]> git.ipfire.org Git - thirdparty/curl.git/log
thirdparty/curl.git
3 weeks agomqtt: reject overly big messages
Daniel Stenberg [Sun, 9 Nov 2025 10:12:06 +0000 (11:12 +0100)] 
mqtt: reject overly big messages

Reported-by: Jiyong Yang
Closes #19415

3 weeks agoftp: refactor a piece of code by merging the repeated part
x2018 [Sat, 8 Nov 2025 17:54:51 +0000 (01:54 +0800)] 
ftp: refactor a piece of code by merging the repeated part

Closes #19411

3 weeks agoasyn-thrdd: release rrname if ares_init_options fails
x2018 [Sat, 8 Nov 2025 16:37:16 +0000 (00:37 +0800)] 
asyn-thrdd: release rrname if ares_init_options fails

Closes #19410

3 weeks agotftp: release filename if conn_get_remote_addr fails
x2018 [Sat, 8 Nov 2025 15:44:58 +0000 (23:44 +0800)] 
tftp: release filename if conn_get_remote_addr fails

Closes #19409

3 weeks agoopenssl: release ssl_session if sess_reuse_cb fails
x2018 [Sat, 8 Nov 2025 12:07:54 +0000 (20:07 +0800)] 
openssl: release ssl_session if sess_reuse_cb fails

Closes #19405

3 weeks agoGHA/linux: stop disabling TLS-SRP tests in event-based & duphandle jobs
Viktor Szakats [Sat, 8 Nov 2025 21:54:08 +0000 (22:54 +0100)] 
GHA/linux: stop disabling TLS-SRP tests in event-based & duphandle jobs

They were disabled since these jobs ran in Zuul. The tests are 320, 321,
322, 323, 324. Of which, 323 runs in CI, the rest needs `gnutls-serv`
with SRP enabled, which is not available in current jobs and no longer
offered by Ubuntu's `gnutls-bin` package. 324 doesn't appear as
a skipped test, 323 seems to be running fine, the rest are logged as
skipped. This suggests it's safe to drop the exceptions.

Closes #19413

3 weeks agoGHA/linux: build and test LibreSSL with Fil-C curl, enable pytests
Viktor Szakats [Sat, 8 Nov 2025 12:54:11 +0000 (13:54 +0100)] 
GHA/linux: build and test LibreSSL with Fil-C curl, enable pytests

Build and cache LibreSSL locally with Fil-C and enable it in the Fil-C
job.

Also:
- disable test 776 in the Fil-C job. It fails consistently, and due to
  flakiness seen earlier its result is disabled. In this job it seems to
  be adding 1 to 9 minues to the test run step and fails consistently.
- include Fil-C version in the LibreSSL cache key to prepare for Fil-C
  ABI changes.
- GHA/linux: fully quote `tflags` values to avoid breaking YAML.

Tested and confirmed working with OpenSSL too, but ended up with
LibreSSL for faster, smaller builds.

Closes #19407

3 weeks agopytest: skip H2 tests if feature missing from curl
Viktor Szakats [Sat, 8 Nov 2025 17:23:32 +0000 (18:23 +0100)] 
pytest: skip H2 tests if feature missing from curl

To allow running pytests on more curl configurations.

Also delete a redundant H3 feature check from test_17_14_expired_cert.

Cherry-picked from #19407
Closes #19412

3 weeks agoCURLOPT_SSH_KEYFUNCTION.md: fix minor indent mistake in example
Daniel Stenberg [Sat, 8 Nov 2025 16:49:30 +0000 (17:49 +0100)] 
CURLOPT_SSH_KEYFUNCTION.md: fix minor indent mistake in example

3 weeks agoCURLOPT_READFUNCTION.md: clarify the size of the buffer
Daniel Stenberg [Sat, 8 Nov 2025 11:43:24 +0000 (12:43 +0100)] 
CURLOPT_READFUNCTION.md: clarify the size of the buffer

No need to multiply with size as size is always 1 - and documented so.

Closes #19402

3 weeks agoCURLINFO_SCHEME/PROTOCOL: they return the "scheme" for a "transfer"
Daniel Stenberg [Sat, 8 Nov 2025 16:09:31 +0000 (17:09 +0100)] 
CURLINFO_SCHEME/PROTOCOL: they return the "scheme" for a "transfer"

Not protocol. Not for connection.

Closes #19403

3 weeks agoCURLINFO_TLS_SSL_PTR.md: remove CURLINFO_TLS_SESSION text
Daniel Stenberg [Sat, 8 Nov 2025 11:58:13 +0000 (12:58 +0100)] 
CURLINFO_TLS_SSL_PTR.md: remove CURLINFO_TLS_SESSION text

That option is properly documented in its own page.

Closes #19404

3 weeks agoCURLINFO: remove 'get' and 'get the' from each short desc
Daniel Stenberg [Sat, 8 Nov 2025 12:09:50 +0000 (13:09 +0100)] 
CURLINFO: remove 'get' and 'get the' from each short desc

The short descriptions describe the data each info retrieves. The info
itself does not 'get' the data.

This simplifies and shortens the descriptions and make them more
consistent.

Closes #19406

3 weeks agovtls: fix CURLOPT_CAPATH use
Stefan Eissing [Sat, 8 Nov 2025 13:28:38 +0000 (14:28 +0100)] 
vtls: fix CURLOPT_CAPATH use

A regression in curl 8.17.0 led to a customer CAPATH set by the
application (or the curl command) to be ignored unless licurl was built
with a default CAPATH.

Add test cases using `--capath` on the custom pytest CA, generated with
the help of the openssl command when available.

Fixes #19401
Reported-by: Brad King
Closes #19308

3 weeks agoGHA/linux: add minimal Fil-C build with tests
Viktor Szakats [Thu, 6 Nov 2025 22:59:16 +0000 (23:59 +0100)] 
GHA/linux: add minimal Fil-C build with tests

Requirements for Fil-C:
- not to accidentally pick up system headers. E.g. from `/usr/include`
  on Linux. It can happen when any dependency is auto-detected on this
  header path. This makes Fil-C find the wrong system headers, which
  in turn breaks the configuration step in subtle ways (with CMake) and
  less subtle ways (autotools). Then CMake ends up running into an error
  while compiling.
- build all dependencies with Fil-C too.
  (this patch doesn't build any dependencies yet.)
- "unity" mode disabled. It should work, but needs a lot of memory and
  slower than a standard compiler, or a Fil-C non-unity build.
- x86_64 Linux host platform when using the pre-built toolchain.

Observations on a minimal, static build made with no dependencies and
Fil-C 0.674 (based on clang 20.1.8).
- curl tool sizes:
  - cmake, default, w/o -O: 30 MB (gcc 14.2.0: 1.7 MB)
  - cmake, default, w/o -O, stripped: 29.6 MB (gcc: 1.4 MB)
  - cmake, Release, -O3: 7.2 MB (gcc: 1 MB)
  - cmake, Release, -O3, stripped: 6.8 MB (gcc: 0.93 MB)
  - autotools, default, -O2: 7 MB
- libcurl.a size is 32 MB (cmake, default, w/o -O) (gcc: 2.7 MB)
- build times 3-3.5x longer (compared to system gcc 14.2.0):
- all runtests available pass OK.
- all pytests skipped due to missing features/dependencies.
- shared libcurl builds also work (cmake, default: 25 MB libcurl.so and
  5.75 MB (5.6 stripped) curl tool)
- autotools works fine too, with dependencies disabled or set to avoid
  `/usr/include`.

Closes #19391

3 weeks agovtls: handle possible malicious certs_num from peer
x2018 [Fri, 7 Nov 2025 12:51:22 +0000 (20:51 +0800)] 
vtls: handle possible malicious certs_num from peer

For GnuTLS, mbedTLS, Rustls, Schannel and wolfSSL

This check was previously added for OpenSSL in 3df71e6dc23e80466c2d448

Closes #19397

3 weeks agortmp: precaution for a potential integer truncation
x2018 [Fri, 7 Nov 2025 16:43:51 +0000 (00:43 +0800)] 
rtmp: precaution for a potential integer truncation

On some platforms, socket descriptors may use types larger than int.
When these values exceed INT_MAX, conversion to int can truncate to
negative values causing RTMP connection failures, and even accidentally
affect other socket when high-value descriptors map to existing
lower-value sockets after integer conversion. This check ensures socket
values are within the safe range before passing them to the RTMP
library.

Closes #19399

3 weeks agolib: fix gssapi.h include on IBMi
Andrew [Mon, 3 Nov 2025 13:53:00 +0000 (13:53 +0000)] 
lib: fix gssapi.h include on IBMi

Fixes #19336
Closes #19337

3 weeks agogtls: skip session resumption when verifystatus is set
Daniel Stenberg [Fri, 7 Nov 2025 12:10:48 +0000 (13:10 +0100)] 
gtls: skip session resumption when verifystatus is set

Resumed TLS sessions skip OCSP stapled-response verification. Force a
full handshake so verifystatus() runs.

Follow-up to 4bfd7a961521e1fd6aab7610e931d82a342781

Pointed out by ZeroPath

3 weeks agodocs: mention umask need when curl creates files
Daniel Stenberg [Fri, 7 Nov 2025 12:42:36 +0000 (13:42 +0100)] 
docs: mention umask need when curl creates files

for cookies, alt-svc and HSTS, command line and library

Closes #19396

3 weeks agoRELEASE-NOTES: synced
Daniel Stenberg [Fri, 7 Nov 2025 12:06:28 +0000 (13:06 +0100)] 
RELEASE-NOTES: synced

3 weeks agolib: refactor the type of funcs which have useless return and checks
x2018 [Thu, 6 Nov 2025 17:59:00 +0000 (01:59 +0800)] 
lib: refactor the type of funcs which have useless return and checks

Some internal functions always return CURLE_OK.

- Curl_http_proxy_get_destination() does that from bb4032a, (2 years
  ago) And the original inline code does not need to check the status.

- Curl_wildcard_init() does that from e60fe20. (8 years ago)

- Curl_initinfo() does that from a very beginning.

- Curl_pgrsSetDownloadCounter() did not have a return before 914e49b,
  ad051e1 recovered its content (2 years ago) but did not completely
  recovered the changes related to it.

- auth_digest_get_qop_values() does that from 676de7f.

This directly changes their type to void and cleaned the remaining
checks for their return value.

Closes #19386

3 weeks agoosslq: code readability
Stefan Eissing [Fri, 7 Nov 2025 10:50:02 +0000 (11:50 +0100)] 
osslq: code readability

- remove assertions that are unnecessary
- lookup stream after assertions

Closes #19394

3 weeks agorenovate: update ruff less often
Viktor Szakats [Thu, 6 Nov 2025 23:52:12 +0000 (00:52 +0100)] 
renovate: update ruff less often

`ruff` seems to be getting a new release every week. Make renovate bump
it once every month.

Closes #19392

3 weeks agotool: consider (some) curl_easy_setopt errors fatal
Daniel Stenberg [Thu, 6 Nov 2025 15:34:23 +0000 (16:34 +0100)] 
tool: consider (some) curl_easy_setopt errors fatal

Instead of happily ignoring return codes.

Calls that allocate data, like duplicating strings, can fail because of
lack of memory which could then leave the option unset and curl would
unknowingly continue (if the memory shortage was momentary).

Closes #19385

3 weeks agombedtls: fix potential use of uninitialized `nread`
Daniel Stenberg [Fri, 7 Nov 2025 09:39:26 +0000 (10:39 +0100)] 
mbedtls: fix potential use of uninitialized `nread`

When Curl_conn_cf_recv() returns error, the variable might not be
assigned and the tracing output may (harmlessly) use it uninitialized.

Also add a comment about the typecast from size_t to int being fine.

Pointed out by ZeroPath

Closes #19393

3 weeks agosetopt: when setting bad protocols, don't store them
Daniel Stenberg [Thu, 6 Nov 2025 22:39:56 +0000 (23:39 +0100)] 
setopt: when setting bad protocols, don't store them

Both CURLOPT_PROTOCOLS_STR and CURLOPT_REDIR_PROTOCOLS_STR would
previously return error on bad input but would wrongly still store and
keep the partial (unacceptable) result in the handle.

Closes #19389

3 weeks agotool_paramhlp: refuse --proto remove all protocols
Daniel Stenberg [Thu, 6 Nov 2025 22:14:04 +0000 (23:14 +0100)] 
tool_paramhlp: refuse --proto remove all protocols

curl is for transfers so disabling all protocols has to be a mistake.
Previously it would allow this to get set (even if curl_easy_setopt()
returns an error for it) and then let libcurl return error instead.

Updated 1474 accordingly.

Closes #19388

3 weeks agoGHA/macos: replace deleted gcc-12 with gcc-13/gcc-14
Viktor Szakats [Thu, 6 Nov 2025 20:10:12 +0000 (21:10 +0100)] 
GHA/macos: replace deleted gcc-12 with gcc-13/gcc-14

GitHub dropped gcc-12 for the remaining two macos runner images.
Replace it with gcc-13 in normal jobs, and gcc-14 in combination jobs.

Ref: https://github.com/actions/runner-images/commit/f7e2c3f34b4985282b39ba42de9f6862a2f8a242
Ref: https://github.com/actions/runner-images/pull/13249

Ref: https://github.com/actions/runner-images/commit/1c1351b6350d920e6c5c524f3eb80cc48c8069a4
Ref: https://github.com/actions/runner-images/pull/13253

Closes #19387

3 weeks agotests: replace remaining CR bytes with the new macro `%CR`
Viktor Szakats [Mon, 3 Nov 2025 20:38:35 +0000 (21:38 +0100)] 
tests: replace remaining CR bytes with the new macro `%CR`

There is no more mixed-newline file in the repository after this patch.
Except for`.bat` and `.sln` files (4 in total), all files use LF
newlines.

Also:
- `spacecheck.pl`: drop mixed-EOL exception for test data.
- runtests: add option `-w` to check if test data has stray CR bytes in
  them.
- build: enable the option above in test targets, except the CI-specific
  one where `spacecheck.pl` does this job already.
- tested OK (with expected failures) in CI with stray CRs added.
- cmake: enable option `-a` for the `tests` target. To continue testing
  after a failed test.

Follow-up to 63e9721b63d01518db83a664bc1e8373c352879e #19313
Follow-up to 6cf3d7b1b161bc45501d17b401225befe3c43943 #19318
Follow-up to 4d2a05d3fe8ba4db9168b03057029ea5ce7dab77 #19284

Closes #19347

3 weeks agowolfSSL: able to differentiate between IP and DNS in alt names
Juliusz Sosinowicz [Tue, 4 Nov 2025 17:39:43 +0000 (18:39 +0100)] 
wolfSSL: able to differentiate between IP and DNS in alt names

Fix implemented in https://github.com/wolfSSL/wolfssl/pull/9380

Closes #19364

3 weeks agotool_help: add checks to avoid unsigned wrap around
x2018 [Wed, 5 Nov 2025 15:50:51 +0000 (23:50 +0800)] 
tool_help: add checks to avoid unsigned wrap around

Closes #19377

3 weeks agocurl: fix progress meter in parallel mode
Stefan Eissing [Thu, 6 Nov 2025 11:47:33 +0000 (12:47 +0100)] 
curl: fix progress meter in parallel mode

With `check_finished()` triggered by notifications now, the
`progress_meter()` was no longer called at regular intervals.

Move `progress_meter()` out of `check_finished()` into the perform loop
and event callbacks.

Closes #19383

3 weeks agocmake: disable `CURL_CA_PATH` auto-detection if `USE_APPLE_SECTRUST=ON`
Viktor Szakats [Thu, 6 Nov 2025 01:20:12 +0000 (02:20 +0100)] 
cmake: disable `CURL_CA_PATH` auto-detection if `USE_APPLE_SECTRUST=ON`

Syncing behavior with `CURL_CA_BUNDLE` and autotools.

`/etc/ssl/certs` is empty by default on macOS systems, thus no likely
auto-detection finds something there.

Follow-up to eefd03c572996e5de4dec4fe295ad6f103e0eefc #18703

Closes #19380

3 weeks agoconncache: silence `-Wnull-dereference` on gcc 14 RISC-V 64
Viktor Szakats [Wed, 5 Nov 2025 18:53:30 +0000 (19:53 +0100)] 
conncache: silence `-Wnull-dereference` on gcc 14 RISC-V 64

A false positive that appeared after a recent patch for no reason.

Seen in curl-for-win unity native Linux builds on debian:testing and
debian:trixie with gcc 14.3.0 and 14.2.0 respectively:
```
-- curl version=[8.17.1-DEV]
-- The C compiler identification is GNU 14.2.0
-- Cross-compiling: Linux/x86_64 -> Linux/riscv64
[...]
lib/conncache.c: In function 'Curl_cpool_conn_now_idle':
lib/conncache.c:539:11: error: null pointer dereference [-Werror=null-dereference]
  539 |   if(!data->multi->maxconnects) {
      |       ~~~~^~~~~~~
```
Ref: https://github.com/curl/curl-for-win/actions/runs/19111497271/job/54609512969#step:3:5788

```
-- The C compiler identification is GNU 14.3.0
```
Ref: https://github.com/curl/curl-for-win/actions/runs/19111497271/job/54609512899#step:3:5801

Patch confirmed silencing:
https://github.com/curl/curl-for-win/actions/runs/19112580362/job/54613288202

Follow-up to fbc4d59151dc4a56052f3a92da3682dc97b32148 #19271

Closes #19378

3 weeks agognutls: report accurate error when TLS-SRP is not built-in
Viktor Szakats [Tue, 4 Nov 2025 17:37:49 +0000 (18:37 +0100)] 
gnutls: report accurate error when TLS-SRP is not built-in

With GnuTLS 3.8.0+ the build-time SRP feature detection always succeeds.
It's also disabled by default in these GnuTLS versions.

When using TLS-SRP without it being available in GnuTLS, report
the correct error code `CURLE_NOT_BUILT_IN`, replacing the out of memory
error reported before this patch.

Also add comments to autotools and cmake scripts about this feature
detection property.

Detecting it at build-time would need to run code which doesn't work
in cross-builds. Once curl requires 3.8.0 as minimum, the build-time
checks can be deleted.

```
# before:
curl: (27) gnutls_srp_allocate_client_cred() failed: An unimplemented or disabled feature has been requested.
# after:
curl: (4) GnuTLS: TLS-SRP support not built in: An unimplemented or disabled feature has been requested.
```

Ref: https://github.com/gnutls/gnutls/commit/dab063fca2eecb9ff1db73234108315c5b713756
Ref: https://github.com/gnutls/gnutls/commit/a21e89edacfe4ec3c501b030fff59c11fd20dcf0

Closes #19365

3 weeks agotool_operate: remove redundant condition
Daniel Stenberg [Thu, 6 Nov 2025 08:52:30 +0000 (09:52 +0100)] 
tool_operate: remove redundant condition

And avoid an early return.

Pointed out by CodeSonar

Closes #19381

3 weeks agotool_ipfs: check return codes better
Daniel Stenberg [Thu, 6 Nov 2025 09:00:52 +0000 (10:00 +0100)] 
tool_ipfs: check return codes better

Closes #19382

3 weeks agodocs: fix checksrc `EQUALSPACE` warnings
Viktor Szakats [Wed, 5 Nov 2025 20:21:34 +0000 (21:21 +0100)] 
docs: fix checksrc `EQUALSPACE` warnings

```
docs/libcurl/opts/CURLOPT_SSL_CTX_DATA.md:86:16
docs/libcurl/opts/CURLOPT_SSL_CTX_FUNCTION.md:139:16
```

Also sync `CURL *` and result variable names with rest of docs.

Follow-up to 6d7e924e80096a7e2cebad16235674fd3d3012af #19375

Closes #19379

3 weeks agotests/Makefile.am: fix 'checksrc' target
Daniel Stenberg [Wed, 5 Nov 2025 14:16:00 +0000 (15:16 +0100)] 
tests/Makefile.am: fix 'checksrc' target

Skip the http and client subdirs as they contain no code to check. The
http clients are in libtests/ now.

Closes #19376

3 weeks agotests: Add tests to validate that path is ignored with -J
Dan Fandrich [Tue, 4 Nov 2025 16:25:41 +0000 (08:25 -0800)] 
tests: Add tests to validate that path is ignored with -J

curl is correctly dropping the Content-Disposition: filename path, but
there was no test ensuring that.

Ref: https://hackerone.com/reports/3408126

3 weeks agoGHA/checksrc: add actionlint, fix or silence issues found
Viktor Szakats [Wed, 5 Nov 2025 01:50:42 +0000 (02:50 +0100)] 
GHA/checksrc: add actionlint, fix or silence issues found

It also does shellcheck on `run:` elements, overlapping with
the homegrown `shellcheck-ci.sh` with the same purpose. But it also does
more and perhaps could replace the script too, especially in curl
sub-repos.

Also:
- GHA/macos: delete potentially useful, but commented, and ultimately
  unused, non-default Xcode-testing logic. It's causing unused matrix
  exceptions, upsetting actionlint.

Ref: https://github.com/rhysd/actionlint

Closes #19373

3 weeks agotests: shorten space and tab macro names
Viktor Szakats [Tue, 4 Nov 2025 03:05:19 +0000 (04:05 +0100)] 
tests: shorten space and tab macro names

Easier to write and read.

Follow-up to d29f14b9cf0d38f3887b6eadc71af16903bc7f5b #19300

Closes #19349

3 weeks agoDockerfile: update debian:bookworm-slim digest to 936abff
renovate[bot] [Tue, 4 Nov 2025 10:52:26 +0000 (10:52 +0000)] 
Dockerfile: update debian:bookworm-slim digest to 936abff

Closes #19348

3 weeks agochecksrc.pl: detect assign followed by more than one space
Daniel Stenberg [Wed, 5 Nov 2025 13:40:18 +0000 (14:40 +0100)] 
checksrc.pl: detect assign followed by more than one space

And fix some code previously doing this.

Closes #19375

3 weeks agotests: remove trailing spaces in server responses
Daniel Stenberg [Wed, 5 Nov 2025 13:30:41 +0000 (14:30 +0100)] 
tests: remove trailing spaces in server responses

Allows us to drop lots of %spc% from test cases making them easier on
the eye.

Closes #19374

3 weeks agoopenssl: remove code handling default version
Daniel Stenberg [Tue, 4 Nov 2025 09:17:28 +0000 (10:17 +0100)] 
openssl: remove code handling default version

Since it is no longer actually kept as default internally, that's just
dead code.

Follow-up to 9d8998c99408e1adf8eba629fad9f
Closes #19354

3 weeks agoRELEASE-NOTES: synced
Daniel Stenberg [Wed, 5 Nov 2025 13:12:23 +0000 (14:12 +0100)] 
RELEASE-NOTES: synced

bumped to 8.17.1 for now

fixed typo in THANKS-filter

3 weeks agogtls: add return checks and optimize the code
x2018 [Tue, 4 Nov 2025 18:16:54 +0000 (02:16 +0800)] 
gtls: add return checks and optimize the code

This commit does the following things:

1. Update the description of gtls_init()

2. In gtls_client_init(), check the invaild SSLVERSION at first. Note
   that this part refactors the duplicate/incompatible checks and removes
   the useless local variable `sni`.

3. Check the return value of gnutls_ocsp_resp_init(). Although the
   original code is safe because gnutls_ocsp_resp_import() will check
   the validity of `ocsp_resp`, it is better to catch the error in time
   and record the proper message to output log.

Closes #19366

3 weeks agolib: cleanup for some typos about spaces and code style
x2018 [Wed, 5 Nov 2025 07:28:56 +0000 (15:28 +0800)] 
lib: cleanup for some typos about spaces and code style

Closes #19370

3 weeks agom4/sectrust: fix test(1) operator
Thomas Klausner [Wed, 5 Nov 2025 09:10:03 +0000 (10:10 +0100)] 
m4/sectrust: fix test(1) operator

'=' is the operator defined by POSIX, only bash supports '=='

Closes #19371

3 weeks agoGHA/non-native: revert to OpenBSD 7.7 due to test hangs with 7.8
Viktor Szakats [Wed, 5 Nov 2025 09:18:32 +0000 (10:18 +0100)] 
GHA/non-native: revert to OpenBSD 7.7 due to test hangs with 7.8

test 701 (SOCKS5) and 708 (SOCKS4) started hanging occasionally, and
most likely others too.

https://github.com/curl/curl/actions/runs/19081279902/job/54510279013 (701 hangs) https://github.com/curl/curl/actions/runs/19095657593/job/54555001348?pr=19370 (708 hangs)
https://github.com/curl/curl/actions/runs/19097996671/job/54562669865?pr=19371 (unknown test hangs)

Reverts c3b890b2c005401e18b54dacf9e63d33412e2b4f #19368

Closes #19372

3 weeks agoGHA/windows: switch a dl-mingw job to skeeto/w64devkit gcc 15.1.0
Viktor Szakats [Tue, 4 Nov 2025 23:57:21 +0000 (00:57 +0100)] 
GHA/windows: switch a dl-mingw job to skeeto/w64devkit gcc 15.1.0

To add another, so far untested standalone toolchain variant to the mix.
This distro is a fairly compact, GCC mingw-w64.

Replacing an existing 15.0.1 snapshot toolchain build job.

Ref: https://github.com/skeeto/w64devkit/releases

Closes #19369

3 weeks agoRELEASE-NOTES: synced curl-8_17_0
Daniel Stenberg [Mon, 3 Nov 2025 21:41:16 +0000 (22:41 +0100)] 
RELEASE-NOTES: synced

version 8.17.0 relese

3 weeks agoVERSIONS: 8.18.0 is now pending
Daniel Stenberg [Wed, 5 Nov 2025 06:59:38 +0000 (07:59 +0100)] 
VERSIONS: 8.18.0 is now pending

3 weeks agoTHANKS: add contributors from 8.17.0
Daniel Stenberg [Wed, 5 Nov 2025 06:57:10 +0000 (07:57 +0100)] 
THANKS: add contributors from 8.17.0

4 weeks agoGHA/non-native: bump to OpenBSD 7.8
Viktor Szakats [Tue, 4 Nov 2025 20:01:32 +0000 (21:01 +0100)] 
GHA/non-native: bump to OpenBSD 7.8

Follow-up to e5cc5640b37672bd18d7561bc45c5dd91271753a #19367

Closes #19368

4 weeks agoGHA: update cross-platform-actions/action action to v0.30.0
renovate[bot] [Tue, 4 Nov 2025 19:41:30 +0000 (19:41 +0000)] 
GHA: update cross-platform-actions/action action to v0.30.0

Closes #19367

4 weeks agogtls: check the return value of gnutls_pubkey_init()
x2018 [Tue, 4 Nov 2025 15:27:49 +0000 (23:27 +0800)] 
gtls: check the return value of gnutls_pubkey_init()

Closes #19362

4 weeks agocmake: fix `HAVE_GNUTLS_SRP` detection after adding local FindGnuTLS module
Viktor Szakats [Tue, 4 Nov 2025 14:26:29 +0000 (15:26 +0100)] 
cmake: fix `HAVE_GNUTLS_SRP` detection after adding local FindGnuTLS module

When GnuTLS is detected via pkg-config on a non-default path, e.g. with
Homebrew arm64 (`/opt/homebrew/`).

This was a regression from a commit made in this release cycle.

The Find module doesn't return an absolute path to the detected library
(as the former solution did), but a bare libname and a libpath. We thus
need to explicitly use the libpath while detecting a feature in GnuTLS
found this way. Syncing this with other dependencies.

Follow-up to 1966c86d71eb90beeeb3ccbefd6321bd64992553 #19163

Closes #19360

4 weeks agoHISTORY: extend
Daniel Stenberg [Tue, 4 Nov 2025 15:10:09 +0000 (16:10 +0100)] 
HISTORY: extend

With recent events and some more in the past

Closes #19361

4 weeks agowcurl: import v2025.11.04
Samuel Henrique [Tue, 4 Nov 2025 08:33:28 +0000 (08:33 +0000)] 
wcurl: import v2025.11.04

Closes #19353

4 weeks agotool_ipfs: check the return value of curl_url_get for gwpath
x2018 [Tue, 4 Nov 2025 08:49:28 +0000 (16:49 +0800)] 
tool_ipfs: check the return value of curl_url_get for gwpath

Closes #19358

4 weeks agotests: remove most user-agent filters
Daniel Stenberg [Tue, 4 Nov 2025 10:22:32 +0000 (11:22 +0100)] 
tests: remove most user-agent filters

Use the %VERSION instead. The user-agent stripping was introduced at the
time before we had %VERSION (introduced in e6b21d4). The tests would
then remove the user-agent header to make them possible to be compared
in a version independent way.

Fixes #19355
Reported-by: Stefan Eissing
Closes #19356

4 weeks agoCURLOPT_COPYPOSTFIELDS.md: used with MQTT and RTSP as well
Daniel Stenberg [Tue, 4 Nov 2025 07:44:28 +0000 (08:44 +0100)] 
CURLOPT_COPYPOSTFIELDS.md: used with MQTT and RTSP as well

Follow-up to 5ec87346a9bfad1a24f97c3785

Closes #19351

4 weeks agoBINDINGS: change dead link to archive.org version
Daniel Stenberg [Tue, 4 Nov 2025 07:50:11 +0000 (08:50 +0100)] 
BINDINGS: change dead link to archive.org version

The Hollywood binding host name www.hollywood-mal.com does not seem to
work anymore.

Closes #19352

4 weeks agoREADME.md: use the first paragraph from the man page
Daniel Stenberg [Mon, 3 Nov 2025 13:40:36 +0000 (14:40 +0100)] 
README.md: use the first paragraph from the man page

Which also mentions all protocols

Closes #19335

4 weeks agoCURLOPT_POSTFIELDSIZE*: these also work for MQTT and RTSP
Daniel Stenberg [Mon, 3 Nov 2025 19:25:43 +0000 (20:25 +0100)] 
CURLOPT_POSTFIELDSIZE*: these also work for MQTT and RTSP

Closes #19346

4 weeks agortsp: use explicit postfieldsize if specified
Joshua Rogers [Mon, 3 Nov 2025 16:15:22 +0000 (00:15 +0800)] 
rtsp: use explicit postfieldsize if specified

Signed-off-by: Joshua Rogers <MegaManSec@users.noreply.github.com>
Closes #19345

4 weeks agotests: avoid hard-coded CRLFs in more sections
Viktor Szakats [Fri, 31 Oct 2025 14:50:01 +0000 (15:50 +0100)] 
tests: avoid hard-coded CRLFs in more sections

- `reply/data*`, `verify/stdout`, `verify/stderr`, `verify/file*`,
  `verify/proxy`:
  - make `crlf="yes"` force CRLF to all lines, instead of just applying
    to HTTP protocol headers.
  - add support for `crlf="headers"` that only converts HTTP protocol
    header lines to CRLF. (previously done via `crlf="yes"`.)
  - use `crlf="headers"` where possible.

- `reply/connect*`:
  - add support for `crlf="yes"` and `crlf="headers"`.
  - use them where possible.

- `client/file*`, `client/stdin`:
  - add support for `crlf="yes"`.
  - use it where possible.

- `reply/data*`, `verify/protocol`:
  - replace existing uses of `crlf="yes"` with `crlf="headers`" where it
    does not change the result.

Reducing the number of `tests/data/test*`:
- CRLF newlines from 10295 to 1985. (119985 lines total)
- files with mixed newlines from 656 to 113. (1890 files total)

After this patch there remain 141 sections with mixed newlines, where
the mixing is not split between headers/non-headers. There is no obvious
pattern here. Some of the CRLF uses might be accidental, or
non-significant. They will be tackled in a future patch.

Follow-up to 6cf3d7b1b161bc45501d17b401225befe3c43943 #19318
Follow-up to 4d2a05d3fe8ba4db9168b03057029ea5ce7dab77 #19284

Closes #19313

4 weeks agomulti: check the return value of strdup()
x2018 [Mon, 3 Nov 2025 17:12:42 +0000 (01:12 +0800)] 
multi: check the return value of strdup()

Closes #19344

4 weeks agohttp: check the return value of strdup
x2018 [Mon, 3 Nov 2025 16:47:36 +0000 (00:47 +0800)] 
http: check the return value of strdup

Closes #19343

4 weeks agovtls: check final cfilter node in find_ssl_filter
Joshua Rogers [Fri, 24 Oct 2025 19:49:58 +0000 (03:49 +0800)] 
vtls: check final cfilter node in find_ssl_filter

find_ssl_filter used while(cf && cf->next) and skipped the last node.
If the SSL filter was last, channel binding lookup failed and we returned
CURLE_BAD_FUNCTION_ARGUMENT. Switch to while(cf) so the tail is examined.

This bug was found with ZeroPath.

Closes #19229

4 weeks agogssapi: make channel binding conditional on GSS_C_CHANNEL_BOUND_FLAG
Devdatta Talele [Mon, 20 Oct 2025 15:21:43 +0000 (20:51 +0530)] 
gssapi: make channel binding conditional on GSS_C_CHANNEL_BOUND_FLAG

Fixes #19109 - GSSAPI authentication fails on macOS with Apple's Heimdal
implementation which lacks GSS_C_CHANNEL_BOUND_FLAG support for TLS
channel binding.

Commit 0a5ea09a910e introduced TLS channel binding for SPNEGO/GSSAPI
authentication unconditionally, but Apple's Heimdal fork (used on macOS)
does not support this feature, causing "unsupported mechanism" errors
when authenticating to corporate HTTP services with Kerberos.

Solution:
- Add CURL_GSSAPI_HAS_CHANNEL_BINDING detection in curl_gssapi.h based
  on GSS_C_CHANNEL_BOUND_FLAG presence (MIT Kerberos >= 1.19)
- Make negotiatedata.channel_binding_data field conditional in vauth.h
- Guard channel binding collection/cleanup in http_negotiate.c
- Guard channel binding usage in spnego_gssapi.c

This follows the same pattern as GSS_C_DELEG_POLICY_FLAG detection and
ensures graceful degradation when channel binding is unavailable while
maintaining full support for implementations that have it.

Changes:
- lib/curl_gssapi.h: Add feature detection macro
- lib/vauth/vauth.h: Make struct field conditional
- lib/http_negotiate.c: Conditional init/cleanup (2 locations)
- lib/vauth/spnego_gssapi.c: Conditional channel binding usage

Tested on macOS with Apple Heimdal (no channel binding) and Linux with
MIT Kerberos (with channel binding). Both configurations authenticate
successfully without errors.

Closes #19164

4 weeks agoopenssl: check CURL_SSLVERSION_MAX_DEFAULT properly
Stefan Eissing [Mon, 3 Nov 2025 15:01:56 +0000 (16:01 +0100)] 
openssl: check CURL_SSLVERSION_MAX_DEFAULT properly

The definition of these constants does not give a numeric ordering
and MAX_DEFAULT needs to be checked in addition of ciphers and QUIC
checks to apply correctly.

Fixes #19340
Reported-by: Peter Piekarski
Closes #19341

4 weeks agocw-out: fix EAGAIN handling on pause
Stefan Eissing [Mon, 3 Nov 2025 14:07:57 +0000 (15:07 +0100)] 
cw-out: fix EAGAIN handling on pause

The interim CURLE_AGAIN result was not always converted to a
CURLE_OK and then caused write callers to report a failure.

Fixes #19334
Reported-by: pennae on github
Closes #19338

4 weeks agoREUSE: add copyright header to two files
Viktor Szakats [Mon, 3 Nov 2025 14:24:27 +0000 (15:24 +0100)] 
REUSE: add copyright header to two files

`.mailmap` supports comments and empty lines since at least 2.31.0:
https://git-scm.com/docs/gitmailmap/2.31.0

Closes #19339

4 weeks agoftpserver.pl: fix newlines in 227/229 replies, adjust tests
Viktor Szakats [Sun, 2 Nov 2025 22:25:27 +0000 (23:25 +0100)] 
ftpserver.pl: fix newlines in 227/229 replies, adjust tests

The test FTP server returned LF newlines for 227/229 replies, instead of
the CRLF used for the rest.

Test data added later were explicitly made to expect an LF in these
response lines.

After this patch the FTP server returns CRLF newlines, allowing
to delete this special case in test data.

Follow-up to 3bfff57e1f604ef0b2bc84243014eb9c9a1e6acc
Follow-up to a7937ed49c8f201444323b8a721e1d04e0306105

Closes #19330

4 weeks agotest696: decouple from test556 data
Viktor Szakats [Sun, 2 Nov 2025 15:00:24 +0000 (16:00 +0100)] 
test696: decouple from test556 data

Test 696 and 556 share the same libtest code. Make sure to issue
the `GET` request to the correct runtime test number instead of using
the hard-wired "556".

It makes the `sws` test server read the response string from `test696`
`<data>` section, instead of reading it from `test556`. To avoid this
hidden interaction between test data.

AFAICS there is no other similar hard-coded string in reused libtests.

Ref: https://github.com/curl/curl/pull/19313#issuecomment-3477448933
Follow-up to be82a3605a4b539580b3de776ffcca25b8770e43 #16003
Closes #19329

4 weeks agoimap: avoid integer overflow
Daniel Stenberg [Mon, 3 Nov 2025 07:05:35 +0000 (08:05 +0100)] 
imap: avoid integer overflow

Follow-up to e64c28e243d797da4ef76d6e8959

Spotted by OSS-Fuzz

Closes #19332

4 weeks agoRELEASE-NOTES: synced
Daniel Stenberg [Sun, 2 Nov 2025 22:09:54 +0000 (23:09 +0100)] 
RELEASE-NOTES: synced

4 weeks agohttp_aws_sigv4: check the return value of curl_maprintf()
x2018 [Sun, 2 Nov 2025 05:28:34 +0000 (13:28 +0800)] 
http_aws_sigv4: check the return value of curl_maprintf()

Closes #9328

4 weeks agoGHA/dependabot: tidy-ups
Viktor Szakats [Sun, 2 Nov 2025 16:45:33 +0000 (17:45 +0100)] 
GHA/dependabot: tidy-ups

4 weeks agoGHA: bump the pip-dependencies group across 2 directories with 3 updates
dependabot[bot] [Sat, 1 Nov 2025 14:19:58 +0000 (14:19 +0000)] 
GHA: bump the pip-dependencies group across 2 directories with 3 updates

Closes #19321

4 weeks agoGHA/dependabot: fix update group names
Viktor Szakats [Sun, 2 Nov 2025 16:10:00 +0000 (17:10 +0100)] 
GHA/dependabot: fix update group names

4 weeks agoruntests: fix Perl warning after recent patch
Viktor Szakats [Sun, 2 Nov 2025 03:45:05 +0000 (04:45 +0100)] 
runtests: fix Perl warning after recent patch

```
Use of uninitialized value $hash{"crlf"} in string eq at tests/runtests.pl line 1406.
```

Follow-up to 6cf3d7b1b161bc45501d17b401225befe3c43943 #19318
Closes #19327

4 weeks agotests: avoid more hard-coded CRLFs in `protocol` sections
Viktor Szakats [Sat, 1 Nov 2025 00:28:55 +0000 (01:28 +0100)] 
tests: avoid more hard-coded CRLFs in `protocol` sections

- fix regex to not catch CR (from CRLF), in `PORT`, `EPRT`
  commands, allowing to use `crlf="yes"` more.
- add `crlf="headers"` mode for `protocol` sections.
  To call `subnewlines()` without its force option.
  This is the mode used in `data` sections when `crlf="yes"`.
  (This confusion may be subject to a future commit.)
- subnewlines: apply CRLF to `HEAD` and `CONNECT` HTTP requests.
- subnewlines: apply CRLF to RTSP requests.
- delete remaining empty `protocol` sections.

Reducing the number of `tests/data/test*`:
- CRLF newlines from 11325 to 10295. (119984 lines total)
- files with mixed newlines from 707 to 656. (1890 files total)

Follow-up to 4d2a05d3fe8ba4db9168b03057029ea5ce7dab77 #19284

Closes #19318

4 weeks agovtls: properly handle SSL shutdown timeout
x2018 [Sat, 1 Nov 2025 15:28:07 +0000 (23:28 +0800)] 
vtls: properly handle SSL shutdown timeout

Closes #19323

4 weeks agodoh: cleanup resources on error paths
x2018 [Fri, 31 Oct 2025 19:43:26 +0000 (03:43 +0800)] 
doh: cleanup resources on error paths

Closes #19310

4 weeks agohttpsrr: send HTTPS query to the right target
Daniel Stenberg [Sat, 1 Nov 2025 19:21:25 +0000 (20:21 +0100)] 
httpsrr: send HTTPS query to the right target

When the target host is on a different port than 443, the name
"_[port]._https.[name]" shall be used.

Fixes #19301
Reported-by: Gunni on github
Closes #19324

4 weeks agotests/data: delete stray CRLFs in markup lines
Viktor Szakats [Sat, 1 Nov 2025 03:02:13 +0000 (04:02 +0100)] 
tests/data: delete stray CRLFs in markup lines

Closes #19317

4 weeks agoCI: two display name tweaks
Viktor Szakats [Fri, 31 Oct 2025 23:47:21 +0000 (00:47 +0100)] 
CI: two display name tweaks

- use `AM`/`CM` where missing.
  In GHA/linux-old and AppVeyor CI.
  To denote autotools and CMake, and to align with rest of the jobs.

- rename `Old Linux` to `Linux Old` to align with the rest of Linux
  jobs on GitHub web views sorted by name.

Closes #19316

4 weeks agotests: replace standalone significant tabs with macro
Viktor Szakats [Fri, 31 Oct 2025 22:25:17 +0000 (23:25 +0100)] 
tests: replace standalone significant tabs with macro

Follow-up to d29f14b9cf0d38f3887b6eadc71af16903bc7f5b #19300

Closes #19315

4 weeks agotests/libtest: consistent variable naming for easy/multi/share handles
Viktor Szakats [Fri, 31 Oct 2025 17:36:43 +0000 (18:36 +0100)] 
tests/libtest: consistent variable naming for easy/multi/share handles

Follow-up to 928363f28ca533d743adcb70597c3e30917c6cbd #19299

Closes #19311

4 weeks agoscorecard: more params for upload tests
Stefan Eissing [Fri, 31 Oct 2025 15:44:38 +0000 (16:44 +0100)] 
scorecard: more params for upload tests

Add --upload-parallel=n for controlling upload parallelism. Make upload
processing similar to download processing.

Closes #19302

4 weeks agoschannel: properly close the certfile on error
x2018 [Fri, 31 Oct 2025 15:58:52 +0000 (23:58 +0800)] 
schannel: properly close the certfile on error

Closes #19304

4 weeks agoopenssl: combine all the x509-store flags
Daniel Stenberg [Fri, 31 Oct 2025 16:22:36 +0000 (17:22 +0100)] 
openssl: combine all the x509-store flags

... intead of overwriting the previous ones in ossl_populate_x509_store()

Pointed out by ZeroPath

Closes #19306

4 weeks agosectrust: fix verifystatus via sectrust
Stefan Eissing [Fri, 31 Oct 2025 16:57:53 +0000 (17:57 +0100)] 
sectrust: fix verifystatus via sectrust

When openssl does not verify the certificate, but apple sectrust
does, we also pass it the ocsp stapled response when configured and
available.

When openssl does not verify the cert chain, it will also not be able
to verify the ocsp stapling. Do not call it if sectrust is the
verifier of the cert chain.

Fixes #19307
Reported-by: Harry Sintonen
Closes #19308

4 weeks agoopenssl: fix the ocsp len arg to Curl_vtls_apple_verify
Daniel Stenberg [Fri, 31 Oct 2025 16:09:31 +0000 (17:09 +0100)] 
openssl: fix the ocsp len arg to Curl_vtls_apple_verify

If it has no data, pass in a zero.

Fixes #19303
Reported-by: Harry Sintonen
Closes #19305

4 weeks agospelling: fix new finds by typos-cli 1.39.0
Viktor Szakats [Fri, 31 Oct 2025 20:17:31 +0000 (21:17 +0100)] 
spelling: fix new finds by typos-cli 1.39.0

Closes #19312