Re: [PATCH net-next v3 1/2] ipv6: check mc_forwarding before calling ip6mr_get_route()

From: Wenbo Sun

Date: Thu Nov 19 2026 - 22:22:55 EST


Thank you for the detailed review on v3 1/2. We reproduced and verified each
of the findings in a User-Mode Linux (UML) environment (`CONFIG_IPV6_MROUTE=y`)
and prepared a focused single-patch v4 to address them:

1. Root-Cause Analysis & Local UML Reproduction of v3 1/2 Findings:

Base UML topology:
ip link add d0 type veth peer name d1
ip link set d0 up && ip link set d1 up
ip -6 addr add 2001:db8::2/64 dev d0
ip -6 addr add ff0e::1 dev d0 autojoin

(a) [High] Routable multicast query when mc_forwarding=1:
In `inet6_rtm_getroute()`, `skb->dev` is never set on the newly
allocated reply skb, and `ip6mr_get_route()` looks up
`ip6mr_cache_find(mrt, &rt->rt6i_src.addr, &rt->rt6i_dst.addr)`, where
`rt` is the per-CPU route copied from `f6i->fib6_dst` (`ff00::/8`).
We reproduced this in UML by opening an `AF_INET6` `SOCK_RAW`
(`IPPROTO_ICMPV6`) socket, enabling `MRT6_INIT`, registering `d0` via
`MRT6_ADD_MIF`, and adding `(S=2001:db8::3, G=ff0e::1)` via
`MRT6_ADD_MFC`:
MRT6_ADD_MIF(mifi=0, pifi=3) OK
MRT6_ADD_MFC(S=2001:db8::3, G=ff0e::1) OK
mc_forwarding=1
$ ip -6 route get ff0e::1 from 2001:db8::3 iif d0
RTNETLINK answers: No such device
ip_exit=512
Because `ip6mr_get_route()` checks `(::, ff00::)` instead of the
queried `(S,G)` and `skb->dev` is NULL, it misses even when the MFC
entry exists. Making `mc_forwarding=1` MFC queries work requires
passing `(src, dest)` and handling `skb->dev` / `skb_realloc_headroom()`
across `inet6_rtm_getroute()` and `ip6mr_get_route()`, which we now
explicitly state as out of scope in the commit message and drop patch
2/2 in v4 so we do not alter the uAPI errno to -ENODEV prematurely.

(b) [Medium 1] Regression when a multicast query matches a non-multicast
route (e.g. `::/0`):
In v3, replacing `ipv6_addr_is_multicast(&rt6_dst->addr)` with
`ipv6_addr_is_multicast(mc_dst)` caused queries matching a non-multicast
FIB route (when `ff00::/8` is absent, such as on `lo`, an l3mdev
master, or after deleting `ff00::/8`) to enter `ip6mr_get_route()` when
`mc_forwarding=1`.
Reproduced on v3 in UML:
# ip -6 route del multicast ff00::/8 dev d0 table local
# ip -6 route add ::/0 dev d0
# (with MRT6_INIT / mc_forwarding=1):
$ ip -6 route get ff0e::1 from 2001:db8::3 iif d0
RTNETLINK answers: No such device
ip_exit=512

(c) [Medium 2] `RTM_F_FIB_MATCH` (`fibmatch`) missing link-local exclusion:
In `inet6_rtm_getroute()`, the `fibmatch` branch calls `rt6_fill_node()`
with `dst = NULL, dest = NULL`. In v3, `mc_dst` fell back to
`&rt6_dst->addr` (`ff00::`, scope 0), so link-local `fibmatch` queries
still entered `ip6mr_get_route()` when `mc_forwarding=1` (whereas IPv4
`fibmatch` uses `fib_dump_info()` and never calls `ipmr_get_route()`).
Reproduced on v3 in UML (with `MRT6_INIT` / `mc_forwarding=1`):
$ ip -6 route get fibmatch ff02::1 from 2001:db8::3 iif d0
RTNETLINK answers: No such device
ip_exit=512

(d) [Low] Fixes tag / net vs net-next:
Because this patch aligns long-standing non-crash route-get reporting
semantics with IPv4 (while full `mc_forwarding=1` MFC lookup in
`ip6mr_get_route()` remains a separate effort), we keep targeting
net-next without a Fixes tag, and have softened the commit message
accordingly.

2. Concrete Fix in v4 & UML Verification Results:

In v4, we converge the series into a single patch that checks `dest != NULL`
(so `fibmatch` queries always emit the matched FIB6 entry and `RTA_IIF`),
keeps `ipv6_addr_is_multicast(&rt6_dst->addr)` (avoiding the non-multicast
route regression), and tests `ipv6_addr_type(dest)` against
`(IPV6_ADDR_LOOPBACK | IPV6_ADDR_LINKLOCAL)`:

if (dest && ipv6_addr_is_multicast(&rt6_dst->addr) &&
!(ipv6_addr_type(dest) &
(IPV6_ADDR_LOOPBACK | IPV6_ADDR_LINKLOCAL)) &&
atomic_read(&net->ipv6.devconf_all->mc_forwarding))

Verified on v4 in UML (`FAILURES=0, TEST_EXIT_CODE=0`):
- Test 1 (`mc_forwarding=0`, `ff0e::1` with `iif d0`):
`ff0e::1 dev d0 proto kernel metric 256 iif d0 pref medium` (PASS)
- Test 2 (`mc_forwarding=1`, link-local `ff02::1` with `iif d0`):
`multicast ff02::1 dev d0 proto kernel metric 256 iif d0 pref medium` (PASS)
- Test 3 (`mc_forwarding=1`, interface-local `ff01::1` with `iif d0`):
`multicast ff01::1 dev d0 proto kernel metric 256 iif d0 pref medium` (PASS)
- Test 4 (`mc_forwarding=1`, `fibmatch` for `ff02::1` and `ff0e::1` with
`iif d0`):
`multicast ff00::/8 dev d0 table local proto kernel metric 256 iif d0 pref medium`
(`ip_exit=0`, PASS)
- Test 5 (`mc_forwarding=1`, `ff0e::1` matching `::/0` after removing
`ff00::/8`):
`ff0e::1 from 2001:db8::3 dev d0 proto boot src 2001:db8::2 metric 1024 iif d0 pref medium`
(`ip_exit=0`, PASS)
- Test 6 (IPv6 unicast with `iif`, IPv6 multicast without `iif`, IPv4
multicast with `iif` regressions): all PASS.

Thanks,
Wenbo

pw-bot: cr