To simplify and to remove an exception where `errno` was reused to
return a socket error codes on Windows.
The error was used by one call site (`sockaddr2string()` in
`cf-socket.c`), but it was in practice always propagated as
`SOCKEAFNOSUPPORT` to callers.
Also:
- cf-socket: update 3 error messages to show `CURLcode` accordingly.
- if2ip: handle `curlx_inet_ntop()` error in `Curl_if2ip()`.
- dnsd: display `CURLcode` on two errors.
Follow-up to 39dec13ec0#22170Closes#22229
- drop redundant casts for `CURL_UNCONST()` pointers (Windows).
- GHA/linux: delete redundant/dupe `-g` C flag in memory sanitizer job.
Spotted-by: Daniel Stenberg
- examples/http2-upload: drop local macro.
- examples/http2-upload: drop leading underscore from union name.
- examples/http2-upload: drop interim macro.
- tests/server/util: reapply patch lost in a rebase.
- sync `main()` declarations across the codebase.
Closes#22559
curl already built like this in UWP mode; add a macro to allow building
regular Win32 targets too, meaning without the hidden GUI window, window
handler, and thus without support for shutting down `servers.exe`
gracefully. To test the server in CI as a plain console app and remove
one component/variable from the system, in an effort chasing flaky runs.
The graceful option was used by `taskkill` (without `-f`) in runtests,
and it has been disabled (in favor of always `-f` (forced) kill), in an
attempt to avoid taskkill hanging. Those resolved, but general CI
flakiness remained. We see `taskkill` invoked seldom per job, and in
most of these cases it doesn't find the PID requested. It means the
graceful option is not actually used since 2025-04, and rather rarely
before that. Also just today, the callback and main loop received a few
updates to avoid potentially unsafe calls and interaction with the ctrl
handler.
Not expecting anything breakthrough here. Also graceful shutdown
is a useful feature to cleanup properly and to be in sync with other
platforms.
Also: GHA/windows: enable in Windows CI jobs.
Ref: https://learn.microsoft.com/windows/console/registering-a-control-handler-function#listen-with-hidden-window-example
Follow-up to 7db9947fcf#22045
Follow-up to 9ea48811fe#22487
Follow-up to b11e0026f7#22495
Follow-up to 208b87744e#21039
Follow-up to f450f3801b#19897
Follow-up to 2701ac6a4d#19421
Follow-up to 4e203f65a1#17054Closes#22496
- decode results when individual requests are done
- makes happy eyeballing start asap
- remove doh_resp structures as no longer needed
- remove CURL_DNS_TYPE_NS, CURL_DNS_TYPE_CNAME and
CURL_DNS_TYPE_DNAME from DoH
- DoH: do not set PIPEWAIT and SSL OPTS when url starts with http:
- mark Doh master handle as dirty after every sub-request, not only the last
- Doh: start probe on AAAA before A (was the other way).
cf-dns: set EXPIRE_HAPPY_EYEBALLS timer when waiting 50ms on AAAA
result or progress will not be triggered in time.
Add debug env var CURL_DBG_HE_AAAA_AWAIT_MS to override the
default 50ms on waiting for the AAAA result to arrive.
test2100: set CURL_DBG_HE_AAAA_AWAIT_MS to 60 seconds to
provide enough time for slow CI runs to sent all DoH requests.
closes#22514
dnsd now opens UDP+TCP sockets and accepts http: DoH requests to obtain
the same, configured answers (records, delays, error codes) as over UDP.
DoH: use `async->queries_ongoing` like all other resolvers instead of
the internal `pending` counter. Fixes waiting for results.
Tests: in pytest, parameterize dnsd tests to use both DNS and DoH.
Closes#22506
On Windows, the init code calls `SetConsoleCtrlHandler()`, and before
this patch also set handlers for all Unixy signals. Of these, `SIGBREAK`
(used on Windows-only), `SIGINT`, `SIGABRT` and `SIGTERM` were also
setting up a `SetConsoleCtrlHandler()`, in addition to the call made
directly. (The rest, `SIGHUP`, `SIGPIPE`, `SIGALRM` are either missing
the macros, or ignored by `signal()` on Windows.)
As per WINE sources, `SetConsolCtrlHandler(<h>, TRUE)` calls are
additive, which means the test server set up two console ctrl handlers.
Then the ctrl handler set directly (`ctrl_event_handler()`), was
triggering the other signal handler via `raise()`, for the 'initiate
exit' logic, which in turn triggered exiting a wait within `select_ws()`
and other loops. The Windows window handler also made use of the
`SIGTERM` event to initiate exit via `raise()` and the second signal
handler.
To simplify, de-duplicate the ctrl handlers by dropping `signal()` calls
and keeping the direct Win32 call with `ctrl_event_handler()` doing all
the signal handling on Windows. Break out the 'initiate exit' logic into
a function and call it from both Unix and Windows signal/ctrl/window
handlers. Also drop calling `raise()` on exit, because it's a no-op
without a `signal()` pair.
Also:
- drop logging the actual ctrl type number, replace with just logging
whether we handled the event, in `ctrl_event_handler()`. To avoid
using non-signal-safe functions (e.g. `fprintf()`) from the handler.
- also replace `logmsg()` with `WriteFile()` to prevent regressions.
Ref: #22045
- replace `logmsg()` with `WriteFile()` in `main_window_proc()`.
- fix to forward ctrl handling to the OS in the rare case of failed
`exit_event` initialization on startup. To swap a possible hang
(within `WaitForMultipleObjectsEx()`) with an ungraceful shutdown.
- add support for an 'exit message' string, set by signal/ctrl handlers,
and log it on app exit. To avoid the need to deal with logging within
the handlers, yet have a static trace message about the event.
Complementing the already logged signal number.
- drop stderr trace message from `exit_signal_handler()` in favor of an
exit message. runtests triggers it frequantly, which added much noise
to stderr. As a bonus, this also allows dropping the compiler warning
suppression.
Reported-by: Stefan Eissing
Bug: https://github.com/curl/curl/pull/22487#issuecomment-5204092974
Follow-up to 3aae64e4fb#22507
Refs:
https://learn.microsoft.com/windows/console/setconsolectrlhandlerhttps://learn.microsoft.com/windows/console/registering-a-control-handler-functionhttps://learn.microsoft.com/cpp/c-runtime-library/reference/raisehttps://learn.microsoft.com/cpp/c-runtime-library/reference/signalhttps://gitlab.winehq.org/wine/wine/-/blob/wine-11.14/dlls/kernelbase/console.c#L1517-1526d6e817a4cc/misc/signal.cpp (L286-L348)
Follow-up to fe28fcf04c7dc8a981fa0e058776c0#5260Closes#22487
In year 2020 the Unixy signal handler received a `logmsg()` call to log
the signal number, but at the same time it already saved it to a global
variable and logged it on exit, meaning this extra `logmsg()` was
redundant. Because `logmsg()` is not signal-safe, this call was replaced
in 2025 with signal-safe logging, but without the signal number, while
also adding complexity, spent on trying to open the log file and handle
errors. All for nothing, because the signal number was logged all along.
This patch removes all this, and simplifies it to a single, signal-safe
`write()` to STDERR to say that the signal handler triggered. This is
also non-critical, but may help debugging.
Also: point the POSIX documentation to the 2004 revision, which has a
shorter list of safe functions. (was: 2018)
Follow-up to e95f509c66#16852
Follow-up to 9869f6dc5a#5218Closes#22507
To share more common code between servers, and to log the same set of
information for all of them.
Also:
- move server unix socket to global variable.
Follow-up to d1eca3861c#22501
Follow-up to 1637bbc9ce#22498Closes#22503
To sync with the rest of code which already handled `got_exit_signal`
and did the restoration on exit. Also syncing with rest of servers.
Also: turn off `keep_sigalrm` to sync with most servers.
Closes#22499
Replace `ipv_inuse` and `use_ipv6` with `socket_type` and
`socket_domain` (where missing) to avoid dupliicate globals with
overlapping purposes. The replacement variables also support Unix
sockets.
Also:
- simplify/reduce IPv6 guards.
- socksd: fix to reset `socket_domain` for `--ipv4` option.
Closes#22498
Before this patch modern systems used `sigaction()` and `SA_RESTART` to
install signal handlers, but the signal handler function itself still
made a call to the legacy `signal()` function to re-register itself
before returning.
Re-registering the handler is not necessary with `sigaction()`. It's
also undesired to use the legacy API when the modern one is available.
Fix by guarding off this call in builds that support the modern API.
Follow-up to 3fb6e5a010#6529
Follow-up to 18cbb4d7d6Closes#22497
Historically servers used the deprecated `siginterrupt()` function to
configure restart behavior on specific signals. It accepts a flag, where
1 means to remove the `SA_RESTART` option, and 0 means to enable it.
In year 2021 3fb6e5a010 introduced the
modern alternative to the codebase, replacing `siginterrupt()` with
`sigaction()`. After this patch, supporting, modern, systems reacted on
the same flag, but, by accident, set the `SA_RESTART` bit when flag is
1, and did not set it when 0. This reversed the previous behavior, and
the one still used on the `siginterrupt()` legacy codepath.
Fix it by revesring the `SA_RESTART` logic for the `sigaction()`
codepath, syncing it with the pre-existing behavior.
I find it odd this did not cause any perceivable issue for 5 years, even
though it's the active one in most Unix envs.
Spotted by GitHub Code Quality, though suggesting to fix
`siginterrupt()` calls. But looking into the history, those were correct
all along.
Refs:
https://pubs.opengroup.org/onlinepubs/9699919799/functions/siginterrupt.htmlhttps://pubs.opengroup.org/onlinepubs/9699919799/functions/sigaction.htmlhttps://www.man7.org/linux/man-pages/man3/siginterrupt.3.htmlhttps://www.man7.org/linux/man-pages/man2/sigaction.2.html
Follow-up to 3fb6e5a010#6529Closes#22037
"The check `i > 256` permits `i == 256` to pass through. When `i` is
then cast to `uint8_t` in `blob_add(b, (uint8_t)i)`, the value wraps to
0, silently encoding a zero-length ALPN entry instead of rejecting it.
The condition should be `i > 255` (or equivalently `i >= 256`) to
correctly reject any length that does not fit in a single byte."
Reported by GitHub Code Quality
Follow-up to 86f1e5b3f6#21299Closes#22420
- make `const` data `static`, where missing and possible.
- replace `strlen()` on literal or const strings with `sizeof()`.
While the latter is optimized by popular C compiler, e.g. MSVC only
does it with `/O2`.
- replace magic numbers with `sizeof()`, where missing.
- introduce `CURL_CSTRLEN()` macro for `sizeof(char[]) - 1`.
- use `CURL_CSTRLEN()` macro.
- move `const` before integer types, where missing.
- replace `char *var` with `var[]`, where missing and possible.
- use double const, where missing.
`static const char *` -> `static const char * const`.
- lib1514: constify pointers.
- unit3205: drop redundant cast, avoid another one.
- unit1666: map `OID()` macro to identical `STRCONST()`.
Closes#22406
Also repurpose existing build-time feature checks into unit test 1961,
to verify.
Prior to this patch these functions were auto-detected with both
autotools and cmake. In case of autotools there was an extra
verification phase ensuring the functions work as expected. This step
required running the function, thus was limited to non-cross-builds. For
cross-builds and CMake it always used the system implementation if
present. On Windows it always used the local implementation, because
availability/use is complicated there.
After this patch all platforms, always use the local implementation,
which is known to be accurate. This makes curl behave more consistently,
and simplifies the build process, a fixes cross-builds and CMake
auto-detection differences.
Also:
- test1960: enable unconditionally.
- checksrc: disallow globally, allowlist in `block_ip.c` example.
- dnsd: verify ntop result for NULL before passing to printf.
Ref: https://github.com/curl/curl/pull/22137#issuecomment-4797440983
Ref: #22137
Ref: 8537a5b0bc#16577Closes#22170
- cmp-pkg-config.sh: replace `-r -f` with `-rf` to match rest of repo.
- configure.ac: add double quotes for robustness (not a bug).
- curl-openssl.m4: merge nested `if`s.
- CurlTests.c: drop `!= 0`, also to sync with m4.
- CurlTests.c: replace `example.com` with `localhost` in
`gethostbyname()` feature test code. (compile-only, not a bug)
- GHA/http3-linux: drop literal `true` from bool expression.
- lib650: drop redundant `&`.
- move variable/call to left-hand side of equality checks, where
missing.
- perl: detach `<`/`>` from filename in `open()`, where missing.
- schannel: apply two nit fixes lost in rebase.
- scripts/verify-release: drop redundant double quotes.
- scripts/verify-release: exit with error code on error.
- synctime: replace magic numbers with `sizeof()`.
- telnet: add missing parentheses to macro value.
- tests/Makefile.am: use single quotes.
- tool_operate: drop redundant `break` after `return` in VMS code.
- unit2413: drop unused NULL pointer + free call.
- unit2413: fix duplicate test case name.
- urlapi: drop redundant parentheses.
- urlapi: drop `CURL_UNCONST()` that became redundant.
Closes#22186
- GHA/windows: mention `IgnoreStandardErrorWarningFormat=true`
in comment.
- cmake: fix MIT/GNU GSS order in messages.
- drop some exclamation marks from messages.
- drop redundant ending newlines from messages.
- fold/unfold where possible.
- fix indent, whitespace, typos and other nits.
Closes#22185
Remove the suppression of connmon after the first disconnect event. The
connmon flag is set per-test via the 'connection-monitor' server command
and should remain active for the lifetime of that test's connections to
properly detect connection reuse failures.
The suppression was introduced in 510fdad to work around a connection
reuse regression that has since been resolved by the credentials
refactoring in 8f71d0f. With the underlying issue fixed, restoring full
disconnect logging strengthens tests like 338 to catch future
regressions in connection reuse logic.
Fixes#22158Closes#22159
In `bind_unix_socket()`, before retrying `bind()`.
This patch uses `ENOENT`. This was last time in source between
d25b050379 (2018) and
dffd996e3b (2023), and also earlier. Also
defined by supported Windows envs. Seems safe to use.
Reported-by CodeQL
Follow-up to 99fb36797a
Cherry-picked from #22010Closes#22028
The reason to use them seems to be that just (void) before a function
call is not enough to silence compiler warnings when return codes are
ignored and -Werror=unused-result is used.
While (void)! apparently works to silence those warnings, it is just too
weird and surprising to readers to use.
It is rather a reason to reconsider the usefulness of the warning.
Closes#22023
On Windows there is no `lstat()`, which was later substituted with
normal `stat()`, but on Windows `S_IFSOCK` is never defined, which meant
the output of stat was not actually used, reducing this to checking for
the presence of the file, and bailing out without retry if missing.
Follow-up to 30e491e5c9#7034
Follow-up to 99fb36797a
Cherry-picked from #22010Closes#22021
- `N byte` -> `N-byte` or `N bytes`.
- INTERNALS.md: language tweaks.
- schannel: language tweak in comment/error message.
- socks_gssapi, socks_sspi: simplify composing an error message.
(at a cost of 8 extra constant string bytes.)
- m4/curl-compilers.m4: fix typo in link (in comment).
- contrithanks.sh: fix indent, drop stray `;` terminator.
- lib, src, tests: drop/fix a bunch of badwords.
- fix typos in comments.
- fix indent, stray spaces.
Some of these spotted by GitHub Code Quality and Copilot
Closes#22009
- sws: fix socket error code in `select()` failure message.
Spotted by Copilot
Bug: https://github.com/curl/curl/pull/21998#discussion_r3409469444
- sws: do not call `SOCKERRNO` twice on error.
- dnsd: do not call `SOCKERRNO` twice on error.
- dnsd: replace `goto` with `while()` to sync with rest of code.
- dnsd: `sendto()` fail message fixes:
- replace `int` cast with `%zu` mask.
- drop redundant newline.
- show socket error string like rest of code.
- report not-fully-sent error separately from socket errors.
Closes#22007
Check for `!got_exit_signal` as part of the `while()` expression,
instead of doing it after calling `curlx_wait_ms()`. To simplify and
improve consistency with rest of code.
Follow-up to 0653fa107f
Follow-up to 123c92c904Closes#22006
To contain the logic of checking for both `EWOULDBLOCK` and/or `EAGAIN`
depending on platform/availability. Also to avoid checking for both if
they mapp to the same value, and to avoid PP guards around use.
This also ensures `EAGAIN` is consistently not checked on Windows, where
headers defined it, but `SOCKERRNO` never returns it, because curl maps
it to `WSAGetLastError()`.
If they map to the same value, checking them both in an `if` expression
trips GCC warning `-Wlogical-op` (the same way it triggers duplicate
case value error in `switch`).
Also:
- replace two `switch()` statements with the new macro.
- tests/server/sws: make two outliers use the new macro that were only
checking for `EWOULDBLOCK` before this patch, in `connect_to()`.
- move variables to the left-side of expressions, where missing.
- rustls: use a variant of this macro that uses raw `EWOULDBLOCK`.
Tried tracing it back to the origins, but I couldn't figure out if
this is working as expected on all supported Windows versions in
Rust. It seems to be using `GetLastError()`, according to
https://docs.rs/system_error/0.2.0/system_error/, which would be
probably incorrect.
Notes:
- it's probably a good idea to assign `SOCKERRNO` to a variable before
passing it to this macro.
Cherry-picked from #21893Closes#21992
Fix build issues:
- src: adjust `toolx_ftruncate()`.
- libtests/cli_ftp_upload: make `struct timeval` initialization portable.
- libtests/lib1960: do unconst in local `inet_pton()` macro.
- tests/server/dnsd: make it stub instead of failing the build.
- tests/server: make them link AmiSSL for `SocketBase`.
Also:
- bump AmiSSL to the latest release.
- add download hash checks and toolchain cache.
- sync restored code with local updates made since last year.
It takes 30 seconds per job, 1 minute in total.
Thanks-to: Bebbo
Thanks-to: Darren Banfi
Ref: https://github.com/curl/curl/pull/21989#issuecomment-4691380623
Follow-up to 7a14898264#17486Closes#21993
Adjust code to avoid `-Wformat-signedness` warnings, while making sure
that enums are always cast to a known type when passing them to `printf`
functions, to support compilers and compiler settings where enums are
not default-size signed ints.
- cast integers printed as hex to `unsigned`. (63 times, 20 of them in
`mbedtls.c`)
- cast misc enums to `int` for printing. (31 times)
- cast `CURL_LOCK_DATA_*` enums to `int`. (4 times)
- cast `CURL_FORMADD_*` enums to `int`. (13 times)
- cast `CURLSHE_*` enums to `int`. (3 times)
- cast `CURLUE_*` enums to `int`. (33 times)
- cast `CURLMSG_*` enums to `int`. (6 times)
- cast `CURLE_*` enums to `int`. (~380 times)
- unit1675: fix mask.
Follow-up to 7c34365cce#21879
Ref: #18343 (initial attempt)
Closes#20848
