xref: /linux/kernel/bpf/core.c (revision 1f2367a39f17bd553a75e179a747f9b257bc9478)
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_no_stats(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 
108 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
109 {
110 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
111 	struct bpf_prog *prog;
112 	int cpu;
113 
114 	prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
115 	if (!prog)
116 		return NULL;
117 
118 	prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
119 	if (!prog->aux->stats) {
120 		kfree(prog->aux);
121 		vfree(prog);
122 		return NULL;
123 	}
124 
125 	for_each_possible_cpu(cpu) {
126 		struct bpf_prog_stats *pstats;
127 
128 		pstats = per_cpu_ptr(prog->aux->stats, cpu);
129 		u64_stats_init(&pstats->syncp);
130 	}
131 	return prog;
132 }
133 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
134 
135 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
136 {
137 	if (!prog->aux->nr_linfo || !prog->jit_requested)
138 		return 0;
139 
140 	prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo,
141 					 sizeof(*prog->aux->jited_linfo),
142 					 GFP_KERNEL | __GFP_NOWARN);
143 	if (!prog->aux->jited_linfo)
144 		return -ENOMEM;
145 
146 	return 0;
147 }
148 
149 void bpf_prog_free_jited_linfo(struct bpf_prog *prog)
150 {
151 	kfree(prog->aux->jited_linfo);
152 	prog->aux->jited_linfo = NULL;
153 }
154 
155 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog)
156 {
157 	if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0])
158 		bpf_prog_free_jited_linfo(prog);
159 }
160 
161 /* The jit engine is responsible to provide an array
162  * for insn_off to the jited_off mapping (insn_to_jit_off).
163  *
164  * The idx to this array is the insn_off.  Hence, the insn_off
165  * here is relative to the prog itself instead of the main prog.
166  * This array has one entry for each xlated bpf insn.
167  *
168  * jited_off is the byte off to the last byte of the jited insn.
169  *
170  * Hence, with
171  * insn_start:
172  *      The first bpf insn off of the prog.  The insn off
173  *      here is relative to the main prog.
174  *      e.g. if prog is a subprog, insn_start > 0
175  * linfo_idx:
176  *      The prog's idx to prog->aux->linfo and jited_linfo
177  *
178  * jited_linfo[linfo_idx] = prog->bpf_func
179  *
180  * For i > linfo_idx,
181  *
182  * jited_linfo[i] = prog->bpf_func +
183  *	insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
184  */
185 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
186 			       const u32 *insn_to_jit_off)
187 {
188 	u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
189 	const struct bpf_line_info *linfo;
190 	void **jited_linfo;
191 
192 	if (!prog->aux->jited_linfo)
193 		/* Userspace did not provide linfo */
194 		return;
195 
196 	linfo_idx = prog->aux->linfo_idx;
197 	linfo = &prog->aux->linfo[linfo_idx];
198 	insn_start = linfo[0].insn_off;
199 	insn_end = insn_start + prog->len;
200 
201 	jited_linfo = &prog->aux->jited_linfo[linfo_idx];
202 	jited_linfo[0] = prog->bpf_func;
203 
204 	nr_linfo = prog->aux->nr_linfo - linfo_idx;
205 
206 	for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
207 		/* The verifier ensures that linfo[i].insn_off is
208 		 * strictly increasing
209 		 */
210 		jited_linfo[i] = prog->bpf_func +
211 			insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
212 }
213 
214 void bpf_prog_free_linfo(struct bpf_prog *prog)
215 {
216 	bpf_prog_free_jited_linfo(prog);
217 	kvfree(prog->aux->linfo);
218 }
219 
220 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
221 				  gfp_t gfp_extra_flags)
222 {
223 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
224 	struct bpf_prog *fp;
225 	u32 pages, delta;
226 	int ret;
227 
228 	BUG_ON(fp_old == NULL);
229 
230 	size = round_up(size, PAGE_SIZE);
231 	pages = size / PAGE_SIZE;
232 	if (pages <= fp_old->pages)
233 		return fp_old;
234 
235 	delta = pages - fp_old->pages;
236 	ret = __bpf_prog_charge(fp_old->aux->user, delta);
237 	if (ret)
238 		return NULL;
239 
240 	fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
241 	if (fp == NULL) {
242 		__bpf_prog_uncharge(fp_old->aux->user, delta);
243 	} else {
244 		memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
245 		fp->pages = pages;
246 		fp->aux->prog = fp;
247 
248 		/* We keep fp->aux from fp_old around in the new
249 		 * reallocated structure.
250 		 */
251 		fp_old->aux = NULL;
252 		__bpf_prog_free(fp_old);
253 	}
254 
255 	return fp;
256 }
257 
258 void __bpf_prog_free(struct bpf_prog *fp)
259 {
260 	if (fp->aux) {
261 		free_percpu(fp->aux->stats);
262 		kfree(fp->aux);
263 	}
264 	vfree(fp);
265 }
266 
267 int bpf_prog_calc_tag(struct bpf_prog *fp)
268 {
269 	const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64);
270 	u32 raw_size = bpf_prog_tag_scratch_size(fp);
271 	u32 digest[SHA_DIGEST_WORDS];
272 	u32 ws[SHA_WORKSPACE_WORDS];
273 	u32 i, bsize, psize, blocks;
274 	struct bpf_insn *dst;
275 	bool was_ld_map;
276 	u8 *raw, *todo;
277 	__be32 *result;
278 	__be64 *bits;
279 
280 	raw = vmalloc(raw_size);
281 	if (!raw)
282 		return -ENOMEM;
283 
284 	sha_init(digest);
285 	memset(ws, 0, sizeof(ws));
286 
287 	/* We need to take out the map fd for the digest calculation
288 	 * since they are unstable from user space side.
289 	 */
290 	dst = (void *)raw;
291 	for (i = 0, was_ld_map = false; i < fp->len; i++) {
292 		dst[i] = fp->insnsi[i];
293 		if (!was_ld_map &&
294 		    dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
295 		    dst[i].src_reg == BPF_PSEUDO_MAP_FD) {
296 			was_ld_map = true;
297 			dst[i].imm = 0;
298 		} else if (was_ld_map &&
299 			   dst[i].code == 0 &&
300 			   dst[i].dst_reg == 0 &&
301 			   dst[i].src_reg == 0 &&
302 			   dst[i].off == 0) {
303 			was_ld_map = false;
304 			dst[i].imm = 0;
305 		} else {
306 			was_ld_map = false;
307 		}
308 	}
309 
310 	psize = bpf_prog_insn_size(fp);
311 	memset(&raw[psize], 0, raw_size - psize);
312 	raw[psize++] = 0x80;
313 
314 	bsize  = round_up(psize, SHA_MESSAGE_BYTES);
315 	blocks = bsize / SHA_MESSAGE_BYTES;
316 	todo   = raw;
317 	if (bsize - psize >= sizeof(__be64)) {
318 		bits = (__be64 *)(todo + bsize - sizeof(__be64));
319 	} else {
320 		bits = (__be64 *)(todo + bsize + bits_offset);
321 		blocks++;
322 	}
323 	*bits = cpu_to_be64((psize - 1) << 3);
324 
325 	while (blocks--) {
326 		sha_transform(digest, todo, ws);
327 		todo += SHA_MESSAGE_BYTES;
328 	}
329 
330 	result = (__force __be32 *)digest;
331 	for (i = 0; i < SHA_DIGEST_WORDS; i++)
332 		result[i] = cpu_to_be32(digest[i]);
333 	memcpy(fp->tag, result, sizeof(fp->tag));
334 
335 	vfree(raw);
336 	return 0;
337 }
338 
339 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
340 				s32 end_new, u32 curr, const bool probe_pass)
341 {
342 	const s64 imm_min = S32_MIN, imm_max = S32_MAX;
343 	s32 delta = end_new - end_old;
344 	s64 imm = insn->imm;
345 
346 	if (curr < pos && curr + imm + 1 >= end_old)
347 		imm += delta;
348 	else if (curr >= end_new && curr + imm + 1 < end_new)
349 		imm -= delta;
350 	if (imm < imm_min || imm > imm_max)
351 		return -ERANGE;
352 	if (!probe_pass)
353 		insn->imm = imm;
354 	return 0;
355 }
356 
357 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
358 				s32 end_new, u32 curr, const bool probe_pass)
359 {
360 	const s32 off_min = S16_MIN, off_max = S16_MAX;
361 	s32 delta = end_new - end_old;
362 	s32 off = insn->off;
363 
364 	if (curr < pos && curr + off + 1 >= end_old)
365 		off += delta;
366 	else if (curr >= end_new && curr + off + 1 < end_new)
367 		off -= delta;
368 	if (off < off_min || off > off_max)
369 		return -ERANGE;
370 	if (!probe_pass)
371 		insn->off = off;
372 	return 0;
373 }
374 
375 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
376 			    s32 end_new, const bool probe_pass)
377 {
378 	u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
379 	struct bpf_insn *insn = prog->insnsi;
380 	int ret = 0;
381 
382 	for (i = 0; i < insn_cnt; i++, insn++) {
383 		u8 code;
384 
385 		/* In the probing pass we still operate on the original,
386 		 * unpatched image in order to check overflows before we
387 		 * do any other adjustments. Therefore skip the patchlet.
388 		 */
389 		if (probe_pass && i == pos) {
390 			i = end_new;
391 			insn = prog->insnsi + end_old;
392 		}
393 		code = insn->code;
394 		if ((BPF_CLASS(code) != BPF_JMP &&
395 		     BPF_CLASS(code) != BPF_JMP32) ||
396 		    BPF_OP(code) == BPF_EXIT)
397 			continue;
398 		/* Adjust offset of jmps if we cross patch boundaries. */
399 		if (BPF_OP(code) == BPF_CALL) {
400 			if (insn->src_reg != BPF_PSEUDO_CALL)
401 				continue;
402 			ret = bpf_adj_delta_to_imm(insn, pos, end_old,
403 						   end_new, i, probe_pass);
404 		} else {
405 			ret = bpf_adj_delta_to_off(insn, pos, end_old,
406 						   end_new, i, probe_pass);
407 		}
408 		if (ret)
409 			break;
410 	}
411 
412 	return ret;
413 }
414 
415 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
416 {
417 	struct bpf_line_info *linfo;
418 	u32 i, nr_linfo;
419 
420 	nr_linfo = prog->aux->nr_linfo;
421 	if (!nr_linfo || !delta)
422 		return;
423 
424 	linfo = prog->aux->linfo;
425 
426 	for (i = 0; i < nr_linfo; i++)
427 		if (off < linfo[i].insn_off)
428 			break;
429 
430 	/* Push all off < linfo[i].insn_off by delta */
431 	for (; i < nr_linfo; i++)
432 		linfo[i].insn_off += delta;
433 }
434 
435 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
436 				       const struct bpf_insn *patch, u32 len)
437 {
438 	u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
439 	const u32 cnt_max = S16_MAX;
440 	struct bpf_prog *prog_adj;
441 
442 	/* Since our patchlet doesn't expand the image, we're done. */
443 	if (insn_delta == 0) {
444 		memcpy(prog->insnsi + off, patch, sizeof(*patch));
445 		return prog;
446 	}
447 
448 	insn_adj_cnt = prog->len + insn_delta;
449 
450 	/* Reject anything that would potentially let the insn->off
451 	 * target overflow when we have excessive program expansions.
452 	 * We need to probe here before we do any reallocation where
453 	 * we afterwards may not fail anymore.
454 	 */
455 	if (insn_adj_cnt > cnt_max &&
456 	    bpf_adj_branches(prog, off, off + 1, off + len, true))
457 		return NULL;
458 
459 	/* Several new instructions need to be inserted. Make room
460 	 * for them. Likely, there's no need for a new allocation as
461 	 * last page could have large enough tailroom.
462 	 */
463 	prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
464 				    GFP_USER);
465 	if (!prog_adj)
466 		return NULL;
467 
468 	prog_adj->len = insn_adj_cnt;
469 
470 	/* Patching happens in 3 steps:
471 	 *
472 	 * 1) Move over tail of insnsi from next instruction onwards,
473 	 *    so we can patch the single target insn with one or more
474 	 *    new ones (patching is always from 1 to n insns, n > 0).
475 	 * 2) Inject new instructions at the target location.
476 	 * 3) Adjust branch offsets if necessary.
477 	 */
478 	insn_rest = insn_adj_cnt - off - len;
479 
480 	memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
481 		sizeof(*patch) * insn_rest);
482 	memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
483 
484 	/* We are guaranteed to not fail at this point, otherwise
485 	 * the ship has sailed to reverse to the original state. An
486 	 * overflow cannot happen at this point.
487 	 */
488 	BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
489 
490 	bpf_adj_linfo(prog_adj, off, insn_delta);
491 
492 	return prog_adj;
493 }
494 
495 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
496 {
497 	/* Branch offsets can't overflow when program is shrinking, no need
498 	 * to call bpf_adj_branches(..., true) here
499 	 */
500 	memmove(prog->insnsi + off, prog->insnsi + off + cnt,
501 		sizeof(struct bpf_insn) * (prog->len - off - cnt));
502 	prog->len -= cnt;
503 
504 	return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
505 }
506 
507 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
508 {
509 	int i;
510 
511 	for (i = 0; i < fp->aux->func_cnt; i++)
512 		bpf_prog_kallsyms_del(fp->aux->func[i]);
513 }
514 
515 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
516 {
517 	bpf_prog_kallsyms_del_subprogs(fp);
518 	bpf_prog_kallsyms_del(fp);
519 }
520 
521 #ifdef CONFIG_BPF_JIT
522 /* All BPF JIT sysctl knobs here. */
523 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
524 int bpf_jit_harden   __read_mostly;
525 int bpf_jit_kallsyms __read_mostly;
526 long bpf_jit_limit   __read_mostly;
527 
528 static __always_inline void
529 bpf_get_prog_addr_region(const struct bpf_prog *prog,
530 			 unsigned long *symbol_start,
531 			 unsigned long *symbol_end)
532 {
533 	const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
534 	unsigned long addr = (unsigned long)hdr;
535 
536 	WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
537 
538 	*symbol_start = addr;
539 	*symbol_end   = addr + hdr->pages * PAGE_SIZE;
540 }
541 
542 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
543 {
544 	const char *end = sym + KSYM_NAME_LEN;
545 	const struct btf_type *type;
546 	const char *func_name;
547 
548 	BUILD_BUG_ON(sizeof("bpf_prog_") +
549 		     sizeof(prog->tag) * 2 +
550 		     /* name has been null terminated.
551 		      * We should need +1 for the '_' preceding
552 		      * the name.  However, the null character
553 		      * is double counted between the name and the
554 		      * sizeof("bpf_prog_") above, so we omit
555 		      * the +1 here.
556 		      */
557 		     sizeof(prog->aux->name) > KSYM_NAME_LEN);
558 
559 	sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
560 	sym  = bin2hex(sym, prog->tag, sizeof(prog->tag));
561 
562 	/* prog->aux->name will be ignored if full btf name is available */
563 	if (prog->aux->func_info_cnt) {
564 		type = btf_type_by_id(prog->aux->btf,
565 				      prog->aux->func_info[prog->aux->func_idx].type_id);
566 		func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
567 		snprintf(sym, (size_t)(end - sym), "_%s", func_name);
568 		return;
569 	}
570 
571 	if (prog->aux->name[0])
572 		snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
573 	else
574 		*sym = 0;
575 }
576 
577 static __always_inline unsigned long
578 bpf_get_prog_addr_start(struct latch_tree_node *n)
579 {
580 	unsigned long symbol_start, symbol_end;
581 	const struct bpf_prog_aux *aux;
582 
583 	aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
584 	bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
585 
586 	return symbol_start;
587 }
588 
589 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
590 					  struct latch_tree_node *b)
591 {
592 	return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b);
593 }
594 
595 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
596 {
597 	unsigned long val = (unsigned long)key;
598 	unsigned long symbol_start, symbol_end;
599 	const struct bpf_prog_aux *aux;
600 
601 	aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
602 	bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
603 
604 	if (val < symbol_start)
605 		return -1;
606 	if (val >= symbol_end)
607 		return  1;
608 
609 	return 0;
610 }
611 
612 static const struct latch_tree_ops bpf_tree_ops = {
613 	.less	= bpf_tree_less,
614 	.comp	= bpf_tree_comp,
615 };
616 
617 static DEFINE_SPINLOCK(bpf_lock);
618 static LIST_HEAD(bpf_kallsyms);
619 static struct latch_tree_root bpf_tree __cacheline_aligned;
620 
621 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
622 {
623 	WARN_ON_ONCE(!list_empty(&aux->ksym_lnode));
624 	list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms);
625 	latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
626 }
627 
628 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
629 {
630 	if (list_empty(&aux->ksym_lnode))
631 		return;
632 
633 	latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
634 	list_del_rcu(&aux->ksym_lnode);
635 }
636 
637 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
638 {
639 	return fp->jited && !bpf_prog_was_classic(fp);
640 }
641 
642 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
643 {
644 	return list_empty(&fp->aux->ksym_lnode) ||
645 	       fp->aux->ksym_lnode.prev == LIST_POISON2;
646 }
647 
648 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
649 {
650 	if (!bpf_prog_kallsyms_candidate(fp) ||
651 	    !capable(CAP_SYS_ADMIN))
652 		return;
653 
654 	spin_lock_bh(&bpf_lock);
655 	bpf_prog_ksym_node_add(fp->aux);
656 	spin_unlock_bh(&bpf_lock);
657 }
658 
659 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
660 {
661 	if (!bpf_prog_kallsyms_candidate(fp))
662 		return;
663 
664 	spin_lock_bh(&bpf_lock);
665 	bpf_prog_ksym_node_del(fp->aux);
666 	spin_unlock_bh(&bpf_lock);
667 }
668 
669 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
670 {
671 	struct latch_tree_node *n;
672 
673 	if (!bpf_jit_kallsyms_enabled())
674 		return NULL;
675 
676 	n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
677 	return n ?
678 	       container_of(n, struct bpf_prog_aux, ksym_tnode)->prog :
679 	       NULL;
680 }
681 
682 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
683 				 unsigned long *off, char *sym)
684 {
685 	unsigned long symbol_start, symbol_end;
686 	struct bpf_prog *prog;
687 	char *ret = NULL;
688 
689 	rcu_read_lock();
690 	prog = bpf_prog_kallsyms_find(addr);
691 	if (prog) {
692 		bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end);
693 		bpf_get_prog_name(prog, sym);
694 
695 		ret = sym;
696 		if (size)
697 			*size = symbol_end - symbol_start;
698 		if (off)
699 			*off  = addr - symbol_start;
700 	}
701 	rcu_read_unlock();
702 
703 	return ret;
704 }
705 
706 bool is_bpf_text_address(unsigned long addr)
707 {
708 	bool ret;
709 
710 	rcu_read_lock();
711 	ret = bpf_prog_kallsyms_find(addr) != NULL;
712 	rcu_read_unlock();
713 
714 	return ret;
715 }
716 
717 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
718 		    char *sym)
719 {
720 	struct bpf_prog_aux *aux;
721 	unsigned int it = 0;
722 	int ret = -ERANGE;
723 
724 	if (!bpf_jit_kallsyms_enabled())
725 		return ret;
726 
727 	rcu_read_lock();
728 	list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) {
729 		if (it++ != symnum)
730 			continue;
731 
732 		bpf_get_prog_name(aux->prog, sym);
733 
734 		*value = (unsigned long)aux->prog->bpf_func;
735 		*type  = BPF_SYM_ELF_TYPE;
736 
737 		ret = 0;
738 		break;
739 	}
740 	rcu_read_unlock();
741 
742 	return ret;
743 }
744 
745 static atomic_long_t bpf_jit_current;
746 
747 /* Can be overridden by an arch's JIT compiler if it has a custom,
748  * dedicated BPF backend memory area, or if neither of the two
749  * below apply.
750  */
751 u64 __weak bpf_jit_alloc_exec_limit(void)
752 {
753 #if defined(MODULES_VADDR)
754 	return MODULES_END - MODULES_VADDR;
755 #else
756 	return VMALLOC_END - VMALLOC_START;
757 #endif
758 }
759 
760 static int __init bpf_jit_charge_init(void)
761 {
762 	/* Only used as heuristic here to derive limit. */
763 	bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2,
764 					    PAGE_SIZE), LONG_MAX);
765 	return 0;
766 }
767 pure_initcall(bpf_jit_charge_init);
768 
769 static int bpf_jit_charge_modmem(u32 pages)
770 {
771 	if (atomic_long_add_return(pages, &bpf_jit_current) >
772 	    (bpf_jit_limit >> PAGE_SHIFT)) {
773 		if (!capable(CAP_SYS_ADMIN)) {
774 			atomic_long_sub(pages, &bpf_jit_current);
775 			return -EPERM;
776 		}
777 	}
778 
779 	return 0;
780 }
781 
782 static void bpf_jit_uncharge_modmem(u32 pages)
783 {
784 	atomic_long_sub(pages, &bpf_jit_current);
785 }
786 
787 void *__weak bpf_jit_alloc_exec(unsigned long size)
788 {
789 	return module_alloc(size);
790 }
791 
792 void __weak bpf_jit_free_exec(void *addr)
793 {
794 	module_memfree(addr);
795 }
796 
797 struct bpf_binary_header *
798 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
799 		     unsigned int alignment,
800 		     bpf_jit_fill_hole_t bpf_fill_ill_insns)
801 {
802 	struct bpf_binary_header *hdr;
803 	u32 size, hole, start, pages;
804 
805 	/* Most of BPF filters are really small, but if some of them
806 	 * fill a page, allow at least 128 extra bytes to insert a
807 	 * random section of illegal instructions.
808 	 */
809 	size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
810 	pages = size / PAGE_SIZE;
811 
812 	if (bpf_jit_charge_modmem(pages))
813 		return NULL;
814 	hdr = bpf_jit_alloc_exec(size);
815 	if (!hdr) {
816 		bpf_jit_uncharge_modmem(pages);
817 		return NULL;
818 	}
819 
820 	/* Fill space with illegal/arch-dep instructions. */
821 	bpf_fill_ill_insns(hdr, size);
822 
823 	hdr->pages = pages;
824 	hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
825 		     PAGE_SIZE - sizeof(*hdr));
826 	start = (get_random_int() % hole) & ~(alignment - 1);
827 
828 	/* Leave a random number of instructions before BPF code. */
829 	*image_ptr = &hdr->image[start];
830 
831 	return hdr;
832 }
833 
834 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
835 {
836 	u32 pages = hdr->pages;
837 
838 	bpf_jit_free_exec(hdr);
839 	bpf_jit_uncharge_modmem(pages);
840 }
841 
842 /* This symbol is only overridden by archs that have different
843  * requirements than the usual eBPF JITs, f.e. when they only
844  * implement cBPF JIT, do not set images read-only, etc.
845  */
846 void __weak bpf_jit_free(struct bpf_prog *fp)
847 {
848 	if (fp->jited) {
849 		struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
850 
851 		bpf_jit_binary_unlock_ro(hdr);
852 		bpf_jit_binary_free(hdr);
853 
854 		WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
855 	}
856 
857 	bpf_prog_unlock_free(fp);
858 }
859 
860 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
861 			  const struct bpf_insn *insn, bool extra_pass,
862 			  u64 *func_addr, bool *func_addr_fixed)
863 {
864 	s16 off = insn->off;
865 	s32 imm = insn->imm;
866 	u8 *addr;
867 
868 	*func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
869 	if (!*func_addr_fixed) {
870 		/* Place-holder address till the last pass has collected
871 		 * all addresses for JITed subprograms in which case we
872 		 * can pick them up from prog->aux.
873 		 */
874 		if (!extra_pass)
875 			addr = NULL;
876 		else if (prog->aux->func &&
877 			 off >= 0 && off < prog->aux->func_cnt)
878 			addr = (u8 *)prog->aux->func[off]->bpf_func;
879 		else
880 			return -EINVAL;
881 	} else {
882 		/* Address of a BPF helper call. Since part of the core
883 		 * kernel, it's always at a fixed location. __bpf_call_base
884 		 * and the helper with imm relative to it are both in core
885 		 * kernel.
886 		 */
887 		addr = (u8 *)__bpf_call_base + imm;
888 	}
889 
890 	*func_addr = (unsigned long)addr;
891 	return 0;
892 }
893 
894 static int bpf_jit_blind_insn(const struct bpf_insn *from,
895 			      const struct bpf_insn *aux,
896 			      struct bpf_insn *to_buff)
897 {
898 	struct bpf_insn *to = to_buff;
899 	u32 imm_rnd = get_random_int();
900 	s16 off;
901 
902 	BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
903 	BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
904 
905 	/* Constraints on AX register:
906 	 *
907 	 * AX register is inaccessible from user space. It is mapped in
908 	 * all JITs, and used here for constant blinding rewrites. It is
909 	 * typically "stateless" meaning its contents are only valid within
910 	 * the executed instruction, but not across several instructions.
911 	 * There are a few exceptions however which are further detailed
912 	 * below.
913 	 *
914 	 * Constant blinding is only used by JITs, not in the interpreter.
915 	 * The interpreter uses AX in some occasions as a local temporary
916 	 * register e.g. in DIV or MOD instructions.
917 	 *
918 	 * In restricted circumstances, the verifier can also use the AX
919 	 * register for rewrites as long as they do not interfere with
920 	 * the above cases!
921 	 */
922 	if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
923 		goto out;
924 
925 	if (from->imm == 0 &&
926 	    (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
927 	     from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
928 		*to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
929 		goto out;
930 	}
931 
932 	switch (from->code) {
933 	case BPF_ALU | BPF_ADD | BPF_K:
934 	case BPF_ALU | BPF_SUB | BPF_K:
935 	case BPF_ALU | BPF_AND | BPF_K:
936 	case BPF_ALU | BPF_OR  | BPF_K:
937 	case BPF_ALU | BPF_XOR | BPF_K:
938 	case BPF_ALU | BPF_MUL | BPF_K:
939 	case BPF_ALU | BPF_MOV | BPF_K:
940 	case BPF_ALU | BPF_DIV | BPF_K:
941 	case BPF_ALU | BPF_MOD | BPF_K:
942 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
943 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
944 		*to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
945 		break;
946 
947 	case BPF_ALU64 | BPF_ADD | BPF_K:
948 	case BPF_ALU64 | BPF_SUB | BPF_K:
949 	case BPF_ALU64 | BPF_AND | BPF_K:
950 	case BPF_ALU64 | BPF_OR  | BPF_K:
951 	case BPF_ALU64 | BPF_XOR | BPF_K:
952 	case BPF_ALU64 | BPF_MUL | BPF_K:
953 	case BPF_ALU64 | BPF_MOV | BPF_K:
954 	case BPF_ALU64 | BPF_DIV | BPF_K:
955 	case BPF_ALU64 | BPF_MOD | BPF_K:
956 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
957 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
958 		*to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
959 		break;
960 
961 	case BPF_JMP | BPF_JEQ  | BPF_K:
962 	case BPF_JMP | BPF_JNE  | BPF_K:
963 	case BPF_JMP | BPF_JGT  | BPF_K:
964 	case BPF_JMP | BPF_JLT  | BPF_K:
965 	case BPF_JMP | BPF_JGE  | BPF_K:
966 	case BPF_JMP | BPF_JLE  | BPF_K:
967 	case BPF_JMP | BPF_JSGT | BPF_K:
968 	case BPF_JMP | BPF_JSLT | BPF_K:
969 	case BPF_JMP | BPF_JSGE | BPF_K:
970 	case BPF_JMP | BPF_JSLE | BPF_K:
971 	case BPF_JMP | BPF_JSET | BPF_K:
972 		/* Accommodate for extra offset in case of a backjump. */
973 		off = from->off;
974 		if (off < 0)
975 			off -= 2;
976 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
977 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
978 		*to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
979 		break;
980 
981 	case BPF_JMP32 | BPF_JEQ  | BPF_K:
982 	case BPF_JMP32 | BPF_JNE  | BPF_K:
983 	case BPF_JMP32 | BPF_JGT  | BPF_K:
984 	case BPF_JMP32 | BPF_JLT  | BPF_K:
985 	case BPF_JMP32 | BPF_JGE  | BPF_K:
986 	case BPF_JMP32 | BPF_JLE  | BPF_K:
987 	case BPF_JMP32 | BPF_JSGT | BPF_K:
988 	case BPF_JMP32 | BPF_JSLT | BPF_K:
989 	case BPF_JMP32 | BPF_JSGE | BPF_K:
990 	case BPF_JMP32 | BPF_JSLE | BPF_K:
991 	case BPF_JMP32 | BPF_JSET | BPF_K:
992 		/* Accommodate for extra offset in case of a backjump. */
993 		off = from->off;
994 		if (off < 0)
995 			off -= 2;
996 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
997 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
998 		*to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
999 				      off);
1000 		break;
1001 
1002 	case BPF_LD | BPF_IMM | BPF_DW:
1003 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1004 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1005 		*to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1006 		*to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1007 		break;
1008 	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1009 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1010 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1011 		*to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
1012 		break;
1013 
1014 	case BPF_ST | BPF_MEM | BPF_DW:
1015 	case BPF_ST | BPF_MEM | BPF_W:
1016 	case BPF_ST | BPF_MEM | BPF_H:
1017 	case BPF_ST | BPF_MEM | BPF_B:
1018 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1019 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1020 		*to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1021 		break;
1022 	}
1023 out:
1024 	return to - to_buff;
1025 }
1026 
1027 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1028 					      gfp_t gfp_extra_flags)
1029 {
1030 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1031 	struct bpf_prog *fp;
1032 
1033 	fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
1034 	if (fp != NULL) {
1035 		/* aux->prog still points to the fp_other one, so
1036 		 * when promoting the clone to the real program,
1037 		 * this still needs to be adapted.
1038 		 */
1039 		memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1040 	}
1041 
1042 	return fp;
1043 }
1044 
1045 static void bpf_prog_clone_free(struct bpf_prog *fp)
1046 {
1047 	/* aux was stolen by the other clone, so we cannot free
1048 	 * it from this path! It will be freed eventually by the
1049 	 * other program on release.
1050 	 *
1051 	 * At this point, we don't need a deferred release since
1052 	 * clone is guaranteed to not be locked.
1053 	 */
1054 	fp->aux = NULL;
1055 	__bpf_prog_free(fp);
1056 }
1057 
1058 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1059 {
1060 	/* We have to repoint aux->prog to self, as we don't
1061 	 * know whether fp here is the clone or the original.
1062 	 */
1063 	fp->aux->prog = fp;
1064 	bpf_prog_clone_free(fp_other);
1065 }
1066 
1067 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1068 {
1069 	struct bpf_insn insn_buff[16], aux[2];
1070 	struct bpf_prog *clone, *tmp;
1071 	int insn_delta, insn_cnt;
1072 	struct bpf_insn *insn;
1073 	int i, rewritten;
1074 
1075 	if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1076 		return prog;
1077 
1078 	clone = bpf_prog_clone_create(prog, GFP_USER);
1079 	if (!clone)
1080 		return ERR_PTR(-ENOMEM);
1081 
1082 	insn_cnt = clone->len;
1083 	insn = clone->insnsi;
1084 
1085 	for (i = 0; i < insn_cnt; i++, insn++) {
1086 		/* We temporarily need to hold the original ld64 insn
1087 		 * so that we can still access the first part in the
1088 		 * second blinding run.
1089 		 */
1090 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1091 		    insn[1].code == 0)
1092 			memcpy(aux, insn, sizeof(aux));
1093 
1094 		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
1095 		if (!rewritten)
1096 			continue;
1097 
1098 		tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1099 		if (!tmp) {
1100 			/* Patching may have repointed aux->prog during
1101 			 * realloc from the original one, so we need to
1102 			 * fix it up here on error.
1103 			 */
1104 			bpf_jit_prog_release_other(prog, clone);
1105 			return ERR_PTR(-ENOMEM);
1106 		}
1107 
1108 		clone = tmp;
1109 		insn_delta = rewritten - 1;
1110 
1111 		/* Walk new program and skip insns we just inserted. */
1112 		insn = clone->insnsi + i + insn_delta;
1113 		insn_cnt += insn_delta;
1114 		i        += insn_delta;
1115 	}
1116 
1117 	clone->blinded = 1;
1118 	return clone;
1119 }
1120 #endif /* CONFIG_BPF_JIT */
1121 
1122 /* Base function for offset calculation. Needs to go into .text section,
1123  * therefore keeping it non-static as well; will also be used by JITs
1124  * anyway later on, so do not let the compiler omit it. This also needs
1125  * to go into kallsyms for correlation from e.g. bpftool, so naming
1126  * must not change.
1127  */
1128 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1129 {
1130 	return 0;
1131 }
1132 EXPORT_SYMBOL_GPL(__bpf_call_base);
1133 
1134 /* All UAPI available opcodes. */
1135 #define BPF_INSN_MAP(INSN_2, INSN_3)		\
1136 	/* 32 bit ALU operations. */		\
1137 	/*   Register based. */			\
1138 	INSN_3(ALU, ADD,  X),			\
1139 	INSN_3(ALU, SUB,  X),			\
1140 	INSN_3(ALU, AND,  X),			\
1141 	INSN_3(ALU, OR,   X),			\
1142 	INSN_3(ALU, LSH,  X),			\
1143 	INSN_3(ALU, RSH,  X),			\
1144 	INSN_3(ALU, XOR,  X),			\
1145 	INSN_3(ALU, MUL,  X),			\
1146 	INSN_3(ALU, MOV,  X),			\
1147 	INSN_3(ALU, ARSH, X),			\
1148 	INSN_3(ALU, DIV,  X),			\
1149 	INSN_3(ALU, MOD,  X),			\
1150 	INSN_2(ALU, NEG),			\
1151 	INSN_3(ALU, END, TO_BE),		\
1152 	INSN_3(ALU, END, TO_LE),		\
1153 	/*   Immediate based. */		\
1154 	INSN_3(ALU, ADD,  K),			\
1155 	INSN_3(ALU, SUB,  K),			\
1156 	INSN_3(ALU, AND,  K),			\
1157 	INSN_3(ALU, OR,   K),			\
1158 	INSN_3(ALU, LSH,  K),			\
1159 	INSN_3(ALU, RSH,  K),			\
1160 	INSN_3(ALU, XOR,  K),			\
1161 	INSN_3(ALU, MUL,  K),			\
1162 	INSN_3(ALU, MOV,  K),			\
1163 	INSN_3(ALU, ARSH, K),			\
1164 	INSN_3(ALU, DIV,  K),			\
1165 	INSN_3(ALU, MOD,  K),			\
1166 	/* 64 bit ALU operations. */		\
1167 	/*   Register based. */			\
1168 	INSN_3(ALU64, ADD,  X),			\
1169 	INSN_3(ALU64, SUB,  X),			\
1170 	INSN_3(ALU64, AND,  X),			\
1171 	INSN_3(ALU64, OR,   X),			\
1172 	INSN_3(ALU64, LSH,  X),			\
1173 	INSN_3(ALU64, RSH,  X),			\
1174 	INSN_3(ALU64, XOR,  X),			\
1175 	INSN_3(ALU64, MUL,  X),			\
1176 	INSN_3(ALU64, MOV,  X),			\
1177 	INSN_3(ALU64, ARSH, X),			\
1178 	INSN_3(ALU64, DIV,  X),			\
1179 	INSN_3(ALU64, MOD,  X),			\
1180 	INSN_2(ALU64, NEG),			\
1181 	/*   Immediate based. */		\
1182 	INSN_3(ALU64, ADD,  K),			\
1183 	INSN_3(ALU64, SUB,  K),			\
1184 	INSN_3(ALU64, AND,  K),			\
1185 	INSN_3(ALU64, OR,   K),			\
1186 	INSN_3(ALU64, LSH,  K),			\
1187 	INSN_3(ALU64, RSH,  K),			\
1188 	INSN_3(ALU64, XOR,  K),			\
1189 	INSN_3(ALU64, MUL,  K),			\
1190 	INSN_3(ALU64, MOV,  K),			\
1191 	INSN_3(ALU64, ARSH, K),			\
1192 	INSN_3(ALU64, DIV,  K),			\
1193 	INSN_3(ALU64, MOD,  K),			\
1194 	/* Call instruction. */			\
1195 	INSN_2(JMP, CALL),			\
1196 	/* Exit instruction. */			\
1197 	INSN_2(JMP, EXIT),			\
1198 	/* 32-bit Jump instructions. */		\
1199 	/*   Register based. */			\
1200 	INSN_3(JMP32, JEQ,  X),			\
1201 	INSN_3(JMP32, JNE,  X),			\
1202 	INSN_3(JMP32, JGT,  X),			\
1203 	INSN_3(JMP32, JLT,  X),			\
1204 	INSN_3(JMP32, JGE,  X),			\
1205 	INSN_3(JMP32, JLE,  X),			\
1206 	INSN_3(JMP32, JSGT, X),			\
1207 	INSN_3(JMP32, JSLT, X),			\
1208 	INSN_3(JMP32, JSGE, X),			\
1209 	INSN_3(JMP32, JSLE, X),			\
1210 	INSN_3(JMP32, JSET, X),			\
1211 	/*   Immediate based. */		\
1212 	INSN_3(JMP32, JEQ,  K),			\
1213 	INSN_3(JMP32, JNE,  K),			\
1214 	INSN_3(JMP32, JGT,  K),			\
1215 	INSN_3(JMP32, JLT,  K),			\
1216 	INSN_3(JMP32, JGE,  K),			\
1217 	INSN_3(JMP32, JLE,  K),			\
1218 	INSN_3(JMP32, JSGT, K),			\
1219 	INSN_3(JMP32, JSLT, K),			\
1220 	INSN_3(JMP32, JSGE, K),			\
1221 	INSN_3(JMP32, JSLE, K),			\
1222 	INSN_3(JMP32, JSET, K),			\
1223 	/* Jump instructions. */		\
1224 	/*   Register based. */			\
1225 	INSN_3(JMP, JEQ,  X),			\
1226 	INSN_3(JMP, JNE,  X),			\
1227 	INSN_3(JMP, JGT,  X),			\
1228 	INSN_3(JMP, JLT,  X),			\
1229 	INSN_3(JMP, JGE,  X),			\
1230 	INSN_3(JMP, JLE,  X),			\
1231 	INSN_3(JMP, JSGT, X),			\
1232 	INSN_3(JMP, JSLT, X),			\
1233 	INSN_3(JMP, JSGE, X),			\
1234 	INSN_3(JMP, JSLE, X),			\
1235 	INSN_3(JMP, JSET, X),			\
1236 	/*   Immediate based. */		\
1237 	INSN_3(JMP, JEQ,  K),			\
1238 	INSN_3(JMP, JNE,  K),			\
1239 	INSN_3(JMP, JGT,  K),			\
1240 	INSN_3(JMP, JLT,  K),			\
1241 	INSN_3(JMP, JGE,  K),			\
1242 	INSN_3(JMP, JLE,  K),			\
1243 	INSN_3(JMP, JSGT, K),			\
1244 	INSN_3(JMP, JSLT, K),			\
1245 	INSN_3(JMP, JSGE, K),			\
1246 	INSN_3(JMP, JSLE, K),			\
1247 	INSN_3(JMP, JSET, K),			\
1248 	INSN_2(JMP, JA),			\
1249 	/* Store instructions. */		\
1250 	/*   Register based. */			\
1251 	INSN_3(STX, MEM,  B),			\
1252 	INSN_3(STX, MEM,  H),			\
1253 	INSN_3(STX, MEM,  W),			\
1254 	INSN_3(STX, MEM,  DW),			\
1255 	INSN_3(STX, XADD, W),			\
1256 	INSN_3(STX, XADD, DW),			\
1257 	/*   Immediate based. */		\
1258 	INSN_3(ST, MEM, B),			\
1259 	INSN_3(ST, MEM, H),			\
1260 	INSN_3(ST, MEM, W),			\
1261 	INSN_3(ST, MEM, DW),			\
1262 	/* Load instructions. */		\
1263 	/*   Register based. */			\
1264 	INSN_3(LDX, MEM, B),			\
1265 	INSN_3(LDX, MEM, H),			\
1266 	INSN_3(LDX, MEM, W),			\
1267 	INSN_3(LDX, MEM, DW),			\
1268 	/*   Immediate based. */		\
1269 	INSN_3(LD, IMM, DW)
1270 
1271 bool bpf_opcode_in_insntable(u8 code)
1272 {
1273 #define BPF_INSN_2_TBL(x, y)    [BPF_##x | BPF_##y] = true
1274 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1275 	static const bool public_insntable[256] = {
1276 		[0 ... 255] = false,
1277 		/* Now overwrite non-defaults ... */
1278 		BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1279 		/* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1280 		[BPF_LD | BPF_ABS | BPF_B] = true,
1281 		[BPF_LD | BPF_ABS | BPF_H] = true,
1282 		[BPF_LD | BPF_ABS | BPF_W] = true,
1283 		[BPF_LD | BPF_IND | BPF_B] = true,
1284 		[BPF_LD | BPF_IND | BPF_H] = true,
1285 		[BPF_LD | BPF_IND | BPF_W] = true,
1286 	};
1287 #undef BPF_INSN_3_TBL
1288 #undef BPF_INSN_2_TBL
1289 	return public_insntable[code];
1290 }
1291 
1292 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1293 /**
1294  *	__bpf_prog_run - run eBPF program on a given context
1295  *	@regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1296  *	@insn: is the array of eBPF instructions
1297  *	@stack: is the eBPF storage stack
1298  *
1299  * Decode and execute eBPF instructions.
1300  */
1301 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1302 {
1303 #define BPF_INSN_2_LBL(x, y)    [BPF_##x | BPF_##y] = &&x##_##y
1304 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1305 	static const void *jumptable[256] = {
1306 		[0 ... 255] = &&default_label,
1307 		/* Now overwrite non-defaults ... */
1308 		BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1309 		/* Non-UAPI available opcodes. */
1310 		[BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1311 		[BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1312 	};
1313 #undef BPF_INSN_3_LBL
1314 #undef BPF_INSN_2_LBL
1315 	u32 tail_call_cnt = 0;
1316 
1317 #define CONT	 ({ insn++; goto select_insn; })
1318 #define CONT_JMP ({ insn++; goto select_insn; })
1319 
1320 select_insn:
1321 	goto *jumptable[insn->code];
1322 
1323 	/* ALU */
1324 #define ALU(OPCODE, OP)			\
1325 	ALU64_##OPCODE##_X:		\
1326 		DST = DST OP SRC;	\
1327 		CONT;			\
1328 	ALU_##OPCODE##_X:		\
1329 		DST = (u32) DST OP (u32) SRC;	\
1330 		CONT;			\
1331 	ALU64_##OPCODE##_K:		\
1332 		DST = DST OP IMM;		\
1333 		CONT;			\
1334 	ALU_##OPCODE##_K:		\
1335 		DST = (u32) DST OP (u32) IMM;	\
1336 		CONT;
1337 
1338 	ALU(ADD,  +)
1339 	ALU(SUB,  -)
1340 	ALU(AND,  &)
1341 	ALU(OR,   |)
1342 	ALU(LSH, <<)
1343 	ALU(RSH, >>)
1344 	ALU(XOR,  ^)
1345 	ALU(MUL,  *)
1346 #undef ALU
1347 	ALU_NEG:
1348 		DST = (u32) -DST;
1349 		CONT;
1350 	ALU64_NEG:
1351 		DST = -DST;
1352 		CONT;
1353 	ALU_MOV_X:
1354 		DST = (u32) SRC;
1355 		CONT;
1356 	ALU_MOV_K:
1357 		DST = (u32) IMM;
1358 		CONT;
1359 	ALU64_MOV_X:
1360 		DST = SRC;
1361 		CONT;
1362 	ALU64_MOV_K:
1363 		DST = IMM;
1364 		CONT;
1365 	LD_IMM_DW:
1366 		DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1367 		insn++;
1368 		CONT;
1369 	ALU_ARSH_X:
1370 		DST = (u64) (u32) ((*(s32 *) &DST) >> SRC);
1371 		CONT;
1372 	ALU_ARSH_K:
1373 		DST = (u64) (u32) ((*(s32 *) &DST) >> IMM);
1374 		CONT;
1375 	ALU64_ARSH_X:
1376 		(*(s64 *) &DST) >>= SRC;
1377 		CONT;
1378 	ALU64_ARSH_K:
1379 		(*(s64 *) &DST) >>= IMM;
1380 		CONT;
1381 	ALU64_MOD_X:
1382 		div64_u64_rem(DST, SRC, &AX);
1383 		DST = AX;
1384 		CONT;
1385 	ALU_MOD_X:
1386 		AX = (u32) DST;
1387 		DST = do_div(AX, (u32) SRC);
1388 		CONT;
1389 	ALU64_MOD_K:
1390 		div64_u64_rem(DST, IMM, &AX);
1391 		DST = AX;
1392 		CONT;
1393 	ALU_MOD_K:
1394 		AX = (u32) DST;
1395 		DST = do_div(AX, (u32) IMM);
1396 		CONT;
1397 	ALU64_DIV_X:
1398 		DST = div64_u64(DST, SRC);
1399 		CONT;
1400 	ALU_DIV_X:
1401 		AX = (u32) DST;
1402 		do_div(AX, (u32) SRC);
1403 		DST = (u32) AX;
1404 		CONT;
1405 	ALU64_DIV_K:
1406 		DST = div64_u64(DST, IMM);
1407 		CONT;
1408 	ALU_DIV_K:
1409 		AX = (u32) DST;
1410 		do_div(AX, (u32) IMM);
1411 		DST = (u32) AX;
1412 		CONT;
1413 	ALU_END_TO_BE:
1414 		switch (IMM) {
1415 		case 16:
1416 			DST = (__force u16) cpu_to_be16(DST);
1417 			break;
1418 		case 32:
1419 			DST = (__force u32) cpu_to_be32(DST);
1420 			break;
1421 		case 64:
1422 			DST = (__force u64) cpu_to_be64(DST);
1423 			break;
1424 		}
1425 		CONT;
1426 	ALU_END_TO_LE:
1427 		switch (IMM) {
1428 		case 16:
1429 			DST = (__force u16) cpu_to_le16(DST);
1430 			break;
1431 		case 32:
1432 			DST = (__force u32) cpu_to_le32(DST);
1433 			break;
1434 		case 64:
1435 			DST = (__force u64) cpu_to_le64(DST);
1436 			break;
1437 		}
1438 		CONT;
1439 
1440 	/* CALL */
1441 	JMP_CALL:
1442 		/* Function call scratches BPF_R1-BPF_R5 registers,
1443 		 * preserves BPF_R6-BPF_R9, and stores return value
1444 		 * into BPF_R0.
1445 		 */
1446 		BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1447 						       BPF_R4, BPF_R5);
1448 		CONT;
1449 
1450 	JMP_CALL_ARGS:
1451 		BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1452 							    BPF_R3, BPF_R4,
1453 							    BPF_R5,
1454 							    insn + insn->off + 1);
1455 		CONT;
1456 
1457 	JMP_TAIL_CALL: {
1458 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1459 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1460 		struct bpf_prog *prog;
1461 		u32 index = BPF_R3;
1462 
1463 		if (unlikely(index >= array->map.max_entries))
1464 			goto out;
1465 		if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1466 			goto out;
1467 
1468 		tail_call_cnt++;
1469 
1470 		prog = READ_ONCE(array->ptrs[index]);
1471 		if (!prog)
1472 			goto out;
1473 
1474 		/* ARG1 at this point is guaranteed to point to CTX from
1475 		 * the verifier side due to the fact that the tail call is
1476 		 * handeled like a helper, that is, bpf_tail_call_proto,
1477 		 * where arg1_type is ARG_PTR_TO_CTX.
1478 		 */
1479 		insn = prog->insnsi;
1480 		goto select_insn;
1481 out:
1482 		CONT;
1483 	}
1484 	JMP_JA:
1485 		insn += insn->off;
1486 		CONT;
1487 	JMP_EXIT:
1488 		return BPF_R0;
1489 	/* JMP */
1490 #define COND_JMP(SIGN, OPCODE, CMP_OP)				\
1491 	JMP_##OPCODE##_X:					\
1492 		if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) {	\
1493 			insn += insn->off;			\
1494 			CONT_JMP;				\
1495 		}						\
1496 		CONT;						\
1497 	JMP32_##OPCODE##_X:					\
1498 		if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) {	\
1499 			insn += insn->off;			\
1500 			CONT_JMP;				\
1501 		}						\
1502 		CONT;						\
1503 	JMP_##OPCODE##_K:					\
1504 		if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) {	\
1505 			insn += insn->off;			\
1506 			CONT_JMP;				\
1507 		}						\
1508 		CONT;						\
1509 	JMP32_##OPCODE##_K:					\
1510 		if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) {	\
1511 			insn += insn->off;			\
1512 			CONT_JMP;				\
1513 		}						\
1514 		CONT;
1515 	COND_JMP(u, JEQ, ==)
1516 	COND_JMP(u, JNE, !=)
1517 	COND_JMP(u, JGT, >)
1518 	COND_JMP(u, JLT, <)
1519 	COND_JMP(u, JGE, >=)
1520 	COND_JMP(u, JLE, <=)
1521 	COND_JMP(u, JSET, &)
1522 	COND_JMP(s, JSGT, >)
1523 	COND_JMP(s, JSLT, <)
1524 	COND_JMP(s, JSGE, >=)
1525 	COND_JMP(s, JSLE, <=)
1526 #undef COND_JMP
1527 	/* STX and ST and LDX*/
1528 #define LDST(SIZEOP, SIZE)						\
1529 	STX_MEM_##SIZEOP:						\
1530 		*(SIZE *)(unsigned long) (DST + insn->off) = SRC;	\
1531 		CONT;							\
1532 	ST_MEM_##SIZEOP:						\
1533 		*(SIZE *)(unsigned long) (DST + insn->off) = IMM;	\
1534 		CONT;							\
1535 	LDX_MEM_##SIZEOP:						\
1536 		DST = *(SIZE *)(unsigned long) (SRC + insn->off);	\
1537 		CONT;
1538 
1539 	LDST(B,   u8)
1540 	LDST(H,  u16)
1541 	LDST(W,  u32)
1542 	LDST(DW, u64)
1543 #undef LDST
1544 	STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1545 		atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1546 			   (DST + insn->off));
1547 		CONT;
1548 	STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1549 		atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1550 			     (DST + insn->off));
1551 		CONT;
1552 
1553 	default_label:
1554 		/* If we ever reach this, we have a bug somewhere. Die hard here
1555 		 * instead of just returning 0; we could be somewhere in a subprog,
1556 		 * so execution could continue otherwise which we do /not/ want.
1557 		 *
1558 		 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1559 		 */
1560 		pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1561 		BUG_ON(1);
1562 		return 0;
1563 }
1564 STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */
1565 
1566 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1567 #define DEFINE_BPF_PROG_RUN(stack_size) \
1568 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1569 { \
1570 	u64 stack[stack_size / sizeof(u64)]; \
1571 	u64 regs[MAX_BPF_EXT_REG]; \
1572 \
1573 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1574 	ARG1 = (u64) (unsigned long) ctx; \
1575 	return ___bpf_prog_run(regs, insn, stack); \
1576 }
1577 
1578 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1579 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1580 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1581 				      const struct bpf_insn *insn) \
1582 { \
1583 	u64 stack[stack_size / sizeof(u64)]; \
1584 	u64 regs[MAX_BPF_EXT_REG]; \
1585 \
1586 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1587 	BPF_R1 = r1; \
1588 	BPF_R2 = r2; \
1589 	BPF_R3 = r3; \
1590 	BPF_R4 = r4; \
1591 	BPF_R5 = r5; \
1592 	return ___bpf_prog_run(regs, insn, stack); \
1593 }
1594 
1595 #define EVAL1(FN, X) FN(X)
1596 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1597 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1598 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1599 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1600 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1601 
1602 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1603 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1604 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1605 
1606 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1607 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1608 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1609 
1610 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1611 
1612 static unsigned int (*interpreters[])(const void *ctx,
1613 				      const struct bpf_insn *insn) = {
1614 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1615 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1616 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1617 };
1618 #undef PROG_NAME_LIST
1619 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1620 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1621 				  const struct bpf_insn *insn) = {
1622 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1623 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1624 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1625 };
1626 #undef PROG_NAME_LIST
1627 
1628 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1629 {
1630 	stack_depth = max_t(u32, stack_depth, 1);
1631 	insn->off = (s16) insn->imm;
1632 	insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1633 		__bpf_call_base_args;
1634 	insn->code = BPF_JMP | BPF_CALL_ARGS;
1635 }
1636 
1637 #else
1638 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1639 					 const struct bpf_insn *insn)
1640 {
1641 	/* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1642 	 * is not working properly, so warn about it!
1643 	 */
1644 	WARN_ON_ONCE(1);
1645 	return 0;
1646 }
1647 #endif
1648 
1649 bool bpf_prog_array_compatible(struct bpf_array *array,
1650 			       const struct bpf_prog *fp)
1651 {
1652 	if (fp->kprobe_override)
1653 		return false;
1654 
1655 	if (!array->owner_prog_type) {
1656 		/* There's no owner yet where we could check for
1657 		 * compatibility.
1658 		 */
1659 		array->owner_prog_type = fp->type;
1660 		array->owner_jited = fp->jited;
1661 
1662 		return true;
1663 	}
1664 
1665 	return array->owner_prog_type == fp->type &&
1666 	       array->owner_jited == fp->jited;
1667 }
1668 
1669 static int bpf_check_tail_call(const struct bpf_prog *fp)
1670 {
1671 	struct bpf_prog_aux *aux = fp->aux;
1672 	int i;
1673 
1674 	for (i = 0; i < aux->used_map_cnt; i++) {
1675 		struct bpf_map *map = aux->used_maps[i];
1676 		struct bpf_array *array;
1677 
1678 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1679 			continue;
1680 
1681 		array = container_of(map, struct bpf_array, map);
1682 		if (!bpf_prog_array_compatible(array, fp))
1683 			return -EINVAL;
1684 	}
1685 
1686 	return 0;
1687 }
1688 
1689 static void bpf_prog_select_func(struct bpf_prog *fp)
1690 {
1691 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1692 	u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1693 
1694 	fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1695 #else
1696 	fp->bpf_func = __bpf_prog_ret0_warn;
1697 #endif
1698 }
1699 
1700 /**
1701  *	bpf_prog_select_runtime - select exec runtime for BPF program
1702  *	@fp: bpf_prog populated with internal BPF program
1703  *	@err: pointer to error variable
1704  *
1705  * Try to JIT eBPF program, if JIT is not available, use interpreter.
1706  * The BPF program will be executed via BPF_PROG_RUN() macro.
1707  */
1708 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1709 {
1710 	/* In case of BPF to BPF calls, verifier did all the prep
1711 	 * work with regards to JITing, etc.
1712 	 */
1713 	if (fp->bpf_func)
1714 		goto finalize;
1715 
1716 	bpf_prog_select_func(fp);
1717 
1718 	/* eBPF JITs can rewrite the program in case constant
1719 	 * blinding is active. However, in case of error during
1720 	 * blinding, bpf_int_jit_compile() must always return a
1721 	 * valid program, which in this case would simply not
1722 	 * be JITed, but falls back to the interpreter.
1723 	 */
1724 	if (!bpf_prog_is_dev_bound(fp->aux)) {
1725 		*err = bpf_prog_alloc_jited_linfo(fp);
1726 		if (*err)
1727 			return fp;
1728 
1729 		fp = bpf_int_jit_compile(fp);
1730 		if (!fp->jited) {
1731 			bpf_prog_free_jited_linfo(fp);
1732 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1733 			*err = -ENOTSUPP;
1734 			return fp;
1735 #endif
1736 		} else {
1737 			bpf_prog_free_unused_jited_linfo(fp);
1738 		}
1739 	} else {
1740 		*err = bpf_prog_offload_compile(fp);
1741 		if (*err)
1742 			return fp;
1743 	}
1744 
1745 finalize:
1746 	bpf_prog_lock_ro(fp);
1747 
1748 	/* The tail call compatibility check can only be done at
1749 	 * this late stage as we need to determine, if we deal
1750 	 * with JITed or non JITed program concatenations and not
1751 	 * all eBPF JITs might immediately support all features.
1752 	 */
1753 	*err = bpf_check_tail_call(fp);
1754 
1755 	return fp;
1756 }
1757 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1758 
1759 static unsigned int __bpf_prog_ret1(const void *ctx,
1760 				    const struct bpf_insn *insn)
1761 {
1762 	return 1;
1763 }
1764 
1765 static struct bpf_prog_dummy {
1766 	struct bpf_prog prog;
1767 } dummy_bpf_prog = {
1768 	.prog = {
1769 		.bpf_func = __bpf_prog_ret1,
1770 	},
1771 };
1772 
1773 /* to avoid allocating empty bpf_prog_array for cgroups that
1774  * don't have bpf program attached use one global 'empty_prog_array'
1775  * It will not be modified the caller of bpf_prog_array_alloc()
1776  * (since caller requested prog_cnt == 0)
1777  * that pointer should be 'freed' by bpf_prog_array_free()
1778  */
1779 static struct {
1780 	struct bpf_prog_array hdr;
1781 	struct bpf_prog *null_prog;
1782 } empty_prog_array = {
1783 	.null_prog = NULL,
1784 };
1785 
1786 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1787 {
1788 	if (prog_cnt)
1789 		return kzalloc(sizeof(struct bpf_prog_array) +
1790 			       sizeof(struct bpf_prog_array_item) *
1791 			       (prog_cnt + 1),
1792 			       flags);
1793 
1794 	return &empty_prog_array.hdr;
1795 }
1796 
1797 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs)
1798 {
1799 	if (!progs ||
1800 	    progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr)
1801 		return;
1802 	kfree_rcu(progs, rcu);
1803 }
1804 
1805 int bpf_prog_array_length(struct bpf_prog_array __rcu *array)
1806 {
1807 	struct bpf_prog_array_item *item;
1808 	u32 cnt = 0;
1809 
1810 	rcu_read_lock();
1811 	item = rcu_dereference(array)->items;
1812 	for (; item->prog; item++)
1813 		if (item->prog != &dummy_bpf_prog.prog)
1814 			cnt++;
1815 	rcu_read_unlock();
1816 	return cnt;
1817 }
1818 
1819 
1820 static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array,
1821 				     u32 *prog_ids,
1822 				     u32 request_cnt)
1823 {
1824 	struct bpf_prog_array_item *item;
1825 	int i = 0;
1826 
1827 	item = rcu_dereference_check(array, 1)->items;
1828 	for (; item->prog; item++) {
1829 		if (item->prog == &dummy_bpf_prog.prog)
1830 			continue;
1831 		prog_ids[i] = item->prog->aux->id;
1832 		if (++i == request_cnt) {
1833 			item++;
1834 			break;
1835 		}
1836 	}
1837 
1838 	return !!(item->prog);
1839 }
1840 
1841 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array,
1842 				__u32 __user *prog_ids, u32 cnt)
1843 {
1844 	unsigned long err = 0;
1845 	bool nospc;
1846 	u32 *ids;
1847 
1848 	/* users of this function are doing:
1849 	 * cnt = bpf_prog_array_length();
1850 	 * if (cnt > 0)
1851 	 *     bpf_prog_array_copy_to_user(..., cnt);
1852 	 * so below kcalloc doesn't need extra cnt > 0 check, but
1853 	 * bpf_prog_array_length() releases rcu lock and
1854 	 * prog array could have been swapped with empty or larger array,
1855 	 * so always copy 'cnt' prog_ids to the user.
1856 	 * In a rare race the user will see zero prog_ids
1857 	 */
1858 	ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1859 	if (!ids)
1860 		return -ENOMEM;
1861 	rcu_read_lock();
1862 	nospc = bpf_prog_array_copy_core(array, ids, cnt);
1863 	rcu_read_unlock();
1864 	err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1865 	kfree(ids);
1866 	if (err)
1867 		return -EFAULT;
1868 	if (nospc)
1869 		return -ENOSPC;
1870 	return 0;
1871 }
1872 
1873 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array,
1874 				struct bpf_prog *old_prog)
1875 {
1876 	struct bpf_prog_array_item *item = array->items;
1877 
1878 	for (; item->prog; item++)
1879 		if (item->prog == old_prog) {
1880 			WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1881 			break;
1882 		}
1883 }
1884 
1885 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array,
1886 			struct bpf_prog *exclude_prog,
1887 			struct bpf_prog *include_prog,
1888 			struct bpf_prog_array **new_array)
1889 {
1890 	int new_prog_cnt, carry_prog_cnt = 0;
1891 	struct bpf_prog_array_item *existing;
1892 	struct bpf_prog_array *array;
1893 	bool found_exclude = false;
1894 	int new_prog_idx = 0;
1895 
1896 	/* Figure out how many existing progs we need to carry over to
1897 	 * the new array.
1898 	 */
1899 	if (old_array) {
1900 		existing = old_array->items;
1901 		for (; existing->prog; existing++) {
1902 			if (existing->prog == exclude_prog) {
1903 				found_exclude = true;
1904 				continue;
1905 			}
1906 			if (existing->prog != &dummy_bpf_prog.prog)
1907 				carry_prog_cnt++;
1908 			if (existing->prog == include_prog)
1909 				return -EEXIST;
1910 		}
1911 	}
1912 
1913 	if (exclude_prog && !found_exclude)
1914 		return -ENOENT;
1915 
1916 	/* How many progs (not NULL) will be in the new array? */
1917 	new_prog_cnt = carry_prog_cnt;
1918 	if (include_prog)
1919 		new_prog_cnt += 1;
1920 
1921 	/* Do we have any prog (not NULL) in the new array? */
1922 	if (!new_prog_cnt) {
1923 		*new_array = NULL;
1924 		return 0;
1925 	}
1926 
1927 	/* +1 as the end of prog_array is marked with NULL */
1928 	array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
1929 	if (!array)
1930 		return -ENOMEM;
1931 
1932 	/* Fill in the new prog array */
1933 	if (carry_prog_cnt) {
1934 		existing = old_array->items;
1935 		for (; existing->prog; existing++)
1936 			if (existing->prog != exclude_prog &&
1937 			    existing->prog != &dummy_bpf_prog.prog) {
1938 				array->items[new_prog_idx++].prog =
1939 					existing->prog;
1940 			}
1941 	}
1942 	if (include_prog)
1943 		array->items[new_prog_idx++].prog = include_prog;
1944 	array->items[new_prog_idx].prog = NULL;
1945 	*new_array = array;
1946 	return 0;
1947 }
1948 
1949 int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array,
1950 			     u32 *prog_ids, u32 request_cnt,
1951 			     u32 *prog_cnt)
1952 {
1953 	u32 cnt = 0;
1954 
1955 	if (array)
1956 		cnt = bpf_prog_array_length(array);
1957 
1958 	*prog_cnt = cnt;
1959 
1960 	/* return early if user requested only program count or nothing to copy */
1961 	if (!request_cnt || !cnt)
1962 		return 0;
1963 
1964 	/* this function is called under trace/bpf_trace.c: bpf_event_mutex */
1965 	return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
1966 								     : 0;
1967 }
1968 
1969 static void bpf_prog_free_deferred(struct work_struct *work)
1970 {
1971 	struct bpf_prog_aux *aux;
1972 	int i;
1973 
1974 	aux = container_of(work, struct bpf_prog_aux, work);
1975 	if (bpf_prog_is_dev_bound(aux))
1976 		bpf_prog_offload_destroy(aux->prog);
1977 #ifdef CONFIG_PERF_EVENTS
1978 	if (aux->prog->has_callchain_buf)
1979 		put_callchain_buffers();
1980 #endif
1981 	for (i = 0; i < aux->func_cnt; i++)
1982 		bpf_jit_free(aux->func[i]);
1983 	if (aux->func_cnt) {
1984 		kfree(aux->func);
1985 		bpf_prog_unlock_free(aux->prog);
1986 	} else {
1987 		bpf_jit_free(aux->prog);
1988 	}
1989 }
1990 
1991 /* Free internal BPF program */
1992 void bpf_prog_free(struct bpf_prog *fp)
1993 {
1994 	struct bpf_prog_aux *aux = fp->aux;
1995 
1996 	INIT_WORK(&aux->work, bpf_prog_free_deferred);
1997 	schedule_work(&aux->work);
1998 }
1999 EXPORT_SYMBOL_GPL(bpf_prog_free);
2000 
2001 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2002 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2003 
2004 void bpf_user_rnd_init_once(void)
2005 {
2006 	prandom_init_once(&bpf_user_rnd_state);
2007 }
2008 
2009 BPF_CALL_0(bpf_user_rnd_u32)
2010 {
2011 	/* Should someone ever have the rather unwise idea to use some
2012 	 * of the registers passed into this function, then note that
2013 	 * this function is called from native eBPF and classic-to-eBPF
2014 	 * transformations. Register assignments from both sides are
2015 	 * different, f.e. classic always sets fn(ctx, A, X) here.
2016 	 */
2017 	struct rnd_state *state;
2018 	u32 res;
2019 
2020 	state = &get_cpu_var(bpf_user_rnd_state);
2021 	res = prandom_u32_state(state);
2022 	put_cpu_var(bpf_user_rnd_state);
2023 
2024 	return res;
2025 }
2026 
2027 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2028 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2029 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2030 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2031 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2032 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2033 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2034 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2035 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2036 
2037 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2038 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2039 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2040 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2041 
2042 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2043 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2044 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2045 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2046 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2047 
2048 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2049 {
2050 	return NULL;
2051 }
2052 
2053 u64 __weak
2054 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2055 		 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2056 {
2057 	return -ENOTSUPP;
2058 }
2059 EXPORT_SYMBOL_GPL(bpf_event_output);
2060 
2061 /* Always built-in helper functions. */
2062 const struct bpf_func_proto bpf_tail_call_proto = {
2063 	.func		= NULL,
2064 	.gpl_only	= false,
2065 	.ret_type	= RET_VOID,
2066 	.arg1_type	= ARG_PTR_TO_CTX,
2067 	.arg2_type	= ARG_CONST_MAP_PTR,
2068 	.arg3_type	= ARG_ANYTHING,
2069 };
2070 
2071 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2072  * It is encouraged to implement bpf_int_jit_compile() instead, so that
2073  * eBPF and implicitly also cBPF can get JITed!
2074  */
2075 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2076 {
2077 	return prog;
2078 }
2079 
2080 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2081  * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2082  */
2083 void __weak bpf_jit_compile(struct bpf_prog *prog)
2084 {
2085 }
2086 
2087 bool __weak bpf_helper_changes_pkt_data(void *func)
2088 {
2089 	return false;
2090 }
2091 
2092 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2093  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2094  */
2095 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2096 			 int len)
2097 {
2098 	return -EFAULT;
2099 }
2100 
2101 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2102 EXPORT_SYMBOL(bpf_stats_enabled_key);
2103 int sysctl_bpf_stats_enabled __read_mostly;
2104 
2105 /* All definitions of tracepoints related to BPF. */
2106 #define CREATE_TRACE_POINTS
2107 #include <linux/bpf_trace.h>
2108 
2109 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2110