1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
5 #include <linux/slab.h>
6 #include <linux/vmalloc.h>
7 #include <linux/etherdevice.h>
8 #include <linux/filter.h>
9 #include <linux/sched/signal.h>
10 #include <net/bpf_sk_storage.h>
13 #include <net/net_namespace.h>
14 #include <linux/error-injection.h>
15 #include <linux/smp.h>
16 #include <linux/sock_diag.h>
18 #define CREATE_TRACE_POINTS
19 #include <trace/events/bpf_test_run.h>
21 struct bpf_test_timer {
22 enum { NO_PREEMPT, NO_MIGRATE } mode;
24 u64 time_start, time_spent;
27 static void bpf_test_timer_enter(struct bpf_test_timer *t)
31 if (t->mode == NO_PREEMPT)
36 t->time_start = ktime_get_ns();
39 static void bpf_test_timer_leave(struct bpf_test_timer *t)
44 if (t->mode == NO_PREEMPT)
51 static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration)
57 t->time_spent += ktime_get_ns() - t->time_start;
58 do_div(t->time_spent, t->i);
59 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
64 if (signal_pending(current)) {
65 /* During iteration: we've been cancelled, abort. */
71 /* During iteration: we need to reschedule between runs. */
72 t->time_spent += ktime_get_ns() - t->time_start;
73 bpf_test_timer_leave(t);
75 bpf_test_timer_enter(t);
78 /* Do another round. */
86 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
87 u32 *retval, u32 *time, bool xdp)
89 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
90 struct bpf_test_timer t = { NO_MIGRATE };
91 enum bpf_cgroup_storage_type stype;
94 for_each_cgroup_storage_type(stype) {
95 storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
96 if (IS_ERR(storage[stype])) {
97 storage[stype] = NULL;
98 for_each_cgroup_storage_type(stype)
99 bpf_cgroup_storage_free(storage[stype]);
107 bpf_test_timer_enter(&t);
109 ret = bpf_cgroup_storage_set(storage);
114 *retval = bpf_prog_run_xdp(prog, ctx);
116 *retval = BPF_PROG_RUN(prog, ctx);
118 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)
219 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
221 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
222 u32 headroom, u32 tailroom)
224 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
225 u32 user_size = kattr->test.data_size_in;
228 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
229 return ERR_PTR(-EINVAL);
231 if (user_size > size)
232 return ERR_PTR(-EMSGSIZE);
234 data = kzalloc(size + headroom + tailroom, GFP_USER);
236 return ERR_PTR(-ENOMEM);
238 if (copy_from_user(data + headroom, data_in, user_size)) {
240 return ERR_PTR(-EFAULT);
246 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
247 const union bpf_attr *kattr,
248 union bpf_attr __user *uattr)
250 struct bpf_fentry_test_t arg = {};
251 u16 side_effect = 0, ret = 0;
252 int b = 2, err = -EFAULT;
255 if (kattr->test.flags || kattr->test.cpu)
258 switch (prog->expected_attach_type) {
259 case BPF_TRACE_FENTRY:
260 case BPF_TRACE_FEXIT:
261 if (bpf_fentry_test1(1) != 2 ||
262 bpf_fentry_test2(2, 3) != 5 ||
263 bpf_fentry_test3(4, 5, 6) != 15 ||
264 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
265 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
266 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
267 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
268 bpf_fentry_test8(&arg) != 0)
271 case BPF_MODIFY_RETURN:
272 ret = bpf_modify_return_test(1, &b);
280 retval = ((u32)side_effect << 16) | ret;
281 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
286 trace_bpf_test_finish(&err);
290 struct bpf_raw_tp_test_run_info {
291 struct bpf_prog *prog;
297 __bpf_prog_test_run_raw_tp(void *data)
299 struct bpf_raw_tp_test_run_info *info = data;
302 info->retval = BPF_PROG_RUN(info->prog, info->ctx);
306 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
307 const union bpf_attr *kattr,
308 union bpf_attr __user *uattr)
310 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
311 __u32 ctx_size_in = kattr->test.ctx_size_in;
312 struct bpf_raw_tp_test_run_info info;
313 int cpu = kattr->test.cpu, err = 0;
316 /* doesn't support data_in/out, ctx_out, duration, or repeat */
317 if (kattr->test.data_in || kattr->test.data_out ||
318 kattr->test.ctx_out || kattr->test.duration ||
322 if (ctx_size_in < prog->aux->max_ctx_offset ||
323 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
326 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
330 info.ctx = kzalloc(ctx_size_in, GFP_USER);
333 if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
343 current_cpu = get_cpu();
344 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
345 cpu == current_cpu) {
346 __bpf_prog_test_run_raw_tp(&info);
347 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
348 /* smp_call_function_single() also checks cpu_online()
349 * after csd_lock(). However, since cpu is from user
350 * space, let's do an extra quick check to filter out
351 * invalid value before smp_call_function_single().
355 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
361 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
369 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
371 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
372 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
373 u32 size = kattr->test.ctx_size_in;
377 if (!data_in && !data_out)
380 data = kzalloc(max_size, GFP_USER);
382 return ERR_PTR(-ENOMEM);
385 err = bpf_check_uarg_tail_zero(data_in, max_size, size);
391 size = min_t(u32, max_size, size);
392 if (copy_from_user(data, data_in, size)) {
394 return ERR_PTR(-EFAULT);
400 static int bpf_ctx_finish(const union bpf_attr *kattr,
401 union bpf_attr __user *uattr, const void *data,
404 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
406 u32 copy_size = size;
408 if (!data || !data_out)
411 if (copy_size > kattr->test.ctx_size_out) {
412 copy_size = kattr->test.ctx_size_out;
416 if (copy_to_user(data_out, data, copy_size))
418 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
427 * range_is_zero - test whether buffer is initialized
428 * @buf: buffer to check
429 * @from: check from this position
430 * @to: check up until (excluding) this position
432 * This function returns true if the there is a non-zero byte
433 * in the buf in the range [from,to).
435 static inline bool range_is_zero(void *buf, size_t from, size_t to)
437 return !memchr_inv((u8 *)buf + from, 0, to - from);
440 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
442 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
450 /* make sure the fields we don't use are zeroed */
451 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
454 /* mark is allowed */
456 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
457 offsetof(struct __sk_buff, priority)))
460 /* priority is allowed */
462 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
463 offsetof(struct __sk_buff, ifindex)))
466 /* ifindex is allowed */
468 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
469 offsetof(struct __sk_buff, cb)))
474 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
475 offsetof(struct __sk_buff, tstamp)))
478 /* tstamp is allowed */
479 /* wire_len is allowed */
480 /* gso_segs is allowed */
482 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
483 offsetof(struct __sk_buff, gso_size)))
486 /* gso_size is allowed */
488 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
489 sizeof(struct __sk_buff)))
492 skb->mark = __skb->mark;
493 skb->priority = __skb->priority;
494 skb->tstamp = __skb->tstamp;
495 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
497 if (__skb->wire_len == 0) {
498 cb->pkt_len = skb->len;
500 if (__skb->wire_len < skb->len ||
501 __skb->wire_len > GSO_MAX_SIZE)
503 cb->pkt_len = __skb->wire_len;
506 if (__skb->gso_segs > GSO_MAX_SEGS)
508 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
509 skb_shinfo(skb)->gso_size = __skb->gso_size;
514 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
516 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
521 __skb->mark = skb->mark;
522 __skb->priority = skb->priority;
523 __skb->ifindex = skb->dev->ifindex;
524 __skb->tstamp = skb->tstamp;
525 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
526 __skb->wire_len = cb->pkt_len;
527 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
530 static struct proto bpf_dummy_proto = {
532 .owner = THIS_MODULE,
533 .obj_size = sizeof(struct sock),
536 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
537 union bpf_attr __user *uattr)
539 bool is_l2 = false, is_direct_pkt_access = false;
540 struct net *net = current->nsproxy->net_ns;
541 struct net_device *dev = net->loopback_dev;
542 u32 size = kattr->test.data_size_in;
543 u32 repeat = kattr->test.repeat;
544 struct __sk_buff *ctx = NULL;
545 u32 retval, duration;
546 int hh_len = ETH_HLEN;
552 if (kattr->test.flags || kattr->test.cpu)
555 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
556 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
558 return PTR_ERR(data);
560 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
566 switch (prog->type) {
567 case BPF_PROG_TYPE_SCHED_CLS:
568 case BPF_PROG_TYPE_SCHED_ACT:
571 case BPF_PROG_TYPE_LWT_IN:
572 case BPF_PROG_TYPE_LWT_OUT:
573 case BPF_PROG_TYPE_LWT_XMIT:
574 is_direct_pkt_access = true;
580 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
586 sock_init_data(NULL, sk);
588 skb = build_skb(data, 0);
597 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
598 __skb_put(skb, size);
599 if (ctx && ctx->ifindex > 1) {
600 dev = dev_get_by_index(net, ctx->ifindex);
606 skb->protocol = eth_type_trans(skb, dev);
607 skb_reset_network_header(skb);
609 switch (skb->protocol) {
610 case htons(ETH_P_IP):
611 sk->sk_family = AF_INET;
612 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
613 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
614 sk->sk_daddr = ip_hdr(skb)->daddr;
617 #if IS_ENABLED(CONFIG_IPV6)
618 case htons(ETH_P_IPV6):
619 sk->sk_family = AF_INET6;
620 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
621 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
622 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
631 __skb_push(skb, hh_len);
632 if (is_direct_pkt_access)
633 bpf_compute_data_pointers(skb);
634 ret = convert___skb_to_skb(skb, ctx);
637 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
641 if (skb_headroom(skb) < hh_len) {
642 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
644 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
649 memset(__skb_push(skb, hh_len), 0, hh_len);
651 convert_skb_to___skb(skb, ctx);
654 /* bpf program can never convert linear skb to non-linear */
655 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
656 size = skb_headlen(skb);
657 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
659 ret = bpf_ctx_finish(kattr, uattr, ctx,
660 sizeof(struct __sk_buff));
662 if (dev && dev != net->loopback_dev)
670 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
671 union bpf_attr __user *uattr)
673 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
674 u32 headroom = XDP_PACKET_HEADROOM;
675 u32 size = kattr->test.data_size_in;
676 u32 repeat = kattr->test.repeat;
677 struct netdev_rx_queue *rxqueue;
678 struct xdp_buff xdp = {};
679 u32 retval, duration;
684 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
685 prog->expected_attach_type == BPF_XDP_CPUMAP)
687 if (kattr->test.ctx_in || kattr->test.ctx_out)
690 /* XDP have extra tailroom as (most) drivers use full page */
691 max_data_sz = 4096 - headroom - tailroom;
693 data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
695 return PTR_ERR(data);
697 xdp.data_hard_start = data;
698 xdp.data = data + headroom;
699 xdp.data_meta = xdp.data;
700 xdp.data_end = xdp.data + size;
701 xdp.frame_sz = headroom + max_data_sz + tailroom;
703 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
704 xdp.rxq = &rxqueue->xdp_rxq;
705 bpf_prog_change_xdp(NULL, prog);
706 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
709 if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
710 size = xdp.data_end - xdp.data;
711 ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
713 bpf_prog_change_xdp(prog, NULL);
718 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
720 /* make sure the fields we don't use are zeroed */
721 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
724 /* flags is allowed */
726 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
727 sizeof(struct bpf_flow_keys)))
733 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
734 const union bpf_attr *kattr,
735 union bpf_attr __user *uattr)
737 struct bpf_test_timer t = { NO_PREEMPT };
738 u32 size = kattr->test.data_size_in;
739 struct bpf_flow_dissector ctx = {};
740 u32 repeat = kattr->test.repeat;
741 struct bpf_flow_keys *user_ctx;
742 struct bpf_flow_keys flow_keys;
743 const struct ethhdr *eth;
744 unsigned int flags = 0;
745 u32 retval, duration;
749 if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
752 if (kattr->test.flags || kattr->test.cpu)
758 data = bpf_test_init(kattr, size, 0, 0);
760 return PTR_ERR(data);
762 eth = (struct ethhdr *)data;
767 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
768 if (IS_ERR(user_ctx)) {
770 return PTR_ERR(user_ctx);
773 ret = verify_user_bpf_flow_keys(user_ctx);
776 flags = user_ctx->flags;
779 ctx.flow_keys = &flow_keys;
781 ctx.data_end = (__u8 *)data + size;
783 bpf_test_timer_enter(&t);
785 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
787 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
788 bpf_test_timer_leave(&t);
793 ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
796 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
797 sizeof(struct bpf_flow_keys));
805 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
806 union bpf_attr __user *uattr)
808 struct bpf_test_timer t = { NO_PREEMPT };
809 struct bpf_prog_array *progs = NULL;
810 struct bpf_sk_lookup_kern ctx = {};
811 u32 repeat = kattr->test.repeat;
812 struct bpf_sk_lookup *user_ctx;
813 u32 retval, duration;
816 if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
819 if (kattr->test.flags || kattr->test.cpu)
822 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
823 kattr->test.data_size_out)
829 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
830 if (IS_ERR(user_ctx))
831 return PTR_ERR(user_ctx);
839 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
842 if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
847 ctx.family = (u16)user_ctx->family;
848 ctx.protocol = (u16)user_ctx->protocol;
849 ctx.dport = (u16)user_ctx->local_port;
850 ctx.sport = (__force __be16)user_ctx->remote_port;
852 switch (ctx.family) {
854 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
855 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
858 #if IS_ENABLED(CONFIG_IPV6)
860 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
861 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
870 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
876 progs->items[0].prog = prog;
878 bpf_test_timer_enter(&t);
880 ctx.selected_sk = NULL;
881 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
882 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
883 bpf_test_timer_leave(&t);
888 user_ctx->cookie = 0;
889 if (ctx.selected_sk) {
890 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
895 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
898 ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
900 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
903 bpf_prog_array_free(progs);