* Update the user timing implementation to conform to
User Timing Level 3.
* Reimplement user timing and timerify with pure JavaScript
implementations
* Simplify the C++ implementation for gc and http2 perf
* Runtime deprecate additional perf entry properties
in favor of the standard detail argument
* Disable the `buffered` option on PerformanceObserver,
all entries are queued and dispatched on setImmediate.
Only entries with active observers are buffered.
* This does remove the user timing and timerify
trace events. Because the trace_events are still
considered experimental, those can be removed without
a deprecation cycle. They are removed to improve
performance and reduce complexity.
Old: `perf_hooks/usertiming.js n=100000: 92,378.01249733355`
New: perf_hooks/usertiming.js n=100000: 270,393.5280638482`
PR-URL: https://github.com/nodejs/node/pull/37136
Refs: https://github.com/nodejs/diagnostics/issues/464
Reviewed-By: Antoine du Hamel <duhamelantoine1995@gmail.com>
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Michaël Zasso <targos@protonmail.com>
Create weak `WriteWrap` and `ShutdownWrap` objects, and when
referencing them in C++ is necessary, use `BaseObjectPtr<>`
instead of plain pointers to keep these objects from being
garbage-collected.
This solves issues that arise when the underlying `StreamBase`
instance is weak, but the `WriteWrap` or `ShutdownWrap` instances
are not; in that case, they would otherwise potentially stick
around in memory after the stream that they originally belong
to is long gone.
It probably makes sense to use `BaseObjectptr<>` more extensively
in `StreamBase` in the long run as well.
PR-URL: https://github.com/nodejs/node/pull/35488
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
* cleanup constants in http2 binding
The error constants were just doing some weird things. Cleanup
and improve maintainability.
* simplify settings to reduce duplicate code
* improve style consistency and correctness
Use snake_case for getters and setters at c++ level,
avoid unnecessary use of enums,
use consistent style for exported vs. internal constants,
avoid unnecessary memory info reporting,
use setters/getters for flags_ for improved code readability
* make EmitStatistics function private
* un-nest Http2Settings and Http2Ping
* refactoring and cleanup of Http2Settings and Http2Ping
* avoid ** syntax for readability
The **session and **stream syntax for getting the underlying
nghttp2 pointers is not ideal for readability
* use const references for Http2Priority
* remove unnecessary GetStream function
* refactor Http2Scope to use BaseObjectPtr
* move utility function to anonymous namespace
* refactor and simplify Origins
* Use an AllocatedBuffer instead of MaybeStackBuffer
* Use a const reference instead of pointer
* use BaseObjectPtr for Http2Streams map
* move MemoryInfo impl to cc
Signed-off-by: James M Snell <jasnell@gmail.com>
PR-URL: https://github.com/nodejs/node/pull/32884
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Franziska Hinkelmann <franziska.hinkelmann@gmail.com>
For http2 (and eventually QUIC) we have a struct that is backed
by a v8::BackingStore and exposed to the JavaScript side as an
ArrayBuffer and TypedArray. This is similar to AliasedBuffer
except that it is fronted by a struct on the C++ side.
```c++
struct foo {
uint32_t ex1;
uint32_t ex2;
};
AliasedStruct<foo> foo_;
foo_->ex1 = 1;
foo_->ex2 = 2;
foo_.GetArrayBuffer();
```
Signed-off-by: James M Snell <jasnell@gmail.com>
PR-URL: https://github.com/nodejs/node/pull/32778
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Gerhard Stöbich <deb2001-github@yahoo.de>
Fixes warning on clang 11:
In file included from ../../src/node_http2.cc:6:
../../src/node_http2.h:508:15: warning: anonymous non-C-compatible type given name for linkage purposes by typedef declaration; add a tag name here [-Wnon-c-typedef-for-linkage]
typedef struct {
^
SessionJSFields
../../src/node_http2.h:512:33: note: type is not C-compatible due to this default member initializer
uint32_t max_invalid_frames = 1000;
^~~~
../../src/node_http2.h:514:3: note: type is given name 'SessionJSFields' for linkage purposes by this typedef declaration
} SessionJSFields;
^
PR-URL: https://github.com/nodejs/node/pull/32134
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Tobias Nießen <tniessen@tnie.de>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: David Carlier <devnexen@gmail.com>
The nghttp2 and nghttp3 (used in the QUIC implementation) share nearly
identical structs for header handling. However, they differ enough that
they need to be handled slightly different in each case. This PR
includes some elements introduced in the QUIC PR separated out to
make them independently reviewable, and updates the http2 implementation
to use the shared utilities.
Signed-off-by: James M Snell <jasnell@gmail.com>
PR-URL: https://github.com/nodejs/node/pull/32069
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
It looks like it’s virtually impossible at this point to
create “fake” backing stores for objects that don’t fully
own their memory allocations, like the sub-field `js_fields_`
of `Http2Session`. In particular, it turns out that an
`ArrayBuffer` cannot always be easily separated from its
backing store in that situation through by detaching it.
This commit gives the JS-exposed parts of the class its own
memory allocation and its own backing store, simplifying the
code a bit and fixing flakiness coming from it, at the cost
of one additional layer of indirection when accessing the data.
Refs: https://github.com/nodejs/node/pull/30782
Fixes: https://github.com/nodejs/node/issues/31107
PR-URL: https://github.com/nodejs/node/pull/31648
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Denys Otrishko <shishugi@gmail.com>
`ShutdownWrap` instances are being used to carry context between the
start and the asynchronous end of shutting down the writable side of
a `StreamBase`. HTTP/2 streams always perform this action
synchronously, so no such object is necessary.
PR-URL: https://github.com/nodejs/node/pull/31283
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Yongsheng Zhang <zyszys98@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
This option is not useful in practice, as mentioned in comments and the
documentation, because the overhead of calling into JS makes it
unreasonably expensive.
PR-URL: https://github.com/nodejs/node/pull/29144
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
If we are waiting for the ability to send more output, we should not
process more input. This commit a) makes us send output earlier,
during processing of input, if we accumulate a lot and b) allows
interrupting the call into nghttp2 that processes input data
and resuming it at a later time, if we do find ourselves in a position
where we are waiting to be able to send more output.
This is part of mitigating CVE-2019-9511/CVE-2019-9517.
PR-URL: https://github.com/nodejs/node/pull/29122
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
If a write to the underlying socket finishes asynchronously, that
means that we cannot write any more data at that point without waiting
for it to finish. If this happens, we should also not be producing any
more input.
This is part of mitigating CVE-2019-9511/CVE-2019-9517.
PR-URL: https://github.com/nodejs/node/pull/29122
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Limit the number of invalid input frames, as they may be pointing
towards a misbehaving peer. The limit is currently set to 1000 but
could be changed or made configurable.
This is intended to mitigate CVE-2019-9514.
PR-URL: https://github.com/nodejs/node/pull/29122
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Limit the number of streams that are rejected upon creation. Since
each such rejection is associated with an `NGHTTP2_ENHANCE_YOUR_CALM`
error that should tell the peer to not open any more streams,
continuing to open streams should be read as a sign of a misbehaving
peer. The limit is currently set to 100 but could be changed or made
configurable.
This is intended to mitigate CVE-2019-9514.
PR-URL: https://github.com/nodejs/node/pull/29122
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Lazily allocate `ArrayBuffer`s for the contents of DATA frames.
Creating `ArrayBuffer`s is, sadly, not a cheap operation with V8.
This is part of performance improvements to mitigate CVE-2019-9513.
Together with the previous commit, these changes improve throughput
in the adversarial case by about 100 %, and there is little more
that we can do besides artificially limiting the rate of incoming
metadata frames (i.e. after this patch, CPU usage is virtually
exclusively in libnghttp2).
PR-URL: https://github.com/nodejs/node/pull/29122
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
For some JS events, it only makes sense to call into JS when there
are listeners for the event in question.
The overhead is noticeable if a lot of these events are emitted during
the lifetime of a session. To reduce this overhead, keep track of
whether any/how many JS listeners are present, and if there are none,
skip calls into JS altogether.
This is part of performance improvements to mitigate CVE-2019-9513.
PR-URL: https://github.com/nodejs/node/pull/29122
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Have clearer ownership relations between the `Http2Ping`,
`Http2Settings` and `Http2Session` objects.
Ping and Settings objects are now owned by the `Http2Session`
instance, and deleted along with it, so neither type of object
refers to the session after it is gone.
In the case of `Http2Ping`s, that deletion is slightly delayed,
so we explicitly reset its `session_` property.
Fixes: https://github.com/nodejs/node/issues/28088
PR-URL: https://github.com/nodejs/node/pull/28150
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Ujjwal Sharma <usharma1998@gmail.com>
Reviewed-By: Rich Trott <rtrott@gmail.com>
Where appropriate, use a helper that wraps around
`ArrayBufferView::Buffer()` or `ArrayBufferView::CopyContents()`
rather than `Buffer::Data()`, as that may help to avoid materializing
the underlying `ArrayBuffer` when reading small typed arrays from C++.
This allows keeping the performance benefits of the faster creation of
heap-allocated small typed arrays in many cases.
PR-URL: https://github.com/nodejs/node/pull/26301
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Darshan Sen