VB.net 2010 视频教程 VB.net 2010 视频教程 python基础视频教程
SQL Server 2008 视频教程 c#入门经典教程 Visual Basic从门到精通视频教程
当前位置:
首页 > c#编程 >
  • 记一次 .NET某报关系统 非托管泄露分析

一:背景

1. 讲故事

前段时间有位朋友找到我,说他的程序内存会出现暴涨,让我看下是怎么事情?而且还告诉我是在 Linux 环境下,说实话在Linux上分析.NET程序难度会很大,难度大的原因在于Linux上的各种开源工具主要是针对 C/C++, 和 .NET 一毛钱关系都没有,说到底微软在 Linux 上的调试领域支持度还远远不够。

虽然知道分析起来难度可能会很大,但该分析还是要分析的,让朋友抓一个 dump 过来,上 WinDbg 说话。

二:WinDbg 分析

1. 到底是哪里的泄露

只要是进程都会有内存段的,所以分析Linux的dump一样可以使用 !address -summary 命令来观察。


0:000> !address -summary

--- Usage Summary ---------------- RgnCount ----------- Total Size -------- %ofBusy %ofTotal
<unknown>                              1607 ffffffff`cd7a9e00 (  16.000 EB) 100.00%  100.00%
Image                                 41699        0`31e57200 ( 798.340 MB)   0.00%    0.00%

--- Type Summary (for busy) ------ RgnCount ----------- Total Size -------- %ofBusy %ofTotal
                                       1247 fffffffe`1c910000 (  16.000 EB)          100.00%
MEM_PRIVATE                           42059        1`e2cf1000 (   7.544 GB)   0.00%    0.00%

--- State Summary ---------------- RgnCount ----------- Total Size -------- %ofBusy %ofTotal
                                       1247 fffffffe`1c910000 (  16.000 EB) 100.00%  100.00%
MEM_COMMIT                            42059        1`e2cf1000 (   7.544 GB)   0.00%    0.00%

--- Protect Summary (for commit) - RgnCount ----------- Total Size -------- %ofBusy %ofTotal
PAGE_READWRITE                        41067        1`cff54000 (   7.249 GB)   0.00%    0.00%
PAGE_READONLY                           644        0`07268000 ( 114.406 MB)   0.00%    0.00%
PAGE_EXECUTE_READ                       223        0`06d1f000 ( 109.121 MB)   0.00%    0.00%
PAGE_EXECUTE_WRITECOPY                  125        0`04e16000 (  78.086 MB)   0.00%    0.00%

--- Largest Region by Usage ----------- Base Address -------- Region Size ----------
<unknown>                              7ffc`78f8e000 ffff8003`86672000 (  16.000 EB)
Image                                  7ff8`49102000        0`10000000 ( 256.000 MB)

这里简单提一下,我发现有很多朋友搞不清楚这里的 16.000 EB 是什么意思,它其实是2的64次方,即程序的用户态空间的寻址范围。

从卦中的 MEM_COMMIT=7.544G 来看当前提交内存不小,接下来用 !eeheap -gc 观察下托管堆内存占用。


0:000> !eeheap -gc

========================================
Number of GC Heaps: 1
----------------------------------------
generation 0 starts at 7ff688f78f10
generation 1 starts at 7ff688484e70
generation 2 starts at 7ff8f7fff000
ephemeral segment allocation context: none
Small object heap
         segment            begin        allocated        committed allocated size         committed size        
    7ff63fffa000     7ff63fffb000     7ff64fe51d80     7ff64fe5d000 0xfe56d80 (266694016)  0xfe63000 (266743808) 
    7ff6772c8000     7ff6772c9000     7ff6872c2d38     7ff6872c8000 0xfff9d38 (268410168)  0x10000000 (268435456)
    7ff74bffe000     7ff74bfff000     7ff75bffdfc0     7ff75bffe000 0xfffefc0 (268431296)  0x10000000 (268435456)
    7ff773ffe000     7ff773fff000     7ff783ffdfc8     7ff783ffe000 0xfffefc8 (268431304)  0x10000000 (268435456)
    7ff849102000     7ff849103000     7ff859101fe8     7ff859102000 0xfffefe8 (268431336)  0x10000000 (268435456)
    7ff8f7ffe000     7ff8f7fff000     7ff907ffce88     7ff907ffe000 0xfffde88 (268426888)  0x10000000 (268435456)
    7ff6872ca000     7ff6872cb000     7ff68a768438     7ff68aa4b000 0x349d438 (55170104)   0x3781000 (58200064)  
Large object heap starts at 7ff907fff000
         segment            begin        allocated        committed allocated size         committed size        
    7ff733ff8000     7ff733ff9000     7ff73aedd058     7ff73aefe000 0x6ee4058 (116277336)  0x6f06000 (116416512) 
    7ff743ffc000     7ff743ffd000     7ff744358f10     7ff744379000 0x35bf10 (3522320)     0x37d000 (3657728)    
    7ff7a3ffe000     7ff7a3fff000     7ff7a9d63ee0     7ff7a9d84000 0x5d64ee0 (97930976)   0x5d86000 (98066432)  
    7ff7bbffe000     7ff7bbfff000     7ff7c3dc1090     7ff7c3de2000 0x7dc2090 (131866768)  0x7de4000 (132005888) 
    7ff907ffe000     7ff907fff000     7ff90f048b30     7ff90f069000 0x7049b30 (117742384)  0x706b000 (117878784) 
Pinned object heap starts at 7ff90ffff000
         segment            begin        allocated        committed allocated size         committed size        
    7ff90fffe000     7ff90ffff000     7ff9102d15b0     7ff9102d2000 0x2d25b0 (2958768)     0x2d4000 (2965504)    
------------------------------
GC Allocated Heap Size:    Size: 0x7f36bca0 (2134293664) bytes.
GC Committed Heap Size:    Size: 0x7f710000 (2138112000) bytes.

从卦中看当前提交内存也仅有 2.13G,这和 7.5G 相距甚远,说明这是最复杂的 非托管内存泄漏

2. 非托管泄露分析

作为一个.NET调试者,需要像医生一样尽自己最大可能救治病人,那接下来我们的研究方向在哪里呢?大家需要知道所有的内存占用的基本盘都在 虚拟地址 上,结果一搜索,发现有大概 4w+ 的 dll,一个程序怎么可能会有这么多动态链接库呢?截图如下:

既然找到了可疑之处那就继续挖吧,接下来就是要考虑这个dll是托管代码创建的还是非托管代码创建的,用排除法就好了,如果是托管代码创建的,那就肯定属于 Assembly 下的某一个module,可以查下加载堆看看。


0:000> !eeheap -loader
Loader Heap:
--------------------------------------
...
Module 00007ff95e265778: Size: 0x0 (0) bytes.
Module 00007ff95e2661e0: Size: 0x0 (0) bytes.
Module 00007ff95e266c48: Size: 0x0 (0) bytes.
Module 00007ff95e2676b0: Size: 0x0 (0) bytes.
Module 00007ff95e268118: Size: 0x0 (0) bytes.
Module 00007ff95e268b80: Size: 0x0 (0) bytes.
Module 00007ff95e2695e8: Size: 0x0 (0) bytes.
Total size:      Size: 0x0 (0) bytes.
--------------------------------------
Total LoaderHeap size:   Size: 0x4bf4b000 (1274327040) bytes total, 0x4da000 (5087232) bytes wasted.
=======================================

0:000> !dumpmodule 00007ff95e2695e8
Name: *75db8939-8b3a-4075-94ac-e9bb52acf9d1#40147-0.dll
Attributes:              PEFile IsInMemory IsFileLayout 
...
MetaData start address:  00007FF85BBCD330 (2068 bytes)

0:000> !DumpAssembly /d 00007ff80d71bba0
Parent Domain:      0000559009249080
Name:               Unknown
ClassLoader:        00007FF80D71BC00
  Module
  00007ff95e2695e8    *75db8939-8b3a-4075-94ac-e9bb52acf9d1#40147-0.dll

从卦中看虽然加载堆只有 1.27G,但它还有很多关联的内存,而且动态module高多4w+,接下来用 !dumpmodule -mt 观察内部是什么类型。


0:000> !dumpmodule -mt 00007ff95e2695e8
Name: *75db8939-8b3a-4075-94ac-e9bb52acf9d1#40147-0.dll
Attributes:              PEFile IsInMemory IsFileLayout 
...
Types defined in this module

              MT          TypeDef Name
------------------------------------------------------------------------------
00007ff95e269d80 0x02000004 Submission#0
00007ff95e269ec0 0x02000005 Submission#0+<>d__0

Types referenced in this module

              MT            TypeRef Name
------------------------------------------------------------------------------
00007ff92d6152a8 0x0200000c System.Object
00007ff92ead4d18 0x0200000d System.Threading.Tasks.Task`1
00007ff93133f298 0x0200000e System.Runtime.CompilerServices.IAsyncStateMachine
00007ff92d6cec90 0x0200000f System.Exception
00007ff93133f648 0x02000010 System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1

0:000> !dumpmt -md 00007ff95e269ec0
...
MethodDesc Table
           Entry       MethodDesc    JIT Name
00007FF92D620030 00007ff92d615238    JIT System.Object.Finalize()
00007FF92D620038 00007ff92d615248 PreJIT System.Object.ToString()
00007FF92D620040 00007ff92d615258    JIT System.Object.Equals(System.Object)
00007FF92D620058 00007ff92d615298    JIT System.Object.GetHashCode()
00007FF95DFFB0E0 00007ff95e269e58    JIT Submission#0+<>d__0.MoveNext()
00007FF95DFF9B70 00007ff95e269e78   NONE Submission#0+<>d__0.SetStateMachine(System.Runtime.CompilerServices.IAsyncStateMachine)
00007FF95DFF9B60 00007ff95e269e48    JIT Submission#0+<>d__0..ctor()

从卦中看也只能看到一些 Submission 为前缀的类与之相关的状态机类,也看不出来是谁创建的,结果又入了困境。

3. 到底是谁作的孽

要想获取动态程序集的创建事件,有一个好办法就是用跨平台的 dotnet-trace,让它捕获程序集的加载事件即可,详情可参考:https://learn.microsoft.com/en-us/dotnet/core/dependency-loading/collect-details ,然后让朋友跑 30min 看看,参考命令如下:


dotnet-trace collect -p 4108 --clrevents loader --duration 00:00:30:00

有了生成好的 dotnet_xxxx.nettrace 之后就可以用 perfview 观察了,打开 Event视图,搜索 AssemblyLoad 事件,截图如下:

通过 Time MSec 的748前缀来看,这1s种能生成几十个动态程序集,接下来右键选择 Open Any Stacks 观察是什么代码调用的,截图如下:

从 perfivew 的输出看,原来是 XXXCusDis 方法内部调用 Microsoft.CodeAnalysis.CSharp.Scripting.CSharpScript.EvaluateAsync 生成了非常多的程序集。

最后就是把 CSharpScript.EvaluateAsync 告诉朋友,能不能给剔除掉做个排查?

三:总结

网上查了下 Microsoft.CodeAnalysis.CSharp.Scripting 可以用来生成C#脚本代码,大家在用的时候小心点吧。


出处:https://www.cnblogs.com/huangxincheng/p/17614463.html


相关教程