2020-02-19 10:06:49

by Amol Grover

[permalink] [raw]
Subject: [PATCH] tcp: Pass lockdep expression to RCU lists

tcp_cong_list is traversed using list_for_each_entry_rcu
outside an RCU read-side critical section but under the protection
of tcp_cong_list_lock.

Hence, add corresponding lockdep expression to silence false-positive
warnings, and harden RCU lists.

Signed-off-by: Amol Grover <[email protected]>
---
net/ipv4/tcp_cong.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
index 3737ec096650..8d4446ed309e 100644
--- a/net/ipv4/tcp_cong.c
+++ b/net/ipv4/tcp_cong.c
@@ -25,7 +25,8 @@ static struct tcp_congestion_ops *tcp_ca_find(const char *name)
{
struct tcp_congestion_ops *e;

- list_for_each_entry_rcu(e, &tcp_cong_list, list) {
+ list_for_each_entry_rcu(e, &tcp_cong_list, list,
+ lockdep_is_held(&tcp_cong_list_lock)) {
if (strcmp(e->name, name) == 0)
return e;
}
@@ -55,7 +56,8 @@ struct tcp_congestion_ops *tcp_ca_find_key(u32 key)
{
struct tcp_congestion_ops *e;

- list_for_each_entry_rcu(e, &tcp_cong_list, list) {
+ list_for_each_entry_rcu(e, &tcp_cong_list, list,
+ lockdep_is_held(&tcp_cong_list_lock)) {
if (e->key == key)
return e;
}
@@ -317,7 +319,8 @@ int tcp_set_allowed_congestion_control(char *val)
}

/* pass 2 clear old values */
- list_for_each_entry_rcu(ca, &tcp_cong_list, list)
+ list_for_each_entry_rcu(ca, &tcp_cong_list, list,
+ lockdep_is_held(&tcp_cong_list_lock))
ca->flags &= ~TCP_CONG_NON_RESTRICTED;

/* pass 3 mark as allowed */
--
2.24.1


2020-02-19 19:35:49

by Eric Dumazet

[permalink] [raw]
Subject: Re: [PATCH] tcp: Pass lockdep expression to RCU lists



On 2/19/20 2:05 AM, Amol Grover wrote:
> tcp_cong_list is traversed using list_for_each_entry_rcu
> outside an RCU read-side critical section but under the protection
> of tcp_cong_list_lock.
>

This is not true.

There are cases where RCU read lock is held,
and others where the tcp_cong_list_lock is held.

I believe you need to be more precise in the changelog.

If there was a bug, net tree would be the target for this patch,
with a required Fixes: tag.

Otherwise, if net-next tree is the intended target, you have to signal
it, as instructed in Documentation/networking/netdev-FAQ.rst

Thanks.


> Hence, add corresponding lockdep expression to silence false-positive
> warnings, and harden RCU lists.
>
> Signed-off-by: Amol Grover <[email protected]>
> ---
> net/ipv4/tcp_cong.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
> index 3737ec096650..8d4446ed309e 100644
> --- a/net/ipv4/tcp_cong.c
> +++ b/net/ipv4/tcp_cong.c
> @@ -25,7 +25,8 @@ static struct tcp_congestion_ops *tcp_ca_find(const char *name)
> {
> struct tcp_congestion_ops *e;
>
> - list_for_each_entry_rcu(e, &tcp_cong_list, list) {
> + list_for_each_entry_rcu(e, &tcp_cong_list, list,
> + lockdep_is_held(&tcp_cong_list_lock)) {
> if (strcmp(e->name, name) == 0)
> return e;
> }
> @@ -55,7 +56,8 @@ struct tcp_congestion_ops *tcp_ca_find_key(u32 key)
> {
> struct tcp_congestion_ops *e;
>
> - list_for_each_entry_rcu(e, &tcp_cong_list, list) {
> + list_for_each_entry_rcu(e, &tcp_cong_list, list,
> + lockdep_is_held(&tcp_cong_list_lock)) {
> if (e->key == key)
> return e;
> }
> @@ -317,7 +319,8 @@ int tcp_set_allowed_congestion_control(char *val)
> }
>
> /* pass 2 clear old values */
> - list_for_each_entry_rcu(ca, &tcp_cong_list, list)
> + list_for_each_entry_rcu(ca, &tcp_cong_list, list,
> + lockdep_is_held(&tcp_cong_list_lock))
> ca->flags &= ~TCP_CONG_NON_RESTRICTED;
>
> /* pass 3 mark as allowed */
>

2020-02-20 03:47:28

by Amol Grover

[permalink] [raw]
Subject: Re: [PATCH] tcp: Pass lockdep expression to RCU lists

On Wed, Feb 19, 2020 at 11:35:21AM -0800, Eric Dumazet wrote:
>
>
> On 2/19/20 2:05 AM, Amol Grover wrote:
> > tcp_cong_list is traversed using list_for_each_entry_rcu
> > outside an RCU read-side critical section but under the protection
> > of tcp_cong_list_lock.
> >
>
> This is not true.
>
> There are cases where RCU read lock is held,
> and others where the tcp_cong_list_lock is held.
>

That's true but this patch specifically fixes those occurences of
list_for_each_entry_rcu() that are traversed under tcp_cong_list_lock.
Moreover, an implicit check is done for being inside RCU read-side
critical section along with testing for this newly added lockdep
expression.

> I believe you need to be more precise in the changelog.
>
> If there was a bug, net tree would be the target for this patch,
> with a required Fixes: tag.
>
> Otherwise, if net-next tree is the intended target, you have to signal
> it, as instructed in Documentation/networking/netdev-FAQ.rst
>

I don't think this fixes a "bug". However, this may fix potential bugs
that may creep in. Should I send it against net-next tree?

Thanks
Amol

> Thanks.
>
>
> > Hence, add corresponding lockdep expression to silence false-positive
> > warnings, and harden RCU lists.
> >
> > Signed-off-by: Amol Grover <[email protected]>
> > ---
> > net/ipv4/tcp_cong.c | 9 ++++++---
> > 1 file changed, 6 insertions(+), 3 deletions(-)
> >
> > diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
> > index 3737ec096650..8d4446ed309e 100644
> > --- a/net/ipv4/tcp_cong.c
> > +++ b/net/ipv4/tcp_cong.c
> > @@ -25,7 +25,8 @@ static struct tcp_congestion_ops *tcp_ca_find(const char *name)
> > {
> > struct tcp_congestion_ops *e;
> >
> > - list_for_each_entry_rcu(e, &tcp_cong_list, list) {
> > + list_for_each_entry_rcu(e, &tcp_cong_list, list,
> > + lockdep_is_held(&tcp_cong_list_lock)) {
> > if (strcmp(e->name, name) == 0)
> > return e;
> > }
> > @@ -55,7 +56,8 @@ struct tcp_congestion_ops *tcp_ca_find_key(u32 key)
> > {
> > struct tcp_congestion_ops *e;
> >
> > - list_for_each_entry_rcu(e, &tcp_cong_list, list) {
> > + list_for_each_entry_rcu(e, &tcp_cong_list, list,
> > + lockdep_is_held(&tcp_cong_list_lock)) {
> > if (e->key == key)
> > return e;
> > }
> > @@ -317,7 +319,8 @@ int tcp_set_allowed_congestion_control(char *val)
> > }
> >
> > /* pass 2 clear old values */
> > - list_for_each_entry_rcu(ca, &tcp_cong_list, list)
> > + list_for_each_entry_rcu(ca, &tcp_cong_list, list,
> > + lockdep_is_held(&tcp_cong_list_lock))
> > ca->flags &= ~TCP_CONG_NON_RESTRICTED;
> >
> > /* pass 3 mark as allowed */
> >