- refs #11157 and #11175 where uploads get stuck or lead to RST streams
- fixes our h2 send behaviour to continue sending in the nghttp2 session
as long as it wants to. This will empty our send buffer as long as
the remote stream/connection window allows.
- in case the window is exhausted, the data remaining in the send buffer
will wait for a WINDOW_UPDATE from the server. Which is a socket event
that engages our transfer loop again
- the problem in the issue was that we did not exhaust the window, but
left data in the sendbuffer and no further socket events did happen.
The server was just waiting for us to send more.
- relatedly, there was an issue fixed that closing a stream with KEEP_HOLD
set kept the transfer from shutting down - as it should have - leading
to a timeout.
Closes#11176
This reverts commit df6c2f7b54.
(It only keep the test case that checks redirection to an absolute URL
without hostname and CURLU_NO_AUTHORITY).
I originally wanted to make CURLU_ALLOW_SPACE accept spaces in the
hostname only because I thought
curl_url_set(CURLUPART_URL, CURLU_ALLOW_SPACE) was already accepting
them, and they were only not being accepted in the hostname when
curl_url_set(CURLUPART_URL) was used for a redirection.
That is not actually the case, urlapi never accepted hostnames with
spaces, and a hostname with a space in it never makes sense.
I probably misread the output of my original test when I they were
normally accepted when using CURLU_ALLOW_SPACE, and not redirecting.
Some other URL parsers seems to allow space in the host part of the URL,
e.g. both python3's urllib.parse module, and Chromium's javascript URL
object allow spaces (chromium percent escapes the spaces with %20),
(they also both ignore TABs, and other whitespace characters), but those
URLs with spaces in the hostname are useless, neither python3's requests
module nor Chromium's window.location can actually use them.
There is no reason to add support for URLs with spaces in the host,
since it was not a inconsistency bug; let's revert that patch before it
makes it into release. Sorry about that.
I also reverted the extra check for CURLU_NO_AUTHORITY since that does
not seem to be necessary, CURLU_NO_AUTHORITY already worked for
redirects.
Closes#11169
These can be interrupted by signals, especially SIGINT to shut down, and
must be restarted so the IPC call arrives correctly. If the read just
returns an error instead, the IPC calling state will go out of sync and
a proper shutdown won't happen.
Ref: #10818
Make send buffer smaller to have progress and "upload done" reporting
closer to reality. Fix handling of send "drain" condition to no longer
trigger once the transfer loop reports it is done sending. Also do not
trigger the send "drain" on RST streams.
Background:
- a upload stall was reported in #11157 that timed out
- test_07_33a reproduces a problem with such a stall if the
server 404s the request and RSTs the stream.
- test_07_33b verifies a successful PUT, using the parameters
from #11157 and checks success
Ref: #11157Closes#11165
It can only be an IPv4 address if all parts are all digits and no more than
four parts, otherwise it is a host name. Even slightly wrong IPv4 will now be
passed through as a host name.
Regression from 17a15d8846 shipped in 8.1.0
Extended test 1560 accordingly.
Reported-by: Pavel Kalyugin
Fixes#11129Closes#11131
curl_url_set(uh, CURLUPART_URL, redirurl, flags) was not respecing
CURLU_ALLOW_SPACE and CURLU_NO_AUTHORITY in the host part of redirurl
when redirecting to an absolute URL.
Closes#11136
Out of 415 labels throughout the code base, 86 of those labels were
not at the start of the line. Which means labels always at the start of
the line is the favoured style overall with 329 instances.
Out of the 86 labels not at the start of the line:
* 75 were indented with the same indentation level of the following line
* 8 were indented with exactly one space
* 2 were indented with one fewer indentation level then the following
line
* 1 was indented with the indentation level of the following line minus
three space (probably unintentional)
Co-Authored-By: Viktor Szakats
Closes#11134
libcurl assumes that a --continue-at resumption is done to continue an
upload using the read callback and neither --data nor --form use
that and thus won't do what the user wants. Whatever the user wants
with this strange combination.
Add test 426 to verify.
Reported-by: Smackd0wn on github
Fixes#11081Closes#11083
The curl binary argument wasn't being quoted properly. This seems to
have broken at some point after quoting was added in commit 606b29fe.
Reported-by: Daniel Stenberg
Ref: #11073Fixes#11074Closes#11076
When the -j option is given, a new process is spawned in which the test
programs are run and from which test servers are started. Only one
process can be started at once, but this is sufficient to test that the
infrastructure can isolate those functions in a new task. There should
be no visible difference between the two modes at the moment.
Ref: #10818Closes#11064
Program arguments are marshalled and then written to the end of a pipe
which is later read from and the arguments unmarshalled before the
desired function is called normally. The function return values are
then marshalled and written into another pipe when is later read from
and unmarshalled before being returned to the caller.
The implementation is currently blocking but can be made non-blocking
without any changes to the API. This allows calling multiple runners
without blocking in the future.
Ref: #10818
There will soon be multiple log directories so the paths will no longer
be static in runtests.pl. Also, get rid of $SERVER2IN which was not
used.
Ref: #10818
Some recent refactoring made these export no longer necessary. Also,
stop displaying the Unix socket paths at startup since there will soon
be many of them and they're not that interesting.
Ref: #10818
If the previous line starts with if/while/for AND ends with a closed
parenthesis and there's an equal number of open and closed parentheses
on that line, verify that this line is indented $indent more steps, if
not a cpp line.
Also adjust the fall-out from this fix.
Closes#11054
- nghttp2 does not free connection level window flow for
aborted streams
- when closing transfers, make sure that any buffered
response data is "given back" to the flow control window
- add tests test_02_22 and test_02_23 to reproduce
Closes#11052
This is required for the macros there to take effect for system
libraries. Specifically, including the system libraries first led to
warnings about `_FILE_OFFSET_BITS` being redefined in curl_config.h on
the Solaris autobuilds for ws-data.c and ws-pingpong.c.
Also make the curl includes come first for the other source files here
for consistency.
Closes https://github.com/curl/curl/pull/11046
Add both lib/ directories (src & build) to the search path so
curl_setup.h and its dependencies can be found.
Followup-to acd82c8b
Ref: #11006Closes#11036
- Makefile support for building test specific clients in tests/http/clients
- auto-make of clients when invoking pytest
- added test_09_02 for server PUSH_PROMISEs using clients/h2-serverpush
- added test_02_21 for lib based downloads and pausing/unpausing transfers
curl url parser:
- added internal method `curl_url_set_authority()` for setting the
authority part of a url (used for PUSH_PROMISE)
http2:
- made logging of PUSH_PROMISE handling nicer
Placing python test requirements in requirements.txt files
- separate files to base test suite and http tests since use
and module lists differ
- using the files in the gh workflows
websocket test cases, fixes for we and bufq
- bufq: account for spare chunks in space calculation
- bufq: reset chunks that are skipped empty
- ws: correctly encode frames with 126 bytes payload
- ws: update frame meta information on first call of collect
callback that fills user buffer
- test client ws-data: some test/reporting improvements
Closes#11006
- remove the version numbers
- simplify the texts
The date and version number will be put there for releases when maketgz
runs the updatemanpages.pl script.
Closes#11029
The leftmost "label" of the host name can now only match against single
'*'. Like the browsers have worked for a long time.
- extended unit test 1397 for this
- move some SOURCE variables from unit/Makefile.am to unit/Makefile.inc
Reported-by: Hiroki Kurosawa
Closes#11018
Each execution of test 1451 would leave a file in /tmp before. Since
Windows can't delete a file while it's open, all the temporary file
names are stored and deleted on exit.
Closes#10990
- state is fully kept at connection, since curl_ws_send() and
curl_ws_rec() have lifetime beyond usual transfers
- no more limit on frame sizes
Reported-by: simplerobot on github
Fixes#10962Closes#10999
- `drain` was used by http/2 and http/3 implementations to indicate
that the transfer requires send/recv independant from its socket
poll state. Intended as a counter, it was used as bool flag only.
- a similar mechanism exists on `connectdata->cselect_bits` where
specific protocols can indicate something similar, only for the
whole connection.
- `cselect_bits` are cleard in transfer.c on use and, importantly,
also set when the transfer loop expended its `maxloops` tries.
`drain` was not cleared by transfer and the http2/3 implementations
had to take care of that.
- `dselect_bits` is cleared *and* set by the transfer loop. http2/3
does no longer clear it, only set when new events happen.
This change unifies the handling of socket poll overrides, extending
`cselect_bits` by a easy handle specific value and a common treatment in
transfers.
Closes#11005
Log messages generated with logmsg can now be buffered and returned from
the runner as a return value. This will be needed with parallel testing
to allow all messages for one test to be displayed together instead of
interspersed with messages of multiple tests. Buffering can be disabled
by setting a logging callback function with setlogfunc, which is
currently being done to preserve existing logging behaviour for now.
Some additional output is generated in verbose and debugprotocol modes,
which don't always use logmsg. These modes also impact some servers
which generate extra messages. No attempt is made to buffer everything
if these modes are enabled.
Ref: #10818Closes#11016
This is closer to the place where logs are displayed on test failure.
Also, only display these logs if -p is given, which is the same flag
that controls display of test failure logs. Some server log files
need to be deleted later so that they stay around long enough to be
displayed on failure.
Ref: #10818
This is currently useful for starting a test server on its own without
an associated test, which can be used for interactive curl testing or
for validating parts of the test harness itself. More commands can be
added to perform additional functions in the future.
Ref: #10818Closes#11008
The test loop now has an initial loop that first runs through all
possible tests to build a set of those to attempt on this run based on
features and keywords and only then goes through that new list to run
them. This actually makes it three loops through all tests cases, as
there is an existing loop that gathers possible test numbers from the
test files on disk.
This has two minor effects on the output: all the tests that will be
skipped are displayed at the start (instead of being interspersed with
other tests) and the -l option no longer shows a count of tests at the
end or a (misleading) statement that tests have run successfully. The
skipped tests are also omitted from the test results sent to AppVeyor
and Azure in CI builds.
Another effect is a reduction in the amount of work considered part of
the "Test definition reading and preparation time" reported with -r
making those figures slightly lower than before.
Ref: #10818
Test 1442's name was not quoted correctly so wasn't registered in
Appveyor and it had the wrong name in Azure. The JSON string quotes were
also invalid, even though both servers happened to accept it regardless.
Closes#11010
The server ports are chosen randomly for each server, but the random
ranges chosen were inconsistently-sized and overlapping. Now, they are
spread out more so at least the first random port chosen for each server
is guaranteed to not also be chosen by another server. The starting port
numbers are also raised to put them in the Ephemeral Port range—not the
range defined by RFC 6335 but the one used by Linux, which starts lower
and gives us more room to work with.
Reported-by: Daniel Stenberg
Calling stopserver() before retrying stunnel due to an error would stop
the dependent server (such as HTTP) meaning stunnel would have nothing
to talk to when it came up. Don't try to force a stop when it didn't
actually start. Also, don't mark the server as bad for future use when
it starts up on a retry.
Reported-by: eaglegai at github
Tested-by: eaglegai at github
Fixes#10976
If a server couldn't be started on a port, a new one is randomly chosen
and the server is tried again. Avoid accidentally using a
randomly-chosen 0 port offset by adding 1 to the random number.
Found-by: Daniel Stenberg
This reduces the startup time when there is a known conflict on the
random port chosen for a server. This was already done for stunnel, but
now it's done for all servers.
testutil.pm now contains a few miscellaneous functions that are used in
several places but have no better place to live. subvariables moves to
servers.pm since most variables that it substitutes relate to servers,
so this is the most appropriate place. Rename a few functions for better
naming consistency.
Ref: #10818Closes#10995
This is code that is directly responsible for running a single test.
This will eventually run in a separate process as part of the parallel
testing project.
Ref: #10818
Otherwise, an HTTP test closely following this one with a tight time
constraint (e.g. 672) could fail because the test server stays sitting
with the wait command for a while.
As the comment states "Address family must be specified before ListenAddress", otherwise the tests fail to run
`"failed starting SSH server" 52 times (582, 583, 600, 601, 602, 603, 604, 605, 606 and 43 more)`
Closes#10983
This script does not work since the introduction of the test
preprocessing. If we need this functionality, it probably needs to be
moved into the runtests tool or similar.
Reported-by: Dan Fandrich
Fixes#10895Closes#10987
- with `--proxy-http2` allow h2 ALPN negotiation to
forward proxies
- applies to http: requests against a https: proxy only,
as https: requests will auto-tunnel
- adding a HTTP/1 request parser in http1.c
- removed h2h3.c
- using new request parser in nghttp2 and all h3 backends
- adding test 2603 for request parser
- adding h2 proxy test cases to test_10_*
scorecard.py: request scoring accidentally always run curl
with '-v'. Removed that, expect double numbers.
labeller: added http1.* and h2-proxy sources to detection
Closes#10967
- move host checks together
- simplify the scheme parser loop and the end of host name parser
- avoid itermediate buffer storing in multiple places
- reduce scope for several variables
- skip the Curl_dyn_tail() call for speed
- detect IPv6 earlier and skip extra checks for such hosts
- normalize directly in dynbuf instead of itermediate buffer
- split out the IPv6 parser into its own funciton
- call the IPv6 parser directly for ipv6 addresses
- remove (unused) special treatment of % in host names
- junkscan() once in the beginning instead of scattered
- make junkscan return error code
- remove unused query management from dedotdotify()
- make Curl_parse_login_details use memchr
- more use of memchr() instead of strchr() and less strlen() calls
- make junkscan check and return the URL length
An optimized build runs one of my benchmark URL parsing programs ~41%
faster using this branch. (compared against the shipped 7.88.1 library
in Debian)
Closes#10935
This helps enforce more modularization and encapsulation. Enable and fix
warnings on a few packages. Also, rename ftp.pm to processhelp.pm since
there's really nothing ftp-specific in it.
Ref: #10818
- Use an absolute path for the -L option since the module isn't in the
perl path
- Create the needed test file in a <file> section; <precheck> isn't
intended for this
- Fix the test number in the file name, which was wrong
Follow-up to f754990a
Ref: #10818Fixes#10889Closes#10917
It's not available on Solaris 10, for example. Since this is just test
code that doesn't need to use an optimized system version, replace it
with the implementation copied from tool_cb_hdr.c.
A typical mistake would be to try to set "https://" - including the
separator - this is now rejected as that would then lead to
url_get(... URL...) would get an invalid URL extracted.
Extended test 1560 to verify.
Closes#10911
The only user of this define was 'chkdecimalpoint' - a special purpose
test tool that was built but not used anymore (since 17c18fbc3 - Apr
2020).
Closes#10908
- for https CONNECT forwarding, this was fixed at 5 seconds
which led to spurious CI test failures
- add --keepalive parameter to sws to control this
- let httpserver use 30 seconds
Closes#10898
- currently only on debug build and when env variable
CURL_PROXY_TUNNEL_H2 is present.
- will ALPN negotiate with the proxy server and switch
tunnel filter based on the protocol negotiated.
- http/1.1 tunnel code moved into cf-h1-proxy.[ch]
- http/2 tunnel code implemented in cf-h2-proxy.[ch]
- tunnel start and ALPN set remains in http_proxy.c
- moving all haproxy related code into cf-haproxy.[ch]
VTLS changes
- SSL filters rely solely on the "alpn" specification they
are created with and no longer check conn->bits.tls_enable_alpn.
- checks on which ALPN specification to use (or none at all) are
done in vtls.c when creating the filter.
Testing
- added a nghttpx forward proxy to the pytest setup that
speaks HTTP/2 and forwards all requests to the Apache httpd
forward proxy server.
- extending test coverage in test_10 cases
- adding proxy tests for direct/tunnel h1/h2 use of basic auth.
- adding test for http/1.1 and h2 proxy tunneling to pytest
Closes#10780
- eliminate receive loop in vtls to fill buffer. This may
lead to partial reads of data which is counter productive
- let http2 instead loop smarter to process pending network
data without transfer switches
scorecard improvements
- do not start caddy when only httpd is requested
- allow curl -v to stderr file on --curl-verbose
Closes#10891
Using bad numbers in an IPv4 numerical address now returns
CURLUE_BAD_HOSTNAME.
I noticed while working on trurl and it was originally reported here:
https://github.com/curl/trurl/issues/78
Updated test 1560 accordingly.
Closes#10894
Meaning that it would wrongly still store the fragment using spaces
instead of %20 if allowing space while also asking for URL encoding.
Discovered when playing with trurl.
Added test to lib1560 to verify the fix.
Closes#10887
Output specific components from the used URL. The following variables
are added for this purpose:
url.scheme, url.user, url.password, url.options, url.host, url.port,
url.path, url.query, url.fragment, url.zoneid
Add the following for outputting parts of the "effective URL":
urle.scheme, urle.user, urle.password, urle.options, urle.host, urle.port,
urle.path, urle.query, urle.fragment, urle.zoneid
Added test 423 and 424 to verify.
Closes#10853
Allow UNITTEST_STOP to return the error code, use the fail & abort
macros to indicate test failure and return success instead of fail if
the unit test can't test anything because of missing features at
compile-time. A couple of tests could never fail because they were
overriding the failure return code.
These files are generated by the test servers and must therefore be
found in the log directory to make them available to only those servers
once multiple test runners are executing in parallel. They must also not
be deleted with the log files, so they are stored in the pidfile
directory.
Ref: #10818Closes#10875
- when rustls is told to recieve more TLS data and its internal
plaintext buffers are full, it returns an IOERROR
- avoid receiving TLS data while plaintext is not read empty
pytest:
- increase curl run timeout when invoking pytest with higher verbosity
Closes#10876
- ngtcp2: using bufq for recv stream data
- internal stream_ctx instead of `struct HTTP` members
for quiche, ngtcp2 and msh3
- no more QUIC related members in `struct HTTP`
- experimental use of recvmmsg(), disabled by default
- testing on my old debian box shows no throughput improvements.
- leaving it in, but disabled, for future revisit
- vquic: common UDP receive code for ngtcp2 and quiche
- vquic: common UDP send code for ngtcp2 and quiche
- added pytest skips for known msh3 failures
- fix unit2601 to survive torture testing
- quiche: using latest `master` from quiche and enabling large download
tests, now that key change is supported
- fixing test_07_21 where retry handling of starting a stream
was faulty
- msh3: use bufq for recv buffering headers and data
- msh3: replace fprintf debug logging with LOG_CF where possible
- msh3: force QUIC expire timers on recv/send to have more than
1 request per second served
Closes#10772
Values greater than 125 have special meanings, so cap it there. Unit
tests and lib tests use the number of failures as the return code, so a
large number of failures (such as test 2601 as a torture test) can
exceed this causing the test to be erroneously reported as having
failed.
Ref: #10720
Otherwise, it might find the binary in .libs which can cause it to use
the system libcurl which can fail. This error is only visible by
noticing that the test is skipped.
Follow-up to e4dfe6fc
Ref: #10651
- use bufq for send/receive of network data
- usd bufq for send/receive of stream data
- use HTTP/2 flow control with no-auto updates to control the
amount of data we are buffering for a stream
HTTP/2 stream window set to 128K after local tests, defined
code constant for now
- elminiating PAUSEing nghttp2 processing when receiving data
since a stream can now take in all DATA nghttp2 forwards
Improved scorecard and adjuste http2 stream window sizes
- scorecard improved output formatting and options default
- scorecard now also benchmarks small requests / second
Closes#10771
Not all warnings are fixed; many are as much stylistic suggestions than
anything and IMHO don't do much to actually improve the code.
Ref: #10818Closes#10861
It's actually been this way since at least 2012 (when a 3-argument open
was added to runtests.pl). Given the lack of complaints in the interim,
it's safe to call this 23 year old perl version the minimum.
The refactored code calls these functions with the same arguments more
often, so this prevents redundant test case file parsing.
Approved-by: Daniel Stenberg
Ref: #10818Closes#10833
This takes it from a 1200 line behemoth into something more manageable.
The content and order of the functions is taken almost directly from
singletest() so the diff sans whitespace is quite short.
Ref: #10818
Namely:
- Verify that this test case should be run
- Start the servers needed to run this test case
- Check that test environment is fine to run this test case
- Prepare the test environment to run this test case
- Run the test command
- Clean up after test command
- Verify test succeeded
Ref: #10818
RFC 7686 states that:
> Applications that do not implement the Tor
> protocol SHOULD generate an error upon the use of .onion and
> SHOULD NOT perform a DNS lookup.
Let's do that.
https://www.rfc-editor.org/rfc/rfc7686#section-2
Add test 1471 and 1472 to verify
Fixes#543Closes#10705
Some IP cameras send malformed RTSP interleaved frames sometimes, which
can cause curl_easy_perform return 1 (CURLE_UNSUPPORTED_PROTOCOL). This
change attempts to skip clearly incorrect RTSP interleaving frame data.
Closes#10808
Adding `bufq`:
- at init() time configured to hold up to `n` chunks of `m` bytes each.
- various methods for reading from and writing to it.
- `peek` support to get access to buffered data without copy
- `pass` support to allow buffer flushing on write if it becomes full
- use case: IO buffers for dynamic reads and writes that do not blow up
- distinct from `dynbuf` in that:
- it maintains a read position
- writes on a full bufq return CURLE_AGAIN instead of nuking itself
- Init options:
- SOFT_LIMIT: allow writes into a full bufq
- NO_SPARES: free empty chunks right away
- a `bufc_pool` that can keep a number of spare chunks to
be shared between different `bufq` instances
Adding `dynhds`:
- a straightforward list of name+value pairs as used for HTTP headers
- headers can be appended dynamically
- headers can be removed again
- headers can be replaced
- headers can be looked up
- http/1.1 formatting into a `dynbuf`
- configured at init() with limits on header counts and total string
sizes
- use case: pass a HTTP request or response around without being version
specific
- express a HTTP request without a curl easy handle (used in h2 proxy
tunnels)
- future extension possibilities:
- conversions of `dynhds` to nghttp2/nghttp3 name+value arrays
Closes#10720
- will check built curl for http and https support and
skip all tests if not there
- will dump stdout/stderr/trace output on errored responses
Closes#10829
- we had a CI case once where `curl` seemingly did not
return and it was hard to guess what happened.
- make curl execution in test cases time out after 60 seconds
Closes#10783
The test does a slightly ugly busy-loop for this case but should be
managable due to it likely being a very short moment.
Mention CURLE_AGAIN in curl_ws_recv.3
Fixes#10760
Reported-by: Jay Satiro
Closes#10781
all s3 requests default to UNSIGNED-PAYLOAD and add the required
x-amz-content-sha256 header. this allows CURLAUTH_AWS_SIGV4 to correctly
sign s3 requests to amazon with no additional configuration
Signed-off-by: Casey Bodley <cbodley@redhat.com>
Closes#9995
- added to: ngtcp2-quictls, ngtcp2-gnutls and the linux varians
quiche, bearssl, libressl, mbedtls, openssl3, rustls
- added disabled in ngtcp2-wolfssl due to weird SSL_connect() errors
not reproducable locally
Improvements on pytest:
- handling of systems with nghttpx in $PATH
- configure will seach $PATH got nghttpx used in pytest
- pytest fixes for managing nghttpx without h3 support
- ngtcp2-wolfssl: use a fully enabled wolfssl build
- lower parallel count for http/1.1 tests, since we do not
want to test excessive connections.
- check built curl for HTTPS-proxy support in proxy tests
- bearssl does not like one of our critical cert extensions, making
it non-critical now
- bearssl is too slow for test_12, skipping
- making sure we do h3 tests only when curl and server support is there
Closes#10699
The only difference used to be global variable used in unittest tests.
After cb7ed5a removed individual flag overrides for the unittests, first.c
was no longer recompiled for unit tests to include the flag, so whether it
worked or gave a link error depended on whether it was compiled in
libtest or unittest first. This way also speeds up the build by
eliminating 40 identical compile invocations.
Fixes#10749
Using CPPFLAGS sometimes caused odd compile issues when building tests
with parallel make and AM_CPPFILES is the right flag, anyway.
Follow-up to cb7ed5a
Ref #10749
This is already how curl is documented to behave in Everything curl, but
in actuality only short POSTs skip this. This should knock 30 seconds
off a full run of the test suite since the 100-continue timeout will no
longer be hit.
Closes#10740
These are only built when a libcurl static library is available, since
we're not building a special libcurlu library yet and these tests rely
on private symbols that aren't available in the shared library. A few
unit tests do require libcurlu, so those are not built.
Closes#10722
Put only the test names into Makefile.inc so they can be used by both
cmake and automake. This will prevent the list of tests from becoming
out of date when they are also built under cmake.
It results in error "NSS error -5985 (PR_ADDRESS_NOT_SUPPORTED_ERROR)"
Disabled test 1470 for NSS builds and documented the restriction.
Reported-by: Dan Fandrich
Fixes#10723Closes#10734
- add parameter to `conn_is_alive()` cfilter method that returns
if there is input data waiting on the connection
- refrain from re-using connnection from the cache that have
input pending
- adapt http/2 and http/3 alive checks to digest pending input
to check the connection state
- remove check_cxn method from openssl as that was just doing
what the socket filter now does.
- add tests for connection reuse with special server configs
Closes#10690
- a reset transfer (HTTP/2 RST) did not always lead to the proper
error message on receiving its response, leading to wrong reports
of a successful transfer
- test_05_02 was able to trigger this condition with increased transfer
count. The simulated response errors did not carry a 'Content-Length'
so only proper RST handling could detect the abort
- When doing such transfers in parallel, a connection could enter the
state where
a) it had been closed (GOAWAY received)
b) the RST had not been "seen" for the transfer yet
or c) the GOAWAY announced an error and the last successful
stream id was not checked against ongoing transfers
Closes#10693
- necessary ports are bound at start of test suite and then
given to server fixtures for use.
- this make parallel use of pytest (in separate directories),
practically safe for use as OS tend to not reuse such port numbers
for a while
Closes#10692
- time_connect was not updated when the overall connection failed,
e.g. when SSL verification was unsuccessful, refs #10670
- rework gather those values to interrogate involved filters,
also from all eyeballing attempts, to report the maximum of
those values.
- added 3 test cases in test_06 to check reported values on
successful, partially failed and totally failed connections.
Reported-by: Master Inspire
Fixes#10670Closes#10671
When returned from the CURLOPT_SOCKOPTFUNCTION, like when we have a
custom socket connected in the app, passed in to libcurl.
Verifies the fix in #10648Closes#10651
- httpd is only one server we test with
- the suite coveres the HTTP protocol in general where
the default test cases need a more beefy environment
Closes#10654
- refs #10634 where errors in the HTTP/2 framing layer are observed.
- the bug was that on connection reuse, the code attempted to switch
in yet another layer of HTTP/2 handling instead of detecting that
this was already in place.
- added pytest testcase reproducing the issue.
Reported-by: rwmjones on github
Fixes#10634Closes#10643
- when h2/h3 eyeballing was involved, unix domain socket
configurations were not honoured
- configuring --unix-socket will disable HTTP/3 as candidate for eyeballing
- combinatino of --unix-socket and --http3-only will fail during initialisation
- adding pytest test_11 to reproduce
Reported-by: Jelle van der Waa
Fixes#10633Closes#10641
As tested in test_02_07, when firing off 200 urls with --parallel, 199
wait for the first connection to be established. if that is multiuse,
urls are added up to its capacity.
The first url over capacity opens another connection. But subsequent
urls found the same situation and open a connection too. They should
have waited for the second connection to actually connect and make its
capacity known.
This change fixes that by
- setting `connkeep()` early in the HTTP setup handler. as otherwise
a new connection is marked as closeit by default and not considered
for multiuse at all
- checking the "connected" status for a candidate always and continuing
to PIPEWAIT if no alternative is found.
pytest:
- removed "skip" from test_02_07
- added test_02_07b to check that http/1.1 continues to work as before
Closes#10456
- httpserver.pl -> http-server.pl for consistency
- add http3-server.pl to EXTRA_DIST; alphabetise for maintainability
- nghttpx proxy invocation scripts should not use getcwd
Closes#10568
- sscanf() is rather complex and slow, strchr() much simpler
- the port number function does not need to fully verify the IPv6 address
anyway as it is done later in the hostname_check() function and doing
it twice is unnecessary.
Closes#10541
- adding test_02_20 for reproducing the situation
- using recently released mod_h2 Apache module
- skipping test if an older version is installed
- adding installation of current mod_h2 to github pytest workflow
This reproduces the error reliable (for me) on the lib/http2.c version
of curl 7.88.0. And passes with the recent curl master.
Closes#10534
- connect timeout was used at half the configured value, if the
destination had 1 ip version 4 and other version 6 addresses
(or the other way around)
- extended test2600 to reproduce these cases
Reported-by: Michael Kaufmann
Fixes#10514Closes#10517
- adding pytest test_10 cases for proxy httpd setup tests
- fixing openssl bug in https: proxy hostname verification that
used the hostname of the request and not the proxy name.
Closes#10498
vquic stabilization
- udp send code shared between ngtcp2 and quiche
- quiche handling of data and events improved
ngtcp2 and pytest improvements
- fixes handling of "drain" situations, discovered in scorecard
tests with the Caddy server.
- improvements in handling transfers that have already data or
are already closed to make an early return on recv
pytest
- adding caddy tests when available
scorecard improvemnts.
- using correct caddy port
- allowing tests for only httpd or caddy
Closes#10451
- as reported in #10433, HTTP/2 uploads may stall when a response is
received before the upload is done. This happens when the
data->state.drain is set for such a transfer, as the special handling
in transfer.c from then on only cared about downloads.
- add continuation of uploads, if applicable, in this case.
- add pytest case test_07_12_upload_seq_large to reproduce this scenario
(although, current nghttp2 implementation is using drain less often)
Reported-by: Lucas Pardue
Fixes#10433Closes#10443
Prior to this change a workaround for Windows to recv before every send
was enabled by default. The way it works is a recv is called before
every send and saves the received data, in case send fails because in
Windows apparently that can wipe out the socket's internal received
data buffer.
This feature has led to several bugs because the way libcurl operates
it waits on a socket to read or to write, and may not at all times
check for buffered receive data.
Two recent significant bugs this workaround caused:
- Broken Schannel TLS 1.3 connections (#9431)
- HTTP/2 arbitrary hangs (#10253)
The actual code remains though it is disabled by default. Though future
changes to connection filter buffering could improve the situation IMO
it's just not tenable to manage this workaround.
Ref: https://github.com/curl/curl/issues/657
Ref: https://github.com/curl/curl/pull/668
Ref: https://github.com/curl/curl/pull/720
Ref: https://github.com/curl/curl/issues/9431
Ref: https://github.com/curl/curl/issues/10253
Closes https://github.com/curl/curl/pull/10409
curl would erroneously report CURLE_OUT_OF_MEMORY in some cases instead
of CURLE_URL_MALFORMAT. In other cases, it would erroneously return
CURLE_URL_MALFORMAT instead of CURLE_OUT_OF_MEMORY. Add a test case to
test the former condition.
Fixes#10130Closes#10414
New cfilter HTTP-CONNECT for h3/h2/http1.1 eyeballing.
- filter is installed when `--http3` in the tool is used (or
the equivalent CURLOPT_ done in the library)
- starts a QUIC/HTTP/3 connect right away. Should that not
succeed after 100ms (subject to change), a parallel attempt
is started for HTTP/2 and HTTP/1.1 via TCP
- both attempts are subject to IPv6/IPv4 eyeballing, same
as happens for other connections
- tie timeout to the ip-version HAPPY_EYEBALLS_TIMEOUT
- use a `soft` timeout at half the value. When the soft timeout
expires, the HTTPS-CONNECT filter checks if the QUIC filter
has received any data from the server. If not, it will start
the HTTP/2 attempt.
HTTP/3(ngtcp2) improvements.
- setting call_data in all cfilter calls similar to http/2 and vtls filters
for use in callback where no stream data is available.
- returning CURLE_PARTIAL_FILE for prematurely terminated transfers
- enabling pytest test_05 for h3
- shifting functionality to "connect" UDP sockets from ngtcp2
implementation into the udp socket cfilter. Because unconnected
UDP sockets are weird. For example they error when adding to a
pollset.
HTTP/3(quiche) improvements.
- fixed upload bug in quiche implementation, now passes 251 and pytest
- error codes on stream RESET
- improved debug logs
- handling of DRAIN during connect
- limiting pending event queue
HTTP/2 cfilter improvements.
- use LOG_CF macros for dynamic logging in debug build
- fix CURLcode on RST streams to be CURLE_PARTIAL_FILE
- enable pytest test_05 for h2
- fix upload pytests and improve parallel transfer performance.
GOAWAY handling for ngtcp2/quiche
- during connect, when the remote server refuses to accept new connections
and closes immediately (so the local conn goes into DRAIN phase), the
connection is torn down and a another attempt is made after a short grace
period.
This is the behaviour observed with nghttpx when we tell it to shut
down gracefully. Tested in pytest test_03_02.
TLS improvements
- ALPN selection for SSL/SSL-PROXY filters in one vtls set of functions, replaces
copy of logic in all tls backends.
- standardized the infof logging of offered ALPNs
- ALPN negotiated: have common function for all backends that sets alpn proprty
and connection related things based on the negotiated protocol (or lack thereof).
- new tests/tests-httpd/scorecard.py for testing h3/h2 protocol implementation.
Invoke:
python3 tests/tests-httpd/scorecard.py --help
for usage.
Improvements on gathering connect statistics and socket access.
- new CF_CTRL_CONN_REPORT_STATS cfilter control for having cfilters
report connection statistics. This is triggered when the connection
has completely connected.
- new void Curl_pgrsTimeWas(..) method to report a timer update with
a timestamp of when it happend. This allows for updating timers
"later", e.g. a connect statistic after full connectivity has been
reached.
- in case of HTTP eyeballing, the previous changes will update
statistics only from the filter chain that "won" the eyeballing.
- new cfilter query CF_QUERY_SOCKET for retrieving the socket used
by a filter chain.
Added methods Curl_conn_cf_get_socket() and Curl_conn_get_socket()
for convenient use of this query.
- Change VTLS backend to query their sub-filters for the socket when
checks during the handshake are made.
HTTP/3 documentation on how https eyeballing works.
TLS improvements
- ALPN selection for SSL/SSL-PROXY filters in one vtls set of functions, replaces
copy of logic in all tls backends.
- standardized the infof logging of offered ALPNs
- ALPN negotiated: have common function for all backends that sets alpn proprty
and connection related things based on the negotiated protocol (or lack thereof).
Scorecard with Caddy.
- configure can be run with `--with-test-caddy=path` to specify which caddy to use for testing
- tests/tests-httpd/scorecard.py now measures download speeds with caddy
pytest improvements
- adding Makfile to clean gen dir
- adding nghttpx rundir creation on start
- checking httpd version 2.4.55 for test_05 cases where it is needed. Skipping with message if too old.
- catch exception when checking for caddy existance on system.
Closes#10349
- add test2600 as a unit test that triggers various connect conditions
and monitors behaviour, available in a debug build only.
- this exposed edge cases in connect.c that have been fixed
Closes#10312
- adding '--with-test-httpd=<path>' to configure non-standard apache2
install
- python env and base classes for running httpd
- basic tests for connectivity with h1/h2/h3
- adding test cases for truncated responses in http versions.
- adding goaway test for HTTP/3.
- adding "stuttering" tests with parallel downloads in chunks with
varying delays between chunks.
- adding a curltest module to the httpd server, adding GOAWAY test.
- mod_curltest now installs 2 handlers
- 'echo': writing as response body what came as request body
- 'tweak': with query parameters to tweak response behaviour
- marked known fails as skip for now
Closes#10175
Built-in TLS backends are detected at test time by scanning for their
names in the version string line returned by the cli tool: as this line
may also list the libssh configuration that mentions its own backend,
the curl backend may be wrongly determined.
In example, if the version line contains "libssh/0.10.4/openssl/zlib",
OpenSSL is detected as a curl-configured backend even if not.
This fix requires the backend names to appear as full words preceded by
spacing in the version line to be recognized as curl TLS backends.
Closes#10236
- Set all scripts +x, ie 644 => 755.
Prior to this change some scripts were not executable and therefore
could not be called directly.
~~~
git ls-files -s \*.{sh,pl,py} | grep -v 100755
~~~
Closes https://github.com/curl/curl/pull/10219
