* [PATCH 1/2] seq_file: add RCU versions of new hlist/list iterators (v2)
[not found] <20100219054145.959067404@vyatta.com>
@ 2010-02-19 5:41 ` Stephen Hemminger
2010-02-19 5:41 ` [PATCH 2/2] packet: convert socket list to RCU Stephen Hemminger
1 sibling, 0 replies; 3+ messages in thread
From: Stephen Hemminger @ 2010-02-19 5:41 UTC (permalink / raw)
To: David S. Miller, Li Zefan; +Cc: netdev, linux-kernel
[-- Attachment #1: seq_hlist_rcu.patch --]
[-- Type: text/plain, Size: 5194 bytes --]
Many usages of seq_file use RCU protected lists, so non RCU
iterators will not work safely. Needed to make __hlist_for_each_rcu
macro (like __list_for_each_rcu).
Also indented prototypes to match existing style.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
---
fs/seq_file.c | 88 +++++++++++++++++++++++++++++++++++++++++++++++
include/linux/rculist.h | 5 ++
include/linux/seq_file.h | 27 +++++++++-----
3 files changed, 111 insertions(+), 9 deletions(-)
--- a/fs/seq_file.c 2010-02-18 15:08:53.228872265 -0800
+++ b/fs/seq_file.c 2010-02-18 15:37:09.660996801 -0800
@@ -695,6 +695,38 @@ struct list_head *seq_list_next(void *v,
}
EXPORT_SYMBOL(seq_list_next);
+struct list_head *seq_list_start_rcu(struct list_head *head, loff_t pos)
+{
+ struct list_head *lh;
+
+ __list_for_each_rcu(lh, head)
+ if (pos-- == 0)
+ return lh;
+
+ return NULL;
+}
+EXPORT_SYMBOL(seq_list_start_rcu);
+
+struct list_head *seq_list_start_head_rcu(struct list_head *head, loff_t pos)
+{
+ if (!pos)
+ return head;
+
+ return seq_list_start_rcu(head, pos - 1);
+}
+EXPORT_SYMBOL(seq_list_start_head_rcu);
+
+struct list_head *seq_list_next_rcu(void *v, struct list_head *head,
+ loff_t *ppos)
+{
+ struct list_head *lh = v;
+
+ lh = rcu_dereference(lh->next);
+ ++*ppos;
+ return lh == head ? NULL : lh;
+}
+EXPORT_SYMBOL(seq_list_next_rcu);
+
/**
* seq_hlist_start - start an iteration of a hlist
* @head: the head of the hlist
@@ -750,3 +782,59 @@ struct hlist_node *seq_hlist_next(void *
return node->next;
}
EXPORT_SYMBOL(seq_hlist_next);
+
+/**
+ * seq_hlist_start_rcu - start an iteration of a hlist protected by RCU
+ * @head: the head of the hlist
+ * @pos: the start position of the sequence
+ *
+ * Called at seq_file->op->start().
+ */
+struct hlist_node *seq_hlist_start_rcu(struct hlist_head *head, loff_t pos)
+{
+ struct hlist_node *node;
+
+ __hlist_for_each_rcu(node, head)
+ if (pos-- == 0)
+ return node;
+ return NULL;
+}
+EXPORT_SYMBOL(seq_hlist_start_rcu);
+
+/**
+ * seq_hlist_start_head_rcu - start an iteration of a hlist protected by RCU
+ * @head: the head of the hlist
+ * @pos: the start position of the sequence
+ *
+ * Called at seq_file->op->start(). Call this function if you want to
+ * print a header at the top of the output.
+ */
+struct hlist_node *seq_hlist_start_head_rcu(struct hlist_head *head, loff_t pos)
+{
+ if (!pos)
+ return SEQ_START_TOKEN;
+
+ return seq_hlist_start_rcu(head, pos - 1);
+}
+EXPORT_SYMBOL(seq_hlist_start_head_rcu);
+
+/**
+ * seq_hlist_next_rcu - move to the next position of the hlist protected by RCU
+ * @v: the current iterator
+ * @head: the head of the hlist
+ * @pos: the current posision
+ *
+ * Called at seq_file->op->next().
+ */
+struct hlist_node *seq_hlist_next_rcu(void *v, struct hlist_head *head,
+ loff_t *ppos)
+{
+ struct hlist_node *node = v;
+
+ ++*ppos;
+ if (v == SEQ_START_TOKEN)
+ return rcu_dereference(head->first);
+ else
+ return rcu_dereference(node->next);
+}
+EXPORT_SYMBOL(seq_hlist_next_rcu);
--- a/include/linux/seq_file.h 2010-02-18 15:08:53.256872028 -0800
+++ b/include/linux/seq_file.h 2010-02-18 15:09:14.128997914 -0800
@@ -127,23 +127,32 @@ int seq_release_private(struct inode *,
/*
* Helpers for iteration over list_head-s in seq_files
*/
-
extern struct list_head *seq_list_start(struct list_head *head,
- loff_t pos);
+ loff_t pos);
extern struct list_head *seq_list_start_head(struct list_head *head,
- loff_t pos);
+ loff_t pos);
extern struct list_head *seq_list_next(void *v, struct list_head *head,
- loff_t *ppos);
-
+ loff_t *ppos);
+extern struct list_head *seq_list_start_rcu(struct list_head *head,
+ loff_t pos);
+extern struct list_head *seq_list_start_head_rcu(struct list_head *head,
+ loff_t pos);
+extern struct list_head *seq_list_next_rcu(void *v, struct list_head *head,
+ loff_t *ppos);
/*
* Helpers for iteration over hlist_head-s in seq_files
*/
-
extern struct hlist_node *seq_hlist_start(struct hlist_head *head,
- loff_t pos);
+ loff_t pos);
extern struct hlist_node *seq_hlist_start_head(struct hlist_head *head,
- loff_t pos);
+ loff_t pos);
extern struct hlist_node *seq_hlist_next(void *v, struct hlist_head *head,
- loff_t *ppos);
+ loff_t *ppos);
+extern struct hlist_node *seq_hlist_start_rcu(struct hlist_head *head,
+ loff_t pos);
+extern struct hlist_node *seq_hlist_start_head_rcu(struct hlist_head *head,
+ loff_t pos);
+extern struct hlist_node *seq_hlist_next_rcu(void *v, struct hlist_head *head,
+ loff_t *ppos);
#endif
--- a/include/linux/rculist.h 2010-02-18 15:25:03.681496691 -0800
+++ b/include/linux/rculist.h 2010-02-18 15:33:26.352872829 -0800
@@ -406,6 +406,11 @@ static inline void hlist_add_after_rcu(s
n->next->pprev = &n->next;
}
+#define __hlist_for_each_rcu(pos, head) \
+ for (pos = rcu_dereference((head)->first); \
+ pos && ({ prefetch(pos->next); 1; }); \
+ pos = rcu_dereference(pos->next))
+
/**
* hlist_for_each_entry_rcu - iterate over rcu list of given type
* @tpos: the type * to use as a loop cursor.
--
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH 2/2] packet: convert socket list to RCU
[not found] <20100219054145.959067404@vyatta.com>
2010-02-19 5:41 ` [PATCH 1/2] seq_file: add RCU versions of new hlist/list iterators (v2) Stephen Hemminger
@ 2010-02-19 5:41 ` Stephen Hemminger
2010-02-19 15:16 ` Eric Dumazet
1 sibling, 1 reply; 3+ messages in thread
From: Stephen Hemminger @ 2010-02-19 5:41 UTC (permalink / raw)
To: David S. Miller, Li Zefan; +Cc: netdev, linux-kernel
[-- Attachment #1: packet-list-rcu.patch --]
[-- Type: text/plain, Size: 5413 bytes --]
Convert AF_PACKET to use RCU, eliminating one more reader/writer lock.
I needed to create some minor additional socket list RCU infrastructure
to make this work. Note: there is no need for a real sk_del_node_init_rcu(),
because sk_del_node_init is doing the equivalent thing to
hlst_del_init_rcu already; but added some comments to try and make that obvious.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
---
include/net/netns/packet.h | 4 ++--
include/net/sock.h | 10 ++++++++++
net/packet/af_packet.c | 42 ++++++++++++++++++++----------------------
3 files changed, 32 insertions(+), 24 deletions(-)
--- a/include/net/netns/packet.h 2010-02-18 15:08:59.532872158 -0800
+++ b/include/net/netns/packet.h 2010-02-18 15:09:16.433496523 -0800
@@ -4,11 +4,11 @@
#ifndef __NETNS_PACKET_H__
#define __NETNS_PACKET_H__
-#include <linux/list.h>
+#include <linux/rculist.h>
#include <linux/spinlock.h>
struct netns_packet {
- rwlock_t sklist_lock;
+ spinlock_t sklist_lock;
struct hlist_head sklist;
};
--- a/net/packet/af_packet.c 2010-02-18 15:08:59.520873035 -0800
+++ b/net/packet/af_packet.c 2010-02-18 15:10:33.908871539 -0800
@@ -1262,18 +1262,15 @@ static int packet_release(struct socket
net = sock_net(sk);
po = pkt_sk(sk);
- write_lock_bh(&net->packet.sklist_lock);
- sk_del_node_init(sk);
+ spin_lock_bh(&net->packet.sklist_lock);
+ sk_del_node_init_rcu(sk);
sock_prot_inuse_add(net, sk->sk_prot, -1);
- write_unlock_bh(&net->packet.sklist_lock);
-
- /*
- * Unhook packet receive handler.
- */
+ spin_unlock_bh(&net->packet.sklist_lock);
if (po->running) {
/*
- * Remove the protocol hook
+ * Remove from protocol table
+ * does synchronize_net()
*/
dev_remove_pack(&po->prot_hook);
po->running = 0;
@@ -1478,10 +1475,11 @@ static int packet_create(struct net *net
po->running = 1;
}
- write_lock_bh(&net->packet.sklist_lock);
- sk_add_node(sk, &net->packet.sklist);
+ spin_lock_bh(&net->packet.sklist_lock);
+ sk_add_node_rcu(sk, &net->packet.sklist);
sock_prot_inuse_add(net, &packet_proto, 1);
- write_unlock_bh(&net->packet.sklist_lock);
+ spin_unlock_bh(&net->packet.sklist_lock);
+
return 0;
out:
return err;
@@ -2075,8 +2073,8 @@ static int packet_notifier(struct notifi
struct net_device *dev = data;
struct net *net = dev_net(dev);
- read_lock(&net->packet.sklist_lock);
- sk_for_each(sk, node, &net->packet.sklist) {
+ rcu_read_lock();
+ sk_for_each_rcu(sk, node, &net->packet.sklist) {
struct packet_sock *po = pkt_sk(sk);
switch (msg) {
@@ -2115,7 +2113,7 @@ static int packet_notifier(struct notifi
break;
}
}
- read_unlock(&net->packet.sklist_lock);
+ rcu_read_unlock();
return NOTIFY_DONE;
}
@@ -2512,24 +2510,24 @@ static struct notifier_block packet_netd
#ifdef CONFIG_PROC_FS
static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
- __acquires(seq_file_net(seq)->packet.sklist_lock)
+ __acquires(RCU)
{
struct net *net = seq_file_net(seq);
- read_lock(&net->packet.sklist_lock);
- return seq_hlist_start_head(&net->packet.sklist, *pos);
+
+ rcu_read_lock();
+ return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
}
static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
struct net *net = seq_file_net(seq);
- return seq_hlist_next(v, &net->packet.sklist, pos);
+ return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
}
static void packet_seq_stop(struct seq_file *seq, void *v)
- __releases(seq_file_net(seq)->packet.sklist_lock)
+ __releases(RCU)
{
- struct net *net = seq_file_net(seq);
- read_unlock(&net->packet.sklist_lock);
+ rcu_read_unlock();
}
static int packet_seq_show(struct seq_file *seq, void *v)
@@ -2581,7 +2579,7 @@ static const struct file_operations pack
static int __net_init packet_net_init(struct net *net)
{
- rwlock_init(&net->packet.sklist_lock);
+ spin_lock_init(&net->packet.sklist_lock);
INIT_HLIST_HEAD(&net->packet.sklist);
if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
--- a/include/net/sock.h 2010-02-18 15:08:59.548872240 -0800
+++ b/include/net/sock.h 2010-02-18 15:09:16.437496392 -0800
@@ -381,6 +381,7 @@ static __inline__ void __sk_del_node(str
__hlist_del(&sk->sk_node);
}
+/* NB: equivalent to hlist_del_init_rcu */
static __inline__ int __sk_del_node_init(struct sock *sk)
{
if (sk_hashed(sk)) {
@@ -421,6 +422,7 @@ static __inline__ int sk_del_node_init(s
}
return rc;
}
+#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
{
@@ -454,6 +456,12 @@ static __inline__ void sk_add_node(struc
__sk_add_node(sk, list);
}
+static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
+{
+ sock_hold(sk);
+ hlist_add_head_rcu(&sk->sk_node, list);
+}
+
static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
{
hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
@@ -478,6 +486,8 @@ static __inline__ void sk_add_bind_node(
#define sk_for_each(__sk, node, list) \
hlist_for_each_entry(__sk, node, list, sk_node)
+#define sk_for_each_rcu(__sk, node, list) \
+ hlist_for_each_entry_rcu(__sk, node, list, sk_node)
#define sk_nulls_for_each(__sk, node, list) \
hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
#define sk_nulls_for_each_rcu(__sk, node, list) \
--
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH 2/2] packet: convert socket list to RCU
2010-02-19 5:41 ` [PATCH 2/2] packet: convert socket list to RCU Stephen Hemminger
@ 2010-02-19 15:16 ` Eric Dumazet
0 siblings, 0 replies; 3+ messages in thread
From: Eric Dumazet @ 2010-02-19 15:16 UTC (permalink / raw)
To: Stephen Hemminger; +Cc: David S. Miller, Li Zefan, netdev, linux-kernel
Le jeudi 18 février 2010 à 21:41 -0800, Stephen Hemminger a écrit :
> Convert AF_PACKET to use RCU, eliminating one more reader/writer lock.
>
> I needed to create some minor additional socket list RCU infrastructure
> to make this work. Note: there is no need for a real sk_del_node_init_rcu(),
> because sk_del_node_init is doing the equivalent thing to
> hlst_del_init_rcu already; but added some comments to try and make that obvious.
>
> Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
>
Stephen, I am a bit worried by the interaction between packet_release()
and packet_notifier()
With your version, packet_notifier() can run and let another cpu run
packet_release() un-contented. Both cpus could manipulate same po (and
particularly po->running)
Before your patch, the read_lock() done in packet_notifier() was
preventing packet_release() runnning at the same time.
Maybe packet_release() should lock po->bind_lock before manipulating
po->running, avoiding a refcount error.
Something like this preliminary patch :
[PATCH] packet: fix a race in packet_release
packet_release() has a potential race with packet_notifier(NETDEV_DOWN),
leading to a double __sock_put(). (dev_remove_pack() is safe)
Fix is to always use po->bind_lock before accessing po->running
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 10f7295..b706031 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -1271,15 +1271,15 @@ static int packet_release(struct socket *sock)
* Unhook packet receive handler.
*/
+ spin_lock(&po->bind_lock);
if (po->running) {
- /*
- * Remove the protocol hook
- */
- dev_remove_pack(&po->prot_hook);
+ __sock_put(sk);
po->running = 0;
po->num = 0;
- __sock_put(sk);
- }
+ spin_unlock(&po->bind_lock);
+ dev_remove_pack(&po->prot_hook);
+ } else
+ spin_unlock(&po->bind_lock);
packet_flush_mclist(sk);
^ permalink raw reply related [flat|nested] 3+ messages in thread
end of thread, other threads:[~2010-02-19 15:16 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <20100219054145.959067404@vyatta.com>
2010-02-19 5:41 ` [PATCH 1/2] seq_file: add RCU versions of new hlist/list iterators (v2) Stephen Hemminger
2010-02-19 5:41 ` [PATCH 2/2] packet: convert socket list to RCU Stephen Hemminger
2010-02-19 15:16 ` Eric Dumazet
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).