projects/linux·blob master · refs

1910 lines · 44.6 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2017 Facebook
3 */
4#include <linux/bpf.h>
5#include <linux/btf.h>
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>
16#include <net/sock.h>
17#include <net/tcp.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>
25#include <net/xdp.h>
26#include <net/netfilter/nf_bpf_link.h>
27
28#define CREATE_TRACE_POINTS
29#include <trace/events/bpf_test_run.h>
30
31struct bpf_test_timer {
32 u32 i;
33 u64 time_start, time_spent;
34};
35
36static void bpf_test_timer_enter(struct bpf_test_timer *t)
37 __acquires(rcu)
38{
39 rcu_read_lock_dont_migrate();
40 t->time_start = ktime_get_ns();
41}
42
43static void bpf_test_timer_leave(struct bpf_test_timer *t)
44 __releases(rcu)
45{
46 t->time_start = 0;
47 rcu_read_unlock_migrate();
48}
49
50static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
51 u32 repeat, int *err, u32 *duration)
52 __must_hold(rcu)
53{
54 t->i += iterations;
55 if (t->i >= repeat) {
56 /* We're done. */
57 t->time_spent += ktime_get_ns() - t->time_start;
58 do_div(t->time_spent, t->i);
59 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
60 *err = 0;
61 goto reset;
62 }
63
64 if (signal_pending(current)) {
65 /* During iteration: we've been cancelled, abort. */
66 *err = -EINTR;
67 goto reset;
68 }
69
70 if (need_resched()) {
71 /* During iteration: we need to reschedule between runs. */
72 t->time_spent += ktime_get_ns() - t->time_start;
73 bpf_test_timer_leave(t);
74 cond_resched();
75 bpf_test_timer_enter(t);
76 }
77
78 /* Do another round. */
79 return true;
80
81reset:
82 t->i = 0;
83 return false;
84}
85
86/* We put this struct at the head of each page with a context and frame
87 * initialised when the page is allocated, so we don't have to do this on each
88 * repetition of the test run.
89 */
90struct xdp_page_head {
91 struct xdp_buff orig_ctx;
92 struct xdp_buff ctx;
93 union {
94 /* ::data_hard_start starts here */
95 DECLARE_FLEX_ARRAY(struct xdp_frame, frame);
96 DECLARE_FLEX_ARRAY(u8, data);
97 };
98};
99
100struct xdp_test_data {
101 struct xdp_buff *orig_ctx;
102 struct xdp_rxq_info rxq;
103 struct net_device *dev;
104 struct page_pool *pp;
105 struct xdp_frame **frames;
106 struct sk_buff **skbs;
107 struct xdp_mem_info mem;
108 u32 batch_size;
109 u32 frame_cnt;
110};
111
112/* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE
113 * must be updated accordingly this gets changed, otherwise BPF selftests
114 * will fail.
115 */
116#define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
117#define TEST_XDP_MAX_BATCH 256
118
119static void xdp_test_run_init_page(netmem_ref netmem, void *arg)
120{
121 struct xdp_page_head *head =
122 phys_to_virt(page_to_phys(netmem_to_page(netmem)));
123 struct xdp_buff *new_ctx, *orig_ctx;
124 u32 headroom = XDP_PACKET_HEADROOM;
125 struct xdp_test_data *xdp = arg;
126 size_t frm_len, meta_len;
127 struct xdp_frame *frm;
128 void *data;
129
130 orig_ctx = xdp->orig_ctx;
131 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
132 meta_len = orig_ctx->data - orig_ctx->data_meta;
133 headroom -= meta_len;
134
135 new_ctx = &head->ctx;
136 frm = head->frame;
137 data = head->data;
138 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
139
140 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
141 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
142 new_ctx->data = new_ctx->data_meta + meta_len;
143
144 xdp_update_frame_from_buff(new_ctx, frm);
145 frm->mem_type = new_ctx->rxq->mem.type;
146
147 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
148}
149
150static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
151{
152 struct page_pool *pp;
153 int err = -ENOMEM;
154 struct page_pool_params pp_params = {
155 .order = 0,
156 .flags = 0,
157 .pool_size = xdp->batch_size,
158 .nid = NUMA_NO_NODE,
159 .init_callback = xdp_test_run_init_page,
160 .init_arg = xdp,
161 };
162
163 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
164 if (!xdp->frames)
165 return -ENOMEM;
166
167 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
168 if (!xdp->skbs)
169 goto err_skbs;
170
171 pp = page_pool_create(&pp_params);
172 if (IS_ERR(pp)) {
173 err = PTR_ERR(pp);
174 goto err_pp;
175 }
176
177 /* will copy 'mem.id' into pp->xdp_mem_id */
178 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
179 if (err)
180 goto err_mmodel;
181
182 xdp->pp = pp;
183
184 /* We create a 'fake' RXQ referencing the original dev, but with an
185 * xdp_mem_info pointing to our page_pool
186 */
187 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
188 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
189 xdp->rxq.mem.id = pp->xdp_mem_id;
190 xdp->dev = orig_ctx->rxq->dev;
191 xdp->orig_ctx = orig_ctx;
192
193 return 0;
194
195err_mmodel:
196 page_pool_destroy(pp);
197err_pp:
198 kvfree(xdp->skbs);
199err_skbs:
200 kvfree(xdp->frames);
201 return err;
202}
203
204static void xdp_test_run_teardown(struct xdp_test_data *xdp)
205{
206 xdp_unreg_mem_model(&xdp->mem);
207 page_pool_destroy(xdp->pp);
208 kvfree(xdp->frames);
209 kvfree(xdp->skbs);
210}
211
212static bool frame_was_changed(const struct xdp_page_head *head)
213{
214 /* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
215 * i.e. has the highest chances to be overwritten. If those two are
216 * untouched, it's most likely safe to skip the context reset.
217 */
218 return head->frame->data != head->orig_ctx.data ||
219 head->frame->flags != head->orig_ctx.flags;
220}
221
222static bool ctx_was_changed(struct xdp_page_head *head)
223{
224 return head->orig_ctx.data != head->ctx.data ||
225 head->orig_ctx.data_meta != head->ctx.data_meta ||
226 head->orig_ctx.data_end != head->ctx.data_end;
227}
228
229static void reset_ctx(struct xdp_page_head *head)
230{
231 if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
232 return;
233
234 head->ctx.data = head->orig_ctx.data;
235 head->ctx.data_meta = head->orig_ctx.data_meta;
236 head->ctx.data_end = head->orig_ctx.data_end;
237 xdp_update_frame_from_buff(&head->ctx, head->frame);
238 head->frame->mem_type = head->orig_ctx.rxq->mem.type;
239}
240
241static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
242 struct sk_buff **skbs,
243 struct net_device *dev)
244{
245 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
246 int i;
247 LIST_HEAD(list);
248
249 if (unlikely(!kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp,
250 nframes, (void **)skbs))) {
251 for (i = 0; i < nframes; i++)
252 xdp_return_frame(frames[i]);
253 return -ENOMEM;
254 }
255
256 for (i = 0; i < nframes; i++) {
257 struct xdp_frame *xdpf = frames[i];
258 struct sk_buff *skb = skbs[i];
259
260 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
261 if (!skb) {
262 xdp_return_frame(xdpf);
263 continue;
264 }
265
266 list_add_tail(&skb->list, &list);
267 }
268 netif_receive_skb_list(&list);
269
270 return 0;
271}
272
273static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
274 u32 repeat)
275{
276 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
277 int err = 0, act, ret, i, nframes = 0, batch_sz;
278 struct xdp_frame **frames = xdp->frames;
279 struct bpf_redirect_info *ri;
280 struct xdp_page_head *head;
281 struct xdp_frame *frm;
282 bool redirect = false;
283 struct xdp_buff *ctx;
284 struct page *page;
285
286 batch_sz = min_t(u32, repeat, xdp->batch_size);
287
288 local_bh_disable();
289 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
290 ri = bpf_net_ctx_get_ri();
291 xdp_set_return_frame_no_direct();
292
293 for (i = 0; i < batch_sz; i++) {
294 page = page_pool_dev_alloc_pages(xdp->pp);
295 if (!page) {
296 err = -ENOMEM;
297 goto out;
298 }
299
300 head = phys_to_virt(page_to_phys(page));
301 reset_ctx(head);
302 ctx = &head->ctx;
303 frm = head->frame;
304 xdp->frame_cnt++;
305
306 act = bpf_prog_run_xdp(prog, ctx);
307
308 /* if program changed pkt bounds we need to update the xdp_frame */
309 if (unlikely(ctx_was_changed(head))) {
310 ret = xdp_update_frame_from_buff(ctx, frm);
311 if (ret) {
312 xdp_return_buff(ctx);
313 continue;
314 }
315 }
316
317 switch (act) {
318 case XDP_TX:
319 /* we can't do a real XDP_TX since we're not in the
320 * driver, so turn it into a REDIRECT back to the same
321 * index
322 */
323 ri->tgt_index = xdp->dev->ifindex;
324 ri->map_id = INT_MAX;
325 ri->map_type = BPF_MAP_TYPE_UNSPEC;
326 fallthrough;
327 case XDP_REDIRECT:
328 redirect = true;
329 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
330 if (ret)
331 xdp_return_buff(ctx);
332 break;
333 case XDP_PASS:
334 frames[nframes++] = frm;
335 break;
336 default:
337 bpf_warn_invalid_xdp_action(NULL, prog, act);
338 fallthrough;
339 case XDP_DROP:
340 xdp_return_buff(ctx);
341 break;
342 }
343 }
344
345out:
346 if (redirect)
347 xdp_do_flush();
348 if (nframes) {
349 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
350 if (ret)
351 err = ret;
352 }
353
354 xdp_clear_return_frame_no_direct();
355 bpf_net_ctx_clear(bpf_net_ctx);
356 local_bh_enable();
357 return err;
358}
359
360static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
361 u32 repeat, u32 batch_size, u32 *time)
362
363{
364 struct xdp_test_data xdp = { .batch_size = batch_size };
365 struct bpf_test_timer t = {};
366 int ret;
367
368 if (!repeat)
369 repeat = 1;
370
371 ret = xdp_test_run_setup(&xdp, ctx);
372 if (ret)
373 return ret;
374
375 bpf_test_timer_enter(&t);
376 do {
377 xdp.frame_cnt = 0;
378 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
379 if (unlikely(ret < 0))
380 break;
381 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
382 bpf_test_timer_leave(&t);
383
384 xdp_test_run_teardown(&xdp);
385 return ret;
386}
387
388static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
389 u32 *retval, u32 *time, bool xdp)
390{
391 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
392 struct bpf_prog_array_item item = {.prog = prog};
393 struct bpf_run_ctx *old_ctx;
394 struct bpf_cg_run_ctx run_ctx;
395 struct bpf_test_timer t = {};
396 enum bpf_cgroup_storage_type stype;
397 int ret;
398
399 for_each_cgroup_storage_type(stype) {
400 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
401 if (IS_ERR(item.cgroup_storage[stype])) {
402 item.cgroup_storage[stype] = NULL;
403 for_each_cgroup_storage_type(stype)
404 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
405 return -ENOMEM;
406 }
407 }
408
409 if (!repeat)
410 repeat = 1;
411
412 bpf_test_timer_enter(&t);
413 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
414 do {
415 run_ctx.prog_item = &item;
416 local_bh_disable();
417 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
418
419 if (xdp)
420 *retval = bpf_prog_run_xdp(prog, ctx);
421 else
422 *retval = bpf_prog_run(prog, ctx);
423
424 bpf_net_ctx_clear(bpf_net_ctx);
425 local_bh_enable();
426 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
427 bpf_reset_run_ctx(old_ctx);
428 bpf_test_timer_leave(&t);
429
430 for_each_cgroup_storage_type(stype)
431 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
432
433 return ret;
434}
435
436static int bpf_test_finish(const union bpf_attr *kattr,
437 union bpf_attr __user *uattr, const void *data,
438 struct skb_shared_info *sinfo, u32 size, u32 frag_size,
439 u32 retval, u32 duration)
440{
441 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
442 int err = -EFAULT;
443 u32 copy_size = size;
444
445 /* Clamp copy if the user has provided a size hint, but copy the full
446 * buffer if not to retain old behaviour.
447 */
448 if (kattr->test.data_size_out &&
449 copy_size > kattr->test.data_size_out) {
450 copy_size = kattr->test.data_size_out;
451 err = -ENOSPC;
452 }
453
454 if (data_out) {
455 u32 head_len = size - frag_size;
456 u32 len = min(copy_size, head_len);
457
458 if (copy_to_user(data_out, data, len))
459 goto out;
460
461 if (sinfo) {
462 int i, offset = len;
463 u32 data_len;
464
465 for (i = 0; i < sinfo->nr_frags; i++) {
466 skb_frag_t *frag = &sinfo->frags[i];
467
468 if (offset >= copy_size) {
469 err = -ENOSPC;
470 break;
471 }
472
473 data_len = min_t(u32, copy_size - offset,
474 skb_frag_size(frag));
475
476 if (copy_to_user(data_out + offset,
477 skb_frag_address(frag),
478 data_len))
479 goto out;
480
481 offset += data_len;
482 }
483 }
484 }
485
486 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
487 goto out;
488 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
489 goto out;
490 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
491 goto out;
492 if (err != -ENOSPC)
493 err = 0;
494out:
495 trace_bpf_test_finish(&err);
496 return err;
497}
498
499/* Integer types of various sizes and pointer combinations cover variety of
500 * architecture dependent calling conventions. 7+ can be supported in the
501 * future.
502 */
503__bpf_kfunc_start_defs();
504
505__bpf_kfunc int bpf_fentry_test1(int a)
506{
507 return a + 1;
508}
509EXPORT_SYMBOL_GPL(bpf_fentry_test1);
510
511noinline int bpf_fentry_test2(int a, u64 b)
512{
513 return a + b;
514}
515
516noinline int bpf_fentry_test3(char a, int b, u64 c)
517{
518 return a + b + c;
519}
520
521noinline int bpf_fentry_test4(void *a, char b, int c, u64 d)
522{
523 return (long)a + b + c + d;
524}
525
526noinline int bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
527{
528 return a + (long)b + c + d + e;
529}
530
531noinline int bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
532{
533 return a + (long)b + c + d + (long)e + f;
534}
535
536struct bpf_fentry_test_t {
537 struct bpf_fentry_test_t *a;
538};
539
540noinline int bpf_fentry_test7(struct bpf_fentry_test_t *arg)
541{
542 asm volatile ("" : "+r"(arg));
543 return (long)arg;
544}
545
546noinline int bpf_fentry_test8(struct bpf_fentry_test_t *arg)
547{
548 return (long)arg->a;
549}
550
551__bpf_kfunc u32 bpf_fentry_test9(u32 *a)
552{
553 return *a;
554}
555
556noinline int bpf_fentry_test10(const void *a)
557{
558 return (long)a;
559}
560
561noinline void bpf_fentry_test_sinfo(struct skb_shared_info *sinfo)
562{
563}
564
565noinline void bpf_fentry_test_ppvoid(void **pp)
566{
567}
568
569noinline void bpf_fentry_test_pppvoid(void ***ppp)
570{
571}
572
573noinline void bpf_fentry_test_ppfile(struct file **ppf)
574{
575}
576
577noinline struct file **bpf_fexit_test_ret_ppfile(void)
578{
579 return (struct file **)NULL;
580}
581
582__bpf_kfunc int bpf_modify_return_test(int a, int *b)
583{
584 *b += 1;
585 return a + *b;
586}
587
588__bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
589 void *e, char f, int g)
590{
591 *b += 1;
592 return a + *b + c + d + (long)e + f + g;
593}
594
595__bpf_kfunc int bpf_modify_return_test_tp(int nonce)
596{
597 trace_bpf_trigger_tp(nonce);
598
599 return nonce;
600}
601
602noinline int bpf_fentry_shadow_test(int a)
603{
604 return a + 1;
605}
606
607struct prog_test_member1 {
608 int a;
609};
610
611struct prog_test_member {
612 struct prog_test_member1 m;
613 int c;
614};
615
616struct prog_test_ref_kfunc {
617 int a;
618 int b;
619 struct prog_test_member memb;
620 struct prog_test_ref_kfunc *next;
621 refcount_t cnt;
622};
623
624__bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
625{
626 refcount_dec(&p->cnt);
627}
628
629__bpf_kfunc void bpf_kfunc_call_test_release_dtor(void *p)
630{
631 bpf_kfunc_call_test_release(p);
632}
633CFI_NOSEAL(bpf_kfunc_call_test_release_dtor);
634
635__bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
636{
637}
638
639__bpf_kfunc void bpf_kfunc_call_memb_release_dtor(void *p)
640{
641}
642CFI_NOSEAL(bpf_kfunc_call_memb_release_dtor);
643
644__bpf_kfunc_end_defs();
645
646BTF_KFUNCS_START(bpf_test_modify_return_ids)
647BTF_ID_FLAGS(func, bpf_modify_return_test)
648BTF_ID_FLAGS(func, bpf_modify_return_test2)
649BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
650BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
651BTF_KFUNCS_END(bpf_test_modify_return_ids)
652
653static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
654 .owner = THIS_MODULE,
655 .set = &bpf_test_modify_return_ids,
656};
657
658BTF_KFUNCS_START(test_sk_check_kfunc_ids)
659BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
660BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
661BTF_KFUNCS_END(test_sk_check_kfunc_ids)
662
663static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
664 u32 size, u32 headroom, u32 tailroom)
665{
666 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
667 void *data;
668
669 if (user_size > PAGE_SIZE - headroom - tailroom)
670 return ERR_PTR(-EINVAL);
671
672 size = SKB_DATA_ALIGN(size);
673 data = kzalloc(size + headroom + tailroom, GFP_USER);
674 if (!data)
675 return ERR_PTR(-ENOMEM);
676
677 if (copy_from_user(data + headroom, data_in, user_size)) {
678 kfree(data);
679 return ERR_PTR(-EFAULT);
680 }
681
682 return data;
683}
684
685int bpf_prog_test_run_tracing(struct bpf_prog *prog,
686 const union bpf_attr *kattr,
687 union bpf_attr __user *uattr)
688{
689 struct bpf_fentry_test_t arg = {};
690 u16 side_effect = 0, ret = 0;
691 int b = 2, err = -EFAULT;
692 u32 retval = 0;
693
694 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
695 return -EINVAL;
696
697 switch (prog->expected_attach_type) {
698 case BPF_TRACE_FENTRY:
699 case BPF_TRACE_FEXIT:
700 case BPF_TRACE_FSESSION:
701 case BPF_TRACE_FENTRY_MULTI:
702 case BPF_TRACE_FEXIT_MULTI:
703 case BPF_TRACE_FSESSION_MULTI:
704 if (bpf_fentry_test1(1) != 2 ||
705 bpf_fentry_test2(2, 3) != 5 ||
706 bpf_fentry_test3(4, 5, 6) != 15 ||
707 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
708 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
709 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
710 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
711 bpf_fentry_test8(&arg) != 0 ||
712 bpf_fentry_test9(&retval) != 0 ||
713 bpf_fentry_test10((void *)0) != 0)
714 goto out;
715 break;
716 case BPF_MODIFY_RETURN:
717 ret = bpf_modify_return_test(1, &b);
718 if (b != 2)
719 side_effect++;
720 b = 2;
721 ret += bpf_modify_return_test2(1, &b, 3, 4, (void *)5, 6, 7);
722 if (b != 2)
723 side_effect++;
724 break;
725 default:
726 goto out;
727 }
728
729 retval = ((u32)side_effect << 16) | ret;
730 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
731 goto out;
732
733 err = 0;
734out:
735 trace_bpf_test_finish(&err);
736 return err;
737}
738
739struct bpf_raw_tp_test_run_info {
740 struct bpf_prog *prog;
741 void *ctx;
742 u32 retval;
743};
744
745static void
746__bpf_prog_test_run_raw_tp(void *data)
747{
748 struct bpf_raw_tp_test_run_info *info = data;
749 struct srcu_ctr __percpu *scp = NULL;
750 struct bpf_trace_run_ctx run_ctx = {};
751 struct bpf_run_ctx *old_run_ctx;
752
753 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
754
755 if (info->prog->sleepable) {
756 scp = rcu_read_lock_tasks_trace();
757 migrate_disable();
758 } else {
759 rcu_read_lock();
760 }
761
762 if (unlikely(!bpf_prog_get_recursion_context(info->prog))) {
763 bpf_prog_inc_misses_counter(info->prog);
764 goto out;
765 }
766
767 info->retval = bpf_prog_run(info->prog, info->ctx);
768
769out:
770 bpf_prog_put_recursion_context(info->prog);
771
772 if (info->prog->sleepable) {
773 migrate_enable();
774 rcu_read_unlock_tasks_trace(scp);
775 } else {
776 rcu_read_unlock();
777 }
778
779 bpf_reset_run_ctx(old_run_ctx);
780}
781
782int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
783 const union bpf_attr *kattr,
784 union bpf_attr __user *uattr)
785{
786 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
787 __u32 ctx_size_in = kattr->test.ctx_size_in;
788 struct bpf_raw_tp_test_run_info info;
789 int cpu = kattr->test.cpu, err = 0;
790 int current_cpu;
791
792 /* doesn't support data_in/out, ctx_out, duration, or repeat */
793 if (kattr->test.data_in || kattr->test.data_out ||
794 kattr->test.ctx_out || kattr->test.duration ||
795 kattr->test.repeat || kattr->test.batch_size)
796 return -EINVAL;
797
798 if (ctx_size_in < prog->aux->max_ctx_offset ||
799 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
800 return -EINVAL;
801
802 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
803 return -EINVAL;
804
805 /*
806 * Sleepable programs cannot run with preemption disabled or in
807 * hardirq context (smp_call_function_single), reject the flag.
808 */
809 if (prog->sleepable && (kattr->test.flags & BPF_F_TEST_RUN_ON_CPU))
810 return -EINVAL;
811
812 if (ctx_size_in) {
813 info.ctx = memdup_user(ctx_in, ctx_size_in);
814 if (IS_ERR(info.ctx))
815 return PTR_ERR(info.ctx);
816 } else {
817 info.ctx = NULL;
818 }
819
820 info.retval = 0;
821 info.prog = prog;
822
823 if (prog->sleepable) {
824 __bpf_prog_test_run_raw_tp(&info);
825 } else {
826 current_cpu = get_cpu();
827 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
828 cpu == current_cpu) {
829 __bpf_prog_test_run_raw_tp(&info);
830 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
831 /*
832 * smp_call_function_single() also checks cpu_online()
833 * after csd_lock(). However, since cpu is from user
834 * space, let's do an extra quick check to filter out
835 * invalid value before smp_call_function_single().
836 */
837 err = -ENXIO;
838 } else {
839 err = smp_call_function_single(cpu,
840 __bpf_prog_test_run_raw_tp,
841 &info, 1);
842 }
843 put_cpu();
844 }
845
846 if (!err &&
847 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
848 err = -EFAULT;
849
850 kfree(info.ctx);
851 return err;
852}
853
854static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
855{
856 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
857 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
858 u32 size = kattr->test.ctx_size_in;
859 void *data;
860 int err;
861
862 if (!data_in && !data_out)
863 return NULL;
864
865 data = kzalloc(max_size, GFP_USER);
866 if (!data)
867 return ERR_PTR(-ENOMEM);
868
869 if (data_in) {
870 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
871 if (err) {
872 kfree(data);
873 return ERR_PTR(err);
874 }
875
876 size = min_t(u32, max_size, size);
877 if (copy_from_user(data, data_in, size)) {
878 kfree(data);
879 return ERR_PTR(-EFAULT);
880 }
881 }
882 return data;
883}
884
885static int bpf_ctx_finish(const union bpf_attr *kattr,
886 union bpf_attr __user *uattr, const void *data,
887 u32 size)
888{
889 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
890 int err = -EFAULT;
891 u32 copy_size = size;
892
893 if (!data || !data_out)
894 return 0;
895
896 if (copy_size > kattr->test.ctx_size_out) {
897 copy_size = kattr->test.ctx_size_out;
898 err = -ENOSPC;
899 }
900
901 if (copy_to_user(data_out, data, copy_size))
902 goto out;
903 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
904 goto out;
905 if (err != -ENOSPC)
906 err = 0;
907out:
908 return err;
909}
910
911/**
912 * range_is_zero - test whether buffer is initialized
913 * @buf: buffer to check
914 * @from: check from this position
915 * @to: check up until (excluding) this position
916 *
917 * This function returns true if the there is a non-zero byte
918 * in the buf in the range [from,to).
919 */
920static inline bool range_is_zero(void *buf, size_t from, size_t to)
921{
922 return !memchr_inv((u8 *)buf + from, 0, to - from);
923}
924
925static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
926{
927 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
928
929 if (!__skb)
930 return 0;
931
932 /* make sure the fields we don't use are zeroed */
933 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
934 return -EINVAL;
935
936 /* mark is allowed */
937
938 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
939 offsetof(struct __sk_buff, priority)))
940 return -EINVAL;
941
942 /* priority is allowed */
943 /* ingress_ifindex is allowed */
944 /* ifindex is allowed */
945
946 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
947 offsetof(struct __sk_buff, cb)))
948 return -EINVAL;
949
950 /* cb is allowed */
951
952 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
953 offsetof(struct __sk_buff, data_end)))
954 return -EINVAL;
955
956 /* data_end is allowed, but not copied to skb */
957
958 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, data_end),
959 offsetof(struct __sk_buff, tstamp)))
960 return -EINVAL;
961
962 /* tstamp is allowed */
963 /* wire_len is allowed */
964 /* gso_segs is allowed */
965
966 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
967 offsetof(struct __sk_buff, gso_size)))
968 return -EINVAL;
969
970 /* gso_size is allowed */
971
972 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
973 offsetof(struct __sk_buff, hwtstamp)))
974 return -EINVAL;
975
976 /* hwtstamp is allowed */
977
978 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
979 sizeof(struct __sk_buff)))
980 return -EINVAL;
981
982 skb->mark = __skb->mark;
983 skb->priority = __skb->priority;
984 skb->skb_iif = __skb->ingress_ifindex;
985 skb->tstamp = __skb->tstamp;
986 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
987
988 if (__skb->wire_len == 0) {
989 cb->pkt_len = skb->len;
990 } else {
991 if (__skb->wire_len < skb->len ||
992 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
993 return -EINVAL;
994 cb->pkt_len = __skb->wire_len;
995 }
996
997 if (__skb->gso_segs > GSO_MAX_SEGS)
998 return -EINVAL;
999
1000 /* Currently GSO type is zero/unset. If this gets extended with
1001 * a small list of accepted GSO types in future, the filter for
1002 * an unset GSO type in bpf_clone_redirect() can be lifted.
1003 */
1004 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1005 skb_shinfo(skb)->gso_size = __skb->gso_size;
1006 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1007
1008 return 0;
1009}
1010
1011static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1012{
1013 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1014
1015 if (!__skb)
1016 return;
1017
1018 __skb->mark = skb->mark;
1019 __skb->priority = skb->priority;
1020 __skb->ingress_ifindex = skb->skb_iif;
1021 __skb->ifindex = skb->dev->ifindex;
1022 __skb->tstamp = skb->tstamp;
1023 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1024 __skb->wire_len = cb->pkt_len;
1025 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
1026 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1027}
1028
1029static struct proto bpf_dummy_proto = {
1030 .name = "bpf_dummy",
1031 .owner = THIS_MODULE,
1032 .obj_size = sizeof(struct sock),
1033};
1034
1035int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1036 union bpf_attr __user *uattr)
1037{
1038 bool is_l2 = false, is_direct_pkt_access = false, is_lwt = false;
1039 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1040 struct net *net = current->nsproxy->net_ns;
1041 struct net_device *dev = net->loopback_dev;
1042 u32 headroom = NET_SKB_PAD + NET_IP_ALIGN;
1043 u32 linear_sz = kattr->test.data_size_in;
1044 u32 repeat = kattr->test.repeat;
1045 struct __sk_buff *ctx = NULL;
1046 struct sk_buff *skb = NULL;
1047 struct sock *sk = NULL;
1048 u32 retval, duration;
1049 int hh_len = ETH_HLEN;
1050 void *data = NULL;
1051 int ret;
1052
1053 if ((kattr->test.flags & ~BPF_F_TEST_SKB_CHECKSUM_COMPLETE) ||
1054 kattr->test.cpu || kattr->test.batch_size)
1055 return -EINVAL;
1056
1057 if (kattr->test.data_size_in < ETH_HLEN)
1058 return -EINVAL;
1059
1060 switch (prog->type) {
1061 case BPF_PROG_TYPE_SCHED_CLS:
1062 case BPF_PROG_TYPE_SCHED_ACT:
1063 is_direct_pkt_access = true;
1064 is_l2 = true;
1065 break;
1066 case BPF_PROG_TYPE_LWT_IN:
1067 case BPF_PROG_TYPE_LWT_OUT:
1068 case BPF_PROG_TYPE_LWT_XMIT:
1069 is_lwt = true;
1070 fallthrough;
1071 case BPF_PROG_TYPE_CGROUP_SKB:
1072 is_direct_pkt_access = true;
1073 break;
1074 default:
1075 break;
1076 }
1077
1078 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1079 if (IS_ERR(ctx))
1080 return PTR_ERR(ctx);
1081
1082 if (ctx) {
1083 if (ctx->data_end > kattr->test.data_size_in || ctx->data || ctx->data_meta) {
1084 ret = -EINVAL;
1085 goto out;
1086 }
1087 if (ctx->data_end) {
1088 /* Non-linear LWT test_run is unsupported for now. */
1089 if (is_lwt) {
1090 ret = -EINVAL;
1091 goto out;
1092 }
1093 linear_sz = max(ETH_HLEN, ctx->data_end);
1094 }
1095 }
1096
1097 linear_sz = min_t(u32, linear_sz, PAGE_SIZE - headroom - tailroom);
1098
1099 data = bpf_test_init(kattr, linear_sz, linear_sz, headroom, tailroom);
1100 if (IS_ERR(data)) {
1101 ret = PTR_ERR(data);
1102 data = NULL;
1103 goto out;
1104 }
1105
1106 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1107 if (!sk) {
1108 ret = -ENOMEM;
1109 goto out;
1110 }
1111 sock_init_data(NULL, sk);
1112
1113 skb = slab_build_skb(data);
1114 if (!skb) {
1115 ret = -ENOMEM;
1116 goto out;
1117 }
1118 skb->sk = sk;
1119
1120 data = NULL; /* data released via kfree_skb */
1121
1122 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1123 __skb_put(skb, linear_sz);
1124
1125 if (unlikely(kattr->test.data_size_in > linear_sz)) {
1126 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1127 struct skb_shared_info *sinfo = skb_shinfo(skb);
1128 u32 copied = linear_sz;
1129
1130 while (copied < kattr->test.data_size_in) {
1131 struct page *page;
1132 u32 data_len;
1133
1134 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1135 ret = -ENOMEM;
1136 goto out;
1137 }
1138
1139 page = alloc_page(GFP_KERNEL);
1140 if (!page) {
1141 ret = -ENOMEM;
1142 goto out;
1143 }
1144
1145 data_len = min_t(u32, kattr->test.data_size_in - copied,
1146 PAGE_SIZE);
1147 skb_fill_page_desc(skb, sinfo->nr_frags, page, 0, data_len);
1148
1149 if (copy_from_user(page_address(page), data_in + copied,
1150 data_len)) {
1151 ret = -EFAULT;
1152 goto out;
1153 }
1154 skb->data_len += data_len;
1155 skb->truesize += PAGE_SIZE;
1156 skb->len += data_len;
1157 copied += data_len;
1158 }
1159 }
1160
1161 if (ctx && ctx->ifindex > 1) {
1162 dev = dev_get_by_index(net, ctx->ifindex);
1163 if (!dev) {
1164 ret = -ENODEV;
1165 goto out;
1166 }
1167 }
1168 skb->protocol = eth_type_trans(skb, dev);
1169 skb_reset_network_header(skb);
1170
1171 switch (skb->protocol) {
1172 case htons(ETH_P_IP):
1173 if (skb_headlen(skb) < sizeof(struct iphdr)) {
1174 ret = -EINVAL;
1175 goto out;
1176 }
1177 sk->sk_family = AF_INET;
1178 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1179 sk->sk_daddr = ip_hdr(skb)->daddr;
1180 break;
1181#if IS_ENABLED(CONFIG_IPV6)
1182 case htons(ETH_P_IPV6):
1183 if (skb_headlen(skb) < sizeof(struct ipv6hdr)) {
1184 ret = -EINVAL;
1185 goto out;
1186 }
1187 sk->sk_family = AF_INET6;
1188 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1189 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1190 break;
1191#endif
1192 default:
1193 break;
1194 }
1195
1196 if (is_l2)
1197 __skb_push(skb, hh_len);
1198 if (is_direct_pkt_access)
1199 bpf_compute_data_pointers(skb);
1200
1201 ret = convert___skb_to_skb(skb, ctx);
1202 if (ret)
1203 goto out;
1204
1205 if (kattr->test.flags & BPF_F_TEST_SKB_CHECKSUM_COMPLETE) {
1206 const int off = skb_network_offset(skb);
1207 int len = skb->len - off;
1208
1209 skb->csum = skb_checksum(skb, off, len, 0);
1210 skb->ip_summed = CHECKSUM_COMPLETE;
1211 }
1212
1213 if (prog->type == BPF_PROG_TYPE_LWT_XMIT) {
1214 if (!ipv6_mod_enabled()) {
1215 pr_warn_once("Please test this program with IPv6 enabled kernel\n");
1216 ret = -EOPNOTSUPP;
1217 goto out;
1218 }
1219#if IS_ENABLED(CONFIG_IPV6)
1220 dst_hold(&net->ipv6.ip6_null_entry->dst);
1221 skb_dst_set(skb, &net->ipv6.ip6_null_entry->dst);
1222#endif
1223 }
1224
1225 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1226 if (ret)
1227 goto out;
1228 if (!is_l2) {
1229 if (skb_headroom(skb) < hh_len) {
1230 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1231
1232 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1233 ret = -ENOMEM;
1234 goto out;
1235 }
1236 }
1237 memset(__skb_push(skb, hh_len), 0, hh_len);
1238 }
1239
1240 if (kattr->test.flags & BPF_F_TEST_SKB_CHECKSUM_COMPLETE) {
1241 const int off = skb_network_offset(skb);
1242 int len = skb->len - off;
1243 __wsum csum;
1244
1245 csum = skb_checksum(skb, off, len, 0);
1246
1247 if (csum_fold(skb->csum) != csum_fold(csum)) {
1248 ret = -EBADMSG;
1249 goto out;
1250 }
1251 }
1252
1253 convert_skb_to___skb(skb, ctx);
1254
1255 if (skb_is_nonlinear(skb))
1256 /* bpf program can never convert linear skb to non-linear */
1257 WARN_ON_ONCE(linear_sz == kattr->test.data_size_in);
1258 ret = bpf_test_finish(kattr, uattr, skb->data, skb_shinfo(skb), skb->len,
1259 skb->data_len, retval, duration);
1260 if (!ret)
1261 ret = bpf_ctx_finish(kattr, uattr, ctx,
1262 sizeof(struct __sk_buff));
1263out:
1264 if (dev && dev != net->loopback_dev)
1265 dev_put(dev);
1266 kfree_skb(skb);
1267 kfree(data);
1268 if (sk)
1269 sk_free(sk);
1270 kfree(ctx);
1271 return ret;
1272}
1273
1274static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1275{
1276 unsigned int ingress_ifindex, rx_queue_index;
1277 struct netdev_rx_queue *rxqueue;
1278 struct net_device *device;
1279
1280 if (!xdp_md)
1281 return 0;
1282
1283 if (xdp_md->egress_ifindex != 0)
1284 return -EINVAL;
1285
1286 ingress_ifindex = xdp_md->ingress_ifindex;
1287 rx_queue_index = xdp_md->rx_queue_index;
1288
1289 if (!ingress_ifindex && rx_queue_index)
1290 return -EINVAL;
1291
1292 if (ingress_ifindex) {
1293 device = dev_get_by_index(current->nsproxy->net_ns,
1294 ingress_ifindex);
1295 if (!device)
1296 return -ENODEV;
1297
1298 if (rx_queue_index >= device->real_num_rx_queues)
1299 goto free_dev;
1300
1301 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1302
1303 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1304 goto free_dev;
1305
1306 xdp->rxq = &rxqueue->xdp_rxq;
1307 /* The device is now tracked in the xdp->rxq for later
1308 * dev_put()
1309 */
1310 }
1311
1312 xdp->data = xdp->data_meta + xdp_md->data;
1313 return 0;
1314
1315free_dev:
1316 dev_put(device);
1317 return -EINVAL;
1318}
1319
1320static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1321{
1322 if (!xdp_md)
1323 return;
1324
1325 xdp_md->data = xdp->data - xdp->data_meta;
1326 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1327
1328 if (xdp_md->ingress_ifindex)
1329 dev_put(xdp->rxq->dev);
1330}
1331
1332int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1333 union bpf_attr __user *uattr)
1334{
1335 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1336 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1337 u32 retval = 0, meta_sz = 0, duration, max_linear_sz, size;
1338 u32 linear_sz = kattr->test.data_size_in;
1339 u32 batch_size = kattr->test.batch_size;
1340 u32 headroom = XDP_PACKET_HEADROOM;
1341 u32 repeat = kattr->test.repeat;
1342 struct netdev_rx_queue *rxqueue;
1343 struct skb_shared_info *sinfo;
1344 struct xdp_buff xdp = {};
1345 int i, ret = -EINVAL;
1346 struct xdp_md *ctx;
1347 void *data;
1348
1349 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1350 prog->expected_attach_type == BPF_XDP_CPUMAP)
1351 return -EINVAL;
1352
1353 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1354 return -EINVAL;
1355
1356 if (bpf_prog_is_dev_bound(prog->aux))
1357 return -EINVAL;
1358
1359 if (do_live) {
1360 if (!batch_size)
1361 batch_size = NAPI_POLL_WEIGHT;
1362 else if (batch_size > TEST_XDP_MAX_BATCH)
1363 return -E2BIG;
1364 } else if (batch_size) {
1365 return -EINVAL;
1366 }
1367
1368 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1369 if (IS_ERR(ctx))
1370 return PTR_ERR(ctx);
1371
1372 if (ctx) {
1373 /* There can't be user provided data before the meta data */
1374 if (ctx->data_meta || ctx->data_end > kattr->test.data_size_in ||
1375 ctx->data > ctx->data_end ||
1376 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1377 goto free_ctx;
1378
1379 meta_sz = ctx->data;
1380 if (xdp_metalen_invalid(meta_sz) || meta_sz > headroom - sizeof(struct xdp_frame))
1381 goto free_ctx;
1382
1383 /* Meta data is allocated from the headroom */
1384 headroom -= meta_sz;
1385 linear_sz = ctx->data_end;
1386 }
1387
1388 /* The xdp_page_head structure takes up space in each page, limiting the
1389 * size of the packet data; add the extra size to headroom here to make
1390 * sure it's accounted in the length checks below, but not in the
1391 * metadata size check above.
1392 */
1393 if (do_live)
1394 headroom += sizeof(struct xdp_page_head);
1395
1396 max_linear_sz = PAGE_SIZE - headroom - tailroom;
1397 linear_sz = min_t(u32, linear_sz, max_linear_sz);
1398
1399 /* disallow live data mode for jumbo frames */
1400 if (do_live && kattr->test.data_size_in > linear_sz)
1401 goto free_ctx;
1402
1403 if (kattr->test.data_size_in - meta_sz < ETH_HLEN)
1404 goto free_ctx;
1405
1406 data = bpf_test_init(kattr, linear_sz, max_linear_sz, headroom, tailroom);
1407 if (IS_ERR(data)) {
1408 ret = PTR_ERR(data);
1409 goto free_ctx;
1410 }
1411
1412 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1413 rxqueue->xdp_rxq.frag_size = PAGE_SIZE;
1414 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1415 xdp_prepare_buff(&xdp, data, headroom, linear_sz, true);
1416 sinfo = xdp_get_shared_info_from_buff(&xdp);
1417
1418 ret = xdp_convert_md_to_buff(ctx, &xdp);
1419 if (ret)
1420 goto free_data;
1421
1422 size = linear_sz;
1423 if (unlikely(kattr->test.data_size_in > size)) {
1424 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1425
1426 while (size < kattr->test.data_size_in) {
1427 struct page *page;
1428 skb_frag_t *frag;
1429 u32 data_len;
1430
1431 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1432 ret = -ENOMEM;
1433 goto out_put_dev;
1434 }
1435
1436 page = alloc_page(GFP_KERNEL);
1437 if (!page) {
1438 ret = -ENOMEM;
1439 goto out_put_dev;
1440 }
1441
1442 frag = &sinfo->frags[sinfo->nr_frags++];
1443
1444 data_len = min_t(u32, kattr->test.data_size_in - size,
1445 PAGE_SIZE);
1446 skb_frag_fill_page_desc(frag, page, 0, data_len);
1447
1448 if (copy_from_user(page_address(page), data_in + size,
1449 data_len)) {
1450 ret = -EFAULT;
1451 goto out_put_dev;
1452 }
1453 sinfo->xdp_frags_size += data_len;
1454 size += data_len;
1455 }
1456 xdp_buff_set_frags_flag(&xdp);
1457 }
1458
1459 if (repeat > 1)
1460 bpf_prog_change_xdp(NULL, prog);
1461
1462 if (do_live)
1463 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1464 else
1465 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1466out_put_dev:
1467 /* We convert the xdp_buff back to an xdp_md before checking the return
1468 * code so the reference count of any held netdevice will be decremented
1469 * even if the test run failed.
1470 */
1471 xdp_convert_buff_to_md(&xdp, ctx);
1472 if (ret)
1473 goto out;
1474
1475 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1476 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size, sinfo->xdp_frags_size,
1477 retval, duration);
1478 if (!ret)
1479 ret = bpf_ctx_finish(kattr, uattr, ctx,
1480 sizeof(struct xdp_md));
1481
1482out:
1483 if (repeat > 1)
1484 bpf_prog_change_xdp(prog, NULL);
1485free_data:
1486 for (i = 0; i < sinfo->nr_frags; i++)
1487 __free_page(skb_frag_page(&sinfo->frags[i]));
1488 kfree(data);
1489free_ctx:
1490 kfree(ctx);
1491 return ret;
1492}
1493
1494static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1495{
1496 /* make sure the fields we don't use are zeroed */
1497 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1498 return -EINVAL;
1499
1500 /* flags is allowed */
1501
1502 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1503 sizeof(struct bpf_flow_keys)))
1504 return -EINVAL;
1505
1506 return 0;
1507}
1508
1509int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1510 const union bpf_attr *kattr,
1511 union bpf_attr __user *uattr)
1512{
1513 struct bpf_test_timer t = {};
1514 u32 size = kattr->test.data_size_in;
1515 struct bpf_flow_dissector ctx = {};
1516 u32 repeat = kattr->test.repeat;
1517 struct bpf_flow_keys *user_ctx;
1518 struct bpf_flow_keys flow_keys;
1519 const struct ethhdr *eth;
1520 unsigned int flags = 0;
1521 u32 retval, duration;
1522 void *data;
1523 int ret;
1524
1525 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1526 return -EINVAL;
1527
1528 if (size < ETH_HLEN)
1529 return -EINVAL;
1530
1531 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1532 if (IS_ERR(data))
1533 return PTR_ERR(data);
1534
1535 eth = (struct ethhdr *)data;
1536
1537 if (!repeat)
1538 repeat = 1;
1539
1540 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1541 if (IS_ERR(user_ctx)) {
1542 kfree(data);
1543 return PTR_ERR(user_ctx);
1544 }
1545 if (user_ctx) {
1546 ret = verify_user_bpf_flow_keys(user_ctx);
1547 if (ret)
1548 goto out;
1549 flags = user_ctx->flags;
1550 }
1551
1552 ctx.flow_keys = &flow_keys;
1553 ctx.data = data;
1554 ctx.data_end = (__u8 *)data + size;
1555
1556 bpf_test_timer_enter(&t);
1557 do {
1558 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1559 size, flags);
1560 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1561 bpf_test_timer_leave(&t);
1562
1563 if (ret < 0)
1564 goto out;
1565
1566 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1567 sizeof(flow_keys), 0, retval, duration);
1568 if (!ret)
1569 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1570 sizeof(struct bpf_flow_keys));
1571
1572out:
1573 kfree(user_ctx);
1574 kfree(data);
1575 return ret;
1576}
1577
1578int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1579 union bpf_attr __user *uattr)
1580{
1581 struct bpf_test_timer t = {};
1582 struct bpf_prog_array *progs = NULL;
1583 struct bpf_sk_lookup_kern ctx = {};
1584 u32 repeat = kattr->test.repeat;
1585 struct bpf_sk_lookup *user_ctx;
1586 u32 retval, duration;
1587 int ret = -EINVAL;
1588
1589 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1590 return -EINVAL;
1591
1592 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1593 kattr->test.data_size_out)
1594 return -EINVAL;
1595
1596 if (!repeat)
1597 repeat = 1;
1598
1599 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1600 if (IS_ERR(user_ctx))
1601 return PTR_ERR(user_ctx);
1602
1603 if (!user_ctx)
1604 return -EINVAL;
1605
1606 if (user_ctx->sk)
1607 goto out;
1608
1609 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1610 goto out;
1611
1612 if (user_ctx->local_port > U16_MAX) {
1613 ret = -ERANGE;
1614 goto out;
1615 }
1616
1617 ctx.family = (u16)user_ctx->family;
1618 ctx.protocol = (u16)user_ctx->protocol;
1619 ctx.dport = (u16)user_ctx->local_port;
1620 ctx.sport = user_ctx->remote_port;
1621
1622 switch (ctx.family) {
1623 case AF_INET:
1624 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1625 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1626 break;
1627
1628#if IS_ENABLED(CONFIG_IPV6)
1629 case AF_INET6:
1630 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1631 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1632 break;
1633#endif
1634
1635 default:
1636 ret = -EAFNOSUPPORT;
1637 goto out;
1638 }
1639
1640 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1641 if (!progs) {
1642 ret = -ENOMEM;
1643 goto out;
1644 }
1645
1646 progs->items[0].prog = prog;
1647
1648 bpf_test_timer_enter(&t);
1649 do {
1650 ctx.selected_sk = NULL;
1651 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1652 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1653 bpf_test_timer_leave(&t);
1654
1655 if (ret < 0)
1656 goto out;
1657
1658 user_ctx->cookie = 0;
1659 if (ctx.selected_sk) {
1660 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1661 ret = -EOPNOTSUPP;
1662 goto out;
1663 }
1664
1665 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1666 }
1667
1668 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, 0, retval, duration);
1669 if (!ret)
1670 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1671
1672out:
1673 bpf_prog_array_free(progs);
1674 kfree(user_ctx);
1675 return ret;
1676}
1677
1678int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1679 const union bpf_attr *kattr,
1680 union bpf_attr __user *uattr)
1681{
1682 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1683 __u32 ctx_size_in = kattr->test.ctx_size_in;
1684 void *ctx = NULL;
1685 u32 retval;
1686 int err = 0;
1687
1688 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1689 if (kattr->test.data_in || kattr->test.data_out ||
1690 kattr->test.ctx_out || kattr->test.duration ||
1691 kattr->test.repeat || kattr->test.flags ||
1692 kattr->test.batch_size)
1693 return -EINVAL;
1694
1695 if (ctx_size_in < prog->aux->max_ctx_offset ||
1696 ctx_size_in > U16_MAX)
1697 return -EINVAL;
1698
1699 if (ctx_size_in) {
1700 ctx = memdup_user(ctx_in, ctx_size_in);
1701 if (IS_ERR(ctx))
1702 return PTR_ERR(ctx);
1703 }
1704
1705 rcu_read_lock_trace();
1706 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1707 rcu_read_unlock_trace();
1708
1709 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1710 err = -EFAULT;
1711 goto out;
1712 }
1713 if (ctx_size_in)
1714 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1715 err = -EFAULT;
1716out:
1717 kfree(ctx);
1718 return err;
1719}
1720
1721static int verify_and_copy_hook_state(struct nf_hook_state *state,
1722 const struct nf_hook_state *user,
1723 struct net_device *dev)
1724{
1725 if (user->in || user->out)
1726 return -EINVAL;
1727
1728 if (user->net || user->sk || user->okfn)
1729 return -EINVAL;
1730
1731 switch (user->pf) {
1732 case NFPROTO_IPV4:
1733 case NFPROTO_IPV6:
1734 switch (state->hook) {
1735 case NF_INET_PRE_ROUTING:
1736 state->in = dev;
1737 break;
1738 case NF_INET_LOCAL_IN:
1739 state->in = dev;
1740 break;
1741 case NF_INET_FORWARD:
1742 state->in = dev;
1743 state->out = dev;
1744 break;
1745 case NF_INET_LOCAL_OUT:
1746 state->out = dev;
1747 break;
1748 case NF_INET_POST_ROUTING:
1749 state->out = dev;
1750 break;
1751 }
1752
1753 break;
1754 default:
1755 return -EINVAL;
1756 }
1757
1758 state->pf = user->pf;
1759 state->hook = user->hook;
1760
1761 return 0;
1762}
1763
1764static __be16 nfproto_eth(int nfproto)
1765{
1766 switch (nfproto) {
1767 case NFPROTO_IPV4:
1768 return htons(ETH_P_IP);
1769 case NFPROTO_IPV6:
1770 break;
1771 }
1772
1773 return htons(ETH_P_IPV6);
1774}
1775
1776int bpf_prog_test_run_nf(struct bpf_prog *prog,
1777 const union bpf_attr *kattr,
1778 union bpf_attr __user *uattr)
1779{
1780 struct net *net = current->nsproxy->net_ns;
1781 struct net_device *dev = net->loopback_dev;
1782 struct nf_hook_state *user_ctx, hook_state = {
1783 .pf = NFPROTO_IPV4,
1784 .hook = NF_INET_LOCAL_OUT,
1785 };
1786 u32 size = kattr->test.data_size_in;
1787 u32 repeat = kattr->test.repeat;
1788 struct bpf_nf_ctx ctx = {
1789 .state = &hook_state,
1790 };
1791 struct sk_buff *skb = NULL;
1792 u32 retval, duration;
1793 void *data;
1794 int ret;
1795
1796 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1797 return -EINVAL;
1798
1799 if (size < sizeof(struct iphdr))
1800 return -EINVAL;
1801
1802 data = bpf_test_init(kattr, kattr->test.data_size_in, size,
1803 NET_SKB_PAD + NET_IP_ALIGN,
1804 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1805 if (IS_ERR(data))
1806 return PTR_ERR(data);
1807
1808 if (!repeat)
1809 repeat = 1;
1810
1811 user_ctx = bpf_ctx_init(kattr, sizeof(struct nf_hook_state));
1812 if (IS_ERR(user_ctx)) {
1813 kfree(data);
1814 return PTR_ERR(user_ctx);
1815 }
1816
1817 if (user_ctx) {
1818 ret = verify_and_copy_hook_state(&hook_state, user_ctx, dev);
1819 if (ret)
1820 goto out;
1821 }
1822
1823 skb = slab_build_skb(data);
1824 if (!skb) {
1825 ret = -ENOMEM;
1826 goto out;
1827 }
1828
1829 data = NULL; /* data released via kfree_skb */
1830
1831 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1832 __skb_put(skb, size);
1833
1834 ret = -EINVAL;
1835
1836 if (hook_state.hook != NF_INET_LOCAL_OUT) {
1837 if (size < ETH_HLEN + sizeof(struct iphdr))
1838 goto out;
1839
1840 skb->protocol = eth_type_trans(skb, dev);
1841 switch (skb->protocol) {
1842 case htons(ETH_P_IP):
1843 if (hook_state.pf == NFPROTO_IPV4)
1844 break;
1845 goto out;
1846 case htons(ETH_P_IPV6):
1847 if (size < ETH_HLEN + sizeof(struct ipv6hdr))
1848 goto out;
1849 if (hook_state.pf == NFPROTO_IPV6)
1850 break;
1851 goto out;
1852 default:
1853 ret = -EPROTO;
1854 goto out;
1855 }
1856
1857 skb_reset_network_header(skb);
1858 } else {
1859 skb->protocol = nfproto_eth(hook_state.pf);
1860 }
1861
1862 ctx.skb = skb;
1863
1864 ret = bpf_test_run(prog, &ctx, repeat, &retval, &duration, false);
1865 if (ret)
1866 goto out;
1867
1868 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, 0, retval, duration);
1869
1870out:
1871 kfree(user_ctx);
1872 kfree_skb(skb);
1873 kfree(data);
1874 return ret;
1875}
1876
1877static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1878 .owner = THIS_MODULE,
1879 .set = &test_sk_check_kfunc_ids,
1880};
1881
1882BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1883BTF_ID(struct, prog_test_ref_kfunc)
1884BTF_ID(func, bpf_kfunc_call_test_release_dtor)
1885BTF_ID(struct, prog_test_member)
1886BTF_ID(func, bpf_kfunc_call_memb_release_dtor)
1887
1888static int __init bpf_prog_test_run_init(void)
1889{
1890 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1891 {
1892 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1893 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1894 },
1895 {
1896 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1897 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1898 },
1899 };
1900 int ret;
1901
1902 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1903 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1904 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1905 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1906 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1907 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1908 THIS_MODULE);
1909}
1910late_initcall(bpf_prog_test_run_init);