- H3 proxy: re-sync code with original source `curl_ngtcp2.c` to reduce
differences, and to apply missed minor fixes. Also apply clang-format.
Drop redundant `#undef`s, casts, `#endif` comments, includes, drop
intermediate variables, sync include and macro order.
Follow-up to e78b1b3ecc#21153
- INSTALL-CMAKE.md: move `CURL_ENABLE_SMB` to the enable section.
- tests/http/env: rename `tcpdmp` to `tcpdump` to match object variable.
- mbedtls: drop incorrect `mbedTLS 4+` comments.
(features are also supported by 3+, meaning it's always supported.)
- lib1648: rename a variable to match purpose.
- CIPHERS.md: alpha-sort link list.
- replace rare `X''` hex markup with `0x`.
- `IP v4/6` -> `IPv4/6`.
- 'version X.Y' -> 'vX.Y', where sensible.
- 'VX.Y' -> 'vX.Y', where sensible.
- fix indents, casing, newlines, typos.
Closes#21772
The message triggered earlier than intended and did not take the
transfer/content type into account.
Ref #21603
Reported-by: Joshua Rogers
Closes#21756
This patch adds two major proxy capabilities to curl (ngtcp2 QUIC):
- HTTP/3 Proxy CONNECT: Tunnel HTTP/1.1 or HTTP/2 traffic through an
HTTPS proxy that speaks HTTP/3 (QUIC) using the standard CONNECT
method over an HTTP/3 connection.
- MASQUE CONNECT-UDP: Tunnel HTTP/3 (QUIC) traffic through an HTTP
proxy (speaking HTTP/1.1, HTTP/2, or HTTP/3) using the extended
CONNECT method with the CONNECT-UDP protocol (RFC9297 & RFC9298).
Public API additions:
- `CURLPROXY_HTTPS3`: new proxy type constant for HTTP/3 proxy
- `--proxy-http3`: new CLI flag to negotiate HTTP/3 with HTTPS proxy
The implementation adds two new filters:
- `H3-PROXY` - enables negotiating HTTP/3 (QUIC) to the proxy and
running CONNECT/CONNECT-UDP through that proxy transport.
- `CAPSULE` - dedicated filter inserted between QUIC transport and
HTTP-PROXY to handle datagram capsule encapsulation/decapsulation.
Here is how the curl filter chaining looks in different scenarios:
- HTTP/3 Proxy CONNECT (tunneling TCP protocols over QUIC proxy):
conn -> HTTP/1.1 or HTTP/2 -> SSL -> HTTP-PROXY ->
H3-PROXY -> HAPPY-EYEBALLS -> UDP
- MASQUE CONNECT-UDP (tunneling QUIC over any proxy):
conn -> HTTP/3 -> CAPSULE -> HTTP-PROXY -> H3-PROXY ->
HAPPY-EYEBALLS -> UDP
conn -> HTTP/3 -> CAPSULE -> HTTP-PROXY -> H1-PROXY or H2-PROXY ->
SSL -> HAPPY-EYEBALLS -> TCP
- Both features currently require the ngtcp2 QUIC backend.
- Both features are experimental (disabled by default). Enable with
`--enable-proxy-http3`(autotools) or `-DUSE_PROXY_HTTP3=ON`(CMake).
Tests:
- tests/unit/unit3400.c: Unit tests for capsule protocol encode/decode
- tests/http/test_60_h3_proxy.py: Comprehensive pytest integration suite
- tests/http/testenv/h2o.py: Managing h2o instances with HTTP/1.1, HTTP/2,
and HTTP/3 (QUIC) listeners, proxy.connect and proxy.connect-udp enabled.
References:
RFC 9297 - HTTP Datagrams and the Capsule Protocol
RFC 9298 - Proxying UDP in HTTP
RFC 9000 §16 — Variable-Length Integer Encoding
Signed-off-by: Aritra Basu <aritrbas+gh@cisco.com>
Closes#21153
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
- 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
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
"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 #21646Closes#21672
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
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
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
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 47f411c6d8#21554
Follow-up to 39542f0993#20667Closes#21656
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
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 956e1ae84f#20705Closes#21641
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 e0e56e9ae4#21550
Follow-up to 5fa5cb3825#20593Closes#21554
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
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
Or consisting of just a single dot.
Such names cannot be resolved with DNS.
While they *can* still be resolved with /etc/hosts or --resolve tricks,
they easily cause internal problems because their trailing dots.
Let's not allow them anymore.
Closes#21622
The SASL service name, used in authentication, is part of curl's credentials
when authenticating to a server/proxy. Make it part of `struct Curl_creds`.
Change code to use `creds` to obtain a service name. By tying creds used
to the connection, connection reuse is also only allowed when the service
name matches.
Closes#21585
Use parts of text from the upload filename field when that uses globbing
by giving it a name the same way we do it for URL globs. For example, if
you upload three files to a HTTP URL and want to save the corresponding
responses in separate files:
curl -T 'file{<num>1,2,3}' https://upload.example/ -o 'response-#<num>'
Verified by test 2014
Closes#21407
This now points to where the duplicate name ends, not where it starts.
Also fixes test 2410 to use a fixed hostname so that the error position
remains the same.
Reported-by: Viktor Szakats
Fixes#21567Closes#21568
Authorizdation credentials are kept in `struct Curl_creds`. This contains:
* `user`: the username, maybe the empty string
* `passwd`: the password, maybe the empty string
* `sasl_authzid`: the SASL authz value, maybe the empty string
* `oauth_bearer`: the OAUTH bearer token, maybe the empty string
* `source`: where the credentials from from
* `refcount`: a reference counter to link/unkink creds
A `creds` with all values empty is equivalent to NULL, e.g. no `creds`
instance. With reference counting, `creds` can be linked/unlinked
in several places.
See docs/internals/CREDENTIALS.md for use.
Closes#21548
