persistent memory management

18 years ago by Alexey Zakhlestin — view source

unread

Hi.

I am trying to make my first php extension :)
I am sorry, if the question would be too newbiesh, but I just couldn't
find a more correct place to ask

here is the code:
http://loccache.googlecode.com/svn/trunk/loccache.c

I am trying to do the following:
at the minit-phase I am allocating a persistent HashTable
extension has 4 funcitons:
* loc_set($namespace, $varname, $value); // stores $value
* loc_get($namespace, $varname) // returns $value from storage
* loc_isset($namespace, $varname) // returns bool
* loc_unset($namespace, $varname) // removes varname from storage
during loc_set, I chech if there is "$namespace" item in my persistent
hashtable, and if there isn't, I create it as a child hashtable
$varname is a key in that second HashTable and $value is the value

something like:
$big_persistent_hash = array(
$namespace => array(
$varname => $value
)
);

extension build and even tends to work... kinda..

unfortunately, persistent values are not too persistent in my case, and i
keep getting memory errors.
I would be grateful for any help. the source code is not too big :)

18 years ago by Andrey Hristov — view source

unread

Hi,
Alexey Zakhlestin wrote:

Hi.

I am trying to make my first php extension :)
I am sorry, if the question would be too newbiesh, but I just couldn't
find a more correct place to ask

here is the code:
http://loccache.googlecode.com/svn/trunk/loccache.c

I am trying to do the following:
at the minit-phase I am allocating a persistent HashTable
extension has 4 funcitons:

loc_set($namespace, $varname, $value); // stores $value

loc_get($namespace, $varname) // returns $value from storage

loc_isset($namespace, $varname) // returns bool

loc_unset($namespace, $varname) // removes varname from storage
during loc_set, I chech if there is "$namespace" item in my persistent
hashtable, and if there isn't, I create it as a child hashtable
$varname is a key in that second HashTable and $value is the value

something like:
$big_persistent_hash = array(
$namespace => array(
$varname => $value
)
);

extension build and even tends to work... kinda..

unfortunately, persistent values are not too persistent in my case, and
i keep getting memory errors.
I would be grateful for any help. the source code is not too big :)

Tried running the web server under memory checker like valgrind?
Maybe instead of peamlloc() just do malloc()
Andrey

18 years ago by Alexey Zakhlestin — view source

unread

Tried running the web server under memory checker like valgrind?

Valgrind is linux only.. But I guess I still have one linux box.. I
will give it a try.

Maybe instead of peamlloc() just do malloc()
well.. I can do that for clarity, but zend_alloc.h has this:
#define pemalloc(size, persistent) ((persistent)?malloc(size):emalloc(size))

so, it basically maps

--
Alexey Zakhlestin
http://blog.milkfarmsoft.com/

18 years ago by Alexey Zakhlestin — view source

unread

Here is valgrind output. It definitely shows that there is a problem,
but I still can't guess where it is.

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
==25913== Invalid read of size 4
==25913== at 0x82AC068: zend_hash_exists (in /usr/bin/php5)
==25913== by 0x404EE42: zif_loc_isset (loccache.c:147)
==25913== by 0x82CF6CE: (within /usr/bin/php5)
==25913== by 0x82BF0C7: execute (in /usr/bin/php5)
==25913== by 0x82A029B: zend_execute_scripts (in /usr/bin/php5)
==25913== by 0x825B6E1: php_execute_script (in /usr/bin/php5)
==25913== by 0x832F43D: main (in /usr/bin/php5)
==25913== Address 0x4B00B90 is 0 bytes after a block of size 40 alloc'd
==25913== at 0x401C38B: malloc (vg_replace_malloc.c:149)
==25913== by 0x82AB82A: _zend_hash_add_or_update (in /usr/bin/php5)
==25913== by 0x404EDB0: zif_loc_set (loccache.c:86)
==25913== by 0x82CF6CE: (within /usr/bin/php5)
==25913== by 0x82BF0C7: execute (in /usr/bin/php5)
==25913== by 0x82A029B: zend_execute_scripts (in /usr/bin/php5)
==25913== by 0x825B6E1: php_execute_script (in /usr/bin/php5)
==25913== by 0x832F43D: main (in /usr/bin/php5)
==25913==
==25913== Invalid read of size 4
==25913== at 0x82AC06D: zend_hash_exists (in /usr/bin/php5)
==25913== by 0x404EE42: zif_loc_isset (loccache.c:147)
==25913== by 0x82CF6CE: (within /usr/bin/php5)
==25913== by 0x82BF0C7: execute (in /usr/bin/php5)
==25913== by 0x82A029B: zend_execute_scripts (in /usr/bin/php5)
==25913== by 0x825B6E1: php_execute_script (in /usr/bin/php5)
==25913== by 0x832F43D: main (in /usr/bin/php5)
==25913== Address 0x0 is not stack'd, malloc'd or (recently) free'd
==25913==
==25913== Process terminating with default action of signal 11 (SIGSEGV)
==25913== Access not within mapped region at address 0x0
==25913== at 0x82AC06D: zend_hash_exists (in /usr/bin/php5)
==25913== by 0x404EE42: zif_loc_isset (loccache.c:147)
==25913== by 0x82CF6CE: (within /usr/bin/php5)
==25913== by 0x82BF0C7: execute (in /usr/bin/php5)
==25913== by 0x82A029B: zend_execute_scripts (in /usr/bin/php5)
==25913== by 0x825B6E1: php_execute_script (in /usr/bin/php5)
==25913== by 0x832F43D: main (in /usr/bin/php5)
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

--
Alexey Zakhlestin
http://blog.milkfarmsoft.com/

18 years ago by Alexey Zakhlestin — view source

unread

And here is the command I used:
valgrind which php -n tests/3.php

18 years ago by Antony Dovgal — view source

unread

Here is valgrind output. It definitely shows that there is a problem,
but I still can't guess where it is.

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
==25913== Invalid read of size 4
==25913== at 0x82AC068: zend_hash_exists (in /usr/bin/php5)
==25913== by 0x404EE42: zif_loc_isset (loccache.c:147)
==25913== by 0x82CF6CE: (within /usr/bin/php5)
==25913== by 0x82BF0C7: execute (in /usr/bin/php5)
==25913== by 0x82A029B: zend_execute_scripts (in /usr/bin/php5)
==25913== by 0x825B6E1: php_execute_script (in /usr/bin/php5)
==25913== by 0x832F43D: main (in /usr/bin/php5)

--enable-debug would help to get some more info.

--
Wbr,
Antony Dovgal

18 years ago by Alexey Zakhlestin — view source

unread

I know.
but it will be very problematic to build debug-version of php there.
and that is the only linux machine around. :-(

Here is valgrind output. It definitely shows that there is a problem,
but I still can't guess where it is.

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
==25913== Invalid read of size 4
==25913== at 0x82AC068: zend_hash_exists (in /usr/bin/php5)
==25913== by 0x404EE42: zif_loc_isset (loccache.c:147)
==25913== by 0x82CF6CE: (within /usr/bin/php5)
==25913== by 0x82BF0C7: execute (in /usr/bin/php5)
==25913== by 0x82A029B: zend_execute_scripts (in /usr/bin/php5)
==25913== by 0x825B6E1: php_execute_script (in /usr/bin/php5)
==25913== by 0x832F43D: main (in /usr/bin/php5)

--enable-debug would help to get some more info.

--
Wbr,
Antony Dovgal

--
Alexey Zakhlestin
http://blog.milkfarmsoft.com/

18 years ago by Alexey Zakhlestin — view source

unread

I believe, I sorted out most of the issues.
Now, the last problem: what is the official way to deep-copy
non-persistent zval into persistent zval?

I expected zval_copy_ctor to have a flag. Unfortunately, it doesn't have one.

the way I see it:
void zval_copy_ctor(zval *value, zend_bool persistent)

such declaration would break compatibility, so, probably, there's a
need to introduce

zval_pcopy_ctor(zval *value, zend_bool persistent)
{
//…
}

and
#define zval_copy_ctor(v) zval_pcopy_ctor(v, 0)

similiar to how it is done with *alloc functions

--
Alexey Zakhlestin
http://blog.milkfarmsoft.com/

18 years ago by Antony Dovgal — view source

unread

Hi.

I am trying to make my first php extension :)
I am sorry, if the question would be too newbiesh, but I just couldn't
find a more correct place to ask

No need to use _exists here, zend_hash_find() returns FAILURE if it's failed to find the value.
You're doing the hash lookup twice here:

if (!zend_hash_exists(ht, ns_name, ns_len + 1)) {
	return NULL;
} 

zend_hash_find(ht, ns_name, ns_len + 1, (void**)&res);

Just this is enough:

if (zend_hash_find() == SUCCESS) {
return res;
}
return NULL;

And Andrey is right, valgrind is priceless when it comes to debugging memory errors.

--
Wbr,
Antony Dovgal

18 years ago by Alexey Zakhlestin — view source

unread

thanks.
fixed it in subversion

No need to use _exists here, zend_hash_find() returns FAILURE if it's failed to find the value.
You're doing the hash lookup twice here:
    if (!zend_hash_exists(ht, ns_name, ns_len + 1)) {
            return NULL;
    }

    zend_hash_find(ht, ns_name, ns_len + 1, (void**)&res);
Just this is enough:

if (zend_hash_find() == SUCCESS) {
return res;
}
return NULL;

And Andrey is right, valgrind is priceless when it comes to debugging memory errors.

--
Wbr,
Antony Dovgal

--
Alexey Zakhlestin
http://blog.milkfarmsoft.com/