1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
5 #include <linux/btf_ids.h>
6 #include <linux/slab.h>
7 #include <linux/vmalloc.h>
8 #include <linux/etherdevice.h>
9 #include <linux/filter.h>
10 #include <linux/sched/signal.h>
11 #include <net/bpf_sk_storage.h>
14 #include <net/net_namespace.h>
15 #include <linux/error-injection.h>
16 #include <linux/smp.h>
17 #include <linux/sock_diag.h>
19 #define CREATE_TRACE_POINTS
20 #include <trace/events/bpf_test_run.h>
22 struct bpf_test_timer {
23 enum { NO_PREEMPT, NO_MIGRATE } mode;
25 u64 time_start, time_spent;
28 static void bpf_test_timer_enter(struct bpf_test_timer *t)
32 if (t->mode == NO_PREEMPT)
37 t->time_start = ktime_get_ns();
40 static void bpf_test_timer_leave(struct bpf_test_timer *t)
45 if (t->mode == NO_PREEMPT)
52 static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration)
58 t->time_spent += ktime_get_ns() - t->time_start;
59 do_div(t->time_spent, t->i);
60 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
65 if (signal_pending(current)) {
66 /* During iteration: we've been cancelled, abort. */
72 /* During iteration: we need to reschedule between runs. */
73 t->time_spent += ktime_get_ns() - t->time_start;
74 bpf_test_timer_leave(t);
76 bpf_test_timer_enter(t);
79 /* Do another round. */
87 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
88 u32 *retval, u32 *time, bool xdp)
90 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
91 struct bpf_test_timer t = { NO_MIGRATE };
92 enum bpf_cgroup_storage_type stype;
95 for_each_cgroup_storage_type(stype) {
96 storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
97 if (IS_ERR(storage[stype])) {
98 storage[stype] = NULL;
99 for_each_cgroup_storage_type(stype)
100 bpf_cgroup_storage_free(storage[stype]);
108 bpf_test_timer_enter(&t);
110 ret = bpf_cgroup_storage_set(storage);
115 *retval = bpf_prog_run_xdp(prog, ctx);
117 *retval = BPF_PROG_RUN(prog, ctx);
119 bpf_cgroup_storage_unset();
120 } while (bpf_test_timer_continue(&t, repeat, &ret, time));
121 bpf_test_timer_leave(&t);
123 for_each_cgroup_storage_type(stype)
124 bpf_cgroup_storage_free(storage[stype]);
129 static int bpf_test_finish(const union bpf_attr *kattr,
130 union bpf_attr __user *uattr, const void *data,
131 u32 size, u32 retval, u32 duration)
133 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
135 u32 copy_size = size;
137 /* Clamp copy if the user has provided a size hint, but copy the full
138 * buffer if not to retain old behaviour.
140 if (kattr->test.data_size_out &&
141 copy_size > kattr->test.data_size_out) {
142 copy_size = kattr->test.data_size_out;
146 if (data_out && copy_to_user(data_out, data, copy_size))
148 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
150 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
152 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
157 trace_bpf_test_finish(&err);
161 /* Integer types of various sizes and pointer combinations cover variety of
162 * architecture dependent calling conventions. 7+ can be supported in the
166 __diag_ignore(GCC, 8, "-Wmissing-prototypes",
167 "Global functions as their definitions will be in vmlinux BTF");
168 int noinline bpf_fentry_test1(int a)
173 int noinline bpf_fentry_test2(int a, u64 b)
178 int noinline bpf_fentry_test3(char a, int b, u64 c)
183 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
185 return (long)a + b + c + d;
188 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
190 return a + (long)b + c + d + e;
193 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
195 return a + (long)b + c + d + (long)e + f;
198 struct bpf_fentry_test_t {
199 struct bpf_fentry_test_t *a;
202 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
207 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
212 int noinline bpf_modify_return_test(int a, int *b)
218 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
220 return a + b + c + d;
223 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
228 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
235 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
237 BTF_SET_START(test_sk_kfunc_ids)
238 BTF_ID(func, bpf_kfunc_call_test1)
239 BTF_ID(func, bpf_kfunc_call_test2)
240 BTF_ID(func, bpf_kfunc_call_test3)
241 BTF_SET_END(test_sk_kfunc_ids)
243 bool bpf_prog_test_check_kfunc_call(u32 kfunc_id)
245 return btf_id_set_contains(&test_sk_kfunc_ids, kfunc_id);
248 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
249 u32 headroom, u32 tailroom)
251 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
252 u32 user_size = kattr->test.data_size_in;
255 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
256 return ERR_PTR(-EINVAL);
258 if (user_size > size)
259 return ERR_PTR(-EMSGSIZE);
261 data = kzalloc(size + headroom + tailroom, GFP_USER);
263 return ERR_PTR(-ENOMEM);
265 if (copy_from_user(data + headroom, data_in, user_size)) {
267 return ERR_PTR(-EFAULT);
273 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
274 const union bpf_attr *kattr,
275 union bpf_attr __user *uattr)
277 struct bpf_fentry_test_t arg = {};
278 u16 side_effect = 0, ret = 0;
279 int b = 2, err = -EFAULT;
282 if (kattr->test.flags || kattr->test.cpu)
285 switch (prog->expected_attach_type) {
286 case BPF_TRACE_FENTRY:
287 case BPF_TRACE_FEXIT:
288 if (bpf_fentry_test1(1) != 2 ||
289 bpf_fentry_test2(2, 3) != 5 ||
290 bpf_fentry_test3(4, 5, 6) != 15 ||
291 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
292 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
293 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
294 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
295 bpf_fentry_test8(&arg) != 0)
298 case BPF_MODIFY_RETURN:
299 ret = bpf_modify_return_test(1, &b);
307 retval = ((u32)side_effect << 16) | ret;
308 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
313 trace_bpf_test_finish(&err);
317 struct bpf_raw_tp_test_run_info {
318 struct bpf_prog *prog;
324 __bpf_prog_test_run_raw_tp(void *data)
326 struct bpf_raw_tp_test_run_info *info = data;
329 info->retval = BPF_PROG_RUN(info->prog, info->ctx);
333 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
334 const union bpf_attr *kattr,
335 union bpf_attr __user *uattr)
337 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
338 __u32 ctx_size_in = kattr->test.ctx_size_in;
339 struct bpf_raw_tp_test_run_info info;
340 int cpu = kattr->test.cpu, err = 0;
343 /* doesn't support data_in/out, ctx_out, duration, or repeat */
344 if (kattr->test.data_in || kattr->test.data_out ||
345 kattr->test.ctx_out || kattr->test.duration ||
349 if (ctx_size_in < prog->aux->max_ctx_offset ||
350 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
353 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
357 info.ctx = kzalloc(ctx_size_in, GFP_USER);
360 if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
370 current_cpu = get_cpu();
371 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
372 cpu == current_cpu) {
373 __bpf_prog_test_run_raw_tp(&info);
374 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
375 /* smp_call_function_single() also checks cpu_online()
376 * after csd_lock(). However, since cpu is from user
377 * space, let's do an extra quick check to filter out
378 * invalid value before smp_call_function_single().
382 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
388 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
396 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
398 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
399 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
400 u32 size = kattr->test.ctx_size_in;
404 if (!data_in && !data_out)
407 data = kzalloc(max_size, GFP_USER);
409 return ERR_PTR(-ENOMEM);
412 err = bpf_check_uarg_tail_zero(data_in, max_size, size);
418 size = min_t(u32, max_size, size);
419 if (copy_from_user(data, data_in, size)) {
421 return ERR_PTR(-EFAULT);
427 static int bpf_ctx_finish(const union bpf_attr *kattr,
428 union bpf_attr __user *uattr, const void *data,
431 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
433 u32 copy_size = size;
435 if (!data || !data_out)
438 if (copy_size > kattr->test.ctx_size_out) {
439 copy_size = kattr->test.ctx_size_out;
443 if (copy_to_user(data_out, data, copy_size))
445 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
454 * range_is_zero - test whether buffer is initialized
455 * @buf: buffer to check
456 * @from: check from this position
457 * @to: check up until (excluding) this position
459 * This function returns true if the there is a non-zero byte
460 * in the buf in the range [from,to).
462 static inline bool range_is_zero(void *buf, size_t from, size_t to)
464 return !memchr_inv((u8 *)buf + from, 0, to - from);
467 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
469 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
474 /* make sure the fields we don't use are zeroed */
475 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
478 /* mark is allowed */
480 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
481 offsetof(struct __sk_buff, priority)))
484 /* priority is allowed */
486 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
487 offsetof(struct __sk_buff, ifindex)))
490 /* ifindex is allowed */
492 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
493 offsetof(struct __sk_buff, cb)))
498 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
499 offsetof(struct __sk_buff, tstamp)))
502 /* tstamp is allowed */
503 /* wire_len is allowed */
504 /* gso_segs is allowed */
506 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
507 offsetof(struct __sk_buff, gso_size)))
510 /* gso_size is allowed */
512 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
513 sizeof(struct __sk_buff)))
516 skb->mark = __skb->mark;
517 skb->priority = __skb->priority;
518 skb->tstamp = __skb->tstamp;
519 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
521 if (__skb->wire_len == 0) {
522 cb->pkt_len = skb->len;
524 if (__skb->wire_len < skb->len ||
525 __skb->wire_len > GSO_MAX_SIZE)
527 cb->pkt_len = __skb->wire_len;
530 if (__skb->gso_segs > GSO_MAX_SEGS)
532 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
533 skb_shinfo(skb)->gso_size = __skb->gso_size;
538 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
540 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
545 __skb->mark = skb->mark;
546 __skb->priority = skb->priority;
547 __skb->ifindex = skb->dev->ifindex;
548 __skb->tstamp = skb->tstamp;
549 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
550 __skb->wire_len = cb->pkt_len;
551 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
554 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
555 union bpf_attr __user *uattr)
557 bool is_l2 = false, is_direct_pkt_access = false;
558 struct net *net = current->nsproxy->net_ns;
559 struct net_device *dev = net->loopback_dev;
560 u32 size = kattr->test.data_size_in;
561 u32 repeat = kattr->test.repeat;
562 struct __sk_buff *ctx = NULL;
563 u32 retval, duration;
564 int hh_len = ETH_HLEN;
570 if (kattr->test.flags || kattr->test.cpu)
573 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
574 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
576 return PTR_ERR(data);
578 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
584 switch (prog->type) {
585 case BPF_PROG_TYPE_SCHED_CLS:
586 case BPF_PROG_TYPE_SCHED_ACT:
589 case BPF_PROG_TYPE_LWT_IN:
590 case BPF_PROG_TYPE_LWT_OUT:
591 case BPF_PROG_TYPE_LWT_XMIT:
592 is_direct_pkt_access = true;
598 sk = kzalloc(sizeof(struct sock), GFP_USER);
604 sock_net_set(sk, net);
605 sock_init_data(NULL, sk);
607 skb = build_skb(data, 0);
616 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
617 __skb_put(skb, size);
618 if (ctx && ctx->ifindex > 1) {
619 dev = dev_get_by_index(net, ctx->ifindex);
625 skb->protocol = eth_type_trans(skb, dev);
626 skb_reset_network_header(skb);
628 switch (skb->protocol) {
629 case htons(ETH_P_IP):
630 sk->sk_family = AF_INET;
631 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
632 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
633 sk->sk_daddr = ip_hdr(skb)->daddr;
636 #if IS_ENABLED(CONFIG_IPV6)
637 case htons(ETH_P_IPV6):
638 sk->sk_family = AF_INET6;
639 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
640 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
641 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
650 __skb_push(skb, hh_len);
651 if (is_direct_pkt_access)
652 bpf_compute_data_pointers(skb);
653 ret = convert___skb_to_skb(skb, ctx);
656 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
660 if (skb_headroom(skb) < hh_len) {
661 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
663 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
668 memset(__skb_push(skb, hh_len), 0, hh_len);
670 convert_skb_to___skb(skb, ctx);
673 /* bpf program can never convert linear skb to non-linear */
674 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
675 size = skb_headlen(skb);
676 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
678 ret = bpf_ctx_finish(kattr, uattr, ctx,
679 sizeof(struct __sk_buff));
681 if (dev && dev != net->loopback_dev)
684 bpf_sk_storage_free(sk);
690 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
691 union bpf_attr __user *uattr)
693 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
694 u32 headroom = XDP_PACKET_HEADROOM;
695 u32 size = kattr->test.data_size_in;
696 u32 repeat = kattr->test.repeat;
697 struct netdev_rx_queue *rxqueue;
698 struct xdp_buff xdp = {};
699 u32 retval, duration;
704 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
705 prog->expected_attach_type == BPF_XDP_CPUMAP)
707 if (kattr->test.ctx_in || kattr->test.ctx_out)
710 /* XDP have extra tailroom as (most) drivers use full page */
711 max_data_sz = 4096 - headroom - tailroom;
713 data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
715 return PTR_ERR(data);
717 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
718 xdp_init_buff(&xdp, headroom + max_data_sz + tailroom,
720 xdp_prepare_buff(&xdp, data, headroom, size, true);
722 bpf_prog_change_xdp(NULL, prog);
723 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
726 if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
727 size = xdp.data_end - xdp.data;
728 ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
730 bpf_prog_change_xdp(prog, NULL);
735 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
737 /* make sure the fields we don't use are zeroed */
738 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
741 /* flags is allowed */
743 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
744 sizeof(struct bpf_flow_keys)))
750 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
751 const union bpf_attr *kattr,
752 union bpf_attr __user *uattr)
754 struct bpf_test_timer t = { NO_PREEMPT };
755 u32 size = kattr->test.data_size_in;
756 struct bpf_flow_dissector ctx = {};
757 u32 repeat = kattr->test.repeat;
758 struct bpf_flow_keys *user_ctx;
759 struct bpf_flow_keys flow_keys;
760 const struct ethhdr *eth;
761 unsigned int flags = 0;
762 u32 retval, duration;
766 if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
769 if (kattr->test.flags || kattr->test.cpu)
775 data = bpf_test_init(kattr, size, 0, 0);
777 return PTR_ERR(data);
779 eth = (struct ethhdr *)data;
784 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
785 if (IS_ERR(user_ctx)) {
787 return PTR_ERR(user_ctx);
790 ret = verify_user_bpf_flow_keys(user_ctx);
793 flags = user_ctx->flags;
796 ctx.flow_keys = &flow_keys;
798 ctx.data_end = (__u8 *)data + size;
800 bpf_test_timer_enter(&t);
802 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
804 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
805 bpf_test_timer_leave(&t);
810 ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
813 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
814 sizeof(struct bpf_flow_keys));
822 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
823 union bpf_attr __user *uattr)
825 struct bpf_test_timer t = { NO_PREEMPT };
826 struct bpf_prog_array *progs = NULL;
827 struct bpf_sk_lookup_kern ctx = {};
828 u32 repeat = kattr->test.repeat;
829 struct bpf_sk_lookup *user_ctx;
830 u32 retval, duration;
833 if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
836 if (kattr->test.flags || kattr->test.cpu)
839 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
840 kattr->test.data_size_out)
846 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
847 if (IS_ERR(user_ctx))
848 return PTR_ERR(user_ctx);
856 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
859 if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
864 ctx.family = (u16)user_ctx->family;
865 ctx.protocol = (u16)user_ctx->protocol;
866 ctx.dport = (u16)user_ctx->local_port;
867 ctx.sport = (__force __be16)user_ctx->remote_port;
869 switch (ctx.family) {
871 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
872 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
875 #if IS_ENABLED(CONFIG_IPV6)
877 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
878 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
887 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
893 progs->items[0].prog = prog;
895 bpf_test_timer_enter(&t);
897 ctx.selected_sk = NULL;
898 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
899 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
900 bpf_test_timer_leave(&t);
905 user_ctx->cookie = 0;
906 if (ctx.selected_sk) {
907 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
912 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
915 ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
917 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
920 bpf_prog_array_free(progs);