Create or access a named semaphore


#include <semaphore.h>
#include <fcntl.h>

sem_t * sem_open( const char * sem_name,
                  int oflags,
                  ... );


The name of the semaphore that you want to create or access; see below.
Flags that affect how the function creates a new semaphore. This argument is a combination of:
  • O_EXCL
Note: Don't set oflags to O_RDONLY, O_RDWR, or O_WRONLY. According to POSIX, a semaphore's behavior is undefined with these flags. The QNX Neutrino libraries silently ignore these options, but they may reduce your code's portability.

For more information, see below.

If you set O_CREAT in oflags, you must also pass the following arguments:

mode_t mode
The semaphore's mode (just like file modes). For portability, you should set the read, write, and execute bits to the same value. An easy way of doing this is to use the constants from <sys/stat.h>:
  • S_IRWXG for group access.
  • S_IRWXO for other's access.
  • S_IRWXU for your own access.

For more information, see the entry for struct stat.

unsigned int value
The initial value of the semaphore. A positive value indicates the number of semaphore wait operations (e.g., sem_wait()) that will succeed without blocking, and a value of zero indicates that the next semaphore wait operation will block the calling thread. This value must not exceed SEM_VALUE_MAX.



Use the -l c option to qcc to link against this library. This library is usually included automatically.


The sem_open() function creates or accesses a named semaphore. Named semaphores are slower than the unnamed semaphores created with sem_init(). Semaphores persist as long as the system is up.

The sem_open() function returns a semaphore descriptor that you can use with sem_wait(), sem_trywait(), and sem_post(). You can use it until you call sem_close(). You can unlink the semaphore by calling sem_unlink(); when all processes have unlinked the semaphore, it's destroyed.

Named semaphores are always created under /dev/sem:

In either case, slash characters other than the leading slash character aren't interpreted, and the specified name, including these slash characters, is used to identify the semaphore. In other words, additional slashes don't create a directory structure under /dev/sem.

For example, if your current directory is /tmp:

name Pathname space entry
/my_sem /dev/sem/my_sem
my_sem /dev/sem/tmp/my_sem
Note: If you want to create or access a semaphore on another node, specify the name as /net/node/sem_location.

The oflags argument is used only for semaphore creation. When creating a new semaphore, you can set oflags to O_CREAT or (O_CREAT|O_EXCL):

Create a new named semaphore. If you set this bit, you must provide the mode and value arguments to sem_open().
When you're creating a new named semaphore, O_EXCL causes sem_open() to fail if a semaphore with sem_name already exists. Without O_EXCL, sem_open() attaches to an existing semaphore or creates a new one if sem_name doesn't exist.
Don't mix named semaphore operations (sem_open() and sem_close()) with unnamed semaphore operations (sem_init() and sem_destroy()) on the same semaphore.


A pointer to the created or accessed semaphore, or SEM_FAILED for failure (errno is set).


Either the named semaphore exists and you don't have permission to access it, or you're trying to create a new semaphore and you don't have permission.
You specified O_CREAT and O_EXCL in oflags, but the semaphore already exists.
The call was interrupted by a signal.
The sem_name argument is invalid or, when you're creating a semaphore, value is greater than SEM_VALUE_MAX.
Too many levels of symbolic links or prefixes.
The process is using too many files or semaphores.
The sem_name argument is longer than (NAME_MAX - 8).
The system ran out of resources and couldn't open the semaphore.
The semaphore identified by the sem_name argument doesn't exist and you didn't specify O_CREAT in oflags.
There isn't enough memory to create the new named semaphore.


POSIX 1003.1

Cancellation point No
Interrupt handler No
Signal handler No
Thread Yes