]> git.saurik.com Git - apt.git/blob - methods/server.cc
add gzip test and todo
[apt.git] / methods / server.cc
1 // -*- mode: cpp; mode: fold -*-
2 // Description /*{{{*/
3 /* ######################################################################
4
5 HTTP and HTTPS share a lot of common code and these classes are
6 exactly the dumping ground for this common code
7
8 ##################################################################### */
9 /*}}}*/
10 // Include Files /*{{{*/
11 #include <config.h>
12
13 #include <apt-pkg/acquire-method.h>
14 #include <apt-pkg/configuration.h>
15 #include <apt-pkg/error.h>
16 #include <apt-pkg/fileutl.h>
17 #include <apt-pkg/strutl.h>
18
19 #include <ctype.h>
20 #include <signal.h>
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <sys/stat.h>
24 #include <sys/time.h>
25 #include <time.h>
26 #include <unistd.h>
27 #include <iostream>
28 #include <limits>
29 #include <map>
30 #include <string>
31 #include <vector>
32
33 #include "server.h"
34
35 #include <apti18n.h>
36 /*}}}*/
37 using namespace std;
38
39 string ServerMethod::FailFile;
40 int ServerMethod::FailFd = -1;
41 time_t ServerMethod::FailTime = 0;
42
43 // ServerState::RunHeaders - Get the headers before the data /*{{{*/
44 // ---------------------------------------------------------------------
45 /* Returns 0 if things are OK, 1 if an IO error occurred and 2 if a header
46 parse error occurred */
47 ServerState::RunHeadersResult ServerState::RunHeaders(FileFd * const File)
48 {
49 State = Header;
50
51 Owner->Status(_("Waiting for headers"));
52
53 Major = 0;
54 Minor = 0;
55 Result = 0;
56 Size = 0;
57 StartPos = 0;
58 Encoding = Closes;
59 HaveContent = false;
60 time(&Date);
61
62 do
63 {
64 string Data;
65 if (ReadHeaderLines(Data) == false)
66 continue;
67
68 if (Owner->Debug == true)
69 clog << Data;
70
71 for (string::const_iterator I = Data.begin(); I < Data.end(); ++I)
72 {
73 string::const_iterator J = I;
74 for (; J != Data.end() && *J != '\n' && *J != '\r'; ++J);
75 if (HeaderLine(string(I,J)) == false)
76 return RUN_HEADERS_PARSE_ERROR;
77 I = J;
78 }
79
80 // 100 Continue is a Nop...
81 if (Result == 100)
82 continue;
83
84 // Tidy up the connection persistence state.
85 if (Encoding == Closes && HaveContent == true)
86 Persistent = false;
87
88 return RUN_HEADERS_OK;
89 }
90 while (LoadNextResponse(false, File) == true);
91
92 return RUN_HEADERS_IO_ERROR;
93 }
94 /*}}}*/
95 // ServerState::HeaderLine - Process a header line /*{{{*/
96 // ---------------------------------------------------------------------
97 /* */
98 bool ServerState::HeaderLine(string Line)
99 {
100 if (Line.empty() == true)
101 return true;
102
103 string::size_type Pos = Line.find(' ');
104 if (Pos == string::npos || Pos+1 > Line.length())
105 {
106 // Blah, some servers use "connection:closes", evil.
107 Pos = Line.find(':');
108 if (Pos == string::npos || Pos + 2 > Line.length())
109 return _error->Error(_("Bad header line"));
110 Pos++;
111 }
112
113 // Parse off any trailing spaces between the : and the next word.
114 string::size_type Pos2 = Pos;
115 while (Pos2 < Line.length() && isspace(Line[Pos2]) != 0)
116 Pos2++;
117
118 string Tag = string(Line,0,Pos);
119 string Val = string(Line,Pos2);
120
121 if (stringcasecmp(Tag.c_str(),Tag.c_str()+4,"HTTP") == 0)
122 {
123 // Evil servers return no version
124 if (Line[4] == '/')
125 {
126 int const elements = sscanf(Line.c_str(),"HTTP/%3u.%3u %3u%359[^\n]",&Major,&Minor,&Result,Code);
127 if (elements == 3)
128 {
129 Code[0] = '\0';
130 if (Owner->Debug == true)
131 clog << "HTTP server doesn't give Reason-Phrase for " << Result << std::endl;
132 }
133 else if (elements != 4)
134 return _error->Error(_("The HTTP server sent an invalid reply header"));
135 }
136 else
137 {
138 Major = 0;
139 Minor = 9;
140 if (sscanf(Line.c_str(),"HTTP %3u%359[^\n]",&Result,Code) != 2)
141 return _error->Error(_("The HTTP server sent an invalid reply header"));
142 }
143
144 /* Check the HTTP response header to get the default persistence
145 state. */
146 if (Major < 1)
147 Persistent = false;
148 else
149 {
150 if (Major == 1 && Minor == 0)
151 Persistent = false;
152 else
153 Persistent = true;
154 }
155
156 return true;
157 }
158
159 if (stringcasecmp(Tag,"Content-Length:") == 0)
160 {
161 if (Encoding == Closes)
162 Encoding = Stream;
163 HaveContent = true;
164
165 // The length is already set from the Content-Range header
166 if (StartPos != 0)
167 return true;
168
169 Size = strtoull(Val.c_str(), NULL, 10);
170 if (Size >= std::numeric_limits<unsigned long long>::max())
171 return _error->Errno("HeaderLine", _("The HTTP server sent an invalid Content-Length header"));
172 else if (Size == 0)
173 HaveContent = false;
174 return true;
175 }
176
177 if (stringcasecmp(Tag,"Content-Type:") == 0)
178 {
179 HaveContent = true;
180 return true;
181 }
182
183 if (stringcasecmp(Tag,"Content-Range:") == 0)
184 {
185 HaveContent = true;
186
187 // §14.16 says 'byte-range-resp-spec' should be a '*' in case of 416
188 if (Result == 416 && sscanf(Val.c_str(), "bytes */%llu",&Size) == 1)
189 {
190 StartPos = 1; // ignore Content-Length, it would override Size
191 HaveContent = false;
192 }
193 else if (sscanf(Val.c_str(),"bytes %llu-%*u/%llu",&StartPos,&Size) != 2)
194 return _error->Error(_("The HTTP server sent an invalid Content-Range header"));
195 if ((unsigned long long)StartPos > Size)
196 return _error->Error(_("This HTTP server has broken range support"));
197 return true;
198 }
199
200 if (stringcasecmp(Tag,"Transfer-Encoding:") == 0)
201 {
202 HaveContent = true;
203 if (stringcasecmp(Val,"chunked") == 0)
204 Encoding = Chunked;
205 return true;
206 }
207
208 if (stringcasecmp(Tag,"Connection:") == 0)
209 {
210 if (stringcasecmp(Val,"close") == 0)
211 Persistent = false;
212 if (stringcasecmp(Val,"keep-alive") == 0)
213 Persistent = true;
214 return true;
215 }
216
217 if (stringcasecmp(Tag,"Last-Modified:") == 0)
218 {
219 if (RFC1123StrToTime(Val.c_str(), Date) == false)
220 return _error->Error(_("Unknown date format"));
221 return true;
222 }
223
224 if (stringcasecmp(Tag,"Location:") == 0)
225 {
226 Location = Val;
227 return true;
228 }
229
230 return true;
231 }
232 /*}}}*/
233 // ServerState::ServerState - Constructor /*{{{*/
234 ServerState::ServerState(URI Srv, ServerMethod *Owner) : ServerName(Srv), TimeOut(120), Owner(Owner)
235 {
236 Reset();
237 }
238 /*}}}*/
239
240 bool ServerMethod::Configuration(string Message) /*{{{*/
241 {
242 return pkgAcqMethod::Configuration(Message);
243 }
244 /*}}}*/
245
246 // ServerMethod::DealWithHeaders - Handle the retrieved header data /*{{{*/
247 // ---------------------------------------------------------------------
248 /* We look at the header data we got back from the server and decide what
249 to do. Returns DealWithHeadersResult (see http.h for details).
250 */
251 ServerMethod::DealWithHeadersResult
252 ServerMethod::DealWithHeaders(FetchResult &Res)
253 {
254 // Not Modified
255 if (Server->Result == 304)
256 {
257 unlink(Queue->DestFile.c_str());
258 Res.IMSHit = true;
259 Res.LastModified = Queue->LastModified;
260 return IMS_HIT;
261 }
262
263 /* Redirect
264 *
265 * Note that it is only OK for us to treat all redirection the same
266 * because we *always* use GET, not other HTTP methods. There are
267 * three redirection codes for which it is not appropriate that we
268 * redirect. Pass on those codes so the error handling kicks in.
269 */
270 if (AllowRedirect
271 && (Server->Result > 300 && Server->Result < 400)
272 && (Server->Result != 300 // Multiple Choices
273 && Server->Result != 304 // Not Modified
274 && Server->Result != 306)) // (Not part of HTTP/1.1, reserved)
275 {
276 if (Server->Location.empty() == true);
277 else if (Server->Location[0] == '/' && Queue->Uri.empty() == false)
278 {
279 URI Uri = Queue->Uri;
280 if (Uri.Host.empty() == false)
281 NextURI = URI::SiteOnly(Uri);
282 else
283 NextURI.clear();
284 NextURI.append(DeQuoteString(Server->Location));
285 return TRY_AGAIN_OR_REDIRECT;
286 }
287 else
288 {
289 NextURI = DeQuoteString(Server->Location);
290 URI tmpURI = NextURI;
291 URI Uri = Queue->Uri;
292 // same protocol redirects are okay
293 if (tmpURI.Access == Uri.Access)
294 return TRY_AGAIN_OR_REDIRECT;
295 // as well as http to https
296 else if (Uri.Access == "http" && tmpURI.Access == "https")
297 return TRY_AGAIN_OR_REDIRECT;
298 }
299 /* else pass through for error message */
300 }
301 // retry after an invalid range response without partial data
302 else if (Server->Result == 416)
303 {
304 struct stat SBuf;
305 if (stat(Queue->DestFile.c_str(),&SBuf) >= 0 && SBuf.st_size > 0)
306 {
307 if ((unsigned long long)SBuf.st_size == Server->Size)
308 {
309 // the file is completely downloaded, but was not moved
310 Server->StartPos = Server->Size;
311 Server->Result = 200;
312 Server->HaveContent = false;
313 }
314 else if (unlink(Queue->DestFile.c_str()) == 0)
315 {
316 NextURI = Queue->Uri;
317 return TRY_AGAIN_OR_REDIRECT;
318 }
319 }
320 }
321
322 /* We have a reply we dont handle. This should indicate a perm server
323 failure */
324 if (Server->Result < 200 || Server->Result >= 300)
325 {
326 char err[255];
327 snprintf(err,sizeof(err)-1,"HttpError%i",Server->Result);
328 SetFailReason(err);
329 _error->Error("%u %s",Server->Result,Server->Code);
330 if (Server->HaveContent == true)
331 return ERROR_WITH_CONTENT_PAGE;
332 return ERROR_UNRECOVERABLE;
333 }
334
335 // This is some sort of 2xx 'data follows' reply
336 Res.LastModified = Server->Date;
337 Res.Size = Server->Size;
338
339 // Open the file
340 delete File;
341 File = new FileFd(Queue->DestFile,FileFd::WriteAny);
342 if (_error->PendingError() == true)
343 return ERROR_NOT_FROM_SERVER;
344
345 FailFile = Queue->DestFile;
346 FailFile.c_str(); // Make sure we dont do a malloc in the signal handler
347 FailFd = File->Fd();
348 FailTime = Server->Date;
349
350 if (Server->InitHashes(*File) == false)
351 {
352 _error->Errno("read",_("Problem hashing file"));
353 return ERROR_NOT_FROM_SERVER;
354 }
355 if (Server->StartPos > 0)
356 Res.ResumePoint = Server->StartPos;
357
358 SetNonBlock(File->Fd(),true);
359 return FILE_IS_OPEN;
360 }
361 /*}}}*/
362 // ServerMethod::SigTerm - Handle a fatal signal /*{{{*/
363 // ---------------------------------------------------------------------
364 /* This closes and timestamps the open file. This is necessary to get
365 resume behavoir on user abort */
366 void ServerMethod::SigTerm(int)
367 {
368 if (FailFd == -1)
369 _exit(100);
370
371 struct timeval times[2];
372 times[0].tv_sec = FailTime;
373 times[1].tv_sec = FailTime;
374 times[0].tv_usec = times[1].tv_usec = 0;
375 utimes(FailFile.c_str(), times);
376 close(FailFd);
377
378 _exit(100);
379 }
380 /*}}}*/
381 // ServerMethod::Fetch - Fetch an item /*{{{*/
382 // ---------------------------------------------------------------------
383 /* This adds an item to the pipeline. We keep the pipeline at a fixed
384 depth. */
385 bool ServerMethod::Fetch(FetchItem *)
386 {
387 if (Server == 0)
388 return true;
389
390 // Queue the requests
391 int Depth = -1;
392 for (FetchItem *I = Queue; I != 0 && Depth < (signed)PipelineDepth;
393 I = I->Next, Depth++)
394 {
395 if (Depth >= 0)
396 {
397 // If pipelining is disabled, we only queue 1 request
398 if (Server->Pipeline == false)
399 break;
400 // if we have no hashes, do at most one such request
401 // as we can't fixup pipeling misbehaviors otherwise
402 else if (I->ExpectedHashes.usable() == false)
403 break;
404 }
405
406 // Make sure we stick with the same server
407 if (Server->Comp(I->Uri) == false)
408 break;
409 if (QueueBack == I)
410 {
411 QueueBack = I->Next;
412 SendReq(I);
413 continue;
414 }
415 }
416
417 return true;
418 }
419 /*}}}*/
420 // ServerMethod::Loop - Main loop /*{{{*/
421 int ServerMethod::Loop()
422 {
423 typedef vector<string> StringVector;
424 typedef vector<string>::iterator StringVectorIterator;
425 map<string, StringVector> Redirected;
426
427 signal(SIGTERM,SigTerm);
428 signal(SIGINT,SigTerm);
429
430 Server = 0;
431
432 int FailCounter = 0;
433 while (1)
434 {
435 // We have no commands, wait for some to arrive
436 if (Queue == 0)
437 {
438 if (WaitFd(STDIN_FILENO) == false)
439 return 0;
440 }
441
442 /* Run messages, we can accept 0 (no message) if we didn't
443 do a WaitFd above.. Otherwise the FD is closed. */
444 int Result = Run(true);
445 if (Result != -1 && (Result != 0 || Queue == 0))
446 {
447 if(FailReason.empty() == false ||
448 _config->FindB("Acquire::http::DependOnSTDIN", true) == true)
449 return 100;
450 else
451 return 0;
452 }
453
454 if (Queue == 0)
455 continue;
456
457 // Connect to the server
458 if (Server == 0 || Server->Comp(Queue->Uri) == false)
459 {
460 delete Server;
461 Server = CreateServerState(Queue->Uri);
462 }
463 /* If the server has explicitly said this is the last connection
464 then we pre-emptively shut down the pipeline and tear down
465 the connection. This will speed up HTTP/1.0 servers a tad
466 since we don't have to wait for the close sequence to
467 complete */
468 if (Server->Persistent == false)
469 Server->Close();
470
471 // Reset the pipeline
472 if (Server->IsOpen() == false)
473 QueueBack = Queue;
474
475 // Connnect to the host
476 if (Server->Open() == false)
477 {
478 Fail(true);
479 delete Server;
480 Server = 0;
481 continue;
482 }
483
484 // Fill the pipeline.
485 Fetch(0);
486
487 // Fetch the next URL header data from the server.
488 switch (Server->RunHeaders(File))
489 {
490 case ServerState::RUN_HEADERS_OK:
491 break;
492
493 // The header data is bad
494 case ServerState::RUN_HEADERS_PARSE_ERROR:
495 {
496 _error->Error(_("Bad header data"));
497 Fail(true);
498 RotateDNS();
499 continue;
500 }
501
502 // The server closed a connection during the header get..
503 default:
504 case ServerState::RUN_HEADERS_IO_ERROR:
505 {
506 FailCounter++;
507 _error->Discard();
508 Server->Close();
509 Server->Pipeline = false;
510
511 if (FailCounter >= 2)
512 {
513 Fail(_("Connection failed"),true);
514 FailCounter = 0;
515 }
516
517 RotateDNS();
518 continue;
519 }
520 };
521
522 // Decide what to do.
523 FetchResult Res;
524 Res.Filename = Queue->DestFile;
525 switch (DealWithHeaders(Res))
526 {
527 // Ok, the file is Open
528 case FILE_IS_OPEN:
529 {
530 URIStart(Res);
531
532 // Run the data
533 bool Result = true;
534 if (Server->HaveContent)
535 Result = Server->RunData(File);
536
537 /* If the server is sending back sizeless responses then fill in
538 the size now */
539 if (Res.Size == 0)
540 Res.Size = File->Size();
541
542 // Close the file, destroy the FD object and timestamp it
543 FailFd = -1;
544 delete File;
545 File = 0;
546
547 // Timestamp
548 struct timeval times[2];
549 times[0].tv_sec = times[1].tv_sec = Server->Date;
550 times[0].tv_usec = times[1].tv_usec = 0;
551 utimes(Queue->DestFile.c_str(), times);
552
553 // Send status to APT
554 if (Result == true)
555 {
556 Hashes * const resultHashes = Server->GetHashes();
557 HashStringList const hashList = resultHashes->GetHashStringList();
558 if (PipelineDepth != 0 && Queue->ExpectedHashes.usable() == true && Queue->ExpectedHashes != hashList)
559 {
560 // we did not get the expected hash… mhhh:
561 // could it be that server/proxy messed up pipelining?
562 FetchItem * BeforeI = Queue;
563 for (FetchItem *I = Queue->Next; I != 0 && I != QueueBack; I = I->Next)
564 {
565 if (I->ExpectedHashes.usable() == true && I->ExpectedHashes == hashList)
566 {
567 // yes, he did! Disable pipelining and rewrite queue
568 if (Server->Pipeline == true)
569 {
570 // FIXME: fake a warning message as we have no proper way of communicating here
571 std::string out;
572 strprintf(out, _("Automatically disabled %s due to incorrect response from server/proxy. (man 5 apt.conf)"), "Acquire::http::PipelineDepth");
573 std::cerr << "W: " << out << std::endl;
574 Server->Pipeline = false;
575 // we keep the PipelineDepth value so that the rest of the queue can be fixed up as well
576 }
577 Rename(Res.Filename, I->DestFile);
578 Res.Filename = I->DestFile;
579 BeforeI->Next = I->Next;
580 I->Next = Queue;
581 Queue = I;
582 break;
583 }
584 BeforeI = I;
585 }
586 }
587 Res.TakeHashes(*resultHashes);
588 URIDone(Res);
589 }
590 else
591 {
592 if (Server->IsOpen() == false)
593 {
594 FailCounter++;
595 _error->Discard();
596 Server->Close();
597
598 if (FailCounter >= 2)
599 {
600 Fail(_("Connection failed"),true);
601 FailCounter = 0;
602 }
603
604 QueueBack = Queue;
605 }
606 else
607 Fail(true);
608 }
609 break;
610 }
611
612 // IMS hit
613 case IMS_HIT:
614 {
615 URIDone(Res);
616 break;
617 }
618
619 // Hard server error, not found or something
620 case ERROR_UNRECOVERABLE:
621 {
622 Fail();
623 break;
624 }
625
626 // Hard internal error, kill the connection and fail
627 case ERROR_NOT_FROM_SERVER:
628 {
629 delete File;
630 File = 0;
631
632 Fail();
633 RotateDNS();
634 Server->Close();
635 break;
636 }
637
638 // We need to flush the data, the header is like a 404 w/ error text
639 case ERROR_WITH_CONTENT_PAGE:
640 {
641 Fail();
642
643 // Send to content to dev/null
644 File = new FileFd("/dev/null",FileFd::WriteExists);
645 Server->RunData(File);
646 delete File;
647 File = 0;
648 break;
649 }
650
651 // Try again with a new URL
652 case TRY_AGAIN_OR_REDIRECT:
653 {
654 // Clear rest of response if there is content
655 if (Server->HaveContent)
656 {
657 File = new FileFd("/dev/null",FileFd::WriteExists);
658 Server->RunData(File);
659 delete File;
660 File = 0;
661 }
662
663 /* Detect redirect loops. No more redirects are allowed
664 after the same URI is seen twice in a queue item. */
665 StringVector &R = Redirected[Queue->DestFile];
666 bool StopRedirects = false;
667 if (R.empty() == true)
668 R.push_back(Queue->Uri);
669 else if (R[0] == "STOP" || R.size() > 10)
670 StopRedirects = true;
671 else
672 {
673 for (StringVectorIterator I = R.begin(); I != R.end(); ++I)
674 if (Queue->Uri == *I)
675 {
676 R[0] = "STOP";
677 break;
678 }
679
680 R.push_back(Queue->Uri);
681 }
682
683 if (StopRedirects == false)
684 Redirect(NextURI);
685 else
686 Fail();
687
688 break;
689 }
690
691 default:
692 Fail(_("Internal error"));
693 break;
694 }
695
696 FailCounter = 0;
697 }
698
699 return 0;
700 }
701 /*}}}*/