xref: /linux/kernel/bpf/bpf_struct_ops.c (revision 249ebf3f65f8530beb2cbfb91bff1d83ba88d23c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2019 Facebook */
3 
4 #include <linux/bpf.h>
5 #include <linux/bpf_verifier.h>
6 #include <linux/btf.h>
7 #include <linux/filter.h>
8 #include <linux/slab.h>
9 #include <linux/numa.h>
10 #include <linux/seq_file.h>
11 #include <linux/refcount.h>
12 #include <linux/mutex.h>
13 #include <linux/btf_ids.h>
14 #include <linux/rcupdate_wait.h>
15 
16 struct bpf_struct_ops_value {
17 	struct bpf_struct_ops_common_value common;
18 	char data[] ____cacheline_aligned_in_smp;
19 };
20 
21 #define MAX_TRAMP_IMAGE_PAGES 8
22 
23 struct bpf_struct_ops_map {
24 	struct bpf_map map;
25 	struct rcu_head rcu;
26 	const struct bpf_struct_ops_desc *st_ops_desc;
27 	/* protect map_update */
28 	struct mutex lock;
29 	/* link has all the bpf_links that is populated
30 	 * to the func ptr of the kernel's struct
31 	 * (in kvalue.data).
32 	 */
33 	struct bpf_link **links;
34 	u32 links_cnt;
35 	u32 image_pages_cnt;
36 	/* image_pages is an array of pages that has all the trampolines
37 	 * that stores the func args before calling the bpf_prog.
38 	 */
39 	void *image_pages[MAX_TRAMP_IMAGE_PAGES];
40 	/* The owner moduler's btf. */
41 	struct btf *btf;
42 	/* uvalue->data stores the kernel struct
43 	 * (e.g. tcp_congestion_ops) that is more useful
44 	 * to userspace than the kvalue.  For example,
45 	 * the bpf_prog's id is stored instead of the kernel
46 	 * address of a func ptr.
47 	 */
48 	struct bpf_struct_ops_value *uvalue;
49 	/* kvalue.data stores the actual kernel's struct
50 	 * (e.g. tcp_congestion_ops) that will be
51 	 * registered to the kernel subsystem.
52 	 */
53 	struct bpf_struct_ops_value kvalue;
54 };
55 
56 struct bpf_struct_ops_link {
57 	struct bpf_link link;
58 	struct bpf_map __rcu *map;
59 };
60 
61 static DEFINE_MUTEX(update_mutex);
62 
63 #define VALUE_PREFIX "bpf_struct_ops_"
64 #define VALUE_PREFIX_LEN (sizeof(VALUE_PREFIX) - 1)
65 
66 const struct bpf_verifier_ops bpf_struct_ops_verifier_ops = {
67 };
68 
69 const struct bpf_prog_ops bpf_struct_ops_prog_ops = {
70 #ifdef CONFIG_NET
71 	.test_run = bpf_struct_ops_test_run,
72 #endif
73 };
74 
75 BTF_ID_LIST(st_ops_ids)
76 BTF_ID(struct, module)
77 BTF_ID(struct, bpf_struct_ops_common_value)
78 
79 enum {
80 	IDX_MODULE_ID,
81 	IDX_ST_OPS_COMMON_VALUE_ID,
82 };
83 
84 extern struct btf *btf_vmlinux;
85 
86 static bool is_valid_value_type(struct btf *btf, s32 value_id,
87 				const struct btf_type *type,
88 				const char *value_name)
89 {
90 	const struct btf_type *common_value_type;
91 	const struct btf_member *member;
92 	const struct btf_type *vt, *mt;
93 
94 	vt = btf_type_by_id(btf, value_id);
95 	if (btf_vlen(vt) != 2) {
96 		pr_warn("The number of %s's members should be 2, but we get %d\n",
97 			value_name, btf_vlen(vt));
98 		return false;
99 	}
100 	member = btf_type_member(vt);
101 	mt = btf_type_by_id(btf, member->type);
102 	common_value_type = btf_type_by_id(btf_vmlinux,
103 					   st_ops_ids[IDX_ST_OPS_COMMON_VALUE_ID]);
104 	if (mt != common_value_type) {
105 		pr_warn("The first member of %s should be bpf_struct_ops_common_value\n",
106 			value_name);
107 		return false;
108 	}
109 	member++;
110 	mt = btf_type_by_id(btf, member->type);
111 	if (mt != type) {
112 		pr_warn("The second member of %s should be %s\n",
113 			value_name, btf_name_by_offset(btf, type->name_off));
114 		return false;
115 	}
116 
117 	return true;
118 }
119 
120 static void *bpf_struct_ops_image_alloc(void)
121 {
122 	void *image;
123 	int err;
124 
125 	err = bpf_jit_charge_modmem(PAGE_SIZE);
126 	if (err)
127 		return ERR_PTR(err);
128 	image = arch_alloc_bpf_trampoline(PAGE_SIZE);
129 	if (!image) {
130 		bpf_jit_uncharge_modmem(PAGE_SIZE);
131 		return ERR_PTR(-ENOMEM);
132 	}
133 
134 	return image;
135 }
136 
137 void bpf_struct_ops_image_free(void *image)
138 {
139 	if (image) {
140 		arch_free_bpf_trampoline(image, PAGE_SIZE);
141 		bpf_jit_uncharge_modmem(PAGE_SIZE);
142 	}
143 }
144 
145 #define MAYBE_NULL_SUFFIX "__nullable"
146 #define MAX_STUB_NAME 128
147 
148 /* Return the type info of a stub function, if it exists.
149  *
150  * The name of a stub function is made up of the name of the struct_ops and
151  * the name of the function pointer member, separated by "__". For example,
152  * if the struct_ops type is named "foo_ops" and the function pointer
153  * member is named "bar", the stub function name would be "foo_ops__bar".
154  */
155 static const struct btf_type *
156 find_stub_func_proto(const struct btf *btf, const char *st_op_name,
157 		     const char *member_name)
158 {
159 	char stub_func_name[MAX_STUB_NAME];
160 	const struct btf_type *func_type;
161 	s32 btf_id;
162 	int cp;
163 
164 	cp = snprintf(stub_func_name, MAX_STUB_NAME, "%s__%s",
165 		      st_op_name, member_name);
166 	if (cp >= MAX_STUB_NAME) {
167 		pr_warn("Stub function name too long\n");
168 		return NULL;
169 	}
170 	btf_id = btf_find_by_name_kind(btf, stub_func_name, BTF_KIND_FUNC);
171 	if (btf_id < 0)
172 		return NULL;
173 	func_type = btf_type_by_id(btf, btf_id);
174 	if (!func_type)
175 		return NULL;
176 
177 	return btf_type_by_id(btf, func_type->type); /* FUNC_PROTO */
178 }
179 
180 /* Prepare argument info for every nullable argument of a member of a
181  * struct_ops type.
182  *
183  * Initialize a struct bpf_struct_ops_arg_info according to type info of
184  * the arguments of a stub function. (Check kCFI for more information about
185  * stub functions.)
186  *
187  * Each member in the struct_ops type has a struct bpf_struct_ops_arg_info
188  * to provide an array of struct bpf_ctx_arg_aux, which in turn provides
189  * the information that used by the verifier to check the arguments of the
190  * BPF struct_ops program assigned to the member. Here, we only care about
191  * the arguments that are marked as __nullable.
192  *
193  * The array of struct bpf_ctx_arg_aux is eventually assigned to
194  * prog->aux->ctx_arg_info of BPF struct_ops programs and passed to the
195  * verifier. (See check_struct_ops_btf_id())
196  *
197  * arg_info->info will be the list of struct bpf_ctx_arg_aux if success. If
198  * fails, it will be kept untouched.
199  */
200 static int prepare_arg_info(struct btf *btf,
201 			    const char *st_ops_name,
202 			    const char *member_name,
203 			    const struct btf_type *func_proto,
204 			    struct bpf_struct_ops_arg_info *arg_info)
205 {
206 	const struct btf_type *stub_func_proto, *pointed_type;
207 	const struct btf_param *stub_args, *args;
208 	struct bpf_ctx_arg_aux *info, *info_buf;
209 	u32 nargs, arg_no, info_cnt = 0;
210 	u32 arg_btf_id;
211 	int offset;
212 
213 	stub_func_proto = find_stub_func_proto(btf, st_ops_name, member_name);
214 	if (!stub_func_proto)
215 		return 0;
216 
217 	/* Check if the number of arguments of the stub function is the same
218 	 * as the number of arguments of the function pointer.
219 	 */
220 	nargs = btf_type_vlen(func_proto);
221 	if (nargs != btf_type_vlen(stub_func_proto)) {
222 		pr_warn("the number of arguments of the stub function %s__%s does not match the number of arguments of the member %s of struct %s\n",
223 			st_ops_name, member_name, member_name, st_ops_name);
224 		return -EINVAL;
225 	}
226 
227 	if (!nargs)
228 		return 0;
229 
230 	args = btf_params(func_proto);
231 	stub_args = btf_params(stub_func_proto);
232 
233 	info_buf = kcalloc(nargs, sizeof(*info_buf), GFP_KERNEL);
234 	if (!info_buf)
235 		return -ENOMEM;
236 
237 	/* Prepare info for every nullable argument */
238 	info = info_buf;
239 	for (arg_no = 0; arg_no < nargs; arg_no++) {
240 		/* Skip arguments that is not suffixed with
241 		 * "__nullable".
242 		 */
243 		if (!btf_param_match_suffix(btf, &stub_args[arg_no],
244 					    MAYBE_NULL_SUFFIX))
245 			continue;
246 
247 		/* Should be a pointer to struct */
248 		pointed_type = btf_type_resolve_ptr(btf,
249 						    args[arg_no].type,
250 						    &arg_btf_id);
251 		if (!pointed_type ||
252 		    !btf_type_is_struct(pointed_type)) {
253 			pr_warn("stub function %s__%s has %s tagging to an unsupported type\n",
254 				st_ops_name, member_name, MAYBE_NULL_SUFFIX);
255 			goto err_out;
256 		}
257 
258 		offset = btf_ctx_arg_offset(btf, func_proto, arg_no);
259 		if (offset < 0) {
260 			pr_warn("stub function %s__%s has an invalid trampoline ctx offset for arg#%u\n",
261 				st_ops_name, member_name, arg_no);
262 			goto err_out;
263 		}
264 
265 		if (args[arg_no].type != stub_args[arg_no].type) {
266 			pr_warn("arg#%u type in stub function %s__%s does not match with its original func_proto\n",
267 				arg_no, st_ops_name, member_name);
268 			goto err_out;
269 		}
270 
271 		/* Fill the information of the new argument */
272 		info->reg_type =
273 			PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
274 		info->btf_id = arg_btf_id;
275 		info->btf = btf;
276 		info->offset = offset;
277 
278 		info++;
279 		info_cnt++;
280 	}
281 
282 	if (info_cnt) {
283 		arg_info->info = info_buf;
284 		arg_info->cnt = info_cnt;
285 	} else {
286 		kfree(info_buf);
287 	}
288 
289 	return 0;
290 
291 err_out:
292 	kfree(info_buf);
293 
294 	return -EINVAL;
295 }
296 
297 /* Clean up the arg_info in a struct bpf_struct_ops_desc. */
298 void bpf_struct_ops_desc_release(struct bpf_struct_ops_desc *st_ops_desc)
299 {
300 	struct bpf_struct_ops_arg_info *arg_info;
301 	int i;
302 
303 	arg_info = st_ops_desc->arg_info;
304 	for (i = 0; i < btf_type_vlen(st_ops_desc->type); i++)
305 		kfree(arg_info[i].info);
306 
307 	kfree(arg_info);
308 }
309 
310 int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
311 			     struct btf *btf,
312 			     struct bpf_verifier_log *log)
313 {
314 	struct bpf_struct_ops *st_ops = st_ops_desc->st_ops;
315 	struct bpf_struct_ops_arg_info *arg_info;
316 	const struct btf_member *member;
317 	const struct btf_type *t;
318 	s32 type_id, value_id;
319 	char value_name[128];
320 	const char *mname;
321 	int i, err;
322 
323 	if (strlen(st_ops->name) + VALUE_PREFIX_LEN >=
324 	    sizeof(value_name)) {
325 		pr_warn("struct_ops name %s is too long\n",
326 			st_ops->name);
327 		return -EINVAL;
328 	}
329 	sprintf(value_name, "%s%s", VALUE_PREFIX, st_ops->name);
330 
331 	if (!st_ops->cfi_stubs) {
332 		pr_warn("struct_ops for %s has no cfi_stubs\n", st_ops->name);
333 		return -EINVAL;
334 	}
335 
336 	type_id = btf_find_by_name_kind(btf, st_ops->name,
337 					BTF_KIND_STRUCT);
338 	if (type_id < 0) {
339 		pr_warn("Cannot find struct %s in %s\n",
340 			st_ops->name, btf_get_name(btf));
341 		return -EINVAL;
342 	}
343 	t = btf_type_by_id(btf, type_id);
344 	if (btf_type_vlen(t) > BPF_STRUCT_OPS_MAX_NR_MEMBERS) {
345 		pr_warn("Cannot support #%u members in struct %s\n",
346 			btf_type_vlen(t), st_ops->name);
347 		return -EINVAL;
348 	}
349 
350 	value_id = btf_find_by_name_kind(btf, value_name,
351 					 BTF_KIND_STRUCT);
352 	if (value_id < 0) {
353 		pr_warn("Cannot find struct %s in %s\n",
354 			value_name, btf_get_name(btf));
355 		return -EINVAL;
356 	}
357 	if (!is_valid_value_type(btf, value_id, t, value_name))
358 		return -EINVAL;
359 
360 	arg_info = kcalloc(btf_type_vlen(t), sizeof(*arg_info),
361 			   GFP_KERNEL);
362 	if (!arg_info)
363 		return -ENOMEM;
364 
365 	st_ops_desc->arg_info = arg_info;
366 	st_ops_desc->type = t;
367 	st_ops_desc->type_id = type_id;
368 	st_ops_desc->value_id = value_id;
369 	st_ops_desc->value_type = btf_type_by_id(btf, value_id);
370 
371 	for_each_member(i, t, member) {
372 		const struct btf_type *func_proto;
373 
374 		mname = btf_name_by_offset(btf, member->name_off);
375 		if (!*mname) {
376 			pr_warn("anon member in struct %s is not supported\n",
377 				st_ops->name);
378 			err = -EOPNOTSUPP;
379 			goto errout;
380 		}
381 
382 		if (__btf_member_bitfield_size(t, member)) {
383 			pr_warn("bit field member %s in struct %s is not supported\n",
384 				mname, st_ops->name);
385 			err = -EOPNOTSUPP;
386 			goto errout;
387 		}
388 
389 		func_proto = btf_type_resolve_func_ptr(btf,
390 						       member->type,
391 						       NULL);
392 		if (!func_proto)
393 			continue;
394 
395 		if (btf_distill_func_proto(log, btf,
396 					   func_proto, mname,
397 					   &st_ops->func_models[i])) {
398 			pr_warn("Error in parsing func ptr %s in struct %s\n",
399 				mname, st_ops->name);
400 			err = -EINVAL;
401 			goto errout;
402 		}
403 
404 		err = prepare_arg_info(btf, st_ops->name, mname,
405 				       func_proto,
406 				       arg_info + i);
407 		if (err)
408 			goto errout;
409 	}
410 
411 	if (st_ops->init(btf)) {
412 		pr_warn("Error in init bpf_struct_ops %s\n",
413 			st_ops->name);
414 		err = -EINVAL;
415 		goto errout;
416 	}
417 
418 	return 0;
419 
420 errout:
421 	bpf_struct_ops_desc_release(st_ops_desc);
422 
423 	return err;
424 }
425 
426 static int bpf_struct_ops_map_get_next_key(struct bpf_map *map, void *key,
427 					   void *next_key)
428 {
429 	if (key && *(u32 *)key == 0)
430 		return -ENOENT;
431 
432 	*(u32 *)next_key = 0;
433 	return 0;
434 }
435 
436 int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key,
437 				       void *value)
438 {
439 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
440 	struct bpf_struct_ops_value *uvalue, *kvalue;
441 	enum bpf_struct_ops_state state;
442 	s64 refcnt;
443 
444 	if (unlikely(*(u32 *)key != 0))
445 		return -ENOENT;
446 
447 	kvalue = &st_map->kvalue;
448 	/* Pair with smp_store_release() during map_update */
449 	state = smp_load_acquire(&kvalue->common.state);
450 	if (state == BPF_STRUCT_OPS_STATE_INIT) {
451 		memset(value, 0, map->value_size);
452 		return 0;
453 	}
454 
455 	/* No lock is needed.  state and refcnt do not need
456 	 * to be updated together under atomic context.
457 	 */
458 	uvalue = value;
459 	memcpy(uvalue, st_map->uvalue, map->value_size);
460 	uvalue->common.state = state;
461 
462 	/* This value offers the user space a general estimate of how
463 	 * many sockets are still utilizing this struct_ops for TCP
464 	 * congestion control. The number might not be exact, but it
465 	 * should sufficiently meet our present goals.
466 	 */
467 	refcnt = atomic64_read(&map->refcnt) - atomic64_read(&map->usercnt);
468 	refcount_set(&uvalue->common.refcnt, max_t(s64, refcnt, 0));
469 
470 	return 0;
471 }
472 
473 static void *bpf_struct_ops_map_lookup_elem(struct bpf_map *map, void *key)
474 {
475 	return ERR_PTR(-EINVAL);
476 }
477 
478 static void bpf_struct_ops_map_put_progs(struct bpf_struct_ops_map *st_map)
479 {
480 	u32 i;
481 
482 	for (i = 0; i < st_map->links_cnt; i++) {
483 		if (st_map->links[i]) {
484 			bpf_link_put(st_map->links[i]);
485 			st_map->links[i] = NULL;
486 		}
487 	}
488 }
489 
490 static void bpf_struct_ops_map_free_image(struct bpf_struct_ops_map *st_map)
491 {
492 	int i;
493 
494 	for (i = 0; i < st_map->image_pages_cnt; i++)
495 		bpf_struct_ops_image_free(st_map->image_pages[i]);
496 	st_map->image_pages_cnt = 0;
497 }
498 
499 static int check_zero_holes(const struct btf *btf, const struct btf_type *t, void *data)
500 {
501 	const struct btf_member *member;
502 	u32 i, moff, msize, prev_mend = 0;
503 	const struct btf_type *mtype;
504 
505 	for_each_member(i, t, member) {
506 		moff = __btf_member_bit_offset(t, member) / 8;
507 		if (moff > prev_mend &&
508 		    memchr_inv(data + prev_mend, 0, moff - prev_mend))
509 			return -EINVAL;
510 
511 		mtype = btf_type_by_id(btf, member->type);
512 		mtype = btf_resolve_size(btf, mtype, &msize);
513 		if (IS_ERR(mtype))
514 			return PTR_ERR(mtype);
515 		prev_mend = moff + msize;
516 	}
517 
518 	if (t->size > prev_mend &&
519 	    memchr_inv(data + prev_mend, 0, t->size - prev_mend))
520 		return -EINVAL;
521 
522 	return 0;
523 }
524 
525 static void bpf_struct_ops_link_release(struct bpf_link *link)
526 {
527 }
528 
529 static void bpf_struct_ops_link_dealloc(struct bpf_link *link)
530 {
531 	struct bpf_tramp_link *tlink = container_of(link, struct bpf_tramp_link, link);
532 
533 	kfree(tlink);
534 }
535 
536 const struct bpf_link_ops bpf_struct_ops_link_lops = {
537 	.release = bpf_struct_ops_link_release,
538 	.dealloc = bpf_struct_ops_link_dealloc,
539 };
540 
541 int bpf_struct_ops_prepare_trampoline(struct bpf_tramp_links *tlinks,
542 				      struct bpf_tramp_link *link,
543 				      const struct btf_func_model *model,
544 				      void *stub_func,
545 				      void **_image, u32 *_image_off,
546 				      bool allow_alloc)
547 {
548 	u32 image_off = *_image_off, flags = BPF_TRAMP_F_INDIRECT;
549 	void *image = *_image;
550 	int size;
551 
552 	tlinks[BPF_TRAMP_FENTRY].links[0] = link;
553 	tlinks[BPF_TRAMP_FENTRY].nr_links = 1;
554 
555 	if (model->ret_size > 0)
556 		flags |= BPF_TRAMP_F_RET_FENTRY_RET;
557 
558 	size = arch_bpf_trampoline_size(model, flags, tlinks, NULL);
559 	if (size <= 0)
560 		return size ? : -EFAULT;
561 
562 	/* Allocate image buffer if necessary */
563 	if (!image || size > PAGE_SIZE - image_off) {
564 		if (!allow_alloc)
565 			return -E2BIG;
566 
567 		image = bpf_struct_ops_image_alloc();
568 		if (IS_ERR(image))
569 			return PTR_ERR(image);
570 		image_off = 0;
571 	}
572 
573 	size = arch_prepare_bpf_trampoline(NULL, image + image_off,
574 					   image + PAGE_SIZE,
575 					   model, flags, tlinks, stub_func);
576 	if (size <= 0) {
577 		if (image != *_image)
578 			bpf_struct_ops_image_free(image);
579 		return size ? : -EFAULT;
580 	}
581 
582 	*_image = image;
583 	*_image_off = image_off + size;
584 	return 0;
585 }
586 
587 static long bpf_struct_ops_map_update_elem(struct bpf_map *map, void *key,
588 					   void *value, u64 flags)
589 {
590 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
591 	const struct bpf_struct_ops_desc *st_ops_desc = st_map->st_ops_desc;
592 	const struct bpf_struct_ops *st_ops = st_ops_desc->st_ops;
593 	struct bpf_struct_ops_value *uvalue, *kvalue;
594 	const struct btf_type *module_type;
595 	const struct btf_member *member;
596 	const struct btf_type *t = st_ops_desc->type;
597 	struct bpf_tramp_links *tlinks;
598 	void *udata, *kdata;
599 	int prog_fd, err;
600 	u32 i, trampoline_start, image_off = 0;
601 	void *cur_image = NULL, *image = NULL;
602 
603 	if (flags)
604 		return -EINVAL;
605 
606 	if (*(u32 *)key != 0)
607 		return -E2BIG;
608 
609 	err = check_zero_holes(st_map->btf, st_ops_desc->value_type, value);
610 	if (err)
611 		return err;
612 
613 	uvalue = value;
614 	err = check_zero_holes(st_map->btf, t, uvalue->data);
615 	if (err)
616 		return err;
617 
618 	if (uvalue->common.state || refcount_read(&uvalue->common.refcnt))
619 		return -EINVAL;
620 
621 	tlinks = kcalloc(BPF_TRAMP_MAX, sizeof(*tlinks), GFP_KERNEL);
622 	if (!tlinks)
623 		return -ENOMEM;
624 
625 	uvalue = (struct bpf_struct_ops_value *)st_map->uvalue;
626 	kvalue = (struct bpf_struct_ops_value *)&st_map->kvalue;
627 
628 	mutex_lock(&st_map->lock);
629 
630 	if (kvalue->common.state != BPF_STRUCT_OPS_STATE_INIT) {
631 		err = -EBUSY;
632 		goto unlock;
633 	}
634 
635 	memcpy(uvalue, value, map->value_size);
636 
637 	udata = &uvalue->data;
638 	kdata = &kvalue->data;
639 
640 	module_type = btf_type_by_id(btf_vmlinux, st_ops_ids[IDX_MODULE_ID]);
641 	for_each_member(i, t, member) {
642 		const struct btf_type *mtype, *ptype;
643 		struct bpf_prog *prog;
644 		struct bpf_tramp_link *link;
645 		u32 moff;
646 
647 		moff = __btf_member_bit_offset(t, member) / 8;
648 		ptype = btf_type_resolve_ptr(st_map->btf, member->type, NULL);
649 		if (ptype == module_type) {
650 			if (*(void **)(udata + moff))
651 				goto reset_unlock;
652 			*(void **)(kdata + moff) = BPF_MODULE_OWNER;
653 			continue;
654 		}
655 
656 		err = st_ops->init_member(t, member, kdata, udata);
657 		if (err < 0)
658 			goto reset_unlock;
659 
660 		/* The ->init_member() has handled this member */
661 		if (err > 0)
662 			continue;
663 
664 		/* If st_ops->init_member does not handle it,
665 		 * we will only handle func ptrs and zero-ed members
666 		 * here.  Reject everything else.
667 		 */
668 
669 		/* All non func ptr member must be 0 */
670 		if (!ptype || !btf_type_is_func_proto(ptype)) {
671 			u32 msize;
672 
673 			mtype = btf_type_by_id(st_map->btf, member->type);
674 			mtype = btf_resolve_size(st_map->btf, mtype, &msize);
675 			if (IS_ERR(mtype)) {
676 				err = PTR_ERR(mtype);
677 				goto reset_unlock;
678 			}
679 
680 			if (memchr_inv(udata + moff, 0, msize)) {
681 				err = -EINVAL;
682 				goto reset_unlock;
683 			}
684 
685 			continue;
686 		}
687 
688 		prog_fd = (int)(*(unsigned long *)(udata + moff));
689 		/* Similar check as the attr->attach_prog_fd */
690 		if (!prog_fd)
691 			continue;
692 
693 		prog = bpf_prog_get(prog_fd);
694 		if (IS_ERR(prog)) {
695 			err = PTR_ERR(prog);
696 			goto reset_unlock;
697 		}
698 
699 		if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
700 		    prog->aux->attach_btf_id != st_ops_desc->type_id ||
701 		    prog->expected_attach_type != i) {
702 			bpf_prog_put(prog);
703 			err = -EINVAL;
704 			goto reset_unlock;
705 		}
706 
707 		link = kzalloc(sizeof(*link), GFP_USER);
708 		if (!link) {
709 			bpf_prog_put(prog);
710 			err = -ENOMEM;
711 			goto reset_unlock;
712 		}
713 		bpf_link_init(&link->link, BPF_LINK_TYPE_STRUCT_OPS,
714 			      &bpf_struct_ops_link_lops, prog);
715 		st_map->links[i] = &link->link;
716 
717 		trampoline_start = image_off;
718 		err = bpf_struct_ops_prepare_trampoline(tlinks, link,
719 						&st_ops->func_models[i],
720 						*(void **)(st_ops->cfi_stubs + moff),
721 						&image, &image_off,
722 						st_map->image_pages_cnt < MAX_TRAMP_IMAGE_PAGES);
723 		if (err)
724 			goto reset_unlock;
725 
726 		if (cur_image != image) {
727 			st_map->image_pages[st_map->image_pages_cnt++] = image;
728 			cur_image = image;
729 			trampoline_start = 0;
730 		}
731 
732 		*(void **)(kdata + moff) = image + trampoline_start + cfi_get_offset();
733 
734 		/* put prog_id to udata */
735 		*(unsigned long *)(udata + moff) = prog->aux->id;
736 	}
737 
738 	if (st_ops->validate) {
739 		err = st_ops->validate(kdata);
740 		if (err)
741 			goto reset_unlock;
742 	}
743 	for (i = 0; i < st_map->image_pages_cnt; i++) {
744 		err = arch_protect_bpf_trampoline(st_map->image_pages[i],
745 						  PAGE_SIZE);
746 		if (err)
747 			goto reset_unlock;
748 	}
749 
750 	if (st_map->map.map_flags & BPF_F_LINK) {
751 		err = 0;
752 		/* Let bpf_link handle registration & unregistration.
753 		 *
754 		 * Pair with smp_load_acquire() during lookup_elem().
755 		 */
756 		smp_store_release(&kvalue->common.state, BPF_STRUCT_OPS_STATE_READY);
757 		goto unlock;
758 	}
759 
760 	err = st_ops->reg(kdata);
761 	if (likely(!err)) {
762 		/* This refcnt increment on the map here after
763 		 * 'st_ops->reg()' is secure since the state of the
764 		 * map must be set to INIT at this moment, and thus
765 		 * bpf_struct_ops_map_delete_elem() can't unregister
766 		 * or transition it to TOBEFREE concurrently.
767 		 */
768 		bpf_map_inc(map);
769 		/* Pair with smp_load_acquire() during lookup_elem().
770 		 * It ensures the above udata updates (e.g. prog->aux->id)
771 		 * can be seen once BPF_STRUCT_OPS_STATE_INUSE is set.
772 		 */
773 		smp_store_release(&kvalue->common.state, BPF_STRUCT_OPS_STATE_INUSE);
774 		goto unlock;
775 	}
776 
777 	/* Error during st_ops->reg(). Can happen if this struct_ops needs to be
778 	 * verified as a whole, after all init_member() calls. Can also happen if
779 	 * there was a race in registering the struct_ops (under the same name) to
780 	 * a sub-system through different struct_ops's maps.
781 	 */
782 
783 reset_unlock:
784 	bpf_struct_ops_map_free_image(st_map);
785 	bpf_struct_ops_map_put_progs(st_map);
786 	memset(uvalue, 0, map->value_size);
787 	memset(kvalue, 0, map->value_size);
788 unlock:
789 	kfree(tlinks);
790 	mutex_unlock(&st_map->lock);
791 	return err;
792 }
793 
794 static long bpf_struct_ops_map_delete_elem(struct bpf_map *map, void *key)
795 {
796 	enum bpf_struct_ops_state prev_state;
797 	struct bpf_struct_ops_map *st_map;
798 
799 	st_map = (struct bpf_struct_ops_map *)map;
800 	if (st_map->map.map_flags & BPF_F_LINK)
801 		return -EOPNOTSUPP;
802 
803 	prev_state = cmpxchg(&st_map->kvalue.common.state,
804 			     BPF_STRUCT_OPS_STATE_INUSE,
805 			     BPF_STRUCT_OPS_STATE_TOBEFREE);
806 	switch (prev_state) {
807 	case BPF_STRUCT_OPS_STATE_INUSE:
808 		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data);
809 		bpf_map_put(map);
810 		return 0;
811 	case BPF_STRUCT_OPS_STATE_TOBEFREE:
812 		return -EINPROGRESS;
813 	case BPF_STRUCT_OPS_STATE_INIT:
814 		return -ENOENT;
815 	default:
816 		WARN_ON_ONCE(1);
817 		/* Should never happen.  Treat it as not found. */
818 		return -ENOENT;
819 	}
820 }
821 
822 static void bpf_struct_ops_map_seq_show_elem(struct bpf_map *map, void *key,
823 					     struct seq_file *m)
824 {
825 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
826 	void *value;
827 	int err;
828 
829 	value = kmalloc(map->value_size, GFP_USER | __GFP_NOWARN);
830 	if (!value)
831 		return;
832 
833 	err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
834 	if (!err) {
835 		btf_type_seq_show(st_map->btf,
836 				  map->btf_vmlinux_value_type_id,
837 				  value, m);
838 		seq_puts(m, "\n");
839 	}
840 
841 	kfree(value);
842 }
843 
844 static void __bpf_struct_ops_map_free(struct bpf_map *map)
845 {
846 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
847 
848 	if (st_map->links)
849 		bpf_struct_ops_map_put_progs(st_map);
850 	bpf_map_area_free(st_map->links);
851 	bpf_struct_ops_map_free_image(st_map);
852 	bpf_map_area_free(st_map->uvalue);
853 	bpf_map_area_free(st_map);
854 }
855 
856 static void bpf_struct_ops_map_free(struct bpf_map *map)
857 {
858 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
859 
860 	/* st_ops->owner was acquired during map_alloc to implicitly holds
861 	 * the btf's refcnt. The acquire was only done when btf_is_module()
862 	 * st_map->btf cannot be NULL here.
863 	 */
864 	if (btf_is_module(st_map->btf))
865 		module_put(st_map->st_ops_desc->st_ops->owner);
866 
867 	/* The struct_ops's function may switch to another struct_ops.
868 	 *
869 	 * For example, bpf_tcp_cc_x->init() may switch to
870 	 * another tcp_cc_y by calling
871 	 * setsockopt(TCP_CONGESTION, "tcp_cc_y").
872 	 * During the switch,  bpf_struct_ops_put(tcp_cc_x) is called
873 	 * and its refcount may reach 0 which then free its
874 	 * trampoline image while tcp_cc_x is still running.
875 	 *
876 	 * A vanilla rcu gp is to wait for all bpf-tcp-cc prog
877 	 * to finish. bpf-tcp-cc prog is non sleepable.
878 	 * A rcu_tasks gp is to wait for the last few insn
879 	 * in the tramopline image to finish before releasing
880 	 * the trampoline image.
881 	 */
882 	synchronize_rcu_mult(call_rcu, call_rcu_tasks);
883 
884 	__bpf_struct_ops_map_free(map);
885 }
886 
887 static int bpf_struct_ops_map_alloc_check(union bpf_attr *attr)
888 {
889 	if (attr->key_size != sizeof(unsigned int) || attr->max_entries != 1 ||
890 	    (attr->map_flags & ~(BPF_F_LINK | BPF_F_VTYPE_BTF_OBJ_FD)) ||
891 	    !attr->btf_vmlinux_value_type_id)
892 		return -EINVAL;
893 	return 0;
894 }
895 
896 static struct bpf_map *bpf_struct_ops_map_alloc(union bpf_attr *attr)
897 {
898 	const struct bpf_struct_ops_desc *st_ops_desc;
899 	size_t st_map_size;
900 	struct bpf_struct_ops_map *st_map;
901 	const struct btf_type *t, *vt;
902 	struct module *mod = NULL;
903 	struct bpf_map *map;
904 	struct btf *btf;
905 	int ret;
906 
907 	if (attr->map_flags & BPF_F_VTYPE_BTF_OBJ_FD) {
908 		/* The map holds btf for its whole life time. */
909 		btf = btf_get_by_fd(attr->value_type_btf_obj_fd);
910 		if (IS_ERR(btf))
911 			return ERR_CAST(btf);
912 		if (!btf_is_module(btf)) {
913 			btf_put(btf);
914 			return ERR_PTR(-EINVAL);
915 		}
916 
917 		mod = btf_try_get_module(btf);
918 		/* mod holds a refcnt to btf. We don't need an extra refcnt
919 		 * here.
920 		 */
921 		btf_put(btf);
922 		if (!mod)
923 			return ERR_PTR(-EINVAL);
924 	} else {
925 		btf = bpf_get_btf_vmlinux();
926 		if (IS_ERR(btf))
927 			return ERR_CAST(btf);
928 		if (!btf)
929 			return ERR_PTR(-ENOTSUPP);
930 	}
931 
932 	st_ops_desc = bpf_struct_ops_find_value(btf, attr->btf_vmlinux_value_type_id);
933 	if (!st_ops_desc) {
934 		ret = -ENOTSUPP;
935 		goto errout;
936 	}
937 
938 	vt = st_ops_desc->value_type;
939 	if (attr->value_size != vt->size) {
940 		ret = -EINVAL;
941 		goto errout;
942 	}
943 
944 	t = st_ops_desc->type;
945 
946 	st_map_size = sizeof(*st_map) +
947 		/* kvalue stores the
948 		 * struct bpf_struct_ops_tcp_congestions_ops
949 		 */
950 		(vt->size - sizeof(struct bpf_struct_ops_value));
951 
952 	st_map = bpf_map_area_alloc(st_map_size, NUMA_NO_NODE);
953 	if (!st_map) {
954 		ret = -ENOMEM;
955 		goto errout;
956 	}
957 
958 	st_map->st_ops_desc = st_ops_desc;
959 	map = &st_map->map;
960 
961 	st_map->uvalue = bpf_map_area_alloc(vt->size, NUMA_NO_NODE);
962 	st_map->links_cnt = btf_type_vlen(t);
963 	st_map->links =
964 		bpf_map_area_alloc(st_map->links_cnt * sizeof(struct bpf_links *),
965 				   NUMA_NO_NODE);
966 	if (!st_map->uvalue || !st_map->links) {
967 		ret = -ENOMEM;
968 		goto errout_free;
969 	}
970 	st_map->btf = btf;
971 
972 	mutex_init(&st_map->lock);
973 	bpf_map_init_from_attr(map, attr);
974 
975 	return map;
976 
977 errout_free:
978 	__bpf_struct_ops_map_free(map);
979 errout:
980 	module_put(mod);
981 
982 	return ERR_PTR(ret);
983 }
984 
985 static u64 bpf_struct_ops_map_mem_usage(const struct bpf_map *map)
986 {
987 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
988 	const struct bpf_struct_ops_desc *st_ops_desc = st_map->st_ops_desc;
989 	const struct btf_type *vt = st_ops_desc->value_type;
990 	u64 usage;
991 
992 	usage = sizeof(*st_map) +
993 			vt->size - sizeof(struct bpf_struct_ops_value);
994 	usage += vt->size;
995 	usage += btf_type_vlen(vt) * sizeof(struct bpf_links *);
996 	usage += PAGE_SIZE;
997 	return usage;
998 }
999 
1000 BTF_ID_LIST_SINGLE(bpf_struct_ops_map_btf_ids, struct, bpf_struct_ops_map)
1001 const struct bpf_map_ops bpf_struct_ops_map_ops = {
1002 	.map_alloc_check = bpf_struct_ops_map_alloc_check,
1003 	.map_alloc = bpf_struct_ops_map_alloc,
1004 	.map_free = bpf_struct_ops_map_free,
1005 	.map_get_next_key = bpf_struct_ops_map_get_next_key,
1006 	.map_lookup_elem = bpf_struct_ops_map_lookup_elem,
1007 	.map_delete_elem = bpf_struct_ops_map_delete_elem,
1008 	.map_update_elem = bpf_struct_ops_map_update_elem,
1009 	.map_seq_show_elem = bpf_struct_ops_map_seq_show_elem,
1010 	.map_mem_usage = bpf_struct_ops_map_mem_usage,
1011 	.map_btf_id = &bpf_struct_ops_map_btf_ids[0],
1012 };
1013 
1014 /* "const void *" because some subsystem is
1015  * passing a const (e.g. const struct tcp_congestion_ops *)
1016  */
1017 bool bpf_struct_ops_get(const void *kdata)
1018 {
1019 	struct bpf_struct_ops_value *kvalue;
1020 	struct bpf_struct_ops_map *st_map;
1021 	struct bpf_map *map;
1022 
1023 	kvalue = container_of(kdata, struct bpf_struct_ops_value, data);
1024 	st_map = container_of(kvalue, struct bpf_struct_ops_map, kvalue);
1025 
1026 	map = __bpf_map_inc_not_zero(&st_map->map, false);
1027 	return !IS_ERR(map);
1028 }
1029 
1030 void bpf_struct_ops_put(const void *kdata)
1031 {
1032 	struct bpf_struct_ops_value *kvalue;
1033 	struct bpf_struct_ops_map *st_map;
1034 
1035 	kvalue = container_of(kdata, struct bpf_struct_ops_value, data);
1036 	st_map = container_of(kvalue, struct bpf_struct_ops_map, kvalue);
1037 
1038 	bpf_map_put(&st_map->map);
1039 }
1040 
1041 static bool bpf_struct_ops_valid_to_reg(struct bpf_map *map)
1042 {
1043 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
1044 
1045 	return map->map_type == BPF_MAP_TYPE_STRUCT_OPS &&
1046 		map->map_flags & BPF_F_LINK &&
1047 		/* Pair with smp_store_release() during map_update */
1048 		smp_load_acquire(&st_map->kvalue.common.state) == BPF_STRUCT_OPS_STATE_READY;
1049 }
1050 
1051 static void bpf_struct_ops_map_link_dealloc(struct bpf_link *link)
1052 {
1053 	struct bpf_struct_ops_link *st_link;
1054 	struct bpf_struct_ops_map *st_map;
1055 
1056 	st_link = container_of(link, struct bpf_struct_ops_link, link);
1057 	st_map = (struct bpf_struct_ops_map *)
1058 		rcu_dereference_protected(st_link->map, true);
1059 	if (st_map) {
1060 		/* st_link->map can be NULL if
1061 		 * bpf_struct_ops_link_create() fails to register.
1062 		 */
1063 		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data);
1064 		bpf_map_put(&st_map->map);
1065 	}
1066 	kfree(st_link);
1067 }
1068 
1069 static void bpf_struct_ops_map_link_show_fdinfo(const struct bpf_link *link,
1070 					    struct seq_file *seq)
1071 {
1072 	struct bpf_struct_ops_link *st_link;
1073 	struct bpf_map *map;
1074 
1075 	st_link = container_of(link, struct bpf_struct_ops_link, link);
1076 	rcu_read_lock();
1077 	map = rcu_dereference(st_link->map);
1078 	seq_printf(seq, "map_id:\t%d\n", map->id);
1079 	rcu_read_unlock();
1080 }
1081 
1082 static int bpf_struct_ops_map_link_fill_link_info(const struct bpf_link *link,
1083 					       struct bpf_link_info *info)
1084 {
1085 	struct bpf_struct_ops_link *st_link;
1086 	struct bpf_map *map;
1087 
1088 	st_link = container_of(link, struct bpf_struct_ops_link, link);
1089 	rcu_read_lock();
1090 	map = rcu_dereference(st_link->map);
1091 	info->struct_ops.map_id = map->id;
1092 	rcu_read_unlock();
1093 	return 0;
1094 }
1095 
1096 static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map *new_map,
1097 					  struct bpf_map *expected_old_map)
1098 {
1099 	struct bpf_struct_ops_map *st_map, *old_st_map;
1100 	struct bpf_map *old_map;
1101 	struct bpf_struct_ops_link *st_link;
1102 	int err;
1103 
1104 	st_link = container_of(link, struct bpf_struct_ops_link, link);
1105 	st_map = container_of(new_map, struct bpf_struct_ops_map, map);
1106 
1107 	if (!bpf_struct_ops_valid_to_reg(new_map))
1108 		return -EINVAL;
1109 
1110 	if (!st_map->st_ops_desc->st_ops->update)
1111 		return -EOPNOTSUPP;
1112 
1113 	mutex_lock(&update_mutex);
1114 
1115 	old_map = rcu_dereference_protected(st_link->map, lockdep_is_held(&update_mutex));
1116 	if (expected_old_map && old_map != expected_old_map) {
1117 		err = -EPERM;
1118 		goto err_out;
1119 	}
1120 
1121 	old_st_map = container_of(old_map, struct bpf_struct_ops_map, map);
1122 	/* The new and old struct_ops must be the same type. */
1123 	if (st_map->st_ops_desc != old_st_map->st_ops_desc) {
1124 		err = -EINVAL;
1125 		goto err_out;
1126 	}
1127 
1128 	err = st_map->st_ops_desc->st_ops->update(st_map->kvalue.data, old_st_map->kvalue.data);
1129 	if (err)
1130 		goto err_out;
1131 
1132 	bpf_map_inc(new_map);
1133 	rcu_assign_pointer(st_link->map, new_map);
1134 	bpf_map_put(old_map);
1135 
1136 err_out:
1137 	mutex_unlock(&update_mutex);
1138 
1139 	return err;
1140 }
1141 
1142 static const struct bpf_link_ops bpf_struct_ops_map_lops = {
1143 	.dealloc = bpf_struct_ops_map_link_dealloc,
1144 	.show_fdinfo = bpf_struct_ops_map_link_show_fdinfo,
1145 	.fill_link_info = bpf_struct_ops_map_link_fill_link_info,
1146 	.update_map = bpf_struct_ops_map_link_update,
1147 };
1148 
1149 int bpf_struct_ops_link_create(union bpf_attr *attr)
1150 {
1151 	struct bpf_struct_ops_link *link = NULL;
1152 	struct bpf_link_primer link_primer;
1153 	struct bpf_struct_ops_map *st_map;
1154 	struct bpf_map *map;
1155 	int err;
1156 
1157 	map = bpf_map_get(attr->link_create.map_fd);
1158 	if (IS_ERR(map))
1159 		return PTR_ERR(map);
1160 
1161 	st_map = (struct bpf_struct_ops_map *)map;
1162 
1163 	if (!bpf_struct_ops_valid_to_reg(map)) {
1164 		err = -EINVAL;
1165 		goto err_out;
1166 	}
1167 
1168 	link = kzalloc(sizeof(*link), GFP_USER);
1169 	if (!link) {
1170 		err = -ENOMEM;
1171 		goto err_out;
1172 	}
1173 	bpf_link_init(&link->link, BPF_LINK_TYPE_STRUCT_OPS, &bpf_struct_ops_map_lops, NULL);
1174 
1175 	err = bpf_link_prime(&link->link, &link_primer);
1176 	if (err)
1177 		goto err_out;
1178 
1179 	err = st_map->st_ops_desc->st_ops->reg(st_map->kvalue.data);
1180 	if (err) {
1181 		bpf_link_cleanup(&link_primer);
1182 		link = NULL;
1183 		goto err_out;
1184 	}
1185 	RCU_INIT_POINTER(link->map, map);
1186 
1187 	return bpf_link_settle(&link_primer);
1188 
1189 err_out:
1190 	bpf_map_put(map);
1191 	kfree(link);
1192 	return err;
1193 }
1194 
1195 void bpf_map_struct_ops_info_fill(struct bpf_map_info *info, struct bpf_map *map)
1196 {
1197 	struct bpf_struct_ops_map *st_map = (struct bpf_struct_ops_map *)map;
1198 
1199 	info->btf_vmlinux_id = btf_obj_id(st_map->btf);
1200 }
1201