1 概述
在用C/C++編程的時候，經常會出現下面三種內存問題：
- 內存泄漏
- 懸掛指針
- 多次釋放同一塊內存
本系列文章簡要介紹排查這三個問題的工具和方法，先看看Valgrind
2 Valgrind
Valgrind是一款可以監測內存使用情況、監測內存泄漏的工具。對於一些規模不是很大的應用程序，Valgrind是一把利器。
3 內存泄漏監測
3.1 示例代碼
1: int main()
2: { 3: char *p = malloc(sizeof(char) * 10); 4: if (p == NULL) { 5: return 0; 6: } 7: 8: *p++ = 'a'; 9: *p++ = 'b'; 10: 11: printf("%s\n", *p); 12: 13: return 0; 14: }
3.2 編譯它
1: gcc -g -o core1 core1.c
3.3 用Valgrind監測進程的內存泄漏
1: valgrind --leak-check=yes --show-reachable=yes ./core
Valgrind的輸出為為：
1: ==25500== Memcheck, a memory error detector
2: ==25500== Copyright (C) 2002-2009, and GNU GPL'd, by Julian Seward et al. 3: ==25500== Using Valgrind-3.5.0 and LibVEX; rerun with -h for copyright info 4: ==25500== Command: ./core1 5: ==25500== 6: ==25500== Conditional jump or move depends on uninitialised value(s) 7: ==25500== at 0x36A104546A: vfprintf (in /lib64/libc-2.12.so) 8: ==25500== by 0x36A104EAC9: printf (in /lib64/libc-2.12.so) 9: ==25500== by 0x40055D: main (core1.c:13) 10: ==25500== 11: (null) 12: ==25500== 13: ==25500== HEAP SUMMARY: 14: ==25500== in use at exit: 10 bytes in 1 blocks 15: ==25500== total heap usage: 1 allocs, 0 frees, 10 bytes allocated 16: ==25500== 17: ==25500== 10 bytes in 1 blocks are definitely lost in loss record 1 of 1 18: ==25500== at 0x4A0515D: malloc (vg_replace_malloc.c:195) 19: ==25500== by 0x400515: main (core1.c:5) 20: ==25500== 21: ==25500== LEAK SUMMARY: 22: ==25500== definitely lost: 10 bytes in 1 blocks 23: ==25500== indirectly lost: 0 bytes in 0 blocks 24: ==25500== possibly lost: 0 bytes in 0 blocks 25: ==25500== still reachable: 0 bytes in 0 blocks 26: ==25500== suppressed: 0 bytes in 0 blocks 27: ==25500== 28: ==25500== For counts of detected and suppressed errors, rerun with: -v 29: ==25500== Use --track-origins=yes to see where uninitialised values come from 30: ==25500== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 6 from 6)
可以看到，Valgrind提示在第五行分配的內存未被釋放
4 懸掛指針
4.1 示例代碼
1: struct elem {
2: int a; 3: double b; 4: }; 5: 6: int main() 7: { 8: struct elem *e = malloc(sizeof(struct elem)); 9: if (e == NULL) { 10: return 0; 11: } 12: 13: e->a = 10; 14: e->b = 10.10; 15: 16: double *xx = &e->b; 17: 18: printf("%f\n", *xx); 19: 20: free(e); 21: 22: printf("%f\n", *xx); 23: 24: return 0; 25: }
4.2 Valgrind運行結果
同樣用-g編譯后valgrind運行的結果：
1: [cobbliu@MacBook]$ valgrind --leak-check=yes --show-reachable=yes ./core2
2: ==26148== Memcheck, a memory error detector 3: ==26148== Copyright (C) 2002-2009, and GNU GPL'd, by Julian Seward et al. 4: ==26148== Using Valgrind-3.5.0 and LibVEX; rerun with -h for copyright info 5: ==26148== Command: ./core2 6: ==26148== 7: 10.100000 8: ==26148== Invalid read of size 8 9: ==26148== at 0x4005CA: main (core2.c:26) 10: ==26148== Address 0x502a048 is 8 bytes inside a block of size 16 free'd 11: ==26148== at 0x4A04D72: free (vg_replace_malloc.c:325) 12: ==26148== by 0x4005C5: main (core2.c:24) 13: ==26148== 14: 10.100000 15: ==26148== 16: ==26148== HEAP SUMMARY: 17: ==26148== in use at exit: 0 bytes in 0 blocks 18: ==26148== total heap usage: 1 allocs, 1 frees, 16 bytes allocated 19: ==26148== 20: ==26148== All heap blocks were freed -- no leaks are possible 21: ==26148== 22: ==26148== For counts of detected and suppressed errors, rerun with: -v 23: ==26148== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 6 from 6)
可以看到在free(e)后，指針xx成為了懸掛指針，此后對xx的讀，如果xx指向的內存還未被glibc回收，進程不會core掉。valgrind提示在26行做了對xx的 Invalid read.
5 多次釋放同一個指針
5.1 示例代碼
1: int main()
2: { 3: char *p = malloc(sizeof(char) * 10); 4: if (p == NULL) { 5: return 0; 6: } 7: 8: char *q = p; 9: 10: *p++ = 'a'; 11: *p++ = 'b'; 12: 13: printf("%s\n", *p); 14: 15: free(p); 16: free(q); 17: return 0; 18: }
5.2 Valgrind 監測
1: [cobbliu@MacBook]$ valgrind --leak-check=yes --show-reachable=yes ./core1
2: ==26874== Memcheck, a memory error detector 3: ==26874== Copyright (C) 2002-2009, and GNU GPL'd, by Julian Seward et al. 4: ==26874== Using Valgrind-3.5.0 and LibVEX; rerun with -h for copyright info 5: ==26874== Command: ./core1 6: ==26874== 7: ==26874== Conditional jump or move depends on uninitialised value(s) 8: ==26874== at 0x36A104546A: vfprintf (in /lib64/libc-2.12.so) 9: ==26874== by 0x36A104EAC9: printf (in /lib64/libc-2.12.so) 10: ==26874== by 0x4005B5: main (core1.c:15) 11: ==26874== 12: (null) 13: ==26874== Invalid free() / delete / delete[] 14: ==26874== at 0x4A04D72: free (vg_replace_malloc.c:325) 15: ==26874== by 0x4005C1: main (core1.c:17) 16: ==26874== Address 0x502a042 is 2 bytes inside a block of size 10 alloc'd 17: ==26874== at 0x4A0515D: malloc (vg_replace_malloc.c:195) 18: ==26874== by 0x400565: main (core1.c:5) 19: ==26874== 20: ==26874== 21: ==26874== HEAP SUMMARY: 22: ==26874== in use at exit: 0 bytes in 0 blocks 23: ==26874== total heap usage: 1 allocs, 2 frees, 10 bytes allocated 24: ==26874== 25: ==26874== All heap blocks were freed -- no leaks are possible 26: ==26874== 27: ==26874== For counts of detected and suppressed errors, rerun with: -v 28: ==26874== Use --track-origins=yes to see where uninitialised values come from 29: ==26874== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 6 from 6)
可以看到，valgrind提示在17行有一個Invalid Free()
6 Valgrind的優缺點
valgrind默認使用memcheck工具做內存監測。
6.1 Advantages
用valgrind監測內存泄漏，不用重新編譯應用程序，不用重新鏈接應用程序，不用對應用進程做任何修改。如果想查看詳細的出錯信息，只需要在編譯時加上-g選項。
6.2 Disadvantages
不管valgrind在使用memcheck工具監測內存時，它會接管應用程序，並且讀取應用程序可執行文件和庫文件中的debug信息來顯示詳細的出錯位置。當valgrind啟動后，應用 進程實際上在valgrind的虛擬環境中執行，valgrind會將每行代碼傳遞給memcheck工具，memcheck工具再加入自己的調試信息，之后再將合成的代碼真正運行。memcheck工具在 應用進程每個防存操作和每個變量賦值操作時加入額外的統計代碼，通常情況下，使用memcheck工具后應用程序的運行時間會比原生代碼慢大約10-50倍。
其次，對於一些不停機運行的服務器程序的內存問題，valgrind無能為力。不僅僅是因為valgrind無法使之停止，還有可能是因為服務器進程本身就被設計為申請一些生命周期 與進程生命周期一樣長的內存，永遠不釋放，這些內存會被valgrind報泄漏錯誤。
再次，valgrind對多線程程序支持得不夠好。在多線程程序執行時，valgrind在同一時刻只讓其中一個線程執行，它不會充分利用多核的環境。在用valgrind運行您的多線程程序 時，您的寶貴程序的運行情況可能跟不使用valgrind的運行情況千差萬別。
7 Valgrind的其他工具
除了memcheck工具外，valgrind工具包還有一些別的好用的工具
7.1 Cachegrind
這個工具模擬 CPU中的一級緩存I1,D1和L2二級緩存，能夠精確地指出程序中 cache的丟失和命中。如果需要，它還能夠為我們提供cache丟失次數，內存引用次數，以及每行 代碼，每個函數，每個模塊，整個程序產生的指令數。這對優化程序有很大的幫助。 詳情見這里
7.2 Callgrind
Callgrind收集程序運行時的一些數據，函數調用關系等信息，還可以有選擇地進行cache 模擬。在運行結束時，它會把分析數據寫入一個文件。callgrindannotate可以把 這個文件的內容轉化成可讀的形式。 詳情見這里
7.3 Helgrind
它主要用來檢查C/C++多線程程序（使用POSIX線程）中出現的同步問題。Helgrind 尋找內存中被多個線程訪問，而又沒有一貫加鎖的區域，這些區域往往是線程之間失去同 步的地方，而且會導致難以發掘的錯誤。Helgrind實現了名為"Eraser" 的競爭檢測算法，並做了進一步改進，減少了報告錯誤的次數。 詳情見這里
7.4 DRD
這也使一款多線程程序監測工具，它提供的監測信息比Helgrind更豐富。 詳情見這里
7.5 Massif
堆棧分析器，它能測量程序在堆棧中使用了多少內存。告訴我們堆塊，堆管理塊和棧的大小。對於那些被應用進程釋放但是還沒有交還給操作系統的內存，memcheck是監測 不出來的，而Massif能有效第監測到這類內存。 詳情見這里
7.6 DHAT
這個工具能詳細地顯示應用進程如何使用堆棧，以使用戶更好地評估程序的設計。 詳情見這里