Fix GH-15034: Integer overflow on stream_notification_callback byte_max parameter with files bigger than 2GB

[nielsdos]

We were using atoi, which is only for integers. When the size does not fit in an integer this breaks. Use ZEND_STRTOUL instead. Also make sure invalid data isn't accidentally parsed into a file size.

[yortx]

Hi @nielsdos . Thank you for reviewing this so quick!

I've seen your patch, and I have a little suggestion: to check that "parsed" is less than PHP_INT_MAX before setting it as the "file_size".
Since you are parsing the size to an "unsigned long" the result may not fit in a php integer, which is signed. I don't actually know what would receive the php function callback, but it will probably not be the correct result if the result doesn't fit in an php integer. I think it would be better not to set a file_size than set a wrong one.

That check would also avoid incorrect results when running on 32bit systems with PHP_INT_SIZE = 4. And since ZEND_STRTOUL seems to return MAX_ULONG in case the parsed number doesn't fit in a 64 ulong. That check would cover this case, too.

	if (endptr && !*endptr && parsed < ZEND_LONG_MAX) {
		file_size = parsed;
		php_stream_notify_file_size(context, file_size, http_header_line, 0);
	}

I'm not sure if that's the constant to use. It should be the equivalent of PHP_INT_MAX in the C side.

I'm not familiar with the php code, so please, double check my assumptions.
Hope this helps.

[nielsdos]

You are indeed right that it will cause issues on e.g. 32-bit platforms, as it already does now. I had thought about adding such a check when I wrote the patch but wasn't too sure about it. The reason is that adding the check will stop a notification event from happening which is a behaviour change, so I assumed we can only add that extra check on the master branch.
I'll let @bukka decide what's the most appropriate here.

I'm not sure if that's the constant to use. It should be the equivalent of PHP_INT_MAX in the C side.

Yes this constant is right.

[yortx]

I haven't thought about that. I agree that the lack of notification is not the better idea. What about just setting file_size to PHP_INT_MAX in the cases it would overflow?

• Shouldn't break any current code (I hope nobody relies in getting a negative file size).
• The value would be "more correct" than the current result. The best we can do with the current API.
• If documented, it would be easy to check for that condition in the php code.
• It's the behavior of the ZEND_STRTOUL, so it shouldn't be unexpected.

if (endptr && !*endptr) {
    file_size = parsed < ZEND_LONG_MAX ? parsed :  ZEND_LONG_MAX;
    php_stream_notify_file_size(context, file_size, http_header_line, 0);
}

[nielsdos]

What about just setting file_size to PHP_INT_MAX in the cases it would overflow?

I'm open to this, I think this would indeed be better, although a behaviour change as well... I'll leave the choice to Jakub.

[bukka]

I think it's fine to trim it to PHP_INT_MAX. No one should be using 32bit anyway so I don't particularly care that much... :)

[nielsdos]

I think it's fine to trim it to PHP_INT_MAX. No one should be using 32bit anyway so I don't particularly care that much... :)

That's fair, I pushed that now.

[bukka]

LGTM