When the thread started is too fast, the signal will come before the
wait. Add an additional check before the wait to catch the thread
having started or already ended.
Closes#18344
- During handshake, do not require reading more data if unprocessed
encrypted data that may be a complete TLS record is already available.
- During handshake, check that the socket is writeable before processing
encrypted data that may require an immediate reply to the server.
These two fixes are for issues that were found during renegotiation
testing but could affect any handshake.
Prior to this change it was possible in some abnormal network conditions
for the Schannel TLS handshake procedure to erroneously wait or error.
Ref: https://github.com/curl/curl/pull/18125
Closes https://github.com/curl/curl/pull/18323
It had a typo, but it wasn't causing an issue, because `TLS1_3_VERSION`
is enough to detect this feature and the version check remained unused.
Follow-up to 0d3b5937b3#16477
Cherry-picked from #18330Closes#18333
Changed strategy to start up and terminate resolver thread.
When starting up:
Start the thread with mutex acquired, wait for signal from thread that
it started and has incremented the ref counter. Thread set
pthread_cancel() to disabled before that and only enables cancelling
during resolving itself. This assure that the ref counter is correct and
the unlinking of the resolve context always happens.
When shutting down resolving:
If ref counting shows thread has finished, join it, free everything. If
thread has not finished, try pthread_cancel() (non Windows), but keep
the thread handle around.
When destroying resolving:
Shutdown first, then, if the thread is still there and 'quick_exit' is
not set, join it and free everything. This might occur a delay if
getaddrinfo() hangs and cannot be interrupted by pthread_cancel().
Destroying resolving happens when another resolve is started on an
easy handle or when the easy handle is closed.
Add test795 to check that connect timeout triggers correctly
when resolving is delayed. Add debug env var `CURL_DNS_DELAY_MS`
to simulate delays in resolving.
Fix test1557 to set `quick_exit` and use `xxx.invalid` as domain
instead of `nothing` that was leading to hangers in CI.
Closes#18263
`runtests.pl` defaults to `-` if a command is not set, since
c43ad0f972.
Also:
- drop a stray `</file>`.
- replace a `nothing` with a guaranteed invalid hostname.
Ref: 4334033b43
- replace unused URLs with `-`.
Closes#18319
We are destroying the ares channel already when we shutdown
resolving. Querying the pollset afterwards is still happening,
especially in event based processing and needs to work in the
absence of a channel.
Fixes#18317
Reported-by: Natris on github
Closes#18318
To make the CI jobs use native Win32 API calls instead of calling
external tools to look up and kill PIDs of native Windows test server
processes.
Follow-up to 2388b0e587#18308Closes#18296
It's somewhat flaky, slow (6-8 minutes), needs maintenance, and did not
turn up real issues to justify keeping.
Notably it did not help catch a regression seen on Solaris OS: #16915
Follow-up to 90e644f944#13583Closes#18314
`Win32::Process::List` and `Win32::Process`.
To replace external calls to `tasklist.exe` and `taskkill.exe`.
The perl modules are wrappers/binding to Win32 API calls. They avoid
launching external processes with a command shell (including MSYS2),
the external tool, and creating command-lines and parsing tool output.
According to local tests and the CI, one test session calls
`tasklist.exe` 350-400 times. `taskkill.exe` is rarely called:
https://github.com/curl/curl/actions/runs/17012376726?pr=18296
It's hard to predict any possible side-effect of dynamically loading
the two necessary, module DLLs into the Perl process. The MSYS2 runtime
if prone to fail when doing this in fork operations, as seen earlier
with the Win32.DLL module. But, is looks like a symptom, not the root
cause for these failures, because the failures are present with or
without perl.exe loading the Win32.DLL.
Ref: be01b60ce5#18287
Cherry-picked from #18296Closes#18308
This patch fixes flakiness caused by MSBuild scanning the runtests.pl
output for regex patterns. When finding a hit, it returns an error code
to cmake, making the build test CI step fail. This happens rarely after
an earlier mitigation tweaking outputs, but, as expected, it did not
resolve it completely.
MSBuild doesn't have an option to disable this behavior. To fix, this
patch migrates the two affected jobs from MSBuild to Ninja. To align
with existing multi-config logic, it uses the `Ninja Multi-Config`
generator, which hasn't been tested before in CI.
Switching to Ninja was not trivial. Visual Studio to this day relies on
an MS-DOS batch file stored at an unstable location (containing spaces
and parenthesis), to initialize its environment. Without this env,
`cl.exe` is unable to find its own components. GHA does not initialize
it (even if it did, it could only default to a single specific target).
CMake helps with this when using a Visual Studio generator, but doesn't
when using Ninja. (On local machines the VS installer adds a couple
of Start menu items for launching pre-configured command prompts.)
Ref: https://learn.microsoft.com/cpp/build/building-on-the-command-line
The MS-DOS batches don't integrate well with CI envs and even less so
with shell scripts. To avoid it, this patch uses manual configuration.
Also without using environment variables, to make it easy to use and
easy to debug and trace in logs. Configuring Visual Studio is relatively
stable across releases and hasn't changed a whole lot in the last 2
decades, but still may need more maintenance compared to llvm, or pretty
much any other toolchain out there. On the upside, it allows to manually
select compiler version, SDK version, cross-combinations, and allows
choosing clang-cl. The configuration aims to find the latest of these
automatically.
Some traps that had to be avoided:
- need to switch to MS-DOS short names to avoid spaces in the VS
component paths.
- need to switch to forward slashes to avoid confusing downstream tools
with backslashes.
- need to pass either MSYS2 for Windows-style path depending on setting.
- need to use a trick to retrieve the oddly named `ProgramFiles(x86)`
Windows env from shell script.
- need to match VS version (2022) and edition (Enterprise), found on GHA
runners.
- need to pass the CMake generator via env so that the space in the name
doesn't trip the shell when passed via a variable.
- trash and unexpected dirs when detecting SDK/toolchain versions.
- need to pass `-external:W0` to the C compiler to avoid MSVC warning:
`D9007: '/external:I' requires '/external:W'; option ignored`
- using cmake options only, to make it run without relying on envs and
work out-of-the-box when running subsequent cmake sessions.
- some others discovered while making work clang-cl locally in
cross-builds.
Ninja also improves performance in most cases (though wasn't a goal
here). After this patch configure is significantly faster (1.5-2x),
builds are a tiny bit faster, except examples which was twice as fast
with MSBuild. Disk space use is 10% lower.
MSBuild builds remain tested in AppVeyor CI and the UWP job.
Before: https://github.com/curl/curl/actions/runs/17025737223/job/48260856051
After: https://github.com/curl/curl/actions/runs/17027981486/job/48266133301
Fixes:
```
=== Start of file stderr1635
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 4 100 4 0 0 449 0 --:--:-- --:--:-- --:--:-- 500
curl : (22) The requested URL returned error : 429 [D:\a\curl\curl\bld\tests\test-ci.vcxproj]
CUSTOMBUILD : warning : Problem : HTTP error. Will retry in 1 second. 1 retry left. [D:\a\curl\curl\bld\tests\test-ci.vcxproj]
[...]
C:\Program Files\Microsoft Visual Studio\2022\Enterprise\MSBuild\Microsoft\VC\v170\Microsoft.CppCommon.targets(254,5): error MSB8066: Custom build for 'D:\a\curl\curl\bld\CMakeFiles\621f80ddbb0fa48179f056ca77842ff0\test-ci.rule;D:\a\curl\curl\tests\CMakeLists.txt' exited with code -1. [D:\a\curl\curl\bld\tests\test-ci.vcxproj]
Error: Process completed with exit code 1.
```
Ref: https://github.com/curl/curl/actions/runs/16966304797/job/48091058271?pr=18287#step:13:3471
Bug: https://github.com/curl/curl/discussions/14854#discussioncomment-14104166
Ref: a19bd43210#18307
Follow-up to 9463769f2e#16583Closes#18301
Pass curl's memory functions to the nghttp3 and ngtcp2 functions that
allow them. This allows custom memory functions passed by the curl user
to be used in nghttp3 and ngtcp2.
Closes#18196
- remove a superfluous if()
- edit the #ifdef to only do the timeout checks when socketpair support
is missing, as that code flow could not be reached otherways anyway
CodeSonar pointed out possible issue that led me to see this.
Closes#18306
`CMAKE_C_FLAGS` is not set by curl, but may contain custom options
required for a successful compiler run, when invoked by these tests.
One such case is when configuring Visual Studio or clang-cl via compiler
options, instead of envs.
Cherry-picked from #18301Closes#18307
When an application install its own OpenSSL verify callback and that
callback invokes `SSL_set_retry_verify()`, the transfer is automatically
paused and does not progress the connect attempt any further until
unpaused via `curl_easy_pause().
Added test758 to verify.
Ref: #18284
Original PR by @Natris
Bug: https://curl.se/mail/lib-2025-08/0012.htmlCloses#18288
Easily triggered when doing an FTP ranged download with the end range
larger than the file size.
A regression since 30de937bda (shipped in curl 8.9.0)
Add test 2307 to verify.
Reported-by: kupavcevdenis on github
Fixes#18283Closes#18295
- they are fake anyway
- these ones look more obvious fake and in need of replacement
- avoids the risk that they might decode to something of value
Closes#18261
Make the:
- mbedTLS valgrind job finish under 14m, vs 15m before.
- OpenSSL -O3 valgrind job finish in 14m30, vs 16m17.
- OpenSSL libssh2 valgrind job finish in 16m, vs 17m30.
- long valgrind rustls job finish 1 minute earlier, in return
for spending 30s more on the other rustls job.
Keep using autotools for the less slow valgrind job to test this combo.
Closes#18290
`Time::HiRes` was already used unconditionally before this patch in
`servers.pm`. This package, and functions used by runtests (`sleep` and
`gettimeofday`) are supported by the minimum Perl version required for
curl:
https://perldoc.perl.org/5.8.0/Time::HiRes
- Drop the `portable_sleep()` wrapper in favor of `Time::HiRes::sleep()`.
- Use `Time::HiRes` unconditionally in `serverhelp.pm`.
- Stop using the `Win32` package where available. It was included
to provide a Windows fallback for `Time::HiRes::sleep()`. It was never
actually called, but the dependency may have loaded `Win32.dll`, which
often appears in failed fork operations in GHA logs.
Ref: a6fed41f6f#5054#5034
Ref: https://github.com/curl/curl/discussions/14854Closes#18287
To not create a file named `nul` in the `tests` directory when running
tests with MSYS2/Cygwin Perl (the only supported Perl for Windows).
I imagine this may cause issues when being written in parallel. Also
`nul` is a reserved filename on Windows.
Closes#18282
