XiaoMo/frp
1
0
Fork 0
mirror of https://github.com/fatedier/frp.git synced 2026-01-11 22:23:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

update kcp-go package

Browse Source
This commit is contained in:
fatedier
2019-03-17 17:09:54 +08:00
parent 87a4de4370
commit fdcdccb0c2
122 changed files with 14490 additions and 2469 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vendor/github.com/fatedier/kcp-go/.travis.yml generated vendored
View File

@@ -1,6 +1,8 @@
language: go
go:
- 1.9
- 1.9.x
- 1.10.x
- 1.11.x
before_install:
- go get -t -v ./...

98
vendor/github.com/fatedier/kcp-go/README.md generated vendored
View File

@@ -20,24 +20,20 @@
**kcp-go** is a **Production-Grade Reliable-UDP** library for [golang](https://golang.org/).
It provides **fast, ordered and error-checked** delivery of streams over **UDP** packets, has been well tested with opensource project [kcptun](https://github.com/xtaci/kcptun). Millions of devices(from low-end MIPS routers to high-end servers) are running with **kcp-go** at present, including applications like **online games, live broadcasting, file synchronization and network acceleration**.
This library intents to provide a **smooth, resilient, ordered, error-checked and anonymous** delivery of streams over **UDP** packets, it has been battle-tested with opensource project [kcptun](https://github.com/xtaci/kcptun). Millions of devices(from low-end MIPS routers to high-end servers) have deployed **kcp-go** powered program in a variety of forms like **online games, live broadcasting, file synchronization and network acceleration**.
[Lastest Release](https://github.com/xtaci/kcp-go/releases)
## Features
1. Optimized for **Realtime Online Games, Audio/Video Streaming and Latency-Sensitive Distributed Consensus**.
1. Compatible with [skywind3000's](https://github.com/skywind3000) C version with language specific optimizations.
1. Designed for **Latency-sensitive** scenarios.
1. **Cache friendly** and **Memory optimized** design, offers extremely **High Performance** core.
1. Handles **>5K concurrent connections** on a single commodity server.
1. Compatible with [net.Conn](https://golang.org/pkg/net/#Conn) and [net.Listener](https://golang.org/pkg/net/#Listener), a drop-in replacement for [net.TCPConn](https://golang.org/pkg/net/#TCPConn).
1. [FEC(Forward Error Correction)](https://en.wikipedia.org/wiki/Forward_error_correction) Support with [Reed-Solomon Codes](https://en.wikipedia.org/wiki/Reed%E2%80%93Solomon_error_correction)
1. Packet level encryption support with [AES](https://en.wikipedia.org/wiki/Advanced_Encryption_Standard), [TEA](https://en.wikipedia.org/wiki/Tiny_Encryption_Algorithm), [3DES](https://en.wikipedia.org/wiki/Triple_DES), [Blowfish](https://en.wikipedia.org/wiki/Blowfish_(cipher)), [Cast5](https://en.wikipedia.org/wiki/CAST-128), [Salsa20]( https://en.wikipedia.org/wiki/Salsa20), etc. in [CFB](https://en.wikipedia.org/wiki/Block_cipher_mode_of_operation#Cipher_Feedback_.28CFB.29) mode.
1. **Fixed number of goroutines** created for the entire server application, minimized goroutine context switch.
## Conventions
Control messages like **SYN/FIN/RST** in TCP **are not defined** in KCP, you need some **keepalive/heartbeat mechanism** in the application-level. A real world example is to use some **multiplexing** protocol over session, such as [smux](https://github.com/xtaci/smux)(with embedded keepalive mechanism), see [kcptun](https://github.com/xtaci/kcptun) for example.
1. Packet level encryption support with [AES](https://en.wikipedia.org/wiki/Advanced_Encryption_Standard), [TEA](https://en.wikipedia.org/wiki/Tiny_Encryption_Algorithm), [3DES](https://en.wikipedia.org/wiki/Triple_DES), [Blowfish](https://en.wikipedia.org/wiki/Blowfish_(cipher)), [Cast5](https://en.wikipedia.org/wiki/CAST-128), [Salsa20]( https://en.wikipedia.org/wiki/Salsa20), etc. in [CFB](https://en.wikipedia.org/wiki/Block_cipher_mode_of_operation#Cipher_Feedback_.28CFB.29) mode, which generates completely anonymous packet.
1. Only **A fixed number of goroutines** will be created for the entire server application, costs in **context switch** between goroutines have been taken into consideration.
1. Compatible with [skywind3000's](https://github.com/skywind3000) C version with various improvements.
## Documentation
@@ -80,47 +76,59 @@ Server: [full demo](https://github.com/xtaci/kcptun/blob/master/server/main.go
lis, err := kcp.ListenWithOptions(":10000", nil, 10, 3)
```
## Performance
## Benchmark
```
Model Name: MacBook Pro
Model Identifier: MacBookPro12,1
Model Identifier: MacBookPro14,1
Processor Name: Intel Core i5
Processor Speed: 2.7 GHz
Processor Speed: 3.1 GHz
Number of Processors: 1
Total Number of Cores: 2
L2 Cache (per Core): 256 KB
L3 Cache: 3 MB
L3 Cache: 4 MB
Memory: 8 GB
```
```
$ go test -v -run=^$ -bench .
beginning tests, encryption:salsa20, fec:10/3
BenchmarkAES128-4 200000 8256 ns/op 363.33 MB/s 0 B/op 0 allocs/op
BenchmarkAES192-4 200000 9153 ns/op 327.74 MB/s 0 B/op 0 allocs/op
BenchmarkAES256-4 200000 10079 ns/op 297.64 MB/s 0 B/op 0 allocs/op
BenchmarkTEA-4 100000 18643 ns/op 160.91 MB/s 0 B/op 0 allocs/op
BenchmarkXOR-4 5000000 316 ns/op 9486.46 MB/s 0 B/op 0 allocs/op
BenchmarkBlowfish-4 50000 35643 ns/op 84.17 MB/s 0 B/op 0 allocs/op
BenchmarkNone-4 30000000 56.2 ns/op 53371.83 MB/s 0 B/op 0 allocs/op
BenchmarkCast5-4 30000 44744 ns/op 67.05 MB/s 0 B/op 0 allocs/op
Benchmark3DES-4 2000 639839 ns/op 4.69 MB/s 2 B/op 0 allocs/op
BenchmarkTwofish-4 30000 43368 ns/op 69.17 MB/s 0 B/op 0 allocs/op
BenchmarkXTEA-4 30000 57673 ns/op 52.02 MB/s 0 B/op 0 allocs/op
BenchmarkSalsa20-4 300000 3917 ns/op 765.80 MB/s 0 B/op 0 allocs/op
BenchmarkFlush-4 10000000 226 ns/op 0 B/op 0 allocs/op
BenchmarkEchoSpeed4K-4 5000 300030 ns/op 13.65 MB/s 5672 B/op 177 allocs/op
BenchmarkEchoSpeed64K-4 500 3202335 ns/op 20.47 MB/s 73295 B/op 2198 allocs/op
BenchmarkEchoSpeed512K-4 50 24926924 ns/op 21.03 MB/s 659339 B/op 17602 allocs/op
BenchmarkEchoSpeed1M-4 20 64857821 ns/op 16.17 MB/s 1772437 B/op 42869 allocs/op
BenchmarkSinkSpeed4K-4 30000 50230 ns/op 81.54 MB/s 2058 B/op 48 allocs/op
BenchmarkSinkSpeed64K-4 2000 648718 ns/op 101.02 MB/s 31165 B/op 687 allocs/op
BenchmarkSinkSpeed256K-4 300 4635905 ns/op 113.09 MB/s 286229 B/op 5516 allocs/op
BenchmarkSinkSpeed1M-4 200 9566933 ns/op 109.60 MB/s 463771 B/op 10701 allocs/op
goos: darwin
goarch: amd64
pkg: github.com/xtaci/kcp-go
BenchmarkSM4-4 50000 32180 ns/op 93.23 MB/s 0 B/op 0 allocs/op
BenchmarkAES128-4 500000 3285 ns/op 913.21 MB/s 0 B/op 0 allocs/op
BenchmarkAES192-4 300000 3623 ns/op 827.85 MB/s 0 B/op 0 allocs/op
BenchmarkAES256-4 300000 3874 ns/op 774.20 MB/s 0 B/op 0 allocs/op
BenchmarkTEA-4 100000 15384 ns/op 195.00 MB/s 0 B/op 0 allocs/op
BenchmarkXOR-4 20000000 89.9 ns/op 33372.00 MB/s 0 B/op 0 allocs/op
BenchmarkBlowfish-4 50000 26927 ns/op 111.41 MB/s 0 B/op 0 allocs/op
BenchmarkNone-4 30000000 45.7 ns/op 65597.94 MB/s 0 B/op 0 allocs/op
BenchmarkCast5-4 50000 34258 ns/op 87.57 MB/s 0 B/op 0 allocs/op
Benchmark3DES-4 10000 117149 ns/op 25.61 MB/s 0 B/op 0 allocs/op
BenchmarkTwofish-4 50000 33538 ns/op 89.45 MB/s 0 B/op 0 allocs/op
BenchmarkXTEA-4 30000 45666 ns/op 65.69 MB/s 0 B/op 0 allocs/op
BenchmarkSalsa20-4 500000 3308 ns/op 906.76 MB/s 0 B/op 0 allocs/op
BenchmarkCRC32-4 20000000 65.2 ns/op 15712.43 MB/s
BenchmarkCsprngSystem-4 1000000 1150 ns/op 13.91 MB/s
BenchmarkCsprngMD5-4 10000000 145 ns/op 110.26 MB/s
BenchmarkCsprngSHA1-4 10000000 158 ns/op 126.54 MB/s
BenchmarkCsprngNonceMD5-4 10000000 153 ns/op 104.22 MB/s
BenchmarkCsprngNonceAES128-4 100000000 19.1 ns/op 837.81 MB/s
BenchmarkFECDecode-4 1000000 1119 ns/op 1339.61 MB/s 1606 B/op 2 allocs/op
BenchmarkFECEncode-4 2000000 832 ns/op 1801.83 MB/s 17 B/op 0 allocs/op
BenchmarkFlush-4 5000000 272 ns/op 0 B/op 0 allocs/op
BenchmarkEchoSpeed4K-4 5000 259617 ns/op 15.78 MB/s 5451 B/op 149 allocs/op
BenchmarkEchoSpeed64K-4 1000 1706084 ns/op 38.41 MB/s 56002 B/op 1604 allocs/op
BenchmarkEchoSpeed512K-4 100 14345505 ns/op 36.55 MB/s 482597 B/op 13045 allocs/op
BenchmarkEchoSpeed1M-4 30 34859104 ns/op 30.08 MB/s 1143773 B/op 27186 allocs/op
BenchmarkSinkSpeed4K-4 50000 31369 ns/op 130.57 MB/s 1566 B/op 30 allocs/op
BenchmarkSinkSpeed64K-4 5000 329065 ns/op 199.16 MB/s 21529 B/op 453 allocs/op
BenchmarkSinkSpeed256K-4 500 2373354 ns/op 220.91 MB/s 166332 B/op 3554 allocs/op
BenchmarkSinkSpeed1M-4 300 5117927 ns/op 204.88 MB/s 310378 B/op 6988 allocs/op
PASS
ok _/Users/xtaci/.godeps/src/github.com/xtaci/kcp-go 39.689s
ok github.com/xtaci/kcp-go 50.349s
```
## Design Considerations
## Key Design Considerations
1. slice vs. container/list
@@ -139,7 +147,9 @@ List structure introduces **heavy cache misses** compared to slice which owns be
2. Timing accuracy vs. syscall clock_gettime
Timing is **critical** to **RTT estimator**, inaccurate timing introduces false retransmissions in KCP, but calling `time.Now()` costs 42 cycles(10.5ns on 4GHz CPU, 15.6ns on my MacBook Pro 2.7GHz), the benchmark for time.Now():
Timing is **critical** to **RTT estimator**, inaccurate timing leads to false retransmissions in KCP, but calling `time.Now()` costs 42 cycles(10.5ns on 4GHz CPU, 15.6ns on my MacBook Pro 2.7GHz).
The benchmark for time.Now() lies here:
https://github.com/xtaci/notes/blob/master/golang/benchmark2/syscall_test.go
@@ -147,14 +157,17 @@ https://github.com/xtaci/notes/blob/master/golang/benchmark2/syscall_test.go
BenchmarkNow-4 100000000 15.6 ns/op
```
In kcp-go, after each `kcp.output()` function call, current time will be updated upon return, and each `kcp.flush()` will get current time once. For most of the time, 5000 connections costs 5000 * 15.6ns = 78us(no packet needs to be sent by `kcp.output()`), as for 10MB/s data transfering with 1400 MTU, `kcp.output()` will be called around 7500 times and costs 117us for `time.Now()` in **every second**.
In kcp-go, after each `kcp.output()` function call, current clock time will be updated upon return, and for a single `kcp.flush()` operation, current time will be queried from system once. For most of the time, 5000 connections costs 5000 * 15.6ns = 78us(a fixed cost while no packet needs to be sent), as for 10MB/s data transfering with 1400 MTU, `kcp.output()` will be called around 7500 times and costs 117us for `time.Now()` in **every second**.
## Connection Termination
## Tuning
Control messages like **SYN/FIN/RST** in TCP **are not defined** in KCP, you need some **keepalive/heartbeat mechanism** in the application-level. A real world example is to use some **multiplexing** protocol over session, such as [smux](https://github.com/xtaci/smux)(with embedded keepalive mechanism), see [kcptun](https://github.com/xtaci/kcptun) for example.
Q: I'm handling >5K connections on my server. the CPU utilization is high.
## FAQ
A: A standalone `agent` or `gate` server for kcp-go is suggested, not only for CPU utilization, but also important to the **precision** of RTT measurements which indirectly affects retransmission. By increasing update `interval` with `SetNoDelay` like `conn.SetNoDelay(1, 40, 1, 1)` will dramatically reduce system load.
Q: I'm handling >5K connections on my server, the CPU utilization is so high.
A: A standalone `agent` or `gate` server for running kcp-go is suggested, not only for CPU utilization, but also important to the **precision** of RTT measurements(timing) which indirectly affects retransmission. By increasing update `interval` with `SetNoDelay` like `conn.SetNoDelay(1, 40, 1, 1)` will dramatically reduce system load, but lower the performance.
## Who is using this?
@@ -163,10 +176,9 @@ A: A standalone `agent` or `gate` server for kcp-go is suggested, not only for C
3. https://github.com/smallnest/rpcx -- A RPC service framework based on net/rpc like alibaba Dubbo and weibo Motan.
4. https://github.com/gonet2/agent -- A gateway for games with stream multiplexing.
5. https://github.com/syncthing/syncthing -- Open Source Continuous File Synchronization.
6. https://play.google.com/store/apps/details?id=com.k17game.k3 -- Battle Zone - Earth 2048, a world-wide strategy game.
## Links
1. https://github.com/xtaci/libkcp -- FEC enhanced KCP session library for iOS/Android in C++
2. https://github.com/skywind3000/kcp -- A Fast and Reliable ARQ Protocol
3. https://github.com/templexxx/reedsolomon -- Reed-Solomon Erasure Coding in Go
3. https://github.com/klauspost/reedsolomon -- Reed-Solomon Erasure Coding in Go

601
vendor/github.com/fatedier/kcp-go/crypt.go generated vendored
View File

@@ -57,8 +57,8 @@ func (c *salsa20BlockCrypt) Decrypt(dst, src []byte) {
}
type sm4BlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [sm4.BlockSize]byte
decbuf [2 * sm4.BlockSize]byte
block cipher.Block
}
@@ -70,17 +70,15 @@ func NewSM4BlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, sm4.BlockSize)
c.decbuf = make([]byte, 2*sm4.BlockSize)
return c, nil
}
func (c *sm4BlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *sm4BlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *sm4BlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *sm4BlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type twofishBlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [twofish.BlockSize]byte
decbuf [2 * twofish.BlockSize]byte
block cipher.Block
}
@@ -92,17 +90,15 @@ func NewTwofishBlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, twofish.BlockSize)
c.decbuf = make([]byte, 2*twofish.BlockSize)
return c, nil
}
func (c *twofishBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *twofishBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *twofishBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *twofishBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type tripleDESBlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [des.BlockSize]byte
decbuf [2 * des.BlockSize]byte
block cipher.Block
}
@@ -114,17 +110,15 @@ func NewTripleDESBlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, des.BlockSize)
c.decbuf = make([]byte, 2*des.BlockSize)
return c, nil
}
func (c *tripleDESBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *tripleDESBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *tripleDESBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *tripleDESBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type cast5BlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [cast5.BlockSize]byte
decbuf [2 * cast5.BlockSize]byte
block cipher.Block
}
@@ -136,17 +130,15 @@ func NewCast5BlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, cast5.BlockSize)
c.decbuf = make([]byte, 2*cast5.BlockSize)
return c, nil
}
func (c *cast5BlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *cast5BlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *cast5BlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *cast5BlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type blowfishBlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [blowfish.BlockSize]byte
decbuf [2 * blowfish.BlockSize]byte
block cipher.Block
}
@@ -158,17 +150,15 @@ func NewBlowfishBlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, blowfish.BlockSize)
c.decbuf = make([]byte, 2*blowfish.BlockSize)
return c, nil
}
func (c *blowfishBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *blowfishBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *blowfishBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *blowfishBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type aesBlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [aes.BlockSize]byte
decbuf [2 * aes.BlockSize]byte
block cipher.Block
}
@@ -180,17 +170,15 @@ func NewAESBlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, aes.BlockSize)
c.decbuf = make([]byte, 2*aes.BlockSize)
return c, nil
}
func (c *aesBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *aesBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *aesBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *aesBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type teaBlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [tea.BlockSize]byte
decbuf [2 * tea.BlockSize]byte
block cipher.Block
}
@@ -202,17 +190,15 @@ func NewTEABlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, tea.BlockSize)
c.decbuf = make([]byte, 2*tea.BlockSize)
return c, nil
}
func (c *teaBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *teaBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *teaBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *teaBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type xteaBlockCrypt struct {
encbuf []byte
decbuf []byte
encbuf [xtea.BlockSize]byte
decbuf [2 * xtea.BlockSize]byte
block cipher.Block
}
@@ -224,13 +210,11 @@ func NewXTEABlockCrypt(key []byte) (BlockCrypt, error) {
return nil, err
}
c.block = block
c.encbuf = make([]byte, xtea.BlockSize)
c.decbuf = make([]byte, 2*xtea.BlockSize)
return c, nil
}
func (c *xteaBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf) }
func (c *xteaBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf) }
func (c *xteaBlockCrypt) Encrypt(dst, src []byte) { encrypt(c.block, dst, src, c.encbuf[:]) }
func (c *xteaBlockCrypt) Decrypt(dst, src []byte) { decrypt(c.block, dst, src, c.decbuf[:]) }
type simpleXORBlockCrypt struct {
xortbl []byte
@@ -258,31 +242,544 @@ func (c *noneBlockCrypt) Decrypt(dst, src []byte) { copy(dst, src) }
// packet encryption with local CFB mode
func encrypt(block cipher.Block, dst, src, buf []byte) {
switch block.BlockSize() {
case 8:
encrypt8(block, dst, src, buf)
case 16:
encrypt16(block, dst, src, buf)
default:
encryptVariant(block, dst, src, buf)
}
}
// optimized encryption for the ciphers which works in 8-bytes
func encrypt8(block cipher.Block, dst, src, buf []byte) {
tbl := buf[:8]
block.Encrypt(tbl, initialVector)
n := len(src) / 8
base := 0
repeat := n / 8
left := n % 8
for i := 0; i < repeat; i++ {
s := src[base:][0:64]
d := dst[base:][0:64]
// 1
xor.BytesSrc1(d[0:8], s[0:8], tbl)
block.Encrypt(tbl, d[0:8])
// 2
xor.BytesSrc1(d[8:16], s[8:16], tbl)
block.Encrypt(tbl, d[8:16])
// 3
xor.BytesSrc1(d[16:24], s[16:24], tbl)
block.Encrypt(tbl, d[16:24])
// 4
xor.BytesSrc1(d[24:32], s[24:32], tbl)
block.Encrypt(tbl, d[24:32])
// 5
xor.BytesSrc1(d[32:40], s[32:40], tbl)
block.Encrypt(tbl, d[32:40])
// 6
xor.BytesSrc1(d[40:48], s[40:48], tbl)
block.Encrypt(tbl, d[40:48])
// 7
xor.BytesSrc1(d[48:56], s[48:56], tbl)
block.Encrypt(tbl, d[48:56])
// 8
xor.BytesSrc1(d[56:64], s[56:64], tbl)
block.Encrypt(tbl, d[56:64])
base += 64
}
switch left {
case 7:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 6:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 5:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 4:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 3:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 2:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 1:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 8
fallthrough
case 0:
xor.BytesSrc0(dst[base:], src[base:], tbl)
}
}
// optimized encryption for the ciphers which works in 16-bytes
func encrypt16(block cipher.Block, dst, src, buf []byte) {
tbl := buf[:16]
block.Encrypt(tbl, initialVector)
n := len(src) / 16
base := 0
repeat := n / 8
left := n % 8
for i := 0; i < repeat; i++ {
s := src[base:][0:128]
d := dst[base:][0:128]
// 1
xor.BytesSrc1(d[0:16], s[0:16], tbl)
block.Encrypt(tbl, d[0:16])
// 2
xor.BytesSrc1(d[16:32], s[16:32], tbl)
block.Encrypt(tbl, d[16:32])
// 3
xor.BytesSrc1(d[32:48], s[32:48], tbl)
block.Encrypt(tbl, d[32:48])
// 4
xor.BytesSrc1(d[48:64], s[48:64], tbl)
block.Encrypt(tbl, d[48:64])
// 5
xor.BytesSrc1(d[64:80], s[64:80], tbl)
block.Encrypt(tbl, d[64:80])
// 6
xor.BytesSrc1(d[80:96], s[80:96], tbl)
block.Encrypt(tbl, d[80:96])
// 7
xor.BytesSrc1(d[96:112], s[96:112], tbl)
block.Encrypt(tbl, d[96:112])
// 8
xor.BytesSrc1(d[112:128], s[112:128], tbl)
block.Encrypt(tbl, d[112:128])
base += 128
}
switch left {
case 7:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 6:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 5:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 4:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 3:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 2:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 1:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += 16
fallthrough
case 0:
xor.BytesSrc0(dst[base:], src[base:], tbl)
}
}
func encryptVariant(block cipher.Block, dst, src, buf []byte) {
blocksize := block.BlockSize()
tbl := buf[:blocksize]
block.Encrypt(tbl, initialVector)
n := len(src) / blocksize
base := 0
for i := 0; i < n; i++ {
repeat := n / 8
left := n % 8
for i := 0; i < repeat; i++ {
// 1
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 2
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 3
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 4
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 5
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 6
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 7
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
// 8
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
}
xor.BytesSrc0(dst[base:], src[base:], tbl)
switch left {
case 7:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 6:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 5:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 4:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 3:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 2:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 1:
xor.BytesSrc1(dst[base:], src[base:], tbl)
block.Encrypt(tbl, dst[base:])
base += blocksize
fallthrough
case 0:
xor.BytesSrc0(dst[base:], src[base:], tbl)
}
}
// decryption
func decrypt(block cipher.Block, dst, src, buf []byte) {
switch block.BlockSize() {
case 8:
decrypt8(block, dst, src, buf)
case 16:
decrypt16(block, dst, src, buf)
default:
decryptVariant(block, dst, src, buf)
}
}
func decrypt8(block cipher.Block, dst, src, buf []byte) {
tbl := buf[0:8]
next := buf[8:16]
block.Encrypt(tbl, initialVector)
n := len(src) / 8
base := 0
repeat := n / 8
left := n % 8
for i := 0; i < repeat; i++ {
s := src[base:][0:64]
d := dst[base:][0:64]
// 1
block.Encrypt(next, s[0:8])
xor.BytesSrc1(d[0:8], s[0:8], tbl)
// 2
block.Encrypt(tbl, s[8:16])
xor.BytesSrc1(d[8:16], s[8:16], next)
// 3
block.Encrypt(next, s[16:24])
xor.BytesSrc1(d[16:24], s[16:24], tbl)
// 4
block.Encrypt(tbl, s[24:32])
xor.BytesSrc1(d[24:32], s[24:32], next)
// 5
block.Encrypt(next, s[32:40])
xor.BytesSrc1(d[32:40], s[32:40], tbl)
// 6
block.Encrypt(tbl, s[40:48])
xor.BytesSrc1(d[40:48], s[40:48], next)
// 7
block.Encrypt(next, s[48:56])
xor.BytesSrc1(d[48:56], s[48:56], tbl)
// 8
block.Encrypt(tbl, s[56:64])
xor.BytesSrc1(d[56:64], s[56:64], next)
base += 64
}
switch left {
case 7:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 6:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 5:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 4:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 3:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 2:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 1:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 8
fallthrough
case 0:
xor.BytesSrc0(dst[base:], src[base:], tbl)
}
}
func decrypt16(block cipher.Block, dst, src, buf []byte) {
tbl := buf[0:16]
next := buf[16:32]
block.Encrypt(tbl, initialVector)
n := len(src) / 16
base := 0
repeat := n / 8
left := n % 8
for i := 0; i < repeat; i++ {
s := src[base:][0:128]
d := dst[base:][0:128]
// 1
block.Encrypt(next, s[0:16])
xor.BytesSrc1(d[0:16], s[0:16], tbl)
// 2
block.Encrypt(tbl, s[16:32])
xor.BytesSrc1(d[16:32], s[16:32], next)
// 3
block.Encrypt(next, s[32:48])
xor.BytesSrc1(d[32:48], s[32:48], tbl)
// 4
block.Encrypt(tbl, s[48:64])
xor.BytesSrc1(d[48:64], s[48:64], next)
// 5
block.Encrypt(next, s[64:80])
xor.BytesSrc1(d[64:80], s[64:80], tbl)
// 6
block.Encrypt(tbl, s[80:96])
xor.BytesSrc1(d[80:96], s[80:96], next)
// 7
block.Encrypt(next, s[96:112])
xor.BytesSrc1(d[96:112], s[96:112], tbl)
// 8
block.Encrypt(tbl, s[112:128])
xor.BytesSrc1(d[112:128], s[112:128], next)
base += 128
}
switch left {
case 7:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 6:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 5:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 4:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 3:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 2:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 1:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += 16
fallthrough
case 0:
xor.BytesSrc0(dst[base:], src[base:], tbl)
}
}
func decryptVariant(block cipher.Block, dst, src, buf []byte) {
blocksize := block.BlockSize()
tbl := buf[:blocksize]
next := buf[blocksize:]
block.Encrypt(tbl, initialVector)
n := len(src) / blocksize
base := 0
for i := 0; i < n; i++ {
repeat := n / 8
left := n % 8
for i := 0; i < repeat; i++ {
// 1
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
base += blocksize
// 2
block.Encrypt(tbl, src[base:])
xor.BytesSrc1(dst[base:], src[base:], next)
base += blocksize
// 3
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
base += blocksize
// 4
block.Encrypt(tbl, src[base:])
xor.BytesSrc1(dst[base:], src[base:], next)
base += blocksize
// 5
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
base += blocksize
// 6
block.Encrypt(tbl, src[base:])
xor.BytesSrc1(dst[base:], src[base:], next)
base += blocksize
// 7
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
base += blocksize
// 8
block.Encrypt(tbl, src[base:])
xor.BytesSrc1(dst[base:], src[base:], next)
base += blocksize
}
switch left {
case 7:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 6:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 5:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 4:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 3:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 2:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 1:
block.Encrypt(next, src[base:])
xor.BytesSrc1(dst[base:], src[base:], tbl)
tbl, next = next, tbl
base += blocksize
fallthrough
case 0:
xor.BytesSrc0(dst[base:], src[base:], tbl)
}
xor.BytesSrc0(dst[base:], src[base:], tbl)
}

52
vendor/github.com/fatedier/kcp-go/entropy.go generated vendored Normal file
View File

@@ -0,0 +1,52 @@
package kcp
import (
"crypto/aes"
"crypto/cipher"
"crypto/md5"
"crypto/rand"
"io"
)
// Entropy defines a entropy source
type Entropy interface {
Init()
Fill(nonce []byte)
}
// nonceMD5 nonce generator for packet header
type nonceMD5 struct {
seed [md5.Size]byte
}
func (n *nonceMD5) Init() { /*nothing required*/ }
func (n *nonceMD5) Fill(nonce []byte) {
if n.seed[0] == 0 { // entropy update
io.ReadFull(rand.Reader, n.seed[:])
}
n.seed = md5.Sum(n.seed[:])
copy(nonce, n.seed[:])
}
// nonceAES128 nonce generator for packet headers
type nonceAES128 struct {
seed [aes.BlockSize]byte
block cipher.Block
}
func (n *nonceAES128) Init() {
var key [16]byte //aes-128
io.ReadFull(rand.Reader, key[:])
io.ReadFull(rand.Reader, n.seed[:])
block, _ := aes.NewCipher(key[:])
n.block = block
}
func (n *nonceAES128) Fill(nonce []byte) {
if n.seed[0] == 0 { // entropy update
io.ReadFull(rand.Reader, n.seed[:])
}
n.block.Encrypt(n.seed[:], n.seed[:])
copy(nonce, n.seed[:])
}

88
vendor/github.com/fatedier/kcp-go/fec.go generated vendored
View File

@@ -4,7 +4,7 @@ import (
"encoding/binary"
"sync/atomic"
"github.com/templexxx/reedsolomon"
"github.com/klauspost/reedsolomon"
)
const (
@@ -34,6 +34,9 @@ type (
decodeCache [][]byte
flagCache []bool
// zeros
zeros []byte
// RS decoder
codec reedsolomon.Encoder
}
@@ -47,19 +50,20 @@ func newFECDecoder(rxlimit, dataShards, parityShards int) *fecDecoder {
return nil
}
fec := new(fecDecoder)
fec.rxlimit = rxlimit
fec.dataShards = dataShards
fec.parityShards = parityShards
fec.shardSize = dataShards + parityShards
enc, err := reedsolomon.New(dataShards, parityShards)
dec := new(fecDecoder)
dec.rxlimit = rxlimit
dec.dataShards = dataShards
dec.parityShards = parityShards
dec.shardSize = dataShards + parityShards
codec, err := reedsolomon.New(dataShards, parityShards)
if err != nil {
return nil
}
fec.codec = enc
fec.decodeCache = make([][]byte, fec.shardSize)
fec.flagCache = make([]bool, fec.shardSize)
return fec
dec.codec = codec
dec.decodeCache = make([][]byte, dec.shardSize)
dec.flagCache = make([]bool, dec.shardSize)
dec.zeros = make([]byte, mtuLimit)
return dec
}
// decodeBytes a fec packet
@@ -116,7 +120,7 @@ func (dec *fecDecoder) decode(pkt fecPacket) (recovered [][]byte) {
if searchEnd-searchBegin+1 >= dec.dataShards {
var numshard, numDataShard, first, maxlen int
// zero cache
// zero caches
shards := dec.decodeCache
shardsflag := dec.flagCache
for k := range dec.decodeCache {
@@ -146,15 +150,15 @@ func (dec *fecDecoder) decode(pkt fecPacket) (recovered [][]byte) {
}
if numDataShard == dec.dataShards {
// case 1: no lost data shards
// case 1: no loss on data shards
dec.rx = dec.freeRange(first, numshard, dec.rx)
} else if numshard >= dec.dataShards {
// case 2: data shard lost, but recoverable from parity shard
// case 2: loss on data shards, but it's recoverable from parity shards
for k := range shards {
if shards[k] != nil {
dlen := len(shards[k])
shards[k] = shards[k][:maxlen]
xorBytes(shards[k][dlen:], shards[k][dlen:], shards[k][dlen:])
copy(shards[k][dlen:], dec.zeros)
}
}
if err := dec.codec.ReconstructData(shards); err == nil {
@@ -170,7 +174,7 @@ func (dec *fecDecoder) decode(pkt fecPacket) (recovered [][]byte) {
// keep rxlimit
if len(dec.rx) > dec.rxlimit {
if dec.rx[0].flag == typeData { // record unrecoverable data
if dec.rx[0].flag == typeData { // track the unrecoverable data
atomic.AddUint64(&DefaultSnmp.FECShortShards, 1)
}
dec.rx = dec.freeRange(0, 1, dec.rx)
@@ -180,7 +184,7 @@ func (dec *fecDecoder) decode(pkt fecPacket) (recovered [][]byte) {
// free a range of fecPacket, and zero for GC recycling
func (dec *fecDecoder) freeRange(first, n int, q []fecPacket) []fecPacket {
for i := first; i < first+n; i++ { // free
for i := first; i < first+n; i++ { // recycle buffer
xmitBuf.Put(q[i].data)
}
copy(q[first:], q[first+n:])
@@ -200,7 +204,7 @@ type (
next uint32 // next seqid
shardCount int // count the number of datashards collected
maxSize int // record maximum data length in datashard
maxSize int // track maximum data length in datashard
headerOffset int // FEC header offset
payloadOffset int // FEC payload offset
@@ -209,6 +213,9 @@ type (
shardCache [][]byte
encodeCache [][]byte
// zeros
zeros []byte
// RS encoder
codec reedsolomon.Encoder
}
@@ -218,31 +225,32 @@ func newFECEncoder(dataShards, parityShards, offset int) *fecEncoder {
if dataShards <= 0 || parityShards <= 0 {
return nil
}
fec := new(fecEncoder)
fec.dataShards = dataShards
fec.parityShards = parityShards
fec.shardSize = dataShards + parityShards
fec.paws = (0xffffffff/uint32(fec.shardSize) - 1) * uint32(fec.shardSize)
fec.headerOffset = offset
fec.payloadOffset = fec.headerOffset + fecHeaderSize
enc := new(fecEncoder)
enc.dataShards = dataShards
enc.parityShards = parityShards
enc.shardSize = dataShards + parityShards
enc.paws = (0xffffffff/uint32(enc.shardSize) - 1) * uint32(enc.shardSize)
enc.headerOffset = offset
enc.payloadOffset = enc.headerOffset + fecHeaderSize
enc, err := reedsolomon.New(dataShards, parityShards)
codec, err := reedsolomon.New(dataShards, parityShards)
if err != nil {
return nil
}
fec.codec = enc
enc.codec = codec
// caches
fec.encodeCache = make([][]byte, fec.shardSize)
fec.shardCache = make([][]byte, fec.shardSize)
for k := range fec.shardCache {
fec.shardCache[k] = make([]byte, mtuLimit)
enc.encodeCache = make([][]byte, enc.shardSize)
enc.shardCache = make([][]byte, enc.shardSize)
for k := range enc.shardCache {
enc.shardCache[k] = make([]byte, mtuLimit)
}
return fec
enc.zeros = make([]byte, mtuLimit)
return enc
}
// encode the packet, output parity shards if we have enough datashards
// the content of returned parityshards will change in next encode
// encodes the packet, outputs parity shards if we have collected quorum datashards
// notice: the contents of 'ps' will be re-written in successive calling
func (enc *fecEncoder) encode(b []byte) (ps [][]byte) {
enc.markData(b[enc.headerOffset:])
binary.LittleEndian.PutUint16(b[enc.payloadOffset:], uint16(len(b[enc.payloadOffset:])))
@@ -253,18 +261,18 @@ func (enc *fecEncoder) encode(b []byte) (ps [][]byte) {
copy(enc.shardCache[enc.shardCount], b)
enc.shardCount++
// record max datashard length
// track max datashard length
if sz > enc.maxSize {
enc.maxSize = sz
}
// calculate Reed-Solomon Erasure Code
// Generation of Reed-Solomon Erasure Code
if enc.shardCount == enc.dataShards {
// bzero each datashard's tail
// fill '0' into the tail of each datashard
for i := 0; i < enc.dataShards; i++ {
shard := enc.shardCache[i]
slen := len(shard)
xorBytes(shard[slen:enc.maxSize], shard[slen:enc.maxSize], shard[slen:enc.maxSize])
copy(shard[slen:enc.maxSize], enc.zeros)
}
// construct equal-sized slice with stripped header
@@ -273,7 +281,7 @@ func (enc *fecEncoder) encode(b []byte) (ps [][]byte) {
cache[k] = enc.shardCache[k][enc.payloadOffset:enc.maxSize]
}
// rs encode
// encoding
if err := enc.codec.Encode(cache); err == nil {
ps = enc.shardCache[enc.dataShards:]
for k := range ps {
@@ -282,7 +290,7 @@ func (enc *fecEncoder) encode(b []byte) (ps [][]byte) {
}
}
// reset counters to zero
// counters resetting
enc.shardCount = 0
enc.maxSize = 0
}

186
vendor/github.com/fatedier/kcp-go/kcp.go generated vendored
View File

@@ -104,6 +104,7 @@ type segment struct {
xmit uint32
resendts uint32
fastack uint32
acked uint32 // mark if the seg has acked
data []byte
}
@@ -181,8 +182,11 @@ func (kcp *KCP) newSegment(size int) (seg segment) {
}
// delSegment recycles a KCP segment
func (kcp *KCP) delSegment(seg segment) {
xmitBuf.Put(seg.data)
func (kcp *KCP) delSegment(seg *segment) {
if seg.data != nil {
xmitBuf.Put(seg.data)
seg.data = nil
}
}
// PeekSize checks the size of next message in the recv queue
@@ -238,7 +242,7 @@ func (kcp *KCP) Recv(buffer []byte) (n int) {
buffer = buffer[len(seg.data):]
n += len(seg.data)
count++
kcp.delSegment(*seg)
kcp.delSegment(seg)
if seg.frg == 0 {
break
}
@@ -382,10 +386,8 @@ func (kcp *KCP) parse_ack(sn uint32) {
for k := range kcp.snd_buf {
seg := &kcp.snd_buf[k]
if sn == seg.sn {
kcp.delSegment(*seg)
copy(kcp.snd_buf[k:], kcp.snd_buf[k+1:])
kcp.snd_buf[len(kcp.snd_buf)-1] = segment{}
kcp.snd_buf = kcp.snd_buf[:len(kcp.snd_buf)-1]
seg.acked = 1
kcp.delSegment(seg)
break
}
if _itimediff(sn, seg.sn) < 0 {
@@ -394,7 +396,7 @@ func (kcp *KCP) parse_ack(sn uint32) {
}
}
func (kcp *KCP) parse_fastack(sn uint32) {
func (kcp *KCP) parse_fastack(sn, ts uint32) {
if _itimediff(sn, kcp.snd_una) < 0 || _itimediff(sn, kcp.snd_nxt) >= 0 {
return
}
@@ -403,7 +405,7 @@ func (kcp *KCP) parse_fastack(sn uint32) {
seg := &kcp.snd_buf[k]
if _itimediff(sn, seg.sn) < 0 {
break
} else if sn != seg.sn {
} else if sn != seg.sn && _itimediff(seg.ts, ts) <= 0 {
seg.fastack++
}
}
@@ -414,7 +416,7 @@ func (kcp *KCP) parse_una(una uint32) {
for k := range kcp.snd_buf {
seg := &kcp.snd_buf[k]
if _itimediff(una, seg.sn) > 0 {
kcp.delSegment(*seg)
kcp.delSegment(seg)
count++
} else {
break
@@ -430,12 +432,12 @@ func (kcp *KCP) ack_push(sn, ts uint32) {
kcp.acklist = append(kcp.acklist, ackItem{sn, ts})
}
func (kcp *KCP) parse_data(newseg segment) {
// returns true if data has repeated
func (kcp *KCP) parse_data(newseg segment) bool {
sn := newseg.sn
if _itimediff(sn, kcp.rcv_nxt+kcp.rcv_wnd) >= 0 ||
_itimediff(sn, kcp.rcv_nxt) < 0 {
kcp.delSegment(newseg)
return
return true
}
n := len(kcp.rcv_buf) - 1
@@ -445,7 +447,6 @@ func (kcp *KCP) parse_data(newseg segment) {
seg := &kcp.rcv_buf[i]
if seg.sn == sn {
repeat = true
atomic.AddUint64(&DefaultSnmp.RepeatSegs, 1)
break
}
if _itimediff(sn, seg.sn) > 0 {
@@ -455,6 +456,11 @@ func (kcp *KCP) parse_data(newseg segment) {
}
if !repeat {
// replicate the content if it's new
dataCopy := xmitBuf.Get().([]byte)[:len(newseg.data)]
copy(dataCopy, newseg.data)
newseg.data = dataCopy
if insert_idx == n+1 {
kcp.rcv_buf = append(kcp.rcv_buf, newseg)
} else {
@@ -462,8 +468,6 @@ func (kcp *KCP) parse_data(newseg segment) {
copy(kcp.rcv_buf[insert_idx+1:], kcp.rcv_buf[insert_idx:])
kcp.rcv_buf[insert_idx] = newseg
}
} else {
kcp.delSegment(newseg)
}
// move available data from rcv_buf -> rcv_queue
@@ -481,18 +485,19 @@ func (kcp *KCP) parse_data(newseg segment) {
kcp.rcv_queue = append(kcp.rcv_queue, kcp.rcv_buf[:count]...)
kcp.rcv_buf = kcp.remove_front(kcp.rcv_buf, count)
}
return repeat
}
// Input when you received a low level packet (eg. UDP packet), call it
// regular indicates a regular packet has received(not from FEC)
func (kcp *KCP) Input(data []byte, regular, ackNoDelay bool) int {
una := kcp.snd_una
snd_una := kcp.snd_una
if len(data) < IKCP_OVERHEAD {
return -1
}
var maxack uint32
var lastackts uint32
var latest uint32 // the latest ack packet
var flag int
var inSegs uint64
@@ -535,19 +540,15 @@ func (kcp *KCP) Input(data []byte, regular, ackNoDelay bool) int {
if cmd == IKCP_CMD_ACK {
kcp.parse_ack(sn)
kcp.shrink_buf()
if flag == 0 {
flag = 1
maxack = sn
} else if _itimediff(sn, maxack) > 0 {
maxack = sn
}
lastackts = ts
kcp.parse_fastack(sn, ts)
flag |= 1
latest = ts
} else if cmd == IKCP_CMD_PUSH {
repeat := true
if _itimediff(sn, kcp.rcv_nxt+kcp.rcv_wnd) < 0 {
kcp.ack_push(sn, ts)
if _itimediff(sn, kcp.rcv_nxt) >= 0 {
seg := kcp.newSegment(int(length))
var seg segment
seg.conv = conv
seg.cmd = cmd
seg.frg = frg
@@ -555,12 +556,11 @@ func (kcp *KCP) Input(data []byte, regular, ackNoDelay bool) int {
seg.ts = ts
seg.sn = sn
seg.una = una
copy(seg.data, data[:length])
kcp.parse_data(seg)
} else {
atomic.AddUint64(&DefaultSnmp.RepeatSegs, 1)
seg.data = data[:length] // delayed data copying
repeat = kcp.parse_data(seg)
}
} else {
}
if regular && repeat {
atomic.AddUint64(&DefaultSnmp.RepeatSegs, 1)
}
} else if cmd == IKCP_CMD_WASK {
@@ -578,40 +578,42 @@ func (kcp *KCP) Input(data []byte, regular, ackNoDelay bool) int {
}
atomic.AddUint64(&DefaultSnmp.InSegs, inSegs)
// update rtt with the latest ts
// ignore the FEC packet
if flag != 0 && regular {
kcp.parse_fastack(maxack)
current := currentMs()
if _itimediff(current, lastackts) >= 0 {
kcp.update_ack(_itimediff(current, lastackts))
if _itimediff(current, latest) >= 0 {
kcp.update_ack(_itimediff(current, latest))
}
}
if _itimediff(kcp.snd_una, una) > 0 {
if kcp.cwnd < kcp.rmt_wnd {
mss := kcp.mss
if kcp.cwnd < kcp.ssthresh {
kcp.cwnd++
kcp.incr += mss
} else {
if kcp.incr < mss {
kcp.incr = mss
}
kcp.incr += (mss*mss)/kcp.incr + (mss / 16)
if (kcp.cwnd+1)*mss <= kcp.incr {
// cwnd update when packet arrived
if kcp.nocwnd == 0 {
if _itimediff(kcp.snd_una, snd_una) > 0 {
if kcp.cwnd < kcp.rmt_wnd {
mss := kcp.mss
if kcp.cwnd < kcp.ssthresh {
kcp.cwnd++
kcp.incr += mss
} else {
if kcp.incr < mss {
kcp.incr = mss
}
kcp.incr += (mss*mss)/kcp.incr + (mss / 16)
if (kcp.cwnd+1)*mss <= kcp.incr {
kcp.cwnd++
}
}
if kcp.cwnd > kcp.rmt_wnd {
kcp.cwnd = kcp.rmt_wnd
kcp.incr = kcp.rmt_wnd * mss
}
}
if kcp.cwnd > kcp.rmt_wnd {
kcp.cwnd = kcp.rmt_wnd
kcp.incr = kcp.rmt_wnd * mss
}
}
}
if ackNoDelay && len(kcp.acklist) > 0 { // ack immediately
kcp.flush(true)
} else if kcp.rmt_wnd == 0 && len(kcp.acklist) > 0 { // window zero
kcp.flush(true)
}
return 0
}
@@ -624,7 +626,7 @@ func (kcp *KCP) wnd_unused() uint16 {
}
// flush pending data
func (kcp *KCP) flush(ackOnly bool) {
func (kcp *KCP) flush(ackOnly bool) uint32 {
var seg segment
seg.conv = kcp.conv
seg.cmd = IKCP_CMD_ACK
@@ -653,7 +655,7 @@ func (kcp *KCP) flush(ackOnly bool) {
if size > 0 {
kcp.output(buffer, size)
}
return
return kcp.interval
}
// probe window size (if remote window size equals zero)
@@ -723,7 +725,6 @@ func (kcp *KCP) flush(ackOnly bool) {
kcp.snd_buf = append(kcp.snd_buf, newseg)
kcp.snd_nxt++
newSegsCount++
kcp.snd_queue[k].data = nil
}
if newSegsCount > 0 {
kcp.snd_queue = kcp.remove_front(kcp.snd_queue, newSegsCount)
@@ -738,9 +739,15 @@ func (kcp *KCP) flush(ackOnly bool) {
// check for retransmissions
current := currentMs()
var change, lost, lostSegs, fastRetransSegs, earlyRetransSegs uint64
for k := range kcp.snd_buf {
segment := &kcp.snd_buf[k]
minrto := int32(kcp.interval)
ref := kcp.snd_buf[:len(kcp.snd_buf)] // for bounds check elimination
for k := range ref {
segment := &ref[k]
needsend := false
if segment.acked == 1 {
continue
}
if segment.xmit == 0 { // initial transmit
needsend = true
segment.rto = kcp.rx_rto
@@ -772,6 +779,7 @@ func (kcp *KCP) flush(ackOnly bool) {
}
if needsend {
current = currentMs() // time update for a blocking call
segment.xmit++
segment.ts = current
segment.wnd = seg.wnd
@@ -782,7 +790,6 @@ func (kcp *KCP) flush(ackOnly bool) {
if size+need > int(kcp.mtu) {
kcp.output(buffer, size)
current = currentMs() // time update for a blocking call
ptr = buffer
}
@@ -794,6 +801,11 @@ func (kcp *KCP) flush(ackOnly bool) {
kcp.state = 0xFFFFFFFF
}
}
// get the nearest rto
if rto := _itimediff(segment.resendts, current); rto > 0 && rto < minrto {
minrto = rto
}
}
// flash remain segments
@@ -819,32 +831,37 @@ func (kcp *KCP) flush(ackOnly bool) {
atomic.AddUint64(&DefaultSnmp.RetransSegs, sum)
}
// update ssthresh
// rate halving, https://tools.ietf.org/html/rfc6937
if change > 0 {
inflight := kcp.snd_nxt - kcp.snd_una
kcp.ssthresh = inflight / 2
if kcp.ssthresh < IKCP_THRESH_MIN {
kcp.ssthresh = IKCP_THRESH_MIN
// cwnd update
if kcp.nocwnd == 0 {
// update ssthresh
// rate halving, https://tools.ietf.org/html/rfc6937
if change > 0 {
inflight := kcp.snd_nxt - kcp.snd_una
kcp.ssthresh = inflight / 2
if kcp.ssthresh < IKCP_THRESH_MIN {
kcp.ssthresh = IKCP_THRESH_MIN
}
kcp.cwnd = kcp.ssthresh + resent
kcp.incr = kcp.cwnd * kcp.mss
}
// congestion control, https://tools.ietf.org/html/rfc5681
if lost > 0 {
kcp.ssthresh = cwnd / 2
if kcp.ssthresh < IKCP_THRESH_MIN {
kcp.ssthresh = IKCP_THRESH_MIN
}
kcp.cwnd = 1
kcp.incr = kcp.mss
}
if kcp.cwnd < 1 {
kcp.cwnd = 1
kcp.incr = kcp.mss
}
kcp.cwnd = kcp.ssthresh + resent
kcp.incr = kcp.cwnd * kcp.mss
}
// congestion control, https://tools.ietf.org/html/rfc5681
if lost > 0 {
kcp.ssthresh = cwnd / 2
if kcp.ssthresh < IKCP_THRESH_MIN {
kcp.ssthresh = IKCP_THRESH_MIN
}
kcp.cwnd = 1
kcp.incr = kcp.mss
}
if kcp.cwnd < 1 {
kcp.cwnd = 1
kcp.incr = kcp.mss
}
return uint32(minrto)
}
// Update updates state (call it repeatedly, every 10ms-100ms), or you can ask
@@ -991,8 +1008,5 @@ func (kcp *KCP) WaitSnd() int {
// remove front n elements from queue
func (kcp *KCP) remove_front(q []segment, n int) []segment {
newn := copy(q, q[n:])
for i := newn; i < len(q); i++ {
q[i] = segment{} // manual set nil for GC
}
return q[:newn]
}

472
vendor/github.com/fatedier/kcp-go/sess.go generated vendored
View File

@@ -4,7 +4,6 @@ import (
"crypto/rand"
"encoding/binary"
"hash/crc32"
"io"
"net"
"sync"
"sync/atomic"
@@ -12,6 +11,7 @@ import (
"github.com/pkg/errors"
"golang.org/x/net/ipv4"
"golang.org/x/net/ipv6"
)
type errTimeout struct {
@@ -23,7 +23,7 @@ func (errTimeout) Temporary() bool { return true }
func (errTimeout) Error() string { return "i/o timeout" }
const (
// 16-bytes magic number for each packet
// 16-bytes nonce for each packet
nonceSize = 16
// 4-bytes packet checksum
@@ -40,9 +40,6 @@ const (
// accept backlog
acceptBacklog = 128
// prerouting(to session) queue
qlen = 128
)
const (
@@ -51,8 +48,8 @@ const (
)
var (
// global packet buffer
// shared among sending/receiving/FEC
// a system-wide packet buffer shared among sending, receiving and FEC
// to mitigate high-frequency memory allocation for packets
xmitBuf sync.Pool
)
@@ -68,17 +65,17 @@ type (
updaterIdx int // record slice index in updater
conn net.PacketConn // the underlying packet connection
kcp *KCP // KCP ARQ protocol
l *Listener // point to the Listener if it's accepted by Listener
block BlockCrypt // block encryption
l *Listener // pointing to the Listener object if it's been accepted by a Listener
block BlockCrypt // block encryption object
// kcp receiving is based on packets
// recvbuf turns packets into stream
recvbuf []byte
bufptr []byte
// extended output buffer(with header)
// header extended output buffer, if has header
ext []byte
// FEC
// FEC codec
fecDecoder *fecDecoder
fecEncoder *fecEncoder
@@ -86,16 +83,20 @@ type (
remote net.Addr // remote peer address
rd time.Time // read deadline
wd time.Time // write deadline
headerSize int // the overall header size added before KCP frame
ackNoDelay bool // send ack immediately for each incoming packet
headerSize int // the header size additional to a KCP frame
ackNoDelay bool // send ack immediately for each incoming packet(testing purpose)
writeDelay bool // delay kcp.flush() for Write() for bulk transfer
dup int // duplicate udp packets
dup int // duplicate udp packets(testing purpose)
// notifications
die chan struct{} // notify session has Closed
die chan struct{} // notify current session has Closed
chReadEvent chan struct{} // notify Read() can be called without blocking
chWriteEvent chan struct{} // notify Write() can be called without blocking
chErrorEvent chan error // notify Read() have an error
chReadError chan error // notify PacketConn.Read() have an error
chWriteError chan error // notify PacketConn.Write() have an error
// nonce generator
nonce Entropy
isClosed bool // flag the session has Closed
mu sync.Mutex
@@ -114,16 +115,19 @@ type (
func newUDPSession(conv uint32, dataShards, parityShards int, l *Listener, conn net.PacketConn, remote net.Addr, block BlockCrypt) *UDPSession {
sess := new(UDPSession)
sess.die = make(chan struct{})
sess.nonce = new(nonceAES128)
sess.nonce.Init()
sess.chReadEvent = make(chan struct{}, 1)
sess.chWriteEvent = make(chan struct{}, 1)
sess.chErrorEvent = make(chan error, 1)
sess.chReadError = make(chan error, 1)
sess.chWriteError = make(chan error, 1)
sess.remote = remote
sess.conn = conn
sess.l = l
sess.block = block
sess.recvbuf = make([]byte, mtuLimit)
// FEC initialization
// FEC codec initialization
sess.fecDecoder = newFECDecoder(rxFECMulti*(dataShards+parityShards), dataShards, parityShards)
if sess.block != nil {
sess.fecEncoder = newFECEncoder(dataShards, parityShards, cryptHeaderSize)
@@ -131,7 +135,7 @@ func newUDPSession(conv uint32, dataShards, parityShards int, l *Listener, conn
sess.fecEncoder = newFECEncoder(dataShards, parityShards, 0)
}
// calculate header size
// calculate additional header size introduced by FEC and encryption
if sess.block != nil {
sess.headerSize += cryptHeaderSize
}
@@ -139,8 +143,7 @@ func newUDPSession(conv uint32, dataShards, parityShards int, l *Listener, conn
sess.headerSize += fecHeaderSizePlus2
}
// only allocate extended packet buffer
// when the extra header is required
// we only need to allocate extended packet buffer if we have the additional header
if sess.headerSize > 0 {
sess.ext = make([]byte, mtuLimit)
}
@@ -152,8 +155,8 @@ func newUDPSession(conv uint32, dataShards, parityShards int, l *Listener, conn
})
sess.kcp.SetMtu(IKCP_MTU_DEF - sess.headerSize)
// add current session to the global updater,
// which periodically calls sess.update()
// register current session to the global updater,
// which call sess.update() periodically.
updater.addSession(sess)
if sess.l == nil { // it's a client connection
@@ -179,6 +182,7 @@ func (s *UDPSession) Read(b []byte) (n int, err error) {
n = copy(b, s.bufptr)
s.bufptr = s.bufptr[n:]
s.mu.Unlock()
atomic.AddUint64(&DefaultSnmp.BytesReceived, uint64(n))
return n, nil
}
@@ -188,29 +192,29 @@ func (s *UDPSession) Read(b []byte) (n int, err error) {
}
if size := s.kcp.PeekSize(); size > 0 { // peek data size from kcp
atomic.AddUint64(&DefaultSnmp.BytesReceived, uint64(size))
if len(b) >= size { // direct write to b
if len(b) >= size { // receive data into 'b' directly
s.kcp.Recv(b)
s.mu.Unlock()
atomic.AddUint64(&DefaultSnmp.BytesReceived, uint64(size))
return size, nil
}
// resize kcp receive buffer
// to make sure recvbuf has enough capacity
// if necessary resize the stream buffer to guarantee a sufficent buffer space
if cap(s.recvbuf) < size {
s.recvbuf = make([]byte, size)
}
// resize recvbuf slice length
// resize the length of recvbuf to correspond to data size
s.recvbuf = s.recvbuf[:size]
s.kcp.Recv(s.recvbuf)
n = copy(b, s.recvbuf) // copy to b
s.bufptr = s.recvbuf[n:] // update pointer
n = copy(b, s.recvbuf) // copy to 'b'
s.bufptr = s.recvbuf[n:] // pointer update
s.mu.Unlock()
atomic.AddUint64(&DefaultSnmp.BytesReceived, uint64(n))
return n, nil
}
// read deadline
// deadline for current reading operation
var timeout *time.Timer
var c <-chan time.Time
if !s.rd.IsZero() {
@@ -230,7 +234,7 @@ func (s *UDPSession) Read(b []byte) (n int, err error) {
case <-s.chReadEvent:
case <-c:
case <-s.die:
case err = <-s.chErrorEvent:
case err = <-s.chReadError:
if timeout != nil {
timeout.Stop()
}
@@ -252,7 +256,8 @@ func (s *UDPSession) Write(b []byte) (n int, err error) {
return 0, errors.New(errBrokenPipe)
}
// api flow control
// controls how much data will be sent to kcp core
// to prevent the memory from exhuasting
if s.kcp.WaitSnd() < int(s.kcp.snd_wnd) {
n = len(b)
for {
@@ -265,7 +270,8 @@ func (s *UDPSession) Write(b []byte) (n int, err error) {
}
}
if !s.writeDelay {
// flush immediately if the queue is full
if s.kcp.WaitSnd() >= int(s.kcp.snd_wnd) || !s.writeDelay {
s.kcp.flush(false)
}
s.mu.Unlock()
@@ -273,7 +279,7 @@ func (s *UDPSession) Write(b []byte) (n int, err error) {
return n, nil
}
// write deadline
// deadline for current writing operation
var timeout *time.Timer
var c <-chan time.Time
if !s.wd.IsZero() {
@@ -292,6 +298,11 @@ func (s *UDPSession) Write(b []byte) (n int, err error) {
case <-s.chWriteEvent:
case <-c:
case <-s.die:
case err = <-s.chWriteError:
if timeout != nil {
timeout.Stop()
}
return n, err
}
if timeout != nil {
@@ -302,13 +313,10 @@ func (s *UDPSession) Write(b []byte) (n int, err error) {
// Close closes the connection.
func (s *UDPSession) Close() error {
// remove this session from updater & listener(if necessary)
// remove current session from updater & listener(if necessary)
updater.removeSession(s)
if s.l != nil { // notify listener
s.l.closeSession(sessionKey{
addr: s.remote.String(),
convID: s.kcp.conv,
})
s.l.closeSession(s.remote)
}
s.mu.Lock()
@@ -337,6 +345,8 @@ func (s *UDPSession) SetDeadline(t time.Time) error {
defer s.mu.Unlock()
s.rd = t
s.wd = t
s.notifyReadEvent()
s.notifyWriteEvent()
return nil
}
@@ -345,6 +355,7 @@ func (s *UDPSession) SetReadDeadline(t time.Time) error {
s.mu.Lock()
defer s.mu.Unlock()
s.rd = t
s.notifyReadEvent()
return nil
}
@@ -353,6 +364,7 @@ func (s *UDPSession) SetWriteDeadline(t time.Time) error {
s.mu.Lock()
defer s.mu.Unlock()
s.wd = t
s.notifyWriteEvent()
return nil
}
@@ -420,10 +432,11 @@ func (s *UDPSession) SetDSCP(dscp int) error {
s.mu.Lock()
defer s.mu.Unlock()
if s.l == nil {
if nc, ok := s.conn.(*connectedUDPConn); ok {
return ipv4.NewConn(nc.UDPConn).SetTOS(dscp << 2)
} else if nc, ok := s.conn.(net.Conn); ok {
return ipv4.NewConn(nc).SetTOS(dscp << 2)
if nc, ok := s.conn.(net.Conn); ok {
if err := ipv4.NewConn(nc).SetTOS(dscp << 2); err != nil {
return ipv6.NewConn(nc).SetTrafficClass(dscp)
}
return nil
}
}
return errors.New(errInvalidOperation)
@@ -453,11 +466,11 @@ func (s *UDPSession) SetWriteBuffer(bytes int) error {
return errors.New(errInvalidOperation)
}
// output pipeline entry
// steps for output data processing:
// 0. Header extends
// 1. FEC
// 2. CRC32
// post-processing for sending a packet from kcp core
// steps:
// 0. Header extending
// 1. FEC packet generation
// 2. CRC32 integrity
// 3. Encryption
// 4. WriteTo kernel
func (s *UDPSession) output(buf []byte) {
@@ -477,13 +490,13 @@ func (s *UDPSession) output(buf []byte) {
// 2&3. crc32 & encryption
if s.block != nil {
io.ReadFull(rand.Reader, ext[:nonceSize])
s.nonce.Fill(ext[:nonceSize])
checksum := crc32.ChecksumIEEE(ext[cryptHeaderSize:])
binary.LittleEndian.PutUint32(ext[nonceSize:], checksum)
s.block.Encrypt(ext, ext)
for k := range ecc {
io.ReadFull(rand.Reader, ecc[k][:nonceSize])
s.nonce.Fill(ecc[k][:nonceSize])
checksum := crc32.ChecksumIEEE(ecc[k][cryptHeaderSize:])
binary.LittleEndian.PutUint32(ecc[k][nonceSize:], checksum)
s.block.Encrypt(ecc[k], ecc[k])
@@ -497,6 +510,8 @@ func (s *UDPSession) output(buf []byte) {
if n, err := s.conn.WriteTo(ext, s.remote); err == nil {
nbytes += n
npkts++
} else {
s.notifyWriteError(err)
}
}
@@ -504,6 +519,8 @@ func (s *UDPSession) output(buf []byte) {
if n, err := s.conn.WriteTo(ecc[k], s.remote); err == nil {
nbytes += n
npkts++
} else {
s.notifyWriteError(err)
}
}
atomic.AddUint64(&DefaultSnmp.OutPkts, uint64(npkts))
@@ -513,11 +530,11 @@ func (s *UDPSession) output(buf []byte) {
// kcp update, returns interval for next calling
func (s *UDPSession) update() (interval time.Duration) {
s.mu.Lock()
s.kcp.flush(false)
if s.kcp.WaitSnd() < int(s.kcp.snd_wnd) {
waitsnd := s.kcp.WaitSnd()
interval = time.Duration(s.kcp.flush(false)) * time.Millisecond
if s.kcp.WaitSnd() < waitsnd {
s.notifyWriteEvent()
}
interval = time.Duration(s.kcp.interval) * time.Millisecond
s.mu.Unlock()
return
}
@@ -539,56 +556,77 @@ func (s *UDPSession) notifyWriteEvent() {
}
}
func (s *UDPSession) notifyWriteError(err error) {
select {
case s.chWriteError <- err:
default:
}
}
func (s *UDPSession) kcpInput(data []byte) {
var kcpInErrors, fecErrs, fecRecovered, fecParityShards uint64
if s.fecDecoder != nil {
f := s.fecDecoder.decodeBytes(data)
s.mu.Lock()
if f.flag == typeData {
if ret := s.kcp.Input(data[fecHeaderSizePlus2:], true, s.ackNoDelay); ret != 0 {
kcpInErrors++
}
}
if len(data) > fecHeaderSize { // must be larger than fec header size
f := s.fecDecoder.decodeBytes(data)
if f.flag == typeData || f.flag == typeFEC { // header check
if f.flag == typeFEC {
fecParityShards++
}
recovers := s.fecDecoder.decode(f)
if f.flag == typeData || f.flag == typeFEC {
if f.flag == typeFEC {
fecParityShards++
}
s.mu.Lock()
waitsnd := s.kcp.WaitSnd()
if f.flag == typeData {
if ret := s.kcp.Input(data[fecHeaderSizePlus2:], true, s.ackNoDelay); ret != 0 {
kcpInErrors++
}
}
recovers := s.fecDecoder.decode(f)
for _, r := range recovers {
if len(r) >= 2 { // must be larger than 2bytes
sz := binary.LittleEndian.Uint16(r)
if int(sz) <= len(r) && sz >= 2 {
if ret := s.kcp.Input(r[2:sz], false, s.ackNoDelay); ret == 0 {
fecRecovered++
for _, r := range recovers {
if len(r) >= 2 { // must be larger than 2bytes
sz := binary.LittleEndian.Uint16(r)
if int(sz) <= len(r) && sz >= 2 {
if ret := s.kcp.Input(r[2:sz], false, s.ackNoDelay); ret == 0 {
fecRecovered++
} else {
kcpInErrors++
}
} else {
kcpInErrors++
fecErrs++
}
} else {
fecErrs++
}
} else {
fecErrs++
}
}
}
// notify reader
if n := s.kcp.PeekSize(); n > 0 {
s.notifyReadEvent()
// to notify the readers to receive the data
if n := s.kcp.PeekSize(); n > 0 {
s.notifyReadEvent()
}
// to notify the writers when queue is shorter(e.g. ACKed)
if s.kcp.WaitSnd() < waitsnd {
s.notifyWriteEvent()
}
s.mu.Unlock()
} else {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
}
} else {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
}
s.mu.Unlock()
} else {
s.mu.Lock()
waitsnd := s.kcp.WaitSnd()
if ret := s.kcp.Input(data, true, s.ackNoDelay); ret != 0 {
kcpInErrors++
}
// notify reader
if n := s.kcp.PeekSize(); n > 0 {
s.notifyReadEvent()
}
if s.kcp.WaitSnd() < waitsnd {
s.notifyWriteEvent()
}
s.mu.Unlock()
}
@@ -608,65 +646,52 @@ func (s *UDPSession) kcpInput(data []byte) {
}
}
func (s *UDPSession) receiver(ch chan<- []byte) {
for {
data := xmitBuf.Get().([]byte)[:mtuLimit]
if n, _, err := s.conn.ReadFrom(data); err == nil && n >= s.headerSize+IKCP_OVERHEAD {
select {
case ch <- data[:n]:
case <-s.die:
return
}
} else if err != nil {
s.chErrorEvent <- err
return
} else {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
}
}
}
// read loop for client session
// the read loop for a client session
func (s *UDPSession) readLoop() {
chPacket := make(chan []byte, qlen)
go s.receiver(chPacket)
buf := make([]byte, mtuLimit)
var src string
for {
select {
case data := <-chPacket:
raw := data
dataValid := false
if s.block != nil {
s.block.Decrypt(data, data)
data = data[nonceSize:]
checksum := crc32.ChecksumIEEE(data[crcSize:])
if checksum == binary.LittleEndian.Uint32(data) {
data = data[crcSize:]
dataValid = true
} else {
atomic.AddUint64(&DefaultSnmp.InCsumErrors, 1)
}
} else if s.block == nil {
dataValid = true
if n, addr, err := s.conn.ReadFrom(buf); err == nil {
// make sure the packet is from the same source
if src == "" { // set source address
src = addr.String()
} else if addr.String() != src {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
continue
}
if dataValid {
s.kcpInput(data)
if n >= s.headerSize+IKCP_OVERHEAD {
data := buf[:n]
dataValid := false
if s.block != nil {
s.block.Decrypt(data, data)
data = data[nonceSize:]
checksum := crc32.ChecksumIEEE(data[crcSize:])
if checksum == binary.LittleEndian.Uint32(data) {
data = data[crcSize:]
dataValid = true
} else {
atomic.AddUint64(&DefaultSnmp.InCsumErrors, 1)
}
} else if s.block == nil {
dataValid = true
}
if dataValid {
s.kcpInput(data)
}
} else {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
}
xmitBuf.Put(raw)
case <-s.die:
} else {
s.chReadError <- err
return
}
}
}
type (
sessionKey struct {
addr string
convID uint32
}
// Listener defines a server listening for connections
// Listener defines a server which will be waiting to accept incoming connections
Listener struct {
block BlockCrypt // block encryption
dataShards int // FEC data shard
@@ -674,120 +699,93 @@ type (
fecDecoder *fecDecoder // FEC mock initialization
conn net.PacketConn // the underlying packet connection
sessions map[sessionKey]*UDPSession // all sessions accepted by this Listener
chAccepts chan *UDPSession // Listen() backlog
chSessionClosed chan sessionKey // session close queue
headerSize int // the overall header size added before KCP frame
die chan struct{} // notify the listener has closed
rd atomic.Value // read deadline for Accept()
sessions map[string]*UDPSession // all sessions accepted by this Listener
sessionLock sync.Mutex
chAccepts chan *UDPSession // Listen() backlog
chSessionClosed chan net.Addr // session close queue
headerSize int // the additional header to a KCP frame
die chan struct{} // notify the listener has closed
rd atomic.Value // read deadline for Accept()
wd atomic.Value
}
// incoming packet
inPacket struct {
from net.Addr
data []byte
}
)
// monitor incoming data for all connections of server
func (l *Listener) monitor() {
// cache last session
var lastKey sessionKey
// a cache for session object last used
var lastAddr string
var lastSession *UDPSession
chPacket := make(chan inPacket, qlen)
go l.receiver(chPacket)
buf := make([]byte, mtuLimit)
for {
select {
case p := <-chPacket:
raw := p.data
data := p.data
from := p.from
dataValid := false
if l.block != nil {
l.block.Decrypt(data, data)
data = data[nonceSize:]
checksum := crc32.ChecksumIEEE(data[crcSize:])
if checksum == binary.LittleEndian.Uint32(data) {
data = data[crcSize:]
if n, from, err := l.conn.ReadFrom(buf); err == nil {
if n >= l.headerSize+IKCP_OVERHEAD {
data := buf[:n]
dataValid := false
if l.block != nil {
l.block.Decrypt(data, data)
data = data[nonceSize:]
checksum := crc32.ChecksumIEEE(data[crcSize:])
if checksum == binary.LittleEndian.Uint32(data) {
data = data[crcSize:]
dataValid = true
} else {
atomic.AddUint64(&DefaultSnmp.InCsumErrors, 1)
}
} else if l.block == nil {
dataValid = true
} else {
atomic.AddUint64(&DefaultSnmp.InCsumErrors, 1)
}
} else if l.block == nil {
dataValid = true
}
if dataValid {
var conv uint32
convValid := false
if l.fecDecoder != nil {
isfec := binary.LittleEndian.Uint16(data[4:])
if isfec == typeData {
conv = binary.LittleEndian.Uint32(data[fecHeaderSizePlus2:])
convValid = true
}
} else {
conv = binary.LittleEndian.Uint32(data)
convValid = true
}
if convValid {
key := sessionKey{
addr: from.String(),
convID: conv,
}
if dataValid {
addr := from.String()
var s *UDPSession
var ok bool
// packets received from an address always come in batch.
// the packets received from an address always come in batch,
// cache the session for next packet, without querying map.
if key == lastKey {
if addr == lastAddr {
s, ok = lastSession, true
} else if s, ok = l.sessions[key]; ok {
lastSession = s
lastKey = key
} else {
l.sessionLock.Lock()
if s, ok = l.sessions[addr]; ok {
lastSession = s
lastAddr = addr
}
l.sessionLock.Unlock()
}
if !ok { // new session
if len(l.chAccepts) < cap(l.chAccepts) && len(l.sessions) < 4096 { // do not let new session overwhelm accept queue and connection count
s := newUDPSession(conv, l.dataShards, l.parityShards, l, l.conn, from, l.block)
s.kcpInput(data)
l.sessions[key] = s
l.chAccepts <- s
if len(l.chAccepts) < cap(l.chAccepts) { // do not let the new sessions overwhelm accept queue
var conv uint32
convValid := false
if l.fecDecoder != nil {
isfec := binary.LittleEndian.Uint16(data[4:])
if isfec == typeData {
conv = binary.LittleEndian.Uint32(data[fecHeaderSizePlus2:])
convValid = true
}
} else {
conv = binary.LittleEndian.Uint32(data)
convValid = true
}
if convValid { // creates a new session only if the 'conv' field in kcp is accessible
s := newUDPSession(conv, l.dataShards, l.parityShards, l, l.conn, from, l.block)
s.kcpInput(data)
l.sessionLock.Lock()
l.sessions[addr] = s
l.sessionLock.Unlock()
l.chAccepts <- s
}
}
} else {
s.kcpInput(data)
}
}
} else {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
}
xmitBuf.Put(raw)
case key := <-l.chSessionClosed:
if key == lastKey {
lastKey = sessionKey{}
}
delete(l.sessions, key)
case <-l.die:
return
}
}
}
func (l *Listener) receiver(ch chan<- inPacket) {
for {
data := xmitBuf.Get().([]byte)[:mtuLimit]
if n, from, err := l.conn.ReadFrom(data); err == nil && n >= l.headerSize+IKCP_OVERHEAD {
select {
case ch <- inPacket{from, data[:n]}:
case <-l.die:
return
}
} else if err != nil {
return
} else {
atomic.AddUint64(&DefaultSnmp.InErrs, 1)
return
}
}
}
@@ -811,7 +809,10 @@ func (l *Listener) SetWriteBuffer(bytes int) error {
// SetDSCP sets the 6bit DSCP field of IP header
func (l *Listener) SetDSCP(dscp int) error {
if nc, ok := l.conn.(net.Conn); ok {
return ipv4.NewConn(nc).SetTOS(dscp << 2)
if err := ipv4.NewConn(nc).SetTOS(dscp << 2); err != nil {
return ipv6.NewConn(nc).SetTrafficClass(dscp)
}
return nil
}
return errors.New(errInvalidOperation)
}
@@ -864,13 +865,14 @@ func (l *Listener) Close() error {
}
// closeSession notify the listener that a session has closed
func (l *Listener) closeSession(key sessionKey) bool {
select {
case l.chSessionClosed <- key:
func (l *Listener) closeSession(remote net.Addr) (ret bool) {
l.sessionLock.Lock()
defer l.sessionLock.Unlock()
if _, ok := l.sessions[remote.String()]; ok {
delete(l.sessions, remote.String())
return true
case <-l.die:
return false
}
return false
}
// Addr returns the listener's network address, The Addr returned is shared by all invocations of Addr, so do not modify it.
@@ -898,9 +900,9 @@ func ListenWithOptions(laddr string, block BlockCrypt, dataShards, parityShards
func ServeConn(block BlockCrypt, dataShards, parityShards int, conn net.PacketConn) (*Listener, error) {
l := new(Listener)
l.conn = conn
l.sessions = make(map[sessionKey]*UDPSession)
l.sessions = make(map[string]*UDPSession)
l.chAccepts = make(chan *UDPSession, acceptBacklog)
l.chSessionClosed = make(chan sessionKey)
l.chSessionClosed = make(chan net.Addr)
l.die = make(chan struct{})
l.dataShards = dataShards
l.parityShards = parityShards
@@ -924,17 +926,22 @@ func Dial(raddr string) (net.Conn, error) { return DialWithOptions(raddr, nil, 0
// DialWithOptions connects to the remote address "raddr" on the network "udp" with packet encryption
func DialWithOptions(raddr string, block BlockCrypt, dataShards, parityShards int) (*UDPSession, error) {
// network type detection
udpaddr, err := net.ResolveUDPAddr("udp", raddr)
if err != nil {
return nil, errors.Wrap(err, "net.ResolveUDPAddr")
}
network := "udp4"
if udpaddr.IP.To4() == nil {
network = "udp"
}
udpconn, err := net.DialUDP("udp", nil, udpaddr)
conn, err := net.ListenUDP(network, nil)
if err != nil {
return nil, errors.Wrap(err, "net.DialUDP")
}
return NewConn(raddr, block, dataShards, parityShards, &connectedUDPConn{udpconn})
return NewConn(raddr, block, dataShards, parityShards, conn)
}
// NewConn establishes a session and talks KCP protocol over a packet connection.
@@ -949,6 +956,12 @@ func NewConn(raddr string, block BlockCrypt, dataShards, parityShards int, conn
return newUDPSession(convid, dataShards, parityShards, nil, conn, udpaddr, block), nil
}
// monotonic reference time point
var refTime time.Time = time.Now()
// currentMs returns current elasped monotonic milliseconds since program startup
func currentMs() uint32 { return uint32(time.Now().Sub(refTime) / time.Millisecond) }
func NewConnEx(convid uint32, connected bool, raddr string, block BlockCrypt, dataShards, parityShards int, conn *net.UDPConn) (*UDPSession, error) {
udpaddr, err := net.ResolveUDPAddr("udp", raddr)
if err != nil {
@@ -963,9 +976,6 @@ func NewConnEx(convid uint32, connected bool, raddr string, block BlockCrypt, da
return newUDPSession(convid, dataShards, parityShards, nil, pConn, udpaddr, block), nil
}
// returns current time in milliseconds
func currentMs() uint32 { return uint32(time.Now().UnixNano() / int64(time.Millisecond)) }
// connectedUDPConn is a wrapper for net.UDPConn which converts WriteTo syscalls
// to Write syscalls that are 4 times faster on some OS'es. This should only be
// used for connections that were produced by a net.Dial* call.

13
vendor/github.com/fatedier/kcp-go/updater.go generated vendored
View File

@@ -85,20 +85,19 @@ func (h *updateHeap) updateTask() {
h.mu.Lock()
hlen := h.Len()
now := time.Now()
for i := 0; i < hlen; i++ {
entry := heap.Pop(h).(entry)
if now.After(entry.ts) {
entry.ts = now.Add(entry.s.update())
heap.Push(h, entry)
entry := &h.entries[0]
if time.Now().After(entry.ts) {
interval := entry.s.update()
entry.ts = time.Now().Add(interval)
heap.Fix(h, 0)
} else {
heap.Push(h, entry)
break
}
}
if hlen > 0 {
timer = time.After(h.entries[0].ts.Sub(now))
timer = time.After(h.entries[0].ts.Sub(time.Now()))
}
h.mu.Unlock()
}

110
vendor/github.com/fatedier/kcp-go/xor.go generated vendored
View File

@@ -1,110 +0,0 @@
// Copyright 2013 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package kcp
import (
"runtime"
"unsafe"
)
const wordSize = int(unsafe.Sizeof(uintptr(0)))
const supportsUnaligned = runtime.GOARCH == "386" || runtime.GOARCH == "amd64" || runtime.GOARCH == "ppc64" || runtime.GOARCH == "ppc64le" || runtime.GOARCH == "s390x"
// fastXORBytes xors in bulk. It only works on architectures that
// support unaligned read/writes.
func fastXORBytes(dst, a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
w := n / wordSize
if w > 0 {
wordBytes := w * wordSize
fastXORWords(dst[:wordBytes], a[:wordBytes], b[:wordBytes])
}
for i := (n - n%wordSize); i < n; i++ {
dst[i] = a[i] ^ b[i]
}
return n
}
func safeXORBytes(dst, a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
ex := n % 8
for i := 0; i < ex; i++ {
dst[i] = a[i] ^ b[i]
}
for i := ex; i < n; i += 8 {
_dst := dst[i : i+8]
_a := a[i : i+8]
_b := b[i : i+8]
_dst[0] = _a[0] ^ _b[0]
_dst[1] = _a[1] ^ _b[1]
_dst[2] = _a[2] ^ _b[2]
_dst[3] = _a[3] ^ _b[3]
_dst[4] = _a[4] ^ _b[4]
_dst[5] = _a[5] ^ _b[5]
_dst[6] = _a[6] ^ _b[6]
_dst[7] = _a[7] ^ _b[7]
}
return n
}
// xorBytes xors the bytes in a and b. The destination is assumed to have enough
// space. Returns the number of bytes xor'd.
func xorBytes(dst, a, b []byte) int {
if supportsUnaligned {
return fastXORBytes(dst, a, b)
}
// TODO(hanwen): if (dst, a, b) have common alignment
// we could still try fastXORBytes. It is not clear
// how often this happens, and it's only worth it if
// the block encryption itself is hardware
// accelerated.
return safeXORBytes(dst, a, b)
}
// fastXORWords XORs multiples of 4 or 8 bytes (depending on architecture.)
// The arguments are assumed to be of equal length.
func fastXORWords(dst, a, b []byte) {
dw := *(*[]uintptr)(unsafe.Pointer(&dst))
aw := *(*[]uintptr)(unsafe.Pointer(&a))
bw := *(*[]uintptr)(unsafe.Pointer(&b))
n := len(b) / wordSize
ex := n % 8
for i := 0; i < ex; i++ {
dw[i] = aw[i] ^ bw[i]
}
for i := ex; i < n; i += 8 {
_dw := dw[i : i+8]
_aw := aw[i : i+8]
_bw := bw[i : i+8]
_dw[0] = _aw[0] ^ _bw[0]
_dw[1] = _aw[1] ^ _bw[1]
_dw[2] = _aw[2] ^ _bw[2]
_dw[3] = _aw[3] ^ _bw[3]
_dw[4] = _aw[4] ^ _bw[4]
_dw[5] = _aw[5] ^ _bw[5]
_dw[6] = _aw[6] ^ _bw[6]
_dw[7] = _aw[7] ^ _bw[7]
}
}
func xorWords(dst, a, b []byte) {
if supportsUnaligned {
fastXORWords(dst, a, b)
} else {
safeXORBytes(dst, a, b)
}
}