[issue6791] httplib read status memory usage
Senthil Kumaran orsent...@gmail.com added the comment: In the morning, I had a comment on the patch wondering why read _MAXLENGH + 1 and then check for len of header _MAXLENGH. Instead of just reading _MAXLENGH (and if the length matched rejecting). ( Looks like it did not go through). I think that either way is okay. I am taking the privilege of committing the patch. Fixed for py3k in 87373. So it is be available in the next beta. Shall merge the changes to other codelines. -- resolution: - fixed ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: Partially backported in r87382 (3.1) and r87383 (2.7). Not everything could be merged in because of HTTP 0.9 support and (in 2.7) a slightly different architecture. Thank you. -- stage: patch review - committed/rejected status: open - closed ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: Now that 0.9 client support has been removed, this can proceed (at least for 3.2). -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: Here is a patch limiting line length everywhere in http.client, + tests (it also affects http.server since the header parsing routine is shared). -- stage: needs patch - patch review Added file: http://bugs.python.org/file20100/httplinelength.patch ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: Well, removing 0.9 support doesn't make this obsolete, does it? -- status: pending - open ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Senthil Kumaran orsent...@gmail.com added the comment: On Thu, Dec 16, 2010 at 01:18:30PM +, Antoine Pitrou wrote: Well, removing 0.9 support doesn't make this obsolete, does it? It does. Doesn't it? Because I saw in your patch that you fall back on HTTP 1.0 behaviour when the server does not return a status line and in which case a Exception will be raise and this issue won't be observed. -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: It does. Doesn't it? Because I saw in your patch that you fall back on HTTP 1.0 behaviour when the server does not return a status line and in which case a Exception will be raise and this issue won't be observed. I don't think you understood the issue here. Calling readline() without a maximum length means the process memory potentially explodes, if the server sends gigabytes of data without a single \n. -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Senthil Kumaran orsent...@gmail.com added the comment: On Thu, Dec 16, 2010 at 02:02:10PM +, Antoine Pitrou wrote: I don't think you understood the issue here. Calling readline() without a maximum length means the process memory potentially explodes, if the server sends gigabytes of data without a single \n. Yeah, I seem to have misunderstood the issue. Even if the response wa s an *invalid* one but it was huge data without \n, the readline call would just explode. - reading chunked response is doing a readline call too. Both this need to be addressed by having a limit on reading. I thought readline() is being called only when parsing headers which should almost always have CRLF (or at least LF) and thought valid responses always start with headers. -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Ross Lagerwall rosslagerw...@gmail.com added the comment: Attached is a unit test which tests the issue. Unfortunately, since it uses the resource module to limit memory to a workable size, it will only work on Unix. The given patch appears to fix the issue well. I think this should be taken as a security issue (even if a rather odd one) since a malicious http server could be set up in place of the normal one and crash any http python clients that connect to it. Eg: Run: dd if=/dev/zero bs=10M count=1000 | nc -l And then: import httplib h = httplib.HTTPConnection('localhost', ) h.connect() h.request('GET', '/') r = h.getresponse() This should cause python to use up all the memory available. -- nosy: +rosslagerwall Added file: http://bugs.python.org/file20048/i6791_unittest.patch ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Ross Lagerwall rosslagerw...@gmail.com added the comment: A py3k patch against revision 87228. -- Added file: http://bugs.python.org/file20049/i6791_py3k.patch ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: First, I don't think the resource module needs to be used here. Second, I don't see why getcode() would return 200. If no valid response was received then some kind of error should certainly be raised, shouldn't it? -- nosy: +pitrou ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: By the way, looking at the code, readline() without any parameter is used all over http.client, so fixing only this one use case doesn't really make sense. -- stage: unit test needed - needs patch ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Ross Lagerwall rosslagerw...@gmail.com added the comment: That's true. Near the bottom of the code, it says: # The status-line parsing code calls readline(), which normally # get the HTTP status line. For a 0.9 response, however, this is # actually the first line of the body! Limiting the length of the status line would break 0.9 responses so maybe this issue should be closed? -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Antoine Pitrou pit...@free.fr added the comment: That's true. Near the bottom of the code, it says: # The status-line parsing code calls readline(), which normally # get the HTTP status line. For a 0.9 response, however, this is # actually the first line of the body! Limiting the length of the status line would break 0.9 responses so maybe this issue should be closed? Well, the HTTP 1.0 RFC was filed in 1996 and HTTP 1.1 is most commonly used today. I don't think we need to support 0.9 anymore. I'll open a separate issue for ripping off 0.9 support, though. -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Senthil Kumaran orsent...@gmail.com added the comment: I just read the whole discussion and it seems that code was in place so that client can tolerant of a BAD HTTP 0.9 Server response. http://www.w3.org/Protocols/HTTP/OldServers.html Given that issue10711 talks about removing HTTP/0.9 support (+1 to that), this issue will become obsolete. I too support removing HTTP/0.9. There are hardly any advantages in keeping it around. -- nosy: -BreamoreBoy status: open - pending ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Mark Lawrence breamore...@yahoo.co.uk added the comment: Sumar, to get this moved forward could you please provide a unit test. -- nosy: +BreamoreBoy stage: - unit test needed versions: +Python 2.7, Python 3.1, Python 3.2 -Python 2.4, Python 2.5, Python 2.6 ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Changes by Senthil Kumaran orsent...@gmail.com: -- assignee: - orsenthil nosy: +orsenthil ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
Changes by STINNER Victor victor.stin...@haypocalc.com: -- nosy: +haypo ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
New submission from sumar m.sucaj...@gmail.com: During writing some code I discovered some behaviour of httplib. When we connect to host, which doesn’t respond with status line, but it just sending data, httplib may consume more and more memory, becouce when we execute h = httplib.HTTPConnection(‘host’) h.conect() h.request(‘GET’, ‘/’) r = h.getresponse() httplib tries to read one line from host. If host doesn’t send new line character (‘\n’), httplib reads more and more data. On my tests httplib could consume all of 4GB of memory and the python process was killed by oom_killer. The resolution is to limit maximum amount of data read on getting response. I have performed some test: I received 3438293 from hosts located in the network. The longest valid response line is HTTP/1.1 500 ( The specified Secure Sockets Layer (SSL) port is not allowed. ISA Server is not configured to allow SSL requests from this port. Most Web browsers use port 443 for SSL requests. )\r\n and it has 197 characters. In RFC2616 in section 6.1 we have: “The first line of a Response message is the Status-Line, consisting of the protocol version followed by a numeric status code and its associated textual phrase, with each element separated by SP characters. No CR or LF is allowed except in the final CRLF sequence. Status-Line = HTTP-Version SP Status-Code SP Reason-Phrase CRLF (..)The Reason-Phrase is intended to give a short textual description of the Status-Code.” So limiting maximum status line length to 256 characters is a solution of this problem. It doesn’t break compatibility withc RFC 2616. My patch was written originally on python2.4, but I’ve tested it on python2.6: [...@host python2.6]$ patch --dry-run -i /home/ms/httplib.patch patching file httplib.py Hunk #1 succeeded at 209 (offset 54 lines). -- components: Library (Lib) files: httplib.patch keywords: patch messages: 92027 nosy: m.sucajtys severity: normal status: open title: httplib read status memory usage type: resource usage versions: Python 2.4, Python 2.5, Python 2.6 Added file: http://bugs.python.org/file14795/httplib.patch ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue6791] httplib read status memory usage
sumar m.sucaj...@gmail.com added the comment: I've also check patch against code in svn tree: wget http://svn.python.org/projects/python/trunk/Lib/httplib.py patch -p0 -i httplib.patch --dry-run patching file httplib.py Hunk #1 succeeded at 209 (offset 54 lines). Hunk #2 succeeded at 303 (offset 10 lines). -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue6791 ___ ___ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com