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;
100 struct xdp_frame frm;
104 struct xdp_test_data {
105 struct xdp_buff *orig_ctx;
106 struct xdp_rxq_info rxq;
107 struct net_device *dev;
108 struct page_pool *pp;
109 struct xdp_frame **frames;
110 struct sk_buff **skbs;
111 struct xdp_mem_info mem;
116 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
117 #define TEST_XDP_MAX_BATCH 256
119 static void xdp_test_run_init_page(struct page *page, void *arg)
121 struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
122 struct xdp_buff *new_ctx, *orig_ctx;
123 u32 headroom = XDP_PACKET_HEADROOM;
124 struct xdp_test_data *xdp = arg;
125 size_t frm_len, meta_len;
126 struct xdp_frame *frm;
129 orig_ctx = xdp->orig_ctx;
130 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
131 meta_len = orig_ctx->data - orig_ctx->data_meta;
132 headroom -= meta_len;
134 new_ctx = &head->ctx;
137 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
139 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
140 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
141 new_ctx->data = new_ctx->data_meta + meta_len;
143 xdp_update_frame_from_buff(new_ctx, frm);
144 frm->mem = new_ctx->rxq->mem;
146 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
149 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
151 struct page_pool *pp;
153 struct page_pool_params pp_params = {
156 .pool_size = xdp->batch_size,
158 .init_callback = xdp_test_run_init_page,
162 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
166 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
170 pp = page_pool_create(&pp_params);
176 /* will copy 'mem.id' into pp->xdp_mem_id */
177 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
183 /* We create a 'fake' RXQ referencing the original dev, but with an
184 * xdp_mem_info pointing to our page_pool
186 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
187 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
188 xdp->rxq.mem.id = pp->xdp_mem_id;
189 xdp->dev = orig_ctx->rxq->dev;
190 xdp->orig_ctx = orig_ctx;
195 page_pool_destroy(pp);
203 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
205 xdp_unreg_mem_model(&xdp->mem);
206 page_pool_destroy(xdp->pp);
211 static bool ctx_was_changed(struct xdp_page_head *head)
213 return head->orig_ctx.data != head->ctx.data ||
214 head->orig_ctx.data_meta != head->ctx.data_meta ||
215 head->orig_ctx.data_end != head->ctx.data_end;
218 static void reset_ctx(struct xdp_page_head *head)
220 if (likely(!ctx_was_changed(head)))
223 head->ctx.data = head->orig_ctx.data;
224 head->ctx.data_meta = head->orig_ctx.data_meta;
225 head->ctx.data_end = head->orig_ctx.data_end;
226 xdp_update_frame_from_buff(&head->ctx, &head->frm);
229 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
230 struct sk_buff **skbs,
231 struct net_device *dev)
233 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
237 n = kmem_cache_alloc_bulk(skbuff_head_cache, gfp, nframes, (void **)skbs);
238 if (unlikely(n == 0)) {
239 for (i = 0; i < nframes; i++)
240 xdp_return_frame(frames[i]);
244 for (i = 0; i < nframes; i++) {
245 struct xdp_frame *xdpf = frames[i];
246 struct sk_buff *skb = skbs[i];
248 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
250 xdp_return_frame(xdpf);
254 list_add_tail(&skb->list, &list);
256 netif_receive_skb_list(&list);
261 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
264 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
265 int err = 0, act, ret, i, nframes = 0, batch_sz;
266 struct xdp_frame **frames = xdp->frames;
267 struct xdp_page_head *head;
268 struct xdp_frame *frm;
269 bool redirect = false;
270 struct xdp_buff *ctx;
273 batch_sz = min_t(u32, repeat, xdp->batch_size);
276 xdp_set_return_frame_no_direct();
278 for (i = 0; i < batch_sz; i++) {
279 page = page_pool_dev_alloc_pages(xdp->pp);
285 head = phys_to_virt(page_to_phys(page));
291 act = bpf_prog_run_xdp(prog, ctx);
293 /* if program changed pkt bounds we need to update the xdp_frame */
294 if (unlikely(ctx_was_changed(head))) {
295 ret = xdp_update_frame_from_buff(ctx, frm);
297 xdp_return_buff(ctx);
304 /* we can't do a real XDP_TX since we're not in the
305 * driver, so turn it into a REDIRECT back to the same
308 ri->tgt_index = xdp->dev->ifindex;
309 ri->map_id = INT_MAX;
310 ri->map_type = BPF_MAP_TYPE_UNSPEC;
314 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
316 xdp_return_buff(ctx);
319 frames[nframes++] = frm;
322 bpf_warn_invalid_xdp_action(NULL, prog, act);
325 xdp_return_buff(ctx);
334 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
339 xdp_clear_return_frame_no_direct();
344 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
345 u32 repeat, u32 batch_size, u32 *time)
348 struct xdp_test_data xdp = { .batch_size = batch_size };
349 struct bpf_test_timer t = { .mode = NO_MIGRATE };
355 ret = xdp_test_run_setup(&xdp, ctx);
359 bpf_test_timer_enter(&t);
362 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
363 if (unlikely(ret < 0))
365 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
366 bpf_test_timer_leave(&t);
368 xdp_test_run_teardown(&xdp);
372 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
373 u32 *retval, u32 *time, bool xdp)
375 struct bpf_prog_array_item item = {.prog = prog};
376 struct bpf_run_ctx *old_ctx;
377 struct bpf_cg_run_ctx run_ctx;
378 struct bpf_test_timer t = { NO_MIGRATE };
379 enum bpf_cgroup_storage_type stype;
382 for_each_cgroup_storage_type(stype) {
383 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
384 if (IS_ERR(item.cgroup_storage[stype])) {
385 item.cgroup_storage[stype] = NULL;
386 for_each_cgroup_storage_type(stype)
387 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
395 bpf_test_timer_enter(&t);
396 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
398 run_ctx.prog_item = &item;
400 *retval = bpf_prog_run_xdp(prog, ctx);
402 *retval = bpf_prog_run(prog, ctx);
403 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
404 bpf_reset_run_ctx(old_ctx);
405 bpf_test_timer_leave(&t);
407 for_each_cgroup_storage_type(stype)
408 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
413 static int bpf_test_finish(const union bpf_attr *kattr,
414 union bpf_attr __user *uattr, const void *data,
415 struct skb_shared_info *sinfo, u32 size,
416 u32 retval, u32 duration)
418 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
420 u32 copy_size = size;
422 /* Clamp copy if the user has provided a size hint, but copy the full
423 * buffer if not to retain old behaviour.
425 if (kattr->test.data_size_out &&
426 copy_size > kattr->test.data_size_out) {
427 copy_size = kattr->test.data_size_out;
432 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
439 if (copy_to_user(data_out, data, len))
446 for (i = 0; i < sinfo->nr_frags; i++) {
447 skb_frag_t *frag = &sinfo->frags[i];
449 if (offset >= copy_size) {
454 data_len = min_t(u32, copy_size - offset,
455 skb_frag_size(frag));
457 if (copy_to_user(data_out + offset,
458 skb_frag_address(frag),
467 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
469 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
471 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
476 trace_bpf_test_finish(&err);
480 /* Integer types of various sizes and pointer combinations cover variety of
481 * architecture dependent calling conventions. 7+ can be supported in the
485 __diag_ignore_all("-Wmissing-prototypes",
486 "Global functions as their definitions will be in vmlinux BTF");
487 int noinline bpf_fentry_test1(int a)
491 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
492 ALLOW_ERROR_INJECTION(bpf_fentry_test1, ERRNO);
494 int noinline bpf_fentry_test2(int a, u64 b)
499 int noinline bpf_fentry_test3(char a, int b, u64 c)
504 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
506 return (long)a + b + c + d;
509 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
511 return a + (long)b + c + d + e;
514 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
516 return a + (long)b + c + d + (long)e + f;
519 struct bpf_fentry_test_t {
520 struct bpf_fentry_test_t *a;
523 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
528 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
533 int noinline bpf_modify_return_test(int a, int *b)
539 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
541 return a + b + c + d;
544 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
549 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
554 struct prog_test_member1 {
558 struct prog_test_member {
559 struct prog_test_member1 m;
563 struct prog_test_ref_kfunc {
566 struct prog_test_member memb;
567 struct prog_test_ref_kfunc *next;
571 static struct prog_test_ref_kfunc prog_test_struct = {
574 .next = &prog_test_struct,
575 .cnt = REFCOUNT_INIT(1),
578 noinline struct prog_test_ref_kfunc *
579 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
581 refcount_inc(&prog_test_struct.cnt);
582 return &prog_test_struct;
585 noinline struct prog_test_member *
586 bpf_kfunc_call_memb_acquire(void)
592 noinline void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
597 refcount_dec(&p->cnt);
600 noinline void bpf_kfunc_call_memb_release(struct prog_test_member *p)
604 noinline void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
609 noinline struct prog_test_ref_kfunc *
610 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
612 struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
616 refcount_inc(&p->cnt);
620 struct prog_test_pass1 {
633 struct prog_test_pass2 {
638 unsigned long arr3[8];
642 struct prog_test_fail1 {
647 struct prog_test_fail2 {
649 struct prog_test_pass1 x;
652 struct prog_test_fail3 {
658 noinline void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
662 noinline void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
666 noinline void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
670 noinline void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
674 noinline void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
678 noinline void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
682 noinline void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
686 noinline void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
690 noinline void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
694 noinline void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
700 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
702 BTF_SET8_START(test_sk_check_kfunc_ids)
703 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
704 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
705 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
706 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
707 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
708 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
709 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
710 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
711 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
712 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
713 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
714 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
715 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
716 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
717 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
718 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
719 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
720 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
721 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS)
722 BTF_SET8_END(test_sk_check_kfunc_ids)
724 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
725 u32 size, u32 headroom, u32 tailroom)
727 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
730 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
731 return ERR_PTR(-EINVAL);
733 if (user_size > size)
734 return ERR_PTR(-EMSGSIZE);
736 data = kzalloc(size + headroom + tailroom, GFP_USER);
738 return ERR_PTR(-ENOMEM);
740 if (copy_from_user(data + headroom, data_in, user_size)) {
742 return ERR_PTR(-EFAULT);
748 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
749 const union bpf_attr *kattr,
750 union bpf_attr __user *uattr)
752 struct bpf_fentry_test_t arg = {};
753 u16 side_effect = 0, ret = 0;
754 int b = 2, err = -EFAULT;
757 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
760 switch (prog->expected_attach_type) {
761 case BPF_TRACE_FENTRY:
762 case BPF_TRACE_FEXIT:
763 if (bpf_fentry_test1(1) != 2 ||
764 bpf_fentry_test2(2, 3) != 5 ||
765 bpf_fentry_test3(4, 5, 6) != 15 ||
766 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
767 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
768 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
769 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
770 bpf_fentry_test8(&arg) != 0)
773 case BPF_MODIFY_RETURN:
774 ret = bpf_modify_return_test(1, &b);
782 retval = ((u32)side_effect << 16) | ret;
783 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
788 trace_bpf_test_finish(&err);
792 struct bpf_raw_tp_test_run_info {
793 struct bpf_prog *prog;
799 __bpf_prog_test_run_raw_tp(void *data)
801 struct bpf_raw_tp_test_run_info *info = data;
804 info->retval = bpf_prog_run(info->prog, info->ctx);
808 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
809 const union bpf_attr *kattr,
810 union bpf_attr __user *uattr)
812 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
813 __u32 ctx_size_in = kattr->test.ctx_size_in;
814 struct bpf_raw_tp_test_run_info info;
815 int cpu = kattr->test.cpu, err = 0;
818 /* doesn't support data_in/out, ctx_out, duration, or repeat */
819 if (kattr->test.data_in || kattr->test.data_out ||
820 kattr->test.ctx_out || kattr->test.duration ||
821 kattr->test.repeat || kattr->test.batch_size)
824 if (ctx_size_in < prog->aux->max_ctx_offset ||
825 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
828 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
832 info.ctx = memdup_user(ctx_in, ctx_size_in);
833 if (IS_ERR(info.ctx))
834 return PTR_ERR(info.ctx);
841 current_cpu = get_cpu();
842 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
843 cpu == current_cpu) {
844 __bpf_prog_test_run_raw_tp(&info);
845 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
846 /* smp_call_function_single() also checks cpu_online()
847 * after csd_lock(). However, since cpu is from user
848 * space, let's do an extra quick check to filter out
849 * invalid value before smp_call_function_single().
853 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
859 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
866 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
868 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
869 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
870 u32 size = kattr->test.ctx_size_in;
874 if (!data_in && !data_out)
877 data = kzalloc(max_size, GFP_USER);
879 return ERR_PTR(-ENOMEM);
882 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
888 size = min_t(u32, max_size, size);
889 if (copy_from_user(data, data_in, size)) {
891 return ERR_PTR(-EFAULT);
897 static int bpf_ctx_finish(const union bpf_attr *kattr,
898 union bpf_attr __user *uattr, const void *data,
901 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
903 u32 copy_size = size;
905 if (!data || !data_out)
908 if (copy_size > kattr->test.ctx_size_out) {
909 copy_size = kattr->test.ctx_size_out;
913 if (copy_to_user(data_out, data, copy_size))
915 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
924 * range_is_zero - test whether buffer is initialized
925 * @buf: buffer to check
926 * @from: check from this position
927 * @to: check up until (excluding) this position
929 * This function returns true if the there is a non-zero byte
930 * in the buf in the range [from,to).
932 static inline bool range_is_zero(void *buf, size_t from, size_t to)
934 return !memchr_inv((u8 *)buf + from, 0, to - from);
937 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
939 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
947 /* make sure the fields we don't use are zeroed */
948 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
951 /* mark is allowed */
953 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
954 offsetof(struct __sk_buff, priority)))
957 /* priority is allowed */
958 /* ingress_ifindex is allowed */
959 /* ifindex is allowed */
961 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
962 offsetof(struct __sk_buff, cb)))
967 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
968 offsetof(struct __sk_buff, tstamp)))
971 /* tstamp is allowed */
972 /* wire_len is allowed */
973 /* gso_segs is allowed */
975 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
976 offsetof(struct __sk_buff, gso_size)))
979 /* gso_size is allowed */
981 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
982 offsetof(struct __sk_buff, hwtstamp)))
985 /* hwtstamp is allowed */
987 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
988 sizeof(struct __sk_buff)))
991 skb->mark = __skb->mark;
992 skb->priority = __skb->priority;
993 skb->skb_iif = __skb->ingress_ifindex;
994 skb->tstamp = __skb->tstamp;
995 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
997 if (__skb->wire_len == 0) {
998 cb->pkt_len = skb->len;
1000 if (__skb->wire_len < skb->len ||
1001 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1003 cb->pkt_len = __skb->wire_len;
1006 if (__skb->gso_segs > GSO_MAX_SEGS)
1008 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1009 skb_shinfo(skb)->gso_size = __skb->gso_size;
1010 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1015 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1017 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1022 __skb->mark = skb->mark;
1023 __skb->priority = skb->priority;
1024 __skb->ingress_ifindex = skb->skb_iif;
1025 __skb->ifindex = skb->dev->ifindex;
1026 __skb->tstamp = skb->tstamp;
1027 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1028 __skb->wire_len = cb->pkt_len;
1029 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
1030 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1033 static struct proto bpf_dummy_proto = {
1034 .name = "bpf_dummy",
1035 .owner = THIS_MODULE,
1036 .obj_size = sizeof(struct sock),
1039 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1040 union bpf_attr __user *uattr)
1042 bool is_l2 = false, is_direct_pkt_access = false;
1043 struct net *net = current->nsproxy->net_ns;
1044 struct net_device *dev = net->loopback_dev;
1045 u32 size = kattr->test.data_size_in;
1046 u32 repeat = kattr->test.repeat;
1047 struct __sk_buff *ctx = NULL;
1048 u32 retval, duration;
1049 int hh_len = ETH_HLEN;
1050 struct sk_buff *skb;
1055 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1058 data = bpf_test_init(kattr, kattr->test.data_size_in,
1059 size, NET_SKB_PAD + NET_IP_ALIGN,
1060 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1062 return PTR_ERR(data);
1064 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1067 return PTR_ERR(ctx);
1070 switch (prog->type) {
1071 case BPF_PROG_TYPE_SCHED_CLS:
1072 case BPF_PROG_TYPE_SCHED_ACT:
1075 case BPF_PROG_TYPE_LWT_IN:
1076 case BPF_PROG_TYPE_LWT_OUT:
1077 case BPF_PROG_TYPE_LWT_XMIT:
1078 is_direct_pkt_access = true;
1084 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1090 sock_init_data(NULL, sk);
1092 skb = build_skb(data, 0);
1101 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1102 __skb_put(skb, size);
1103 if (ctx && ctx->ifindex > 1) {
1104 dev = dev_get_by_index(net, ctx->ifindex);
1110 skb->protocol = eth_type_trans(skb, dev);
1111 skb_reset_network_header(skb);
1113 switch (skb->protocol) {
1114 case htons(ETH_P_IP):
1115 sk->sk_family = AF_INET;
1116 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1117 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1118 sk->sk_daddr = ip_hdr(skb)->daddr;
1121 #if IS_ENABLED(CONFIG_IPV6)
1122 case htons(ETH_P_IPV6):
1123 sk->sk_family = AF_INET6;
1124 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1125 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1126 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1135 __skb_push(skb, hh_len);
1136 if (is_direct_pkt_access)
1137 bpf_compute_data_pointers(skb);
1138 ret = convert___skb_to_skb(skb, ctx);
1141 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1145 if (skb_headroom(skb) < hh_len) {
1146 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1148 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1153 memset(__skb_push(skb, hh_len), 0, hh_len);
1155 convert_skb_to___skb(skb, ctx);
1158 /* bpf program can never convert linear skb to non-linear */
1159 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1160 size = skb_headlen(skb);
1161 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1164 ret = bpf_ctx_finish(kattr, uattr, ctx,
1165 sizeof(struct __sk_buff));
1167 if (dev && dev != net->loopback_dev)
1175 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1177 unsigned int ingress_ifindex, rx_queue_index;
1178 struct netdev_rx_queue *rxqueue;
1179 struct net_device *device;
1184 if (xdp_md->egress_ifindex != 0)
1187 ingress_ifindex = xdp_md->ingress_ifindex;
1188 rx_queue_index = xdp_md->rx_queue_index;
1190 if (!ingress_ifindex && rx_queue_index)
1193 if (ingress_ifindex) {
1194 device = dev_get_by_index(current->nsproxy->net_ns,
1199 if (rx_queue_index >= device->real_num_rx_queues)
1202 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1204 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1207 xdp->rxq = &rxqueue->xdp_rxq;
1208 /* The device is now tracked in the xdp->rxq for later
1213 xdp->data = xdp->data_meta + xdp_md->data;
1221 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1226 xdp_md->data = xdp->data - xdp->data_meta;
1227 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1229 if (xdp_md->ingress_ifindex)
1230 dev_put(xdp->rxq->dev);
1233 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1234 union bpf_attr __user *uattr)
1236 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1237 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1238 u32 batch_size = kattr->test.batch_size;
1239 u32 retval = 0, duration, max_data_sz;
1240 u32 size = kattr->test.data_size_in;
1241 u32 headroom = XDP_PACKET_HEADROOM;
1242 u32 repeat = kattr->test.repeat;
1243 struct netdev_rx_queue *rxqueue;
1244 struct skb_shared_info *sinfo;
1245 struct xdp_buff xdp = {};
1246 int i, ret = -EINVAL;
1250 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1251 prog->expected_attach_type == BPF_XDP_CPUMAP)
1254 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1259 batch_size = NAPI_POLL_WEIGHT;
1260 else if (batch_size > TEST_XDP_MAX_BATCH)
1263 headroom += sizeof(struct xdp_page_head);
1264 } else if (batch_size) {
1268 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1270 return PTR_ERR(ctx);
1273 /* There can't be user provided data before the meta data */
1274 if (ctx->data_meta || ctx->data_end != size ||
1275 ctx->data > ctx->data_end ||
1276 unlikely(xdp_metalen_invalid(ctx->data)) ||
1277 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1279 /* Meta data is allocated from the headroom */
1280 headroom -= ctx->data;
1283 max_data_sz = 4096 - headroom - tailroom;
1284 if (size > max_data_sz) {
1285 /* disallow live data mode for jumbo frames */
1291 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1293 ret = PTR_ERR(data);
1297 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1298 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1299 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1300 xdp_prepare_buff(&xdp, data, headroom, size, true);
1301 sinfo = xdp_get_shared_info_from_buff(&xdp);
1303 ret = xdp_convert_md_to_buff(ctx, &xdp);
1307 if (unlikely(kattr->test.data_size_in > size)) {
1308 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1310 while (size < kattr->test.data_size_in) {
1315 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1320 page = alloc_page(GFP_KERNEL);
1326 frag = &sinfo->frags[sinfo->nr_frags++];
1327 __skb_frag_set_page(frag, page);
1329 data_len = min_t(u32, kattr->test.data_size_in - size,
1331 skb_frag_size_set(frag, data_len);
1333 if (copy_from_user(page_address(page), data_in + size,
1338 sinfo->xdp_frags_size += data_len;
1341 xdp_buff_set_frags_flag(&xdp);
1345 bpf_prog_change_xdp(NULL, prog);
1348 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1350 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1351 /* We convert the xdp_buff back to an xdp_md before checking the return
1352 * code so the reference count of any held netdevice will be decremented
1353 * even if the test run failed.
1355 xdp_convert_buff_to_md(&xdp, ctx);
1359 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1360 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1363 ret = bpf_ctx_finish(kattr, uattr, ctx,
1364 sizeof(struct xdp_md));
1368 bpf_prog_change_xdp(prog, NULL);
1370 for (i = 0; i < sinfo->nr_frags; i++)
1371 __free_page(skb_frag_page(&sinfo->frags[i]));
1378 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1380 /* make sure the fields we don't use are zeroed */
1381 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1384 /* flags is allowed */
1386 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1387 sizeof(struct bpf_flow_keys)))
1393 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1394 const union bpf_attr *kattr,
1395 union bpf_attr __user *uattr)
1397 struct bpf_test_timer t = { NO_PREEMPT };
1398 u32 size = kattr->test.data_size_in;
1399 struct bpf_flow_dissector ctx = {};
1400 u32 repeat = kattr->test.repeat;
1401 struct bpf_flow_keys *user_ctx;
1402 struct bpf_flow_keys flow_keys;
1403 const struct ethhdr *eth;
1404 unsigned int flags = 0;
1405 u32 retval, duration;
1409 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1412 if (size < ETH_HLEN)
1415 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1417 return PTR_ERR(data);
1419 eth = (struct ethhdr *)data;
1424 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1425 if (IS_ERR(user_ctx)) {
1427 return PTR_ERR(user_ctx);
1430 ret = verify_user_bpf_flow_keys(user_ctx);
1433 flags = user_ctx->flags;
1436 ctx.flow_keys = &flow_keys;
1438 ctx.data_end = (__u8 *)data + size;
1440 bpf_test_timer_enter(&t);
1442 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1444 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1445 bpf_test_timer_leave(&t);
1450 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1451 sizeof(flow_keys), retval, duration);
1453 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1454 sizeof(struct bpf_flow_keys));
1462 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1463 union bpf_attr __user *uattr)
1465 struct bpf_test_timer t = { NO_PREEMPT };
1466 struct bpf_prog_array *progs = NULL;
1467 struct bpf_sk_lookup_kern ctx = {};
1468 u32 repeat = kattr->test.repeat;
1469 struct bpf_sk_lookup *user_ctx;
1470 u32 retval, duration;
1473 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1476 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1477 kattr->test.data_size_out)
1483 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1484 if (IS_ERR(user_ctx))
1485 return PTR_ERR(user_ctx);
1493 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1496 if (user_ctx->local_port > U16_MAX) {
1501 ctx.family = (u16)user_ctx->family;
1502 ctx.protocol = (u16)user_ctx->protocol;
1503 ctx.dport = (u16)user_ctx->local_port;
1504 ctx.sport = user_ctx->remote_port;
1506 switch (ctx.family) {
1508 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1509 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1512 #if IS_ENABLED(CONFIG_IPV6)
1514 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1515 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1520 ret = -EAFNOSUPPORT;
1524 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1530 progs->items[0].prog = prog;
1532 bpf_test_timer_enter(&t);
1534 ctx.selected_sk = NULL;
1535 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1536 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1537 bpf_test_timer_leave(&t);
1542 user_ctx->cookie = 0;
1543 if (ctx.selected_sk) {
1544 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1549 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1552 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1554 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1557 bpf_prog_array_free(progs);
1562 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1563 const union bpf_attr *kattr,
1564 union bpf_attr __user *uattr)
1566 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1567 __u32 ctx_size_in = kattr->test.ctx_size_in;
1572 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1573 if (kattr->test.data_in || kattr->test.data_out ||
1574 kattr->test.ctx_out || kattr->test.duration ||
1575 kattr->test.repeat || kattr->test.flags ||
1576 kattr->test.batch_size)
1579 if (ctx_size_in < prog->aux->max_ctx_offset ||
1580 ctx_size_in > U16_MAX)
1584 ctx = memdup_user(ctx_in, ctx_size_in);
1586 return PTR_ERR(ctx);
1589 rcu_read_lock_trace();
1590 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1591 rcu_read_unlock_trace();
1593 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1598 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1605 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1606 .owner = THIS_MODULE,
1607 .set = &test_sk_check_kfunc_ids,
1610 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1611 BTF_ID(struct, prog_test_ref_kfunc)
1612 BTF_ID(func, bpf_kfunc_call_test_release)
1613 BTF_ID(struct, prog_test_member)
1614 BTF_ID(func, bpf_kfunc_call_memb_release)
1616 static int __init bpf_prog_test_run_init(void)
1618 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1620 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1621 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1624 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1625 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1630 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1631 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1632 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1633 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1636 late_initcall(bpf_prog_test_run_init);