ID: Q117223
The information in this article applies to:
It is possible for a thread in a multithreaded Win32-based application to block while doing a LockFile() or LockFileEx() API call, even when the region of the file the thread has requested is not locked by another thread.
NOTE: If performance-monitoring tools (such as PERFMON) are used to examine the status of existing threads and the Thread State indicates that the thread is waiting and the Thread Wait Reason shows that it is the executive that the thread is waiting on, this is probably not an indication that a deadlock has occurred, because threads are often in this state for other reasons. Also, if the Thread State indicates that the thread is not waiting, then a deadlock has probably not occurred.
There is a small window of time during which a multithreaded application is vulnerable to this condition. Specifically, if one thread (call it Thread1) is in the process of unlocking a currently locked byte range within a file while a second thread (Thread2) is in the process of obtaining a lock on that same byte range using the same file handle and without specifying the flag LOCKFILE_FAIL_IMMEDIATELY, Thread1 can block, waiting for the region to become available. Ordinarily, when unlocking takes place, blocked threads are released; but in this critical window of time, it is possible for Thread2 to unlock the byte range without Thread1 being released. Thus, Thread1 never resumes operation despite the fact that there is no apparent fault in the logic of the program.
The deadlock condition described above can only come about if multiple threads are concurrently doing synchronous I/O using the same file handle.
To avoid the problem, you have three options:
-or-
-or-
For more information about threads, files, and file handles, see the following sections in the "Win32 SDK Programmer's Reference," volume 2, part 3, "System Services":
Keywords : kbprg kbnokeyword kbKernBase kbGrpKernBase
Version : 3.10 3.50
Platform : NT WINDOWS
Last Reviewed: April 20, 1997