ltrace能够跟踪进程的库函数调用,它会显现出哪个库函数被调用,而strace则是跟踪程序的每个系统调用.
下面是一个ltrace与strace的对比
1)系统调用的输出对比
我们用输出hello world的程序做如下测试:
#include <stdio.h>
int
main ()
{
printf("Hello world!
");
return 0;
}
gcc hello.c -o hello
用ltrace跟踪hello程序,如下:
ltrace ./hello
__libc_start_main(0x8048354, 1, 0xbf869aa4, 0x8048390, 0x8048380 <unfinished ...>
puts("Hello world!"Hello world!
) = 13
+++ exited (status 0) +++
注:我们看到程序调用了puts();库函数做了输出.
用strace跟踪hello程序,如下:
strace ./hello
execve("./hello", ["./hello"], [/* 30 vars */]) = 0
brk(0) = 0x83d4000
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7f8a000
access("/etc/ld.so.preload", R_OK) = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY) = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=80846, ...}) = 0
mmap2(NULL, 80846, PROT_READ, MAP_PRIVATE, 3, 0) = 0xb7f76000
close(3) = 0
open("/lib/libc.so.6", O_RDONLY) = 3
read(3, "177ELF111 3 3 1 000?270"..., 512) = 512
fstat64(3, {st_mode=S_IFREG|0755, st_size=1576952, ...}) = 0
mmap2(0xb6e000, 1295780, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0xb6e000
mmap2(0xca5000, 12288, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x137) = 0xca5000
mmap2(0xca8000, 9636, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0xca8000
close(3) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7f75000
set_thread_area({entry_number:-1 -> 6, base_addr:0xb7f756c0, limit:1048575, seg_32bit:1, contents:0, read_exec_only:0, limit_in_pages:1, seg_not_present:0, useable:1}) = 0
mprotect(0xca5000, 8192, PROT_READ) = 0
mprotect(0xb6a000, 4096, PROT_READ) = 0
munmap(0xb7f76000, 80846) = 0
fstat64(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(136, 0), ...}) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7f89000
write(1, "Hello world!
", 13Hello world!
) = 13
exit_group(0) = ?
Process 2874 detached
注:我们看到程序调用write()系统调用做了输出,同时strace还把hello程序运行时所做的系统调用都打印出来了.
同样的ltrace也可以把系统调用都打印出来,如下:
ltrace -S ./hello
SYS_execve(NULL, NULL, NULL) = 0xffffffda
SYS_brk(NULL) = -38
SYS_mmap2(0, 4096, 3, 34, -1) = -38
SYS_access(0xb6798f, 4, 0xb6afc0, 0, 0xb6b6b4) = -38
SYS_open("/etc/ld.so.cache", 0, 00) = -38
SYS_fstat64(3, 0xbfba5414, 0xb6afc0, -1, 3) = -38
SYS_mmap2(0, 80846, 1, 2, 3) = -38
SYS_close(3) = -38
SYS_open("/lib/libc.so.6", 0, 027756452364???, 512) = -38
SYS_read(3, ) = -38
SYS_fstat64(3, 0xbfba5478, 0xb6afc0, 4, 1) = -38
SYS_mmap2(0xb6e000, 0x13c5a4, 5, 2050, 3) = -38
SYS_mmap2(0xca5000, 12288, 3, 2066, 3) = -38
SYS_mmap2(0xca8000, 9636, 3, 50, -1) = -38
SYS_close(3) = -38
SYS_mmap2(0, 4096, 3, 34, -1) = -38
SYS_set_thread_area(0xbfba5960, 0xb7f5e6c0, 243, 0xb6afc0, 0) = -38
SYS_mprotect(0xca5000, 8192, 1, 7676, 0xca6e74) = -38
SYS_mprotect(0xb6a000, 4096, 1, 896, 0) = -38
SYS_munmap(0xb7f5f000, 80846 <unfinished ...>
__libc_start_main(0x8048354, 1, 0xbfba5dd4, 0x8048390, 0x8048380 <unfinished ...>
puts("Hello world!" <unfinished ...>
SYS_fstat64(1, 0xbfba5c20, 0xca6ff4, 0xca74c0, 0xca74c0) = 0
SYS_mmap2(0, 4096, 3, 34, -1) = 0xb7f72000
SYS_write(1, "Hello world!
", 13Hello world!
) = 13
<... puts resumed> ) = 13
SYS_exit_group(0 <no return ...>
+++ exited (status 0) +++
注:我们看到它实际是用SYS_write系统调用来做打印输出,其实write()函数是SYS_write的封装,SYS_write是真正的系统调用.
二)ltrace/strace的耗时
ltrace -c dd if=/dev/urandom of=/dev/null count=1000
1000+0 records in
1000+0 records out
512000 bytes (512 kB) copied, 2.31346 seconds, 221 kB/s
% time seconds usecs/call calls function
------ ----------- ----------- --------- --------------------
84.88 4.942763 4942 1000 read
9.41 0.548195 548 1000 write
5.06 0.294716 294 1001 memcpy
0.11 0.006365 2121 3 __fprintf_chk
0.09 0.004969 4969 1 dcgettext
0.08 0.004850 808 6 strlen
0.05 0.002667 2667 1 setlocale
0.04 0.002579 644 4 sigaction
0.03 0.001869 467 4 close
0.03 0.001825 912 2 open64
0.03 0.001519 759 2 malloc
0.02 0.001187 593 2 __sprintf_chk
0.02 0.001176 588 2 clock_gettime
0.02 0.001169 389 3 __errno_location
0.02 0.001012 506 2 dcngettext
0.01 0.000814 814 1 lseek64
0.01 0.000757 757 1 getopt_long
0.01 0.000744 744 1 textdomain
0.01 0.000742 247 3 strchr
0.01 0.000634 634 1 __strtoull_internal
0.01 0.000602 602 1 getpagesize
0.01 0.000542 271 2 localeconv
0.01 0.000340 340 1 fclose
0.01 0.000300 300 1 memmove
0.00 0.000228 114 2 sigismember
0.00 0.000184 184 1 getenv
0.00 0.000170 85 2 sigaddset
0.00 0.000148 74 2 free
0.00 0.000093 93 1 bindtextdomain
0.00 0.000090 90 1 sigemptyset
0.00 0.000090 90 1 __cxa_atexit
0.00 0.000088 88 1 __ctype_b_loc
0.00 0.000074 74 1 __fpending
------ ----------- ----------- --------- --------------------
100.00 5.823501 3057 total
注:
使用-c选项,ltrace输出由进程创建的库调用,输出结果以调用过程的时间为准进行排序,因为是从urandom设备上读,这是一种产生随机数的设备,完成后,写入null设备.
所以读过程花费了较多的时间.
使用ltrace去捕获运行时函数,就好像在进程上系上了一个调试工具,它占据了ltrace大量的时间,这里ltrace一共消耗了5.8秒
我们再来看一下strace所花费的时间,如下:
strace -c dd if=/dev/urandom of=/dev/null count=1000
1000+0 records in
1000+0 records out
512000 bytes (512 kB) copied, 0.894482 seconds, 572 kB/s
Process 3049 detached
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
82.85 0.159393 159 1005 read
15.07 0.028995 29 1003 write
0.78 0.001494 1494 1 execve
0.42 0.000814 136 6 rt_sigaction
0.23 0.000446 41 11 1 close
0.23 0.000435 73 6 fstat64
0.21 0.000412 32 13 mmap2
0.21 0.000408 29 14 6 open
0.00 0.000000 0 1 1 access
0.00 0.000000 0 3 brk
0.00 0.000000 0 2 munmap
0.00 0.000000 0 1 uname
0.00 0.000000 0 4 mprotect
0.00 0.000000 0 1 _llseek
0.00 0.000000 0 1 rt_sigprocmask
0.00 0.000000 0 1 getrlimit
0.00 0.000000 0 1 set_thread_area
0.00 0.000000 0 1 set_tid_address
0.00 0.000000 0 2 clock_gettime
0.00 0.000000 0 1 set_robust_list
------ ----------- ----------- --------- --------- ----------------
100.00 0.192397 2078 8 total
注:
strace一共消耗了0.19秒,strace把性能提升了30倍,这主要是strace在跟踪系统调用的时候不需要动态库,而ltrace是根据动态库来分析程序运行的.
所以ltrace也只能跟踪动态库,不能跟踪静态库.
事实上我们用ltrace和strace都可以发现程序在哪个系统调用时发生了性能瓶径.
ltrace用-T,而strace也用-T.
三)ltrace与strace的相同点
ltrace与strace都可以指定PID,即对运行中的程序进行跟踪.
ltrace -p PID与strace -p PID
ltrace与strace都可以跟踪程序fork或clone子进程.
ltrace是用-f参数,而strace是用-f(fork/clone)和-F(vfork).