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>
17 #include <net/net_namespace.h>
18 #include <net/page_pool.h>
19 #include <linux/error-injection.h>
20 #include <linux/smp.h>
21 #include <linux/sock_diag.h>
22 #include <linux/netfilter.h>
24 #include <net/netfilter/nf_bpf_link.h>
26 #define CREATE_TRACE_POINTS
27 #include <trace/events/bpf_test_run.h>
29 struct bpf_test_timer {
30 enum { NO_PREEMPT, NO_MIGRATE } mode;
32 u64 time_start, time_spent;
35 static void bpf_test_timer_enter(struct bpf_test_timer *t)
39 if (t->mode == NO_PREEMPT)
44 t->time_start = ktime_get_ns();
47 static void bpf_test_timer_leave(struct bpf_test_timer *t)
52 if (t->mode == NO_PREEMPT)
59 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
60 u32 repeat, int *err, u32 *duration)
66 t->time_spent += ktime_get_ns() - t->time_start;
67 do_div(t->time_spent, t->i);
68 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
73 if (signal_pending(current)) {
74 /* During iteration: we've been cancelled, abort. */
80 /* During iteration: we need to reschedule between runs. */
81 t->time_spent += ktime_get_ns() - t->time_start;
82 bpf_test_timer_leave(t);
84 bpf_test_timer_enter(t);
87 /* Do another round. */
95 /* We put this struct at the head of each page with a context and frame
96 * initialised when the page is allocated, so we don't have to do this on each
97 * repetition of the test run.
99 struct xdp_page_head {
100 struct xdp_buff orig_ctx;
103 /* ::data_hard_start starts here */
104 DECLARE_FLEX_ARRAY(struct xdp_frame, frame);
105 DECLARE_FLEX_ARRAY(u8, data);
109 struct xdp_test_data {
110 struct xdp_buff *orig_ctx;
111 struct xdp_rxq_info rxq;
112 struct net_device *dev;
113 struct page_pool *pp;
114 struct xdp_frame **frames;
115 struct sk_buff **skbs;
116 struct xdp_mem_info mem;
121 /* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE
122 * must be updated accordingly this gets changed, otherwise BPF selftests
125 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
126 #define TEST_XDP_MAX_BATCH 256
128 static void xdp_test_run_init_page(struct page *page, void *arg)
130 struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
131 struct xdp_buff *new_ctx, *orig_ctx;
132 u32 headroom = XDP_PACKET_HEADROOM;
133 struct xdp_test_data *xdp = arg;
134 size_t frm_len, meta_len;
135 struct xdp_frame *frm;
138 orig_ctx = xdp->orig_ctx;
139 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
140 meta_len = orig_ctx->data - orig_ctx->data_meta;
141 headroom -= meta_len;
143 new_ctx = &head->ctx;
146 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
148 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
149 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
150 new_ctx->data = new_ctx->data_meta + meta_len;
152 xdp_update_frame_from_buff(new_ctx, frm);
153 frm->mem = new_ctx->rxq->mem;
155 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
158 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
160 struct page_pool *pp;
162 struct page_pool_params pp_params = {
165 .pool_size = xdp->batch_size,
167 .init_callback = xdp_test_run_init_page,
171 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
175 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
179 pp = page_pool_create(&pp_params);
185 /* will copy 'mem.id' into pp->xdp_mem_id */
186 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
192 /* We create a 'fake' RXQ referencing the original dev, but with an
193 * xdp_mem_info pointing to our page_pool
195 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
196 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
197 xdp->rxq.mem.id = pp->xdp_mem_id;
198 xdp->dev = orig_ctx->rxq->dev;
199 xdp->orig_ctx = orig_ctx;
204 page_pool_destroy(pp);
212 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
214 xdp_unreg_mem_model(&xdp->mem);
215 page_pool_destroy(xdp->pp);
220 static bool frame_was_changed(const struct xdp_page_head *head)
222 /* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
223 * i.e. has the highest chances to be overwritten. If those two are
224 * untouched, it's most likely safe to skip the context reset.
226 return head->frame->data != head->orig_ctx.data ||
227 head->frame->flags != head->orig_ctx.flags;
230 static bool ctx_was_changed(struct xdp_page_head *head)
232 return head->orig_ctx.data != head->ctx.data ||
233 head->orig_ctx.data_meta != head->ctx.data_meta ||
234 head->orig_ctx.data_end != head->ctx.data_end;
237 static void reset_ctx(struct xdp_page_head *head)
239 if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
242 head->ctx.data = head->orig_ctx.data;
243 head->ctx.data_meta = head->orig_ctx.data_meta;
244 head->ctx.data_end = head->orig_ctx.data_end;
245 xdp_update_frame_from_buff(&head->ctx, head->frame);
248 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
249 struct sk_buff **skbs,
250 struct net_device *dev)
252 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
256 n = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, (void **)skbs);
257 if (unlikely(n == 0)) {
258 for (i = 0; i < nframes; i++)
259 xdp_return_frame(frames[i]);
263 for (i = 0; i < nframes; i++) {
264 struct xdp_frame *xdpf = frames[i];
265 struct sk_buff *skb = skbs[i];
267 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
269 xdp_return_frame(xdpf);
273 list_add_tail(&skb->list, &list);
275 netif_receive_skb_list(&list);
280 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
283 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
284 int err = 0, act, ret, i, nframes = 0, batch_sz;
285 struct xdp_frame **frames = xdp->frames;
286 struct xdp_page_head *head;
287 struct xdp_frame *frm;
288 bool redirect = false;
289 struct xdp_buff *ctx;
292 batch_sz = min_t(u32, repeat, xdp->batch_size);
295 xdp_set_return_frame_no_direct();
297 for (i = 0; i < batch_sz; i++) {
298 page = page_pool_dev_alloc_pages(xdp->pp);
304 head = phys_to_virt(page_to_phys(page));
310 act = bpf_prog_run_xdp(prog, ctx);
312 /* if program changed pkt bounds we need to update the xdp_frame */
313 if (unlikely(ctx_was_changed(head))) {
314 ret = xdp_update_frame_from_buff(ctx, frm);
316 xdp_return_buff(ctx);
323 /* we can't do a real XDP_TX since we're not in the
324 * driver, so turn it into a REDIRECT back to the same
327 ri->tgt_index = xdp->dev->ifindex;
328 ri->map_id = INT_MAX;
329 ri->map_type = BPF_MAP_TYPE_UNSPEC;
333 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
335 xdp_return_buff(ctx);
338 frames[nframes++] = frm;
341 bpf_warn_invalid_xdp_action(NULL, prog, act);
344 xdp_return_buff(ctx);
353 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
358 xdp_clear_return_frame_no_direct();
363 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
364 u32 repeat, u32 batch_size, u32 *time)
367 struct xdp_test_data xdp = { .batch_size = batch_size };
368 struct bpf_test_timer t = { .mode = NO_MIGRATE };
374 ret = xdp_test_run_setup(&xdp, ctx);
378 bpf_test_timer_enter(&t);
381 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
382 if (unlikely(ret < 0))
384 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
385 bpf_test_timer_leave(&t);
387 xdp_test_run_teardown(&xdp);
391 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
392 u32 *retval, u32 *time, bool xdp)
394 struct bpf_prog_array_item item = {.prog = prog};
395 struct bpf_run_ctx *old_ctx;
396 struct bpf_cg_run_ctx run_ctx;
397 struct bpf_test_timer t = { NO_MIGRATE };
398 enum bpf_cgroup_storage_type stype;
401 for_each_cgroup_storage_type(stype) {
402 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
403 if (IS_ERR(item.cgroup_storage[stype])) {
404 item.cgroup_storage[stype] = NULL;
405 for_each_cgroup_storage_type(stype)
406 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
414 bpf_test_timer_enter(&t);
415 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
417 run_ctx.prog_item = &item;
420 *retval = bpf_prog_run_xdp(prog, ctx);
422 *retval = bpf_prog_run(prog, ctx);
424 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
425 bpf_reset_run_ctx(old_ctx);
426 bpf_test_timer_leave(&t);
428 for_each_cgroup_storage_type(stype)
429 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
434 static int bpf_test_finish(const union bpf_attr *kattr,
435 union bpf_attr __user *uattr, const void *data,
436 struct skb_shared_info *sinfo, u32 size,
437 u32 retval, u32 duration)
439 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
441 u32 copy_size = size;
443 /* Clamp copy if the user has provided a size hint, but copy the full
444 * buffer if not to retain old behaviour.
446 if (kattr->test.data_size_out &&
447 copy_size > kattr->test.data_size_out) {
448 copy_size = kattr->test.data_size_out;
453 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
460 if (copy_to_user(data_out, data, len))
467 for (i = 0; i < sinfo->nr_frags; i++) {
468 skb_frag_t *frag = &sinfo->frags[i];
470 if (offset >= copy_size) {
475 data_len = min_t(u32, copy_size - offset,
476 skb_frag_size(frag));
478 if (copy_to_user(data_out + offset,
479 skb_frag_address(frag),
488 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
490 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
492 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
497 trace_bpf_test_finish(&err);
501 /* Integer types of various sizes and pointer combinations cover variety of
502 * architecture dependent calling conventions. 7+ can be supported in the
506 __diag_ignore_all("-Wmissing-prototypes",
507 "Global functions as their definitions will be in vmlinux BTF");
508 __bpf_kfunc int bpf_fentry_test1(int a)
512 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
514 int noinline bpf_fentry_test2(int a, u64 b)
519 int noinline bpf_fentry_test3(char a, int b, u64 c)
524 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
526 return (long)a + b + c + d;
529 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
531 return a + (long)b + c + d + e;
534 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
536 return a + (long)b + c + d + (long)e + f;
539 struct bpf_fentry_test_t {
540 struct bpf_fentry_test_t *a;
543 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
548 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
553 __bpf_kfunc u32 bpf_fentry_test9(u32 *a)
558 __bpf_kfunc int bpf_modify_return_test(int a, int *b)
564 int noinline bpf_fentry_shadow_test(int a)
569 struct prog_test_member1 {
573 struct prog_test_member {
574 struct prog_test_member1 m;
578 struct prog_test_ref_kfunc {
581 struct prog_test_member memb;
582 struct prog_test_ref_kfunc *next;
586 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
588 refcount_dec(&p->cnt);
591 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
597 BTF_SET8_START(bpf_test_modify_return_ids)
598 BTF_ID_FLAGS(func, bpf_modify_return_test)
599 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
600 BTF_SET8_END(bpf_test_modify_return_ids)
602 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
603 .owner = THIS_MODULE,
604 .set = &bpf_test_modify_return_ids,
607 BTF_SET8_START(test_sk_check_kfunc_ids)
608 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
609 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
610 BTF_SET8_END(test_sk_check_kfunc_ids)
612 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
613 u32 size, u32 headroom, u32 tailroom)
615 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
618 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
619 return ERR_PTR(-EINVAL);
621 if (user_size > size)
622 return ERR_PTR(-EMSGSIZE);
624 size = SKB_DATA_ALIGN(size);
625 data = kzalloc(size + headroom + tailroom, GFP_USER);
627 return ERR_PTR(-ENOMEM);
629 if (copy_from_user(data + headroom, data_in, user_size)) {
631 return ERR_PTR(-EFAULT);
637 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
638 const union bpf_attr *kattr,
639 union bpf_attr __user *uattr)
641 struct bpf_fentry_test_t arg = {};
642 u16 side_effect = 0, ret = 0;
643 int b = 2, err = -EFAULT;
646 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
649 switch (prog->expected_attach_type) {
650 case BPF_TRACE_FENTRY:
651 case BPF_TRACE_FEXIT:
652 if (bpf_fentry_test1(1) != 2 ||
653 bpf_fentry_test2(2, 3) != 5 ||
654 bpf_fentry_test3(4, 5, 6) != 15 ||
655 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
656 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
657 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
658 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
659 bpf_fentry_test8(&arg) != 0 ||
660 bpf_fentry_test9(&retval) != 0)
663 case BPF_MODIFY_RETURN:
664 ret = bpf_modify_return_test(1, &b);
672 retval = ((u32)side_effect << 16) | ret;
673 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
678 trace_bpf_test_finish(&err);
682 struct bpf_raw_tp_test_run_info {
683 struct bpf_prog *prog;
689 __bpf_prog_test_run_raw_tp(void *data)
691 struct bpf_raw_tp_test_run_info *info = data;
694 info->retval = bpf_prog_run(info->prog, info->ctx);
698 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
699 const union bpf_attr *kattr,
700 union bpf_attr __user *uattr)
702 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
703 __u32 ctx_size_in = kattr->test.ctx_size_in;
704 struct bpf_raw_tp_test_run_info info;
705 int cpu = kattr->test.cpu, err = 0;
708 /* doesn't support data_in/out, ctx_out, duration, or repeat */
709 if (kattr->test.data_in || kattr->test.data_out ||
710 kattr->test.ctx_out || kattr->test.duration ||
711 kattr->test.repeat || kattr->test.batch_size)
714 if (ctx_size_in < prog->aux->max_ctx_offset ||
715 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
718 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
722 info.ctx = memdup_user(ctx_in, ctx_size_in);
723 if (IS_ERR(info.ctx))
724 return PTR_ERR(info.ctx);
731 current_cpu = get_cpu();
732 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
733 cpu == current_cpu) {
734 __bpf_prog_test_run_raw_tp(&info);
735 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
736 /* smp_call_function_single() also checks cpu_online()
737 * after csd_lock(). However, since cpu is from user
738 * space, let's do an extra quick check to filter out
739 * invalid value before smp_call_function_single().
743 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
749 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
756 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
758 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
759 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
760 u32 size = kattr->test.ctx_size_in;
764 if (!data_in && !data_out)
767 data = kzalloc(max_size, GFP_USER);
769 return ERR_PTR(-ENOMEM);
772 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
778 size = min_t(u32, max_size, size);
779 if (copy_from_user(data, data_in, size)) {
781 return ERR_PTR(-EFAULT);
787 static int bpf_ctx_finish(const union bpf_attr *kattr,
788 union bpf_attr __user *uattr, const void *data,
791 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
793 u32 copy_size = size;
795 if (!data || !data_out)
798 if (copy_size > kattr->test.ctx_size_out) {
799 copy_size = kattr->test.ctx_size_out;
803 if (copy_to_user(data_out, data, copy_size))
805 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
814 * range_is_zero - test whether buffer is initialized
815 * @buf: buffer to check
816 * @from: check from this position
817 * @to: check up until (excluding) this position
819 * This function returns true if the there is a non-zero byte
820 * in the buf in the range [from,to).
822 static inline bool range_is_zero(void *buf, size_t from, size_t to)
824 return !memchr_inv((u8 *)buf + from, 0, to - from);
827 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
829 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
834 /* make sure the fields we don't use are zeroed */
835 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
838 /* mark is allowed */
840 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
841 offsetof(struct __sk_buff, priority)))
844 /* priority is allowed */
845 /* ingress_ifindex is allowed */
846 /* ifindex is allowed */
848 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
849 offsetof(struct __sk_buff, cb)))
854 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
855 offsetof(struct __sk_buff, tstamp)))
858 /* tstamp is allowed */
859 /* wire_len is allowed */
860 /* gso_segs is allowed */
862 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
863 offsetof(struct __sk_buff, gso_size)))
866 /* gso_size is allowed */
868 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
869 offsetof(struct __sk_buff, hwtstamp)))
872 /* hwtstamp is allowed */
874 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
875 sizeof(struct __sk_buff)))
878 skb->mark = __skb->mark;
879 skb->priority = __skb->priority;
880 skb->skb_iif = __skb->ingress_ifindex;
881 skb->tstamp = __skb->tstamp;
882 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
884 if (__skb->wire_len == 0) {
885 cb->pkt_len = skb->len;
887 if (__skb->wire_len < skb->len ||
888 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
890 cb->pkt_len = __skb->wire_len;
893 if (__skb->gso_segs > GSO_MAX_SEGS)
895 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
896 skb_shinfo(skb)->gso_size = __skb->gso_size;
897 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
902 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
904 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
909 __skb->mark = skb->mark;
910 __skb->priority = skb->priority;
911 __skb->ingress_ifindex = skb->skb_iif;
912 __skb->ifindex = skb->dev->ifindex;
913 __skb->tstamp = skb->tstamp;
914 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
915 __skb->wire_len = cb->pkt_len;
916 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
917 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
920 static struct proto bpf_dummy_proto = {
922 .owner = THIS_MODULE,
923 .obj_size = sizeof(struct sock),
926 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
927 union bpf_attr __user *uattr)
929 bool is_l2 = false, is_direct_pkt_access = false;
930 struct net *net = current->nsproxy->net_ns;
931 struct net_device *dev = net->loopback_dev;
932 u32 size = kattr->test.data_size_in;
933 u32 repeat = kattr->test.repeat;
934 struct __sk_buff *ctx = NULL;
935 u32 retval, duration;
936 int hh_len = ETH_HLEN;
942 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
945 data = bpf_test_init(kattr, kattr->test.data_size_in,
946 size, NET_SKB_PAD + NET_IP_ALIGN,
947 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
949 return PTR_ERR(data);
951 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
957 switch (prog->type) {
958 case BPF_PROG_TYPE_SCHED_CLS:
959 case BPF_PROG_TYPE_SCHED_ACT:
962 case BPF_PROG_TYPE_LWT_IN:
963 case BPF_PROG_TYPE_LWT_OUT:
964 case BPF_PROG_TYPE_LWT_XMIT:
965 is_direct_pkt_access = true;
971 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
977 sock_init_data(NULL, sk);
979 skb = slab_build_skb(data);
988 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
989 __skb_put(skb, size);
990 if (ctx && ctx->ifindex > 1) {
991 dev = dev_get_by_index(net, ctx->ifindex);
997 skb->protocol = eth_type_trans(skb, dev);
998 skb_reset_network_header(skb);
1000 switch (skb->protocol) {
1001 case htons(ETH_P_IP):
1002 sk->sk_family = AF_INET;
1003 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1004 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1005 sk->sk_daddr = ip_hdr(skb)->daddr;
1008 #if IS_ENABLED(CONFIG_IPV6)
1009 case htons(ETH_P_IPV6):
1010 sk->sk_family = AF_INET6;
1011 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1012 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1013 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1022 __skb_push(skb, hh_len);
1023 if (is_direct_pkt_access)
1024 bpf_compute_data_pointers(skb);
1025 ret = convert___skb_to_skb(skb, ctx);
1028 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1032 if (skb_headroom(skb) < hh_len) {
1033 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1035 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1040 memset(__skb_push(skb, hh_len), 0, hh_len);
1042 convert_skb_to___skb(skb, ctx);
1045 /* bpf program can never convert linear skb to non-linear */
1046 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1047 size = skb_headlen(skb);
1048 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1051 ret = bpf_ctx_finish(kattr, uattr, ctx,
1052 sizeof(struct __sk_buff));
1054 if (dev && dev != net->loopback_dev)
1062 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1064 unsigned int ingress_ifindex, rx_queue_index;
1065 struct netdev_rx_queue *rxqueue;
1066 struct net_device *device;
1071 if (xdp_md->egress_ifindex != 0)
1074 ingress_ifindex = xdp_md->ingress_ifindex;
1075 rx_queue_index = xdp_md->rx_queue_index;
1077 if (!ingress_ifindex && rx_queue_index)
1080 if (ingress_ifindex) {
1081 device = dev_get_by_index(current->nsproxy->net_ns,
1086 if (rx_queue_index >= device->real_num_rx_queues)
1089 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1091 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1094 xdp->rxq = &rxqueue->xdp_rxq;
1095 /* The device is now tracked in the xdp->rxq for later
1100 xdp->data = xdp->data_meta + xdp_md->data;
1108 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1113 xdp_md->data = xdp->data - xdp->data_meta;
1114 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1116 if (xdp_md->ingress_ifindex)
1117 dev_put(xdp->rxq->dev);
1120 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1121 union bpf_attr __user *uattr)
1123 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1124 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1125 u32 batch_size = kattr->test.batch_size;
1126 u32 retval = 0, duration, max_data_sz;
1127 u32 size = kattr->test.data_size_in;
1128 u32 headroom = XDP_PACKET_HEADROOM;
1129 u32 repeat = kattr->test.repeat;
1130 struct netdev_rx_queue *rxqueue;
1131 struct skb_shared_info *sinfo;
1132 struct xdp_buff xdp = {};
1133 int i, ret = -EINVAL;
1137 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1138 prog->expected_attach_type == BPF_XDP_CPUMAP)
1141 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1144 if (bpf_prog_is_dev_bound(prog->aux))
1149 batch_size = NAPI_POLL_WEIGHT;
1150 else if (batch_size > TEST_XDP_MAX_BATCH)
1153 headroom += sizeof(struct xdp_page_head);
1154 } else if (batch_size) {
1158 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1160 return PTR_ERR(ctx);
1163 /* There can't be user provided data before the meta data */
1164 if (ctx->data_meta || ctx->data_end != size ||
1165 ctx->data > ctx->data_end ||
1166 unlikely(xdp_metalen_invalid(ctx->data)) ||
1167 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1169 /* Meta data is allocated from the headroom */
1170 headroom -= ctx->data;
1173 max_data_sz = 4096 - headroom - tailroom;
1174 if (size > max_data_sz) {
1175 /* disallow live data mode for jumbo frames */
1181 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1183 ret = PTR_ERR(data);
1187 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1188 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1189 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1190 xdp_prepare_buff(&xdp, data, headroom, size, true);
1191 sinfo = xdp_get_shared_info_from_buff(&xdp);
1193 ret = xdp_convert_md_to_buff(ctx, &xdp);
1197 if (unlikely(kattr->test.data_size_in > size)) {
1198 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1200 while (size < kattr->test.data_size_in) {
1205 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1210 page = alloc_page(GFP_KERNEL);
1216 frag = &sinfo->frags[sinfo->nr_frags++];
1218 data_len = min_t(u32, kattr->test.data_size_in - size,
1220 skb_frag_fill_page_desc(frag, page, 0, data_len);
1222 if (copy_from_user(page_address(page), data_in + size,
1227 sinfo->xdp_frags_size += data_len;
1230 xdp_buff_set_frags_flag(&xdp);
1234 bpf_prog_change_xdp(NULL, prog);
1237 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1239 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1240 /* We convert the xdp_buff back to an xdp_md before checking the return
1241 * code so the reference count of any held netdevice will be decremented
1242 * even if the test run failed.
1244 xdp_convert_buff_to_md(&xdp, ctx);
1248 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1249 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1252 ret = bpf_ctx_finish(kattr, uattr, ctx,
1253 sizeof(struct xdp_md));
1257 bpf_prog_change_xdp(prog, NULL);
1259 for (i = 0; i < sinfo->nr_frags; i++)
1260 __free_page(skb_frag_page(&sinfo->frags[i]));
1267 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1269 /* make sure the fields we don't use are zeroed */
1270 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1273 /* flags is allowed */
1275 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1276 sizeof(struct bpf_flow_keys)))
1282 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1283 const union bpf_attr *kattr,
1284 union bpf_attr __user *uattr)
1286 struct bpf_test_timer t = { NO_PREEMPT };
1287 u32 size = kattr->test.data_size_in;
1288 struct bpf_flow_dissector ctx = {};
1289 u32 repeat = kattr->test.repeat;
1290 struct bpf_flow_keys *user_ctx;
1291 struct bpf_flow_keys flow_keys;
1292 const struct ethhdr *eth;
1293 unsigned int flags = 0;
1294 u32 retval, duration;
1298 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1301 if (size < ETH_HLEN)
1304 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1306 return PTR_ERR(data);
1308 eth = (struct ethhdr *)data;
1313 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1314 if (IS_ERR(user_ctx)) {
1316 return PTR_ERR(user_ctx);
1319 ret = verify_user_bpf_flow_keys(user_ctx);
1322 flags = user_ctx->flags;
1325 ctx.flow_keys = &flow_keys;
1327 ctx.data_end = (__u8 *)data + size;
1329 bpf_test_timer_enter(&t);
1331 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1333 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1334 bpf_test_timer_leave(&t);
1339 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1340 sizeof(flow_keys), retval, duration);
1342 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1343 sizeof(struct bpf_flow_keys));
1351 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1352 union bpf_attr __user *uattr)
1354 struct bpf_test_timer t = { NO_PREEMPT };
1355 struct bpf_prog_array *progs = NULL;
1356 struct bpf_sk_lookup_kern ctx = {};
1357 u32 repeat = kattr->test.repeat;
1358 struct bpf_sk_lookup *user_ctx;
1359 u32 retval, duration;
1362 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1365 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1366 kattr->test.data_size_out)
1372 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1373 if (IS_ERR(user_ctx))
1374 return PTR_ERR(user_ctx);
1382 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1385 if (user_ctx->local_port > U16_MAX) {
1390 ctx.family = (u16)user_ctx->family;
1391 ctx.protocol = (u16)user_ctx->protocol;
1392 ctx.dport = (u16)user_ctx->local_port;
1393 ctx.sport = user_ctx->remote_port;
1395 switch (ctx.family) {
1397 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1398 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1401 #if IS_ENABLED(CONFIG_IPV6)
1403 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1404 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1409 ret = -EAFNOSUPPORT;
1413 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1419 progs->items[0].prog = prog;
1421 bpf_test_timer_enter(&t);
1423 ctx.selected_sk = NULL;
1424 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1425 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1426 bpf_test_timer_leave(&t);
1431 user_ctx->cookie = 0;
1432 if (ctx.selected_sk) {
1433 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1438 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1441 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1443 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1446 bpf_prog_array_free(progs);
1451 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1452 const union bpf_attr *kattr,
1453 union bpf_attr __user *uattr)
1455 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1456 __u32 ctx_size_in = kattr->test.ctx_size_in;
1461 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1462 if (kattr->test.data_in || kattr->test.data_out ||
1463 kattr->test.ctx_out || kattr->test.duration ||
1464 kattr->test.repeat || kattr->test.flags ||
1465 kattr->test.batch_size)
1468 if (ctx_size_in < prog->aux->max_ctx_offset ||
1469 ctx_size_in > U16_MAX)
1473 ctx = memdup_user(ctx_in, ctx_size_in);
1475 return PTR_ERR(ctx);
1478 rcu_read_lock_trace();
1479 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1480 rcu_read_unlock_trace();
1482 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1487 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1494 static int verify_and_copy_hook_state(struct nf_hook_state *state,
1495 const struct nf_hook_state *user,
1496 struct net_device *dev)
1498 if (user->in || user->out)
1501 if (user->net || user->sk || user->okfn)
1507 switch (state->hook) {
1508 case NF_INET_PRE_ROUTING:
1511 case NF_INET_LOCAL_IN:
1514 case NF_INET_FORWARD:
1518 case NF_INET_LOCAL_OUT:
1521 case NF_INET_POST_ROUTING:
1531 state->pf = user->pf;
1532 state->hook = user->hook;
1537 static __be16 nfproto_eth(int nfproto)
1541 return htons(ETH_P_IP);
1546 return htons(ETH_P_IPV6);
1549 int bpf_prog_test_run_nf(struct bpf_prog *prog,
1550 const union bpf_attr *kattr,
1551 union bpf_attr __user *uattr)
1553 struct net *net = current->nsproxy->net_ns;
1554 struct net_device *dev = net->loopback_dev;
1555 struct nf_hook_state *user_ctx, hook_state = {
1557 .hook = NF_INET_LOCAL_OUT,
1559 u32 size = kattr->test.data_size_in;
1560 u32 repeat = kattr->test.repeat;
1561 struct bpf_nf_ctx ctx = {
1562 .state = &hook_state,
1564 struct sk_buff *skb = NULL;
1565 u32 retval, duration;
1569 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1572 if (size < sizeof(struct iphdr))
1575 data = bpf_test_init(kattr, kattr->test.data_size_in, size,
1576 NET_SKB_PAD + NET_IP_ALIGN,
1577 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1579 return PTR_ERR(data);
1584 user_ctx = bpf_ctx_init(kattr, sizeof(struct nf_hook_state));
1585 if (IS_ERR(user_ctx)) {
1587 return PTR_ERR(user_ctx);
1591 ret = verify_and_copy_hook_state(&hook_state, user_ctx, dev);
1596 skb = slab_build_skb(data);
1602 data = NULL; /* data released via kfree_skb */
1604 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1605 __skb_put(skb, size);
1609 if (hook_state.hook != NF_INET_LOCAL_OUT) {
1610 if (size < ETH_HLEN + sizeof(struct iphdr))
1613 skb->protocol = eth_type_trans(skb, dev);
1614 switch (skb->protocol) {
1615 case htons(ETH_P_IP):
1616 if (hook_state.pf == NFPROTO_IPV4)
1619 case htons(ETH_P_IPV6):
1620 if (size < ETH_HLEN + sizeof(struct ipv6hdr))
1622 if (hook_state.pf == NFPROTO_IPV6)
1630 skb_reset_network_header(skb);
1632 skb->protocol = nfproto_eth(hook_state.pf);
1637 ret = bpf_test_run(prog, &ctx, repeat, &retval, &duration, false);
1641 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1650 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1651 .owner = THIS_MODULE,
1652 .set = &test_sk_check_kfunc_ids,
1655 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1656 BTF_ID(struct, prog_test_ref_kfunc)
1657 BTF_ID(func, bpf_kfunc_call_test_release)
1658 BTF_ID(struct, prog_test_member)
1659 BTF_ID(func, bpf_kfunc_call_memb_release)
1661 static int __init bpf_prog_test_run_init(void)
1663 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1665 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1666 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1669 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1670 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1675 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1676 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1677 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1678 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1679 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1680 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1683 late_initcall(bpf_prog_test_run_init);