xref: /openbmc/linux/kernel/bpf/core.c (revision 4419617e)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Linux Socket Filter - Kernel level socket filtering
4  *
5  * Based on the design of the Berkeley Packet Filter. The new
6  * internal format has been designed by PLUMgrid:
7  *
8  *	Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9  *
10  * Authors:
11  *
12  *	Jay Schulist <jschlst@samba.org>
13  *	Alexei Starovoitov <ast@plumgrid.com>
14  *	Daniel Borkmann <dborkman@redhat.com>
15  *
16  * Andi Kleen - Fix a few bad bugs and races.
17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18  */
19 
20 #include <uapi/linux/btf.h>
21 #include <linux/filter.h>
22 #include <linux/skbuff.h>
23 #include <linux/vmalloc.h>
24 #include <linux/random.h>
25 #include <linux/moduleloader.h>
26 #include <linux/bpf.h>
27 #include <linux/btf.h>
28 #include <linux/frame.h>
29 #include <linux/rbtree_latch.h>
30 #include <linux/kallsyms.h>
31 #include <linux/rcupdate.h>
32 #include <linux/perf_event.h>
33 
34 #include <asm/unaligned.h>
35 
36 /* Registers */
37 #define BPF_R0	regs[BPF_REG_0]
38 #define BPF_R1	regs[BPF_REG_1]
39 #define BPF_R2	regs[BPF_REG_2]
40 #define BPF_R3	regs[BPF_REG_3]
41 #define BPF_R4	regs[BPF_REG_4]
42 #define BPF_R5	regs[BPF_REG_5]
43 #define BPF_R6	regs[BPF_REG_6]
44 #define BPF_R7	regs[BPF_REG_7]
45 #define BPF_R8	regs[BPF_REG_8]
46 #define BPF_R9	regs[BPF_REG_9]
47 #define BPF_R10	regs[BPF_REG_10]
48 
49 /* Named registers */
50 #define DST	regs[insn->dst_reg]
51 #define SRC	regs[insn->src_reg]
52 #define FP	regs[BPF_REG_FP]
53 #define AX	regs[BPF_REG_AX]
54 #define ARG1	regs[BPF_REG_ARG1]
55 #define CTX	regs[BPF_REG_CTX]
56 #define IMM	insn->imm
57 
58 /* No hurry in this branch
59  *
60  * Exported for the bpf jit load helper.
61  */
62 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
63 {
64 	u8 *ptr = NULL;
65 
66 	if (k >= SKF_NET_OFF)
67 		ptr = skb_network_header(skb) + k - SKF_NET_OFF;
68 	else if (k >= SKF_LL_OFF)
69 		ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
70 
71 	if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
72 		return ptr;
73 
74 	return NULL;
75 }
76 
77 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
78 {
79 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
80 	struct bpf_prog_aux *aux;
81 	struct bpf_prog *fp;
82 
83 	size = round_up(size, PAGE_SIZE);
84 	fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
85 	if (fp == NULL)
86 		return NULL;
87 
88 	aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
89 	if (aux == NULL) {
90 		vfree(fp);
91 		return NULL;
92 	}
93 
94 	fp->pages = size / PAGE_SIZE;
95 	fp->aux = aux;
96 	fp->aux->prog = fp;
97 	fp->jit_requested = ebpf_jit_enabled();
98 
99 	INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode);
100 
101 	return fp;
102 }
103 
104 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
105 {
106 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
107 	struct bpf_prog *prog;
108 	int cpu;
109 
110 	prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
111 	if (!prog)
112 		return NULL;
113 
114 	prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
115 	if (!prog->aux->stats) {
116 		kfree(prog->aux);
117 		vfree(prog);
118 		return NULL;
119 	}
120 
121 	for_each_possible_cpu(cpu) {
122 		struct bpf_prog_stats *pstats;
123 
124 		pstats = per_cpu_ptr(prog->aux->stats, cpu);
125 		u64_stats_init(&pstats->syncp);
126 	}
127 	return prog;
128 }
129 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
130 
131 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
132 {
133 	if (!prog->aux->nr_linfo || !prog->jit_requested)
134 		return 0;
135 
136 	prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo,
137 					 sizeof(*prog->aux->jited_linfo),
138 					 GFP_KERNEL | __GFP_NOWARN);
139 	if (!prog->aux->jited_linfo)
140 		return -ENOMEM;
141 
142 	return 0;
143 }
144 
145 void bpf_prog_free_jited_linfo(struct bpf_prog *prog)
146 {
147 	kfree(prog->aux->jited_linfo);
148 	prog->aux->jited_linfo = NULL;
149 }
150 
151 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog)
152 {
153 	if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0])
154 		bpf_prog_free_jited_linfo(prog);
155 }
156 
157 /* The jit engine is responsible to provide an array
158  * for insn_off to the jited_off mapping (insn_to_jit_off).
159  *
160  * The idx to this array is the insn_off.  Hence, the insn_off
161  * here is relative to the prog itself instead of the main prog.
162  * This array has one entry for each xlated bpf insn.
163  *
164  * jited_off is the byte off to the last byte of the jited insn.
165  *
166  * Hence, with
167  * insn_start:
168  *      The first bpf insn off of the prog.  The insn off
169  *      here is relative to the main prog.
170  *      e.g. if prog is a subprog, insn_start > 0
171  * linfo_idx:
172  *      The prog's idx to prog->aux->linfo and jited_linfo
173  *
174  * jited_linfo[linfo_idx] = prog->bpf_func
175  *
176  * For i > linfo_idx,
177  *
178  * jited_linfo[i] = prog->bpf_func +
179  *	insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
180  */
181 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
182 			       const u32 *insn_to_jit_off)
183 {
184 	u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
185 	const struct bpf_line_info *linfo;
186 	void **jited_linfo;
187 
188 	if (!prog->aux->jited_linfo)
189 		/* Userspace did not provide linfo */
190 		return;
191 
192 	linfo_idx = prog->aux->linfo_idx;
193 	linfo = &prog->aux->linfo[linfo_idx];
194 	insn_start = linfo[0].insn_off;
195 	insn_end = insn_start + prog->len;
196 
197 	jited_linfo = &prog->aux->jited_linfo[linfo_idx];
198 	jited_linfo[0] = prog->bpf_func;
199 
200 	nr_linfo = prog->aux->nr_linfo - linfo_idx;
201 
202 	for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
203 		/* The verifier ensures that linfo[i].insn_off is
204 		 * strictly increasing
205 		 */
206 		jited_linfo[i] = prog->bpf_func +
207 			insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
208 }
209 
210 void bpf_prog_free_linfo(struct bpf_prog *prog)
211 {
212 	bpf_prog_free_jited_linfo(prog);
213 	kvfree(prog->aux->linfo);
214 }
215 
216 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
217 				  gfp_t gfp_extra_flags)
218 {
219 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
220 	struct bpf_prog *fp;
221 	u32 pages, delta;
222 	int ret;
223 
224 	BUG_ON(fp_old == NULL);
225 
226 	size = round_up(size, PAGE_SIZE);
227 	pages = size / PAGE_SIZE;
228 	if (pages <= fp_old->pages)
229 		return fp_old;
230 
231 	delta = pages - fp_old->pages;
232 	ret = __bpf_prog_charge(fp_old->aux->user, delta);
233 	if (ret)
234 		return NULL;
235 
236 	fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
237 	if (fp == NULL) {
238 		__bpf_prog_uncharge(fp_old->aux->user, delta);
239 	} else {
240 		memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
241 		fp->pages = pages;
242 		fp->aux->prog = fp;
243 
244 		/* We keep fp->aux from fp_old around in the new
245 		 * reallocated structure.
246 		 */
247 		fp_old->aux = NULL;
248 		__bpf_prog_free(fp_old);
249 	}
250 
251 	return fp;
252 }
253 
254 void __bpf_prog_free(struct bpf_prog *fp)
255 {
256 	if (fp->aux) {
257 		free_percpu(fp->aux->stats);
258 		kfree(fp->aux);
259 	}
260 	vfree(fp);
261 }
262 
263 int bpf_prog_calc_tag(struct bpf_prog *fp)
264 {
265 	const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64);
266 	u32 raw_size = bpf_prog_tag_scratch_size(fp);
267 	u32 digest[SHA_DIGEST_WORDS];
268 	u32 ws[SHA_WORKSPACE_WORDS];
269 	u32 i, bsize, psize, blocks;
270 	struct bpf_insn *dst;
271 	bool was_ld_map;
272 	u8 *raw, *todo;
273 	__be32 *result;
274 	__be64 *bits;
275 
276 	raw = vmalloc(raw_size);
277 	if (!raw)
278 		return -ENOMEM;
279 
280 	sha_init(digest);
281 	memset(ws, 0, sizeof(ws));
282 
283 	/* We need to take out the map fd for the digest calculation
284 	 * since they are unstable from user space side.
285 	 */
286 	dst = (void *)raw;
287 	for (i = 0, was_ld_map = false; i < fp->len; i++) {
288 		dst[i] = fp->insnsi[i];
289 		if (!was_ld_map &&
290 		    dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
291 		    (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
292 		     dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
293 			was_ld_map = true;
294 			dst[i].imm = 0;
295 		} else if (was_ld_map &&
296 			   dst[i].code == 0 &&
297 			   dst[i].dst_reg == 0 &&
298 			   dst[i].src_reg == 0 &&
299 			   dst[i].off == 0) {
300 			was_ld_map = false;
301 			dst[i].imm = 0;
302 		} else {
303 			was_ld_map = false;
304 		}
305 	}
306 
307 	psize = bpf_prog_insn_size(fp);
308 	memset(&raw[psize], 0, raw_size - psize);
309 	raw[psize++] = 0x80;
310 
311 	bsize  = round_up(psize, SHA_MESSAGE_BYTES);
312 	blocks = bsize / SHA_MESSAGE_BYTES;
313 	todo   = raw;
314 	if (bsize - psize >= sizeof(__be64)) {
315 		bits = (__be64 *)(todo + bsize - sizeof(__be64));
316 	} else {
317 		bits = (__be64 *)(todo + bsize + bits_offset);
318 		blocks++;
319 	}
320 	*bits = cpu_to_be64((psize - 1) << 3);
321 
322 	while (blocks--) {
323 		sha_transform(digest, todo, ws);
324 		todo += SHA_MESSAGE_BYTES;
325 	}
326 
327 	result = (__force __be32 *)digest;
328 	for (i = 0; i < SHA_DIGEST_WORDS; i++)
329 		result[i] = cpu_to_be32(digest[i]);
330 	memcpy(fp->tag, result, sizeof(fp->tag));
331 
332 	vfree(raw);
333 	return 0;
334 }
335 
336 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
337 				s32 end_new, s32 curr, const bool probe_pass)
338 {
339 	const s64 imm_min = S32_MIN, imm_max = S32_MAX;
340 	s32 delta = end_new - end_old;
341 	s64 imm = insn->imm;
342 
343 	if (curr < pos && curr + imm + 1 >= end_old)
344 		imm += delta;
345 	else if (curr >= end_new && curr + imm + 1 < end_new)
346 		imm -= delta;
347 	if (imm < imm_min || imm > imm_max)
348 		return -ERANGE;
349 	if (!probe_pass)
350 		insn->imm = imm;
351 	return 0;
352 }
353 
354 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
355 				s32 end_new, s32 curr, const bool probe_pass)
356 {
357 	const s32 off_min = S16_MIN, off_max = S16_MAX;
358 	s32 delta = end_new - end_old;
359 	s32 off = insn->off;
360 
361 	if (curr < pos && curr + off + 1 >= end_old)
362 		off += delta;
363 	else if (curr >= end_new && curr + off + 1 < end_new)
364 		off -= delta;
365 	if (off < off_min || off > off_max)
366 		return -ERANGE;
367 	if (!probe_pass)
368 		insn->off = off;
369 	return 0;
370 }
371 
372 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
373 			    s32 end_new, const bool probe_pass)
374 {
375 	u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
376 	struct bpf_insn *insn = prog->insnsi;
377 	int ret = 0;
378 
379 	for (i = 0; i < insn_cnt; i++, insn++) {
380 		u8 code;
381 
382 		/* In the probing pass we still operate on the original,
383 		 * unpatched image in order to check overflows before we
384 		 * do any other adjustments. Therefore skip the patchlet.
385 		 */
386 		if (probe_pass && i == pos) {
387 			i = end_new;
388 			insn = prog->insnsi + end_old;
389 		}
390 		code = insn->code;
391 		if ((BPF_CLASS(code) != BPF_JMP &&
392 		     BPF_CLASS(code) != BPF_JMP32) ||
393 		    BPF_OP(code) == BPF_EXIT)
394 			continue;
395 		/* Adjust offset of jmps if we cross patch boundaries. */
396 		if (BPF_OP(code) == BPF_CALL) {
397 			if (insn->src_reg != BPF_PSEUDO_CALL)
398 				continue;
399 			ret = bpf_adj_delta_to_imm(insn, pos, end_old,
400 						   end_new, i, probe_pass);
401 		} else {
402 			ret = bpf_adj_delta_to_off(insn, pos, end_old,
403 						   end_new, i, probe_pass);
404 		}
405 		if (ret)
406 			break;
407 	}
408 
409 	return ret;
410 }
411 
412 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
413 {
414 	struct bpf_line_info *linfo;
415 	u32 i, nr_linfo;
416 
417 	nr_linfo = prog->aux->nr_linfo;
418 	if (!nr_linfo || !delta)
419 		return;
420 
421 	linfo = prog->aux->linfo;
422 
423 	for (i = 0; i < nr_linfo; i++)
424 		if (off < linfo[i].insn_off)
425 			break;
426 
427 	/* Push all off < linfo[i].insn_off by delta */
428 	for (; i < nr_linfo; i++)
429 		linfo[i].insn_off += delta;
430 }
431 
432 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
433 				       const struct bpf_insn *patch, u32 len)
434 {
435 	u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
436 	const u32 cnt_max = S16_MAX;
437 	struct bpf_prog *prog_adj;
438 	int err;
439 
440 	/* Since our patchlet doesn't expand the image, we're done. */
441 	if (insn_delta == 0) {
442 		memcpy(prog->insnsi + off, patch, sizeof(*patch));
443 		return prog;
444 	}
445 
446 	insn_adj_cnt = prog->len + insn_delta;
447 
448 	/* Reject anything that would potentially let the insn->off
449 	 * target overflow when we have excessive program expansions.
450 	 * We need to probe here before we do any reallocation where
451 	 * we afterwards may not fail anymore.
452 	 */
453 	if (insn_adj_cnt > cnt_max &&
454 	    (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
455 		return ERR_PTR(err);
456 
457 	/* Several new instructions need to be inserted. Make room
458 	 * for them. Likely, there's no need for a new allocation as
459 	 * last page could have large enough tailroom.
460 	 */
461 	prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
462 				    GFP_USER);
463 	if (!prog_adj)
464 		return ERR_PTR(-ENOMEM);
465 
466 	prog_adj->len = insn_adj_cnt;
467 
468 	/* Patching happens in 3 steps:
469 	 *
470 	 * 1) Move over tail of insnsi from next instruction onwards,
471 	 *    so we can patch the single target insn with one or more
472 	 *    new ones (patching is always from 1 to n insns, n > 0).
473 	 * 2) Inject new instructions at the target location.
474 	 * 3) Adjust branch offsets if necessary.
475 	 */
476 	insn_rest = insn_adj_cnt - off - len;
477 
478 	memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
479 		sizeof(*patch) * insn_rest);
480 	memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
481 
482 	/* We are guaranteed to not fail at this point, otherwise
483 	 * the ship has sailed to reverse to the original state. An
484 	 * overflow cannot happen at this point.
485 	 */
486 	BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
487 
488 	bpf_adj_linfo(prog_adj, off, insn_delta);
489 
490 	return prog_adj;
491 }
492 
493 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
494 {
495 	/* Branch offsets can't overflow when program is shrinking, no need
496 	 * to call bpf_adj_branches(..., true) here
497 	 */
498 	memmove(prog->insnsi + off, prog->insnsi + off + cnt,
499 		sizeof(struct bpf_insn) * (prog->len - off - cnt));
500 	prog->len -= cnt;
501 
502 	return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
503 }
504 
505 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
506 {
507 	int i;
508 
509 	for (i = 0; i < fp->aux->func_cnt; i++)
510 		bpf_prog_kallsyms_del(fp->aux->func[i]);
511 }
512 
513 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
514 {
515 	bpf_prog_kallsyms_del_subprogs(fp);
516 	bpf_prog_kallsyms_del(fp);
517 }
518 
519 #ifdef CONFIG_BPF_JIT
520 /* All BPF JIT sysctl knobs here. */
521 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
522 int bpf_jit_harden   __read_mostly;
523 int bpf_jit_kallsyms __read_mostly;
524 long bpf_jit_limit   __read_mostly;
525 
526 static __always_inline void
527 bpf_get_prog_addr_region(const struct bpf_prog *prog,
528 			 unsigned long *symbol_start,
529 			 unsigned long *symbol_end)
530 {
531 	const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
532 	unsigned long addr = (unsigned long)hdr;
533 
534 	WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
535 
536 	*symbol_start = addr;
537 	*symbol_end   = addr + hdr->pages * PAGE_SIZE;
538 }
539 
540 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
541 {
542 	const char *end = sym + KSYM_NAME_LEN;
543 	const struct btf_type *type;
544 	const char *func_name;
545 
546 	BUILD_BUG_ON(sizeof("bpf_prog_") +
547 		     sizeof(prog->tag) * 2 +
548 		     /* name has been null terminated.
549 		      * We should need +1 for the '_' preceding
550 		      * the name.  However, the null character
551 		      * is double counted between the name and the
552 		      * sizeof("bpf_prog_") above, so we omit
553 		      * the +1 here.
554 		      */
555 		     sizeof(prog->aux->name) > KSYM_NAME_LEN);
556 
557 	sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
558 	sym  = bin2hex(sym, prog->tag, sizeof(prog->tag));
559 
560 	/* prog->aux->name will be ignored if full btf name is available */
561 	if (prog->aux->func_info_cnt) {
562 		type = btf_type_by_id(prog->aux->btf,
563 				      prog->aux->func_info[prog->aux->func_idx].type_id);
564 		func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
565 		snprintf(sym, (size_t)(end - sym), "_%s", func_name);
566 		return;
567 	}
568 
569 	if (prog->aux->name[0])
570 		snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
571 	else
572 		*sym = 0;
573 }
574 
575 static __always_inline unsigned long
576 bpf_get_prog_addr_start(struct latch_tree_node *n)
577 {
578 	unsigned long symbol_start, symbol_end;
579 	const struct bpf_prog_aux *aux;
580 
581 	aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
582 	bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
583 
584 	return symbol_start;
585 }
586 
587 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
588 					  struct latch_tree_node *b)
589 {
590 	return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b);
591 }
592 
593 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
594 {
595 	unsigned long val = (unsigned long)key;
596 	unsigned long symbol_start, symbol_end;
597 	const struct bpf_prog_aux *aux;
598 
599 	aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
600 	bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
601 
602 	if (val < symbol_start)
603 		return -1;
604 	if (val >= symbol_end)
605 		return  1;
606 
607 	return 0;
608 }
609 
610 static const struct latch_tree_ops bpf_tree_ops = {
611 	.less	= bpf_tree_less,
612 	.comp	= bpf_tree_comp,
613 };
614 
615 static DEFINE_SPINLOCK(bpf_lock);
616 static LIST_HEAD(bpf_kallsyms);
617 static struct latch_tree_root bpf_tree __cacheline_aligned;
618 
619 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
620 {
621 	WARN_ON_ONCE(!list_empty(&aux->ksym_lnode));
622 	list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms);
623 	latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
624 }
625 
626 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
627 {
628 	if (list_empty(&aux->ksym_lnode))
629 		return;
630 
631 	latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
632 	list_del_rcu(&aux->ksym_lnode);
633 }
634 
635 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
636 {
637 	return fp->jited && !bpf_prog_was_classic(fp);
638 }
639 
640 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
641 {
642 	return list_empty(&fp->aux->ksym_lnode) ||
643 	       fp->aux->ksym_lnode.prev == LIST_POISON2;
644 }
645 
646 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
647 {
648 	if (!bpf_prog_kallsyms_candidate(fp) ||
649 	    !capable(CAP_SYS_ADMIN))
650 		return;
651 
652 	spin_lock_bh(&bpf_lock);
653 	bpf_prog_ksym_node_add(fp->aux);
654 	spin_unlock_bh(&bpf_lock);
655 }
656 
657 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
658 {
659 	if (!bpf_prog_kallsyms_candidate(fp))
660 		return;
661 
662 	spin_lock_bh(&bpf_lock);
663 	bpf_prog_ksym_node_del(fp->aux);
664 	spin_unlock_bh(&bpf_lock);
665 }
666 
667 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
668 {
669 	struct latch_tree_node *n;
670 
671 	if (!bpf_jit_kallsyms_enabled())
672 		return NULL;
673 
674 	n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
675 	return n ?
676 	       container_of(n, struct bpf_prog_aux, ksym_tnode)->prog :
677 	       NULL;
678 }
679 
680 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
681 				 unsigned long *off, char *sym)
682 {
683 	unsigned long symbol_start, symbol_end;
684 	struct bpf_prog *prog;
685 	char *ret = NULL;
686 
687 	rcu_read_lock();
688 	prog = bpf_prog_kallsyms_find(addr);
689 	if (prog) {
690 		bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end);
691 		bpf_get_prog_name(prog, sym);
692 
693 		ret = sym;
694 		if (size)
695 			*size = symbol_end - symbol_start;
696 		if (off)
697 			*off  = addr - symbol_start;
698 	}
699 	rcu_read_unlock();
700 
701 	return ret;
702 }
703 
704 bool is_bpf_text_address(unsigned long addr)
705 {
706 	bool ret;
707 
708 	rcu_read_lock();
709 	ret = bpf_prog_kallsyms_find(addr) != NULL;
710 	rcu_read_unlock();
711 
712 	return ret;
713 }
714 
715 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
716 		    char *sym)
717 {
718 	struct bpf_prog_aux *aux;
719 	unsigned int it = 0;
720 	int ret = -ERANGE;
721 
722 	if (!bpf_jit_kallsyms_enabled())
723 		return ret;
724 
725 	rcu_read_lock();
726 	list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) {
727 		if (it++ != symnum)
728 			continue;
729 
730 		bpf_get_prog_name(aux->prog, sym);
731 
732 		*value = (unsigned long)aux->prog->bpf_func;
733 		*type  = BPF_SYM_ELF_TYPE;
734 
735 		ret = 0;
736 		break;
737 	}
738 	rcu_read_unlock();
739 
740 	return ret;
741 }
742 
743 static atomic_long_t bpf_jit_current;
744 
745 /* Can be overridden by an arch's JIT compiler if it has a custom,
746  * dedicated BPF backend memory area, or if neither of the two
747  * below apply.
748  */
749 u64 __weak bpf_jit_alloc_exec_limit(void)
750 {
751 #if defined(MODULES_VADDR)
752 	return MODULES_END - MODULES_VADDR;
753 #else
754 	return VMALLOC_END - VMALLOC_START;
755 #endif
756 }
757 
758 static int __init bpf_jit_charge_init(void)
759 {
760 	/* Only used as heuristic here to derive limit. */
761 	bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2,
762 					    PAGE_SIZE), LONG_MAX);
763 	return 0;
764 }
765 pure_initcall(bpf_jit_charge_init);
766 
767 static int bpf_jit_charge_modmem(u32 pages)
768 {
769 	if (atomic_long_add_return(pages, &bpf_jit_current) >
770 	    (bpf_jit_limit >> PAGE_SHIFT)) {
771 		if (!capable(CAP_SYS_ADMIN)) {
772 			atomic_long_sub(pages, &bpf_jit_current);
773 			return -EPERM;
774 		}
775 	}
776 
777 	return 0;
778 }
779 
780 static void bpf_jit_uncharge_modmem(u32 pages)
781 {
782 	atomic_long_sub(pages, &bpf_jit_current);
783 }
784 
785 void *__weak bpf_jit_alloc_exec(unsigned long size)
786 {
787 	return module_alloc(size);
788 }
789 
790 void __weak bpf_jit_free_exec(void *addr)
791 {
792 	module_memfree(addr);
793 }
794 
795 struct bpf_binary_header *
796 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
797 		     unsigned int alignment,
798 		     bpf_jit_fill_hole_t bpf_fill_ill_insns)
799 {
800 	struct bpf_binary_header *hdr;
801 	u32 size, hole, start, pages;
802 
803 	/* Most of BPF filters are really small, but if some of them
804 	 * fill a page, allow at least 128 extra bytes to insert a
805 	 * random section of illegal instructions.
806 	 */
807 	size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
808 	pages = size / PAGE_SIZE;
809 
810 	if (bpf_jit_charge_modmem(pages))
811 		return NULL;
812 	hdr = bpf_jit_alloc_exec(size);
813 	if (!hdr) {
814 		bpf_jit_uncharge_modmem(pages);
815 		return NULL;
816 	}
817 
818 	/* Fill space with illegal/arch-dep instructions. */
819 	bpf_fill_ill_insns(hdr, size);
820 
821 	hdr->pages = pages;
822 	hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
823 		     PAGE_SIZE - sizeof(*hdr));
824 	start = (get_random_int() % hole) & ~(alignment - 1);
825 
826 	/* Leave a random number of instructions before BPF code. */
827 	*image_ptr = &hdr->image[start];
828 
829 	return hdr;
830 }
831 
832 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
833 {
834 	u32 pages = hdr->pages;
835 
836 	bpf_jit_free_exec(hdr);
837 	bpf_jit_uncharge_modmem(pages);
838 }
839 
840 /* This symbol is only overridden by archs that have different
841  * requirements than the usual eBPF JITs, f.e. when they only
842  * implement cBPF JIT, do not set images read-only, etc.
843  */
844 void __weak bpf_jit_free(struct bpf_prog *fp)
845 {
846 	if (fp->jited) {
847 		struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
848 
849 		bpf_jit_binary_free(hdr);
850 
851 		WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
852 	}
853 
854 	bpf_prog_unlock_free(fp);
855 }
856 
857 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
858 			  const struct bpf_insn *insn, bool extra_pass,
859 			  u64 *func_addr, bool *func_addr_fixed)
860 {
861 	s16 off = insn->off;
862 	s32 imm = insn->imm;
863 	u8 *addr;
864 
865 	*func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
866 	if (!*func_addr_fixed) {
867 		/* Place-holder address till the last pass has collected
868 		 * all addresses for JITed subprograms in which case we
869 		 * can pick them up from prog->aux.
870 		 */
871 		if (!extra_pass)
872 			addr = NULL;
873 		else if (prog->aux->func &&
874 			 off >= 0 && off < prog->aux->func_cnt)
875 			addr = (u8 *)prog->aux->func[off]->bpf_func;
876 		else
877 			return -EINVAL;
878 	} else {
879 		/* Address of a BPF helper call. Since part of the core
880 		 * kernel, it's always at a fixed location. __bpf_call_base
881 		 * and the helper with imm relative to it are both in core
882 		 * kernel.
883 		 */
884 		addr = (u8 *)__bpf_call_base + imm;
885 	}
886 
887 	*func_addr = (unsigned long)addr;
888 	return 0;
889 }
890 
891 static int bpf_jit_blind_insn(const struct bpf_insn *from,
892 			      const struct bpf_insn *aux,
893 			      struct bpf_insn *to_buff)
894 {
895 	struct bpf_insn *to = to_buff;
896 	u32 imm_rnd = get_random_int();
897 	s16 off;
898 
899 	BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
900 	BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
901 
902 	/* Constraints on AX register:
903 	 *
904 	 * AX register is inaccessible from user space. It is mapped in
905 	 * all JITs, and used here for constant blinding rewrites. It is
906 	 * typically "stateless" meaning its contents are only valid within
907 	 * the executed instruction, but not across several instructions.
908 	 * There are a few exceptions however which are further detailed
909 	 * below.
910 	 *
911 	 * Constant blinding is only used by JITs, not in the interpreter.
912 	 * The interpreter uses AX in some occasions as a local temporary
913 	 * register e.g. in DIV or MOD instructions.
914 	 *
915 	 * In restricted circumstances, the verifier can also use the AX
916 	 * register for rewrites as long as they do not interfere with
917 	 * the above cases!
918 	 */
919 	if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
920 		goto out;
921 
922 	if (from->imm == 0 &&
923 	    (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
924 	     from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
925 		*to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
926 		goto out;
927 	}
928 
929 	switch (from->code) {
930 	case BPF_ALU | BPF_ADD | BPF_K:
931 	case BPF_ALU | BPF_SUB | BPF_K:
932 	case BPF_ALU | BPF_AND | BPF_K:
933 	case BPF_ALU | BPF_OR  | BPF_K:
934 	case BPF_ALU | BPF_XOR | BPF_K:
935 	case BPF_ALU | BPF_MUL | BPF_K:
936 	case BPF_ALU | BPF_MOV | BPF_K:
937 	case BPF_ALU | BPF_DIV | BPF_K:
938 	case BPF_ALU | BPF_MOD | BPF_K:
939 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
940 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
941 		*to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
942 		break;
943 
944 	case BPF_ALU64 | BPF_ADD | BPF_K:
945 	case BPF_ALU64 | BPF_SUB | BPF_K:
946 	case BPF_ALU64 | BPF_AND | BPF_K:
947 	case BPF_ALU64 | BPF_OR  | BPF_K:
948 	case BPF_ALU64 | BPF_XOR | BPF_K:
949 	case BPF_ALU64 | BPF_MUL | BPF_K:
950 	case BPF_ALU64 | BPF_MOV | BPF_K:
951 	case BPF_ALU64 | BPF_DIV | BPF_K:
952 	case BPF_ALU64 | BPF_MOD | BPF_K:
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_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
956 		break;
957 
958 	case BPF_JMP | BPF_JEQ  | BPF_K:
959 	case BPF_JMP | BPF_JNE  | BPF_K:
960 	case BPF_JMP | BPF_JGT  | BPF_K:
961 	case BPF_JMP | BPF_JLT  | BPF_K:
962 	case BPF_JMP | BPF_JGE  | BPF_K:
963 	case BPF_JMP | BPF_JLE  | BPF_K:
964 	case BPF_JMP | BPF_JSGT | BPF_K:
965 	case BPF_JMP | BPF_JSLT | BPF_K:
966 	case BPF_JMP | BPF_JSGE | BPF_K:
967 	case BPF_JMP | BPF_JSLE | BPF_K:
968 	case BPF_JMP | BPF_JSET | BPF_K:
969 		/* Accommodate for extra offset in case of a backjump. */
970 		off = from->off;
971 		if (off < 0)
972 			off -= 2;
973 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
974 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
975 		*to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
976 		break;
977 
978 	case BPF_JMP32 | BPF_JEQ  | BPF_K:
979 	case BPF_JMP32 | BPF_JNE  | BPF_K:
980 	case BPF_JMP32 | BPF_JGT  | BPF_K:
981 	case BPF_JMP32 | BPF_JLT  | BPF_K:
982 	case BPF_JMP32 | BPF_JGE  | BPF_K:
983 	case BPF_JMP32 | BPF_JLE  | BPF_K:
984 	case BPF_JMP32 | BPF_JSGT | BPF_K:
985 	case BPF_JMP32 | BPF_JSLT | BPF_K:
986 	case BPF_JMP32 | BPF_JSGE | BPF_K:
987 	case BPF_JMP32 | BPF_JSLE | BPF_K:
988 	case BPF_JMP32 | BPF_JSET | BPF_K:
989 		/* Accommodate for extra offset in case of a backjump. */
990 		off = from->off;
991 		if (off < 0)
992 			off -= 2;
993 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
994 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
995 		*to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
996 				      off);
997 		break;
998 
999 	case BPF_LD | BPF_IMM | BPF_DW:
1000 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1001 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1002 		*to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1003 		*to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1004 		break;
1005 	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1006 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1007 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1008 		*to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
1009 		break;
1010 
1011 	case BPF_ST | BPF_MEM | BPF_DW:
1012 	case BPF_ST | BPF_MEM | BPF_W:
1013 	case BPF_ST | BPF_MEM | BPF_H:
1014 	case BPF_ST | BPF_MEM | BPF_B:
1015 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1016 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1017 		*to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1018 		break;
1019 	}
1020 out:
1021 	return to - to_buff;
1022 }
1023 
1024 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1025 					      gfp_t gfp_extra_flags)
1026 {
1027 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1028 	struct bpf_prog *fp;
1029 
1030 	fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
1031 	if (fp != NULL) {
1032 		/* aux->prog still points to the fp_other one, so
1033 		 * when promoting the clone to the real program,
1034 		 * this still needs to be adapted.
1035 		 */
1036 		memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1037 	}
1038 
1039 	return fp;
1040 }
1041 
1042 static void bpf_prog_clone_free(struct bpf_prog *fp)
1043 {
1044 	/* aux was stolen by the other clone, so we cannot free
1045 	 * it from this path! It will be freed eventually by the
1046 	 * other program on release.
1047 	 *
1048 	 * At this point, we don't need a deferred release since
1049 	 * clone is guaranteed to not be locked.
1050 	 */
1051 	fp->aux = NULL;
1052 	__bpf_prog_free(fp);
1053 }
1054 
1055 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1056 {
1057 	/* We have to repoint aux->prog to self, as we don't
1058 	 * know whether fp here is the clone or the original.
1059 	 */
1060 	fp->aux->prog = fp;
1061 	bpf_prog_clone_free(fp_other);
1062 }
1063 
1064 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1065 {
1066 	struct bpf_insn insn_buff[16], aux[2];
1067 	struct bpf_prog *clone, *tmp;
1068 	int insn_delta, insn_cnt;
1069 	struct bpf_insn *insn;
1070 	int i, rewritten;
1071 
1072 	if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1073 		return prog;
1074 
1075 	clone = bpf_prog_clone_create(prog, GFP_USER);
1076 	if (!clone)
1077 		return ERR_PTR(-ENOMEM);
1078 
1079 	insn_cnt = clone->len;
1080 	insn = clone->insnsi;
1081 
1082 	for (i = 0; i < insn_cnt; i++, insn++) {
1083 		/* We temporarily need to hold the original ld64 insn
1084 		 * so that we can still access the first part in the
1085 		 * second blinding run.
1086 		 */
1087 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1088 		    insn[1].code == 0)
1089 			memcpy(aux, insn, sizeof(aux));
1090 
1091 		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
1092 		if (!rewritten)
1093 			continue;
1094 
1095 		tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1096 		if (IS_ERR(tmp)) {
1097 			/* Patching may have repointed aux->prog during
1098 			 * realloc from the original one, so we need to
1099 			 * fix it up here on error.
1100 			 */
1101 			bpf_jit_prog_release_other(prog, clone);
1102 			return tmp;
1103 		}
1104 
1105 		clone = tmp;
1106 		insn_delta = rewritten - 1;
1107 
1108 		/* Walk new program and skip insns we just inserted. */
1109 		insn = clone->insnsi + i + insn_delta;
1110 		insn_cnt += insn_delta;
1111 		i        += insn_delta;
1112 	}
1113 
1114 	clone->blinded = 1;
1115 	return clone;
1116 }
1117 #endif /* CONFIG_BPF_JIT */
1118 
1119 /* Base function for offset calculation. Needs to go into .text section,
1120  * therefore keeping it non-static as well; will also be used by JITs
1121  * anyway later on, so do not let the compiler omit it. This also needs
1122  * to go into kallsyms for correlation from e.g. bpftool, so naming
1123  * must not change.
1124  */
1125 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1126 {
1127 	return 0;
1128 }
1129 EXPORT_SYMBOL_GPL(__bpf_call_base);
1130 
1131 /* All UAPI available opcodes. */
1132 #define BPF_INSN_MAP(INSN_2, INSN_3)		\
1133 	/* 32 bit ALU operations. */		\
1134 	/*   Register based. */			\
1135 	INSN_3(ALU, ADD,  X),			\
1136 	INSN_3(ALU, SUB,  X),			\
1137 	INSN_3(ALU, AND,  X),			\
1138 	INSN_3(ALU, OR,   X),			\
1139 	INSN_3(ALU, LSH,  X),			\
1140 	INSN_3(ALU, RSH,  X),			\
1141 	INSN_3(ALU, XOR,  X),			\
1142 	INSN_3(ALU, MUL,  X),			\
1143 	INSN_3(ALU, MOV,  X),			\
1144 	INSN_3(ALU, ARSH, X),			\
1145 	INSN_3(ALU, DIV,  X),			\
1146 	INSN_3(ALU, MOD,  X),			\
1147 	INSN_2(ALU, NEG),			\
1148 	INSN_3(ALU, END, TO_BE),		\
1149 	INSN_3(ALU, END, TO_LE),		\
1150 	/*   Immediate based. */		\
1151 	INSN_3(ALU, ADD,  K),			\
1152 	INSN_3(ALU, SUB,  K),			\
1153 	INSN_3(ALU, AND,  K),			\
1154 	INSN_3(ALU, OR,   K),			\
1155 	INSN_3(ALU, LSH,  K),			\
1156 	INSN_3(ALU, RSH,  K),			\
1157 	INSN_3(ALU, XOR,  K),			\
1158 	INSN_3(ALU, MUL,  K),			\
1159 	INSN_3(ALU, MOV,  K),			\
1160 	INSN_3(ALU, ARSH, K),			\
1161 	INSN_3(ALU, DIV,  K),			\
1162 	INSN_3(ALU, MOD,  K),			\
1163 	/* 64 bit ALU operations. */		\
1164 	/*   Register based. */			\
1165 	INSN_3(ALU64, ADD,  X),			\
1166 	INSN_3(ALU64, SUB,  X),			\
1167 	INSN_3(ALU64, AND,  X),			\
1168 	INSN_3(ALU64, OR,   X),			\
1169 	INSN_3(ALU64, LSH,  X),			\
1170 	INSN_3(ALU64, RSH,  X),			\
1171 	INSN_3(ALU64, XOR,  X),			\
1172 	INSN_3(ALU64, MUL,  X),			\
1173 	INSN_3(ALU64, MOV,  X),			\
1174 	INSN_3(ALU64, ARSH, X),			\
1175 	INSN_3(ALU64, DIV,  X),			\
1176 	INSN_3(ALU64, MOD,  X),			\
1177 	INSN_2(ALU64, NEG),			\
1178 	/*   Immediate based. */		\
1179 	INSN_3(ALU64, ADD,  K),			\
1180 	INSN_3(ALU64, SUB,  K),			\
1181 	INSN_3(ALU64, AND,  K),			\
1182 	INSN_3(ALU64, OR,   K),			\
1183 	INSN_3(ALU64, LSH,  K),			\
1184 	INSN_3(ALU64, RSH,  K),			\
1185 	INSN_3(ALU64, XOR,  K),			\
1186 	INSN_3(ALU64, MUL,  K),			\
1187 	INSN_3(ALU64, MOV,  K),			\
1188 	INSN_3(ALU64, ARSH, K),			\
1189 	INSN_3(ALU64, DIV,  K),			\
1190 	INSN_3(ALU64, MOD,  K),			\
1191 	/* Call instruction. */			\
1192 	INSN_2(JMP, CALL),			\
1193 	/* Exit instruction. */			\
1194 	INSN_2(JMP, EXIT),			\
1195 	/* 32-bit Jump instructions. */		\
1196 	/*   Register based. */			\
1197 	INSN_3(JMP32, JEQ,  X),			\
1198 	INSN_3(JMP32, JNE,  X),			\
1199 	INSN_3(JMP32, JGT,  X),			\
1200 	INSN_3(JMP32, JLT,  X),			\
1201 	INSN_3(JMP32, JGE,  X),			\
1202 	INSN_3(JMP32, JLE,  X),			\
1203 	INSN_3(JMP32, JSGT, X),			\
1204 	INSN_3(JMP32, JSLT, X),			\
1205 	INSN_3(JMP32, JSGE, X),			\
1206 	INSN_3(JMP32, JSLE, X),			\
1207 	INSN_3(JMP32, JSET, X),			\
1208 	/*   Immediate based. */		\
1209 	INSN_3(JMP32, JEQ,  K),			\
1210 	INSN_3(JMP32, JNE,  K),			\
1211 	INSN_3(JMP32, JGT,  K),			\
1212 	INSN_3(JMP32, JLT,  K),			\
1213 	INSN_3(JMP32, JGE,  K),			\
1214 	INSN_3(JMP32, JLE,  K),			\
1215 	INSN_3(JMP32, JSGT, K),			\
1216 	INSN_3(JMP32, JSLT, K),			\
1217 	INSN_3(JMP32, JSGE, K),			\
1218 	INSN_3(JMP32, JSLE, K),			\
1219 	INSN_3(JMP32, JSET, K),			\
1220 	/* Jump instructions. */		\
1221 	/*   Register based. */			\
1222 	INSN_3(JMP, JEQ,  X),			\
1223 	INSN_3(JMP, JNE,  X),			\
1224 	INSN_3(JMP, JGT,  X),			\
1225 	INSN_3(JMP, JLT,  X),			\
1226 	INSN_3(JMP, JGE,  X),			\
1227 	INSN_3(JMP, JLE,  X),			\
1228 	INSN_3(JMP, JSGT, X),			\
1229 	INSN_3(JMP, JSLT, X),			\
1230 	INSN_3(JMP, JSGE, X),			\
1231 	INSN_3(JMP, JSLE, X),			\
1232 	INSN_3(JMP, JSET, X),			\
1233 	/*   Immediate based. */		\
1234 	INSN_3(JMP, JEQ,  K),			\
1235 	INSN_3(JMP, JNE,  K),			\
1236 	INSN_3(JMP, JGT,  K),			\
1237 	INSN_3(JMP, JLT,  K),			\
1238 	INSN_3(JMP, JGE,  K),			\
1239 	INSN_3(JMP, JLE,  K),			\
1240 	INSN_3(JMP, JSGT, K),			\
1241 	INSN_3(JMP, JSLT, K),			\
1242 	INSN_3(JMP, JSGE, K),			\
1243 	INSN_3(JMP, JSLE, K),			\
1244 	INSN_3(JMP, JSET, K),			\
1245 	INSN_2(JMP, JA),			\
1246 	/* Store instructions. */		\
1247 	/*   Register based. */			\
1248 	INSN_3(STX, MEM,  B),			\
1249 	INSN_3(STX, MEM,  H),			\
1250 	INSN_3(STX, MEM,  W),			\
1251 	INSN_3(STX, MEM,  DW),			\
1252 	INSN_3(STX, XADD, W),			\
1253 	INSN_3(STX, XADD, DW),			\
1254 	/*   Immediate based. */		\
1255 	INSN_3(ST, MEM, B),			\
1256 	INSN_3(ST, MEM, H),			\
1257 	INSN_3(ST, MEM, W),			\
1258 	INSN_3(ST, MEM, DW),			\
1259 	/* Load instructions. */		\
1260 	/*   Register based. */			\
1261 	INSN_3(LDX, MEM, B),			\
1262 	INSN_3(LDX, MEM, H),			\
1263 	INSN_3(LDX, MEM, W),			\
1264 	INSN_3(LDX, MEM, DW),			\
1265 	/*   Immediate based. */		\
1266 	INSN_3(LD, IMM, DW)
1267 
1268 bool bpf_opcode_in_insntable(u8 code)
1269 {
1270 #define BPF_INSN_2_TBL(x, y)    [BPF_##x | BPF_##y] = true
1271 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1272 	static const bool public_insntable[256] = {
1273 		[0 ... 255] = false,
1274 		/* Now overwrite non-defaults ... */
1275 		BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1276 		/* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1277 		[BPF_LD | BPF_ABS | BPF_B] = true,
1278 		[BPF_LD | BPF_ABS | BPF_H] = true,
1279 		[BPF_LD | BPF_ABS | BPF_W] = true,
1280 		[BPF_LD | BPF_IND | BPF_B] = true,
1281 		[BPF_LD | BPF_IND | BPF_H] = true,
1282 		[BPF_LD | BPF_IND | BPF_W] = true,
1283 	};
1284 #undef BPF_INSN_3_TBL
1285 #undef BPF_INSN_2_TBL
1286 	return public_insntable[code];
1287 }
1288 
1289 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1290 /**
1291  *	__bpf_prog_run - run eBPF program on a given context
1292  *	@regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1293  *	@insn: is the array of eBPF instructions
1294  *	@stack: is the eBPF storage stack
1295  *
1296  * Decode and execute eBPF instructions.
1297  */
1298 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1299 {
1300 #define BPF_INSN_2_LBL(x, y)    [BPF_##x | BPF_##y] = &&x##_##y
1301 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1302 	static const void *jumptable[256] = {
1303 		[0 ... 255] = &&default_label,
1304 		/* Now overwrite non-defaults ... */
1305 		BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1306 		/* Non-UAPI available opcodes. */
1307 		[BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1308 		[BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1309 	};
1310 #undef BPF_INSN_3_LBL
1311 #undef BPF_INSN_2_LBL
1312 	u32 tail_call_cnt = 0;
1313 
1314 #define CONT	 ({ insn++; goto select_insn; })
1315 #define CONT_JMP ({ insn++; goto select_insn; })
1316 
1317 select_insn:
1318 	goto *jumptable[insn->code];
1319 
1320 	/* ALU */
1321 #define ALU(OPCODE, OP)			\
1322 	ALU64_##OPCODE##_X:		\
1323 		DST = DST OP SRC;	\
1324 		CONT;			\
1325 	ALU_##OPCODE##_X:		\
1326 		DST = (u32) DST OP (u32) SRC;	\
1327 		CONT;			\
1328 	ALU64_##OPCODE##_K:		\
1329 		DST = DST OP IMM;		\
1330 		CONT;			\
1331 	ALU_##OPCODE##_K:		\
1332 		DST = (u32) DST OP (u32) IMM;	\
1333 		CONT;
1334 
1335 	ALU(ADD,  +)
1336 	ALU(SUB,  -)
1337 	ALU(AND,  &)
1338 	ALU(OR,   |)
1339 	ALU(LSH, <<)
1340 	ALU(RSH, >>)
1341 	ALU(XOR,  ^)
1342 	ALU(MUL,  *)
1343 #undef ALU
1344 	ALU_NEG:
1345 		DST = (u32) -DST;
1346 		CONT;
1347 	ALU64_NEG:
1348 		DST = -DST;
1349 		CONT;
1350 	ALU_MOV_X:
1351 		DST = (u32) SRC;
1352 		CONT;
1353 	ALU_MOV_K:
1354 		DST = (u32) IMM;
1355 		CONT;
1356 	ALU64_MOV_X:
1357 		DST = SRC;
1358 		CONT;
1359 	ALU64_MOV_K:
1360 		DST = IMM;
1361 		CONT;
1362 	LD_IMM_DW:
1363 		DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1364 		insn++;
1365 		CONT;
1366 	ALU_ARSH_X:
1367 		DST = (u64) (u32) ((*(s32 *) &DST) >> SRC);
1368 		CONT;
1369 	ALU_ARSH_K:
1370 		DST = (u64) (u32) ((*(s32 *) &DST) >> IMM);
1371 		CONT;
1372 	ALU64_ARSH_X:
1373 		(*(s64 *) &DST) >>= SRC;
1374 		CONT;
1375 	ALU64_ARSH_K:
1376 		(*(s64 *) &DST) >>= IMM;
1377 		CONT;
1378 	ALU64_MOD_X:
1379 		div64_u64_rem(DST, SRC, &AX);
1380 		DST = AX;
1381 		CONT;
1382 	ALU_MOD_X:
1383 		AX = (u32) DST;
1384 		DST = do_div(AX, (u32) SRC);
1385 		CONT;
1386 	ALU64_MOD_K:
1387 		div64_u64_rem(DST, IMM, &AX);
1388 		DST = AX;
1389 		CONT;
1390 	ALU_MOD_K:
1391 		AX = (u32) DST;
1392 		DST = do_div(AX, (u32) IMM);
1393 		CONT;
1394 	ALU64_DIV_X:
1395 		DST = div64_u64(DST, SRC);
1396 		CONT;
1397 	ALU_DIV_X:
1398 		AX = (u32) DST;
1399 		do_div(AX, (u32) SRC);
1400 		DST = (u32) AX;
1401 		CONT;
1402 	ALU64_DIV_K:
1403 		DST = div64_u64(DST, IMM);
1404 		CONT;
1405 	ALU_DIV_K:
1406 		AX = (u32) DST;
1407 		do_div(AX, (u32) IMM);
1408 		DST = (u32) AX;
1409 		CONT;
1410 	ALU_END_TO_BE:
1411 		switch (IMM) {
1412 		case 16:
1413 			DST = (__force u16) cpu_to_be16(DST);
1414 			break;
1415 		case 32:
1416 			DST = (__force u32) cpu_to_be32(DST);
1417 			break;
1418 		case 64:
1419 			DST = (__force u64) cpu_to_be64(DST);
1420 			break;
1421 		}
1422 		CONT;
1423 	ALU_END_TO_LE:
1424 		switch (IMM) {
1425 		case 16:
1426 			DST = (__force u16) cpu_to_le16(DST);
1427 			break;
1428 		case 32:
1429 			DST = (__force u32) cpu_to_le32(DST);
1430 			break;
1431 		case 64:
1432 			DST = (__force u64) cpu_to_le64(DST);
1433 			break;
1434 		}
1435 		CONT;
1436 
1437 	/* CALL */
1438 	JMP_CALL:
1439 		/* Function call scratches BPF_R1-BPF_R5 registers,
1440 		 * preserves BPF_R6-BPF_R9, and stores return value
1441 		 * into BPF_R0.
1442 		 */
1443 		BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1444 						       BPF_R4, BPF_R5);
1445 		CONT;
1446 
1447 	JMP_CALL_ARGS:
1448 		BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1449 							    BPF_R3, BPF_R4,
1450 							    BPF_R5,
1451 							    insn + insn->off + 1);
1452 		CONT;
1453 
1454 	JMP_TAIL_CALL: {
1455 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1456 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1457 		struct bpf_prog *prog;
1458 		u32 index = BPF_R3;
1459 
1460 		if (unlikely(index >= array->map.max_entries))
1461 			goto out;
1462 		if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1463 			goto out;
1464 
1465 		tail_call_cnt++;
1466 
1467 		prog = READ_ONCE(array->ptrs[index]);
1468 		if (!prog)
1469 			goto out;
1470 
1471 		/* ARG1 at this point is guaranteed to point to CTX from
1472 		 * the verifier side due to the fact that the tail call is
1473 		 * handeled like a helper, that is, bpf_tail_call_proto,
1474 		 * where arg1_type is ARG_PTR_TO_CTX.
1475 		 */
1476 		insn = prog->insnsi;
1477 		goto select_insn;
1478 out:
1479 		CONT;
1480 	}
1481 	JMP_JA:
1482 		insn += insn->off;
1483 		CONT;
1484 	JMP_EXIT:
1485 		return BPF_R0;
1486 	/* JMP */
1487 #define COND_JMP(SIGN, OPCODE, CMP_OP)				\
1488 	JMP_##OPCODE##_X:					\
1489 		if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) {	\
1490 			insn += insn->off;			\
1491 			CONT_JMP;				\
1492 		}						\
1493 		CONT;						\
1494 	JMP32_##OPCODE##_X:					\
1495 		if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) {	\
1496 			insn += insn->off;			\
1497 			CONT_JMP;				\
1498 		}						\
1499 		CONT;						\
1500 	JMP_##OPCODE##_K:					\
1501 		if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) {	\
1502 			insn += insn->off;			\
1503 			CONT_JMP;				\
1504 		}						\
1505 		CONT;						\
1506 	JMP32_##OPCODE##_K:					\
1507 		if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) {	\
1508 			insn += insn->off;			\
1509 			CONT_JMP;				\
1510 		}						\
1511 		CONT;
1512 	COND_JMP(u, JEQ, ==)
1513 	COND_JMP(u, JNE, !=)
1514 	COND_JMP(u, JGT, >)
1515 	COND_JMP(u, JLT, <)
1516 	COND_JMP(u, JGE, >=)
1517 	COND_JMP(u, JLE, <=)
1518 	COND_JMP(u, JSET, &)
1519 	COND_JMP(s, JSGT, >)
1520 	COND_JMP(s, JSLT, <)
1521 	COND_JMP(s, JSGE, >=)
1522 	COND_JMP(s, JSLE, <=)
1523 #undef COND_JMP
1524 	/* STX and ST and LDX*/
1525 #define LDST(SIZEOP, SIZE)						\
1526 	STX_MEM_##SIZEOP:						\
1527 		*(SIZE *)(unsigned long) (DST + insn->off) = SRC;	\
1528 		CONT;							\
1529 	ST_MEM_##SIZEOP:						\
1530 		*(SIZE *)(unsigned long) (DST + insn->off) = IMM;	\
1531 		CONT;							\
1532 	LDX_MEM_##SIZEOP:						\
1533 		DST = *(SIZE *)(unsigned long) (SRC + insn->off);	\
1534 		CONT;
1535 
1536 	LDST(B,   u8)
1537 	LDST(H,  u16)
1538 	LDST(W,  u32)
1539 	LDST(DW, u64)
1540 #undef LDST
1541 	STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1542 		atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1543 			   (DST + insn->off));
1544 		CONT;
1545 	STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1546 		atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1547 			     (DST + insn->off));
1548 		CONT;
1549 
1550 	default_label:
1551 		/* If we ever reach this, we have a bug somewhere. Die hard here
1552 		 * instead of just returning 0; we could be somewhere in a subprog,
1553 		 * so execution could continue otherwise which we do /not/ want.
1554 		 *
1555 		 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1556 		 */
1557 		pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1558 		BUG_ON(1);
1559 		return 0;
1560 }
1561 STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */
1562 
1563 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1564 #define DEFINE_BPF_PROG_RUN(stack_size) \
1565 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1566 { \
1567 	u64 stack[stack_size / sizeof(u64)]; \
1568 	u64 regs[MAX_BPF_EXT_REG]; \
1569 \
1570 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1571 	ARG1 = (u64) (unsigned long) ctx; \
1572 	return ___bpf_prog_run(regs, insn, stack); \
1573 }
1574 
1575 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1576 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1577 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1578 				      const struct bpf_insn *insn) \
1579 { \
1580 	u64 stack[stack_size / sizeof(u64)]; \
1581 	u64 regs[MAX_BPF_EXT_REG]; \
1582 \
1583 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1584 	BPF_R1 = r1; \
1585 	BPF_R2 = r2; \
1586 	BPF_R3 = r3; \
1587 	BPF_R4 = r4; \
1588 	BPF_R5 = r5; \
1589 	return ___bpf_prog_run(regs, insn, stack); \
1590 }
1591 
1592 #define EVAL1(FN, X) FN(X)
1593 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1594 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1595 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1596 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1597 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1598 
1599 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1600 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1601 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1602 
1603 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1604 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1605 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1606 
1607 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1608 
1609 static unsigned int (*interpreters[])(const void *ctx,
1610 				      const struct bpf_insn *insn) = {
1611 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1612 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1613 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1614 };
1615 #undef PROG_NAME_LIST
1616 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1617 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
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 
1625 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1626 {
1627 	stack_depth = max_t(u32, stack_depth, 1);
1628 	insn->off = (s16) insn->imm;
1629 	insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1630 		__bpf_call_base_args;
1631 	insn->code = BPF_JMP | BPF_CALL_ARGS;
1632 }
1633 
1634 #else
1635 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1636 					 const struct bpf_insn *insn)
1637 {
1638 	/* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1639 	 * is not working properly, so warn about it!
1640 	 */
1641 	WARN_ON_ONCE(1);
1642 	return 0;
1643 }
1644 #endif
1645 
1646 bool bpf_prog_array_compatible(struct bpf_array *array,
1647 			       const struct bpf_prog *fp)
1648 {
1649 	if (fp->kprobe_override)
1650 		return false;
1651 
1652 	if (!array->owner_prog_type) {
1653 		/* There's no owner yet where we could check for
1654 		 * compatibility.
1655 		 */
1656 		array->owner_prog_type = fp->type;
1657 		array->owner_jited = fp->jited;
1658 
1659 		return true;
1660 	}
1661 
1662 	return array->owner_prog_type == fp->type &&
1663 	       array->owner_jited == fp->jited;
1664 }
1665 
1666 static int bpf_check_tail_call(const struct bpf_prog *fp)
1667 {
1668 	struct bpf_prog_aux *aux = fp->aux;
1669 	int i;
1670 
1671 	for (i = 0; i < aux->used_map_cnt; i++) {
1672 		struct bpf_map *map = aux->used_maps[i];
1673 		struct bpf_array *array;
1674 
1675 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1676 			continue;
1677 
1678 		array = container_of(map, struct bpf_array, map);
1679 		if (!bpf_prog_array_compatible(array, fp))
1680 			return -EINVAL;
1681 	}
1682 
1683 	return 0;
1684 }
1685 
1686 static void bpf_prog_select_func(struct bpf_prog *fp)
1687 {
1688 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1689 	u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1690 
1691 	fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1692 #else
1693 	fp->bpf_func = __bpf_prog_ret0_warn;
1694 #endif
1695 }
1696 
1697 /**
1698  *	bpf_prog_select_runtime - select exec runtime for BPF program
1699  *	@fp: bpf_prog populated with internal BPF program
1700  *	@err: pointer to error variable
1701  *
1702  * Try to JIT eBPF program, if JIT is not available, use interpreter.
1703  * The BPF program will be executed via BPF_PROG_RUN() macro.
1704  */
1705 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1706 {
1707 	/* In case of BPF to BPF calls, verifier did all the prep
1708 	 * work with regards to JITing, etc.
1709 	 */
1710 	if (fp->bpf_func)
1711 		goto finalize;
1712 
1713 	bpf_prog_select_func(fp);
1714 
1715 	/* eBPF JITs can rewrite the program in case constant
1716 	 * blinding is active. However, in case of error during
1717 	 * blinding, bpf_int_jit_compile() must always return a
1718 	 * valid program, which in this case would simply not
1719 	 * be JITed, but falls back to the interpreter.
1720 	 */
1721 	if (!bpf_prog_is_dev_bound(fp->aux)) {
1722 		*err = bpf_prog_alloc_jited_linfo(fp);
1723 		if (*err)
1724 			return fp;
1725 
1726 		fp = bpf_int_jit_compile(fp);
1727 		if (!fp->jited) {
1728 			bpf_prog_free_jited_linfo(fp);
1729 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1730 			*err = -ENOTSUPP;
1731 			return fp;
1732 #endif
1733 		} else {
1734 			bpf_prog_free_unused_jited_linfo(fp);
1735 		}
1736 	} else {
1737 		*err = bpf_prog_offload_compile(fp);
1738 		if (*err)
1739 			return fp;
1740 	}
1741 
1742 finalize:
1743 	bpf_prog_lock_ro(fp);
1744 
1745 	/* The tail call compatibility check can only be done at
1746 	 * this late stage as we need to determine, if we deal
1747 	 * with JITed or non JITed program concatenations and not
1748 	 * all eBPF JITs might immediately support all features.
1749 	 */
1750 	*err = bpf_check_tail_call(fp);
1751 
1752 	return fp;
1753 }
1754 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1755 
1756 static unsigned int __bpf_prog_ret1(const void *ctx,
1757 				    const struct bpf_insn *insn)
1758 {
1759 	return 1;
1760 }
1761 
1762 static struct bpf_prog_dummy {
1763 	struct bpf_prog prog;
1764 } dummy_bpf_prog = {
1765 	.prog = {
1766 		.bpf_func = __bpf_prog_ret1,
1767 	},
1768 };
1769 
1770 /* to avoid allocating empty bpf_prog_array for cgroups that
1771  * don't have bpf program attached use one global 'empty_prog_array'
1772  * It will not be modified the caller of bpf_prog_array_alloc()
1773  * (since caller requested prog_cnt == 0)
1774  * that pointer should be 'freed' by bpf_prog_array_free()
1775  */
1776 static struct {
1777 	struct bpf_prog_array hdr;
1778 	struct bpf_prog *null_prog;
1779 } empty_prog_array = {
1780 	.null_prog = NULL,
1781 };
1782 
1783 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1784 {
1785 	if (prog_cnt)
1786 		return kzalloc(sizeof(struct bpf_prog_array) +
1787 			       sizeof(struct bpf_prog_array_item) *
1788 			       (prog_cnt + 1),
1789 			       flags);
1790 
1791 	return &empty_prog_array.hdr;
1792 }
1793 
1794 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs)
1795 {
1796 	if (!progs ||
1797 	    progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr)
1798 		return;
1799 	kfree_rcu(progs, rcu);
1800 }
1801 
1802 int bpf_prog_array_length(struct bpf_prog_array __rcu *array)
1803 {
1804 	struct bpf_prog_array_item *item;
1805 	u32 cnt = 0;
1806 
1807 	rcu_read_lock();
1808 	item = rcu_dereference(array)->items;
1809 	for (; item->prog; item++)
1810 		if (item->prog != &dummy_bpf_prog.prog)
1811 			cnt++;
1812 	rcu_read_unlock();
1813 	return cnt;
1814 }
1815 
1816 
1817 static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array,
1818 				     u32 *prog_ids,
1819 				     u32 request_cnt)
1820 {
1821 	struct bpf_prog_array_item *item;
1822 	int i = 0;
1823 
1824 	item = rcu_dereference_check(array, 1)->items;
1825 	for (; item->prog; item++) {
1826 		if (item->prog == &dummy_bpf_prog.prog)
1827 			continue;
1828 		prog_ids[i] = item->prog->aux->id;
1829 		if (++i == request_cnt) {
1830 			item++;
1831 			break;
1832 		}
1833 	}
1834 
1835 	return !!(item->prog);
1836 }
1837 
1838 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array,
1839 				__u32 __user *prog_ids, u32 cnt)
1840 {
1841 	unsigned long err = 0;
1842 	bool nospc;
1843 	u32 *ids;
1844 
1845 	/* users of this function are doing:
1846 	 * cnt = bpf_prog_array_length();
1847 	 * if (cnt > 0)
1848 	 *     bpf_prog_array_copy_to_user(..., cnt);
1849 	 * so below kcalloc doesn't need extra cnt > 0 check, but
1850 	 * bpf_prog_array_length() releases rcu lock and
1851 	 * prog array could have been swapped with empty or larger array,
1852 	 * so always copy 'cnt' prog_ids to the user.
1853 	 * In a rare race the user will see zero prog_ids
1854 	 */
1855 	ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1856 	if (!ids)
1857 		return -ENOMEM;
1858 	rcu_read_lock();
1859 	nospc = bpf_prog_array_copy_core(array, ids, cnt);
1860 	rcu_read_unlock();
1861 	err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1862 	kfree(ids);
1863 	if (err)
1864 		return -EFAULT;
1865 	if (nospc)
1866 		return -ENOSPC;
1867 	return 0;
1868 }
1869 
1870 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array,
1871 				struct bpf_prog *old_prog)
1872 {
1873 	struct bpf_prog_array_item *item = array->items;
1874 
1875 	for (; item->prog; item++)
1876 		if (item->prog == old_prog) {
1877 			WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1878 			break;
1879 		}
1880 }
1881 
1882 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array,
1883 			struct bpf_prog *exclude_prog,
1884 			struct bpf_prog *include_prog,
1885 			struct bpf_prog_array **new_array)
1886 {
1887 	int new_prog_cnt, carry_prog_cnt = 0;
1888 	struct bpf_prog_array_item *existing;
1889 	struct bpf_prog_array *array;
1890 	bool found_exclude = false;
1891 	int new_prog_idx = 0;
1892 
1893 	/* Figure out how many existing progs we need to carry over to
1894 	 * the new array.
1895 	 */
1896 	if (old_array) {
1897 		existing = old_array->items;
1898 		for (; existing->prog; existing++) {
1899 			if (existing->prog == exclude_prog) {
1900 				found_exclude = true;
1901 				continue;
1902 			}
1903 			if (existing->prog != &dummy_bpf_prog.prog)
1904 				carry_prog_cnt++;
1905 			if (existing->prog == include_prog)
1906 				return -EEXIST;
1907 		}
1908 	}
1909 
1910 	if (exclude_prog && !found_exclude)
1911 		return -ENOENT;
1912 
1913 	/* How many progs (not NULL) will be in the new array? */
1914 	new_prog_cnt = carry_prog_cnt;
1915 	if (include_prog)
1916 		new_prog_cnt += 1;
1917 
1918 	/* Do we have any prog (not NULL) in the new array? */
1919 	if (!new_prog_cnt) {
1920 		*new_array = NULL;
1921 		return 0;
1922 	}
1923 
1924 	/* +1 as the end of prog_array is marked with NULL */
1925 	array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
1926 	if (!array)
1927 		return -ENOMEM;
1928 
1929 	/* Fill in the new prog array */
1930 	if (carry_prog_cnt) {
1931 		existing = old_array->items;
1932 		for (; existing->prog; existing++)
1933 			if (existing->prog != exclude_prog &&
1934 			    existing->prog != &dummy_bpf_prog.prog) {
1935 				array->items[new_prog_idx++].prog =
1936 					existing->prog;
1937 			}
1938 	}
1939 	if (include_prog)
1940 		array->items[new_prog_idx++].prog = include_prog;
1941 	array->items[new_prog_idx].prog = NULL;
1942 	*new_array = array;
1943 	return 0;
1944 }
1945 
1946 int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array,
1947 			     u32 *prog_ids, u32 request_cnt,
1948 			     u32 *prog_cnt)
1949 {
1950 	u32 cnt = 0;
1951 
1952 	if (array)
1953 		cnt = bpf_prog_array_length(array);
1954 
1955 	*prog_cnt = cnt;
1956 
1957 	/* return early if user requested only program count or nothing to copy */
1958 	if (!request_cnt || !cnt)
1959 		return 0;
1960 
1961 	/* this function is called under trace/bpf_trace.c: bpf_event_mutex */
1962 	return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
1963 								     : 0;
1964 }
1965 
1966 static void bpf_prog_free_deferred(struct work_struct *work)
1967 {
1968 	struct bpf_prog_aux *aux;
1969 	int i;
1970 
1971 	aux = container_of(work, struct bpf_prog_aux, work);
1972 	if (bpf_prog_is_dev_bound(aux))
1973 		bpf_prog_offload_destroy(aux->prog);
1974 #ifdef CONFIG_PERF_EVENTS
1975 	if (aux->prog->has_callchain_buf)
1976 		put_callchain_buffers();
1977 #endif
1978 	for (i = 0; i < aux->func_cnt; i++)
1979 		bpf_jit_free(aux->func[i]);
1980 	if (aux->func_cnt) {
1981 		kfree(aux->func);
1982 		bpf_prog_unlock_free(aux->prog);
1983 	} else {
1984 		bpf_jit_free(aux->prog);
1985 	}
1986 }
1987 
1988 /* Free internal BPF program */
1989 void bpf_prog_free(struct bpf_prog *fp)
1990 {
1991 	struct bpf_prog_aux *aux = fp->aux;
1992 
1993 	INIT_WORK(&aux->work, bpf_prog_free_deferred);
1994 	schedule_work(&aux->work);
1995 }
1996 EXPORT_SYMBOL_GPL(bpf_prog_free);
1997 
1998 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
1999 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2000 
2001 void bpf_user_rnd_init_once(void)
2002 {
2003 	prandom_init_once(&bpf_user_rnd_state);
2004 }
2005 
2006 BPF_CALL_0(bpf_user_rnd_u32)
2007 {
2008 	/* Should someone ever have the rather unwise idea to use some
2009 	 * of the registers passed into this function, then note that
2010 	 * this function is called from native eBPF and classic-to-eBPF
2011 	 * transformations. Register assignments from both sides are
2012 	 * different, f.e. classic always sets fn(ctx, A, X) here.
2013 	 */
2014 	struct rnd_state *state;
2015 	u32 res;
2016 
2017 	state = &get_cpu_var(bpf_user_rnd_state);
2018 	res = prandom_u32_state(state);
2019 	put_cpu_var(bpf_user_rnd_state);
2020 
2021 	return res;
2022 }
2023 
2024 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2025 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2026 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2027 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2028 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2029 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2030 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2031 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2032 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2033 
2034 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2035 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2036 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2037 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2038 
2039 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2040 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2041 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2042 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2043 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2044 
2045 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2046 {
2047 	return NULL;
2048 }
2049 
2050 u64 __weak
2051 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2052 		 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2053 {
2054 	return -ENOTSUPP;
2055 }
2056 EXPORT_SYMBOL_GPL(bpf_event_output);
2057 
2058 /* Always built-in helper functions. */
2059 const struct bpf_func_proto bpf_tail_call_proto = {
2060 	.func		= NULL,
2061 	.gpl_only	= false,
2062 	.ret_type	= RET_VOID,
2063 	.arg1_type	= ARG_PTR_TO_CTX,
2064 	.arg2_type	= ARG_CONST_MAP_PTR,
2065 	.arg3_type	= ARG_ANYTHING,
2066 };
2067 
2068 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2069  * It is encouraged to implement bpf_int_jit_compile() instead, so that
2070  * eBPF and implicitly also cBPF can get JITed!
2071  */
2072 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2073 {
2074 	return prog;
2075 }
2076 
2077 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2078  * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2079  */
2080 void __weak bpf_jit_compile(struct bpf_prog *prog)
2081 {
2082 }
2083 
2084 bool __weak bpf_helper_changes_pkt_data(void *func)
2085 {
2086 	return false;
2087 }
2088 
2089 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2090  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2091  */
2092 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2093 			 int len)
2094 {
2095 	return -EFAULT;
2096 }
2097 
2098 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2099 EXPORT_SYMBOL(bpf_stats_enabled_key);
2100 int sysctl_bpf_stats_enabled __read_mostly;
2101 
2102 /* All definitions of tracepoints related to BPF. */
2103 #define CREATE_TRACE_POINTS
2104 #include <linux/bpf_trace.h>
2105 
2106 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2107