2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
24 #include <linux/filter.h>
25 #include <linux/skbuff.h>
26 #include <linux/vmalloc.h>
27 #include <linux/random.h>
28 #include <linux/moduleloader.h>
29 #include <linux/bpf.h>
30 #include <linux/frame.h>
31 #include <linux/rbtree_latch.h>
32 #include <linux/kallsyms.h>
33 #include <linux/rcupdate.h>
34 #include <linux/perf_event.h>
36 #include <asm/barrier.h>
37 #include <asm/unaligned.h>
40 #define BPF_R0 regs[BPF_REG_0]
41 #define BPF_R1 regs[BPF_REG_1]
42 #define BPF_R2 regs[BPF_REG_2]
43 #define BPF_R3 regs[BPF_REG_3]
44 #define BPF_R4 regs[BPF_REG_4]
45 #define BPF_R5 regs[BPF_REG_5]
46 #define BPF_R6 regs[BPF_REG_6]
47 #define BPF_R7 regs[BPF_REG_7]
48 #define BPF_R8 regs[BPF_REG_8]
49 #define BPF_R9 regs[BPF_REG_9]
50 #define BPF_R10 regs[BPF_REG_10]
53 #define DST regs[insn->dst_reg]
54 #define SRC regs[insn->src_reg]
55 #define FP regs[BPF_REG_FP]
56 #define AX regs[BPF_REG_AX]
57 #define ARG1 regs[BPF_REG_ARG1]
58 #define CTX regs[BPF_REG_CTX]
61 /* No hurry in this branch
63 * Exported for the bpf jit load helper.
65 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
70 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
71 else if (k >= SKF_LL_OFF)
72 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
74 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
80 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
82 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
83 struct bpf_prog_aux *aux;
86 size = round_up(size, PAGE_SIZE);
87 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
91 aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
97 fp->pages = size / PAGE_SIZE;
100 fp->jit_requested = ebpf_jit_enabled();
102 INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode);
106 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
108 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
109 gfp_t gfp_extra_flags)
111 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
116 BUG_ON(fp_old == NULL);
118 size = round_up(size, PAGE_SIZE);
119 pages = size / PAGE_SIZE;
120 if (pages <= fp_old->pages)
123 delta = pages - fp_old->pages;
124 ret = __bpf_prog_charge(fp_old->aux->user, delta);
128 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
130 __bpf_prog_uncharge(fp_old->aux->user, delta);
132 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
136 /* We keep fp->aux from fp_old around in the new
137 * reallocated structure.
140 __bpf_prog_free(fp_old);
146 void __bpf_prog_free(struct bpf_prog *fp)
152 int bpf_prog_calc_tag(struct bpf_prog *fp)
154 const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64);
155 u32 raw_size = bpf_prog_tag_scratch_size(fp);
156 u32 digest[SHA_DIGEST_WORDS];
157 u32 ws[SHA_WORKSPACE_WORDS];
158 u32 i, bsize, psize, blocks;
159 struct bpf_insn *dst;
165 raw = vmalloc(raw_size);
170 memset(ws, 0, sizeof(ws));
172 /* We need to take out the map fd for the digest calculation
173 * since they are unstable from user space side.
176 for (i = 0, was_ld_map = false; i < fp->len; i++) {
177 dst[i] = fp->insnsi[i];
179 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
180 dst[i].src_reg == BPF_PSEUDO_MAP_FD) {
183 } else if (was_ld_map &&
185 dst[i].dst_reg == 0 &&
186 dst[i].src_reg == 0 &&
195 psize = bpf_prog_insn_size(fp);
196 memset(&raw[psize], 0, raw_size - psize);
199 bsize = round_up(psize, SHA_MESSAGE_BYTES);
200 blocks = bsize / SHA_MESSAGE_BYTES;
202 if (bsize - psize >= sizeof(__be64)) {
203 bits = (__be64 *)(todo + bsize - sizeof(__be64));
205 bits = (__be64 *)(todo + bsize + bits_offset);
208 *bits = cpu_to_be64((psize - 1) << 3);
211 sha_transform(digest, todo, ws);
212 todo += SHA_MESSAGE_BYTES;
215 result = (__force __be32 *)digest;
216 for (i = 0; i < SHA_DIGEST_WORDS; i++)
217 result[i] = cpu_to_be32(digest[i]);
218 memcpy(fp->tag, result, sizeof(fp->tag));
224 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, u32 delta,
225 u32 curr, const bool probe_pass)
227 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
230 if (curr < pos && curr + imm + 1 > pos)
232 else if (curr > pos + delta && curr + imm + 1 <= pos + delta)
234 if (imm < imm_min || imm > imm_max)
241 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, u32 delta,
242 u32 curr, const bool probe_pass)
244 const s32 off_min = S16_MIN, off_max = S16_MAX;
247 if (curr < pos && curr + off + 1 > pos)
249 else if (curr > pos + delta && curr + off + 1 <= pos + delta)
251 if (off < off_min || off > off_max)
258 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta,
259 const bool probe_pass)
261 u32 i, insn_cnt = prog->len + (probe_pass ? delta : 0);
262 struct bpf_insn *insn = prog->insnsi;
265 for (i = 0; i < insn_cnt; i++, insn++) {
268 /* In the probing pass we still operate on the original,
269 * unpatched image in order to check overflows before we
270 * do any other adjustments. Therefore skip the patchlet.
272 if (probe_pass && i == pos) {
277 if (BPF_CLASS(code) != BPF_JMP ||
278 BPF_OP(code) == BPF_EXIT)
280 /* Adjust offset of jmps if we cross patch boundaries. */
281 if (BPF_OP(code) == BPF_CALL) {
282 if (insn->src_reg != BPF_PSEUDO_CALL)
284 ret = bpf_adj_delta_to_imm(insn, pos, delta, i,
287 ret = bpf_adj_delta_to_off(insn, pos, delta, i,
297 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
298 const struct bpf_insn *patch, u32 len)
300 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
301 const u32 cnt_max = S16_MAX;
302 struct bpf_prog *prog_adj;
304 /* Since our patchlet doesn't expand the image, we're done. */
305 if (insn_delta == 0) {
306 memcpy(prog->insnsi + off, patch, sizeof(*patch));
310 insn_adj_cnt = prog->len + insn_delta;
312 /* Reject anything that would potentially let the insn->off
313 * target overflow when we have excessive program expansions.
314 * We need to probe here before we do any reallocation where
315 * we afterwards may not fail anymore.
317 if (insn_adj_cnt > cnt_max &&
318 bpf_adj_branches(prog, off, insn_delta, true))
321 /* Several new instructions need to be inserted. Make room
322 * for them. Likely, there's no need for a new allocation as
323 * last page could have large enough tailroom.
325 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
330 prog_adj->len = insn_adj_cnt;
332 /* Patching happens in 3 steps:
334 * 1) Move over tail of insnsi from next instruction onwards,
335 * so we can patch the single target insn with one or more
336 * new ones (patching is always from 1 to n insns, n > 0).
337 * 2) Inject new instructions at the target location.
338 * 3) Adjust branch offsets if necessary.
340 insn_rest = insn_adj_cnt - off - len;
342 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
343 sizeof(*patch) * insn_rest);
344 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
346 /* We are guaranteed to not fail at this point, otherwise
347 * the ship has sailed to reverse to the original state. An
348 * overflow cannot happen at this point.
350 BUG_ON(bpf_adj_branches(prog_adj, off, insn_delta, false));
355 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
359 for (i = 0; i < fp->aux->func_cnt; i++)
360 bpf_prog_kallsyms_del(fp->aux->func[i]);
363 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
365 bpf_prog_kallsyms_del_subprogs(fp);
366 bpf_prog_kallsyms_del(fp);
369 #ifdef CONFIG_BPF_JIT
370 /* All BPF JIT sysctl knobs here. */
371 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
372 int bpf_jit_harden __read_mostly;
373 int bpf_jit_kallsyms __read_mostly;
374 long bpf_jit_limit __read_mostly;
375 long bpf_jit_limit_max __read_mostly;
377 static __always_inline void
378 bpf_get_prog_addr_region(const struct bpf_prog *prog,
379 unsigned long *symbol_start,
380 unsigned long *symbol_end)
382 const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
383 unsigned long addr = (unsigned long)hdr;
385 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
387 *symbol_start = addr;
388 *symbol_end = addr + hdr->pages * PAGE_SIZE;
391 static void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
393 const char *end = sym + KSYM_NAME_LEN;
395 BUILD_BUG_ON(sizeof("bpf_prog_") +
396 sizeof(prog->tag) * 2 +
397 /* name has been null terminated.
398 * We should need +1 for the '_' preceding
399 * the name. However, the null character
400 * is double counted between the name and the
401 * sizeof("bpf_prog_") above, so we omit
404 sizeof(prog->aux->name) > KSYM_NAME_LEN);
406 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
407 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
408 if (prog->aux->name[0])
409 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
414 static __always_inline unsigned long
415 bpf_get_prog_addr_start(struct latch_tree_node *n)
417 unsigned long symbol_start, symbol_end;
418 const struct bpf_prog_aux *aux;
420 aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
421 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
426 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
427 struct latch_tree_node *b)
429 return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b);
432 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
434 unsigned long val = (unsigned long)key;
435 unsigned long symbol_start, symbol_end;
436 const struct bpf_prog_aux *aux;
438 aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
439 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
441 if (val < symbol_start)
443 if (val >= symbol_end)
449 static const struct latch_tree_ops bpf_tree_ops = {
450 .less = bpf_tree_less,
451 .comp = bpf_tree_comp,
454 static DEFINE_SPINLOCK(bpf_lock);
455 static LIST_HEAD(bpf_kallsyms);
456 static struct latch_tree_root bpf_tree __cacheline_aligned;
458 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
460 WARN_ON_ONCE(!list_empty(&aux->ksym_lnode));
461 list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms);
462 latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
465 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
467 if (list_empty(&aux->ksym_lnode))
470 latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
471 list_del_rcu(&aux->ksym_lnode);
474 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
476 return fp->jited && !bpf_prog_was_classic(fp);
479 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
481 return list_empty(&fp->aux->ksym_lnode) ||
482 fp->aux->ksym_lnode.prev == LIST_POISON2;
485 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
487 if (!bpf_prog_kallsyms_candidate(fp) ||
488 !capable(CAP_SYS_ADMIN))
491 spin_lock_bh(&bpf_lock);
492 bpf_prog_ksym_node_add(fp->aux);
493 spin_unlock_bh(&bpf_lock);
496 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
498 if (!bpf_prog_kallsyms_candidate(fp))
501 spin_lock_bh(&bpf_lock);
502 bpf_prog_ksym_node_del(fp->aux);
503 spin_unlock_bh(&bpf_lock);
506 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
508 struct latch_tree_node *n;
510 if (!bpf_jit_kallsyms_enabled())
513 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
515 container_of(n, struct bpf_prog_aux, ksym_tnode)->prog :
519 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
520 unsigned long *off, char *sym)
522 unsigned long symbol_start, symbol_end;
523 struct bpf_prog *prog;
527 prog = bpf_prog_kallsyms_find(addr);
529 bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end);
530 bpf_get_prog_name(prog, sym);
534 *size = symbol_end - symbol_start;
536 *off = addr - symbol_start;
543 bool is_bpf_text_address(unsigned long addr)
548 ret = bpf_prog_kallsyms_find(addr) != NULL;
554 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
557 unsigned long symbol_start, symbol_end;
558 struct bpf_prog_aux *aux;
562 if (!bpf_jit_kallsyms_enabled())
566 list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) {
570 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
571 bpf_get_prog_name(aux->prog, sym);
573 *value = symbol_start;
574 *type = BPF_SYM_ELF_TYPE;
584 static atomic_long_t bpf_jit_current;
586 /* Can be overridden by an arch's JIT compiler if it has a custom,
587 * dedicated BPF backend memory area, or if neither of the two
590 u64 __weak bpf_jit_alloc_exec_limit(void)
592 #if defined(MODULES_VADDR)
593 return MODULES_END - MODULES_VADDR;
595 return VMALLOC_END - VMALLOC_START;
599 static int __init bpf_jit_charge_init(void)
601 /* Only used as heuristic here to derive limit. */
602 bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
603 bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
604 PAGE_SIZE), LONG_MAX);
607 pure_initcall(bpf_jit_charge_init);
609 static int bpf_jit_charge_modmem(u32 pages)
611 if (atomic_long_add_return(pages, &bpf_jit_current) >
612 (bpf_jit_limit >> PAGE_SHIFT)) {
613 if (!capable(CAP_SYS_ADMIN)) {
614 atomic_long_sub(pages, &bpf_jit_current);
622 static void bpf_jit_uncharge_modmem(u32 pages)
624 atomic_long_sub(pages, &bpf_jit_current);
627 struct bpf_binary_header *
628 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
629 unsigned int alignment,
630 bpf_jit_fill_hole_t bpf_fill_ill_insns)
632 struct bpf_binary_header *hdr;
633 u32 size, hole, start, pages;
635 /* Most of BPF filters are really small, but if some of them
636 * fill a page, allow at least 128 extra bytes to insert a
637 * random section of illegal instructions.
639 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
640 pages = size / PAGE_SIZE;
642 if (bpf_jit_charge_modmem(pages))
644 hdr = module_alloc(size);
646 bpf_jit_uncharge_modmem(pages);
650 /* Fill space with illegal/arch-dep instructions. */
651 bpf_fill_ill_insns(hdr, size);
654 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
655 PAGE_SIZE - sizeof(*hdr));
656 start = (get_random_int() % hole) & ~(alignment - 1);
658 /* Leave a random number of instructions before BPF code. */
659 *image_ptr = &hdr->image[start];
664 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
666 u32 pages = hdr->pages;
669 bpf_jit_uncharge_modmem(pages);
672 /* This symbol is only overridden by archs that have different
673 * requirements than the usual eBPF JITs, f.e. when they only
674 * implement cBPF JIT, do not set images read-only, etc.
676 void __weak bpf_jit_free(struct bpf_prog *fp)
679 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
681 bpf_jit_binary_unlock_ro(hdr);
682 bpf_jit_binary_free(hdr);
684 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
687 bpf_prog_unlock_free(fp);
690 static int bpf_jit_blind_insn(const struct bpf_insn *from,
691 const struct bpf_insn *aux,
692 struct bpf_insn *to_buff)
694 struct bpf_insn *to = to_buff;
695 u32 imm_rnd = get_random_int();
698 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
699 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
701 /* Constraints on AX register:
703 * AX register is inaccessible from user space. It is mapped in
704 * all JITs, and used here for constant blinding rewrites. It is
705 * typically "stateless" meaning its contents are only valid within
706 * the executed instruction, but not across several instructions.
707 * There are a few exceptions however which are further detailed
710 * Constant blinding is only used by JITs, not in the interpreter.
711 * In restricted circumstances, the verifier can also use the AX
712 * register for rewrites as long as they do not interfere with
715 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
718 if (from->imm == 0 &&
719 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
720 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
721 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
725 switch (from->code) {
726 case BPF_ALU | BPF_ADD | BPF_K:
727 case BPF_ALU | BPF_SUB | BPF_K:
728 case BPF_ALU | BPF_AND | BPF_K:
729 case BPF_ALU | BPF_OR | BPF_K:
730 case BPF_ALU | BPF_XOR | BPF_K:
731 case BPF_ALU | BPF_MUL | BPF_K:
732 case BPF_ALU | BPF_MOV | BPF_K:
733 case BPF_ALU | BPF_DIV | BPF_K:
734 case BPF_ALU | BPF_MOD | BPF_K:
735 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
736 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
737 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
740 case BPF_ALU64 | BPF_ADD | BPF_K:
741 case BPF_ALU64 | BPF_SUB | BPF_K:
742 case BPF_ALU64 | BPF_AND | BPF_K:
743 case BPF_ALU64 | BPF_OR | BPF_K:
744 case BPF_ALU64 | BPF_XOR | BPF_K:
745 case BPF_ALU64 | BPF_MUL | BPF_K:
746 case BPF_ALU64 | BPF_MOV | BPF_K:
747 case BPF_ALU64 | BPF_DIV | BPF_K:
748 case BPF_ALU64 | BPF_MOD | BPF_K:
749 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
750 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
751 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
754 case BPF_JMP | BPF_JEQ | BPF_K:
755 case BPF_JMP | BPF_JNE | BPF_K:
756 case BPF_JMP | BPF_JGT | BPF_K:
757 case BPF_JMP | BPF_JLT | BPF_K:
758 case BPF_JMP | BPF_JGE | BPF_K:
759 case BPF_JMP | BPF_JLE | BPF_K:
760 case BPF_JMP | BPF_JSGT | BPF_K:
761 case BPF_JMP | BPF_JSLT | BPF_K:
762 case BPF_JMP | BPF_JSGE | BPF_K:
763 case BPF_JMP | BPF_JSLE | BPF_K:
764 case BPF_JMP | BPF_JSET | BPF_K:
765 /* Accommodate for extra offset in case of a backjump. */
769 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
770 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
771 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
774 case BPF_LD | BPF_IMM | BPF_DW:
775 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
776 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
777 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
778 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
780 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
781 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
782 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
783 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
786 case BPF_ST | BPF_MEM | BPF_DW:
787 case BPF_ST | BPF_MEM | BPF_W:
788 case BPF_ST | BPF_MEM | BPF_H:
789 case BPF_ST | BPF_MEM | BPF_B:
790 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
791 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
792 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
799 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
800 gfp_t gfp_extra_flags)
802 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
805 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
807 /* aux->prog still points to the fp_other one, so
808 * when promoting the clone to the real program,
809 * this still needs to be adapted.
811 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
817 static void bpf_prog_clone_free(struct bpf_prog *fp)
819 /* aux was stolen by the other clone, so we cannot free
820 * it from this path! It will be freed eventually by the
821 * other program on release.
823 * At this point, we don't need a deferred release since
824 * clone is guaranteed to not be locked.
830 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
832 /* We have to repoint aux->prog to self, as we don't
833 * know whether fp here is the clone or the original.
836 bpf_prog_clone_free(fp_other);
839 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
841 struct bpf_insn insn_buff[16], aux[2];
842 struct bpf_prog *clone, *tmp;
843 int insn_delta, insn_cnt;
844 struct bpf_insn *insn;
847 if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
850 clone = bpf_prog_clone_create(prog, GFP_USER);
852 return ERR_PTR(-ENOMEM);
854 insn_cnt = clone->len;
855 insn = clone->insnsi;
857 for (i = 0; i < insn_cnt; i++, insn++) {
858 /* We temporarily need to hold the original ld64 insn
859 * so that we can still access the first part in the
860 * second blinding run.
862 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
864 memcpy(aux, insn, sizeof(aux));
866 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
870 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
872 /* Patching may have repointed aux->prog during
873 * realloc from the original one, so we need to
874 * fix it up here on error.
876 bpf_jit_prog_release_other(prog, clone);
877 return ERR_PTR(-ENOMEM);
881 insn_delta = rewritten - 1;
883 /* Walk new program and skip insns we just inserted. */
884 insn = clone->insnsi + i + insn_delta;
885 insn_cnt += insn_delta;
892 #endif /* CONFIG_BPF_JIT */
894 /* Base function for offset calculation. Needs to go into .text section,
895 * therefore keeping it non-static as well; will also be used by JITs
896 * anyway later on, so do not let the compiler omit it. This also needs
897 * to go into kallsyms for correlation from e.g. bpftool, so naming
900 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
904 EXPORT_SYMBOL_GPL(__bpf_call_base);
906 /* All UAPI available opcodes. */
907 #define BPF_INSN_MAP(INSN_2, INSN_3) \
908 /* 32 bit ALU operations. */ \
909 /* Register based. */ \
910 INSN_3(ALU, ADD, X), \
911 INSN_3(ALU, SUB, X), \
912 INSN_3(ALU, AND, X), \
913 INSN_3(ALU, OR, X), \
914 INSN_3(ALU, LSH, X), \
915 INSN_3(ALU, RSH, X), \
916 INSN_3(ALU, XOR, X), \
917 INSN_3(ALU, MUL, X), \
918 INSN_3(ALU, MOV, X), \
919 INSN_3(ALU, DIV, X), \
920 INSN_3(ALU, MOD, X), \
922 INSN_3(ALU, END, TO_BE), \
923 INSN_3(ALU, END, TO_LE), \
924 /* Immediate based. */ \
925 INSN_3(ALU, ADD, K), \
926 INSN_3(ALU, SUB, K), \
927 INSN_3(ALU, AND, K), \
928 INSN_3(ALU, OR, K), \
929 INSN_3(ALU, LSH, K), \
930 INSN_3(ALU, RSH, K), \
931 INSN_3(ALU, XOR, K), \
932 INSN_3(ALU, MUL, K), \
933 INSN_3(ALU, MOV, K), \
934 INSN_3(ALU, DIV, K), \
935 INSN_3(ALU, MOD, K), \
936 /* 64 bit ALU operations. */ \
937 /* Register based. */ \
938 INSN_3(ALU64, ADD, X), \
939 INSN_3(ALU64, SUB, X), \
940 INSN_3(ALU64, AND, X), \
941 INSN_3(ALU64, OR, X), \
942 INSN_3(ALU64, LSH, X), \
943 INSN_3(ALU64, RSH, X), \
944 INSN_3(ALU64, XOR, X), \
945 INSN_3(ALU64, MUL, X), \
946 INSN_3(ALU64, MOV, X), \
947 INSN_3(ALU64, ARSH, X), \
948 INSN_3(ALU64, DIV, X), \
949 INSN_3(ALU64, MOD, X), \
950 INSN_2(ALU64, NEG), \
951 /* Immediate based. */ \
952 INSN_3(ALU64, ADD, K), \
953 INSN_3(ALU64, SUB, K), \
954 INSN_3(ALU64, AND, K), \
955 INSN_3(ALU64, OR, K), \
956 INSN_3(ALU64, LSH, K), \
957 INSN_3(ALU64, RSH, K), \
958 INSN_3(ALU64, XOR, K), \
959 INSN_3(ALU64, MUL, K), \
960 INSN_3(ALU64, MOV, K), \
961 INSN_3(ALU64, ARSH, K), \
962 INSN_3(ALU64, DIV, K), \
963 INSN_3(ALU64, MOD, K), \
964 /* Call instruction. */ \
966 /* Exit instruction. */ \
968 /* Jump instructions. */ \
969 /* Register based. */ \
970 INSN_3(JMP, JEQ, X), \
971 INSN_3(JMP, JNE, X), \
972 INSN_3(JMP, JGT, X), \
973 INSN_3(JMP, JLT, X), \
974 INSN_3(JMP, JGE, X), \
975 INSN_3(JMP, JLE, X), \
976 INSN_3(JMP, JSGT, X), \
977 INSN_3(JMP, JSLT, X), \
978 INSN_3(JMP, JSGE, X), \
979 INSN_3(JMP, JSLE, X), \
980 INSN_3(JMP, JSET, X), \
981 /* Immediate based. */ \
982 INSN_3(JMP, JEQ, K), \
983 INSN_3(JMP, JNE, K), \
984 INSN_3(JMP, JGT, K), \
985 INSN_3(JMP, JLT, K), \
986 INSN_3(JMP, JGE, K), \
987 INSN_3(JMP, JLE, K), \
988 INSN_3(JMP, JSGT, K), \
989 INSN_3(JMP, JSLT, K), \
990 INSN_3(JMP, JSGE, K), \
991 INSN_3(JMP, JSLE, K), \
992 INSN_3(JMP, JSET, K), \
994 /* Store instructions. */ \
995 /* Register based. */ \
996 INSN_3(STX, MEM, B), \
997 INSN_3(STX, MEM, H), \
998 INSN_3(STX, MEM, W), \
999 INSN_3(STX, MEM, DW), \
1000 INSN_3(STX, XADD, W), \
1001 INSN_3(STX, XADD, DW), \
1002 /* Immediate based. */ \
1003 INSN_3(ST, MEM, B), \
1004 INSN_3(ST, MEM, H), \
1005 INSN_3(ST, MEM, W), \
1006 INSN_3(ST, MEM, DW), \
1007 /* Load instructions. */ \
1008 /* Register based. */ \
1009 INSN_3(LDX, MEM, B), \
1010 INSN_3(LDX, MEM, H), \
1011 INSN_3(LDX, MEM, W), \
1012 INSN_3(LDX, MEM, DW), \
1013 /* Immediate based. */ \
1016 bool bpf_opcode_in_insntable(u8 code)
1018 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1019 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1020 static const bool public_insntable[256] = {
1021 [0 ... 255] = false,
1022 /* Now overwrite non-defaults ... */
1023 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1024 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1025 [BPF_LD | BPF_ABS | BPF_B] = true,
1026 [BPF_LD | BPF_ABS | BPF_H] = true,
1027 [BPF_LD | BPF_ABS | BPF_W] = true,
1028 [BPF_LD | BPF_IND | BPF_B] = true,
1029 [BPF_LD | BPF_IND | BPF_H] = true,
1030 [BPF_LD | BPF_IND | BPF_W] = true,
1032 #undef BPF_INSN_3_TBL
1033 #undef BPF_INSN_2_TBL
1034 return public_insntable[code];
1037 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1039 * __bpf_prog_run - run eBPF program on a given context
1040 * @ctx: is the data we are operating on
1041 * @insn: is the array of eBPF instructions
1043 * Decode and execute eBPF instructions.
1045 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1047 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1048 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1049 static const void *jumptable[256] = {
1050 [0 ... 255] = &&default_label,
1051 /* Now overwrite non-defaults ... */
1052 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1053 /* Non-UAPI available opcodes. */
1054 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1055 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1056 [BPF_ST | BPF_NOSPEC] = &&ST_NOSPEC,
1058 #undef BPF_INSN_3_LBL
1059 #undef BPF_INSN_2_LBL
1060 u32 tail_call_cnt = 0;
1063 #define CONT ({ insn++; goto select_insn; })
1064 #define CONT_JMP ({ insn++; goto select_insn; })
1067 goto *jumptable[insn->code];
1070 #define ALU(OPCODE, OP) \
1071 ALU64_##OPCODE##_X: \
1075 DST = (u32) DST OP (u32) SRC; \
1077 ALU64_##OPCODE##_K: \
1081 DST = (u32) DST OP (u32) IMM; \
1112 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1116 (*(s64 *) &DST) >>= SRC;
1119 (*(s64 *) &DST) >>= IMM;
1122 div64_u64_rem(DST, SRC, &tmp);
1127 DST = do_div(tmp, (u32) SRC);
1130 div64_u64_rem(DST, IMM, &tmp);
1135 DST = do_div(tmp, (u32) IMM);
1138 DST = div64_u64(DST, SRC);
1142 do_div(tmp, (u32) SRC);
1146 DST = div64_u64(DST, IMM);
1150 do_div(tmp, (u32) IMM);
1156 DST = (__force u16) cpu_to_be16(DST);
1159 DST = (__force u32) cpu_to_be32(DST);
1162 DST = (__force u64) cpu_to_be64(DST);
1169 DST = (__force u16) cpu_to_le16(DST);
1172 DST = (__force u32) cpu_to_le32(DST);
1175 DST = (__force u64) cpu_to_le64(DST);
1182 /* Function call scratches BPF_R1-BPF_R5 registers,
1183 * preserves BPF_R6-BPF_R9, and stores return value
1186 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1191 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1194 insn + insn->off + 1);
1198 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1199 struct bpf_array *array = container_of(map, struct bpf_array, map);
1200 struct bpf_prog *prog;
1203 if (unlikely(index >= array->map.max_entries))
1205 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1210 prog = READ_ONCE(array->ptrs[index]);
1214 /* ARG1 at this point is guaranteed to point to CTX from
1215 * the verifier side due to the fact that the tail call is
1216 * handeled like a helper, that is, bpf_tail_call_proto,
1217 * where arg1_type is ARG_PTR_TO_CTX.
1219 insn = prog->insnsi;
1301 if (((s64) DST) > ((s64) SRC)) {
1307 if (((s64) DST) > ((s64) IMM)) {
1313 if (((s64) DST) < ((s64) SRC)) {
1319 if (((s64) DST) < ((s64) IMM)) {
1325 if (((s64) DST) >= ((s64) SRC)) {
1331 if (((s64) DST) >= ((s64) IMM)) {
1337 if (((s64) DST) <= ((s64) SRC)) {
1343 if (((s64) DST) <= ((s64) IMM)) {
1363 /* ST, STX and LDX*/
1365 /* Speculation barrier for mitigating Speculative Store Bypass.
1366 * In case of arm64, we rely on the firmware mitigation as
1367 * controlled via the ssbd kernel parameter. Whenever the
1368 * mitigation is enabled, it works for all of the kernel code
1369 * with no need to provide any additional instructions here.
1370 * In case of x86, we use 'lfence' insn for mitigation. We
1371 * reuse preexisting logic from Spectre v1 mitigation that
1372 * happens to produce the required code on x86 for v4 as well.
1378 #define LDST(SIZEOP, SIZE) \
1380 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1383 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1386 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1394 STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1395 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1398 STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1399 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1404 /* If we ever reach this, we have a bug somewhere. Die hard here
1405 * instead of just returning 0; we could be somewhere in a subprog,
1406 * so execution could continue otherwise which we do /not/ want.
1408 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1410 pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1414 STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */
1416 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1417 #define DEFINE_BPF_PROG_RUN(stack_size) \
1418 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1420 u64 stack[stack_size / sizeof(u64)]; \
1421 u64 regs[MAX_BPF_EXT_REG]; \
1423 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1424 ARG1 = (u64) (unsigned long) ctx; \
1425 return ___bpf_prog_run(regs, insn, stack); \
1428 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1429 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1430 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1431 const struct bpf_insn *insn) \
1433 u64 stack[stack_size / sizeof(u64)]; \
1434 u64 regs[MAX_BPF_EXT_REG]; \
1436 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1442 return ___bpf_prog_run(regs, insn, stack); \
1445 #define EVAL1(FN, X) FN(X)
1446 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1447 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1448 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1449 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1450 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1452 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1453 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1454 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1456 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1457 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1458 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1460 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1462 static unsigned int (*interpreters[])(const void *ctx,
1463 const struct bpf_insn *insn) = {
1464 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1465 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1466 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1468 #undef PROG_NAME_LIST
1469 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1470 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1471 const struct bpf_insn *insn) = {
1472 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1473 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1474 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1476 #undef PROG_NAME_LIST
1478 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1480 stack_depth = max_t(u32, stack_depth, 1);
1481 insn->off = (s16) insn->imm;
1482 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1483 __bpf_call_base_args;
1484 insn->code = BPF_JMP | BPF_CALL_ARGS;
1488 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1489 const struct bpf_insn *insn)
1491 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1492 * is not working properly, so warn about it!
1499 bool bpf_prog_array_compatible(struct bpf_array *array,
1500 const struct bpf_prog *fp)
1502 if (fp->kprobe_override)
1505 if (!array->owner_prog_type) {
1506 /* There's no owner yet where we could check for
1509 array->owner_prog_type = fp->type;
1510 array->owner_jited = fp->jited;
1515 return array->owner_prog_type == fp->type &&
1516 array->owner_jited == fp->jited;
1519 static int bpf_check_tail_call(const struct bpf_prog *fp)
1521 struct bpf_prog_aux *aux = fp->aux;
1524 for (i = 0; i < aux->used_map_cnt; i++) {
1525 struct bpf_map *map = aux->used_maps[i];
1526 struct bpf_array *array;
1528 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1531 array = container_of(map, struct bpf_array, map);
1532 if (!bpf_prog_array_compatible(array, fp))
1539 static void bpf_prog_select_func(struct bpf_prog *fp)
1541 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1542 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1544 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1546 fp->bpf_func = __bpf_prog_ret0_warn;
1551 * bpf_prog_select_runtime - select exec runtime for BPF program
1552 * @fp: bpf_prog populated with internal BPF program
1553 * @err: pointer to error variable
1555 * Try to JIT eBPF program, if JIT is not available, use interpreter.
1556 * The BPF program will be executed via BPF_PROG_RUN() macro.
1558 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1560 /* In case of BPF to BPF calls, verifier did all the prep
1561 * work with regards to JITing, etc.
1566 bpf_prog_select_func(fp);
1568 /* eBPF JITs can rewrite the program in case constant
1569 * blinding is active. However, in case of error during
1570 * blinding, bpf_int_jit_compile() must always return a
1571 * valid program, which in this case would simply not
1572 * be JITed, but falls back to the interpreter.
1574 if (!bpf_prog_is_dev_bound(fp->aux)) {
1575 fp = bpf_int_jit_compile(fp);
1576 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1583 *err = bpf_prog_offload_compile(fp);
1589 bpf_prog_lock_ro(fp);
1591 /* The tail call compatibility check can only be done at
1592 * this late stage as we need to determine, if we deal
1593 * with JITed or non JITed program concatenations and not
1594 * all eBPF JITs might immediately support all features.
1596 *err = bpf_check_tail_call(fp);
1600 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1602 static unsigned int __bpf_prog_ret1(const void *ctx,
1603 const struct bpf_insn *insn)
1608 static struct bpf_prog_dummy {
1609 struct bpf_prog prog;
1610 } dummy_bpf_prog = {
1612 .bpf_func = __bpf_prog_ret1,
1616 /* to avoid allocating empty bpf_prog_array for cgroups that
1617 * don't have bpf program attached use one global 'empty_prog_array'
1618 * It will not be modified the caller of bpf_prog_array_alloc()
1619 * (since caller requested prog_cnt == 0)
1620 * that pointer should be 'freed' by bpf_prog_array_free()
1623 struct bpf_prog_array hdr;
1624 struct bpf_prog *null_prog;
1625 } empty_prog_array = {
1629 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1632 return kzalloc(sizeof(struct bpf_prog_array) +
1633 sizeof(struct bpf_prog_array_item) *
1637 return &empty_prog_array.hdr;
1640 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs)
1643 progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr)
1645 kfree_rcu(progs, rcu);
1648 int bpf_prog_array_length(struct bpf_prog_array __rcu *array)
1650 struct bpf_prog_array_item *item;
1654 item = rcu_dereference(array)->items;
1655 for (; item->prog; item++)
1656 if (item->prog != &dummy_bpf_prog.prog)
1663 static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array,
1667 struct bpf_prog_array_item *item;
1670 item = rcu_dereference_check(array, 1)->items;
1671 for (; item->prog; item++) {
1672 if (item->prog == &dummy_bpf_prog.prog)
1674 prog_ids[i] = item->prog->aux->id;
1675 if (++i == request_cnt) {
1681 return !!(item->prog);
1684 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array,
1685 __u32 __user *prog_ids, u32 cnt)
1687 unsigned long err = 0;
1691 /* users of this function are doing:
1692 * cnt = bpf_prog_array_length();
1694 * bpf_prog_array_copy_to_user(..., cnt);
1695 * so below kcalloc doesn't need extra cnt > 0 check, but
1696 * bpf_prog_array_length() releases rcu lock and
1697 * prog array could have been swapped with empty or larger array,
1698 * so always copy 'cnt' prog_ids to the user.
1699 * In a rare race the user will see zero prog_ids
1701 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1705 nospc = bpf_prog_array_copy_core(array, ids, cnt);
1707 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1716 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array,
1717 struct bpf_prog *old_prog)
1719 struct bpf_prog_array_item *item = array->items;
1721 for (; item->prog; item++)
1722 if (item->prog == old_prog) {
1723 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1728 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array,
1729 struct bpf_prog *exclude_prog,
1730 struct bpf_prog *include_prog,
1731 struct bpf_prog_array **new_array)
1733 int new_prog_cnt, carry_prog_cnt = 0;
1734 struct bpf_prog_array_item *existing;
1735 struct bpf_prog_array *array;
1736 bool found_exclude = false;
1737 int new_prog_idx = 0;
1739 /* Figure out how many existing progs we need to carry over to
1743 existing = old_array->items;
1744 for (; existing->prog; existing++) {
1745 if (existing->prog == exclude_prog) {
1746 found_exclude = true;
1749 if (existing->prog != &dummy_bpf_prog.prog)
1751 if (existing->prog == include_prog)
1756 if (exclude_prog && !found_exclude)
1759 /* How many progs (not NULL) will be in the new array? */
1760 new_prog_cnt = carry_prog_cnt;
1764 /* Do we have any prog (not NULL) in the new array? */
1765 if (!new_prog_cnt) {
1770 /* +1 as the end of prog_array is marked with NULL */
1771 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
1775 /* Fill in the new prog array */
1776 if (carry_prog_cnt) {
1777 existing = old_array->items;
1778 for (; existing->prog; existing++)
1779 if (existing->prog != exclude_prog &&
1780 existing->prog != &dummy_bpf_prog.prog) {
1781 array->items[new_prog_idx++].prog =
1786 array->items[new_prog_idx++].prog = include_prog;
1787 array->items[new_prog_idx].prog = NULL;
1792 int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array,
1793 u32 *prog_ids, u32 request_cnt,
1799 cnt = bpf_prog_array_length(array);
1803 /* return early if user requested only program count or nothing to copy */
1804 if (!request_cnt || !cnt)
1807 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
1808 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
1812 static void bpf_prog_free_deferred(struct work_struct *work)
1814 struct bpf_prog_aux *aux;
1817 aux = container_of(work, struct bpf_prog_aux, work);
1818 if (bpf_prog_is_dev_bound(aux))
1819 bpf_prog_offload_destroy(aux->prog);
1820 #ifdef CONFIG_PERF_EVENTS
1821 if (aux->prog->has_callchain_buf)
1822 put_callchain_buffers();
1824 for (i = 0; i < aux->func_cnt; i++)
1825 bpf_jit_free(aux->func[i]);
1826 if (aux->func_cnt) {
1828 bpf_prog_unlock_free(aux->prog);
1830 bpf_jit_free(aux->prog);
1834 /* Free internal BPF program */
1835 void bpf_prog_free(struct bpf_prog *fp)
1837 struct bpf_prog_aux *aux = fp->aux;
1839 INIT_WORK(&aux->work, bpf_prog_free_deferred);
1840 schedule_work(&aux->work);
1842 EXPORT_SYMBOL_GPL(bpf_prog_free);
1844 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
1845 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
1847 void bpf_user_rnd_init_once(void)
1849 prandom_init_once(&bpf_user_rnd_state);
1852 BPF_CALL_0(bpf_user_rnd_u32)
1854 /* Should someone ever have the rather unwise idea to use some
1855 * of the registers passed into this function, then note that
1856 * this function is called from native eBPF and classic-to-eBPF
1857 * transformations. Register assignments from both sides are
1858 * different, f.e. classic always sets fn(ctx, A, X) here.
1860 struct rnd_state *state;
1863 state = &get_cpu_var(bpf_user_rnd_state);
1864 res = prandom_u32_state(state);
1865 put_cpu_var(bpf_user_rnd_state);
1870 /* Weak definitions of helper functions in case we don't have bpf syscall. */
1871 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
1872 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
1873 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
1875 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
1876 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
1877 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
1878 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
1880 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
1881 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
1882 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
1883 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
1884 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
1885 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
1886 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
1888 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
1894 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
1895 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
1899 EXPORT_SYMBOL_GPL(bpf_event_output);
1901 /* Always built-in helper functions. */
1902 const struct bpf_func_proto bpf_tail_call_proto = {
1905 .ret_type = RET_VOID,
1906 .arg1_type = ARG_PTR_TO_CTX,
1907 .arg2_type = ARG_CONST_MAP_PTR,
1908 .arg3_type = ARG_ANYTHING,
1911 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
1912 * It is encouraged to implement bpf_int_jit_compile() instead, so that
1913 * eBPF and implicitly also cBPF can get JITed!
1915 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
1920 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
1921 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
1923 void __weak bpf_jit_compile(struct bpf_prog *prog)
1927 bool __weak bpf_helper_changes_pkt_data(void *func)
1932 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
1933 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
1935 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
1941 /* All definitions of tracepoints related to BPF. */
1942 #define CREATE_TRACE_POINTS
1943 #include <linux/bpf_trace.h>
1945 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);