In curl 8.12 I tried to improve the logic on how we handle connections
that "upgrade" to TLS later, e.g. with a STARTTLS. I found the existing
code hard to read in this regard. But of course, the "improvements" blew
up in my face.
We fixed issues with imap, opo3, smtp in 8.12.1, but ftp was no longer
reusing existing, upgraded control connections as before. This PR adds
checks in our pytest FTP tests that verify reuse is happening as
intended.
I rewrote the logic in url.c again, so that the new test checks now pass.
Reported-by: Zenju on github
Fixes#16384Closes#16392
Adds a "cw-pause" client writer in the PROTOCOL phase that buffers
output when the client paused the transfer. This prevents content
decoding from blowing the buffer in the "cw-out" writer.
Added test_02_35 that downloads 2 100MB gzip bombs in parallel and
pauses after 1MB of decoded 0's.
This is a solution to issue #16280, with some limitations:
- cw-out still needs buffering of its own, since it can be paused
"in the middle" of a write that started with some KB of gzipped
zeros and exploded into several MB of calls to cw-out.
- cw-pause will then start buffering on its own *after* the write
that caused the pause. cw-pause has no buffer limits, but the
data it buffers is still content-encoded.
Protocols like http/1.1 stop receiving, h2/h3 have window sizes,
so the cw-pause buffer should not grow out of control, at least
for these protocols.
- the current limit on cw-out's buffer is ~75MB (for whatever
historical reason). A potential content-encoding that blows 16KB
(the common h2 chunk size) into > 75MB would still blow the buffer,
making the transfer fail. A gzip of 0's makes 16KB into ~16MB, so
that still works.
A better solution would be to allow CURLE_AGAIN handling in the client
writer chain and make all content encoders handle that. This would stop
explosion of encoding on a pause right away. But this is a large change
of the deocoder operations.
Reported-by: lf- on github
Fixes#16280Closes#16296
Since the script 'apachectl' from the httpd project is severly mutilated
on several distros, use the executable httpd/apache2 directly in pytest
runs.
Remove detection of apachectl form autoconf and cmake.
Closes#16000
When a QUIC TLS session announced early data support and
'CURLSSLOPT_EARLYDATA' is set for the transfer, send initial request and
body (up to the 128k we buffer) as 0RTT when curl is built with
ngtcp2+gnutls.
QUIC 0RTT needs not only the TLS session but the QUIC transport
paramters as well. Store those and the earlydata max value together with
the session in the cache.
Add test case for h3 use of this. Enable quic early data in nghttpx for
testing.
Closes#15667
Fix regression that no longer printed the error messages about expired
certificates in openssl. Add test case for openssl/gnutls/wolfssl.
Fixes#15612
Reported-by: hiimmat on github
Closes#15613
Changes to make a curl built with OpenSSL + GnuTLS to run successfully
in our pytests. Run
CURL_SSL_BACKEND=openssl pytest
to test a TLS backend other than the default.
Closes#15443
Add two test cases that connection using a hostname the server has no
certificate for. First, verify that the peer verification fail, as
expected. Second, provide '--insecure' to test that the connection
succeeded and returned some data.
Closes#15429
New args for scorecard.py:
* --remote ip:port to run tests against a remote server
* --start-only: to only start the servers, listing their ports
Start the server using "scorecard.py --start-only" on one machine and
then run the tests with "scorecard.py --remote ip:port" against that
machine.
Closes#15415
When a server signals EOS from its side and the curl upload is
unfinished and the server has not given a positive HTTP status response,
auto RST the stream to signal that the upload is incomplete and that the
whole transfer can be stopped.
Fixes the case where the server responds with 413 on an upload but does
not RST the stream from its side, as httpd and others do.
Reported-by: jkamp-aws on github
Fixes#15316Closes#15325
vsftpd 3.0 at least writes its version number to stdin (!) instead of
stderr. This works due for backwards compatibility reasons in UNIX, so
we must check stdin for anything written there to reliably parse the
version string.
Closes#15278
Based on #14135, implement TLSv1.3 earlydata support for the curl
command line, libcurl and its implementation in GnuTLS.
If a known TLS session announces early data support, and the feature is
enabled *and* it is not a "connect-only" transfer, delay the TLS
handshake until the first request is being sent.
- Add --tls-earldata as new boolean command line option for curl.
- Add CURLSSLOPT_EARLYDATA to libcurl to enable use of the feature.
- Add CURLINFO_EARLYDATA_SENT_T to libcurl, reporting the amount of
bytes sent and accepted/rejected by the server.
Implementation details:
- store the ALPN protocol selected at the SSL session.
- When reusing the session and enabling earlydata, use exactly
that ALPN protocol for negoptiation with the server. When the
sessions ALPN does not match the connections ALPN, earlydata
will not be enabled.
- Check that the server selected the correct ALPN protocol for
an earlydata connect. If the server does not confirm or reports
something different, the connect fails.
- HTTP/2: delay sending the initial SETTINGS frames during connect,
if not connect-only.
Verification:
- add test_02_32 to verify earlydata GET with nghttpx.
- add test_07_70 to verify earlydata PUT with nghttpx.
- add support in 'hx-download', 'hx-upload' clients for the feature
Assisted-by: ad-chaos on github
Closes#15211
When uploading FTP with unknown length, we write a last 0-length chunk
with the EOS flag set. OpenSSL's SSL_write() errors on such a write.
Skip writing 0-length data to TLS backends instead.
Add test in FTPS for such uploads to verify.
Fixes#15101
Reported-by: Denis Goleshchikhin
Closes#15102
It enables running pytests in cmake jobs, regardless of underlying build
tool choice (= makes it work with ninja.)
Also:
- drop pytest logic launching `make` and exiting in case of failure.
Maybe there is a better way and keep this functionality somehow, bind
it to a command-line option? make it fail softly?
- GHA/linux: invoke pytest via the build, not directly.
- autotools: add missing dummy runtests targets when cross-compiling.
Closes#15034
Adds test cases to check that plain http: with HTTP/2 works
via 'Upgrade: h2c' or --http2-prior-knowledge'.
Also added tests to check connection reused in these situations.
Closes#14694
Update IP related information at the connection and the transfer in two
places only: once the filter chain connects and when a transfer is added
to a connection. The latter only updates on reuse when the filters
already are connected.
The only user of that information before a full connect is the HAProxy
filter. Add cfilter CF_QUERY_IP_INFO query to let it find the
information from the filters "below".
This solves two issues with the previous version:
- updates where often done twice with the same info
- happy eyeballing filter "forks" could overwrite each others
updates before the full winner was determined.
Closes#14699
Change TLS proto version on the test httpd server to test setting
combinations of --tls13-ciphers and --ciphers.
To not let the changed config of the httpd server bleed into the next
test, clean and reload on each test. Because a reload is slow, only
do this if the config is different than the loaded config. For this
the httpd.reload_if_config_changed() method is added.
Overloading of autouse fixtures does not seem to work. For the test
httpd server to be reloaded with a clean config in test_18_methods,
to not be affected by the config changes in test_17_ssl_use, the two
class scope fixtures of test_18_methods are now combined.
Closes#14589
Configure the AES 256 instead of the AES 128 cipher in the test httpd to
make scorecard testing between httpd and caddy more comparable.
Adapt test_17 expectations, now that AES 128 can no longer, but 256 can
now be negotiated.
Closes#14502
(in debug-builds)
Fix implementation in curl using libuv to process parallel transfers.
Add pytest capabilities to run test cases with --test-event.
- fix uv_timer handling to carry correct 'data' pointing to uv context.
- fix uv_loop handling to reap and add transfers when possible
- fix return code when a transfer errored
Closes#14413
revert f6cb3c63#14338
Setting SSLHonorCipherOrder to on means it honors the server cipher
order. From the documentation: "When choosing a cipher during an SSLv3
or TLSv1 handshake, normally the client's preference is used. If this
directive is enabled, the server's preference will be used instead."
Also the commit inhibits test_17_07_ssl_ciphers. The test tries to
tests if all the ciphers specified, and only those, are properly set
in curl. For that to work we need have cases where some or all ciphers
do no intersect with the cipher-set of the server. We need to be able
to assert a failed connection based on a cipher set mismatch.
That is why a restricted set of ciphers is used on the server. This
set is so chosen that it contains the well known most secure ciphers.
Except with the slower aes256 variant intentionally left out, to be
able to test above described.
As test_17_07_ssl_ciphers is currently the only test that tests the
functioning of the --ciphers and --tls13-ciphers options, it is
important that its coverage is as good as possible.
Closes#14381
Let the client, e.g. curl, influence the cipher selected in a TLS
handshake. TLS backends have different preferences and honor that
in httpd the same as Caddy does.
Also makes for a more fair compare of different TLS backends.
Closes#14338
- add upload tests to scorecard, invoke with
> python3 tests/http/scorecard.py -u h1|h2|h3
- add a reverse proxy setup from Caddy to httpd for
upload tests since Caddy does not have other PUT/POST handling
- add caddy tests in test_08 for POST/PUT
- increase read buffer in mod_curltest for larger reads
Closes#14208
When libcurl discards a connection there are two phases this may go
through: "shutdown" and "closing". If a connection is aborted, the
shutdown phase is skipped and it is closed right away.
The connection filters attached to the connection implement the phases
in their `do_shutdown()` and `do_close()` callbacks. Filters carry now a
`shutdown` flags next to `connected` to keep track of the shutdown
operation.
Filters are shut down from top to bottom. If a filter is not connected,
its shutdown is skipped. Notable filters that *do* something during
shutdown are HTTP/2 and TLS. HTTP/2 sends the GOAWAY frame. TLS sends
its close notify and expects to receive a close notify from the server.
As sends and receives may EAGAIN on the network, a shutdown is often not
successful right away and needs to poll the connection's socket(s). To
facilitate this, such connections are placed on a new shutdown list
inside the connection cache.
Since managing this list requires the cooperation of a multi handle,
only the connection cache belonging to a multi handle is used. If a
connection was in another cache when being discarded, it is removed
there and added to the multi's cache. If no multi handle is available at
that time, the connection is shutdown and closed in a one-time,
best-effort attempt.
When a multi handle is destroyed, all connection still on the shutdown
list are discarded with a final shutdown attempt and close. In curl
debug builds, the environment variable `CURL_GRACEFUL_SHUTDOWN` can be
set to make this graceful with a timeout in milliseconds given by the
variable.
The shutdown list is limited to the max number of connections configured
for a multi cache. Set via CURLMOPT_MAX_TOTAL_CONNECTIONS. When the
limit is reached, the oldest connection on the shutdown list is
discarded.
- In multi_wait() and multi_waitfds(), collect all connection caches
involved (each transfer might carry its own) into a temporary list.
Let each connection cache on the list contribute sockets and
POLLIN/OUT events it's connections are waiting for.
- in multi_perform() collect the connection caches the same way and let
them peform their maintenance. This will make another non-blocking
attempt to shutdown all connections on its shutdown list.
- for event based multis (multi->socket_cb set), add the sockets and
their poll events via the callback. When `multi_socket()` is invoked
for a socket not known by an active transfer, forward this to the
multi's cache for processing. On closing a connection, remove its
socket(s) via the callback.
TLS connection filters MUST NOT send close nofity messages in their
`do_close()` implementation. The reason is that a TLS close notify
signals a success. When a connection is aborted and skips its shutdown
phase, the server needs to see a missing close notify to detect
something has gone wrong.
A graceful shutdown of FTP's data connection is performed implicitly
before regarding the upload/download as complete and continuing on the
control connection. For FTP without TLS, there is just the socket close
happening. But with TLS, the sent/received close notify signals that the
transfer is complete and healthy. Servers like `vsftpd` verify that and
reject uploads without a TLS close notify.
- added test_19_* for shutdown related tests
- test_19_01 and test_19_02 test for TCP RST packets
which happen without a graceful shutdown and should
no longer appear otherwise.
- add test_19_03 for handling shutdowns by the server
- add test_19_04 for handling shutdowns by curl
- add test_19_05 for event based shutdowny by server
- add test_30_06/07 and test_31_06/07 for shutdown checks
on FTP up- and downloads.
Closes#13976
- add 2 variations on test_07_42 which PAUSEs uploads
and response connections terminating either right away
or after the 100-continue response
- when detecting the connection being closed in transfer.c
readwrite_data(), clear ALL send bits in data->req.keepon.
It no longer makes send to wait for a KEEP_SEND_PAUSE or HOLD.
- in the protocol client writer add the check for incomplete
response bodies. When an EOS is seen and the length is known,
check that and fail if bytes are missing.
Reported-by: Sergey Bronnikov
Fixes#13740Closes#13750
- refs #13556
- allow anon uploads on vsftpd test server
- add test_30_05 for plain upload of 1k, 100k, 1m
- add test_31_05 for SSL upload of 1k, 100k, 1m
- verify file size and contents
Closes#13734
Fixes:
- in uds tests, abort also silently on os errors
- be conservative on the h3 goaway duration
- detect curl debug build and use in checks
- fix caddy version check for slight difference under linux
- set caddy default path fitting for linux
- fix deprecation warnings in valid time checks
FTP tests:
- add '--with-test-vsftpd=path' to configure
- use vsftpd default path suitable for linux
- add test_30 with plain FTP tests
- add test_31 with --ssl-reqd FTP tests
- add vsftpd to linux GHA for pytest workflows
Closes#13661
- connect to DNS names with trailing dot
- connect to DNS names with double trailing dot
- rustls, always give `peer->hostname` and let it
figure out SNI itself
- add SNI tests for ip address and localhost
- document in code and TODO that QUIC with ngtcp2+wolfssl
does not do proper peer verification of the certificate
- mbedtls, skip tests with ip address verification as not
supported by the library
Closes#13486
- add session with destructor callback
- remove vtls `session_free` method
- let `Curl_ssl_addsessionid()` take ownership
of session object, freeing it also on failures
- change tls backend use
- test_17, add tests for SSL session resumption
Closes#13386
- fix flow handling in ngtcp2 to ACK data on streams
we abort ourself.
- extend test_02_23* cases to also run for h3
- skip test_02_23* for OpenSSL QUIC as it gets stalled
on progressing the connection
Closes#13374
A transfer with a completed download that is still uploading needs to
check the connection state when it is PAUSEd, since connection
close/errors would otherwise go unnoticed.
Reported-by: Sergey Bronnikov
Fixes#13260Closes#13271
- curl's transfer handling may write 0-length chunks at the end of the
download with an EOS flag. (HTTP/2 does this commonly)
- content encoders need to pass-through such a write and not count this
as error in case they are finished decoding
Fixes#13209Fixes#13212Closes#13219
Move all handling of HTTP's `Expect: 100-continue` feature into a client
reader. Add sending flag `KEEP_SEND_TIMED` that triggers transfer
sending on general events like a timer.
HTTP installs a `CURL_CR_PROTOCOL` reader when announcing `Expect:
100-continue`. That reader works as follows:
- on first invocation, records time, starts the `EXPIRE_100_TIMEOUT`
timer, disables `KEEP_SEND`, enables `KEEP_SEND_TIMER` and returns 0,
eos=FALSE like a paused upload.
- on subsequent invocation it checks if the timer has expired. If so, it
enables `KEEP_SEND` and switches to passing through reads to the
underlying readers.
Transfer handling's `readwrite()` will be invoked when a timer expires
(like `EXPIRE_100_TIMEOUT`) or when data from the server arrives. Seeing
`KEEP_SEND_TIMER`, it will try to upload more data, which triggers
reading from the client readers again. Which then may lead to a new
pausing or cause the upload to start.
Flags and timestamps connected to this have been moved from
`SingleRequest` into the reader's context.
Closes#13110
- when data arrived in several chunks, the collection into
the passed buffer always started at offset 0, overwriting
the data already there.
adding test_20_07 to verify fix
- debug environment var CURL_WS_CHUNK_SIZE can be used to
influence the buffer chunk size used for en-/decoding.
Closes#12945
