persistentConnStatus()) {
1439 case INCOMPLETE_MSG: {
1440 debugs(11, 5, "processReplyBody: INCOMPLETE_MSG from " << serverConnection);
1441 /* Wait for more data or EOF condition */
1442 AsyncCall::Pointer nil;
1443 if (flags.keepalive_broken) {
1444 commSetConnTimeout(serverConnection, 10, nil);
1445 } else {
1446 commSetConnTimeout(serverConnection, Config.Timeout.read, nil);
1447 }
1448
1449 flags.do_next_read = true;
1450 }
1451 break;
1452
1453 case COMPLETE_PERSISTENT_MSG:
1454 debugs(11, 5, "processReplyBody: COMPLETE_PERSISTENT_MSG from " << serverConnection);
1455 /* yes we have to clear all these! */
1456 commUnsetConnTimeout(serverConnection);
1457 flags.do_next_read = false;
1458
1459 comm_remove_close_handler(serverConnection->fd, closeHandler);
1460 closeHandler = NULL;
1461 fwd->unregister(serverConnection);
1462
1463 if (request->flags.spoofClientIp)
1464 client_addr = request->client_addr;
1465
1466 if (request->flags.pinned) {
1467 ispinned = true;
1468 } else if (request->flags.connectionAuth && request->flags.authSent) {
1469 ispinned = true;
1470 }
1471
1472 if (ispinned && request->clientConnectionManager.valid()) {
1473 request->clientConnectionManager->pinConnection(serverConnection, request.getRaw(), _peer,
1474 (request->flags.connectionAuth));
1475 } else {
1476 fwd->pconnPush(serverConnection, request->url.host());
1477 }
1478
1479 serverConnection = NULL;
1480 serverComplete();
1481 return;
1482
1483 case COMPLETE_NONPERSISTENT_MSG:
1484 debugs(11, 5, "processReplyBody: COMPLETE_NONPERSISTENT_MSG from " << serverConnection);
1485 serverComplete();
1486 return;
1487 }
1488
1489 maybeReadVirginBody();
1490 }
1491
1492 bool
1493 HttpStateData::mayReadVirginReplyBody() const
1494 {
1495 // TODO: Be more precise here. For example, if/when reading trailer, we may
1496 // not be doneWithServer() yet, but we should return false. Similarly, we
1497 // could still be writing the request body after receiving the whole reply.
1498 return !doneWithServer();
1499 }
1500
1501 void
1502 HttpStateData::maybeReadVirginBody()
1503 {
1504 // too late to read
1505 if (!Comm::IsConnOpen(serverConnection) || fd_table[serverConnection->fd].closing())
1506 return;
1507
1508 if (!maybeMakeSpaceAvailable(false))
1509 return;
1510
1511 // XXX: get rid of the do_next_read flag
1512 // check for the proper reasons preventing read(2)
1513 if (!flags.do_next_read)
1514 return;
1515
1516 flags.do_next_read = false;
1517
1518 // must not already be waiting for read(2) ...
1519 assert(!Comm::MonitorsRead(serverConnection->fd));
1520
1521 // wait for read(2) to be possible.
1522 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
1523 AsyncCall::Pointer call = JobCallback(11, 5, Dialer, this, HttpStateData::readReply);
1524 Comm::Read(serverConnection, call);
1525 }
1526
1527 bool
1528 HttpStateData::maybeMakeSpaceAvailable(bool doGrow)
1529 {
1530 // how much we are allowed to buffer
1531 const int limitBuffer = (flags.headers_parsed ? Config.readAheadGap : Config.maxReplyHeaderSize);
1532
1533 if (limitBuffer < 0 || inBuf.length() >= (SBuf::size_type)limitBuffer) {
1534 // when buffer is at or over limit already
1535 debugs(11, 7, "wont read up to " << limitBuffer << ". buffer has (" << inBuf.length() << "/" << inBuf.spaceSize() << ") from " << serverConnection);
1536 debugs(11, DBG_DATA, "buffer has {" << inBuf << "}");
1537 // Process next response from buffer
1538 processReply();
1539 return false;
1540 }
1541
1542 // how much we want to read
1543 const size_t read_size = calcBufferSpaceToReserve(inBuf.spaceSize(), (limitBuffer - inBuf.length()));
1544
1545 if (!read_size) {
1546 debugs(11, 7, "wont read up to " << read_size << " into buffer (" << inBuf.length() << "/" << inBuf.spaceSize() << ") from " << serverConnection);
1547 return false;
1548 }
1549
1550 // just report whether we could grow or not, dont actually do it
1551 if (doGrow)
1552 return (read_size >= 2);
1553
1554 // we may need to grow the buffer
1555 inBuf.reserveSpace(read_size);
1556 debugs(11, 8, (!flags.do_next_read ? "wont" : "may") <<
1557 " read up to " << read_size << " bytes info buf(" << inBuf.length() << "/" << inBuf.spaceSize() <<
1558 ") from " << serverConnection);
1559
1560 return (inBuf.spaceSize() >= 2); // only read if there is 1+ bytes of space available
1561 }
1562
1563 /// called after writing the very last request byte (body, last-chunk, etc)
1564 void
1565 HttpStateData::wroteLast(const CommIoCbParams &io)
1566 {
1567 debugs(11, 5, HERE << serverConnection << ": size " << io.size << ": errflag " << io.flag << ".");
1568 #if URL_CHECKSUM_DEBUG
1569
1570 entry->mem_obj->checkUrlChecksum();
1571 #endif
1572
1573 if (io.size > 0) {
1574 fd_bytes(io.fd, io.size, FD_WRITE);
1575 statCounter.server.all.kbytes_out += io.size;
1576 statCounter.server.http.kbytes_out += io.size;
1577 }
1578
1579 if (io.flag == Comm::ERR_CLOSING)
1580 return;
1581
1582 if (io.flag) {
1583 ErrorState *err = new ErrorState(ERR_WRITE_ERROR, Http::scBadGateway, fwd->request);
1584 err->xerrno = io.xerrno;
1585 fwd->fail(err);
1586 closeServer();
1587 mustStop("HttpStateData::wroteLast");
1588 return;
1589 }
1590
1591 sendComplete();
1592 }
1593
1594 /// successfully wrote the entire request (including body, last-chunk, etc.)
1595 void
1596 HttpStateData::sendComplete()
1597 {
1598 /*
1599 * Set the read timeout here because it hasn't been set yet.
1600 * We only set the read timeout after the request has been
1601 * fully written to the peer. If we start the timeout
1602 * after connection establishment, then we are likely to hit
1603 * the timeout for POST/PUT requests that have very large
1604 * request bodies.
1605 */
1606 typedef CommCbMemFunT<HttpStateData, CommTimeoutCbParams> TimeoutDialer;
1607 AsyncCall::Pointer timeoutCall = JobCallback(11, 5,
1608 TimeoutDialer, this, HttpStateData::httpTimeout);
1609
1610 commSetConnTimeout(serverConnection, Config.Timeout.read, timeoutCall);
1611 flags.request_sent = true;
1612 request->hier.peer_http_request_sent = current_time;
1613 }
1614
1615 void
1616 HttpStateData::closeServer()
1617 {
1618 debugs(11,5, HERE << "closing HTTP server " << serverConnection << " this " << this);
1619
1620 if (Comm::IsConnOpen(serverConnection)) {
1621 fwd->unregister(serverConnection);
1622 comm_remove_close_handler(serverConnection->fd, closeHandler);
1623 closeHandler = NULL;
1624 serverConnection->close();
1625 }
1626 }
1627
1628 bool
1629 HttpStateData::doneWithServer() const
1630 {
1631 return !Comm::IsConnOpen(serverConnection);
1632 }
1633
1634 /*
1635 * Fixup authentication request headers for special cases
1636 */
1637 static void
1638 httpFixupAuthentication(HttpRequest * request, const HttpHeader * hdr_in, HttpHeader * hdr_out, const Http::StateFlags &flags)
1639 {
1640 Http::HdrType header = flags.originpeer ? Http::HdrType::AUTHORIZATION : Http::HdrType::PROXY_AUTHORIZATION;
1641
1642 /* Nothing to do unless we are forwarding to a peer */
1643 if (!request->flags.proxying)
1644 return;
1645
1646 /* Needs to be explicitly enabled */
1647 if (!request->peer_login)
1648 return;
1649
1650 /* Maybe already dealt with? */
1651 if (hdr_out->has(header))
1652 return;
1653
1654 /* Nothing to do here for PASSTHRU */
1655 if (strcmp(request->peer_login, "PASSTHRU") == 0)
1656 return;
1657
1658 /* PROXYPASS is a special case, single-signon to servers with the proxy password (basic only) */
1659 if (flags.originpeer && strcmp(request->peer_login, "PROXYPASS") == 0 && hdr_in->has(Http::HdrType::PROXY_AUTHORIZATION)) {
1660 const char *auth = hdr_in->getStr(Http::HdrType::PROXY_AUTHORIZATION);
1661
1662 if (auth && strncasecmp(auth, "basic ", 6) == 0) {
1663 hdr_out->putStr(header, auth);
1664 return;
1665 }
1666 }
1667
1668 uint8_t loginbuf[base64_encode_len(MAX_LOGIN_SZ)];
1669 size_t blen;
1670 struct base64_encode_ctx ctx;
1671 base64_encode_init(&ctx);
1672
1673 /* Special mode to pass the username to the upstream cache */
1674 if (*request->peer_login == '*') {
1675 const char *username = "-";
1676
1677 if (request->extacl_user.size())
1678 username = request->extacl_user.termedBuf();
1679 #if USE_AUTH
1680 else if (request->auth_user_request != NULL)
1681 username = request->auth_user_request->username();
1682 #endif
1683
1684 blen = base64_encode_update(&ctx, loginbuf, strlen(username), reinterpret_cast<const uint8_t*>(username));
1685 blen += base64_encode_update(&ctx, loginbuf+blen, strlen(request->peer_login +1), reinterpret_cast<const uint8_t*>(request->peer_login +1));
1686 blen += base64_encode_final(&ctx, loginbuf+blen);
1687 httpHeaderPutStrf(hdr_out, header, "Basic %.*s", (int)blen, loginbuf);
1688 return;
1689 }
1690
1691 /* external_acl provided credentials */
1692 if (request->extacl_user.size() && request->extacl_passwd.size() &&
1693 (strcmp(request->peer_login, "PASS") == 0 ||
1694 strcmp(request->peer_login, "PROXYPASS") == 0)) {
1695
1696 blen = base64_encode_update(&ctx, loginbuf, request->extacl_user.size(), reinterpret_cast<const uint8_t*>(request->extacl_user.rawBuf()));
1697 blen += base64_encode_update(&ctx, loginbuf+blen, 1, reinterpret_cast<const uint8_t*>(":"));
1698 blen += base64_encode_update(&ctx, loginbuf+blen, request->extacl_passwd.size(), reinterpret_cast<const uint8_t*>(request->extacl_passwd.rawBuf()));
1699 blen += base64_encode_final(&ctx, loginbuf+blen);
1700 httpHeaderPutStrf(hdr_out, header, "Basic %.*s", (int)blen, loginbuf);
1701 return;
1702 }
1703 // if no external user credentials are available to fake authentication with PASS acts like PASSTHRU
1704 if (strcmp(request->peer_login, "PASS") == 0)
1705 return;
1706
1707 /* Kerberos login to peer */
1708 #if HAVE_AUTH_MODULE_NEGOTIATE && HAVE_KRB5 && HAVE_GSSAPI
1709 if (strncmp(request->peer_login, "NEGOTIATE",strlen("NEGOTIATE")) == 0) {
1710 char *Token=NULL;
1711 char *PrincipalName=NULL,*p;
1712 int negotiate_flags = 0;
1713
1714 if ((p=strchr(request->peer_login,':')) != NULL ) {
1715 PrincipalName=++p;
1716 }
1717 if (request->flags.auth_no_keytab) {
1718 negotiate_flags |= PEER_PROXY_NEGOTIATE_NOKEYTAB;
1719 }
1720 Token = peer_proxy_negotiate_auth(PrincipalName, request->peer_host, negotiate_flags);
1721 if (Token) {
1722 httpHeaderPutStrf(hdr_out, header, "Negotiate %s",Token);
1723 }
1724 return;
1725 }
1726 #endif /* HAVE_KRB5 && HAVE_GSSAPI */
1727
1728 blen = base64_encode_update(&ctx, loginbuf, strlen(request->peer_login), reinterpret_cast<const uint8_t*>(request->peer_login));
1729 blen += base64_encode_final(&ctx, loginbuf+blen);
1730 httpHeaderPutStrf(hdr_out, header, "Basic %.*s", (int)blen, loginbuf);
1731 return;
1732 }
1733
1734 /*
1735 * build request headers and append them to a given MemBuf
1736 * used by buildRequestPrefix()
1737 * note: initialised the HttpHeader, the caller is responsible for Clean()-ing
1738 */
1739 void
1740 HttpStateData::httpBuildRequestHeader(HttpRequest * request,
1741 StoreEntry * entry,
1742 const AccessLogEntryPointer &al,
1743 HttpHeader * hdr_out,
1744 const Http::StateFlags &flags)
1745 {
1746 /* building buffer for complex strings */
1747 #define BBUF_SZ (MAX_URL+32)
1748 LOCAL_ARRAY(char, bbuf, BBUF_SZ);
1749 LOCAL_ARRAY(char, ntoabuf, MAX_IPSTRLEN);
1750 const HttpHeader *hdr_in = &request->header;
1751 const HttpHeaderEntry *e = NULL;
1752 HttpHeaderPos pos = HttpHeaderInitPos;
1753 assert (hdr_out->owner == hoRequest);
1754
1755 /* use our IMS header if the cached entry has Last-Modified time */
1756 if (request->lastmod > -1)
1757 hdr_out->putTime(Http::HdrType::IF_MODIFIED_SINCE, request->lastmod);
1758
1759 // Add our own If-None-Match field if the cached entry has a strong ETag.
1760 // copyOneHeaderFromClientsideRequestToUpstreamRequest() adds client ones.
1761 if (request->etag.size() > 0) {
1762 hdr_out->addEntry(new HttpHeaderEntry(Http::HdrType::IF_NONE_MATCH, SBuf(),
1763 request->etag.termedBuf()));
1764 }
1765
1766 bool we_do_ranges = decideIfWeDoRanges (request);
1767
1768 String strConnection (hdr_in->getList(Http::HdrType::CONNECTION));
1769
1770 while ((e = hdr_in->getEntry(&pos)))
1771 copyOneHeaderFromClientsideRequestToUpstreamRequest(e, strConnection, request, hdr_out, we_do_ranges, flags);
1772
1773 /* Abstraction break: We should interpret multipart/byterange responses
1774 * into offset-length data, and this works around our inability to do so.
1775 */
1776 if (!we_do_ranges && request->multipartRangeRequest()) {
1777 /* don't cache the result */
1778 request->flags.cachable = false;
1779 /* pretend it's not a range request */
1780 request->ignoreRange("want to request the whole object");
1781 request->flags.isRanged = false;
1782 }
1783
1784 hdr_out->addVia(request->http_ver, hdr_in);
1785
1786 if (request->flags.accelerated) {
1787 /* Append Surrogate-Capabilities */
1788 String strSurrogate(hdr_in->getList(Http::HdrType::SURROGATE_CAPABILITY));
1789 #if USE_SQUID_ESI
1790 snprintf(bbuf, BBUF_SZ, "%s=\"Surrogate/1.0 ESI/1.0\"", Config.Accel.surrogate_id);
1791 #else
1792 snprintf(bbuf, BBUF_SZ, "%s=\"Surrogate/1.0\"", Config.Accel.surrogate_id);
1793 #endif
1794 strListAdd(&strSurrogate, bbuf, ',');
1795 hdr_out->putStr(Http::HdrType::SURROGATE_CAPABILITY, strSurrogate.termedBuf());
1796 }
1797
1798 /** \pre Handle X-Forwarded-For */
1799 if (strcmp(opt_forwarded_for, "delete") != 0) {
1800
1801 String strFwd = hdr_in->getList(Http::HdrType::X_FORWARDED_FOR);
1802
1803 // if we cannot double strFwd size, then it grew past 50% of the limit
1804 if (!strFwd.canGrowBy(strFwd.size())) {
1805 // There is probably a forwarding loop with Via detection disabled.
1806 // If we do nothing, String will assert on overflow soon.
1807 // TODO: Terminate all transactions with huge XFF?
1808 strFwd = "error";
1809
1810 static int warnedCount = 0;
1811 if (warnedCount++ < 100) {
1812 const SBuf url(entry ? SBuf(entry->url()) : request->effectiveRequestUri());
1813 debugs(11, DBG_IMPORTANT, "Warning: likely forwarding loop with " << url);
1814 }
1815 }
1816
1817 if (strcmp(opt_forwarded_for, "on") == 0) {
1818 /** If set to ON - append client IP or 'unknown'. */
1819 if ( request->client_addr.isNoAddr() )
1820 strListAdd(&strFwd, "unknown", ',');
1821 else
1822 strListAdd(&strFwd, request->client_addr.toStr(ntoabuf, MAX_IPSTRLEN), ',');
1823 } else if (strcmp(opt_forwarded_for, "off") == 0) {
1824 /** If set to OFF - append 'unknown'. */
1825 strListAdd(&strFwd, "unknown", ',');
1826 } else if (strcmp(opt_forwarded_for, "transparent") == 0) {
1827 /** If set to TRANSPARENT - pass through unchanged. */
1828 } else if (strcmp(opt_forwarded_for, "truncate") == 0) {
1829 /** If set to TRUNCATE - drop existing list and replace with client IP or 'unknown'. */
1830 if ( request->client_addr.isNoAddr() )
1831 strFwd = "unknown";
1832 else
1833 strFwd = request->client_addr.toStr(ntoabuf, MAX_IPSTRLEN);
1834 }
1835 if (strFwd.size() > 0)
1836 hdr_out->putStr(Http::HdrType::X_FORWARDED_FOR, strFwd.termedBuf());
1837 }
1838 /** If set to DELETE - do not copy through. */
1839
1840 /* append Host if not there already */
1841 if (!hdr_out->has(Http::HdrType::HOST)) {
1842 if (request->peer_domain) {
1843 hdr_out->putStr(Http::HdrType::HOST, request->peer_domain);
1844 } else {
1845 SBuf authority = request->url.authority();
1846 hdr_out->putStr(Http::HdrType::HOST, authority.c_str());
1847 }
1848 }
1849
1850 /* append Authorization if known in URL, not in header and going direct */
1851 if (!hdr_out->has(Http::HdrType::AUTHORIZATION)) {
1852 if (!request->flags.proxying && !request->url.userInfo().isEmpty()) {
1853 static uint8_t result[base64_encode_len(MAX_URL*2)]; // should be big enough for a single URI segment
1854 struct base64_encode_ctx ctx;
1855 base64_encode_init(&ctx);
1856 size_t blen = base64_encode_update(&ctx, result, request->url.userInfo().length(), reinterpret_cast<const uint8_t*>(request->url.userInfo().rawContent()));
1857 blen += base64_encode_final(&ctx, result+blen);
1858 result[blen] = '\0';
1859 if (blen)
1860 httpHeaderPutStrf(hdr_out, Http::HdrType::AUTHORIZATION, "Basic %.*s", (int)blen, result);
1861 }
1862 }
1863
1864 /* Fixup (Proxy-)Authorization special cases. Plain relaying dealt with above */
1865 httpFixupAuthentication(request, hdr_in, hdr_out, flags);
1866
1867 /* append Cache-Control, add max-age if not there already */
1868 {
1869 HttpHdrCc *cc = hdr_in->getCc();
1870
1871 if (!cc)
1872 cc = new HttpHdrCc();
1873
1874 #if 0 /* see bug 2330 */
1875 /* Set no-cache if determined needed but not found */
1876 if (request->flags.nocache)
1877 EBIT_SET(cc->mask, HttpHdrCcType::CC_NO_CACHE);
1878 #endif
1879
1880 /* Add max-age only without no-cache */
1881 if (!cc->hasMaxAge() && !cc->hasNoCache()) {
1882 // XXX: performance regression. c_str() reallocates
1883 SBuf tmp(request->effectiveRequestUri());
1884 cc->maxAge(getMaxAge(entry ? entry->url() : tmp.c_str()));
1885 }
1886
1887 /* Enforce sibling relations */
1888 if (flags.only_if_cached)
1889 cc->onlyIfCached(true);
1890
1891 hdr_out->putCc(cc);
1892
1893 delete cc;
1894 }
1895
1896 /* maybe append Connection: keep-alive */
1897 if (flags.keepalive) {
1898 hdr_out->putStr(Http::HdrType::CONNECTION, "keep-alive");
1899 }
1900
1901 /* append Front-End-Https */
1902 if (flags.front_end_https) {
1903 if (flags.front_end_https == 1 || request->url.getScheme() == AnyP::PROTO_HTTPS)
1904 hdr_out->putStr(Http::HdrType::FRONT_END_HTTPS, "On");
1905 }
1906
1907 if (flags.chunked_request) {
1908 // Do not just copy the original value so that if the client-side
1909 // starts decode other encodings, this code may remain valid.
1910 hdr_out->putStr(Http::HdrType::TRANSFER_ENCODING, "chunked");
1911 }
1912
1913 /* Now mangle the headers. */
1914 httpHdrMangleList(hdr_out, request, al, ROR_REQUEST);
1915
1916 strConnection.clean();
1917 }
1918
1919 /**
1920 * Decides whether a particular header may be cloned from the received Clients request
1921 * to our outgoing fetch request.
1922 */
1923 void
1924 copyOneHeaderFromClientsideRequestToUpstreamRequest(const HttpHeaderEntry *e, const String strConnection, const HttpRequest * request, HttpHeader * hdr_out, const int we_do_ranges, const Http::StateFlags &flags)
1925 {
1926 debugs(11, 5, "httpBuildRequestHeader: " << e->name << ": " << e->value );
1927
1928 switch (e->id) {
1929
1930 /** \par RFC 2616 sect 13.5.1 - Hop-by-Hop headers which Squid should not pass on. */
1931
1932 case Http::HdrType::PROXY_AUTHORIZATION:
1933 /** \par Proxy-Authorization:
1934 * Only pass on proxy authentication to peers for which
1935 * authentication forwarding is explicitly enabled
1936 */
1937 if (!flags.originpeer && flags.proxying && request->peer_login &&
1938 (strcmp(request->peer_login, "PASS") == 0 ||
1939 strcmp(request->peer_login, "PROXYPASS") == 0 ||
1940 strcmp(request->peer_login, "PASSTHRU") == 0)) {
1941 hdr_out->addEntry(e->clone());
1942 }
1943 break;
1944
1945 /** \par RFC 2616 sect 13.5.1 - Hop-by-Hop headers which Squid does not pass on. */
1946
1947 case Http::HdrType::CONNECTION: /** \par Connection: */
1948 case Http::HdrType::TE: /** \par TE: */
1949 case Http::HdrType::KEEP_ALIVE: /** \par Keep-Alive: */
1950 case Http::HdrType::PROXY_AUTHENTICATE: /** \par Proxy-Authenticate: */
1951 case Http::HdrType::TRAILER: /** \par Trailer: */
1952 case Http::HdrType::UPGRADE: /** \par Upgrade: */
1953 case Http::HdrType::TRANSFER_ENCODING: /** \par Transfer-Encoding: */
1954 break;
1955
1956 /** \par OTHER headers I haven't bothered to track down yet. */
1957
1958 case Http::HdrType::AUTHORIZATION:
1959 /** \par WWW-Authorization:
1960 * Pass on WWW authentication */
1961
1962 if (!flags.originpeer) {
1963 hdr_out->addEntry(e->clone());
1964 } else {
1965 /** \note In accelerators, only forward authentication if enabled
1966 * (see also httpFixupAuthentication for special cases)
1967 */
1968 if (request->peer_login &&
1969 (strcmp(request->peer_login, "PASS") == 0 ||
1970 strcmp(request->peer_login, "PASSTHRU") == 0 ||
1971 strcmp(request->peer_login, "PROXYPASS") == 0)) {
1972 hdr_out->addEntry(e->clone());
1973 }
1974 }
1975
1976 break;
1977
1978 case Http::HdrType::HOST:
1979 /** \par Host:
1980 * Normally Squid rewrites the Host: header.
1981 * However, there is one case when we don't: If the URL
1982 * went through our redirector and the admin configured
1983 * 'redir_rewrites_host' to be off.
1984 */
1985 if (request->peer_domain)
1986 hdr_out->putStr(Http::HdrType::HOST, request->peer_domain);
1987 else if (request->flags.redirected && !Config.onoff.redir_rewrites_host)
1988 hdr_out->addEntry(e->clone());
1989 else {
1990 SBuf authority = request->url.authority();
1991 hdr_out->putStr(Http::HdrType::HOST, authority.c_str());
1992 }
1993
1994 break;
1995
1996 case Http::HdrType::IF_MODIFIED_SINCE:
1997 /** \par If-Modified-Since:
1998 * append unless we added our own,
1999 * but only if cache_miss_revalidate is enabled, or
2000 * the request is not cacheable, or
2001 * the request contains authentication credentials.
2002 * \note at most one client's If-Modified-Since header can pass through
2003 */
2004 // XXX: need to check and cleanup the auth case so cacheable auth requests get cached.
2005 if (hdr_out->has(Http::HdrType::IF_MODIFIED_SINCE))
2006 break;
2007 else if (Config.onoff.cache_miss_revalidate || !request->flags.cachable || request->flags.auth)
2008 hdr_out->addEntry(e->clone());
2009 break;
2010
2011 case Http::HdrType::IF_NONE_MATCH:
2012 /** \par If-None-Match:
2013 * append if the wildcard '*' special case value is present, or
2014 * cache_miss_revalidate is disabled, or
2015 * the request is not cacheable in this proxy, or
2016 * the request contains authentication credentials.
2017 * \note this header lists a set of responses for the server to elide sending. Squid added values are extending that set.
2018 */
2019 // XXX: need to check and cleanup the auth case so cacheable auth requests get cached.
2020 if (hdr_out->hasListMember(Http::HdrType::IF_MATCH, "*", ',') || Config.onoff.cache_miss_revalidate || !request->flags.cachable || request->flags.auth)
2021 hdr_out->addEntry(e->clone());
2022 break;
2023
2024 case Http::HdrType::MAX_FORWARDS:
2025 /** \par Max-Forwards:
2026 * pass only on TRACE or OPTIONS requests */
2027 if (request->method == Http::METHOD_TRACE || request->method == Http::METHOD_OPTIONS) {
2028 const int64_t hops = e->getInt64();
2029
2030 if (hops > 0)
2031 hdr_out->putInt64(Http::HdrType::MAX_FORWARDS, hops - 1);
2032 }
2033
2034 break;
2035
2036 case Http::HdrType::VIA:
2037 /** \par Via:
2038 * If Via is disabled then forward any received header as-is.
2039 * Otherwise leave for explicit updated addition later. */
2040
2041 if (!Config.onoff.via)
2042 hdr_out->addEntry(e->clone());
2043
2044 break;
2045
2046 case Http::HdrType::RANGE:
2047
2048 case Http::HdrType::IF_RANGE:
2049
2050 case Http::HdrType::REQUEST_RANGE:
2051 /** \par Range:, If-Range:, Request-Range:
2052 * Only pass if we accept ranges */
2053 if (!we_do_ranges)
2054 hdr_out->addEntry(e->clone());
2055
2056 break;
2057
2058 case Http::HdrType::PROXY_CONNECTION: // SHOULD ignore. But doing so breaks things.
2059 break;
2060
2061 case Http::HdrType::CONTENT_LENGTH:
2062 // pass through unless we chunk; also, keeping this away from default
2063 // prevents request smuggling via Connection: Content-Length tricks
2064 if (!flags.chunked_request)
2065 hdr_out->addEntry(e->clone());
2066 break;
2067
2068 case Http::HdrType::X_FORWARDED_FOR:
2069
2070 case Http::HdrType::CACHE_CONTROL:
2071 /** \par X-Forwarded-For:, Cache-Control:
2072 * handled specially by Squid, so leave off for now.
2073 * append these after the loop if needed */
2074 break;
2075
2076 case Http::HdrType::FRONT_END_HTTPS:
2077 /** \par Front-End-Https:
2078 * Pass thru only if peer is configured with front-end-https */
2079 if (!flags.front_end_https)
2080 hdr_out->addEntry(e->clone());
2081
2082 break;
2083
2084 default:
2085 /** \par default.
2086 * pass on all other header fields
2087 * which are NOT listed by the special Connection: header. */
2088 if (strConnection.size()>0 && strListIsMember(&strConnection, e->name, ',')) {
2089 debugs(11, 2, "'" << e->name << "' header cropped by Connection: definition");
2090 return;
2091 }
2092
2093 hdr_out->addEntry(e->clone());
2094 }
2095 }
2096
2097 bool
2098 HttpStateData::decideIfWeDoRanges (HttpRequest * request)
2099 {
2100 bool result = true;
2101 /* decide if we want to do Ranges ourselves
2102 * and fetch the whole object now)
2103 * We want to handle Ranges ourselves iff
2104 * - we can actually parse client Range specs
2105 * - the specs are expected to be simple enough (e.g. no out-of-order ranges)
2106 * - reply will be cachable
2107 * (If the reply will be uncachable we have to throw it away after
2108 * serving this request, so it is better to forward ranges to
2109 * the server and fetch only the requested content)
2110 */
2111
2112 int64_t roffLimit = request->getRangeOffsetLimit();
2113
2114 if (NULL == request->range || !request->flags.cachable
2115 || request->range->offsetLimitExceeded(roffLimit) || request->flags.connectionAuth)
2116 result = false;
2117
2118 debugs(11, 8, "decideIfWeDoRanges: range specs: " <<
2119 request->range << ", cachable: " <<
2120 request->flags.cachable << "; we_do_ranges: " << result);
2121
2122 return result;
2123 }
2124
2125 /* build request prefix and append it to a given MemBuf;
2126 * return the length of the prefix */
2127 mb_size_t
2128 HttpStateData::buildRequestPrefix(MemBuf * mb)
2129 {
2130 const int offset = mb->size;
2131 /* Uses a local httpver variable to print the HTTP label
2132 * since the HttpRequest may have an older version label.
2133 * XXX: This could create protocol bugs as the headers sent and
2134 * flow control should all be based on the HttpRequest version
2135 * not the one we are sending. Needs checking.
2136 */
2137 const AnyP::ProtocolVersion httpver = Http::ProtocolVersion();
2138 const SBuf url(_peer && !_peer->options.originserver ? request->effectiveRequestUri() : request->url.path());
2139 mb->appendf(SQUIDSBUFPH " " SQUIDSBUFPH " %s/%d.%d\r\n",
2140 SQUIDSBUFPRINT(request->method.image()),
2141 SQUIDSBUFPRINT(url),
2142 AnyP::ProtocolType_str[httpver.protocol],
2143 httpver.major,httpver.minor);
2144 /* build and pack headers */
2145 {
2146 HttpHeader hdr(hoRequest);
2147 httpBuildRequestHeader(request.getRaw(), entry, fwd->al, &hdr, flags);
2148
2149 if (request->flags.pinned && request->flags.connectionAuth)
2150 request->flags.authSent = true;
2151 else if (hdr.has(Http::HdrType::AUTHORIZATION))
2152 request->flags.authSent = true;
2153
2154 hdr.packInto(mb);
2155 hdr.clean();
2156 }
2157 /* append header terminator */
2158 mb->append(crlf, 2);
2159 return mb->size - offset;
2160 }
2161
2162 /* This will be called when connect completes. Write request. */
2163 bool
2164 HttpStateData::sendRequest()
2165 {
2166 MemBuf mb;
2167
2168 debugs(11, 5, HERE << serverConnection << ", request " << request << ", this " << this << ".");
2169
2170 if (!Comm::IsConnOpen(serverConnection)) {
2171 debugs(11,3, HERE << "cannot send request to closing " << serverConnection);
2172 assert(closeHandler != NULL);
2173 return false;
2174 }
2175
2176 typedef CommCbMemFunT<HttpStateData, CommTimeoutCbParams> TimeoutDialer;
2177 AsyncCall::Pointer timeoutCall = JobCallback(11, 5,
2178 TimeoutDialer, this, HttpStateData::httpTimeout);
2179 commSetConnTimeout(serverConnection, Config.Timeout.lifetime, timeoutCall);
2180 flags.do_next_read = true;
2181 maybeReadVirginBody();
2182
2183 if (request->body_pipe != NULL) {
2184 if (!startRequestBodyFlow()) // register to receive body data
2185 return false;
2186 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2187 requestSender = JobCallback(11,5,
2188 Dialer, this, HttpStateData::sentRequestBody);
2189
2190 Must(!flags.chunked_request);
2191 // use chunked encoding if we do not know the length
2192 if (request->content_length < 0)
2193 flags.chunked_request = true;
2194 } else {
2195 assert(!requestBodySource);
2196 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2197 requestSender = JobCallback(11,5,
2198 Dialer, this, HttpStateData::wroteLast);
2199 }
2200
2201 flags.originpeer = (_peer != NULL && _peer->options.originserver);
2202 flags.proxying = (_peer != NULL && !flags.originpeer);
2203
2204 /*
2205 * Is keep-alive okay for all request methods?
2206 */
2207 if (request->flags.mustKeepalive)
2208 flags.keepalive = true;
2209 else if (request->flags.pinned)
2210 flags.keepalive = request->persistent();
2211 else if (!Config.onoff.server_pconns)
2212 flags.keepalive = false;
2213 else if (_peer == NULL)
2214 flags.keepalive = true;
2215 else if (_peer->stats.n_keepalives_sent < 10)
2216 flags.keepalive = true;
2217 else if ((double) _peer->stats.n_keepalives_recv /
2218 (double) _peer->stats.n_keepalives_sent > 0.50)
2219 flags.keepalive = true;
2220
2221 if (_peer) {
2222 /*The old code here was
2223 if (neighborType(_peer, request->url) == PEER_SIBLING && ...
2224 which is equivalent to:
2225 if (neighborType(_peer, URL()) == PEER_SIBLING && ...
2226 or better:
2227 if (((_peer->type == PEER_MULTICAST && p->options.mcast_siblings) ||
2228 _peer->type == PEER_SIBLINGS ) && _peer->options.allow_miss)
2229 flags.only_if_cached = 1;
2230
2231 But I suppose it was a bug
2232 */
2233 if (neighborType(_peer, request->url) == PEER_SIBLING && !_peer->options.allow_miss)
2234 flags.only_if_cached = true;
2235
2236 flags.front_end_https = _peer->front_end_https;
2237 }
2238
2239 mb.init();
2240 request->peer_host=_peer?_peer->host:NULL;
2241 buildRequestPrefix(&mb);
2242
2243 debugs(11, 2, "HTTP Server " << serverConnection);
2244 debugs(11, 2, "HTTP Server REQUEST:\n---------\n" << mb.buf << "\n----------");
2245
2246 Comm::Write(serverConnection, &mb, requestSender);
2247 return true;
2248 }
2249
2250 bool
2251 HttpStateData::getMoreRequestBody(MemBuf &buf)
2252 {
2253 // parent's implementation can handle the no-encoding case
2254 if (!flags.chunked_request)
2255 return Client::getMoreRequestBody(buf);
2256
2257 MemBuf raw;
2258
2259 Must(requestBodySource != NULL);
2260 if (!requestBodySource->getMoreData(raw))
2261 return false; // no request body bytes to chunk yet
2262
2263 // optimization: pre-allocate buffer size that should be enough
2264 const mb_size_t rawDataSize = raw.contentSize();
2265 // we may need to send: hex-chunk-size CRLF raw-data CRLF last-chunk
2266 buf.init(16 + 2 + rawDataSize + 2 + 5, raw.max_capacity);
2267
2268 buf.appendf("%x\r\n", static_cast<unsigned int>(rawDataSize));
2269 buf.append(raw.content(), rawDataSize);
2270 buf.append("\r\n", 2);
2271
2272 Must(rawDataSize > 0); // we did not accidently created last-chunk above
2273
2274 // Do not send last-chunk unless we successfully received everything
2275 if (receivedWholeRequestBody) {
2276 Must(!flags.sentLastChunk);
2277 flags.sentLastChunk = true;
2278 buf.append("0\r\n\r\n", 5);
2279 }
2280
2281 return true;
2282 }
2283
2284 void
2285 httpStart(FwdState *fwd)
2286 {
2287 debugs(11, 3, fwd->request->method << ' ' << fwd->entry->url());
2288 AsyncJob::Start(new HttpStateData(fwd));
2289 }
2290
2291 void
2292 HttpStateData::start()
2293 {
2294 if (!sendRequest()) {
2295 debugs(11, 3, "httpStart: aborted");
2296 mustStop("HttpStateData::start failed");
2297 return;
2298 }
2299
2300 ++ statCounter.server.all.requests;
2301 ++ statCounter.server.http.requests;
2302
2303 /*
2304 * We used to set the read timeout here, but not any more.
2305 * Now its set in httpSendComplete() after the full request,
2306 * including request body, has been written to the server.
2307 */
2308 }
2309
2310 /// if broken posts are enabled for the request, try to fix and return true
2311 bool
2312 HttpStateData::finishingBrokenPost()
2313 {
2314 #if USE_HTTP_VIOLATIONS
2315 if (!Config.accessList.brokenPosts) {
2316 debugs(11, 5, HERE << "No brokenPosts list");
2317 return false;
2318 }
2319
2320 ACLFilledChecklist ch(Config.accessList.brokenPosts, originalRequest().getRaw());
2321 if (ch.fastCheck() != ACCESS_ALLOWED) {
2322 debugs(11, 5, HERE << "didn't match brokenPosts");
2323 return false;
2324 }
2325
2326 if (!Comm::IsConnOpen(serverConnection)) {
2327 debugs(11, 3, HERE << "ignoring broken POST for closed " << serverConnection);
2328 assert(closeHandler != NULL);
2329 return true; // prevent caller from proceeding as if nothing happened
2330 }
2331
2332 debugs(11, 3, "finishingBrokenPost: fixing broken POST");
2333 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2334 requestSender = JobCallback(11,5,
2335 Dialer, this, HttpStateData::wroteLast);
2336 Comm::Write(serverConnection, "\r\n", 2, requestSender, NULL);
2337 return true;
2338 #else
2339 return false;
2340 #endif /* USE_HTTP_VIOLATIONS */
2341 }
2342
2343 /// if needed, write last-chunk to end the request body and return true
2344 bool
2345 HttpStateData::finishingChunkedRequest()
2346 {
2347 if (flags.sentLastChunk) {
2348 debugs(11, 5, HERE << "already sent last-chunk");
2349 return false;
2350 }
2351
2352 Must(receivedWholeRequestBody); // or we should not be sending last-chunk
2353 flags.sentLastChunk = true;
2354
2355 typedef CommCbMemFunT<HttpStateData, CommIoCbParams> Dialer;
2356 requestSender = JobCallback(11,5, Dialer, this, HttpStateData::wroteLast);
2357 Comm::Write(serverConnection, "0\r\n\r\n", 5, requestSender, NULL);
2358 return true;
2359 }
2360
2361 void
2362 HttpStateData::doneSendingRequestBody()
2363 {
2364 Client::doneSendingRequestBody();
2365 debugs(11,5, HERE << serverConnection);
2366
2367 // do we need to write something after the last body byte?
2368 if (flags.chunked_request && finishingChunkedRequest())
2369 return;
2370 if (!flags.chunked_request && finishingBrokenPost())
2371 return;
2372
2373 sendComplete();
2374 }
2375
2376 // more origin request body data is available
2377 void
2378 HttpStateData::handleMoreRequestBodyAvailable()
2379 {
2380 if (eof || !Comm::IsConnOpen(serverConnection)) {
2381 // XXX: we should check this condition in other callbacks then!
2382 // TODO: Check whether this can actually happen: We should unsubscribe
2383 // as a body consumer when the above condition(s) are detected.
2384 debugs(11, DBG_IMPORTANT, HERE << "Transaction aborted while reading HTTP body");
2385 return;
2386 }
2387
2388 assert(requestBodySource != NULL);
2389
2390 if (requestBodySource->buf().hasContent()) {
2391 // XXX: why does not this trigger a debug message on every request?
2392
2393 if (flags.headers_parsed && !flags.abuse_detected) {
2394 flags.abuse_detected = true;
2395 debugs(11, DBG_IMPORTANT, "http handleMoreRequestBodyAvailable: Likely proxy abuse detected '" << request->client_addr << "' -> '" << entry->url() << "'" );
2396
2397 if (virginReply()->sline.status() == Http::scInvalidHeader) {
2398 closeServer();
2399 mustStop("HttpStateData::handleMoreRequestBodyAvailable");
2400 return;
2401 }
2402 }
2403 }
2404
2405 HttpStateData::handleMoreRequestBodyAvailable();
2406 }
2407
2408 // premature end of the request body
2409 void
2410 HttpStateData::handleRequestBodyProducerAborted()
2411 {
2412 Client::handleRequestBodyProducerAborted();
2413 if (entry->isEmpty()) {
2414 debugs(11, 3, "request body aborted: " << serverConnection);
2415 // We usually get here when ICAP REQMOD aborts during body processing.
2416 // We might also get here if client-side aborts, but then our response
2417 // should not matter because either client-side will provide its own or
2418 // there will be no response at all (e.g., if the the client has left).
2419 ErrorState *err = new ErrorState(ERR_ICAP_FAILURE, Http::scInternalServerError, fwd->request);
2420 err->detailError(ERR_DETAIL_SRV_REQMOD_REQ_BODY);
2421 fwd->fail(err);
2422 }
2423
2424 abortTransaction("request body producer aborted");
2425 }
2426
2427 // called when we wrote request headers(!) or a part of the body
2428 void
2429 HttpStateData::sentRequestBody(const CommIoCbParams &io)
2430 {
2431 if (io.size > 0)
2432 statCounter.server.http.kbytes_out += io.size;
2433
2434 Client::sentRequestBody(io);
2435 }
2436
2437 void
2438 HttpStateData::abortAll(const char *reason)
2439 {
2440 debugs(11,5, HERE << "aborting transaction for " << reason <<
2441 "; " << serverConnection << ", this " << this);
2442 mustStop(reason);
2443 }
2444
2445 HttpStateData::ReuseDecision::ReuseDecision(const StoreEntry *e, const Http::StatusCode code)
2446 : answer(HttpStateData::ReuseDecision::reuseNot), reason(nullptr), entry(e), statusCode(code) {}
2447
2448 HttpStateData::ReuseDecision::Answers
2449 HttpStateData::ReuseDecision::make(const HttpStateData::ReuseDecision::Answers ans, const char *why)
2450 {
2451 answer = ans;
2452 reason = why;
2453 return answer;
2454 }
2455
2456 std::ostream &operator <<(std::ostream &os, const HttpStateData::ReuseDecision &d)
2457 {
2458 static const char *ReuseMessages[] = {
2459 "do not cache and do not share", // reuseNot
2460 "cache positively and share", // cachePositively
2461 "cache negatively and share", // cacheNegatively
2462 "do not cache but share" // doNotCacheButShare
2463 };
2464
2465 assert(d.answer >= HttpStateData::ReuseDecision::reuseNot &&
2466 d.answer <= HttpStateData::ReuseDecision::doNotCacheButShare);
2467 return os << ReuseMessages[d.answer] << " because " << d.reason <<
2468 "; HTTP status " << d.statusCode << " " << *(d.entry);
2469 }
2470