* [PATCH net] ipv6: fix suspecious RCU usage warning
@ 2021-03-08 19:21 Wei Wang
2021-03-09 2:47 ` David Ahern
0 siblings, 1 reply; 6+ messages in thread
From: Wei Wang @ 2021-03-08 19:21 UTC (permalink / raw)
To: David S . Miller, Jakub Kicinski, netdev
Cc: syzbot, David Ahern, Eric Dumazet
Syzbot reported the suspecious RCU usage in nexthop_fib6_nh() when
called from ipv6_route_seq_show(). The reason is ipv6_route_seq_start()
calls rcu_read_lock_bh(), while nexthop_fib6_nh() calls
rcu_dereference_rtnl().
The fix proposed is to add an extra parameter in nexthop_fib6_nh() to
indicate if bh is disabled or not, and then use the corresponding
dereference function.
The reported trace is as follows:
./include/net/nexthop.h:416 suspicious rcu_dereference_check() usage!
other info that might help us debug this:
rcu_scheduler_active = 2, debug_locks = 1
2 locks held by syz-executor.0/17895:
#0: ffff8801db098e20 (&p->lock){+.+.},
at: seq_read+0x71/0x12a0 fs/seq_file.c:169
#1: ffffffff85e18de0 (rcu_read_lock_bh){....},
at: seq_file_net include/linux/seq_file_net.h:19 [inline]
#1: ffffffff85e18de0 (rcu_read_lock_bh){....},
at: ipv6_route_seq_start+0xaf/0x300 net/ipv6/ip6_fib.c:2616
stack backtrace:
CPU: 1 PID: 17895 Comm: syz-executor.0 Not tainted 4.15.0-syzkaller #0
Call Trace:
[<ffffffff849edf9e>] __dump_stack lib/dump_stack.c:17 [inline]
[<ffffffff849edf9e>] dump_stack+0xd8/0x147 lib/dump_stack.c:53
[<ffffffff8480b7fa>] lockdep_rcu_suspicious+0x153/0x15d kernel/locking/lockdep.c:5745
[<ffffffff8459ada6>] nexthop_fib6_nh include/net/nexthop.h:416 [inline]
[<ffffffff8459ada6>] ipv6_route_native_seq_show net/ipv6/ip6_fib.c:2488 [inline]
[<ffffffff8459ada6>] ipv6_route_seq_show+0x436/0x7a0 net/ipv6/ip6_fib.c:2673
[<ffffffff81c556df>] seq_read+0xccf/0x12a0 fs/seq_file.c:276
[<ffffffff81dbc62c>] proc_reg_read+0x10c/0x1d0 fs/proc/inode.c:231
[<ffffffff81bc28ae>] do_loop_readv_writev fs/read_write.c:714 [inline]
[<ffffffff81bc28ae>] do_loop_readv_writev fs/read_write.c:701 [inline]
[<ffffffff81bc28ae>] do_iter_read+0x49e/0x660 fs/read_write.c:935
[<ffffffff81bc81ab>] vfs_readv+0xfb/0x170 fs/read_write.c:997
[<ffffffff81c88847>] kernel_readv fs/splice.c:361 [inline]
[<ffffffff81c88847>] default_file_splice_read+0x487/0x9c0 fs/splice.c:416
[<ffffffff81c86189>] do_splice_to+0x129/0x190 fs/splice.c:879
[<ffffffff81c86f66>] splice_direct_to_actor+0x256/0x890 fs/splice.c:951
[<ffffffff81c8777d>] do_splice_direct+0x1dd/0x2b0 fs/splice.c:1060
[<ffffffff81bc4747>] do_sendfile+0x597/0xce0 fs/read_write.c:1459
[<ffffffff81bca205>] SYSC_sendfile64 fs/read_write.c:1520 [inline]
[<ffffffff81bca205>] SyS_sendfile64+0x155/0x170 fs/read_write.c:1506
[<ffffffff81015fcf>] do_syscall_64+0x1ff/0x310 arch/x86/entry/common.c:305
[<ffffffff84a00076>] entry_SYSCALL_64_after_hwframe+0x42/0xb7
Fixes: f88d8ea67fbdb ("ipv6: Plumb support for nexthop object in a fib6_info")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Wei Wang <weiwan@google.com>
Cc: David Ahern <dsahern@kernel.org>
Cc: Eric Dumazet <edumazet@google.com>
---
include/net/nexthop.h | 18 +++++++++++++-----
net/ipv6/ip6_fib.c | 2 +-
net/ipv6/route.c | 10 +++++-----
3 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/include/net/nexthop.h b/include/net/nexthop.h
index 7bc057aee40b..48956b144689 100644
--- a/include/net/nexthop.h
+++ b/include/net/nexthop.h
@@ -410,31 +410,39 @@ static inline struct fib_nh *fib_info_nh(struct fib_info *fi, int nhsel)
int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
struct netlink_ext_ack *extack);
-static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
+static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh,
+ bool bh_disabled)
{
struct nh_info *nhi;
if (nh->is_group) {
struct nh_group *nh_grp;
- nh_grp = rcu_dereference_rtnl(nh->nh_grp);
+ if (bh_disabled)
+ nh_grp = rcu_dereference_bh_rtnl(nh->nh_grp);
+ else
+ nh_grp = rcu_dereference_rtnl(nh->nh_grp);
nh = nexthop_mpath_select(nh_grp, 0);
if (!nh)
return NULL;
}
- nhi = rcu_dereference_rtnl(nh->nh_info);
+ if (bh_disabled)
+ nhi = rcu_dereference_bh_rtnl(nh->nh_info);
+ else
+ nhi = rcu_dereference_rtnl(nh->nh_info);
if (nhi->family == AF_INET6)
return &nhi->fib6_nh;
return NULL;
}
+// Called with rcu_read_lock()
static inline struct net_device *fib6_info_nh_dev(struct fib6_info *f6i)
{
struct fib6_nh *fib6_nh;
- fib6_nh = f6i->nh ? nexthop_fib6_nh(f6i->nh) : f6i->fib6_nh;
+ fib6_nh = f6i->nh ? nexthop_fib6_nh(f6i->nh, false) : f6i->fib6_nh;
return fib6_nh->fib_nh_dev;
}
@@ -449,7 +457,7 @@ static inline void nexthop_path_fib6_result(struct fib6_result *res, int hash)
if (nhi->reject_nh) {
res->fib6_type = RTN_BLACKHOLE;
res->fib6_flags |= RTF_REJECT;
- res->nh = nexthop_fib6_nh(nh);
+ res->nh = nexthop_fib6_nh(nh, false);
} else {
res->nh = &nhi->fib6_nh;
}
diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c
index ef9d022e693f..70246a7e1344 100644
--- a/net/ipv6/ip6_fib.c
+++ b/net/ipv6/ip6_fib.c
@@ -2486,7 +2486,7 @@ static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
const struct net_device *dev;
if (rt->nh)
- fib6_nh = nexthop_fib6_nh(rt->nh);
+ fib6_nh = nexthop_fib6_nh(rt->nh, true);
seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 1536f4948e86..2439f6904806 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -547,7 +547,7 @@ static void rt6_device_match(struct net *net, struct fib6_result *res,
if (!oif && ipv6_addr_any(saddr)) {
if (unlikely(f6i->nh)) {
- nh = nexthop_fib6_nh(f6i->nh);
+ nh = nexthop_fib6_nh(f6i->nh, false);
if (nexthop_is_blackhole(f6i->nh))
goto out_blackhole;
} else {
@@ -583,7 +583,7 @@ static void rt6_device_match(struct net *net, struct fib6_result *res,
}
if (unlikely(f6i->nh)) {
- nh = nexthop_fib6_nh(f6i->nh);
+ nh = nexthop_fib6_nh(f6i->nh, false);
if (nexthop_is_blackhole(f6i->nh))
goto out_blackhole;
} else {
@@ -830,7 +830,7 @@ static void __find_rr_leaf(struct fib6_info *f6i_start,
res->fib6_flags = RTF_REJECT;
res->fib6_type = RTN_BLACKHOLE;
res->f6i = f6i;
- res->nh = nexthop_fib6_nh(f6i->nh);
+ res->nh = nexthop_fib6_nh(f6i->nh, false);
return;
}
if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
@@ -3701,7 +3701,7 @@ static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
goto out;
}
rt->nh = nh;
- fib6_nh = nexthop_fib6_nh(rt->nh);
+ fib6_nh = nexthop_fib6_nh(rt->nh, false);
} else {
err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
if (err)
@@ -5443,7 +5443,7 @@ static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
} else {
struct fib6_nh *fib6_nh;
- fib6_nh = nexthop_fib6_nh(nh);
+ fib6_nh = nexthop_fib6_nh(nh, false);
if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
flags, false) < 0)
goto nla_put_failure;
--
2.30.1.766.gb4fecdf3b7-goog
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH net] ipv6: fix suspecious RCU usage warning
2021-03-08 19:21 [PATCH net] ipv6: fix suspecious RCU usage warning Wei Wang
@ 2021-03-09 2:47 ` David Ahern
2021-03-09 8:15 ` Ido Schimmel
0 siblings, 1 reply; 6+ messages in thread
From: David Ahern @ 2021-03-09 2:47 UTC (permalink / raw)
To: Wei Wang, David S . Miller, Jakub Kicinski, netdev, Ido Schimmel,
Petr Machata
Cc: syzbot, David Ahern, Eric Dumazet
[ cc Ido and Petr ]
On 3/8/21 12:21 PM, Wei Wang wrote:
> diff --git a/include/net/nexthop.h b/include/net/nexthop.h
> index 7bc057aee40b..48956b144689 100644
> --- a/include/net/nexthop.h
> +++ b/include/net/nexthop.h
> @@ -410,31 +410,39 @@ static inline struct fib_nh *fib_info_nh(struct fib_info *fi, int nhsel)
> int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
> struct netlink_ext_ack *extack);
>
> -static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
> +static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh,
> + bool bh_disabled)
Hi Wei: I would prefer not to have a second argument to nexthop_fib6_nh
for 1 code path, and a control path at that.
> {
> struct nh_info *nhi;
>
> if (nh->is_group) {
> struct nh_group *nh_grp;
>
> - nh_grp = rcu_dereference_rtnl(nh->nh_grp);
> + if (bh_disabled)
> + nh_grp = rcu_dereference_bh_rtnl(nh->nh_grp);
> + else
> + nh_grp = rcu_dereference_rtnl(nh->nh_grp);
> nh = nexthop_mpath_select(nh_grp, 0);
> if (!nh)
> return NULL;
> }
>
> - nhi = rcu_dereference_rtnl(nh->nh_info);
> + if (bh_disabled)
> + nhi = rcu_dereference_bh_rtnl(nh->nh_info);
> + else
> + nhi = rcu_dereference_rtnl(nh->nh_info);
> if (nhi->family == AF_INET6)
> return &nhi->fib6_nh;
>
> return NULL;
> }
>
I am wary of duplicating code, but this helper is simple enough that it
should be ok with proper documentation.
Ido/Petr: I think your resilient hashing patch set touches this helper.
How ugly does it get to have a second version?
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] ipv6: fix suspecious RCU usage warning
2021-03-09 2:47 ` David Ahern
@ 2021-03-09 8:15 ` Ido Schimmel
2021-03-09 17:32 ` Wei Wang
0 siblings, 1 reply; 6+ messages in thread
From: Ido Schimmel @ 2021-03-09 8:15 UTC (permalink / raw)
To: David Ahern
Cc: Wei Wang, David S . Miller, Jakub Kicinski, netdev, Ido Schimmel,
Petr Machata, syzbot, David Ahern, Eric Dumazet
On Mon, Mar 08, 2021 at 07:47:31PM -0700, David Ahern wrote:
> [ cc Ido and Petr ]
>
> On 3/8/21 12:21 PM, Wei Wang wrote:
> > diff --git a/include/net/nexthop.h b/include/net/nexthop.h
> > index 7bc057aee40b..48956b144689 100644
> > --- a/include/net/nexthop.h
> > +++ b/include/net/nexthop.h
> > @@ -410,31 +410,39 @@ static inline struct fib_nh *fib_info_nh(struct fib_info *fi, int nhsel)
> > int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
> > struct netlink_ext_ack *extack);
> >
> > -static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
> > +static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh,
> > + bool bh_disabled)
>
> Hi Wei: I would prefer not to have a second argument to nexthop_fib6_nh
> for 1 code path, and a control path at that.
>
> > {
> > struct nh_info *nhi;
> >
> > if (nh->is_group) {
> > struct nh_group *nh_grp;
> >
> > - nh_grp = rcu_dereference_rtnl(nh->nh_grp);
> > + if (bh_disabled)
> > + nh_grp = rcu_dereference_bh_rtnl(nh->nh_grp);
> > + else
> > + nh_grp = rcu_dereference_rtnl(nh->nh_grp);
> > nh = nexthop_mpath_select(nh_grp, 0);
> > if (!nh)
> > return NULL;
> > }
> >
> > - nhi = rcu_dereference_rtnl(nh->nh_info);
> > + if (bh_disabled)
> > + nhi = rcu_dereference_bh_rtnl(nh->nh_info);
> > + else
> > + nhi = rcu_dereference_rtnl(nh->nh_info);
> > if (nhi->family == AF_INET6)
> > return &nhi->fib6_nh;
> >
> > return NULL;
> > }
> >
>
> I am wary of duplicating code, but this helper is simple enough that it
> should be ok with proper documentation.
>
> Ido/Petr: I think your resilient hashing patch set touches this helper.
> How ugly does it get to have a second version?
It actually doesn't touch this helper. Looks fine to me:
diff --git a/include/net/nexthop.h b/include/net/nexthop.h
index ba94868a21d5..6df9c12546fd 100644
--- a/include/net/nexthop.h
+++ b/include/net/nexthop.h
@@ -496,6 +496,26 @@ static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
return NULL;
}
+static inline struct fib6_nh *nexthop_fib6_nh_bh(struct nexthop *nh)
+{
+ struct nh_info *nhi;
+
+ if (nh->is_group) {
+ struct nh_group *nh_grp;
+
+ nh_grp = rcu_dereference_bh(nh->nh_grp);
+ nh = nexthop_mpath_select(nh_grp, 0);
+ if (!nh)
+ return NULL;
+ }
+
+ nhi = rcu_dereference_bh(nh->nh_info);
+ if (nhi->family == AF_INET6)
+ return &nhi->fib6_nh;
+
+ return NULL;
+}
+
static inline struct net_device *fib6_info_nh_dev(struct fib6_info *f6i)
{
struct fib6_nh *fib6_nh;
diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c
index ef9d022e693f..679699e953f1 100644
--- a/net/ipv6/ip6_fib.c
+++ b/net/ipv6/ip6_fib.c
@@ -2486,7 +2486,7 @@ static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
const struct net_device *dev;
if (rt->nh)
- fib6_nh = nexthop_fib6_nh(rt->nh);
+ fib6_nh = nexthop_fib6_nh_bh(rt->nh);
seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH net] ipv6: fix suspecious RCU usage warning
2021-03-09 8:15 ` Ido Schimmel
@ 2021-03-09 17:32 ` Wei Wang
2021-03-09 19:33 ` David Ahern
0 siblings, 1 reply; 6+ messages in thread
From: Wei Wang @ 2021-03-09 17:32 UTC (permalink / raw)
To: Ido Schimmel
Cc: David Ahern, David S . Miller, Jakub Kicinski,
Linux Kernel Network Developers, Ido Schimmel, Petr Machata,
syzbot, David Ahern, Eric Dumazet
On Tue, Mar 9, 2021 at 12:15 AM Ido Schimmel <idosch@idosch.org> wrote:
>
> On Mon, Mar 08, 2021 at 07:47:31PM -0700, David Ahern wrote:
> > [ cc Ido and Petr ]
> >
> > On 3/8/21 12:21 PM, Wei Wang wrote:
> > > diff --git a/include/net/nexthop.h b/include/net/nexthop.h
> > > index 7bc057aee40b..48956b144689 100644
> > > --- a/include/net/nexthop.h
> > > +++ b/include/net/nexthop.h
> > > @@ -410,31 +410,39 @@ static inline struct fib_nh *fib_info_nh(struct fib_info *fi, int nhsel)
> > > int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
> > > struct netlink_ext_ack *extack);
> > >
> > > -static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
> > > +static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh,
> > > + bool bh_disabled)
> >
> > Hi Wei: I would prefer not to have a second argument to nexthop_fib6_nh
> > for 1 code path, and a control path at that.
> >
> > > {
> > > struct nh_info *nhi;
> > >
> > > if (nh->is_group) {
> > > struct nh_group *nh_grp;
> > >
> > > - nh_grp = rcu_dereference_rtnl(nh->nh_grp);
> > > + if (bh_disabled)
> > > + nh_grp = rcu_dereference_bh_rtnl(nh->nh_grp);
> > > + else
> > > + nh_grp = rcu_dereference_rtnl(nh->nh_grp);
> > > nh = nexthop_mpath_select(nh_grp, 0);
> > > if (!nh)
> > > return NULL;
> > > }
> > >
> > > - nhi = rcu_dereference_rtnl(nh->nh_info);
> > > + if (bh_disabled)
> > > + nhi = rcu_dereference_bh_rtnl(nh->nh_info);
> > > + else
> > > + nhi = rcu_dereference_rtnl(nh->nh_info);
> > > if (nhi->family == AF_INET6)
> > > return &nhi->fib6_nh;
> > >
> > > return NULL;
> > > }
> > >
> >
> > I am wary of duplicating code, but this helper is simple enough that it
> > should be ok with proper documentation.
> >
> > Ido/Petr: I think your resilient hashing patch set touches this helper.
> > How ugly does it get to have a second version?
>
> It actually doesn't touch this helper. Looks fine to me:
Thanks David and Ido.
To clarify, David, you suggest we add a separate function instead of
adding an extra parameter, right?
>
>
> diff --git a/include/net/nexthop.h b/include/net/nexthop.h
> index ba94868a21d5..6df9c12546fd 100644
> --- a/include/net/nexthop.h
> +++ b/include/net/nexthop.h
> @@ -496,6 +496,26 @@ static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
> return NULL;
> }
>
> +static inline struct fib6_nh *nexthop_fib6_nh_bh(struct nexthop *nh)
> +{
> + struct nh_info *nhi;
> +
> + if (nh->is_group) {
> + struct nh_group *nh_grp;
> +
> + nh_grp = rcu_dereference_bh(nh->nh_grp);
> + nh = nexthop_mpath_select(nh_grp, 0);
> + if (!nh)
> + return NULL;
> + }
> +
> + nhi = rcu_dereference_bh(nh->nh_info);
> + if (nhi->family == AF_INET6)
> + return &nhi->fib6_nh;
> +
> + return NULL;
> +}
> +
> static inline struct net_device *fib6_info_nh_dev(struct fib6_info *f6i)
> {
> struct fib6_nh *fib6_nh;
> diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c
> index ef9d022e693f..679699e953f1 100644
> --- a/net/ipv6/ip6_fib.c
> +++ b/net/ipv6/ip6_fib.c
> @@ -2486,7 +2486,7 @@ static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
> const struct net_device *dev;
>
> if (rt->nh)
> - fib6_nh = nexthop_fib6_nh(rt->nh);
> + fib6_nh = nexthop_fib6_nh_bh(rt->nh);
>
> seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] ipv6: fix suspecious RCU usage warning
2021-03-09 17:32 ` Wei Wang
@ 2021-03-09 19:33 ` David Ahern
2021-03-09 19:37 ` Wei Wang
0 siblings, 1 reply; 6+ messages in thread
From: David Ahern @ 2021-03-09 19:33 UTC (permalink / raw)
To: Wei Wang, Ido Schimmel
Cc: David S . Miller, Jakub Kicinski,
Linux Kernel Network Developers, Ido Schimmel, Petr Machata,
syzbot, David Ahern, Eric Dumazet
On 3/9/21 10:32 AM, Wei Wang wrote:
> Thanks David and Ido.
> To clarify, David, you suggest we add a separate function instead of
> adding an extra parameter, right?
for this case I think it is the better way to go.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] ipv6: fix suspecious RCU usage warning
2021-03-09 19:33 ` David Ahern
@ 2021-03-09 19:37 ` Wei Wang
0 siblings, 0 replies; 6+ messages in thread
From: Wei Wang @ 2021-03-09 19:37 UTC (permalink / raw)
To: David Ahern
Cc: Ido Schimmel, David S . Miller, Jakub Kicinski,
Linux Kernel Network Developers, Ido Schimmel, Petr Machata,
syzbot, David Ahern, Eric Dumazet
On Tue, Mar 9, 2021 at 11:33 AM David Ahern <dsahern@gmail.com> wrote:
>
> On 3/9/21 10:32 AM, Wei Wang wrote:
> > Thanks David and Ido.
> > To clarify, David, you suggest we add a separate function instead of
> > adding an extra parameter, right?
>
> for this case I think it is the better way to go.
OK. Will send out v2 for this.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2021-03-09 19:38 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-03-08 19:21 [PATCH net] ipv6: fix suspecious RCU usage warning Wei Wang
2021-03-09 2:47 ` David Ahern
2021-03-09 8:15 ` Ido Schimmel
2021-03-09 17:32 ` Wei Wang
2021-03-09 19:33 ` David Ahern
2021-03-09 19:37 ` Wei Wang
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).