/* * Provide an simpler and easier to understand interface to the System V * semaphore system calls. There are 7 routines available to the user: * * id = sem_create(key, initval); # create with initial value or open * id = sem_open(key); # open (must already exist) * sem_wait(id); # wait = P = down by 1 * sem_signal(id); # signal = V = up by 1 * sem_op(id, amount); # wait if (amount < 0) * # signal if (amount > 0) * sem_close(id); # close * sem_rm(id); # remove (delete) * * We create and use a 3-member set for the requested semaphore. * The first member, [0], is the actual semaphore value, and the second * member, [1], is a counter used to know when all processes have finished * with the semaphore. The counter is initialized to a large number, * decremented on every create or open and incremented on every close. * This way we can use the "adjust" feature provided by System V so that * any process that exit's without calling sem_close() is accounted * for. It doesn't help us if the last process does this (as we have * no way of getting control to remove the semaphore) but it will * work if any process other than the last does an exit (intentional * or unintentional). * The third member, [2], of the semaphore set is used as a lock variable * to avoid any race conditions in the sem_create() and sem_close() * functions. */ #ifndef BOOST_INTERPROCESS_SYNC_XSI_ADVANCED_XSI_SEMAPHORE_HPP #define BOOST_INTERPROCESS_SYNC_XSI_ADVANCED_XSI_SEMAPHORE_HPP #ifndef BOOST_CONFIG_HPP # include #endif # #if defined(BOOST_HAS_PRAGMA_ONCE) # pragma once #endif #include #include #include namespace boost { namespace interprocess { namespace xsi { // Create a semaphore with a specified initial value. // If the semaphore already exists, we don't initialize it (of course). // We return the semaphore ID if all OK, else -1. inline bool advanced_sem_open_or_create(::key_t key, int initval, int &semid, int perm) { semid = -1; int id, semval; union semun { int val; ::semid_ds *buf; ushort *array; } semctl_arg; if (key == IPC_PRIVATE) return false; //not intended for private semaphores else if (key == (::key_t) -1) return false; //probably an ftok() error by caller again: if ((id = ::semget(key, 3, (perm & 0x01FF) | IPC_CREAT)) < 0) return false; //permission problem or tables full // When the semaphore is created, we know that the value of all // 3 members is 0. // Get a lock on the semaphore by waiting for [2] to equal 0, // then increment it. // // There is a race condition here. There is a possibility that // between the semget() above and the ::semop() below, another // process can call our sem_close() function which can remove // the semaphore if that process is the last one using it. // Therefore, we handle the error condition of an invalid // semaphore ID specially below, and if it does happen, we just // go back and create it again. struct sembuf op_lock[2] = { {2, 0, 0}, // wait for [2] (lock) to equal 0 {2, 1, SEM_UNDO} // then increment [2] to 1 - this locks it // UNDO to release the lock if processes exits // before explicitly unlocking }; if (::semop(id, &op_lock[0], 2) < 0) { if (errno == EINVAL) goto again; } // Get the value of the process counter. If it equals 0, // then no one has initialized the semaphore yet. if ((semval = ::semctl(id, 1, GETVAL, 0)) < 0) return false; if (semval == 0) { // We could initialize by doing a SETALL, but that // would clear the adjust value that we set when we // locked the semaphore above. Instead, we'll do 2 // system calls to initialize [0] and [1]. semctl_arg.val = initval; if (::semctl(id, 0, SETVAL, semctl_arg) < 0) return false; semctl_arg.val = 1; if (::semctl(id, 1, SETVAL, semctl_arg) < 0) return false; } // Decrement the process counter and then release the lock. struct sembuf op_unlock[1] = { 2, -1, 0/*SEM_UNDO*/ // decrement [2] (lock) back to 0 }; if (::semop(id, &op_unlock[0], 1) < 0) return false; semid = id; return true; } // Open a semaphore that must already exist. // This function should be used, instead of sem_create(), if the caller // knows that the semaphore must already exist. For example a client // from a client-server pair would use this, if its the server's // responsibility to create the semaphore. // We return the semaphore ID if all OK, else -1. /* inline bool advanced_sem_open(key_t key, int &semid) { semid = -1; if (key == IPC_PRIVATE) return false; // not intended for private semaphores else if (key == (::key_t) -1) return false; // probably an ftok() error by caller if ((semid = ::semget(key, 3, 0)) < 0) return false; // doesn't exist, or tables full // Decrement the process counter. We don't need a lock struct sembuf op_open[1] = { 1, -1, SEM_UNDO // decrement [1] (proc counter) with undo on exit }; if (::semop(id, &op_open[0], 1) < 0) return false; return true; } */ /**************************************************************************** * Remove a semaphore. * This call is intended to be called by a server, for example, * when it is being shut down, as we do an IPC_RMID on the semaphore, * regardless whether other processes may be using it or not. * Most other processes should use sem_close() below. */ inline bool advanced_sem_rm(int id) { if (::semctl(id, 0, IPC_RMID, 0) < 0) return false; return true; } /**************************************************************************** * General semaphore operation. Increment or decrement by a user-specified * amount (positive or negative; amount can't be zero). */ inline bool advanced_sem_op(int id, int value, bool undo = true) { ::sembuf op_op[1] = { 0, 99, 0 // decrement or increment [0] with undo on exit // the 99 is set to the actual amount to add // or subtract (positive or negative) }; if(undo){ op_op[0].sem_flg = SEM_UNDO; } if ((op_op[0].sem_op = value) == 0) return false; if (::semop(id, &op_op[0], 1) < 0) return false; return true; } } //namespace xsi { } //namespace interprocess { } //namespace boost { #endif //BOOST_INTERPROCESS_SYNC_XSI_ADVANCED_XSI_SEMAPHORE_HPP