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