1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
6 #include <linux/btf_ids.h>
7 #include <linux/slab.h>
8 #include <linux/init.h>
9 #include <linux/vmalloc.h>
10 #include <linux/etherdevice.h>
11 #include <linux/filter.h>
12 #include <linux/rcupdate_trace.h>
13 #include <linux/sched/signal.h>
14 #include <net/bpf_sk_storage.h>
15 #include <net/hotdata.h>
18 #include <net/net_namespace.h>
19 #include <net/page_pool/helpers.h>
20 #include <linux/error-injection.h>
21 #include <linux/smp.h>
22 #include <linux/sock_diag.h>
23 #include <linux/netfilter.h>
24 #include <net/netdev_rx_queue.h>
26 #include <net/netfilter/nf_bpf_link.h>
28 #define CREATE_TRACE_POINTS
29 #include <trace/events/bpf_test_run.h>
31 struct bpf_test_timer {
32 enum { NO_PREEMPT, NO_MIGRATE } mode;
34 u64 time_start, time_spent;
37 static void bpf_test_timer_enter(struct bpf_test_timer *t)
41 if (t->mode == NO_PREEMPT)
46 t->time_start = ktime_get_ns();
49 static void bpf_test_timer_leave(struct bpf_test_timer *t)
54 if (t->mode == NO_PREEMPT)
61 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
62 u32 repeat, int *err, u32 *duration)
68 t->time_spent += ktime_get_ns() - t->time_start;
69 do_div(t->time_spent, t->i);
70 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
75 if (signal_pending(current)) {
76 /* During iteration: we've been cancelled, abort. */
82 /* During iteration: we need to reschedule between runs. */
83 t->time_spent += ktime_get_ns() - t->time_start;
84 bpf_test_timer_leave(t);
86 bpf_test_timer_enter(t);
89 /* Do another round. */
97 /* We put this struct at the head of each page with a context and frame
98 * initialised when the page is allocated, so we don't have to do this on each
99 * repetition of the test run.
101 struct xdp_page_head {
102 struct xdp_buff orig_ctx;
105 /* ::data_hard_start starts here */
106 DECLARE_FLEX_ARRAY(struct xdp_frame, frame);
107 DECLARE_FLEX_ARRAY(u8, data);
111 struct xdp_test_data {
112 struct xdp_buff *orig_ctx;
113 struct xdp_rxq_info rxq;
114 struct net_device *dev;
115 struct page_pool *pp;
116 struct xdp_frame **frames;
117 struct sk_buff **skbs;
118 struct xdp_mem_info mem;
123 /* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE
124 * must be updated accordingly this gets changed, otherwise BPF selftests
127 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
128 #define TEST_XDP_MAX_BATCH 256
130 static void xdp_test_run_init_page(struct page *page, void *arg)
132 struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
133 struct xdp_buff *new_ctx, *orig_ctx;
134 u32 headroom = XDP_PACKET_HEADROOM;
135 struct xdp_test_data *xdp = arg;
136 size_t frm_len, meta_len;
137 struct xdp_frame *frm;
140 orig_ctx = xdp->orig_ctx;
141 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
142 meta_len = orig_ctx->data - orig_ctx->data_meta;
143 headroom -= meta_len;
145 new_ctx = &head->ctx;
148 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
150 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
151 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
152 new_ctx->data = new_ctx->data_meta + meta_len;
154 xdp_update_frame_from_buff(new_ctx, frm);
155 frm->mem = new_ctx->rxq->mem;
157 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
160 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
162 struct page_pool *pp;
164 struct page_pool_params pp_params = {
167 .pool_size = xdp->batch_size,
169 .init_callback = xdp_test_run_init_page,
173 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
177 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
181 pp = page_pool_create(&pp_params);
187 /* will copy 'mem.id' into pp->xdp_mem_id */
188 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
194 /* We create a 'fake' RXQ referencing the original dev, but with an
195 * xdp_mem_info pointing to our page_pool
197 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
198 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
199 xdp->rxq.mem.id = pp->xdp_mem_id;
200 xdp->dev = orig_ctx->rxq->dev;
201 xdp->orig_ctx = orig_ctx;
206 page_pool_destroy(pp);
214 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
216 xdp_unreg_mem_model(&xdp->mem);
217 page_pool_destroy(xdp->pp);
222 static bool frame_was_changed(const struct xdp_page_head *head)
224 /* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
225 * i.e. has the highest chances to be overwritten. If those two are
226 * untouched, it's most likely safe to skip the context reset.
228 return head->frame->data != head->orig_ctx.data ||
229 head->frame->flags != head->orig_ctx.flags;
232 static bool ctx_was_changed(struct xdp_page_head *head)
234 return head->orig_ctx.data != head->ctx.data ||
235 head->orig_ctx.data_meta != head->ctx.data_meta ||
236 head->orig_ctx.data_end != head->ctx.data_end;
239 static void reset_ctx(struct xdp_page_head *head)
241 if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
244 head->ctx.data = head->orig_ctx.data;
245 head->ctx.data_meta = head->orig_ctx.data_meta;
246 head->ctx.data_end = head->orig_ctx.data_end;
247 xdp_update_frame_from_buff(&head->ctx, head->frame);
250 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
251 struct sk_buff **skbs,
252 struct net_device *dev)
254 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
258 n = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, nframes,
260 if (unlikely(n == 0)) {
261 for (i = 0; i < nframes; i++)
262 xdp_return_frame(frames[i]);
266 for (i = 0; i < nframes; i++) {
267 struct xdp_frame *xdpf = frames[i];
268 struct sk_buff *skb = skbs[i];
270 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
272 xdp_return_frame(xdpf);
276 list_add_tail(&skb->list, &list);
278 netif_receive_skb_list(&list);
283 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
286 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
287 int err = 0, act, ret, i, nframes = 0, batch_sz;
288 struct xdp_frame **frames = xdp->frames;
289 struct xdp_page_head *head;
290 struct xdp_frame *frm;
291 bool redirect = false;
292 struct xdp_buff *ctx;
295 batch_sz = min_t(u32, repeat, xdp->batch_size);
298 xdp_set_return_frame_no_direct();
300 for (i = 0; i < batch_sz; i++) {
301 page = page_pool_dev_alloc_pages(xdp->pp);
307 head = phys_to_virt(page_to_phys(page));
313 act = bpf_prog_run_xdp(prog, ctx);
315 /* if program changed pkt bounds we need to update the xdp_frame */
316 if (unlikely(ctx_was_changed(head))) {
317 ret = xdp_update_frame_from_buff(ctx, frm);
319 xdp_return_buff(ctx);
326 /* we can't do a real XDP_TX since we're not in the
327 * driver, so turn it into a REDIRECT back to the same
330 ri->tgt_index = xdp->dev->ifindex;
331 ri->map_id = INT_MAX;
332 ri->map_type = BPF_MAP_TYPE_UNSPEC;
336 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
338 xdp_return_buff(ctx);
341 frames[nframes++] = frm;
344 bpf_warn_invalid_xdp_action(NULL, prog, act);
347 xdp_return_buff(ctx);
356 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
361 xdp_clear_return_frame_no_direct();
366 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
367 u32 repeat, u32 batch_size, u32 *time)
370 struct xdp_test_data xdp = { .batch_size = batch_size };
371 struct bpf_test_timer t = { .mode = NO_MIGRATE };
377 ret = xdp_test_run_setup(&xdp, ctx);
381 bpf_test_timer_enter(&t);
384 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
385 if (unlikely(ret < 0))
387 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
388 bpf_test_timer_leave(&t);
390 xdp_test_run_teardown(&xdp);
394 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
395 u32 *retval, u32 *time, bool xdp)
397 struct bpf_prog_array_item item = {.prog = prog};
398 struct bpf_run_ctx *old_ctx;
399 struct bpf_cg_run_ctx run_ctx;
400 struct bpf_test_timer t = { NO_MIGRATE };
401 enum bpf_cgroup_storage_type stype;
404 for_each_cgroup_storage_type(stype) {
405 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
406 if (IS_ERR(item.cgroup_storage[stype])) {
407 item.cgroup_storage[stype] = NULL;
408 for_each_cgroup_storage_type(stype)
409 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
417 bpf_test_timer_enter(&t);
418 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
420 run_ctx.prog_item = &item;
423 *retval = bpf_prog_run_xdp(prog, ctx);
425 *retval = bpf_prog_run(prog, ctx);
427 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
428 bpf_reset_run_ctx(old_ctx);
429 bpf_test_timer_leave(&t);
431 for_each_cgroup_storage_type(stype)
432 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
437 static int bpf_test_finish(const union bpf_attr *kattr,
438 union bpf_attr __user *uattr, const void *data,
439 struct skb_shared_info *sinfo, u32 size,
440 u32 retval, u32 duration)
442 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
444 u32 copy_size = size;
446 /* Clamp copy if the user has provided a size hint, but copy the full
447 * buffer if not to retain old behaviour.
449 if (kattr->test.data_size_out &&
450 copy_size > kattr->test.data_size_out) {
451 copy_size = kattr->test.data_size_out;
456 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
463 if (copy_to_user(data_out, data, len))
470 for (i = 0; i < sinfo->nr_frags; i++) {
471 skb_frag_t *frag = &sinfo->frags[i];
473 if (offset >= copy_size) {
478 data_len = min_t(u32, copy_size - offset,
479 skb_frag_size(frag));
481 if (copy_to_user(data_out + offset,
482 skb_frag_address(frag),
491 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
493 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
495 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
500 trace_bpf_test_finish(&err);
504 /* Integer types of various sizes and pointer combinations cover variety of
505 * architecture dependent calling conventions. 7+ can be supported in the
508 __bpf_kfunc_start_defs();
510 __bpf_kfunc int bpf_fentry_test1(int a)
514 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
516 int noinline bpf_fentry_test2(int a, u64 b)
521 int noinline bpf_fentry_test3(char a, int b, u64 c)
526 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
528 return (long)a + b + c + d;
531 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
533 return a + (long)b + c + d + e;
536 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
538 return a + (long)b + c + d + (long)e + f;
541 struct bpf_fentry_test_t {
542 struct bpf_fentry_test_t *a;
545 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
547 asm volatile ("": "+r"(arg));
551 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
556 __bpf_kfunc u32 bpf_fentry_test9(u32 *a)
561 void noinline bpf_fentry_test_sinfo(struct skb_shared_info *sinfo)
565 __bpf_kfunc int bpf_modify_return_test(int a, int *b)
571 __bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
572 void *e, char f, int g)
575 return a + *b + c + d + (long)e + f + g;
578 __bpf_kfunc int bpf_modify_return_test_tp(int nonce)
580 trace_bpf_trigger_tp(nonce);
585 int noinline bpf_fentry_shadow_test(int a)
590 struct prog_test_member1 {
594 struct prog_test_member {
595 struct prog_test_member1 m;
599 struct prog_test_ref_kfunc {
602 struct prog_test_member memb;
603 struct prog_test_ref_kfunc *next;
607 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
609 refcount_dec(&p->cnt);
612 __bpf_kfunc void bpf_kfunc_call_test_release_dtor(void *p)
614 bpf_kfunc_call_test_release(p);
616 CFI_NOSEAL(bpf_kfunc_call_test_release_dtor);
618 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
622 __bpf_kfunc void bpf_kfunc_call_memb_release_dtor(void *p)
625 CFI_NOSEAL(bpf_kfunc_call_memb_release_dtor);
627 __bpf_kfunc_end_defs();
629 BTF_KFUNCS_START(bpf_test_modify_return_ids)
630 BTF_ID_FLAGS(func, bpf_modify_return_test)
631 BTF_ID_FLAGS(func, bpf_modify_return_test2)
632 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
633 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
634 BTF_KFUNCS_END(bpf_test_modify_return_ids)
636 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
637 .owner = THIS_MODULE,
638 .set = &bpf_test_modify_return_ids,
641 BTF_KFUNCS_START(test_sk_check_kfunc_ids)
642 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
643 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
644 BTF_KFUNCS_END(test_sk_check_kfunc_ids)
646 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
647 u32 size, u32 headroom, u32 tailroom)
649 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
652 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
653 return ERR_PTR(-EINVAL);
655 if (user_size > size)
656 return ERR_PTR(-EMSGSIZE);
658 size = SKB_DATA_ALIGN(size);
659 data = kzalloc(size + headroom + tailroom, GFP_USER);
661 return ERR_PTR(-ENOMEM);
663 if (copy_from_user(data + headroom, data_in, user_size)) {
665 return ERR_PTR(-EFAULT);
671 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
672 const union bpf_attr *kattr,
673 union bpf_attr __user *uattr)
675 struct bpf_fentry_test_t arg = {};
676 u16 side_effect = 0, ret = 0;
677 int b = 2, err = -EFAULT;
680 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
683 switch (prog->expected_attach_type) {
684 case BPF_TRACE_FENTRY:
685 case BPF_TRACE_FEXIT:
686 if (bpf_fentry_test1(1) != 2 ||
687 bpf_fentry_test2(2, 3) != 5 ||
688 bpf_fentry_test3(4, 5, 6) != 15 ||
689 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
690 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
691 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
692 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
693 bpf_fentry_test8(&arg) != 0 ||
694 bpf_fentry_test9(&retval) != 0)
697 case BPF_MODIFY_RETURN:
698 ret = bpf_modify_return_test(1, &b);
702 ret += bpf_modify_return_test2(1, &b, 3, 4, (void *)5, 6, 7);
710 retval = ((u32)side_effect << 16) | ret;
711 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
716 trace_bpf_test_finish(&err);
720 struct bpf_raw_tp_test_run_info {
721 struct bpf_prog *prog;
727 __bpf_prog_test_run_raw_tp(void *data)
729 struct bpf_raw_tp_test_run_info *info = data;
732 info->retval = bpf_prog_run(info->prog, info->ctx);
736 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
737 const union bpf_attr *kattr,
738 union bpf_attr __user *uattr)
740 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
741 __u32 ctx_size_in = kattr->test.ctx_size_in;
742 struct bpf_raw_tp_test_run_info info;
743 int cpu = kattr->test.cpu, err = 0;
746 /* doesn't support data_in/out, ctx_out, duration, or repeat */
747 if (kattr->test.data_in || kattr->test.data_out ||
748 kattr->test.ctx_out || kattr->test.duration ||
749 kattr->test.repeat || kattr->test.batch_size)
752 if (ctx_size_in < prog->aux->max_ctx_offset ||
753 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
756 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
760 info.ctx = memdup_user(ctx_in, ctx_size_in);
761 if (IS_ERR(info.ctx))
762 return PTR_ERR(info.ctx);
769 current_cpu = get_cpu();
770 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
771 cpu == current_cpu) {
772 __bpf_prog_test_run_raw_tp(&info);
773 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
774 /* smp_call_function_single() also checks cpu_online()
775 * after csd_lock(). However, since cpu is from user
776 * space, let's do an extra quick check to filter out
777 * invalid value before smp_call_function_single().
781 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
787 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
794 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
796 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
797 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
798 u32 size = kattr->test.ctx_size_in;
802 if (!data_in && !data_out)
805 data = kzalloc(max_size, GFP_USER);
807 return ERR_PTR(-ENOMEM);
810 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
816 size = min_t(u32, max_size, size);
817 if (copy_from_user(data, data_in, size)) {
819 return ERR_PTR(-EFAULT);
825 static int bpf_ctx_finish(const union bpf_attr *kattr,
826 union bpf_attr __user *uattr, const void *data,
829 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
831 u32 copy_size = size;
833 if (!data || !data_out)
836 if (copy_size > kattr->test.ctx_size_out) {
837 copy_size = kattr->test.ctx_size_out;
841 if (copy_to_user(data_out, data, copy_size))
843 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
852 * range_is_zero - test whether buffer is initialized
853 * @buf: buffer to check
854 * @from: check from this position
855 * @to: check up until (excluding) this position
857 * This function returns true if the there is a non-zero byte
858 * in the buf in the range [from,to).
860 static inline bool range_is_zero(void *buf, size_t from, size_t to)
862 return !memchr_inv((u8 *)buf + from, 0, to - from);
865 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
867 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
872 /* make sure the fields we don't use are zeroed */
873 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
876 /* mark is allowed */
878 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
879 offsetof(struct __sk_buff, priority)))
882 /* priority is allowed */
883 /* ingress_ifindex is allowed */
884 /* ifindex is allowed */
886 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
887 offsetof(struct __sk_buff, cb)))
892 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
893 offsetof(struct __sk_buff, tstamp)))
896 /* tstamp is allowed */
897 /* wire_len is allowed */
898 /* gso_segs is allowed */
900 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
901 offsetof(struct __sk_buff, gso_size)))
904 /* gso_size is allowed */
906 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
907 offsetof(struct __sk_buff, hwtstamp)))
910 /* hwtstamp is allowed */
912 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
913 sizeof(struct __sk_buff)))
916 skb->mark = __skb->mark;
917 skb->priority = __skb->priority;
918 skb->skb_iif = __skb->ingress_ifindex;
919 skb->tstamp = __skb->tstamp;
920 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
922 if (__skb->wire_len == 0) {
923 cb->pkt_len = skb->len;
925 if (__skb->wire_len < skb->len ||
926 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
928 cb->pkt_len = __skb->wire_len;
931 if (__skb->gso_segs > GSO_MAX_SEGS)
933 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
934 skb_shinfo(skb)->gso_size = __skb->gso_size;
935 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
940 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
942 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
947 __skb->mark = skb->mark;
948 __skb->priority = skb->priority;
949 __skb->ingress_ifindex = skb->skb_iif;
950 __skb->ifindex = skb->dev->ifindex;
951 __skb->tstamp = skb->tstamp;
952 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
953 __skb->wire_len = cb->pkt_len;
954 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
955 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
958 static struct proto bpf_dummy_proto = {
960 .owner = THIS_MODULE,
961 .obj_size = sizeof(struct sock),
964 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
965 union bpf_attr __user *uattr)
967 bool is_l2 = false, is_direct_pkt_access = false;
968 struct net *net = current->nsproxy->net_ns;
969 struct net_device *dev = net->loopback_dev;
970 u32 size = kattr->test.data_size_in;
971 u32 repeat = kattr->test.repeat;
972 struct __sk_buff *ctx = NULL;
973 u32 retval, duration;
974 int hh_len = ETH_HLEN;
980 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
983 data = bpf_test_init(kattr, kattr->test.data_size_in,
984 size, NET_SKB_PAD + NET_IP_ALIGN,
985 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
987 return PTR_ERR(data);
989 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
995 switch (prog->type) {
996 case BPF_PROG_TYPE_SCHED_CLS:
997 case BPF_PROG_TYPE_SCHED_ACT:
1000 case BPF_PROG_TYPE_LWT_IN:
1001 case BPF_PROG_TYPE_LWT_OUT:
1002 case BPF_PROG_TYPE_LWT_XMIT:
1003 is_direct_pkt_access = true;
1009 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1015 sock_init_data(NULL, sk);
1017 skb = slab_build_skb(data);
1026 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1027 __skb_put(skb, size);
1028 if (ctx && ctx->ifindex > 1) {
1029 dev = dev_get_by_index(net, ctx->ifindex);
1035 skb->protocol = eth_type_trans(skb, dev);
1036 skb_reset_network_header(skb);
1038 switch (skb->protocol) {
1039 case htons(ETH_P_IP):
1040 sk->sk_family = AF_INET;
1041 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1042 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1043 sk->sk_daddr = ip_hdr(skb)->daddr;
1046 #if IS_ENABLED(CONFIG_IPV6)
1047 case htons(ETH_P_IPV6):
1048 sk->sk_family = AF_INET6;
1049 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1050 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1051 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1060 __skb_push(skb, hh_len);
1061 if (is_direct_pkt_access)
1062 bpf_compute_data_pointers(skb);
1063 ret = convert___skb_to_skb(skb, ctx);
1066 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1070 if (skb_headroom(skb) < hh_len) {
1071 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1073 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1078 memset(__skb_push(skb, hh_len), 0, hh_len);
1080 convert_skb_to___skb(skb, ctx);
1083 /* bpf program can never convert linear skb to non-linear */
1084 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1085 size = skb_headlen(skb);
1086 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1089 ret = bpf_ctx_finish(kattr, uattr, ctx,
1090 sizeof(struct __sk_buff));
1092 if (dev && dev != net->loopback_dev)
1100 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1102 unsigned int ingress_ifindex, rx_queue_index;
1103 struct netdev_rx_queue *rxqueue;
1104 struct net_device *device;
1109 if (xdp_md->egress_ifindex != 0)
1112 ingress_ifindex = xdp_md->ingress_ifindex;
1113 rx_queue_index = xdp_md->rx_queue_index;
1115 if (!ingress_ifindex && rx_queue_index)
1118 if (ingress_ifindex) {
1119 device = dev_get_by_index(current->nsproxy->net_ns,
1124 if (rx_queue_index >= device->real_num_rx_queues)
1127 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1129 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1132 xdp->rxq = &rxqueue->xdp_rxq;
1133 /* The device is now tracked in the xdp->rxq for later
1138 xdp->data = xdp->data_meta + xdp_md->data;
1146 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1151 xdp_md->data = xdp->data - xdp->data_meta;
1152 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1154 if (xdp_md->ingress_ifindex)
1155 dev_put(xdp->rxq->dev);
1158 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1159 union bpf_attr __user *uattr)
1161 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1162 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1163 u32 batch_size = kattr->test.batch_size;
1164 u32 retval = 0, duration, max_data_sz;
1165 u32 size = kattr->test.data_size_in;
1166 u32 headroom = XDP_PACKET_HEADROOM;
1167 u32 repeat = kattr->test.repeat;
1168 struct netdev_rx_queue *rxqueue;
1169 struct skb_shared_info *sinfo;
1170 struct xdp_buff xdp = {};
1171 int i, ret = -EINVAL;
1175 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1176 prog->expected_attach_type == BPF_XDP_CPUMAP)
1179 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1182 if (bpf_prog_is_dev_bound(prog->aux))
1187 batch_size = NAPI_POLL_WEIGHT;
1188 else if (batch_size > TEST_XDP_MAX_BATCH)
1191 headroom += sizeof(struct xdp_page_head);
1192 } else if (batch_size) {
1196 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1198 return PTR_ERR(ctx);
1201 /* There can't be user provided data before the meta data */
1202 if (ctx->data_meta || ctx->data_end != size ||
1203 ctx->data > ctx->data_end ||
1204 unlikely(xdp_metalen_invalid(ctx->data)) ||
1205 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1207 /* Meta data is allocated from the headroom */
1208 headroom -= ctx->data;
1211 max_data_sz = 4096 - headroom - tailroom;
1212 if (size > max_data_sz) {
1213 /* disallow live data mode for jumbo frames */
1219 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1221 ret = PTR_ERR(data);
1225 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1226 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1227 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1228 xdp_prepare_buff(&xdp, data, headroom, size, true);
1229 sinfo = xdp_get_shared_info_from_buff(&xdp);
1231 ret = xdp_convert_md_to_buff(ctx, &xdp);
1235 if (unlikely(kattr->test.data_size_in > size)) {
1236 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1238 while (size < kattr->test.data_size_in) {
1243 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1248 page = alloc_page(GFP_KERNEL);
1254 frag = &sinfo->frags[sinfo->nr_frags++];
1256 data_len = min_t(u32, kattr->test.data_size_in - size,
1258 skb_frag_fill_page_desc(frag, page, 0, data_len);
1260 if (copy_from_user(page_address(page), data_in + size,
1265 sinfo->xdp_frags_size += data_len;
1268 xdp_buff_set_frags_flag(&xdp);
1272 bpf_prog_change_xdp(NULL, prog);
1275 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1277 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1278 /* We convert the xdp_buff back to an xdp_md before checking the return
1279 * code so the reference count of any held netdevice will be decremented
1280 * even if the test run failed.
1282 xdp_convert_buff_to_md(&xdp, ctx);
1286 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1287 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1290 ret = bpf_ctx_finish(kattr, uattr, ctx,
1291 sizeof(struct xdp_md));
1295 bpf_prog_change_xdp(prog, NULL);
1297 for (i = 0; i < sinfo->nr_frags; i++)
1298 __free_page(skb_frag_page(&sinfo->frags[i]));
1305 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1307 /* make sure the fields we don't use are zeroed */
1308 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1311 /* flags is allowed */
1313 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1314 sizeof(struct bpf_flow_keys)))
1320 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1321 const union bpf_attr *kattr,
1322 union bpf_attr __user *uattr)
1324 struct bpf_test_timer t = { NO_PREEMPT };
1325 u32 size = kattr->test.data_size_in;
1326 struct bpf_flow_dissector ctx = {};
1327 u32 repeat = kattr->test.repeat;
1328 struct bpf_flow_keys *user_ctx;
1329 struct bpf_flow_keys flow_keys;
1330 const struct ethhdr *eth;
1331 unsigned int flags = 0;
1332 u32 retval, duration;
1336 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1339 if (size < ETH_HLEN)
1342 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1344 return PTR_ERR(data);
1346 eth = (struct ethhdr *)data;
1351 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1352 if (IS_ERR(user_ctx)) {
1354 return PTR_ERR(user_ctx);
1357 ret = verify_user_bpf_flow_keys(user_ctx);
1360 flags = user_ctx->flags;
1363 ctx.flow_keys = &flow_keys;
1365 ctx.data_end = (__u8 *)data + size;
1367 bpf_test_timer_enter(&t);
1369 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1371 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1372 bpf_test_timer_leave(&t);
1377 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1378 sizeof(flow_keys), retval, duration);
1380 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1381 sizeof(struct bpf_flow_keys));
1389 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1390 union bpf_attr __user *uattr)
1392 struct bpf_test_timer t = { NO_PREEMPT };
1393 struct bpf_prog_array *progs = NULL;
1394 struct bpf_sk_lookup_kern ctx = {};
1395 u32 repeat = kattr->test.repeat;
1396 struct bpf_sk_lookup *user_ctx;
1397 u32 retval, duration;
1400 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1403 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1404 kattr->test.data_size_out)
1410 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1411 if (IS_ERR(user_ctx))
1412 return PTR_ERR(user_ctx);
1420 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1423 if (user_ctx->local_port > U16_MAX) {
1428 ctx.family = (u16)user_ctx->family;
1429 ctx.protocol = (u16)user_ctx->protocol;
1430 ctx.dport = (u16)user_ctx->local_port;
1431 ctx.sport = user_ctx->remote_port;
1433 switch (ctx.family) {
1435 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1436 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1439 #if IS_ENABLED(CONFIG_IPV6)
1441 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1442 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1447 ret = -EAFNOSUPPORT;
1451 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1457 progs->items[0].prog = prog;
1459 bpf_test_timer_enter(&t);
1461 ctx.selected_sk = NULL;
1462 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1463 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1464 bpf_test_timer_leave(&t);
1469 user_ctx->cookie = 0;
1470 if (ctx.selected_sk) {
1471 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1476 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1479 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1481 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1484 bpf_prog_array_free(progs);
1489 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1490 const union bpf_attr *kattr,
1491 union bpf_attr __user *uattr)
1493 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1494 __u32 ctx_size_in = kattr->test.ctx_size_in;
1499 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1500 if (kattr->test.data_in || kattr->test.data_out ||
1501 kattr->test.ctx_out || kattr->test.duration ||
1502 kattr->test.repeat || kattr->test.flags ||
1503 kattr->test.batch_size)
1506 if (ctx_size_in < prog->aux->max_ctx_offset ||
1507 ctx_size_in > U16_MAX)
1511 ctx = memdup_user(ctx_in, ctx_size_in);
1513 return PTR_ERR(ctx);
1516 rcu_read_lock_trace();
1517 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1518 rcu_read_unlock_trace();
1520 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1525 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1532 static int verify_and_copy_hook_state(struct nf_hook_state *state,
1533 const struct nf_hook_state *user,
1534 struct net_device *dev)
1536 if (user->in || user->out)
1539 if (user->net || user->sk || user->okfn)
1545 switch (state->hook) {
1546 case NF_INET_PRE_ROUTING:
1549 case NF_INET_LOCAL_IN:
1552 case NF_INET_FORWARD:
1556 case NF_INET_LOCAL_OUT:
1559 case NF_INET_POST_ROUTING:
1569 state->pf = user->pf;
1570 state->hook = user->hook;
1575 static __be16 nfproto_eth(int nfproto)
1579 return htons(ETH_P_IP);
1584 return htons(ETH_P_IPV6);
1587 int bpf_prog_test_run_nf(struct bpf_prog *prog,
1588 const union bpf_attr *kattr,
1589 union bpf_attr __user *uattr)
1591 struct net *net = current->nsproxy->net_ns;
1592 struct net_device *dev = net->loopback_dev;
1593 struct nf_hook_state *user_ctx, hook_state = {
1595 .hook = NF_INET_LOCAL_OUT,
1597 u32 size = kattr->test.data_size_in;
1598 u32 repeat = kattr->test.repeat;
1599 struct bpf_nf_ctx ctx = {
1600 .state = &hook_state,
1602 struct sk_buff *skb = NULL;
1603 u32 retval, duration;
1607 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1610 if (size < sizeof(struct iphdr))
1613 data = bpf_test_init(kattr, kattr->test.data_size_in, size,
1614 NET_SKB_PAD + NET_IP_ALIGN,
1615 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1617 return PTR_ERR(data);
1622 user_ctx = bpf_ctx_init(kattr, sizeof(struct nf_hook_state));
1623 if (IS_ERR(user_ctx)) {
1625 return PTR_ERR(user_ctx);
1629 ret = verify_and_copy_hook_state(&hook_state, user_ctx, dev);
1634 skb = slab_build_skb(data);
1640 data = NULL; /* data released via kfree_skb */
1642 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1643 __skb_put(skb, size);
1647 if (hook_state.hook != NF_INET_LOCAL_OUT) {
1648 if (size < ETH_HLEN + sizeof(struct iphdr))
1651 skb->protocol = eth_type_trans(skb, dev);
1652 switch (skb->protocol) {
1653 case htons(ETH_P_IP):
1654 if (hook_state.pf == NFPROTO_IPV4)
1657 case htons(ETH_P_IPV6):
1658 if (size < ETH_HLEN + sizeof(struct ipv6hdr))
1660 if (hook_state.pf == NFPROTO_IPV6)
1668 skb_reset_network_header(skb);
1670 skb->protocol = nfproto_eth(hook_state.pf);
1675 ret = bpf_test_run(prog, &ctx, repeat, &retval, &duration, false);
1679 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1688 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1689 .owner = THIS_MODULE,
1690 .set = &test_sk_check_kfunc_ids,
1693 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1694 BTF_ID(struct, prog_test_ref_kfunc)
1695 BTF_ID(func, bpf_kfunc_call_test_release_dtor)
1696 BTF_ID(struct, prog_test_member)
1697 BTF_ID(func, bpf_kfunc_call_memb_release_dtor)
1699 static int __init bpf_prog_test_run_init(void)
1701 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1703 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1704 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1707 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1708 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1713 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1714 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1715 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1716 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1717 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1718 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1721 late_initcall(bpf_prog_test_run_init);