[RFC 4/5] Page allocation fault injection script

[Anup C Shan <anupcshan at gmail dot com>]

#!/usr/local/bin/stap

/*
 * This script is used to fail __alloc_pages() function in mm/page_alloc.c
 */

probe begin(0)
{
	fij_load_param(20000,0,100,0,0,1,1000);
	fij_add_option("min_order",0,"Minimum order(base 2) of the allocation")
	fij_add_gfp_wait_param()
	fij_add_gfp_highmem_param()
}

global min_order

probe begin(2000)
{
	min_order = fij_params["min_order"]
}

function should_fail_alloc(order:long,flags:long)
{
	if (order < min_order) {
		fij_logger(100,sprintf("Skipping on order %d",order))
		return 0
	} else
		fij_logger(100,sprintf("Continuing on order %d",order))

	if (fij_should_fail_gfp_wait(flags) == 0)
		return 0

	if (fij_should_fail_gfp_highmem(flags) == 0)
		return 0

	if (fij_should_fail() == 0)
		return 0

	return 1
}

global zones_save,under_fail

/*
 * zones_save : Temporary variable to save zone information that will be 
 * assigned fake value to induce fault.
 * under_fail : Flag to denote whether system is under fault injected condition.
 * Both these variables are 'per-cpu' variables to prevent race conditions.
 */

function nullify(zptr:long,cpu:long)
%{
	struct zones **z = (struct zones **)(THIS->zptr);

	THIS->__retvalue = (long)*z;
	*z = NULL;
%}

/*
 * Method of fault injection : 
 * The function __alloc_pages() is failed by using an if condition on
 * variable z (=zonelist->zones).
 * To fail, the value of *(zonelist->zones) is saved in zones_save and
 * set to NULL to cause the function to return NULL value.
 * Once the fault is injected, the value of zonelist->zones is restored.
 */

probe kernel.function("__alloc_pages@mm/page_alloc.c")
{
	if (should_fail_alloc($order,$gfp_mask) == 0)
		next

	zones_save[cpu()] = nullify($zonelist->zones,cpu())
	under_fail[cpu()] = 1
	fij_done_fail()
}

function restore(zones:long,zones_save:long)
%{
	struct zones **z = (struct zones **)(THIS->zones);
	*z = (struct zones *)THIS->zones_save;
%}

probe kernel.function("__alloc_pages@mm/page_alloc.c").return
{
	if (under_fail[cpu()]) {
		restore($zonelist->zones,zones_save[cpu()])
		under_fail[cpu()] = 0
	}
}