This still allows users to explictily ask for 1.0 or 1.1 as the minimum
version. If the TLS library allows it.
Starting with this change, the CURL_SSLVERSION_DEFAULT value is no
longer used as minimum version when the TLS backend are called.
This also makes curl set the minimum version to 1.2 independently of
libcurl for the rare case where a newer curl tool would use an older
libcurl.
URL: https://curl.se/mail/lib-2025-07/0007.html
Assisted-by: Stefan Eissing
Closes#17894
- vtls: fix unused variable and symbols.
- ftp: fix unused variable.
- http: fix unused variables.
- smtp: fix unsued variable.
- wolfssl: fix unused variable with !proxy.
- libssh: fix unused argument.
- curl_trc: sync guards between declaration and definition.
- curl_trc: add missing guard for `Curl_trc_ssls` when !verbose.
- curl_trc: fix errors with !http + http3.
- curl_trc: fix missing function with !http + nghttp2.
- cf-h2-proxy: disable when !http + nghttp2, to avoid calling undeclared
functions.
- sha256: fix missing declaration in rare configs.
- md4: fix symbol conflict when building GnuTLS together with AWS-LC or
wolfSSL. By prioritizing the latter two. AWS-LC has no option
to disable the clashing symbol. wolfSSL does, but the most seamless is
to skip including GnuTLS's standalone `md4.h` to avoid the clash.
- build: fix errors with !http + nghttp2.
- build: catch !ssl + ssls-export combination in source. Convert
build-level errors to warnings.
- build: fix errors with !http + http3.
- build: fix building curl tool and unit1302 in rare combinations.
By always compiling base64 curlx functions.
- cmake: add `_CURL_SKIP_BUILD_CERTS` internal option.
To disable automatically building certs with the testdeps target.
To improve performance when testing builds.
(used locally to find the failing builds fixed in this PR.)
Closes#17962
- codespell: break logic out into its own runnable script. Allowing
to run it on local machines.
- codespell: install via `pip`, bump to latest version.
- codespell: show version number in CI log.
- codespell: drop no longer needed word exception: `msdos`.
- codespell: include all curl source tree, except `packages` and
`winbuild`. Drop an obsolete file exclusion.
- add new spellchecker job using the `typos` tool. It includes
the codespell dictionary and a couple more. Use linuxbrew to install
it. This takes 10 seconds, while installing via `cargo` from source
would take over a minute.
- codespell: introduce an inline ignore filter compatible with `cspell`
Make `typos` recognize it, too. Move single exceptions inline.
Fix new typos found. Also rename variables and words to keep
spellchecking exceptions at minumum. This involves touching some tests.
Also switch base64 strings to `%b64[]` to avoid false positives.
Ref: https://github.com/crate-ci/typos/blob/master/docs/reference.md
Ref: https://github.com/codespell-project/codespell?tab=readme-ov-file#inline-ignore
Ref: https://github.com/codespell-project/codespell/issues/1212#issuecomment-1721152455
Ref: https://cspell.org/docs/Configuration/document-settingsCloses#17905
To fix test 1308 in MultiSSL builds.
Failure was caused by the random number generator virtual function being
NULL, instead of pointing to the implementation in the runtime-selected
TLS backend. This could happen in MultiSSL builds when a functionality
was asking for a random number without triggering a VTLS function table
initialization first. Such functionality is MIME, or form data via MIME.
The reason CI did not catch it in an earlier MultiSSL GHA/windows job,
is that it was a debug-enabled one. In debug-enabled builds the test
runner was overriding the random number generator for all tests.
Fixed this by moving the override to the tests requiring it, via
1fcf22585f#17971, enabling debug builds
to catch this issue.
Enable MultiSSL in two CI jobs, to verify this patch.
Fixing:
```
test 1308...[formpost tests]
libtests returned 44, when expecting 0
1308: exit FAILED
[...]
=== Start of file stderr1308
URL: log/3/test-1308
tests/libtest/lib1308.c:70 Assertion 'res == 0' FAILED: curl_formget returned error
tests/libtest/lib1308.c:72 Assertion 'total_size == 518' FAILED: curl_formget got wrong size back
tests/libtest/lib1308.c:88 Assertion 'res == 0' FAILED: curl_formget returned error
tests/libtest/lib1308.c:89 Assertion 'total_size == 899' FAILED: curl_formget got wrong size back
```
Ref: https://github.com/curl/curl/actions/runs/16387693424/job/46309536359?pr=17963#step:16:2515
Bug: https://github.com/curl/curl/pull/17963#issuecomment-3092282057Closes#17970
- de-complex Curl_gtls_verifyserver() by splitting of static
functions for parts of it.
- follow the `goto out` style with common deallocation code
Closes#17941
- rename Curl_oss_check_peer_cert() to Curl_ossl_check_peer_cert()
- leave altname match loop after the first success when the match
was an ip address
- remove static subj_alt_hostcheck() since it did not really do much
- use length based infof() output of altname, even though it does
seem always to be nul terminated
Closes#17940
The `transport` to use for a transfer, e.g. TCP/QUIC/UNIX/UDP, is
initially selected by options and protocol used. This is set at the
`struct connectdata` as `transport` member.
During connection establishment, this transport may change due to
Alt-Svc or Happy-Eyeballing. Most common is the switch from TCP to QUIC.
Rename the connection member to `transport_wanted` and add a way to
query the connection for the transport in use via a new connection
filter query.
The filter query can also be used in the happy eyeballing attempts when
code needs to know which transport is used by the "filter below". This
happens in wolfssl initialization, as one example.
Closes#17923
In gtls.c there was a typo'd comment that I think was missing the word
"no" to indicate there's "no GnuTLS way to signal no EarlyData".
This commit fixes that typo, and also updates a copy-pasted instance
that made it into wolfssl.c where it should refer to the WolfSSL API
missing the capability, not GnuTLS.
Closes#17907
This callback was permanently mapped to libcurl's internal
`Curl_wcsdup()`, which always uses the customizable malloc for
allocation, thus making a custom mapping redundant anyway.
To simplify, drop the callback and map `_tcsdup()` in Unicode mode
directly to `Curl_wcsdup()`.
Also fixes:
- `curl_global_init()` which, before this patch, (re)initialized its
mapping to `_wcsdup()`, returning buffers potentially incompatible
with a custom allocator.
Bug: https://github.com/curl/curl/pull/17840#issuecomment-3044361245
Bug: https://github.com/curl/curl/pull/7540#issuecomment-2380995349
Co-reported-by: Luca Kellermann
Follow-up to 76e047fc27#7540
Assisted-by: Jay Satiro
Closes#17843
Replace the old Curl_ssl_get_internals() with a new connection filter
query to retrieve the information. Implement that filter query for TCP
and QUIC TLS filter types.
Add tests in client tls_session_reuse to use the info option and check
that pointers are returned.
Reported-by: Larry Campbell
Fixes#17801Closes#17809
Fixing:
- a raw `free()` in ECH code that's malloced in lib code, causing
an invalid free, also reported by valgrind (in non-unity builds).
And in unity builds adjusted to behave like non-unity via #17827:
Ref: https://github.com/curl/curl/actions/runs/16093372427/job/45421778472?pr=17827#step:39:3321
- a local pair of `malloc()`/`free()` to use curl's memory allocators,
and participate in memory tracking when enabled.
Cherry-picked from #17827Closes#17830
Commit f2ce6c46 among other things added the use of own library context
instead of the default context. Default context has access to OpenSSL
configuration file, own context doesn't have it.
Therefore if a pkcs11 provider is loaded via config file, the function
OSSL_PROVIDER_available() incorrectly detects the provider as
unavailable.
Fix this by loading the OpenSSL config to the library context according
to OpenSSL documentation:
"OSSL_LIB_CTX_load_config() loads a configuration file using the given
ctx. This can be used to associate a library context with providers that
are loaded from a configuration."
Moreover use the provider_loaded flag instead of provider pointer to
determine if a provider is available, as the latter is not set when the
provider is loaded from a configuration.
Closes#17804
Drop `strcasecompare` and `strncasecompare` in favor of libcurl API
calls `curl_strequal` and `curl_strnequal` respectively.
Also drop unnecessary `strcase.h` includes. Include `curl/curl.h`
instead where it wasn't included before.
Closes#17772
- tests/libtest: move exception to `stub_gssapi.h`.
- tests/libtest: move remaining exception to `testtrace.c`.
- tests/server: drop obsolete exception.
- docs/examples: move `BANNEDFUNC` exceptions to local files (3 lines).
- docs/examples: move `ERRNOVAR` exception to `ephiperfifo.c`.
- docs/examples: drop `typedef struct` (8 files).
- lib/curlx: add `.checksrc` with banned funcs copied from lib.
- checksrc: ban `strncpy`, `strtok_r`, `strtoul` by default.
Drop local bans. Add exception for `strtoul` to `tests/server'.
- lib, src: sync banned funcs.
Also:
- REUSE: drop `stunnel.pem`, it no longer exists.
- docs/examples: formatting.
- docs/examples: simplify some `sizeof()`s.
Closes#17764
Change the signature of `bufq` functions from
* `ssize_t Curl_bufq_*(..., CURLcode *err)` to
* `CURLcode Curl_bufq_*(..., size_t *pn)`
This allows us to write slightly less code and avoids the ssize_t/size_t
conversions in many cases. Also, it gets the function in line with all
the other send/recv signatures.
Added helper functions in `cfilters.h` for sending from/receving into
a bufq.
Fuzzer now fails to build due to these changes and its testing of
the bufq API.
Closes#17396
For consistency. RFC 3986 section 2.1 says:
"URI producers and normalizers should use uppercase hexadecimal digits
for all percent-encodings."
Reported-by: Jeroen Ooms
Fixes#17685Closes#17739
Caught by gcc (Ubuntu 12.3.0-1ubuntu1~22.04) 12.3.0:
```
vtls/rustls.c: In function ‘cr_connect’:
vtls/rustls.c:857:61: warning: ‘%s’ directive argument is null [-Wformat-overflow=]
857 | failf(data, "rustls: must provide certificate with key '%s'",
| ^~
```
Closes#17704
Connection filters had a method `get_host()` which had not really been
documented. Since then, the cf had the `query()` method added. Replace
the separate get_host with query.
Add `CF_QUERY_HOST_PORT` as query to connection filters to retrieve
which remote hostname and port the filter (or its sub-filter) is talking
to. The query is implemented by HTTP and SOCKS filters, all others pass
it through.
Add `Curl_conn_get_current_host()` to retrieve the remote host and port
for a connection. During connect, this will return the host the
connection is talking to right now. Before/After connect, this will
return `conn->host.name`.
This is used by SASL authentication.
Closes#17419
- Return CURLE_OK if the engine successfully loaded.
Prior to this change:
When loading an openssl engine, the result code is initialized to
CURLE_SSL_ENGINE_NOTFOUND, but is never set to CURLE_OK when the engine
was successfully loaded. This causes curl to error out, falsely stating
engine not found when it actually was.
Broken since f2ce6c46 (precedes 8.14.0) which added support for using
engines and providers at the same time.
Fixes https://github.com/curl/curl/issues/17617
Closes https://github.com/curl/curl/pull/17618
- Align --cacert behaviour with OpenSSL and LibreSSL.
This changes the default behavior of Schannel manual certificate
verification, which is used when the user provides their own CA
certificates for verification, to accept partial chains. In other words,
the user may provide an intermediate certificate without having to
provide the root CA.
Win8/Server2012 widened the PKIX chain traversal API to allow
certificate traversal to terminate at an intermediate.
This behaviour (terminate at the fist matching intermediate) is the
default for LibreSSL and OpenSSL (with OpenSSL allowing control via
CURLSSLOPT_NO_PARTIALCHAIN).
This change uses the new API if it is available, and also allows the
behaviour to revert legacy if CURLSSLOPT_NO_PARTIALCHAIN is present.
Closes https://github.com/curl/curl/pull/17418
`SSL_pending()` only checks if the *current* TLS packet has more data.
There might be more data in SSL's buffers.
`SSL_has_pending()` only checks if there is data in buffers, but does
*not* check if there is a complete TLS packet that can be decoded.
If we only check the first, we will poll on socket events without having
processed all data and may stall. If we only check the second, we would
busy loop without SSL_read() ever giving something.
Add the flag `connssl->input_pending` that is set on incoming data in
the BIO receive. Clear the flag when encountering a CURLE_AGAIN on
the filters receive (via SSL_read()) or see an EOF.
Ref: #17596Closes#17601
The secure transport vTLS backend was the only Curl_ssl struct instance
that populated the false_start field. Since its removed, we can now
remove that field entirely. This was a protocol feature specific to TLS
1.2 that has been replaced by the more widely adopted TLS 1.3 early data
mechanisms.
--false-start is now deprecated
Closes#17595
cfilter/conn: change send/recv function signatures. Unify the
calling/return conventions in our send/receive handling.
Curl_conn_recv(), adjust pnread type
Parameter `pnread` was a `ssize_t *`, but `size_t *` is better since the
function returns any error in its `CURLcode` return value.
Closes#17546
Convert the debug-only handling of SSL_ERROR_SYSCALL so that it is
enabled in all builds with openssl. This should not make a difference in
supported OpenSSL versions, but if whatever version or fork we link
against *does* return SSL_ERROR_SYSCALL, handle this as a fatal error.
Fixes#17471
Reported-by: Michael Kaufmann
Closes#17531
These libraries do not support TLS 1.3 and have been marked for removal
for over a year. We want to help users select a TLS dependency that is
future-proof and reliable, and not supporting TLS 1.3 in 2025 does not
infer confidence. Users who build libcurl are likely to be served better
and get something more future-proof with a TLS library that supports
1.3.
Closes#16677
With a dash, using two Ls. Also for different forms of the word.
Use NULL in all uppercase if it means a zero pointer.
Follow-up to 307b7543eaCloses#17489
Early data was reported as being sent, but was not. While everything was
aligned with the Gods and early data was reported as accepted, the
actual sending required another call to wolfSSL.
Fixes#17481
Reported-by: Ethan Everett
Closes#17488
