Prior to this change cookies with an expiry date that failed parsing
and were converted to session cookies could be purged in remove_expired.
Bug: https://github.com/curl/curl/issues/697
Reported-by: Seth Mos
Prevent a crash if 2 (or more) requests are made to the same host and
pipelining is enabled and the connection does not complete.
Bug: https://github.com/curl/curl/pull/690
Some systems have special files that report as 0 bytes big, but still
contain data that can be read (for example /proc/cpuinfo on
Linux). Starting now, a zero byte size is considered "unknown" size and
will be read as far as possible anyway.
Reported-by: Jesse Tan
Closes#681
... as when pipelining is used, we read things into a unified buffer and
we don't do that with HTTP/2. This could then easily make programs that
set CURLMOPT_PIPELINING = CURLPIPE_HTTP1|CURLPIPE_MULTIPLEX to get data
intermixed or plain broken between HTTP/2 streams.
Reported-by: Anders Bakken
The two options are almost the same, except in the case of OpenSSL:
CURLINFO_TLS_SESSION OpenSSL session internals is SSL_CTX *.
CURLINFO_TLS_SSL_PTR OpenSSL session internals is SSL *.
For backwards compatibility we couldn't modify CURLINFO_TLS_SESSION to
return an SSL pointer for OpenSSL.
Also, add support for the 'internals' member to point to SSL object for
the other backends axTLS, PolarSSL, Secure Channel, Secure Transport and
wolfSSL.
Bug: https://github.com/curl/curl/issues/234
Reported-by: dkjjr89@users.noreply.github.com
Bug: https://curl.haxx.se/mail/lib-2015-09/0127.html
Reported-by: Michael König
The internal Curl_done() function uses Curl_expire() at times and that
uses the timeout list. Better clean up the list once we're done using
it. This caused a segfault.
Reported-by: 蔡文凱
Bug: https://curl.haxx.se/mail/lib-2016-02/0097.html
- Add tests.
- Add an example to CURLOPT_TFTP_NO_OPTIONS.3.
- Add --tftp-no-options to expose CURLOPT_TFTP_NO_OPTIONS.
Bug: https://github.com/curl/curl/issues/481
Some TFTP server implementations ignore the "TFTP Option extension"
(RFC 1782-1784, 2347-2349), or implement it in a flawed way, causing
problems with libcurl. Another switch for curl_easy_setopt
"CURLOPT_TFTP_NO_OPTIONS" is introduced which prevents libcurl from
sending TFTP option requests to a server, avoiding many problems caused
by faulty implementations.
Bug: https://github.com/curl/curl/issues/481
If any parameter in a HTTP DIGEST challenge message is present multiple
times, memory allocated for all but the last entry should be freed.
Bug: https://github.com/curl/curl/pull/667
At one point during the development of HTTP/2, the commit 133cdd29ea
introduced automatic decompression of Content-Encoding as that was what
the spec said then. Now however, HTTP/2 should work the same way as
HTTP/1 in this regard.
Reported-by: Kazuho Oku
Closes#661
nghttp2 callback deals with TLS layer and therefore the header does not
need to be broken into chunks.
Bug: https://github.com/curl/curl/issues/659
Reported-by: Kazuho Oku
Since we didn't keep the input argument around after having called
mbedtls, it could end up accessing the wrong memory when figuring out
the ALPN protocols.
Closes#642
As of https://boringssl-review.googlesource.com/#/c/6980/, almost all of
BoringSSL #ifdefs in cURL should be unnecessary:
- BoringSSL provides no-op stubs for compatibility which replaces most
#ifdefs.
- DES_set_odd_parity has been in BoringSSL for nearly a year now. Remove
the compatibility codepath.
- With a small tweak to an extend_key_56_to_64 call, the NTLM code
builds fine.
- Switch OCSP-related #ifdefs to the more generally useful
OPENSSL_NO_OCSP.
The only #ifdefs which remain are Curl_ossl_version and the #undefs to
work around OpenSSL and wincrypt.h name conflicts. (BoringSSL leaves
that to the consumer. The in-header workaround makes things sensitive to
include order.)
This change errs on the side of removing conditionals despite many of
the restored codepaths being no-ops. (BoringSSL generally adds no-op
compatibility stubs when possible. OPENSSL_VERSION_NUMBER #ifdefs are
bad enough!)
Closes#640
It turns out Firefox and Chrome both allow spaces in cookie names and
there are sites out there using that.
Turned out the code meant to strip off trailing space from cookie names
didn't work. Fixed now.
Test case 8 modified to verify both these changes.
Closes#639
When trying to verify a peer without having any root CA certificates
set, this makes libcurl use the TLS library's built in default as
fallback.
Closes#569
.. also fix a conversion bug in the unused function
curl_win32_ascii_to_idn().
And remove wprintfs on error (Jay).
Bug: https://github.com/curl/curl/pull/637
It isn't used by the code in current conditions but for safety it seems
sensible to at least not crash on such input.
Extended unit test 1395 to verify this too as well as a plain "/" input.
- Switch from verifying a pinned public key in a callback during the
certificate verification to inline after the certificate verification.
The callback method had three problems:
1. If a pinned public key didn't match, CURLE_SSL_PINNEDPUBKEYNOTMATCH
was not returned.
2. If peer certificate verification was disabled the pinned key
verification did not take place as it should.
3. (related to #2) If there was no certificate of depth 0 the callback
would not have checked the pinned public key.
Though all those problems could have been fixed it would have made the
code more complex. Instead we now verify inline after the certificate
verification in mbedtls_connect_step2.
Ref: http://curl.haxx.se/mail/lib-2016-01/0047.html
Ref: https://github.com/bagder/curl/pull/601
The CURLOPT_SSH_PUBLIC_KEYFILE option has been documented to handle
empty strings specially since curl-7_25_0-31-g05a443a but the behavior
was unintentionally removed in curl-7_38_0-47-gfa7d04f.
This commit restores the original behavior and clarifies it in the
documentation that NULL and "" have both the same meaning when passed
to CURLOPT_SSH_PUBLIC_KEYFILE.
Bug: http://curl.haxx.se/mail/lib-2016-01/0072.html
... by extracting the LIB + REASON from the OpenSSL error code. OpenSSL
1.1.0+ returned a new func number of another cerfificate fail so this
required a fix and this is the better way to catch this error anyway.
When an HTTP/2 upgrade request fails (no protocol switch), it would
previously detect that as still possible to pipeline on (which is
acorrect) and do that when PIPEWAIT was enabled even if pipelining was
not explictily enabled.
It should only pipelined if explicitly asked to.
Closes#584
Before this patch, if a URL does not start with the protocol
name/scheme, effective URLs would be prefixed with upper-case protocol
names/schemes. This behavior might not be expected by library users or
end users.
For example, if `CURLOPT_DEFAULT_PROTOCOL` is set to "https". And the
URL is "hostname/path". The effective URL would be
"HTTPS://hostname/path" instead of "https://hostname/path".
After this patch, effective URLs would be prefixed with a lower-case
protocol name/scheme.
Closes#597
Signed-off-by: Mohammad AlSaleh <CE.Mohammad.AlSaleh@gmail.com>
Previously, when HTTP/2 is enabled and used, and stream has content
length known, Curl_read was not called when there was no bytes left to
read. Because of this, we could not make sure that
http2_handle_stream_close was called for every stream. Since we use
http2_handle_stream_close to emit trailer fields, they were
effectively ignored. This commit changes the code so that Curl_read is
called even if no bytes left to read, to ensure that
http2_handle_stream_close is called for every stream.
Discussed in https://github.com/bagder/curl/pull/564
Check that the trailer buffer exists before attempting a client write
for trailers on stream close.
Refer to comments in https://github.com/bagder/curl/pull/564
To make sure curl doesn't allow multiplexing before a connection is
upgraded to HTTP/2 (like when Upgrade: h2c fails), we must make sure the
connection uses HTTP/2 as well and not only check what's wanted.
Closes#584
Patch-by: c0ff
Previously file.txt[CR][LF] would have been returned as file.tx
(without the last t) if filetype is symlink. Now the t is
included and the internal item_length includes the zero byte.
Spotted using test 576 on Windows.
Try harder to prevent libcurl from opening up an additional socket when
CURLOPT_PIPEWAIT is set. Accomplished by letting ongoing TCP and TLS
handshakes complete first before the decision is made.
Closes#575
The function is only present in wolfssl/cyassl if it was built with
--enable-opensslextra. With these checks added, pinning support is disabled
unless the TLS lib has that function available.
Also fix the mistake in configure that checks for the wrong lib name.
Closes#566
This commit adds trailer support in HTTP/2. In HTTP/1.1, chunked
encoding must be used to send trialer fields. HTTP/2 deprecated any
trandfer-encoding, including chunked. But trailer fields are now
always available.
Since trailer fields are relatively rare these days (gRPC uses them
extensively though), allocating buffer for trailer fields is done when
we detect that HEADERS frame containing trailer fields is started. We
use Curl_add_buffer_* functions to buffer all trailers, just like we
do for regular header fields. And then deliver them when stream is
closed. We have to be careful here so that all data are delivered to
upper layer before sending trailers to the application.
We can deliver trailer field one by one using NGHTTP2_ERR_PAUSE
mechanism, but current method is far more simple.
Another possibility is use chunked encoding internally for HTTP/2
traffic. I have not tested it, but it could add another overhead.
Closes#564
- In Curl_verifyhost check all altnames in the certificate.
Prior to this change only the first altname was checked. Only the GSKit
SSL backend was affected by this bug.
Bug: http://curl.haxx.se/mail/lib-2015-12/0062.html
Reported-by: John Kohl
When NGHTTP2_ERR_PAUSE is returned from data_source_read_callback, we
might not process DATA frame fully. Calling nghttp2_session_mem_recv()
again will continue to process DATA frame, but if there is no incoming
frames, then we have to call it again with 0-length data. Without this,
on_stream_close callback will not be called, and stream could be hanged.
Bug: http://curl.haxx.se/mail/lib-2015-11/0103.html
Reported-by: Francisco Moraes
The name of the header guard in lwIP's <lwip/opt.h> has changed from
'__LWIP_OPT_H__' to 'LWIP_HDR_OPT_H' (bug #35874 in May 2015).
Other fixes:
- In curl_setup.h, the problem with an old PSDK doesn't apply if lwIP is
used.
- In memdebug.h, the 'socket' should be undefined first due to lwIP's
lwip_socket() macro.
- In curl_addrinfo.c lwIP's getaddrinfo() + freeaddrinfo() macros need
special handling because they were undef'ed in memdebug.h.
- In select.c we can't use preprocessor conditionals inside select if
MSVC and select is a macro, as it is with lwIP.
http://curl.haxx.se/mail/lib-2015-12/0023.htmlhttp://curl.haxx.se/mail/lib-2015-12/0024.html
- If the size of the length type (curl_off_t) is greater than the size
of the size_t type then check before allocating memory to make sure the
value of length will fit in a size_t without overflow. If it doesn't
then return CURLE_BAD_FUNCTION_ARGUMENT.
Bug: https://github.com/bagder/curl/issues/425#issuecomment-154518679
Reported-by: Steve Holme
IoctlSocket() apparently wants a pointer to a long, passed as a char *
in its third parameter. This bug was introduced already back in commit
c5fdeef41d from October 1 2001!
Bug: http://curl.haxx.se/mail/lib-2015-11/0088.html
Reported-by: Norbert Kett
It would previously be skipped if an existing error was returned, but
would lead to a previous value being left there and later used.
CURLINFO_TOTAL_TIME for example.
Still it avoids that final progress update if we reached DONE as the
result of a callback abort to avoid another callback to be called after
an abort-by-callback.
Reported-by: Lukas Ruzicka
Closes#538
smb.c:134:3: warning: conversion to 'short unsigned int' from 'int' may
alter its value
smb.c:146:42: warning: conversion to 'unsigned int' from 'long long
unsigned int' may alter its value
smb.c:146:65: warning: conversion to 'unsigned int' from 'long long
unsigned int' may alter its value
The push headers are freed after the push callback has been invoked,
meaning this code should only free the headers if the callback was never
invoked and thus the headers weren't freed at that time.
Reported-by: Davey Shafik
According to RFC7628 a failure message may be sent by the server in a
base64 encoded JSON string as a continuation response.
Currently only implemented for OAUTHBEARER and not XAUTH2.
OAUTHBEARER is now the official "registered" SASL mechanism name for
OAuth 2.0. However, we don't want to drop support for XOAUTH2 as some
servers won't support the new mechanism yet.
They tend to never get updated anyway so they're frequently inaccurate
and we never go back to revisit them anyway. We document issues to work
on properly in KNOWN_BUGS and TODO instead.
Following the fix in commit d6d58dd558 it is necessary to re-introduce
XOAUTH2 in the default enabled authentication mechanism, which was
removed in commit 7b2012f262, otherwise users will have to specify
AUTH=XOAUTH2 in the URL.
Note: OAuth 2.0 will only be used when the bearer is specified.
Regression from commit 9e8ced9890 which meant if --oauth2-bearer was
specified but the SASL mechanism wasn't supported by the server then
the mechanism would be chosen.
Added support to the OAuth 2.0 message function for host and port, in
order to accommodate the official OAUTHBEARER SASL mechanism which is
to be added shortly.
The curl_config.h file can be generated either from curl_config.h.cmake
or curl_config.h.in, depending on whether you're building using CMake or
the autotools. The CMake template header doesn't include entries for
all of the protocols that you can disable, which (I think) means that
you can't actually disable those protocols when building via CMake.
Closes#523
BoringSSL implements `BIO_get_mem_data` as a function, instead of a
macro, and expects the output pointer to be a `char **`. We have to add
an explicit cast to grab the pointer as a `const char **`.
Closes#524
- Set user info param to the socket returned by Curl_getconnectinfo,
regardless of if the socket is bad. Effectively this means the user info
param now will receive CURL_SOCKET_BAD instead of -1 on bad socket.
- Remove incorrect comments.
CURLINFO_ACTIVESOCKET is documented to write CURL_SOCKET_BAD to user
info param but prior to this change it wrote -1.
Bug: https://github.com/bagder/curl/pull/518
Reported-by: Marcel Raad
Rationale: when starting up a curl-using app, all cookies from the jar
are checked against each other. This was causing a startup delay in the
Fifth browser.
All tests pass.
Signed-off-by: Lauri Kasanen <cand@gmx.com>
Apparently there are sites out there that do redirects to URLs they
provide in plain UTF-8 or similar. Browsers and wget %-encode such
headers when doing a subsequent request. Now libcurl does too.
Added test 1138 to verify.
Closes#473
Fixes a name space pollution at the cost of programs using one of these
defines will no longer compile. However, the vast majority of libcurl
programs that do multipart formposts use curl_formadd() to build this
list.
Closes#506
This reverts commit 370ee919b3.
Issue #509 has all the details but it was confirmed that the crash was
not due to this, so the previous commit was wrong.
Removed wrong assert()s
The 'conn' passed in as userdata can be used and there can be other
sessionhandles ('data') than the single one this checked for.
introduced in c6aedf680f. It needs to be CURLM_STATE_LAST big since it
must hande the range 0 .. CURLM_STATE_MSGSENT (18) and CURLM_STATE_LAST
is 19 right now.
Reported-by: Dan Fandrich
Bug: http://curl.haxx.se/mail/lib-2015-10/0069.html
... and assign it from the set.fread_func_set pointer in the
Curl_init_CONNECT function. This A) avoids that we have code that
assigns fields in the 'set' struct (which we always knew was bad) and
more importantly B) it makes it impossibly to accidentally leave the
wrong value for when the handle is re-used etc.
Introducing a state-init functionality in multi.c, so that we can set a
specific function to get called when we enter a state. The
Curl_init_CONNECT is thus called when switching to the CONNECT state.
Bug: https://github.com/bagder/curl/issues/346Closes#346
sk_X509_pop will decrease the size of the stack which means that the loop would
end after having added only half of the certificates.
Also make sure that the X509 certificate is freed in case
SSL_CTX_add_extra_chain_cert fails.
- If a CURLINFO option is unknown return CURLE_UNKNOWN_OPTION.
Prior to this change CURLE_BAD_FUNCTION_ARGUMENT was returned on
unknown. That return value is contradicted by the CURLINFO option
documentation which specifies a return of CURLE_UNKNOWN_OPTION on
unknown.
- Change algorithm init to happen after OpenSSL config load.
Additional algorithms may be available due to the user's config so we
initialize the algorithms after the user's config is loaded.
Bug: https://github.com/bagder/curl/issues/447
Reported-by: Denis Feklushkin
For a single-stream download from localhost, we managed to increase
transfer speed from 1.6MB/sec to around 400MB/sec, mostly because of
this single fix.
... only call it when there is data arriving for another handle than the
one that is currently driving it.
Improves single-stream download performance quite a lot.
Thanks-to: Tatsuhiro Tsujikawa
Bug: http://curl.haxx.se/mail/lib-2015-09/0097.html
If GnuTLS fails to read the certificate then include whatever reason it
provides in the failure message reported to the client.
Signed-off-by: Mike Crowe <mac@mcrowe.com>
The gnutls vtls back-end was previously ignoring any password set via
CURLOPT_KEYPASSWD. Presumably this was because
gnutls_certificate_set_x509_key_file did not support encrypted keys.
gnutls now has a gnutls_certificate_set_x509_key_file2 function that
does support encrypted keys. Let's determine at compile time whether the
available gnutls supports this new function. If it does then use it to
pass the password. If it does not then emit a helpful diagnostic if a
password is set. This is preferable to the previous behaviour of just
failing to read the certificate without giving a reason in that case.
Signed-off-by: Mike Crowe <mac@mcrowe.com>
... even for those that don't support providing anything in the
'internals' struct member since it offers a convenient way for
applications to figure this out.
- Change the designator name we use to show the base64 encoded sha256
hash of the server's public key from 'pinnedpubkey' to
'public key hash'.
Though the server's public key hash is only shown when comparing pinned
public key hashes, the server's hash may not match one of the pinned.
Without this workaround, NSS re-uses a session cache entry despite the
server name does not match. This causes SNI host name to differ from
the actual host name. Consequently, certain servers (e.g. github.com)
respond by 400 to such requests.
Bug: https://bugzilla.mozilla.org/1202264
If the port number in the proxy string ended weirdly or the number is
too large, skip it. Mostly as a means to bail out early if a "bare" IPv6
numerical address is used without enclosing brackets.
Also mention the bracket requirement for IPv6 numerical addresses to the
man page for CURLOPT_PROXY.
Closes#415
Reported-by: Marcel Raad
In some timing-dependnt cases when a 4xx response immediately followed
after a 150 when a STOR was issued, this function would wrongly return
'complete == true' while 'wait_data_conn' was still set.
Closes#405
Reported-by: Patricia Muscalu
RFC 7540 section 8.1.2.2 states: "An endpoint MUST NOT generate an
HTTP/2 message containing connection-specific header fields; any message
containing connection-specific header fields MUST be treated as
malformed"
Closes#401
Introduced in commit 59f3f92ba6 this function is only implemented when
CURL_DISABLE_CRYPTO_AUTH is not defined. As such we shouldn't define
the function in the header file either.
This patch addresses known bug #76, where on 64-bit Windows SOCKET is 64
bits wide, but long is only 32, making CURLINFO_LASTSOCKET unreliable.
Signed-off-by: Razvan Cojocaru <rcojocaru@bitdefender.com>
- Add new option CURLOPT_DEFAULT_PROTOCOL to allow specifying a default
protocol for schemeless URLs.
- Add new tool option --proto-default to expose
CURLOPT_DEFAULT_PROTOCOL.
In the case of schemeless URLs libcurl will behave in this way:
When the option is used libcurl will use the supplied default.
When the option is not used, libcurl will follow its usual plan of
guessing from the hostname and falling back to 'http'.
If strict certificate checking is disabled (CURLOPT_SSL_VERIFYPEER
and CURLOPT_SSL_VERIFYHOST are disabled) do not fail if the server
doesn't present a certificate at all.
Closes#392
The multi state machine would otherwise go into the DO_MORE state after
DO, even for the case when the FTP state machine had already performed
those duties, which caused libcurl to get stuck in that state and fail
miserably. This occured for for active ftp uploads.
Reported-by: Patricia Muscalu
Visual Studio complains with a message box:
"Run-Time Check Failure #1 - A cast to a smaller data type has caused a
loss of data. If this was intentional, you should mask the source of
the cast with the appropriate bitmask.
For example:
char c = (i & 0xFF);
Changing the code in this way will not affect the quality of the
resulting optimized code."
This is because only 'val' is cast to unsigned char, so the "& 0xff" has
no effect.
Closes#387
Return 0 instead of NGHTTP2_ERR_CALLBACK_FAILURE if we can't locate the
SessionHandle. Apparently mod_h2 will sometimes send a frame for a
stream_id we're finished with.
Use nghttp2_session_get_stream_user_data and
nghttp2_session_set_stream_user_data to identify SessionHandles instead
of a hash.
Closes#372
Currently when the server responds with 401 on NTLM authenticated
connection (re-used) we consider it to have failed. However this is
legitimate and may happen when for example IIS is set configured to
'authPersistSingleRequest' or when the request goes thru a proxy (with
'via' header).
Implemented by imploying an additional state once a connection is
re-used to indicate that if we receive 401 we need to restart
authentication.
Closes#363
The SSH state machine didn't clear the 'rc' variable appropriately in a
two places which prevented it from looping the way it should. And it
lacked an 'else' statement that made it possible to erroneously get
stuck in the SSH_AUTH_AGENT state.
Reported-by: Tim Stack
Closes#357
connect.c:953:5: warning: initializer element is not computable at load
time
connect.c:953:5: warning: missing initializer for field 'dwMinorVersion'
of 'OSVERSIONINFOEX'
curl_sspi.c:97:5: warning: initializer element is not computable at load
time
curl_sspi.c:97:5: warning: missing initializer for field 'szCSDVersion'
of 'OSVERSIONINFOEX'
Otherwise it would never be called for an HTTP/2 connection, which has
its own disconnect handler.
I spotted this while debugging <https://bugzilla.redhat.com/1248389>
where the http_disconnect() handler was called on an FTP session handle
causing 'dnf' to crash. conn->data->req.protop of type (struct FTP *)
was reinterpreted as type (struct HTTP *) which resulted in SIGSEGV in
Curl_add_buffer_free() after printing the "Connection cache is full,
closing the oldest one." message.
A previously working version of libcurl started to crash after it was
recompiled with the HTTP/2 support despite the HTTP/2 protocol was not
actually used. This commit makes it work again although I suspect the
root cause (reinterpreting session handle data of incompatible protocol)
still has to be fixed. Otherwise the same will happen when mixing FTP
and HTTP/2 connections and exceeding the connection cache limit.
Reported-by: Tomas Tomecek
Bug: https://bugzilla.redhat.com/1248389
Currently, libcurl rejects responses with "Content-Encoding: compress"
when CURLOPT_ACCEPT_ENCODING is set to "". I think that libcurl should
treat the Content-Encoding "compress" the same as other
Content-Encodings that it does not support, e.g. "bzip2". That means
just ignoring it.
MSVC 12 complains:
lib\vtls\openssl.c(1554): warning C4701: potentially uninitialized local
variable 'verstr' used It's a false positive, but as it's normally not,
I have enabled warning-as-error for that warning.
New tool option --ssl-no-revoke.
New value CURLSSLOPT_NO_REVOKE for CURLOPT_SSL_OPTIONS.
Currently this option applies only to WinSSL where we have automatic
certificate revocation checking by default. According to the
ssl-compared chart there are other backends that have automatic checking
(NSS, wolfSSL and DarwinSSL) so we could possibly accommodate them at
some later point.
Bug: https://github.com/bagder/curl/issues/264
Reported-by: zenden2k <zenden2k@gmail.com>
Static analysis indicated that my commit 9008f3d564 ("ntlm_wb: Fix
hard-coded limit on NTLM auth packet size") introduced a potential
memory leak on an error path, because we forget to free the buffer
before returning an error.
Fix this.
Although actually, it never happens in practice because we never *get*
here with state == NTLMSTATE_TYPE1. The state is always zero. That
might want cleaning up in a separate patch.
Reported-by: Terri Oda
setup-vms.h: More symbols for SHA256, hacks for older VAX
openssl.h: Use OpenSSL OPENSSL_NO_SHA256 macro to allow building on VAX.
openssl.c: Use OpenSSL version checks and OPENSSL_NO_SHA256 macro to
allow building on VAX and 64 bit VMS.
setup-vms.h: Symbol case fixups submitted by Michael Steve
build_gnv_curl_pcsi_desc.com: VSI aka as VMS Software, is now the
supplier of new versions of VMS. The install kit needs to accept
VSI as a producer.
Since we do prefix match using given header by application code
against header name pair in format "NAME:VALUE", and VALUE part can
contain ":", we have to careful about existence of ":" in header
parameter. ":" should be allowed to match HTTP/2 pseudo-header field,
and other use of ":" in header must be treated as error, and
curl_pushheader_byname should return NULL. This commit implements
this behaviour.
In 3013bb6 I had changed cookie export to ignore any-domain cookies,
however the logic I used to do so was incorrect, and would lead to a
busy loop in the case of exporting a cookie list that contained
any-domain cookies. The result of that is worse though, because in that
case the other cookies would not be written resulting in an empty file
once the application is terminated to stop the busy loop.
