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(netmem_ref netmem, void *arg)
132 struct xdp_page_head *head =
133 phys_to_virt(page_to_phys(netmem_to_page(netmem)));
134 struct xdp_buff *new_ctx, *orig_ctx;
135 u32 headroom = XDP_PACKET_HEADROOM;
136 struct xdp_test_data *xdp = arg;
137 size_t frm_len, meta_len;
138 struct xdp_frame *frm;
141 orig_ctx = xdp->orig_ctx;
142 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
143 meta_len = orig_ctx->data - orig_ctx->data_meta;
144 headroom -= meta_len;
146 new_ctx = &head->ctx;
149 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
151 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
152 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
153 new_ctx->data = new_ctx->data_meta + meta_len;
155 xdp_update_frame_from_buff(new_ctx, frm);
156 frm->mem = new_ctx->rxq->mem;
158 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
161 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
163 struct page_pool *pp;
165 struct page_pool_params pp_params = {
168 .pool_size = xdp->batch_size,
170 .init_callback = xdp_test_run_init_page,
174 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
178 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
182 pp = page_pool_create(&pp_params);
188 /* will copy 'mem.id' into pp->xdp_mem_id */
189 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
195 /* We create a 'fake' RXQ referencing the original dev, but with an
196 * xdp_mem_info pointing to our page_pool
198 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
199 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
200 xdp->rxq.mem.id = pp->xdp_mem_id;
201 xdp->dev = orig_ctx->rxq->dev;
202 xdp->orig_ctx = orig_ctx;
207 page_pool_destroy(pp);
215 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
217 xdp_unreg_mem_model(&xdp->mem);
218 page_pool_destroy(xdp->pp);
223 static bool frame_was_changed(const struct xdp_page_head *head)
225 /* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
226 * i.e. has the highest chances to be overwritten. If those two are
227 * untouched, it's most likely safe to skip the context reset.
229 return head->frame->data != head->orig_ctx.data ||
230 head->frame->flags != head->orig_ctx.flags;
233 static bool ctx_was_changed(struct xdp_page_head *head)
235 return head->orig_ctx.data != head->ctx.data ||
236 head->orig_ctx.data_meta != head->ctx.data_meta ||
237 head->orig_ctx.data_end != head->ctx.data_end;
240 static void reset_ctx(struct xdp_page_head *head)
242 if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
245 head->ctx.data = head->orig_ctx.data;
246 head->ctx.data_meta = head->orig_ctx.data_meta;
247 head->ctx.data_end = head->orig_ctx.data_end;
248 xdp_update_frame_from_buff(&head->ctx, head->frame);
249 head->frame->mem = head->orig_ctx.rxq->mem;
252 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
253 struct sk_buff **skbs,
254 struct net_device *dev)
256 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
260 n = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, nframes,
262 if (unlikely(n == 0)) {
263 for (i = 0; i < nframes; i++)
264 xdp_return_frame(frames[i]);
268 for (i = 0; i < nframes; i++) {
269 struct xdp_frame *xdpf = frames[i];
270 struct sk_buff *skb = skbs[i];
272 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
274 xdp_return_frame(xdpf);
278 list_add_tail(&skb->list, &list);
280 netif_receive_skb_list(&list);
285 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
288 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
289 int err = 0, act, ret, i, nframes = 0, batch_sz;
290 struct xdp_frame **frames = xdp->frames;
291 struct bpf_redirect_info *ri;
292 struct xdp_page_head *head;
293 struct xdp_frame *frm;
294 bool redirect = false;
295 struct xdp_buff *ctx;
298 batch_sz = min_t(u32, repeat, xdp->batch_size);
301 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
302 ri = bpf_net_ctx_get_ri();
303 xdp_set_return_frame_no_direct();
305 for (i = 0; i < batch_sz; i++) {
306 page = page_pool_dev_alloc_pages(xdp->pp);
312 head = phys_to_virt(page_to_phys(page));
318 act = bpf_prog_run_xdp(prog, ctx);
320 /* if program changed pkt bounds we need to update the xdp_frame */
321 if (unlikely(ctx_was_changed(head))) {
322 ret = xdp_update_frame_from_buff(ctx, frm);
324 xdp_return_buff(ctx);
331 /* we can't do a real XDP_TX since we're not in the
332 * driver, so turn it into a REDIRECT back to the same
335 ri->tgt_index = xdp->dev->ifindex;
336 ri->map_id = INT_MAX;
337 ri->map_type = BPF_MAP_TYPE_UNSPEC;
341 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
343 xdp_return_buff(ctx);
346 frames[nframes++] = frm;
349 bpf_warn_invalid_xdp_action(NULL, prog, act);
352 xdp_return_buff(ctx);
361 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
366 xdp_clear_return_frame_no_direct();
367 bpf_net_ctx_clear(bpf_net_ctx);
372 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
373 u32 repeat, u32 batch_size, u32 *time)
376 struct xdp_test_data xdp = { .batch_size = batch_size };
377 struct bpf_test_timer t = { .mode = NO_MIGRATE };
383 ret = xdp_test_run_setup(&xdp, ctx);
387 bpf_test_timer_enter(&t);
390 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
391 if (unlikely(ret < 0))
393 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
394 bpf_test_timer_leave(&t);
396 xdp_test_run_teardown(&xdp);
400 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
401 u32 *retval, u32 *time, bool xdp)
403 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
404 struct bpf_prog_array_item item = {.prog = prog};
405 struct bpf_run_ctx *old_ctx;
406 struct bpf_cg_run_ctx run_ctx;
407 struct bpf_test_timer t = { NO_MIGRATE };
408 enum bpf_cgroup_storage_type stype;
411 for_each_cgroup_storage_type(stype) {
412 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
413 if (IS_ERR(item.cgroup_storage[stype])) {
414 item.cgroup_storage[stype] = NULL;
415 for_each_cgroup_storage_type(stype)
416 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
424 bpf_test_timer_enter(&t);
425 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
427 run_ctx.prog_item = &item;
429 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
432 *retval = bpf_prog_run_xdp(prog, ctx);
434 *retval = bpf_prog_run(prog, ctx);
436 bpf_net_ctx_clear(bpf_net_ctx);
438 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
439 bpf_reset_run_ctx(old_ctx);
440 bpf_test_timer_leave(&t);
442 for_each_cgroup_storage_type(stype)
443 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
448 static int bpf_test_finish(const union bpf_attr *kattr,
449 union bpf_attr __user *uattr, const void *data,
450 struct skb_shared_info *sinfo, u32 size,
451 u32 retval, u32 duration)
453 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
455 u32 copy_size = size;
457 /* Clamp copy if the user has provided a size hint, but copy the full
458 * buffer if not to retain old behaviour.
460 if (kattr->test.data_size_out &&
461 copy_size > kattr->test.data_size_out) {
462 copy_size = kattr->test.data_size_out;
467 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
474 if (copy_to_user(data_out, data, len))
481 for (i = 0; i < sinfo->nr_frags; i++) {
482 skb_frag_t *frag = &sinfo->frags[i];
484 if (offset >= copy_size) {
489 data_len = min_t(u32, copy_size - offset,
490 skb_frag_size(frag));
492 if (copy_to_user(data_out + offset,
493 skb_frag_address(frag),
502 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
504 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
506 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
511 trace_bpf_test_finish(&err);
515 /* Integer types of various sizes and pointer combinations cover variety of
516 * architecture dependent calling conventions. 7+ can be supported in the
519 __bpf_kfunc_start_defs();
521 __bpf_kfunc int bpf_fentry_test1(int a)
525 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
527 int noinline bpf_fentry_test2(int a, u64 b)
532 int noinline bpf_fentry_test3(char a, int b, u64 c)
537 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
539 return (long)a + b + c + d;
542 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
544 return a + (long)b + c + d + e;
547 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
549 return a + (long)b + c + d + (long)e + f;
552 struct bpf_fentry_test_t {
553 struct bpf_fentry_test_t *a;
556 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
558 asm volatile ("": "+r"(arg));
562 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
567 __bpf_kfunc u32 bpf_fentry_test9(u32 *a)
572 void noinline bpf_fentry_test_sinfo(struct skb_shared_info *sinfo)
576 __bpf_kfunc int bpf_modify_return_test(int a, int *b)
582 __bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
583 void *e, char f, int g)
586 return a + *b + c + d + (long)e + f + g;
589 __bpf_kfunc int bpf_modify_return_test_tp(int nonce)
591 trace_bpf_trigger_tp(nonce);
596 int noinline bpf_fentry_shadow_test(int a)
601 struct prog_test_member1 {
605 struct prog_test_member {
606 struct prog_test_member1 m;
610 struct prog_test_ref_kfunc {
613 struct prog_test_member memb;
614 struct prog_test_ref_kfunc *next;
618 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
620 refcount_dec(&p->cnt);
623 __bpf_kfunc void bpf_kfunc_call_test_release_dtor(void *p)
625 bpf_kfunc_call_test_release(p);
627 CFI_NOSEAL(bpf_kfunc_call_test_release_dtor);
629 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
633 __bpf_kfunc void bpf_kfunc_call_memb_release_dtor(void *p)
636 CFI_NOSEAL(bpf_kfunc_call_memb_release_dtor);
638 __bpf_kfunc_end_defs();
640 BTF_KFUNCS_START(bpf_test_modify_return_ids)
641 BTF_ID_FLAGS(func, bpf_modify_return_test)
642 BTF_ID_FLAGS(func, bpf_modify_return_test2)
643 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
644 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
645 BTF_KFUNCS_END(bpf_test_modify_return_ids)
647 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
648 .owner = THIS_MODULE,
649 .set = &bpf_test_modify_return_ids,
652 BTF_KFUNCS_START(test_sk_check_kfunc_ids)
653 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
654 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
655 BTF_KFUNCS_END(test_sk_check_kfunc_ids)
657 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
658 u32 size, u32 headroom, u32 tailroom)
660 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
663 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
664 return ERR_PTR(-EINVAL);
666 if (user_size > size)
667 return ERR_PTR(-EMSGSIZE);
669 size = SKB_DATA_ALIGN(size);
670 data = kzalloc(size + headroom + tailroom, GFP_USER);
672 return ERR_PTR(-ENOMEM);
674 if (copy_from_user(data + headroom, data_in, user_size)) {
676 return ERR_PTR(-EFAULT);
682 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
683 const union bpf_attr *kattr,
684 union bpf_attr __user *uattr)
686 struct bpf_fentry_test_t arg = {};
687 u16 side_effect = 0, ret = 0;
688 int b = 2, err = -EFAULT;
691 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
694 switch (prog->expected_attach_type) {
695 case BPF_TRACE_FENTRY:
696 case BPF_TRACE_FEXIT:
697 if (bpf_fentry_test1(1) != 2 ||
698 bpf_fentry_test2(2, 3) != 5 ||
699 bpf_fentry_test3(4, 5, 6) != 15 ||
700 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
701 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
702 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
703 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
704 bpf_fentry_test8(&arg) != 0 ||
705 bpf_fentry_test9(&retval) != 0)
708 case BPF_MODIFY_RETURN:
709 ret = bpf_modify_return_test(1, &b);
713 ret += bpf_modify_return_test2(1, &b, 3, 4, (void *)5, 6, 7);
721 retval = ((u32)side_effect << 16) | ret;
722 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
727 trace_bpf_test_finish(&err);
731 struct bpf_raw_tp_test_run_info {
732 struct bpf_prog *prog;
738 __bpf_prog_test_run_raw_tp(void *data)
740 struct bpf_raw_tp_test_run_info *info = data;
741 struct bpf_trace_run_ctx run_ctx = {};
742 struct bpf_run_ctx *old_run_ctx;
744 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
747 info->retval = bpf_prog_run(info->prog, info->ctx);
750 bpf_reset_run_ctx(old_run_ctx);
753 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
754 const union bpf_attr *kattr,
755 union bpf_attr __user *uattr)
757 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
758 __u32 ctx_size_in = kattr->test.ctx_size_in;
759 struct bpf_raw_tp_test_run_info info;
760 int cpu = kattr->test.cpu, err = 0;
763 /* doesn't support data_in/out, ctx_out, duration, or repeat */
764 if (kattr->test.data_in || kattr->test.data_out ||
765 kattr->test.ctx_out || kattr->test.duration ||
766 kattr->test.repeat || kattr->test.batch_size)
769 if (ctx_size_in < prog->aux->max_ctx_offset ||
770 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
773 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
777 info.ctx = memdup_user(ctx_in, ctx_size_in);
778 if (IS_ERR(info.ctx))
779 return PTR_ERR(info.ctx);
786 current_cpu = get_cpu();
787 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
788 cpu == current_cpu) {
789 __bpf_prog_test_run_raw_tp(&info);
790 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
791 /* smp_call_function_single() also checks cpu_online()
792 * after csd_lock(). However, since cpu is from user
793 * space, let's do an extra quick check to filter out
794 * invalid value before smp_call_function_single().
798 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
804 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
811 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
813 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
814 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
815 u32 size = kattr->test.ctx_size_in;
819 if (!data_in && !data_out)
822 data = kzalloc(max_size, GFP_USER);
824 return ERR_PTR(-ENOMEM);
827 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
833 size = min_t(u32, max_size, size);
834 if (copy_from_user(data, data_in, size)) {
836 return ERR_PTR(-EFAULT);
842 static int bpf_ctx_finish(const union bpf_attr *kattr,
843 union bpf_attr __user *uattr, const void *data,
846 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
848 u32 copy_size = size;
850 if (!data || !data_out)
853 if (copy_size > kattr->test.ctx_size_out) {
854 copy_size = kattr->test.ctx_size_out;
858 if (copy_to_user(data_out, data, copy_size))
860 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
869 * range_is_zero - test whether buffer is initialized
870 * @buf: buffer to check
871 * @from: check from this position
872 * @to: check up until (excluding) this position
874 * This function returns true if the there is a non-zero byte
875 * in the buf in the range [from,to).
877 static inline bool range_is_zero(void *buf, size_t from, size_t to)
879 return !memchr_inv((u8 *)buf + from, 0, to - from);
882 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
884 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
889 /* make sure the fields we don't use are zeroed */
890 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
893 /* mark is allowed */
895 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
896 offsetof(struct __sk_buff, priority)))
899 /* priority is allowed */
900 /* ingress_ifindex is allowed */
901 /* ifindex is allowed */
903 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
904 offsetof(struct __sk_buff, cb)))
909 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
910 offsetof(struct __sk_buff, tstamp)))
913 /* tstamp is allowed */
914 /* wire_len is allowed */
915 /* gso_segs is allowed */
917 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
918 offsetof(struct __sk_buff, gso_size)))
921 /* gso_size is allowed */
923 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
924 offsetof(struct __sk_buff, hwtstamp)))
927 /* hwtstamp is allowed */
929 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
930 sizeof(struct __sk_buff)))
933 skb->mark = __skb->mark;
934 skb->priority = __skb->priority;
935 skb->skb_iif = __skb->ingress_ifindex;
936 skb->tstamp = __skb->tstamp;
937 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
939 if (__skb->wire_len == 0) {
940 cb->pkt_len = skb->len;
942 if (__skb->wire_len < skb->len ||
943 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
945 cb->pkt_len = __skb->wire_len;
948 if (__skb->gso_segs > GSO_MAX_SEGS)
950 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
951 skb_shinfo(skb)->gso_size = __skb->gso_size;
952 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
957 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
959 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
964 __skb->mark = skb->mark;
965 __skb->priority = skb->priority;
966 __skb->ingress_ifindex = skb->skb_iif;
967 __skb->ifindex = skb->dev->ifindex;
968 __skb->tstamp = skb->tstamp;
969 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
970 __skb->wire_len = cb->pkt_len;
971 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
972 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
975 static struct proto bpf_dummy_proto = {
977 .owner = THIS_MODULE,
978 .obj_size = sizeof(struct sock),
981 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
982 union bpf_attr __user *uattr)
984 bool is_l2 = false, is_direct_pkt_access = false;
985 struct net *net = current->nsproxy->net_ns;
986 struct net_device *dev = net->loopback_dev;
987 u32 size = kattr->test.data_size_in;
988 u32 repeat = kattr->test.repeat;
989 struct __sk_buff *ctx = NULL;
990 u32 retval, duration;
991 int hh_len = ETH_HLEN;
997 if ((kattr->test.flags & ~BPF_F_TEST_SKB_CHECKSUM_COMPLETE) ||
998 kattr->test.cpu || kattr->test.batch_size)
1001 data = bpf_test_init(kattr, kattr->test.data_size_in,
1002 size, NET_SKB_PAD + NET_IP_ALIGN,
1003 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1005 return PTR_ERR(data);
1007 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1010 return PTR_ERR(ctx);
1013 switch (prog->type) {
1014 case BPF_PROG_TYPE_SCHED_CLS:
1015 case BPF_PROG_TYPE_SCHED_ACT:
1018 case BPF_PROG_TYPE_LWT_IN:
1019 case BPF_PROG_TYPE_LWT_OUT:
1020 case BPF_PROG_TYPE_LWT_XMIT:
1021 is_direct_pkt_access = true;
1027 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1033 sock_init_data(NULL, sk);
1035 skb = slab_build_skb(data);
1044 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1045 __skb_put(skb, size);
1047 if (ctx && ctx->ifindex > 1) {
1048 dev = dev_get_by_index(net, ctx->ifindex);
1054 skb->protocol = eth_type_trans(skb, dev);
1055 skb_reset_network_header(skb);
1057 switch (skb->protocol) {
1058 case htons(ETH_P_IP):
1059 sk->sk_family = AF_INET;
1060 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1061 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1062 sk->sk_daddr = ip_hdr(skb)->daddr;
1065 #if IS_ENABLED(CONFIG_IPV6)
1066 case htons(ETH_P_IPV6):
1067 sk->sk_family = AF_INET6;
1068 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1069 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1070 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1079 __skb_push(skb, hh_len);
1080 if (is_direct_pkt_access)
1081 bpf_compute_data_pointers(skb);
1083 ret = convert___skb_to_skb(skb, ctx);
1087 if (kattr->test.flags & BPF_F_TEST_SKB_CHECKSUM_COMPLETE) {
1088 const int off = skb_network_offset(skb);
1089 int len = skb->len - off;
1091 skb->csum = skb_checksum(skb, off, len, 0);
1092 skb->ip_summed = CHECKSUM_COMPLETE;
1095 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1099 if (skb_headroom(skb) < hh_len) {
1100 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1102 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1107 memset(__skb_push(skb, hh_len), 0, hh_len);
1110 if (kattr->test.flags & BPF_F_TEST_SKB_CHECKSUM_COMPLETE) {
1111 const int off = skb_network_offset(skb);
1112 int len = skb->len - off;
1115 csum = skb_checksum(skb, off, len, 0);
1117 if (csum_fold(skb->csum) != csum_fold(csum)) {
1123 convert_skb_to___skb(skb, ctx);
1126 /* bpf program can never convert linear skb to non-linear */
1127 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1128 size = skb_headlen(skb);
1129 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1132 ret = bpf_ctx_finish(kattr, uattr, ctx,
1133 sizeof(struct __sk_buff));
1135 if (dev && dev != net->loopback_dev)
1143 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1145 unsigned int ingress_ifindex, rx_queue_index;
1146 struct netdev_rx_queue *rxqueue;
1147 struct net_device *device;
1152 if (xdp_md->egress_ifindex != 0)
1155 ingress_ifindex = xdp_md->ingress_ifindex;
1156 rx_queue_index = xdp_md->rx_queue_index;
1158 if (!ingress_ifindex && rx_queue_index)
1161 if (ingress_ifindex) {
1162 device = dev_get_by_index(current->nsproxy->net_ns,
1167 if (rx_queue_index >= device->real_num_rx_queues)
1170 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1172 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1175 xdp->rxq = &rxqueue->xdp_rxq;
1176 /* The device is now tracked in the xdp->rxq for later
1181 xdp->data = xdp->data_meta + xdp_md->data;
1189 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1194 xdp_md->data = xdp->data - xdp->data_meta;
1195 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1197 if (xdp_md->ingress_ifindex)
1198 dev_put(xdp->rxq->dev);
1201 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1202 union bpf_attr __user *uattr)
1204 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1205 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1206 u32 batch_size = kattr->test.batch_size;
1207 u32 retval = 0, duration, max_data_sz;
1208 u32 size = kattr->test.data_size_in;
1209 u32 headroom = XDP_PACKET_HEADROOM;
1210 u32 repeat = kattr->test.repeat;
1211 struct netdev_rx_queue *rxqueue;
1212 struct skb_shared_info *sinfo;
1213 struct xdp_buff xdp = {};
1214 int i, ret = -EINVAL;
1218 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1219 prog->expected_attach_type == BPF_XDP_CPUMAP)
1222 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1225 if (bpf_prog_is_dev_bound(prog->aux))
1230 batch_size = NAPI_POLL_WEIGHT;
1231 else if (batch_size > TEST_XDP_MAX_BATCH)
1234 headroom += sizeof(struct xdp_page_head);
1235 } else if (batch_size) {
1239 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1241 return PTR_ERR(ctx);
1244 /* There can't be user provided data before the meta data */
1245 if (ctx->data_meta || ctx->data_end != size ||
1246 ctx->data > ctx->data_end ||
1247 unlikely(xdp_metalen_invalid(ctx->data)) ||
1248 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1250 /* Meta data is allocated from the headroom */
1251 headroom -= ctx->data;
1254 max_data_sz = 4096 - headroom - tailroom;
1255 if (size > max_data_sz) {
1256 /* disallow live data mode for jumbo frames */
1262 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1264 ret = PTR_ERR(data);
1268 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1269 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1270 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1271 xdp_prepare_buff(&xdp, data, headroom, size, true);
1272 sinfo = xdp_get_shared_info_from_buff(&xdp);
1274 ret = xdp_convert_md_to_buff(ctx, &xdp);
1278 if (unlikely(kattr->test.data_size_in > size)) {
1279 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1281 while (size < kattr->test.data_size_in) {
1286 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1291 page = alloc_page(GFP_KERNEL);
1297 frag = &sinfo->frags[sinfo->nr_frags++];
1299 data_len = min_t(u32, kattr->test.data_size_in - size,
1301 skb_frag_fill_page_desc(frag, page, 0, data_len);
1303 if (copy_from_user(page_address(page), data_in + size,
1308 sinfo->xdp_frags_size += data_len;
1311 xdp_buff_set_frags_flag(&xdp);
1315 bpf_prog_change_xdp(NULL, prog);
1318 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1320 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1321 /* We convert the xdp_buff back to an xdp_md before checking the return
1322 * code so the reference count of any held netdevice will be decremented
1323 * even if the test run failed.
1325 xdp_convert_buff_to_md(&xdp, ctx);
1329 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1330 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1333 ret = bpf_ctx_finish(kattr, uattr, ctx,
1334 sizeof(struct xdp_md));
1338 bpf_prog_change_xdp(prog, NULL);
1340 for (i = 0; i < sinfo->nr_frags; i++)
1341 __free_page(skb_frag_page(&sinfo->frags[i]));
1348 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1350 /* make sure the fields we don't use are zeroed */
1351 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1354 /* flags is allowed */
1356 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1357 sizeof(struct bpf_flow_keys)))
1363 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1364 const union bpf_attr *kattr,
1365 union bpf_attr __user *uattr)
1367 struct bpf_test_timer t = { NO_PREEMPT };
1368 u32 size = kattr->test.data_size_in;
1369 struct bpf_flow_dissector ctx = {};
1370 u32 repeat = kattr->test.repeat;
1371 struct bpf_flow_keys *user_ctx;
1372 struct bpf_flow_keys flow_keys;
1373 const struct ethhdr *eth;
1374 unsigned int flags = 0;
1375 u32 retval, duration;
1379 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1382 if (size < ETH_HLEN)
1385 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1387 return PTR_ERR(data);
1389 eth = (struct ethhdr *)data;
1394 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1395 if (IS_ERR(user_ctx)) {
1397 return PTR_ERR(user_ctx);
1400 ret = verify_user_bpf_flow_keys(user_ctx);
1403 flags = user_ctx->flags;
1406 ctx.flow_keys = &flow_keys;
1408 ctx.data_end = (__u8 *)data + size;
1410 bpf_test_timer_enter(&t);
1412 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1414 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1415 bpf_test_timer_leave(&t);
1420 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1421 sizeof(flow_keys), retval, duration);
1423 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1424 sizeof(struct bpf_flow_keys));
1432 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1433 union bpf_attr __user *uattr)
1435 struct bpf_test_timer t = { NO_PREEMPT };
1436 struct bpf_prog_array *progs = NULL;
1437 struct bpf_sk_lookup_kern ctx = {};
1438 u32 repeat = kattr->test.repeat;
1439 struct bpf_sk_lookup *user_ctx;
1440 u32 retval, duration;
1443 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1446 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1447 kattr->test.data_size_out)
1453 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1454 if (IS_ERR(user_ctx))
1455 return PTR_ERR(user_ctx);
1463 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1466 if (user_ctx->local_port > U16_MAX) {
1471 ctx.family = (u16)user_ctx->family;
1472 ctx.protocol = (u16)user_ctx->protocol;
1473 ctx.dport = (u16)user_ctx->local_port;
1474 ctx.sport = user_ctx->remote_port;
1476 switch (ctx.family) {
1478 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1479 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1482 #if IS_ENABLED(CONFIG_IPV6)
1484 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1485 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1490 ret = -EAFNOSUPPORT;
1494 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1500 progs->items[0].prog = prog;
1502 bpf_test_timer_enter(&t);
1504 ctx.selected_sk = NULL;
1505 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1506 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1507 bpf_test_timer_leave(&t);
1512 user_ctx->cookie = 0;
1513 if (ctx.selected_sk) {
1514 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1519 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1522 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1524 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1527 bpf_prog_array_free(progs);
1532 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1533 const union bpf_attr *kattr,
1534 union bpf_attr __user *uattr)
1536 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1537 __u32 ctx_size_in = kattr->test.ctx_size_in;
1542 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1543 if (kattr->test.data_in || kattr->test.data_out ||
1544 kattr->test.ctx_out || kattr->test.duration ||
1545 kattr->test.repeat || kattr->test.flags ||
1546 kattr->test.batch_size)
1549 if (ctx_size_in < prog->aux->max_ctx_offset ||
1550 ctx_size_in > U16_MAX)
1554 ctx = memdup_user(ctx_in, ctx_size_in);
1556 return PTR_ERR(ctx);
1559 rcu_read_lock_trace();
1560 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1561 rcu_read_unlock_trace();
1563 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1568 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1575 static int verify_and_copy_hook_state(struct nf_hook_state *state,
1576 const struct nf_hook_state *user,
1577 struct net_device *dev)
1579 if (user->in || user->out)
1582 if (user->net || user->sk || user->okfn)
1588 switch (state->hook) {
1589 case NF_INET_PRE_ROUTING:
1592 case NF_INET_LOCAL_IN:
1595 case NF_INET_FORWARD:
1599 case NF_INET_LOCAL_OUT:
1602 case NF_INET_POST_ROUTING:
1612 state->pf = user->pf;
1613 state->hook = user->hook;
1618 static __be16 nfproto_eth(int nfproto)
1622 return htons(ETH_P_IP);
1627 return htons(ETH_P_IPV6);
1630 int bpf_prog_test_run_nf(struct bpf_prog *prog,
1631 const union bpf_attr *kattr,
1632 union bpf_attr __user *uattr)
1634 struct net *net = current->nsproxy->net_ns;
1635 struct net_device *dev = net->loopback_dev;
1636 struct nf_hook_state *user_ctx, hook_state = {
1638 .hook = NF_INET_LOCAL_OUT,
1640 u32 size = kattr->test.data_size_in;
1641 u32 repeat = kattr->test.repeat;
1642 struct bpf_nf_ctx ctx = {
1643 .state = &hook_state,
1645 struct sk_buff *skb = NULL;
1646 u32 retval, duration;
1650 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1653 if (size < sizeof(struct iphdr))
1656 data = bpf_test_init(kattr, kattr->test.data_size_in, size,
1657 NET_SKB_PAD + NET_IP_ALIGN,
1658 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1660 return PTR_ERR(data);
1665 user_ctx = bpf_ctx_init(kattr, sizeof(struct nf_hook_state));
1666 if (IS_ERR(user_ctx)) {
1668 return PTR_ERR(user_ctx);
1672 ret = verify_and_copy_hook_state(&hook_state, user_ctx, dev);
1677 skb = slab_build_skb(data);
1683 data = NULL; /* data released via kfree_skb */
1685 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1686 __skb_put(skb, size);
1690 if (hook_state.hook != NF_INET_LOCAL_OUT) {
1691 if (size < ETH_HLEN + sizeof(struct iphdr))
1694 skb->protocol = eth_type_trans(skb, dev);
1695 switch (skb->protocol) {
1696 case htons(ETH_P_IP):
1697 if (hook_state.pf == NFPROTO_IPV4)
1700 case htons(ETH_P_IPV6):
1701 if (size < ETH_HLEN + sizeof(struct ipv6hdr))
1703 if (hook_state.pf == NFPROTO_IPV6)
1711 skb_reset_network_header(skb);
1713 skb->protocol = nfproto_eth(hook_state.pf);
1718 ret = bpf_test_run(prog, &ctx, repeat, &retval, &duration, false);
1722 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1731 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1732 .owner = THIS_MODULE,
1733 .set = &test_sk_check_kfunc_ids,
1736 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1737 BTF_ID(struct, prog_test_ref_kfunc)
1738 BTF_ID(func, bpf_kfunc_call_test_release_dtor)
1739 BTF_ID(struct, prog_test_member)
1740 BTF_ID(func, bpf_kfunc_call_memb_release_dtor)
1742 static int __init bpf_prog_test_run_init(void)
1744 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1746 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1747 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1750 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1751 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1756 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1757 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1758 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1759 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1760 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1761 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1764 late_initcall(bpf_prog_test_run_init);