C dynamic memory allocation

C dynamic memory allocation refers to performing manual memory management for dynamic memory allocation in the C programming language via a group of functions in the C standard library, namely malloc, realloc, calloc and free.[1][2][3]

The C++ programming language includes these functions; however, the operators new and delete provide similar functionality and are recommended by that language's authors.[4] Still, there are several situations in which using new/delete is not applicable, such as garbage collection code or performance-sensitive code, and a combination of malloc and placement new may be required instead of the higher-level new operator.

Many different implementations of the actual memory allocation mechanism, used by malloc, are available. Their performance varies in both execution time and required memory.

The C programming language manages memory statically, automatically, or dynamically. Static-duration variables are allocated in main memory, usually along with the executable code of the program, and persist for the lifetime of the program; automatic-duration variables are allocated on the stack and come and go as functions are called and return. For static-duration and automatic-duration variables, the size of the allocation must be compile-time constant (except for the case of variable-length automatic arrays[5]). If the required size is not known until run-time (for example, if data of arbitrary size is being read from the user or from a disk file), then using fixed-size data objects is inadequate.

The lifetime of allocated memory can also cause concern. Neither static- nor automatic-duration memory is adequate for all situations. Automatic-allocated data cannot persist across multiple function calls, while static data persists for the life of the program whether it is needed or not. In many situations the programmer requires greater flexibility in managing the lifetime of allocated memory.

These limitations are avoided by using dynamic memory allocation, in which memory is more explicitly (but more flexibly) managed, typically by allocating it from the free store (informally called the "heap"), an area of memory structured for this purpose. In C, the library function malloc is used to allocate a block of memory on the heap. The program accesses this block of memory via a pointer that malloc returns. When the memory is no longer needed, the pointer is passed to free which deallocates the memory so that it can be used for other purposes.

The original description of C indicated that calloc and cfree were in the standard library, but not malloc. Code for a simple model implementation of a storage manager for Unix was given with alloc and free as the user interface functions, and using the sbrk system call to request memory from the operating system.[6] The 6th Edition Unix documentation gives alloc and free as the low-level memory allocation functions.[7] The malloc and free routines in their modern form are completely described in the 7th Edition Unix manual.[8][9]

Some platforms provide library or intrinsic function calls which allow run-time dynamic allocation from the C stack rather than the heap (e.g. alloca()[10]). This memory is automatically freed when the calling function ends.

The C dynamic memory allocation functions are defined in stdlib.h header (cstdlib header in C++).[1]

Creating an array of ten integers with automatic scope is straightforward in C:

However, the size of the array is fixed at compile time. If one wishes to allocate a similar array dynamically, the following code can be used:

This computes the number of bytes that ten integers occupy in memory, then requests that many bytes from malloc and assigns the result to a pointer named array (due to C syntax, pointers and arrays can be used interchangeably in some situations).

Because malloc might not be able to service the request, it might return a null pointer and it is good programming practice to check for this:

When the program no longer needs the dynamic array, it must eventually call free to return the memory it occupies to the free store:

The memory set aside by malloc is not initialized and may contain cruft: the remnants of previously used and discarded data. After allocation with malloc, elements of the array are uninitialized variables. The command calloc will return an allocation that has already been cleared:

Note that realloc must be assumed to have changed the base address of the block (i.e. if it has failed to extend the size of the original block, and has therefore allocated a new larger block elsewhere and copied the old contents into it). Therefore, any pointers to addresses within the original block are also no longer valid.

malloc returns a void pointer (void *), which indicates that it is a pointer to a region of unknown data type. The use of casting is required in C++ due to the strong type system, whereas this is not the case in C. One may "cast" (see type conversion) this pointer to a specific type:

The improper use of dynamic memory allocation can frequently be a source of bugs. These can include security bugs or program crashes, most often due to segmentation faults.

Memory allocation is not guaranteed to succeed, and may instead return a null pointer. Using the returned value, without checking if the allocation is successful, invokes undefined behavior. This usually leads to crash (due to the resulting segmentation fault on the null pointer dereference), but there is no guarantee that a crash will happen so relying on that can also lead to problems.

In addition, as an interface that precedes ANSI C standardization, malloc and its associated functions have behaviors that were intentionally left to the implementation to define for themselves. One of them is the zero-length allocation, which is more of a problem with realloc since it is more common to resize to zero.[15] Although both POSIX and the Single Unix Specification require proper handling of 0-size allocations by either returning NULL or something else that can be safely freed,[16] not all platforms are required to abide by these rules. Among the many double-free errors that it has led to, the 2019 WhatsApp RCE was especially prominent.[17] A way to wrap these functions to make them safer is by simply checking for 0-size allocations and turning them into those of size 1. (Returning NULL has its own problems: it otherwise indicates an out-of-memory failure. In the case of realloc it would have signaled that the original memory was not moved and freed, which again is not the case for size 0, leading to the double-free.)[18]

The implementation of memory management depends greatly upon operating system and architecture. Some operating systems supply an allocator for malloc, while others supply functions to control certain regions of data. The same dynamic memory allocator is often used to implement both malloc and the operator new in C++.[19]

Implementation of the allocator is commonly done using the heap, or data segment. The allocator will usually expand and contract the heap to fulfill allocation requests.

The heap method suffers from a few inherent flaws, stemming entirely from fragmentation. Like any method of memory allocation, the heap will become fragmented; that is, there will be sections of used and unused memory in the allocated space on the heap. A good allocator will attempt to find an unused area of already allocated memory to use before resorting to expanding the heap. The major problem with this method is that the heap has only two significant attributes: base, or the beginning of the heap in virtual memory space; and length, or its size. The heap requires enough system memory to fill its entire length, and its base can never change. Thus, any large areas of unused memory are wasted. The heap can get "stuck" in this position if a small used segment exists at the end of the heap, which could waste any amount of address space. On lazy memory allocation schemes, such as those often found in the Linux operating system, a large heap does not necessarily reserve the equivalent system memory; it will only do so at the first write time (reads of non-mapped memory pages return zero). The granularity of this depends on page size.

Doug Lea has developed the public domain dlmalloc ("Doug Lea's Malloc") as a general-purpose allocator, starting in 1987. The GNU C library (glibc) is derived from Wolfram Gloger's ptmalloc ("pthreads malloc"), a fork of dlmalloc with threading-related improvements.[20][21][22] As of November 2019, the latest version of dlmalloc is version 2.8.6 from August 2012.[23]

dlmalloc is a boundary tag allocator. Memory on the heap is allocated as "chunks", an 8-byte aligned data structure which contains a header, and usable memory. Allocated memory contains an 8- or 16-byte overhead for the size of the chunk and usage flags (similar to a dope vector). Unallocated chunks also store pointers to other free chunks in the usable space area, making the minimum chunk size 16 bytes on 32-bit systems and 24/32 (depends on alignment) bytes on 64-bit systems.[21][23]: 2.8.6, Minimum allocated size 

Unallocated memory is grouped into "bins" of similar sizes, implemented by using a double-linked list of chunks (with pointers stored in the unallocated space inside the chunk). Bins are sorted by size into three classes:[21][23]: Overlaid data structures 

Game developer Adrian Stone argues that dlmalloc, as a boundary-tag allocator, is unfriendly for console systems that have virtual memory but do not have demand paging. This is because its pool-shrinking and growing callbacks (sysmalloc/systrim) cannot be used to allocate and commit individual pages of virtual memory. In the absence of demand paging, fragmentation becomes a greater concern.[26]

Since FreeBSD 7.0 and NetBSD 5.0, the old malloc implementation (phkmalloc) was replaced by , written by Jason Evans. The main reason for this was a lack of scalability of phkmalloc in terms of multithreading. In order to avoid lock contention, jemalloc uses separate "arenas" for each CPU. Experiments measuring number of allocations per second in multithreading application have shown that this makes it scale linearly with the number of threads, while for both phkmalloc and dlmalloc performance was inversely proportional to the number of threads.[27]

OpenBSD's implementation of the malloc function makes use of mmap. For requests greater in size than one page, the entire allocation is retrieved using mmap; smaller sizes are assigned from memory pools maintained by malloc within a number of "bucket pages," also allocated with mmap.[28][better source needed] On a call to free, memory is released and unmapped from the process address space using munmap. This system is designed to improve security by taking advantage of the address space layout randomization and gap page features implemented as part of OpenBSD's mmap system call, and to detect use-after-free bugs—as a large memory allocation is completely unmapped after it is freed, further use causes a segmentation fault and termination of the program.

Hoard is an allocator whose goal is scalable memory allocation performance. Like OpenBSD's allocator, Hoard uses mmap exclusively, but manages memory in chunks of 64 kilobytes called superblocks. Hoard's heap is logically divided into a single global heap and a number of per-processor heaps. In addition, there is a thread-local cache that can hold a limited number of superblocks. By allocating only from superblocks on the local per-thread or per-processor heap, and moving mostly-empty superblocks to the global heap so they can be reused by other processors, Hoard keeps fragmentation low while achieving near linear scalability with the number of threads.[29]

An open-source compact general-purpose memory allocator from Microsoft Research with focus on performance.[30] The library is about 11,000 lines of code.

Every thread has a thread-local storage for small allocations. For large allocations mmap or sbrk can be used. , a malloc developed by Google,[31] has garbage-collection for local storage of dead threads. The TCMalloc is considered to be more than twice as fast as glibc's ptmalloc for multithreaded programs.[32][33]

Operating system kernels need to allocate memory just as application programs do. The implementation of malloc within a kernel often differs significantly from the implementations used by C libraries, however. For example, memory buffers might need to conform to special restrictions imposed by DMA, or the memory allocation function might be called from interrupt context.[34] This necessitates a malloc implementation tightly integrated with the virtual memory subsystem of the operating system kernel.

Because malloc and its relatives can have a strong impact on the performance of a program, it is not uncommon to override the functions for a specific application by custom implementations that are optimized for application's allocation patterns. The C standard provides no way of doing this, but operating systems have found various ways to do this by exploiting dynamic linking. One way is to simply link in a different library to override the symbols. Another, employed by Unix System V.3, is to make malloc and free function pointers that an application can reset to custom functions.[35]

The largest possible memory block malloc can allocate depends on the host system, particularly the size of physical memory and the operating system implementation.

Theoretically, the largest number should be the maximum value that can be held in a size_t type, which is an implementation-dependent unsigned integer representing the size of an area of memory. In the C99 standard and later, it is available as the SIZE_MAX constant from <stdint.h>. Although not guaranteed by ISO C, it is usually 2^(CHAR_BIT * sizeof(size_t)) - 1.

On glibc systems, the largest possible memory block malloc can allocate is only half this size, namely 2^(CHAR_BIT * sizeof(ptrdiff_t) - 1) - 1.[36]

The C library implementations shipping with various operating systems and compilers may come with alternatives and extensions to the standard malloc interface. Notable among these is: