On Fri, May 17, 2019 at 4:32 PM Eric Dumazet <eric.duma...@gmail.com> wrote:
>
>
>
> On 5/17/19 8:56 AM, Willem de Bruijn wrote:
> > From: Willem de Bruijn <will...@google.com>
> >
> > Processes can request ipv6 flowlabels with cmsg IPV6_FLOWINFO.
> > If not set, by default an autogenerated flowlabel is selected.
> >
> > Explicit flowlabels require a control operation per label plus a
> > datapath check on every connection (every datagram if unconnected).
> >
> > This is particularly expensive on unconnected sockets with many
> > connections, such as QUIC.
> >
> > In the common case, where no lease is exclusive, the check can be
> > safely elided, as both lease request and check trivially succeed.
> > Indeed, autoflowlabel does the same (even with exclusive leases).
> >
> > Elide the check if no process has requested an exclusive lease.
> >
> > This is an optimization. Robust applications still have to revert to
> > requesting leases if the fast path fails due to an exclusive lease.
> >
> > This is decidedly an RFC patch:
> > - need to update all fl6_sock_lookup callers, not just udp
> > - behavior should be per-netns isolated
> >
> > Other approaches considered:
> > - a single "get all flowlabels, non-exclusive" flowlabel get request
> > if set, elide fl6_sock_lookup and fail exclusive lease requests
> >
> > - sysctls (only useful if on by default, with static_branch)
> > A) "non-exclusive mode", failing all exclusive lease requests:
> > processes already have to be robust against lease failure
> > B) just bypass check in fl6_sock_lookup, like autoflowlabel
> >
> > Signed-off-by: Willem de Bruijn <will...@google.com>
> > ---
> > include/net/ipv6.h | 11 +++++++++++
> > net/ipv6/ip6_flowlabel.c | 6 ++++++
> > net/ipv6/udp.c | 8 ++++----
> > 3 files changed, 21 insertions(+), 4 deletions(-)
> >
> > diff --git a/include/net/ipv6.h b/include/net/ipv6.h
> > index daf80863d3a50..8881cee572410 100644
> > --- a/include/net/ipv6.h
> > +++ b/include/net/ipv6.h
> > @@ -17,6 +17,7 @@
> > #include <linux/hardirq.h>
> > #include <linux/jhash.h>
> > #include <linux/refcount.h>
> > +#include <linux/jump_label.h>
> > #include <net/if_inet6.h>
> > #include <net/ndisc.h>
> > #include <net/flow.h>
> > @@ -343,7 +344,17 @@ static inline void txopt_put(struct ipv6_txoptions
> > *opt)
> > kfree_rcu(opt, rcu);
> > }
> >
> > +extern struct static_key_false ipv6_flowlabel_exclusive;
> > struct ip6_flowlabel *fl6_sock_lookup(struct sock *sk, __be32 label);
> > +static inline struct ip6_flowlabel *fl6_sock_verify(struct sock *sk,
> > + __be32 label)
> > +{
> > + if (static_branch_unlikely(&ipv6_flowlabel_exclusive))
> > + return fl6_sock_lookup(sk, label) ? : ERR_PTR(-ENOENT);
> > +
> > + return NULL;
> > +}
> > +
> > struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
> > struct ip6_flowlabel *fl,
> > struct ipv6_txoptions *fopt);
> > diff --git a/net/ipv6/ip6_flowlabel.c b/net/ipv6/ip6_flowlabel.c
> > index be5f3d7ceb966..d5f4233b04e0c 100644
> > --- a/net/ipv6/ip6_flowlabel.c
> > +++ b/net/ipv6/ip6_flowlabel.c
> > @@ -57,6 +57,8 @@ static DEFINE_SPINLOCK(ip6_fl_lock);
> >
> > static DEFINE_SPINLOCK(ip6_sk_fl_lock);
> >
> > +DEFINE_STATIC_KEY_FALSE(ipv6_flowlabel_exclusive);
> > +
> > #define for_each_fl_rcu(hash, fl) \
> > for (fl = rcu_dereference_bh(fl_ht[(hash)]); \
> > fl != NULL; \
> > @@ -98,6 +100,8 @@ static void fl_free_rcu(struct rcu_head *head)
> > {
> > struct ip6_flowlabel *fl = container_of(head, struct ip6_flowlabel,
> > rcu);
> >
> > + if (fl->share != IPV6_FL_S_NONE && fl->share != IPV6_FL_S_ANY)
> > + static_branch_dec(&ipv6_flowlabel_exclusive);
>
> static_branch_dec() can not be invoked from a rcu call back.
>
> > if (fl->share == IPV6_FL_S_PROCESS)
> > put_pid(fl->owner.pid);
> > kfree(fl->opt);
> > @@ -423,6 +427,8 @@ fl_create(struct net *net, struct sock *sk, struct
> > in6_flowlabel_req *freq,
> > }
> > fl->dst = freq->flr_dst;
> > atomic_set(&fl->users, 1);
> > + if (fl->share != IPV6_FL_S_ANY)
> > + static_branch_inc(&ipv6_flowlabel_exclusive);
>
>
> Can this be used by unpriv users ?
>
> If yes, then you want to use static_key_false_deferred instead
Ah of course. Yes, any user can exercise this API. Thanks, Eric. I'll
take a look at both points.
