]>
git.saurik.com Git - apt.git/blob - methods/http.cc
1 // -*- mode: cpp; mode: fold -*-
3 // $Id: http.cc,v 1.59 2004/05/08 19:42:35 mdz Exp $
4 /* ######################################################################
6 HTTP Acquire Method - This is the HTTP acquire method for APT.
8 It uses HTTP/1.1 and many of the fancy options there-in, such as
9 pipelining, range, if-range and so on.
11 It is based on a doubly buffered select loop. A groupe of requests are
12 fed into a single output buffer that is constantly fed out the
13 socket. This provides ideal pipelining as in many cases all of the
14 requests will fit into a single packet. The input socket is buffered
15 the same way and fed into the fd for the file (may be a pipe in future).
17 This double buffering provides fairly substantial transfer rates,
18 compared to wget the http method is about 4% faster. Most importantly,
19 when HTTP is compared with FTP as a protocol the speed difference is
20 huge. In tests over the internet from two sites to llug (via ATM) this
21 program got 230k/s sustained http transfer rates. FTP on the other
22 hand topped out at 170k/s. That combined with the time to setup the
23 FTP connection makes HTTP a vastly superior protocol.
25 ##################################################################### */
27 // Include Files /*{{{*/
30 #include <apt-pkg/fileutl.h>
31 #include <apt-pkg/acquire-method.h>
32 #include <apt-pkg/configuration.h>
33 #include <apt-pkg/error.h>
34 #include <apt-pkg/hashes.h>
35 #include <apt-pkg/netrc.h>
36 #include <apt-pkg/strutl.h>
40 #include <sys/select.h>
58 unsigned long long CircleBuf::BwReadLimit
=0;
59 unsigned long long CircleBuf::BwTickReadData
=0;
60 struct timeval
CircleBuf::BwReadTick
={0,0};
61 const unsigned int CircleBuf::BW_HZ
=10;
63 // CircleBuf::CircleBuf - Circular input buffer /*{{{*/
64 // ---------------------------------------------------------------------
66 CircleBuf::CircleBuf(unsigned long long Size
)
67 : Size(Size
), Hash(0), TotalWriten(0)
69 Buf
= new unsigned char[Size
];
72 CircleBuf::BwReadLimit
= _config
->FindI("Acquire::http::Dl-Limit",0)*1024;
75 // CircleBuf::Reset - Reset to the default state /*{{{*/
76 // ---------------------------------------------------------------------
78 void CircleBuf::Reset()
84 MaxGet
= (unsigned long long)-1;
93 // CircleBuf::Read - Read from a FD into the circular buffer /*{{{*/
94 // ---------------------------------------------------------------------
95 /* This fills up the buffer with as much data as is in the FD, assuming it
97 bool CircleBuf::Read(int Fd
)
101 // Woops, buffer is full
102 if (InP
- OutP
== Size
)
105 // what's left to read in this tick
106 unsigned long long const BwReadMax
= CircleBuf::BwReadLimit
/BW_HZ
;
108 if(CircleBuf::BwReadLimit
) {
110 gettimeofday(&now
,0);
112 unsigned long long d
= (now
.tv_sec
-CircleBuf::BwReadTick
.tv_sec
)*1000000 +
113 now
.tv_usec
-CircleBuf::BwReadTick
.tv_usec
;
114 if(d
> 1000000/BW_HZ
) {
115 CircleBuf::BwReadTick
= now
;
116 CircleBuf::BwTickReadData
= 0;
119 if(CircleBuf::BwTickReadData
>= BwReadMax
) {
120 usleep(1000000/BW_HZ
);
125 // Write the buffer segment
127 if(CircleBuf::BwReadLimit
) {
128 Res
= read(Fd
,Buf
+ (InP%Size
),
129 BwReadMax
> LeftRead() ? LeftRead() : BwReadMax
);
131 Res
= read(Fd
,Buf
+ (InP%Size
),LeftRead());
133 if(Res
> 0 && BwReadLimit
> 0)
134 CircleBuf::BwTickReadData
+= Res
;
146 gettimeofday(&Start
,0);
151 // CircleBuf::Read - Put the string into the buffer /*{{{*/
152 // ---------------------------------------------------------------------
153 /* This will hold the string in and fill the buffer with it as it empties */
154 bool CircleBuf::Read(string Data
)
161 // CircleBuf::FillOut - Fill the buffer from the output queue /*{{{*/
162 // ---------------------------------------------------------------------
164 void CircleBuf::FillOut()
166 if (OutQueue
.empty() == true)
170 // Woops, buffer is full
171 if (InP
- OutP
== Size
)
174 // Write the buffer segment
175 unsigned long long Sz
= LeftRead();
176 if (OutQueue
.length() - StrPos
< Sz
)
177 Sz
= OutQueue
.length() - StrPos
;
178 memcpy(Buf
+ (InP%Size
),OutQueue
.c_str() + StrPos
,Sz
);
183 if (OutQueue
.length() == StrPos
)
192 // CircleBuf::Write - Write from the buffer into a FD /*{{{*/
193 // ---------------------------------------------------------------------
194 /* This empties the buffer into the FD. */
195 bool CircleBuf::Write(int Fd
)
201 // Woops, buffer is empty
208 // Write the buffer segment
210 Res
= write(Fd
,Buf
+ (OutP%Size
),LeftWrite());
225 Hash
->Add(Buf
+ (OutP%Size
),Res
);
231 // CircleBuf::WriteTillEl - Write from the buffer to a string /*{{{*/
232 // ---------------------------------------------------------------------
233 /* This copies till the first empty line */
234 bool CircleBuf::WriteTillEl(string
&Data
,bool Single
)
236 // We cheat and assume it is unneeded to have more than one buffer load
237 for (unsigned long long I
= OutP
; I
< InP
; I
++)
239 if (Buf
[I%Size
] != '\n')
245 if (I
< InP
&& Buf
[I%Size
] == '\r')
247 if (I
>= InP
|| Buf
[I%Size
] != '\n')
255 unsigned long long Sz
= LeftWrite();
260 Data
+= string((char *)(Buf
+ (OutP%Size
)),Sz
);
268 // CircleBuf::Stats - Print out stats information /*{{{*/
269 // ---------------------------------------------------------------------
271 void CircleBuf::Stats()
277 gettimeofday(&Stop
,0);
278 /* float Diff = Stop.tv_sec - Start.tv_sec +
279 (float)(Stop.tv_usec - Start.tv_usec)/1000000;
280 clog << "Got " << InP << " in " << Diff << " at " << InP/Diff << endl;*/
283 CircleBuf::~CircleBuf()
289 // HttpServerState::HttpServerState - Constructor /*{{{*/
290 HttpServerState::HttpServerState(URI Srv
,HttpMethod
*Owner
) : ServerState(Srv
, Owner
), In(64*1024), Out(4*1024)
292 TimeOut
= _config
->FindI("Acquire::http::Timeout",TimeOut
);
296 // HttpServerState::Open - Open a connection to the server /*{{{*/
297 // ---------------------------------------------------------------------
298 /* This opens a connection to the server. */
299 bool HttpServerState::Open()
301 // Use the already open connection if possible.
310 // Determine the proxy setting
311 string SpecificProxy
= _config
->Find("Acquire::http::Proxy::" + ServerName
.Host
);
312 if (!SpecificProxy
.empty())
314 if (SpecificProxy
== "DIRECT")
317 Proxy
= SpecificProxy
;
321 string DefProxy
= _config
->Find("Acquire::http::Proxy");
322 if (!DefProxy
.empty())
328 char* result
= getenv("http_proxy");
329 Proxy
= result
? result
: "";
333 // Parse no_proxy, a , separated list of domains
334 if (getenv("no_proxy") != 0)
336 if (CheckDomainList(ServerName
.Host
,getenv("no_proxy")) == true)
340 // Determine what host and port to use based on the proxy settings
343 if (Proxy
.empty() == true || Proxy
.Host
.empty() == true)
345 if (ServerName
.Port
!= 0)
346 Port
= ServerName
.Port
;
347 Host
= ServerName
.Host
;
356 // Connect to the remote server
357 if (Connect(Host
,Port
,"http",80,ServerFd
,TimeOut
,Owner
) == false)
363 // HttpServerState::Close - Close a connection to the server /*{{{*/
364 // ---------------------------------------------------------------------
366 bool HttpServerState::Close()
373 // HttpServerState::RunData - Transfer the data from the socket /*{{{*/
374 bool HttpServerState::RunData(FileFd
* const File
)
378 // Chunked transfer encoding is fun..
379 if (Encoding
== Chunked
)
383 // Grab the block size
389 if (In
.WriteTillEl(Data
,true) == true)
392 while ((Last
= Go(false, File
)) == true);
397 // See if we are done
398 unsigned long long Len
= strtoull(Data
.c_str(),0,16);
403 // We have to remove the entity trailer
407 if (In
.WriteTillEl(Data
,true) == true && Data
.length() <= 2)
410 while ((Last
= Go(false, File
)) == true);
413 return !_error
->PendingError();
416 // Transfer the block
418 while (Go(true, File
) == true)
419 if (In
.IsLimit() == true)
423 if (In
.IsLimit() == false)
426 // The server sends an extra new line before the next block specifier..
431 if (In
.WriteTillEl(Data
,true) == true)
434 while ((Last
= Go(false, File
)) == true);
441 /* Closes encoding is used when the server did not specify a size, the
442 loss of the connection means we are done */
443 if (Encoding
== Closes
)
446 In
.Limit(Size
- StartPos
);
448 // Just transfer the whole block.
451 if (In
.IsLimit() == false)
455 return !_error
->PendingError();
457 while (Go(true, File
) == true);
460 return Owner
->Flush() && !_error
->PendingError();
463 bool HttpServerState::ReadHeaderLines(std::string
&Data
) /*{{{*/
465 return In
.WriteTillEl(Data
);
468 bool HttpServerState::LoadNextResponse(bool const ToFile
, FileFd
* const File
)/*{{{*/
470 return Go(ToFile
, File
);
473 bool HttpServerState::WriteResponse(const std::string
&Data
) /*{{{*/
475 return Out
.Read(Data
);
478 APT_PURE
bool HttpServerState::IsOpen() /*{{{*/
480 return (ServerFd
!= -1);
483 bool HttpServerState::InitHashes(FileFd
&File
) /*{{{*/
486 In
.Hash
= new Hashes
;
488 // Set the expected size and read file for the hashes
489 File
.Truncate(StartPos
);
490 return In
.Hash
->AddFD(File
, StartPos
);
493 APT_PURE Hashes
* HttpServerState::GetHashes() /*{{{*/
498 // HttpServerState::Die - The server has closed the connection. /*{{{*/
499 bool HttpServerState::Die(FileFd
&File
)
501 unsigned int LErrno
= errno
;
503 // Dump the buffer to the file
504 if (State
== ServerState::Data
)
506 // on GNU/kFreeBSD, apt dies on /dev/null because non-blocking
508 if (File
.Name() != "/dev/null")
509 SetNonBlock(File
.Fd(),false);
510 while (In
.WriteSpace() == true)
512 if (In
.Write(File
.Fd()) == false)
513 return _error
->Errno("write",_("Error writing to the file"));
516 if (In
.IsLimit() == true)
521 // See if this is because the server finished the data stream
522 if (In
.IsLimit() == false && State
!= HttpServerState::Header
&&
523 Encoding
!= HttpServerState::Closes
)
527 return _error
->Error(_("Error reading from server. Remote end closed connection"));
529 return _error
->Errno("read",_("Error reading from server"));
535 // Nothing left in the buffer
536 if (In
.WriteSpace() == false)
539 // We may have got multiple responses back in one packet..
547 // HttpServerState::Flush - Dump the buffer into the file /*{{{*/
548 // ---------------------------------------------------------------------
549 /* This takes the current input buffer from the Server FD and writes it
551 bool HttpServerState::Flush(FileFd
* const File
)
555 // on GNU/kFreeBSD, apt dies on /dev/null because non-blocking
557 if (File
->Name() != "/dev/null")
558 SetNonBlock(File
->Fd(),false);
559 if (In
.WriteSpace() == false)
562 while (In
.WriteSpace() == true)
564 if (In
.Write(File
->Fd()) == false)
565 return _error
->Errno("write",_("Error writing to file"));
566 if (In
.IsLimit() == true)
570 if (In
.IsLimit() == true || Encoding
== ServerState::Closes
)
576 // HttpServerState::Go - Run a single loop /*{{{*/
577 // ---------------------------------------------------------------------
578 /* This runs the select loop over the server FDs, Output file FDs and
580 bool HttpServerState::Go(bool ToFile
, FileFd
* const File
)
582 // Server has closed the connection
583 if (ServerFd
== -1 && (In
.WriteSpace() == false ||
591 /* Add the server. We only send more requests if the connection will
593 if (Out
.WriteSpace() == true && ServerFd
!= -1
594 && Persistent
== true)
595 FD_SET(ServerFd
,&wfds
);
596 if (In
.ReadSpace() == true && ServerFd
!= -1)
597 FD_SET(ServerFd
,&rfds
);
604 if (In
.WriteSpace() == true && ToFile
== true && FileFD
!= -1)
605 FD_SET(FileFD
,&wfds
);
608 if (_config
->FindB("Acquire::http::DependOnSTDIN", true) == true)
609 FD_SET(STDIN_FILENO
,&rfds
);
611 // Figure out the max fd
613 if (MaxFd
< ServerFd
)
621 if ((Res
= select(MaxFd
+1,&rfds
,&wfds
,0,&tv
)) < 0)
625 return _error
->Errno("select",_("Select failed"));
630 _error
->Error(_("Connection timed out"));
635 if (ServerFd
!= -1 && FD_ISSET(ServerFd
,&rfds
))
638 if (In
.Read(ServerFd
) == false)
642 if (ServerFd
!= -1 && FD_ISSET(ServerFd
,&wfds
))
645 if (Out
.Write(ServerFd
) == false)
649 // Send data to the file
650 if (FileFD
!= -1 && FD_ISSET(FileFD
,&wfds
))
652 if (In
.Write(FileFD
) == false)
653 return _error
->Errno("write",_("Error writing to output file"));
656 if (ExpectedSize
> 0 && In
.TotalWriten
> ExpectedSize
)
657 return _error
->Error("Writing more data than expected (%llu > %llu)",
658 In
.TotalWriten
, ExpectedSize
);
660 // Handle commands from APT
661 if (FD_ISSET(STDIN_FILENO
,&rfds
))
663 if (Owner
->Run(true) != -1)
671 // HttpMethod::SendReq - Send the HTTP request /*{{{*/
672 // ---------------------------------------------------------------------
673 /* This places the http request in the outbound buffer */
674 void HttpMethod::SendReq(FetchItem
*Itm
)
678 // The HTTP server expects a hostname with a trailing :port
679 std::stringstream Req
;
682 if (Uri
.Host
.find(':') != string::npos
)
683 ProperHost
= '[' + Uri
.Host
+ ']';
685 ProperHost
= Uri
.Host
;
687 /* RFC 2616 ยง5.1.2 requires absolute URIs for requests to proxies,
688 but while its a must for all servers to accept absolute URIs,
689 it is assumed clients will sent an absolute path for non-proxies */
690 std::string requesturi
;
691 if (Server
->Proxy
.empty() == true || Server
->Proxy
.Host
.empty())
692 requesturi
= Uri
.Path
;
694 requesturi
= Itm
->Uri
;
696 // The "+" is encoded as a workaround for a amazon S3 bug
697 // see LP bugs #1003633 and #1086997.
698 requesturi
= QuoteString(requesturi
, "+~ ");
700 /* Build the request. No keep-alive is included as it is the default
701 in 1.1, can cause problems with proxies, and we are an HTTP/1.1
703 C.f. https://tools.ietf.org/wg/httpbis/trac/ticket/158 */
704 Req
<< "GET " << requesturi
<< " HTTP/1.1\r\n";
706 Req
<< "Host: " << ProperHost
<< ":" << Uri
.Port
<< "\r\n";
708 Req
<< "Host: " << ProperHost
<< "\r\n";
710 // generate a cache control header (if needed)
711 if (_config
->FindB("Acquire::http::No-Cache",false) == true)
712 Req
<< "Cache-Control: no-cache\r\n"
713 << "Pragma: no-cache\r\n";
714 else if (Itm
->IndexFile
== true)
715 Req
<< "Cache-Control: max-age=" << _config
->FindI("Acquire::http::Max-Age",0) << "\r\n";
716 else if (_config
->FindB("Acquire::http::No-Store",false) == true)
717 Req
<< "Cache-Control: no-store\r\n";
719 // If we ask for uncompressed files servers might respond with content-
720 // negotiation which lets us end up with compressed files we do not support,
721 // see 657029, 657560 and co, so if we have no extension on the request
722 // ask for text only. As a sidenote: If there is nothing to negotate servers
723 // seem to be nice and ignore it.
724 if (_config
->FindB("Acquire::http::SendAccept", true) == true)
726 size_t const filepos
= Itm
->Uri
.find_last_of('/');
727 string
const file
= Itm
->Uri
.substr(filepos
+ 1);
728 if (flExtension(file
) == file
)
729 Req
<< "Accept: text/*\r\n";
732 // Check for a partial file and send if-queries accordingly
734 if (stat(Itm
->DestFile
.c_str(),&SBuf
) >= 0 && SBuf
.st_size
> 0)
735 Req
<< "Range: bytes=" << SBuf
.st_size
<< "-\r\n"
736 << "If-Range: " << TimeRFC1123(SBuf
.st_mtime
) << "\r\n";
737 else if (Itm
->LastModified
!= 0)
738 Req
<< "If-Modified-Since: " << TimeRFC1123(Itm
->LastModified
).c_str() << "\r\n";
740 if (Server
->Proxy
.User
.empty() == false || Server
->Proxy
.Password
.empty() == false)
741 Req
<< "Proxy-Authorization: Basic "
742 << Base64Encode(Server
->Proxy
.User
+ ":" + Server
->Proxy
.Password
) << "\r\n";
744 maybe_add_auth (Uri
, _config
->FindFile("Dir::Etc::netrc"));
745 if (Uri
.User
.empty() == false || Uri
.Password
.empty() == false)
746 Req
<< "Authorization: Basic "
747 << Base64Encode(Uri
.User
+ ":" + Uri
.Password
) << "\r\n";
749 Req
<< "User-Agent: " << _config
->Find("Acquire::http::User-Agent",
750 "Debian APT-HTTP/1.3 (" PACKAGE_VERSION
")") << "\r\n";
757 Server
->WriteResponse(Req
.str());
760 // HttpMethod::Configuration - Handle a configuration message /*{{{*/
761 // ---------------------------------------------------------------------
762 /* We stash the desired pipeline depth */
763 bool HttpMethod::Configuration(string Message
)
765 if (ServerMethod::Configuration(Message
) == false)
768 AllowRedirect
= _config
->FindB("Acquire::http::AllowRedirect",true);
769 PipelineDepth
= _config
->FindI("Acquire::http::Pipeline-Depth",
771 Debug
= _config
->FindB("Debug::Acquire::http",false);
773 // Get the proxy to use
779 // HttpMethod::AutoDetectProxy - auto detect proxy /*{{{*/
780 // ---------------------------------------------------------------------
782 bool HttpMethod::AutoDetectProxy()
784 // option is "Acquire::http::Proxy-Auto-Detect" but we allow the old
785 // name without the dash ("-")
786 AutoDetectProxyCmd
= _config
->Find("Acquire::http::Proxy-Auto-Detect",
787 _config
->Find("Acquire::http::ProxyAutoDetect"));
789 if (AutoDetectProxyCmd
.empty())
793 clog
<< "Using auto proxy detect command: " << AutoDetectProxyCmd
<< endl
;
795 int Pipes
[2] = {-1,-1};
796 if (pipe(Pipes
) != 0)
797 return _error
->Errno("pipe", "Failed to create Pipe");
799 pid_t Process
= ExecFork();
803 dup2(Pipes
[1],STDOUT_FILENO
);
804 SetCloseExec(STDOUT_FILENO
,false);
807 Args
[0] = AutoDetectProxyCmd
.c_str();
809 execv(Args
[0],(char **)Args
);
810 cerr
<< "Failed to exec method " << Args
[0] << endl
;
816 int res
= read(InFd
, buf
, sizeof(buf
)-1);
817 ExecWait(Process
, "ProxyAutoDetect", true);
820 return _error
->Errno("read", "Failed to read");
822 return _error
->Warning("ProxyAutoDetect returned no data");
828 clog
<< "auto detect command returned: '" << buf
<< "'" << endl
;
830 if (strstr(buf
, "http://") == buf
)
831 _config
->Set("Acquire::http::proxy", _strstrip(buf
));
836 ServerState
* HttpMethod::CreateServerState(URI uri
) /*{{{*/
838 return new HttpServerState(uri
, this);
841 void HttpMethod::RotateDNS() /*{{{*/