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 size = SKB_DATA_ALIGN(size);
235 data = kzalloc(size + headroom + tailroom, GFP_USER);
237 return ERR_PTR(-ENOMEM);
239 if (copy_from_user(data + headroom, data_in, user_size)) {
241 return ERR_PTR(-EFAULT);
247 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
248 const union bpf_attr *kattr,
249 union bpf_attr __user *uattr)
251 struct bpf_fentry_test_t arg = {};
252 u16 side_effect = 0, ret = 0;
253 int b = 2, err = -EFAULT;
256 if (kattr->test.flags || kattr->test.cpu)
259 switch (prog->expected_attach_type) {
260 case BPF_TRACE_FENTRY:
261 case BPF_TRACE_FEXIT:
262 if (bpf_fentry_test1(1) != 2 ||
263 bpf_fentry_test2(2, 3) != 5 ||
264 bpf_fentry_test3(4, 5, 6) != 15 ||
265 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
266 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
267 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
268 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
269 bpf_fentry_test8(&arg) != 0)
272 case BPF_MODIFY_RETURN:
273 ret = bpf_modify_return_test(1, &b);
281 retval = ((u32)side_effect << 16) | ret;
282 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
287 trace_bpf_test_finish(&err);
291 struct bpf_raw_tp_test_run_info {
292 struct bpf_prog *prog;
298 __bpf_prog_test_run_raw_tp(void *data)
300 struct bpf_raw_tp_test_run_info *info = data;
303 info->retval = BPF_PROG_RUN(info->prog, info->ctx);
307 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
308 const union bpf_attr *kattr,
309 union bpf_attr __user *uattr)
311 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
312 __u32 ctx_size_in = kattr->test.ctx_size_in;
313 struct bpf_raw_tp_test_run_info info;
314 int cpu = kattr->test.cpu, err = 0;
317 /* doesn't support data_in/out, ctx_out, duration, or repeat */
318 if (kattr->test.data_in || kattr->test.data_out ||
319 kattr->test.ctx_out || kattr->test.duration ||
323 if (ctx_size_in < prog->aux->max_ctx_offset ||
324 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
327 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
331 info.ctx = kzalloc(ctx_size_in, GFP_USER);
334 if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
344 current_cpu = get_cpu();
345 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
346 cpu == current_cpu) {
347 __bpf_prog_test_run_raw_tp(&info);
348 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
349 /* smp_call_function_single() also checks cpu_online()
350 * after csd_lock(). However, since cpu is from user
351 * space, let's do an extra quick check to filter out
352 * invalid value before smp_call_function_single().
356 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
362 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
370 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
372 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
373 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
374 u32 size = kattr->test.ctx_size_in;
378 if (!data_in && !data_out)
381 data = kzalloc(max_size, GFP_USER);
383 return ERR_PTR(-ENOMEM);
386 err = bpf_check_uarg_tail_zero(data_in, max_size, size);
392 size = min_t(u32, max_size, size);
393 if (copy_from_user(data, data_in, size)) {
395 return ERR_PTR(-EFAULT);
401 static int bpf_ctx_finish(const union bpf_attr *kattr,
402 union bpf_attr __user *uattr, const void *data,
405 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
407 u32 copy_size = size;
409 if (!data || !data_out)
412 if (copy_size > kattr->test.ctx_size_out) {
413 copy_size = kattr->test.ctx_size_out;
417 if (copy_to_user(data_out, data, copy_size))
419 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
428 * range_is_zero - test whether buffer is initialized
429 * @buf: buffer to check
430 * @from: check from this position
431 * @to: check up until (excluding) this position
433 * This function returns true if the there is a non-zero byte
434 * in the buf in the range [from,to).
436 static inline bool range_is_zero(void *buf, size_t from, size_t to)
438 return !memchr_inv((u8 *)buf + from, 0, to - from);
441 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
443 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
448 /* make sure the fields we don't use are zeroed */
449 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
452 /* mark is allowed */
454 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
455 offsetof(struct __sk_buff, priority)))
458 /* priority is allowed */
460 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
461 offsetof(struct __sk_buff, ifindex)))
464 /* ifindex is allowed */
466 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
467 offsetof(struct __sk_buff, cb)))
472 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
473 offsetof(struct __sk_buff, tstamp)))
476 /* tstamp is allowed */
477 /* wire_len is allowed */
478 /* gso_segs is allowed */
480 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
481 offsetof(struct __sk_buff, gso_size)))
484 /* gso_size is allowed */
486 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
487 sizeof(struct __sk_buff)))
490 skb->mark = __skb->mark;
491 skb->priority = __skb->priority;
492 skb->tstamp = __skb->tstamp;
493 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
495 if (__skb->wire_len == 0) {
496 cb->pkt_len = skb->len;
498 if (__skb->wire_len < skb->len ||
499 __skb->wire_len > GSO_MAX_SIZE)
501 cb->pkt_len = __skb->wire_len;
504 if (__skb->gso_segs > GSO_MAX_SEGS)
506 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
507 skb_shinfo(skb)->gso_size = __skb->gso_size;
512 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
514 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
519 __skb->mark = skb->mark;
520 __skb->priority = skb->priority;
521 __skb->ifindex = skb->dev->ifindex;
522 __skb->tstamp = skb->tstamp;
523 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
524 __skb->wire_len = cb->pkt_len;
525 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
528 static struct proto bpf_dummy_proto = {
530 .owner = THIS_MODULE,
531 .obj_size = sizeof(struct sock),
534 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
535 union bpf_attr __user *uattr)
537 bool is_l2 = false, is_direct_pkt_access = false;
538 struct net *net = current->nsproxy->net_ns;
539 struct net_device *dev = net->loopback_dev;
540 u32 size = kattr->test.data_size_in;
541 u32 repeat = kattr->test.repeat;
542 struct __sk_buff *ctx = NULL;
543 u32 retval, duration;
544 int hh_len = ETH_HLEN;
550 if (kattr->test.flags || kattr->test.cpu)
553 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
554 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
556 return PTR_ERR(data);
558 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
564 switch (prog->type) {
565 case BPF_PROG_TYPE_SCHED_CLS:
566 case BPF_PROG_TYPE_SCHED_ACT:
569 case BPF_PROG_TYPE_LWT_IN:
570 case BPF_PROG_TYPE_LWT_OUT:
571 case BPF_PROG_TYPE_LWT_XMIT:
572 is_direct_pkt_access = true;
578 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
584 sock_init_data(NULL, sk);
586 skb = build_skb(data, 0);
595 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
596 __skb_put(skb, size);
597 if (ctx && ctx->ifindex > 1) {
598 dev = dev_get_by_index(net, ctx->ifindex);
604 skb->protocol = eth_type_trans(skb, dev);
605 skb_reset_network_header(skb);
607 switch (skb->protocol) {
608 case htons(ETH_P_IP):
609 sk->sk_family = AF_INET;
610 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
611 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
612 sk->sk_daddr = ip_hdr(skb)->daddr;
615 #if IS_ENABLED(CONFIG_IPV6)
616 case htons(ETH_P_IPV6):
617 sk->sk_family = AF_INET6;
618 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
619 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
620 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
629 __skb_push(skb, hh_len);
630 if (is_direct_pkt_access)
631 bpf_compute_data_pointers(skb);
632 ret = convert___skb_to_skb(skb, ctx);
635 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
639 if (skb_headroom(skb) < hh_len) {
640 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
642 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
647 memset(__skb_push(skb, hh_len), 0, hh_len);
649 convert_skb_to___skb(skb, ctx);
652 /* bpf program can never convert linear skb to non-linear */
653 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
654 size = skb_headlen(skb);
655 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
657 ret = bpf_ctx_finish(kattr, uattr, ctx,
658 sizeof(struct __sk_buff));
660 if (dev && dev != net->loopback_dev)
668 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
669 union bpf_attr __user *uattr)
671 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
672 u32 headroom = XDP_PACKET_HEADROOM;
673 u32 size = kattr->test.data_size_in;
674 u32 repeat = kattr->test.repeat;
675 struct netdev_rx_queue *rxqueue;
676 struct xdp_buff xdp = {};
677 u32 retval, duration;
682 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
683 prog->expected_attach_type == BPF_XDP_CPUMAP)
685 if (kattr->test.ctx_in || kattr->test.ctx_out)
688 /* XDP have extra tailroom as (most) drivers use full page */
689 max_data_sz = 4096 - headroom - tailroom;
691 data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
693 return PTR_ERR(data);
695 xdp.data_hard_start = data;
696 xdp.data = data + headroom;
697 xdp.data_meta = xdp.data;
698 xdp.data_end = xdp.data + size;
699 xdp.frame_sz = headroom + max_data_sz + tailroom;
701 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
702 xdp.rxq = &rxqueue->xdp_rxq;
703 bpf_prog_change_xdp(NULL, prog);
704 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
707 if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
708 size = xdp.data_end - xdp.data;
709 ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
711 bpf_prog_change_xdp(prog, NULL);
716 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
718 /* make sure the fields we don't use are zeroed */
719 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
722 /* flags is allowed */
724 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
725 sizeof(struct bpf_flow_keys)))
731 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
732 const union bpf_attr *kattr,
733 union bpf_attr __user *uattr)
735 struct bpf_test_timer t = { NO_PREEMPT };
736 u32 size = kattr->test.data_size_in;
737 struct bpf_flow_dissector ctx = {};
738 u32 repeat = kattr->test.repeat;
739 struct bpf_flow_keys *user_ctx;
740 struct bpf_flow_keys flow_keys;
741 const struct ethhdr *eth;
742 unsigned int flags = 0;
743 u32 retval, duration;
747 if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
750 if (kattr->test.flags || kattr->test.cpu)
756 data = bpf_test_init(kattr, size, 0, 0);
758 return PTR_ERR(data);
760 eth = (struct ethhdr *)data;
765 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
766 if (IS_ERR(user_ctx)) {
768 return PTR_ERR(user_ctx);
771 ret = verify_user_bpf_flow_keys(user_ctx);
774 flags = user_ctx->flags;
777 ctx.flow_keys = &flow_keys;
779 ctx.data_end = (__u8 *)data + size;
781 bpf_test_timer_enter(&t);
783 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
785 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
786 bpf_test_timer_leave(&t);
791 ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
794 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
795 sizeof(struct bpf_flow_keys));
803 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
804 union bpf_attr __user *uattr)
806 struct bpf_test_timer t = { NO_PREEMPT };
807 struct bpf_prog_array *progs = NULL;
808 struct bpf_sk_lookup_kern ctx = {};
809 u32 repeat = kattr->test.repeat;
810 struct bpf_sk_lookup *user_ctx;
811 u32 retval, duration;
814 if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
817 if (kattr->test.flags || kattr->test.cpu)
820 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
821 kattr->test.data_size_out)
827 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
828 if (IS_ERR(user_ctx))
829 return PTR_ERR(user_ctx);
837 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
840 if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
845 ctx.family = (u16)user_ctx->family;
846 ctx.protocol = (u16)user_ctx->protocol;
847 ctx.dport = (u16)user_ctx->local_port;
848 ctx.sport = (__force __be16)user_ctx->remote_port;
850 switch (ctx.family) {
852 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
853 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
856 #if IS_ENABLED(CONFIG_IPV6)
858 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
859 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
868 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
874 progs->items[0].prog = prog;
876 bpf_test_timer_enter(&t);
878 ctx.selected_sk = NULL;
879 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
880 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
881 bpf_test_timer_leave(&t);
886 user_ctx->cookie = 0;
887 if (ctx.selected_sk) {
888 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
893 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
896 ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
898 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
901 bpf_prog_array_free(progs);