Catch different invalid XDP_REDIRECT and bpf_redirect_map API usage. It is fairly easy to create a dangling redirect_info->map pointer, which (until John or Daniel fix this) can crash the kernel.
The intended usage of the BPF helper bpf_redirect_map(), is to return XDP_REDIRECT action after invoking it, but there is nothing stopping the bpf_prog to return anything else. When XDP_REDIRECT isn't returned, then a dangling ->map pointer is left behind, as xdp_do_redirect() isn't called. This also happens for drivers not implementing XDP_REDIRECT, as they are not aware of this new XDP_REDIRECT return code, they leave the map pointer dangling. The simply solution to check for a dangling ->map pointer after each driver napi->poll() invocation, see xdp_do_map_check(). This patch also add a check for a dangling ->map_to_flush pointer. This should be considered a driver bug, as the driver contract is that a pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the same cpu context. Note, we need to check after each drivers napi->poll call, as: 1. DevA poll call bpf_redirect_map() but not xdp_do_redirect() 2. DevB bpf_prog uses bpf_redirect() and call xdp_do_redirect() which now use map from DevA Signed-off-by: Jesper Dangaard Brouer <bro...@redhat.com> --- include/linux/filter.h | 1 + net/core/dev.c | 3 +++ net/core/filter.c | 25 +++++++++++++++++++++++++ 3 files changed, 29 insertions(+) diff --git a/include/linux/filter.h b/include/linux/filter.h index d29e58fde364..0c48941e0022 100644 --- a/include/linux/filter.h +++ b/include/linux/filter.h @@ -724,6 +724,7 @@ int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, struct bpf_prog *prog); void xdp_do_flush_map(void); +void xdp_do_map_check(struct napi_struct *napi); void bpf_warn_invalid_xdp_action(u32 act); void bpf_warn_invalid_xdp_redirect(u32 ifindex); diff --git a/net/core/dev.c b/net/core/dev.c index 6f845e4fec17..7eac642b469f 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -5320,6 +5320,7 @@ static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) */ rc = napi->poll(napi, BUSY_POLL_BUDGET); trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); + xdp_do_map_check(napi); netpoll_poll_unlock(have_poll_lock); if (rc == BUSY_POLL_BUDGET) __napi_schedule(napi); @@ -5367,6 +5368,7 @@ void napi_busy_loop(unsigned int napi_id, } work = napi_poll(napi, BUSY_POLL_BUDGET); trace_napi_poll(napi, work, BUSY_POLL_BUDGET); + xdp_do_map_check(napi); count: if (work > 0) __NET_ADD_STATS(dev_net(napi->dev), @@ -5529,6 +5531,7 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll) if (test_bit(NAPI_STATE_SCHED, &n->state)) { work = n->poll(n, weight); trace_napi_poll(n, work, weight); + xdp_do_map_check(n); } WARN_ON_ONCE(work > weight); diff --git a/net/core/filter.c b/net/core/filter.c index 3767470cab6c..f0e1135eeb9d 100644 --- a/net/core/filter.c +++ b/net/core/filter.c @@ -2500,6 +2500,31 @@ void xdp_do_flush_map(void) } EXPORT_SYMBOL_GPL(xdp_do_flush_map); +void xdp_do_map_check(struct napi_struct *napi) +{ + struct redirect_info *ri = this_cpu_ptr(&redirect_info); + + /* XDP drivers (and XDP-generic) must invoke xdp_do_redirect() + * when bpf_prog use helper bpf_redirect_map(), else the map + * pointer can be left dangling. Catch this invalid API + * usage, instead of potentially crashing. + */ + if (ri->map) { + ri->map = NULL; + net_err_ratelimited("%s: caught invalid XDP bpf_redirect_map\n", + napi->dev->name); + trace_xdp_exception(napi->dev, NULL, XDP_REDIRECT); + } + if (ri->map_to_flush) { /* Driver bug */ + net_err_ratelimited("%s: XDP driver miss xdp_do_flush_map\n", + napi->dev->name); + trace_xdp_exception(napi->dev, NULL, XDP_REDIRECT); + /* Flush map, else pkts can be stuck on XDP TXq */ + xdp_do_flush_map(); + } +} +EXPORT_SYMBOL_GPL(xdp_do_map_check); + static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp, struct bpf_prog *xdp_prog) {