Skip Navigation Links | |
Exit Print View | |
![]() |
man pages section 3: Basic Library Functions Oracle Solaris 11 Express 11/10 |
enable_extended_FILE_stdio(3C)
posix_spawnattr_getschedparam(3C)
posix_spawnattr_getschedpolicy(3C)
posix_spawnattr_getsigdefault(3C)
posix_spawnattr_getsigignore_np(3C)
posix_spawnattr_getsigmask(3C)
posix_spawnattr_setschedparam(3C)
posix_spawnattr_setschedpolicy(3C)
posix_spawnattr_setsigdefault(3C)
posix_spawnattr_setsigignore_np(3C)
posix_spawnattr_setsigmask(3C)
posix_spawn_file_actions_addclose(3C)
posix_spawn_file_actions_addclosefrom_np(3C)
posix_spawn_file_actions_adddup2(3C)
posix_spawn_file_actions_addopen(3C)
posix_spawn_file_actions_destroy(3C)
posix_spawn_file_actions_init(3C)
pthread_attr_getdetachstate(3C)
pthread_attr_getinheritsched(3C)
pthread_attr_getschedparam(3C)
pthread_attr_getschedpolicy(3C)
pthread_attr_setdetachstate(3C)
pthread_attr_setinheritsched(3C)
pthread_attr_setschedparam(3C)
pthread_attr_setschedpolicy(3C)
pthread_barrierattr_destroy(3C)
pthread_barrierattr_getpshared(3C)
pthread_barrierattr_setpshared(3C)
pthread_condattr_getpshared(3C)
pthread_condattr_setpshared(3C)
pthread_cond_reltimedwait_np(3C)
pthread_key_create_once_np(3C)
pthread_mutexattr_getprioceiling(3C)
pthread_mutexattr_getprotocol(3C)
pthread_mutexattr_getpshared(3C)
pthread_mutexattr_getrobust(3C)
pthread_mutexattr_setprioceiling(3C)
pthread_mutexattr_setprotocol(3C)
pthread_mutexattr_setpshared(3C)
pthread_mutexattr_setrobust(3C)
pthread_mutex_getprioceiling(3C)
pthread_mutex_reltimedlock_np(3C)
pthread_mutex_setprioceiling(3C)
pthread_rwlockattr_destroy(3C)
pthread_rwlockattr_getpshared(3C)
pthread_rwlockattr_setpshared(3C)
pthread_rwlock_reltimedrdlock_np(3C)
pthread_rwlock_reltimedwrlock_np(3C)
pthread_rwlock_timedrdlock(3C)
pthread_rwlock_timedwrlock(3C)
rctlblk_get_enforced_value(3C)
- allocation cache manipulation
cc [ flag … ] file… -lumem [ library … ] #include <umem.h> umem_cache_t *umem_cache_create(char *debug_name, size_t bufsize, size_t align, umem_constructor_t *constructor, umem_destructor_t *destructor, umem_reclaim_t *reclaim, void *callback_data, vmem_t *source, int cflags);
void umem_cache_destroy(umem_cache_t *cache);
void *umem_cache_alloc(umem_cache_t *cache, int flags);
void umem_cache_free(umem_cache_t *cache, void *buffer);
These functions create, destroy, and use an “object cache” An object cache is a collection of buffers of a single size, with optional content caching enabled by the use of callbacks (see Cache Callbacks). Object caches are MT-Safe. Multiple allocations and freeing of memory from different threads can proceed simultaneously. Object caches are faster and use less space per buffer than malloc(3MALLOC) and umem_alloc(3MALLOC). For more information about object caching, see “The Slab Allocator: An Object-Caching Kernel Memory Allocator” and “Magazines and vmem: Extending the Slab Allocator to Many CPUs and Arbitrary Resources”.
The umem_cache_create() function creates object caches. Once a cache has been created, objects can be requested from and returned to the cache using umem_cache_alloc() and umem_cache_free(), respectively. A cache with no outstanding buffers can be destroyed with umem_cache_destroy().
The umem_cache_create() function creates a cache of objects and takes as arguments the following:
A human-readable name for debugging purposes.
The size, in bytes, of the buffers in this cache.
The minimum alignment required for buffers in this cache. This parameter must be a power of 2. If 0, it is replaced with the minimum required alignment for the current architecture.
The callback to construct an object.
The callback to destroy an object.
The callback to reclaim objects.
An opaque pointer passed to the callbacks.
This parameter must be NULL.
This parameter must be either 0 or UMC_NODEBUG. If UMC_NODEBUG, all debugging features are disabled for this cache. See umem_debug(3MALLOC).
Each cache can have up to three associated callbacks:
int constructor(void *buffer, void *callback_data, int flags); void destructor(void *buffer, void *callback_data); void reclaim(void *callback_data);
The callback_data argument is always equal to the value passed to umem_cache_create(), thereby allowing a client to use the same callback functions for multiple caches, but with customized behavior.
The reclaim callback is called when the umem function is requesting more memory from the operating system. This callback can be used by clients who retain objects longer than they are strictly needed (for example, caching non-active state). A typical reclaim callback might return to the cache ten per cent of the unneeded buffers.
The constructor and destructor callbacks enable the management of buffers with the constructed state. The constructor takes as arguments a buffer with undefined contents, some callback data, and the flags to use for any allocations. This callback should transform the buffer into the constructed state.
The destructor callback takes as an argument a constructed object and prepares it for return to the general pool of memory. The destructor should undo any state that the constructor created. For debugging, the destructor can also check that the buffer is in the constructed state, to catch incorrectly freed buffers. See umem_debug(3MALLOC) for further information on debugging support.
The umem_cache_destroy() function destroys an object cache. If the cache has any outstanding allocations, the behavior is undefined.
The umem_cache_alloc() function takes as arguments:
a cache pointer
flags that determine the behavior if umem_cache_alloc() is unable to fulfill the allocation request
If successful, umem_cache_alloc() returns a pointer to the beginning of an object of bufsize length.
There are three cases to consider:
A new buffer needed to be allocated. If the cache was created with a constructor, it is applied to the buffer and the resulting object is returned.
The object cache was able to use a previously freed buffer. If the cache was created with a constructor, the object is returned unchanged from when it was freed.
The allocation of a new buffer failed. The flags argument determines the behavior:
The umem_cache_alloc() function returns NULL if the allocation fails.
The umem_cache_alloc() function cannot return NULL. A callback is used to determine what action occurs. See umem_alloc(3MALLOC) for more information.
The umem_cache_free() function takes as arguments:
a cache pointer
a pointer previously returned from umem_cache_alloc(). This argument must not be NULL.
If the cache was created with a constructor callback, the object must be returned to the constructed state before it is freed.
Undefined behavior results if an object is freed multiple times, if an object is modified after it is freed, or if an object is freed to a cache other than the one from which it was allocated.
When a constructor callback is in use, there is essentially a contract between the cache and its clients. The cache guarantees that all objects returned from umem_cache_alloc() will be in the constructed state, and the client guarantees that it will return the object to the constructed state before handing it to umem_cache_free().
Upon failure, the umem_cache_create() function returns a null pointer.
The umem_cache_create() function will fail if:
There is not enough memory available to allocate the cache data structure.
The debug_name argument is NULL, the align argument is not a power of two or is larger than the system pagesize, or the bufsize argument is 0.
The libumem library could not be initialized, or the bufsize argument is too large and its use would cause integer overflow to occur.
Example 1 Use a fixed-size structure with no constructor callback.
#include <umem.h> typedef struct my_obj { long my_data1; } my_obj_t; /* * my_objs can be freed at any time. The contents of * my_data1 is undefined at allocation time. */ umem_cache_t *my_obj_cache; ... my_obj_cache = umem_cache_create("my_obj", sizeof (my_obj_t), 0, NULL, NULL, NULL, NULL, NULL, 0); ... my_obj_t *cur = umem_cache_alloc(my_obj_cache, UMEM_DEFAULT); ... /* use cur */ ... umem_cache_free(my_obj_cache, cur); ...
Example 2 Use an object with a mutex.
#define _REENTRANT #include <synch.h> #include <umem.h> typedef struct my_obj { mutex_t my_mutex; long my_data; } my_obj_t; /* * my_objs can only be freed when my_mutex is unlocked. */ int my_obj_constructor(void *buf, void *ignored, int flags) { my_obj_t *myobj = buf; (void) mutex_init(&my_obj->my_mutex, USYNC_THREAD, NULL); return (0); } void my_obj_destructor(void *buf, void *ignored) { my_obj_t *myobj = buf; (void) mutex_destroy(&my_obj->my_mutex); } umem_cache_t *my_obj_cache; ... my_obj_cache = umem_cache_create("my_obj", sizeof (my_obj_t), 0, my_obj_constructor, my_obj_destructor, NULL, NULL, NULL, 0); ... my_obj_t *cur = umem_cache_alloc(my_obj_cache, UMEM_DEFAULT); cur->my_data = 0; /* cannot assume anything about my_data */ ... umem_cache_free(my_obj_cache, cur); ...
Example 3 Use a more complex object with a mutex.
#define _REENTRANT #include <assert.h> #include <synch.h> #include <umem.h> typedef struct my_obj { mutex_t my_mutex; cond_t my_cv; struct bar *my_barlist; unsigned my_refcount; } my_obj_t; /* * my_objs can only be freed when my_barlist == NULL, * my_refcount == 0, there are no waiters on my_cv, and * my_mutex is unlocked. */ int my_obj_constructor(void *buf, void *ignored, int flags) { my_obj_t *myobj = buf; (void) mutex_init(&my_obj->my_mutex, USYNC_THREAD, NULL); (void) cond_init(&my_obj->my_cv, USYNC_THREAD, NULL); myobj->my_barlist = NULL; myobj->my_refcount = 0; return (0); } void my_obj_destructor(void *buf, void *ignored) { my_obj_t *myobj = buf; assert(myobj->my_refcount == 0); assert(myobj->my_barlist == NULL); (void) cond_destroy(&my_obj->my_cv); (void) mutex_destroy(&my_obj->my_mutex); } umem_cache_t *my_obj_cache; ... my_obj_cache = umem_cache_create("my_obj", sizeof (my_obj_t), 0, my_obj_constructor, my_obj_destructor, NULL, NULL, NULL, 0); ... my_obj_t *cur = umem_cache_alloc(my_obj_cache, UMEM_DEFAULT); ... /* use cur */ ... umem_cache_free(my_obj_cache, cur); ...
Example 4 Use objects with a subordinate buffer while reusing callbacks.
#include assert.h> #include umem.h> typedef struct my_obj { char *my_buffer; size_t my_size; } my_obj_t; /* * my_size and the my_buffer pointer should never be changed */ int my_obj_constructor(void *buf, void *arg, int flags) { size_t sz = (size_t)arg; my_obj_t *myobj = buf; if ((myobj->my_buffer = umem_alloc(sz, flags)) == NULL) return (1); my_size = sz; return (0); } void my_obj_destructor(void *buf, void *arg) { size_t sz = (size_t)arg; my_obj_t *myobj = buf; assert(sz == buf->my_size); umem_free(myobj->my_buffer, sz); } ... umem_cache_t *my_obj_4k_cache; umem_cache_t *my_obj_8k_cache; ... my_obj_cache_4k = umem_cache_create("my_obj_4k", sizeof (my_obj_t), 0, my_obj_constructor, my_obj_destructor, NULL, (void *)4096, NULL, 0); my_obj_cache_8k = umem_cache_create("my_obj_8k", sizeof (my_obj_t), 0, my_obj_constructor, my_obj_destructor, NULL, (void *)8192, NULL, 0); ... my_obj_t *my_obj_4k = umem_cache_alloc(my_obj_4k_cache, UMEM_DEFAULT); my_obj_t *my_obj_8k = umem_cache_alloc(my_obj_8k_cache, UMEM_DEFAULT); /* no assumptions should be made about the contents of the buffers */ ... /* make sure to return them to the correct cache */ umem_cache_free(my_obj_4k_cache, my_obj_4k); umem_cache_free(my_obj_8k_cache, my_obj_8k); ...
See the EXAMPLES section of umem_alloc(3MALLOC) for examples involving the UMEM_NOFAIL flag.
See attributes(5) for descriptions of the following attributes:
|
setcontext(2), atexit(3C), libumem(3LIB), longjmp(3C), swapcontext(3C), thr_exit(3C), umem_alloc(3MALLOC), umem_debug(3MALLOC), attributes(5)
Bonwick, Jeff, “The Slab Allocator: An Object-Caching Kernel Memory Allocator”, Proceedings of the Summer 1994 Usenix Conference.
Bonwick, Jeff and Jonathan Adams, “Magazines and vmem: Extending the Slab Allocator to Many CPUs and Arbitrary Resources”, Proceedings of the Summer 2001 Usenix Conference.
Any of the following can cause undefined results:
Destroying a cache that has outstanding allocated buffers.
Using a cache after it has been destroyed.
Calling umem_cache_free() on the same buffer multiple times.
Passing a NULL pointer to umem_cache_free().
Writing past the end of a buffer.
Reading from or writing to a buffer after it has been freed.
Performing UMEM_NOFAIL allocations from an atexit(3C) handler.
Per-cache callbacks can be called from a variety of contexts. The use of functions that modify the active context, such as setcontext(2), swapcontext(3C), and thr_exit(3C), or functions that are unsafe for use in multithreaded applications, such as longjmp(3C) and siglongjmp(3C), result in undefined behavior.
A constructor callback that performs allocations must pass its flags argument unchanged to umem_alloc(3MALLOC) and umem_cache_alloc(). Any allocations made with a different flags argument results in undefined behavior. The constructor must correctly handle the failure of any allocations it makes.
Object caches make the following guarantees about objects:
If the cache has a constructor callback, it is applied to every object before it is returned from umem_cache_alloc() for the first time.
If the cache has a constructor callback, an object passed to umem_cache_free() and later returned from umem_cache_alloc() is not modified between the two events.
If the cache has a destructor, it is applied to all objects before their underlying storage is returned.
No other guarantees are made. In particular, even if there are buffers recently freed to the cache, umem_cache_alloc() can fail.