xref: /openbmc/linux/kernel/bpf/core.c (revision ca637c0e)
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/objtool.h>
29 #include <linux/rbtree_latch.h>
30 #include <linux/kallsyms.h>
31 #include <linux/rcupdate.h>
32 #include <linux/perf_event.h>
33 #include <linux/extable.h>
34 #include <linux/log2.h>
35 #include <linux/bpf_verifier.h>
36 #include <linux/nodemask.h>
37 
38 #include <asm/barrier.h>
39 #include <asm/unaligned.h>
40 
41 /* Registers */
42 #define BPF_R0	regs[BPF_REG_0]
43 #define BPF_R1	regs[BPF_REG_1]
44 #define BPF_R2	regs[BPF_REG_2]
45 #define BPF_R3	regs[BPF_REG_3]
46 #define BPF_R4	regs[BPF_REG_4]
47 #define BPF_R5	regs[BPF_REG_5]
48 #define BPF_R6	regs[BPF_REG_6]
49 #define BPF_R7	regs[BPF_REG_7]
50 #define BPF_R8	regs[BPF_REG_8]
51 #define BPF_R9	regs[BPF_REG_9]
52 #define BPF_R10	regs[BPF_REG_10]
53 
54 /* Named registers */
55 #define DST	regs[insn->dst_reg]
56 #define SRC	regs[insn->src_reg]
57 #define FP	regs[BPF_REG_FP]
58 #define AX	regs[BPF_REG_AX]
59 #define ARG1	regs[BPF_REG_ARG1]
60 #define CTX	regs[BPF_REG_CTX]
61 #define IMM	insn->imm
62 
63 /* No hurry in this branch
64  *
65  * Exported for the bpf jit load helper.
66  */
67 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
68 {
69 	u8 *ptr = NULL;
70 
71 	if (k >= SKF_NET_OFF) {
72 		ptr = skb_network_header(skb) + k - SKF_NET_OFF;
73 	} else if (k >= SKF_LL_OFF) {
74 		if (unlikely(!skb_mac_header_was_set(skb)))
75 			return NULL;
76 		ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
77 	}
78 	if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
79 		return ptr;
80 
81 	return NULL;
82 }
83 
84 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
85 {
86 	gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
87 	struct bpf_prog_aux *aux;
88 	struct bpf_prog *fp;
89 
90 	size = round_up(size, PAGE_SIZE);
91 	fp = __vmalloc(size, gfp_flags);
92 	if (fp == NULL)
93 		return NULL;
94 
95 	aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT | gfp_extra_flags);
96 	if (aux == NULL) {
97 		vfree(fp);
98 		return NULL;
99 	}
100 	fp->active = alloc_percpu_gfp(int, GFP_KERNEL_ACCOUNT | gfp_extra_flags);
101 	if (!fp->active) {
102 		vfree(fp);
103 		kfree(aux);
104 		return NULL;
105 	}
106 
107 	fp->pages = size / PAGE_SIZE;
108 	fp->aux = aux;
109 	fp->aux->prog = fp;
110 	fp->jit_requested = ebpf_jit_enabled();
111 	fp->blinding_requested = bpf_jit_blinding_enabled(fp);
112 #ifdef CONFIG_CGROUP_BPF
113 	aux->cgroup_atype = CGROUP_BPF_ATTACH_TYPE_INVALID;
114 #endif
115 
116 	INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
117 	mutex_init(&fp->aux->used_maps_mutex);
118 	mutex_init(&fp->aux->dst_mutex);
119 
120 	return fp;
121 }
122 
123 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
124 {
125 	gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
126 	struct bpf_prog *prog;
127 	int cpu;
128 
129 	prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
130 	if (!prog)
131 		return NULL;
132 
133 	prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
134 	if (!prog->stats) {
135 		free_percpu(prog->active);
136 		kfree(prog->aux);
137 		vfree(prog);
138 		return NULL;
139 	}
140 
141 	for_each_possible_cpu(cpu) {
142 		struct bpf_prog_stats *pstats;
143 
144 		pstats = per_cpu_ptr(prog->stats, cpu);
145 		u64_stats_init(&pstats->syncp);
146 	}
147 	return prog;
148 }
149 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
150 
151 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
152 {
153 	if (!prog->aux->nr_linfo || !prog->jit_requested)
154 		return 0;
155 
156 	prog->aux->jited_linfo = kvcalloc(prog->aux->nr_linfo,
157 					  sizeof(*prog->aux->jited_linfo),
158 					  GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
159 	if (!prog->aux->jited_linfo)
160 		return -ENOMEM;
161 
162 	return 0;
163 }
164 
165 void bpf_prog_jit_attempt_done(struct bpf_prog *prog)
166 {
167 	if (prog->aux->jited_linfo &&
168 	    (!prog->jited || !prog->aux->jited_linfo[0])) {
169 		kvfree(prog->aux->jited_linfo);
170 		prog->aux->jited_linfo = NULL;
171 	}
172 
173 	kfree(prog->aux->kfunc_tab);
174 	prog->aux->kfunc_tab = NULL;
175 }
176 
177 /* The jit engine is responsible to provide an array
178  * for insn_off to the jited_off mapping (insn_to_jit_off).
179  *
180  * The idx to this array is the insn_off.  Hence, the insn_off
181  * here is relative to the prog itself instead of the main prog.
182  * This array has one entry for each xlated bpf insn.
183  *
184  * jited_off is the byte off to the end of the jited insn.
185  *
186  * Hence, with
187  * insn_start:
188  *      The first bpf insn off of the prog.  The insn off
189  *      here is relative to the main prog.
190  *      e.g. if prog is a subprog, insn_start > 0
191  * linfo_idx:
192  *      The prog's idx to prog->aux->linfo and jited_linfo
193  *
194  * jited_linfo[linfo_idx] = prog->bpf_func
195  *
196  * For i > linfo_idx,
197  *
198  * jited_linfo[i] = prog->bpf_func +
199  *	insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
200  */
201 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
202 			       const u32 *insn_to_jit_off)
203 {
204 	u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
205 	const struct bpf_line_info *linfo;
206 	void **jited_linfo;
207 
208 	if (!prog->aux->jited_linfo)
209 		/* Userspace did not provide linfo */
210 		return;
211 
212 	linfo_idx = prog->aux->linfo_idx;
213 	linfo = &prog->aux->linfo[linfo_idx];
214 	insn_start = linfo[0].insn_off;
215 	insn_end = insn_start + prog->len;
216 
217 	jited_linfo = &prog->aux->jited_linfo[linfo_idx];
218 	jited_linfo[0] = prog->bpf_func;
219 
220 	nr_linfo = prog->aux->nr_linfo - linfo_idx;
221 
222 	for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
223 		/* The verifier ensures that linfo[i].insn_off is
224 		 * strictly increasing
225 		 */
226 		jited_linfo[i] = prog->bpf_func +
227 			insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
228 }
229 
230 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
231 				  gfp_t gfp_extra_flags)
232 {
233 	gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
234 	struct bpf_prog *fp;
235 	u32 pages;
236 
237 	size = round_up(size, PAGE_SIZE);
238 	pages = size / PAGE_SIZE;
239 	if (pages <= fp_old->pages)
240 		return fp_old;
241 
242 	fp = __vmalloc(size, gfp_flags);
243 	if (fp) {
244 		memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
245 		fp->pages = pages;
246 		fp->aux->prog = fp;
247 
248 		/* We keep fp->aux from fp_old around in the new
249 		 * reallocated structure.
250 		 */
251 		fp_old->aux = NULL;
252 		fp_old->stats = NULL;
253 		fp_old->active = NULL;
254 		__bpf_prog_free(fp_old);
255 	}
256 
257 	return fp;
258 }
259 
260 void __bpf_prog_free(struct bpf_prog *fp)
261 {
262 	if (fp->aux) {
263 		mutex_destroy(&fp->aux->used_maps_mutex);
264 		mutex_destroy(&fp->aux->dst_mutex);
265 		kfree(fp->aux->poke_tab);
266 		kfree(fp->aux);
267 	}
268 	free_percpu(fp->stats);
269 	free_percpu(fp->active);
270 	vfree(fp);
271 }
272 
273 int bpf_prog_calc_tag(struct bpf_prog *fp)
274 {
275 	const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
276 	u32 raw_size = bpf_prog_tag_scratch_size(fp);
277 	u32 digest[SHA1_DIGEST_WORDS];
278 	u32 ws[SHA1_WORKSPACE_WORDS];
279 	u32 i, bsize, psize, blocks;
280 	struct bpf_insn *dst;
281 	bool was_ld_map;
282 	u8 *raw, *todo;
283 	__be32 *result;
284 	__be64 *bits;
285 
286 	raw = vmalloc(raw_size);
287 	if (!raw)
288 		return -ENOMEM;
289 
290 	sha1_init(digest);
291 	memset(ws, 0, sizeof(ws));
292 
293 	/* We need to take out the map fd for the digest calculation
294 	 * since they are unstable from user space side.
295 	 */
296 	dst = (void *)raw;
297 	for (i = 0, was_ld_map = false; i < fp->len; i++) {
298 		dst[i] = fp->insnsi[i];
299 		if (!was_ld_map &&
300 		    dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
301 		    (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
302 		     dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
303 			was_ld_map = true;
304 			dst[i].imm = 0;
305 		} else if (was_ld_map &&
306 			   dst[i].code == 0 &&
307 			   dst[i].dst_reg == 0 &&
308 			   dst[i].src_reg == 0 &&
309 			   dst[i].off == 0) {
310 			was_ld_map = false;
311 			dst[i].imm = 0;
312 		} else {
313 			was_ld_map = false;
314 		}
315 	}
316 
317 	psize = bpf_prog_insn_size(fp);
318 	memset(&raw[psize], 0, raw_size - psize);
319 	raw[psize++] = 0x80;
320 
321 	bsize  = round_up(psize, SHA1_BLOCK_SIZE);
322 	blocks = bsize / SHA1_BLOCK_SIZE;
323 	todo   = raw;
324 	if (bsize - psize >= sizeof(__be64)) {
325 		bits = (__be64 *)(todo + bsize - sizeof(__be64));
326 	} else {
327 		bits = (__be64 *)(todo + bsize + bits_offset);
328 		blocks++;
329 	}
330 	*bits = cpu_to_be64((psize - 1) << 3);
331 
332 	while (blocks--) {
333 		sha1_transform(digest, todo, ws);
334 		todo += SHA1_BLOCK_SIZE;
335 	}
336 
337 	result = (__force __be32 *)digest;
338 	for (i = 0; i < SHA1_DIGEST_WORDS; i++)
339 		result[i] = cpu_to_be32(digest[i]);
340 	memcpy(fp->tag, result, sizeof(fp->tag));
341 
342 	vfree(raw);
343 	return 0;
344 }
345 
346 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
347 				s32 end_new, s32 curr, const bool probe_pass)
348 {
349 	const s64 imm_min = S32_MIN, imm_max = S32_MAX;
350 	s32 delta = end_new - end_old;
351 	s64 imm = insn->imm;
352 
353 	if (curr < pos && curr + imm + 1 >= end_old)
354 		imm += delta;
355 	else if (curr >= end_new && curr + imm + 1 < end_new)
356 		imm -= delta;
357 	if (imm < imm_min || imm > imm_max)
358 		return -ERANGE;
359 	if (!probe_pass)
360 		insn->imm = imm;
361 	return 0;
362 }
363 
364 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
365 				s32 end_new, s32 curr, const bool probe_pass)
366 {
367 	const s32 off_min = S16_MIN, off_max = S16_MAX;
368 	s32 delta = end_new - end_old;
369 	s32 off = insn->off;
370 
371 	if (curr < pos && curr + off + 1 >= end_old)
372 		off += delta;
373 	else if (curr >= end_new && curr + off + 1 < end_new)
374 		off -= delta;
375 	if (off < off_min || off > off_max)
376 		return -ERANGE;
377 	if (!probe_pass)
378 		insn->off = off;
379 	return 0;
380 }
381 
382 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
383 			    s32 end_new, const bool probe_pass)
384 {
385 	u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
386 	struct bpf_insn *insn = prog->insnsi;
387 	int ret = 0;
388 
389 	for (i = 0; i < insn_cnt; i++, insn++) {
390 		u8 code;
391 
392 		/* In the probing pass we still operate on the original,
393 		 * unpatched image in order to check overflows before we
394 		 * do any other adjustments. Therefore skip the patchlet.
395 		 */
396 		if (probe_pass && i == pos) {
397 			i = end_new;
398 			insn = prog->insnsi + end_old;
399 		}
400 		if (bpf_pseudo_func(insn)) {
401 			ret = bpf_adj_delta_to_imm(insn, pos, end_old,
402 						   end_new, i, probe_pass);
403 			if (ret)
404 				return ret;
405 			continue;
406 		}
407 		code = insn->code;
408 		if ((BPF_CLASS(code) != BPF_JMP &&
409 		     BPF_CLASS(code) != BPF_JMP32) ||
410 		    BPF_OP(code) == BPF_EXIT)
411 			continue;
412 		/* Adjust offset of jmps if we cross patch boundaries. */
413 		if (BPF_OP(code) == BPF_CALL) {
414 			if (insn->src_reg != BPF_PSEUDO_CALL)
415 				continue;
416 			ret = bpf_adj_delta_to_imm(insn, pos, end_old,
417 						   end_new, i, probe_pass);
418 		} else {
419 			ret = bpf_adj_delta_to_off(insn, pos, end_old,
420 						   end_new, i, probe_pass);
421 		}
422 		if (ret)
423 			break;
424 	}
425 
426 	return ret;
427 }
428 
429 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
430 {
431 	struct bpf_line_info *linfo;
432 	u32 i, nr_linfo;
433 
434 	nr_linfo = prog->aux->nr_linfo;
435 	if (!nr_linfo || !delta)
436 		return;
437 
438 	linfo = prog->aux->linfo;
439 
440 	for (i = 0; i < nr_linfo; i++)
441 		if (off < linfo[i].insn_off)
442 			break;
443 
444 	/* Push all off < linfo[i].insn_off by delta */
445 	for (; i < nr_linfo; i++)
446 		linfo[i].insn_off += delta;
447 }
448 
449 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
450 				       const struct bpf_insn *patch, u32 len)
451 {
452 	u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
453 	const u32 cnt_max = S16_MAX;
454 	struct bpf_prog *prog_adj;
455 	int err;
456 
457 	/* Since our patchlet doesn't expand the image, we're done. */
458 	if (insn_delta == 0) {
459 		memcpy(prog->insnsi + off, patch, sizeof(*patch));
460 		return prog;
461 	}
462 
463 	insn_adj_cnt = prog->len + insn_delta;
464 
465 	/* Reject anything that would potentially let the insn->off
466 	 * target overflow when we have excessive program expansions.
467 	 * We need to probe here before we do any reallocation where
468 	 * we afterwards may not fail anymore.
469 	 */
470 	if (insn_adj_cnt > cnt_max &&
471 	    (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
472 		return ERR_PTR(err);
473 
474 	/* Several new instructions need to be inserted. Make room
475 	 * for them. Likely, there's no need for a new allocation as
476 	 * last page could have large enough tailroom.
477 	 */
478 	prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
479 				    GFP_USER);
480 	if (!prog_adj)
481 		return ERR_PTR(-ENOMEM);
482 
483 	prog_adj->len = insn_adj_cnt;
484 
485 	/* Patching happens in 3 steps:
486 	 *
487 	 * 1) Move over tail of insnsi from next instruction onwards,
488 	 *    so we can patch the single target insn with one or more
489 	 *    new ones (patching is always from 1 to n insns, n > 0).
490 	 * 2) Inject new instructions at the target location.
491 	 * 3) Adjust branch offsets if necessary.
492 	 */
493 	insn_rest = insn_adj_cnt - off - len;
494 
495 	memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
496 		sizeof(*patch) * insn_rest);
497 	memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
498 
499 	/* We are guaranteed to not fail at this point, otherwise
500 	 * the ship has sailed to reverse to the original state. An
501 	 * overflow cannot happen at this point.
502 	 */
503 	BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
504 
505 	bpf_adj_linfo(prog_adj, off, insn_delta);
506 
507 	return prog_adj;
508 }
509 
510 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
511 {
512 	/* Branch offsets can't overflow when program is shrinking, no need
513 	 * to call bpf_adj_branches(..., true) here
514 	 */
515 	memmove(prog->insnsi + off, prog->insnsi + off + cnt,
516 		sizeof(struct bpf_insn) * (prog->len - off - cnt));
517 	prog->len -= cnt;
518 
519 	return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
520 }
521 
522 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
523 {
524 	int i;
525 
526 	for (i = 0; i < fp->aux->func_cnt; i++)
527 		bpf_prog_kallsyms_del(fp->aux->func[i]);
528 }
529 
530 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
531 {
532 	bpf_prog_kallsyms_del_subprogs(fp);
533 	bpf_prog_kallsyms_del(fp);
534 }
535 
536 #ifdef CONFIG_BPF_JIT
537 /* All BPF JIT sysctl knobs here. */
538 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
539 int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
540 int bpf_jit_harden   __read_mostly;
541 long bpf_jit_limit   __read_mostly;
542 long bpf_jit_limit_max __read_mostly;
543 
544 static void
545 bpf_prog_ksym_set_addr(struct bpf_prog *prog)
546 {
547 	WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
548 
549 	prog->aux->ksym.start = (unsigned long) prog->bpf_func;
550 	prog->aux->ksym.end   = prog->aux->ksym.start + prog->jited_len;
551 }
552 
553 static void
554 bpf_prog_ksym_set_name(struct bpf_prog *prog)
555 {
556 	char *sym = prog->aux->ksym.name;
557 	const char *end = sym + KSYM_NAME_LEN;
558 	const struct btf_type *type;
559 	const char *func_name;
560 
561 	BUILD_BUG_ON(sizeof("bpf_prog_") +
562 		     sizeof(prog->tag) * 2 +
563 		     /* name has been null terminated.
564 		      * We should need +1 for the '_' preceding
565 		      * the name.  However, the null character
566 		      * is double counted between the name and the
567 		      * sizeof("bpf_prog_") above, so we omit
568 		      * the +1 here.
569 		      */
570 		     sizeof(prog->aux->name) > KSYM_NAME_LEN);
571 
572 	sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
573 	sym  = bin2hex(sym, prog->tag, sizeof(prog->tag));
574 
575 	/* prog->aux->name will be ignored if full btf name is available */
576 	if (prog->aux->func_info_cnt) {
577 		type = btf_type_by_id(prog->aux->btf,
578 				      prog->aux->func_info[prog->aux->func_idx].type_id);
579 		func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
580 		snprintf(sym, (size_t)(end - sym), "_%s", func_name);
581 		return;
582 	}
583 
584 	if (prog->aux->name[0])
585 		snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
586 	else
587 		*sym = 0;
588 }
589 
590 static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
591 {
592 	return container_of(n, struct bpf_ksym, tnode)->start;
593 }
594 
595 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
596 					  struct latch_tree_node *b)
597 {
598 	return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
599 }
600 
601 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
602 {
603 	unsigned long val = (unsigned long)key;
604 	const struct bpf_ksym *ksym;
605 
606 	ksym = container_of(n, struct bpf_ksym, tnode);
607 
608 	if (val < ksym->start)
609 		return -1;
610 	if (val >= ksym->end)
611 		return  1;
612 
613 	return 0;
614 }
615 
616 static const struct latch_tree_ops bpf_tree_ops = {
617 	.less	= bpf_tree_less,
618 	.comp	= bpf_tree_comp,
619 };
620 
621 static DEFINE_SPINLOCK(bpf_lock);
622 static LIST_HEAD(bpf_kallsyms);
623 static struct latch_tree_root bpf_tree __cacheline_aligned;
624 
625 void bpf_ksym_add(struct bpf_ksym *ksym)
626 {
627 	spin_lock_bh(&bpf_lock);
628 	WARN_ON_ONCE(!list_empty(&ksym->lnode));
629 	list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
630 	latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
631 	spin_unlock_bh(&bpf_lock);
632 }
633 
634 static void __bpf_ksym_del(struct bpf_ksym *ksym)
635 {
636 	if (list_empty(&ksym->lnode))
637 		return;
638 
639 	latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
640 	list_del_rcu(&ksym->lnode);
641 }
642 
643 void bpf_ksym_del(struct bpf_ksym *ksym)
644 {
645 	spin_lock_bh(&bpf_lock);
646 	__bpf_ksym_del(ksym);
647 	spin_unlock_bh(&bpf_lock);
648 }
649 
650 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
651 {
652 	return fp->jited && !bpf_prog_was_classic(fp);
653 }
654 
655 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
656 {
657 	if (!bpf_prog_kallsyms_candidate(fp) ||
658 	    !bpf_capable())
659 		return;
660 
661 	bpf_prog_ksym_set_addr(fp);
662 	bpf_prog_ksym_set_name(fp);
663 	fp->aux->ksym.prog = true;
664 
665 	bpf_ksym_add(&fp->aux->ksym);
666 }
667 
668 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
669 {
670 	if (!bpf_prog_kallsyms_candidate(fp))
671 		return;
672 
673 	bpf_ksym_del(&fp->aux->ksym);
674 }
675 
676 static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
677 {
678 	struct latch_tree_node *n;
679 
680 	n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
681 	return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
682 }
683 
684 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
685 				 unsigned long *off, char *sym)
686 {
687 	struct bpf_ksym *ksym;
688 	char *ret = NULL;
689 
690 	rcu_read_lock();
691 	ksym = bpf_ksym_find(addr);
692 	if (ksym) {
693 		unsigned long symbol_start = ksym->start;
694 		unsigned long symbol_end = ksym->end;
695 
696 		strncpy(sym, ksym->name, KSYM_NAME_LEN);
697 
698 		ret = sym;
699 		if (size)
700 			*size = symbol_end - symbol_start;
701 		if (off)
702 			*off  = addr - symbol_start;
703 	}
704 	rcu_read_unlock();
705 
706 	return ret;
707 }
708 
709 bool is_bpf_text_address(unsigned long addr)
710 {
711 	bool ret;
712 
713 	rcu_read_lock();
714 	ret = bpf_ksym_find(addr) != NULL;
715 	rcu_read_unlock();
716 
717 	return ret;
718 }
719 
720 static struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
721 {
722 	struct bpf_ksym *ksym = bpf_ksym_find(addr);
723 
724 	return ksym && ksym->prog ?
725 	       container_of(ksym, struct bpf_prog_aux, ksym)->prog :
726 	       NULL;
727 }
728 
729 const struct exception_table_entry *search_bpf_extables(unsigned long addr)
730 {
731 	const struct exception_table_entry *e = NULL;
732 	struct bpf_prog *prog;
733 
734 	rcu_read_lock();
735 	prog = bpf_prog_ksym_find(addr);
736 	if (!prog)
737 		goto out;
738 	if (!prog->aux->num_exentries)
739 		goto out;
740 
741 	e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
742 out:
743 	rcu_read_unlock();
744 	return e;
745 }
746 
747 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
748 		    char *sym)
749 {
750 	struct bpf_ksym *ksym;
751 	unsigned int it = 0;
752 	int ret = -ERANGE;
753 
754 	if (!bpf_jit_kallsyms_enabled())
755 		return ret;
756 
757 	rcu_read_lock();
758 	list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
759 		if (it++ != symnum)
760 			continue;
761 
762 		strncpy(sym, ksym->name, KSYM_NAME_LEN);
763 
764 		*value = ksym->start;
765 		*type  = BPF_SYM_ELF_TYPE;
766 
767 		ret = 0;
768 		break;
769 	}
770 	rcu_read_unlock();
771 
772 	return ret;
773 }
774 
775 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
776 				struct bpf_jit_poke_descriptor *poke)
777 {
778 	struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
779 	static const u32 poke_tab_max = 1024;
780 	u32 slot = prog->aux->size_poke_tab;
781 	u32 size = slot + 1;
782 
783 	if (size > poke_tab_max)
784 		return -ENOSPC;
785 	if (poke->tailcall_target || poke->tailcall_target_stable ||
786 	    poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
787 		return -EINVAL;
788 
789 	switch (poke->reason) {
790 	case BPF_POKE_REASON_TAIL_CALL:
791 		if (!poke->tail_call.map)
792 			return -EINVAL;
793 		break;
794 	default:
795 		return -EINVAL;
796 	}
797 
798 	tab = krealloc(tab, size * sizeof(*poke), GFP_KERNEL);
799 	if (!tab)
800 		return -ENOMEM;
801 
802 	memcpy(&tab[slot], poke, sizeof(*poke));
803 	prog->aux->size_poke_tab = size;
804 	prog->aux->poke_tab = tab;
805 
806 	return slot;
807 }
808 
809 /*
810  * BPF program pack allocator.
811  *
812  * Most BPF programs are pretty small. Allocating a hole page for each
813  * program is sometime a waste. Many small bpf program also adds pressure
814  * to instruction TLB. To solve this issue, we introduce a BPF program pack
815  * allocator. The prog_pack allocator uses HPAGE_PMD_SIZE page (2MB on x86)
816  * to host BPF programs.
817  */
818 #define BPF_PROG_CHUNK_SHIFT	6
819 #define BPF_PROG_CHUNK_SIZE	(1 << BPF_PROG_CHUNK_SHIFT)
820 #define BPF_PROG_CHUNK_MASK	(~(BPF_PROG_CHUNK_SIZE - 1))
821 
822 struct bpf_prog_pack {
823 	struct list_head list;
824 	void *ptr;
825 	unsigned long bitmap[];
826 };
827 
828 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size)
829 {
830 	memset(area, 0, size);
831 }
832 
833 #define BPF_PROG_SIZE_TO_NBITS(size)	(round_up(size, BPF_PROG_CHUNK_SIZE) / BPF_PROG_CHUNK_SIZE)
834 
835 static DEFINE_MUTEX(pack_mutex);
836 static LIST_HEAD(pack_list);
837 
838 /* PMD_SIZE is not available in some special config, e.g. ARCH=arm with
839  * CONFIG_MMU=n. Use PAGE_SIZE in these cases.
840  */
841 #ifdef PMD_SIZE
842 #define BPF_PROG_PACK_SIZE (PMD_SIZE * num_possible_nodes())
843 #else
844 #define BPF_PROG_PACK_SIZE PAGE_SIZE
845 #endif
846 
847 #define BPF_PROG_CHUNK_COUNT (BPF_PROG_PACK_SIZE / BPF_PROG_CHUNK_SIZE)
848 
849 static struct bpf_prog_pack *alloc_new_pack(bpf_jit_fill_hole_t bpf_fill_ill_insns)
850 {
851 	struct bpf_prog_pack *pack;
852 
853 	pack = kzalloc(struct_size(pack, bitmap, BITS_TO_LONGS(BPF_PROG_CHUNK_COUNT)),
854 		       GFP_KERNEL);
855 	if (!pack)
856 		return NULL;
857 	pack->ptr = module_alloc(BPF_PROG_PACK_SIZE);
858 	if (!pack->ptr) {
859 		kfree(pack);
860 		return NULL;
861 	}
862 	bpf_fill_ill_insns(pack->ptr, BPF_PROG_PACK_SIZE);
863 	bitmap_zero(pack->bitmap, BPF_PROG_PACK_SIZE / BPF_PROG_CHUNK_SIZE);
864 	list_add_tail(&pack->list, &pack_list);
865 
866 	set_vm_flush_reset_perms(pack->ptr);
867 	set_memory_ro((unsigned long)pack->ptr, BPF_PROG_PACK_SIZE / PAGE_SIZE);
868 	set_memory_x((unsigned long)pack->ptr, BPF_PROG_PACK_SIZE / PAGE_SIZE);
869 	return pack;
870 }
871 
872 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns)
873 {
874 	unsigned int nbits = BPF_PROG_SIZE_TO_NBITS(size);
875 	struct bpf_prog_pack *pack;
876 	unsigned long pos;
877 	void *ptr = NULL;
878 
879 	mutex_lock(&pack_mutex);
880 	if (size > BPF_PROG_PACK_SIZE) {
881 		size = round_up(size, PAGE_SIZE);
882 		ptr = module_alloc(size);
883 		if (ptr) {
884 			bpf_fill_ill_insns(ptr, size);
885 			set_vm_flush_reset_perms(ptr);
886 			set_memory_ro((unsigned long)ptr, size / PAGE_SIZE);
887 			set_memory_x((unsigned long)ptr, size / PAGE_SIZE);
888 		}
889 		goto out;
890 	}
891 	list_for_each_entry(pack, &pack_list, list) {
892 		pos = bitmap_find_next_zero_area(pack->bitmap, BPF_PROG_CHUNK_COUNT, 0,
893 						 nbits, 0);
894 		if (pos < BPF_PROG_CHUNK_COUNT)
895 			goto found_free_area;
896 	}
897 
898 	pack = alloc_new_pack(bpf_fill_ill_insns);
899 	if (!pack)
900 		goto out;
901 
902 	pos = 0;
903 
904 found_free_area:
905 	bitmap_set(pack->bitmap, pos, nbits);
906 	ptr = (void *)(pack->ptr) + (pos << BPF_PROG_CHUNK_SHIFT);
907 
908 out:
909 	mutex_unlock(&pack_mutex);
910 	return ptr;
911 }
912 
913 void bpf_prog_pack_free(struct bpf_binary_header *hdr)
914 {
915 	struct bpf_prog_pack *pack = NULL, *tmp;
916 	unsigned int nbits;
917 	unsigned long pos;
918 
919 	mutex_lock(&pack_mutex);
920 	if (hdr->size > BPF_PROG_PACK_SIZE) {
921 		module_memfree(hdr);
922 		goto out;
923 	}
924 
925 	list_for_each_entry(tmp, &pack_list, list) {
926 		if ((void *)hdr >= tmp->ptr && (tmp->ptr + BPF_PROG_PACK_SIZE) > (void *)hdr) {
927 			pack = tmp;
928 			break;
929 		}
930 	}
931 
932 	if (WARN_ONCE(!pack, "bpf_prog_pack bug\n"))
933 		goto out;
934 
935 	nbits = BPF_PROG_SIZE_TO_NBITS(hdr->size);
936 	pos = ((unsigned long)hdr - (unsigned long)pack->ptr) >> BPF_PROG_CHUNK_SHIFT;
937 
938 	WARN_ONCE(bpf_arch_text_invalidate(hdr, hdr->size),
939 		  "bpf_prog_pack bug: missing bpf_arch_text_invalidate?\n");
940 
941 	bitmap_clear(pack->bitmap, pos, nbits);
942 	if (bitmap_find_next_zero_area(pack->bitmap, BPF_PROG_CHUNK_COUNT, 0,
943 				       BPF_PROG_CHUNK_COUNT, 0) == 0) {
944 		list_del(&pack->list);
945 		module_memfree(pack->ptr);
946 		kfree(pack);
947 	}
948 out:
949 	mutex_unlock(&pack_mutex);
950 }
951 
952 static atomic_long_t bpf_jit_current;
953 
954 /* Can be overridden by an arch's JIT compiler if it has a custom,
955  * dedicated BPF backend memory area, or if neither of the two
956  * below apply.
957  */
958 u64 __weak bpf_jit_alloc_exec_limit(void)
959 {
960 #if defined(MODULES_VADDR)
961 	return MODULES_END - MODULES_VADDR;
962 #else
963 	return VMALLOC_END - VMALLOC_START;
964 #endif
965 }
966 
967 static int __init bpf_jit_charge_init(void)
968 {
969 	/* Only used as heuristic here to derive limit. */
970 	bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
971 	bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
972 					    PAGE_SIZE), LONG_MAX);
973 	return 0;
974 }
975 pure_initcall(bpf_jit_charge_init);
976 
977 int bpf_jit_charge_modmem(u32 size)
978 {
979 	if (atomic_long_add_return(size, &bpf_jit_current) > READ_ONCE(bpf_jit_limit)) {
980 		if (!bpf_capable()) {
981 			atomic_long_sub(size, &bpf_jit_current);
982 			return -EPERM;
983 		}
984 	}
985 
986 	return 0;
987 }
988 
989 void bpf_jit_uncharge_modmem(u32 size)
990 {
991 	atomic_long_sub(size, &bpf_jit_current);
992 }
993 
994 void *__weak bpf_jit_alloc_exec(unsigned long size)
995 {
996 	return module_alloc(size);
997 }
998 
999 void __weak bpf_jit_free_exec(void *addr)
1000 {
1001 	module_memfree(addr);
1002 }
1003 
1004 struct bpf_binary_header *
1005 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1006 		     unsigned int alignment,
1007 		     bpf_jit_fill_hole_t bpf_fill_ill_insns)
1008 {
1009 	struct bpf_binary_header *hdr;
1010 	u32 size, hole, start;
1011 
1012 	WARN_ON_ONCE(!is_power_of_2(alignment) ||
1013 		     alignment > BPF_IMAGE_ALIGNMENT);
1014 
1015 	/* Most of BPF filters are really small, but if some of them
1016 	 * fill a page, allow at least 128 extra bytes to insert a
1017 	 * random section of illegal instructions.
1018 	 */
1019 	size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
1020 
1021 	if (bpf_jit_charge_modmem(size))
1022 		return NULL;
1023 	hdr = bpf_jit_alloc_exec(size);
1024 	if (!hdr) {
1025 		bpf_jit_uncharge_modmem(size);
1026 		return NULL;
1027 	}
1028 
1029 	/* Fill space with illegal/arch-dep instructions. */
1030 	bpf_fill_ill_insns(hdr, size);
1031 
1032 	hdr->size = size;
1033 	hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
1034 		     PAGE_SIZE - sizeof(*hdr));
1035 	start = prandom_u32_max(hole) & ~(alignment - 1);
1036 
1037 	/* Leave a random number of instructions before BPF code. */
1038 	*image_ptr = &hdr->image[start];
1039 
1040 	return hdr;
1041 }
1042 
1043 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
1044 {
1045 	u32 size = hdr->size;
1046 
1047 	bpf_jit_free_exec(hdr);
1048 	bpf_jit_uncharge_modmem(size);
1049 }
1050 
1051 /* Allocate jit binary from bpf_prog_pack allocator.
1052  * Since the allocated memory is RO+X, the JIT engine cannot write directly
1053  * to the memory. To solve this problem, a RW buffer is also allocated at
1054  * as the same time. The JIT engine should calculate offsets based on the
1055  * RO memory address, but write JITed program to the RW buffer. Once the
1056  * JIT engine finishes, it calls bpf_jit_binary_pack_finalize, which copies
1057  * the JITed program to the RO memory.
1058  */
1059 struct bpf_binary_header *
1060 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **image_ptr,
1061 			  unsigned int alignment,
1062 			  struct bpf_binary_header **rw_header,
1063 			  u8 **rw_image,
1064 			  bpf_jit_fill_hole_t bpf_fill_ill_insns)
1065 {
1066 	struct bpf_binary_header *ro_header;
1067 	u32 size, hole, start;
1068 
1069 	WARN_ON_ONCE(!is_power_of_2(alignment) ||
1070 		     alignment > BPF_IMAGE_ALIGNMENT);
1071 
1072 	/* add 16 bytes for a random section of illegal instructions */
1073 	size = round_up(proglen + sizeof(*ro_header) + 16, BPF_PROG_CHUNK_SIZE);
1074 
1075 	if (bpf_jit_charge_modmem(size))
1076 		return NULL;
1077 	ro_header = bpf_prog_pack_alloc(size, bpf_fill_ill_insns);
1078 	if (!ro_header) {
1079 		bpf_jit_uncharge_modmem(size);
1080 		return NULL;
1081 	}
1082 
1083 	*rw_header = kvmalloc(size, GFP_KERNEL);
1084 	if (!*rw_header) {
1085 		bpf_arch_text_copy(&ro_header->size, &size, sizeof(size));
1086 		bpf_prog_pack_free(ro_header);
1087 		bpf_jit_uncharge_modmem(size);
1088 		return NULL;
1089 	}
1090 
1091 	/* Fill space with illegal/arch-dep instructions. */
1092 	bpf_fill_ill_insns(*rw_header, size);
1093 	(*rw_header)->size = size;
1094 
1095 	hole = min_t(unsigned int, size - (proglen + sizeof(*ro_header)),
1096 		     BPF_PROG_CHUNK_SIZE - sizeof(*ro_header));
1097 	start = prandom_u32_max(hole) & ~(alignment - 1);
1098 
1099 	*image_ptr = &ro_header->image[start];
1100 	*rw_image = &(*rw_header)->image[start];
1101 
1102 	return ro_header;
1103 }
1104 
1105 /* Copy JITed text from rw_header to its final location, the ro_header. */
1106 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1107 				 struct bpf_binary_header *ro_header,
1108 				 struct bpf_binary_header *rw_header)
1109 {
1110 	void *ptr;
1111 
1112 	ptr = bpf_arch_text_copy(ro_header, rw_header, rw_header->size);
1113 
1114 	kvfree(rw_header);
1115 
1116 	if (IS_ERR(ptr)) {
1117 		bpf_prog_pack_free(ro_header);
1118 		return PTR_ERR(ptr);
1119 	}
1120 	return 0;
1121 }
1122 
1123 /* bpf_jit_binary_pack_free is called in two different scenarios:
1124  *   1) when the program is freed after;
1125  *   2) when the JIT engine fails (before bpf_jit_binary_pack_finalize).
1126  * For case 2), we need to free both the RO memory and the RW buffer.
1127  *
1128  * bpf_jit_binary_pack_free requires proper ro_header->size. However,
1129  * bpf_jit_binary_pack_alloc does not set it. Therefore, ro_header->size
1130  * must be set with either bpf_jit_binary_pack_finalize (normal path) or
1131  * bpf_arch_text_copy (when jit fails).
1132  */
1133 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1134 			      struct bpf_binary_header *rw_header)
1135 {
1136 	u32 size = ro_header->size;
1137 
1138 	bpf_prog_pack_free(ro_header);
1139 	kvfree(rw_header);
1140 	bpf_jit_uncharge_modmem(size);
1141 }
1142 
1143 struct bpf_binary_header *
1144 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp)
1145 {
1146 	unsigned long real_start = (unsigned long)fp->bpf_func;
1147 	unsigned long addr;
1148 
1149 	addr = real_start & BPF_PROG_CHUNK_MASK;
1150 	return (void *)addr;
1151 }
1152 
1153 static inline struct bpf_binary_header *
1154 bpf_jit_binary_hdr(const struct bpf_prog *fp)
1155 {
1156 	unsigned long real_start = (unsigned long)fp->bpf_func;
1157 	unsigned long addr;
1158 
1159 	addr = real_start & PAGE_MASK;
1160 	return (void *)addr;
1161 }
1162 
1163 /* This symbol is only overridden by archs that have different
1164  * requirements than the usual eBPF JITs, f.e. when they only
1165  * implement cBPF JIT, do not set images read-only, etc.
1166  */
1167 void __weak bpf_jit_free(struct bpf_prog *fp)
1168 {
1169 	if (fp->jited) {
1170 		struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
1171 
1172 		bpf_jit_binary_free(hdr);
1173 		WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
1174 	}
1175 
1176 	bpf_prog_unlock_free(fp);
1177 }
1178 
1179 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1180 			  const struct bpf_insn *insn, bool extra_pass,
1181 			  u64 *func_addr, bool *func_addr_fixed)
1182 {
1183 	s16 off = insn->off;
1184 	s32 imm = insn->imm;
1185 	u8 *addr;
1186 
1187 	*func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
1188 	if (!*func_addr_fixed) {
1189 		/* Place-holder address till the last pass has collected
1190 		 * all addresses for JITed subprograms in which case we
1191 		 * can pick them up from prog->aux.
1192 		 */
1193 		if (!extra_pass)
1194 			addr = NULL;
1195 		else if (prog->aux->func &&
1196 			 off >= 0 && off < prog->aux->func_cnt)
1197 			addr = (u8 *)prog->aux->func[off]->bpf_func;
1198 		else
1199 			return -EINVAL;
1200 	} else {
1201 		/* Address of a BPF helper call. Since part of the core
1202 		 * kernel, it's always at a fixed location. __bpf_call_base
1203 		 * and the helper with imm relative to it are both in core
1204 		 * kernel.
1205 		 */
1206 		addr = (u8 *)__bpf_call_base + imm;
1207 	}
1208 
1209 	*func_addr = (unsigned long)addr;
1210 	return 0;
1211 }
1212 
1213 static int bpf_jit_blind_insn(const struct bpf_insn *from,
1214 			      const struct bpf_insn *aux,
1215 			      struct bpf_insn *to_buff,
1216 			      bool emit_zext)
1217 {
1218 	struct bpf_insn *to = to_buff;
1219 	u32 imm_rnd = get_random_u32();
1220 	s16 off;
1221 
1222 	BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
1223 	BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
1224 
1225 	/* Constraints on AX register:
1226 	 *
1227 	 * AX register is inaccessible from user space. It is mapped in
1228 	 * all JITs, and used here for constant blinding rewrites. It is
1229 	 * typically "stateless" meaning its contents are only valid within
1230 	 * the executed instruction, but not across several instructions.
1231 	 * There are a few exceptions however which are further detailed
1232 	 * below.
1233 	 *
1234 	 * Constant blinding is only used by JITs, not in the interpreter.
1235 	 * The interpreter uses AX in some occasions as a local temporary
1236 	 * register e.g. in DIV or MOD instructions.
1237 	 *
1238 	 * In restricted circumstances, the verifier can also use the AX
1239 	 * register for rewrites as long as they do not interfere with
1240 	 * the above cases!
1241 	 */
1242 	if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
1243 		goto out;
1244 
1245 	if (from->imm == 0 &&
1246 	    (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
1247 	     from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
1248 		*to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
1249 		goto out;
1250 	}
1251 
1252 	switch (from->code) {
1253 	case BPF_ALU | BPF_ADD | BPF_K:
1254 	case BPF_ALU | BPF_SUB | BPF_K:
1255 	case BPF_ALU | BPF_AND | BPF_K:
1256 	case BPF_ALU | BPF_OR  | BPF_K:
1257 	case BPF_ALU | BPF_XOR | BPF_K:
1258 	case BPF_ALU | BPF_MUL | BPF_K:
1259 	case BPF_ALU | BPF_MOV | BPF_K:
1260 	case BPF_ALU | BPF_DIV | BPF_K:
1261 	case BPF_ALU | BPF_MOD | BPF_K:
1262 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1263 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1264 		*to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
1265 		break;
1266 
1267 	case BPF_ALU64 | BPF_ADD | BPF_K:
1268 	case BPF_ALU64 | BPF_SUB | BPF_K:
1269 	case BPF_ALU64 | BPF_AND | BPF_K:
1270 	case BPF_ALU64 | BPF_OR  | BPF_K:
1271 	case BPF_ALU64 | BPF_XOR | BPF_K:
1272 	case BPF_ALU64 | BPF_MUL | BPF_K:
1273 	case BPF_ALU64 | BPF_MOV | BPF_K:
1274 	case BPF_ALU64 | BPF_DIV | BPF_K:
1275 	case BPF_ALU64 | BPF_MOD | BPF_K:
1276 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1277 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1278 		*to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
1279 		break;
1280 
1281 	case BPF_JMP | BPF_JEQ  | BPF_K:
1282 	case BPF_JMP | BPF_JNE  | BPF_K:
1283 	case BPF_JMP | BPF_JGT  | BPF_K:
1284 	case BPF_JMP | BPF_JLT  | BPF_K:
1285 	case BPF_JMP | BPF_JGE  | BPF_K:
1286 	case BPF_JMP | BPF_JLE  | BPF_K:
1287 	case BPF_JMP | BPF_JSGT | BPF_K:
1288 	case BPF_JMP | BPF_JSLT | BPF_K:
1289 	case BPF_JMP | BPF_JSGE | BPF_K:
1290 	case BPF_JMP | BPF_JSLE | BPF_K:
1291 	case BPF_JMP | BPF_JSET | BPF_K:
1292 		/* Accommodate for extra offset in case of a backjump. */
1293 		off = from->off;
1294 		if (off < 0)
1295 			off -= 2;
1296 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1297 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1298 		*to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1299 		break;
1300 
1301 	case BPF_JMP32 | BPF_JEQ  | BPF_K:
1302 	case BPF_JMP32 | BPF_JNE  | BPF_K:
1303 	case BPF_JMP32 | BPF_JGT  | BPF_K:
1304 	case BPF_JMP32 | BPF_JLT  | BPF_K:
1305 	case BPF_JMP32 | BPF_JGE  | BPF_K:
1306 	case BPF_JMP32 | BPF_JLE  | BPF_K:
1307 	case BPF_JMP32 | BPF_JSGT | BPF_K:
1308 	case BPF_JMP32 | BPF_JSLT | BPF_K:
1309 	case BPF_JMP32 | BPF_JSGE | BPF_K:
1310 	case BPF_JMP32 | BPF_JSLE | BPF_K:
1311 	case BPF_JMP32 | BPF_JSET | BPF_K:
1312 		/* Accommodate for extra offset in case of a backjump. */
1313 		off = from->off;
1314 		if (off < 0)
1315 			off -= 2;
1316 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1317 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1318 		*to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1319 				      off);
1320 		break;
1321 
1322 	case BPF_LD | BPF_IMM | BPF_DW:
1323 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1324 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1325 		*to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1326 		*to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1327 		break;
1328 	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1329 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1330 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1331 		if (emit_zext)
1332 			*to++ = BPF_ZEXT_REG(BPF_REG_AX);
1333 		*to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
1334 		break;
1335 
1336 	case BPF_ST | BPF_MEM | BPF_DW:
1337 	case BPF_ST | BPF_MEM | BPF_W:
1338 	case BPF_ST | BPF_MEM | BPF_H:
1339 	case BPF_ST | BPF_MEM | BPF_B:
1340 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1341 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1342 		*to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1343 		break;
1344 	}
1345 out:
1346 	return to - to_buff;
1347 }
1348 
1349 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1350 					      gfp_t gfp_extra_flags)
1351 {
1352 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1353 	struct bpf_prog *fp;
1354 
1355 	fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
1356 	if (fp != NULL) {
1357 		/* aux->prog still points to the fp_other one, so
1358 		 * when promoting the clone to the real program,
1359 		 * this still needs to be adapted.
1360 		 */
1361 		memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1362 	}
1363 
1364 	return fp;
1365 }
1366 
1367 static void bpf_prog_clone_free(struct bpf_prog *fp)
1368 {
1369 	/* aux was stolen by the other clone, so we cannot free
1370 	 * it from this path! It will be freed eventually by the
1371 	 * other program on release.
1372 	 *
1373 	 * At this point, we don't need a deferred release since
1374 	 * clone is guaranteed to not be locked.
1375 	 */
1376 	fp->aux = NULL;
1377 	fp->stats = NULL;
1378 	fp->active = NULL;
1379 	__bpf_prog_free(fp);
1380 }
1381 
1382 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1383 {
1384 	/* We have to repoint aux->prog to self, as we don't
1385 	 * know whether fp here is the clone or the original.
1386 	 */
1387 	fp->aux->prog = fp;
1388 	bpf_prog_clone_free(fp_other);
1389 }
1390 
1391 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1392 {
1393 	struct bpf_insn insn_buff[16], aux[2];
1394 	struct bpf_prog *clone, *tmp;
1395 	int insn_delta, insn_cnt;
1396 	struct bpf_insn *insn;
1397 	int i, rewritten;
1398 
1399 	if (!prog->blinding_requested || prog->blinded)
1400 		return prog;
1401 
1402 	clone = bpf_prog_clone_create(prog, GFP_USER);
1403 	if (!clone)
1404 		return ERR_PTR(-ENOMEM);
1405 
1406 	insn_cnt = clone->len;
1407 	insn = clone->insnsi;
1408 
1409 	for (i = 0; i < insn_cnt; i++, insn++) {
1410 		if (bpf_pseudo_func(insn)) {
1411 			/* ld_imm64 with an address of bpf subprog is not
1412 			 * a user controlled constant. Don't randomize it,
1413 			 * since it will conflict with jit_subprogs() logic.
1414 			 */
1415 			insn++;
1416 			i++;
1417 			continue;
1418 		}
1419 
1420 		/* We temporarily need to hold the original ld64 insn
1421 		 * so that we can still access the first part in the
1422 		 * second blinding run.
1423 		 */
1424 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1425 		    insn[1].code == 0)
1426 			memcpy(aux, insn, sizeof(aux));
1427 
1428 		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1429 						clone->aux->verifier_zext);
1430 		if (!rewritten)
1431 			continue;
1432 
1433 		tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1434 		if (IS_ERR(tmp)) {
1435 			/* Patching may have repointed aux->prog during
1436 			 * realloc from the original one, so we need to
1437 			 * fix it up here on error.
1438 			 */
1439 			bpf_jit_prog_release_other(prog, clone);
1440 			return tmp;
1441 		}
1442 
1443 		clone = tmp;
1444 		insn_delta = rewritten - 1;
1445 
1446 		/* Walk new program and skip insns we just inserted. */
1447 		insn = clone->insnsi + i + insn_delta;
1448 		insn_cnt += insn_delta;
1449 		i        += insn_delta;
1450 	}
1451 
1452 	clone->blinded = 1;
1453 	return clone;
1454 }
1455 #endif /* CONFIG_BPF_JIT */
1456 
1457 /* Base function for offset calculation. Needs to go into .text section,
1458  * therefore keeping it non-static as well; will also be used by JITs
1459  * anyway later on, so do not let the compiler omit it. This also needs
1460  * to go into kallsyms for correlation from e.g. bpftool, so naming
1461  * must not change.
1462  */
1463 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1464 {
1465 	return 0;
1466 }
1467 EXPORT_SYMBOL_GPL(__bpf_call_base);
1468 
1469 /* All UAPI available opcodes. */
1470 #define BPF_INSN_MAP(INSN_2, INSN_3)		\
1471 	/* 32 bit ALU operations. */		\
1472 	/*   Register based. */			\
1473 	INSN_3(ALU, ADD,  X),			\
1474 	INSN_3(ALU, SUB,  X),			\
1475 	INSN_3(ALU, AND,  X),			\
1476 	INSN_3(ALU, OR,   X),			\
1477 	INSN_3(ALU, LSH,  X),			\
1478 	INSN_3(ALU, RSH,  X),			\
1479 	INSN_3(ALU, XOR,  X),			\
1480 	INSN_3(ALU, MUL,  X),			\
1481 	INSN_3(ALU, MOV,  X),			\
1482 	INSN_3(ALU, ARSH, X),			\
1483 	INSN_3(ALU, DIV,  X),			\
1484 	INSN_3(ALU, MOD,  X),			\
1485 	INSN_2(ALU, NEG),			\
1486 	INSN_3(ALU, END, TO_BE),		\
1487 	INSN_3(ALU, END, TO_LE),		\
1488 	/*   Immediate based. */		\
1489 	INSN_3(ALU, ADD,  K),			\
1490 	INSN_3(ALU, SUB,  K),			\
1491 	INSN_3(ALU, AND,  K),			\
1492 	INSN_3(ALU, OR,   K),			\
1493 	INSN_3(ALU, LSH,  K),			\
1494 	INSN_3(ALU, RSH,  K),			\
1495 	INSN_3(ALU, XOR,  K),			\
1496 	INSN_3(ALU, MUL,  K),			\
1497 	INSN_3(ALU, MOV,  K),			\
1498 	INSN_3(ALU, ARSH, K),			\
1499 	INSN_3(ALU, DIV,  K),			\
1500 	INSN_3(ALU, MOD,  K),			\
1501 	/* 64 bit ALU operations. */		\
1502 	/*   Register based. */			\
1503 	INSN_3(ALU64, ADD,  X),			\
1504 	INSN_3(ALU64, SUB,  X),			\
1505 	INSN_3(ALU64, AND,  X),			\
1506 	INSN_3(ALU64, OR,   X),			\
1507 	INSN_3(ALU64, LSH,  X),			\
1508 	INSN_3(ALU64, RSH,  X),			\
1509 	INSN_3(ALU64, XOR,  X),			\
1510 	INSN_3(ALU64, MUL,  X),			\
1511 	INSN_3(ALU64, MOV,  X),			\
1512 	INSN_3(ALU64, ARSH, X),			\
1513 	INSN_3(ALU64, DIV,  X),			\
1514 	INSN_3(ALU64, MOD,  X),			\
1515 	INSN_2(ALU64, NEG),			\
1516 	/*   Immediate based. */		\
1517 	INSN_3(ALU64, ADD,  K),			\
1518 	INSN_3(ALU64, SUB,  K),			\
1519 	INSN_3(ALU64, AND,  K),			\
1520 	INSN_3(ALU64, OR,   K),			\
1521 	INSN_3(ALU64, LSH,  K),			\
1522 	INSN_3(ALU64, RSH,  K),			\
1523 	INSN_3(ALU64, XOR,  K),			\
1524 	INSN_3(ALU64, MUL,  K),			\
1525 	INSN_3(ALU64, MOV,  K),			\
1526 	INSN_3(ALU64, ARSH, K),			\
1527 	INSN_3(ALU64, DIV,  K),			\
1528 	INSN_3(ALU64, MOD,  K),			\
1529 	/* Call instruction. */			\
1530 	INSN_2(JMP, CALL),			\
1531 	/* Exit instruction. */			\
1532 	INSN_2(JMP, EXIT),			\
1533 	/* 32-bit Jump instructions. */		\
1534 	/*   Register based. */			\
1535 	INSN_3(JMP32, JEQ,  X),			\
1536 	INSN_3(JMP32, JNE,  X),			\
1537 	INSN_3(JMP32, JGT,  X),			\
1538 	INSN_3(JMP32, JLT,  X),			\
1539 	INSN_3(JMP32, JGE,  X),			\
1540 	INSN_3(JMP32, JLE,  X),			\
1541 	INSN_3(JMP32, JSGT, X),			\
1542 	INSN_3(JMP32, JSLT, X),			\
1543 	INSN_3(JMP32, JSGE, X),			\
1544 	INSN_3(JMP32, JSLE, X),			\
1545 	INSN_3(JMP32, JSET, X),			\
1546 	/*   Immediate based. */		\
1547 	INSN_3(JMP32, JEQ,  K),			\
1548 	INSN_3(JMP32, JNE,  K),			\
1549 	INSN_3(JMP32, JGT,  K),			\
1550 	INSN_3(JMP32, JLT,  K),			\
1551 	INSN_3(JMP32, JGE,  K),			\
1552 	INSN_3(JMP32, JLE,  K),			\
1553 	INSN_3(JMP32, JSGT, K),			\
1554 	INSN_3(JMP32, JSLT, K),			\
1555 	INSN_3(JMP32, JSGE, K),			\
1556 	INSN_3(JMP32, JSLE, K),			\
1557 	INSN_3(JMP32, JSET, K),			\
1558 	/* Jump instructions. */		\
1559 	/*   Register based. */			\
1560 	INSN_3(JMP, JEQ,  X),			\
1561 	INSN_3(JMP, JNE,  X),			\
1562 	INSN_3(JMP, JGT,  X),			\
1563 	INSN_3(JMP, JLT,  X),			\
1564 	INSN_3(JMP, JGE,  X),			\
1565 	INSN_3(JMP, JLE,  X),			\
1566 	INSN_3(JMP, JSGT, X),			\
1567 	INSN_3(JMP, JSLT, X),			\
1568 	INSN_3(JMP, JSGE, X),			\
1569 	INSN_3(JMP, JSLE, X),			\
1570 	INSN_3(JMP, JSET, X),			\
1571 	/*   Immediate based. */		\
1572 	INSN_3(JMP, JEQ,  K),			\
1573 	INSN_3(JMP, JNE,  K),			\
1574 	INSN_3(JMP, JGT,  K),			\
1575 	INSN_3(JMP, JLT,  K),			\
1576 	INSN_3(JMP, JGE,  K),			\
1577 	INSN_3(JMP, JLE,  K),			\
1578 	INSN_3(JMP, JSGT, K),			\
1579 	INSN_3(JMP, JSLT, K),			\
1580 	INSN_3(JMP, JSGE, K),			\
1581 	INSN_3(JMP, JSLE, K),			\
1582 	INSN_3(JMP, JSET, K),			\
1583 	INSN_2(JMP, JA),			\
1584 	/* Store instructions. */		\
1585 	/*   Register based. */			\
1586 	INSN_3(STX, MEM,  B),			\
1587 	INSN_3(STX, MEM,  H),			\
1588 	INSN_3(STX, MEM,  W),			\
1589 	INSN_3(STX, MEM,  DW),			\
1590 	INSN_3(STX, ATOMIC, W),			\
1591 	INSN_3(STX, ATOMIC, DW),		\
1592 	/*   Immediate based. */		\
1593 	INSN_3(ST, MEM, B),			\
1594 	INSN_3(ST, MEM, H),			\
1595 	INSN_3(ST, MEM, W),			\
1596 	INSN_3(ST, MEM, DW),			\
1597 	/* Load instructions. */		\
1598 	/*   Register based. */			\
1599 	INSN_3(LDX, MEM, B),			\
1600 	INSN_3(LDX, MEM, H),			\
1601 	INSN_3(LDX, MEM, W),			\
1602 	INSN_3(LDX, MEM, DW),			\
1603 	/*   Immediate based. */		\
1604 	INSN_3(LD, IMM, DW)
1605 
1606 bool bpf_opcode_in_insntable(u8 code)
1607 {
1608 #define BPF_INSN_2_TBL(x, y)    [BPF_##x | BPF_##y] = true
1609 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1610 	static const bool public_insntable[256] = {
1611 		[0 ... 255] = false,
1612 		/* Now overwrite non-defaults ... */
1613 		BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1614 		/* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1615 		[BPF_LD | BPF_ABS | BPF_B] = true,
1616 		[BPF_LD | BPF_ABS | BPF_H] = true,
1617 		[BPF_LD | BPF_ABS | BPF_W] = true,
1618 		[BPF_LD | BPF_IND | BPF_B] = true,
1619 		[BPF_LD | BPF_IND | BPF_H] = true,
1620 		[BPF_LD | BPF_IND | BPF_W] = true,
1621 	};
1622 #undef BPF_INSN_3_TBL
1623 #undef BPF_INSN_2_TBL
1624 	return public_insntable[code];
1625 }
1626 
1627 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1628 u64 __weak bpf_probe_read_kernel(void *dst, u32 size, const void *unsafe_ptr)
1629 {
1630 	memset(dst, 0, size);
1631 	return -EFAULT;
1632 }
1633 
1634 /**
1635  *	___bpf_prog_run - run eBPF program on a given context
1636  *	@regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1637  *	@insn: is the array of eBPF instructions
1638  *
1639  * Decode and execute eBPF instructions.
1640  *
1641  * Return: whatever value is in %BPF_R0 at program exit
1642  */
1643 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
1644 {
1645 #define BPF_INSN_2_LBL(x, y)    [BPF_##x | BPF_##y] = &&x##_##y
1646 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1647 	static const void * const jumptable[256] __annotate_jump_table = {
1648 		[0 ... 255] = &&default_label,
1649 		/* Now overwrite non-defaults ... */
1650 		BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1651 		/* Non-UAPI available opcodes. */
1652 		[BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1653 		[BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1654 		[BPF_ST  | BPF_NOSPEC] = &&ST_NOSPEC,
1655 		[BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1656 		[BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1657 		[BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1658 		[BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1659 	};
1660 #undef BPF_INSN_3_LBL
1661 #undef BPF_INSN_2_LBL
1662 	u32 tail_call_cnt = 0;
1663 
1664 #define CONT	 ({ insn++; goto select_insn; })
1665 #define CONT_JMP ({ insn++; goto select_insn; })
1666 
1667 select_insn:
1668 	goto *jumptable[insn->code];
1669 
1670 	/* Explicitly mask the register-based shift amounts with 63 or 31
1671 	 * to avoid undefined behavior. Normally this won't affect the
1672 	 * generated code, for example, in case of native 64 bit archs such
1673 	 * as x86-64 or arm64, the compiler is optimizing the AND away for
1674 	 * the interpreter. In case of JITs, each of the JIT backends compiles
1675 	 * the BPF shift operations to machine instructions which produce
1676 	 * implementation-defined results in such a case; the resulting
1677 	 * contents of the register may be arbitrary, but program behaviour
1678 	 * as a whole remains defined. In other words, in case of JIT backends,
1679 	 * the AND must /not/ be added to the emitted LSH/RSH/ARSH translation.
1680 	 */
1681 	/* ALU (shifts) */
1682 #define SHT(OPCODE, OP)					\
1683 	ALU64_##OPCODE##_X:				\
1684 		DST = DST OP (SRC & 63);		\
1685 		CONT;					\
1686 	ALU_##OPCODE##_X:				\
1687 		DST = (u32) DST OP ((u32) SRC & 31);	\
1688 		CONT;					\
1689 	ALU64_##OPCODE##_K:				\
1690 		DST = DST OP IMM;			\
1691 		CONT;					\
1692 	ALU_##OPCODE##_K:				\
1693 		DST = (u32) DST OP (u32) IMM;		\
1694 		CONT;
1695 	/* ALU (rest) */
1696 #define ALU(OPCODE, OP)					\
1697 	ALU64_##OPCODE##_X:				\
1698 		DST = DST OP SRC;			\
1699 		CONT;					\
1700 	ALU_##OPCODE##_X:				\
1701 		DST = (u32) DST OP (u32) SRC;		\
1702 		CONT;					\
1703 	ALU64_##OPCODE##_K:				\
1704 		DST = DST OP IMM;			\
1705 		CONT;					\
1706 	ALU_##OPCODE##_K:				\
1707 		DST = (u32) DST OP (u32) IMM;		\
1708 		CONT;
1709 	ALU(ADD,  +)
1710 	ALU(SUB,  -)
1711 	ALU(AND,  &)
1712 	ALU(OR,   |)
1713 	ALU(XOR,  ^)
1714 	ALU(MUL,  *)
1715 	SHT(LSH, <<)
1716 	SHT(RSH, >>)
1717 #undef SHT
1718 #undef ALU
1719 	ALU_NEG:
1720 		DST = (u32) -DST;
1721 		CONT;
1722 	ALU64_NEG:
1723 		DST = -DST;
1724 		CONT;
1725 	ALU_MOV_X:
1726 		DST = (u32) SRC;
1727 		CONT;
1728 	ALU_MOV_K:
1729 		DST = (u32) IMM;
1730 		CONT;
1731 	ALU64_MOV_X:
1732 		DST = SRC;
1733 		CONT;
1734 	ALU64_MOV_K:
1735 		DST = IMM;
1736 		CONT;
1737 	LD_IMM_DW:
1738 		DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1739 		insn++;
1740 		CONT;
1741 	ALU_ARSH_X:
1742 		DST = (u64) (u32) (((s32) DST) >> (SRC & 31));
1743 		CONT;
1744 	ALU_ARSH_K:
1745 		DST = (u64) (u32) (((s32) DST) >> IMM);
1746 		CONT;
1747 	ALU64_ARSH_X:
1748 		(*(s64 *) &DST) >>= (SRC & 63);
1749 		CONT;
1750 	ALU64_ARSH_K:
1751 		(*(s64 *) &DST) >>= IMM;
1752 		CONT;
1753 	ALU64_MOD_X:
1754 		div64_u64_rem(DST, SRC, &AX);
1755 		DST = AX;
1756 		CONT;
1757 	ALU_MOD_X:
1758 		AX = (u32) DST;
1759 		DST = do_div(AX, (u32) SRC);
1760 		CONT;
1761 	ALU64_MOD_K:
1762 		div64_u64_rem(DST, IMM, &AX);
1763 		DST = AX;
1764 		CONT;
1765 	ALU_MOD_K:
1766 		AX = (u32) DST;
1767 		DST = do_div(AX, (u32) IMM);
1768 		CONT;
1769 	ALU64_DIV_X:
1770 		DST = div64_u64(DST, SRC);
1771 		CONT;
1772 	ALU_DIV_X:
1773 		AX = (u32) DST;
1774 		do_div(AX, (u32) SRC);
1775 		DST = (u32) AX;
1776 		CONT;
1777 	ALU64_DIV_K:
1778 		DST = div64_u64(DST, IMM);
1779 		CONT;
1780 	ALU_DIV_K:
1781 		AX = (u32) DST;
1782 		do_div(AX, (u32) IMM);
1783 		DST = (u32) AX;
1784 		CONT;
1785 	ALU_END_TO_BE:
1786 		switch (IMM) {
1787 		case 16:
1788 			DST = (__force u16) cpu_to_be16(DST);
1789 			break;
1790 		case 32:
1791 			DST = (__force u32) cpu_to_be32(DST);
1792 			break;
1793 		case 64:
1794 			DST = (__force u64) cpu_to_be64(DST);
1795 			break;
1796 		}
1797 		CONT;
1798 	ALU_END_TO_LE:
1799 		switch (IMM) {
1800 		case 16:
1801 			DST = (__force u16) cpu_to_le16(DST);
1802 			break;
1803 		case 32:
1804 			DST = (__force u32) cpu_to_le32(DST);
1805 			break;
1806 		case 64:
1807 			DST = (__force u64) cpu_to_le64(DST);
1808 			break;
1809 		}
1810 		CONT;
1811 
1812 	/* CALL */
1813 	JMP_CALL:
1814 		/* Function call scratches BPF_R1-BPF_R5 registers,
1815 		 * preserves BPF_R6-BPF_R9, and stores return value
1816 		 * into BPF_R0.
1817 		 */
1818 		BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1819 						       BPF_R4, BPF_R5);
1820 		CONT;
1821 
1822 	JMP_CALL_ARGS:
1823 		BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1824 							    BPF_R3, BPF_R4,
1825 							    BPF_R5,
1826 							    insn + insn->off + 1);
1827 		CONT;
1828 
1829 	JMP_TAIL_CALL: {
1830 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1831 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1832 		struct bpf_prog *prog;
1833 		u32 index = BPF_R3;
1834 
1835 		if (unlikely(index >= array->map.max_entries))
1836 			goto out;
1837 
1838 		if (unlikely(tail_call_cnt >= MAX_TAIL_CALL_CNT))
1839 			goto out;
1840 
1841 		tail_call_cnt++;
1842 
1843 		prog = READ_ONCE(array->ptrs[index]);
1844 		if (!prog)
1845 			goto out;
1846 
1847 		/* ARG1 at this point is guaranteed to point to CTX from
1848 		 * the verifier side due to the fact that the tail call is
1849 		 * handled like a helper, that is, bpf_tail_call_proto,
1850 		 * where arg1_type is ARG_PTR_TO_CTX.
1851 		 */
1852 		insn = prog->insnsi;
1853 		goto select_insn;
1854 out:
1855 		CONT;
1856 	}
1857 	JMP_JA:
1858 		insn += insn->off;
1859 		CONT;
1860 	JMP_EXIT:
1861 		return BPF_R0;
1862 	/* JMP */
1863 #define COND_JMP(SIGN, OPCODE, CMP_OP)				\
1864 	JMP_##OPCODE##_X:					\
1865 		if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) {	\
1866 			insn += insn->off;			\
1867 			CONT_JMP;				\
1868 		}						\
1869 		CONT;						\
1870 	JMP32_##OPCODE##_X:					\
1871 		if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) {	\
1872 			insn += insn->off;			\
1873 			CONT_JMP;				\
1874 		}						\
1875 		CONT;						\
1876 	JMP_##OPCODE##_K:					\
1877 		if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) {	\
1878 			insn += insn->off;			\
1879 			CONT_JMP;				\
1880 		}						\
1881 		CONT;						\
1882 	JMP32_##OPCODE##_K:					\
1883 		if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) {	\
1884 			insn += insn->off;			\
1885 			CONT_JMP;				\
1886 		}						\
1887 		CONT;
1888 	COND_JMP(u, JEQ, ==)
1889 	COND_JMP(u, JNE, !=)
1890 	COND_JMP(u, JGT, >)
1891 	COND_JMP(u, JLT, <)
1892 	COND_JMP(u, JGE, >=)
1893 	COND_JMP(u, JLE, <=)
1894 	COND_JMP(u, JSET, &)
1895 	COND_JMP(s, JSGT, >)
1896 	COND_JMP(s, JSLT, <)
1897 	COND_JMP(s, JSGE, >=)
1898 	COND_JMP(s, JSLE, <=)
1899 #undef COND_JMP
1900 	/* ST, STX and LDX*/
1901 	ST_NOSPEC:
1902 		/* Speculation barrier for mitigating Speculative Store Bypass.
1903 		 * In case of arm64, we rely on the firmware mitigation as
1904 		 * controlled via the ssbd kernel parameter. Whenever the
1905 		 * mitigation is enabled, it works for all of the kernel code
1906 		 * with no need to provide any additional instructions here.
1907 		 * In case of x86, we use 'lfence' insn for mitigation. We
1908 		 * reuse preexisting logic from Spectre v1 mitigation that
1909 		 * happens to produce the required code on x86 for v4 as well.
1910 		 */
1911 #ifdef CONFIG_X86
1912 		barrier_nospec();
1913 #endif
1914 		CONT;
1915 #define LDST(SIZEOP, SIZE)						\
1916 	STX_MEM_##SIZEOP:						\
1917 		*(SIZE *)(unsigned long) (DST + insn->off) = SRC;	\
1918 		CONT;							\
1919 	ST_MEM_##SIZEOP:						\
1920 		*(SIZE *)(unsigned long) (DST + insn->off) = IMM;	\
1921 		CONT;							\
1922 	LDX_MEM_##SIZEOP:						\
1923 		DST = *(SIZE *)(unsigned long) (SRC + insn->off);	\
1924 		CONT;							\
1925 	LDX_PROBE_MEM_##SIZEOP:						\
1926 		bpf_probe_read_kernel(&DST, sizeof(SIZE),		\
1927 				      (const void *)(long) (SRC + insn->off));	\
1928 		DST = *((SIZE *)&DST);					\
1929 		CONT;
1930 
1931 	LDST(B,   u8)
1932 	LDST(H,  u16)
1933 	LDST(W,  u32)
1934 	LDST(DW, u64)
1935 #undef LDST
1936 
1937 #define ATOMIC_ALU_OP(BOP, KOP)						\
1938 		case BOP:						\
1939 			if (BPF_SIZE(insn->code) == BPF_W)		\
1940 				atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
1941 					     (DST + insn->off));	\
1942 			else						\
1943 				atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
1944 					       (DST + insn->off));	\
1945 			break;						\
1946 		case BOP | BPF_FETCH:					\
1947 			if (BPF_SIZE(insn->code) == BPF_W)		\
1948 				SRC = (u32) atomic_fetch_##KOP(		\
1949 					(u32) SRC,			\
1950 					(atomic_t *)(unsigned long) (DST + insn->off)); \
1951 			else						\
1952 				SRC = (u64) atomic64_fetch_##KOP(	\
1953 					(u64) SRC,			\
1954 					(atomic64_t *)(unsigned long) (DST + insn->off)); \
1955 			break;
1956 
1957 	STX_ATOMIC_DW:
1958 	STX_ATOMIC_W:
1959 		switch (IMM) {
1960 		ATOMIC_ALU_OP(BPF_ADD, add)
1961 		ATOMIC_ALU_OP(BPF_AND, and)
1962 		ATOMIC_ALU_OP(BPF_OR, or)
1963 		ATOMIC_ALU_OP(BPF_XOR, xor)
1964 #undef ATOMIC_ALU_OP
1965 
1966 		case BPF_XCHG:
1967 			if (BPF_SIZE(insn->code) == BPF_W)
1968 				SRC = (u32) atomic_xchg(
1969 					(atomic_t *)(unsigned long) (DST + insn->off),
1970 					(u32) SRC);
1971 			else
1972 				SRC = (u64) atomic64_xchg(
1973 					(atomic64_t *)(unsigned long) (DST + insn->off),
1974 					(u64) SRC);
1975 			break;
1976 		case BPF_CMPXCHG:
1977 			if (BPF_SIZE(insn->code) == BPF_W)
1978 				BPF_R0 = (u32) atomic_cmpxchg(
1979 					(atomic_t *)(unsigned long) (DST + insn->off),
1980 					(u32) BPF_R0, (u32) SRC);
1981 			else
1982 				BPF_R0 = (u64) atomic64_cmpxchg(
1983 					(atomic64_t *)(unsigned long) (DST + insn->off),
1984 					(u64) BPF_R0, (u64) SRC);
1985 			break;
1986 
1987 		default:
1988 			goto default_label;
1989 		}
1990 		CONT;
1991 
1992 	default_label:
1993 		/* If we ever reach this, we have a bug somewhere. Die hard here
1994 		 * instead of just returning 0; we could be somewhere in a subprog,
1995 		 * so execution could continue otherwise which we do /not/ want.
1996 		 *
1997 		 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1998 		 */
1999 		pr_warn("BPF interpreter: unknown opcode %02x (imm: 0x%x)\n",
2000 			insn->code, insn->imm);
2001 		BUG_ON(1);
2002 		return 0;
2003 }
2004 
2005 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
2006 #define DEFINE_BPF_PROG_RUN(stack_size) \
2007 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
2008 { \
2009 	u64 stack[stack_size / sizeof(u64)]; \
2010 	u64 regs[MAX_BPF_EXT_REG] = {}; \
2011 \
2012 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2013 	ARG1 = (u64) (unsigned long) ctx; \
2014 	return ___bpf_prog_run(regs, insn); \
2015 }
2016 
2017 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
2018 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
2019 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
2020 				      const struct bpf_insn *insn) \
2021 { \
2022 	u64 stack[stack_size / sizeof(u64)]; \
2023 	u64 regs[MAX_BPF_EXT_REG]; \
2024 \
2025 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2026 	BPF_R1 = r1; \
2027 	BPF_R2 = r2; \
2028 	BPF_R3 = r3; \
2029 	BPF_R4 = r4; \
2030 	BPF_R5 = r5; \
2031 	return ___bpf_prog_run(regs, insn); \
2032 }
2033 
2034 #define EVAL1(FN, X) FN(X)
2035 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
2036 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
2037 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
2038 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
2039 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
2040 
2041 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
2042 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
2043 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
2044 
2045 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
2046 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
2047 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
2048 
2049 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
2050 
2051 static unsigned int (*interpreters[])(const void *ctx,
2052 				      const struct bpf_insn *insn) = {
2053 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2054 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2055 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2056 };
2057 #undef PROG_NAME_LIST
2058 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
2059 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
2060 				  const struct bpf_insn *insn) = {
2061 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2062 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2063 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2064 };
2065 #undef PROG_NAME_LIST
2066 
2067 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
2068 {
2069 	stack_depth = max_t(u32, stack_depth, 1);
2070 	insn->off = (s16) insn->imm;
2071 	insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
2072 		__bpf_call_base_args;
2073 	insn->code = BPF_JMP | BPF_CALL_ARGS;
2074 }
2075 
2076 #else
2077 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
2078 					 const struct bpf_insn *insn)
2079 {
2080 	/* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
2081 	 * is not working properly, so warn about it!
2082 	 */
2083 	WARN_ON_ONCE(1);
2084 	return 0;
2085 }
2086 #endif
2087 
2088 bool bpf_prog_map_compatible(struct bpf_map *map,
2089 			     const struct bpf_prog *fp)
2090 {
2091 	bool ret;
2092 
2093 	if (fp->kprobe_override)
2094 		return false;
2095 
2096 	spin_lock(&map->owner.lock);
2097 	if (!map->owner.type) {
2098 		/* There's no owner yet where we could check for
2099 		 * compatibility.
2100 		 */
2101 		map->owner.type  = fp->type;
2102 		map->owner.jited = fp->jited;
2103 		map->owner.xdp_has_frags = fp->aux->xdp_has_frags;
2104 		ret = true;
2105 	} else {
2106 		ret = map->owner.type  == fp->type &&
2107 		      map->owner.jited == fp->jited &&
2108 		      map->owner.xdp_has_frags == fp->aux->xdp_has_frags;
2109 	}
2110 	spin_unlock(&map->owner.lock);
2111 
2112 	return ret;
2113 }
2114 
2115 static int bpf_check_tail_call(const struct bpf_prog *fp)
2116 {
2117 	struct bpf_prog_aux *aux = fp->aux;
2118 	int i, ret = 0;
2119 
2120 	mutex_lock(&aux->used_maps_mutex);
2121 	for (i = 0; i < aux->used_map_cnt; i++) {
2122 		struct bpf_map *map = aux->used_maps[i];
2123 
2124 		if (!map_type_contains_progs(map))
2125 			continue;
2126 
2127 		if (!bpf_prog_map_compatible(map, fp)) {
2128 			ret = -EINVAL;
2129 			goto out;
2130 		}
2131 	}
2132 
2133 out:
2134 	mutex_unlock(&aux->used_maps_mutex);
2135 	return ret;
2136 }
2137 
2138 static void bpf_prog_select_func(struct bpf_prog *fp)
2139 {
2140 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
2141 	u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
2142 
2143 	fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
2144 #else
2145 	fp->bpf_func = __bpf_prog_ret0_warn;
2146 #endif
2147 }
2148 
2149 /**
2150  *	bpf_prog_select_runtime - select exec runtime for BPF program
2151  *	@fp: bpf_prog populated with BPF program
2152  *	@err: pointer to error variable
2153  *
2154  * Try to JIT eBPF program, if JIT is not available, use interpreter.
2155  * The BPF program will be executed via bpf_prog_run() function.
2156  *
2157  * Return: the &fp argument along with &err set to 0 for success or
2158  * a negative errno code on failure
2159  */
2160 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
2161 {
2162 	/* In case of BPF to BPF calls, verifier did all the prep
2163 	 * work with regards to JITing, etc.
2164 	 */
2165 	bool jit_needed = false;
2166 
2167 	if (fp->bpf_func)
2168 		goto finalize;
2169 
2170 	if (IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) ||
2171 	    bpf_prog_has_kfunc_call(fp))
2172 		jit_needed = true;
2173 
2174 	bpf_prog_select_func(fp);
2175 
2176 	/* eBPF JITs can rewrite the program in case constant
2177 	 * blinding is active. However, in case of error during
2178 	 * blinding, bpf_int_jit_compile() must always return a
2179 	 * valid program, which in this case would simply not
2180 	 * be JITed, but falls back to the interpreter.
2181 	 */
2182 	if (!bpf_prog_is_dev_bound(fp->aux)) {
2183 		*err = bpf_prog_alloc_jited_linfo(fp);
2184 		if (*err)
2185 			return fp;
2186 
2187 		fp = bpf_int_jit_compile(fp);
2188 		bpf_prog_jit_attempt_done(fp);
2189 		if (!fp->jited && jit_needed) {
2190 			*err = -ENOTSUPP;
2191 			return fp;
2192 		}
2193 	} else {
2194 		*err = bpf_prog_offload_compile(fp);
2195 		if (*err)
2196 			return fp;
2197 	}
2198 
2199 finalize:
2200 	bpf_prog_lock_ro(fp);
2201 
2202 	/* The tail call compatibility check can only be done at
2203 	 * this late stage as we need to determine, if we deal
2204 	 * with JITed or non JITed program concatenations and not
2205 	 * all eBPF JITs might immediately support all features.
2206 	 */
2207 	*err = bpf_check_tail_call(fp);
2208 
2209 	return fp;
2210 }
2211 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
2212 
2213 static unsigned int __bpf_prog_ret1(const void *ctx,
2214 				    const struct bpf_insn *insn)
2215 {
2216 	return 1;
2217 }
2218 
2219 static struct bpf_prog_dummy {
2220 	struct bpf_prog prog;
2221 } dummy_bpf_prog = {
2222 	.prog = {
2223 		.bpf_func = __bpf_prog_ret1,
2224 	},
2225 };
2226 
2227 struct bpf_empty_prog_array bpf_empty_prog_array = {
2228 	.null_prog = NULL,
2229 };
2230 EXPORT_SYMBOL(bpf_empty_prog_array);
2231 
2232 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
2233 {
2234 	if (prog_cnt)
2235 		return kzalloc(sizeof(struct bpf_prog_array) +
2236 			       sizeof(struct bpf_prog_array_item) *
2237 			       (prog_cnt + 1),
2238 			       flags);
2239 
2240 	return &bpf_empty_prog_array.hdr;
2241 }
2242 
2243 void bpf_prog_array_free(struct bpf_prog_array *progs)
2244 {
2245 	if (!progs || progs == &bpf_empty_prog_array.hdr)
2246 		return;
2247 	kfree_rcu(progs, rcu);
2248 }
2249 
2250 static void __bpf_prog_array_free_sleepable_cb(struct rcu_head *rcu)
2251 {
2252 	struct bpf_prog_array *progs;
2253 
2254 	progs = container_of(rcu, struct bpf_prog_array, rcu);
2255 	kfree_rcu(progs, rcu);
2256 }
2257 
2258 void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs)
2259 {
2260 	if (!progs || progs == &bpf_empty_prog_array.hdr)
2261 		return;
2262 	call_rcu_tasks_trace(&progs->rcu, __bpf_prog_array_free_sleepable_cb);
2263 }
2264 
2265 int bpf_prog_array_length(struct bpf_prog_array *array)
2266 {
2267 	struct bpf_prog_array_item *item;
2268 	u32 cnt = 0;
2269 
2270 	for (item = array->items; item->prog; item++)
2271 		if (item->prog != &dummy_bpf_prog.prog)
2272 			cnt++;
2273 	return cnt;
2274 }
2275 
2276 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
2277 {
2278 	struct bpf_prog_array_item *item;
2279 
2280 	for (item = array->items; item->prog; item++)
2281 		if (item->prog != &dummy_bpf_prog.prog)
2282 			return false;
2283 	return true;
2284 }
2285 
2286 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
2287 				     u32 *prog_ids,
2288 				     u32 request_cnt)
2289 {
2290 	struct bpf_prog_array_item *item;
2291 	int i = 0;
2292 
2293 	for (item = array->items; item->prog; item++) {
2294 		if (item->prog == &dummy_bpf_prog.prog)
2295 			continue;
2296 		prog_ids[i] = item->prog->aux->id;
2297 		if (++i == request_cnt) {
2298 			item++;
2299 			break;
2300 		}
2301 	}
2302 
2303 	return !!(item->prog);
2304 }
2305 
2306 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
2307 				__u32 __user *prog_ids, u32 cnt)
2308 {
2309 	unsigned long err = 0;
2310 	bool nospc;
2311 	u32 *ids;
2312 
2313 	/* users of this function are doing:
2314 	 * cnt = bpf_prog_array_length();
2315 	 * if (cnt > 0)
2316 	 *     bpf_prog_array_copy_to_user(..., cnt);
2317 	 * so below kcalloc doesn't need extra cnt > 0 check.
2318 	 */
2319 	ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
2320 	if (!ids)
2321 		return -ENOMEM;
2322 	nospc = bpf_prog_array_copy_core(array, ids, cnt);
2323 	err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
2324 	kfree(ids);
2325 	if (err)
2326 		return -EFAULT;
2327 	if (nospc)
2328 		return -ENOSPC;
2329 	return 0;
2330 }
2331 
2332 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
2333 				struct bpf_prog *old_prog)
2334 {
2335 	struct bpf_prog_array_item *item;
2336 
2337 	for (item = array->items; item->prog; item++)
2338 		if (item->prog == old_prog) {
2339 			WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
2340 			break;
2341 		}
2342 }
2343 
2344 /**
2345  * bpf_prog_array_delete_safe_at() - Replaces the program at the given
2346  *                                   index into the program array with
2347  *                                   a dummy no-op program.
2348  * @array: a bpf_prog_array
2349  * @index: the index of the program to replace
2350  *
2351  * Skips over dummy programs, by not counting them, when calculating
2352  * the position of the program to replace.
2353  *
2354  * Return:
2355  * * 0		- Success
2356  * * -EINVAL	- Invalid index value. Must be a non-negative integer.
2357  * * -ENOENT	- Index out of range
2358  */
2359 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
2360 {
2361 	return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
2362 }
2363 
2364 /**
2365  * bpf_prog_array_update_at() - Updates the program at the given index
2366  *                              into the program array.
2367  * @array: a bpf_prog_array
2368  * @index: the index of the program to update
2369  * @prog: the program to insert into the array
2370  *
2371  * Skips over dummy programs, by not counting them, when calculating
2372  * the position of the program to update.
2373  *
2374  * Return:
2375  * * 0		- Success
2376  * * -EINVAL	- Invalid index value. Must be a non-negative integer.
2377  * * -ENOENT	- Index out of range
2378  */
2379 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2380 			     struct bpf_prog *prog)
2381 {
2382 	struct bpf_prog_array_item *item;
2383 
2384 	if (unlikely(index < 0))
2385 		return -EINVAL;
2386 
2387 	for (item = array->items; item->prog; item++) {
2388 		if (item->prog == &dummy_bpf_prog.prog)
2389 			continue;
2390 		if (!index) {
2391 			WRITE_ONCE(item->prog, prog);
2392 			return 0;
2393 		}
2394 		index--;
2395 	}
2396 	return -ENOENT;
2397 }
2398 
2399 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2400 			struct bpf_prog *exclude_prog,
2401 			struct bpf_prog *include_prog,
2402 			u64 bpf_cookie,
2403 			struct bpf_prog_array **new_array)
2404 {
2405 	int new_prog_cnt, carry_prog_cnt = 0;
2406 	struct bpf_prog_array_item *existing, *new;
2407 	struct bpf_prog_array *array;
2408 	bool found_exclude = false;
2409 
2410 	/* Figure out how many existing progs we need to carry over to
2411 	 * the new array.
2412 	 */
2413 	if (old_array) {
2414 		existing = old_array->items;
2415 		for (; existing->prog; existing++) {
2416 			if (existing->prog == exclude_prog) {
2417 				found_exclude = true;
2418 				continue;
2419 			}
2420 			if (existing->prog != &dummy_bpf_prog.prog)
2421 				carry_prog_cnt++;
2422 			if (existing->prog == include_prog)
2423 				return -EEXIST;
2424 		}
2425 	}
2426 
2427 	if (exclude_prog && !found_exclude)
2428 		return -ENOENT;
2429 
2430 	/* How many progs (not NULL) will be in the new array? */
2431 	new_prog_cnt = carry_prog_cnt;
2432 	if (include_prog)
2433 		new_prog_cnt += 1;
2434 
2435 	/* Do we have any prog (not NULL) in the new array? */
2436 	if (!new_prog_cnt) {
2437 		*new_array = NULL;
2438 		return 0;
2439 	}
2440 
2441 	/* +1 as the end of prog_array is marked with NULL */
2442 	array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2443 	if (!array)
2444 		return -ENOMEM;
2445 	new = array->items;
2446 
2447 	/* Fill in the new prog array */
2448 	if (carry_prog_cnt) {
2449 		existing = old_array->items;
2450 		for (; existing->prog; existing++) {
2451 			if (existing->prog == exclude_prog ||
2452 			    existing->prog == &dummy_bpf_prog.prog)
2453 				continue;
2454 
2455 			new->prog = existing->prog;
2456 			new->bpf_cookie = existing->bpf_cookie;
2457 			new++;
2458 		}
2459 	}
2460 	if (include_prog) {
2461 		new->prog = include_prog;
2462 		new->bpf_cookie = bpf_cookie;
2463 		new++;
2464 	}
2465 	new->prog = NULL;
2466 	*new_array = array;
2467 	return 0;
2468 }
2469 
2470 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2471 			     u32 *prog_ids, u32 request_cnt,
2472 			     u32 *prog_cnt)
2473 {
2474 	u32 cnt = 0;
2475 
2476 	if (array)
2477 		cnt = bpf_prog_array_length(array);
2478 
2479 	*prog_cnt = cnt;
2480 
2481 	/* return early if user requested only program count or nothing to copy */
2482 	if (!request_cnt || !cnt)
2483 		return 0;
2484 
2485 	/* this function is called under trace/bpf_trace.c: bpf_event_mutex */
2486 	return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
2487 								     : 0;
2488 }
2489 
2490 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2491 			  struct bpf_map **used_maps, u32 len)
2492 {
2493 	struct bpf_map *map;
2494 	u32 i;
2495 
2496 	for (i = 0; i < len; i++) {
2497 		map = used_maps[i];
2498 		if (map->ops->map_poke_untrack)
2499 			map->ops->map_poke_untrack(map, aux);
2500 		bpf_map_put(map);
2501 	}
2502 }
2503 
2504 static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2505 {
2506 	__bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
2507 	kfree(aux->used_maps);
2508 }
2509 
2510 void __bpf_free_used_btfs(struct bpf_prog_aux *aux,
2511 			  struct btf_mod_pair *used_btfs, u32 len)
2512 {
2513 #ifdef CONFIG_BPF_SYSCALL
2514 	struct btf_mod_pair *btf_mod;
2515 	u32 i;
2516 
2517 	for (i = 0; i < len; i++) {
2518 		btf_mod = &used_btfs[i];
2519 		if (btf_mod->module)
2520 			module_put(btf_mod->module);
2521 		btf_put(btf_mod->btf);
2522 	}
2523 #endif
2524 }
2525 
2526 static void bpf_free_used_btfs(struct bpf_prog_aux *aux)
2527 {
2528 	__bpf_free_used_btfs(aux, aux->used_btfs, aux->used_btf_cnt);
2529 	kfree(aux->used_btfs);
2530 }
2531 
2532 static void bpf_prog_free_deferred(struct work_struct *work)
2533 {
2534 	struct bpf_prog_aux *aux;
2535 	int i;
2536 
2537 	aux = container_of(work, struct bpf_prog_aux, work);
2538 #ifdef CONFIG_BPF_SYSCALL
2539 	bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
2540 #endif
2541 #ifdef CONFIG_CGROUP_BPF
2542 	if (aux->cgroup_atype != CGROUP_BPF_ATTACH_TYPE_INVALID)
2543 		bpf_cgroup_atype_put(aux->cgroup_atype);
2544 #endif
2545 	bpf_free_used_maps(aux);
2546 	bpf_free_used_btfs(aux);
2547 	if (bpf_prog_is_dev_bound(aux))
2548 		bpf_prog_offload_destroy(aux->prog);
2549 #ifdef CONFIG_PERF_EVENTS
2550 	if (aux->prog->has_callchain_buf)
2551 		put_callchain_buffers();
2552 #endif
2553 	if (aux->dst_trampoline)
2554 		bpf_trampoline_put(aux->dst_trampoline);
2555 	for (i = 0; i < aux->func_cnt; i++) {
2556 		/* We can just unlink the subprog poke descriptor table as
2557 		 * it was originally linked to the main program and is also
2558 		 * released along with it.
2559 		 */
2560 		aux->func[i]->aux->poke_tab = NULL;
2561 		bpf_jit_free(aux->func[i]);
2562 	}
2563 	if (aux->func_cnt) {
2564 		kfree(aux->func);
2565 		bpf_prog_unlock_free(aux->prog);
2566 	} else {
2567 		bpf_jit_free(aux->prog);
2568 	}
2569 }
2570 
2571 void bpf_prog_free(struct bpf_prog *fp)
2572 {
2573 	struct bpf_prog_aux *aux = fp->aux;
2574 
2575 	if (aux->dst_prog)
2576 		bpf_prog_put(aux->dst_prog);
2577 	INIT_WORK(&aux->work, bpf_prog_free_deferred);
2578 	schedule_work(&aux->work);
2579 }
2580 EXPORT_SYMBOL_GPL(bpf_prog_free);
2581 
2582 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2583 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2584 
2585 void bpf_user_rnd_init_once(void)
2586 {
2587 	prandom_init_once(&bpf_user_rnd_state);
2588 }
2589 
2590 BPF_CALL_0(bpf_user_rnd_u32)
2591 {
2592 	/* Should someone ever have the rather unwise idea to use some
2593 	 * of the registers passed into this function, then note that
2594 	 * this function is called from native eBPF and classic-to-eBPF
2595 	 * transformations. Register assignments from both sides are
2596 	 * different, f.e. classic always sets fn(ctx, A, X) here.
2597 	 */
2598 	struct rnd_state *state;
2599 	u32 res;
2600 
2601 	state = &get_cpu_var(bpf_user_rnd_state);
2602 	res = prandom_u32_state(state);
2603 	put_cpu_var(bpf_user_rnd_state);
2604 
2605 	return res;
2606 }
2607 
2608 BPF_CALL_0(bpf_get_raw_cpu_id)
2609 {
2610 	return raw_smp_processor_id();
2611 }
2612 
2613 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2614 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2615 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2616 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2617 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2618 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2619 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2620 const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto __weak;
2621 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2622 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2623 const struct bpf_func_proto bpf_jiffies64_proto __weak;
2624 
2625 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2626 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2627 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2628 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2629 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
2630 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto __weak;
2631 const struct bpf_func_proto bpf_ktime_get_tai_ns_proto __weak;
2632 
2633 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2634 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2635 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2636 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2637 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
2638 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2639 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
2640 const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
2641 const struct bpf_func_proto bpf_seq_printf_btf_proto __weak;
2642 const struct bpf_func_proto bpf_set_retval_proto __weak;
2643 const struct bpf_func_proto bpf_get_retval_proto __weak;
2644 
2645 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2646 {
2647 	return NULL;
2648 }
2649 
2650 const struct bpf_func_proto * __weak bpf_get_trace_vprintk_proto(void)
2651 {
2652 	return NULL;
2653 }
2654 
2655 u64 __weak
2656 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2657 		 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2658 {
2659 	return -ENOTSUPP;
2660 }
2661 EXPORT_SYMBOL_GPL(bpf_event_output);
2662 
2663 /* Always built-in helper functions. */
2664 const struct bpf_func_proto bpf_tail_call_proto = {
2665 	.func		= NULL,
2666 	.gpl_only	= false,
2667 	.ret_type	= RET_VOID,
2668 	.arg1_type	= ARG_PTR_TO_CTX,
2669 	.arg2_type	= ARG_CONST_MAP_PTR,
2670 	.arg3_type	= ARG_ANYTHING,
2671 };
2672 
2673 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2674  * It is encouraged to implement bpf_int_jit_compile() instead, so that
2675  * eBPF and implicitly also cBPF can get JITed!
2676  */
2677 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2678 {
2679 	return prog;
2680 }
2681 
2682 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2683  * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2684  */
2685 void __weak bpf_jit_compile(struct bpf_prog *prog)
2686 {
2687 }
2688 
2689 bool __weak bpf_helper_changes_pkt_data(void *func)
2690 {
2691 	return false;
2692 }
2693 
2694 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2695  * analysis code and wants explicit zero extension inserted by verifier.
2696  * Otherwise, return FALSE.
2697  *
2698  * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
2699  * you don't override this. JITs that don't want these extra insns can detect
2700  * them using insn_is_zext.
2701  */
2702 bool __weak bpf_jit_needs_zext(void)
2703 {
2704 	return false;
2705 }
2706 
2707 /* Return TRUE if the JIT backend supports mixing bpf2bpf and tailcalls. */
2708 bool __weak bpf_jit_supports_subprog_tailcalls(void)
2709 {
2710 	return false;
2711 }
2712 
2713 bool __weak bpf_jit_supports_kfunc_call(void)
2714 {
2715 	return false;
2716 }
2717 
2718 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2719  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2720  */
2721 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2722 			 int len)
2723 {
2724 	return -EFAULT;
2725 }
2726 
2727 int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
2728 			      void *addr1, void *addr2)
2729 {
2730 	return -ENOTSUPP;
2731 }
2732 
2733 void * __weak bpf_arch_text_copy(void *dst, void *src, size_t len)
2734 {
2735 	return ERR_PTR(-ENOTSUPP);
2736 }
2737 
2738 int __weak bpf_arch_text_invalidate(void *dst, size_t len)
2739 {
2740 	return -ENOTSUPP;
2741 }
2742 
2743 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2744 EXPORT_SYMBOL(bpf_stats_enabled_key);
2745 
2746 /* All definitions of tracepoints related to BPF. */
2747 #define CREATE_TRACE_POINTS
2748 #include <linux/bpf_trace.h>
2749 
2750 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2751 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
2752