Advanced Memory Allocation
You can alter the behavior of the memory management functions by adjusting some of the parameters exposed by the mallopt() function (Listings 1 and 2).
The prototype of this function and a basic set of four parameters are part of the SVID/XPG/ANSI standard. The current GNU C library implementation (version 2.3.1 as of this writing) honors only one of them (M_MXFAST), leaving three out. On the other hand, the library provides four additional parameters not specified by the standard. Tunable parameters accepted by mallopt() are described in the Sidebar “Tunable Parameters for mallopt()”.
Allocation tuning is possible even without introducing mallopt() calls inside your program and recompiling it. This may be useful if you want to test values quickly or if you don't have the sources. All you have to do is set the appropriate environment variable before running the application. Table 1 shows the mapping between mallopt() parameters and environment variables, as well as some additional information. If you wish to set the trim threshold to 64KB, for example, you can run this program:
Speaking of trimming, it is possible to trim the memory arena and give any unused memory back to the system by calling malloc_trim(pad). This function resizes the data segment, leaving at least pad bytes at the end of it and failing if less than one page worth of bytes can be freed. Segment size is always a multiple of one page, which is 4,096 bytes on i386. The size of the memory available to be trimmed is stored in the keepcost parameter of the struct returned by mallinfo(). Automatic trimming is done inside the free() function by calling memory_trim(), if the current value of keepcost is higher than the M_TRIM_THRESHOLD value, and by using the value of M_TOP_PAD as the argument.
Debugging memory is often one of the most time-consuming tasks when developing complex programs. The two basic aspects of this problem are checking memory corruption and tracing block allocation and release.
Memory corruption happens when writing to a location lying inside the legal data segment but outside the boundaries of the memory block you intended to use. An example is writing beyond an array's end. In fact, if you were to write outside the legal data segment, a segmentation fault would halt the program immediately or trigger the appropriate signal handler, allowing you to identify the misbehaving instruction. Memory corruption is thus more subtle, because it can pass unnoticed and cause a faulty behavior in a part of the program quite far from the offending part. For this reason, the sooner you detect it in the program, the higher your chances are of catching the bug.
Corruption may affect other memory blocks (messing with the application data) and the heap management structures. In the former case, the only symptom that something is going wrong comes from analyzing your own data structures. In the latter case, you can rely on some specific GNU libc consistency check mechanisms that alert you when something wrong is detected.
Memory checking in a program can be enabled as automatic or manual. The former is done by setting the environment variable MALLOC_CHECK_:
This mechanism is able to catch a fair number of boundary overflows and, in some cases, to protect the program from crashing. The action undertaken when a fault is detected depends on the value of MALLOC_CHECK_: 1 prints a warning message to stderr but does not abort the program; 2 aborts the program without any output; and 3 combines the effects of 1 and 2.
Automatic checking takes place only when memory-related functions are invoked. That is, if you write beyond an array's end, it won't be noticed until the next malloc() or free() call. Also, not all the errors are caught, and the information you obtain is not always extremely useful. In the case of free(), you know which pointer was being freed when the error was detected, but that gives no hint whatsoever as to who trashed the heap. In the case of errors detected during an allocation, you merely receive a “heap corrupted” message.
The alternative is to place manual checkpoints here and there in the program. To do this, you must call the mcheck() function at the beginning of the program. This function allows you to install a custom memory fault handler that can be invoked each time heap corruption is detected. A default handler also is available if you don't provide your own. Once mcheck() has been called, all the consistency checks you get with MALLOC_CHECK_ are in place. Moreover, you can call the mprobe() function manually to force a check on a given memory pointer at any time. Values returned by mprobe() are summarized in the Sidebar “mprobe() Results”.
If you want to check the whole heap and not only one block, you can call mcheck_check_all() to walk through all the active blocks. You also can instruct the memory management routines to use mcheck_check_all(), instead of checking only the current block by initializing mcheck_pedantic() instead of mcheck(). Be aware, though, that this approach is rather time consuming.
A third way to enable memory checking is to link your program with libmcheck:
gcc myprog.c -o myprog -lmcheck
The mcheck() function is called automatically before the first memory allocation takes place—useful in those cases when some dynamic blocks are allocated before entering main().
- DevOps: Better Than the Sum of Its Parts
- Return of the Mac
- Drupageddon: SQL Injection, Database Abstraction and Hundreds of Thousands of Web Sites
- Play for Me, Jarvis
- Non-Linux FOSS: .NET?
- Not So Dynamic Updates
- Designing Foils with XFLR5
- Users, Permissions and Multitenant Sites
- April 2015 Issue of Linux Journal: High-Performance Computing