I added it in commit 57231d5286 ("src: notify V8 profiler when we're
idle") from October 2013 as a stop-gap measure to measure CPU time
rather than wall clock time, otherwise processes that spend a lot
of time sleeping in system calls give a false impression of being
very busy.
That fix is not without drawbacks because the idle flag is set before
libuv makes I/O callbacks and cleared again after. I/O callbacks can
result into calls into JS code and executing JS code is as non-idle
as you can get.
In commit 96ffcb9a21 ("src: reduce cpu profiler overhead") from January
2015, I made Node.js block off the SIGPROF signal that V8's CPU profiler
uses before Node.js goes to sleep. The goal of that commit is to reduce
the overhead from EINTR system call wakeups but it also has the pleasant
side effect of fixing what the idle notifier tried to fix.
This commit removes the idle notifier and turns the JS process object
methods into no-ops.
Fixes: https://github.com/nodejs/node/issues/19009
Refs: https://github.com/nodejs/node/pull/33138
PR-URL: https://github.com/nodejs/node/pull/34010
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Instances coming from different VM contexts don't pass `instanceof`
type checks because each context has its own copy of the built-in
globals.
After review of the relevant code it seems like it should be safe to
relax the type check and that is what this commit does: `wasi.start()`
now accepts any input that walks and quacks like a WebAssembly.Instance
or WebAssembly.Memory instance.
Fixes: https://github.com/nodejs/node/issues/33415
PR-URL: https://github.com/nodejs/node/pull/33431
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Disable the following compiler warning:
warning: visibility attribute not supported in this
configuration; ignored [-Wattributes]
This is gcc complaining about `__attribute((visibility("default"))`
in static library builds. Legitimate but harmless (and uninteresting)
and it drowns out more relevant warnings.
PR-URL: https://github.com/nodejs/node/pull/32419
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Richard Lau <riclau@uk.ibm.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
We might not have sufficient privileges to signal the child process
so don't make assumptions about the return value of `uv_process_kill()`.
Example:
node -e 'child_process.spawnSync("sudo", ["ls"], { maxBuffer: 1 })'
No test because:
1. The test needs to run as root (can't invoke sudo), and
2. The parent needs to drop privileges but can't, because
then the child process won't have sufficient privileges.
Fixes: https://github.com/nodejs/node/issues/31747
PR-URL: https://github.com/nodejs/node/pull/31768
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: David Carlier <devnexen@gmail.com>
Make the cipher/decipher/hash/hmac update() methods ignore the input
encoding when the input is a buffer.
This is the documented behavior but some inputs were rejected, notably
when the specified encoding is 'hex' and the buffer has an odd length
(because a _string_ with an odd length is never a valid hex string.)
The sign/verify update() methods work okay because they use different
validation logic.
Fixes: https://github.com/nodejs/node/issues/31751
PR-URL: https://github.com/nodejs/node/pull/31766
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Richard Lau <riclau@uk.ibm.com>
Reviewed-By: Luigi Pinca <luigipinca@gmail.com>
Reviewed-By: Tobias Nießen <tniessen@tnie.de>
I found it exceedingly hard to figure out if there is a race condition
where one thread reads the inspector agent's HostPort's properties while
another modifies them concurrently.
I think the answer is "no, there isn't" but with this commit use sites
are forced to unwrap the object (and acquire the mutex in the process),
making it a great deal easier to reason about correctness.
PR-URL: https://github.com/nodejs/node/pull/31717
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Tobias Nießen <tniessen@tnie.de>
Reviewed-By: David Carlier <devnexen@gmail.com>
The getter passed a stack-allocated, fixed-size buffer to
uv_get_process_title() but neglected to check the return value.
When the total length of the command line arguments exceeds the size of
the buffer, libuv returns UV_ENOBUFS and doesn't modify the contents of
the buffer. The getter then proceeded to return whatever garbage was on
the stack at the time of the call, quite possibly reading beyond the
end of the buffer.
Add a GetProcessTitle() helper that reads the process title into a
dynamically allocated buffer that is resized when necessary.
Fixes: https://github.com/nodejs/node/issues/31631
PR-URL: https://github.com/nodejs/node/pull/31633
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Gus Caplan <me@gus.host>
Reviewed-By: David Carlier <devnexen@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
`__executable_start` is provided by GNU's and LLVM's default linker
scripts, obviating the need to plug in a custom linker script.
The problem with our bespoke linker script is that it works with ld.bfd
but not ld.gold and cannot easily be ported because the latter linker
doesn't understand the `INSERT BEFORE` directive.
The /proc/self/maps scanner is updated to account for the fact that
there are a number of sections between `&__executable_start` and
the start of the .text section.
Fortunately, those sections are all mapped into the same memory segment
so we only need to look at the next line to find the start of our text
segment.
Fixes: https://github.com/nodejs/node/issues/31520
PR-URL: https://github.com/nodejs/node/pull/31547
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Gus Caplan <me@gus.host>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: David Carlier <devnexen@gmail.com>
This API method was introduced in commit 90ae4bd0c9 ("src: add
InitializeV8Platform function") from July 2018 but wasn't properly
exported and therefore not usable on Windows or with shared library
builds.
The motivation from the commit log is mainly about making it easier
to wire up the cctests and there are better ways to do that.
Refs: https://github.com/nodejs/node/pull/31217
PR-URL: https://github.com/nodejs/node/pull/31245
Reviewed-By: David Carlier <devnexen@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
This requires minor changes to src/env.cc to deal with
`node::tracing::AgentWriterHandle::GetTracingController()` now possibly
returning a nullptr, because the cctest doesn't set one.
It seems plausible to me that embedders won't set one either so that
seems like an okay change to make. It avoids embedders having to track
down nullptr segfaults.
PR-URL: https://github.com/nodejs/node/pull/31245
Refs: https://github.com/nodejs/node/pull/31217
Reviewed-By: David Carlier <devnexen@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Make `distutils.version.StrictVersion` available as a helper to
gyp expressions so they can do proper version checks and update
the gyp files accordingly.
Caveat emptor: `StrictVersion` does *not* like empty strings so
this commit adds truthiness guards. The helper could deal with
those but I felt it better to make it explicit.
Fixes: https://github.com/nodejs/node/issues/29927
PR-URL: https://github.com/nodejs/node/pull/29931
Reviewed-By: Michaël Zasso <targos@protonmail.com>
Reviewed-By: Richard Lau <riclau@uk.ibm.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
`node --prof-process` on macOS calls out to nm(1) to look up C++
symbols. If Xcode hasn't been properly installed or its license
hasn't been accepted yet, it prints out an error and exits.
Before this commit, that error was swallowed and the output of
the tick processor was not showing the C++ entry points.
This commit detects that error message and turns it into an
exception. No regression test because this particular condition
is hard to test for without going to extreme lengths to mock
the output of nm.
Fixes: https://github.com/nodejs/node/issues/29804
PR-URL: https://github.com/nodejs/node/pull/29830
Reviewed-By: David Carlier <devnexen@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: Minwoo Jung <minwoo@nodesource.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Document the `promiseResolve` option to `async_hooks.createHook()`.
This seems to have been overlooked in commit b605b15346 ("async_hooks:
support promise resolve hook") from September 2017.
It is documented elsewhere in the async_hooks API documentation, except
where you actually pass it in.
PR-URL: https://github.com/nodejs/node/pull/29405
Reviewed-By: David Carlier <devnexen@gmail.com>
Reviewed-By: Michaël Zasso <targos@protonmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Luigi Pinca <luigipinca@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Before this commit it was using a tagged union to store the one-byte and
two-byte pointers.
From a `sizeof(UnionBytes)` perspective that makes no difference - there
is a hole between the tag and the union - and it makes the code just a
little harder to reason about, IMO.
PR-URL: https://github.com/nodejs/node/pull/29116
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
Add BSD-x86 to the list of arches and regenerate everything.
Everything in deps/openssl/config/archs/BSD-x86 is auto-generated,
everything else is manual edits to include the right files in the
right places.
I cheated a little: I didn't check in changes to files for other arches
because they contained mostly churn (updated buildstamps and such.)
Fixes: https://github.com/nodejs/node/issues/20326
PR-URL: https://github.com/nodejs/node/pull/28806
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Luigi Pinca <luigipinca@gmail.com>
This commit also adds error handling. A THP-enabled build terminated
with an out-of-memory error on a system without /proc because it cast
the -1 from readlink() to size_t (i.e. ULONG_MAX) and then tried to
allocate a string of that size.
PR-URL: https://github.com/nodejs/node/pull/28333
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Some pty tests persistently hung on the AIX CI buildbots. Fix that by
adding a helper script that properly sets up the pty before spawning
the script under test.
On investigation I discovered that the test runner hung when it tried
to close the slave pty's file descriptor, probably due to a bug in
AIX's pty implementation. I could reproduce it with a short C program.
The test runner also leaked file descriptors to the child process.
I couldn't convince python's `subprocess.Popen()` to do what I wanted
it to do so I opted to move the logic to a helper script that can do
fork/setsid/etc. without having to worry about stomping on state in
tools/test.py.
In the process I also uncovered some bugs in the pty module of the
python distro that ships with macOS 10.14, leading me to reimplement
a sizable chunk of the functionality of that module.
And last but not least, of course there are differences between ptys
on different platforms and the helper script has to paper over that.
Of course.
Really, this commit took me longer to put together than I care to admit.
Caveat emptor: this commit takes the hacky ^D feeding to the slave out
of tools/test.py and puts it in the *.in input files. You can also feed
other control characters to tests, like ^C or ^Z, simply by inserting
them into the corresponding input file. I think that's nice.
Fixes: https://github.com/nodejs/build/issues/1820
Fixes: https://github.com/nodejs/node/issues/28489
PR-URL: https://github.com/nodejs/node/pull/28600
Reviewed-By: Richard Lau <riclau@uk.ibm.com>
Reviewed-By: Sam Roberts <vieuxtech@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
PTHREAD_STACK_MIN is 2 KB with musl, which is too small to safely
receive signals. PTHREAD_STACK_MIN + MINSIGSTKSZ is 8 KB on arm64,
which is the musl architecture with the biggest MINSIGSTKSZ so let's
use that as a lower bound and let's quadruple it just in case.
PR-URL: https://github.com/nodejs/node/pull/27855
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Eugene Ostroukhov <eostroukhov@google.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Rich Trott <rtrott@gmail.com>
It was reported that parallel builds on Windows sometimes error because
of missing intermediate files.
On closer inspection I noticed that some files are copied from src/ to
the intermediate build directory in a way where they don't participate
in dependency resolution. Put another way, the build system doesn't
know to wait for the copy to complete because we don't tell it to.
Fix that by not copying around files but instead making the script that
processes them a little smarter about where to find them and where to
store the results.
PR-URL: https://github.com/nodejs/node/pull/27026
Fixes: https://github.com/nodejs/node/issues/27025
Reviewed-By: Eugene Ostroukhov <eostroukhov@google.com>
Reviewed-By: Richard Lau <riclau@uk.ibm.com>
Implementing the methods out-of-line (i.e., not inline) means we can fix
bugs and have already compiled add-ons pick up the fixes automatically,
something that doesn't work when the methods are inline because then
they get compiled into the add-on instead of the node binary.
PR-URL: https://github.com/nodejs/node/pull/26348
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Refael Ackermann <refack@gmail.com>
Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com>
Before this commit, Node.js left it up to the system resolver or c-ares.
Leaving it to the system resolver introduces platform differences
because:
* some support IDNA 2008
* some only IDNA 2003 (glibc until 2.28), and
* some don't support IDNA at all (musl libc)
c-ares doesn't support IDNA either although curl does, by virtue of
linking against libidn2. Upgrading from libidn1 to libidn2 in order
to get proper IDNA 2008 support was the fix for curl's CVE-2016-8625.
libidn2 is not an option (incompatible license) but ICU has an IDNA API
and we already use that in one place. For non-ICU builds, we fall back
to the bundled punycode.js that also supports IDNA 2008.
Fixes: https://github.com/nodejs-private/security/issues/97
Fixes: https://github.com/nodejs/node/issues/25558
PR-URL: https://github.com/nodejs/node/pull/25679
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Tiancheng "Timothy" Gu <timothygu99@gmail.com>
Ben Noordhuis