xref: /openbmc/linux/kernel/bpf/core.c (revision 5a170e9e)
1 /*
2  * Linux Socket Filter - Kernel level socket filtering
3  *
4  * Based on the design of the Berkeley Packet Filter. The new
5  * internal format has been designed by PLUMgrid:
6  *
7  *	Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
8  *
9  * Authors:
10  *
11  *	Jay Schulist <jschlst@samba.org>
12  *	Alexei Starovoitov <ast@plumgrid.com>
13  *	Daniel Borkmann <dborkman@redhat.com>
14  *
15  * This program is free software; you can redistribute it and/or
16  * modify it under the terms of the GNU General Public License
17  * as published by the Free Software Foundation; either version
18  * 2 of the License, or (at your option) any later version.
19  *
20  * Andi Kleen - Fix a few bad bugs and races.
21  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
22  */
23 
24 #include <uapi/linux/btf.h>
25 #include <linux/filter.h>
26 #include <linux/skbuff.h>
27 #include <linux/vmalloc.h>
28 #include <linux/random.h>
29 #include <linux/moduleloader.h>
30 #include <linux/bpf.h>
31 #include <linux/btf.h>
32 #include <linux/frame.h>
33 #include <linux/rbtree_latch.h>
34 #include <linux/kallsyms.h>
35 #include <linux/rcupdate.h>
36 #include <linux/perf_event.h>
37 
38 #include <asm/unaligned.h>
39 
40 /* Registers */
41 #define BPF_R0	regs[BPF_REG_0]
42 #define BPF_R1	regs[BPF_REG_1]
43 #define BPF_R2	regs[BPF_REG_2]
44 #define BPF_R3	regs[BPF_REG_3]
45 #define BPF_R4	regs[BPF_REG_4]
46 #define BPF_R5	regs[BPF_REG_5]
47 #define BPF_R6	regs[BPF_REG_6]
48 #define BPF_R7	regs[BPF_REG_7]
49 #define BPF_R8	regs[BPF_REG_8]
50 #define BPF_R9	regs[BPF_REG_9]
51 #define BPF_R10	regs[BPF_REG_10]
52 
53 /* Named registers */
54 #define DST	regs[insn->dst_reg]
55 #define SRC	regs[insn->src_reg]
56 #define FP	regs[BPF_REG_FP]
57 #define AX	regs[BPF_REG_AX]
58 #define ARG1	regs[BPF_REG_ARG1]
59 #define CTX	regs[BPF_REG_CTX]
60 #define IMM	insn->imm
61 
62 /* No hurry in this branch
63  *
64  * Exported for the bpf jit load helper.
65  */
66 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
67 {
68 	u8 *ptr = NULL;
69 
70 	if (k >= SKF_NET_OFF)
71 		ptr = skb_network_header(skb) + k - SKF_NET_OFF;
72 	else if (k >= SKF_LL_OFF)
73 		ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
74 
75 	if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
76 		return ptr;
77 
78 	return NULL;
79 }
80 
81 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
82 {
83 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
84 	struct bpf_prog_aux *aux;
85 	struct bpf_prog *fp;
86 
87 	size = round_up(size, PAGE_SIZE);
88 	fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
89 	if (fp == NULL)
90 		return NULL;
91 
92 	aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
93 	if (aux == NULL) {
94 		vfree(fp);
95 		return NULL;
96 	}
97 
98 	fp->pages = size / PAGE_SIZE;
99 	fp->aux = aux;
100 	fp->aux->prog = fp;
101 	fp->jit_requested = ebpf_jit_enabled();
102 
103 	INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode);
104 
105 	return fp;
106 }
107 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
108 
109 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
110 {
111 	if (!prog->aux->nr_linfo || !prog->jit_requested)
112 		return 0;
113 
114 	prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo,
115 					 sizeof(*prog->aux->jited_linfo),
116 					 GFP_KERNEL | __GFP_NOWARN);
117 	if (!prog->aux->jited_linfo)
118 		return -ENOMEM;
119 
120 	return 0;
121 }
122 
123 void bpf_prog_free_jited_linfo(struct bpf_prog *prog)
124 {
125 	kfree(prog->aux->jited_linfo);
126 	prog->aux->jited_linfo = NULL;
127 }
128 
129 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog)
130 {
131 	if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0])
132 		bpf_prog_free_jited_linfo(prog);
133 }
134 
135 /* The jit engine is responsible to provide an array
136  * for insn_off to the jited_off mapping (insn_to_jit_off).
137  *
138  * The idx to this array is the insn_off.  Hence, the insn_off
139  * here is relative to the prog itself instead of the main prog.
140  * This array has one entry for each xlated bpf insn.
141  *
142  * jited_off is the byte off to the last byte of the jited insn.
143  *
144  * Hence, with
145  * insn_start:
146  *      The first bpf insn off of the prog.  The insn off
147  *      here is relative to the main prog.
148  *      e.g. if prog is a subprog, insn_start > 0
149  * linfo_idx:
150  *      The prog's idx to prog->aux->linfo and jited_linfo
151  *
152  * jited_linfo[linfo_idx] = prog->bpf_func
153  *
154  * For i > linfo_idx,
155  *
156  * jited_linfo[i] = prog->bpf_func +
157  *	insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
158  */
159 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
160 			       const u32 *insn_to_jit_off)
161 {
162 	u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
163 	const struct bpf_line_info *linfo;
164 	void **jited_linfo;
165 
166 	if (!prog->aux->jited_linfo)
167 		/* Userspace did not provide linfo */
168 		return;
169 
170 	linfo_idx = prog->aux->linfo_idx;
171 	linfo = &prog->aux->linfo[linfo_idx];
172 	insn_start = linfo[0].insn_off;
173 	insn_end = insn_start + prog->len;
174 
175 	jited_linfo = &prog->aux->jited_linfo[linfo_idx];
176 	jited_linfo[0] = prog->bpf_func;
177 
178 	nr_linfo = prog->aux->nr_linfo - linfo_idx;
179 
180 	for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
181 		/* The verifier ensures that linfo[i].insn_off is
182 		 * strictly increasing
183 		 */
184 		jited_linfo[i] = prog->bpf_func +
185 			insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
186 }
187 
188 void bpf_prog_free_linfo(struct bpf_prog *prog)
189 {
190 	bpf_prog_free_jited_linfo(prog);
191 	kvfree(prog->aux->linfo);
192 }
193 
194 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
195 				  gfp_t gfp_extra_flags)
196 {
197 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
198 	struct bpf_prog *fp;
199 	u32 pages, delta;
200 	int ret;
201 
202 	BUG_ON(fp_old == NULL);
203 
204 	size = round_up(size, PAGE_SIZE);
205 	pages = size / PAGE_SIZE;
206 	if (pages <= fp_old->pages)
207 		return fp_old;
208 
209 	delta = pages - fp_old->pages;
210 	ret = __bpf_prog_charge(fp_old->aux->user, delta);
211 	if (ret)
212 		return NULL;
213 
214 	fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
215 	if (fp == NULL) {
216 		__bpf_prog_uncharge(fp_old->aux->user, delta);
217 	} else {
218 		memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
219 		fp->pages = pages;
220 		fp->aux->prog = fp;
221 
222 		/* We keep fp->aux from fp_old around in the new
223 		 * reallocated structure.
224 		 */
225 		fp_old->aux = NULL;
226 		__bpf_prog_free(fp_old);
227 	}
228 
229 	return fp;
230 }
231 
232 void __bpf_prog_free(struct bpf_prog *fp)
233 {
234 	kfree(fp->aux);
235 	vfree(fp);
236 }
237 
238 int bpf_prog_calc_tag(struct bpf_prog *fp)
239 {
240 	const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64);
241 	u32 raw_size = bpf_prog_tag_scratch_size(fp);
242 	u32 digest[SHA_DIGEST_WORDS];
243 	u32 ws[SHA_WORKSPACE_WORDS];
244 	u32 i, bsize, psize, blocks;
245 	struct bpf_insn *dst;
246 	bool was_ld_map;
247 	u8 *raw, *todo;
248 	__be32 *result;
249 	__be64 *bits;
250 
251 	raw = vmalloc(raw_size);
252 	if (!raw)
253 		return -ENOMEM;
254 
255 	sha_init(digest);
256 	memset(ws, 0, sizeof(ws));
257 
258 	/* We need to take out the map fd for the digest calculation
259 	 * since they are unstable from user space side.
260 	 */
261 	dst = (void *)raw;
262 	for (i = 0, was_ld_map = false; i < fp->len; i++) {
263 		dst[i] = fp->insnsi[i];
264 		if (!was_ld_map &&
265 		    dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
266 		    dst[i].src_reg == BPF_PSEUDO_MAP_FD) {
267 			was_ld_map = true;
268 			dst[i].imm = 0;
269 		} else if (was_ld_map &&
270 			   dst[i].code == 0 &&
271 			   dst[i].dst_reg == 0 &&
272 			   dst[i].src_reg == 0 &&
273 			   dst[i].off == 0) {
274 			was_ld_map = false;
275 			dst[i].imm = 0;
276 		} else {
277 			was_ld_map = false;
278 		}
279 	}
280 
281 	psize = bpf_prog_insn_size(fp);
282 	memset(&raw[psize], 0, raw_size - psize);
283 	raw[psize++] = 0x80;
284 
285 	bsize  = round_up(psize, SHA_MESSAGE_BYTES);
286 	blocks = bsize / SHA_MESSAGE_BYTES;
287 	todo   = raw;
288 	if (bsize - psize >= sizeof(__be64)) {
289 		bits = (__be64 *)(todo + bsize - sizeof(__be64));
290 	} else {
291 		bits = (__be64 *)(todo + bsize + bits_offset);
292 		blocks++;
293 	}
294 	*bits = cpu_to_be64((psize - 1) << 3);
295 
296 	while (blocks--) {
297 		sha_transform(digest, todo, ws);
298 		todo += SHA_MESSAGE_BYTES;
299 	}
300 
301 	result = (__force __be32 *)digest;
302 	for (i = 0; i < SHA_DIGEST_WORDS; i++)
303 		result[i] = cpu_to_be32(digest[i]);
304 	memcpy(fp->tag, result, sizeof(fp->tag));
305 
306 	vfree(raw);
307 	return 0;
308 }
309 
310 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, u32 delta,
311 				u32 curr, const bool probe_pass)
312 {
313 	const s64 imm_min = S32_MIN, imm_max = S32_MAX;
314 	s64 imm = insn->imm;
315 
316 	if (curr < pos && curr + imm + 1 > pos)
317 		imm += delta;
318 	else if (curr > pos + delta && curr + imm + 1 <= pos + delta)
319 		imm -= delta;
320 	if (imm < imm_min || imm > imm_max)
321 		return -ERANGE;
322 	if (!probe_pass)
323 		insn->imm = imm;
324 	return 0;
325 }
326 
327 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, u32 delta,
328 				u32 curr, const bool probe_pass)
329 {
330 	const s32 off_min = S16_MIN, off_max = S16_MAX;
331 	s32 off = insn->off;
332 
333 	if (curr < pos && curr + off + 1 > pos)
334 		off += delta;
335 	else if (curr > pos + delta && curr + off + 1 <= pos + delta)
336 		off -= delta;
337 	if (off < off_min || off > off_max)
338 		return -ERANGE;
339 	if (!probe_pass)
340 		insn->off = off;
341 	return 0;
342 }
343 
344 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta,
345 			    const bool probe_pass)
346 {
347 	u32 i, insn_cnt = prog->len + (probe_pass ? delta : 0);
348 	struct bpf_insn *insn = prog->insnsi;
349 	int ret = 0;
350 
351 	for (i = 0; i < insn_cnt; i++, insn++) {
352 		u8 code;
353 
354 		/* In the probing pass we still operate on the original,
355 		 * unpatched image in order to check overflows before we
356 		 * do any other adjustments. Therefore skip the patchlet.
357 		 */
358 		if (probe_pass && i == pos) {
359 			i += delta + 1;
360 			insn++;
361 		}
362 		code = insn->code;
363 		if (BPF_CLASS(code) != BPF_JMP ||
364 		    BPF_OP(code) == BPF_EXIT)
365 			continue;
366 		/* Adjust offset of jmps if we cross patch boundaries. */
367 		if (BPF_OP(code) == BPF_CALL) {
368 			if (insn->src_reg != BPF_PSEUDO_CALL)
369 				continue;
370 			ret = bpf_adj_delta_to_imm(insn, pos, delta, i,
371 						   probe_pass);
372 		} else {
373 			ret = bpf_adj_delta_to_off(insn, pos, delta, i,
374 						   probe_pass);
375 		}
376 		if (ret)
377 			break;
378 	}
379 
380 	return ret;
381 }
382 
383 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
384 {
385 	struct bpf_line_info *linfo;
386 	u32 i, nr_linfo;
387 
388 	nr_linfo = prog->aux->nr_linfo;
389 	if (!nr_linfo || !delta)
390 		return;
391 
392 	linfo = prog->aux->linfo;
393 
394 	for (i = 0; i < nr_linfo; i++)
395 		if (off < linfo[i].insn_off)
396 			break;
397 
398 	/* Push all off < linfo[i].insn_off by delta */
399 	for (; i < nr_linfo; i++)
400 		linfo[i].insn_off += delta;
401 }
402 
403 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
404 				       const struct bpf_insn *patch, u32 len)
405 {
406 	u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
407 	const u32 cnt_max = S16_MAX;
408 	struct bpf_prog *prog_adj;
409 
410 	/* Since our patchlet doesn't expand the image, we're done. */
411 	if (insn_delta == 0) {
412 		memcpy(prog->insnsi + off, patch, sizeof(*patch));
413 		return prog;
414 	}
415 
416 	insn_adj_cnt = prog->len + insn_delta;
417 
418 	/* Reject anything that would potentially let the insn->off
419 	 * target overflow when we have excessive program expansions.
420 	 * We need to probe here before we do any reallocation where
421 	 * we afterwards may not fail anymore.
422 	 */
423 	if (insn_adj_cnt > cnt_max &&
424 	    bpf_adj_branches(prog, off, insn_delta, true))
425 		return NULL;
426 
427 	/* Several new instructions need to be inserted. Make room
428 	 * for them. Likely, there's no need for a new allocation as
429 	 * last page could have large enough tailroom.
430 	 */
431 	prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
432 				    GFP_USER);
433 	if (!prog_adj)
434 		return NULL;
435 
436 	prog_adj->len = insn_adj_cnt;
437 
438 	/* Patching happens in 3 steps:
439 	 *
440 	 * 1) Move over tail of insnsi from next instruction onwards,
441 	 *    so we can patch the single target insn with one or more
442 	 *    new ones (patching is always from 1 to n insns, n > 0).
443 	 * 2) Inject new instructions at the target location.
444 	 * 3) Adjust branch offsets if necessary.
445 	 */
446 	insn_rest = insn_adj_cnt - off - len;
447 
448 	memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
449 		sizeof(*patch) * insn_rest);
450 	memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
451 
452 	/* We are guaranteed to not fail at this point, otherwise
453 	 * the ship has sailed to reverse to the original state. An
454 	 * overflow cannot happen at this point.
455 	 */
456 	BUG_ON(bpf_adj_branches(prog_adj, off, insn_delta, false));
457 
458 	bpf_adj_linfo(prog_adj, off, insn_delta);
459 
460 	return prog_adj;
461 }
462 
463 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
464 {
465 	int i;
466 
467 	for (i = 0; i < fp->aux->func_cnt; i++)
468 		bpf_prog_kallsyms_del(fp->aux->func[i]);
469 }
470 
471 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
472 {
473 	bpf_prog_kallsyms_del_subprogs(fp);
474 	bpf_prog_kallsyms_del(fp);
475 }
476 
477 #ifdef CONFIG_BPF_JIT
478 /* All BPF JIT sysctl knobs here. */
479 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
480 int bpf_jit_harden   __read_mostly;
481 int bpf_jit_kallsyms __read_mostly;
482 long bpf_jit_limit   __read_mostly;
483 
484 static __always_inline void
485 bpf_get_prog_addr_region(const struct bpf_prog *prog,
486 			 unsigned long *symbol_start,
487 			 unsigned long *symbol_end)
488 {
489 	const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
490 	unsigned long addr = (unsigned long)hdr;
491 
492 	WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
493 
494 	*symbol_start = addr;
495 	*symbol_end   = addr + hdr->pages * PAGE_SIZE;
496 }
497 
498 static void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
499 {
500 	const char *end = sym + KSYM_NAME_LEN;
501 	const struct btf_type *type;
502 	const char *func_name;
503 
504 	BUILD_BUG_ON(sizeof("bpf_prog_") +
505 		     sizeof(prog->tag) * 2 +
506 		     /* name has been null terminated.
507 		      * We should need +1 for the '_' preceding
508 		      * the name.  However, the null character
509 		      * is double counted between the name and the
510 		      * sizeof("bpf_prog_") above, so we omit
511 		      * the +1 here.
512 		      */
513 		     sizeof(prog->aux->name) > KSYM_NAME_LEN);
514 
515 	sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
516 	sym  = bin2hex(sym, prog->tag, sizeof(prog->tag));
517 
518 	/* prog->aux->name will be ignored if full btf name is available */
519 	if (prog->aux->func_info_cnt) {
520 		type = btf_type_by_id(prog->aux->btf,
521 				      prog->aux->func_info[prog->aux->func_idx].type_id);
522 		func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
523 		snprintf(sym, (size_t)(end - sym), "_%s", func_name);
524 		return;
525 	}
526 
527 	if (prog->aux->name[0])
528 		snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
529 	else
530 		*sym = 0;
531 }
532 
533 static __always_inline unsigned long
534 bpf_get_prog_addr_start(struct latch_tree_node *n)
535 {
536 	unsigned long symbol_start, symbol_end;
537 	const struct bpf_prog_aux *aux;
538 
539 	aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
540 	bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
541 
542 	return symbol_start;
543 }
544 
545 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
546 					  struct latch_tree_node *b)
547 {
548 	return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b);
549 }
550 
551 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
552 {
553 	unsigned long val = (unsigned long)key;
554 	unsigned long symbol_start, symbol_end;
555 	const struct bpf_prog_aux *aux;
556 
557 	aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
558 	bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
559 
560 	if (val < symbol_start)
561 		return -1;
562 	if (val >= symbol_end)
563 		return  1;
564 
565 	return 0;
566 }
567 
568 static const struct latch_tree_ops bpf_tree_ops = {
569 	.less	= bpf_tree_less,
570 	.comp	= bpf_tree_comp,
571 };
572 
573 static DEFINE_SPINLOCK(bpf_lock);
574 static LIST_HEAD(bpf_kallsyms);
575 static struct latch_tree_root bpf_tree __cacheline_aligned;
576 
577 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
578 {
579 	WARN_ON_ONCE(!list_empty(&aux->ksym_lnode));
580 	list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms);
581 	latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
582 }
583 
584 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
585 {
586 	if (list_empty(&aux->ksym_lnode))
587 		return;
588 
589 	latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
590 	list_del_rcu(&aux->ksym_lnode);
591 }
592 
593 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
594 {
595 	return fp->jited && !bpf_prog_was_classic(fp);
596 }
597 
598 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
599 {
600 	return list_empty(&fp->aux->ksym_lnode) ||
601 	       fp->aux->ksym_lnode.prev == LIST_POISON2;
602 }
603 
604 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
605 {
606 	if (!bpf_prog_kallsyms_candidate(fp) ||
607 	    !capable(CAP_SYS_ADMIN))
608 		return;
609 
610 	spin_lock_bh(&bpf_lock);
611 	bpf_prog_ksym_node_add(fp->aux);
612 	spin_unlock_bh(&bpf_lock);
613 }
614 
615 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
616 {
617 	if (!bpf_prog_kallsyms_candidate(fp))
618 		return;
619 
620 	spin_lock_bh(&bpf_lock);
621 	bpf_prog_ksym_node_del(fp->aux);
622 	spin_unlock_bh(&bpf_lock);
623 }
624 
625 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
626 {
627 	struct latch_tree_node *n;
628 
629 	if (!bpf_jit_kallsyms_enabled())
630 		return NULL;
631 
632 	n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
633 	return n ?
634 	       container_of(n, struct bpf_prog_aux, ksym_tnode)->prog :
635 	       NULL;
636 }
637 
638 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
639 				 unsigned long *off, char *sym)
640 {
641 	unsigned long symbol_start, symbol_end;
642 	struct bpf_prog *prog;
643 	char *ret = NULL;
644 
645 	rcu_read_lock();
646 	prog = bpf_prog_kallsyms_find(addr);
647 	if (prog) {
648 		bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end);
649 		bpf_get_prog_name(prog, sym);
650 
651 		ret = sym;
652 		if (size)
653 			*size = symbol_end - symbol_start;
654 		if (off)
655 			*off  = addr - symbol_start;
656 	}
657 	rcu_read_unlock();
658 
659 	return ret;
660 }
661 
662 bool is_bpf_text_address(unsigned long addr)
663 {
664 	bool ret;
665 
666 	rcu_read_lock();
667 	ret = bpf_prog_kallsyms_find(addr) != NULL;
668 	rcu_read_unlock();
669 
670 	return ret;
671 }
672 
673 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
674 		    char *sym)
675 {
676 	struct bpf_prog_aux *aux;
677 	unsigned int it = 0;
678 	int ret = -ERANGE;
679 
680 	if (!bpf_jit_kallsyms_enabled())
681 		return ret;
682 
683 	rcu_read_lock();
684 	list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) {
685 		if (it++ != symnum)
686 			continue;
687 
688 		bpf_get_prog_name(aux->prog, sym);
689 
690 		*value = (unsigned long)aux->prog->bpf_func;
691 		*type  = BPF_SYM_ELF_TYPE;
692 
693 		ret = 0;
694 		break;
695 	}
696 	rcu_read_unlock();
697 
698 	return ret;
699 }
700 
701 static atomic_long_t bpf_jit_current;
702 
703 /* Can be overridden by an arch's JIT compiler if it has a custom,
704  * dedicated BPF backend memory area, or if neither of the two
705  * below apply.
706  */
707 u64 __weak bpf_jit_alloc_exec_limit(void)
708 {
709 #if defined(MODULES_VADDR)
710 	return MODULES_END - MODULES_VADDR;
711 #else
712 	return VMALLOC_END - VMALLOC_START;
713 #endif
714 }
715 
716 static int __init bpf_jit_charge_init(void)
717 {
718 	/* Only used as heuristic here to derive limit. */
719 	bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2,
720 					    PAGE_SIZE), LONG_MAX);
721 	return 0;
722 }
723 pure_initcall(bpf_jit_charge_init);
724 
725 static int bpf_jit_charge_modmem(u32 pages)
726 {
727 	if (atomic_long_add_return(pages, &bpf_jit_current) >
728 	    (bpf_jit_limit >> PAGE_SHIFT)) {
729 		if (!capable(CAP_SYS_ADMIN)) {
730 			atomic_long_sub(pages, &bpf_jit_current);
731 			return -EPERM;
732 		}
733 	}
734 
735 	return 0;
736 }
737 
738 static void bpf_jit_uncharge_modmem(u32 pages)
739 {
740 	atomic_long_sub(pages, &bpf_jit_current);
741 }
742 
743 void *__weak bpf_jit_alloc_exec(unsigned long size)
744 {
745 	return module_alloc(size);
746 }
747 
748 void __weak bpf_jit_free_exec(void *addr)
749 {
750 	module_memfree(addr);
751 }
752 
753 struct bpf_binary_header *
754 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
755 		     unsigned int alignment,
756 		     bpf_jit_fill_hole_t bpf_fill_ill_insns)
757 {
758 	struct bpf_binary_header *hdr;
759 	u32 size, hole, start, pages;
760 
761 	/* Most of BPF filters are really small, but if some of them
762 	 * fill a page, allow at least 128 extra bytes to insert a
763 	 * random section of illegal instructions.
764 	 */
765 	size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
766 	pages = size / PAGE_SIZE;
767 
768 	if (bpf_jit_charge_modmem(pages))
769 		return NULL;
770 	hdr = bpf_jit_alloc_exec(size);
771 	if (!hdr) {
772 		bpf_jit_uncharge_modmem(pages);
773 		return NULL;
774 	}
775 
776 	/* Fill space with illegal/arch-dep instructions. */
777 	bpf_fill_ill_insns(hdr, size);
778 
779 	hdr->pages = pages;
780 	hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
781 		     PAGE_SIZE - sizeof(*hdr));
782 	start = (get_random_int() % hole) & ~(alignment - 1);
783 
784 	/* Leave a random number of instructions before BPF code. */
785 	*image_ptr = &hdr->image[start];
786 
787 	return hdr;
788 }
789 
790 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
791 {
792 	u32 pages = hdr->pages;
793 
794 	bpf_jit_free_exec(hdr);
795 	bpf_jit_uncharge_modmem(pages);
796 }
797 
798 /* This symbol is only overridden by archs that have different
799  * requirements than the usual eBPF JITs, f.e. when they only
800  * implement cBPF JIT, do not set images read-only, etc.
801  */
802 void __weak bpf_jit_free(struct bpf_prog *fp)
803 {
804 	if (fp->jited) {
805 		struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
806 
807 		bpf_jit_binary_unlock_ro(hdr);
808 		bpf_jit_binary_free(hdr);
809 
810 		WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
811 	}
812 
813 	bpf_prog_unlock_free(fp);
814 }
815 
816 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
817 			  const struct bpf_insn *insn, bool extra_pass,
818 			  u64 *func_addr, bool *func_addr_fixed)
819 {
820 	s16 off = insn->off;
821 	s32 imm = insn->imm;
822 	u8 *addr;
823 
824 	*func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
825 	if (!*func_addr_fixed) {
826 		/* Place-holder address till the last pass has collected
827 		 * all addresses for JITed subprograms in which case we
828 		 * can pick them up from prog->aux.
829 		 */
830 		if (!extra_pass)
831 			addr = NULL;
832 		else if (prog->aux->func &&
833 			 off >= 0 && off < prog->aux->func_cnt)
834 			addr = (u8 *)prog->aux->func[off]->bpf_func;
835 		else
836 			return -EINVAL;
837 	} else {
838 		/* Address of a BPF helper call. Since part of the core
839 		 * kernel, it's always at a fixed location. __bpf_call_base
840 		 * and the helper with imm relative to it are both in core
841 		 * kernel.
842 		 */
843 		addr = (u8 *)__bpf_call_base + imm;
844 	}
845 
846 	*func_addr = (unsigned long)addr;
847 	return 0;
848 }
849 
850 static int bpf_jit_blind_insn(const struct bpf_insn *from,
851 			      const struct bpf_insn *aux,
852 			      struct bpf_insn *to_buff)
853 {
854 	struct bpf_insn *to = to_buff;
855 	u32 imm_rnd = get_random_int();
856 	s16 off;
857 
858 	BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
859 	BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
860 
861 	/* Constraints on AX register:
862 	 *
863 	 * AX register is inaccessible from user space. It is mapped in
864 	 * all JITs, and used here for constant blinding rewrites. It is
865 	 * typically "stateless" meaning its contents are only valid within
866 	 * the executed instruction, but not across several instructions.
867 	 * There are a few exceptions however which are further detailed
868 	 * below.
869 	 *
870 	 * Constant blinding is only used by JITs, not in the interpreter.
871 	 * The interpreter uses AX in some occasions as a local temporary
872 	 * register e.g. in DIV or MOD instructions.
873 	 *
874 	 * In restricted circumstances, the verifier can also use the AX
875 	 * register for rewrites as long as they do not interfere with
876 	 * the above cases!
877 	 */
878 	if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
879 		goto out;
880 
881 	if (from->imm == 0 &&
882 	    (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
883 	     from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
884 		*to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
885 		goto out;
886 	}
887 
888 	switch (from->code) {
889 	case BPF_ALU | BPF_ADD | BPF_K:
890 	case BPF_ALU | BPF_SUB | BPF_K:
891 	case BPF_ALU | BPF_AND | BPF_K:
892 	case BPF_ALU | BPF_OR  | BPF_K:
893 	case BPF_ALU | BPF_XOR | BPF_K:
894 	case BPF_ALU | BPF_MUL | BPF_K:
895 	case BPF_ALU | BPF_MOV | BPF_K:
896 	case BPF_ALU | BPF_DIV | BPF_K:
897 	case BPF_ALU | BPF_MOD | BPF_K:
898 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
899 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
900 		*to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
901 		break;
902 
903 	case BPF_ALU64 | BPF_ADD | BPF_K:
904 	case BPF_ALU64 | BPF_SUB | BPF_K:
905 	case BPF_ALU64 | BPF_AND | BPF_K:
906 	case BPF_ALU64 | BPF_OR  | BPF_K:
907 	case BPF_ALU64 | BPF_XOR | BPF_K:
908 	case BPF_ALU64 | BPF_MUL | BPF_K:
909 	case BPF_ALU64 | BPF_MOV | BPF_K:
910 	case BPF_ALU64 | BPF_DIV | BPF_K:
911 	case BPF_ALU64 | BPF_MOD | BPF_K:
912 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
913 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
914 		*to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
915 		break;
916 
917 	case BPF_JMP | BPF_JEQ  | BPF_K:
918 	case BPF_JMP | BPF_JNE  | BPF_K:
919 	case BPF_JMP | BPF_JGT  | BPF_K:
920 	case BPF_JMP | BPF_JLT  | BPF_K:
921 	case BPF_JMP | BPF_JGE  | BPF_K:
922 	case BPF_JMP | BPF_JLE  | BPF_K:
923 	case BPF_JMP | BPF_JSGT | BPF_K:
924 	case BPF_JMP | BPF_JSLT | BPF_K:
925 	case BPF_JMP | BPF_JSGE | BPF_K:
926 	case BPF_JMP | BPF_JSLE | BPF_K:
927 	case BPF_JMP | BPF_JSET | BPF_K:
928 		/* Accommodate for extra offset in case of a backjump. */
929 		off = from->off;
930 		if (off < 0)
931 			off -= 2;
932 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
933 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
934 		*to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
935 		break;
936 
937 	case BPF_LD | BPF_IMM | BPF_DW:
938 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
939 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
940 		*to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
941 		*to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
942 		break;
943 	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
944 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
945 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
946 		*to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
947 		break;
948 
949 	case BPF_ST | BPF_MEM | BPF_DW:
950 	case BPF_ST | BPF_MEM | BPF_W:
951 	case BPF_ST | BPF_MEM | BPF_H:
952 	case BPF_ST | BPF_MEM | BPF_B:
953 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
954 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
955 		*to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
956 		break;
957 	}
958 out:
959 	return to - to_buff;
960 }
961 
962 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
963 					      gfp_t gfp_extra_flags)
964 {
965 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
966 	struct bpf_prog *fp;
967 
968 	fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
969 	if (fp != NULL) {
970 		/* aux->prog still points to the fp_other one, so
971 		 * when promoting the clone to the real program,
972 		 * this still needs to be adapted.
973 		 */
974 		memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
975 	}
976 
977 	return fp;
978 }
979 
980 static void bpf_prog_clone_free(struct bpf_prog *fp)
981 {
982 	/* aux was stolen by the other clone, so we cannot free
983 	 * it from this path! It will be freed eventually by the
984 	 * other program on release.
985 	 *
986 	 * At this point, we don't need a deferred release since
987 	 * clone is guaranteed to not be locked.
988 	 */
989 	fp->aux = NULL;
990 	__bpf_prog_free(fp);
991 }
992 
993 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
994 {
995 	/* We have to repoint aux->prog to self, as we don't
996 	 * know whether fp here is the clone or the original.
997 	 */
998 	fp->aux->prog = fp;
999 	bpf_prog_clone_free(fp_other);
1000 }
1001 
1002 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1003 {
1004 	struct bpf_insn insn_buff[16], aux[2];
1005 	struct bpf_prog *clone, *tmp;
1006 	int insn_delta, insn_cnt;
1007 	struct bpf_insn *insn;
1008 	int i, rewritten;
1009 
1010 	if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1011 		return prog;
1012 
1013 	clone = bpf_prog_clone_create(prog, GFP_USER);
1014 	if (!clone)
1015 		return ERR_PTR(-ENOMEM);
1016 
1017 	insn_cnt = clone->len;
1018 	insn = clone->insnsi;
1019 
1020 	for (i = 0; i < insn_cnt; i++, insn++) {
1021 		/* We temporarily need to hold the original ld64 insn
1022 		 * so that we can still access the first part in the
1023 		 * second blinding run.
1024 		 */
1025 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1026 		    insn[1].code == 0)
1027 			memcpy(aux, insn, sizeof(aux));
1028 
1029 		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
1030 		if (!rewritten)
1031 			continue;
1032 
1033 		tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1034 		if (!tmp) {
1035 			/* Patching may have repointed aux->prog during
1036 			 * realloc from the original one, so we need to
1037 			 * fix it up here on error.
1038 			 */
1039 			bpf_jit_prog_release_other(prog, clone);
1040 			return ERR_PTR(-ENOMEM);
1041 		}
1042 
1043 		clone = tmp;
1044 		insn_delta = rewritten - 1;
1045 
1046 		/* Walk new program and skip insns we just inserted. */
1047 		insn = clone->insnsi + i + insn_delta;
1048 		insn_cnt += insn_delta;
1049 		i        += insn_delta;
1050 	}
1051 
1052 	clone->blinded = 1;
1053 	return clone;
1054 }
1055 #endif /* CONFIG_BPF_JIT */
1056 
1057 /* Base function for offset calculation. Needs to go into .text section,
1058  * therefore keeping it non-static as well; will also be used by JITs
1059  * anyway later on, so do not let the compiler omit it. This also needs
1060  * to go into kallsyms for correlation from e.g. bpftool, so naming
1061  * must not change.
1062  */
1063 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1064 {
1065 	return 0;
1066 }
1067 EXPORT_SYMBOL_GPL(__bpf_call_base);
1068 
1069 /* All UAPI available opcodes. */
1070 #define BPF_INSN_MAP(INSN_2, INSN_3)		\
1071 	/* 32 bit ALU operations. */		\
1072 	/*   Register based. */			\
1073 	INSN_3(ALU, ADD,  X),			\
1074 	INSN_3(ALU, SUB,  X),			\
1075 	INSN_3(ALU, AND,  X),			\
1076 	INSN_3(ALU, OR,   X),			\
1077 	INSN_3(ALU, LSH,  X),			\
1078 	INSN_3(ALU, RSH,  X),			\
1079 	INSN_3(ALU, XOR,  X),			\
1080 	INSN_3(ALU, MUL,  X),			\
1081 	INSN_3(ALU, MOV,  X),			\
1082 	INSN_3(ALU, ARSH, X),			\
1083 	INSN_3(ALU, DIV,  X),			\
1084 	INSN_3(ALU, MOD,  X),			\
1085 	INSN_2(ALU, NEG),			\
1086 	INSN_3(ALU, END, TO_BE),		\
1087 	INSN_3(ALU, END, TO_LE),		\
1088 	/*   Immediate based. */		\
1089 	INSN_3(ALU, ADD,  K),			\
1090 	INSN_3(ALU, SUB,  K),			\
1091 	INSN_3(ALU, AND,  K),			\
1092 	INSN_3(ALU, OR,   K),			\
1093 	INSN_3(ALU, LSH,  K),			\
1094 	INSN_3(ALU, RSH,  K),			\
1095 	INSN_3(ALU, XOR,  K),			\
1096 	INSN_3(ALU, MUL,  K),			\
1097 	INSN_3(ALU, MOV,  K),			\
1098 	INSN_3(ALU, ARSH, K),			\
1099 	INSN_3(ALU, DIV,  K),			\
1100 	INSN_3(ALU, MOD,  K),			\
1101 	/* 64 bit ALU operations. */		\
1102 	/*   Register based. */			\
1103 	INSN_3(ALU64, ADD,  X),			\
1104 	INSN_3(ALU64, SUB,  X),			\
1105 	INSN_3(ALU64, AND,  X),			\
1106 	INSN_3(ALU64, OR,   X),			\
1107 	INSN_3(ALU64, LSH,  X),			\
1108 	INSN_3(ALU64, RSH,  X),			\
1109 	INSN_3(ALU64, XOR,  X),			\
1110 	INSN_3(ALU64, MUL,  X),			\
1111 	INSN_3(ALU64, MOV,  X),			\
1112 	INSN_3(ALU64, ARSH, X),			\
1113 	INSN_3(ALU64, DIV,  X),			\
1114 	INSN_3(ALU64, MOD,  X),			\
1115 	INSN_2(ALU64, NEG),			\
1116 	/*   Immediate based. */		\
1117 	INSN_3(ALU64, ADD,  K),			\
1118 	INSN_3(ALU64, SUB,  K),			\
1119 	INSN_3(ALU64, AND,  K),			\
1120 	INSN_3(ALU64, OR,   K),			\
1121 	INSN_3(ALU64, LSH,  K),			\
1122 	INSN_3(ALU64, RSH,  K),			\
1123 	INSN_3(ALU64, XOR,  K),			\
1124 	INSN_3(ALU64, MUL,  K),			\
1125 	INSN_3(ALU64, MOV,  K),			\
1126 	INSN_3(ALU64, ARSH, K),			\
1127 	INSN_3(ALU64, DIV,  K),			\
1128 	INSN_3(ALU64, MOD,  K),			\
1129 	/* Call instruction. */			\
1130 	INSN_2(JMP, CALL),			\
1131 	/* Exit instruction. */			\
1132 	INSN_2(JMP, EXIT),			\
1133 	/* Jump instructions. */		\
1134 	/*   Register based. */			\
1135 	INSN_3(JMP, JEQ,  X),			\
1136 	INSN_3(JMP, JNE,  X),			\
1137 	INSN_3(JMP, JGT,  X),			\
1138 	INSN_3(JMP, JLT,  X),			\
1139 	INSN_3(JMP, JGE,  X),			\
1140 	INSN_3(JMP, JLE,  X),			\
1141 	INSN_3(JMP, JSGT, X),			\
1142 	INSN_3(JMP, JSLT, X),			\
1143 	INSN_3(JMP, JSGE, X),			\
1144 	INSN_3(JMP, JSLE, X),			\
1145 	INSN_3(JMP, JSET, X),			\
1146 	/*   Immediate based. */		\
1147 	INSN_3(JMP, JEQ,  K),			\
1148 	INSN_3(JMP, JNE,  K),			\
1149 	INSN_3(JMP, JGT,  K),			\
1150 	INSN_3(JMP, JLT,  K),			\
1151 	INSN_3(JMP, JGE,  K),			\
1152 	INSN_3(JMP, JLE,  K),			\
1153 	INSN_3(JMP, JSGT, K),			\
1154 	INSN_3(JMP, JSLT, K),			\
1155 	INSN_3(JMP, JSGE, K),			\
1156 	INSN_3(JMP, JSLE, K),			\
1157 	INSN_3(JMP, JSET, K),			\
1158 	INSN_2(JMP, JA),			\
1159 	/* Store instructions. */		\
1160 	/*   Register based. */			\
1161 	INSN_3(STX, MEM,  B),			\
1162 	INSN_3(STX, MEM,  H),			\
1163 	INSN_3(STX, MEM,  W),			\
1164 	INSN_3(STX, MEM,  DW),			\
1165 	INSN_3(STX, XADD, W),			\
1166 	INSN_3(STX, XADD, DW),			\
1167 	/*   Immediate based. */		\
1168 	INSN_3(ST, MEM, B),			\
1169 	INSN_3(ST, MEM, H),			\
1170 	INSN_3(ST, MEM, W),			\
1171 	INSN_3(ST, MEM, DW),			\
1172 	/* Load instructions. */		\
1173 	/*   Register based. */			\
1174 	INSN_3(LDX, MEM, B),			\
1175 	INSN_3(LDX, MEM, H),			\
1176 	INSN_3(LDX, MEM, W),			\
1177 	INSN_3(LDX, MEM, DW),			\
1178 	/*   Immediate based. */		\
1179 	INSN_3(LD, IMM, DW)
1180 
1181 bool bpf_opcode_in_insntable(u8 code)
1182 {
1183 #define BPF_INSN_2_TBL(x, y)    [BPF_##x | BPF_##y] = true
1184 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1185 	static const bool public_insntable[256] = {
1186 		[0 ... 255] = false,
1187 		/* Now overwrite non-defaults ... */
1188 		BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1189 		/* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1190 		[BPF_LD | BPF_ABS | BPF_B] = true,
1191 		[BPF_LD | BPF_ABS | BPF_H] = true,
1192 		[BPF_LD | BPF_ABS | BPF_W] = true,
1193 		[BPF_LD | BPF_IND | BPF_B] = true,
1194 		[BPF_LD | BPF_IND | BPF_H] = true,
1195 		[BPF_LD | BPF_IND | BPF_W] = true,
1196 	};
1197 #undef BPF_INSN_3_TBL
1198 #undef BPF_INSN_2_TBL
1199 	return public_insntable[code];
1200 }
1201 
1202 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1203 /**
1204  *	__bpf_prog_run - run eBPF program on a given context
1205  *	@ctx: is the data we are operating on
1206  *	@insn: is the array of eBPF instructions
1207  *
1208  * Decode and execute eBPF instructions.
1209  */
1210 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1211 {
1212 #define BPF_INSN_2_LBL(x, y)    [BPF_##x | BPF_##y] = &&x##_##y
1213 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1214 	static const void *jumptable[256] = {
1215 		[0 ... 255] = &&default_label,
1216 		/* Now overwrite non-defaults ... */
1217 		BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1218 		/* Non-UAPI available opcodes. */
1219 		[BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1220 		[BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1221 	};
1222 #undef BPF_INSN_3_LBL
1223 #undef BPF_INSN_2_LBL
1224 	u32 tail_call_cnt = 0;
1225 
1226 #define CONT	 ({ insn++; goto select_insn; })
1227 #define CONT_JMP ({ insn++; goto select_insn; })
1228 
1229 select_insn:
1230 	goto *jumptable[insn->code];
1231 
1232 	/* ALU */
1233 #define ALU(OPCODE, OP)			\
1234 	ALU64_##OPCODE##_X:		\
1235 		DST = DST OP SRC;	\
1236 		CONT;			\
1237 	ALU_##OPCODE##_X:		\
1238 		DST = (u32) DST OP (u32) SRC;	\
1239 		CONT;			\
1240 	ALU64_##OPCODE##_K:		\
1241 		DST = DST OP IMM;		\
1242 		CONT;			\
1243 	ALU_##OPCODE##_K:		\
1244 		DST = (u32) DST OP (u32) IMM;	\
1245 		CONT;
1246 
1247 	ALU(ADD,  +)
1248 	ALU(SUB,  -)
1249 	ALU(AND,  &)
1250 	ALU(OR,   |)
1251 	ALU(LSH, <<)
1252 	ALU(RSH, >>)
1253 	ALU(XOR,  ^)
1254 	ALU(MUL,  *)
1255 #undef ALU
1256 	ALU_NEG:
1257 		DST = (u32) -DST;
1258 		CONT;
1259 	ALU64_NEG:
1260 		DST = -DST;
1261 		CONT;
1262 	ALU_MOV_X:
1263 		DST = (u32) SRC;
1264 		CONT;
1265 	ALU_MOV_K:
1266 		DST = (u32) IMM;
1267 		CONT;
1268 	ALU64_MOV_X:
1269 		DST = SRC;
1270 		CONT;
1271 	ALU64_MOV_K:
1272 		DST = IMM;
1273 		CONT;
1274 	LD_IMM_DW:
1275 		DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1276 		insn++;
1277 		CONT;
1278 	ALU_ARSH_X:
1279 		DST = (u64) (u32) ((*(s32 *) &DST) >> SRC);
1280 		CONT;
1281 	ALU_ARSH_K:
1282 		DST = (u64) (u32) ((*(s32 *) &DST) >> IMM);
1283 		CONT;
1284 	ALU64_ARSH_X:
1285 		(*(s64 *) &DST) >>= SRC;
1286 		CONT;
1287 	ALU64_ARSH_K:
1288 		(*(s64 *) &DST) >>= IMM;
1289 		CONT;
1290 	ALU64_MOD_X:
1291 		div64_u64_rem(DST, SRC, &AX);
1292 		DST = AX;
1293 		CONT;
1294 	ALU_MOD_X:
1295 		AX = (u32) DST;
1296 		DST = do_div(AX, (u32) SRC);
1297 		CONT;
1298 	ALU64_MOD_K:
1299 		div64_u64_rem(DST, IMM, &AX);
1300 		DST = AX;
1301 		CONT;
1302 	ALU_MOD_K:
1303 		AX = (u32) DST;
1304 		DST = do_div(AX, (u32) IMM);
1305 		CONT;
1306 	ALU64_DIV_X:
1307 		DST = div64_u64(DST, SRC);
1308 		CONT;
1309 	ALU_DIV_X:
1310 		AX = (u32) DST;
1311 		do_div(AX, (u32) SRC);
1312 		DST = (u32) AX;
1313 		CONT;
1314 	ALU64_DIV_K:
1315 		DST = div64_u64(DST, IMM);
1316 		CONT;
1317 	ALU_DIV_K:
1318 		AX = (u32) DST;
1319 		do_div(AX, (u32) IMM);
1320 		DST = (u32) AX;
1321 		CONT;
1322 	ALU_END_TO_BE:
1323 		switch (IMM) {
1324 		case 16:
1325 			DST = (__force u16) cpu_to_be16(DST);
1326 			break;
1327 		case 32:
1328 			DST = (__force u32) cpu_to_be32(DST);
1329 			break;
1330 		case 64:
1331 			DST = (__force u64) cpu_to_be64(DST);
1332 			break;
1333 		}
1334 		CONT;
1335 	ALU_END_TO_LE:
1336 		switch (IMM) {
1337 		case 16:
1338 			DST = (__force u16) cpu_to_le16(DST);
1339 			break;
1340 		case 32:
1341 			DST = (__force u32) cpu_to_le32(DST);
1342 			break;
1343 		case 64:
1344 			DST = (__force u64) cpu_to_le64(DST);
1345 			break;
1346 		}
1347 		CONT;
1348 
1349 	/* CALL */
1350 	JMP_CALL:
1351 		/* Function call scratches BPF_R1-BPF_R5 registers,
1352 		 * preserves BPF_R6-BPF_R9, and stores return value
1353 		 * into BPF_R0.
1354 		 */
1355 		BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1356 						       BPF_R4, BPF_R5);
1357 		CONT;
1358 
1359 	JMP_CALL_ARGS:
1360 		BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1361 							    BPF_R3, BPF_R4,
1362 							    BPF_R5,
1363 							    insn + insn->off + 1);
1364 		CONT;
1365 
1366 	JMP_TAIL_CALL: {
1367 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1368 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1369 		struct bpf_prog *prog;
1370 		u32 index = BPF_R3;
1371 
1372 		if (unlikely(index >= array->map.max_entries))
1373 			goto out;
1374 		if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1375 			goto out;
1376 
1377 		tail_call_cnt++;
1378 
1379 		prog = READ_ONCE(array->ptrs[index]);
1380 		if (!prog)
1381 			goto out;
1382 
1383 		/* ARG1 at this point is guaranteed to point to CTX from
1384 		 * the verifier side due to the fact that the tail call is
1385 		 * handeled like a helper, that is, bpf_tail_call_proto,
1386 		 * where arg1_type is ARG_PTR_TO_CTX.
1387 		 */
1388 		insn = prog->insnsi;
1389 		goto select_insn;
1390 out:
1391 		CONT;
1392 	}
1393 	/* JMP */
1394 	JMP_JA:
1395 		insn += insn->off;
1396 		CONT;
1397 	JMP_JEQ_X:
1398 		if (DST == SRC) {
1399 			insn += insn->off;
1400 			CONT_JMP;
1401 		}
1402 		CONT;
1403 	JMP_JEQ_K:
1404 		if (DST == IMM) {
1405 			insn += insn->off;
1406 			CONT_JMP;
1407 		}
1408 		CONT;
1409 	JMP_JNE_X:
1410 		if (DST != SRC) {
1411 			insn += insn->off;
1412 			CONT_JMP;
1413 		}
1414 		CONT;
1415 	JMP_JNE_K:
1416 		if (DST != IMM) {
1417 			insn += insn->off;
1418 			CONT_JMP;
1419 		}
1420 		CONT;
1421 	JMP_JGT_X:
1422 		if (DST > SRC) {
1423 			insn += insn->off;
1424 			CONT_JMP;
1425 		}
1426 		CONT;
1427 	JMP_JGT_K:
1428 		if (DST > IMM) {
1429 			insn += insn->off;
1430 			CONT_JMP;
1431 		}
1432 		CONT;
1433 	JMP_JLT_X:
1434 		if (DST < SRC) {
1435 			insn += insn->off;
1436 			CONT_JMP;
1437 		}
1438 		CONT;
1439 	JMP_JLT_K:
1440 		if (DST < IMM) {
1441 			insn += insn->off;
1442 			CONT_JMP;
1443 		}
1444 		CONT;
1445 	JMP_JGE_X:
1446 		if (DST >= SRC) {
1447 			insn += insn->off;
1448 			CONT_JMP;
1449 		}
1450 		CONT;
1451 	JMP_JGE_K:
1452 		if (DST >= IMM) {
1453 			insn += insn->off;
1454 			CONT_JMP;
1455 		}
1456 		CONT;
1457 	JMP_JLE_X:
1458 		if (DST <= SRC) {
1459 			insn += insn->off;
1460 			CONT_JMP;
1461 		}
1462 		CONT;
1463 	JMP_JLE_K:
1464 		if (DST <= IMM) {
1465 			insn += insn->off;
1466 			CONT_JMP;
1467 		}
1468 		CONT;
1469 	JMP_JSGT_X:
1470 		if (((s64) DST) > ((s64) SRC)) {
1471 			insn += insn->off;
1472 			CONT_JMP;
1473 		}
1474 		CONT;
1475 	JMP_JSGT_K:
1476 		if (((s64) DST) > ((s64) IMM)) {
1477 			insn += insn->off;
1478 			CONT_JMP;
1479 		}
1480 		CONT;
1481 	JMP_JSLT_X:
1482 		if (((s64) DST) < ((s64) SRC)) {
1483 			insn += insn->off;
1484 			CONT_JMP;
1485 		}
1486 		CONT;
1487 	JMP_JSLT_K:
1488 		if (((s64) DST) < ((s64) IMM)) {
1489 			insn += insn->off;
1490 			CONT_JMP;
1491 		}
1492 		CONT;
1493 	JMP_JSGE_X:
1494 		if (((s64) DST) >= ((s64) SRC)) {
1495 			insn += insn->off;
1496 			CONT_JMP;
1497 		}
1498 		CONT;
1499 	JMP_JSGE_K:
1500 		if (((s64) DST) >= ((s64) IMM)) {
1501 			insn += insn->off;
1502 			CONT_JMP;
1503 		}
1504 		CONT;
1505 	JMP_JSLE_X:
1506 		if (((s64) DST) <= ((s64) SRC)) {
1507 			insn += insn->off;
1508 			CONT_JMP;
1509 		}
1510 		CONT;
1511 	JMP_JSLE_K:
1512 		if (((s64) DST) <= ((s64) IMM)) {
1513 			insn += insn->off;
1514 			CONT_JMP;
1515 		}
1516 		CONT;
1517 	JMP_JSET_X:
1518 		if (DST & SRC) {
1519 			insn += insn->off;
1520 			CONT_JMP;
1521 		}
1522 		CONT;
1523 	JMP_JSET_K:
1524 		if (DST & IMM) {
1525 			insn += insn->off;
1526 			CONT_JMP;
1527 		}
1528 		CONT;
1529 	JMP_EXIT:
1530 		return BPF_R0;
1531 
1532 	/* STX and ST and LDX*/
1533 #define LDST(SIZEOP, SIZE)						\
1534 	STX_MEM_##SIZEOP:						\
1535 		*(SIZE *)(unsigned long) (DST + insn->off) = SRC;	\
1536 		CONT;							\
1537 	ST_MEM_##SIZEOP:						\
1538 		*(SIZE *)(unsigned long) (DST + insn->off) = IMM;	\
1539 		CONT;							\
1540 	LDX_MEM_##SIZEOP:						\
1541 		DST = *(SIZE *)(unsigned long) (SRC + insn->off);	\
1542 		CONT;
1543 
1544 	LDST(B,   u8)
1545 	LDST(H,  u16)
1546 	LDST(W,  u32)
1547 	LDST(DW, u64)
1548 #undef LDST
1549 	STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1550 		atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1551 			   (DST + insn->off));
1552 		CONT;
1553 	STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1554 		atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1555 			     (DST + insn->off));
1556 		CONT;
1557 
1558 	default_label:
1559 		/* If we ever reach this, we have a bug somewhere. Die hard here
1560 		 * instead of just returning 0; we could be somewhere in a subprog,
1561 		 * so execution could continue otherwise which we do /not/ want.
1562 		 *
1563 		 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1564 		 */
1565 		pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1566 		BUG_ON(1);
1567 		return 0;
1568 }
1569 STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */
1570 
1571 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1572 #define DEFINE_BPF_PROG_RUN(stack_size) \
1573 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1574 { \
1575 	u64 stack[stack_size / sizeof(u64)]; \
1576 	u64 regs[MAX_BPF_EXT_REG]; \
1577 \
1578 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1579 	ARG1 = (u64) (unsigned long) ctx; \
1580 	return ___bpf_prog_run(regs, insn, stack); \
1581 }
1582 
1583 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1584 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1585 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1586 				      const struct bpf_insn *insn) \
1587 { \
1588 	u64 stack[stack_size / sizeof(u64)]; \
1589 	u64 regs[MAX_BPF_EXT_REG]; \
1590 \
1591 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1592 	BPF_R1 = r1; \
1593 	BPF_R2 = r2; \
1594 	BPF_R3 = r3; \
1595 	BPF_R4 = r4; \
1596 	BPF_R5 = r5; \
1597 	return ___bpf_prog_run(regs, insn, stack); \
1598 }
1599 
1600 #define EVAL1(FN, X) FN(X)
1601 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1602 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1603 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1604 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1605 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1606 
1607 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1608 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1609 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1610 
1611 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1612 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1613 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1614 
1615 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1616 
1617 static unsigned int (*interpreters[])(const void *ctx,
1618 				      const struct bpf_insn *insn) = {
1619 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1620 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1621 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1622 };
1623 #undef PROG_NAME_LIST
1624 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1625 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1626 				  const struct bpf_insn *insn) = {
1627 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1628 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1629 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1630 };
1631 #undef PROG_NAME_LIST
1632 
1633 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1634 {
1635 	stack_depth = max_t(u32, stack_depth, 1);
1636 	insn->off = (s16) insn->imm;
1637 	insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1638 		__bpf_call_base_args;
1639 	insn->code = BPF_JMP | BPF_CALL_ARGS;
1640 }
1641 
1642 #else
1643 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1644 					 const struct bpf_insn *insn)
1645 {
1646 	/* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1647 	 * is not working properly, so warn about it!
1648 	 */
1649 	WARN_ON_ONCE(1);
1650 	return 0;
1651 }
1652 #endif
1653 
1654 bool bpf_prog_array_compatible(struct bpf_array *array,
1655 			       const struct bpf_prog *fp)
1656 {
1657 	if (fp->kprobe_override)
1658 		return false;
1659 
1660 	if (!array->owner_prog_type) {
1661 		/* There's no owner yet where we could check for
1662 		 * compatibility.
1663 		 */
1664 		array->owner_prog_type = fp->type;
1665 		array->owner_jited = fp->jited;
1666 
1667 		return true;
1668 	}
1669 
1670 	return array->owner_prog_type == fp->type &&
1671 	       array->owner_jited == fp->jited;
1672 }
1673 
1674 static int bpf_check_tail_call(const struct bpf_prog *fp)
1675 {
1676 	struct bpf_prog_aux *aux = fp->aux;
1677 	int i;
1678 
1679 	for (i = 0; i < aux->used_map_cnt; i++) {
1680 		struct bpf_map *map = aux->used_maps[i];
1681 		struct bpf_array *array;
1682 
1683 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1684 			continue;
1685 
1686 		array = container_of(map, struct bpf_array, map);
1687 		if (!bpf_prog_array_compatible(array, fp))
1688 			return -EINVAL;
1689 	}
1690 
1691 	return 0;
1692 }
1693 
1694 static void bpf_prog_select_func(struct bpf_prog *fp)
1695 {
1696 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1697 	u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1698 
1699 	fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1700 #else
1701 	fp->bpf_func = __bpf_prog_ret0_warn;
1702 #endif
1703 }
1704 
1705 /**
1706  *	bpf_prog_select_runtime - select exec runtime for BPF program
1707  *	@fp: bpf_prog populated with internal BPF program
1708  *	@err: pointer to error variable
1709  *
1710  * Try to JIT eBPF program, if JIT is not available, use interpreter.
1711  * The BPF program will be executed via BPF_PROG_RUN() macro.
1712  */
1713 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1714 {
1715 	/* In case of BPF to BPF calls, verifier did all the prep
1716 	 * work with regards to JITing, etc.
1717 	 */
1718 	if (fp->bpf_func)
1719 		goto finalize;
1720 
1721 	bpf_prog_select_func(fp);
1722 
1723 	/* eBPF JITs can rewrite the program in case constant
1724 	 * blinding is active. However, in case of error during
1725 	 * blinding, bpf_int_jit_compile() must always return a
1726 	 * valid program, which in this case would simply not
1727 	 * be JITed, but falls back to the interpreter.
1728 	 */
1729 	if (!bpf_prog_is_dev_bound(fp->aux)) {
1730 		*err = bpf_prog_alloc_jited_linfo(fp);
1731 		if (*err)
1732 			return fp;
1733 
1734 		fp = bpf_int_jit_compile(fp);
1735 		if (!fp->jited) {
1736 			bpf_prog_free_jited_linfo(fp);
1737 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1738 			*err = -ENOTSUPP;
1739 			return fp;
1740 #endif
1741 		} else {
1742 			bpf_prog_free_unused_jited_linfo(fp);
1743 		}
1744 	} else {
1745 		*err = bpf_prog_offload_compile(fp);
1746 		if (*err)
1747 			return fp;
1748 	}
1749 
1750 finalize:
1751 	bpf_prog_lock_ro(fp);
1752 
1753 	/* The tail call compatibility check can only be done at
1754 	 * this late stage as we need to determine, if we deal
1755 	 * with JITed or non JITed program concatenations and not
1756 	 * all eBPF JITs might immediately support all features.
1757 	 */
1758 	*err = bpf_check_tail_call(fp);
1759 
1760 	return fp;
1761 }
1762 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1763 
1764 static unsigned int __bpf_prog_ret1(const void *ctx,
1765 				    const struct bpf_insn *insn)
1766 {
1767 	return 1;
1768 }
1769 
1770 static struct bpf_prog_dummy {
1771 	struct bpf_prog prog;
1772 } dummy_bpf_prog = {
1773 	.prog = {
1774 		.bpf_func = __bpf_prog_ret1,
1775 	},
1776 };
1777 
1778 /* to avoid allocating empty bpf_prog_array for cgroups that
1779  * don't have bpf program attached use one global 'empty_prog_array'
1780  * It will not be modified the caller of bpf_prog_array_alloc()
1781  * (since caller requested prog_cnt == 0)
1782  * that pointer should be 'freed' by bpf_prog_array_free()
1783  */
1784 static struct {
1785 	struct bpf_prog_array hdr;
1786 	struct bpf_prog *null_prog;
1787 } empty_prog_array = {
1788 	.null_prog = NULL,
1789 };
1790 
1791 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1792 {
1793 	if (prog_cnt)
1794 		return kzalloc(sizeof(struct bpf_prog_array) +
1795 			       sizeof(struct bpf_prog_array_item) *
1796 			       (prog_cnt + 1),
1797 			       flags);
1798 
1799 	return &empty_prog_array.hdr;
1800 }
1801 
1802 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs)
1803 {
1804 	if (!progs ||
1805 	    progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr)
1806 		return;
1807 	kfree_rcu(progs, rcu);
1808 }
1809 
1810 int bpf_prog_array_length(struct bpf_prog_array __rcu *array)
1811 {
1812 	struct bpf_prog_array_item *item;
1813 	u32 cnt = 0;
1814 
1815 	rcu_read_lock();
1816 	item = rcu_dereference(array)->items;
1817 	for (; item->prog; item++)
1818 		if (item->prog != &dummy_bpf_prog.prog)
1819 			cnt++;
1820 	rcu_read_unlock();
1821 	return cnt;
1822 }
1823 
1824 
1825 static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array,
1826 				     u32 *prog_ids,
1827 				     u32 request_cnt)
1828 {
1829 	struct bpf_prog_array_item *item;
1830 	int i = 0;
1831 
1832 	item = rcu_dereference_check(array, 1)->items;
1833 	for (; item->prog; item++) {
1834 		if (item->prog == &dummy_bpf_prog.prog)
1835 			continue;
1836 		prog_ids[i] = item->prog->aux->id;
1837 		if (++i == request_cnt) {
1838 			item++;
1839 			break;
1840 		}
1841 	}
1842 
1843 	return !!(item->prog);
1844 }
1845 
1846 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array,
1847 				__u32 __user *prog_ids, u32 cnt)
1848 {
1849 	unsigned long err = 0;
1850 	bool nospc;
1851 	u32 *ids;
1852 
1853 	/* users of this function are doing:
1854 	 * cnt = bpf_prog_array_length();
1855 	 * if (cnt > 0)
1856 	 *     bpf_prog_array_copy_to_user(..., cnt);
1857 	 * so below kcalloc doesn't need extra cnt > 0 check, but
1858 	 * bpf_prog_array_length() releases rcu lock and
1859 	 * prog array could have been swapped with empty or larger array,
1860 	 * so always copy 'cnt' prog_ids to the user.
1861 	 * In a rare race the user will see zero prog_ids
1862 	 */
1863 	ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1864 	if (!ids)
1865 		return -ENOMEM;
1866 	rcu_read_lock();
1867 	nospc = bpf_prog_array_copy_core(array, ids, cnt);
1868 	rcu_read_unlock();
1869 	err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1870 	kfree(ids);
1871 	if (err)
1872 		return -EFAULT;
1873 	if (nospc)
1874 		return -ENOSPC;
1875 	return 0;
1876 }
1877 
1878 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array,
1879 				struct bpf_prog *old_prog)
1880 {
1881 	struct bpf_prog_array_item *item = array->items;
1882 
1883 	for (; item->prog; item++)
1884 		if (item->prog == old_prog) {
1885 			WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1886 			break;
1887 		}
1888 }
1889 
1890 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array,
1891 			struct bpf_prog *exclude_prog,
1892 			struct bpf_prog *include_prog,
1893 			struct bpf_prog_array **new_array)
1894 {
1895 	int new_prog_cnt, carry_prog_cnt = 0;
1896 	struct bpf_prog_array_item *existing;
1897 	struct bpf_prog_array *array;
1898 	bool found_exclude = false;
1899 	int new_prog_idx = 0;
1900 
1901 	/* Figure out how many existing progs we need to carry over to
1902 	 * the new array.
1903 	 */
1904 	if (old_array) {
1905 		existing = old_array->items;
1906 		for (; existing->prog; existing++) {
1907 			if (existing->prog == exclude_prog) {
1908 				found_exclude = true;
1909 				continue;
1910 			}
1911 			if (existing->prog != &dummy_bpf_prog.prog)
1912 				carry_prog_cnt++;
1913 			if (existing->prog == include_prog)
1914 				return -EEXIST;
1915 		}
1916 	}
1917 
1918 	if (exclude_prog && !found_exclude)
1919 		return -ENOENT;
1920 
1921 	/* How many progs (not NULL) will be in the new array? */
1922 	new_prog_cnt = carry_prog_cnt;
1923 	if (include_prog)
1924 		new_prog_cnt += 1;
1925 
1926 	/* Do we have any prog (not NULL) in the new array? */
1927 	if (!new_prog_cnt) {
1928 		*new_array = NULL;
1929 		return 0;
1930 	}
1931 
1932 	/* +1 as the end of prog_array is marked with NULL */
1933 	array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
1934 	if (!array)
1935 		return -ENOMEM;
1936 
1937 	/* Fill in the new prog array */
1938 	if (carry_prog_cnt) {
1939 		existing = old_array->items;
1940 		for (; existing->prog; existing++)
1941 			if (existing->prog != exclude_prog &&
1942 			    existing->prog != &dummy_bpf_prog.prog) {
1943 				array->items[new_prog_idx++].prog =
1944 					existing->prog;
1945 			}
1946 	}
1947 	if (include_prog)
1948 		array->items[new_prog_idx++].prog = include_prog;
1949 	array->items[new_prog_idx].prog = NULL;
1950 	*new_array = array;
1951 	return 0;
1952 }
1953 
1954 int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array,
1955 			     u32 *prog_ids, u32 request_cnt,
1956 			     u32 *prog_cnt)
1957 {
1958 	u32 cnt = 0;
1959 
1960 	if (array)
1961 		cnt = bpf_prog_array_length(array);
1962 
1963 	*prog_cnt = cnt;
1964 
1965 	/* return early if user requested only program count or nothing to copy */
1966 	if (!request_cnt || !cnt)
1967 		return 0;
1968 
1969 	/* this function is called under trace/bpf_trace.c: bpf_event_mutex */
1970 	return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
1971 								     : 0;
1972 }
1973 
1974 static void bpf_prog_free_deferred(struct work_struct *work)
1975 {
1976 	struct bpf_prog_aux *aux;
1977 	int i;
1978 
1979 	aux = container_of(work, struct bpf_prog_aux, work);
1980 	if (bpf_prog_is_dev_bound(aux))
1981 		bpf_prog_offload_destroy(aux->prog);
1982 #ifdef CONFIG_PERF_EVENTS
1983 	if (aux->prog->has_callchain_buf)
1984 		put_callchain_buffers();
1985 #endif
1986 	for (i = 0; i < aux->func_cnt; i++)
1987 		bpf_jit_free(aux->func[i]);
1988 	if (aux->func_cnt) {
1989 		kfree(aux->func);
1990 		bpf_prog_unlock_free(aux->prog);
1991 	} else {
1992 		bpf_jit_free(aux->prog);
1993 	}
1994 }
1995 
1996 /* Free internal BPF program */
1997 void bpf_prog_free(struct bpf_prog *fp)
1998 {
1999 	struct bpf_prog_aux *aux = fp->aux;
2000 
2001 	INIT_WORK(&aux->work, bpf_prog_free_deferred);
2002 	schedule_work(&aux->work);
2003 }
2004 EXPORT_SYMBOL_GPL(bpf_prog_free);
2005 
2006 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2007 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2008 
2009 void bpf_user_rnd_init_once(void)
2010 {
2011 	prandom_init_once(&bpf_user_rnd_state);
2012 }
2013 
2014 BPF_CALL_0(bpf_user_rnd_u32)
2015 {
2016 	/* Should someone ever have the rather unwise idea to use some
2017 	 * of the registers passed into this function, then note that
2018 	 * this function is called from native eBPF and classic-to-eBPF
2019 	 * transformations. Register assignments from both sides are
2020 	 * different, f.e. classic always sets fn(ctx, A, X) here.
2021 	 */
2022 	struct rnd_state *state;
2023 	u32 res;
2024 
2025 	state = &get_cpu_var(bpf_user_rnd_state);
2026 	res = prandom_u32_state(state);
2027 	put_cpu_var(bpf_user_rnd_state);
2028 
2029 	return res;
2030 }
2031 
2032 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2033 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2034 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2035 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2036 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2037 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2038 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2039 
2040 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2041 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2042 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2043 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2044 
2045 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2046 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2047 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2048 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2049 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2050 
2051 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2052 {
2053 	return NULL;
2054 }
2055 
2056 u64 __weak
2057 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2058 		 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2059 {
2060 	return -ENOTSUPP;
2061 }
2062 EXPORT_SYMBOL_GPL(bpf_event_output);
2063 
2064 /* Always built-in helper functions. */
2065 const struct bpf_func_proto bpf_tail_call_proto = {
2066 	.func		= NULL,
2067 	.gpl_only	= false,
2068 	.ret_type	= RET_VOID,
2069 	.arg1_type	= ARG_PTR_TO_CTX,
2070 	.arg2_type	= ARG_CONST_MAP_PTR,
2071 	.arg3_type	= ARG_ANYTHING,
2072 };
2073 
2074 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2075  * It is encouraged to implement bpf_int_jit_compile() instead, so that
2076  * eBPF and implicitly also cBPF can get JITed!
2077  */
2078 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2079 {
2080 	return prog;
2081 }
2082 
2083 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2084  * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2085  */
2086 void __weak bpf_jit_compile(struct bpf_prog *prog)
2087 {
2088 }
2089 
2090 bool __weak bpf_helper_changes_pkt_data(void *func)
2091 {
2092 	return false;
2093 }
2094 
2095 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2096  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2097  */
2098 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2099 			 int len)
2100 {
2101 	return -EFAULT;
2102 }
2103 
2104 /* All definitions of tracepoints related to BPF. */
2105 #define CREATE_TRACE_POINTS
2106 #include <linux/bpf_trace.h>
2107 
2108 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2109