xref: /openbmc/linux/kernel/bpf/core.c (revision a44e4f3a)
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 static 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 			      bool emit_zext)
895 {
896 	struct bpf_insn *to = to_buff;
897 	u32 imm_rnd = get_random_int();
898 	s16 off;
899 
900 	BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
901 	BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
902 
903 	/* Constraints on AX register:
904 	 *
905 	 * AX register is inaccessible from user space. It is mapped in
906 	 * all JITs, and used here for constant blinding rewrites. It is
907 	 * typically "stateless" meaning its contents are only valid within
908 	 * the executed instruction, but not across several instructions.
909 	 * There are a few exceptions however which are further detailed
910 	 * below.
911 	 *
912 	 * Constant blinding is only used by JITs, not in the interpreter.
913 	 * The interpreter uses AX in some occasions as a local temporary
914 	 * register e.g. in DIV or MOD instructions.
915 	 *
916 	 * In restricted circumstances, the verifier can also use the AX
917 	 * register for rewrites as long as they do not interfere with
918 	 * the above cases!
919 	 */
920 	if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
921 		goto out;
922 
923 	if (from->imm == 0 &&
924 	    (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
925 	     from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
926 		*to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
927 		goto out;
928 	}
929 
930 	switch (from->code) {
931 	case BPF_ALU | BPF_ADD | BPF_K:
932 	case BPF_ALU | BPF_SUB | BPF_K:
933 	case BPF_ALU | BPF_AND | BPF_K:
934 	case BPF_ALU | BPF_OR  | BPF_K:
935 	case BPF_ALU | BPF_XOR | BPF_K:
936 	case BPF_ALU | BPF_MUL | BPF_K:
937 	case BPF_ALU | BPF_MOV | BPF_K:
938 	case BPF_ALU | BPF_DIV | BPF_K:
939 	case BPF_ALU | BPF_MOD | BPF_K:
940 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
941 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
942 		*to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
943 		break;
944 
945 	case BPF_ALU64 | BPF_ADD | BPF_K:
946 	case BPF_ALU64 | BPF_SUB | BPF_K:
947 	case BPF_ALU64 | BPF_AND | BPF_K:
948 	case BPF_ALU64 | BPF_OR  | BPF_K:
949 	case BPF_ALU64 | BPF_XOR | BPF_K:
950 	case BPF_ALU64 | BPF_MUL | BPF_K:
951 	case BPF_ALU64 | BPF_MOV | BPF_K:
952 	case BPF_ALU64 | BPF_DIV | BPF_K:
953 	case BPF_ALU64 | BPF_MOD | BPF_K:
954 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
955 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
956 		*to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
957 		break;
958 
959 	case BPF_JMP | BPF_JEQ  | BPF_K:
960 	case BPF_JMP | BPF_JNE  | BPF_K:
961 	case BPF_JMP | BPF_JGT  | BPF_K:
962 	case BPF_JMP | BPF_JLT  | BPF_K:
963 	case BPF_JMP | BPF_JGE  | BPF_K:
964 	case BPF_JMP | BPF_JLE  | BPF_K:
965 	case BPF_JMP | BPF_JSGT | BPF_K:
966 	case BPF_JMP | BPF_JSLT | BPF_K:
967 	case BPF_JMP | BPF_JSGE | BPF_K:
968 	case BPF_JMP | BPF_JSLE | BPF_K:
969 	case BPF_JMP | BPF_JSET | BPF_K:
970 		/* Accommodate for extra offset in case of a backjump. */
971 		off = from->off;
972 		if (off < 0)
973 			off -= 2;
974 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
975 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
976 		*to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
977 		break;
978 
979 	case BPF_JMP32 | BPF_JEQ  | BPF_K:
980 	case BPF_JMP32 | BPF_JNE  | BPF_K:
981 	case BPF_JMP32 | BPF_JGT  | BPF_K:
982 	case BPF_JMP32 | BPF_JLT  | BPF_K:
983 	case BPF_JMP32 | BPF_JGE  | BPF_K:
984 	case BPF_JMP32 | BPF_JLE  | BPF_K:
985 	case BPF_JMP32 | BPF_JSGT | BPF_K:
986 	case BPF_JMP32 | BPF_JSLT | BPF_K:
987 	case BPF_JMP32 | BPF_JSGE | BPF_K:
988 	case BPF_JMP32 | BPF_JSLE | BPF_K:
989 	case BPF_JMP32 | BPF_JSET | BPF_K:
990 		/* Accommodate for extra offset in case of a backjump. */
991 		off = from->off;
992 		if (off < 0)
993 			off -= 2;
994 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
995 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
996 		*to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
997 				      off);
998 		break;
999 
1000 	case BPF_LD | BPF_IMM | BPF_DW:
1001 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1002 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1003 		*to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1004 		*to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1005 		break;
1006 	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1007 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1008 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1009 		if (emit_zext)
1010 			*to++ = BPF_ZEXT_REG(BPF_REG_AX);
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 						clone->aux->verifier_zext);
1096 		if (!rewritten)
1097 			continue;
1098 
1099 		tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1100 		if (IS_ERR(tmp)) {
1101 			/* Patching may have repointed aux->prog during
1102 			 * realloc from the original one, so we need to
1103 			 * fix it up here on error.
1104 			 */
1105 			bpf_jit_prog_release_other(prog, clone);
1106 			return tmp;
1107 		}
1108 
1109 		clone = tmp;
1110 		insn_delta = rewritten - 1;
1111 
1112 		/* Walk new program and skip insns we just inserted. */
1113 		insn = clone->insnsi + i + insn_delta;
1114 		insn_cnt += insn_delta;
1115 		i        += insn_delta;
1116 	}
1117 
1118 	clone->blinded = 1;
1119 	return clone;
1120 }
1121 #endif /* CONFIG_BPF_JIT */
1122 
1123 /* Base function for offset calculation. Needs to go into .text section,
1124  * therefore keeping it non-static as well; will also be used by JITs
1125  * anyway later on, so do not let the compiler omit it. This also needs
1126  * to go into kallsyms for correlation from e.g. bpftool, so naming
1127  * must not change.
1128  */
1129 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1130 {
1131 	return 0;
1132 }
1133 EXPORT_SYMBOL_GPL(__bpf_call_base);
1134 
1135 /* All UAPI available opcodes. */
1136 #define BPF_INSN_MAP(INSN_2, INSN_3)		\
1137 	/* 32 bit ALU operations. */		\
1138 	/*   Register based. */			\
1139 	INSN_3(ALU, ADD,  X),			\
1140 	INSN_3(ALU, SUB,  X),			\
1141 	INSN_3(ALU, AND,  X),			\
1142 	INSN_3(ALU, OR,   X),			\
1143 	INSN_3(ALU, LSH,  X),			\
1144 	INSN_3(ALU, RSH,  X),			\
1145 	INSN_3(ALU, XOR,  X),			\
1146 	INSN_3(ALU, MUL,  X),			\
1147 	INSN_3(ALU, MOV,  X),			\
1148 	INSN_3(ALU, ARSH, X),			\
1149 	INSN_3(ALU, DIV,  X),			\
1150 	INSN_3(ALU, MOD,  X),			\
1151 	INSN_2(ALU, NEG),			\
1152 	INSN_3(ALU, END, TO_BE),		\
1153 	INSN_3(ALU, END, TO_LE),		\
1154 	/*   Immediate based. */		\
1155 	INSN_3(ALU, ADD,  K),			\
1156 	INSN_3(ALU, SUB,  K),			\
1157 	INSN_3(ALU, AND,  K),			\
1158 	INSN_3(ALU, OR,   K),			\
1159 	INSN_3(ALU, LSH,  K),			\
1160 	INSN_3(ALU, RSH,  K),			\
1161 	INSN_3(ALU, XOR,  K),			\
1162 	INSN_3(ALU, MUL,  K),			\
1163 	INSN_3(ALU, MOV,  K),			\
1164 	INSN_3(ALU, ARSH, K),			\
1165 	INSN_3(ALU, DIV,  K),			\
1166 	INSN_3(ALU, MOD,  K),			\
1167 	/* 64 bit ALU operations. */		\
1168 	/*   Register based. */			\
1169 	INSN_3(ALU64, ADD,  X),			\
1170 	INSN_3(ALU64, SUB,  X),			\
1171 	INSN_3(ALU64, AND,  X),			\
1172 	INSN_3(ALU64, OR,   X),			\
1173 	INSN_3(ALU64, LSH,  X),			\
1174 	INSN_3(ALU64, RSH,  X),			\
1175 	INSN_3(ALU64, XOR,  X),			\
1176 	INSN_3(ALU64, MUL,  X),			\
1177 	INSN_3(ALU64, MOV,  X),			\
1178 	INSN_3(ALU64, ARSH, X),			\
1179 	INSN_3(ALU64, DIV,  X),			\
1180 	INSN_3(ALU64, MOD,  X),			\
1181 	INSN_2(ALU64, NEG),			\
1182 	/*   Immediate based. */		\
1183 	INSN_3(ALU64, ADD,  K),			\
1184 	INSN_3(ALU64, SUB,  K),			\
1185 	INSN_3(ALU64, AND,  K),			\
1186 	INSN_3(ALU64, OR,   K),			\
1187 	INSN_3(ALU64, LSH,  K),			\
1188 	INSN_3(ALU64, RSH,  K),			\
1189 	INSN_3(ALU64, XOR,  K),			\
1190 	INSN_3(ALU64, MUL,  K),			\
1191 	INSN_3(ALU64, MOV,  K),			\
1192 	INSN_3(ALU64, ARSH, K),			\
1193 	INSN_3(ALU64, DIV,  K),			\
1194 	INSN_3(ALU64, MOD,  K),			\
1195 	/* Call instruction. */			\
1196 	INSN_2(JMP, CALL),			\
1197 	/* Exit instruction. */			\
1198 	INSN_2(JMP, EXIT),			\
1199 	/* 32-bit Jump instructions. */		\
1200 	/*   Register based. */			\
1201 	INSN_3(JMP32, JEQ,  X),			\
1202 	INSN_3(JMP32, JNE,  X),			\
1203 	INSN_3(JMP32, JGT,  X),			\
1204 	INSN_3(JMP32, JLT,  X),			\
1205 	INSN_3(JMP32, JGE,  X),			\
1206 	INSN_3(JMP32, JLE,  X),			\
1207 	INSN_3(JMP32, JSGT, X),			\
1208 	INSN_3(JMP32, JSLT, X),			\
1209 	INSN_3(JMP32, JSGE, X),			\
1210 	INSN_3(JMP32, JSLE, X),			\
1211 	INSN_3(JMP32, JSET, X),			\
1212 	/*   Immediate based. */		\
1213 	INSN_3(JMP32, JEQ,  K),			\
1214 	INSN_3(JMP32, JNE,  K),			\
1215 	INSN_3(JMP32, JGT,  K),			\
1216 	INSN_3(JMP32, JLT,  K),			\
1217 	INSN_3(JMP32, JGE,  K),			\
1218 	INSN_3(JMP32, JLE,  K),			\
1219 	INSN_3(JMP32, JSGT, K),			\
1220 	INSN_3(JMP32, JSLT, K),			\
1221 	INSN_3(JMP32, JSGE, K),			\
1222 	INSN_3(JMP32, JSLE, K),			\
1223 	INSN_3(JMP32, JSET, K),			\
1224 	/* Jump instructions. */		\
1225 	/*   Register based. */			\
1226 	INSN_3(JMP, JEQ,  X),			\
1227 	INSN_3(JMP, JNE,  X),			\
1228 	INSN_3(JMP, JGT,  X),			\
1229 	INSN_3(JMP, JLT,  X),			\
1230 	INSN_3(JMP, JGE,  X),			\
1231 	INSN_3(JMP, JLE,  X),			\
1232 	INSN_3(JMP, JSGT, X),			\
1233 	INSN_3(JMP, JSLT, X),			\
1234 	INSN_3(JMP, JSGE, X),			\
1235 	INSN_3(JMP, JSLE, X),			\
1236 	INSN_3(JMP, JSET, X),			\
1237 	/*   Immediate based. */		\
1238 	INSN_3(JMP, JEQ,  K),			\
1239 	INSN_3(JMP, JNE,  K),			\
1240 	INSN_3(JMP, JGT,  K),			\
1241 	INSN_3(JMP, JLT,  K),			\
1242 	INSN_3(JMP, JGE,  K),			\
1243 	INSN_3(JMP, JLE,  K),			\
1244 	INSN_3(JMP, JSGT, K),			\
1245 	INSN_3(JMP, JSLT, K),			\
1246 	INSN_3(JMP, JSGE, K),			\
1247 	INSN_3(JMP, JSLE, K),			\
1248 	INSN_3(JMP, JSET, K),			\
1249 	INSN_2(JMP, JA),			\
1250 	/* Store instructions. */		\
1251 	/*   Register based. */			\
1252 	INSN_3(STX, MEM,  B),			\
1253 	INSN_3(STX, MEM,  H),			\
1254 	INSN_3(STX, MEM,  W),			\
1255 	INSN_3(STX, MEM,  DW),			\
1256 	INSN_3(STX, XADD, W),			\
1257 	INSN_3(STX, XADD, DW),			\
1258 	/*   Immediate based. */		\
1259 	INSN_3(ST, MEM, B),			\
1260 	INSN_3(ST, MEM, H),			\
1261 	INSN_3(ST, MEM, W),			\
1262 	INSN_3(ST, MEM, DW),			\
1263 	/* Load instructions. */		\
1264 	/*   Register based. */			\
1265 	INSN_3(LDX, MEM, B),			\
1266 	INSN_3(LDX, MEM, H),			\
1267 	INSN_3(LDX, MEM, W),			\
1268 	INSN_3(LDX, MEM, DW),			\
1269 	/*   Immediate based. */		\
1270 	INSN_3(LD, IMM, DW)
1271 
1272 bool bpf_opcode_in_insntable(u8 code)
1273 {
1274 #define BPF_INSN_2_TBL(x, y)    [BPF_##x | BPF_##y] = true
1275 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1276 	static const bool public_insntable[256] = {
1277 		[0 ... 255] = false,
1278 		/* Now overwrite non-defaults ... */
1279 		BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1280 		/* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1281 		[BPF_LD | BPF_ABS | BPF_B] = true,
1282 		[BPF_LD | BPF_ABS | BPF_H] = true,
1283 		[BPF_LD | BPF_ABS | BPF_W] = true,
1284 		[BPF_LD | BPF_IND | BPF_B] = true,
1285 		[BPF_LD | BPF_IND | BPF_H] = true,
1286 		[BPF_LD | BPF_IND | BPF_W] = true,
1287 	};
1288 #undef BPF_INSN_3_TBL
1289 #undef BPF_INSN_2_TBL
1290 	return public_insntable[code];
1291 }
1292 
1293 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1294 /**
1295  *	__bpf_prog_run - run eBPF program on a given context
1296  *	@regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1297  *	@insn: is the array of eBPF instructions
1298  *	@stack: is the eBPF storage stack
1299  *
1300  * Decode and execute eBPF instructions.
1301  */
1302 static u64 __no_fgcse ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1303 {
1304 #define BPF_INSN_2_LBL(x, y)    [BPF_##x | BPF_##y] = &&x##_##y
1305 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1306 	static const void * const jumptable[256] __annotate_jump_table = {
1307 		[0 ... 255] = &&default_label,
1308 		/* Now overwrite non-defaults ... */
1309 		BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1310 		/* Non-UAPI available opcodes. */
1311 		[BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1312 		[BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1313 	};
1314 #undef BPF_INSN_3_LBL
1315 #undef BPF_INSN_2_LBL
1316 	u32 tail_call_cnt = 0;
1317 
1318 #define CONT	 ({ insn++; goto select_insn; })
1319 #define CONT_JMP ({ insn++; goto select_insn; })
1320 
1321 select_insn:
1322 	goto *jumptable[insn->code];
1323 
1324 	/* ALU */
1325 #define ALU(OPCODE, OP)			\
1326 	ALU64_##OPCODE##_X:		\
1327 		DST = DST OP SRC;	\
1328 		CONT;			\
1329 	ALU_##OPCODE##_X:		\
1330 		DST = (u32) DST OP (u32) SRC;	\
1331 		CONT;			\
1332 	ALU64_##OPCODE##_K:		\
1333 		DST = DST OP IMM;		\
1334 		CONT;			\
1335 	ALU_##OPCODE##_K:		\
1336 		DST = (u32) DST OP (u32) IMM;	\
1337 		CONT;
1338 
1339 	ALU(ADD,  +)
1340 	ALU(SUB,  -)
1341 	ALU(AND,  &)
1342 	ALU(OR,   |)
1343 	ALU(LSH, <<)
1344 	ALU(RSH, >>)
1345 	ALU(XOR,  ^)
1346 	ALU(MUL,  *)
1347 #undef ALU
1348 	ALU_NEG:
1349 		DST = (u32) -DST;
1350 		CONT;
1351 	ALU64_NEG:
1352 		DST = -DST;
1353 		CONT;
1354 	ALU_MOV_X:
1355 		DST = (u32) SRC;
1356 		CONT;
1357 	ALU_MOV_K:
1358 		DST = (u32) IMM;
1359 		CONT;
1360 	ALU64_MOV_X:
1361 		DST = SRC;
1362 		CONT;
1363 	ALU64_MOV_K:
1364 		DST = IMM;
1365 		CONT;
1366 	LD_IMM_DW:
1367 		DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1368 		insn++;
1369 		CONT;
1370 	ALU_ARSH_X:
1371 		DST = (u64) (u32) (((s32) DST) >> SRC);
1372 		CONT;
1373 	ALU_ARSH_K:
1374 		DST = (u64) (u32) (((s32) DST) >> IMM);
1375 		CONT;
1376 	ALU64_ARSH_X:
1377 		(*(s64 *) &DST) >>= SRC;
1378 		CONT;
1379 	ALU64_ARSH_K:
1380 		(*(s64 *) &DST) >>= IMM;
1381 		CONT;
1382 	ALU64_MOD_X:
1383 		div64_u64_rem(DST, SRC, &AX);
1384 		DST = AX;
1385 		CONT;
1386 	ALU_MOD_X:
1387 		AX = (u32) DST;
1388 		DST = do_div(AX, (u32) SRC);
1389 		CONT;
1390 	ALU64_MOD_K:
1391 		div64_u64_rem(DST, IMM, &AX);
1392 		DST = AX;
1393 		CONT;
1394 	ALU_MOD_K:
1395 		AX = (u32) DST;
1396 		DST = do_div(AX, (u32) IMM);
1397 		CONT;
1398 	ALU64_DIV_X:
1399 		DST = div64_u64(DST, SRC);
1400 		CONT;
1401 	ALU_DIV_X:
1402 		AX = (u32) DST;
1403 		do_div(AX, (u32) SRC);
1404 		DST = (u32) AX;
1405 		CONT;
1406 	ALU64_DIV_K:
1407 		DST = div64_u64(DST, IMM);
1408 		CONT;
1409 	ALU_DIV_K:
1410 		AX = (u32) DST;
1411 		do_div(AX, (u32) IMM);
1412 		DST = (u32) AX;
1413 		CONT;
1414 	ALU_END_TO_BE:
1415 		switch (IMM) {
1416 		case 16:
1417 			DST = (__force u16) cpu_to_be16(DST);
1418 			break;
1419 		case 32:
1420 			DST = (__force u32) cpu_to_be32(DST);
1421 			break;
1422 		case 64:
1423 			DST = (__force u64) cpu_to_be64(DST);
1424 			break;
1425 		}
1426 		CONT;
1427 	ALU_END_TO_LE:
1428 		switch (IMM) {
1429 		case 16:
1430 			DST = (__force u16) cpu_to_le16(DST);
1431 			break;
1432 		case 32:
1433 			DST = (__force u32) cpu_to_le32(DST);
1434 			break;
1435 		case 64:
1436 			DST = (__force u64) cpu_to_le64(DST);
1437 			break;
1438 		}
1439 		CONT;
1440 
1441 	/* CALL */
1442 	JMP_CALL:
1443 		/* Function call scratches BPF_R1-BPF_R5 registers,
1444 		 * preserves BPF_R6-BPF_R9, and stores return value
1445 		 * into BPF_R0.
1446 		 */
1447 		BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1448 						       BPF_R4, BPF_R5);
1449 		CONT;
1450 
1451 	JMP_CALL_ARGS:
1452 		BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1453 							    BPF_R3, BPF_R4,
1454 							    BPF_R5,
1455 							    insn + insn->off + 1);
1456 		CONT;
1457 
1458 	JMP_TAIL_CALL: {
1459 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1460 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1461 		struct bpf_prog *prog;
1462 		u32 index = BPF_R3;
1463 
1464 		if (unlikely(index >= array->map.max_entries))
1465 			goto out;
1466 		if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1467 			goto out;
1468 
1469 		tail_call_cnt++;
1470 
1471 		prog = READ_ONCE(array->ptrs[index]);
1472 		if (!prog)
1473 			goto out;
1474 
1475 		/* ARG1 at this point is guaranteed to point to CTX from
1476 		 * the verifier side due to the fact that the tail call is
1477 		 * handeled like a helper, that is, bpf_tail_call_proto,
1478 		 * where arg1_type is ARG_PTR_TO_CTX.
1479 		 */
1480 		insn = prog->insnsi;
1481 		goto select_insn;
1482 out:
1483 		CONT;
1484 	}
1485 	JMP_JA:
1486 		insn += insn->off;
1487 		CONT;
1488 	JMP_EXIT:
1489 		return BPF_R0;
1490 	/* JMP */
1491 #define COND_JMP(SIGN, OPCODE, CMP_OP)				\
1492 	JMP_##OPCODE##_X:					\
1493 		if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) {	\
1494 			insn += insn->off;			\
1495 			CONT_JMP;				\
1496 		}						\
1497 		CONT;						\
1498 	JMP32_##OPCODE##_X:					\
1499 		if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) {	\
1500 			insn += insn->off;			\
1501 			CONT_JMP;				\
1502 		}						\
1503 		CONT;						\
1504 	JMP_##OPCODE##_K:					\
1505 		if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) {	\
1506 			insn += insn->off;			\
1507 			CONT_JMP;				\
1508 		}						\
1509 		CONT;						\
1510 	JMP32_##OPCODE##_K:					\
1511 		if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) {	\
1512 			insn += insn->off;			\
1513 			CONT_JMP;				\
1514 		}						\
1515 		CONT;
1516 	COND_JMP(u, JEQ, ==)
1517 	COND_JMP(u, JNE, !=)
1518 	COND_JMP(u, JGT, >)
1519 	COND_JMP(u, JLT, <)
1520 	COND_JMP(u, JGE, >=)
1521 	COND_JMP(u, JLE, <=)
1522 	COND_JMP(u, JSET, &)
1523 	COND_JMP(s, JSGT, >)
1524 	COND_JMP(s, JSLT, <)
1525 	COND_JMP(s, JSGE, >=)
1526 	COND_JMP(s, JSLE, <=)
1527 #undef COND_JMP
1528 	/* STX and ST and LDX*/
1529 #define LDST(SIZEOP, SIZE)						\
1530 	STX_MEM_##SIZEOP:						\
1531 		*(SIZE *)(unsigned long) (DST + insn->off) = SRC;	\
1532 		CONT;							\
1533 	ST_MEM_##SIZEOP:						\
1534 		*(SIZE *)(unsigned long) (DST + insn->off) = IMM;	\
1535 		CONT;							\
1536 	LDX_MEM_##SIZEOP:						\
1537 		DST = *(SIZE *)(unsigned long) (SRC + insn->off);	\
1538 		CONT;
1539 
1540 	LDST(B,   u8)
1541 	LDST(H,  u16)
1542 	LDST(W,  u32)
1543 	LDST(DW, u64)
1544 #undef LDST
1545 	STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1546 		atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1547 			   (DST + insn->off));
1548 		CONT;
1549 	STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1550 		atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1551 			     (DST + insn->off));
1552 		CONT;
1553 
1554 	default_label:
1555 		/* If we ever reach this, we have a bug somewhere. Die hard here
1556 		 * instead of just returning 0; we could be somewhere in a subprog,
1557 		 * so execution could continue otherwise which we do /not/ want.
1558 		 *
1559 		 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1560 		 */
1561 		pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1562 		BUG_ON(1);
1563 		return 0;
1564 }
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 *progs)
1798 {
1799 	if (!progs || progs == &empty_prog_array.hdr)
1800 		return;
1801 	kfree_rcu(progs, rcu);
1802 }
1803 
1804 int bpf_prog_array_length(struct bpf_prog_array *array)
1805 {
1806 	struct bpf_prog_array_item *item;
1807 	u32 cnt = 0;
1808 
1809 	for (item = array->items; item->prog; item++)
1810 		if (item->prog != &dummy_bpf_prog.prog)
1811 			cnt++;
1812 	return cnt;
1813 }
1814 
1815 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
1816 {
1817 	struct bpf_prog_array_item *item;
1818 
1819 	for (item = array->items; item->prog; item++)
1820 		if (item->prog != &dummy_bpf_prog.prog)
1821 			return false;
1822 	return true;
1823 }
1824 
1825 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
1826 				     u32 *prog_ids,
1827 				     u32 request_cnt)
1828 {
1829 	struct bpf_prog_array_item *item;
1830 	int i = 0;
1831 
1832 	for (item = array->items; item->prog; item++) {
1833 		if (item->prog == &dummy_bpf_prog.prog)
1834 			continue;
1835 		prog_ids[i] = item->prog->aux->id;
1836 		if (++i == request_cnt) {
1837 			item++;
1838 			break;
1839 		}
1840 	}
1841 
1842 	return !!(item->prog);
1843 }
1844 
1845 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
1846 				__u32 __user *prog_ids, u32 cnt)
1847 {
1848 	unsigned long err = 0;
1849 	bool nospc;
1850 	u32 *ids;
1851 
1852 	/* users of this function are doing:
1853 	 * cnt = bpf_prog_array_length();
1854 	 * if (cnt > 0)
1855 	 *     bpf_prog_array_copy_to_user(..., cnt);
1856 	 * so below kcalloc doesn't need extra cnt > 0 check.
1857 	 */
1858 	ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1859 	if (!ids)
1860 		return -ENOMEM;
1861 	nospc = bpf_prog_array_copy_core(array, ids, cnt);
1862 	err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1863 	kfree(ids);
1864 	if (err)
1865 		return -EFAULT;
1866 	if (nospc)
1867 		return -ENOSPC;
1868 	return 0;
1869 }
1870 
1871 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
1872 				struct bpf_prog *old_prog)
1873 {
1874 	struct bpf_prog_array_item *item;
1875 
1876 	for (item = array->items; item->prog; item++)
1877 		if (item->prog == old_prog) {
1878 			WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1879 			break;
1880 		}
1881 }
1882 
1883 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
1884 			struct bpf_prog *exclude_prog,
1885 			struct bpf_prog *include_prog,
1886 			struct bpf_prog_array **new_array)
1887 {
1888 	int new_prog_cnt, carry_prog_cnt = 0;
1889 	struct bpf_prog_array_item *existing;
1890 	struct bpf_prog_array *array;
1891 	bool found_exclude = false;
1892 	int new_prog_idx = 0;
1893 
1894 	/* Figure out how many existing progs we need to carry over to
1895 	 * the new array.
1896 	 */
1897 	if (old_array) {
1898 		existing = old_array->items;
1899 		for (; existing->prog; existing++) {
1900 			if (existing->prog == exclude_prog) {
1901 				found_exclude = true;
1902 				continue;
1903 			}
1904 			if (existing->prog != &dummy_bpf_prog.prog)
1905 				carry_prog_cnt++;
1906 			if (existing->prog == include_prog)
1907 				return -EEXIST;
1908 		}
1909 	}
1910 
1911 	if (exclude_prog && !found_exclude)
1912 		return -ENOENT;
1913 
1914 	/* How many progs (not NULL) will be in the new array? */
1915 	new_prog_cnt = carry_prog_cnt;
1916 	if (include_prog)
1917 		new_prog_cnt += 1;
1918 
1919 	/* Do we have any prog (not NULL) in the new array? */
1920 	if (!new_prog_cnt) {
1921 		*new_array = NULL;
1922 		return 0;
1923 	}
1924 
1925 	/* +1 as the end of prog_array is marked with NULL */
1926 	array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
1927 	if (!array)
1928 		return -ENOMEM;
1929 
1930 	/* Fill in the new prog array */
1931 	if (carry_prog_cnt) {
1932 		existing = old_array->items;
1933 		for (; existing->prog; existing++)
1934 			if (existing->prog != exclude_prog &&
1935 			    existing->prog != &dummy_bpf_prog.prog) {
1936 				array->items[new_prog_idx++].prog =
1937 					existing->prog;
1938 			}
1939 	}
1940 	if (include_prog)
1941 		array->items[new_prog_idx++].prog = include_prog;
1942 	array->items[new_prog_idx].prog = NULL;
1943 	*new_array = array;
1944 	return 0;
1945 }
1946 
1947 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
1948 			     u32 *prog_ids, u32 request_cnt,
1949 			     u32 *prog_cnt)
1950 {
1951 	u32 cnt = 0;
1952 
1953 	if (array)
1954 		cnt = bpf_prog_array_length(array);
1955 
1956 	*prog_cnt = cnt;
1957 
1958 	/* return early if user requested only program count or nothing to copy */
1959 	if (!request_cnt || !cnt)
1960 		return 0;
1961 
1962 	/* this function is called under trace/bpf_trace.c: bpf_event_mutex */
1963 	return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
1964 								     : 0;
1965 }
1966 
1967 static void bpf_prog_free_deferred(struct work_struct *work)
1968 {
1969 	struct bpf_prog_aux *aux;
1970 	int i;
1971 
1972 	aux = container_of(work, struct bpf_prog_aux, work);
1973 	if (bpf_prog_is_dev_bound(aux))
1974 		bpf_prog_offload_destroy(aux->prog);
1975 #ifdef CONFIG_PERF_EVENTS
1976 	if (aux->prog->has_callchain_buf)
1977 		put_callchain_buffers();
1978 #endif
1979 	for (i = 0; i < aux->func_cnt; i++)
1980 		bpf_jit_free(aux->func[i]);
1981 	if (aux->func_cnt) {
1982 		kfree(aux->func);
1983 		bpf_prog_unlock_free(aux->prog);
1984 	} else {
1985 		bpf_jit_free(aux->prog);
1986 	}
1987 }
1988 
1989 /* Free internal BPF program */
1990 void bpf_prog_free(struct bpf_prog *fp)
1991 {
1992 	struct bpf_prog_aux *aux = fp->aux;
1993 
1994 	INIT_WORK(&aux->work, bpf_prog_free_deferred);
1995 	schedule_work(&aux->work);
1996 }
1997 EXPORT_SYMBOL_GPL(bpf_prog_free);
1998 
1999 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2000 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2001 
2002 void bpf_user_rnd_init_once(void)
2003 {
2004 	prandom_init_once(&bpf_user_rnd_state);
2005 }
2006 
2007 BPF_CALL_0(bpf_user_rnd_u32)
2008 {
2009 	/* Should someone ever have the rather unwise idea to use some
2010 	 * of the registers passed into this function, then note that
2011 	 * this function is called from native eBPF and classic-to-eBPF
2012 	 * transformations. Register assignments from both sides are
2013 	 * different, f.e. classic always sets fn(ctx, A, X) here.
2014 	 */
2015 	struct rnd_state *state;
2016 	u32 res;
2017 
2018 	state = &get_cpu_var(bpf_user_rnd_state);
2019 	res = prandom_u32_state(state);
2020 	put_cpu_var(bpf_user_rnd_state);
2021 
2022 	return res;
2023 }
2024 
2025 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2026 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2027 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2028 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2029 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2030 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2031 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2032 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2033 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2034 
2035 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2036 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2037 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2038 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2039 
2040 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2041 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2042 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2043 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2044 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2045 
2046 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2047 {
2048 	return NULL;
2049 }
2050 
2051 u64 __weak
2052 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2053 		 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2054 {
2055 	return -ENOTSUPP;
2056 }
2057 EXPORT_SYMBOL_GPL(bpf_event_output);
2058 
2059 /* Always built-in helper functions. */
2060 const struct bpf_func_proto bpf_tail_call_proto = {
2061 	.func		= NULL,
2062 	.gpl_only	= false,
2063 	.ret_type	= RET_VOID,
2064 	.arg1_type	= ARG_PTR_TO_CTX,
2065 	.arg2_type	= ARG_CONST_MAP_PTR,
2066 	.arg3_type	= ARG_ANYTHING,
2067 };
2068 
2069 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2070  * It is encouraged to implement bpf_int_jit_compile() instead, so that
2071  * eBPF and implicitly also cBPF can get JITed!
2072  */
2073 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2074 {
2075 	return prog;
2076 }
2077 
2078 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2079  * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2080  */
2081 void __weak bpf_jit_compile(struct bpf_prog *prog)
2082 {
2083 }
2084 
2085 bool __weak bpf_helper_changes_pkt_data(void *func)
2086 {
2087 	return false;
2088 }
2089 
2090 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2091  * analysis code and wants explicit zero extension inserted by verifier.
2092  * Otherwise, return FALSE.
2093  */
2094 bool __weak bpf_jit_needs_zext(void)
2095 {
2096 	return false;
2097 }
2098 
2099 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2100  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2101  */
2102 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2103 			 int len)
2104 {
2105 	return -EFAULT;
2106 }
2107 
2108 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2109 EXPORT_SYMBOL(bpf_stats_enabled_key);
2110 
2111 /* All definitions of tracepoints related to BPF. */
2112 #define CREATE_TRACE_POINTS
2113 #include <linux/bpf_trace.h>
2114 
2115 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2116 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
2117