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/sock.h>
16 #include <net/tcp.h>
17 #include <net/net_namespace.h>
18 #include <net/page_pool.h>
19 #include <linux/error-injection.h>
20 #include <linux/smp.h>
21 #include <linux/sock_diag.h>
22 #include <net/xdp.h>
23
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/bpf_test_run.h>
26
27 struct bpf_test_timer {
28 enum { NO_PREEMPT, NO_MIGRATE } mode;
29 u32 i;
30 u64 time_start, time_spent;
31 };
32
bpf_test_timer_enter(struct bpf_test_timer * t)33 static void bpf_test_timer_enter(struct bpf_test_timer *t)
34 __acquires(rcu)
35 {
36 rcu_read_lock();
37 if (t->mode == NO_PREEMPT)
38 preempt_disable();
39 else
40 migrate_disable();
41
42 t->time_start = ktime_get_ns();
43 }
44
bpf_test_timer_leave(struct bpf_test_timer * t)45 static void bpf_test_timer_leave(struct bpf_test_timer *t)
46 __releases(rcu)
47 {
48 t->time_start = 0;
49
50 if (t->mode == NO_PREEMPT)
51 preempt_enable();
52 else
53 migrate_enable();
54 rcu_read_unlock();
55 }
56
bpf_test_timer_continue(struct bpf_test_timer * t,int iterations,u32 repeat,int * err,u32 * duration)57 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
58 u32 repeat, int *err, u32 *duration)
59 __must_hold(rcu)
60 {
61 t->i += iterations;
62 if (t->i >= repeat) {
63 /* We're done. */
64 t->time_spent += ktime_get_ns() - t->time_start;
65 do_div(t->time_spent, t->i);
66 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
67 *err = 0;
68 goto reset;
69 }
70
71 if (signal_pending(current)) {
72 /* During iteration: we've been cancelled, abort. */
73 *err = -EINTR;
74 goto reset;
75 }
76
77 if (need_resched()) {
78 /* During iteration: we need to reschedule between runs. */
79 t->time_spent += ktime_get_ns() - t->time_start;
80 bpf_test_timer_leave(t);
81 cond_resched();
82 bpf_test_timer_enter(t);
83 }
84
85 /* Do another round. */
86 return true;
87
88 reset:
89 t->i = 0;
90 return false;
91 }
92
93 /* We put this struct at the head of each page with a context and frame
94 * initialised when the page is allocated, so we don't have to do this on each
95 * repetition of the test run.
96 */
97 struct xdp_page_head {
98 struct xdp_buff orig_ctx;
99 struct xdp_buff ctx;
100 struct xdp_frame frm;
101 u8 data[];
102 };
103
104 struct xdp_test_data {
105 struct xdp_buff *orig_ctx;
106 struct xdp_rxq_info rxq;
107 struct net_device *dev;
108 struct page_pool *pp;
109 struct xdp_frame **frames;
110 struct sk_buff **skbs;
111 struct xdp_mem_info mem;
112 u32 batch_size;
113 u32 frame_cnt;
114 };
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(struct page * page,void * arg)119 static void xdp_test_run_init_page(struct page *page, void *arg)
120 {
121 struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
122 struct xdp_buff *new_ctx, *orig_ctx;
123 u32 headroom = XDP_PACKET_HEADROOM;
124 struct xdp_test_data *xdp = arg;
125 size_t frm_len, meta_len;
126 struct xdp_frame *frm;
127 void *data;
128
129 orig_ctx = xdp->orig_ctx;
130 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
131 meta_len = orig_ctx->data - orig_ctx->data_meta;
132 headroom -= meta_len;
133
134 new_ctx = &head->ctx;
135 frm = &head->frm;
136 data = &head->data;
137 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
138
139 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
140 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
141 new_ctx->data = new_ctx->data_meta + meta_len;
142
143 xdp_update_frame_from_buff(new_ctx, frm);
144 frm->mem = new_ctx->rxq->mem;
145
146 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
147 }
148
xdp_test_run_setup(struct xdp_test_data * xdp,struct xdp_buff * orig_ctx)149 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
150 {
151 struct page_pool *pp;
152 int err = -ENOMEM;
153 struct page_pool_params pp_params = {
154 .order = 0,
155 .flags = 0,
156 .pool_size = xdp->batch_size,
157 .nid = NUMA_NO_NODE,
158 .init_callback = xdp_test_run_init_page,
159 .init_arg = xdp,
160 };
161
162 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
163 if (!xdp->frames)
164 return -ENOMEM;
165
166 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
167 if (!xdp->skbs)
168 goto err_skbs;
169
170 pp = page_pool_create(&pp_params);
171 if (IS_ERR(pp)) {
172 err = PTR_ERR(pp);
173 goto err_pp;
174 }
175
176 /* will copy 'mem.id' into pp->xdp_mem_id */
177 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
178 if (err)
179 goto err_mmodel;
180
181 xdp->pp = pp;
182
183 /* We create a 'fake' RXQ referencing the original dev, but with an
184 * xdp_mem_info pointing to our page_pool
185 */
186 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
187 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
188 xdp->rxq.mem.id = pp->xdp_mem_id;
189 xdp->dev = orig_ctx->rxq->dev;
190 xdp->orig_ctx = orig_ctx;
191
192 return 0;
193
194 err_mmodel:
195 page_pool_destroy(pp);
196 err_pp:
197 kvfree(xdp->skbs);
198 err_skbs:
199 kvfree(xdp->frames);
200 return err;
201 }
202
xdp_test_run_teardown(struct xdp_test_data * xdp)203 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
204 {
205 xdp_unreg_mem_model(&xdp->mem);
206 page_pool_destroy(xdp->pp);
207 kfree(xdp->frames);
208 kfree(xdp->skbs);
209 }
210
ctx_was_changed(struct xdp_page_head * head)211 static bool ctx_was_changed(struct xdp_page_head *head)
212 {
213 return head->orig_ctx.data != head->ctx.data ||
214 head->orig_ctx.data_meta != head->ctx.data_meta ||
215 head->orig_ctx.data_end != head->ctx.data_end;
216 }
217
reset_ctx(struct xdp_page_head * head)218 static void reset_ctx(struct xdp_page_head *head)
219 {
220 if (likely(!ctx_was_changed(head)))
221 return;
222
223 head->ctx.data = head->orig_ctx.data;
224 head->ctx.data_meta = head->orig_ctx.data_meta;
225 head->ctx.data_end = head->orig_ctx.data_end;
226 xdp_update_frame_from_buff(&head->ctx, &head->frm);
227 }
228
xdp_recv_frames(struct xdp_frame ** frames,int nframes,struct sk_buff ** skbs,struct net_device * dev)229 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
230 struct sk_buff **skbs,
231 struct net_device *dev)
232 {
233 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
234 int i, n;
235 LIST_HEAD(list);
236
237 n = kmem_cache_alloc_bulk(skbuff_head_cache, gfp, nframes, (void **)skbs);
238 if (unlikely(n == 0)) {
239 for (i = 0; i < nframes; i++)
240 xdp_return_frame(frames[i]);
241 return -ENOMEM;
242 }
243
244 for (i = 0; i < nframes; i++) {
245 struct xdp_frame *xdpf = frames[i];
246 struct sk_buff *skb = skbs[i];
247
248 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
249 if (!skb) {
250 xdp_return_frame(xdpf);
251 continue;
252 }
253
254 list_add_tail(&skb->list, &list);
255 }
256 netif_receive_skb_list(&list);
257
258 return 0;
259 }
260
xdp_test_run_batch(struct xdp_test_data * xdp,struct bpf_prog * prog,u32 repeat)261 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
262 u32 repeat)
263 {
264 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
265 int err = 0, act, ret, i, nframes = 0, batch_sz;
266 struct xdp_frame **frames = xdp->frames;
267 struct xdp_page_head *head;
268 struct xdp_frame *frm;
269 bool redirect = false;
270 struct xdp_buff *ctx;
271 struct page *page;
272
273 batch_sz = min_t(u32, repeat, xdp->batch_size);
274
275 local_bh_disable();
276 xdp_set_return_frame_no_direct();
277
278 for (i = 0; i < batch_sz; i++) {
279 page = page_pool_dev_alloc_pages(xdp->pp);
280 if (!page) {
281 err = -ENOMEM;
282 goto out;
283 }
284
285 head = phys_to_virt(page_to_phys(page));
286 reset_ctx(head);
287 ctx = &head->ctx;
288 frm = &head->frm;
289 xdp->frame_cnt++;
290
291 act = bpf_prog_run_xdp(prog, ctx);
292
293 /* if program changed pkt bounds we need to update the xdp_frame */
294 if (unlikely(ctx_was_changed(head))) {
295 ret = xdp_update_frame_from_buff(ctx, frm);
296 if (ret) {
297 xdp_return_buff(ctx);
298 continue;
299 }
300 }
301
302 switch (act) {
303 case XDP_TX:
304 /* we can't do a real XDP_TX since we're not in the
305 * driver, so turn it into a REDIRECT back to the same
306 * index
307 */
308 ri->tgt_index = xdp->dev->ifindex;
309 ri->map_id = INT_MAX;
310 ri->map_type = BPF_MAP_TYPE_UNSPEC;
311 fallthrough;
312 case XDP_REDIRECT:
313 redirect = true;
314 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
315 if (ret)
316 xdp_return_buff(ctx);
317 break;
318 case XDP_PASS:
319 frames[nframes++] = frm;
320 break;
321 default:
322 bpf_warn_invalid_xdp_action(NULL, prog, act);
323 fallthrough;
324 case XDP_DROP:
325 xdp_return_buff(ctx);
326 break;
327 }
328 }
329
330 out:
331 if (redirect)
332 xdp_do_flush();
333 if (nframes) {
334 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
335 if (ret)
336 err = ret;
337 }
338
339 xdp_clear_return_frame_no_direct();
340 local_bh_enable();
341 return err;
342 }
343
bpf_test_run_xdp_live(struct bpf_prog * prog,struct xdp_buff * ctx,u32 repeat,u32 batch_size,u32 * time)344 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
345 u32 repeat, u32 batch_size, u32 *time)
346
347 {
348 struct xdp_test_data xdp = { .batch_size = batch_size };
349 struct bpf_test_timer t = { .mode = NO_MIGRATE };
350 int ret;
351
352 if (!repeat)
353 repeat = 1;
354
355 ret = xdp_test_run_setup(&xdp, ctx);
356 if (ret)
357 return ret;
358
359 bpf_test_timer_enter(&t);
360 do {
361 xdp.frame_cnt = 0;
362 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
363 if (unlikely(ret < 0))
364 break;
365 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
366 bpf_test_timer_leave(&t);
367
368 xdp_test_run_teardown(&xdp);
369 return ret;
370 }
371
bpf_test_run(struct bpf_prog * prog,void * ctx,u32 repeat,u32 * retval,u32 * time,bool xdp)372 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
373 u32 *retval, u32 *time, bool xdp)
374 {
375 struct bpf_prog_array_item item = {.prog = prog};
376 struct bpf_run_ctx *old_ctx;
377 struct bpf_cg_run_ctx run_ctx;
378 struct bpf_test_timer t = { NO_MIGRATE };
379 enum bpf_cgroup_storage_type stype;
380 int ret;
381
382 for_each_cgroup_storage_type(stype) {
383 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
384 if (IS_ERR(item.cgroup_storage[stype])) {
385 item.cgroup_storage[stype] = NULL;
386 for_each_cgroup_storage_type(stype)
387 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
388 return -ENOMEM;
389 }
390 }
391
392 if (!repeat)
393 repeat = 1;
394
395 bpf_test_timer_enter(&t);
396 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
397 do {
398 run_ctx.prog_item = &item;
399 if (xdp)
400 *retval = bpf_prog_run_xdp(prog, ctx);
401 else
402 *retval = bpf_prog_run(prog, ctx);
403 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
404 bpf_reset_run_ctx(old_ctx);
405 bpf_test_timer_leave(&t);
406
407 for_each_cgroup_storage_type(stype)
408 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
409
410 return ret;
411 }
412
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)413 static int bpf_test_finish(const union bpf_attr *kattr,
414 union bpf_attr __user *uattr, const void *data,
415 struct skb_shared_info *sinfo, u32 size,
416 u32 retval, u32 duration)
417 {
418 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
419 int err = -EFAULT;
420 u32 copy_size = size;
421
422 /* Clamp copy if the user has provided a size hint, but copy the full
423 * buffer if not to retain old behaviour.
424 */
425 if (kattr->test.data_size_out &&
426 copy_size > kattr->test.data_size_out) {
427 copy_size = kattr->test.data_size_out;
428 err = -ENOSPC;
429 }
430
431 if (data_out) {
432 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
433
434 if (len < 0) {
435 err = -ENOSPC;
436 goto out;
437 }
438
439 if (copy_to_user(data_out, data, len))
440 goto out;
441
442 if (sinfo) {
443 int i, offset = len;
444 u32 data_len;
445
446 for (i = 0; i < sinfo->nr_frags; i++) {
447 skb_frag_t *frag = &sinfo->frags[i];
448
449 if (offset >= copy_size) {
450 err = -ENOSPC;
451 break;
452 }
453
454 data_len = min_t(u32, copy_size - offset,
455 skb_frag_size(frag));
456
457 if (copy_to_user(data_out + offset,
458 skb_frag_address(frag),
459 data_len))
460 goto out;
461
462 offset += data_len;
463 }
464 }
465 }
466
467 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
468 goto out;
469 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
470 goto out;
471 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
472 goto out;
473 if (err != -ENOSPC)
474 err = 0;
475 out:
476 trace_bpf_test_finish(&err);
477 return err;
478 }
479
480 /* Integer types of various sizes and pointer combinations cover variety of
481 * architecture dependent calling conventions. 7+ can be supported in the
482 * future.
483 */
484 __diag_push();
485 __diag_ignore_all("-Wmissing-prototypes",
486 "Global functions as their definitions will be in vmlinux BTF");
bpf_fentry_test1(int a)487 int noinline bpf_fentry_test1(int a)
488 {
489 return a + 1;
490 }
491 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
492 ALLOW_ERROR_INJECTION(bpf_fentry_test1, ERRNO);
493
bpf_fentry_test2(int a,u64 b)494 int noinline bpf_fentry_test2(int a, u64 b)
495 {
496 return a + b;
497 }
498
bpf_fentry_test3(char a,int b,u64 c)499 int noinline bpf_fentry_test3(char a, int b, u64 c)
500 {
501 return a + b + c;
502 }
503
bpf_fentry_test4(void * a,char b,int c,u64 d)504 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
505 {
506 return (long)a + b + c + d;
507 }
508
bpf_fentry_test5(u64 a,void * b,short c,int d,u64 e)509 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
510 {
511 return a + (long)b + c + d + e;
512 }
513
bpf_fentry_test6(u64 a,void * b,short c,int d,void * e,u64 f)514 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
515 {
516 return a + (long)b + c + d + (long)e + f;
517 }
518
519 struct bpf_fentry_test_t {
520 struct bpf_fentry_test_t *a;
521 };
522
bpf_fentry_test7(struct bpf_fentry_test_t * arg)523 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
524 {
525 return (long)arg;
526 }
527
bpf_fentry_test8(struct bpf_fentry_test_t * arg)528 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
529 {
530 return (long)arg->a;
531 }
532
bpf_modify_return_test(int a,int * b)533 int noinline bpf_modify_return_test(int a, int *b)
534 {
535 *b += 1;
536 return a + *b;
537 }
538
bpf_kfunc_call_test1(struct sock * sk,u32 a,u64 b,u32 c,u64 d)539 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
540 {
541 return a + b + c + d;
542 }
543
bpf_kfunc_call_test2(struct sock * sk,u32 a,u32 b)544 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
545 {
546 return a + b;
547 }
548
bpf_kfunc_call_test3(struct sock * sk)549 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
550 {
551 return sk;
552 }
553
554 struct prog_test_member1 {
555 int a;
556 };
557
558 struct prog_test_member {
559 struct prog_test_member1 m;
560 int c;
561 };
562
563 struct prog_test_ref_kfunc {
564 int a;
565 int b;
566 struct prog_test_member memb;
567 struct prog_test_ref_kfunc *next;
568 refcount_t cnt;
569 };
570
571 static struct prog_test_ref_kfunc prog_test_struct = {
572 .a = 42,
573 .b = 108,
574 .next = &prog_test_struct,
575 .cnt = REFCOUNT_INIT(1),
576 };
577
578 noinline struct prog_test_ref_kfunc *
bpf_kfunc_call_test_acquire(unsigned long * scalar_ptr)579 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
580 {
581 refcount_inc(&prog_test_struct.cnt);
582 return &prog_test_struct;
583 }
584
585 noinline struct prog_test_member *
bpf_kfunc_call_memb_acquire(void)586 bpf_kfunc_call_memb_acquire(void)
587 {
588 WARN_ON_ONCE(1);
589 return NULL;
590 }
591
bpf_kfunc_call_test_release(struct prog_test_ref_kfunc * p)592 noinline void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
593 {
594 if (!p)
595 return;
596
597 refcount_dec(&p->cnt);
598 }
599
bpf_kfunc_call_memb_release(struct prog_test_member * p)600 noinline void bpf_kfunc_call_memb_release(struct prog_test_member *p)
601 {
602 }
603
bpf_kfunc_call_memb1_release(struct prog_test_member1 * p)604 noinline void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
605 {
606 WARN_ON_ONCE(1);
607 }
608
__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc * p,const int size)609 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size)
610 {
611 if (size > 2 * sizeof(int))
612 return NULL;
613
614 return (int *)p;
615 }
616
bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc * p,const int rdwr_buf_size)617 noinline int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p, const int rdwr_buf_size)
618 {
619 return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size);
620 }
621
bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc * p,const int rdonly_buf_size)622 noinline int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p, const int rdonly_buf_size)
623 {
624 return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
625 }
626
627 /* the next 2 ones can't be really used for testing expect to ensure
628 * that the verifier rejects the call.
629 * Acquire functions must return struct pointers, so these ones are
630 * failing.
631 */
bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc * p,const int rdonly_buf_size)632 noinline int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p, const int rdonly_buf_size)
633 {
634 return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
635 }
636
bpf_kfunc_call_int_mem_release(int * p)637 noinline void bpf_kfunc_call_int_mem_release(int *p)
638 {
639 }
640
641 noinline struct prog_test_ref_kfunc *
bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc ** pp,int a,int b)642 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
643 {
644 struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
645
646 if (!p)
647 return NULL;
648 refcount_inc(&p->cnt);
649 return p;
650 }
651
652 struct prog_test_pass1 {
653 int x0;
654 struct {
655 int x1;
656 struct {
657 int x2;
658 struct {
659 int x3;
660 };
661 };
662 };
663 };
664
665 struct prog_test_pass2 {
666 int len;
667 short arr1[4];
668 struct {
669 char arr2[4];
670 unsigned long arr3[8];
671 } x;
672 };
673
674 struct prog_test_fail1 {
675 void *p;
676 int x;
677 };
678
679 struct prog_test_fail2 {
680 int x8;
681 struct prog_test_pass1 x;
682 };
683
684 struct prog_test_fail3 {
685 int len;
686 char arr1[2];
687 char arr2[];
688 };
689
bpf_kfunc_call_test_pass_ctx(struct __sk_buff * skb)690 noinline void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
691 {
692 }
693
bpf_kfunc_call_test_pass1(struct prog_test_pass1 * p)694 noinline void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
695 {
696 }
697
bpf_kfunc_call_test_pass2(struct prog_test_pass2 * p)698 noinline void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
699 {
700 }
701
bpf_kfunc_call_test_fail1(struct prog_test_fail1 * p)702 noinline void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
703 {
704 }
705
bpf_kfunc_call_test_fail2(struct prog_test_fail2 * p)706 noinline void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
707 {
708 }
709
bpf_kfunc_call_test_fail3(struct prog_test_fail3 * p)710 noinline void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
711 {
712 }
713
bpf_kfunc_call_test_mem_len_pass1(void * mem,int mem__sz)714 noinline void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
715 {
716 }
717
bpf_kfunc_call_test_mem_len_fail1(void * mem,int len)718 noinline void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
719 {
720 }
721
bpf_kfunc_call_test_mem_len_fail2(u64 * mem,int len)722 noinline void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
723 {
724 }
725
bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc * p)726 noinline void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
727 {
728 }
729
bpf_kfunc_call_test_destructive(void)730 noinline void bpf_kfunc_call_test_destructive(void)
731 {
732 }
733
734 __diag_pop();
735
736 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
737
738 BTF_SET8_START(test_sk_check_kfunc_ids)
BTF_ID_FLAGS(func,bpf_kfunc_call_test1)739 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
740 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
741 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
742 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
743 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
744 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
745 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
746 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
747 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL)
748 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL)
749 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL)
750 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE)
751 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
752 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
753 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
754 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
755 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
756 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
757 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
758 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
759 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
760 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
761 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS)
762 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
763 BTF_SET8_END(test_sk_check_kfunc_ids)
764
765 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
766 u32 size, u32 headroom, u32 tailroom)
767 {
768 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
769 void *data;
770
771 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
772 return ERR_PTR(-EINVAL);
773
774 if (user_size > size)
775 return ERR_PTR(-EMSGSIZE);
776
777 size = SKB_DATA_ALIGN(size);
778 data = kzalloc(size + headroom + tailroom, GFP_USER);
779 if (!data)
780 return ERR_PTR(-ENOMEM);
781
782 if (copy_from_user(data + headroom, data_in, user_size)) {
783 kfree(data);
784 return ERR_PTR(-EFAULT);
785 }
786
787 return data;
788 }
789
bpf_prog_test_run_tracing(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)790 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
791 const union bpf_attr *kattr,
792 union bpf_attr __user *uattr)
793 {
794 struct bpf_fentry_test_t arg = {};
795 u16 side_effect = 0, ret = 0;
796 int b = 2, err = -EFAULT;
797 u32 retval = 0;
798
799 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
800 return -EINVAL;
801
802 switch (prog->expected_attach_type) {
803 case BPF_TRACE_FENTRY:
804 case BPF_TRACE_FEXIT:
805 if (bpf_fentry_test1(1) != 2 ||
806 bpf_fentry_test2(2, 3) != 5 ||
807 bpf_fentry_test3(4, 5, 6) != 15 ||
808 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
809 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
810 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
811 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
812 bpf_fentry_test8(&arg) != 0)
813 goto out;
814 break;
815 case BPF_MODIFY_RETURN:
816 ret = bpf_modify_return_test(1, &b);
817 if (b != 2)
818 side_effect = 1;
819 break;
820 default:
821 goto out;
822 }
823
824 retval = ((u32)side_effect << 16) | ret;
825 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
826 goto out;
827
828 err = 0;
829 out:
830 trace_bpf_test_finish(&err);
831 return err;
832 }
833
834 struct bpf_raw_tp_test_run_info {
835 struct bpf_prog *prog;
836 void *ctx;
837 u32 retval;
838 };
839
840 static void
__bpf_prog_test_run_raw_tp(void * data)841 __bpf_prog_test_run_raw_tp(void *data)
842 {
843 struct bpf_raw_tp_test_run_info *info = data;
844
845 rcu_read_lock();
846 info->retval = bpf_prog_run(info->prog, info->ctx);
847 rcu_read_unlock();
848 }
849
bpf_prog_test_run_raw_tp(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)850 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
851 const union bpf_attr *kattr,
852 union bpf_attr __user *uattr)
853 {
854 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
855 __u32 ctx_size_in = kattr->test.ctx_size_in;
856 struct bpf_raw_tp_test_run_info info;
857 int cpu = kattr->test.cpu, err = 0;
858 int current_cpu;
859
860 /* doesn't support data_in/out, ctx_out, duration, or repeat */
861 if (kattr->test.data_in || kattr->test.data_out ||
862 kattr->test.ctx_out || kattr->test.duration ||
863 kattr->test.repeat || kattr->test.batch_size)
864 return -EINVAL;
865
866 if (ctx_size_in < prog->aux->max_ctx_offset ||
867 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
868 return -EINVAL;
869
870 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
871 return -EINVAL;
872
873 if (ctx_size_in) {
874 info.ctx = memdup_user(ctx_in, ctx_size_in);
875 if (IS_ERR(info.ctx))
876 return PTR_ERR(info.ctx);
877 } else {
878 info.ctx = NULL;
879 }
880
881 info.prog = prog;
882
883 current_cpu = get_cpu();
884 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
885 cpu == current_cpu) {
886 __bpf_prog_test_run_raw_tp(&info);
887 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
888 /* smp_call_function_single() also checks cpu_online()
889 * after csd_lock(). However, since cpu is from user
890 * space, let's do an extra quick check to filter out
891 * invalid value before smp_call_function_single().
892 */
893 err = -ENXIO;
894 } else {
895 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
896 &info, 1);
897 }
898 put_cpu();
899
900 if (!err &&
901 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
902 err = -EFAULT;
903
904 kfree(info.ctx);
905 return err;
906 }
907
bpf_ctx_init(const union bpf_attr * kattr,u32 max_size)908 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
909 {
910 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
911 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
912 u32 size = kattr->test.ctx_size_in;
913 void *data;
914 int err;
915
916 if (!data_in && !data_out)
917 return NULL;
918
919 data = kzalloc(max_size, GFP_USER);
920 if (!data)
921 return ERR_PTR(-ENOMEM);
922
923 if (data_in) {
924 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
925 if (err) {
926 kfree(data);
927 return ERR_PTR(err);
928 }
929
930 size = min_t(u32, max_size, size);
931 if (copy_from_user(data, data_in, size)) {
932 kfree(data);
933 return ERR_PTR(-EFAULT);
934 }
935 }
936 return data;
937 }
938
bpf_ctx_finish(const union bpf_attr * kattr,union bpf_attr __user * uattr,const void * data,u32 size)939 static int bpf_ctx_finish(const union bpf_attr *kattr,
940 union bpf_attr __user *uattr, const void *data,
941 u32 size)
942 {
943 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
944 int err = -EFAULT;
945 u32 copy_size = size;
946
947 if (!data || !data_out)
948 return 0;
949
950 if (copy_size > kattr->test.ctx_size_out) {
951 copy_size = kattr->test.ctx_size_out;
952 err = -ENOSPC;
953 }
954
955 if (copy_to_user(data_out, data, copy_size))
956 goto out;
957 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
958 goto out;
959 if (err != -ENOSPC)
960 err = 0;
961 out:
962 return err;
963 }
964
965 /**
966 * range_is_zero - test whether buffer is initialized
967 * @buf: buffer to check
968 * @from: check from this position
969 * @to: check up until (excluding) this position
970 *
971 * This function returns true if the there is a non-zero byte
972 * in the buf in the range [from,to).
973 */
range_is_zero(void * buf,size_t from,size_t to)974 static inline bool range_is_zero(void *buf, size_t from, size_t to)
975 {
976 return !memchr_inv((u8 *)buf + from, 0, to - from);
977 }
978
convert___skb_to_skb(struct sk_buff * skb,struct __sk_buff * __skb)979 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
980 {
981 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
982
983 if (!__skb)
984 return 0;
985
986 /* make sure the fields we don't use are zeroed */
987 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
988 return -EINVAL;
989
990 /* mark is allowed */
991
992 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
993 offsetof(struct __sk_buff, priority)))
994 return -EINVAL;
995
996 /* priority is allowed */
997 /* ingress_ifindex is allowed */
998 /* ifindex is allowed */
999
1000 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1001 offsetof(struct __sk_buff, cb)))
1002 return -EINVAL;
1003
1004 /* cb is allowed */
1005
1006 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1007 offsetof(struct __sk_buff, tstamp)))
1008 return -EINVAL;
1009
1010 /* tstamp is allowed */
1011 /* wire_len is allowed */
1012 /* gso_segs is allowed */
1013
1014 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1015 offsetof(struct __sk_buff, gso_size)))
1016 return -EINVAL;
1017
1018 /* gso_size is allowed */
1019
1020 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1021 offsetof(struct __sk_buff, hwtstamp)))
1022 return -EINVAL;
1023
1024 /* hwtstamp is allowed */
1025
1026 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1027 sizeof(struct __sk_buff)))
1028 return -EINVAL;
1029
1030 skb->mark = __skb->mark;
1031 skb->priority = __skb->priority;
1032 skb->skb_iif = __skb->ingress_ifindex;
1033 skb->tstamp = __skb->tstamp;
1034 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1035
1036 if (__skb->wire_len == 0) {
1037 cb->pkt_len = skb->len;
1038 } else {
1039 if (__skb->wire_len < skb->len ||
1040 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1041 return -EINVAL;
1042 cb->pkt_len = __skb->wire_len;
1043 }
1044
1045 if (__skb->gso_segs > GSO_MAX_SEGS)
1046 return -EINVAL;
1047 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1048 skb_shinfo(skb)->gso_size = __skb->gso_size;
1049 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1050
1051 return 0;
1052 }
1053
convert_skb_to___skb(struct sk_buff * skb,struct __sk_buff * __skb)1054 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1055 {
1056 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1057
1058 if (!__skb)
1059 return;
1060
1061 __skb->mark = skb->mark;
1062 __skb->priority = skb->priority;
1063 __skb->ingress_ifindex = skb->skb_iif;
1064 __skb->ifindex = skb->dev->ifindex;
1065 __skb->tstamp = skb->tstamp;
1066 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1067 __skb->wire_len = cb->pkt_len;
1068 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
1069 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1070 }
1071
1072 static struct proto bpf_dummy_proto = {
1073 .name = "bpf_dummy",
1074 .owner = THIS_MODULE,
1075 .obj_size = sizeof(struct sock),
1076 };
1077
bpf_prog_test_run_skb(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)1078 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1079 union bpf_attr __user *uattr)
1080 {
1081 bool is_l2 = false, is_direct_pkt_access = false;
1082 struct net *net = current->nsproxy->net_ns;
1083 struct net_device *dev = net->loopback_dev;
1084 u32 size = kattr->test.data_size_in;
1085 u32 repeat = kattr->test.repeat;
1086 struct __sk_buff *ctx = NULL;
1087 u32 retval, duration;
1088 int hh_len = ETH_HLEN;
1089 struct sk_buff *skb;
1090 struct sock *sk;
1091 void *data;
1092 int ret;
1093
1094 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1095 return -EINVAL;
1096
1097 data = bpf_test_init(kattr, kattr->test.data_size_in,
1098 size, NET_SKB_PAD + NET_IP_ALIGN,
1099 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1100 if (IS_ERR(data))
1101 return PTR_ERR(data);
1102
1103 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1104 if (IS_ERR(ctx)) {
1105 kfree(data);
1106 return PTR_ERR(ctx);
1107 }
1108
1109 switch (prog->type) {
1110 case BPF_PROG_TYPE_SCHED_CLS:
1111 case BPF_PROG_TYPE_SCHED_ACT:
1112 is_l2 = true;
1113 fallthrough;
1114 case BPF_PROG_TYPE_LWT_IN:
1115 case BPF_PROG_TYPE_LWT_OUT:
1116 case BPF_PROG_TYPE_LWT_XMIT:
1117 is_direct_pkt_access = true;
1118 break;
1119 default:
1120 break;
1121 }
1122
1123 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1124 if (!sk) {
1125 kfree(data);
1126 kfree(ctx);
1127 return -ENOMEM;
1128 }
1129 sock_init_data(NULL, sk);
1130
1131 skb = build_skb(data, 0);
1132 if (!skb) {
1133 kfree(data);
1134 kfree(ctx);
1135 sk_free(sk);
1136 return -ENOMEM;
1137 }
1138 skb->sk = sk;
1139
1140 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1141 __skb_put(skb, size);
1142 if (ctx && ctx->ifindex > 1) {
1143 dev = dev_get_by_index(net, ctx->ifindex);
1144 if (!dev) {
1145 ret = -ENODEV;
1146 goto out;
1147 }
1148 }
1149 skb->protocol = eth_type_trans(skb, dev);
1150 skb_reset_network_header(skb);
1151
1152 switch (skb->protocol) {
1153 case htons(ETH_P_IP):
1154 sk->sk_family = AF_INET;
1155 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1156 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1157 sk->sk_daddr = ip_hdr(skb)->daddr;
1158 }
1159 break;
1160 #if IS_ENABLED(CONFIG_IPV6)
1161 case htons(ETH_P_IPV6):
1162 sk->sk_family = AF_INET6;
1163 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1164 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1165 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1166 }
1167 break;
1168 #endif
1169 default:
1170 break;
1171 }
1172
1173 if (is_l2)
1174 __skb_push(skb, hh_len);
1175 if (is_direct_pkt_access)
1176 bpf_compute_data_pointers(skb);
1177 ret = convert___skb_to_skb(skb, ctx);
1178 if (ret)
1179 goto out;
1180 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1181 if (ret)
1182 goto out;
1183 if (!is_l2) {
1184 if (skb_headroom(skb) < hh_len) {
1185 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1186
1187 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1188 ret = -ENOMEM;
1189 goto out;
1190 }
1191 }
1192 memset(__skb_push(skb, hh_len), 0, hh_len);
1193 }
1194 convert_skb_to___skb(skb, ctx);
1195
1196 size = skb->len;
1197 /* bpf program can never convert linear skb to non-linear */
1198 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1199 size = skb_headlen(skb);
1200 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1201 duration);
1202 if (!ret)
1203 ret = bpf_ctx_finish(kattr, uattr, ctx,
1204 sizeof(struct __sk_buff));
1205 out:
1206 if (dev && dev != net->loopback_dev)
1207 dev_put(dev);
1208 kfree_skb(skb);
1209 sk_free(sk);
1210 kfree(ctx);
1211 return ret;
1212 }
1213
xdp_convert_md_to_buff(struct xdp_md * xdp_md,struct xdp_buff * xdp)1214 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1215 {
1216 unsigned int ingress_ifindex, rx_queue_index;
1217 struct netdev_rx_queue *rxqueue;
1218 struct net_device *device;
1219
1220 if (!xdp_md)
1221 return 0;
1222
1223 if (xdp_md->egress_ifindex != 0)
1224 return -EINVAL;
1225
1226 ingress_ifindex = xdp_md->ingress_ifindex;
1227 rx_queue_index = xdp_md->rx_queue_index;
1228
1229 if (!ingress_ifindex && rx_queue_index)
1230 return -EINVAL;
1231
1232 if (ingress_ifindex) {
1233 device = dev_get_by_index(current->nsproxy->net_ns,
1234 ingress_ifindex);
1235 if (!device)
1236 return -ENODEV;
1237
1238 if (rx_queue_index >= device->real_num_rx_queues)
1239 goto free_dev;
1240
1241 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1242
1243 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1244 goto free_dev;
1245
1246 xdp->rxq = &rxqueue->xdp_rxq;
1247 /* The device is now tracked in the xdp->rxq for later
1248 * dev_put()
1249 */
1250 }
1251
1252 xdp->data = xdp->data_meta + xdp_md->data;
1253 return 0;
1254
1255 free_dev:
1256 dev_put(device);
1257 return -EINVAL;
1258 }
1259
xdp_convert_buff_to_md(struct xdp_buff * xdp,struct xdp_md * xdp_md)1260 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1261 {
1262 if (!xdp_md)
1263 return;
1264
1265 xdp_md->data = xdp->data - xdp->data_meta;
1266 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1267
1268 if (xdp_md->ingress_ifindex)
1269 dev_put(xdp->rxq->dev);
1270 }
1271
bpf_prog_test_run_xdp(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)1272 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1273 union bpf_attr __user *uattr)
1274 {
1275 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1276 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1277 u32 batch_size = kattr->test.batch_size;
1278 u32 retval = 0, duration, max_data_sz;
1279 u32 size = kattr->test.data_size_in;
1280 u32 headroom = XDP_PACKET_HEADROOM;
1281 u32 repeat = kattr->test.repeat;
1282 struct netdev_rx_queue *rxqueue;
1283 struct skb_shared_info *sinfo;
1284 struct xdp_buff xdp = {};
1285 int i, ret = -EINVAL;
1286 struct xdp_md *ctx;
1287 void *data;
1288
1289 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1290 prog->expected_attach_type == BPF_XDP_CPUMAP)
1291 return -EINVAL;
1292
1293 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1294 return -EINVAL;
1295
1296 if (do_live) {
1297 if (!batch_size)
1298 batch_size = NAPI_POLL_WEIGHT;
1299 else if (batch_size > TEST_XDP_MAX_BATCH)
1300 return -E2BIG;
1301
1302 headroom += sizeof(struct xdp_page_head);
1303 } else if (batch_size) {
1304 return -EINVAL;
1305 }
1306
1307 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1308 if (IS_ERR(ctx))
1309 return PTR_ERR(ctx);
1310
1311 if (ctx) {
1312 /* There can't be user provided data before the meta data */
1313 if (ctx->data_meta || ctx->data_end != size ||
1314 ctx->data > ctx->data_end ||
1315 unlikely(xdp_metalen_invalid(ctx->data)) ||
1316 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1317 goto free_ctx;
1318 /* Meta data is allocated from the headroom */
1319 headroom -= ctx->data;
1320 }
1321
1322 max_data_sz = 4096 - headroom - tailroom;
1323 if (size > max_data_sz) {
1324 /* disallow live data mode for jumbo frames */
1325 if (do_live)
1326 goto free_ctx;
1327 size = max_data_sz;
1328 }
1329
1330 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1331 if (IS_ERR(data)) {
1332 ret = PTR_ERR(data);
1333 goto free_ctx;
1334 }
1335
1336 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1337 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1338 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1339 xdp_prepare_buff(&xdp, data, headroom, size, true);
1340 sinfo = xdp_get_shared_info_from_buff(&xdp);
1341
1342 ret = xdp_convert_md_to_buff(ctx, &xdp);
1343 if (ret)
1344 goto free_data;
1345
1346 if (unlikely(kattr->test.data_size_in > size)) {
1347 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1348
1349 while (size < kattr->test.data_size_in) {
1350 struct page *page;
1351 skb_frag_t *frag;
1352 u32 data_len;
1353
1354 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1355 ret = -ENOMEM;
1356 goto out;
1357 }
1358
1359 page = alloc_page(GFP_KERNEL);
1360 if (!page) {
1361 ret = -ENOMEM;
1362 goto out;
1363 }
1364
1365 frag = &sinfo->frags[sinfo->nr_frags++];
1366 __skb_frag_set_page(frag, page);
1367
1368 data_len = min_t(u32, kattr->test.data_size_in - size,
1369 PAGE_SIZE);
1370 skb_frag_size_set(frag, data_len);
1371
1372 if (copy_from_user(page_address(page), data_in + size,
1373 data_len)) {
1374 ret = -EFAULT;
1375 goto out;
1376 }
1377 sinfo->xdp_frags_size += data_len;
1378 size += data_len;
1379 }
1380 xdp_buff_set_frags_flag(&xdp);
1381 }
1382
1383 if (repeat > 1)
1384 bpf_prog_change_xdp(NULL, prog);
1385
1386 if (do_live)
1387 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1388 else
1389 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1390 /* We convert the xdp_buff back to an xdp_md before checking the return
1391 * code so the reference count of any held netdevice will be decremented
1392 * even if the test run failed.
1393 */
1394 xdp_convert_buff_to_md(&xdp, ctx);
1395 if (ret)
1396 goto out;
1397
1398 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1399 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1400 retval, duration);
1401 if (!ret)
1402 ret = bpf_ctx_finish(kattr, uattr, ctx,
1403 sizeof(struct xdp_md));
1404
1405 out:
1406 if (repeat > 1)
1407 bpf_prog_change_xdp(prog, NULL);
1408 free_data:
1409 for (i = 0; i < sinfo->nr_frags; i++)
1410 __free_page(skb_frag_page(&sinfo->frags[i]));
1411 kfree(data);
1412 free_ctx:
1413 kfree(ctx);
1414 return ret;
1415 }
1416
verify_user_bpf_flow_keys(struct bpf_flow_keys * ctx)1417 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1418 {
1419 /* make sure the fields we don't use are zeroed */
1420 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1421 return -EINVAL;
1422
1423 /* flags is allowed */
1424
1425 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1426 sizeof(struct bpf_flow_keys)))
1427 return -EINVAL;
1428
1429 return 0;
1430 }
1431
bpf_prog_test_run_flow_dissector(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)1432 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1433 const union bpf_attr *kattr,
1434 union bpf_attr __user *uattr)
1435 {
1436 struct bpf_test_timer t = { NO_PREEMPT };
1437 u32 size = kattr->test.data_size_in;
1438 struct bpf_flow_dissector ctx = {};
1439 u32 repeat = kattr->test.repeat;
1440 struct bpf_flow_keys *user_ctx;
1441 struct bpf_flow_keys flow_keys;
1442 const struct ethhdr *eth;
1443 unsigned int flags = 0;
1444 u32 retval, duration;
1445 void *data;
1446 int ret;
1447
1448 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1449 return -EINVAL;
1450
1451 if (size < ETH_HLEN)
1452 return -EINVAL;
1453
1454 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1455 if (IS_ERR(data))
1456 return PTR_ERR(data);
1457
1458 eth = (struct ethhdr *)data;
1459
1460 if (!repeat)
1461 repeat = 1;
1462
1463 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1464 if (IS_ERR(user_ctx)) {
1465 kfree(data);
1466 return PTR_ERR(user_ctx);
1467 }
1468 if (user_ctx) {
1469 ret = verify_user_bpf_flow_keys(user_ctx);
1470 if (ret)
1471 goto out;
1472 flags = user_ctx->flags;
1473 }
1474
1475 ctx.flow_keys = &flow_keys;
1476 ctx.data = data;
1477 ctx.data_end = (__u8 *)data + size;
1478
1479 bpf_test_timer_enter(&t);
1480 do {
1481 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1482 size, flags);
1483 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1484 bpf_test_timer_leave(&t);
1485
1486 if (ret < 0)
1487 goto out;
1488
1489 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1490 sizeof(flow_keys), retval, duration);
1491 if (!ret)
1492 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1493 sizeof(struct bpf_flow_keys));
1494
1495 out:
1496 kfree(user_ctx);
1497 kfree(data);
1498 return ret;
1499 }
1500
bpf_prog_test_run_sk_lookup(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)1501 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1502 union bpf_attr __user *uattr)
1503 {
1504 struct bpf_test_timer t = { NO_PREEMPT };
1505 struct bpf_prog_array *progs = NULL;
1506 struct bpf_sk_lookup_kern ctx = {};
1507 u32 repeat = kattr->test.repeat;
1508 struct bpf_sk_lookup *user_ctx;
1509 u32 retval, duration;
1510 int ret = -EINVAL;
1511
1512 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1513 return -EINVAL;
1514
1515 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1516 kattr->test.data_size_out)
1517 return -EINVAL;
1518
1519 if (!repeat)
1520 repeat = 1;
1521
1522 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1523 if (IS_ERR(user_ctx))
1524 return PTR_ERR(user_ctx);
1525
1526 if (!user_ctx)
1527 return -EINVAL;
1528
1529 if (user_ctx->sk)
1530 goto out;
1531
1532 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1533 goto out;
1534
1535 if (user_ctx->local_port > U16_MAX) {
1536 ret = -ERANGE;
1537 goto out;
1538 }
1539
1540 ctx.family = (u16)user_ctx->family;
1541 ctx.protocol = (u16)user_ctx->protocol;
1542 ctx.dport = (u16)user_ctx->local_port;
1543 ctx.sport = user_ctx->remote_port;
1544
1545 switch (ctx.family) {
1546 case AF_INET:
1547 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1548 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1549 break;
1550
1551 #if IS_ENABLED(CONFIG_IPV6)
1552 case AF_INET6:
1553 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1554 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1555 break;
1556 #endif
1557
1558 default:
1559 ret = -EAFNOSUPPORT;
1560 goto out;
1561 }
1562
1563 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1564 if (!progs) {
1565 ret = -ENOMEM;
1566 goto out;
1567 }
1568
1569 progs->items[0].prog = prog;
1570
1571 bpf_test_timer_enter(&t);
1572 do {
1573 ctx.selected_sk = NULL;
1574 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1575 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1576 bpf_test_timer_leave(&t);
1577
1578 if (ret < 0)
1579 goto out;
1580
1581 user_ctx->cookie = 0;
1582 if (ctx.selected_sk) {
1583 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1584 ret = -EOPNOTSUPP;
1585 goto out;
1586 }
1587
1588 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1589 }
1590
1591 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1592 if (!ret)
1593 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1594
1595 out:
1596 bpf_prog_array_free(progs);
1597 kfree(user_ctx);
1598 return ret;
1599 }
1600
bpf_prog_test_run_syscall(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)1601 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1602 const union bpf_attr *kattr,
1603 union bpf_attr __user *uattr)
1604 {
1605 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1606 __u32 ctx_size_in = kattr->test.ctx_size_in;
1607 void *ctx = NULL;
1608 u32 retval;
1609 int err = 0;
1610
1611 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1612 if (kattr->test.data_in || kattr->test.data_out ||
1613 kattr->test.ctx_out || kattr->test.duration ||
1614 kattr->test.repeat || kattr->test.flags ||
1615 kattr->test.batch_size)
1616 return -EINVAL;
1617
1618 if (ctx_size_in < prog->aux->max_ctx_offset ||
1619 ctx_size_in > U16_MAX)
1620 return -EINVAL;
1621
1622 if (ctx_size_in) {
1623 ctx = memdup_user(ctx_in, ctx_size_in);
1624 if (IS_ERR(ctx))
1625 return PTR_ERR(ctx);
1626 }
1627
1628 rcu_read_lock_trace();
1629 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1630 rcu_read_unlock_trace();
1631
1632 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1633 err = -EFAULT;
1634 goto out;
1635 }
1636 if (ctx_size_in)
1637 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1638 err = -EFAULT;
1639 out:
1640 kfree(ctx);
1641 return err;
1642 }
1643
1644 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1645 .owner = THIS_MODULE,
1646 .set = &test_sk_check_kfunc_ids,
1647 };
1648
1649 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
BTF_ID(struct,prog_test_ref_kfunc)1650 BTF_ID(struct, prog_test_ref_kfunc)
1651 BTF_ID(func, bpf_kfunc_call_test_release)
1652 BTF_ID(struct, prog_test_member)
1653 BTF_ID(func, bpf_kfunc_call_memb_release)
1654
1655 static int __init bpf_prog_test_run_init(void)
1656 {
1657 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1658 {
1659 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1660 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1661 },
1662 {
1663 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1664 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1665 },
1666 };
1667 int ret;
1668
1669 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1670 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1671 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1672 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1673 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1674 THIS_MODULE);
1675 }
1676 late_initcall(bpf_prog_test_run_init);
1677