- Save the last set SO_SNDBUF size to compare against so that we can
avoid setsockopt calls every second.
This is a follow-up to 0b520e12 which moved the SO_SNDBUF update check
into cf-socket. This change improves it further by making the function
names easier to understand and reducing the amount of setsockopt calls.
Adjusts the `make -j` flag to match the latest GitHub-hosted runner
hardware specs[^1]:
- `ubuntu-latest` on 4 CPU cores
- `macos-latest` on 3 CPU cores
The processor count is ideally obtained from `nproc`, but setting env
vars from the current CI yaml files is not possible because they expect
literal strings.
Stefan Eissing [Wed, 29 May 2024 15:13:34 +0000 (17:13 +0200)]
multi: fix multi_wait() timeout handling
- determine the actual poll timeout *after* all sockets
have been collected. Protocols and connection filters may
install new timeouts during collection.
- add debug logging to test1533 where the mistake was noticed
Reported-by: Matt Jolly
Fixes #13782
Closes #13825
Viktor Szakats [Tue, 28 May 2024 14:16:01 +0000 (16:16 +0200)]
lib: prefer `var = time(NULL)` over `time(&var)`
Following up on previous occurrences showing up as gcc warnings, replace
the remaining `time(&var)` calls with `var = time(NULL)`, though these
aren't specifically causing compiler warnings. These are in the TFTP
client code (`lib/tftp.c`), except one which is in a debug branch in
`lib/http_aws_sigv4.c`.
What's unexplainable is that this patch seems to mitigate TFTP tests
often hanging or going into an infinite loop on GHA windows workflows
with MSYS2, mingw-w64 and MSVC (Cygwin is unaffected):
https://github.com/curl/curl/pull/13599#issuecomment-2119372376
TFTP hangs did not entirely disappear though, so could be unrelated.
Stefan Eissing [Fri, 24 May 2024 08:09:32 +0000 (10:09 +0200)]
winsock: move SO_SNDBUF update into cf-socket
- Move the code that updates the SO_SNDBUF size for Windows to
cf_socket_send.
Prior to this change the code was in readwrite_upload but the socket
filter is the more appropriate place because it applies to all sends.
Background:
For Windows users SO_SNDBUF (the total per-socket buffer size reserved
by Winsock for sends) is updated dynamically by libcurl during the
transfer. This is because Windows does not do it automatically for
non-blocking sockets and without it the performance of large transfers
may suffer.
Matthias Gatto [Mon, 27 May 2024 12:58:11 +0000 (14:58 +0200)]
aws-sigv4: url encode the canonical path
Refactors canon_query, so it could use the encoding part of the function
to use it in the path.
As the path doesn't encode '/', but encode '=', I had to add some
conditions to know If I was doing the query or path encoding.
Also, instead of adding a `bool in_path` variable, I use `bool
*found_equals` to know if the function was called for the query or path,
as found_equals is used only in query_encoding.
Test 472 verifies.
Reported-by: Alexander Shtuchkin
Fixes #13754
Closes #13814
Daniel Stenberg [Tue, 28 May 2024 06:32:19 +0000 (08:32 +0200)]
cd2nroff: use an empty "##" to signal end of .IP sequence
Like when we list a series of options and then want to add "normal" text
again afterwards.
Without this, the indentation level wrongly continues even after the
final "##" header, making following text wrongly appear to belong to the
header above.
Adjusted several curldown files to use this.
Fixes #13803 Reported-by: Jay Satiro
Closes #13806
Daniel Stenberg [Mon, 27 May 2024 21:12:27 +0000 (23:12 +0200)]
openssl/gnutls: rectify the TLS version checks for QUIC
The versions check wrongly complained and return error if the *minimum*
version was set to something less than 1.3. QUIC is always TLS 1.3, but
that means minimum 1.2 is still fine to ask for.
This also renames the local variable to make the mistake harder to make
in the future.
Stefan Eissing [Mon, 27 May 2024 09:51:03 +0000 (11:51 +0200)]
gnutls: improve TLS shutdown
local ftp upload tests sometimes failed with an invalid TLS record being
reported by gnutls. vsftp did log that the shutdown was not regarded as
clean, failing the control connection thereafter.
These changes make test_31_05 work reliable locally.
- on closing the SSL filter, shutdown READ *and* WRITE
- on closing, try a receive after shutdown is sent
- convert to DEBUGF to CURL_TRC_CF
Viktor Szakats [Mon, 20 May 2024 12:21:05 +0000 (14:21 +0200)]
build: untangle `CURLDEBUG` and `DEBUGBUILD` macros
`CURLDEBUG` is meant to enable memory tracking, but in a bunch of cases,
it was protecting debug features that were supposed to be guarded with
`DEBUGBUILD`.
Replace these uses with `DEBUGBUILD`.
This leaves `CURLDEBUG` uses solely for its intended purpose: to enable
the memory tracking debug feature.
Also:
- autotools: rely on `DEBUGBUILD` to enable `checksrc`.
Instead of `CURLDEBUG`, which worked in most cases because debug
builds enable `CURLDEBUG` by default, but it's not accurate.
- include `lib/easyif.h` instead of keeping a copy of a declaration.
- add CI test jobs for the build issues discovered.
Viktor Szakats [Sat, 18 May 2024 00:15:32 +0000 (02:15 +0200)]
cmake: always build unit tests with the `testdeps` target
Before this patch, the `testdeps` build target required `-DCURLDEBUG`
be set either via `ENABLE_DEBUG=ON` or `ENABLE_CURLDEBUG=ON` to build
the curl unit tests.
After fixing build issues in #13694, we can drop this requirement and
build unit tests unconditionally.
Depends-on: #13694
Depends-on: #13697 (fix unit test issue revealed by Old Linux CI job)
Follow-up to 39e7c22bb459c2e818f079984989a26a09741860 #11446
Closes #13698
Viktor Szakats [Mon, 27 May 2024 13:19:35 +0000 (15:19 +0200)]
CI: disable dependency tracking in most autotools builds
For better build performance. Dependency tracking causes a build
overhead while compiling to help a subsequent build, but in CI there is
never one and the extra work is discarded.
Viktor Szakats [Thu, 16 May 2024 11:49:22 +0000 (13:49 +0200)]
build: untangle `UNITTESTS` and `DEBUGBUILD` macros
- fix `DEBUGBUILD` guards that should be `UNITTESTS`, in libcurl code
used by unit tests.
- fix guards for libcurl functions used in unit tests only.
- sync `UNITTEST` attribute between declarations and definitions.
- drop `DEBUGBUILD` guard from test `unit2600`.
- fix guards for libcurl HSTS code used by both a unit test (`unit1660`)
and `test0446`.
- update an existing AppVeyor CI job to test the issues fixed.
This fixes building tests with `CURLDEBUG` enabled but `DEBUGBUILD`
disabled. This can happen when building tests with CMake with
`ENABLE_DEBUG=ON` in Release config, or with `ENABLE_CURLDEBUG=ON`
and _without_ `ENABLE_DEBUG=ON`. Possibly also with autotools
when using `--enable-curldebug` without `--enable-debug`.
Test results:
- before:
https://ci.appveyor.com/project/curlorg/curl/builds/49835609
https://ci.appveyor.com/project/curlorg/curl/builds/49898529/job/k8qpbs8idby70smw
https://github.com/curl/curl/actions/runs/9259078835/job/25470318167?pr=13798#step:13:821
- after: https://ci.appveyor.com/project/curlorg/curl/builds/49839255
(the two failures are unrelated, subject to PR #13705)
Ref: #13592 (issue discovery)
Ref: #13689 (CI testing this PR with `DEBUGBUILD`/`CURLDEBUG` combinations)
Closes #13694
Viktor Szakats [Mon, 27 May 2024 16:48:32 +0000 (18:48 +0200)]
CI: tidy up skipping tests build/run in Windows jobs
Simplify controlling whether to build and/run tests in a CI job.
Apply the TFLAGS='skipall' (do not build nor run tests) or
'skiprun' (build, but do not run) method already used with old-mingw-w64
and msvc jobs to existing Windows jobs in GHA and AppVeyor.
Also:
- add Cygwin/cmake test build and run steps while here.
- replace `DISABLED_TESTS` with `TFLAGS` in AppVeyor.
Viktor Szakats [Mon, 27 May 2024 12:33:54 +0000 (14:33 +0200)]
cmake: allow `ENABLE_CURLDEBUG=OFF` with `ENABLE_DEBUG=ON`
Before this patch, `ENABLE_CURLDEBUG` (memory tracking) was
unconditionally enabled when `ENABLE_DEBUGBUILD` was set. This made
testing some build configurations complicated. To fix it, this patch
makes `ENABLE_CURLDEBUG` to receive the value of `ENABLE_DEBUG` by
default, while allowing free override by the user.
This allows to use the config:
`ENABLE_DEBUGBUILD=ON ENABLE_CURLDEBUG=OFF`
to enable debug features, without also enabling memory tracking.
This is important because some other build methods allow to set one of
these features but not the other. This patch allows to test any
combination with CMake.
This makes it unnecessary to use the workaround of passing
`-DDEBUGBUILD` via `CMAKE_C_FLAGS`. Which has the disadvantage that our
CMake logic cannot easily detect it, e.g. for disabling symbol hiding on
Windows for `ENABLE_DEBUG`/`DEBUGBUILD` builds.
Viktor Szakats [Sat, 11 May 2024 00:44:10 +0000 (02:44 +0200)]
cmake: `ENABLE_DEBUG=ON` to always set `-DDEBUGBUILD`
Before this patch `ENABLE_DEBUG=ON` always enabled the TrackMemory
(aka `ENABLE_CURLDEBUG=ON`) feature, but required the `Debug` CMake
configration to actually enable curl debug features
(aka `-DDEBUGBUILD`).
Curl debug features do not require compiling with C debug options. This
also made enabling debug features unintuitive and complicated to use.
Due to other issues (subject to PR #13694) it also caused an error in
default (and `Release`/`MinSizeRel`/`RelWithDebInfo`) configs, when
building the `testdeps` target:
```
ld: CMakeFiles/unit1395.dir/unit1395.c.o: in function `test':
unit1395.c:(.text+0x1a0): undefined reference to `dedotdotify'
```
Ref: https://github.com/curl/curl/actions/runs/9037287098/job/24835990826#step:3:2483
Fix it by always defining `DEBUGBUILD` when setting `ENABLE_DEBUG=ON`.
Decoupling this option from the selected CMake configuration.
Note that after this patch `ENABLE_DEBUG=ON` unconditionally enables
curl debug features. These features are insecure and unsuited for
production. Make sure to omit this option when building for production
in default, `Release` (and other not-`Debug`) modes.
Also delete a workaround no longer necessary in GHA CI jobs.
Viktor Szakats [Fri, 24 May 2024 15:53:27 +0000 (17:53 +0200)]
GHA: add three MSVC jobs
Continuing the theme, add 3 MSVC jobs with tests, matching
configurations used on AppVeyor. MSVC versions are identical:
19.39.33523.0 + Windows SDK 10.0.22621.0.
Also enable websockets, and build examples. Tests are run in parallel
(`-j14`), with improved performance.
Job performance:
```
AppVeyor GHA
w/examples
-------- ----------
CMake, VS2022, Debug, x64, Schannel, Static, Unicode 38m 4s 11m57s
CMake, VS2022, Debug, x64, no SSL, Static 35m15s 12m 6s
CMake, VS2022, Debug, x64, no SSL, Static, HTTP only 25m25s 10m36s
```
Based on these runs:
https://ci.appveyor.com/project/curlorg/curl/builds/49884748
https://github.com/curl/curl/actions/runs/9229448468
Viktor Szakats [Thu, 23 May 2024 12:54:49 +0000 (14:54 +0200)]
GHA: add three old (gcc 6, 7, 9) mingw-w64 jobs
Re-implement old mingw-w64 jobs in GHA. This allows to use the latest
Windows runners, replacing Windows Server 2012 R2 (gcc 6) and Windows
Server 2016 (gcc 7, 9) with Windows Server 2022.
GHA runners are also significantly faster, and allow running tests in
parallel (`-j14`). It also offloads 3 more long-running jobs from
AppVeyor CI.
These jobs download (then cache) the mingw-w64 packages from their
original location, which allows flexibility in choosing which versions
and flavours (win32/POSIX, SEH/DWARF, 64/32-bit) we want to test in CI.
The new jobs use these distros:
- https://sourceforge.net/projects/mingw-w64/files/Toolchains%20targetting%20Win64/Personal%20Builds/mingw-builds/ (for gcc 7, same as on AppVeyor)
- https://sourceforge.net/projects/mingw-w64/files/Toolchains%20targetting%20Win32/Personal%20Builds/mingw-builds/ (for gcc 6, same as on AppVeyor)
- https://winlibs.com/ (for gcc 9)
I matched existing AppVeyor job configs, with these differences:
- gcc 6.4.0 instead of 6.3.0.
(same distro as on AppVeyor, but the latest bugfix release)
- gcc 9.5.0 instead of 9.1.0 and a different (but compatible) binary distro.
(in AppVeyor this relies on an old MSYS2 pre-installed on the runner)
- using win32 builds instead of posix for gcc 6.4.0 and 7.3.0.
- websockets enabled.
- always build examples.
- always build tests (this wasn't done for 6.4.0 with AppVeyor CI).
I did not replicate existing test exclusions, and oddly enough the few
failures (so far) were different from MSYS2 jobs and also from their
AppVeyor CI counterparts.
Also:
- delete redundant (default) `-u` option from `cygpath` calls.
- allow matrix options to override default ones in CMake.
- detect and use Windows-supplied curl for `TFLAGS` `-ac` option.
(it's available in modern runners.)
- delete the 3 AppVeyor CI jobs now replicated in GHA.
- appveyor: prefer `SYSTEMROOT` over `WINDIR`.
- tidy-up quotes.
Viktor Szakats [Fri, 24 May 2024 16:38:32 +0000 (18:38 +0200)]
examples: fix compiling with MSVC
- `websocket.c`: use `Sleep()` on Windows.
`sleep()` and `unistd.h` are not available in MSVC.
- `http2-upload.c`: use local `gettimeofday()` implementation when
compiled with MSVC.
(Alternate solution is to disable the trace function for MSVC.)
Public domain code copied and adapted from libssh2:
https://github.com/libssh2/libssh2/blob/e973493f992313b3be73f51d3f7ca6d52e288558/src/misc.c#L719-L743
- silence compiler warning for deprecated `inet_addr()`.
Also drop duplicate winsock2 include.
```
curl\docs\examples\externalsocket.c(125,32): error C2220: the following warning is treated as an error [curl\bld\docs\examples\curl-example-externalsocket.vcxproj]
curl\docs\examples\externalsocket.c(125,32): warning C4996: 'inet_addr': Use inet_pton() or InetPton() instead or define _WINSOCK_DEPRECATED_NO_WARNINGS to disable deprecated API warnings [curl\bld\docs\examples\curl-example-e
```
Ref: https://github.com/curl/curl/actions/runs/9227337318/job/25389073450#step:4:95
- silence an MSVC compiler warning. This is in conflict with `checksrc`
rules, so silence the rule in favour of the warning-free C syntax.
```
curl\docs\examples\multi-legacy.c(152,1): error C2220: the following warning is treated as an error [curl\bld\docs\examples\curl-example-multi-legacy.vcxproj]
curl\docs\examples\multi-legacy.c(152,1): warning C4706: assignment within conditional expression [curl\bld\docs\examples\curl-example-multi-legacy.vcxproj]
```
Ref: https://github.com/curl/curl/actions/runs/9227337318/job/25389073450#step:4:226
- do not use `sys/time.h` and `unistd.h` in Windows builds.
Some of these includes look unnecessary. Subject to another PR.
Stefan Eissing [Thu, 11 Apr 2024 10:34:40 +0000 (12:34 +0200)]
multi: add multi->proto_hash, a key-value store for protocol data
- add `Curl_hash_add2()` that passes a destructor function for
the element added. Call element destructor instead of hash
destructor if present.
- multi: add `proto_hash` for protocol related information,
remove `struct multi_ssl_backend_data`.
- openssl: use multi->proto_hash to keep x509 shared store
- schannel: use multi->proto_hash to keep x509 shared store
- vtls: remove Curl_free_multi_ssl_backend_data() and its
equivalents in the TLS backends
Stefan Eissing [Thu, 23 May 2024 10:21:46 +0000 (12:21 +0200)]
http: write last header line late
- HEADERFUNCTIONS might inspect response properties like
CURLINFO_CONTENT_LENGTH_DOWNLOAD_T on seeing the last header line. If
the line is being written before this is initialized, values are not
available.
- write the last header line late when analyzing a HTTP response so that
all information is available at the time of the writing.
- add test1485 to verify that CURLINFO_CONTENT_LENGTH_DOWNLOAD_T works
on seeing the last header.
Fixes #13752 Reported-by: Harry Sintonen
Closes #13757
Dan Fandrich [Fri, 24 May 2024 21:46:19 +0000 (14:46 -0700)]
tests: use exec when spawning nghttpx
This stops keeping perl and shell processes around that are no longer
needed, plus it eliminates an unneeded shell message when the server is
later terminated.
Viktor Szakats [Thu, 23 May 2024 12:54:49 +0000 (14:54 +0200)]
GHA: stop deleting curl in non-native workflows
We do it in Cirrus CI, but for some platforms it's not possible to
delete it and tests work anyway.
The test runner also runs `../src/curl` by default, which is always the
one freshly built. The runner may also need the system curl to talk to
APIs when needed.
Also:
- stop setting `CURL` env. This isn't picked up by the runners,
and works out of the box anyway.
- quote an option just in case.
Jay Satiro [Fri, 17 May 2024 08:01:35 +0000 (04:01 -0400)]
openssl: stop duplicate ssl key logging for legacy OpenSSL
- Don't call the keylog function if it has already logged the key.
For old OpenSSL versions and its forks that do not have support for
OpenSSL's keylog callback, libcurl has its own legacy key logging
function that logs the TLS 1.2 (and earlier) key (client random + master
key) on a single line.
Prior to this change, since e7de80e8 (precedes 8.8.0), the legacy key
logging function could write the same key line more than once (usually
twice) due to some incorrect logic.
Viktor Szakats [Thu, 23 May 2024 14:44:40 +0000 (16:44 +0200)]
cmake: fix brotli lib order
Fix root cause that caused missing symbols when linking brotli
statically with e.g. binutils `ld` (and any other "picky" linker,
or "traditional" linker as CMake now calls them).
Also drop existing workaround that added brotli libs twice to the lib
list.
```
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(decode.c.obj):decode.c:(.text$ProcessCommands[ProcessCommands]+0xbb5): undefined reference to `BrotliTransformDictionaryWord'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(decode.c.obj):decode.c:(.text$SafeProcessCommands[SafeProcessCommands]+0xe8a): undefined reference to `BrotliTransformDictionaryWord'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(decode.c.obj):decode.c:(.rdata$.refptr._kBrotliContextLookupTable[.refptr._kBrotliContextLookupTable]+0x0): undefined reference to `_kBrotliContextLookupTable'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(decode.c.obj):decode.c:(.rdata$.refptr._kBrotliPrefixCodeRanges[.refptr._kBrotliPrefixCodeRanges]+0x0): undefined reference to `_kBrotliPrefixCodeRanges'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(state.c.obj):state.c:(.text$BrotliDecoderStateInit[BrotliDecoderStateInit]+0x21): undefined reference to `BrotliDefaultAllocFunc'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(state.c.obj):state.c:(.text$BrotliDecoderStateInit[BrotliDecoderStateInit]+0x2f): undefined reference to `BrotliDefaultFreeFunc'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(state.c.obj):state.c:(.text$BrotliDecoderStateInit[BrotliDecoderStateInit]+0x10e): undefined reference to `BrotliSharedDictionaryCreateInstance'
x86_64-w64-mingw32-ld: .../curl/brotli/_bld/usr/lib/libbrotlidec.a(state.c.obj):state.c:(.text$BrotliDecoderStateCleanup[BrotliDecoderStateCleanup]+0xf4): undefined reference to `BrotliSharedDictionaryDestroyInstance'
collect2: error: ld returned 1 exit status
```
Breakage reproducible with curl-for-win config "`win-gcc`" and deleting
the `LDFLAGS+=' -Wl,--start-group'` line from its `curl.sh` script.
(Above line still required for some non-brotli cases, e.g. libssh2 and
zlib.)
Assisted-by: Kai Pastor
Ref: https://github.com/curl/curl/pull/10857#discussion_r1611714989
Follow-up to 1e3319a167d2f32d295603167486e9e88af9bb4e #10857
Closes #13761
Jay Satiro [Wed, 22 May 2024 06:46:13 +0000 (02:46 -0400)]
winbuild: remove outdated WIN32 defines
- Remove all instances in the makefile of compiler option /DWIN32.
This is a follow-up to e9a7d4a1 which replaced all defined(WIN32) checks
with defined(_WIN32) in the codebase, since only the latter is
automatically defined by all compilers for Windows builds.
Bug: https://github.com/curl/curl/pull/13739#issuecomment-2123937859 Reported-by: Viktor Szakats
Closes https://github.com/curl/curl/pull/13742
Pavel P [Tue, 21 May 2024 23:23:34 +0000 (01:23 +0200)]
asyn-thread: avoid using GetAddrInfoExW with impersonation
Multiple reports suggest that GetAddrInfoExW fails when impersonation is
used. This PR checks if thread is impersonating and avoids using
GetAddrInfoExW api.
Stefan Eissing [Wed, 22 May 2024 14:52:16 +0000 (16:52 +0200)]
transfer: conn close on paused upload
- 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.
Stefan Eissing [Wed, 22 May 2024 08:35:51 +0000 (10:35 +0200)]
mbedtls: send close-notify on close
- send the TLS close notify message when cloding down
the mbedtls connection filter
- this is a "least" effort version and, as other TLS filters,
is lacking a graceful send/receive/timeout for a really
clean shutdown.
Stefan Eissing [Tue, 21 May 2024 10:17:33 +0000 (12:17 +0200)]
pytest: add ftp upload tests
- 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
Jay Satiro [Wed, 22 May 2024 05:39:44 +0000 (01:39 -0400)]
winbuild: fix PE version info debug flag
- Only set PE file flag VS_FF_DEBUG if curl.exe and libcurl.dll were
built with winbuild option DEBUG=yes which builds with debug info.
VS_FF_DEBUG is a PE flag (Portable Executable file flag - dll, exe, etc)
that indicates the file contains or was built with debug info.
Prior to this change when winbuild was used to build curl, curl.exe
and libcurl.dll always had VS_FF_DEBUG set, regardless of build option
DEBUG=yes/no, due to some bad logic.
Stefan Eissing [Tue, 21 May 2024 09:21:14 +0000 (11:21 +0200)]
http: HEAD response body tolerance
- as reported in #13725, some servers wrongly send body bytes in
responses to a HEAD request. This used to be tolerated in curl
8.4 and before and leads to failed transfers in newer versions.
- restore previous behaviour for HTTP/1.1 and HTTP/2:
* 1.1: do not add 'Transfer-Encoding' writers from HEAD
responses. RFC 9112 says they do not apply.
* 2: when the transfer expects 'no_body', to not report stream
resets as error when all response headers have been received.