Revision 9e2b7fa2df4365e99934901da4fb4af52d81e820 authored by Martin Willi on 06 November 2020, 07:30:30 UTC, committed by Jakub Kicinski on 12 November 2020, 15:47:06 UTC
VRF devices use an optimized direct path on output if a default qdisc
is involved, calling Netfilter hooks directly. This path, however, does
not consider Netfilter rules completing asynchronously, such as with
NFQUEUE. The Netfilter okfn() is called for asynchronously accepted
packets, but the VRF never passes that packet down the stack to send
it out over the slave device. Using the slower redirect path for this
seems not feasible, as we do not know beforehand if a Netfilter hook
has asynchronously completing rules.

Fix the use of asynchronously completing Netfilter rules in OUTPUT and
POSTROUTING by using a special completion function that additionally
calls dst_output() to pass the packet down the stack. Also, slightly
adjust the use of nf_reset_ct() so that is called in the asynchronous
case, too.

Fixes: dcdd43c41e60 ("net: vrf: performance improvements for IPv4")
Fixes: a9ec54d1b0cd ("net: vrf: performance improvements for IPv6")
Signed-off-by: Martin Willi <martin@strongswan.org>
Link: https://lore.kernel.org/r/20201106073030.3974927-1-martin@strongswan.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
1 parent 52755b6
Raw File
llist.c
// SPDX-License-Identifier: GPL-2.0-only
/*
 * Lock-less NULL terminated single linked list
 *
 * The basic atomic operation of this list is cmpxchg on long.  On
 * architectures that don't have NMI-safe cmpxchg implementation, the
 * list can NOT be used in NMI handlers.  So code that uses the list in
 * an NMI handler should depend on CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG.
 *
 * Copyright 2010,2011 Intel Corp.
 *   Author: Huang Ying <ying.huang@intel.com>
 */
#include <linux/kernel.h>
#include <linux/export.h>
#include <linux/llist.h>


/**
 * llist_add_batch - add several linked entries in batch
 * @new_first:	first entry in batch to be added
 * @new_last:	last entry in batch to be added
 * @head:	the head for your lock-less list
 *
 * Return whether list is empty before adding.
 */
bool llist_add_batch(struct llist_node *new_first, struct llist_node *new_last,
		     struct llist_head *head)
{
	struct llist_node *first;

	do {
		new_last->next = first = READ_ONCE(head->first);
	} while (cmpxchg(&head->first, first, new_first) != first);

	return !first;
}
EXPORT_SYMBOL_GPL(llist_add_batch);

/**
 * llist_del_first - delete the first entry of lock-less list
 * @head:	the head for your lock-less list
 *
 * If list is empty, return NULL, otherwise, return the first entry
 * deleted, this is the newest added one.
 *
 * Only one llist_del_first user can be used simultaneously with
 * multiple llist_add users without lock.  Because otherwise
 * llist_del_first, llist_add, llist_add (or llist_del_all, llist_add,
 * llist_add) sequence in another user may change @head->first->next,
 * but keep @head->first.  If multiple consumers are needed, please
 * use llist_del_all or use lock between consumers.
 */
struct llist_node *llist_del_first(struct llist_head *head)
{
	struct llist_node *entry, *old_entry, *next;

	entry = smp_load_acquire(&head->first);
	for (;;) {
		if (entry == NULL)
			return NULL;
		old_entry = entry;
		next = READ_ONCE(entry->next);
		entry = cmpxchg(&head->first, old_entry, next);
		if (entry == old_entry)
			break;
	}

	return entry;
}
EXPORT_SYMBOL_GPL(llist_del_first);

/**
 * llist_reverse_order - reverse order of a llist chain
 * @head:	first item of the list to be reversed
 *
 * Reverse the order of a chain of llist entries and return the
 * new first entry.
 */
struct llist_node *llist_reverse_order(struct llist_node *head)
{
	struct llist_node *new_head = NULL;

	while (head) {
		struct llist_node *tmp = head;
		head = head->next;
		tmp->next = new_head;
		new_head = tmp;
	}

	return new_head;
}
EXPORT_SYMBOL_GPL(llist_reverse_order);
back to top