libtalloc_pools (3) - Linux Manuals

libtalloc_pools: Chapter 5: Memory pools

NAME

libtalloc_pools - Chapter 5: Memory pools

Memory pools

Allocation of a new memory is an expensive operation and large programs can contain thousands of calls of malloc() for a single computation, where every call allocates only a very small amount of the memory. This can result in an undesirable slowdown of the application. We can avoid this slowdown by decreasing the number of malloc() calls by using a memory pool.

A memory pool is a preallocated memory space with a fixed size. If we need to allocate new data we will take the desired amount of the memory from the pool instead of requesting a new memory from the system. This is done by creating a pointer that points inside the preallocated memory. Such a pool must not be reallocated as it would change its location - pointers that were pointing inside the pool would become invalid. Therefore, a memory pool requires a very good estimate of the required memory space.

The talloc library contains its own implementation of a memory pool. It is highly transparent for the programmer. The only thing that needs to be done is an initialization of a new pool context using talloc_pool() - which can be used in the same way as any other context.

Refactoring of existing code (that uses talloc) to take the advantage of a memory pool is quite simple due to the following properties of the pool context:

if we are allocating data on a pool context, it takes the desired amount of memory from the pool,
if the context is a descendant of the pool context, it takes the space from the pool as well,
if the pool does not have sufficient portion of memory left, it will create a new non-pool context, leaving the pool intact

/* allocate 1KiB in a pool */
TALLOC_CTX *pool_ctx = talloc_pool(NULL, 1024);

/* Take 512B from the pool, 512B is left there */
void *ptr = talloc_size(pool_ctx, 512);

/* 1024B > 512B, this will create new talloc chunk outside
   the pool */
void *ptr2 = talloc_size(ptr, 1024);

/* The pool still contains 512 free bytes
 * this will take 200B from them. */
void *ptr3 = talloc_size(ptr, 200);

/* This will destroy context 'ptr3' but the memory
 * is not freed, the available space in the pool
 * will increase to 512B. */
talloc_free(ptr3);

/* This will free memory taken by 'pool_ctx'
 * and 'ptr2' as well. */
talloc_free(pool_ctx);

The above given is very convenient, but there is one big issue to be kept in mind. If the parent of a talloc pool child is changed to a parent that is outside of this pool, the whole pool memory will not be freed until the child is freed. For this reason we must be very careful when stealing a descendant of a pool context.

TALLOC_CTX *mem_ctx = talloc_new(NULL);
TALLOC_CTX *pool_ctx = talloc_pool(NULL, 1024);
struct foo *foo = talloc(pool_ctx, struct foo);

/* mem_ctx is not in the pool */
talloc_steal(mem_ctx, foo);

/* pool_ctx is marked as freed but the memory is not
   deallocated, accessing the pool_ctx again will cause
   an error */
talloc_free(pool_ctx);

/* This deallocates the pool_ctx. */
talloc_free(mem_ctx);

It may often be better to copy the memory we want instead of stealing it to avoid this problem. If we do not need to retain the context name (to keep the type information), we can use talloc_memdup() to do this.

Copying the memory out of the pool may, however, discard all the performance boost given by the pool, depending on the size of the copied memory. Therefore, the code should be well profiled before taking this path. In general, the golden rule is: if we need to steal from the pool context, we should not use a pool context.