#47540 [Opn]: CLI can go into an infinite write() loop when ignore_user_abort(true)
ID: 47540
User updated by: tstarling at wikimedia dot org
Reported By: tstarling at wikimedia dot org
Status: Open
-Bug Type: *General Issues
+Bug Type: CGI related
Operating System: Linux
PHP Version: 5.2CVS-2009-03-02 (CVS)
New Comment:
Fixed category.
Previous Comments:
------------------------------------------------------------------------
[2009-03-02 04:01:46] tstarling at wikimedia dot org
Description:
------------
sapi_cli_ub_write() is faulty and may go into a tight infinite loop
under certain circumstances. If write() produces an error, it calls
php_handle_aborted_connection(), but if ignore_user_abort(true) has been
called, this does not exit the request. Thus it enters an infinite
loop.
Bug #20053 is a similar bug in a different SAPI. It's likely that many
SAPIs have copied this code and suffer the same problem.
Reproduce code:
---------------
<?php
ignore_user_abort(true);
while ( true ) {
fwrite( STDERR, microtime(true) . "\n" );
echo "Hello\n";
usleep(500000);
}
?>
Run it like this:
$ php nfs-test.php > /mnt/some-nfs-mount/test-file
Then while that is running, on another server:
$ rm /mnt/some-nfs-mount/test-file
Expected result:
----------------
It should exit when the file is removed, like it does when
ignore_user_abort(true) is not called.
Actual result:
--------------
After a few seconds, the stderr output will stop ticking, and strace
will show a flood of:
write(1, "Hello\n", 6) = -1 ESTALE (Stale NFS file
handle)
write(1, "Hello\n", 6) = -1 ESTALE (Stale NFS file
handle)
write(1, "Hello\n", 6) = -1 ESTALE (Stale NFS file
handle)
write(1, "Hello\n", 6) = -1 ESTALE (Stale NFS file
handle)
write(1, "Hello\n", 6) = -1 ESTALE (Stale NFS file
handle)
write(1, "Hello\n", 6) = -1 ESTALE (Stale NFS file
handle)
------------------------------------------------------------------------
--
Edit this bug report at http://bugs.php.net/?id=47540&edit=1
Thread (4 messages)