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>
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/bpf_test_run.h>
27 struct bpf_test_timer {
28 enum { NO_PREEMPT, NO_MIGRATE } mode;
30 u64 time_start, time_spent;
33 static void bpf_test_timer_enter(struct bpf_test_timer *t)
37 if (t->mode == NO_PREEMPT)
42 t->time_start = ktime_get_ns();
45 static void bpf_test_timer_leave(struct bpf_test_timer *t)
50 if (t->mode == NO_PREEMPT)
57 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
58 u32 repeat, int *err, u32 *duration)
64 t->time_spent += ktime_get_ns() - t->time_start;
65 do_div(t->time_spent, t->i);
66 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
71 if (signal_pending(current)) {
72 /* During iteration: we've been cancelled, abort. */
78 /* During iteration: we need to reschedule between runs. */
79 t->time_spent += ktime_get_ns() - t->time_start;
80 bpf_test_timer_leave(t);
82 bpf_test_timer_enter(t);
85 /* Do another round. */
93 /* We put this struct at the head of each page with a context and frame
94 * initialised when the page is allocated, so we don't have to do this on each
95 * repetition of the test run.
97 struct xdp_page_head {
98 struct xdp_buff orig_ctx;
101 /* ::data_hard_start starts here */
102 DECLARE_FLEX_ARRAY(struct xdp_frame, frame);
103 DECLARE_FLEX_ARRAY(u8, data);
107 struct xdp_test_data {
108 struct xdp_buff *orig_ctx;
109 struct xdp_rxq_info rxq;
110 struct net_device *dev;
111 struct page_pool *pp;
112 struct xdp_frame **frames;
113 struct sk_buff **skbs;
114 struct xdp_mem_info mem;
119 /* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE
120 * must be updated accordingly this gets changed, otherwise BPF selftests
123 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
124 #define TEST_XDP_MAX_BATCH 256
126 static void xdp_test_run_init_page(struct page *page, void *arg)
128 struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
129 struct xdp_buff *new_ctx, *orig_ctx;
130 u32 headroom = XDP_PACKET_HEADROOM;
131 struct xdp_test_data *xdp = arg;
132 size_t frm_len, meta_len;
133 struct xdp_frame *frm;
136 orig_ctx = xdp->orig_ctx;
137 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
138 meta_len = orig_ctx->data - orig_ctx->data_meta;
139 headroom -= meta_len;
141 new_ctx = &head->ctx;
144 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
146 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
147 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
148 new_ctx->data = new_ctx->data_meta + meta_len;
150 xdp_update_frame_from_buff(new_ctx, frm);
151 frm->mem = new_ctx->rxq->mem;
153 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
156 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
158 struct page_pool *pp;
160 struct page_pool_params pp_params = {
163 .pool_size = xdp->batch_size,
165 .init_callback = xdp_test_run_init_page,
169 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
173 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
177 pp = page_pool_create(&pp_params);
183 /* will copy 'mem.id' into pp->xdp_mem_id */
184 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
190 /* We create a 'fake' RXQ referencing the original dev, but with an
191 * xdp_mem_info pointing to our page_pool
193 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
194 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
195 xdp->rxq.mem.id = pp->xdp_mem_id;
196 xdp->dev = orig_ctx->rxq->dev;
197 xdp->orig_ctx = orig_ctx;
202 page_pool_destroy(pp);
210 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
212 xdp_unreg_mem_model(&xdp->mem);
213 page_pool_destroy(xdp->pp);
218 static bool ctx_was_changed(struct xdp_page_head *head)
220 return head->orig_ctx.data != head->ctx.data ||
221 head->orig_ctx.data_meta != head->ctx.data_meta ||
222 head->orig_ctx.data_end != head->ctx.data_end;
225 static void reset_ctx(struct xdp_page_head *head)
227 if (likely(!ctx_was_changed(head)))
230 head->ctx.data = head->orig_ctx.data;
231 head->ctx.data_meta = head->orig_ctx.data_meta;
232 head->ctx.data_end = head->orig_ctx.data_end;
233 xdp_update_frame_from_buff(&head->ctx, head->frame);
236 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
237 struct sk_buff **skbs,
238 struct net_device *dev)
240 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
244 n = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, (void **)skbs);
245 if (unlikely(n == 0)) {
246 for (i = 0; i < nframes; i++)
247 xdp_return_frame(frames[i]);
251 for (i = 0; i < nframes; i++) {
252 struct xdp_frame *xdpf = frames[i];
253 struct sk_buff *skb = skbs[i];
255 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
257 xdp_return_frame(xdpf);
261 list_add_tail(&skb->list, &list);
263 netif_receive_skb_list(&list);
268 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
271 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
272 int err = 0, act, ret, i, nframes = 0, batch_sz;
273 struct xdp_frame **frames = xdp->frames;
274 struct xdp_page_head *head;
275 struct xdp_frame *frm;
276 bool redirect = false;
277 struct xdp_buff *ctx;
280 batch_sz = min_t(u32, repeat, xdp->batch_size);
283 xdp_set_return_frame_no_direct();
285 for (i = 0; i < batch_sz; i++) {
286 page = page_pool_dev_alloc_pages(xdp->pp);
292 head = phys_to_virt(page_to_phys(page));
298 act = bpf_prog_run_xdp(prog, ctx);
300 /* if program changed pkt bounds we need to update the xdp_frame */
301 if (unlikely(ctx_was_changed(head))) {
302 ret = xdp_update_frame_from_buff(ctx, frm);
304 xdp_return_buff(ctx);
311 /* we can't do a real XDP_TX since we're not in the
312 * driver, so turn it into a REDIRECT back to the same
315 ri->tgt_index = xdp->dev->ifindex;
316 ri->map_id = INT_MAX;
317 ri->map_type = BPF_MAP_TYPE_UNSPEC;
321 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
323 xdp_return_buff(ctx);
326 frames[nframes++] = frm;
329 bpf_warn_invalid_xdp_action(NULL, prog, act);
332 xdp_return_buff(ctx);
341 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
346 xdp_clear_return_frame_no_direct();
351 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
352 u32 repeat, u32 batch_size, u32 *time)
355 struct xdp_test_data xdp = { .batch_size = batch_size };
356 struct bpf_test_timer t = { .mode = NO_MIGRATE };
362 ret = xdp_test_run_setup(&xdp, ctx);
366 bpf_test_timer_enter(&t);
369 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
370 if (unlikely(ret < 0))
372 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
373 bpf_test_timer_leave(&t);
375 xdp_test_run_teardown(&xdp);
379 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
380 u32 *retval, u32 *time, bool xdp)
382 struct bpf_prog_array_item item = {.prog = prog};
383 struct bpf_run_ctx *old_ctx;
384 struct bpf_cg_run_ctx run_ctx;
385 struct bpf_test_timer t = { NO_MIGRATE };
386 enum bpf_cgroup_storage_type stype;
389 for_each_cgroup_storage_type(stype) {
390 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
391 if (IS_ERR(item.cgroup_storage[stype])) {
392 item.cgroup_storage[stype] = NULL;
393 for_each_cgroup_storage_type(stype)
394 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
402 bpf_test_timer_enter(&t);
403 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
405 run_ctx.prog_item = &item;
408 *retval = bpf_prog_run_xdp(prog, ctx);
410 *retval = bpf_prog_run(prog, ctx);
412 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
413 bpf_reset_run_ctx(old_ctx);
414 bpf_test_timer_leave(&t);
416 for_each_cgroup_storage_type(stype)
417 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
422 static int bpf_test_finish(const union bpf_attr *kattr,
423 union bpf_attr __user *uattr, const void *data,
424 struct skb_shared_info *sinfo, u32 size,
425 u32 retval, u32 duration)
427 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
429 u32 copy_size = size;
431 /* Clamp copy if the user has provided a size hint, but copy the full
432 * buffer if not to retain old behaviour.
434 if (kattr->test.data_size_out &&
435 copy_size > kattr->test.data_size_out) {
436 copy_size = kattr->test.data_size_out;
441 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
448 if (copy_to_user(data_out, data, len))
455 for (i = 0; i < sinfo->nr_frags; i++) {
456 skb_frag_t *frag = &sinfo->frags[i];
458 if (offset >= copy_size) {
463 data_len = min_t(u32, copy_size - offset,
464 skb_frag_size(frag));
466 if (copy_to_user(data_out + offset,
467 skb_frag_address(frag),
476 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
478 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
480 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
485 trace_bpf_test_finish(&err);
489 /* Integer types of various sizes and pointer combinations cover variety of
490 * architecture dependent calling conventions. 7+ can be supported in the
494 __diag_ignore_all("-Wmissing-prototypes",
495 "Global functions as their definitions will be in vmlinux BTF");
496 __bpf_kfunc int bpf_fentry_test1(int a)
500 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
502 int noinline bpf_fentry_test2(int a, u64 b)
507 int noinline bpf_fentry_test3(char a, int b, u64 c)
512 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
514 return (long)a + b + c + d;
517 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
519 return a + (long)b + c + d + e;
522 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
524 return a + (long)b + c + d + (long)e + f;
527 struct bpf_fentry_test_t {
528 struct bpf_fentry_test_t *a;
531 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
536 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
541 __bpf_kfunc int bpf_modify_return_test(int a, int *b)
547 __bpf_kfunc u64 bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
549 return a + b + c + d;
552 __bpf_kfunc int bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
557 __bpf_kfunc struct sock *bpf_kfunc_call_test3(struct sock *sk)
562 long noinline bpf_kfunc_call_test4(signed char a, short b, int c, long d)
564 /* Provoke the compiler to assume that the caller has sign-extended a,
565 * b and c on platforms where this is required (e.g. s390x).
567 return (long)a + (long)b + (long)c + d;
570 struct prog_test_member1 {
574 struct prog_test_member {
575 struct prog_test_member1 m;
579 struct prog_test_ref_kfunc {
582 struct prog_test_member memb;
583 struct prog_test_ref_kfunc *next;
587 static struct prog_test_ref_kfunc prog_test_struct = {
590 .next = &prog_test_struct,
591 .cnt = REFCOUNT_INIT(1),
594 __bpf_kfunc struct prog_test_ref_kfunc *
595 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
597 refcount_inc(&prog_test_struct.cnt);
598 return &prog_test_struct;
601 __bpf_kfunc struct prog_test_member *
602 bpf_kfunc_call_memb_acquire(void)
608 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
613 refcount_dec(&p->cnt);
616 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
620 __bpf_kfunc void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
625 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size)
627 if (size > 2 * sizeof(int))
633 __bpf_kfunc int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p,
634 const int rdwr_buf_size)
636 return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size);
639 __bpf_kfunc int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p,
640 const int rdonly_buf_size)
642 return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
645 /* the next 2 ones can't be really used for testing expect to ensure
646 * that the verifier rejects the call.
647 * Acquire functions must return struct pointers, so these ones are
650 __bpf_kfunc int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p,
651 const int rdonly_buf_size)
653 return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
656 __bpf_kfunc void bpf_kfunc_call_int_mem_release(int *p)
660 __bpf_kfunc struct prog_test_ref_kfunc *
661 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
663 struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
667 refcount_inc(&p->cnt);
671 struct prog_test_pass1 {
684 struct prog_test_pass2 {
689 unsigned long arr3[8];
693 struct prog_test_fail1 {
698 struct prog_test_fail2 {
700 struct prog_test_pass1 x;
703 struct prog_test_fail3 {
709 __bpf_kfunc void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
713 __bpf_kfunc void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
717 __bpf_kfunc void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
721 __bpf_kfunc void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
725 __bpf_kfunc void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
729 __bpf_kfunc void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
733 __bpf_kfunc void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
737 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
741 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
745 __bpf_kfunc void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
749 __bpf_kfunc void bpf_kfunc_call_test_destructive(void)
753 __bpf_kfunc static u32 bpf_kfunc_call_test_static_unused_arg(u32 arg, u32 unused)
760 BTF_SET8_START(bpf_test_modify_return_ids)
761 BTF_ID_FLAGS(func, bpf_modify_return_test)
762 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
763 BTF_SET8_END(bpf_test_modify_return_ids)
765 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
766 .owner = THIS_MODULE,
767 .set = &bpf_test_modify_return_ids,
770 BTF_SET8_START(test_sk_check_kfunc_ids)
771 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
772 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
773 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
774 BTF_ID_FLAGS(func, bpf_kfunc_call_test4)
775 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
776 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
777 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
778 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
779 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
780 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL)
781 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL)
782 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL)
783 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE)
784 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
785 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
786 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
787 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
788 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
789 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
790 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
791 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
792 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
793 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
794 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS)
795 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
796 BTF_ID_FLAGS(func, bpf_kfunc_call_test_static_unused_arg)
797 BTF_SET8_END(test_sk_check_kfunc_ids)
799 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
800 u32 size, u32 headroom, u32 tailroom)
802 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
805 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
806 return ERR_PTR(-EINVAL);
808 if (user_size > size)
809 return ERR_PTR(-EMSGSIZE);
811 size = SKB_DATA_ALIGN(size);
812 data = kzalloc(size + headroom + tailroom, GFP_USER);
814 return ERR_PTR(-ENOMEM);
816 if (copy_from_user(data + headroom, data_in, user_size)) {
818 return ERR_PTR(-EFAULT);
824 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
825 const union bpf_attr *kattr,
826 union bpf_attr __user *uattr)
828 struct bpf_fentry_test_t arg = {};
829 u16 side_effect = 0, ret = 0;
830 int b = 2, err = -EFAULT;
833 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
836 switch (prog->expected_attach_type) {
837 case BPF_TRACE_FENTRY:
838 case BPF_TRACE_FEXIT:
839 if (bpf_fentry_test1(1) != 2 ||
840 bpf_fentry_test2(2, 3) != 5 ||
841 bpf_fentry_test3(4, 5, 6) != 15 ||
842 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
843 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
844 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
845 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
846 bpf_fentry_test8(&arg) != 0)
849 case BPF_MODIFY_RETURN:
850 ret = bpf_modify_return_test(1, &b);
858 retval = ((u32)side_effect << 16) | ret;
859 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
864 trace_bpf_test_finish(&err);
868 struct bpf_raw_tp_test_run_info {
869 struct bpf_prog *prog;
875 __bpf_prog_test_run_raw_tp(void *data)
877 struct bpf_raw_tp_test_run_info *info = data;
880 info->retval = bpf_prog_run(info->prog, info->ctx);
884 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
885 const union bpf_attr *kattr,
886 union bpf_attr __user *uattr)
888 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
889 __u32 ctx_size_in = kattr->test.ctx_size_in;
890 struct bpf_raw_tp_test_run_info info;
891 int cpu = kattr->test.cpu, err = 0;
894 /* doesn't support data_in/out, ctx_out, duration, or repeat */
895 if (kattr->test.data_in || kattr->test.data_out ||
896 kattr->test.ctx_out || kattr->test.duration ||
897 kattr->test.repeat || kattr->test.batch_size)
900 if (ctx_size_in < prog->aux->max_ctx_offset ||
901 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
904 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
908 info.ctx = memdup_user(ctx_in, ctx_size_in);
909 if (IS_ERR(info.ctx))
910 return PTR_ERR(info.ctx);
917 current_cpu = get_cpu();
918 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
919 cpu == current_cpu) {
920 __bpf_prog_test_run_raw_tp(&info);
921 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
922 /* smp_call_function_single() also checks cpu_online()
923 * after csd_lock(). However, since cpu is from user
924 * space, let's do an extra quick check to filter out
925 * invalid value before smp_call_function_single().
929 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
935 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
942 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
944 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
945 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
946 u32 size = kattr->test.ctx_size_in;
950 if (!data_in && !data_out)
953 data = kzalloc(max_size, GFP_USER);
955 return ERR_PTR(-ENOMEM);
958 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
964 size = min_t(u32, max_size, size);
965 if (copy_from_user(data, data_in, size)) {
967 return ERR_PTR(-EFAULT);
973 static int bpf_ctx_finish(const union bpf_attr *kattr,
974 union bpf_attr __user *uattr, const void *data,
977 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
979 u32 copy_size = size;
981 if (!data || !data_out)
984 if (copy_size > kattr->test.ctx_size_out) {
985 copy_size = kattr->test.ctx_size_out;
989 if (copy_to_user(data_out, data, copy_size))
991 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
1000 * range_is_zero - test whether buffer is initialized
1001 * @buf: buffer to check
1002 * @from: check from this position
1003 * @to: check up until (excluding) this position
1005 * This function returns true if the there is a non-zero byte
1006 * in the buf in the range [from,to).
1008 static inline bool range_is_zero(void *buf, size_t from, size_t to)
1010 return !memchr_inv((u8 *)buf + from, 0, to - from);
1013 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
1015 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1020 /* make sure the fields we don't use are zeroed */
1021 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
1024 /* mark is allowed */
1026 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
1027 offsetof(struct __sk_buff, priority)))
1030 /* priority is allowed */
1031 /* ingress_ifindex is allowed */
1032 /* ifindex is allowed */
1034 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1035 offsetof(struct __sk_buff, cb)))
1040 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1041 offsetof(struct __sk_buff, tstamp)))
1044 /* tstamp is allowed */
1045 /* wire_len is allowed */
1046 /* gso_segs is allowed */
1048 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1049 offsetof(struct __sk_buff, gso_size)))
1052 /* gso_size is allowed */
1054 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1055 offsetof(struct __sk_buff, hwtstamp)))
1058 /* hwtstamp is allowed */
1060 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1061 sizeof(struct __sk_buff)))
1064 skb->mark = __skb->mark;
1065 skb->priority = __skb->priority;
1066 skb->skb_iif = __skb->ingress_ifindex;
1067 skb->tstamp = __skb->tstamp;
1068 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1070 if (__skb->wire_len == 0) {
1071 cb->pkt_len = skb->len;
1073 if (__skb->wire_len < skb->len ||
1074 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1076 cb->pkt_len = __skb->wire_len;
1079 if (__skb->gso_segs > GSO_MAX_SEGS)
1081 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1082 skb_shinfo(skb)->gso_size = __skb->gso_size;
1083 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1088 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1090 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1095 __skb->mark = skb->mark;
1096 __skb->priority = skb->priority;
1097 __skb->ingress_ifindex = skb->skb_iif;
1098 __skb->ifindex = skb->dev->ifindex;
1099 __skb->tstamp = skb->tstamp;
1100 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1101 __skb->wire_len = cb->pkt_len;
1102 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
1103 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1106 static struct proto bpf_dummy_proto = {
1107 .name = "bpf_dummy",
1108 .owner = THIS_MODULE,
1109 .obj_size = sizeof(struct sock),
1112 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1113 union bpf_attr __user *uattr)
1115 bool is_l2 = false, is_direct_pkt_access = false;
1116 struct net *net = current->nsproxy->net_ns;
1117 struct net_device *dev = net->loopback_dev;
1118 u32 size = kattr->test.data_size_in;
1119 u32 repeat = kattr->test.repeat;
1120 struct __sk_buff *ctx = NULL;
1121 u32 retval, duration;
1122 int hh_len = ETH_HLEN;
1123 struct sk_buff *skb;
1128 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1131 data = bpf_test_init(kattr, kattr->test.data_size_in,
1132 size, NET_SKB_PAD + NET_IP_ALIGN,
1133 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1135 return PTR_ERR(data);
1137 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1140 return PTR_ERR(ctx);
1143 switch (prog->type) {
1144 case BPF_PROG_TYPE_SCHED_CLS:
1145 case BPF_PROG_TYPE_SCHED_ACT:
1148 case BPF_PROG_TYPE_LWT_IN:
1149 case BPF_PROG_TYPE_LWT_OUT:
1150 case BPF_PROG_TYPE_LWT_XMIT:
1151 is_direct_pkt_access = true;
1157 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1163 sock_init_data(NULL, sk);
1165 skb = slab_build_skb(data);
1174 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1175 __skb_put(skb, size);
1176 if (ctx && ctx->ifindex > 1) {
1177 dev = dev_get_by_index(net, ctx->ifindex);
1183 skb->protocol = eth_type_trans(skb, dev);
1184 skb_reset_network_header(skb);
1186 switch (skb->protocol) {
1187 case htons(ETH_P_IP):
1188 sk->sk_family = AF_INET;
1189 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1190 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1191 sk->sk_daddr = ip_hdr(skb)->daddr;
1194 #if IS_ENABLED(CONFIG_IPV6)
1195 case htons(ETH_P_IPV6):
1196 sk->sk_family = AF_INET6;
1197 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1198 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1199 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1208 __skb_push(skb, hh_len);
1209 if (is_direct_pkt_access)
1210 bpf_compute_data_pointers(skb);
1211 ret = convert___skb_to_skb(skb, ctx);
1214 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1218 if (skb_headroom(skb) < hh_len) {
1219 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1221 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1226 memset(__skb_push(skb, hh_len), 0, hh_len);
1228 convert_skb_to___skb(skb, ctx);
1231 /* bpf program can never convert linear skb to non-linear */
1232 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1233 size = skb_headlen(skb);
1234 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1237 ret = bpf_ctx_finish(kattr, uattr, ctx,
1238 sizeof(struct __sk_buff));
1240 if (dev && dev != net->loopback_dev)
1248 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1250 unsigned int ingress_ifindex, rx_queue_index;
1251 struct netdev_rx_queue *rxqueue;
1252 struct net_device *device;
1257 if (xdp_md->egress_ifindex != 0)
1260 ingress_ifindex = xdp_md->ingress_ifindex;
1261 rx_queue_index = xdp_md->rx_queue_index;
1263 if (!ingress_ifindex && rx_queue_index)
1266 if (ingress_ifindex) {
1267 device = dev_get_by_index(current->nsproxy->net_ns,
1272 if (rx_queue_index >= device->real_num_rx_queues)
1275 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1277 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1280 xdp->rxq = &rxqueue->xdp_rxq;
1281 /* The device is now tracked in the xdp->rxq for later
1286 xdp->data = xdp->data_meta + xdp_md->data;
1294 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1299 xdp_md->data = xdp->data - xdp->data_meta;
1300 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1302 if (xdp_md->ingress_ifindex)
1303 dev_put(xdp->rxq->dev);
1306 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1307 union bpf_attr __user *uattr)
1309 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1310 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1311 u32 batch_size = kattr->test.batch_size;
1312 u32 retval = 0, duration, max_data_sz;
1313 u32 size = kattr->test.data_size_in;
1314 u32 headroom = XDP_PACKET_HEADROOM;
1315 u32 repeat = kattr->test.repeat;
1316 struct netdev_rx_queue *rxqueue;
1317 struct skb_shared_info *sinfo;
1318 struct xdp_buff xdp = {};
1319 int i, ret = -EINVAL;
1323 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1324 prog->expected_attach_type == BPF_XDP_CPUMAP)
1327 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1330 if (bpf_prog_is_dev_bound(prog->aux))
1335 batch_size = NAPI_POLL_WEIGHT;
1336 else if (batch_size > TEST_XDP_MAX_BATCH)
1339 headroom += sizeof(struct xdp_page_head);
1340 } else if (batch_size) {
1344 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1346 return PTR_ERR(ctx);
1349 /* There can't be user provided data before the meta data */
1350 if (ctx->data_meta || ctx->data_end != size ||
1351 ctx->data > ctx->data_end ||
1352 unlikely(xdp_metalen_invalid(ctx->data)) ||
1353 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1355 /* Meta data is allocated from the headroom */
1356 headroom -= ctx->data;
1359 max_data_sz = 4096 - headroom - tailroom;
1360 if (size > max_data_sz) {
1361 /* disallow live data mode for jumbo frames */
1367 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1369 ret = PTR_ERR(data);
1373 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1374 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1375 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1376 xdp_prepare_buff(&xdp, data, headroom, size, true);
1377 sinfo = xdp_get_shared_info_from_buff(&xdp);
1379 ret = xdp_convert_md_to_buff(ctx, &xdp);
1383 if (unlikely(kattr->test.data_size_in > size)) {
1384 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1386 while (size < kattr->test.data_size_in) {
1391 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1396 page = alloc_page(GFP_KERNEL);
1402 frag = &sinfo->frags[sinfo->nr_frags++];
1403 __skb_frag_set_page(frag, page);
1405 data_len = min_t(u32, kattr->test.data_size_in - size,
1407 skb_frag_size_set(frag, data_len);
1409 if (copy_from_user(page_address(page), data_in + size,
1414 sinfo->xdp_frags_size += data_len;
1417 xdp_buff_set_frags_flag(&xdp);
1421 bpf_prog_change_xdp(NULL, prog);
1424 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1426 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1427 /* We convert the xdp_buff back to an xdp_md before checking the return
1428 * code so the reference count of any held netdevice will be decremented
1429 * even if the test run failed.
1431 xdp_convert_buff_to_md(&xdp, ctx);
1435 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1436 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1439 ret = bpf_ctx_finish(kattr, uattr, ctx,
1440 sizeof(struct xdp_md));
1444 bpf_prog_change_xdp(prog, NULL);
1446 for (i = 0; i < sinfo->nr_frags; i++)
1447 __free_page(skb_frag_page(&sinfo->frags[i]));
1454 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1456 /* make sure the fields we don't use are zeroed */
1457 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1460 /* flags is allowed */
1462 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1463 sizeof(struct bpf_flow_keys)))
1469 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1470 const union bpf_attr *kattr,
1471 union bpf_attr __user *uattr)
1473 struct bpf_test_timer t = { NO_PREEMPT };
1474 u32 size = kattr->test.data_size_in;
1475 struct bpf_flow_dissector ctx = {};
1476 u32 repeat = kattr->test.repeat;
1477 struct bpf_flow_keys *user_ctx;
1478 struct bpf_flow_keys flow_keys;
1479 const struct ethhdr *eth;
1480 unsigned int flags = 0;
1481 u32 retval, duration;
1485 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1488 if (size < ETH_HLEN)
1491 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1493 return PTR_ERR(data);
1495 eth = (struct ethhdr *)data;
1500 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1501 if (IS_ERR(user_ctx)) {
1503 return PTR_ERR(user_ctx);
1506 ret = verify_user_bpf_flow_keys(user_ctx);
1509 flags = user_ctx->flags;
1512 ctx.flow_keys = &flow_keys;
1514 ctx.data_end = (__u8 *)data + size;
1516 bpf_test_timer_enter(&t);
1518 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1520 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1521 bpf_test_timer_leave(&t);
1526 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1527 sizeof(flow_keys), retval, duration);
1529 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1530 sizeof(struct bpf_flow_keys));
1538 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1539 union bpf_attr __user *uattr)
1541 struct bpf_test_timer t = { NO_PREEMPT };
1542 struct bpf_prog_array *progs = NULL;
1543 struct bpf_sk_lookup_kern ctx = {};
1544 u32 repeat = kattr->test.repeat;
1545 struct bpf_sk_lookup *user_ctx;
1546 u32 retval, duration;
1549 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1552 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1553 kattr->test.data_size_out)
1559 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1560 if (IS_ERR(user_ctx))
1561 return PTR_ERR(user_ctx);
1569 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1572 if (user_ctx->local_port > U16_MAX) {
1577 ctx.family = (u16)user_ctx->family;
1578 ctx.protocol = (u16)user_ctx->protocol;
1579 ctx.dport = (u16)user_ctx->local_port;
1580 ctx.sport = user_ctx->remote_port;
1582 switch (ctx.family) {
1584 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1585 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1588 #if IS_ENABLED(CONFIG_IPV6)
1590 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1591 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1596 ret = -EAFNOSUPPORT;
1600 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1606 progs->items[0].prog = prog;
1608 bpf_test_timer_enter(&t);
1610 ctx.selected_sk = NULL;
1611 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1612 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1613 bpf_test_timer_leave(&t);
1618 user_ctx->cookie = 0;
1619 if (ctx.selected_sk) {
1620 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1625 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1628 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1630 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1633 bpf_prog_array_free(progs);
1638 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1639 const union bpf_attr *kattr,
1640 union bpf_attr __user *uattr)
1642 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1643 __u32 ctx_size_in = kattr->test.ctx_size_in;
1648 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1649 if (kattr->test.data_in || kattr->test.data_out ||
1650 kattr->test.ctx_out || kattr->test.duration ||
1651 kattr->test.repeat || kattr->test.flags ||
1652 kattr->test.batch_size)
1655 if (ctx_size_in < prog->aux->max_ctx_offset ||
1656 ctx_size_in > U16_MAX)
1660 ctx = memdup_user(ctx_in, ctx_size_in);
1662 return PTR_ERR(ctx);
1665 rcu_read_lock_trace();
1666 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1667 rcu_read_unlock_trace();
1669 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1674 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1681 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1682 .owner = THIS_MODULE,
1683 .set = &test_sk_check_kfunc_ids,
1686 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1687 BTF_ID(struct, prog_test_ref_kfunc)
1688 BTF_ID(func, bpf_kfunc_call_test_release)
1689 BTF_ID(struct, prog_test_member)
1690 BTF_ID(func, bpf_kfunc_call_memb_release)
1692 static int __init bpf_prog_test_run_init(void)
1694 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1696 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1697 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1700 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1701 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1706 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1707 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1708 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1709 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1710 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1711 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1714 late_initcall(bpf_prog_test_run_init);