hc
2024-12-19 9370bb92b2d16684ee45cf24e879c93c509162da
kernel/net/bpf/test_run.c
....@@ -1,8 +1,5 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /* Copyright (c) 2017 Facebook
2
- *
3
- * This program is free software; you can redistribute it and/or
4
- * modify it under the terms of version 2 of the GNU General Public
5
- * License as published by the Free Software Foundation.
63 */
74 #include <linux/bpf.h>
85 #include <linux/slab.h>
....@@ -10,50 +7,121 @@
107 #include <linux/etherdevice.h>
118 #include <linux/filter.h>
129 #include <linux/sched/signal.h>
10
+#include <net/bpf_sk_storage.h>
11
+#include <net/sock.h>
12
+#include <net/tcp.h>
13
+#include <net/net_namespace.h>
14
+#include <linux/error-injection.h>
15
+#include <linux/smp.h>
16
+#include <linux/sock_diag.h>
1317
14
-static __always_inline u32 bpf_test_run_one(struct bpf_prog *prog, void *ctx,
15
- struct bpf_cgroup_storage *storage)
18
+#define CREATE_TRACE_POINTS
19
+#include <trace/events/bpf_test_run.h>
20
+
21
+struct bpf_test_timer {
22
+ enum { NO_PREEMPT, NO_MIGRATE } mode;
23
+ u32 i;
24
+ u64 time_start, time_spent;
25
+};
26
+
27
+static void bpf_test_timer_enter(struct bpf_test_timer *t)
28
+ __acquires(rcu)
1629 {
17
- u32 ret;
18
-
19
- preempt_disable();
2030 rcu_read_lock();
21
- bpf_cgroup_storage_set(storage);
22
- ret = BPF_PROG_RUN(prog, ctx);
23
- rcu_read_unlock();
24
- preempt_enable();
31
+ if (t->mode == NO_PREEMPT)
32
+ preempt_disable();
33
+ else
34
+ migrate_disable();
2535
26
- return ret;
36
+ t->time_start = ktime_get_ns();
2737 }
2838
29
-static u32 bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat, u32 *time)
39
+static void bpf_test_timer_leave(struct bpf_test_timer *t)
40
+ __releases(rcu)
3041 {
31
- struct bpf_cgroup_storage *storage = NULL;
32
- u64 time_start, time_spent = 0;
33
- u32 ret = 0, i;
42
+ t->time_start = 0;
3443
35
- storage = bpf_cgroup_storage_alloc(prog);
36
- if (IS_ERR(storage))
37
- return PTR_ERR(storage);
44
+ if (t->mode == NO_PREEMPT)
45
+ preempt_enable();
46
+ else
47
+ migrate_enable();
48
+ rcu_read_unlock();
49
+}
50
+
51
+static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration)
52
+ __must_hold(rcu)
53
+{
54
+ t->i++;
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
+static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
87
+ u32 *retval, u32 *time, bool xdp)
88
+{
89
+ struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
90
+ struct bpf_test_timer t = { NO_MIGRATE };
91
+ enum bpf_cgroup_storage_type stype;
92
+ int ret;
93
+
94
+ for_each_cgroup_storage_type(stype) {
95
+ storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
96
+ if (IS_ERR(storage[stype])) {
97
+ storage[stype] = NULL;
98
+ for_each_cgroup_storage_type(stype)
99
+ bpf_cgroup_storage_free(storage[stype]);
100
+ return -ENOMEM;
101
+ }
102
+ }
38103
39104 if (!repeat)
40105 repeat = 1;
41
- time_start = ktime_get_ns();
42
- for (i = 0; i < repeat; i++) {
43
- ret = bpf_test_run_one(prog, ctx, storage);
44
- if (need_resched()) {
45
- if (signal_pending(current))
46
- break;
47
- time_spent += ktime_get_ns() - time_start;
48
- cond_resched();
49
- time_start = ktime_get_ns();
50
- }
51
- }
52
- time_spent += ktime_get_ns() - time_start;
53
- do_div(time_spent, repeat);
54
- *time = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
55106
56
- bpf_cgroup_storage_free(storage);
107
+ bpf_test_timer_enter(&t);
108
+ do {
109
+ ret = bpf_cgroup_storage_set(storage);
110
+ if (ret)
111
+ break;
112
+
113
+ if (xdp)
114
+ *retval = bpf_prog_run_xdp(prog, ctx);
115
+ else
116
+ *retval = BPF_PROG_RUN(prog, ctx);
117
+
118
+ bpf_cgroup_storage_unset();
119
+
120
+ } while (bpf_test_timer_continue(&t, repeat, &ret, time));
121
+ bpf_test_timer_leave(&t);
122
+
123
+ for_each_cgroup_storage_type(stype)
124
+ bpf_cgroup_storage_free(storage[stype]);
57125
58126 return ret;
59127 }
....@@ -64,8 +132,18 @@
64132 {
65133 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
66134 int err = -EFAULT;
135
+ u32 copy_size = size;
67136
68
- if (data_out && copy_to_user(data_out, data, size))
137
+ /* Clamp copy if the user has provided a size hint, but copy the full
138
+ * buffer if not to retain old behaviour.
139
+ */
140
+ if (kattr->test.data_size_out &&
141
+ copy_size > kattr->test.data_size_out) {
142
+ copy_size = kattr->test.data_size_out;
143
+ err = -ENOSPC;
144
+ }
145
+
146
+ if (data_out && copy_to_user(data_out, data, copy_size))
69147 goto out;
70148 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
71149 goto out;
....@@ -73,53 +151,421 @@
73151 goto out;
74152 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
75153 goto out;
76
- err = 0;
154
+ if (err != -ENOSPC)
155
+ err = 0;
77156 out:
157
+ trace_bpf_test_finish(&err);
78158 return err;
79159 }
160
+
161
+/* Integer types of various sizes and pointer combinations cover variety of
162
+ * architecture dependent calling conventions. 7+ can be supported in the
163
+ * future.
164
+ */
165
+__diag_push();
166
+__diag_ignore(GCC, 8, "-Wmissing-prototypes",
167
+ "Global functions as their definitions will be in vmlinux BTF");
168
+int noinline bpf_fentry_test1(int a)
169
+{
170
+ return a + 1;
171
+}
172
+
173
+int noinline bpf_fentry_test2(int a, u64 b)
174
+{
175
+ return a + b;
176
+}
177
+
178
+int noinline bpf_fentry_test3(char a, int b, u64 c)
179
+{
180
+ return a + b + c;
181
+}
182
+
183
+int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
184
+{
185
+ return (long)a + b + c + d;
186
+}
187
+
188
+int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
189
+{
190
+ return a + (long)b + c + d + e;
191
+}
192
+
193
+int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
194
+{
195
+ return a + (long)b + c + d + (long)e + f;
196
+}
197
+
198
+struct bpf_fentry_test_t {
199
+ struct bpf_fentry_test_t *a;
200
+};
201
+
202
+int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
203
+{
204
+ return (long)arg;
205
+}
206
+
207
+int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
208
+{
209
+ return (long)arg->a;
210
+}
211
+
212
+int noinline bpf_modify_return_test(int a, int *b)
213
+{
214
+ *b += 1;
215
+ return a + *b;
216
+}
217
+__diag_pop();
218
+
219
+ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
80220
81221 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
82222 u32 headroom, u32 tailroom)
83223 {
84224 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
225
+ u32 user_size = kattr->test.data_size_in;
85226 void *data;
86227
87228 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
88229 return ERR_PTR(-EINVAL);
89230
231
+ if (user_size > size)
232
+ return ERR_PTR(-EMSGSIZE);
233
+
234
+ size = SKB_DATA_ALIGN(size);
90235 data = kzalloc(size + headroom + tailroom, GFP_USER);
91236 if (!data)
92237 return ERR_PTR(-ENOMEM);
93238
94
- if (copy_from_user(data + headroom, data_in, size)) {
239
+ if (copy_from_user(data + headroom, data_in, user_size)) {
95240 kfree(data);
96241 return ERR_PTR(-EFAULT);
97242 }
243
+
98244 return data;
99245 }
246
+
247
+int bpf_prog_test_run_tracing(struct bpf_prog *prog,
248
+ const union bpf_attr *kattr,
249
+ union bpf_attr __user *uattr)
250
+{
251
+ struct bpf_fentry_test_t arg = {};
252
+ u16 side_effect = 0, ret = 0;
253
+ int b = 2, err = -EFAULT;
254
+ u32 retval = 0;
255
+
256
+ if (kattr->test.flags || kattr->test.cpu)
257
+ return -EINVAL;
258
+
259
+ switch (prog->expected_attach_type) {
260
+ case BPF_TRACE_FENTRY:
261
+ case BPF_TRACE_FEXIT:
262
+ if (bpf_fentry_test1(1) != 2 ||
263
+ bpf_fentry_test2(2, 3) != 5 ||
264
+ bpf_fentry_test3(4, 5, 6) != 15 ||
265
+ bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
266
+ bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
267
+ bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
268
+ bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
269
+ bpf_fentry_test8(&arg) != 0)
270
+ goto out;
271
+ break;
272
+ case BPF_MODIFY_RETURN:
273
+ ret = bpf_modify_return_test(1, &b);
274
+ if (b != 2)
275
+ side_effect = 1;
276
+ break;
277
+ default:
278
+ goto out;
279
+ }
280
+
281
+ retval = ((u32)side_effect << 16) | ret;
282
+ if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
283
+ goto out;
284
+
285
+ err = 0;
286
+out:
287
+ trace_bpf_test_finish(&err);
288
+ return err;
289
+}
290
+
291
+struct bpf_raw_tp_test_run_info {
292
+ struct bpf_prog *prog;
293
+ void *ctx;
294
+ u32 retval;
295
+};
296
+
297
+static void
298
+__bpf_prog_test_run_raw_tp(void *data)
299
+{
300
+ struct bpf_raw_tp_test_run_info *info = data;
301
+
302
+ rcu_read_lock();
303
+ info->retval = BPF_PROG_RUN(info->prog, info->ctx);
304
+ rcu_read_unlock();
305
+}
306
+
307
+int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
308
+ const union bpf_attr *kattr,
309
+ union bpf_attr __user *uattr)
310
+{
311
+ void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
312
+ __u32 ctx_size_in = kattr->test.ctx_size_in;
313
+ struct bpf_raw_tp_test_run_info info;
314
+ int cpu = kattr->test.cpu, err = 0;
315
+ int current_cpu;
316
+
317
+ /* doesn't support data_in/out, ctx_out, duration, or repeat */
318
+ if (kattr->test.data_in || kattr->test.data_out ||
319
+ kattr->test.ctx_out || kattr->test.duration ||
320
+ kattr->test.repeat)
321
+ return -EINVAL;
322
+
323
+ if (ctx_size_in < prog->aux->max_ctx_offset ||
324
+ ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
325
+ return -EINVAL;
326
+
327
+ if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
328
+ return -EINVAL;
329
+
330
+ if (ctx_size_in) {
331
+ info.ctx = kzalloc(ctx_size_in, GFP_USER);
332
+ if (!info.ctx)
333
+ return -ENOMEM;
334
+ if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
335
+ err = -EFAULT;
336
+ goto out;
337
+ }
338
+ } else {
339
+ info.ctx = NULL;
340
+ }
341
+
342
+ info.prog = prog;
343
+
344
+ current_cpu = get_cpu();
345
+ if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
346
+ cpu == current_cpu) {
347
+ __bpf_prog_test_run_raw_tp(&info);
348
+ } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
349
+ /* smp_call_function_single() also checks cpu_online()
350
+ * after csd_lock(). However, since cpu is from user
351
+ * space, let's do an extra quick check to filter out
352
+ * invalid value before smp_call_function_single().
353
+ */
354
+ err = -ENXIO;
355
+ } else {
356
+ err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
357
+ &info, 1);
358
+ }
359
+ put_cpu();
360
+
361
+ if (!err &&
362
+ copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
363
+ err = -EFAULT;
364
+
365
+out:
366
+ kfree(info.ctx);
367
+ return err;
368
+}
369
+
370
+static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
371
+{
372
+ void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
373
+ void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
374
+ u32 size = kattr->test.ctx_size_in;
375
+ void *data;
376
+ int err;
377
+
378
+ if (!data_in && !data_out)
379
+ return NULL;
380
+
381
+ data = kzalloc(max_size, GFP_USER);
382
+ if (!data)
383
+ return ERR_PTR(-ENOMEM);
384
+
385
+ if (data_in) {
386
+ err = bpf_check_uarg_tail_zero(data_in, max_size, size);
387
+ if (err) {
388
+ kfree(data);
389
+ return ERR_PTR(err);
390
+ }
391
+
392
+ size = min_t(u32, max_size, size);
393
+ if (copy_from_user(data, data_in, size)) {
394
+ kfree(data);
395
+ return ERR_PTR(-EFAULT);
396
+ }
397
+ }
398
+ return data;
399
+}
400
+
401
+static int bpf_ctx_finish(const union bpf_attr *kattr,
402
+ union bpf_attr __user *uattr, const void *data,
403
+ u32 size)
404
+{
405
+ void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
406
+ int err = -EFAULT;
407
+ u32 copy_size = size;
408
+
409
+ if (!data || !data_out)
410
+ return 0;
411
+
412
+ if (copy_size > kattr->test.ctx_size_out) {
413
+ copy_size = kattr->test.ctx_size_out;
414
+ err = -ENOSPC;
415
+ }
416
+
417
+ if (copy_to_user(data_out, data, copy_size))
418
+ goto out;
419
+ if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
420
+ goto out;
421
+ if (err != -ENOSPC)
422
+ err = 0;
423
+out:
424
+ return err;
425
+}
426
+
427
+/**
428
+ * range_is_zero - test whether buffer is initialized
429
+ * @buf: buffer to check
430
+ * @from: check from this position
431
+ * @to: check up until (excluding) this position
432
+ *
433
+ * This function returns true if the there is a non-zero byte
434
+ * in the buf in the range [from,to).
435
+ */
436
+static inline bool range_is_zero(void *buf, size_t from, size_t to)
437
+{
438
+ return !memchr_inv((u8 *)buf + from, 0, to - from);
439
+}
440
+
441
+static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
442
+{
443
+ struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
444
+
445
+ if (!__skb)
446
+ return 0;
447
+
448
+ /* make sure the fields we don't use are zeroed */
449
+ if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
450
+ return -EINVAL;
451
+
452
+ /* mark is allowed */
453
+
454
+ if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
455
+ offsetof(struct __sk_buff, priority)))
456
+ return -EINVAL;
457
+
458
+ /* priority is allowed */
459
+
460
+ if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
461
+ offsetof(struct __sk_buff, ifindex)))
462
+ return -EINVAL;
463
+
464
+ /* ifindex is allowed */
465
+
466
+ if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
467
+ offsetof(struct __sk_buff, cb)))
468
+ return -EINVAL;
469
+
470
+ /* cb is allowed */
471
+
472
+ if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
473
+ offsetof(struct __sk_buff, tstamp)))
474
+ return -EINVAL;
475
+
476
+ /* tstamp is allowed */
477
+ /* wire_len is allowed */
478
+ /* gso_segs is allowed */
479
+
480
+ if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
481
+ offsetof(struct __sk_buff, gso_size)))
482
+ return -EINVAL;
483
+
484
+ /* gso_size is allowed */
485
+
486
+ if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
487
+ sizeof(struct __sk_buff)))
488
+ return -EINVAL;
489
+
490
+ skb->mark = __skb->mark;
491
+ skb->priority = __skb->priority;
492
+ skb->tstamp = __skb->tstamp;
493
+ memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
494
+
495
+ if (__skb->wire_len == 0) {
496
+ cb->pkt_len = skb->len;
497
+ } else {
498
+ if (__skb->wire_len < skb->len ||
499
+ __skb->wire_len > GSO_MAX_SIZE)
500
+ return -EINVAL;
501
+ cb->pkt_len = __skb->wire_len;
502
+ }
503
+
504
+ if (__skb->gso_segs > GSO_MAX_SEGS)
505
+ return -EINVAL;
506
+ skb_shinfo(skb)->gso_segs = __skb->gso_segs;
507
+ skb_shinfo(skb)->gso_size = __skb->gso_size;
508
+
509
+ return 0;
510
+}
511
+
512
+static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
513
+{
514
+ struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
515
+
516
+ if (!__skb)
517
+ return;
518
+
519
+ __skb->mark = skb->mark;
520
+ __skb->priority = skb->priority;
521
+ __skb->ifindex = skb->dev->ifindex;
522
+ __skb->tstamp = skb->tstamp;
523
+ memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
524
+ __skb->wire_len = cb->pkt_len;
525
+ __skb->gso_segs = skb_shinfo(skb)->gso_segs;
526
+}
527
+
528
+static struct proto bpf_dummy_proto = {
529
+ .name = "bpf_dummy",
530
+ .owner = THIS_MODULE,
531
+ .obj_size = sizeof(struct sock),
532
+};
100533
101534 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
102535 union bpf_attr __user *uattr)
103536 {
104537 bool is_l2 = false, is_direct_pkt_access = false;
538
+ struct net *net = current->nsproxy->net_ns;
539
+ struct net_device *dev = net->loopback_dev;
105540 u32 size = kattr->test.data_size_in;
106541 u32 repeat = kattr->test.repeat;
542
+ struct __sk_buff *ctx = NULL;
107543 u32 retval, duration;
108544 int hh_len = ETH_HLEN;
109545 struct sk_buff *skb;
546
+ struct sock *sk;
110547 void *data;
111548 int ret;
549
+
550
+ if (kattr->test.flags || kattr->test.cpu)
551
+ return -EINVAL;
112552
113553 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
114554 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
115555 if (IS_ERR(data))
116556 return PTR_ERR(data);
117557
558
+ ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
559
+ if (IS_ERR(ctx)) {
560
+ kfree(data);
561
+ return PTR_ERR(ctx);
562
+ }
563
+
118564 switch (prog->type) {
119565 case BPF_PROG_TYPE_SCHED_CLS:
120566 case BPF_PROG_TYPE_SCHED_ACT:
121567 is_l2 = true;
122
- /* fall through */
568
+ fallthrough;
123569 case BPF_PROG_TYPE_LWT_IN:
124570 case BPF_PROG_TYPE_LWT_OUT:
125571 case BPF_PROG_TYPE_LWT_XMIT:
....@@ -129,71 +575,330 @@
129575 break;
130576 }
131577
578
+ sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
579
+ if (!sk) {
580
+ kfree(data);
581
+ kfree(ctx);
582
+ return -ENOMEM;
583
+ }
584
+ sock_init_data(NULL, sk);
585
+
132586 skb = build_skb(data, 0);
133587 if (!skb) {
134588 kfree(data);
589
+ kfree(ctx);
590
+ sk_free(sk);
135591 return -ENOMEM;
136592 }
593
+ skb->sk = sk;
137594
138595 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
139596 __skb_put(skb, size);
140
- skb->protocol = eth_type_trans(skb, current->nsproxy->net_ns->loopback_dev);
597
+ if (ctx && ctx->ifindex > 1) {
598
+ dev = dev_get_by_index(net, ctx->ifindex);
599
+ if (!dev) {
600
+ ret = -ENODEV;
601
+ goto out;
602
+ }
603
+ }
604
+ skb->protocol = eth_type_trans(skb, dev);
141605 skb_reset_network_header(skb);
606
+
607
+ switch (skb->protocol) {
608
+ case htons(ETH_P_IP):
609
+ sk->sk_family = AF_INET;
610
+ if (sizeof(struct iphdr) <= skb_headlen(skb)) {
611
+ sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
612
+ sk->sk_daddr = ip_hdr(skb)->daddr;
613
+ }
614
+ break;
615
+#if IS_ENABLED(CONFIG_IPV6)
616
+ case htons(ETH_P_IPV6):
617
+ sk->sk_family = AF_INET6;
618
+ if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
619
+ sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
620
+ sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
621
+ }
622
+ break;
623
+#endif
624
+ default:
625
+ break;
626
+ }
142627
143628 if (is_l2)
144629 __skb_push(skb, hh_len);
145630 if (is_direct_pkt_access)
146631 bpf_compute_data_pointers(skb);
147
- retval = bpf_test_run(prog, skb, repeat, &duration);
632
+ ret = convert___skb_to_skb(skb, ctx);
633
+ if (ret)
634
+ goto out;
635
+ ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
636
+ if (ret)
637
+ goto out;
148638 if (!is_l2) {
149639 if (skb_headroom(skb) < hh_len) {
150640 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
151641
152642 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
153
- kfree_skb(skb);
154
- return -ENOMEM;
643
+ ret = -ENOMEM;
644
+ goto out;
155645 }
156646 }
157647 memset(__skb_push(skb, hh_len), 0, hh_len);
158648 }
649
+ convert_skb_to___skb(skb, ctx);
159650
160651 size = skb->len;
161652 /* bpf program can never convert linear skb to non-linear */
162653 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
163654 size = skb_headlen(skb);
164655 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
656
+ if (!ret)
657
+ ret = bpf_ctx_finish(kattr, uattr, ctx,
658
+ sizeof(struct __sk_buff));
659
+out:
660
+ if (dev && dev != net->loopback_dev)
661
+ dev_put(dev);
165662 kfree_skb(skb);
663
+ sk_free(sk);
664
+ kfree(ctx);
166665 return ret;
167666 }
168667
169668 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
170669 union bpf_attr __user *uattr)
171670 {
671
+ u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
672
+ u32 headroom = XDP_PACKET_HEADROOM;
172673 u32 size = kattr->test.data_size_in;
173674 u32 repeat = kattr->test.repeat;
174675 struct netdev_rx_queue *rxqueue;
175676 struct xdp_buff xdp = {};
176677 u32 retval, duration;
678
+ u32 max_data_sz;
177679 void *data;
178680 int ret;
179681
180
- data = bpf_test_init(kattr, size, XDP_PACKET_HEADROOM + NET_IP_ALIGN, 0);
682
+ if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
683
+ prog->expected_attach_type == BPF_XDP_CPUMAP)
684
+ return -EINVAL;
685
+ if (kattr->test.ctx_in || kattr->test.ctx_out)
686
+ return -EINVAL;
687
+
688
+ /* XDP have extra tailroom as (most) drivers use full page */
689
+ max_data_sz = 4096 - headroom - tailroom;
690
+
691
+ data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
181692 if (IS_ERR(data))
182693 return PTR_ERR(data);
183694
184695 xdp.data_hard_start = data;
185
- xdp.data = data + XDP_PACKET_HEADROOM + NET_IP_ALIGN;
696
+ xdp.data = data + headroom;
186697 xdp.data_meta = xdp.data;
187698 xdp.data_end = xdp.data + size;
699
+ xdp.frame_sz = headroom + max_data_sz + tailroom;
188700
189701 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
190702 xdp.rxq = &rxqueue->xdp_rxq;
191
-
192
- retval = bpf_test_run(prog, &xdp, repeat, &duration);
193
- if (xdp.data != data + XDP_PACKET_HEADROOM + NET_IP_ALIGN ||
194
- xdp.data_end != xdp.data + size)
703
+ bpf_prog_change_xdp(NULL, prog);
704
+ ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
705
+ if (ret)
706
+ goto out;
707
+ if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
195708 size = xdp.data_end - xdp.data;
196709 ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
710
+out:
711
+ bpf_prog_change_xdp(prog, NULL);
197712 kfree(data);
198713 return ret;
199714 }
715
+
716
+static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
717
+{
718
+ /* make sure the fields we don't use are zeroed */
719
+ if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
720
+ return -EINVAL;
721
+
722
+ /* flags is allowed */
723
+
724
+ if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
725
+ sizeof(struct bpf_flow_keys)))
726
+ return -EINVAL;
727
+
728
+ return 0;
729
+}
730
+
731
+int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
732
+ const union bpf_attr *kattr,
733
+ union bpf_attr __user *uattr)
734
+{
735
+ struct bpf_test_timer t = { NO_PREEMPT };
736
+ u32 size = kattr->test.data_size_in;
737
+ struct bpf_flow_dissector ctx = {};
738
+ u32 repeat = kattr->test.repeat;
739
+ struct bpf_flow_keys *user_ctx;
740
+ struct bpf_flow_keys flow_keys;
741
+ const struct ethhdr *eth;
742
+ unsigned int flags = 0;
743
+ u32 retval, duration;
744
+ void *data;
745
+ int ret;
746
+
747
+ if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
748
+ return -EINVAL;
749
+
750
+ if (kattr->test.flags || kattr->test.cpu)
751
+ return -EINVAL;
752
+
753
+ if (size < ETH_HLEN)
754
+ return -EINVAL;
755
+
756
+ data = bpf_test_init(kattr, size, 0, 0);
757
+ if (IS_ERR(data))
758
+ return PTR_ERR(data);
759
+
760
+ eth = (struct ethhdr *)data;
761
+
762
+ if (!repeat)
763
+ repeat = 1;
764
+
765
+ user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
766
+ if (IS_ERR(user_ctx)) {
767
+ kfree(data);
768
+ return PTR_ERR(user_ctx);
769
+ }
770
+ if (user_ctx) {
771
+ ret = verify_user_bpf_flow_keys(user_ctx);
772
+ if (ret)
773
+ goto out;
774
+ flags = user_ctx->flags;
775
+ }
776
+
777
+ ctx.flow_keys = &flow_keys;
778
+ ctx.data = data;
779
+ ctx.data_end = (__u8 *)data + size;
780
+
781
+ bpf_test_timer_enter(&t);
782
+ do {
783
+ retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
784
+ size, flags);
785
+ } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
786
+ bpf_test_timer_leave(&t);
787
+
788
+ if (ret < 0)
789
+ goto out;
790
+
791
+ ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
792
+ retval, duration);
793
+ if (!ret)
794
+ ret = bpf_ctx_finish(kattr, uattr, user_ctx,
795
+ sizeof(struct bpf_flow_keys));
796
+
797
+out:
798
+ kfree(user_ctx);
799
+ kfree(data);
800
+ return ret;
801
+}
802
+
803
+int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
804
+ union bpf_attr __user *uattr)
805
+{
806
+ struct bpf_test_timer t = { NO_PREEMPT };
807
+ struct bpf_prog_array *progs = NULL;
808
+ struct bpf_sk_lookup_kern ctx = {};
809
+ u32 repeat = kattr->test.repeat;
810
+ struct bpf_sk_lookup *user_ctx;
811
+ u32 retval, duration;
812
+ int ret = -EINVAL;
813
+
814
+ if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
815
+ return -EINVAL;
816
+
817
+ if (kattr->test.flags || kattr->test.cpu)
818
+ return -EINVAL;
819
+
820
+ if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
821
+ kattr->test.data_size_out)
822
+ return -EINVAL;
823
+
824
+ if (!repeat)
825
+ repeat = 1;
826
+
827
+ user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
828
+ if (IS_ERR(user_ctx))
829
+ return PTR_ERR(user_ctx);
830
+
831
+ if (!user_ctx)
832
+ return -EINVAL;
833
+
834
+ if (user_ctx->sk)
835
+ goto out;
836
+
837
+ if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
838
+ goto out;
839
+
840
+ if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
841
+ ret = -ERANGE;
842
+ goto out;
843
+ }
844
+
845
+ ctx.family = (u16)user_ctx->family;
846
+ ctx.protocol = (u16)user_ctx->protocol;
847
+ ctx.dport = (u16)user_ctx->local_port;
848
+ ctx.sport = (__force __be16)user_ctx->remote_port;
849
+
850
+ switch (ctx.family) {
851
+ case AF_INET:
852
+ ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
853
+ ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
854
+ break;
855
+
856
+#if IS_ENABLED(CONFIG_IPV6)
857
+ case AF_INET6:
858
+ ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
859
+ ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
860
+ break;
861
+#endif
862
+
863
+ default:
864
+ ret = -EAFNOSUPPORT;
865
+ goto out;
866
+ }
867
+
868
+ progs = bpf_prog_array_alloc(1, GFP_KERNEL);
869
+ if (!progs) {
870
+ ret = -ENOMEM;
871
+ goto out;
872
+ }
873
+
874
+ progs->items[0].prog = prog;
875
+
876
+ bpf_test_timer_enter(&t);
877
+ do {
878
+ ctx.selected_sk = NULL;
879
+ retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
880
+ } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
881
+ bpf_test_timer_leave(&t);
882
+
883
+ if (ret < 0)
884
+ goto out;
885
+
886
+ user_ctx->cookie = 0;
887
+ if (ctx.selected_sk) {
888
+ if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
889
+ ret = -EOPNOTSUPP;
890
+ goto out;
891
+ }
892
+
893
+ user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
894
+ }
895
+
896
+ ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
897
+ if (!ret)
898
+ ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
899
+
900
+out:
901
+ bpf_prog_array_free(progs);
902
+ kfree(user_ctx);
903
+ return ret;
904
+}