Replace `write_reqs_pending` (unsigned int counter) in
`uv_stream_connection_fields` with `write_queue` (struct uv__queue),
mirroring the Unix stream implementation. Add `struct uv__queue queue`
to `UV_WRITE_PRIVATE_FIELDS` so each `uv_write_t` can be linked.
This enables iterating over in-flight write requests individually, which
is used in `uv__tcp_try_cancel_reqs` to replace the blanket `CancelIo`
call with per-request `CancelIoEx` calls (resolving the TODO comment
that anticipated this change).
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove the uv_fs_poll_t handle and all associated implementation, tests,
and APIs. Users should migrate to managing uv_fs_stat calls themselves
with their own uv_timer. See #4543 for explanation.
Closes: #4543Closes: #4942
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Named pipes do not actually support `shutdown`, and this hack to fake it
was unsound. Return `UV_ENOTSOCK` from `uv_shutdown` immediately for
`UV_NAMED_PIPE` handles, matching the behavior on unix where `shutdown`
on a pipe fd fails with `ENOTSOCK` (as oppposed to a unix domain socket
which is also a `uv_pipe_t`, but does supposed `shutdown`).
Fixes tests to work correctly whether or not shutdown works.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
uv_spawn currently unsets all supplementary groups if the caller sets
either a gid or uid to run the process under. This commit adds support
for specifying the supplementary groups to set instead.
Co-authored-by: Ben Noordhuis <info@bnoordhuis.nl>
Co-authored-by: Joe Dollard <jdollard@gmail.com>
Signed-off-by: bryopsida <8363252+bryopsida@users.noreply.github.com>
Add a version of uv_udp_try_send that can send multiple datagrams.
Uses sendmmsg(2) on platforms that support it (Linux, FreeBSD, macOS),
falls back to a regular sendmsg(2) loop elsewhere.
This work was sponsored by ISC, the Internet Systems Consortium.
Shuffle around and DRY the sendmsg logic in preparation for
uv_udp_try_send2(). NFC barring bugs.
This work was sponsored by ISC, the Internet Systems Consortium.
io_uring support was default-disabled because of numerous kernel bugs
but those are all in the sqpoll (file i/o) parts of io_uring.
Batching of epoll_ctl calls through io_uring works fine, is a nice
optimization, and is therefore unconditionally enabled again.
The UV_USE_IO_URING environment variable now only affects sqpoll, and
only when the UV_LOOP_ENABLE_IO_URING_SQPOLL event loop flag is set.
Fixes: https://github.com/libuv/libuv/issues/4616
Libuv looks for "Processor" in /proc/cpuinfo but it's been reported
that on at least some Raspberry Pi models, it's called "model name".
Look for both.
Fixes: https://github.com/nodejs/node/issues/56105
`uv_thread_setname()` sets the name of the current thread. Different
platforms define different limits on the max number of characters
a thread name can be: Linux, IBMi (16), macOS (64), Windows (32767),
and NetBSD (32), etc. `uv_thread_setname()` will truncate it in case
`name` is larger than the limit of the platform.
`uv_thread_getname()` gets the name of the thread specified by `tid`.
The thread name is copied into the buffer pointed to by `name`. The
`size` parameter specifies the size of the buffer pointed to by `name`.
The buffer should be large enough to hold the name of the thread plus
the trailing NUL, or it will be truncated to fit.
This commit introduces the `uv_thread_detach` for thread detaching,
allowing threads to be detached state on both UNIX and Windows platforms.
Signed-off-by: Juan José Arboleda <soyjuanarbol@gmail.com>
For any API that takes a buffer and size pointer, check both pointers
and the pointed-to size and return UV_EINVAL in case of error.
Example:
```
int uv_foo(char* buffer, size_t* size) {
if (buffer == NULL || size == NULL || *size == 0)
return UV_EINVAL;
...
}
```
In order to "peek" the necessary size for dynamic allocation, the
following pattern can be used:
```
char *buf;
char scratch[1];
size_t len = sizeof(scratch);
int r;
r = uv_foo(scratch, &len);
assert(r == UV_ENOBUFS);
buf = malloc(len);
r = uv_foo(buf, &len);
...
```
The compile-time detection check from commit 7b75935 ("kqueue: use
EVFILT_USER for async if available") was not being used, breaking
numerous operating systems. This commit hopefully unbreaks them.
Fixes; https://github.com/libuv/libuv/issues/4608
Signed-off-by: Jeffrey H. Johnson <trnsz@pobox.com>
uv_wtf8_length_as_utf16() checks if codepoints are > 0xFFFF (to see if
it should be encoded as a surrogate pair), therefore uv_wtf8_to_utf16()
should too. Instead it checked > 0x1000. Harmonize the checks.
Fixes: https://github.com/nodejs/node/issues/55914
Establishes a user event for kqueue to eliminate the overhead
of the pipe and the system call read(2) per wakeup event.
Relands commit 27134547ff using VSCode merge, since it shows the
conflict is just on the order of #ifdef calls.
Co-authored-by: Andy Pan <panjf2000@gmail.com>
Add UV_PROCESS_WINDOWS_USE_PARENT_ERROR_MODE to allow the user to
propagate user-controlled process-wide SetErrorMode to child processes
(for cmake). Default behavior is unchanged.
It seemed incorrect to map a segfault to EACCES, since posix would typically
map this to EFAULT. The ERROR_BUFFER_OVERFLOW is literally "the filename is too
long", and is not typically an invalid parameter in posix.
Test originally added in #1060 to test the API, not the value.
This reverts commit e5cb1d3d3d.
Reason: bisecting says it breaks dnstap.
Also revert commit 27134547ff ("kqueue: use EVFILT_USER for async if
available") because otherwise the first commit doesn't revert cleanly,
with enough conflicts in src/unix/async.c that I'm not comfortable
fixing those up manually.
Fixes: https://github.com/libuv/libuv/issues/4584
Make sure `UV__SOLARIS_11_4` is not set for `smartOS`/`illumOS`. In our
codebase is used only twice:
- Detect correct implementation of `SO_REUSEPORT`, which is not even
implemented on `illumOS`.
- Detect the time unit used for the TCP keepalive options. If set to
`0`, which was the case for `illumOS`, it chose milliseconds, which is
not correct for `illumOS` either as it uses seconds.
The implementation of IPC pipe in libuv on Windows does not properly support
async reading. This means we cannot set the more parameter without likely
causing hangs. Sorry this is yet another followup to #4511.
Fixes#4548
After commit 18266a6969, it changed to return `ENOTDIR`, which makes it
consistent with other platforms but it also can be considered a breaking
change.
Route ftruncate() system calls through io_uring instead of the thread
pool when the kernel is new enough to support it (linux >= 6.9).
This commit harmonizes how libuv checks if the kernel is new enough.
Some ops were checking against `uv__kernel_version()` directly while
others stored the result of the version check as a feature flag.
Because the kernel version is cached, and because it is more direct
than a feature flag, I opted for the former approach.
Introduced in Linux 6.6, it tells the kernel to omit the sqarray from
the ring buffer.
Libuv initalizes the array once to an identity mapping and then forgets
about it, so not only does it save a little memory (ca. 1 KiB per ring)
but it also makes things more efficient kernel-side because it removes
a level of indirection.
