xref: /linux/kernel/bpf/core.c (revision 1771c8c9e65a20128f93df107353a5f4cb91546a)
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 		ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
75 
76 	if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
77 		return ptr;
78 
79 	return NULL;
80 }
81 
82 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
83 {
84 	gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
85 	struct bpf_prog_aux *aux;
86 	struct bpf_prog *fp;
87 
88 	size = round_up(size, PAGE_SIZE);
89 	fp = __vmalloc(size, gfp_flags);
90 	if (fp == NULL)
91 		return NULL;
92 
93 	aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT | gfp_extra_flags);
94 	if (aux == NULL) {
95 		vfree(fp);
96 		return NULL;
97 	}
98 	fp->active = alloc_percpu_gfp(int, GFP_KERNEL_ACCOUNT | gfp_extra_flags);
99 	if (!fp->active) {
100 		vfree(fp);
101 		kfree(aux);
102 		return NULL;
103 	}
104 
105 	fp->pages = size / PAGE_SIZE;
106 	fp->aux = aux;
107 	fp->aux->prog = fp;
108 	fp->jit_requested = ebpf_jit_enabled();
109 	fp->blinding_requested = bpf_jit_blinding_enabled(fp);
110 
111 	INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
112 	mutex_init(&fp->aux->used_maps_mutex);
113 	mutex_init(&fp->aux->dst_mutex);
114 
115 	return fp;
116 }
117 
118 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
119 {
120 	gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
121 	struct bpf_prog *prog;
122 	int cpu;
123 
124 	prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
125 	if (!prog)
126 		return NULL;
127 
128 	prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
129 	if (!prog->stats) {
130 		free_percpu(prog->active);
131 		kfree(prog->aux);
132 		vfree(prog);
133 		return NULL;
134 	}
135 
136 	for_each_possible_cpu(cpu) {
137 		struct bpf_prog_stats *pstats;
138 
139 		pstats = per_cpu_ptr(prog->stats, cpu);
140 		u64_stats_init(&pstats->syncp);
141 	}
142 	return prog;
143 }
144 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
145 
146 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
147 {
148 	if (!prog->aux->nr_linfo || !prog->jit_requested)
149 		return 0;
150 
151 	prog->aux->jited_linfo = kvcalloc(prog->aux->nr_linfo,
152 					  sizeof(*prog->aux->jited_linfo),
153 					  GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
154 	if (!prog->aux->jited_linfo)
155 		return -ENOMEM;
156 
157 	return 0;
158 }
159 
160 void bpf_prog_jit_attempt_done(struct bpf_prog *prog)
161 {
162 	if (prog->aux->jited_linfo &&
163 	    (!prog->jited || !prog->aux->jited_linfo[0])) {
164 		kvfree(prog->aux->jited_linfo);
165 		prog->aux->jited_linfo = NULL;
166 	}
167 
168 	kfree(prog->aux->kfunc_tab);
169 	prog->aux->kfunc_tab = NULL;
170 }
171 
172 /* The jit engine is responsible to provide an array
173  * for insn_off to the jited_off mapping (insn_to_jit_off).
174  *
175  * The idx to this array is the insn_off.  Hence, the insn_off
176  * here is relative to the prog itself instead of the main prog.
177  * This array has one entry for each xlated bpf insn.
178  *
179  * jited_off is the byte off to the last byte of the jited insn.
180  *
181  * Hence, with
182  * insn_start:
183  *      The first bpf insn off of the prog.  The insn off
184  *      here is relative to the main prog.
185  *      e.g. if prog is a subprog, insn_start > 0
186  * linfo_idx:
187  *      The prog's idx to prog->aux->linfo and jited_linfo
188  *
189  * jited_linfo[linfo_idx] = prog->bpf_func
190  *
191  * For i > linfo_idx,
192  *
193  * jited_linfo[i] = prog->bpf_func +
194  *	insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
195  */
196 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
197 			       const u32 *insn_to_jit_off)
198 {
199 	u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
200 	const struct bpf_line_info *linfo;
201 	void **jited_linfo;
202 
203 	if (!prog->aux->jited_linfo)
204 		/* Userspace did not provide linfo */
205 		return;
206 
207 	linfo_idx = prog->aux->linfo_idx;
208 	linfo = &prog->aux->linfo[linfo_idx];
209 	insn_start = linfo[0].insn_off;
210 	insn_end = insn_start + prog->len;
211 
212 	jited_linfo = &prog->aux->jited_linfo[linfo_idx];
213 	jited_linfo[0] = prog->bpf_func;
214 
215 	nr_linfo = prog->aux->nr_linfo - linfo_idx;
216 
217 	for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
218 		/* The verifier ensures that linfo[i].insn_off is
219 		 * strictly increasing
220 		 */
221 		jited_linfo[i] = prog->bpf_func +
222 			insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
223 }
224 
225 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
226 				  gfp_t gfp_extra_flags)
227 {
228 	gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
229 	struct bpf_prog *fp;
230 	u32 pages;
231 
232 	size = round_up(size, PAGE_SIZE);
233 	pages = size / PAGE_SIZE;
234 	if (pages <= fp_old->pages)
235 		return fp_old;
236 
237 	fp = __vmalloc(size, gfp_flags);
238 	if (fp) {
239 		memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
240 		fp->pages = pages;
241 		fp->aux->prog = fp;
242 
243 		/* We keep fp->aux from fp_old around in the new
244 		 * reallocated structure.
245 		 */
246 		fp_old->aux = NULL;
247 		fp_old->stats = NULL;
248 		fp_old->active = NULL;
249 		__bpf_prog_free(fp_old);
250 	}
251 
252 	return fp;
253 }
254 
255 void __bpf_prog_free(struct bpf_prog *fp)
256 {
257 	if (fp->aux) {
258 		mutex_destroy(&fp->aux->used_maps_mutex);
259 		mutex_destroy(&fp->aux->dst_mutex);
260 		kfree(fp->aux->poke_tab);
261 		kfree(fp->aux);
262 	}
263 	free_percpu(fp->stats);
264 	free_percpu(fp->active);
265 	vfree(fp);
266 }
267 
268 int bpf_prog_calc_tag(struct bpf_prog *fp)
269 {
270 	const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
271 	u32 raw_size = bpf_prog_tag_scratch_size(fp);
272 	u32 digest[SHA1_DIGEST_WORDS];
273 	u32 ws[SHA1_WORKSPACE_WORDS];
274 	u32 i, bsize, psize, blocks;
275 	struct bpf_insn *dst;
276 	bool was_ld_map;
277 	u8 *raw, *todo;
278 	__be32 *result;
279 	__be64 *bits;
280 
281 	raw = vmalloc(raw_size);
282 	if (!raw)
283 		return -ENOMEM;
284 
285 	sha1_init(digest);
286 	memset(ws, 0, sizeof(ws));
287 
288 	/* We need to take out the map fd for the digest calculation
289 	 * since they are unstable from user space side.
290 	 */
291 	dst = (void *)raw;
292 	for (i = 0, was_ld_map = false; i < fp->len; i++) {
293 		dst[i] = fp->insnsi[i];
294 		if (!was_ld_map &&
295 		    dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
296 		    (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
297 		     dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
298 			was_ld_map = true;
299 			dst[i].imm = 0;
300 		} else if (was_ld_map &&
301 			   dst[i].code == 0 &&
302 			   dst[i].dst_reg == 0 &&
303 			   dst[i].src_reg == 0 &&
304 			   dst[i].off == 0) {
305 			was_ld_map = false;
306 			dst[i].imm = 0;
307 		} else {
308 			was_ld_map = false;
309 		}
310 	}
311 
312 	psize = bpf_prog_insn_size(fp);
313 	memset(&raw[psize], 0, raw_size - psize);
314 	raw[psize++] = 0x80;
315 
316 	bsize  = round_up(psize, SHA1_BLOCK_SIZE);
317 	blocks = bsize / SHA1_BLOCK_SIZE;
318 	todo   = raw;
319 	if (bsize - psize >= sizeof(__be64)) {
320 		bits = (__be64 *)(todo + bsize - sizeof(__be64));
321 	} else {
322 		bits = (__be64 *)(todo + bsize + bits_offset);
323 		blocks++;
324 	}
325 	*bits = cpu_to_be64((psize - 1) << 3);
326 
327 	while (blocks--) {
328 		sha1_transform(digest, todo, ws);
329 		todo += SHA1_BLOCK_SIZE;
330 	}
331 
332 	result = (__force __be32 *)digest;
333 	for (i = 0; i < SHA1_DIGEST_WORDS; i++)
334 		result[i] = cpu_to_be32(digest[i]);
335 	memcpy(fp->tag, result, sizeof(fp->tag));
336 
337 	vfree(raw);
338 	return 0;
339 }
340 
341 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
342 				s32 end_new, s32 curr, const bool probe_pass)
343 {
344 	const s64 imm_min = S32_MIN, imm_max = S32_MAX;
345 	s32 delta = end_new - end_old;
346 	s64 imm = insn->imm;
347 
348 	if (curr < pos && curr + imm + 1 >= end_old)
349 		imm += delta;
350 	else if (curr >= end_new && curr + imm + 1 < end_new)
351 		imm -= delta;
352 	if (imm < imm_min || imm > imm_max)
353 		return -ERANGE;
354 	if (!probe_pass)
355 		insn->imm = imm;
356 	return 0;
357 }
358 
359 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
360 				s32 end_new, s32 curr, const bool probe_pass)
361 {
362 	const s32 off_min = S16_MIN, off_max = S16_MAX;
363 	s32 delta = end_new - end_old;
364 	s32 off = insn->off;
365 
366 	if (curr < pos && curr + off + 1 >= end_old)
367 		off += delta;
368 	else if (curr >= end_new && curr + off + 1 < end_new)
369 		off -= delta;
370 	if (off < off_min || off > off_max)
371 		return -ERANGE;
372 	if (!probe_pass)
373 		insn->off = off;
374 	return 0;
375 }
376 
377 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
378 			    s32 end_new, const bool probe_pass)
379 {
380 	u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
381 	struct bpf_insn *insn = prog->insnsi;
382 	int ret = 0;
383 
384 	for (i = 0; i < insn_cnt; i++, insn++) {
385 		u8 code;
386 
387 		/* In the probing pass we still operate on the original,
388 		 * unpatched image in order to check overflows before we
389 		 * do any other adjustments. Therefore skip the patchlet.
390 		 */
391 		if (probe_pass && i == pos) {
392 			i = end_new;
393 			insn = prog->insnsi + end_old;
394 		}
395 		if (bpf_pseudo_func(insn)) {
396 			ret = bpf_adj_delta_to_imm(insn, pos, end_old,
397 						   end_new, i, probe_pass);
398 			if (ret)
399 				return ret;
400 			continue;
401 		}
402 		code = insn->code;
403 		if ((BPF_CLASS(code) != BPF_JMP &&
404 		     BPF_CLASS(code) != BPF_JMP32) ||
405 		    BPF_OP(code) == BPF_EXIT)
406 			continue;
407 		/* Adjust offset of jmps if we cross patch boundaries. */
408 		if (BPF_OP(code) == BPF_CALL) {
409 			if (insn->src_reg != BPF_PSEUDO_CALL)
410 				continue;
411 			ret = bpf_adj_delta_to_imm(insn, pos, end_old,
412 						   end_new, i, probe_pass);
413 		} else {
414 			ret = bpf_adj_delta_to_off(insn, pos, end_old,
415 						   end_new, i, probe_pass);
416 		}
417 		if (ret)
418 			break;
419 	}
420 
421 	return ret;
422 }
423 
424 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
425 {
426 	struct bpf_line_info *linfo;
427 	u32 i, nr_linfo;
428 
429 	nr_linfo = prog->aux->nr_linfo;
430 	if (!nr_linfo || !delta)
431 		return;
432 
433 	linfo = prog->aux->linfo;
434 
435 	for (i = 0; i < nr_linfo; i++)
436 		if (off < linfo[i].insn_off)
437 			break;
438 
439 	/* Push all off < linfo[i].insn_off by delta */
440 	for (; i < nr_linfo; i++)
441 		linfo[i].insn_off += delta;
442 }
443 
444 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
445 				       const struct bpf_insn *patch, u32 len)
446 {
447 	u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
448 	const u32 cnt_max = S16_MAX;
449 	struct bpf_prog *prog_adj;
450 	int err;
451 
452 	/* Since our patchlet doesn't expand the image, we're done. */
453 	if (insn_delta == 0) {
454 		memcpy(prog->insnsi + off, patch, sizeof(*patch));
455 		return prog;
456 	}
457 
458 	insn_adj_cnt = prog->len + insn_delta;
459 
460 	/* Reject anything that would potentially let the insn->off
461 	 * target overflow when we have excessive program expansions.
462 	 * We need to probe here before we do any reallocation where
463 	 * we afterwards may not fail anymore.
464 	 */
465 	if (insn_adj_cnt > cnt_max &&
466 	    (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
467 		return ERR_PTR(err);
468 
469 	/* Several new instructions need to be inserted. Make room
470 	 * for them. Likely, there's no need for a new allocation as
471 	 * last page could have large enough tailroom.
472 	 */
473 	prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
474 				    GFP_USER);
475 	if (!prog_adj)
476 		return ERR_PTR(-ENOMEM);
477 
478 	prog_adj->len = insn_adj_cnt;
479 
480 	/* Patching happens in 3 steps:
481 	 *
482 	 * 1) Move over tail of insnsi from next instruction onwards,
483 	 *    so we can patch the single target insn with one or more
484 	 *    new ones (patching is always from 1 to n insns, n > 0).
485 	 * 2) Inject new instructions at the target location.
486 	 * 3) Adjust branch offsets if necessary.
487 	 */
488 	insn_rest = insn_adj_cnt - off - len;
489 
490 	memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
491 		sizeof(*patch) * insn_rest);
492 	memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
493 
494 	/* We are guaranteed to not fail at this point, otherwise
495 	 * the ship has sailed to reverse to the original state. An
496 	 * overflow cannot happen at this point.
497 	 */
498 	BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
499 
500 	bpf_adj_linfo(prog_adj, off, insn_delta);
501 
502 	return prog_adj;
503 }
504 
505 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
506 {
507 	/* Branch offsets can't overflow when program is shrinking, no need
508 	 * to call bpf_adj_branches(..., true) here
509 	 */
510 	memmove(prog->insnsi + off, prog->insnsi + off + cnt,
511 		sizeof(struct bpf_insn) * (prog->len - off - cnt));
512 	prog->len -= cnt;
513 
514 	return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
515 }
516 
517 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
518 {
519 	int i;
520 
521 	for (i = 0; i < fp->aux->func_cnt; i++)
522 		bpf_prog_kallsyms_del(fp->aux->func[i]);
523 }
524 
525 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
526 {
527 	bpf_prog_kallsyms_del_subprogs(fp);
528 	bpf_prog_kallsyms_del(fp);
529 }
530 
531 #ifdef CONFIG_BPF_JIT
532 /* All BPF JIT sysctl knobs here. */
533 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
534 int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
535 int bpf_jit_harden   __read_mostly;
536 long bpf_jit_limit   __read_mostly;
537 long bpf_jit_limit_max __read_mostly;
538 
539 static void
540 bpf_prog_ksym_set_addr(struct bpf_prog *prog)
541 {
542 	WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
543 
544 	prog->aux->ksym.start = (unsigned long) prog->bpf_func;
545 	prog->aux->ksym.end   = prog->aux->ksym.start + prog->jited_len;
546 }
547 
548 static void
549 bpf_prog_ksym_set_name(struct bpf_prog *prog)
550 {
551 	char *sym = prog->aux->ksym.name;
552 	const char *end = sym + KSYM_NAME_LEN;
553 	const struct btf_type *type;
554 	const char *func_name;
555 
556 	BUILD_BUG_ON(sizeof("bpf_prog_") +
557 		     sizeof(prog->tag) * 2 +
558 		     /* name has been null terminated.
559 		      * We should need +1 for the '_' preceding
560 		      * the name.  However, the null character
561 		      * is double counted between the name and the
562 		      * sizeof("bpf_prog_") above, so we omit
563 		      * the +1 here.
564 		      */
565 		     sizeof(prog->aux->name) > KSYM_NAME_LEN);
566 
567 	sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
568 	sym  = bin2hex(sym, prog->tag, sizeof(prog->tag));
569 
570 	/* prog->aux->name will be ignored if full btf name is available */
571 	if (prog->aux->func_info_cnt) {
572 		type = btf_type_by_id(prog->aux->btf,
573 				      prog->aux->func_info[prog->aux->func_idx].type_id);
574 		func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
575 		snprintf(sym, (size_t)(end - sym), "_%s", func_name);
576 		return;
577 	}
578 
579 	if (prog->aux->name[0])
580 		snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
581 	else
582 		*sym = 0;
583 }
584 
585 static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
586 {
587 	return container_of(n, struct bpf_ksym, tnode)->start;
588 }
589 
590 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
591 					  struct latch_tree_node *b)
592 {
593 	return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
594 }
595 
596 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
597 {
598 	unsigned long val = (unsigned long)key;
599 	const struct bpf_ksym *ksym;
600 
601 	ksym = container_of(n, struct bpf_ksym, tnode);
602 
603 	if (val < ksym->start)
604 		return -1;
605 	if (val >= ksym->end)
606 		return  1;
607 
608 	return 0;
609 }
610 
611 static const struct latch_tree_ops bpf_tree_ops = {
612 	.less	= bpf_tree_less,
613 	.comp	= bpf_tree_comp,
614 };
615 
616 static DEFINE_SPINLOCK(bpf_lock);
617 static LIST_HEAD(bpf_kallsyms);
618 static struct latch_tree_root bpf_tree __cacheline_aligned;
619 
620 void bpf_ksym_add(struct bpf_ksym *ksym)
621 {
622 	spin_lock_bh(&bpf_lock);
623 	WARN_ON_ONCE(!list_empty(&ksym->lnode));
624 	list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
625 	latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
626 	spin_unlock_bh(&bpf_lock);
627 }
628 
629 static void __bpf_ksym_del(struct bpf_ksym *ksym)
630 {
631 	if (list_empty(&ksym->lnode))
632 		return;
633 
634 	latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
635 	list_del_rcu(&ksym->lnode);
636 }
637 
638 void bpf_ksym_del(struct bpf_ksym *ksym)
639 {
640 	spin_lock_bh(&bpf_lock);
641 	__bpf_ksym_del(ksym);
642 	spin_unlock_bh(&bpf_lock);
643 }
644 
645 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
646 {
647 	return fp->jited && !bpf_prog_was_classic(fp);
648 }
649 
650 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
651 {
652 	return list_empty(&fp->aux->ksym.lnode) ||
653 	       fp->aux->ksym.lnode.prev == LIST_POISON2;
654 }
655 
656 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
657 {
658 	if (!bpf_prog_kallsyms_candidate(fp) ||
659 	    !bpf_capable())
660 		return;
661 
662 	bpf_prog_ksym_set_addr(fp);
663 	bpf_prog_ksym_set_name(fp);
664 	fp->aux->ksym.prog = true;
665 
666 	bpf_ksym_add(&fp->aux->ksym);
667 }
668 
669 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
670 {
671 	if (!bpf_prog_kallsyms_candidate(fp))
672 		return;
673 
674 	bpf_ksym_del(&fp->aux->ksym);
675 }
676 
677 static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
678 {
679 	struct latch_tree_node *n;
680 
681 	n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
682 	return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
683 }
684 
685 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
686 				 unsigned long *off, char *sym)
687 {
688 	struct bpf_ksym *ksym;
689 	char *ret = NULL;
690 
691 	rcu_read_lock();
692 	ksym = bpf_ksym_find(addr);
693 	if (ksym) {
694 		unsigned long symbol_start = ksym->start;
695 		unsigned long symbol_end = ksym->end;
696 
697 		strncpy(sym, ksym->name, KSYM_NAME_LEN);
698 
699 		ret = sym;
700 		if (size)
701 			*size = symbol_end - symbol_start;
702 		if (off)
703 			*off  = addr - symbol_start;
704 	}
705 	rcu_read_unlock();
706 
707 	return ret;
708 }
709 
710 bool is_bpf_text_address(unsigned long addr)
711 {
712 	bool ret;
713 
714 	rcu_read_lock();
715 	ret = bpf_ksym_find(addr) != NULL;
716 	rcu_read_unlock();
717 
718 	return ret;
719 }
720 
721 static struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
722 {
723 	struct bpf_ksym *ksym = bpf_ksym_find(addr);
724 
725 	return ksym && ksym->prog ?
726 	       container_of(ksym, struct bpf_prog_aux, ksym)->prog :
727 	       NULL;
728 }
729 
730 const struct exception_table_entry *search_bpf_extables(unsigned long addr)
731 {
732 	const struct exception_table_entry *e = NULL;
733 	struct bpf_prog *prog;
734 
735 	rcu_read_lock();
736 	prog = bpf_prog_ksym_find(addr);
737 	if (!prog)
738 		goto out;
739 	if (!prog->aux->num_exentries)
740 		goto out;
741 
742 	e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
743 out:
744 	rcu_read_unlock();
745 	return e;
746 }
747 
748 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
749 		    char *sym)
750 {
751 	struct bpf_ksym *ksym;
752 	unsigned int it = 0;
753 	int ret = -ERANGE;
754 
755 	if (!bpf_jit_kallsyms_enabled())
756 		return ret;
757 
758 	rcu_read_lock();
759 	list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
760 		if (it++ != symnum)
761 			continue;
762 
763 		strncpy(sym, ksym->name, KSYM_NAME_LEN);
764 
765 		*value = ksym->start;
766 		*type  = BPF_SYM_ELF_TYPE;
767 
768 		ret = 0;
769 		break;
770 	}
771 	rcu_read_unlock();
772 
773 	return ret;
774 }
775 
776 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
777 				struct bpf_jit_poke_descriptor *poke)
778 {
779 	struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
780 	static const u32 poke_tab_max = 1024;
781 	u32 slot = prog->aux->size_poke_tab;
782 	u32 size = slot + 1;
783 
784 	if (size > poke_tab_max)
785 		return -ENOSPC;
786 	if (poke->tailcall_target || poke->tailcall_target_stable ||
787 	    poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
788 		return -EINVAL;
789 
790 	switch (poke->reason) {
791 	case BPF_POKE_REASON_TAIL_CALL:
792 		if (!poke->tail_call.map)
793 			return -EINVAL;
794 		break;
795 	default:
796 		return -EINVAL;
797 	}
798 
799 	tab = krealloc(tab, size * sizeof(*poke), GFP_KERNEL);
800 	if (!tab)
801 		return -ENOMEM;
802 
803 	memcpy(&tab[slot], poke, sizeof(*poke));
804 	prog->aux->size_poke_tab = size;
805 	prog->aux->poke_tab = tab;
806 
807 	return slot;
808 }
809 
810 /*
811  * BPF program pack allocator.
812  *
813  * Most BPF programs are pretty small. Allocating a hole page for each
814  * program is sometime a waste. Many small bpf program also adds pressure
815  * to instruction TLB. To solve this issue, we introduce a BPF program pack
816  * allocator. The prog_pack allocator uses HPAGE_PMD_SIZE page (2MB on x86)
817  * to host BPF programs.
818  */
819 #define BPF_PROG_CHUNK_SHIFT	6
820 #define BPF_PROG_CHUNK_SIZE	(1 << BPF_PROG_CHUNK_SHIFT)
821 #define BPF_PROG_CHUNK_MASK	(~(BPF_PROG_CHUNK_SIZE - 1))
822 
823 struct bpf_prog_pack {
824 	struct list_head list;
825 	void *ptr;
826 	unsigned long bitmap[];
827 };
828 
829 #define BPF_PROG_SIZE_TO_NBITS(size)	(round_up(size, BPF_PROG_CHUNK_SIZE) / BPF_PROG_CHUNK_SIZE)
830 
831 static size_t bpf_prog_pack_size = -1;
832 static size_t bpf_prog_pack_mask = -1;
833 
834 static int bpf_prog_chunk_count(void)
835 {
836 	WARN_ON_ONCE(bpf_prog_pack_size == -1);
837 	return bpf_prog_pack_size / BPF_PROG_CHUNK_SIZE;
838 }
839 
840 static DEFINE_MUTEX(pack_mutex);
841 static LIST_HEAD(pack_list);
842 
843 /* PMD_SIZE is not available in some special config, e.g. ARCH=arm with
844  * CONFIG_MMU=n. Use PAGE_SIZE in these cases.
845  */
846 #ifdef PMD_SIZE
847 #define BPF_HPAGE_SIZE PMD_SIZE
848 #define BPF_HPAGE_MASK PMD_MASK
849 #else
850 #define BPF_HPAGE_SIZE PAGE_SIZE
851 #define BPF_HPAGE_MASK PAGE_MASK
852 #endif
853 
854 static size_t select_bpf_prog_pack_size(void)
855 {
856 	size_t size;
857 	void *ptr;
858 
859 	size = BPF_HPAGE_SIZE * num_online_nodes();
860 	ptr = module_alloc(size);
861 
862 	/* Test whether we can get huge pages. If not just use PAGE_SIZE
863 	 * packs.
864 	 */
865 	if (!ptr || !is_vm_area_hugepages(ptr)) {
866 		size = PAGE_SIZE;
867 		bpf_prog_pack_mask = PAGE_MASK;
868 	} else {
869 		bpf_prog_pack_mask = BPF_HPAGE_MASK;
870 	}
871 
872 	vfree(ptr);
873 	return size;
874 }
875 
876 static struct bpf_prog_pack *alloc_new_pack(bpf_jit_fill_hole_t bpf_fill_ill_insns)
877 {
878 	struct bpf_prog_pack *pack;
879 
880 	pack = kzalloc(struct_size(pack, bitmap, BITS_TO_LONGS(bpf_prog_chunk_count())),
881 		       GFP_KERNEL);
882 	if (!pack)
883 		return NULL;
884 	pack->ptr = module_alloc(bpf_prog_pack_size);
885 	if (!pack->ptr) {
886 		kfree(pack);
887 		return NULL;
888 	}
889 	bpf_fill_ill_insns(pack->ptr, bpf_prog_pack_size);
890 	bitmap_zero(pack->bitmap, bpf_prog_pack_size / BPF_PROG_CHUNK_SIZE);
891 	list_add_tail(&pack->list, &pack_list);
892 
893 	set_vm_flush_reset_perms(pack->ptr);
894 	set_memory_ro((unsigned long)pack->ptr, bpf_prog_pack_size / PAGE_SIZE);
895 	set_memory_x((unsigned long)pack->ptr, bpf_prog_pack_size / PAGE_SIZE);
896 	return pack;
897 }
898 
899 static void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns)
900 {
901 	unsigned int nbits = BPF_PROG_SIZE_TO_NBITS(size);
902 	struct bpf_prog_pack *pack;
903 	unsigned long pos;
904 	void *ptr = NULL;
905 
906 	mutex_lock(&pack_mutex);
907 	if (bpf_prog_pack_size == -1)
908 		bpf_prog_pack_size = select_bpf_prog_pack_size();
909 
910 	if (size > bpf_prog_pack_size) {
911 		size = round_up(size, PAGE_SIZE);
912 		ptr = module_alloc(size);
913 		if (ptr) {
914 			bpf_fill_ill_insns(ptr, size);
915 			set_vm_flush_reset_perms(ptr);
916 			set_memory_ro((unsigned long)ptr, size / PAGE_SIZE);
917 			set_memory_x((unsigned long)ptr, size / PAGE_SIZE);
918 		}
919 		goto out;
920 	}
921 	list_for_each_entry(pack, &pack_list, list) {
922 		pos = bitmap_find_next_zero_area(pack->bitmap, bpf_prog_chunk_count(), 0,
923 						 nbits, 0);
924 		if (pos < bpf_prog_chunk_count())
925 			goto found_free_area;
926 	}
927 
928 	pack = alloc_new_pack(bpf_fill_ill_insns);
929 	if (!pack)
930 		goto out;
931 
932 	pos = 0;
933 
934 found_free_area:
935 	bitmap_set(pack->bitmap, pos, nbits);
936 	ptr = (void *)(pack->ptr) + (pos << BPF_PROG_CHUNK_SHIFT);
937 
938 out:
939 	mutex_unlock(&pack_mutex);
940 	return ptr;
941 }
942 
943 static void bpf_prog_pack_free(struct bpf_binary_header *hdr)
944 {
945 	struct bpf_prog_pack *pack = NULL, *tmp;
946 	unsigned int nbits;
947 	unsigned long pos;
948 	void *pack_ptr;
949 
950 	mutex_lock(&pack_mutex);
951 	if (hdr->size > bpf_prog_pack_size) {
952 		module_memfree(hdr);
953 		goto out;
954 	}
955 
956 	pack_ptr = (void *)((unsigned long)hdr & bpf_prog_pack_mask);
957 
958 	list_for_each_entry(tmp, &pack_list, list) {
959 		if (tmp->ptr == pack_ptr) {
960 			pack = tmp;
961 			break;
962 		}
963 	}
964 
965 	if (WARN_ONCE(!pack, "bpf_prog_pack bug\n"))
966 		goto out;
967 
968 	nbits = BPF_PROG_SIZE_TO_NBITS(hdr->size);
969 	pos = ((unsigned long)hdr - (unsigned long)pack_ptr) >> BPF_PROG_CHUNK_SHIFT;
970 
971 	WARN_ONCE(bpf_arch_text_invalidate(hdr, hdr->size),
972 		  "bpf_prog_pack bug: missing bpf_arch_text_invalidate?\n");
973 
974 	bitmap_clear(pack->bitmap, pos, nbits);
975 	if (bitmap_find_next_zero_area(pack->bitmap, bpf_prog_chunk_count(), 0,
976 				       bpf_prog_chunk_count(), 0) == 0) {
977 		list_del(&pack->list);
978 		module_memfree(pack->ptr);
979 		kfree(pack);
980 	}
981 out:
982 	mutex_unlock(&pack_mutex);
983 }
984 
985 static atomic_long_t bpf_jit_current;
986 
987 /* Can be overridden by an arch's JIT compiler if it has a custom,
988  * dedicated BPF backend memory area, or if neither of the two
989  * below apply.
990  */
991 u64 __weak bpf_jit_alloc_exec_limit(void)
992 {
993 #if defined(MODULES_VADDR)
994 	return MODULES_END - MODULES_VADDR;
995 #else
996 	return VMALLOC_END - VMALLOC_START;
997 #endif
998 }
999 
1000 static int __init bpf_jit_charge_init(void)
1001 {
1002 	/* Only used as heuristic here to derive limit. */
1003 	bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
1004 	bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
1005 					    PAGE_SIZE), LONG_MAX);
1006 	return 0;
1007 }
1008 pure_initcall(bpf_jit_charge_init);
1009 
1010 int bpf_jit_charge_modmem(u32 size)
1011 {
1012 	if (atomic_long_add_return(size, &bpf_jit_current) > bpf_jit_limit) {
1013 		if (!bpf_capable()) {
1014 			atomic_long_sub(size, &bpf_jit_current);
1015 			return -EPERM;
1016 		}
1017 	}
1018 
1019 	return 0;
1020 }
1021 
1022 void bpf_jit_uncharge_modmem(u32 size)
1023 {
1024 	atomic_long_sub(size, &bpf_jit_current);
1025 }
1026 
1027 void *__weak bpf_jit_alloc_exec(unsigned long size)
1028 {
1029 	return module_alloc(size);
1030 }
1031 
1032 void __weak bpf_jit_free_exec(void *addr)
1033 {
1034 	module_memfree(addr);
1035 }
1036 
1037 struct bpf_binary_header *
1038 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1039 		     unsigned int alignment,
1040 		     bpf_jit_fill_hole_t bpf_fill_ill_insns)
1041 {
1042 	struct bpf_binary_header *hdr;
1043 	u32 size, hole, start;
1044 
1045 	WARN_ON_ONCE(!is_power_of_2(alignment) ||
1046 		     alignment > BPF_IMAGE_ALIGNMENT);
1047 
1048 	/* Most of BPF filters are really small, but if some of them
1049 	 * fill a page, allow at least 128 extra bytes to insert a
1050 	 * random section of illegal instructions.
1051 	 */
1052 	size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
1053 
1054 	if (bpf_jit_charge_modmem(size))
1055 		return NULL;
1056 	hdr = bpf_jit_alloc_exec(size);
1057 	if (!hdr) {
1058 		bpf_jit_uncharge_modmem(size);
1059 		return NULL;
1060 	}
1061 
1062 	/* Fill space with illegal/arch-dep instructions. */
1063 	bpf_fill_ill_insns(hdr, size);
1064 
1065 	hdr->size = size;
1066 	hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
1067 		     PAGE_SIZE - sizeof(*hdr));
1068 	start = (get_random_int() % hole) & ~(alignment - 1);
1069 
1070 	/* Leave a random number of instructions before BPF code. */
1071 	*image_ptr = &hdr->image[start];
1072 
1073 	return hdr;
1074 }
1075 
1076 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
1077 {
1078 	u32 size = hdr->size;
1079 
1080 	bpf_jit_free_exec(hdr);
1081 	bpf_jit_uncharge_modmem(size);
1082 }
1083 
1084 /* Allocate jit binary from bpf_prog_pack allocator.
1085  * Since the allocated memory is RO+X, the JIT engine cannot write directly
1086  * to the memory. To solve this problem, a RW buffer is also allocated at
1087  * as the same time. The JIT engine should calculate offsets based on the
1088  * RO memory address, but write JITed program to the RW buffer. Once the
1089  * JIT engine finishes, it calls bpf_jit_binary_pack_finalize, which copies
1090  * the JITed program to the RO memory.
1091  */
1092 struct bpf_binary_header *
1093 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **image_ptr,
1094 			  unsigned int alignment,
1095 			  struct bpf_binary_header **rw_header,
1096 			  u8 **rw_image,
1097 			  bpf_jit_fill_hole_t bpf_fill_ill_insns)
1098 {
1099 	struct bpf_binary_header *ro_header;
1100 	u32 size, hole, start;
1101 
1102 	WARN_ON_ONCE(!is_power_of_2(alignment) ||
1103 		     alignment > BPF_IMAGE_ALIGNMENT);
1104 
1105 	/* add 16 bytes for a random section of illegal instructions */
1106 	size = round_up(proglen + sizeof(*ro_header) + 16, BPF_PROG_CHUNK_SIZE);
1107 
1108 	if (bpf_jit_charge_modmem(size))
1109 		return NULL;
1110 	ro_header = bpf_prog_pack_alloc(size, bpf_fill_ill_insns);
1111 	if (!ro_header) {
1112 		bpf_jit_uncharge_modmem(size);
1113 		return NULL;
1114 	}
1115 
1116 	*rw_header = kvmalloc(size, GFP_KERNEL);
1117 	if (!*rw_header) {
1118 		bpf_arch_text_copy(&ro_header->size, &size, sizeof(size));
1119 		bpf_prog_pack_free(ro_header);
1120 		bpf_jit_uncharge_modmem(size);
1121 		return NULL;
1122 	}
1123 
1124 	/* Fill space with illegal/arch-dep instructions. */
1125 	bpf_fill_ill_insns(*rw_header, size);
1126 	(*rw_header)->size = size;
1127 
1128 	hole = min_t(unsigned int, size - (proglen + sizeof(*ro_header)),
1129 		     BPF_PROG_CHUNK_SIZE - sizeof(*ro_header));
1130 	start = (get_random_int() % hole) & ~(alignment - 1);
1131 
1132 	*image_ptr = &ro_header->image[start];
1133 	*rw_image = &(*rw_header)->image[start];
1134 
1135 	return ro_header;
1136 }
1137 
1138 /* Copy JITed text from rw_header to its final location, the ro_header. */
1139 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1140 				 struct bpf_binary_header *ro_header,
1141 				 struct bpf_binary_header *rw_header)
1142 {
1143 	void *ptr;
1144 
1145 	ptr = bpf_arch_text_copy(ro_header, rw_header, rw_header->size);
1146 
1147 	kvfree(rw_header);
1148 
1149 	if (IS_ERR(ptr)) {
1150 		bpf_prog_pack_free(ro_header);
1151 		return PTR_ERR(ptr);
1152 	}
1153 	prog->aux->use_bpf_prog_pack = true;
1154 	return 0;
1155 }
1156 
1157 /* bpf_jit_binary_pack_free is called in two different scenarios:
1158  *   1) when the program is freed after;
1159  *   2) when the JIT engine fails (before bpf_jit_binary_pack_finalize).
1160  * For case 2), we need to free both the RO memory and the RW buffer.
1161  *
1162  * bpf_jit_binary_pack_free requires proper ro_header->size. However,
1163  * bpf_jit_binary_pack_alloc does not set it. Therefore, ro_header->size
1164  * must be set with either bpf_jit_binary_pack_finalize (normal path) or
1165  * bpf_arch_text_copy (when jit fails).
1166  */
1167 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1168 			      struct bpf_binary_header *rw_header)
1169 {
1170 	u32 size = ro_header->size;
1171 
1172 	bpf_prog_pack_free(ro_header);
1173 	kvfree(rw_header);
1174 	bpf_jit_uncharge_modmem(size);
1175 }
1176 
1177 static inline struct bpf_binary_header *
1178 bpf_jit_binary_hdr(const struct bpf_prog *fp)
1179 {
1180 	unsigned long real_start = (unsigned long)fp->bpf_func;
1181 	unsigned long addr;
1182 
1183 	if (fp->aux->use_bpf_prog_pack)
1184 		addr = real_start & BPF_PROG_CHUNK_MASK;
1185 	else
1186 		addr = real_start & PAGE_MASK;
1187 
1188 	return (void *)addr;
1189 }
1190 
1191 /* This symbol is only overridden by archs that have different
1192  * requirements than the usual eBPF JITs, f.e. when they only
1193  * implement cBPF JIT, do not set images read-only, etc.
1194  */
1195 void __weak bpf_jit_free(struct bpf_prog *fp)
1196 {
1197 	if (fp->jited) {
1198 		struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
1199 
1200 		if (fp->aux->use_bpf_prog_pack)
1201 			bpf_jit_binary_pack_free(hdr, NULL /* rw_buffer */);
1202 		else
1203 			bpf_jit_binary_free(hdr);
1204 
1205 		WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
1206 	}
1207 
1208 	bpf_prog_unlock_free(fp);
1209 }
1210 
1211 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1212 			  const struct bpf_insn *insn, bool extra_pass,
1213 			  u64 *func_addr, bool *func_addr_fixed)
1214 {
1215 	s16 off = insn->off;
1216 	s32 imm = insn->imm;
1217 	u8 *addr;
1218 
1219 	*func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
1220 	if (!*func_addr_fixed) {
1221 		/* Place-holder address till the last pass has collected
1222 		 * all addresses for JITed subprograms in which case we
1223 		 * can pick them up from prog->aux.
1224 		 */
1225 		if (!extra_pass)
1226 			addr = NULL;
1227 		else if (prog->aux->func &&
1228 			 off >= 0 && off < prog->aux->func_cnt)
1229 			addr = (u8 *)prog->aux->func[off]->bpf_func;
1230 		else
1231 			return -EINVAL;
1232 	} else {
1233 		/* Address of a BPF helper call. Since part of the core
1234 		 * kernel, it's always at a fixed location. __bpf_call_base
1235 		 * and the helper with imm relative to it are both in core
1236 		 * kernel.
1237 		 */
1238 		addr = (u8 *)__bpf_call_base + imm;
1239 	}
1240 
1241 	*func_addr = (unsigned long)addr;
1242 	return 0;
1243 }
1244 
1245 static int bpf_jit_blind_insn(const struct bpf_insn *from,
1246 			      const struct bpf_insn *aux,
1247 			      struct bpf_insn *to_buff,
1248 			      bool emit_zext)
1249 {
1250 	struct bpf_insn *to = to_buff;
1251 	u32 imm_rnd = get_random_int();
1252 	s16 off;
1253 
1254 	BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
1255 	BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
1256 
1257 	/* Constraints on AX register:
1258 	 *
1259 	 * AX register is inaccessible from user space. It is mapped in
1260 	 * all JITs, and used here for constant blinding rewrites. It is
1261 	 * typically "stateless" meaning its contents are only valid within
1262 	 * the executed instruction, but not across several instructions.
1263 	 * There are a few exceptions however which are further detailed
1264 	 * below.
1265 	 *
1266 	 * Constant blinding is only used by JITs, not in the interpreter.
1267 	 * The interpreter uses AX in some occasions as a local temporary
1268 	 * register e.g. in DIV or MOD instructions.
1269 	 *
1270 	 * In restricted circumstances, the verifier can also use the AX
1271 	 * register for rewrites as long as they do not interfere with
1272 	 * the above cases!
1273 	 */
1274 	if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
1275 		goto out;
1276 
1277 	if (from->imm == 0 &&
1278 	    (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
1279 	     from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
1280 		*to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
1281 		goto out;
1282 	}
1283 
1284 	switch (from->code) {
1285 	case BPF_ALU | BPF_ADD | BPF_K:
1286 	case BPF_ALU | BPF_SUB | BPF_K:
1287 	case BPF_ALU | BPF_AND | BPF_K:
1288 	case BPF_ALU | BPF_OR  | BPF_K:
1289 	case BPF_ALU | BPF_XOR | BPF_K:
1290 	case BPF_ALU | BPF_MUL | BPF_K:
1291 	case BPF_ALU | BPF_MOV | BPF_K:
1292 	case BPF_ALU | BPF_DIV | BPF_K:
1293 	case BPF_ALU | BPF_MOD | BPF_K:
1294 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1295 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1296 		*to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
1297 		break;
1298 
1299 	case BPF_ALU64 | BPF_ADD | BPF_K:
1300 	case BPF_ALU64 | BPF_SUB | BPF_K:
1301 	case BPF_ALU64 | BPF_AND | BPF_K:
1302 	case BPF_ALU64 | BPF_OR  | BPF_K:
1303 	case BPF_ALU64 | BPF_XOR | BPF_K:
1304 	case BPF_ALU64 | BPF_MUL | BPF_K:
1305 	case BPF_ALU64 | BPF_MOV | BPF_K:
1306 	case BPF_ALU64 | BPF_DIV | BPF_K:
1307 	case BPF_ALU64 | BPF_MOD | BPF_K:
1308 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1309 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1310 		*to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
1311 		break;
1312 
1313 	case BPF_JMP | BPF_JEQ  | BPF_K:
1314 	case BPF_JMP | BPF_JNE  | BPF_K:
1315 	case BPF_JMP | BPF_JGT  | BPF_K:
1316 	case BPF_JMP | BPF_JLT  | BPF_K:
1317 	case BPF_JMP | BPF_JGE  | BPF_K:
1318 	case BPF_JMP | BPF_JLE  | BPF_K:
1319 	case BPF_JMP | BPF_JSGT | BPF_K:
1320 	case BPF_JMP | BPF_JSLT | BPF_K:
1321 	case BPF_JMP | BPF_JSGE | BPF_K:
1322 	case BPF_JMP | BPF_JSLE | BPF_K:
1323 	case BPF_JMP | BPF_JSET | BPF_K:
1324 		/* Accommodate for extra offset in case of a backjump. */
1325 		off = from->off;
1326 		if (off < 0)
1327 			off -= 2;
1328 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1329 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1330 		*to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1331 		break;
1332 
1333 	case BPF_JMP32 | BPF_JEQ  | BPF_K:
1334 	case BPF_JMP32 | BPF_JNE  | BPF_K:
1335 	case BPF_JMP32 | BPF_JGT  | BPF_K:
1336 	case BPF_JMP32 | BPF_JLT  | BPF_K:
1337 	case BPF_JMP32 | BPF_JGE  | BPF_K:
1338 	case BPF_JMP32 | BPF_JLE  | BPF_K:
1339 	case BPF_JMP32 | BPF_JSGT | BPF_K:
1340 	case BPF_JMP32 | BPF_JSLT | BPF_K:
1341 	case BPF_JMP32 | BPF_JSGE | BPF_K:
1342 	case BPF_JMP32 | BPF_JSLE | BPF_K:
1343 	case BPF_JMP32 | BPF_JSET | BPF_K:
1344 		/* Accommodate for extra offset in case of a backjump. */
1345 		off = from->off;
1346 		if (off < 0)
1347 			off -= 2;
1348 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1349 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1350 		*to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1351 				      off);
1352 		break;
1353 
1354 	case BPF_LD | BPF_IMM | BPF_DW:
1355 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1356 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1357 		*to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1358 		*to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1359 		break;
1360 	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1361 		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1362 		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1363 		if (emit_zext)
1364 			*to++ = BPF_ZEXT_REG(BPF_REG_AX);
1365 		*to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
1366 		break;
1367 
1368 	case BPF_ST | BPF_MEM | BPF_DW:
1369 	case BPF_ST | BPF_MEM | BPF_W:
1370 	case BPF_ST | BPF_MEM | BPF_H:
1371 	case BPF_ST | BPF_MEM | BPF_B:
1372 		*to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1373 		*to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1374 		*to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1375 		break;
1376 	}
1377 out:
1378 	return to - to_buff;
1379 }
1380 
1381 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1382 					      gfp_t gfp_extra_flags)
1383 {
1384 	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1385 	struct bpf_prog *fp;
1386 
1387 	fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
1388 	if (fp != NULL) {
1389 		/* aux->prog still points to the fp_other one, so
1390 		 * when promoting the clone to the real program,
1391 		 * this still needs to be adapted.
1392 		 */
1393 		memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1394 	}
1395 
1396 	return fp;
1397 }
1398 
1399 static void bpf_prog_clone_free(struct bpf_prog *fp)
1400 {
1401 	/* aux was stolen by the other clone, so we cannot free
1402 	 * it from this path! It will be freed eventually by the
1403 	 * other program on release.
1404 	 *
1405 	 * At this point, we don't need a deferred release since
1406 	 * clone is guaranteed to not be locked.
1407 	 */
1408 	fp->aux = NULL;
1409 	fp->stats = NULL;
1410 	fp->active = NULL;
1411 	__bpf_prog_free(fp);
1412 }
1413 
1414 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1415 {
1416 	/* We have to repoint aux->prog to self, as we don't
1417 	 * know whether fp here is the clone or the original.
1418 	 */
1419 	fp->aux->prog = fp;
1420 	bpf_prog_clone_free(fp_other);
1421 }
1422 
1423 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1424 {
1425 	struct bpf_insn insn_buff[16], aux[2];
1426 	struct bpf_prog *clone, *tmp;
1427 	int insn_delta, insn_cnt;
1428 	struct bpf_insn *insn;
1429 	int i, rewritten;
1430 
1431 	if (!prog->blinding_requested || prog->blinded)
1432 		return prog;
1433 
1434 	clone = bpf_prog_clone_create(prog, GFP_USER);
1435 	if (!clone)
1436 		return ERR_PTR(-ENOMEM);
1437 
1438 	insn_cnt = clone->len;
1439 	insn = clone->insnsi;
1440 
1441 	for (i = 0; i < insn_cnt; i++, insn++) {
1442 		if (bpf_pseudo_func(insn)) {
1443 			/* ld_imm64 with an address of bpf subprog is not
1444 			 * a user controlled constant. Don't randomize it,
1445 			 * since it will conflict with jit_subprogs() logic.
1446 			 */
1447 			insn++;
1448 			i++;
1449 			continue;
1450 		}
1451 
1452 		/* We temporarily need to hold the original ld64 insn
1453 		 * so that we can still access the first part in the
1454 		 * second blinding run.
1455 		 */
1456 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1457 		    insn[1].code == 0)
1458 			memcpy(aux, insn, sizeof(aux));
1459 
1460 		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1461 						clone->aux->verifier_zext);
1462 		if (!rewritten)
1463 			continue;
1464 
1465 		tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1466 		if (IS_ERR(tmp)) {
1467 			/* Patching may have repointed aux->prog during
1468 			 * realloc from the original one, so we need to
1469 			 * fix it up here on error.
1470 			 */
1471 			bpf_jit_prog_release_other(prog, clone);
1472 			return tmp;
1473 		}
1474 
1475 		clone = tmp;
1476 		insn_delta = rewritten - 1;
1477 
1478 		/* Walk new program and skip insns we just inserted. */
1479 		insn = clone->insnsi + i + insn_delta;
1480 		insn_cnt += insn_delta;
1481 		i        += insn_delta;
1482 	}
1483 
1484 	clone->blinded = 1;
1485 	return clone;
1486 }
1487 #endif /* CONFIG_BPF_JIT */
1488 
1489 /* Base function for offset calculation. Needs to go into .text section,
1490  * therefore keeping it non-static as well; will also be used by JITs
1491  * anyway later on, so do not let the compiler omit it. This also needs
1492  * to go into kallsyms for correlation from e.g. bpftool, so naming
1493  * must not change.
1494  */
1495 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1496 {
1497 	return 0;
1498 }
1499 EXPORT_SYMBOL_GPL(__bpf_call_base);
1500 
1501 /* All UAPI available opcodes. */
1502 #define BPF_INSN_MAP(INSN_2, INSN_3)		\
1503 	/* 32 bit ALU operations. */		\
1504 	/*   Register based. */			\
1505 	INSN_3(ALU, ADD,  X),			\
1506 	INSN_3(ALU, SUB,  X),			\
1507 	INSN_3(ALU, AND,  X),			\
1508 	INSN_3(ALU, OR,   X),			\
1509 	INSN_3(ALU, LSH,  X),			\
1510 	INSN_3(ALU, RSH,  X),			\
1511 	INSN_3(ALU, XOR,  X),			\
1512 	INSN_3(ALU, MUL,  X),			\
1513 	INSN_3(ALU, MOV,  X),			\
1514 	INSN_3(ALU, ARSH, X),			\
1515 	INSN_3(ALU, DIV,  X),			\
1516 	INSN_3(ALU, MOD,  X),			\
1517 	INSN_2(ALU, NEG),			\
1518 	INSN_3(ALU, END, TO_BE),		\
1519 	INSN_3(ALU, END, TO_LE),		\
1520 	/*   Immediate based. */		\
1521 	INSN_3(ALU, ADD,  K),			\
1522 	INSN_3(ALU, SUB,  K),			\
1523 	INSN_3(ALU, AND,  K),			\
1524 	INSN_3(ALU, OR,   K),			\
1525 	INSN_3(ALU, LSH,  K),			\
1526 	INSN_3(ALU, RSH,  K),			\
1527 	INSN_3(ALU, XOR,  K),			\
1528 	INSN_3(ALU, MUL,  K),			\
1529 	INSN_3(ALU, MOV,  K),			\
1530 	INSN_3(ALU, ARSH, K),			\
1531 	INSN_3(ALU, DIV,  K),			\
1532 	INSN_3(ALU, MOD,  K),			\
1533 	/* 64 bit ALU operations. */		\
1534 	/*   Register based. */			\
1535 	INSN_3(ALU64, ADD,  X),			\
1536 	INSN_3(ALU64, SUB,  X),			\
1537 	INSN_3(ALU64, AND,  X),			\
1538 	INSN_3(ALU64, OR,   X),			\
1539 	INSN_3(ALU64, LSH,  X),			\
1540 	INSN_3(ALU64, RSH,  X),			\
1541 	INSN_3(ALU64, XOR,  X),			\
1542 	INSN_3(ALU64, MUL,  X),			\
1543 	INSN_3(ALU64, MOV,  X),			\
1544 	INSN_3(ALU64, ARSH, X),			\
1545 	INSN_3(ALU64, DIV,  X),			\
1546 	INSN_3(ALU64, MOD,  X),			\
1547 	INSN_2(ALU64, NEG),			\
1548 	/*   Immediate based. */		\
1549 	INSN_3(ALU64, ADD,  K),			\
1550 	INSN_3(ALU64, SUB,  K),			\
1551 	INSN_3(ALU64, AND,  K),			\
1552 	INSN_3(ALU64, OR,   K),			\
1553 	INSN_3(ALU64, LSH,  K),			\
1554 	INSN_3(ALU64, RSH,  K),			\
1555 	INSN_3(ALU64, XOR,  K),			\
1556 	INSN_3(ALU64, MUL,  K),			\
1557 	INSN_3(ALU64, MOV,  K),			\
1558 	INSN_3(ALU64, ARSH, K),			\
1559 	INSN_3(ALU64, DIV,  K),			\
1560 	INSN_3(ALU64, MOD,  K),			\
1561 	/* Call instruction. */			\
1562 	INSN_2(JMP, CALL),			\
1563 	/* Exit instruction. */			\
1564 	INSN_2(JMP, EXIT),			\
1565 	/* 32-bit Jump instructions. */		\
1566 	/*   Register based. */			\
1567 	INSN_3(JMP32, JEQ,  X),			\
1568 	INSN_3(JMP32, JNE,  X),			\
1569 	INSN_3(JMP32, JGT,  X),			\
1570 	INSN_3(JMP32, JLT,  X),			\
1571 	INSN_3(JMP32, JGE,  X),			\
1572 	INSN_3(JMP32, JLE,  X),			\
1573 	INSN_3(JMP32, JSGT, X),			\
1574 	INSN_3(JMP32, JSLT, X),			\
1575 	INSN_3(JMP32, JSGE, X),			\
1576 	INSN_3(JMP32, JSLE, X),			\
1577 	INSN_3(JMP32, JSET, X),			\
1578 	/*   Immediate based. */		\
1579 	INSN_3(JMP32, JEQ,  K),			\
1580 	INSN_3(JMP32, JNE,  K),			\
1581 	INSN_3(JMP32, JGT,  K),			\
1582 	INSN_3(JMP32, JLT,  K),			\
1583 	INSN_3(JMP32, JGE,  K),			\
1584 	INSN_3(JMP32, JLE,  K),			\
1585 	INSN_3(JMP32, JSGT, K),			\
1586 	INSN_3(JMP32, JSLT, K),			\
1587 	INSN_3(JMP32, JSGE, K),			\
1588 	INSN_3(JMP32, JSLE, K),			\
1589 	INSN_3(JMP32, JSET, K),			\
1590 	/* Jump instructions. */		\
1591 	/*   Register based. */			\
1592 	INSN_3(JMP, JEQ,  X),			\
1593 	INSN_3(JMP, JNE,  X),			\
1594 	INSN_3(JMP, JGT,  X),			\
1595 	INSN_3(JMP, JLT,  X),			\
1596 	INSN_3(JMP, JGE,  X),			\
1597 	INSN_3(JMP, JLE,  X),			\
1598 	INSN_3(JMP, JSGT, X),			\
1599 	INSN_3(JMP, JSLT, X),			\
1600 	INSN_3(JMP, JSGE, X),			\
1601 	INSN_3(JMP, JSLE, X),			\
1602 	INSN_3(JMP, JSET, X),			\
1603 	/*   Immediate based. */		\
1604 	INSN_3(JMP, JEQ,  K),			\
1605 	INSN_3(JMP, JNE,  K),			\
1606 	INSN_3(JMP, JGT,  K),			\
1607 	INSN_3(JMP, JLT,  K),			\
1608 	INSN_3(JMP, JGE,  K),			\
1609 	INSN_3(JMP, JLE,  K),			\
1610 	INSN_3(JMP, JSGT, K),			\
1611 	INSN_3(JMP, JSLT, K),			\
1612 	INSN_3(JMP, JSGE, K),			\
1613 	INSN_3(JMP, JSLE, K),			\
1614 	INSN_3(JMP, JSET, K),			\
1615 	INSN_2(JMP, JA),			\
1616 	/* Store instructions. */		\
1617 	/*   Register based. */			\
1618 	INSN_3(STX, MEM,  B),			\
1619 	INSN_3(STX, MEM,  H),			\
1620 	INSN_3(STX, MEM,  W),			\
1621 	INSN_3(STX, MEM,  DW),			\
1622 	INSN_3(STX, ATOMIC, W),			\
1623 	INSN_3(STX, ATOMIC, DW),		\
1624 	/*   Immediate based. */		\
1625 	INSN_3(ST, MEM, B),			\
1626 	INSN_3(ST, MEM, H),			\
1627 	INSN_3(ST, MEM, W),			\
1628 	INSN_3(ST, MEM, DW),			\
1629 	/* Load instructions. */		\
1630 	/*   Register based. */			\
1631 	INSN_3(LDX, MEM, B),			\
1632 	INSN_3(LDX, MEM, H),			\
1633 	INSN_3(LDX, MEM, W),			\
1634 	INSN_3(LDX, MEM, DW),			\
1635 	/*   Immediate based. */		\
1636 	INSN_3(LD, IMM, DW)
1637 
1638 bool bpf_opcode_in_insntable(u8 code)
1639 {
1640 #define BPF_INSN_2_TBL(x, y)    [BPF_##x | BPF_##y] = true
1641 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1642 	static const bool public_insntable[256] = {
1643 		[0 ... 255] = false,
1644 		/* Now overwrite non-defaults ... */
1645 		BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1646 		/* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1647 		[BPF_LD | BPF_ABS | BPF_B] = true,
1648 		[BPF_LD | BPF_ABS | BPF_H] = true,
1649 		[BPF_LD | BPF_ABS | BPF_W] = true,
1650 		[BPF_LD | BPF_IND | BPF_B] = true,
1651 		[BPF_LD | BPF_IND | BPF_H] = true,
1652 		[BPF_LD | BPF_IND | BPF_W] = true,
1653 	};
1654 #undef BPF_INSN_3_TBL
1655 #undef BPF_INSN_2_TBL
1656 	return public_insntable[code];
1657 }
1658 
1659 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1660 u64 __weak bpf_probe_read_kernel(void *dst, u32 size, const void *unsafe_ptr)
1661 {
1662 	memset(dst, 0, size);
1663 	return -EFAULT;
1664 }
1665 
1666 /**
1667  *	___bpf_prog_run - run eBPF program on a given context
1668  *	@regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1669  *	@insn: is the array of eBPF instructions
1670  *
1671  * Decode and execute eBPF instructions.
1672  *
1673  * Return: whatever value is in %BPF_R0 at program exit
1674  */
1675 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
1676 {
1677 #define BPF_INSN_2_LBL(x, y)    [BPF_##x | BPF_##y] = &&x##_##y
1678 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1679 	static const void * const jumptable[256] __annotate_jump_table = {
1680 		[0 ... 255] = &&default_label,
1681 		/* Now overwrite non-defaults ... */
1682 		BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1683 		/* Non-UAPI available opcodes. */
1684 		[BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1685 		[BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1686 		[BPF_ST  | BPF_NOSPEC] = &&ST_NOSPEC,
1687 		[BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1688 		[BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1689 		[BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1690 		[BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1691 	};
1692 #undef BPF_INSN_3_LBL
1693 #undef BPF_INSN_2_LBL
1694 	u32 tail_call_cnt = 0;
1695 
1696 #define CONT	 ({ insn++; goto select_insn; })
1697 #define CONT_JMP ({ insn++; goto select_insn; })
1698 
1699 select_insn:
1700 	goto *jumptable[insn->code];
1701 
1702 	/* Explicitly mask the register-based shift amounts with 63 or 31
1703 	 * to avoid undefined behavior. Normally this won't affect the
1704 	 * generated code, for example, in case of native 64 bit archs such
1705 	 * as x86-64 or arm64, the compiler is optimizing the AND away for
1706 	 * the interpreter. In case of JITs, each of the JIT backends compiles
1707 	 * the BPF shift operations to machine instructions which produce
1708 	 * implementation-defined results in such a case; the resulting
1709 	 * contents of the register may be arbitrary, but program behaviour
1710 	 * as a whole remains defined. In other words, in case of JIT backends,
1711 	 * the AND must /not/ be added to the emitted LSH/RSH/ARSH translation.
1712 	 */
1713 	/* ALU (shifts) */
1714 #define SHT(OPCODE, OP)					\
1715 	ALU64_##OPCODE##_X:				\
1716 		DST = DST OP (SRC & 63);		\
1717 		CONT;					\
1718 	ALU_##OPCODE##_X:				\
1719 		DST = (u32) DST OP ((u32) SRC & 31);	\
1720 		CONT;					\
1721 	ALU64_##OPCODE##_K:				\
1722 		DST = DST OP IMM;			\
1723 		CONT;					\
1724 	ALU_##OPCODE##_K:				\
1725 		DST = (u32) DST OP (u32) IMM;		\
1726 		CONT;
1727 	/* ALU (rest) */
1728 #define ALU(OPCODE, OP)					\
1729 	ALU64_##OPCODE##_X:				\
1730 		DST = DST OP SRC;			\
1731 		CONT;					\
1732 	ALU_##OPCODE##_X:				\
1733 		DST = (u32) DST OP (u32) SRC;		\
1734 		CONT;					\
1735 	ALU64_##OPCODE##_K:				\
1736 		DST = DST OP IMM;			\
1737 		CONT;					\
1738 	ALU_##OPCODE##_K:				\
1739 		DST = (u32) DST OP (u32) IMM;		\
1740 		CONT;
1741 	ALU(ADD,  +)
1742 	ALU(SUB,  -)
1743 	ALU(AND,  &)
1744 	ALU(OR,   |)
1745 	ALU(XOR,  ^)
1746 	ALU(MUL,  *)
1747 	SHT(LSH, <<)
1748 	SHT(RSH, >>)
1749 #undef SHT
1750 #undef ALU
1751 	ALU_NEG:
1752 		DST = (u32) -DST;
1753 		CONT;
1754 	ALU64_NEG:
1755 		DST = -DST;
1756 		CONT;
1757 	ALU_MOV_X:
1758 		DST = (u32) SRC;
1759 		CONT;
1760 	ALU_MOV_K:
1761 		DST = (u32) IMM;
1762 		CONT;
1763 	ALU64_MOV_X:
1764 		DST = SRC;
1765 		CONT;
1766 	ALU64_MOV_K:
1767 		DST = IMM;
1768 		CONT;
1769 	LD_IMM_DW:
1770 		DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1771 		insn++;
1772 		CONT;
1773 	ALU_ARSH_X:
1774 		DST = (u64) (u32) (((s32) DST) >> (SRC & 31));
1775 		CONT;
1776 	ALU_ARSH_K:
1777 		DST = (u64) (u32) (((s32) DST) >> IMM);
1778 		CONT;
1779 	ALU64_ARSH_X:
1780 		(*(s64 *) &DST) >>= (SRC & 63);
1781 		CONT;
1782 	ALU64_ARSH_K:
1783 		(*(s64 *) &DST) >>= IMM;
1784 		CONT;
1785 	ALU64_MOD_X:
1786 		div64_u64_rem(DST, SRC, &AX);
1787 		DST = AX;
1788 		CONT;
1789 	ALU_MOD_X:
1790 		AX = (u32) DST;
1791 		DST = do_div(AX, (u32) SRC);
1792 		CONT;
1793 	ALU64_MOD_K:
1794 		div64_u64_rem(DST, IMM, &AX);
1795 		DST = AX;
1796 		CONT;
1797 	ALU_MOD_K:
1798 		AX = (u32) DST;
1799 		DST = do_div(AX, (u32) IMM);
1800 		CONT;
1801 	ALU64_DIV_X:
1802 		DST = div64_u64(DST, SRC);
1803 		CONT;
1804 	ALU_DIV_X:
1805 		AX = (u32) DST;
1806 		do_div(AX, (u32) SRC);
1807 		DST = (u32) AX;
1808 		CONT;
1809 	ALU64_DIV_K:
1810 		DST = div64_u64(DST, IMM);
1811 		CONT;
1812 	ALU_DIV_K:
1813 		AX = (u32) DST;
1814 		do_div(AX, (u32) IMM);
1815 		DST = (u32) AX;
1816 		CONT;
1817 	ALU_END_TO_BE:
1818 		switch (IMM) {
1819 		case 16:
1820 			DST = (__force u16) cpu_to_be16(DST);
1821 			break;
1822 		case 32:
1823 			DST = (__force u32) cpu_to_be32(DST);
1824 			break;
1825 		case 64:
1826 			DST = (__force u64) cpu_to_be64(DST);
1827 			break;
1828 		}
1829 		CONT;
1830 	ALU_END_TO_LE:
1831 		switch (IMM) {
1832 		case 16:
1833 			DST = (__force u16) cpu_to_le16(DST);
1834 			break;
1835 		case 32:
1836 			DST = (__force u32) cpu_to_le32(DST);
1837 			break;
1838 		case 64:
1839 			DST = (__force u64) cpu_to_le64(DST);
1840 			break;
1841 		}
1842 		CONT;
1843 
1844 	/* CALL */
1845 	JMP_CALL:
1846 		/* Function call scratches BPF_R1-BPF_R5 registers,
1847 		 * preserves BPF_R6-BPF_R9, and stores return value
1848 		 * into BPF_R0.
1849 		 */
1850 		BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1851 						       BPF_R4, BPF_R5);
1852 		CONT;
1853 
1854 	JMP_CALL_ARGS:
1855 		BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1856 							    BPF_R3, BPF_R4,
1857 							    BPF_R5,
1858 							    insn + insn->off + 1);
1859 		CONT;
1860 
1861 	JMP_TAIL_CALL: {
1862 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1863 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1864 		struct bpf_prog *prog;
1865 		u32 index = BPF_R3;
1866 
1867 		if (unlikely(index >= array->map.max_entries))
1868 			goto out;
1869 
1870 		if (unlikely(tail_call_cnt >= MAX_TAIL_CALL_CNT))
1871 			goto out;
1872 
1873 		tail_call_cnt++;
1874 
1875 		prog = READ_ONCE(array->ptrs[index]);
1876 		if (!prog)
1877 			goto out;
1878 
1879 		/* ARG1 at this point is guaranteed to point to CTX from
1880 		 * the verifier side due to the fact that the tail call is
1881 		 * handled like a helper, that is, bpf_tail_call_proto,
1882 		 * where arg1_type is ARG_PTR_TO_CTX.
1883 		 */
1884 		insn = prog->insnsi;
1885 		goto select_insn;
1886 out:
1887 		CONT;
1888 	}
1889 	JMP_JA:
1890 		insn += insn->off;
1891 		CONT;
1892 	JMP_EXIT:
1893 		return BPF_R0;
1894 	/* JMP */
1895 #define COND_JMP(SIGN, OPCODE, CMP_OP)				\
1896 	JMP_##OPCODE##_X:					\
1897 		if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) {	\
1898 			insn += insn->off;			\
1899 			CONT_JMP;				\
1900 		}						\
1901 		CONT;						\
1902 	JMP32_##OPCODE##_X:					\
1903 		if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) {	\
1904 			insn += insn->off;			\
1905 			CONT_JMP;				\
1906 		}						\
1907 		CONT;						\
1908 	JMP_##OPCODE##_K:					\
1909 		if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) {	\
1910 			insn += insn->off;			\
1911 			CONT_JMP;				\
1912 		}						\
1913 		CONT;						\
1914 	JMP32_##OPCODE##_K:					\
1915 		if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) {	\
1916 			insn += insn->off;			\
1917 			CONT_JMP;				\
1918 		}						\
1919 		CONT;
1920 	COND_JMP(u, JEQ, ==)
1921 	COND_JMP(u, JNE, !=)
1922 	COND_JMP(u, JGT, >)
1923 	COND_JMP(u, JLT, <)
1924 	COND_JMP(u, JGE, >=)
1925 	COND_JMP(u, JLE, <=)
1926 	COND_JMP(u, JSET, &)
1927 	COND_JMP(s, JSGT, >)
1928 	COND_JMP(s, JSLT, <)
1929 	COND_JMP(s, JSGE, >=)
1930 	COND_JMP(s, JSLE, <=)
1931 #undef COND_JMP
1932 	/* ST, STX and LDX*/
1933 	ST_NOSPEC:
1934 		/* Speculation barrier for mitigating Speculative Store Bypass.
1935 		 * In case of arm64, we rely on the firmware mitigation as
1936 		 * controlled via the ssbd kernel parameter. Whenever the
1937 		 * mitigation is enabled, it works for all of the kernel code
1938 		 * with no need to provide any additional instructions here.
1939 		 * In case of x86, we use 'lfence' insn for mitigation. We
1940 		 * reuse preexisting logic from Spectre v1 mitigation that
1941 		 * happens to produce the required code on x86 for v4 as well.
1942 		 */
1943 #ifdef CONFIG_X86
1944 		barrier_nospec();
1945 #endif
1946 		CONT;
1947 #define LDST(SIZEOP, SIZE)						\
1948 	STX_MEM_##SIZEOP:						\
1949 		*(SIZE *)(unsigned long) (DST + insn->off) = SRC;	\
1950 		CONT;							\
1951 	ST_MEM_##SIZEOP:						\
1952 		*(SIZE *)(unsigned long) (DST + insn->off) = IMM;	\
1953 		CONT;							\
1954 	LDX_MEM_##SIZEOP:						\
1955 		DST = *(SIZE *)(unsigned long) (SRC + insn->off);	\
1956 		CONT;							\
1957 	LDX_PROBE_MEM_##SIZEOP:						\
1958 		bpf_probe_read_kernel(&DST, sizeof(SIZE),		\
1959 				      (const void *)(long) (SRC + insn->off));	\
1960 		DST = *((SIZE *)&DST);					\
1961 		CONT;
1962 
1963 	LDST(B,   u8)
1964 	LDST(H,  u16)
1965 	LDST(W,  u32)
1966 	LDST(DW, u64)
1967 #undef LDST
1968 
1969 #define ATOMIC_ALU_OP(BOP, KOP)						\
1970 		case BOP:						\
1971 			if (BPF_SIZE(insn->code) == BPF_W)		\
1972 				atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
1973 					     (DST + insn->off));	\
1974 			else						\
1975 				atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
1976 					       (DST + insn->off));	\
1977 			break;						\
1978 		case BOP | BPF_FETCH:					\
1979 			if (BPF_SIZE(insn->code) == BPF_W)		\
1980 				SRC = (u32) atomic_fetch_##KOP(		\
1981 					(u32) SRC,			\
1982 					(atomic_t *)(unsigned long) (DST + insn->off)); \
1983 			else						\
1984 				SRC = (u64) atomic64_fetch_##KOP(	\
1985 					(u64) SRC,			\
1986 					(atomic64_t *)(unsigned long) (DST + insn->off)); \
1987 			break;
1988 
1989 	STX_ATOMIC_DW:
1990 	STX_ATOMIC_W:
1991 		switch (IMM) {
1992 		ATOMIC_ALU_OP(BPF_ADD, add)
1993 		ATOMIC_ALU_OP(BPF_AND, and)
1994 		ATOMIC_ALU_OP(BPF_OR, or)
1995 		ATOMIC_ALU_OP(BPF_XOR, xor)
1996 #undef ATOMIC_ALU_OP
1997 
1998 		case BPF_XCHG:
1999 			if (BPF_SIZE(insn->code) == BPF_W)
2000 				SRC = (u32) atomic_xchg(
2001 					(atomic_t *)(unsigned long) (DST + insn->off),
2002 					(u32) SRC);
2003 			else
2004 				SRC = (u64) atomic64_xchg(
2005 					(atomic64_t *)(unsigned long) (DST + insn->off),
2006 					(u64) SRC);
2007 			break;
2008 		case BPF_CMPXCHG:
2009 			if (BPF_SIZE(insn->code) == BPF_W)
2010 				BPF_R0 = (u32) atomic_cmpxchg(
2011 					(atomic_t *)(unsigned long) (DST + insn->off),
2012 					(u32) BPF_R0, (u32) SRC);
2013 			else
2014 				BPF_R0 = (u64) atomic64_cmpxchg(
2015 					(atomic64_t *)(unsigned long) (DST + insn->off),
2016 					(u64) BPF_R0, (u64) SRC);
2017 			break;
2018 
2019 		default:
2020 			goto default_label;
2021 		}
2022 		CONT;
2023 
2024 	default_label:
2025 		/* If we ever reach this, we have a bug somewhere. Die hard here
2026 		 * instead of just returning 0; we could be somewhere in a subprog,
2027 		 * so execution could continue otherwise which we do /not/ want.
2028 		 *
2029 		 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
2030 		 */
2031 		pr_warn("BPF interpreter: unknown opcode %02x (imm: 0x%x)\n",
2032 			insn->code, insn->imm);
2033 		BUG_ON(1);
2034 		return 0;
2035 }
2036 
2037 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
2038 #define DEFINE_BPF_PROG_RUN(stack_size) \
2039 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
2040 { \
2041 	u64 stack[stack_size / sizeof(u64)]; \
2042 	u64 regs[MAX_BPF_EXT_REG]; \
2043 \
2044 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2045 	ARG1 = (u64) (unsigned long) ctx; \
2046 	return ___bpf_prog_run(regs, insn); \
2047 }
2048 
2049 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
2050 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
2051 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
2052 				      const struct bpf_insn *insn) \
2053 { \
2054 	u64 stack[stack_size / sizeof(u64)]; \
2055 	u64 regs[MAX_BPF_EXT_REG]; \
2056 \
2057 	FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2058 	BPF_R1 = r1; \
2059 	BPF_R2 = r2; \
2060 	BPF_R3 = r3; \
2061 	BPF_R4 = r4; \
2062 	BPF_R5 = r5; \
2063 	return ___bpf_prog_run(regs, insn); \
2064 }
2065 
2066 #define EVAL1(FN, X) FN(X)
2067 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
2068 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
2069 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
2070 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
2071 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
2072 
2073 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
2074 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
2075 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
2076 
2077 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
2078 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
2079 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
2080 
2081 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
2082 
2083 static unsigned int (*interpreters[])(const void *ctx,
2084 				      const struct bpf_insn *insn) = {
2085 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2086 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2087 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2088 };
2089 #undef PROG_NAME_LIST
2090 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
2091 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
2092 				  const struct bpf_insn *insn) = {
2093 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2094 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2095 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2096 };
2097 #undef PROG_NAME_LIST
2098 
2099 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
2100 {
2101 	stack_depth = max_t(u32, stack_depth, 1);
2102 	insn->off = (s16) insn->imm;
2103 	insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
2104 		__bpf_call_base_args;
2105 	insn->code = BPF_JMP | BPF_CALL_ARGS;
2106 }
2107 
2108 #else
2109 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
2110 					 const struct bpf_insn *insn)
2111 {
2112 	/* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
2113 	 * is not working properly, so warn about it!
2114 	 */
2115 	WARN_ON_ONCE(1);
2116 	return 0;
2117 }
2118 #endif
2119 
2120 bool bpf_prog_map_compatible(struct bpf_map *map,
2121 			     const struct bpf_prog *fp)
2122 {
2123 	bool ret;
2124 
2125 	if (fp->kprobe_override)
2126 		return false;
2127 
2128 	spin_lock(&map->owner.lock);
2129 	if (!map->owner.type) {
2130 		/* There's no owner yet where we could check for
2131 		 * compatibility.
2132 		 */
2133 		map->owner.type  = fp->type;
2134 		map->owner.jited = fp->jited;
2135 		map->owner.xdp_has_frags = fp->aux->xdp_has_frags;
2136 		ret = true;
2137 	} else {
2138 		ret = map->owner.type  == fp->type &&
2139 		      map->owner.jited == fp->jited &&
2140 		      map->owner.xdp_has_frags == fp->aux->xdp_has_frags;
2141 	}
2142 	spin_unlock(&map->owner.lock);
2143 
2144 	return ret;
2145 }
2146 
2147 static int bpf_check_tail_call(const struct bpf_prog *fp)
2148 {
2149 	struct bpf_prog_aux *aux = fp->aux;
2150 	int i, ret = 0;
2151 
2152 	mutex_lock(&aux->used_maps_mutex);
2153 	for (i = 0; i < aux->used_map_cnt; i++) {
2154 		struct bpf_map *map = aux->used_maps[i];
2155 
2156 		if (!map_type_contains_progs(map))
2157 			continue;
2158 
2159 		if (!bpf_prog_map_compatible(map, fp)) {
2160 			ret = -EINVAL;
2161 			goto out;
2162 		}
2163 	}
2164 
2165 out:
2166 	mutex_unlock(&aux->used_maps_mutex);
2167 	return ret;
2168 }
2169 
2170 static void bpf_prog_select_func(struct bpf_prog *fp)
2171 {
2172 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
2173 	u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
2174 
2175 	fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
2176 #else
2177 	fp->bpf_func = __bpf_prog_ret0_warn;
2178 #endif
2179 }
2180 
2181 /**
2182  *	bpf_prog_select_runtime - select exec runtime for BPF program
2183  *	@fp: bpf_prog populated with BPF program
2184  *	@err: pointer to error variable
2185  *
2186  * Try to JIT eBPF program, if JIT is not available, use interpreter.
2187  * The BPF program will be executed via bpf_prog_run() function.
2188  *
2189  * Return: the &fp argument along with &err set to 0 for success or
2190  * a negative errno code on failure
2191  */
2192 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
2193 {
2194 	/* In case of BPF to BPF calls, verifier did all the prep
2195 	 * work with regards to JITing, etc.
2196 	 */
2197 	bool jit_needed = false;
2198 
2199 	if (fp->bpf_func)
2200 		goto finalize;
2201 
2202 	if (IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) ||
2203 	    bpf_prog_has_kfunc_call(fp))
2204 		jit_needed = true;
2205 
2206 	bpf_prog_select_func(fp);
2207 
2208 	/* eBPF JITs can rewrite the program in case constant
2209 	 * blinding is active. However, in case of error during
2210 	 * blinding, bpf_int_jit_compile() must always return a
2211 	 * valid program, which in this case would simply not
2212 	 * be JITed, but falls back to the interpreter.
2213 	 */
2214 	if (!bpf_prog_is_dev_bound(fp->aux)) {
2215 		*err = bpf_prog_alloc_jited_linfo(fp);
2216 		if (*err)
2217 			return fp;
2218 
2219 		fp = bpf_int_jit_compile(fp);
2220 		bpf_prog_jit_attempt_done(fp);
2221 		if (!fp->jited && jit_needed) {
2222 			*err = -ENOTSUPP;
2223 			return fp;
2224 		}
2225 	} else {
2226 		*err = bpf_prog_offload_compile(fp);
2227 		if (*err)
2228 			return fp;
2229 	}
2230 
2231 finalize:
2232 	bpf_prog_lock_ro(fp);
2233 
2234 	/* The tail call compatibility check can only be done at
2235 	 * this late stage as we need to determine, if we deal
2236 	 * with JITed or non JITed program concatenations and not
2237 	 * all eBPF JITs might immediately support all features.
2238 	 */
2239 	*err = bpf_check_tail_call(fp);
2240 
2241 	return fp;
2242 }
2243 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
2244 
2245 static unsigned int __bpf_prog_ret1(const void *ctx,
2246 				    const struct bpf_insn *insn)
2247 {
2248 	return 1;
2249 }
2250 
2251 static struct bpf_prog_dummy {
2252 	struct bpf_prog prog;
2253 } dummy_bpf_prog = {
2254 	.prog = {
2255 		.bpf_func = __bpf_prog_ret1,
2256 	},
2257 };
2258 
2259 struct bpf_empty_prog_array bpf_empty_prog_array = {
2260 	.null_prog = NULL,
2261 };
2262 EXPORT_SYMBOL(bpf_empty_prog_array);
2263 
2264 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
2265 {
2266 	if (prog_cnt)
2267 		return kzalloc(sizeof(struct bpf_prog_array) +
2268 			       sizeof(struct bpf_prog_array_item) *
2269 			       (prog_cnt + 1),
2270 			       flags);
2271 
2272 	return &bpf_empty_prog_array.hdr;
2273 }
2274 
2275 void bpf_prog_array_free(struct bpf_prog_array *progs)
2276 {
2277 	if (!progs || progs == &bpf_empty_prog_array.hdr)
2278 		return;
2279 	kfree_rcu(progs, rcu);
2280 }
2281 
2282 int bpf_prog_array_length(struct bpf_prog_array *array)
2283 {
2284 	struct bpf_prog_array_item *item;
2285 	u32 cnt = 0;
2286 
2287 	for (item = array->items; item->prog; item++)
2288 		if (item->prog != &dummy_bpf_prog.prog)
2289 			cnt++;
2290 	return cnt;
2291 }
2292 
2293 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
2294 {
2295 	struct bpf_prog_array_item *item;
2296 
2297 	for (item = array->items; item->prog; item++)
2298 		if (item->prog != &dummy_bpf_prog.prog)
2299 			return false;
2300 	return true;
2301 }
2302 
2303 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
2304 				     u32 *prog_ids,
2305 				     u32 request_cnt)
2306 {
2307 	struct bpf_prog_array_item *item;
2308 	int i = 0;
2309 
2310 	for (item = array->items; item->prog; item++) {
2311 		if (item->prog == &dummy_bpf_prog.prog)
2312 			continue;
2313 		prog_ids[i] = item->prog->aux->id;
2314 		if (++i == request_cnt) {
2315 			item++;
2316 			break;
2317 		}
2318 	}
2319 
2320 	return !!(item->prog);
2321 }
2322 
2323 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
2324 				__u32 __user *prog_ids, u32 cnt)
2325 {
2326 	unsigned long err = 0;
2327 	bool nospc;
2328 	u32 *ids;
2329 
2330 	/* users of this function are doing:
2331 	 * cnt = bpf_prog_array_length();
2332 	 * if (cnt > 0)
2333 	 *     bpf_prog_array_copy_to_user(..., cnt);
2334 	 * so below kcalloc doesn't need extra cnt > 0 check.
2335 	 */
2336 	ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
2337 	if (!ids)
2338 		return -ENOMEM;
2339 	nospc = bpf_prog_array_copy_core(array, ids, cnt);
2340 	err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
2341 	kfree(ids);
2342 	if (err)
2343 		return -EFAULT;
2344 	if (nospc)
2345 		return -ENOSPC;
2346 	return 0;
2347 }
2348 
2349 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
2350 				struct bpf_prog *old_prog)
2351 {
2352 	struct bpf_prog_array_item *item;
2353 
2354 	for (item = array->items; item->prog; item++)
2355 		if (item->prog == old_prog) {
2356 			WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
2357 			break;
2358 		}
2359 }
2360 
2361 /**
2362  * bpf_prog_array_delete_safe_at() - Replaces the program at the given
2363  *                                   index into the program array with
2364  *                                   a dummy no-op program.
2365  * @array: a bpf_prog_array
2366  * @index: the index of the program to replace
2367  *
2368  * Skips over dummy programs, by not counting them, when calculating
2369  * the position of the program to replace.
2370  *
2371  * Return:
2372  * * 0		- Success
2373  * * -EINVAL	- Invalid index value. Must be a non-negative integer.
2374  * * -ENOENT	- Index out of range
2375  */
2376 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
2377 {
2378 	return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
2379 }
2380 
2381 /**
2382  * bpf_prog_array_update_at() - Updates the program at the given index
2383  *                              into the program array.
2384  * @array: a bpf_prog_array
2385  * @index: the index of the program to update
2386  * @prog: the program to insert into the array
2387  *
2388  * Skips over dummy programs, by not counting them, when calculating
2389  * the position of the program to update.
2390  *
2391  * Return:
2392  * * 0		- Success
2393  * * -EINVAL	- Invalid index value. Must be a non-negative integer.
2394  * * -ENOENT	- Index out of range
2395  */
2396 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2397 			     struct bpf_prog *prog)
2398 {
2399 	struct bpf_prog_array_item *item;
2400 
2401 	if (unlikely(index < 0))
2402 		return -EINVAL;
2403 
2404 	for (item = array->items; item->prog; item++) {
2405 		if (item->prog == &dummy_bpf_prog.prog)
2406 			continue;
2407 		if (!index) {
2408 			WRITE_ONCE(item->prog, prog);
2409 			return 0;
2410 		}
2411 		index--;
2412 	}
2413 	return -ENOENT;
2414 }
2415 
2416 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2417 			struct bpf_prog *exclude_prog,
2418 			struct bpf_prog *include_prog,
2419 			u64 bpf_cookie,
2420 			struct bpf_prog_array **new_array)
2421 {
2422 	int new_prog_cnt, carry_prog_cnt = 0;
2423 	struct bpf_prog_array_item *existing, *new;
2424 	struct bpf_prog_array *array;
2425 	bool found_exclude = false;
2426 
2427 	/* Figure out how many existing progs we need to carry over to
2428 	 * the new array.
2429 	 */
2430 	if (old_array) {
2431 		existing = old_array->items;
2432 		for (; existing->prog; existing++) {
2433 			if (existing->prog == exclude_prog) {
2434 				found_exclude = true;
2435 				continue;
2436 			}
2437 			if (existing->prog != &dummy_bpf_prog.prog)
2438 				carry_prog_cnt++;
2439 			if (existing->prog == include_prog)
2440 				return -EEXIST;
2441 		}
2442 	}
2443 
2444 	if (exclude_prog && !found_exclude)
2445 		return -ENOENT;
2446 
2447 	/* How many progs (not NULL) will be in the new array? */
2448 	new_prog_cnt = carry_prog_cnt;
2449 	if (include_prog)
2450 		new_prog_cnt += 1;
2451 
2452 	/* Do we have any prog (not NULL) in the new array? */
2453 	if (!new_prog_cnt) {
2454 		*new_array = NULL;
2455 		return 0;
2456 	}
2457 
2458 	/* +1 as the end of prog_array is marked with NULL */
2459 	array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2460 	if (!array)
2461 		return -ENOMEM;
2462 	new = array->items;
2463 
2464 	/* Fill in the new prog array */
2465 	if (carry_prog_cnt) {
2466 		existing = old_array->items;
2467 		for (; existing->prog; existing++) {
2468 			if (existing->prog == exclude_prog ||
2469 			    existing->prog == &dummy_bpf_prog.prog)
2470 				continue;
2471 
2472 			new->prog = existing->prog;
2473 			new->bpf_cookie = existing->bpf_cookie;
2474 			new++;
2475 		}
2476 	}
2477 	if (include_prog) {
2478 		new->prog = include_prog;
2479 		new->bpf_cookie = bpf_cookie;
2480 		new++;
2481 	}
2482 	new->prog = NULL;
2483 	*new_array = array;
2484 	return 0;
2485 }
2486 
2487 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2488 			     u32 *prog_ids, u32 request_cnt,
2489 			     u32 *prog_cnt)
2490 {
2491 	u32 cnt = 0;
2492 
2493 	if (array)
2494 		cnt = bpf_prog_array_length(array);
2495 
2496 	*prog_cnt = cnt;
2497 
2498 	/* return early if user requested only program count or nothing to copy */
2499 	if (!request_cnt || !cnt)
2500 		return 0;
2501 
2502 	/* this function is called under trace/bpf_trace.c: bpf_event_mutex */
2503 	return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
2504 								     : 0;
2505 }
2506 
2507 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2508 			  struct bpf_map **used_maps, u32 len)
2509 {
2510 	struct bpf_map *map;
2511 	u32 i;
2512 
2513 	for (i = 0; i < len; i++) {
2514 		map = used_maps[i];
2515 		if (map->ops->map_poke_untrack)
2516 			map->ops->map_poke_untrack(map, aux);
2517 		bpf_map_put(map);
2518 	}
2519 }
2520 
2521 static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2522 {
2523 	__bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
2524 	kfree(aux->used_maps);
2525 }
2526 
2527 void __bpf_free_used_btfs(struct bpf_prog_aux *aux,
2528 			  struct btf_mod_pair *used_btfs, u32 len)
2529 {
2530 #ifdef CONFIG_BPF_SYSCALL
2531 	struct btf_mod_pair *btf_mod;
2532 	u32 i;
2533 
2534 	for (i = 0; i < len; i++) {
2535 		btf_mod = &used_btfs[i];
2536 		if (btf_mod->module)
2537 			module_put(btf_mod->module);
2538 		btf_put(btf_mod->btf);
2539 	}
2540 #endif
2541 }
2542 
2543 static void bpf_free_used_btfs(struct bpf_prog_aux *aux)
2544 {
2545 	__bpf_free_used_btfs(aux, aux->used_btfs, aux->used_btf_cnt);
2546 	kfree(aux->used_btfs);
2547 }
2548 
2549 static void bpf_prog_free_deferred(struct work_struct *work)
2550 {
2551 	struct bpf_prog_aux *aux;
2552 	int i;
2553 
2554 	aux = container_of(work, struct bpf_prog_aux, work);
2555 #ifdef CONFIG_BPF_SYSCALL
2556 	bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
2557 #endif
2558 	bpf_free_used_maps(aux);
2559 	bpf_free_used_btfs(aux);
2560 	if (bpf_prog_is_dev_bound(aux))
2561 		bpf_prog_offload_destroy(aux->prog);
2562 #ifdef CONFIG_PERF_EVENTS
2563 	if (aux->prog->has_callchain_buf)
2564 		put_callchain_buffers();
2565 #endif
2566 	if (aux->dst_trampoline)
2567 		bpf_trampoline_put(aux->dst_trampoline);
2568 	for (i = 0; i < aux->func_cnt; i++) {
2569 		/* We can just unlink the subprog poke descriptor table as
2570 		 * it was originally linked to the main program and is also
2571 		 * released along with it.
2572 		 */
2573 		aux->func[i]->aux->poke_tab = NULL;
2574 		bpf_jit_free(aux->func[i]);
2575 	}
2576 	if (aux->func_cnt) {
2577 		kfree(aux->func);
2578 		bpf_prog_unlock_free(aux->prog);
2579 	} else {
2580 		bpf_jit_free(aux->prog);
2581 	}
2582 }
2583 
2584 void bpf_prog_free(struct bpf_prog *fp)
2585 {
2586 	struct bpf_prog_aux *aux = fp->aux;
2587 
2588 	if (aux->dst_prog)
2589 		bpf_prog_put(aux->dst_prog);
2590 	INIT_WORK(&aux->work, bpf_prog_free_deferred);
2591 	schedule_work(&aux->work);
2592 }
2593 EXPORT_SYMBOL_GPL(bpf_prog_free);
2594 
2595 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2596 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2597 
2598 void bpf_user_rnd_init_once(void)
2599 {
2600 	prandom_init_once(&bpf_user_rnd_state);
2601 }
2602 
2603 BPF_CALL_0(bpf_user_rnd_u32)
2604 {
2605 	/* Should someone ever have the rather unwise idea to use some
2606 	 * of the registers passed into this function, then note that
2607 	 * this function is called from native eBPF and classic-to-eBPF
2608 	 * transformations. Register assignments from both sides are
2609 	 * different, f.e. classic always sets fn(ctx, A, X) here.
2610 	 */
2611 	struct rnd_state *state;
2612 	u32 res;
2613 
2614 	state = &get_cpu_var(bpf_user_rnd_state);
2615 	res = prandom_u32_state(state);
2616 	put_cpu_var(bpf_user_rnd_state);
2617 
2618 	return res;
2619 }
2620 
2621 BPF_CALL_0(bpf_get_raw_cpu_id)
2622 {
2623 	return raw_smp_processor_id();
2624 }
2625 
2626 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2627 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2628 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2629 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2630 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2631 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2632 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2633 const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto __weak;
2634 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2635 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2636 const struct bpf_func_proto bpf_jiffies64_proto __weak;
2637 
2638 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2639 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2640 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2641 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2642 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
2643 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto __weak;
2644 
2645 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2646 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2647 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2648 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2649 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
2650 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2651 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
2652 const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
2653 const struct bpf_func_proto bpf_seq_printf_btf_proto __weak;
2654 
2655 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2656 {
2657 	return NULL;
2658 }
2659 
2660 const struct bpf_func_proto * __weak bpf_get_trace_vprintk_proto(void)
2661 {
2662 	return NULL;
2663 }
2664 
2665 u64 __weak
2666 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2667 		 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2668 {
2669 	return -ENOTSUPP;
2670 }
2671 EXPORT_SYMBOL_GPL(bpf_event_output);
2672 
2673 /* Always built-in helper functions. */
2674 const struct bpf_func_proto bpf_tail_call_proto = {
2675 	.func		= NULL,
2676 	.gpl_only	= false,
2677 	.ret_type	= RET_VOID,
2678 	.arg1_type	= ARG_PTR_TO_CTX,
2679 	.arg2_type	= ARG_CONST_MAP_PTR,
2680 	.arg3_type	= ARG_ANYTHING,
2681 };
2682 
2683 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2684  * It is encouraged to implement bpf_int_jit_compile() instead, so that
2685  * eBPF and implicitly also cBPF can get JITed!
2686  */
2687 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2688 {
2689 	return prog;
2690 }
2691 
2692 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2693  * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2694  */
2695 void __weak bpf_jit_compile(struct bpf_prog *prog)
2696 {
2697 }
2698 
2699 bool __weak bpf_helper_changes_pkt_data(void *func)
2700 {
2701 	return false;
2702 }
2703 
2704 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2705  * analysis code and wants explicit zero extension inserted by verifier.
2706  * Otherwise, return FALSE.
2707  *
2708  * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
2709  * you don't override this. JITs that don't want these extra insns can detect
2710  * them using insn_is_zext.
2711  */
2712 bool __weak bpf_jit_needs_zext(void)
2713 {
2714 	return false;
2715 }
2716 
2717 bool __weak bpf_jit_supports_kfunc_call(void)
2718 {
2719 	return false;
2720 }
2721 
2722 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2723  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2724  */
2725 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2726 			 int len)
2727 {
2728 	return -EFAULT;
2729 }
2730 
2731 int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
2732 			      void *addr1, void *addr2)
2733 {
2734 	return -ENOTSUPP;
2735 }
2736 
2737 void * __weak bpf_arch_text_copy(void *dst, void *src, size_t len)
2738 {
2739 	return ERR_PTR(-ENOTSUPP);
2740 }
2741 
2742 int __weak bpf_arch_text_invalidate(void *dst, size_t len)
2743 {
2744 	return -ENOTSUPP;
2745 }
2746 
2747 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2748 EXPORT_SYMBOL(bpf_stats_enabled_key);
2749 
2750 /* All definitions of tracepoints related to BPF. */
2751 #define CREATE_TRACE_POINTS
2752 #include <linux/bpf_trace.h>
2753 
2754 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2755 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
2756