]> git.ipfire.org Git - thirdparty/curl.git/log
thirdparty/curl.git
3 weeks agoGHA: verify tarball downloads
Viktor Szakats [Tue, 26 May 2026 13:56:27 +0000 (15:56 +0200)] 
GHA: verify tarball downloads

Detect latest tarball version via the https://curl.se/downloads.html
page, download the signing key from a public keyserver then verify
source download signatures.

To ensure that public downloads are intact.

Closes #21759

3 weeks agoGHA: require HTTPS protocol in redirections (where missing)
Viktor Szakats [Tue, 26 May 2026 13:59:18 +0000 (15:59 +0200)] 
GHA: require HTTPS protocol in redirections (where missing)

Also:
- drop following redirects on `openldap.org`.

Closes #21757

3 weeks agotest1981: explicitly set the locale
Stefan Eissing [Tue, 26 May 2026 08:25:22 +0000 (10:25 +0200)] 
test1981: explicitly set the locale

Otherwise we may get a different month name in the output.

Closes #21749

3 weeks agoGHA: bump LibreSSL to 4.3.2
Viktor Szakats [Tue, 26 May 2026 11:22:20 +0000 (13:22 +0200)] 
GHA: bump LibreSSL to 4.3.2

Also switch back to ftp.openbsd.org download server. More often than not
the GitHub release entry is missing the download artifacts at the time
of detecting a new version, breaking automatic bumps. We cache the
download so it does not bang the origin server with many requests.

Follow-up to 800b0bec18e9c77e35912fac8321c791d7b57863 #19082
Closes #21742
Closes #21754

3 weeks agoGnuTLS: require 3.7.2 for earlydata
Stefan Eissing [Tue, 26 May 2026 09:25:01 +0000 (11:25 +0200)] 
GnuTLS: require 3.7.2 for earlydata

Since all API features we need for TLSv1.3 earlydata support do exist
only from version 3.7.2 onwards, make that the minimal version required.

Fixes #21750
Reported-by: Johannes Schlatow
Closes #21751

3 weeks agomulti_ev: silence clang-tidy nonsense
Stefan Eissing [Tue, 26 May 2026 09:38:25 +0000 (11:38 +0200)] 
multi_ev: silence clang-tidy nonsense

About a "unnecessary define" - my ass.

Closes #21752

3 weeks agodocs: fix odd wording in CONTRIBUTE.md
mik [Thu, 21 May 2026 08:25:49 +0000 (10:25 +0200)] 
docs: fix odd wording in CONTRIBUTE.md

Found with AI assistance, verified manually

Closes #21705

3 weeks agomulti: handle pause in multi socket callback
Daniel Stenberg [Tue, 26 May 2026 07:52:19 +0000 (09:52 +0200)] 
multi: handle pause in multi socket callback

The mev_sh_entry object might be removed if curl_easy_pause() is called
from within the socket callback.

Introduced a 'magic' struct field to to 'mev_sh_entry' to make it easier
to programmatically detect/assert if the pointer is bad - in debug
builds.

Reported-by: Joshua Rogers
Closes #21748

3 weeks agotransfer: clear referer when set to NULL
Daniel Stenberg [Mon, 25 May 2026 14:43:00 +0000 (16:43 +0200)] 
transfer: clear referer when set to NULL

Verify in test 1649

Closes #21741

3 weeks agoVULN-DISCLOSURE-POLICY.md: emphasize the no email thank you part
Daniel Stenberg [Tue, 26 May 2026 07:09:24 +0000 (09:09 +0200)] 
VULN-DISCLOSURE-POLICY.md: emphasize the no email thank you part

Closes #21747

3 weeks agoldap: switch of chasing referrals
Stefan Eissing [Sat, 23 May 2026 12:40:55 +0000 (14:40 +0200)] 
ldap: switch of chasing referrals

It is switched off in the OpenLDAP backend, so we should do the
same here.

Follow-up to cdc1da912066535680f02eb31

Closes #21732

3 weeks agolibssh: map SSH_KNOWN_HOSTS_OTHER to CURLKHMATCH_MISMATCH
Joshua Rogers [Fri, 22 May 2026 07:48:15 +0000 (09:48 +0200)] 
libssh: map SSH_KNOWN_HOSTS_OTHER to CURLKHMATCH_MISMATCH

Host key type mismatch from libssh was incorrectly reported as missing,
causing key callbacks to accept instead of reject.

Reported by: Joshua Rogers (Aisle Research)
Closes #21724

3 weeks agoGHA/http3-linux: fixup quiche cache
Viktor Szakats [Sun, 24 May 2026 13:12:54 +0000 (15:12 +0200)] 
GHA/http3-linux: fixup quiche cache

Closes #21740

3 weeks agoquiche: bump cloudflare/quiche to v0.29.0, update pytest workaround
Viktor Szakats [Sat, 23 May 2026 09:05:27 +0000 (11:05 +0200)] 
quiche: bump cloudflare/quiche to v0.29.0, update pytest workaround

Also:
- drop no longer necessary quiche build workaround.
- update build for boringssl's new location (since v0.29.0, it's no
  longer vendored) within the quiche tree.
- move boringssl install dir out of quiche tree, and shorten it.

Ref: https://github.com/cloudflare/quiche/issues/2277
Ref: https://github.com/cloudflare/quiche/pull/2278
Ref: #21620

Closes #21730

3 weeks agomailmap: cmeister2@gmail is primary for Max Dymond
Max Dymond [Sun, 24 May 2026 09:22:21 +0000 (10:22 +0100)] 
mailmap: cmeister2@gmail is primary for Max Dymond

I'd rather all my commits be attributed to cmeister2@gmail.com instead
of anything else; especially not my old Microsoft email address!

Closes #21735

3 weeks agoKNOWN_BUGS: remove stale Threads::Threads entry
Max Dymond [Sun, 24 May 2026 08:02:33 +0000 (09:02 +0100)] 
KNOWN_BUGS: remove stale Threads::Threads entry

The old CMake bug about exporting -lpthread instead of Threads::Threads
no longer matches current master. As of 2d546d239ecd455b6459e68b85ef8d4b045c0a00
("cmake: use Threads::Threads imported target for POSIX Threads"),
the build now uses Threads::Threads and the generated CMake package
config resolves the dependency explicitly, so this KNOWN_BUGS entry is stale.

Closes #21734

4 weeks agoschannel: error on TLS 1.3-only with cipher list
Joshua Rogers [Fri, 22 May 2026 07:59:17 +0000 (09:59 +0200)] 
schannel: error on TLS 1.3-only with cipher list

The legacy SCHANNEL_CRED path cannot negotiate TLS 1.3. When TLS 1.3
is the only enabled protocol and a cipher list is set, fail instead of
silently downgrading to TLS 1.2.

Fixes https://github.com/curl/curl/issues/21702
Closes https://github.com/curl/curl/pull/21725

4 weeks agoCIPHERS.md: fix the example that uses only TLS 1.3
Jay Satiro [Thu, 21 May 2026 18:00:09 +0000 (14:00 -0400)] 
CIPHERS.md: fix the example that uses only TLS 1.3

- Add --tls-max 1.3 to set the maximum version to TLS 1.3.

- Remove Schannel because it doesn't support TLS 1.3 ciphers since
  6238888.

Prior to this change the example set the minimum version to TLS 1.3 but
not the maximum version to TLS 1.3.

Ref: https://github.com/curl/curl/issues/21702

Closes https://github.com/curl/curl/pull/21719

4 weeks agongtcp2: fail handshake directly
Stefan Eissing [Thu, 21 May 2026 12:21:59 +0000 (14:21 +0200)] 
ngtcp2: fail handshake directly

When certificate verification fails, error out of the handshake
callback, forcing ngtcp2 to stop processing the connection any further.

Closes #21712

4 weeks agoRELEASE-NOTES: synced
Daniel Stenberg [Fri, 22 May 2026 07:48:59 +0000 (09:48 +0200)] 
RELEASE-NOTES: synced

4 weeks agohttp2: remove stream dependency tracking
Stefan Eissing [Fri, 22 May 2026 07:11:41 +0000 (09:11 +0200)] 
http2: remove stream dependency tracking

The HTTP/2 feature is deprecated, few servers implement it and our
implementation is complicated by its state management. Make the two
CURLOPT_* involved a nop and deprecate them.

Closes #21723

4 weeks agoKNOWN_BUGS.md: remove fixed GnuTLS <-> OpenSSL incompat bug
11soda11 [Thu, 21 May 2026 22:03:51 +0000 (00:03 +0200)] 
KNOWN_BUGS.md: remove fixed GnuTLS <-> OpenSSL incompat bug

The entry is about GnuTLS not sending the client cert when it doesn't
match the `DN` the server requested.  OpenSSL does the opposite.

The issue was already fixed by #4958 and removed from KNOWN_BUGS,
but it was added back to the list by #16677, seemingly by mistake.

The issue is still fixed for GnuTLS >= 3.5.0.
As curl only supports GnuTLS >= 3.6.5, remove the bug entry from
KNOWN_BUGS.md

Fixes #21720
Closes #21722

4 weeks agourlapi: fix redirect handling if CURLU_NO_GUESS_SCHEME is set
Daniel Stenberg [Thu, 21 May 2026 21:00:55 +0000 (23:00 +0200)] 
urlapi: fix redirect handling if CURLU_NO_GUESS_SCHEME is set

Verified by test 1967

Reported-by: Joshua Rogers
Closes #21721

4 weeks agoGHA/checksrc: add auditor-level zizmor (warning-only)
Viktor Szakats [Thu, 21 May 2026 17:09:35 +0000 (19:09 +0200)] 
GHA/checksrc: add auditor-level zizmor (warning-only)

CI time cost is 1s.

It may replace existing pedantic check, if this level isn't bringing
false-positives or annoyance. Officially it's not meant for CI, but curl
has been passing this in the last couple of months when checked locally.

Closes #21718

4 weeks agounits: tidy up begin/end blocks
Viktor Szakats [Thu, 21 May 2026 02:21:33 +0000 (04:21 +0200)] 
units: tidy up begin/end blocks

- use `UNITTEST_BEGIN_SIMPLE`/`UNITTEST_END_SIMPLE` where missing.
- drop redundant `(void)arg;` where using `UNITTEST_BEGIN*`.
- unit1636: drop redundant block after `UNITTEST_BEGIN*`.
- unit1609: fix typo in comment.
- unit1627: merge to `if`s.

Closes #21715

4 weeks agomulti: silence gcc 16 `-Wnull-dereference`, bump CI job to test
Viktor Szakats [Thu, 21 May 2026 09:16:49 +0000 (11:16 +0200)] 
multi: silence gcc 16 `-Wnull-dereference`, bump CI job to test

- GHA/windows: bump dl-mingw job from gcc 15 to 16.

- multi: silence warning while building libcurlu:
  ```
  In function 'multi_ischanged',
      inlined from 'multi_socket.isra' at D:/a/curl/curl/lib/multi.c:3282:6:
  D:/a/curl/curl/lib/multi.c:1710:17: error: null pointer dereference [-Werror=null-dereference]
   1710 |   bool retval = (bool)multi->recheckstate;
        |                 ^~~~~~~~~~~~~~~~~~~~~~~~~
  D:/a/curl/curl/lib/multi.c:1712:25: error: null pointer dereference [-Werror=null-dereference]
   1712 |     multi->recheckstate = FALSE;
        |                         ^
  ```
  Ref: https://github.com/curl/curl/actions/runs/26217071531/job/77142119137?pr=21707

- multi: silence another `-Wnull-dereference`, popping up in libcurl
  with gcc 13 after the previous silencing:
  ```
  In function 'Curl_multi_xfers_running',
      inlined from 'multi_socket.isra' at ../../lib/multi.c:3292:28:
  ../../lib/multi.c:4132:15: error: null pointer dereference [-Werror=null-dereference]
   4132 |   return multi->xfers_alive;
        |          ~~~~~^~~~~~~~~~~~~
  ```
  Ref: https://github.com/curl/curl/actions/runs/26218822231/job/77148186045

- multi: also add `DEBUGASSERT(multi)` to the two updated functions.

Closes #21707

4 weeks agocmake: opt in `MSVC_VERSION` 1951 to picky warnings
Viktor Szakats [Thu, 21 May 2026 12:58:11 +0000 (14:58 +0200)] 
cmake: opt in `MSVC_VERSION` 1951 to picky warnings

Closes #21714

4 weeks agoGHA/windows: bump windows-2025 runners to windows-2025-vs2026
Viktor Szakats [Thu, 21 May 2026 12:49:31 +0000 (14:49 +0200)] 
GHA/windows: bump windows-2025 runners to windows-2025-vs2026

To silence:
```
NOTICE: windows-2025 requests are being redirected to windows-2025-vs2026 by June 15, 2026
```

Closes #21713

4 weeks agoGHA: simplify boolean `if` comparisons
Viktor Szakats [Thu, 21 May 2026 09:28:58 +0000 (11:28 +0200)] 
GHA: simplify boolean `if` comparisons

Closes #21709

4 weeks agotest1185: fix to previous commit
Viktor Szakats [Thu, 21 May 2026 09:24:15 +0000 (11:24 +0200)] 
test1185: fix to previous commit

Follow-up to 419b1c0b751b257bd54787618454d90fe88e7b79 #21703

4 weeks agochecksrc: detect `curlx_safefree()` opportunities
Viktor Szakats [Wed, 20 May 2026 22:51:04 +0000 (00:51 +0200)] 
checksrc: detect `curlx_safefree()` opportunities

Follow-up to bcd0497c8112e05412d2c649e8d9eea2bda8020e #21700
Follow-up to 1c3289c85e1a7a939464d5c5e84382d2e250e611 #21684
Follow-up to c0f0e400e0bc43cbe8c42c6937ed0ac743a8d81a #5968
Follow-up to 0f4a03cbb6fdb84d05cb6aafe50444edad4f4119

Closes #21703

4 weeks agodocs: note CURLOPT_PINNEDPUBLICKEY has no effect on legacy LDAP backend
Joshua Rogers [Tue, 19 May 2026 21:14:01 +0000 (23:14 +0200)] 
docs: note CURLOPT_PINNEDPUBLICKEY has no effect on legacy LDAP backend

Closes #21682

4 weeks agotidy-up: use `curlx_safefree()`
Viktor Szakats [Wed, 20 May 2026 18:43:59 +0000 (20:43 +0200)] 
tidy-up: use `curlx_safefree()`

Closes #21700

4 weeks agotests: add an assert to avoid IPC blocking
Dan Fandrich [Tue, 5 May 2026 03:48:52 +0000 (20:48 -0700)] 
tests: add an assert to avoid IPC blocking

If so much data is being sent over the internal IPC pipe that the pipe
buffer fills and the syscall blocks, the program will hang. Add an
assert to ensure that this limit is never reached. The buffer size is
going to be different on different platforms, so choose 1KB which is
likely to be a reasonable lower bound on just about any system.
Currently, the maximum amount ever written is <100 bytes, so this should
provide plenty of headroom.

Spotted by Codex Security

Closes #21688

4 weeks agoschannel: fix revoke_best_effort setting for proxy
Joshua Rogers [Tue, 19 May 2026 21:32:26 +0000 (23:32 +0200)] 
schannel: fix revoke_best_effort setting for proxy

- Fix revoke_best_effort reading wrong ssl config.

Prior to this change the revoke_best_effort setting for the proxy was
wrongly ignored in favor of the same setting for the destination host.

In other words, CURLSSLOPT_REVOKE_BEST_EFFORT set via
CURLOPT_PROXY_SSL_OPTIONS did not apply to the proxy and
CURLSSLOPT_REVOKE_BEST_EFFORT set via CURLOPT_SSL_OPTIONS wrongly
applied to the proxy.

Closes https://github.com/curl/curl/pull/21683

4 weeks agocfilters: fix busy loop on blocked transfers
Stefan Eissing [Tue, 19 May 2026 13:31:45 +0000 (15:31 +0200)] 
cfilters: fix busy loop on blocked transfers

When a transfer gets paused after the connection has been established,
any data sitting in the kernel socket buffers will no longer get read.

Prevent the sockets form being added to the pollsets, because they
will trigger POLLIN endlessly and cause a busy poll loop. Same in event
based processing.

Reported-by: Harry Sintonen
Fixes https://github.com/curl/curl/issues/21671
Closes https://github.com/curl/curl/pull/21675

4 weeks agosspi: clear SSPI credentials on AcquireCredentialsHandle failure
penpal [Fri, 15 May 2026 17:53:29 +0000 (23:38 +0545)] 
sspi: clear SSPI credentials on AcquireCredentialsHandle failure

- Clear credentials on AcquireCredentialsHandle failure so it is not
  used on a subsequent call.

SSPI initialization may evaluate the credentials pointer to determine
whether or not a prior call to AcquireCredentialsHandle was successful,
therefore we must clear it on a failed call.

Closes https://github.com/curl/curl/pull/21642

4 weeks agocmake: quote `COMPONENTS` string in `curl-config.in.cmake`
Viktor Szakats [Wed, 20 May 2026 16:43:41 +0000 (18:43 +0200)] 
cmake: quote `COMPONENTS` string in `curl-config.in.cmake`

For consistency and for handling an accidental empty more gracefully.

Follow-up to 7d546e52b21c94e1d4f6669d2d4d64f79bff0d7b #21540

Closes #21699

4 weeks agosetopt: clear proxy auth properly on NULL
Daniel Stenberg [Wed, 20 May 2026 11:39:25 +0000 (13:39 +0200)] 
setopt: clear proxy auth properly on NULL

Verify NULLed proxy credentials with test1648

Closes #21696

4 weeks agoDockerfile: fix typo in variable name
Viktor Szakats [Wed, 20 May 2026 10:01:40 +0000 (12:01 +0200)] 
Dockerfile: fix typo in variable name

Did not cause an actual issue.

Spotted by GitHub Code Quality

Follow-up to 41c03b4c98dbc639a32d32486ed5146be2e73ee1 #13250

Closes #21693

4 weeks agoGHA/non-native: drop DragonFlyBSD job, due to unreliable package repo updates
Viktor Szakats [Wed, 20 May 2026 01:42:49 +0000 (03:42 +0200)] 
GHA/non-native: drop DragonFlyBSD job, due to unreliable package repo updates

Example:
```
Wed, 20 May 2026 09:51:48 GMT Updating Avalon repository catalogue...
Wed, 20 May 2026 09:51:48 GMT pkg: An error occurred while fetching package: No error
Wed, 20 May 2026 09:51:48 GMT pkg: An error occurred while fetching package: No error
Wed, 20 May 2026 09:51:48 GMT repository Avalon has no meta file, using default settings
Wed, 20 May 2026 09:51:48 GMT pkg: An error occurred while fetching package: No error
Wed, 20 May 2026 09:51:48 GMT pkg: An error occurred while fetching package: No error
Wed, 20 May 2026 09:51:48 GMT pkg: An error occurred while fetching package: No error
Wed, 20 May 2026 09:51:48 GMT pkg: An error occurred while fetching package: No error
Wed, 20 May 2026 09:51:48 GMT Unable to update repository Avalon
Wed, 20 May 2026 09:51:48 GMT Error updating repositories!
Wed, 20 May 2026 09:51:48 GMT Error: Process completed with exit code 3.
```

As tested over at libssh2, retrying the install command also does not
help, only repeats the same failure.

Also: fix whitespace in matrix.

Follow-up to b158d1c9f7456a8f976c74c08d2dc5a555e9cc77 #21681

Closes #21694

4 weeks agourlapi: compare zone-id in Curl_url_same_origin()
Joshua Rogers [Tue, 19 May 2026 23:03:31 +0000 (01:03 +0200)] 
urlapi: compare zone-id in Curl_url_same_origin()

Closes #21686

4 weeks agowebsockets: auto-tunnel through http proxy
Stefan Eissing [Wed, 20 May 2026 08:30:25 +0000 (10:30 +0200)] 
websockets: auto-tunnel through http proxy

When using a ws: or wss: url with a http proxy, automatically
switch to tunneling operation mode.

Add test_20_10 to check.

Fixes #21663
Closes #21691

4 weeks agoGHA/non-native: move BSDs to a single matrix, add DragonFly and Midnight
Viktor Szakats [Tue, 19 May 2026 17:05:41 +0000 (19:05 +0200)] 
GHA/non-native: move BSDs to a single matrix, add DragonFly and Midnight

- bump cross-platform-actions to v1.1.0.
  Ref: https://github.com/cross-platform-actions/action/releases/tag/v1.1.0

- merge BSD jobs into a single matrix.

- split BSD jobs into build steps as used for other platforms.
  A new feature of cross-platform-actions v1.1.0.

- sync BSD build steps with other platforms.

- add DragonFlyBSD and MidnightBSD to the BSD matrix.
  New features of cross-platform-actions v1.1.0.
  MidnightBSD uses GnuTLS to add variation, also the preinstalled
  OpenSSL is too old (v1.1.1w) for curl.
  Stick with autotools for DragonFlyBSD; I could not figure out how
  to install cmake.
  Refs:
  https://en.wikipedia.org/wiki/DragonFly_BSD
  https://en.wikipedia.org/wiki/MidnightBSD

- bump Intel FreeBSD jobs from v14.3 to v15.0.

- fix to show `gcc` in the NetBSD job name.

All these saved 50 lines of YAML. The two new jobs take 2m15s each. The
bump to FreeBSD 15 needs and extra minute in total.

Note, the DragonFlyBSD job seems to have reliability issues. If it
remains an issue, I'll comment it out or delete it in a future commit.

Closes #21681

4 weeks agoDockerfile: update debian:bookworm-slim Docker digest to 0104b33
renovate[bot] [Wed, 20 May 2026 01:15:47 +0000 (01:15 +0000)] 
Dockerfile: update debian:bookworm-slim Docker digest to 0104b33

Closes #21687

4 weeks agorunner.pm: apply minor correctness fix
Viktor Szakats [Sat, 16 May 2026 16:58:47 +0000 (18:58 +0200)] 
runner.pm: apply minor correctness fix

"Lines 244-245 overwrite global variables `$runnerr` and `$runnerw` that
were already assigned in the child process (lines 205-206). In the
parent process context, these assignments appear incorrect and could
cause issues if `runner_init` is called multiple times. The parent
should only store references in the controller hashes."

It could never cause an actual issue, but clarifies the intent of the
code.

Spotted and fixed by GitHub Code Quality

Cherry-picked from #21646

Closes #21672

4 weeks agotidy-up: miscellaneous
Viktor Szakats [Sat, 16 May 2026 16:47:52 +0000 (18:47 +0200)] 
tidy-up: miscellaneous

- fix typos and wording in documentation and comments.
- KNOWN_BUGS: merge duplicate H1 section.
- test_10_proxy: delete stray expressions.
- Perl: `while()` -> `while(1)`.
- Perl: fix indent, whitespace, drop redundant quotes and parentheses.
- fix casing: URL, SSL, Windows.
- badwords: readd `threadsafe`, add `well-known` (and fix it).
- replace `WinXP` -> `Windows XP` to match other uses.

Closes #21646

4 weeks agounit3303, unit3304: tidy-ups
Viktor Szakats [Tue, 19 May 2026 22:17:58 +0000 (00:17 +0200)] 
unit3303, unit3304: tidy-ups

- use `curlx_safefree()`.
- drop redundant blocks.

Follow-up to 7541ae569d82fb308a5e2d94916027da4fa3ba3e #21667

Closes #21684

4 weeks agotls: fix incomplete mTLS config in conn reuse and session cache
Joshua Rogers [Tue, 19 May 2026 09:47:50 +0000 (11:47 +0200)] 
tls: fix incomplete mTLS config in conn reuse and session cache

cert_type, key, key_type, key_passwd and key_blob lived in
ssl_config_data but not in ssl_primary_config, so they were invisible to
match_ssl_primary_config() and to the TLS session cache peer key.

Two easy handles sharing a connection pool could reuse each other's
authenticated connections when they differed only on SSLKEY, SSLKEYTYPE,
KEYPASSWD, SSLCERTTYPE or SSLKEYBLOB. The second handle would silently
inherit the first handle's authenticated identity.

Promote all five fields into ssl_primary_config so the conn-reuse
predicate and session cache key cover the complete client credential
set. Also replace the fixed ":CCERT" session cache marker with the
actual clientcert path so sessions are not shared across different
client certificates.

Verified by test 3303 and 3304

Reported-By: Joshua Rogers (AISLE Research)
Closes #21667

4 weeks agoGHA/non-native: alpha-sort BSD jobs
Viktor Szakats [Tue, 19 May 2026 16:56:01 +0000 (18:56 +0200)] 
GHA/non-native: alpha-sort BSD jobs

Closes #21680

4 weeks agourl: detect proxy changes read from environment
Daniel Stenberg [Mon, 18 May 2026 21:47:11 +0000 (23:47 +0200)] 
url: detect proxy changes read from environment

When a proxy is set from an environment variable, detect if that proxy
is not the same as previously and flush state.

Verified by test1647: verify changing proxy with env variables and make
sure Digest state is flushed in the second use

Closes #21666

4 weeks agoBUFQ.md: re-sync with source code
Viktor Szakats [Tue, 19 May 2026 13:04:46 +0000 (15:04 +0200)] 
BUFQ.md: re-sync with source code

Also:
- move bullet points out from C code fences.
- fold long lines.

Follow-up to d4983ffc134addd20bea18987dec7c3b771e74a4 #17396

Closes #21678

4 weeks agotidy-up: rename more `CURLcode` variables to `result`
Viktor Szakats [Tue, 19 May 2026 12:49:28 +0000 (14:49 +0200)] 
tidy-up: rename more `CURLcode` variables to `result`

Follow-up to 885b553545a74365f4fc2541a0829f7745e80d37 #21348

Closes #21676

4 weeks agodocs: tidy-up scheme references
Viktor Szakats [Mon, 18 May 2026 16:38:25 +0000 (18:38 +0200)] 
docs: tidy-up scheme references

After this patch `://` schemes are lowercase and enclosed in backticks.

Also:
- docs/libcurl/libcurl-multi.md: drop a stray C code fence.
- docs/libcurl/libcurl-tutorial.md: replace single/double quotes with
  Markdown markup where applicable.

Ref: #21646

Closes #21674

4 weeks agounit1304: use enum type and values, rename `result` to `res`
Viktor Szakats [Tue, 19 May 2026 08:12:35 +0000 (10:12 +0200)] 
unit1304: use enum type and values, rename `result` to `res`

Ref: #21646

Closes #21673

4 weeks agomanagen: apply minor fixes and improvements
Viktor Szakats [Sat, 16 May 2026 16:20:37 +0000 (18:20 +0200)] 
managen: apply minor fixes and improvements

- replace `goto` with `while` loop.
- fix newlines in a warning message.
- handle open error for `curl/curlver.h` header.

Ref: #21646

Closes #21670

4 weeks agovtls: use Curl_safecmp for CRLfile and pinned_key comparison
Joshua Rogers [Tue, 19 May 2026 10:29:34 +0000 (12:29 +0200)] 
vtls: use Curl_safecmp for CRLfile and pinned_key comparison

Both are filesystem paths (or case-sensitive hash strings for
pinned_key). curl_strequal is case-insensitive and would treat
/etc/ssl/Crl.pem and /etc/ssl/crl.pem as the same file, unlike the other
path fields (CApath, CAfile, issuercert, clientcert) which already use
Curl_safecmp.

Closes #21668

4 weeks agodocs/libcurl: fix the version for curl_multi_socket_action
Dan Fandrich [Tue, 19 May 2026 05:30:57 +0000 (22:30 -0700)] 
docs/libcurl: fix the version for curl_multi_socket_action

It was added in 7.16.3, not 7.15.4 (that's when curl_multi_socket was
added).

4 weeks agocreds: mask OAuth bearer token in trace logs
Shintomon Mathew [Mon, 18 May 2026 14:02:45 +0000 (19:32 +0530)] 
creds: mask OAuth bearer token in trace logs

Masked OAuth bearer tokens in credential trace output by emitting ***
when a bearer token is present, matching the existing password redaction
behavior and preventing sensitive token disclosure in verbose/debug
logs.

Closes #21659

4 weeks agotest1588: use %TESTNUMBER, not hard-coded number
Daniel Stenberg [Mon, 18 May 2026 20:59:14 +0000 (22:59 +0200)] 
test1588: use %TESTNUMBER, not hard-coded number

Closes #21662

4 weeks agoVULN-DISCLOSURE-POLICY.md: test code is not secure
Daniel Stenberg [Mon, 18 May 2026 14:05:49 +0000 (16:05 +0200)] 
VULN-DISCLOSURE-POLICY.md: test code is not secure

Don't tell us about it

Closes #21660

4 weeks agocurl_easy_pause.md: rephrase the stream cache when pause clause
Daniel Stenberg [Mon, 18 May 2026 13:41:24 +0000 (15:41 +0200)] 
curl_easy_pause.md: rephrase the stream cache when pause clause

- mention HTTP/3
- it is 10 MB per stream these days

Closes #21658

4 weeks agotest1646: netrc parsing without user match but user in URL
Daniel Stenberg [Fri, 15 May 2026 22:54:13 +0000 (00:54 +0200)] 
test1646: netrc parsing without user match but user in URL

Follow-up to 4ae1d7cc2643e

4 weeks agonetrc: scanner refactor
Stefan Eissing [Fri, 15 May 2026 09:45:49 +0000 (11:45 +0200)] 
netrc: scanner refactor

Refactor the netrc scanner. Add test case for checking that the last
matched machine with unmatched login does not return the password as
success (unit1304).

Closes #21624

4 weeks agocurl_easy_setopt.md: change options when no transfer runs
Daniel Stenberg [Mon, 18 May 2026 12:15:28 +0000 (14:15 +0200)] 
curl_easy_setopt.md: change options when no transfer runs

Underscore this. Changing them mid-transfer may cause problems.

Fixes #21604
Reported-by: Joshua Rogers
Closes #21657

4 weeks agovtls_scache: include signature_algorithms in the SSL peer cache key
Mark Esler [Sat, 16 May 2026 22:07:15 +0000 (15:07 -0700)] 
vtls_scache: include signature_algorithms in the SSL peer cache key

Curl_ssl_peer_key_make() omitted ssl->signature_algorithms, although
match_ssl_primary_config() compares the field. Two handles differing
only in CURLOPT_SSL_SIGNATURE_ALGORITHMS therefore shared a peer key and
could resume each other's sessions across a shared CURLSH SSL session
cache.

Add :SIGALGS-%s next to the other ssl_primary_config fields.

Closes #21651

4 weeks agomk-unity.pl: `#include`, and not concatenate input headers
Viktor Szakats [Mon, 18 May 2026 09:50:37 +0000 (11:50 +0200)] 
mk-unity.pl: `#include`, and not concatenate input headers

When using `-D_CURL_TESTS_CONCAT=ON` with CMake, do not concatenate
`first.h` (or any future header) into the output C file, but `#include`
it instead. This is to play nice with compilers and analyzers which may
apply different checker rules on logic found in headers, vs. the input
source file. As seen for example with `-Wunused-macro` enabled in CI.
After this patch concatenated sources behave closer to regular C
sources.

Also:
- first.h: drop some `-Wunused-macro` silencers that became redundant
  with this patch.

Follow-up to 47f411c6d840dcee63a2ac9cbc0bfbea522ac5cd #21554
Follow-up to 39542f09935aba0b7130c20b6aae0be5cd6ff709 #20667

Closes #21656

4 weeks agoscripts/contri*: fix the Credits-to regex
Daniel Stenberg [Sun, 17 May 2026 21:55:48 +0000 (23:55 +0200)] 
scripts/contri*: fix the Credits-to regex

On my suggestion, the regex turned up wrong when looking for Credits-to
in git logs. This adjustment allows the leading spaces.

Follow-up to 64adc43a6ea07e4d807bbf9b5

Closes #21655

4 weeks agocmake: fix three issues generating lib options in config files
Viktor Szakats [Sun, 17 May 2026 12:04:49 +0000 (14:04 +0200)] 
cmake: fix three issues generating lib options in config files

- drop duplicate libs lists next to each other in `libcurl.pc`.
  Logic copied from libssh2.
  Fixing (seen in a local build):
  ```diff
  -Libs.private: -lssh2 -lz -lz -lldap -llber -lssl -lcrypto -lcrypto -lz -lbrotlidec -lbrotlicommon -lzstd -lnghttp2 -licucore -liconv -lpsl -lbacktrace
  +Libs.private: -lssh2 -lz -lldap -llber -lssl -lcrypto -lz -lbrotlidec -lbrotlicommon -lzstd -lnghttp2 -licucore -liconv -lpsl -lbacktrace
  ```
  Refs:
  https://github.com/libssh2/libssh2/commit/e1da7b2cb89063fc253bf94570c1ccfb3f1c2e81
  https://github.com/libssh2/libssh2/pull/1621
  https://github.com/libssh2/libssh2/commit/6464301820a9ca4a56c5f02717430bbd4150c7b2
  https://github.com/libssh2/libssh2/pull/1131

- handle `$<LINK_ONLY:NAMESPACE::NAME>` references.
  Fixing (seen in a local build using libssh2 v1.11.2-DEV):
  ```diff
  -Libs.private: -lssh2 -l$<LINK_ONLY:OpenSSL::Crypto> -lz -lldap -llber [...]
  +Libs.private: -lssh2 -lcrypto -lz -lldap -llber [...]
  ```

- fix `-l-pthread` sneaking into `libcurl.pc`.
  Fixing (seen with Android):
  ```diff
  -Libs.private:  -lz -l-pthread
  +Libs.private: -pthread -lz
  ```
  Refs:
  https://github.com/microsoft/vcpkg/blob/2b65c20fc66eda893aa15a15a453c3cf09500b19/ports/curl/dependencies.patch#L631-L634
  https://github.com/microsoft/vcpkg/commit/70b941a5d2443e79eeab62507acb41bd22201277#diff-7f2c3b2f93cd3478671a603cbd5ef818c7c403a11dc25e1d3539e9b03495a5d3
Upstream-patch-by: Kai Pastor
Closes #21654

4 weeks agoscripts: catch Credits-to contributors
Viktor Szakats [Sun, 17 May 2026 11:33:07 +0000 (13:33 +0200)] 
scripts: catch Credits-to contributors

Also:
- THANKS: add Credits-to attribution missed earlier.

Closes #21653

4 weeks agoconnect: remove deref of freed pointer in trace call
Daniel Stenberg [Sat, 16 May 2026 21:59:05 +0000 (23:59 +0200)] 
connect: remove deref of freed pointer in trace call

Spotted by CodeSonar

Closes #21649

4 weeks agourlapi: prevent a terminal `.0x` component to normalize IPv4
Daniel Stenberg [Sat, 16 May 2026 22:27:30 +0000 (00:27 +0200)] 
urlapi: prevent a terminal `.0x` component to normalize IPv4

Extend test 1560 to verify

Follow-up to 831a1514843bfa4d4d006627

Spotted by Codex Security
Closes #21652

4 weeks agobuild: omit zlib pkg-config reference for Android
Viktor Szakats [Sat, 16 May 2026 21:38:11 +0000 (23:38 +0200)] 
build: omit zlib pkg-config reference for Android

In both autotools and cmake builds, because Android does not offer
a `zlib.pc`.

Also:
- GHA/non-native: dump config files, to verify.

Reported-by: sfan5 on github
Fixes #21647

Closes #21648

4 weeks agourl: remove superfluous check
Daniel Stenberg [Sat, 16 May 2026 22:02:08 +0000 (00:02 +0200)] 
url: remove superfluous check

This pointer is already verified to be non-NULL some 15 lines above.

Pointed out by CodeSonar
Closes #21650

4 weeks agocf-h2-prox: fix peer leak
Stefan Eissing [Fri, 15 May 2026 11:37:36 +0000 (13:37 +0200)] 
cf-h2-prox: fix peer leak

The unlinking of the new Curl_peer was happening too later after
the struct had been set to zero. Move the unlink to happen before
that.

Fixes #21602
Reported-by: Joshua Rogers
Closes #21627

4 weeks agoRELEASE-NOTES: synced
Daniel Stenberg [Sat, 16 May 2026 21:09:52 +0000 (23:09 +0200)] 
RELEASE-NOTES: synced

4 weeks agounix-sockets: ignore proxy settings
Stefan Eissing [Fri, 15 May 2026 12:11:13 +0000 (14:11 +0200)] 
unix-sockets: ignore proxy settings

Fix a recent regression: when a unix-socket is configured, all proxy
settings must be ignored. The `via_peer` had been checked correctly,
but the connections proxy bits were not cleared.

Add test_11_04 to verify.

Reported-by: Fabian Keil (libcurl mailing list)
Closes #21630

5 weeks agolib678: fix to not be perma-skipped
Viktor Szakats [Sat, 16 May 2026 01:56:33 +0000 (03:56 +0200)] 
lib678: fix to not be perma-skipped

Prior to this patch the test was always skipped due to failing precheck
with `CURLE_BAD_FUNCTION_ARGUMENT`, because of the zero-length blob
passed to setopt. Fix by passing a non-zero long dummy blob as done in
`mk-lib1521.pl`.

Fixing:
```
test 0678 SKIPPED: CURLOPT_CAINFO_BLOB is not supported
```

Follow-up to 956e1ae84f2fec9f027b4ce80999744326b30992 #20705

Closes #21641

5 weeks agorunner.pm: set `CURL_TESTNUM` for `precheck` commands
Viktor Szakats [Sat, 16 May 2026 01:38:19 +0000 (03:38 +0200)] 
runner.pm: set `CURL_TESTNUM` for `precheck` commands

Closes #21640

5 weeks agoSSLCERTS: document 8.19.0 default Native CA builds (Windows)
Viktor Szakats [Fri, 15 May 2026 23:51:17 +0000 (01:51 +0200)] 
SSLCERTS: document 8.19.0 default Native CA builds (Windows)

Ref: https://curl.se/docs/sslcerts.html

Follow-up to 1730407b74f41cfd33f189348be54d0504b7c291 #18279
Reported-by: chrizilla on github
Fixes #21634

Closes #21639

5 weeks agosetopt: fix to honor `CURLOPT_PROXY_CAINFO_BLOB` over Native CA
Viktor Szakats [Fri, 15 May 2026 12:33:07 +0000 (14:33 +0200)] 
setopt: fix to honor `CURLOPT_PROXY_CAINFO_BLOB` over Native CA

In AppleSecTrust or NativeCA-enabled builds, make sure override it when
setting a custom `CURLOPT_PROXY_CAINFO_BLOB`.

Reported-by: Joshua Rogers (Aisle Research)
Follow-up to 1730407b74f41cfd33f189348be54d0504b7c291 #18279
Follow-up to eefd03c572996e5de4dec4fe295ad6f103e0eefc #18703

Closes #21631

5 weeks agoschannel: enforce Extended Key Usage for custom CA roots
Viktor Szakats [Fri, 15 May 2026 11:26:05 +0000 (13:26 +0200)] 
schannel: enforce Extended Key Usage for custom CA roots

Reported-by: Joshua Rogers (Aisle Research)
Closes #21629

5 weeks agodocs: end "...can be used several times..." sentences with period
Tim Martin [Sat, 16 May 2026 09:12:06 +0000 (04:12 -0500)] 
docs: end "...can be used several times..." sentences with period

Closes #21644

5 weeks agourlapi: handle redirect without set scheme with default-scheme
Daniel Stenberg [Fri, 15 May 2026 12:37:58 +0000 (14:37 +0200)] 
urlapi: handle redirect without set scheme with default-scheme

Verify in test 1921

Reported-by: mulan_dh on hackerone
Closes #21632

5 weeks agocookie: trim trailing dots when checking PSL
Daniel Stenberg [Fri, 15 May 2026 22:19:09 +0000 (00:19 +0200)] 
cookie: trim trailing dots when checking PSL

Verified with test 1629

Closes #21636

5 weeks agotidy-up: miscellaneous
Viktor Szakats [Thu, 14 May 2026 18:37:42 +0000 (20:37 +0200)] 
tidy-up: miscellaneous

- avoid "will" in builds scripts, scripts folder, curl_easy_ssls_export.md,
  and few other files.
- badwords: add "initialise", "nul terminated", "thread safety" and
  variations.
- prefer "null-terminat", where missing (two places).
- fix "null-terminat*" missing dash.
- hostip: merge two `#if` blocks.
- tool_doswin: fix comment
  Spotted by GitHub Code Quality
  Follow-up to 9a2663322c330ff11275abafd612e9c99407a94a #17572
- fix stray spaces and newlines.

Closes #21638

5 weeks agoGHA: enable `-Wunused-macros` in clang-tidy jobs
Viktor Szakats [Tue, 12 May 2026 00:26:05 +0000 (02:26 +0200)] 
GHA: enable `-Wunused-macros` in clang-tidy jobs

Also fix fallouts found.

Windows clang-tidy CI job is a little pickier than I'd prefer due to the
`_CURL_TESTS_CONCAT=ON` option used there, and all macros considered
local, thus checked by the compiler. Upside: it revealed macro usage
dynamics in tests. If too annoying, `first.h` may be opted-out from the
concat logic. Some macros may also be deleted instead of `#if 0`-ing.

Follow-up to e0e56e9ae434552bd6ac5570ed91483188d75788 #21550
Follow-up to 5fa5cb382560316a55f0954f1e8cebdbd6568cfb #20593

Closes #21554

5 weeks agourl: compare full origin when setting credentials
jeffhuang [Tue, 12 May 2026 16:13:15 +0000 (16:13 +0000)] 
url: compare full origin when setting credentials

Closes #21575

5 weeks agourl: url_match_destination fix
Stefan Eissing [Tue, 12 May 2026 15:58:03 +0000 (17:58 +0200)] 
url: url_match_destination fix

Match origin/via_peer also for non-SSL schemes.

Closes #21573

5 weeks agox509asn1: fix DH public key parameter extraction
Sergio Correia [Wed, 13 May 2026 18:44:05 +0000 (19:44 +0100)] 
x509asn1: fix DH public key parameter extraction

The dh(g) parameter was read from param->beg instead of from the
cursor p returned by parsing dh(p). This caused dh(g) to always
report the same value as dh(p) when inspecting DH certificates
via CURLOPT_CERTINFO on non-OpenSSL backends.

The DSA branch correctly advances the cursor; the DH branch lost
this during what appears to be a copy-paste.

Add unit1676 to verify that dh(p) and dh(g) report distinct values
using a hand-crafted minimal DER certificate.

Assisted by: Claude Opus 4.6
Signed-off-by: Sergio Correia <scorreia@redhat.com>
Closes #21595

5 weeks agosetopt: more careful cleanup of the HSTS cache
Daniel Stenberg [Thu, 14 May 2026 21:23:08 +0000 (23:23 +0200)] 
setopt: more careful cleanup of the HSTS cache

Reported-by: Joshua Rogers
Closes #21615

5 weeks agoKNOWN_BUGS.md: remove fixed x509asn.1 bug
Emanuel Krollmann [Thu, 14 May 2026 15:27:13 +0000 (17:27 +0200)] 
KNOWN_BUGS.md: remove fixed x509asn.1 bug

KNOWN_BUGS.md contains an entry about a CURLE_OUT_OF_MEMORY
error on a CURLOPT_CERTINFO call when using Schannel.

This bug was fixed by 137a668e8cb42dda1673bf2c79cbb24c8fe0b405.

remove the entry from KNOWN_BUGS.md.

Ref: https://github.com/curl/curl/issues/8741#issuecomment-4445486705

Closes #21611

5 weeks agocf-h2-proxy: drop interim responses
Stefan Eissing [Fri, 15 May 2026 11:03:02 +0000 (13:03 +0200)] 
cf-h2-proxy: drop interim responses

Any 1xx response before the CONNECT final one can be dropped as no one
uses those in the HTTP/2 proxy filter. This eliminates a potential
memory exhaustion by the famous malicious server on the internet.

Closes #21626

5 weeks agoCURLOPT_SSH_HOSTKEYFUNCTION.md: for new connections only
Daniel Stenberg [Fri, 15 May 2026 11:54:41 +0000 (13:54 +0200)] 
CURLOPT_SSH_HOSTKEYFUNCTION.md: for new connections only

curl can then reuse the connection for subsequent transfers without
calling this function again.

Fixes #21606
Reported-by: Joshua Rogers
Closes #21628

5 weeks agosrc: add `curlx_memzero()` to clear buffers securely
Viktor Szakats [Wed, 13 May 2026 16:20:33 +0000 (18:20 +0200)] 
src: add `curlx_memzero()` to clear buffers securely

To safely zero memory, introduce `curlx_memzero()`, and map it to
`memset_s()` (C11) or `memset_explicit()` (C23) if auto-detected, or
`explicit_bzero()` or `explicit_memset()` for platforms opted-in, or
fall back to a local workaround if all unavailable. On Windows, always
use `SecureZeroMemory()`, or `SecureZeroMemory2()` with Visual Studio
and Windows SDK 10.0.26100.0+.

Details above are experimental and may change if they cause issues.

Also add macros/functions that zero memory before freeing a buffer:
- `curlx_safefreezero()`: for buffers with size.
- `curlx_safefreezeroz()`: for null-terminated buffers.
- `curlx_freezero()`: for buffers with size.
- `curlx_freezeroz()`: for null-terminated buffers.

`curlx_memzero()` must not be passed a NULL pointer because in some
implementations it is undefined behavior.

Also:
- curl_sha512_256: Replace hard-wired `explicit_memset()` call with new
  `curlx_memzero()`.

Refs:
https://en.cppreference.com/c/string/byte/memset
https://man7.org/linux/man-pages/man3/explicit_bzero.3.html
https://man.freebsd.org/cgi/man.cgi?query=explicit_bzero
https://man.netbsd.org/NetBSD-7.2/explicit_memset.3
https://learn.microsoft.com/previous-versions/windows/desktop/legacy/aa366877(v=vs.85)
https://learn.microsoft.com/windows/win32/memory/winbase-securezeromemory2
https://learn.microsoft.com/cpp/overview/compiler-versions
https://learn.microsoft.com/windows/apps/windows-sdk/downloads
https://jtsoya539.github.io/windows-sdk-versions/

Credits-to: Daniel Gustafsson
Credits-to: Will Cosgrove and co-authors in libssh2
Ref: #13589 (original attempt)
Ref: #21588

Closes #21598

5 weeks agourlapi: consume trailing dots after IPv4 numerical addresses
Daniel Stenberg [Fri, 15 May 2026 15:04:26 +0000 (17:04 +0200)] 
urlapi: consume trailing dots after IPv4 numerical addresses

If the hostname is specified as an IPv4 numerical address and it is
followed by a single dot, acccept that as a valid IPv4 and remove the
dot when normalizing.

This prevents otherwise legitimate IPv4 hostnames to have trailing dots.
Seems to match what browsers do.

Extended test 1560 to verify.

Closes #21635

5 weeks agoCURLOPT_SHARE: warn about early remove
Stefan Eissing [Fri, 15 May 2026 12:56:24 +0000 (14:56 +0200)] 
CURLOPT_SHARE: warn about early remove

Add a warning to removing a SHARE from an EASY handle before it is
finished.

Closes #21633

5 weeks agocookie: compare path case sensitively
Daniel Stenberg [Thu, 14 May 2026 21:46:45 +0000 (23:46 +0200)] 
cookie: compare path case sensitively

Verify with test 1645

Reported-by: Joshua Rogers
Closes #21616

5 weeks agostrparse: make curlx_str_until() accept zero for 'max'
Daniel Stenberg [Wed, 13 May 2026 22:06:03 +0000 (00:06 +0200)] 
strparse: make curlx_str_until() accept zero for 'max'

When asked to parse for a string with max zero bytes, it will always
return error and no longer trigger an assert. This saves the caller from
having to check for this condition.

Closes #21600

5 weeks agorustls: drop two wrong leftover casts to `ssize_t`
Viktor Szakats [Fri, 15 May 2026 09:54:14 +0000 (11:54 +0200)] 
rustls: drop two wrong leftover casts to `ssize_t`

While both source and target types are already `size_t`.

Spotted by GitHub Code Quality

Follow-up to b7c676d13f0988bde9bb0e4c3cfc688072cdb2e0 #17593

Closes #21625