Repository navigation
Commit 07d11eb
[Experimental] DirectTls transport implementation (#67912)
* Add experimental DirectTls transport for Kestrel
Adds Microsoft.AspNetCore.Server.Kestrel.Transport.DirectTls, an opt-in, per-endpoint transport that terminates TLS directly on the connection socket file descriptor via the runtime TlsContext/TlsSocketSession API, driven by an epoll event pump, bypassing SslStream. Primarily focused on Linux and requires OpenSSL on the host.
Public API (all [Experimental ASPNETCORE_DIRECTTLS_001]): UseDirectTlsTransport(), DirectTlsEndpoint, DirectTlsEndpointOptions, DirectTlsTransportOptions.
Features: per-SNI certificate selection, raw ClientHello bytes callback, ALPN/HTTP-2 (ITlsApplicationProtocolFeature set directly), mutual TLS client-certificate modes (DelayCertificate not supported), and the TLS handshake/connection features at parity with the SslStream path.
Includes a sample app and tests. Test certificates are generated locally via scripts/generate-certs.sh and git-ignored, never committed. Wired into AspNetCore.slnx and Kestrel.slnf.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* DirectTls: apply API-review-approved cosmetic changes
Applies the changes requested during the #67909 public API review:
1. Rename WebHostBuilderDirectTlsExtensions.UseDirectTlsTransport -> UseDirectTls.
2. Remove the (int port, DirectTlsEndpointOptions options) constructor on
DirectTlsEndpoint (the (IPAddress, int, options) overload covers it).
3. Remove HttpProtocols from DirectTlsEndpointOptions; the transport now sources
the HTTP protocols (ALPN) from ListenOptions.Protocols. The factory reads the
matching endpoint via KestrelServerOptions.GetListenOptions() (new one-line
InternalsVisibleTo from Kestrel.Core), keeping the approved options.Listen(endpoint)
surface unchanged.
4. Change TlsClientHelloBytesCallback from Action<ConnectionContext, ReadOnlySequence<byte>>
to a new DirectTlsClientHelloBytesCallback delegate taking ReadOnlySpan<byte>
(the ClientHello bytes are already contiguous in memory).
Updates PublicAPI.Unshipped.txt, the sample app, and tests. Builds clean;
19/19 DirectTls tests pass.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* DirectTls: drop custom ClientHello delegate; move protocols off factory DI
Address API-review follow-ups on the cosmetic-changes commit:
1. Remove the public DirectTlsClientHelloBytesCallback delegate. The
ClientHello-bytes callback is now typed System.Buffers.ReadOnlySpanAction<
byte, ConnectionContext> (a BCL type), so no new public delegate is added.
Action<ConnectionContext, ReadOnlySpan<byte>> is not expressible because
ReadOnlySpan<byte> is a ref struct and cannot be an Action<> type argument.
The callback argument order is now (bytes, connection).
2. Remove the IOptions<KestrelServerOptions> dependency from
DirectTlsTransportFactory. A new internal IPostConfigureOptions<
KestrelServerOptions> (DirectTlsEndpointProtocolsSetup) copies each
DirectTlsEndpoint ListenOptions.Protocols onto a new internal
DirectTlsEndpointOptions.HttpProtocols after all endpoints are configured.
The factory reads that internal field at bind time.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* DirectTls: revert ClientHello callback to Action<ConnectionContext, ReadOnlySequence<byte>>
A ReadOnlySpan<byte> is a poor fit for a stored callback: its lifetime is
undetermined and it cannot be a type argument to Action<> (ref struct).
Use ReadOnlySequence<byte> instead, which is a normal struct and works with a
plain Action<> in the natural (connection, bytes) argument order. No custom
public delegate is needed.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* DirectTls: move DirectTlsEndpoint(+Options) into Kestrel.Core; socket transport refuses it
Move DirectTlsEndpoint and DirectTlsEndpointOptions from the DirectTls transport
assembly into Kestrel.Core (namespace unchanged, so the approved public API names
are identical). This makes the endpoint a first-class Kestrel concept that the
default socket transport can reference cheaply.
SocketTransportFactory.CanBind now returns false for DirectTlsEndpoint. Because
DirectTlsEndpoint derives from IPEndPoint, the socket transport would otherwise
accept it and could bind it as a plaintext socket if it were tried before the
DirectTls transport. Combined with DirectTlsTransportFactory.CanBind (which
returns true ONLY for DirectTlsEndpoint), this guarantees - regardless of
transport registration order - that a DirectTls endpoint is served only by the
DirectTls transport, and no other endpoint is ever served by it.
Transport.Sockets now references Kestrel.Core (acyclic: Core does not reference
Sockets). Added a Sockets.BindTests case locking in the exclusion.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Ship DirectTls transport as a standalone NuGet package
Convert Microsoft.AspNetCore.Server.Kestrel.Transport.DirectTls from a
shared-framework assembly to a standalone shipping NuGet package: remove
IsAspNetCoreApp (so it is no longer part of Microsoft.AspNetCore.App) and
set IsPackable=true.
The transport was never actually enumerated in the generated shared-framework
list (eng/SharedFramework.Local.props) or its baselines, and the sample/tests
already consumed it via ProjectReference, so the IsAspNetCoreApp flag was
effectively dead. Making it a package keeps this experimental transport opt-in
and easy to remove later.
The endpoint/options public types live in Kestrel.Core (shared framework); only
the transport implementation and UseDirectTls extension method ship in this
package. Its references to shared-framework assemblies (Kestrel.Core, etc.)
become a FrameworkReference to Microsoft.AspNetCore.App in the produced nupkg.
Verified: pack produces a nupkg with frameworkReference Microsoft.AspNetCore.App;
sample builds clean; 19/19 DirectTls tests pass.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Add package README and processor-based default WorkerCount for DirectTls
Two small refinements to the DirectTls transport package:
- Add a minimal package README (PackageReadmeFile) so the produced nupkg no
longer warns about a missing readme. Content is intentionally a single
sentence noting this is an experimental, Linux-only, fd-bound native OpenSSL
TLS transport.
- Give DirectTlsTransportOptions.WorkerCount a processor-based default instead
of a hard-coded 4, mirroring the sockets transport IOQueue heuristic (capped
at 16 for up to 32 processors, half the processor count beyond that). The
heuristic is exposed only as an internal static DefaultWorkerCount, so no new
public API is introduced.
Verified: pack produces a nupkg containing README.md with <readme> in the
nuspec; build clean; 19/19 DirectTls tests pass.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Remove package README from DirectTls transport
A package README is not required to ship the DirectTls NuGet package; drop the
README.md and its PackageReadmeFile/None wiring. The processor-based default
WorkerCount change from the previous commit is unaffected.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Fix DirectTls epoll P/Invokes missing SetLastError=true
The libc epoll P/Invokes in Interop/NativeTls.cs used [LibraryImport] without
SetLastError=true, which defaults to false. As a result every
Marshal.GetLastWin32Error() call in TlsEventPump read a stale/garbage errno
rather than the real one from the failed syscall.
The critical impact is in PumpLoop's epoll_wait EINTR handling: on a signal
interrupt epoll_wait returns -1 with errno==EINTR (4), and the loop is supposed
to `continue`. With a garbage errno the check almost never matched, so the pump
logged an error and broke out of the loop, permanently terminating that worker
thread and orphaning its connections. The other errno sites (exception messages
and warning logs) also reported garbage values.
Add SetLastError=true to epoll_create1, both epoll_ctl overloads, epoll_wait and
close so GetLastWin32Error returns the real errno. This matches how the runtime's
own Sockets epoll interop is declared.
Verified: build clean; 19/19 DirectTls tests pass.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* log error on epoll closure fail
* cleanup tlspumpeventpool ; add per-endpoint workercount
* fix renego read/write collapses
* cover connectionIoState with tests
* comment on dispose/shutdown race condition problem
* Fix DirectTls accept loop spin on listen socket teardown
Stop accepting on all pumps (de-register the listen fd from each pump's
epoll set and clear the cached _listenFd) before the listener closes the
listen socket, in both DisposeAsync and UnbindAsync. This prevents a pump
from calling Accept() on a closed fd - or misrouting a client fd that
reuses the closed listen fd's number into the accept path - and spinning.
AcceptConnections now guards on _running/_listenFd and breaks on any accept
error, returning to epoll_wait (level-triggered re-arm) rather than looping,
so a dead listen fd can never tight-spin. AcceptOne/ProcessAcceptedSocket
are split out as test seams; TlsEventPump is unsealed for the scripted fake.
Adds TlsEventPumpAcceptTests covering the accept-loop control flow and
StopAccepting without a live listen socket, epoll, or the pump thread.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Dispose DirectTls native TLS contexts on listener teardown
The per-listener bootstrap TlsContext and the per-SNI TlsContext cache each
acquire OpenSSL server credentials (a long-lived SSL_CTX with the loaded cert
chain and private key) and were never disposed, leaking for the listener's
lifetime.
Wrap them in ServerTlsContexts and dispose it from
DirectTlsConnectionListener.DisposeAsync, after the pump pool is stopped so no
in-flight handshake can be using a context. DisposeAsync is now idempotent.
UnbindAsync intentionally does not free the contexts (established connections
keep serving).
Not disposing established _connections sessions from TlsEventPump.Dispose is
intentional: every entry is owned by a DirectTlsConnection that disposes it, so
a pump-side dispose would double-Shutdown the same native SSL* from two threads.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Dispose accepted client certificate on DirectTls connection teardown
On the fd fast path the runtime's TlsSession transfers ownership of the
accepted client leaf certificate to the caller: SetRemoteCertificateValidationResult(None)
promotes the pending cert into TlsSession._remoteCertificate, which
TlsSession.Dispose() deliberately never frees (unlike _externalPendingCert and
the intermediate certificates). DirectTlsConnection surfaced that leaf as
ITlsConnectionFeature.ClientCertificate but never disposed it, leaking one
native key handle per accepted mTLS connection.
Track the runtime-handed leaf in a private _ownedClientCertificate field (set
alongside the settable feature property so we dispose exactly the cert the
runtime gave us, never an app-reassigned object) and dispose it once in both
teardown paths, mirroring how SslStream disposes RemoteCertificate with the
stream. Disposal is idempotent, so this is safe even if a future runtime also
frees the leaf.
The two GetRemoteCertificate() calls in TlsEventPump return the same
session-owned instance, so there was no double allocation to fix there.
Extends the mTLS functional test to assert the app can read the client
certificate thumbprint during the request, guarding against disposing too early.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* defend for linux OS
* Trim unrelated AspNetCore.slnx reordering churn
Adding the DirectTls projects caused slnx tooling to also re-sort the
Components/Testing and ProjectTemplates folders. Restore their original
ordering so the solution diff shows only the three new DirectTls entries.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Disable AIA certificate downloads during client-cert chain build
chain.Build for client-certificate validation runs on the pump thread.
RevocationMode.NoCheck only suppresses CRL/OCSP; it does not stop the
chain engine from making synchronous AIA fetches for missing
intermediates. An attacker-supplied leaf with an unreachable AIA URL
could stall every connection owned by that pump (a DoS vector).
Set ChainPolicy.DisableCertificateDownloads = true, mirroring SslStream,
which sets the same flag on the server side (SslStream.Protocol.cs).
Legitimate clients send intermediates in the handshake, which are already
loaded into ChainPolicy.ExtraStore, so valid connections are unaffected.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Extract ClientCertificateValidator seam and add unit tests
Factor the completed-handshake client-certificate validation (chain build, policy, callback) out of TlsEventPump into an internal static ClientCertificateValidator so it can be unit tested without epoll or a live TLS session. Behavior is unchanged.
Also collapse the redundant post-validation GetRemoteCertificate() call: on the accept path the runtime promotes the pending cert into its canonical slot as the same instance, so the leaf surfaced to Kestrel reuses presentedCertificate instead of re-reading it.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* Add DirectTls HandshakeTimeout option and enforce it in the pump
Adds a public HandshakeTimeout property to DirectTlsEndpointOptions
(mirroring HttpsConnectionAdapterOptions: 10s default, rejects
non-positive values, normalizes Timeout.InfiniteTimeSpan to
TimeSpan.MaxValue) and enforces it in TlsEventPump: each handshaking
connection gets a deadline set at accept, and PumpLoop sweeps expired
handshakes every ~10ms via a bounded stackalloc buffer. This bounds
slow/stalled handshakes (e.g. slowloris ClientHello) that would
otherwise pin an fd and native TLS session indefinitely.
Also renames the pump test seams off the *ForTests suffix and groups
them into a single internal-for-testing block.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c655
* cleanup
* DirectTls: dispose-race fix, Linux-only test gating, backpressure buffer options
- Fix a lost read-cancellation race in DirectTlsConnection dispose so a
`Connection: close` response reliably reaches the client (ConnectionIoState,
TlsAwaitable).
- Gate the native-OpenSSL-dependent tests to Linux via [ConditionalFact] +
[OSSkipCondition] so they skip instead of failing on Windows/macOS.
- Add MaxReadBufferSize/MaxWriteBufferSize transport options plus the receive/
send plumbing that applies read- and write-side backpressure.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* DirectTls: de-flake e2e backpressure functional tests
Rework both e2e backpressure tests to poll-until-stalled with 100 MiB
payloads sent in 4 KiB chunks (Interlocked/Volatile progress counters +
Assert.InRange), mirroring the canonical MaxRequestBufferSizeTests.LargeUpload
idiom so they are robust on slow machines.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Fix DirectTls transport concurrency races (pump busy-spin, SSL op serialization, teardown)
- Pump busy-spin: the drop paths in TlsEventPump.HandleConnectionEvent now
de-register the fd from epoll (EPOLL_CTL_DEL) in addition to removing it from the
connection table. Established-connection events are level-triggered, so a
still-registered dead fd re-fired on every epoll_wait and was re-dropped at the
_connections lookup, tight-spinning the pump thread at 100% CPU and starving every
other connection it owned until disposal. Adds deterministic Linux-only unit tests.
- Serialize per-connection SSL operations: a per-connection _sslLock now guards the
native RawRead/RawWrite calls and the read/write state machine, so a completing read
on the pump thread and a concurrent app write can no longer interleave into the same
SSL session or drop a parked write's EPOLLOUT.
- Locked Dispose: shutdown (close_notify) is emitted under _sslLock so it cannot race a
concurrent SSL_read/SSL_write and corrupt the record stream.
- Abort unblocks both loops: Abort now cancels the reusable read/write awaitables and
the pending flush in addition to the pending read, so an app-initiated abort
mid-response tears the connection down immediately instead of waiting for the peer's
timeout (DirectTls closes the socket in DisposeAsync, not synchronously).
- Log swallowed AuthenticationExceptions at Debug instead of silently mapping to EOF.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* DirectTls: make _sslLock I/O transitions structurally synchronous
Follow-up hardening for the _sslLock serialization added in the previous commit.
- Inline the read/write initiation and the pump-side completion transitions
(ReadAsyncCore/WriteAsyncCore/OnReadableCore/OnWritableCore) directly into their
lock (_sslLock) blocks in ReadAsync/WriteAsync/OnReadable/OnWritable. The bodies are
moved verbatim (no behavior change), but keeping them lexically inside the lock makes
any future await on these paths a CS1996 compile error instead of silently releasing
the Monitor for the continuation - which would reintroduce the concurrent SSL_read/
SSL_write and epoll-interest races the lock exists to prevent. A plain helper call
could hide such an await behind the method boundary and defeat that guard.
- Rename the completion helpers TryCompleteRead/TryCompleteWrite ->
TryCompleteReadUnsynchronized/TryCompleteWriteUnsynchronized and document that the
caller must already hold _sslLock; they mutate the I/O state machine and issue the
absolute epoll_ctl (via UpdateEvents) without taking the lock themselves.
Validated: ConnectionIoStateTests 15/15 on Windows and the full DirectTls suite 66/66
on Linux, clean build (0 warnings).
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Harden DirectTls resource ownership and pump failures
Use Kestrel's memory pool factory and make each listener own its pool.
Transfer accepted socket handles explicitly and isolate pump-loop failures.
Require non-null logging dependencies and update lifecycle coverage.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Address DirectTls review feedback
Release early connections during pump shutdown and dispose converted client certificates.
Regenerate project reference lists and use resolved references from tests and the sample.
Correct ClientHello fd logging and add ownership regression coverage.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* sync interest for DestinationTooSmall / NeedMoreData
* safeguard connectioniostate disposal
* proper order in complete handshake
* guard against flushasync backpressure
* arch agnostic epoll native structs
* proper LIBC name
* verify version
* simplify
* simplify abort
* drop catching auth exception from session read/write
* ssl lock on the cancel
* handshake tracker + cap
* cache and dispose socket for IConnectionSocketFeature
* refactor to CTS-style pump stop
* react to failed epoll_ctl
* guard acceptedSocket
* fail connection if callback failed
* update interest on specific closed+armed EPOLLOUT case
* dispose allocated resources on transport.Bind() failures
* handle EPOLLERR/EPOLLHUP
* accept4 to simplify creation
* cap accepting connections in a single epoll wakeup iteration
* use min alloc buffer size for memory alloc
* practically select a suitable poll timeout
* dont catch object disposed while tls fatal error
* nit
* document blocking callback behavior
* use SOCK_CLOEXEC
* cap early
* critical epoll error leading to host shutdown
* add explicit lock on socket creation / disposal
* fire and forget conncetion disposal for closed-channel shutdown failure
* nit
---------
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Brennan Conroy <brecon@microsoft.com>
Copilot-Session: 3d22b61a-2dc4-4417-ab8a-54529a87c6551 parent efe577d commit 07d11eb
60 files changed
Lines changed: 7990 additions & 4 deletions
File tree
- eng
- src/Servers/Kestrel
- Core/src
- Transport.DirectTls
- src
- Connection
- Interop
- test
- Transport.Sockets/src
- samples/DirectTlsTransportApp
- scripts
- test/Sockets.BindTests
Some content is hidden
Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
1062 | 1062 | | |
1063 | 1063 | | |
1064 | 1064 | | |
| 1065 | + | |
1065 | 1066 | | |
1066 | 1067 | | |
1067 | 1068 | | |
| |||
1080 | 1081 | | |
1081 | 1082 | | |
1082 | 1083 | | |
| 1084 | + | |
| 1085 | + | |
| 1086 | + | |
| 1087 | + | |
1083 | 1088 | | |
1084 | 1089 | | |
1085 | 1090 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
51 | 51 | | |
52 | 52 | | |
53 | 53 | | |
| 54 | + | |
54 | 55 | | |
55 | 56 | | |
56 | 57 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
120 | 120 | | |
121 | 121 | | |
122 | 122 | | |
| 123 | + | |
123 | 124 | | |
124 | 125 | | |
125 | 126 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
41 | 41 | | |
42 | 42 | | |
43 | 43 | | |
| 44 | + | |
44 | 45 | | |
45 | 46 | | |
46 | 47 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
| 2 | + | |
| 3 | + | |
| 4 | + | |
| 5 | + | |
| 6 | + | |
| 7 | + | |
| 8 | + | |
| 9 | + | |
| 10 | + | |
| 11 | + | |
| 12 | + | |
| 13 | + | |
| 14 | + | |
| 15 | + | |
| 16 | + | |
| 17 | + | |
| 18 | + | |
| 19 | + | |
| 20 | + | |
| 21 | + | |
| 22 | + | |
| 23 | + | |
| 24 | + | |
| 25 | + | |
| 26 | + | |
| 27 | + | |
| 28 | + | |
| 29 | + | |
| 30 | + | |
| 31 | + | |
| 32 | + | |
| 33 | + | |
| 34 | + | |
| 35 | + | |
| 36 | + | |
| 37 | + | |
| 38 | + | |
| 39 | + | |
| 40 | + | |
| 41 | + | |
| 42 | + | |
| 43 | + | |
| 44 | + | |
| 45 | + | |
| 46 | + | |
| 47 | + | |
| 48 | + | |
| 49 | + | |
| 50 | + | |
| 51 | + | |
| 52 | + | |
| 53 | + | |
| 54 | + | |
| 55 | + | |
| 56 | + | |
| 57 | + | |
| 58 | + | |
| 59 | + | |
| 60 | + | |
| 61 | + | |
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
| 2 | + | |
| 3 | + | |
| 4 | + | |
| 5 | + | |
| 6 | + | |
| 7 | + | |
| 8 | + | |
| 9 | + | |
| 10 | + | |
| 11 | + | |
| 12 | + | |
| 13 | + | |
| 14 | + | |
| 15 | + | |
| 16 | + | |
| 17 | + | |
| 18 | + | |
| 19 | + | |
| 20 | + | |
| 21 | + | |
| 22 | + | |
| 23 | + | |
| 24 | + | |
| 25 | + | |
| 26 | + | |
| 27 | + | |
| 28 | + | |
| 29 | + | |
| 30 | + | |
| 31 | + | |
| 32 | + | |
| 33 | + | |
| 34 | + | |
| 35 | + | |
| 36 | + | |
| 37 | + | |
| 38 | + | |
| 39 | + | |
| 40 | + | |
| 41 | + | |
| 42 | + | |
| 43 | + | |
| 44 | + | |
| 45 | + | |
| 46 | + | |
| 47 | + | |
| 48 | + | |
| 49 | + | |
| 50 | + | |
| 51 | + | |
| 52 | + | |
| 53 | + | |
| 54 | + | |
| 55 | + | |
| 56 | + | |
| 57 | + | |
| 58 | + | |
| 59 | + | |
| 60 | + | |
| 61 | + | |
| 62 | + | |
| 63 | + | |
| 64 | + | |
| 65 | + | |
| 66 | + | |
| 67 | + | |
| 68 | + | |
| 69 | + | |
| 70 | + | |
| 71 | + | |
| 72 | + | |
| 73 | + | |
| 74 | + | |
| 75 | + | |
| 76 | + | |
| 77 | + | |
| 78 | + | |
| 79 | + | |
| 80 | + | |
| 81 | + | |
| 82 | + | |
| 83 | + | |
| 84 | + | |
| 85 | + | |
| 86 | + | |
| 87 | + | |
| 88 | + | |
| 89 | + | |
| 90 | + | |
| 91 | + | |
| 92 | + | |
| 93 | + | |
| 94 | + | |
| 95 | + | |
| 96 | + | |
| 97 | + | |
| 98 | + | |
| 99 | + | |
| 100 | + | |
| 101 | + | |
| 102 | + | |
| 103 | + | |
| 104 | + | |
| 105 | + | |
| 106 | + | |
| 107 | + | |
| 108 | + | |
| 109 | + | |
| 110 | + | |
| 111 | + | |
| 112 | + | |
| 113 | + | |
| 114 | + | |
| 115 | + | |
| 116 | + | |
| 117 | + | |
| 118 | + | |
| 119 | + | |
| 120 | + | |
| 121 | + | |
| 122 | + | |
| 123 | + | |
| 124 | + | |
| 125 | + | |
| 126 | + | |
| 127 | + | |
| 128 | + | |
| 129 | + | |
| 130 | + | |
| 131 | + | |
| 132 | + | |
| 133 | + | |
| 134 | + | |
| 135 | + | |
| 136 | + | |
| 137 | + | |
| 138 | + | |
| 139 | + | |
| 140 | + | |
| 141 | + | |
| 142 | + | |
| 143 | + | |
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
10 | 10 | | |
11 | 11 | | |
12 | 12 | | |
| 13 | + | |
13 | 14 | | |
14 | 15 | | |
15 | 16 | | |
| |||
74 | 75 | | |
75 | 76 | | |
76 | 77 | | |
| 78 | + | |
77 | 79 | | |
78 | 80 | | |
79 | 81 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
1 | 1 | | |
| 2 | + | |
| 3 | + | |
| 4 | + | |
| 5 | + | |
| 6 | + | |
| 7 | + | |
| 8 | + | |
| 9 | + | |
| 10 | + | |
| 11 | + | |
| 12 | + | |
| 13 | + | |
| 14 | + | |
| 15 | + | |
| 16 | + | |
| 17 | + | |
| 18 | + | |
| 19 | + | |
| 20 | + | |
| 21 | + | |
| 22 | + | |
| 23 | + | |
| 24 | + | |
2 | 25 | | |
0 commit comments