Patrick Monnerat [Wed, 28 Sep 2022 16:12:15 +0000 (18:12 +0200)]
lib: sanitize conditional exclusion around MIME
The introduction of CURL_DISABLE_MIME came with some additional bugs:
- Disabled MIME is compiled-in anyway if SMTP and/or IMAP is enabled.
- CURLOPT_MIMEPOST, CURLOPT_MIME_OPTIONS and CURLOPT_HTTPHEADER are
conditioned on HTTP, although also needed for SMTP and IMAP MIME mail
uploads.
In addition, the CURLOPT_HTTPHEADER and --header documentation does not
mention their use for MIME mail.
Daniel Stenberg [Tue, 27 Sep 2022 15:20:23 +0000 (17:20 +0200)]
altsvc: reject bad port numbers
The existing code tried but did not properly reject alternative services
using negative or too large port numbers.
With this fix, the logic now also flushes the old entries immediately
before adding a new one, making a following header with an illegal entry
not flush the already stored entry.
Report from the ongoing source code audit by Trail of Bits.
Daniel Stenberg [Tue, 27 Sep 2022 09:13:21 +0000 (11:13 +0200)]
configure: deprecate builds with small curl_off_t
If curl_off_t turns out to be smaller than 8 bytes,
--with-n64-deprecated needs to be used to allow the build to
continue. This is to highlight the fact that support for such builds is
going away next year.
Daniel Stenberg [Mon, 26 Sep 2022 14:19:50 +0000 (16:19 +0200)]
openssl: make certinfo available for QUIC
Curl_ossl_certchain() is now an exported function in lib/vtls/openssl.c that
can also be used from quiche.c and ngtcp2.c to get the cert chain for QUIC
connections as well.
The *certchain function was moved to the top of the file for this reason.
Viktor Szakats [Mon, 26 Sep 2022 22:31:09 +0000 (22:31 +0000)]
windows: adjust name of two internal public functions
According to `docs/INTERNALS.md`, internal function names spanning source
files start with uppercase `Curl_`. Bring these two functions in
alignment with this.
This also stops exporting them from `libcurl.dll` in autotools builds.
Jay Satiro [Fri, 9 Sep 2022 19:33:47 +0000 (15:33 -0400)]
schannel: ban server ALPN change during recv renegotiation
By the time schannel_recv is renegotiating the connection, libcurl has
already decided on a protocol and it is too late for the server to
select a protocol via ALPN except for the originally selected protocol.
Viktor Szakats [Sun, 25 Sep 2022 22:19:13 +0000 (22:19 +0000)]
autotools: allow --enable-symbol-hiding with windows
This local autotools logic was put in place in 9e24b9c7afbcb81120af4cf3f6cdee49a06d8224 (in 2012) which disabled it for
Windows unconditionally. Testing reveals that it actually works with
tested toolchains (mingw-w64 and CI ones), so let's allow this build
feature on that platform. Bringing this in sync with CMake, which already
supported this.
Viktor Szakats [Sun, 25 Sep 2022 21:58:21 +0000 (21:58 +0000)]
autotools: reduce brute-force when detecting recv/send arg list
autotools uses brute-force to detect `recv`/`send`/`select` argument
lists, by interating through _all_ argument type combinations on each
`./configure` run. This logic exists since 01fa02d0b545e1433dced2430561f8c0c72b74a9 (from 2006) and was a bit later
extended with Windows support.
This results in a worst-case number of compile + link cycles as below:
- `recv`: 96
- `send`: 192
- `select`: 60
Total: 348 (the number of curl C source files is 195, for comparison)
Notice that e.g. curl-for-win autotools builds require two `./configure`
invocations, doubling these numbers.
`recv` on Windows was especially unlucky because `SOCKET` (the correct
choice there) was listed _last_ in one of the outer trial loops. This
resulted in lengthy waits while autotools was trying all invalid
combinations first, wasting cycles, disk writes and slowing down
iteration.
This patch reduces the amount of idle work by reordering the tests in
a way to succeed first on a well-known platform such as Windows, and
also on non-Windows by testing for POSIX prototypes first, on the
assumption that these are the most likely candidates these days. (We do
not touch `select`, where the order was already optimal for these
platforms.)
For non-Windows, this means to try a return value of `ssize_t` first,
then `int`, reordering the buffer argument type to try `void *` first,
then `byte *`, and prefer the `const` flavor with `send`. If we are
here, also stop testing for `SOCKET` type in non-Windows builds.
After the patch, detection on Windows is instantaneous. It should also be
faster on popular platforms such as Linux and BSD-based ones.
If there are known-good variations for other platforms, they can also be
fast-tracked like above, given a way to check for that platform inside
the autotools logic.
Daniel Stenberg [Thu, 22 Sep 2022 09:52:08 +0000 (11:52 +0200)]
mprintf: use snprintf if available
This is the single place in libcurl code where it uses the "native"
s(n)printf() function. Used for writing floats. The use has been
reviewed and vetted and uses a HUGE target buffer, but switching to
snprintf() still makes this safer and removes build-time warnings.
Reported-by: Philip Heiduck
Fixes #9569
Closes #9570
Patrick Monnerat [Wed, 21 Sep 2022 08:41:22 +0000 (10:41 +0200)]
tool: remove protocol count limitation
Replace bit mask protocol sets by null-terminated arrays of protocol
tokens. These are the addresses of the protocol names returned by
curl_version_info().
Protocol names are sorted case-insensitively before output to satisfy CI
tests matches consistency.
The protocol list returned by curl_version_info() is augmented with all
RTMP protocol variants.
Daniel Stenberg [Wed, 21 Sep 2022 10:32:16 +0000 (12:32 +0200)]
test972: verify the output without using external tool
It seems too restrictive to assume and use an external tool to verify
the JSON. This now verifies the outut byte per byte. We could consider
building a local "JSON verifyer" in a future.
Daniel Stenberg [Wed, 21 Sep 2022 06:52:57 +0000 (08:52 +0200)]
hostip: lazily wait to figure out if IPv6 works until needed
The check may take many milliseconds, so now it is performed once the
value is first needed. Also, this change makes sure that the value is
not used if the resolve is set to be IPv4-only.
Daniel Stenberg [Tue, 20 Sep 2022 12:50:09 +0000 (14:50 +0200)]
symbol-scan.pl: scan and verify .3 man pages
This script now also finds all .3 man pages in docs/include and
docs/include/opts, extracts all uses of CURL* symbols and verifies that all
symbols mentioned in docs are defined in public headers.
A "global symbol" is one of those matching a known prefix and the script makes
an attempt to check all/most of them. Just using *all* symbols that match
CURL* proved matching a little too many other references as well and turned
difficult turning into something useful.
Daniel Stenberg [Tue, 20 Sep 2022 11:32:58 +0000 (13:32 +0200)]
test1119: scan all public headers
Previously this test only scanned a subset of the headers, which made us
accidentally miss symbols that were provided in the others. Now, the script
iterates over all headers present in include/curl.
Daniel Stenberg [Tue, 20 Sep 2022 21:30:02 +0000 (23:30 +0200)]
GHA: spellcheck
This spellchecker checks markdown files. For this reason this job
converts all man pages in the repository to markdown with pandoc before
the check runs.
The perl script 'cleanspell' filters out details from the man page in
the process, to avoid the spellchecker trying to spellcheck things it
can't. Like curl specific symbols and the SYNOPSIS and EXAMPLE sections
of libcurl man pages.
The spell checker does not check words in sections that are within pre,
strong and em tags.
'spellcheck.words' is a custom word list with additional accepted words.
Daniel Stenberg [Wed, 21 Sep 2022 07:06:49 +0000 (09:06 +0200)]
connect: fix the wrong error message on connect failures
The "Failed to connect to" message after a connection failure would
include the strerror message based on the presumed previous socket
error, but in times it seems that error number is not set when reaching
this code and therefore it would include the wrong error message.
The strerror message is now removed from here and the curl_easy_strerror
error is used instead.
Daniel Stenberg [Wed, 21 Sep 2022 07:32:12 +0000 (09:32 +0200)]
httpput-postfields.c: shorten string for C89 compliance
httpput-postfields.c:41:3: error: string length ‘522’ is greater than the length ‘509’ ISO C90 compilers are required to support [-Woverlength-strings]
41 | "this chapter.";
| ^~~~~~~~~~~~~~~
Patrick Monnerat [Wed, 21 Sep 2022 01:00:30 +0000 (03:00 +0200)]
unit test 1655: make it C89-compliant
Initializations performed in unit test 1655 use automatic variables in
aggregates and thus can only be computed at run-time. Using gcc in C89
dialect mode produces warning messages like:
unit1655.c:96:7: warning: initializer element is not computable at load time [-Wpedantic]
96 | { toolong, DOH_DNS_NAME_TOO_LONG }, /* expect early failure */
| ^~~~~~~
Fix the problem by converting these automatic pointer variables to
static arrays.
resolve: make forced IPv4 resolve only use A queries
This protects IPv4-only transfers from undesired bad IPv6-related side
effects and make IPv4 transfers in dual-stack libcurl behave the same
way as in IPv4 single-stack libcurl.
Marc Hoersken [Sat, 17 Sep 2022 21:45:32 +0000 (23:45 +0200)]
CI/GHA: cancel outdated CI runs on new PR changes
Avoid letting outdated CI runs continue if a PR receives
new changes. Outside a PR we let them continue running
by tying the concurrency to the commit hash instead.
Also only let one CodeQL or Hacktoberfest job run at a time.
Other CI platforms we use have this build in, but GitHub
unfortunately neither by default nor with a simple option.
This saves CI resources and therefore a little energy.
Approved-by: Daniel Stenberg Approved-by: Max Dymond
Closes #9533
Viktor Szakats [Sun, 18 Sep 2022 12:05:35 +0000 (12:05 +0000)]
windows: add .rc support to autotools builds
After this update autotools builds will compile and link `.rc` resources
to Windows executables. Bringing this feature on par with CMake and
Makefile.m32 builds. And also making it unnecessary to improvise these
steps manually, while monkey patching build files, e.g. [0].
You can customize the resource compiler via the `RC` envvar, and its
options via `RCFLAGS`.
This harmless warning may appear throughout the build, even though the
autotools manual documents [1] `RC` as a valid tag, and it fails when
omitting one:
`libtool: error: ignoring unknown tag RC`
Patrick Monnerat [Thu, 15 Sep 2022 12:31:36 +0000 (14:31 +0200)]
cli tool: do not use disabled protocols
As they are now rejected by the library, take care of not passing
disabled protocol names to CURLOPT_PROTOCOLS_STR and
CURLOPT_REDIR_PROTOCOLS_STR.
Rather than using the CURLPROTO_* constants, dynamically assign protocol
numbers based on the order they are listed by curl_version_info().
New type proto_set_t implements prototype bit masks: it should therefore
be large enough to accomodate all library-enabled protocols. If not,
protocol numbers beyond the bit count of proto_set_t are recognized but
"inaccessible": when used, a warning is displayed and the value is
ignored. Should proto_set_t overflows, enabled protocols are reordered to
force those having a public CURLPROTO_* representation to be accessible.
Code has been added to subordinate RTMP?* protocols to the presence of
RTMP in the enabled protocol list, being returned by curl_version_info()
or not.
Daniel Stenberg [Wed, 14 Sep 2022 07:18:30 +0000 (09:18 +0200)]
urlapi: detect scheme better when not guessing
When the parser is not allowed to guess scheme, it should consider the
word ending at the first colon to be the scheme, independently of number
of slashes.
The parser now checks that the scheme is known before it counts slashes,
to improve the error messge for URLs with unknown schemes and maybe no
slashes.
When following redirects, no scheme guessing is allowed and therefore
this change effectively prevents redirects to unknown schemes such as
"data".
Daniel Stenberg [Mon, 12 Sep 2022 07:57:01 +0000 (09:57 +0200)]
urldata: use a curl_prot_t type for storing protocol bits
This internal-use-only storage type can be bumped to a curl_off_t once
we need to use bit 32 as the previous 'unsigned int' can no longer hold
them all then.
The websocket protocols take bit 30 and 31 so they are the last ones
that fit within 32 bits - but cannot properly be exported through APIs
since those use *signed* 32 bit types (long) in places.
formdata: fix warning: 'CURLformoption' is promoted to 'int'
curl/lib/formdata.c: In function 'FormAdd':
curl/lib/formdata.c:249:31: warning: 'CURLformoption' is promoted to 'int' when passed through '...'
249 | option = va_arg(params, CURLformoption);
| ^
curl/lib/formdata.c:249:31: note: (so you should pass 'int' not 'CURLformoption' to 'va_arg')
curl/lib/formdata.c:249:31: note: if this code is reached, the program will abort