Double fil_system->mutex exit in fil_extend_space_to_desired_size if posix_fallocate is used

[Raghavendra D Prabhu]

Here:

 /* At this point it is safe to release fil_system mutex. No
other thread can rename, delete or close the file because
we have set the node->being_extended flag. */
mutex_exit(&fil_system->mutex);

start_page_no = space->size;
file_start_page_no = space->size - node->size;

#ifdef HAVE_POSIX_FALLOCATE
if (srv_use_posix_fallocate) {

mutex_exit(&fil_system->mutex);
success = os_file_set_size(node->name, node->handle,
(size_after_extend
- file_start_page_no) * page_size);
mutex_enter(&fil_system->mutex);
if (success) {
node->size += (size_after_extend - start_page_no);
space->size += (size_after_extend - start_page_no);
os_has_said_disk_full = FALSE;
}
node->being_extended = FALSE;
fil_node_complete_io(node, fil_system, OS_FILE_READ);
goto complete_io;
}
#endif

the mutex_exit and mutex_enter are reversed. (may be a merge regression etc.)

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-766