Never use syslog from within signal handler

今天调试程序的时候发现worker进程收到信号后就似乎“不工作”了。gdb attach上去一看:

原来是锁住了。Google “syslog  signal handler” ,看到:

Non-reentrant functions are functions that cannot safely be called, interrupted, and then recalled before the first call has finished without resulting in memory corruption. This can lead to an unexpected system state,an unpredictable result with a variety of potential consequences depending on context, including denial of service and code execution.

Many functions are not reentrant, but some of them can result in the corruption of memory if they are used in a signal handler. The function call syslog() is an example of this. In order to perform its functionality, it allocates a small amount of memory as “scratch space.” If syslog() is suspended by a signal call and the signal handler calls syslog(), the memory used by both of these functions enters an undefined, and possibly, exploitable state. Implementations of malloc() and free() manage metadata in global structures in order to track which memory is allocated versus which memory is available, but they are non-reentrant. Simultaneous calls to these functions can cause corruption of the metadata.

再翻开TLPI或者APUE.3e中关于signal handler,reentrant and async-signal-safe functions的解释,于这一点上便更明白了。

  • APUE.3e       10.6           Reentrant Functions
  • TLPI               21.1.2        Reentrant and Async-Signal-Safe Functions