1285 lines
56 KiB
Raw Blame History

_ _ ____ _
___| | | | _ \| |
/ __| | | | |_) | |
| (__| |_| | _ <| |___
\___|\___/|_| \_\_____|
Daniel (16 January 2006)
- David Shaw finally removed all traces of Gopher and we are now officially
not supporting it. It hasn't been functioning for years anyway, so this is
just finally stating what already was true. And a cleanup at the same time.
- Bryan Henderson turned the 'initialized' variable for curl_global_init()
into a counter, and thus you can now do multiple curl_global_init() and you
are then supposed to do the same amount of calls to curl_global_cleanup().
Bryan has also updated the docs accordingly.
Daniel (13 January 2006)
- Andrew Benham fixed a race condition in the test suite that could cause the
test script to kill all processes in the current process group!
Daniel (12 January 2006)
- Michael Jahn:
Fixed FTP_SKIP_PASV_IP and FTP_USE_EPSV to "do right" when used on FTP thru
HTTP proxy.
Fixed PROXYTUNNEL to work fine when you do ftp through a proxy. It would
previously overwrite internal memory and cause unpredicted behaviour!
Daniel (11 January 2006)
- I decided to document the "secret option" here now, as I've received *NO*
feedback at all on my mailing list requests from November 2005:
I'm looking for feedback and comments. I added some experimental code the
other day, that allows a libcurl user to select what method libcurl should
use to reach a file on a FTP(S) server.
This functionality is available in CVS code and in recent daily snapshots.
Let me explain...
The current name for the option is CURLOPT_FTP_FILEMETHOD (--ftp-method for
the command line tool) and you set it to a long (there are currenly no
defines for the argument values, just plain numericals). You can set three
different "methods" that do this:
1 multicwd - like today, curl will do a single CWD operation for each path
part in the given URL. For deep hierarchies this means very many
commands. This is how RFC1738 says it should be done. This is the
2 nocwd - no CWD at all is done, curl will do SIZE, RETR, STOR etc and give
a full path to the server.
3 singlecwd - make one CWD with the full target directory and then operate
on the file "normally".
(With the command line tool you do --ftp-method [METHOD], where [METHOD] is
one of "multicwd", "nocwd" or "singlecwd".)
What feedback I'm interested in:
1 - Do they work at all? Do you find servers where one of these don't work?
2 - What would proper names for the option and its arguments be, if we
consider this feature good enough to get included and documented in
upcoming releases?
3 - Should we make libcurl able to "walk through" these options in case of
(path related) failures, or should it fail and let the user redo any
possible retries?
(This option is not documented in any man page just yet since I'm not sure
these names will be used or if the functionality will end up exactly like
this. And for the same reasons we have no test cases for these yet.)
Daniel (10 January 2006)
- When using a bad path over FTP, as in when libcurl couldn't CWD into all
given subdirs, libcurl would still "remember" the full path as if it is the
current directory libcurl is in so that the next curl_easy_perform() would
get really confused if it tried the same path again - as it would not issue
any CWD commands at all, assuming it is already in the "proper" dir.
Starting now, a failed CWD command sets a flag that prevents the path to be
"remembered" after returning.
Daniel (7 January 2006)
- Michael Jahn fixed so that the second CONNECT when doing FTP over a HTTP
proxy actually used a new connection and not sent the second request on the
first socket!
Daniel (6 January 2006)
- Alexander Lazic made the buildconf run the buildconf in the ares dir if that
is present instead of trying to mimic that script in curl's buildconf
Daniel (3 January 2006)
- Andres Garcia made the TFTP test server build with mingw.
Daniel (16 December 2005)
- Jean Jacques Drouin pointed out that you could only have a user name or
password of 127 bytes or less embedded in a URL, where actually the code
uses a 255 byte buffer for it! Modified now to use the full buffer size.
Daniel (12 December 2005)
- Dov Murik corrected the HTTP_ONLY define to disable the TFTP support properly
Version 7.15.1 (7 December 2005)
Daniel (6 December 2005)
- Full text here: Pointed out by
Stefan Esser.
libcurl's URL parser function can overflow a malloced buffer in two ways, if
given a too long URL.
These overflows happen if you
1 - pass in a URL with no protocol (like "http://") prefix, using no slash
and the string is 256 bytes or longer. This leads to a single zero byte
overflow of the malloced buffer.
2 - pass in a URL with only a question mark as separator (no slash) between
the host and the query part of the URL. This leads to a single zero byte
overflow of the malloced buffer.
Both overflows can be made with the same input string, leading to two single
zero byte overwrites.
The affected flaw cannot be triggered by a redirect, but the long URL must
be passed in "directly" to libcurl. It makes this a "local" problem. Of
course, lots of programs may still pass in user-provided URLs to libcurl
without doing much syntax checking of their own, allowing a user to exploit
this vulnerability.
There is no known exploit at the time of this writing.
Daniel (2 December 2005)
- Jamie Newton pointed out that libcurl's file:// code would close() a zero
file descriptor if given a non-existing file.
Daniel (24 November 2005)
- Doug Kaufman provided a set of patches to make curl build fine on DJGPP
again using configure.
- Yang Tse provided a whole series of patches to clear up compiler warnings on
Daniel (17 November 2005)
- I extended a patch from David Shaw to make libcurl _always_ provide an error
string in the given error buffer to address the flaw mention on 21 sep 2005.
Daniel (16 November 2005)
- Applied Albert Chin's patch that makes the libcurl.pc pkgconfig file get
installed on 'make install' time.
Daniel (14 November 2005)
- Quagmire reported that he needed to raise a NTLM buffer for SSPI to work
properly for a case, and so we did. We raised it even for non-SSPI builds
but it should not do any harm.
- Jan Kunder's debian bug report identified a weird
error message for when you try to upload a file and the requested directory
doesn't exist on the target server.
- Yang Tse fixed compiler warnings in lib/ssluse.c with OpenSSL 0.9.8 and in
lib/memdebug.h that showed up in his msvc builds.
Daniel (13 November 2005)
- Debian bug report 338681 by Jan Kunder: make curl better detect and report
bad limit-rate units: Now curl will return
error if a bad unit is used.
- Thanks to this nice summary of poll() implementations: and further tests by Eugene
Kotlyarov, we now know that cygwin's poll returns only POLLHUP on remote
connectin closure so we check for that case (too) and re-enable poll for
cygwin builds.
Daniel (12 November 2005)
- Eugene Kotlyarov found out that cygwin's poll() function isn't doing things
right: so we now disable
poll() and use select() on cygwin too (we already do the same choice on Mac
- Dima Barsky patched problem #1348930: the GnuTLS code completely ignored
client certificates! (
Daniel (10 November 2005)
- David Lang fixed IPv6 support for TFTP!
- Introducing range stepping to the curl globbing support. Now you can specify
step counter by adding :[num] within the brackets when specifying a range:
If no step counter is set, it defaults to 1 as before:
Daniel (8 November 2005)
- Removed the use of AI_CANONNAME in the IPv6-enabled resolver functions since
we really have no use for reverse lookups of the address.
I truly hope these are the last reverse lookups we had lingering in the
- Dmitry Bartsevich discovered some issues in compatibilty of SSPI-enabled
version of libcurl with different Windows versions. Current version of
libcurl imports SSPI functions from secur32.dll. However, under Windows NT
4.0 these functions are located in security.dll, under Windows 9x - in
secur32.dll and Windows 2000 and XP contains both these DLLs (security.dll
just forwards calls to secur32.dll).
Dmitry's patch loads proper library dynamically depending on Windows
version. Function InitSecurityInterface() is used to obtain pointers to all
of SSPI function in one structure.
Daniel (31 October 2005)
- Vilmos Nebehaj improved libcurl's LDAP abilities:
The LDAP code in libcurl can't handle LDAP servers of LDAPv3 nor binary
attributes in LDAP objects. So, I made a quick patch to address these
The solution is simple: if we connect to an LDAP server, first try LDAPv3
(which is the preferred protocol as of now) and then fall back to LDAPv2.
In case of binary attributes, we first convert them to base64, just like the
openldap client does. It uses ldap_get_values_len() instead of
ldap_get_values() to be able to retrieve binary attributes correctly. I
defined the necessary LDAP macros in lib/ldap.c to be able to compile
libcurl without the presence of libldap
Daniel (27 October 2005)
- Nis Jorgensen filed bug report #1338648
( which really is more of a
feature request, but anyway. It pointed out that --max-redirs did not allow
it to be set to 0, which then would return an error code on the first
Location: found. Based on Nis' patch, now libcurl supports CURLOPT_MAXREDIRS
set to 0, or -1 for infinity. Added test case 274 to verify.
- tommink[at] reported in bug report #1337723
( that curl could not upload
binary data from stdin on Windows if the data contained control-Z (hex 1a)
since that is treated as end-of-file when read in text mode. Gisle Vanem
pointed out the fix, and I made both -T and --data-binary take advantage of
- Jaz Fresh pointed out that if you used "-r [number]" as was wrongly described
in the man page, curl would send an invalid HTTP Range: header. The correct
way would be to use "-r [number]-" or even "-r -[number]". Starting now,
curl will warn if this is discovered, and automatically append a dash to the
range before passing it to libcurl.
Daniel (25 October 2005)
- Amol Pattekar reported a bug with great detail and a fine example in bug
#1326306 ( When using the multi
interface and connecting to a host with multiple IP addresses, and one of
the addresses fails to connect (the server must exist and respond, just not
accept connections) libcurl leaks a socket descriptor. Thanks to the fine
report, I could find and fix this.
Daniel (22 October 2005)
- Dima Barsky reported a problem with GnuTLS-enabled libcurl in bug report
#1334338 ( When reading an SSL
stream from a server and the server requests a "rehandshake", the current
code simply returns this as an error. I have no good way to test this, but
I've added a crude attempt of dealing with this situation slightly better -
it makes a blocking handshake if this happens. Done like this because fixing
this the "proper" way (that would handshake asynchronously) will require
quite some work and I really need a good way to test this to do such a
Daniel (21 October 2005)
- "Ofer" reported a problem when libcurl re-used a connection and failed to do
it, it could then accidentally actually crash. Presumably, this concerns FTP
- Temprimus improved the MSVC makefile so that the static debug SSL libs are
linked to the executable and not to the libcurld.lib
- Bradford Bruce made the windows resolver code properly return
errors (as documented).
Daniel (20 October 2005)
- Dave Dribin made libcurl understand and handle cases when the server
(wrongly) sends *two* WWW-Authenticate headers for Digest. While this should
never happen in a sane world, libcurl previously got into an infinite loop
when this occurred. Dave added test 273 to verify this.
- Temprimus improved the MSVC makefile: "makes a build option available so if
you set rtlibcfg=static for the make, then it would build with /MT. The
default behaviour is /MD (the original)."
Daniel (14 October 2005)
- Reverted the LIBCURL_VERSION_NUM change from October 6. As Dave Dribin
reported, the define is used by the configure script and is assumed to use
the 0xYYXXZZ format. This made "curl-config --vernum" fail in the 7.15.0
release version.
Version 7.15.0 (13 October 2005)
Daniel (12 October 2005)
- Michael Sutton of iDEFENSE reported and I fixed a securitfy flaw in the NTLM
code that would overflow a buffer if given a too long user name or domain
name. This would happen if you enable NTLM authentication and either
A - pass in a user name and domain name to libcurl that together are longer
than 192 bytes
B - allow (lib)curl to follow HTTP "redirects" (Location: and the
appropriate HTTP 30x response code) and the new URL contains a URL with
a user name and domain name that together are longer than 192 bytes
See for further details and updates
Daniel (5 October 2005)
- Darryl House reported a problem with using -z to download files from FTP.
It turned out that if the given time stamp was exact the same as the remote
time stamp, the file would still wrongly be downloaded. Added test case 272
to verify.
Daniel (4 October 2005)
- Domenico Andreoli fixed a man page malformat and removed odd (0xa0) bytes
from the configure script.
- Michael Wallner reported that the date parser had wrong offset stored for
the MEST and CEST time zones.
Daniel (27 September 2005)
- David Yan filed bug #1299181 (
that identified a silly problem with Content-Range: headers with the 'bytes'
keyword written in a different case than all lowercase! It would cause a
- TJ Saunders of the proftpd project identified and pointed out problems with
the modified FTPS negotiation change of August 19 2005. Thus, we revert the
change back to pre-7.14.1 status.
Daniel (21 September 2005)
- Fixed "cut off" sentence in the libcurl-tutorial man page:
- Clarified in the curl_easy_setopt man page what the default
- Clarified in the curl_easy_setopt man page that CURLOPT_ERRORBUFFER
sometimes doesn't fill in the buffer even though it is supposed to:
- When CURLE_URL_MALFORMAT is returned due to a missing URL, it now has an
error string set.
Daniel (19 September 2005)
- Dmitry Bartsevich made the SSPI support work on Windows 9x as well.
Daniel (15 September 2005)
- Added a TFTP server to the test suite and made the test suite capable of
using it.
Daniel (7 September 2005)
- Ben Madsen's detailed reports that funnily enough only occurred with certain
glibc versions turned out to be curl using an already closed file handle
during certain conditions (like when saving FTP server "headers").
- Scott Davis helped me track down a problem in the test HTTP server that made
test case 56 wrongly fail at times. It turned out it was due to the server
finding the end of a chunked-encoded POST too early.
Daniel (6 September 2005)
- Now curl warns if an unknown variable is used in the -w/--writeout argument.
Daniel (4 September 2005)
- I applied Nicolas Fran<61>ois' man page patch he posted to the Debian bug
tracker. It corrected two lines that started with apostrophes, which isn't
legal nroff format.
- Added --ftp-skip-pasv-ip to the command line tool, that sets the new
CURLOPT_FTP_SKIP_PASV_IP option. It makes libcurl re-use the control
connection's IP address when setting up the data connection instead of
extractting the IP address from the PASV response. It has turned out this
feature is frequently needed by people to circumvent silly servers and silly
firewalls, especially when FTPS is used and the PASV command-response is
sent encrtyped.
Sponsored by CU*Answers
Daniel (1 September 2005)
- John Kelly added TFTP support to libcurl. A bunch of new error codes was
added. TODO: add them to docs. add TFTP server to test suite. add TFTP to
list of protocols whereever those are mentioned.
Version 7.14.1 (1 September 2005)
Daniel (29 August 2005)
- Kevin Lussier pointed out a problem with curllib.dsp and how to fix it.
- Igor Polyakov fixed a rather nasty problem with the threaded name resolver
for Windows, that could lead to an Access Violation when the multi interface
was used due to an issue with how the resolver thread was and was not
- Simon Josefsson brought a patch that allows curl to get built to use GNU GSS
instead of MIT/Heimdal for GSS capabilities.
Daniel (24 August 2005)
- Toby Peterson added CURLOPT_IGNORE_CONTENT_LENGTH to the library, accessible
from the command line tool with --ignore-content-length. This will make it
easier to download files from Apache 1.x (and similar) servers that are
still having problems serving files larger than 2 or 4 GB. When this option
is enabled, curl will simply have to wait for the server to close the
connection to signal end of transfer. I wrote test case 269 that runs a
simple test to verify that this works.
- (Trying hard to exclude emotions now.) valgrind version 3 suddenly renamed
the --logfile command line option to --log-file, and thus the test script
valgrind autodetection now has yet another version check to do and then it
alters the valgrind command line accordingly.
- Fixed CA cert verification using GnuTLS with the default bundle, which
previously failed due to GnuTLS not allowing x509 v1 CA certs by default.
Ralph Mitchell reported.
Daniel (19 August 2005)
- Norbert Novotny had problems with FTPS and he helped me work out a patch
that made curl run fine in his end. The key was to make sure we do the
SSL/TLS negotiation immediately after the TCP connect is done and not after
a few other commands have been sent like we did previously. I don't consider
this change necessary to obey the standards, I think this server is pickier
than what the specs allow it to be, but I can't see how this modified
libcurl code can add any problems to those who are interpreting the
standards more liberally.
Daniel (17 August 2005)
- Jeff Pohlmeyer found out that if you ask libcurl to load a cookiefile (with
CURLOPT_COOKIEFILE), add a cookie (with CURLOPT_COOKIELIST), tell it to
write the result to a given cookie jar and then never actually call
curl_easy_perform() - the given file(s) to read was never read but the
output file was written and thus it caused a "funny" result.
- While doing some tests for the bug above, I noticed that Firefox generates
large numbers (for the expire time) in the cookies.txt file and libcurl
didn't treat them properly. Now it does.
Daniel (15 August 2005)
- Added more verbose "warning" messages to the curl client for cases where it
fails to open/read files etc to help users diagnose why it doesn't do what
you'd expect it to. Converted lots of old messages to use the new generic
function I wrote for this purpose.
Daniel (13 August 2005)
- James Bursa identified a libcurl HTTP bug and a good way to repeat it. If a
site responds with bad HTTP response that doesn't contain any header at all,
only a response body, and the write callback returns 0 to abort the
transfer, it didn't have any real effect but the write callback would be
called once more anyway.
Daniel (12 August 2005)
- Based on Richard Clayton's reports, I found out that using curl -d @filename
when 'filename' was not possible to access made curl use a GET request
- The time condition illegal syntax warning is now inhibited if -s is used.
Daniel (10 August 2005)
- Mario Schroeder found out that one of the debug callbacks calls that regards
SSL data with the CURLINFO_TEXT type claimed that the data was one byte
larger than it actually is, thus falsely telling the application that the
terminating zero was part of the data.
Daniel (9 August 2005)
- Christopher R. Palmer fixed the offsets used for date parsings when the time
zone name of a daylight savings time was used. For example, PDT vs PDS. This
flaw was introduced with the new date parser (11 sep 2004 - 7.12.2).
Fortunately, no web server or cookie string etc should be using such time
zone names thus limiting the effect of this bug.
Daniel (8 August 2005)
- Jon Grubbs filed bug report #1249962
( which identified a problem
with NTLM on a HTTP proxy if an FTP URL was given. libcurl now properly
switches to pure HTTP internally when an HTTP proxy is used, even for FTP
URLs. The problem would also occur with other multi-pass auth methods.
Daniel (7 August 2005)
- When curl is built with GnuTLS, curl-config didn't include "SSL" when
--features was used.
Daniel (28 July 2005)
set to 1, CURLOPT_NOBODY will now automatically be set to 0.
Daniel (27 July 2005)
- Dan Fandrich changes over the last week: fixed numerous minor configure
option parsing flaws: --without-gnutls, --without-spnego --without-gssapi
and --without-krb4. Spellfixed several error messages.
- Peteris Krumins added CURLOPT_COOKIELIST and CURLINFO_COOKIELIST, which is a
simple interface to extracting and setting cookies in libcurl's internal
"cookie jar". See the new cookie_interface.c example code.
Daniel (13 July 2005)
- Diego Casorran provided patches to make curl build fine on Amiga again.
Daniel (12 July 2005)
- Adrian Schuur added trailer support in the chunked encoding stream. The
trailer is then sent to the normal header callback/stream. I wrote up test
case 266 to verify the basic functionality. Do note that test case 34
contains a flawed chunked encoding stream that still works the same.
Daniel (5 July 2005)
- Gisle Vanem came up with a nice little work-around for bug #1230118
( It seems the Windows (MSVC)
libc time functions may return data one hour off if TZ is not set and
automatic DST adjustment is enabled. This made curl_getdate() return wrong
value, and it also concerned internal cookie expirations etc.
Daniel (4 July 2005)
- Andrew Bushnell provided enough info for me to tell that we badly needed to
fix the CONNECT authentication code with multi-pass auth methods (such as
NTLM) as it didn't previously properly ignore response-bodies - in fact it
stopped reading after all response headers had been received. This could
lead to libcurl sending the next request and reading the body from the first
request as response to the second request. (I also renamed the function,
which wasn't strictly necessary but...)
The best fix would to once and for all make the CONNECT code use the
ordinary request sending/receiving code, treating it as any ordinary request
instead of the special-purpose function we have now. It should make it
better for multi-interface too. And possibly lead to less code...
Added test case 265 for this. It doesn't work as a _really_ good test case
since the test proxy is too stupid, but the test case helps when running the
debugger to verify.
Daniel (30 June 2005)
- Dan Fandrich improved the configure script's ability to figure out what kind
of strerror_r() API that is used when cross-compiling. If __GLIB__ is
defined, it assumes the glibc API. If not, it issues a notice as before that
the user needs to manually edit lib/config.h for this.
Daniel (23 June 2005)
- David Shaw's fix that unifies proxy string treatment so that a proxy given
with CURLOPT_PROXY can use a http:// prefix and user + password. The user
and password fields are now also URL decoded properly. Test case 264 added
to verify.
Daniel (22 June 2005)
- David Shaw updated libcurl.m4
Daniel (14 June 2005)
- Gisle Vanem fixed a potential thread handle leak. Bug report #1216500
( Comment in
Daniel (13 June 2005)
- Made buildconf run libtoolize in the ares dir too (inspired by Tupone's
reverted patch).
Daniel (9 June 2005)
- Incorporated Tupone's findtool fix in buildconf (slightly edited)
- Incorporated Tupone's head -n fix in buildconf.
Daniel (8 June 2005)
- Reverted Tupone's patch again, it broke numerous autobuilds. Let's apply it
in pieces, one by one and see what we need to adjust to work all over.
Daniel (6 June 2005)
- Tupone Alfredo fixed three problems in buildconf:
1) findtool does look per tool in PATH and think ./perl is the perl
executable, while is just a local directory (I have . in the PATH)
2) I got several warning for head -1 deprecated in favour of head -n 1
3) ares directory is missing some file (missing is missing :-) ) because
automake and friends is not run.
Daniel (3 June 2005)
- Added docs/libcurl/getinfo-times, based on feedback from 'Edi':
- Andres Garcia provided yet another text mode patch for several test cases so
that they do text comparisions better on Windows (newline-wise).
Daniel (1 June 2005)
- The configure check for c-ares now adds the cares lib before the other libs,
to make it build fine with mingw. Inspired by Tupone Alfredo's bug report
and patch:
Daniel (31 May 2005)
- Todd Kulesza reported a flaw in the proxy option, since a numerical IPv6
address was not possible to use. It is now, but requires it written
RFC2732-style, within brackets - which incidently is how you enter numerical
IPv6 addresses in URLs. Test case 263 added to verify.
Daniel (30 May 2005)
- Eric Cooper reported about a problem with HTTP servers that responds with
binary zeroes within the headers. They confused libcurl to do wrong so the
downloaded headers become incomplete. The fix is now verified with test case
Daniel (25 May 2005)
- Fixed problems with the test suite, and in particular the FTP test cases
since it previously was failing every now and then in a nonsense manner.
- --trace-time now outputs the full microsecond, all 6 digits.
Daniel (24 May 2005)
- Andres Garcia provided a text mode patch for several test cases so that they
do text comparisions better on Windows (newline-wise).
- Any 2xx response (and not just 200) is now considered a fine response to
TYPE, as some servers obviously sends a 226 there. Added test case 261 to
verify. Based on a question/report by Georg Wicherski.
Daniel (20 May 2005)
- Improved to allow stdout tests to be mode=text as well, just
as file comparisons already supports. Added this info to the FILEFORMAT
Daniel (18 May 2005)
- John McGowan identified a problem in bug report #1204435
( with malformed URLs like
"http://somehost?data" as it added a slash too much in the request ("GET
/?data/"...). Added test case 260 to verify.
- The configure check for strerror_r() failed to detect the proper API at
times, like on HP-UX 10.20. Then lib/strerror.c badly assumed the glibc
version if the posix define wasn't set (since it _had_ found a strerror_r).
Daniel (16 May 2005)
- The gmtime_r() function in HP-UX 10.20 is broken. About 13 test cases fail
due to this. There's now a configure check that attempts to detect the bad
function and not use it on such systems.
Version 7.14.0 (16 May 2005)
Daniel (13 May 2005)
- Grigory Entin reported that curl's configure detects a fine poll() for Mac
OS X 10.4 (while 10.3 or later detected a "bad" one), but the executable
doesn't work as good as if built without poll(). I've adjusted the configure
to always skip the fine-poll() test on Mac OS X (darwin).
Daniel (12 May 2005)
- When doing a second request (after a disconnect) using the same easy handle,
over a proxy that uses NTLM authentication, libcurl failed to use NTLM again
properly (the auth method was accidentally reset to the same as had been set
for host auth, which defaults to Basic). Bug report #1200661
( identified the the problem and
the fix.
- If -z/--time-cond is used with an invalid date syntax, this is no longer
silently discarded. Instead a proper warning message is diplayed that
informs about it. But it still continues without the condition.
Version 7.14.0-pre2 (11 May 2005)
Daniel (11 May 2005)
- Starting now, libcurl sends a little different set of headers in its default
HTTP requests:
A) Normal non-proxy HTTP:
- no more "Pragma: no-cache" (this only makes sense to proxies)
B) Non-CONNECT HTTP request over proxy:
- "Pragma: no-cache" is used (like before)
- "Proxy-Connection: Keep-alive" (for older style 1.0-proxies)
C) CONNECT HTTP request over proxy:
- "Host: [name]:[port]"
- "Proxy-Connection: Keep-alive"
The A) case is mostly to reduce the default header size and remove a
pointless header.
The B) is to address (rare) problems with HTTP 1.0 proxies
The C) headers are both to address (rare) problems with some proxies. The
code in libcurl that deals with CONNECT requests need a rewrite, but it
feels like a too big a job for me to do now. Details are added in the code
comments for now.
Updated a large amount of test cases to reflect the news.
Daniel (10 May 2005)
- Half-baked attempt to bail out if select() returns _only_ errorfds when the
transfer is in progress. An attempt to fix Allan's problem. See and the rest of that thread
for details.
I'm still not sure this is the right fix, but...
Version 7.14.0-pre1 (9 May 2005)
Daniel (2 May 2005)
- Sort of "fixed" KNOWN_BUGS #4: curl now builds IPv6 enabled on AIX 4.3. At
least it should no longer cause a compiler error. However, it does not have
AI_NUMERICHOST so we cannot getaddrinfo() any numerical addresses with it
(we use that for FTP PORT/EPRT)! So, I modified the configure check that
checks if the getaddrinfo() is working, to use AI_NUMERICHOST since then
it'll fail on AIX 4.3 and it will automatically build with IPv6 support
- Added --trace-time that when used adds a time stamp to each trace line that
--trace, --trace-ascii and --verbose output. I also made the '>' display
separate each line on the linefeed so that HTTP requests etc look nicer in
the -v output.
- Made curl recognize the environment variables Lynx (and others?) support for
pointing out the CA cert path/file: SSL_CERT_DIR and SSL_CERT_FILE. If
CURL_CA_BUNDLE is not set, they are checked afterwards.
Like before: on windows if none of these are set, it checks for the ca cert
file like this:
1. application's directory
2. current working directory
3. Windows System directory (e.g. C:\windows\system32)
4. Windows Directory (e.g. C:\windows)
5. all directories along %PATH%
Daniel (1 May 2005)
- The script now starts test servers by doing fork() and exec()
instead of the previous approach. This is less complicated and should
hopefully lead to less "leaked" servers (servers that aren't stopped
properly when the tests are stopped).
- Alexander Zhuravlev found a case when you did "curl -I [URL]" and it
complained on the chunked encoding, even though a HEAD should never return a
body and thus it cannot be a chunked-encoding problem!
Daniel (30 April 2005)
- Alexander Zhuravlev found out that (lib)curl SIGSEGVed when using
--interface on an address that can't be bound.
Daniel (28 April 2005)
- Working on fixing up test cases to mark sections as 'mode=text' for things
that curl writes as text files, since then they can get different line
endings depending on OS. Andr<64>s Garc<72>a helps me work this out.
Did lots of other minor tweaks on the test scripts to work better and more
reliably find test servers and also kill test servers.
- Dan Fandrich pointed out how the script killed the HTTP server
instead of the HTTPS server when closing it down.
Daniel (27 April 2005)
- Paul Moore made curl check for the .curlrc file (_curlrc on windows) on two
more places. First, CURL_HOME is a new environment variable that is used
instead of HOME if it is set, to point out where the default config file
lives. If there's no config file in the dir pointed out by one of the
environment variables, the Windows version will instead check the same
directory the executable curl is located in.
Daniel (26 April 2005)
- Cory Nelson's work on nuking compiler warnings when building on x64 with
Daniel (25 April 2005)
- Fred New reported a bug where we used Basic auth and user name and password
in .netrc, and when following a Location: the subsequent requests didn't
properly use the auth as found in the netrc file. Added test case 257 to
verify my fix.
- Based on feedback from Cory Nelson, I added some preprocessor magic in
*/setup.h and */config-win32.h to build fine with VS2005 on x64.
Daniel (23 April 2005)
- Alex Suykov made the curl tool now assume that uploads using HTTP:// or
HTTPS:// are the only ones that show output and thus motivates a switched
off progress meter if the output is sent to the terminal. This makes FTP
uploads without '>', -o or -O show the progress meter.
Daniel (22 April 2005)
- Dave Dribin's MSVC makefile fix: set CURL_STATICLIB when it builds static
library variants.
- Andres Garcia fixed configure to set the proper define when building static
libcurl on windows.
- --retry-delay didn't work.
Daniel (18 April 2005)
- Olivier reported that even though he used CURLOPT_PORT, libcurl clearly
still used the default port. He was right. I fixed the problem and added the
test cases 521, 522 and 523 to verify the fix.
- Toshiyuki Maezawa reported that when doing a POST with a read callback,
libcurl didn't properly send an Expect: 100-continue header. It does now.
- I committed by mig change in the test suite's FTP server that moves out all
socket/TCP code to a separate C program named sockfilt. And added 4 new
test cases for FTP over IPv6.
Daniel (8 April 2005)
- Cory Nelson reported a problem with a HTTP server that responded with a 304
response containing an "illegal" Content-Length: header, which was not
properly ignored by libcurl. Now it is. Test case 249 verifies.
Daniel (7 April 2005)
- Added ability to build and run with GnuTLS as an alternative to OpenSSL for
the secure layer. configure --with-gnutls enables with. Note that the
previous OpenSSL check still has preference and if it first detects OpenSSL,
it will not check for GnuTLS. You may need to explictly diable OpenSSL with
This work has been sponsored by The Written Word.
Daniel (5 April 2005)
- Christophe Legry fixed the post-upload check for FTP to not complain if the
upload was skipped due to a time-condition as set with
CURLOPT_TIMECONDITION. I added test case 247 and 248 to verify.
Version 7.13.2 (5 April 2005)
Daniel (4 April 2005)
- Marcelo Juchem fixed the MSVC makefile for libcurl
- Gisle Vanem fixed a crash in libcurl, that could happen if the easy handle
was killed before the threading resolver (windows only) still hadn't
- Hardeep Singh reported a problem doing HTTP POST with Digest. (It was
actually also affecting NTLM and Negotiate.) It turned out that if the
server responded with 100 Continue before the initial 401 response, libcurl
didn't take care of the response properly. Test case 245 and 246 added to
verify this.
Daniel (30 March 2005)
- Andres Garcia modified the configure script to check for libgdi32 before
libcrypto, to make the SSL check work fine on msys/mingw.
Daniel (29 March 2005)
- Tom Moers identified a flaw when you sent a POST with Digest authentication,
as in the first request when curl sends a POST with Content-Length: 0, it
still forcibly closed the connection before doing the next step in the auth
- Jesper Jensen found out that FTP-SSL didn't work since my FTP
rewrite. Fixing that was easy, but it also revealed a much worse problem:
the FTP server response reader function didn't properly deal with reading
responses in multiple tiny chunks properly! I modified the FTP server to
allow it to produce such split-up responses to make sure curl deals with
them as it should.
- Based on Augustus Saunders' comments and findings, the HTTP output auth
function was fixed to use the proper proxy authentication when multiple ones
are accepted. test 239 and test 243 were added to repeat the problems and
verify the fixes.
--proxy-anyauth was added to the curl tool
Daniel (16 March 2005)
- Tru64 and some IRIX boxes seem to not like test 237 as it is. Their
inet_addr() functions seems to use &255 on all numericals in a ipv4 dotted
address which makes a different failure... Now I've modified the ipv4
resolve code to use inet_pton() instead in an attempt to make these systems
better detect this as a bad IP address rather than creating a toally bogus
address that is then passed on and used.
Daniel (15 March 2005)
- Dan Fandrich made the code properly use the uClibc's version of
inet_ntoa_r() when built with it.
- Added test 237 and 238: test EPSV and PASV response handling when they get
well- formated data back but using illegal values. In 237 PASV gets an IP
address that is way bad. In 238 EPSV gets a port that is way out of range.
Daniel (14 March 2005)
- Added a few missing features to the curl-config --features list
- Modified to now offer
1 - command line options for all info it previously only read from
file: --name, --email, --desc and --configure
2 - --nocvsup makes it not attempt to do cvs update
3 - --crosscompile informs it and makes it not attempt things it can't do
- Fixed numerous win32 compiler warnings.
- Removed the lib/security.h file since it shadowed the mingw/win32 header
with the same name which is needed for SSPI builds. The contents of the
former security.h is now i krb4.h
- configure --enable-sspi now enables SSPI in the build. It only works for
windows builds (including cross-compiles for windows).
Daniel (12 March 2005)
- David Houlder added --form-string that adds that string to a multipart
formpost part, without special characters having special meanings etc like
--form features.
Daniel (11 March 2005)
- curl_version_info() returns the feature bit CURL_VERSION_SSPI if it was
built with SSPI support.
- Christopher R. Palmer made it possible to build libcurl with the
USE_WINDOWS_SSPI on Windows, and then libcurl will be built to use the
native way to do NTLM. SSPI also allows libcurl to pass on the current user
and its password in the request.
Daniel (9 March 2005)
- Dan F improved the SSL lib setup in configure.
- Nodak Sodak reported a crash when using a SOCKS4 proxy.
- Jean-Marc Ranger pointed out an embarassing debug printf() leftover in the
multi interface code.
- Adjusted the man page for the curl_getdate() return value for dates after
year 2038. For 32 bit time_t it returns 0x7fffffff but for 64bit time_t it
returns either the correct value or even -1 on some systems that still seem
to not deal with this properly. Tor Arntsen found a 64bit AIX system for us
that did the latter. Gwenole Beauchesne's Mandrake patch put the lights on
this problem in the first place.
Daniel (8 March 2005)
- Dominick Meglio reported that using CURLOPT_FILETIME when transferring a FTP
file got a Last-Modified: header written to the data stream, corrupting the
actual data. This was because some conditions from the previous FTP code was
not properly brought into the new FTP code. I fixed and I added test case
520 to verify. (This bug was introduced in 7.13.1)
- Dan Fandrich fixed the configure --with-zlib option to always consider the
given path before any standard paths.
Daniel (6 March 2005)
- Randy McMurchy was the first to report that was missing from the
release archive and thus 'make test' fails.
Daniel (5 March 2005)
- Dan Fandrich added HAVE_FTRUNCATE to several config-*.h files.
- Added test case 235 that makes a resumed upload of a file that isn't present
on the remote side. This then converts the operation to an ordinary STOR
upload. This was requested/pointed out by Ignacio Vazquez-Abrams.
It also proved (and I fixed) a bug in the newly rewritten ftp code (and
present in the 7.13.1 release) when trying to resume an upload and the
servers returns an error to the SIZE command. libcurl then loops and sends
SIZE commands infinitely.
- Dan Fandrich fixed a SSL problem introduced on February 9th that made
libcurl attempt to load the whole random file to seed the PRNG. This is
really bad since this turns out to be using /dev/urandom at times...
Version 7.13.1 (4 March 2005)
Daniel (4 March 2005)
- Dave Dribin made it possible to set CURLOPT_COOKIEFILE to "" to activate
the cookie "engine" without having to provide an empty or non-existing file.
- Rene Rebe fixed a -# crash when more data than expected was retrieved.
Daniel (22 February 2005)
- NTLM and ftp-krb4 buffer overflow fixed, as reported here: and the CAN report here:
If these security guys were serious, we'd been notified in advance and we
could've saved a few of you a little surprise, but now we weren't.
Daniel (19 February 2005)
- Ralph Mitchell reported a flaw when you used a proxy with auth, and you
requested data from a host and then followed a redirect to another
host. libcurl then didn't use the proxy-auth properly in the second request,
due to the host-only check for original host name wrongly being extended to
the proxy auth as well. Added test case 233 to verify the flaw and that the
fix removed the problem.
Daniel (18 February 2005)
- Mike Dobbs reported a mingw build failure due to the lack of
BUILDING_LIBCURL being defined when libcurl is built. Now this is defined by
configure when mingw is used.
Daniel (17 February 2005)
- David in bug report #1124588 found and fixed a socket leak when libcurl
didn't close the socket properly when returning error due to failing
Daniel (16 February 2005)
- Christopher R. Palmer reported a problem with HTTP-POSTing using "anyauth"
that picks NTLM. Thanks to David Byron letting me test NTLM against his
servers, I could quickly repeat and fix the problem. It turned out to be:
When libcurl POSTs without knowing/using an authentication and it gets back
a list of types from which it picks NTLM, it needs to either continue
sending its data if it keeps the connection alive, or not send the data but
close the connection. Then do the first step in the NTLM auth. libcurl
didn't send the data nor close the connection but simply read the
response-body and then sent the first negotiation step. Which then failed
miserably of course. The fixed version forces a connection if there is more
than 2000 bytes left to send.
Daniel (14 February 2005)
- The configure script didn't check for ENGINE_load_builtin_engines() so it
was never used.
Daniel (11 February 2005)
- Removed all uses of strftime() since it uses the localised version of the
week day names and month names and servers don't like that.
Daniel (10 February 2005)
- Now the test script disables valgrind-testing when the test suite runs if
libcurl is built shared. Otherwise valgrind only tests the shell that runs
the wrapper-script named 'curl' that is a front-end to curl in this case.
This should also fix the huge amount of reports of false positives when
valgrind has identified leaks in (ba)sh and not in curl and people report
that as curl bugs. Bug report #1116672 is one example.
Also, the valgrind report parser has been adapted to check that at least one
of the sources in a stack strace is one of (lib)curl's source files or
otherwise it will not consider the problem to concern (lib)curl.
- Marty Kuhrt streamlined the VMS build.
Daniel (9 February 2005)
- David Byron fixed his SSL problems, initially mentioned here: It turned out we didn't use
SSL_pending() as we should.
- Converted lots of FTP code to a statemachine, so that the multi interface
doesn't block while communicating commands-responses with an FTP server.
I've added a comment like BLOCKING in the code on all spots I could find
where we still have blocking operations. When we change curl_easy_perform()
to use the multi interface, we'll also be able to simplify the code since
there will only be one "internal interface".
While doing this, I've now made CURLE_FTP_ACCESS_DENIED separate from the
new CURLE_LOGIN_DENIED. The first one is now access denied to a function,
like changing directory or retrieving a file, while the second means that we
were denied login.
The CVS tag 'before_ftp_statemachine' was set just before this went in, in
case of future need.
- Gisle made the DICT code send CRLF and not just LF as the spec says so.
Daniel (8 February 2005)
- Gisle fixed problems when libcurl runs out of memory, and worked on making
sure the proper error code is returned for those occations.
Daniel (7 February 2005)
- Maruko pointed out a problem with inflate decompressing exactly 64K
Daniel (5 February 2005)
- Eric Vergnaud found a use of an uninitialised variable in the ftp when doing
PORT on ipv6-enabled hosts.
- David Byron pointed out we could use BUFSIZE to read data (in
lib/transfer.c) instead of using BUFSIZE -1.
Version 7.13.0 (1 February 2005)
Daniel (31 January 2005)
- Added Lars Nilsson's example
Daniel (30 January 2005)
- Fixed a memory leak when using the multi interface and the DO operation
failed (as in test case 205).
- Fixed a valgrind warning for file:// operations.
- Fixed a valgrind report in the url globbing code for the curl command line
- Bugfixed the parser that scans the valgrind report outputs (in
I noticed that it previously didn't detect and report the "Conditional jump
or move depends on uninitialised value(s)" error. When I fixed this, I
caught a few curl bugs with it. And then I had to spend time to make the
test suite IGNORE these errors when OpenSSL is used since it produce massive
amounts of valgrind warnings (but only of the "Conditional..." kind it
seems). So, if a test that requires SSL is run, it ignores the
"Conditional..." errors, and you'll get a "valgrind PARTIAL" output instead
of "valgrind OK".
Daniel (29 January 2005)
- Using the multi interface, and doing a requsted a re-used connection that
gets closed just after the request has been sent failed and did not re-issue
a request on a fresh reconnect like the easy interface did. Now it does!
- Define CURL_MULTIEASY when building libcurl (lib/easy.c to be exact), to use
my new curl_easy_perform() that uses the multi interface to run the
request. It is a great testbed for the multi interface and I believe we
shall do it this way for real in the future when we have a successor to
curl_multi_fdset(). I've used this approach to detect and fix several of the
recent multi-interfaces issues.
- Adjusted the KNOWN_BUGS #17 fix a bit more since the FTP code also did some
bad assumptions.
- multi interface: when a request is denied due to "Maximum redirects
followed" libcurl leaked the last Location: URL.
- Connect failures with the multi interface was often returned as "connect()
timed out" even though the reason was different.
Daniel (28 January 2005)
- KNOWN_BUGS #17 fixed. A DNS cache entry may not remain locked between two
curl_easy_perform() invokes. It was previously unlocked at disconnect, which
could mean that it remained locked between multiple transfers. The DNS cache
may not live as long as the connection cache does, as they are separate.
To deal with the lack of DNS (host address) data availability in re-used
connections, libcurl now keeps a copy of the IP adress as a string, to be
able to show it even on subsequent requests on the same connection.
The problem could be made to appear with this stunt:
1. create a multi handle
2. add an easy handle
3. fetch a URL that is persistent (leaves the connection alive)
4. remove the easy handle from the multi
5. kill the multi handle
6. create a multi handle
7. add the same easy handle to the new multi handle
8. fetch a URL from the same server as before (re-using the connection)
- Stephen More pointed out that CURLOPT_FTPPORT and the -P option didn't work
when built ipv6-enabled. I've now made a fix for it. Writing test cases for
custom port hosts turned too tricky so unfortunately there's none.
Daniel (25 January 2005)
- Ian Ford asked about support for the FTP command ACCT, and I discovered it
is present in RFC959... so now (lib)curl supports it as well. --ftp-account
and CURLOPT_FTP_ACCOUNT set the account string. (The server may ask for an
account string after PASS have been sent away. The client responds
with "ACCT [account string]".) Added test case 228 and 229 to verify the
functionality. Updated the test FTP server to support ACCT somewhat.
- David Shaw contributed a fairly complete and detailed autoconf test you can
use to detect libcurl and setup variables for the protocols the installed
libcurl supports: docs/libcurl/libcurl.m4
Daniel (21 January 2005)
- Major FTP third party transfer overhaul.
These four options are now obsolete: CURLOPT_SOURCE_HOST,
CURLOPT_SOURCE_PATH, CURLOPT_SOURCE_PORT (this option didn't work before)
These two options are added: CURLOPT_SOURCE_URL and CURLOPT_SOURCE_QUOTE.
The target-side didn't use the proper path with RETR, and thus this only
worked correctly in the login path (i.e without doing any CWD). The source-
side still uses a wrong path, but the fix for this will need to wait. Verify
the flaw by using a source URL with included %XX-codes.
Made CURLOPT_FTPPORT control weather the target operation should use PORT
(or not). The other side thus uses passive (PASV) mode.
Updated the ftp3rdparty.c example source to use the updated options.
Added support for a second FTP server in the test suite. Named... ftp2.
Added test cases 230, 231 and 232 as a few first basic tests of very simple
3rd party transfers.
Changed the debug output to include 'target' and 'source' when a 3rd party
is being made, to make it clearer what commands/responses came on what
Added three new command line options: --3p-url, --3p-user and --3p-quote.
Documented the command line options and the curl_easy_setopt options related
to third party transfers.
(Temporarily) disabled the ability to re-use an existing connection for the
source connection. This is because it needs to force a new in case the
source and target is the same host, and the host name check is trickier now
when the source is identified with a full URL instead of a plain host name
like before.
TODO (short-term) for 3rd party transfers: quote support. The options are
there, we need to add test cases to verify their functionality.
TODO (long-term) for 3rd party transfers: IPv6 support (EPRT and EPSV etc)
and SSL/TSL support.
Daniel (20 January 2005)
- Philippe Hameau found out that -Q "+[command]" didn't work, although some
code was written for it. I fixed and added test case 227 to verify it.
The curl.1 man page didn't mention the '+' so I added it.
Daniel (19 January 2005)
- Stephan Bergmann made libcurl return CURLE_URL_MALFORMAT if an FTP URL
contains %0a or %0d in the user, password or CWD parts. (A future fix would
include doing it for %00 as well - see KNOWN_BUGS for details.) Test case
225 and 226 were added to verify this
- Stephan Bergmann pointed out two flaws in libcurl built with HTTP disabled:
1) the proxy environment variables are still read and used to set HTTP proxy
2) you couldn't disable http proxy with CURLOPT_PROXY (since the option was
disabled). This is important since apps may want to disable HTTP proxy
without actually knowing if libcurl was built to disable HTTP or not.
Based on Stephan's patch, both these issues should now be fixed.
Daniel (18 January 2005)
- Cody Jones' enhanced version of Samuel D<>az Garc<72>a's MSVC makefile patch was
Daniel (16 January 2005)
- Alex aka WindEagle pointed out that when doing "curl -v", curl
assumed this used the DICT protocol. While guessing protocols will remain
fuzzy, I've now made sure that the host names must start with "[protocol]."
for them to be a valid guessable name. I also removed "https" as a prefix
that indicates HTTPS, since we hardly ever see any host names using that.
Daniel (13 January 2005)
- Inspired by Martijn Koster's patch and example source at, I now made the
select() and poll() calls properly loop if they return -1 and errno is
EINTR. glibc docs for this is found here:
This last link says BSD doesn't have this "effect". Will there be a problem
if we do this unconditionally?
Daniel (11 January 2005)
- Dan Torop cleaned up a few no longer used variables from David Phillips'
select() overhaul fix.
- Cyrill Osterwalder posted a detailed analysis about a bug that occurs when
using a custom Host: header and curl fails to send a request on a re-used
persistent connection and thus creates a new connection and resends it. It
then sent two Host: headers. Cyrill's analysis was posted here:
- Bruce Mitchener identified (bug report #1099640) the never-ending SOCKS5
problem with the version byte and the check for bad versions. Bruce has lots
of clues on this, and based on his suggestion I've now removed the check of
that byte since it seems to be able to contain 1 or 5.
Daniel (10 January 2005)
- Pavel Orehov reported memory problems with the multi interface in bug report
#1098843. In short, a shared DNS cache was setup for a multi handle and when
the shared cache was deleted before the individual easy handles, the latter
cleanups caused read/writes to already freed memory.
- Hzhijun reported a memory leak in the SSL certificate code, that leaked the
remote certificate name when it didn't match the used host name.
Gisle (8 January 2005)
- Added Makefile.Watcom files (src/lib). Updated Makefile.dist.
Daniel (7 January 2005)
- Improved the test script's valgrind log parser to actually work! Also added
the ability to disable the log scanner for specific test cases. Test case
509 results in numerous problems and leaks in OpenSSL and has to get it
Daniel (6 January 2005)
- Fixed a single-byte read out of bounds in test case 39 in the curl tool code
(i.e not in the library).
- Bug report #1097019 identified a problem when doing -d "data" with -G and
sending it to two URLs with {}. Added test 199 to verify the fix.
Daniel (4 January 2005)
- Marty Kuhrt adjusted a VMS build script slightly
- Kai Sommerfeld and Gisle Vanem fixed libcurl to build with IPv6 support on
Daniel (2 January 2005)
- Alex Neblett updated the MSVC makefiles slightly.