Use separate dns cache entries for addresses (A+AAAA) and HTTPS-RR
results. That makes also "negative" results independent of each other.
Dns cache entries, once in use, can no longer be modified safely, as
concurrent use would require each access then to be done under lock. By
using separate entries, we can update a HTTPS entry without needing to
duplicate an existing address entry for the same host+port.
Connection filters can now ask for DNS resolves. This works at any time
during connection setup and while connect is going on.
`Curl_conn_dns_add_addr_resolve()` and
`Curl_conn_dns_add_https_resolve()` are used for that. They check if the
resolve is already ongoing, can be added to a resolve not started yet
or, as last resort, create a new filter instance and add it to the
connection (it's easier to add more filters than making the same filter
handle multiple resolves. Since DNS filters are removed once the
connection is established, there is no later penalty).
HTTPS-RR queries are added by the `HTTPS-CONNECT`, `SSL` and `QUIC`
filters. The latter will only do that when ECH is configured and
supported. That means we trigger HTTPS-RR queries only when the results
matter.
Add test_22_06 for ALPN influenced via HTTPS-RR when tunneling through a
proxy. This did not work before.
Adjust test2100 to use https: as the previous http: URL no longer
triggers HTTPS-RR resolves.
Closes#22216
A flaky CI failure of `test_17_05_bad_ip_addr[h3]` (GnuTLS, event-based)
had curl detect the certificate name mismatch yet exit with
`CURLE_RECV_ERROR` (56) instead of `CURLE_PEER_FAILED_VERIFICATION`
(60). `Curl_cf_ngtcp2_cmn_connect` calls `Curl_cf_ngtcp2_cmn_set_expiry`
after its `ctx->tls_vrfy_result` override and returns the error
unfiltered, so when the server's final handshake flight happens to be
processed by the ingress inside set_expiry, the verify failure surfaces
as a generic receive error.
This PR makes set_expiry prefer `ctx->tls_vrfy_result` over generic
progress errors, as the recv and send paths already do after calling it,
and also covers the `cf-ngtcp2-proxy.c` call sites that lack the
override. Completes #21712. Seen in
https://github.com/curl/curl/actions/runs/29243256619/job/86794119412.
Closes#22317
quiche 0.29.3 started enforcing a default
`SETTINGS_MAX_FIELD_SECTION_SIZE` of 32 KiB, so larger response headers
now shut down the whole connection with `CURLE_HTTP3`. curl accepts up
to 300 KiB of response headers with every other backend and HTTP
version. Tell quiche to allow what curl itself allows. This keeps
test_01_11 passing with quiche 0.29.3 and it also advertises our real
limit to servers, which the RFC encourages.
Ref: 899e2edec9#22333
Ref: #22329
Ref: #22325
Ref: 9be0e4fa18
Ref: https://github.com/cloudflare/quiche/releases/tag/0.29.3Closes#22331
Remove the timestamp passed for ratelimit checks. Let the limit
calculation obtain a timestamp when it needs it. Most transfers run
without active ratelimits and getting a fresh timestamp is unnecessary.
Closes#22292
FTP servers using SSL can be configured to check TLS session reuse on
the DATA connection. They hand out a new session on every CONTROL
connect and require to see the client using exactly that one when
up-/downloading on DATA.
This means:
1. We have to configure the SSL filter on the DATA connection with
exactly the same peers.
2. We have to remember the SSL session on the CONTROL connection -
separately from the session cache. The SSL filter on the DATA
connection then looks for a session at the CONTROL filter and, if
present, uses that.
Tests:
Enable `require_ssl_reuse` in our pytest setup for vsftpd. This
reproduces the problem reported in #22225 and verifies the fix.
Skip ftp+SSL pytests for rustls, as we have no possibility to reuse
sessions in that backend.
Schannel: we do not run these tests with the backend. I expect it has
similar problems but am not able to verify.
Reported-by: Laurent Sabourin
Fixes#22225Closes#22246
When TLS is verified via Apple SecTrust, openssl internal state is a
verification faiure (that is how it works, we use sectrust when openssl
fails to verify and natice ca store is enabled).
OpenSSL stores this verification status inside its TLS session objects.
On reuse, we see an unverified session and do not reuse it for a
verified connect attempt. While this is a performance penalty for most
connections, it *fails* on ftps:// transfers where servers expect
session reuse on DATA connections.
Fix this by remembering that TLS and a session was verified by sectrust
and allow reuse of such sessions for new connect attempts that also use
sectrust.
Closes#22235
The DNS filter knows the peer it resolves and the code parts that want
the results know the peer as well. Pass it to lookup methods to make
sure results match.
Background: when tunneling, the resolved peer is not always the one that
other filters are looking for. Especially when HTTPS-RR results are
accessed in TLS filters, those will differ.
This prevents a HTTPS-RR for a proxy to be used for the origin when ECH
is activated. To make ECH work through a tunnel, we need to start an
additional resolve. Something to be fixed after 8.21.
Closes#22042
Now that `conn->origin` can be the proxy, we need to change how
Curl_xfer_is_secure() and some other places work.
Pointed out by Codex Security
Closes#22015
- `N byte` -> `N-byte` or `N bytes`.
- INTERNALS.md: language tweaks.
- schannel: language tweak in comment/error message.
- socks_gssapi, socks_sspi: simplify composing an error message.
(at a cost of 8 extra constant string bytes.)
- m4/curl-compilers.m4: fix typo in link (in comment).
- contrithanks.sh: fix indent, drop stray `;` terminator.
- lib, src, tests: drop/fix a bunch of badwords.
- fix typos in comments.
- fix indent, stray spaces.
Some of these spotted by GitHub Code Quality and Copilot
Closes#22009
Add `data->state.origin` as the origin the transfer is sending the
current request to/gets the response from. Use it for request specific
properties like authentication, hsts and cookie handling, etc.
Unless talking to a forwarding HTTP proxy (e.g. not tunneling),
`data->state.origin` and `conn->origin` are the same.
With a forwarding HTTP proxy in play, `conn->origin` is set to
`conn->http_proxy.peer` and `conn->bits.origin_is_proxy` (a new bit) is
set.
Remove the connection bits, now replaced with:
* `conn->bits.socksproxy` -> `conn->socks_proy.peer`
* `conn->bits.httpproxy` -> `conn->http_proy.peer`
* `conn->bits.proxy` -> `(conn->socks_proy.peer || conn->http_proy.peer`)
* `conn->bits.tunnel_proxy` -> (`conn->http_proy.peer && !conn->bits.origin_is_proxy`)
* `(conn->bits.httpproxy && !conn->bits.tunnel_proxy)` -> `conn->bits.origin_is_proxy`
Rename `noproxy.[ch]` to `proxy.[ch]`. Move the connection proxy setup
code from `url.c` to `proxy.c`.
Remove `data->info.conn_remote_port` as no one uses it.
Add test_40_02b for a SOCKS connection to a forwarding HTTPS proxy.
Update internal documentation about peers and creds.
Closes#21967
To contain the logic of checking for both `EWOULDBLOCK` and/or `EAGAIN`
depending on platform/availability. Also to avoid checking for both if
they mapp to the same value, and to avoid PP guards around use.
This also ensures `EAGAIN` is consistently not checked on Windows, where
headers defined it, but `SOCKERRNO` never returns it, because curl maps
it to `WSAGetLastError()`.
If they map to the same value, checking them both in an `if` expression
trips GCC warning `-Wlogical-op` (the same way it triggers duplicate
case value error in `switch`).
Also:
- replace two `switch()` statements with the new macro.
- tests/server/sws: make two outliers use the new macro that were only
checking for `EWOULDBLOCK` before this patch, in `connect_to()`.
- move variables to the left-side of expressions, where missing.
- rustls: use a variant of this macro that uses raw `EWOULDBLOCK`.
Tried tracing it back to the origins, but I couldn't figure out if
this is working as expected on all supported Windows versions in
Rust. It seems to be using `GetLastError()`, according to
https://docs.rs/system_error/0.2.0/system_error/, which would be
probably incorrect.
Notes:
- it's probably a good idea to assign `SOCKERRNO` to a variable before
passing it to this macro.
Cherry-picked from #21893Closes#21992
Do not set the easy handle opening a proxy tunnel as userdata on the
stream. The ease handle might go out of scope long before the tunnel
stream is closed.
Closes#21962
Move as sockaddr2string() into cf-socket.c where its only callers are.
Mark as UNITTEST for unit1609.
Move "struct Curl_sockaddr_ex" into sockaddr.h, so connect.h and
cf-socket.h can be included without all the system headers needed.
Closes#21946
Adjust code to avoid `-Wformat-signedness` warnings, while making sure
that enums are always cast to a known type when passing them to `printf`
functions, to support compilers and compiler settings where enums are
not default-size signed ints.
- cast integers printed as hex to `unsigned`. (63 times, 20 of them in
`mbedtls.c`)
- cast misc enums to `int` for printing. (31 times)
- cast `CURL_LOCK_DATA_*` enums to `int`. (4 times)
- cast `CURL_FORMADD_*` enums to `int`. (13 times)
- cast `CURLSHE_*` enums to `int`. (3 times)
- cast `CURLUE_*` enums to `int`. (33 times)
- cast `CURLMSG_*` enums to `int`. (6 times)
- cast `CURLE_*` enums to `int`. (~380 times)
- unit1675: fix mask.
Follow-up to 7c34365cce#21879
Ref: #18343 (initial attempt)
Closes#20848
Share common functions/structs between ngtcp2 HTTP/3 and the proxy
version.
Fix bugs in proxy implementation when it comes to stream and pollset
handling and transfer lifetimes.
Curl_multi_xfer_sockbuf_borrow: work without multi
When a connection gets shutdown by a share, the easy handle used is
share->admin and it does not have a multi handle. In that case let
Curl_multi_xfer_sockbuf_borrow() allocate a buffer to be freed on
release.
This happens when a TLS filter sends its last notify through a HTTP/3
proxy tunnel.
Closes#21871
With a flood of zero lenght UDP packets to curl, the receive loop might
run longer than intended to. Count such packets against the max to
terminate the loop as intended.
URL: https://hackerone.com/reports/3783438
Reported-by: vectorqueue on hackerone
Closes#21869
closing a filter chain and reconnecting it again is a complication
that only the HTTP/1.x proxy filter used. Remove it from all filters.
Instead, a filter can return CURLE_AGAIN during the connect phase
and the cf-setup filter will tear down all "sub filters" and restart
over.
With this, a filter never resets to the initial phase but progresses
through connect -> connected -> shutdown -> destroy once.
Closes#21831
- drop more uses of the word "just". (not enforced here)
- drop some uses of the "will" word.
- "then" -> "than".
- tests/http/testenv/curl.py: fix copy-paste typo in error message.
- pytest: replace `shutdownh` with `shutdown` in test names.
Spotted by GitHub Code Quality.
- comment typos.
- whitespace and newlines fixes.
Closes#21830
code:
- less exception handling in existing code
- true ip happy eyeballing
- enable certificate verification
- cf-h2-proxy: abort connection when server closed connection
tests:
- remove all --insecure and --proxy-insecure args
- make session reuse test_60_12 a working one
- resolve port conflicts between h2o and nghttpx
- use proxy args better
- make test_60_06 run shorter
- kill h2o at the end of tests, normal stop takes too long
Ref: 59213f8248#21789
Follow-up to e78b1b3ecc#21153Closes#21798
- 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
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
`struct Curl_peer` keeps information about a communication endpoint
together. It will replace `conn->host` and `conn->conn_to_host` and
proxyinfo host. It will also become part of `struct ssl_peer`.
It has a reference counter, so an instance can be shared between
connections and filters.
Elminiates `conn->host` and `conn->connect_to_host`, used in the
proxyinfo structures. Passed to DNS resolution and socks filters, etc.
Pass peer to http proxy and socks tunnel filters. Use peer in dns filter
and resolving. Make `Curl_peer` a member in the `struct ssl_peer`.
Add `docs/internals/PEERS.md` for documentation.
Closes#21472
When HTTPS-RR is needed for the HTTP/3 handshake, delay the connect
until it arrives. Relevant only for TLS backends that support ECH, for
now.
Closes#21253
Make cf-https-connect work async correctly:
- only start first baller when at least one A/AAAA address
is available
- select first connect attempt after that with HTTPS-RR info
there or not.
- select second connect attempt only when HTTPS-RR is resolved
(may have resolved to "not known") and select possible ALPN
from things known by then. May not select any second attempt
when first already covers everything.
This means when the HTTPS-RR is known at/before the first address
is resolved, everything behaves as before. When the HTTPS-RR is
late, a first connection attempt will have been started. Any
ALPN preference from the HTTPS-RR that is not already ongoing will
then start the second attempt.
For HTTPS-RRs that recommend 2 or more ALPNs, the first will always
be attempted: either it is already ongong or it will be the ALPN
for the second attempt. The 2nd ALPN recommendation from HTTPS-RR
*may* be honored or not, depending on what is already selected.
The difference in behaviour between early/late HTTPS-RR resolve
cannot be helped - unless we do not perform any attempts before
it arrives. Trade offs.
Closes#21267
New connection filter `cf-dns` that manages DNS queries. If hands
out addresses and HTTPS-RR records to anyone interested. Used by
HTTPS and IP happy eyeballing.
Information may become available *before* the libcurl "dns entry"
is complete, e.g. all queries have been answered. The cf-ip-happy
filter uses this information to start connection attempts as soon
as the first address is available.
The multi MSTATE_RESOLVING was removed. A new connection always
goes to MSTATE_CONNECTING. The connectdata bit `dns_resolved`
indicates when DNS information is complete. This is used for
error reporting and starting the progress meter.
Removed dns entries `data->state.dns[i]`, as the `cf-dns` filter
now keeps the reference now.
Many minor tweaks for making this work and pass address information
around safely.
Closes#21027
Curl_1st_err() does not return the second error if the first result is
CURLE_AGAIN. This may cause errors to not become noticeable when they
should be.
Replace all use of Curl_1st_err() with Curl_1st_fatal(), which handles
CURLE_AGAIN as a not-a-real-error case.
Closes#20980
