bpf.c

  1/*
  2 * Copyright (C) 2017 Netronome Systems, Inc.
  3 *
  4 * This software is licensed under the GNU General License Version 2,
  5 * June 1991 as shown in the file COPYING in the top-level directory of this
  6 * source tree.
  7 *
  8 * THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"
  9 * WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING,
 10 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 11 * FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE
 12 * OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME
 13 * THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
 14 */
 15
 16#include <linux/bpf.h>
 17#include <linux/bpf_verifier.h>
 18#include <linux/debugfs.h>
 19#include <linux/kernel.h>
 20#include <linux/mutex.h>
 21#include <linux/rtnetlink.h>
 22#include <net/pkt_cls.h>
 23
 24#include "netdevsim.h"
 25
 26#define pr_vlog(env, fmt, ...)	\
 27	bpf_verifier_log_write(env, "[netdevsim] " fmt, ##__VA_ARGS__)
 28
 29struct nsim_bpf_bound_prog {
 30	struct nsim_dev *nsim_dev;
 31	struct bpf_prog *prog;
 32	struct dentry *ddir;
 33	const char *state;
 34	bool is_loaded;
 35	struct list_head l;
 36};
 37
 38#define NSIM_BPF_MAX_KEYS		2
 39
 40struct nsim_bpf_bound_map {
 41	struct netdevsim *ns;
 42	struct bpf_offloaded_map *map;
 43	struct mutex mutex;
 44	struct nsim_map_entry {
 45		void *key;
 46		void *value;
 47	} entry[NSIM_BPF_MAX_KEYS];
 48	struct list_head l;
 49};
 50
 51static int nsim_bpf_string_show(struct seq_file *file, void *data)
 52{
 53	const char **str = file->private;
 54
 55	if (*str)
 56		seq_printf(file, "%s\n", *str);
 57
 58	return 0;
 59}
 60DEFINE_SHOW_ATTRIBUTE(nsim_bpf_string);
 61
 62static int
 63nsim_bpf_verify_insn(struct bpf_verifier_env *env, int insn_idx, int prev_insn)
 64{
 65	struct nsim_bpf_bound_prog *state;
 66	int ret = 0;
 67
 68	state = env->prog->aux->offload->dev_priv;
 69	if (state->nsim_dev->bpf_bind_verifier_delay && !insn_idx)
 70		msleep(state->nsim_dev->bpf_bind_verifier_delay);
 71
 72	if (insn_idx == env->prog->len - 1) {
 73		pr_vlog(env, "Hello from netdevsim!\n");
 74
 75		if (!state->nsim_dev->bpf_bind_verifier_accept)
 76			ret = -EOPNOTSUPP;
 77	}
 78
 79	return ret;
 80}
 81
 82static int nsim_bpf_finalize(struct bpf_verifier_env *env)
 83{
 84	return 0;
 85}
 86
 87static bool nsim_xdp_offload_active(struct netdevsim *ns)
 88{
 89	return ns->xdp_hw.prog;
 90}
 91
 92static void nsim_prog_set_loaded(struct bpf_prog *prog, bool loaded)
 93{
 94	struct nsim_bpf_bound_prog *state;
 95
 96	if (!prog || !bpf_prog_is_offloaded(prog->aux))
 97		return;
 98
 99	state = prog->aux->offload->dev_priv;
100	state->is_loaded = loaded;
101}
102
103static int
104nsim_bpf_offload(struct netdevsim *ns, struct bpf_prog *prog, bool oldprog)
105{
106	nsim_prog_set_loaded(ns->bpf_offloaded, false);
107
108	WARN(!!ns->bpf_offloaded != oldprog,
109	     "bad offload state, expected offload %sto be active",
110	     oldprog ? "" : "not ");
111	ns->bpf_offloaded = prog;
112	ns->bpf_offloaded_id = prog ? prog->aux->id : 0;
113	nsim_prog_set_loaded(prog, true);
114
115	return 0;
116}
117
118int nsim_bpf_setup_tc_block_cb(enum tc_setup_type type,
119			       void *type_data, void *cb_priv)
120{
121	struct tc_cls_bpf_offload *cls_bpf = type_data;
122	struct bpf_prog *prog = cls_bpf->prog;
123	struct netdevsim *ns = cb_priv;
124	struct bpf_prog *oldprog;
125
126	if (type != TC_SETUP_CLSBPF) {
127		NSIM_EA(cls_bpf->common.extack,
128			"only offload of BPF classifiers supported");
129		return -EOPNOTSUPP;
130	}
131
132	if (!tc_cls_can_offload_and_chain0(ns->netdev, &cls_bpf->common))
133		return -EOPNOTSUPP;
134
135	if (cls_bpf->common.protocol != htons(ETH_P_ALL)) {
136		NSIM_EA(cls_bpf->common.extack,
137			"only ETH_P_ALL supported as filter protocol");
138		return -EOPNOTSUPP;
139	}
140
141	if (!ns->bpf_tc_accept) {
142		NSIM_EA(cls_bpf->common.extack,
143			"netdevsim configured to reject BPF TC offload");
144		return -EOPNOTSUPP;
145	}
146	/* Note: progs without skip_sw will probably not be dev bound */
147	if (prog && !prog->aux->offload && !ns->bpf_tc_non_bound_accept) {
148		NSIM_EA(cls_bpf->common.extack,
149			"netdevsim configured to reject unbound programs");
150		return -EOPNOTSUPP;
151	}
152
153	if (cls_bpf->command != TC_CLSBPF_OFFLOAD)
154		return -EOPNOTSUPP;
155
156	oldprog = cls_bpf->oldprog;
157
158	/* Don't remove if oldprog doesn't match driver's state */
159	if (ns->bpf_offloaded != oldprog) {
160		oldprog = NULL;
161		if (!cls_bpf->prog)
162			return 0;
163		if (ns->bpf_offloaded) {
164			NSIM_EA(cls_bpf->common.extack,
165				"driver and netdev offload states mismatch");
166			return -EBUSY;
167		}
168	}
169
170	return nsim_bpf_offload(ns, cls_bpf->prog, oldprog);
171}
172
173int nsim_bpf_disable_tc(struct netdevsim *ns)
174{
175	if (ns->bpf_offloaded && !nsim_xdp_offload_active(ns))
176		return -EBUSY;
177	return 0;
178}
179
180static int nsim_xdp_offload_prog(struct netdevsim *ns, struct netdev_bpf *bpf)
181{
182	if (!nsim_xdp_offload_active(ns) && !bpf->prog)
183		return 0;
184	if (!nsim_xdp_offload_active(ns) && bpf->prog && ns->bpf_offloaded) {
185		NSIM_EA(bpf->extack, "TC program is already loaded");
186		return -EBUSY;
187	}
188
189	return nsim_bpf_offload(ns, bpf->prog, nsim_xdp_offload_active(ns));
190}
191
192static int
193nsim_xdp_set_prog(struct netdevsim *ns, struct netdev_bpf *bpf,
194		  struct xdp_attachment_info *xdp)
195{
196	int err;
197
198	if (bpf->command == XDP_SETUP_PROG && !ns->bpf_xdpdrv_accept) {
199		NSIM_EA(bpf->extack, "driver XDP disabled in DebugFS");
200		return -EOPNOTSUPP;
201	}
202	if (bpf->command == XDP_SETUP_PROG_HW && !ns->bpf_xdpoffload_accept) {
203		NSIM_EA(bpf->extack, "XDP offload disabled in DebugFS");
204		return -EOPNOTSUPP;
205	}
206
207	if (bpf->command == XDP_SETUP_PROG_HW) {
208		err = nsim_xdp_offload_prog(ns, bpf);
209		if (err)
210			return err;
211	}
212
213	xdp_attachment_setup(xdp, bpf);
214
215	return 0;
216}
217
218static int nsim_bpf_create_prog(struct nsim_dev *nsim_dev,
219				struct bpf_prog *prog)
220{
221	struct nsim_bpf_bound_prog *state;
222	char name[16];
223	int ret;
224
225	state = kzalloc(sizeof(*state), GFP_KERNEL);
226	if (!state)
227		return -ENOMEM;
228
229	state->nsim_dev = nsim_dev;
230	state->prog = prog;
231	state->state = "verify";
232
233	/* Program id is not populated yet when we create the state. */
234	sprintf(name, "%u", nsim_dev->prog_id_gen++);
235	state->ddir = debugfs_create_dir(name, nsim_dev->ddir_bpf_bound_progs);
236	if (IS_ERR(state->ddir)) {
237		ret = PTR_ERR(state->ddir);
238		kfree(state);
239		return ret;
240	}
241
242	debugfs_create_u32("id", 0400, state->ddir, &prog->aux->id);
243	debugfs_create_file("state", 0400, state->ddir,
244			    &state->state, &nsim_bpf_string_fops);
245	debugfs_create_bool("loaded", 0400, state->ddir, &state->is_loaded);
246
247	mutex_lock(&nsim_dev->progs_list_lock);
248	list_add_tail(&state->l, &nsim_dev->bpf_bound_progs);
249	mutex_unlock(&nsim_dev->progs_list_lock);
250
251	prog->aux->offload->dev_priv = state;
252
253	return 0;
254}
255
256static int nsim_bpf_verifier_prep(struct bpf_prog *prog)
257{
258	struct nsim_dev *nsim_dev =
259			bpf_offload_dev_priv(prog->aux->offload->offdev);
260
261	if (!nsim_dev->bpf_bind_accept)
262		return -EOPNOTSUPP;
263
264	return nsim_bpf_create_prog(nsim_dev, prog);
265}
266
267static int nsim_bpf_translate(struct bpf_prog *prog)
268{
269	struct nsim_bpf_bound_prog *state = prog->aux->offload->dev_priv;
270
271	state->state = "xlated";
272	return 0;
273}
274
275static void nsim_bpf_destroy_prog(struct bpf_prog *prog)
276{
277	struct nsim_bpf_bound_prog *state;
278	struct nsim_dev *nsim_dev;
279
280	state = prog->aux->offload->dev_priv;
281	nsim_dev = state->nsim_dev;
282	WARN(state->is_loaded,
283	     "offload state destroyed while program still bound");
284	debugfs_remove_recursive(state->ddir);
285	mutex_lock(&nsim_dev->progs_list_lock);
286	list_del(&state->l);
287	mutex_unlock(&nsim_dev->progs_list_lock);
288	kfree(state);
289}
290
291static const struct bpf_prog_offload_ops nsim_bpf_dev_ops = {
292	.insn_hook	= nsim_bpf_verify_insn,
293	.finalize	= nsim_bpf_finalize,
294	.prepare	= nsim_bpf_verifier_prep,
295	.translate	= nsim_bpf_translate,
296	.destroy	= nsim_bpf_destroy_prog,
297};
298
299static int nsim_setup_prog_checks(struct netdevsim *ns, struct netdev_bpf *bpf)
300{
301	if (bpf->prog && bpf->prog->aux->offload) {
302		NSIM_EA(bpf->extack, "attempt to load offloaded prog to drv");
303		return -EINVAL;
304	}
305	if (bpf->prog && !bpf->prog->aux->xdp_has_frags &&
306	    ns->netdev->mtu > NSIM_XDP_MAX_MTU) {
307		NSIM_EA(bpf->extack, "MTU too large w/ XDP enabled");
308		return -EINVAL;
309	}
310	return 0;
311}
312
313static int
314nsim_setup_prog_hw_checks(struct netdevsim *ns, struct netdev_bpf *bpf)
315{
316	struct nsim_bpf_bound_prog *state;
317
318	if (!bpf->prog)
319		return 0;
320
321	if (!bpf_prog_is_offloaded(bpf->prog->aux)) {
322		NSIM_EA(bpf->extack, "xdpoffload of non-bound program");
323		return -EINVAL;
324	}
325
326	state = bpf->prog->aux->offload->dev_priv;
327	if (WARN_ON(strcmp(state->state, "xlated"))) {
328		NSIM_EA(bpf->extack, "offloading program in bad state");
329		return -EINVAL;
330	}
331	return 0;
332}
333
334static bool
335nsim_map_key_match(struct bpf_map *map, struct nsim_map_entry *e, void *key)
336{
337	return e->key && !memcmp(key, e->key, map->key_size);
338}
339
340static int nsim_map_key_find(struct bpf_offloaded_map *offmap, void *key)
341{
342	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
343	unsigned int i;
344
345	for (i = 0; i < ARRAY_SIZE(nmap->entry); i++)
346		if (nsim_map_key_match(&offmap->map, &nmap->entry[i], key))
347			return i;
348
349	return -ENOENT;
350}
351
352static int
353nsim_map_alloc_elem(struct bpf_offloaded_map *offmap, unsigned int idx)
354{
355	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
356
357	nmap->entry[idx].key = kmalloc(offmap->map.key_size,
358				       GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
359	if (!nmap->entry[idx].key)
360		return -ENOMEM;
361	nmap->entry[idx].value = kmalloc(offmap->map.value_size,
362					 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
363	if (!nmap->entry[idx].value) {
364		kfree(nmap->entry[idx].key);
365		nmap->entry[idx].key = NULL;
366		return -ENOMEM;
367	}
368
369	return 0;
370}
371
372static int
373nsim_map_get_next_key(struct bpf_offloaded_map *offmap,
374		      void *key, void *next_key)
375{
376	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
377	int idx = -ENOENT;
378
379	mutex_lock(&nmap->mutex);
380
381	if (key)
382		idx = nsim_map_key_find(offmap, key);
383	if (idx == -ENOENT)
384		idx = 0;
385	else
386		idx++;
387
388	for (; idx < ARRAY_SIZE(nmap->entry); idx++) {
389		if (nmap->entry[idx].key) {
390			memcpy(next_key, nmap->entry[idx].key,
391			       offmap->map.key_size);
392			break;
393		}
394	}
395
396	mutex_unlock(&nmap->mutex);
397
398	if (idx == ARRAY_SIZE(nmap->entry))
399		return -ENOENT;
400	return 0;
401}
402
403static int
404nsim_map_lookup_elem(struct bpf_offloaded_map *offmap, void *key, void *value)
405{
406	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
407	int idx;
408
409	mutex_lock(&nmap->mutex);
410
411	idx = nsim_map_key_find(offmap, key);
412	if (idx >= 0)
413		memcpy(value, nmap->entry[idx].value, offmap->map.value_size);
414
415	mutex_unlock(&nmap->mutex);
416
417	return idx < 0 ? idx : 0;
418}
419
420static int
421nsim_map_update_elem(struct bpf_offloaded_map *offmap,
422		     void *key, void *value, u64 flags)
423{
424	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
425	int idx, err = 0;
426
427	mutex_lock(&nmap->mutex);
428
429	idx = nsim_map_key_find(offmap, key);
430	if (idx < 0 && flags == BPF_EXIST) {
431		err = idx;
432		goto exit_unlock;
433	}
434	if (idx >= 0 && flags == BPF_NOEXIST) {
435		err = -EEXIST;
436		goto exit_unlock;
437	}
438
439	if (idx < 0) {
440		for (idx = 0; idx < ARRAY_SIZE(nmap->entry); idx++)
441			if (!nmap->entry[idx].key)
442				break;
443		if (idx == ARRAY_SIZE(nmap->entry)) {
444			err = -E2BIG;
445			goto exit_unlock;
446		}
447
448		err = nsim_map_alloc_elem(offmap, idx);
449		if (err)
450			goto exit_unlock;
451	}
452
453	memcpy(nmap->entry[idx].key, key, offmap->map.key_size);
454	memcpy(nmap->entry[idx].value, value, offmap->map.value_size);
455exit_unlock:
456	mutex_unlock(&nmap->mutex);
457
458	return err;
459}
460
461static int nsim_map_delete_elem(struct bpf_offloaded_map *offmap, void *key)
462{
463	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
464	int idx;
465
466	if (offmap->map.map_type == BPF_MAP_TYPE_ARRAY)
467		return -EINVAL;
468
469	mutex_lock(&nmap->mutex);
470
471	idx = nsim_map_key_find(offmap, key);
472	if (idx >= 0) {
473		kfree(nmap->entry[idx].key);
474		kfree(nmap->entry[idx].value);
475		memset(&nmap->entry[idx], 0, sizeof(nmap->entry[idx]));
476	}
477
478	mutex_unlock(&nmap->mutex);
479
480	return idx < 0 ? idx : 0;
481}
482
483static const struct bpf_map_dev_ops nsim_bpf_map_ops = {
484	.map_get_next_key	= nsim_map_get_next_key,
485	.map_lookup_elem	= nsim_map_lookup_elem,
486	.map_update_elem	= nsim_map_update_elem,
487	.map_delete_elem	= nsim_map_delete_elem,
488};
489
490static int
491nsim_bpf_map_alloc(struct netdevsim *ns, struct bpf_offloaded_map *offmap)
492{
493	struct nsim_bpf_bound_map *nmap;
494	int i, err;
495
496	if (WARN_ON(offmap->map.map_type != BPF_MAP_TYPE_ARRAY &&
497		    offmap->map.map_type != BPF_MAP_TYPE_HASH))
498		return -EINVAL;
499	if (offmap->map.max_entries > NSIM_BPF_MAX_KEYS)
500		return -ENOMEM;
501	if (offmap->map.map_flags)
502		return -EINVAL;
503
504	nmap = kzalloc(sizeof(*nmap), GFP_KERNEL_ACCOUNT);
505	if (!nmap)
506		return -ENOMEM;
507
508	offmap->dev_priv = nmap;
509	nmap->ns = ns;
510	nmap->map = offmap;
511	mutex_init(&nmap->mutex);
512
513	if (offmap->map.map_type == BPF_MAP_TYPE_ARRAY) {
514		for (i = 0; i < ARRAY_SIZE(nmap->entry); i++) {
515			u32 *key;
516
517			err = nsim_map_alloc_elem(offmap, i);
518			if (err)
519				goto err_free;
520			key = nmap->entry[i].key;
521			*key = i;
522			memset(nmap->entry[i].value, 0, offmap->map.value_size);
523		}
524	}
525
526	offmap->dev_ops = &nsim_bpf_map_ops;
527	list_add_tail(&nmap->l, &ns->nsim_dev->bpf_bound_maps);
528
529	return 0;
530
531err_free:
532	while (--i >= 0) {
533		kfree(nmap->entry[i].key);
534		kfree(nmap->entry[i].value);
535	}
536	kfree(nmap);
537	return err;
538}
539
540static void nsim_bpf_map_free(struct bpf_offloaded_map *offmap)
541{
542	struct nsim_bpf_bound_map *nmap = offmap->dev_priv;
543	unsigned int i;
544
545	for (i = 0; i < ARRAY_SIZE(nmap->entry); i++) {
546		kfree(nmap->entry[i].key);
547		kfree(nmap->entry[i].value);
548	}
549	list_del_init(&nmap->l);
550	mutex_destroy(&nmap->mutex);
551	kfree(nmap);
552}
553
554int nsim_bpf(struct net_device *dev, struct netdev_bpf *bpf)
555{
556	struct netdevsim *ns = netdev_priv(dev);
557	int err;
558
559	ASSERT_RTNL();
560
561	switch (bpf->command) {
562	case XDP_SETUP_PROG:
563		err = nsim_setup_prog_checks(ns, bpf);
564		if (err)
565			return err;
566
567		return nsim_xdp_set_prog(ns, bpf, &ns->xdp);
568	case XDP_SETUP_PROG_HW:
569		err = nsim_setup_prog_hw_checks(ns, bpf);
570		if (err)
571			return err;
572
573		return nsim_xdp_set_prog(ns, bpf, &ns->xdp_hw);
574	case BPF_OFFLOAD_MAP_ALLOC:
575		if (!ns->bpf_map_accept)
576			return -EOPNOTSUPP;
577
578		return nsim_bpf_map_alloc(ns, bpf->offmap);
579	case BPF_OFFLOAD_MAP_FREE:
580		nsim_bpf_map_free(bpf->offmap);
581		return 0;
582	default:
583		return -EINVAL;
584	}
585}
586
587int nsim_bpf_dev_init(struct nsim_dev *nsim_dev)
588{
589	int err;
590
591	INIT_LIST_HEAD(&nsim_dev->bpf_bound_progs);
592	INIT_LIST_HEAD(&nsim_dev->bpf_bound_maps);
593
594	nsim_dev->ddir_bpf_bound_progs = debugfs_create_dir("bpf_bound_progs",
595							    nsim_dev->ddir);
596	if (IS_ERR(nsim_dev->ddir_bpf_bound_progs))
597		return PTR_ERR(nsim_dev->ddir_bpf_bound_progs);
598
599	nsim_dev->bpf_dev = bpf_offload_dev_create(&nsim_bpf_dev_ops, nsim_dev);
600	err = PTR_ERR_OR_ZERO(nsim_dev->bpf_dev);
601	if (err)
602		return err;
603
604	nsim_dev->bpf_bind_accept = true;
605	debugfs_create_bool("bpf_bind_accept", 0600, nsim_dev->ddir,
606			    &nsim_dev->bpf_bind_accept);
607	debugfs_create_u32("bpf_bind_verifier_delay", 0600, nsim_dev->ddir,
608			   &nsim_dev->bpf_bind_verifier_delay);
609	nsim_dev->bpf_bind_verifier_accept = true;
610	debugfs_create_bool("bpf_bind_verifier_accept", 0600, nsim_dev->ddir,
611			    &nsim_dev->bpf_bind_verifier_accept);
612	return 0;
613}
614
615void nsim_bpf_dev_exit(struct nsim_dev *nsim_dev)
616{
617	WARN_ON(!list_empty(&nsim_dev->bpf_bound_progs));
618	WARN_ON(!list_empty(&nsim_dev->bpf_bound_maps));
619	bpf_offload_dev_destroy(nsim_dev->bpf_dev);
620}
621
622int nsim_bpf_init(struct netdevsim *ns)
623{
624	struct dentry *ddir = ns->nsim_dev_port->ddir;
625	int err;
626
627	err = bpf_offload_dev_netdev_register(ns->nsim_dev->bpf_dev,
628					      ns->netdev);
629	if (err)
630		return err;
631
632	debugfs_create_u32("bpf_offloaded_id", 0400, ddir,
633			   &ns->bpf_offloaded_id);
634
635	ns->bpf_tc_accept = true;
636	debugfs_create_bool("bpf_tc_accept", 0600, ddir,
637			    &ns->bpf_tc_accept);
638	debugfs_create_bool("bpf_tc_non_bound_accept", 0600, ddir,
639			    &ns->bpf_tc_non_bound_accept);
640	ns->bpf_xdpdrv_accept = true;
641	debugfs_create_bool("bpf_xdpdrv_accept", 0600, ddir,
642			    &ns->bpf_xdpdrv_accept);
643	ns->bpf_xdpoffload_accept = true;
644	debugfs_create_bool("bpf_xdpoffload_accept", 0600, ddir,
645			    &ns->bpf_xdpoffload_accept);
646
647	ns->bpf_map_accept = true;
648	debugfs_create_bool("bpf_map_accept", 0600, ddir,
649			    &ns->bpf_map_accept);
650
651	return 0;
652}
653
654void nsim_bpf_uninit(struct netdevsim *ns)
655{
656	WARN_ON(ns->xdp.prog);
657	WARN_ON(ns->xdp_hw.prog);
658	WARN_ON(ns->bpf_offloaded);
659	bpf_offload_dev_netdev_unregister(ns->nsim_dev->bpf_dev, ns->netdev);
660}