I've had a devil of a time interfacing with non-PHP declared shared memory. DO NOT do this unless you read the php-X.X.X/ext/sysvshm/ code directly. Non-PHP programs need to use exactly the same "shared memory" struct as PHP uses, otherwise this "system V standard" call doesn't work.
Here are the specific deviations from the System-V C-calls:
<?
$shmkey = ftok($shmkey_path, $shmkey_proj_id);
?>
In C the $shmkey_proj_id is an int, but in PHP it's a char* to a single char. This means that your C program is restricted to using 0-255 (in PHP chr(0) through chr(255)) for project ID's, if it used anything else then you have to re-write your C-routine.
<?
$var = shm_get_var($shm_id, 0);
$fail = shm_put_var($shm_id, 0, $var1 )
?>
There are no such things as a system V calls like this at all. In system V you simply have a pointer to the beginning of your shared memory segment after attaching, so as long as you offset to the right spot you get what you want. In PHP, shared memory is ONLY cast as a struct, and the "data" is serialized as a string in the "mem" portion of the struct:
<? //C code
typedef struct {
long key;
long length;
long next;
char mem;
} sysvshm_chunk;
?>
If your C program didn't use this struct you need to re-write it to use this struct and write everything as a string.
This is rather absurd, as the purpose of shared memory is to communicate quickly with other programs resident in memory. But it is what it is.
Here is a simple Mutex class implementation, using semaphore on Linux and flock on windows.
Filename is optionnal, but you can provide it anyway. This way the file will be created on windows, not on linux as it is not needed.
It's fast written and certainly lacks error checking code.
<?php
class Mutex
{
private $id;
private $sem_id;
private $is_acquired = false;
private $is_windows = false;
private $filename = '';
private $filepointer;
function __construct()
{
if(substr(PHP_OS, 0, 3) == 'WIN')
$this->is_windows = true;
}
public function init($id, $filename = '')
{
$this->id = $id;
if($this->is_windows)
{
if(empty($filename)){
print "no filename specified";
return false;
}
else
$this->filename = $filename;
}
else
{
if(!($this->sem_id = sem_get($this->id, 1))){
print "Error getting semaphore";
return false;
}
}
return true;
}
public function acquire()
{
if($this->is_windows)
{
if(($this->filepointer = @fopen($this->filename, "w+")) == false)
{
print "error opening mutex file<br>";
return false;
}
if(flock($this->filepointer, LOCK_EX) == false)
{
print "error locking mutex file<br>";
return false;
}
}
else
{
if (! sem_acquire($this->sem_id)){
print "error acquiring semaphore";
return false;
}
}
$this->is_acquired = true;
return true;
}
public function release()
{
if(!$this->is_acquired)
return true;
if($this->is_windows)
{
if(flock($this->filepointer, LOCK_UN) == false)
{
print "error unlocking mutex file<br>";
return false;
}
fclose($this->filepointer);
}
else
{
if (! sem_release($this->sem_id)){
print "error releasing semaphore";
return false;
}
}
$this->is_acquired = false;
return true;
}
public function getId()
{
return $this->sem_id;
}
}
?>
Example use:
<?php
$mutex = new Mutex();
$mutex->init(1, "mutex_file.txt");
$mutex->acquire();
//Whatever you want single-threaded here...
$mutex->release();
?>
I have been trying to get a php console script and a C application to use a common semaphore for a while. I just got it working, so I thought Id paste the code here incase anyone needs to do this, however, this is not the place for long code examples
I used c code from the php implementation to set up the semaphore set and then mimic the way the php interpreter implements a mutext type locking scheme, using a common semop call.
One has to do the process in the same way as its done in the php implementation, otherwise you run the risk of the php interpreter resetting the semaphore set for you.
The basic idea is.
1) sem_get - use a three semaphore set
1.1) increment the first sem
1.2) check the usage count (sem 3), if only one, set the max_attach using sem 2 for mutex behaviour to sem 3
2) decrement sem 1
3) for locking / unlocking use the first semaphore, but always call the above from your c-code.
If you want a copy of my code, email me and I'll happily send it to you !
I was confused by two things that caused strange behaviour in my use of semaphores with php scripts running under apache.
Often enough page requests will end up being filled by the same process as other simultaneous requests. So semaphores will block when you may not have expected.
Also note that sem_remove() will remove it for all processes, not just the calling one. So you have to be sure that the last process running removes the semaphore and none before. I thought there was some failures occurring when my child processes were dropping out with errors.
So you can't just use get, acquire, release, remove in one script that will be hit by a web user. (1) They may end up in the same process and will wait on the other, and (2) the first one to finish will destroy the semaphore for others.
I left out the remove call, and it works ok, but I still wonder if the semaphore is removed by php when the last script that did a get finishes? Also creating a child process to do the work using proc_open works to ensure seperate processes but to be careful you would want to limit the number somehow as well.
Many (most?) developers use Win32 platform for PHP Web applications development while production servers mostly run Unix/Linux OS. Below is the stub code I use to make it possible to write scripts on Win32 that use semaphores:
<?php
if (substr(PHP_OS, 0, 3) == 'WIN') { // if Windows OS detected
function ftok($pathname, $proj)
{
if (empty($pathname) || !file_exists($pathname)) { // an error occured
return -1;
}
$pathname = $pathname . (string) $proj;
$key = array();
while (sizeof($key) < strlen($pathname)) {
$key[] = ord(substr($pathname, sizeof($key), 1));
}
return dechex(array_sum($key));
}
function sem_acquire($sem_identifier)
{
return true;
}
function sem_get($key, $max_acquire = null, $perm = null, $auto_release = null)
{
return true;
}
function sem_release($sem_identifier)
{
return true;
}
function sem_remove($sem_identifier)
{
return true;
}
}
?>
Of course, there is no way to test semaphores until you have no Unix/Linux test server.
Actually, the way to lock a semaphore from C code appears to be:
<?
struct sembuf semptr[2];
if( (semid = semget(SEM_KEY, PHP_SEM_NEED_NUMBER, 0666 | IPC_CREAT)) < 0 ) {
perror("semget");
return 1;
}
semptr[0].sem_num = 0;
semptr[0].sem_op = -1;
semptr[0].sem_flg = SEM_UNDO;
semptr[1].sem_num = 1;
semptr[1].sem_op = 1;
semptr[1].sem_flg = SEM_UNDO;
if( semop(semid, &semptr[0], 2) < 0 ) {
perror("semop");
}
?>
PS: Disregard the <? and ?>, it's just for pretty printing... this is C code!
Don't use semaphores to serialize access to an undefined number of resources. There is no way (yet) to know before locking if a semaphore is already locked, thus not being able to fully release the semaphore and occupying a semaphore resource for an undefined time.
A possible solution is to build a shared mem pool and store there the current number of locks for a semaphore id.
Cheers,
Horaci Cuevas
If you going to work with semaphore, which was created by some external program, you can try the following code for this program (C example):
#define SVSEM_MODE (SEM_R | SEM_A | SEM_R>>3 | SEM_R>>6) /* 0644 */
#define PHP_SEM_NEED_NUMBER 3
/*.......*/
int semid, semflag = SVSEM_MODE | IPC_CREAT | IPC_EXCL;
struct sembuf semptr;
union semun semopts;
/*.......*/
if( (semid = semget(sempath, PHP_SEM_NEED_NUMBER, semflag)) >= 0 ) {
semopts.val = 1; /* initial value for sem */
if( semctl( semid, 0, SETVAL, semopts) < 0 ) {/*error*/}
if( semctl( semid, 1, SETVAL, semopts) < 0 ) {/*error*/}
/* PHP wanna zero for its own semget at third sem.
* look at ./PHP_SOURCE_PATH/ext/sysvsem/sysvsem.c
*/
semopts.val = 0;
if( semctl( semid, 2, SETVAL, semopts) < 0 ) {/*error*/}
}
else if(errno == EEXIST) { /* connect only */
if( (semid = semget(sempath, PHP_SEM_NEED_NUMBER, SVSEM_MODE | IPC_CREAT)) < 0 ) {/*error*/}
}
else {/*error*/}
/*.......*/
/* If you want acquire the sem */
semptr.sem_num = 0;
semptr.sem_op = -1; /* lock it */
semptr.sem_flg = SEM_UNDO;
while( semop(semid, &semptr, 1) < 0 ) {/*error*/}
/*.......*/
Thanks,
Roma
Samlpe code for using most of the functions here:
$MEMSIZE = 512;// size of shared memory to allocate
$SEMKEY = 1; // Semaphore key
$SHMKEY = 2; // Shared memory key
echo "Start.\n";
// Get semaphore
$sem_id = sem_get($SEMKEY, 1);
if ($sem_id === false)
{
echo "Fail to get semaphore";
exit;
}
else
echo "Got semaphore $sem_id.\n";
// Accuire semaphore
if (! sem_acquire($sem_id))
{
echo "Fail to aquire semaphore $sem_id.\n";
sem_remove($sem_id);
exit;
}
else
echo "Success aquire semaphore $sem_id.\n";
$shm_id = shm_attach($SHMKEY, $MEMSIZE);
if ($shm_id === false)
{
echo "Fail to attach shared memory.\n";
sem_remove($sem_id);
exit;
}
else
echo "Success to attach shared memory : $shm_id.\n";
// Write variable 1
if (!shm_put_var($shm_id, 1, "Variable 1"))
{
echo "Fail to put var 1 on shared memory $shm_id.\n";
sem_remove($sem_id);
shm_remove ($shm_id);
exit;
}
else
echo "Write var1 to shared memory.\n";
// Write variable 2
if (!shm_put_var($shm_id, 2, "Variable 2"))
{
echo "Fail to put var 2 on shared memory $shm_id.\n";
sem_remove($sem_id);
shm_remove ($shm_id);
exit;
}
else
echo "Write var2 to shared memory.\n";
// Read variable 1
$var1 = shm_get_var ($shm_id, 1);
if ($var1 === false)
{
echo "Fail to retrive Var 1 from Shared memory $shm_id, return value=$var1.\n";
}
else
echo "Read var1=$var1.\n";
// Read variable 1
$var2 = shm_get_var ($shm_id, 2);
if ($var1 === false)
{
echo "Fail to retrive Var 2 from Shared memory $shm_id, return value=$var2.\n";
}
else
echo "Read var2=$var2.\n";
// Release semaphore
if (!sem_release($sem_id))
echo "Fail to release $sem_id semaphore.\n";
else
echo "Semaphore $sem_id released.\n";
// remove shared memory segmant from SysV
if (shm_remove ($shm_id))
echo "Shared memory successfully removed from SysV.\n";
else
echo "Fail to remove $shm_id shared memory from SysV.\n";
// Remove semaphore
if (sem_remove($sem_id))
echo "semaphore removed successfully from SysV.\n";
else
echo "Fail to remove $sem_id semaphore from SysV.\n";
echo "End.\n";
As for security, please look at the perm argument to shm_get. Shared Memory blocks has the same permission semantics as unix user/group/other file permissions. As long as your webserver is running as a user that no other users can script to.. and as long as the permissions are set to 600, you should be fine and have no security concerns.
The integer keys for sem_get() and shm_attach() have to be systemwide unique. There is no method to ensure that no other process on the system will use your specific key (security! and possible malfunction). Also shared memory is very seldom used there are possibilities for conflicts! To see the used id's you can use the program 'ipcs' (at least under SuseLinux;) ). Thanks Christian C.