Bugs item #1871269, was opened at 2008-01-14 11:44
Message generated for change (Comment added) made by jmalicki
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=100976&aid=1871269&group_id=976

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: libcurl
Group: hang
Status: Open
Resolution: None
Priority: 7
Private: No
Submitted By: Joe Malicki (jmalicki)
Assigned to: Daniel Stenberg (bagder)
Summary: Curl hangs when sending large request-headers

Initial Comment:
Using libcurl 7.13.2, 7.15.5 or 7.17.1 with software that sends extremely large (> ~40KB, whatever is greater than one syscall) request-headers as part of a post. The following patch fixes it, passes regression tests the same as the old curl, and has made it through several months of QA integration in a larger project with no issues:

This patch is against 7.15.5, and works with patch "fuzz" against 7.13.2 as well (7.17.1 requires a slightly different patch, still haven't tested fix for 7.17.1):

Index: lib/http.c
===================================================================
--- lib/http.c (revision 67598)
+++ lib/http.c (revision 67599)
@@ -778,7 +778,7 @@
     memcpy(buffer, http->postdata, (size_t)http->postsize);
     fullsize = (size_t)http->postsize;

- if(http->backup.postsize) {
+ if(http->backup.postsize || http->backup.fread) {
       /* move backup data into focus and continue on that */
       http->postdata = http->backup.postdata;
       http->postsize = http->backup.postsize;
@@ -787,6 +787,7 @@

       http->sending++; /* move one step up */

+ http->backup.fread=NULL;
       http->backup.postsize=0;
     }
     else

----------------------------------------------------------------------

>Comment By: Joe Malicki (jmalicki)
Date: 2008-01-14 15:49

Message:
Logged In: YES
user_id=1981241
Originator: YES

If you notice, I attached a testcase that fails... does it not fail for
you?

In this testcase, it is clearly posting data. The reason why
http->backup.postsize is 0 is
that http->postsize is only ever initially set in curl 7.17.1 http.c:2643,
which is only
if CURLOPT_POSTFIELDS is set. Because I'm writing the posts with a
callback, this value is
never set (perhaps this is the real bug? The patch I posted was that way
because it was less intrusive).

I provided a testcase (curl-large-header-bug.c) so it could be reproduced
with code rather than description.

Are you proposing that the attached patch is rather the correct one? (it
is untested)

File Added: curl-large-headers2.diff

----------------------------------------------------------------------

Comment By: Daniel Stenberg (bagder)
Date: 2008-01-14 15:34

Message:
Logged In: YES
user_id=1110
Originator: NO

Can you please be more specific on as exact details as possible on what
you do to get this problem so that I can work on writing up a test case for
this?

I don't understand the patch. If http->backup.postsize is zero, then what
exactly is libcurl going to send when that condition equals true?

Also, while the 'fread' is now 'fread_func' in recent libcurls, how come
the patch isn't needed the exact same way there? I can't recall any
particular fixes in this area in a very long time?

----------------------------------------------------------------------

Comment By: Joe Malicki (jmalicki)
Date: 2008-01-14 11:45

Message:
Logged In: YES
user_id=1981241
Originator: YES

Above patch for 7.15.5 attached as a file.
File Added: curl-large-headers.diff

----------------------------------------------------------------------

You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=100976&aid=1871269&group_id=976
Received on 2008-01-14