| Message ID | 20230529201914.69828-4-bpoirier@nvidia.com |
|---|---|
| State | New |
| Headers | show |
| Series | [net-next,1/4] nexthop: Factor out hash threshold fdb nexthop selection | expand |
On 5/29/23 2:19 PM, Benjamin Poirier wrote: > diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c > index c12acbf39659..ca501ced04fb 100644 > --- a/net/ipv4/nexthop.c > +++ b/net/ipv4/nexthop.c > @@ -1186,6 +1186,7 @@ static struct nexthop *nexthop_select_path_fdb(struct nh_group *nhg, int hash) > static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash) > { > struct nexthop *rc = NULL; > + bool first = false; > int i; > > if (nhg->fdb_nh) > @@ -1194,20 +1195,24 @@ static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash) > for (i = 0; i < nhg->num_nh; ++i) { > struct nh_grp_entry *nhge = &nhg->nh_entries[i]; > > - if (hash > atomic_read(&nhge->hthr.upper_bound)) > - continue; > - > /* nexthops always check if it is good and does > * not rely on a sysctl for this behavior > */ > - if (nexthop_is_good_nh(nhge->nh)) > - return nhge->nh; > + if (!nexthop_is_good_nh(nhge->nh)) > + continue; > > - if (!rc) > + if (!first) { Setting 'first' and 'rc' are equivalent, so 'first' is not needed. As I recall it was used in fib_select_multipath before the nexthop refactoring (eba618abacade) because nhsel == 0 is valid, so the loop could not rely on it. > rc = nhge->nh; > + first = true; > + } > + > + if (hash > atomic_read(&nhge->hthr.upper_bound)) > + continue; > + > + return nhge->nh; > } > > - return rc; > + return rc ? : nhg->nh_entries[0].nh; > } > > static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)
diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c index c12acbf39659..ca501ced04fb 100644 --- a/net/ipv4/nexthop.c +++ b/net/ipv4/nexthop.c @@ -1186,6 +1186,7 @@ static struct nexthop *nexthop_select_path_fdb(struct nh_group *nhg, int hash) static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash) { struct nexthop *rc = NULL; + bool first = false; int i; if (nhg->fdb_nh) @@ -1194,20 +1195,24 @@ static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash) for (i = 0; i < nhg->num_nh; ++i) { struct nh_grp_entry *nhge = &nhg->nh_entries[i]; - if (hash > atomic_read(&nhge->hthr.upper_bound)) - continue; - /* nexthops always check if it is good and does * not rely on a sysctl for this behavior */ - if (nexthop_is_good_nh(nhge->nh)) - return nhge->nh; + if (!nexthop_is_good_nh(nhge->nh)) + continue; - if (!rc) + if (!first) { rc = nhge->nh; + first = true; + } + + if (hash > atomic_read(&nhge->hthr.upper_bound)) + continue; + + return nhge->nh; } - return rc; + return rc ? : nhg->nh_entries[0].nh; } static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)
With legacy nexthops, when net.ipv4.fib_multipath_use_neigh is set, fib_select_multipath() will never set res->nhc to a nexthop that is not good (as per fib_good_nh()). OTOH, with nexthop objects, nexthop_select_path_hthr() may return a nexthop that failed the nexthop_is_good_nh() test even if there was one that passed. Refactor nexthop_select_path_hthr() to follow a selection logic more similar to fib_select_multipath(). The issue can be demonstrated with the following sequence of commands. The first block shows that things work as expected with legacy nexthops. The last sequence of `ip rou get` in the second block shows the problem case - some routes still use the .2 nexthop. sysctl net.ipv4.fib_multipath_use_neigh=1 ip link add dummy1 up type dummy ip rou add 198.51.100.0/24 nexthop via 192.0.2.1 dev dummy1 onlink nexthop via 192.0.2.2 dev dummy1 onlink for i in {10..19}; do ip -o rou get 198.51.100.$i; done ip neigh add 192.0.2.1 dev dummy1 nud failed echo ".1 failed:" # results should not use .1 for i in {10..19}; do ip -o rou get 198.51.100.$i; done ip neigh del 192.0.2.1 dev dummy1 ip neigh add 192.0.2.2 dev dummy1 nud failed echo ".2 failed:" # results should not use .2 for i in {10..19}; do ip -o rou get 198.51.100.$i; done ip link del dummy1 ip link add dummy1 up type dummy ip nexthop add id 1 via 192.0.2.1 dev dummy1 onlink ip nexthop add id 2 via 192.0.2.2 dev dummy1 onlink ip nexthop add id 1001 group 1/2 ip rou add 198.51.100.0/24 nhid 1001 for i in {10..19}; do ip -o rou get 198.51.100.$i; done ip neigh add 192.0.2.1 dev dummy1 nud failed echo ".1 failed:" # results should not use .1 for i in {10..19}; do ip -o rou get 198.51.100.$i; done ip neigh del 192.0.2.1 dev dummy1 ip neigh add 192.0.2.2 dev dummy1 nud failed echo ".2 failed:" # results should not use .2 for i in {10..19}; do ip -o rou get 198.51.100.$i; done ip link del dummy1 Signed-off-by: Ido Schimmel <idosch@nvidia.com> Signed-off-by: Benjamin Poirier <bpoirier@nvidia.com> --- net/ipv4/nexthop.c | 19 ++++++++++++------- 1 file changed, 12 insertions(+), 7 deletions(-)