Last modified: Oct. 13, 2001
Please send comments and suggestions directly to me: jmhart@world.std.com.
ADDITIONAL INFORMATION IF YOU ARE USING VISUAL C++ Version 5.0
Author Note: The book was developed primarily using Version 4.0, and Version 5.0 added to the supported API. Several comments mention new features available with Version 5.0, and I will add new material here from time to time. Please feel free to send me additional information at jmhart@world.std.com.
p. 233
The statement regarding critical section performance is not entirely correct; CSs can be slower than mutexes in some circumstances. See details.
p. 281 From Cagdas Ozgenc <cagdasozgenc@hotmail.com>. When performing asynchronous I/O with threads, it is important to note that you cannot have multiple threads use the same file HANDLE, unless the HANDLE is opened in overlapped mode. A thread that is blocked performing I/O on a thread will lock the HANDLE, blocking all other threads that attempt to use the HANDLE concurrently.
There are two solutions:
Note: I have not tested solution one thoroughly on Windows NT/2000/XP, but will do so soon.
p. 42
See atou for additional atou performance comparisons that modify and extend some of the conclusions in the text.
p. 88
The declaration of the variable Exit should be:
This is required as the console control handler runs in a separate thread; see Chapter 11 for a more complete explanation. The code on the disc has this fix.
p. 199 Cagdas Ozgenc cagdasozgenc@hotmail.com pointed out that SrvrBcst.exe (Program 9-4) fails on Windows 9x/Me. The reason is that CreateFile (..., GENERIC_WRITE | GENERIC_READ, ...) should not use GENERIC_READ (the mailslot client only writes to the mailslot). Windows NT/2000/XP, on the other hand, will execute SrvrBcst.exe properly as is.
This bug was never detected because the program is used with the named pipe server and was therefore never tested on Windows 9x/Me. However, there are many situations in which a mailslot client is appropriate on these systems. For example, the client application could have been written to act as a mailslot client, with the named pipe server being the mailslot server.
Comment: I have not retested on Windows NT/2000/XP to determine whether or not the GENERIC_READ flag is required; I suspect that it is not.
p. 228
VC++ Ver 5.0 ONLY - You can test a critical section with the TryCriticalSection function, which will take ownership of the critical section if possible, but will return a FALSE if a different thread owns the critical section. Thus, you can poll a critical section, but you cannot time out.
p. 233
Here is an additional comparison of mutexes with CRITICAL_SECTION objects. Mutex handles can participate in a WaitForMultipleObjects call (the handles in the array can be a mix of any synchronization object handles). This is desirable at times, and careful use of WaitForMultipleObjects can help to avoid the deadlocks that might occur if you waited for the individual objects sequentially. You can only wait for one single CRITICAL_SECTION object at a time, however, and you must use EnterCriticalSection.
p. 234
The remarks on p. 232 regarding names apply to semaphores as well. Creating a semaphore with the name of an existing semaphore means that the initial and maximum counts are ignored.
This applies to CreateMutex, etc.
p. 237
The remarks on p. 232 regarding names applies to events (and semaphores) as well, but, of course, the initial state and the auto/manual reset properties cannot be changed.
p. 246
case CS_RQCOMPLETE is missing a closing bracket } for the if statement. It should go just before the closing bracket for the _try statement and have its own line.
p. 248
The utilization computation on the seventh line is inaccurate because of the way in which it is parenthesized. As it is, the utilization will stay at zero. The fix will allow the multiplication by 100 to precede the division. Alternatively, you could cast the operands to floating point. Here is an improved expression for the utilization.
Utilization = (LONG)(100 *
(TotKerTim.li - OldKerTim +
TotUsrTim.li - OldUsrTim) / (10000 *
CS_TIMEOUT));
p. 293
Eliminate the phrase at the end of the first sentence of Paragraph 4, so that the first sentence now reads:
Obtain the names of subkeys by specifying a key to
RegEnumKeyEx.Likewise, remove the last sentence in the second to last paragraph. The sentence to be removed currently reads "RegistryQueryInfoKey is ... index." Notice that there is a function RegQueryInfoKey, but its job is to retrieve information regarding a specified key; among other things, it will find the number of subkeys.
See atou for additional atou performance comparisons that modify and extend some of the conclusions in the text.
p. 25
Middle of page, last of 5 bullets, fdwShareMode should be fdwAccess.
p. 29
At the end of the very last line, replace _tstrcpy with _tcscpy.
p. 65
For more information on the C Run-Time Library, see Win32 vs. The C Run-Time Library for File I/O. comp.os.ms-windows.programmer.win32 is also an excellent place to go for additional information or to get answers to your questions.
p. 98
The final parameter, lpMem, has type LPCVOID, not LPSTR as shown.
p. 108
The allocation granularity may not necessarily be 64K, contrary to what is stated. The actual granularity is the dwAllocationGranularity member of the SYSTEM_INFO structure, which you can read using the GetSystemInfo() function.
p. 136
Under the heading "Example: Reading File Permissions," the first sentence should begin, "Proram 7-4..." (not 7-3).
p. 155
Add an additional return value, WAIT_FAILED, for the wait functions. This value, of course, indicates that the call failed.
p. 156
In the "Process Security" section, the access right CREATE_PROCESS should be PROCESS_CREATE_PROCESS. DUPLICATE_HANDLE should be PROCESS_DUP_HANDLE, and PROCESS_CREATE_THREAD instead of CREATE_THREAD (Use the PROCESS_ prefix on all process access rights).
p. 187
Windows 95/98 systems cannot be named pipe servers, although this fact is not well documented. Furthermore, under W95, CreateNamedPipe actually returns a valid handle, but a client CreateFile fails to find the named pipe. The new example program reimplementing the pipe program (Program 9-1) with named pipes demonstrates this behavior. [Go to pipeNP.c program.]
p. 206
In the ExitThread box, it should say VOID ExitThread (DWORD dwExitCode)
9 lines up from the bottom of the page should say, "Windows NT Version 3.51 did not allow ..."
p. 216
The last sentence of the first paragraph under the heading "Thread Local Storage" should reference Program 10-1.
p. 217
The last line on the page should be REALTIME_PRIORITY_CLASS (there is no underbar in REALTIME).
p. 222
Exercise 10-7, as stated, does not make sense as TLS cannot pass data between threads.
p. 231
In the sixth line down, _finally { should begin with a small f (not a capital "F").
p. 232
p. 245
In the middle of the page, 4th line down after "Maintaining and Reporting Server Statistics," there is a word missing:
...this code shows some of the essential processing...
p. 288-289
All references to SwitchThreadToFiber should be to SwitchToFiber. There are three references, as well as the index.
Bibliography
Custer's book has been updated by David A Solomon as a second edition of Inside Windows NT. The edition is updated and expanded to include file system and Version 4.0 information. The ISBN is 1-572-31677-2.