1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
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 <asm/unaligned.h>
38 #define BPF_R0 regs[BPF_REG_0]
39 #define BPF_R1 regs[BPF_REG_1]
40 #define BPF_R2 regs[BPF_REG_2]
41 #define BPF_R3 regs[BPF_REG_3]
42 #define BPF_R4 regs[BPF_REG_4]
43 #define BPF_R5 regs[BPF_REG_5]
44 #define BPF_R6 regs[BPF_REG_6]
45 #define BPF_R7 regs[BPF_REG_7]
46 #define BPF_R8 regs[BPF_REG_8]
47 #define BPF_R9 regs[BPF_REG_9]
48 #define BPF_R10 regs[BPF_REG_10]
51 #define DST regs[insn->dst_reg]
52 #define SRC regs[insn->src_reg]
53 #define FP regs[BPF_REG_FP]
54 #define AX regs[BPF_REG_AX]
55 #define ARG1 regs[BPF_REG_ARG1]
56 #define CTX regs[BPF_REG_CTX]
59 /* No hurry in this branch
61 * Exported for the bpf jit load helper.
63 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
68 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
69 else if (k >= SKF_LL_OFF)
70 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
72 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
78 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
80 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
81 struct bpf_prog_aux *aux;
84 size = round_up(size, PAGE_SIZE);
85 fp = __vmalloc(size, gfp_flags);
89 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT | gfp_extra_flags);
94 fp->active = alloc_percpu_gfp(int, GFP_KERNEL_ACCOUNT | gfp_extra_flags);
101 fp->pages = size / PAGE_SIZE;
104 fp->jit_requested = ebpf_jit_enabled();
106 INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
107 mutex_init(&fp->aux->used_maps_mutex);
108 mutex_init(&fp->aux->dst_mutex);
113 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
115 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
116 struct bpf_prog *prog;
119 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
123 prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
125 free_percpu(prog->active);
131 for_each_possible_cpu(cpu) {
132 struct bpf_prog_stats *pstats;
134 pstats = per_cpu_ptr(prog->stats, cpu);
135 u64_stats_init(&pstats->syncp);
139 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
141 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
143 if (!prog->aux->nr_linfo || !prog->jit_requested)
146 prog->aux->jited_linfo = kvcalloc(prog->aux->nr_linfo,
147 sizeof(*prog->aux->jited_linfo),
148 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
149 if (!prog->aux->jited_linfo)
155 void bpf_prog_jit_attempt_done(struct bpf_prog *prog)
157 if (prog->aux->jited_linfo &&
158 (!prog->jited || !prog->aux->jited_linfo[0])) {
159 kvfree(prog->aux->jited_linfo);
160 prog->aux->jited_linfo = NULL;
164 /* The jit engine is responsible to provide an array
165 * for insn_off to the jited_off mapping (insn_to_jit_off).
167 * The idx to this array is the insn_off. Hence, the insn_off
168 * here is relative to the prog itself instead of the main prog.
169 * This array has one entry for each xlated bpf insn.
171 * jited_off is the byte off to the last byte of the jited insn.
175 * The first bpf insn off of the prog. The insn off
176 * here is relative to the main prog.
177 * e.g. if prog is a subprog, insn_start > 0
179 * The prog's idx to prog->aux->linfo and jited_linfo
181 * jited_linfo[linfo_idx] = prog->bpf_func
185 * jited_linfo[i] = prog->bpf_func +
186 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
188 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
189 const u32 *insn_to_jit_off)
191 u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
192 const struct bpf_line_info *linfo;
195 if (!prog->aux->jited_linfo)
196 /* Userspace did not provide linfo */
199 linfo_idx = prog->aux->linfo_idx;
200 linfo = &prog->aux->linfo[linfo_idx];
201 insn_start = linfo[0].insn_off;
202 insn_end = insn_start + prog->len;
204 jited_linfo = &prog->aux->jited_linfo[linfo_idx];
205 jited_linfo[0] = prog->bpf_func;
207 nr_linfo = prog->aux->nr_linfo - linfo_idx;
209 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
210 /* The verifier ensures that linfo[i].insn_off is
211 * strictly increasing
213 jited_linfo[i] = prog->bpf_func +
214 insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
217 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
218 gfp_t gfp_extra_flags)
220 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
224 size = round_up(size, PAGE_SIZE);
225 pages = size / PAGE_SIZE;
226 if (pages <= fp_old->pages)
229 fp = __vmalloc(size, gfp_flags);
231 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
235 /* We keep fp->aux from fp_old around in the new
236 * reallocated structure.
239 fp_old->stats = NULL;
240 fp_old->active = NULL;
241 __bpf_prog_free(fp_old);
247 void __bpf_prog_free(struct bpf_prog *fp)
250 mutex_destroy(&fp->aux->used_maps_mutex);
251 mutex_destroy(&fp->aux->dst_mutex);
252 kfree(fp->aux->poke_tab);
255 free_percpu(fp->stats);
256 free_percpu(fp->active);
260 int bpf_prog_calc_tag(struct bpf_prog *fp)
262 const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
263 u32 raw_size = bpf_prog_tag_scratch_size(fp);
264 u32 digest[SHA1_DIGEST_WORDS];
265 u32 ws[SHA1_WORKSPACE_WORDS];
266 u32 i, bsize, psize, blocks;
267 struct bpf_insn *dst;
273 raw = vmalloc(raw_size);
278 memset(ws, 0, sizeof(ws));
280 /* We need to take out the map fd for the digest calculation
281 * since they are unstable from user space side.
284 for (i = 0, was_ld_map = false; i < fp->len; i++) {
285 dst[i] = fp->insnsi[i];
287 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
288 (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
289 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
292 } else if (was_ld_map &&
294 dst[i].dst_reg == 0 &&
295 dst[i].src_reg == 0 &&
304 psize = bpf_prog_insn_size(fp);
305 memset(&raw[psize], 0, raw_size - psize);
308 bsize = round_up(psize, SHA1_BLOCK_SIZE);
309 blocks = bsize / SHA1_BLOCK_SIZE;
311 if (bsize - psize >= sizeof(__be64)) {
312 bits = (__be64 *)(todo + bsize - sizeof(__be64));
314 bits = (__be64 *)(todo + bsize + bits_offset);
317 *bits = cpu_to_be64((psize - 1) << 3);
320 sha1_transform(digest, todo, ws);
321 todo += SHA1_BLOCK_SIZE;
324 result = (__force __be32 *)digest;
325 for (i = 0; i < SHA1_DIGEST_WORDS; i++)
326 result[i] = cpu_to_be32(digest[i]);
327 memcpy(fp->tag, result, sizeof(fp->tag));
333 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
334 s32 end_new, s32 curr, const bool probe_pass)
336 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
337 s32 delta = end_new - end_old;
340 if (curr < pos && curr + imm + 1 >= end_old)
342 else if (curr >= end_new && curr + imm + 1 < end_new)
344 if (imm < imm_min || imm > imm_max)
351 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
352 s32 end_new, s32 curr, const bool probe_pass)
354 const s32 off_min = S16_MIN, off_max = S16_MAX;
355 s32 delta = end_new - end_old;
358 if (curr < pos && curr + off + 1 >= end_old)
360 else if (curr >= end_new && curr + off + 1 < end_new)
362 if (off < off_min || off > off_max)
369 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
370 s32 end_new, const bool probe_pass)
372 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
373 struct bpf_insn *insn = prog->insnsi;
376 for (i = 0; i < insn_cnt; i++, insn++) {
379 /* In the probing pass we still operate on the original,
380 * unpatched image in order to check overflows before we
381 * do any other adjustments. Therefore skip the patchlet.
383 if (probe_pass && i == pos) {
385 insn = prog->insnsi + end_old;
388 if ((BPF_CLASS(code) != BPF_JMP &&
389 BPF_CLASS(code) != BPF_JMP32) ||
390 BPF_OP(code) == BPF_EXIT)
392 /* Adjust offset of jmps if we cross patch boundaries. */
393 if (BPF_OP(code) == BPF_CALL) {
394 if (insn->src_reg != BPF_PSEUDO_CALL)
396 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
397 end_new, i, probe_pass);
399 ret = bpf_adj_delta_to_off(insn, pos, end_old,
400 end_new, i, probe_pass);
409 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
411 struct bpf_line_info *linfo;
414 nr_linfo = prog->aux->nr_linfo;
415 if (!nr_linfo || !delta)
418 linfo = prog->aux->linfo;
420 for (i = 0; i < nr_linfo; i++)
421 if (off < linfo[i].insn_off)
424 /* Push all off < linfo[i].insn_off by delta */
425 for (; i < nr_linfo; i++)
426 linfo[i].insn_off += delta;
429 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
430 const struct bpf_insn *patch, u32 len)
432 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
433 const u32 cnt_max = S16_MAX;
434 struct bpf_prog *prog_adj;
437 /* Since our patchlet doesn't expand the image, we're done. */
438 if (insn_delta == 0) {
439 memcpy(prog->insnsi + off, patch, sizeof(*patch));
443 insn_adj_cnt = prog->len + insn_delta;
445 /* Reject anything that would potentially let the insn->off
446 * target overflow when we have excessive program expansions.
447 * We need to probe here before we do any reallocation where
448 * we afterwards may not fail anymore.
450 if (insn_adj_cnt > cnt_max &&
451 (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
454 /* Several new instructions need to be inserted. Make room
455 * for them. Likely, there's no need for a new allocation as
456 * last page could have large enough tailroom.
458 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
461 return ERR_PTR(-ENOMEM);
463 prog_adj->len = insn_adj_cnt;
465 /* Patching happens in 3 steps:
467 * 1) Move over tail of insnsi from next instruction onwards,
468 * so we can patch the single target insn with one or more
469 * new ones (patching is always from 1 to n insns, n > 0).
470 * 2) Inject new instructions at the target location.
471 * 3) Adjust branch offsets if necessary.
473 insn_rest = insn_adj_cnt - off - len;
475 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
476 sizeof(*patch) * insn_rest);
477 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
479 /* We are guaranteed to not fail at this point, otherwise
480 * the ship has sailed to reverse to the original state. An
481 * overflow cannot happen at this point.
483 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
485 bpf_adj_linfo(prog_adj, off, insn_delta);
490 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
492 /* Branch offsets can't overflow when program is shrinking, no need
493 * to call bpf_adj_branches(..., true) here
495 memmove(prog->insnsi + off, prog->insnsi + off + cnt,
496 sizeof(struct bpf_insn) * (prog->len - off - cnt));
499 return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
502 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
506 for (i = 0; i < fp->aux->func_cnt; i++)
507 bpf_prog_kallsyms_del(fp->aux->func[i]);
510 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
512 bpf_prog_kallsyms_del_subprogs(fp);
513 bpf_prog_kallsyms_del(fp);
516 #ifdef CONFIG_BPF_JIT
517 /* All BPF JIT sysctl knobs here. */
518 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
519 int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
520 int bpf_jit_harden __read_mostly;
521 long bpf_jit_limit __read_mostly;
524 bpf_prog_ksym_set_addr(struct bpf_prog *prog)
526 const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
527 unsigned long addr = (unsigned long)hdr;
529 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
531 prog->aux->ksym.start = (unsigned long) prog->bpf_func;
532 prog->aux->ksym.end = addr + hdr->pages * PAGE_SIZE;
536 bpf_prog_ksym_set_name(struct bpf_prog *prog)
538 char *sym = prog->aux->ksym.name;
539 const char *end = sym + KSYM_NAME_LEN;
540 const struct btf_type *type;
541 const char *func_name;
543 BUILD_BUG_ON(sizeof("bpf_prog_") +
544 sizeof(prog->tag) * 2 +
545 /* name has been null terminated.
546 * We should need +1 for the '_' preceding
547 * the name. However, the null character
548 * is double counted between the name and the
549 * sizeof("bpf_prog_") above, so we omit
552 sizeof(prog->aux->name) > KSYM_NAME_LEN);
554 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
555 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
557 /* prog->aux->name will be ignored if full btf name is available */
558 if (prog->aux->func_info_cnt) {
559 type = btf_type_by_id(prog->aux->btf,
560 prog->aux->func_info[prog->aux->func_idx].type_id);
561 func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
562 snprintf(sym, (size_t)(end - sym), "_%s", func_name);
566 if (prog->aux->name[0])
567 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
572 static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
574 return container_of(n, struct bpf_ksym, tnode)->start;
577 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
578 struct latch_tree_node *b)
580 return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
583 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
585 unsigned long val = (unsigned long)key;
586 const struct bpf_ksym *ksym;
588 ksym = container_of(n, struct bpf_ksym, tnode);
590 if (val < ksym->start)
592 if (val >= ksym->end)
598 static const struct latch_tree_ops bpf_tree_ops = {
599 .less = bpf_tree_less,
600 .comp = bpf_tree_comp,
603 static DEFINE_SPINLOCK(bpf_lock);
604 static LIST_HEAD(bpf_kallsyms);
605 static struct latch_tree_root bpf_tree __cacheline_aligned;
607 void bpf_ksym_add(struct bpf_ksym *ksym)
609 spin_lock_bh(&bpf_lock);
610 WARN_ON_ONCE(!list_empty(&ksym->lnode));
611 list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
612 latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
613 spin_unlock_bh(&bpf_lock);
616 static void __bpf_ksym_del(struct bpf_ksym *ksym)
618 if (list_empty(&ksym->lnode))
621 latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
622 list_del_rcu(&ksym->lnode);
625 void bpf_ksym_del(struct bpf_ksym *ksym)
627 spin_lock_bh(&bpf_lock);
628 __bpf_ksym_del(ksym);
629 spin_unlock_bh(&bpf_lock);
632 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
634 return fp->jited && !bpf_prog_was_classic(fp);
637 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
639 return list_empty(&fp->aux->ksym.lnode) ||
640 fp->aux->ksym.lnode.prev == LIST_POISON2;
643 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
645 if (!bpf_prog_kallsyms_candidate(fp) ||
649 bpf_prog_ksym_set_addr(fp);
650 bpf_prog_ksym_set_name(fp);
651 fp->aux->ksym.prog = true;
653 bpf_ksym_add(&fp->aux->ksym);
656 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
658 if (!bpf_prog_kallsyms_candidate(fp))
661 bpf_ksym_del(&fp->aux->ksym);
664 static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
666 struct latch_tree_node *n;
668 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
669 return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
672 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
673 unsigned long *off, char *sym)
675 struct bpf_ksym *ksym;
679 ksym = bpf_ksym_find(addr);
681 unsigned long symbol_start = ksym->start;
682 unsigned long symbol_end = ksym->end;
684 strncpy(sym, ksym->name, KSYM_NAME_LEN);
688 *size = symbol_end - symbol_start;
690 *off = addr - symbol_start;
697 bool is_bpf_text_address(unsigned long addr)
702 ret = bpf_ksym_find(addr) != NULL;
708 static struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
710 struct bpf_ksym *ksym = bpf_ksym_find(addr);
712 return ksym && ksym->prog ?
713 container_of(ksym, struct bpf_prog_aux, ksym)->prog :
717 const struct exception_table_entry *search_bpf_extables(unsigned long addr)
719 const struct exception_table_entry *e = NULL;
720 struct bpf_prog *prog;
723 prog = bpf_prog_ksym_find(addr);
726 if (!prog->aux->num_exentries)
729 e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
735 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
738 struct bpf_ksym *ksym;
742 if (!bpf_jit_kallsyms_enabled())
746 list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
750 strncpy(sym, ksym->name, KSYM_NAME_LEN);
752 *value = ksym->start;
753 *type = BPF_SYM_ELF_TYPE;
763 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
764 struct bpf_jit_poke_descriptor *poke)
766 struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
767 static const u32 poke_tab_max = 1024;
768 u32 slot = prog->aux->size_poke_tab;
771 if (size > poke_tab_max)
773 if (poke->tailcall_target || poke->tailcall_target_stable ||
774 poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
777 switch (poke->reason) {
778 case BPF_POKE_REASON_TAIL_CALL:
779 if (!poke->tail_call.map)
786 tab = krealloc(tab, size * sizeof(*poke), GFP_KERNEL);
790 memcpy(&tab[slot], poke, sizeof(*poke));
791 prog->aux->size_poke_tab = size;
792 prog->aux->poke_tab = tab;
797 static atomic_long_t bpf_jit_current;
799 /* Can be overridden by an arch's JIT compiler if it has a custom,
800 * dedicated BPF backend memory area, or if neither of the two
803 u64 __weak bpf_jit_alloc_exec_limit(void)
805 #if defined(MODULES_VADDR)
806 return MODULES_END - MODULES_VADDR;
808 return VMALLOC_END - VMALLOC_START;
812 static int __init bpf_jit_charge_init(void)
814 /* Only used as heuristic here to derive limit. */
815 bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2,
816 PAGE_SIZE), LONG_MAX);
819 pure_initcall(bpf_jit_charge_init);
821 int bpf_jit_charge_modmem(u32 pages)
823 if (atomic_long_add_return(pages, &bpf_jit_current) >
824 (bpf_jit_limit >> PAGE_SHIFT)) {
825 if (!capable(CAP_SYS_ADMIN)) {
826 atomic_long_sub(pages, &bpf_jit_current);
834 void bpf_jit_uncharge_modmem(u32 pages)
836 atomic_long_sub(pages, &bpf_jit_current);
839 void *__weak bpf_jit_alloc_exec(unsigned long size)
841 return module_alloc(size);
844 void __weak bpf_jit_free_exec(void *addr)
846 module_memfree(addr);
849 struct bpf_binary_header *
850 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
851 unsigned int alignment,
852 bpf_jit_fill_hole_t bpf_fill_ill_insns)
854 struct bpf_binary_header *hdr;
855 u32 size, hole, start, pages;
857 WARN_ON_ONCE(!is_power_of_2(alignment) ||
858 alignment > BPF_IMAGE_ALIGNMENT);
860 /* Most of BPF filters are really small, but if some of them
861 * fill a page, allow at least 128 extra bytes to insert a
862 * random section of illegal instructions.
864 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
865 pages = size / PAGE_SIZE;
867 if (bpf_jit_charge_modmem(pages))
869 hdr = bpf_jit_alloc_exec(size);
871 bpf_jit_uncharge_modmem(pages);
875 /* Fill space with illegal/arch-dep instructions. */
876 bpf_fill_ill_insns(hdr, size);
879 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
880 PAGE_SIZE - sizeof(*hdr));
881 start = (get_random_int() % hole) & ~(alignment - 1);
883 /* Leave a random number of instructions before BPF code. */
884 *image_ptr = &hdr->image[start];
889 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
891 u32 pages = hdr->pages;
893 bpf_jit_free_exec(hdr);
894 bpf_jit_uncharge_modmem(pages);
897 /* This symbol is only overridden by archs that have different
898 * requirements than the usual eBPF JITs, f.e. when they only
899 * implement cBPF JIT, do not set images read-only, etc.
901 void __weak bpf_jit_free(struct bpf_prog *fp)
904 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
906 bpf_jit_binary_free(hdr);
908 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
911 bpf_prog_unlock_free(fp);
914 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
915 const struct bpf_insn *insn, bool extra_pass,
916 u64 *func_addr, bool *func_addr_fixed)
922 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
923 if (!*func_addr_fixed) {
924 /* Place-holder address till the last pass has collected
925 * all addresses for JITed subprograms in which case we
926 * can pick them up from prog->aux.
930 else if (prog->aux->func &&
931 off >= 0 && off < prog->aux->func_cnt)
932 addr = (u8 *)prog->aux->func[off]->bpf_func;
936 /* Address of a BPF helper call. Since part of the core
937 * kernel, it's always at a fixed location. __bpf_call_base
938 * and the helper with imm relative to it are both in core
941 addr = (u8 *)__bpf_call_base + imm;
944 *func_addr = (unsigned long)addr;
948 static int bpf_jit_blind_insn(const struct bpf_insn *from,
949 const struct bpf_insn *aux,
950 struct bpf_insn *to_buff,
953 struct bpf_insn *to = to_buff;
954 u32 imm_rnd = get_random_int();
957 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
958 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
960 /* Constraints on AX register:
962 * AX register is inaccessible from user space. It is mapped in
963 * all JITs, and used here for constant blinding rewrites. It is
964 * typically "stateless" meaning its contents are only valid within
965 * the executed instruction, but not across several instructions.
966 * There are a few exceptions however which are further detailed
969 * Constant blinding is only used by JITs, not in the interpreter.
970 * The interpreter uses AX in some occasions as a local temporary
971 * register e.g. in DIV or MOD instructions.
973 * In restricted circumstances, the verifier can also use the AX
974 * register for rewrites as long as they do not interfere with
977 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
980 if (from->imm == 0 &&
981 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
982 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
983 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
987 switch (from->code) {
988 case BPF_ALU | BPF_ADD | BPF_K:
989 case BPF_ALU | BPF_SUB | BPF_K:
990 case BPF_ALU | BPF_AND | BPF_K:
991 case BPF_ALU | BPF_OR | BPF_K:
992 case BPF_ALU | BPF_XOR | BPF_K:
993 case BPF_ALU | BPF_MUL | BPF_K:
994 case BPF_ALU | BPF_MOV | BPF_K:
995 case BPF_ALU | BPF_DIV | BPF_K:
996 case BPF_ALU | BPF_MOD | BPF_K:
997 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
998 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
999 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
1002 case BPF_ALU64 | BPF_ADD | BPF_K:
1003 case BPF_ALU64 | BPF_SUB | BPF_K:
1004 case BPF_ALU64 | BPF_AND | BPF_K:
1005 case BPF_ALU64 | BPF_OR | BPF_K:
1006 case BPF_ALU64 | BPF_XOR | BPF_K:
1007 case BPF_ALU64 | BPF_MUL | BPF_K:
1008 case BPF_ALU64 | BPF_MOV | BPF_K:
1009 case BPF_ALU64 | BPF_DIV | BPF_K:
1010 case BPF_ALU64 | BPF_MOD | BPF_K:
1011 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1012 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1013 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
1016 case BPF_JMP | BPF_JEQ | BPF_K:
1017 case BPF_JMP | BPF_JNE | BPF_K:
1018 case BPF_JMP | BPF_JGT | BPF_K:
1019 case BPF_JMP | BPF_JLT | BPF_K:
1020 case BPF_JMP | BPF_JGE | BPF_K:
1021 case BPF_JMP | BPF_JLE | BPF_K:
1022 case BPF_JMP | BPF_JSGT | BPF_K:
1023 case BPF_JMP | BPF_JSLT | BPF_K:
1024 case BPF_JMP | BPF_JSGE | BPF_K:
1025 case BPF_JMP | BPF_JSLE | BPF_K:
1026 case BPF_JMP | BPF_JSET | BPF_K:
1027 /* Accommodate for extra offset in case of a backjump. */
1031 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1032 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1033 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1036 case BPF_JMP32 | BPF_JEQ | BPF_K:
1037 case BPF_JMP32 | BPF_JNE | BPF_K:
1038 case BPF_JMP32 | BPF_JGT | BPF_K:
1039 case BPF_JMP32 | BPF_JLT | BPF_K:
1040 case BPF_JMP32 | BPF_JGE | BPF_K:
1041 case BPF_JMP32 | BPF_JLE | BPF_K:
1042 case BPF_JMP32 | BPF_JSGT | BPF_K:
1043 case BPF_JMP32 | BPF_JSLT | BPF_K:
1044 case BPF_JMP32 | BPF_JSGE | BPF_K:
1045 case BPF_JMP32 | BPF_JSLE | BPF_K:
1046 case BPF_JMP32 | BPF_JSET | BPF_K:
1047 /* Accommodate for extra offset in case of a backjump. */
1051 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1052 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1053 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1057 case BPF_LD | BPF_IMM | BPF_DW:
1058 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1059 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1060 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1061 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1063 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1064 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1065 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1067 *to++ = BPF_ZEXT_REG(BPF_REG_AX);
1068 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
1071 case BPF_ST | BPF_MEM | BPF_DW:
1072 case BPF_ST | BPF_MEM | BPF_W:
1073 case BPF_ST | BPF_MEM | BPF_H:
1074 case BPF_ST | BPF_MEM | BPF_B:
1075 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1076 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1077 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1081 return to - to_buff;
1084 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1085 gfp_t gfp_extra_flags)
1087 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1088 struct bpf_prog *fp;
1090 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
1092 /* aux->prog still points to the fp_other one, so
1093 * when promoting the clone to the real program,
1094 * this still needs to be adapted.
1096 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1102 static void bpf_prog_clone_free(struct bpf_prog *fp)
1104 /* aux was stolen by the other clone, so we cannot free
1105 * it from this path! It will be freed eventually by the
1106 * other program on release.
1108 * At this point, we don't need a deferred release since
1109 * clone is guaranteed to not be locked.
1114 __bpf_prog_free(fp);
1117 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1119 /* We have to repoint aux->prog to self, as we don't
1120 * know whether fp here is the clone or the original.
1123 bpf_prog_clone_free(fp_other);
1126 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1128 struct bpf_insn insn_buff[16], aux[2];
1129 struct bpf_prog *clone, *tmp;
1130 int insn_delta, insn_cnt;
1131 struct bpf_insn *insn;
1134 if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1137 clone = bpf_prog_clone_create(prog, GFP_USER);
1139 return ERR_PTR(-ENOMEM);
1141 insn_cnt = clone->len;
1142 insn = clone->insnsi;
1144 for (i = 0; i < insn_cnt; i++, insn++) {
1145 /* We temporarily need to hold the original ld64 insn
1146 * so that we can still access the first part in the
1147 * second blinding run.
1149 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1151 memcpy(aux, insn, sizeof(aux));
1153 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1154 clone->aux->verifier_zext);
1158 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1160 /* Patching may have repointed aux->prog during
1161 * realloc from the original one, so we need to
1162 * fix it up here on error.
1164 bpf_jit_prog_release_other(prog, clone);
1169 insn_delta = rewritten - 1;
1171 /* Walk new program and skip insns we just inserted. */
1172 insn = clone->insnsi + i + insn_delta;
1173 insn_cnt += insn_delta;
1180 #endif /* CONFIG_BPF_JIT */
1182 /* Base function for offset calculation. Needs to go into .text section,
1183 * therefore keeping it non-static as well; will also be used by JITs
1184 * anyway later on, so do not let the compiler omit it. This also needs
1185 * to go into kallsyms for correlation from e.g. bpftool, so naming
1188 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1192 EXPORT_SYMBOL_GPL(__bpf_call_base);
1194 /* All UAPI available opcodes. */
1195 #define BPF_INSN_MAP(INSN_2, INSN_3) \
1196 /* 32 bit ALU operations. */ \
1197 /* Register based. */ \
1198 INSN_3(ALU, ADD, X), \
1199 INSN_3(ALU, SUB, X), \
1200 INSN_3(ALU, AND, X), \
1201 INSN_3(ALU, OR, X), \
1202 INSN_3(ALU, LSH, X), \
1203 INSN_3(ALU, RSH, X), \
1204 INSN_3(ALU, XOR, X), \
1205 INSN_3(ALU, MUL, X), \
1206 INSN_3(ALU, MOV, X), \
1207 INSN_3(ALU, ARSH, X), \
1208 INSN_3(ALU, DIV, X), \
1209 INSN_3(ALU, MOD, X), \
1211 INSN_3(ALU, END, TO_BE), \
1212 INSN_3(ALU, END, TO_LE), \
1213 /* Immediate based. */ \
1214 INSN_3(ALU, ADD, K), \
1215 INSN_3(ALU, SUB, K), \
1216 INSN_3(ALU, AND, K), \
1217 INSN_3(ALU, OR, K), \
1218 INSN_3(ALU, LSH, K), \
1219 INSN_3(ALU, RSH, K), \
1220 INSN_3(ALU, XOR, K), \
1221 INSN_3(ALU, MUL, K), \
1222 INSN_3(ALU, MOV, K), \
1223 INSN_3(ALU, ARSH, K), \
1224 INSN_3(ALU, DIV, K), \
1225 INSN_3(ALU, MOD, K), \
1226 /* 64 bit ALU operations. */ \
1227 /* Register based. */ \
1228 INSN_3(ALU64, ADD, X), \
1229 INSN_3(ALU64, SUB, X), \
1230 INSN_3(ALU64, AND, X), \
1231 INSN_3(ALU64, OR, X), \
1232 INSN_3(ALU64, LSH, X), \
1233 INSN_3(ALU64, RSH, X), \
1234 INSN_3(ALU64, XOR, X), \
1235 INSN_3(ALU64, MUL, X), \
1236 INSN_3(ALU64, MOV, X), \
1237 INSN_3(ALU64, ARSH, X), \
1238 INSN_3(ALU64, DIV, X), \
1239 INSN_3(ALU64, MOD, X), \
1240 INSN_2(ALU64, NEG), \
1241 /* Immediate based. */ \
1242 INSN_3(ALU64, ADD, K), \
1243 INSN_3(ALU64, SUB, K), \
1244 INSN_3(ALU64, AND, K), \
1245 INSN_3(ALU64, OR, K), \
1246 INSN_3(ALU64, LSH, K), \
1247 INSN_3(ALU64, RSH, K), \
1248 INSN_3(ALU64, XOR, K), \
1249 INSN_3(ALU64, MUL, K), \
1250 INSN_3(ALU64, MOV, K), \
1251 INSN_3(ALU64, ARSH, K), \
1252 INSN_3(ALU64, DIV, K), \
1253 INSN_3(ALU64, MOD, K), \
1254 /* Call instruction. */ \
1255 INSN_2(JMP, CALL), \
1256 /* Exit instruction. */ \
1257 INSN_2(JMP, EXIT), \
1258 /* 32-bit Jump instructions. */ \
1259 /* Register based. */ \
1260 INSN_3(JMP32, JEQ, X), \
1261 INSN_3(JMP32, JNE, X), \
1262 INSN_3(JMP32, JGT, X), \
1263 INSN_3(JMP32, JLT, X), \
1264 INSN_3(JMP32, JGE, X), \
1265 INSN_3(JMP32, JLE, X), \
1266 INSN_3(JMP32, JSGT, X), \
1267 INSN_3(JMP32, JSLT, X), \
1268 INSN_3(JMP32, JSGE, X), \
1269 INSN_3(JMP32, JSLE, X), \
1270 INSN_3(JMP32, JSET, X), \
1271 /* Immediate based. */ \
1272 INSN_3(JMP32, JEQ, K), \
1273 INSN_3(JMP32, JNE, K), \
1274 INSN_3(JMP32, JGT, K), \
1275 INSN_3(JMP32, JLT, K), \
1276 INSN_3(JMP32, JGE, K), \
1277 INSN_3(JMP32, JLE, K), \
1278 INSN_3(JMP32, JSGT, K), \
1279 INSN_3(JMP32, JSLT, K), \
1280 INSN_3(JMP32, JSGE, K), \
1281 INSN_3(JMP32, JSLE, K), \
1282 INSN_3(JMP32, JSET, K), \
1283 /* Jump instructions. */ \
1284 /* Register based. */ \
1285 INSN_3(JMP, JEQ, X), \
1286 INSN_3(JMP, JNE, X), \
1287 INSN_3(JMP, JGT, X), \
1288 INSN_3(JMP, JLT, X), \
1289 INSN_3(JMP, JGE, X), \
1290 INSN_3(JMP, JLE, X), \
1291 INSN_3(JMP, JSGT, X), \
1292 INSN_3(JMP, JSLT, X), \
1293 INSN_3(JMP, JSGE, X), \
1294 INSN_3(JMP, JSLE, X), \
1295 INSN_3(JMP, JSET, X), \
1296 /* Immediate based. */ \
1297 INSN_3(JMP, JEQ, K), \
1298 INSN_3(JMP, JNE, K), \
1299 INSN_3(JMP, JGT, K), \
1300 INSN_3(JMP, JLT, K), \
1301 INSN_3(JMP, JGE, K), \
1302 INSN_3(JMP, JLE, K), \
1303 INSN_3(JMP, JSGT, K), \
1304 INSN_3(JMP, JSLT, K), \
1305 INSN_3(JMP, JSGE, K), \
1306 INSN_3(JMP, JSLE, K), \
1307 INSN_3(JMP, JSET, K), \
1309 /* Store instructions. */ \
1310 /* Register based. */ \
1311 INSN_3(STX, MEM, B), \
1312 INSN_3(STX, MEM, H), \
1313 INSN_3(STX, MEM, W), \
1314 INSN_3(STX, MEM, DW), \
1315 INSN_3(STX, ATOMIC, W), \
1316 INSN_3(STX, ATOMIC, DW), \
1317 /* Immediate based. */ \
1318 INSN_3(ST, MEM, B), \
1319 INSN_3(ST, MEM, H), \
1320 INSN_3(ST, MEM, W), \
1321 INSN_3(ST, MEM, DW), \
1322 /* Load instructions. */ \
1323 /* Register based. */ \
1324 INSN_3(LDX, MEM, B), \
1325 INSN_3(LDX, MEM, H), \
1326 INSN_3(LDX, MEM, W), \
1327 INSN_3(LDX, MEM, DW), \
1328 /* Immediate based. */ \
1331 bool bpf_opcode_in_insntable(u8 code)
1333 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1334 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1335 static const bool public_insntable[256] = {
1336 [0 ... 255] = false,
1337 /* Now overwrite non-defaults ... */
1338 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1339 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1340 [BPF_LD | BPF_ABS | BPF_B] = true,
1341 [BPF_LD | BPF_ABS | BPF_H] = true,
1342 [BPF_LD | BPF_ABS | BPF_W] = true,
1343 [BPF_LD | BPF_IND | BPF_B] = true,
1344 [BPF_LD | BPF_IND | BPF_H] = true,
1345 [BPF_LD | BPF_IND | BPF_W] = true,
1347 #undef BPF_INSN_3_TBL
1348 #undef BPF_INSN_2_TBL
1349 return public_insntable[code];
1352 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1353 u64 __weak bpf_probe_read_kernel(void *dst, u32 size, const void *unsafe_ptr)
1355 memset(dst, 0, size);
1360 * __bpf_prog_run - run eBPF program on a given context
1361 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1362 * @insn: is the array of eBPF instructions
1363 * @stack: is the eBPF storage stack
1365 * Decode and execute eBPF instructions.
1367 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1369 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1370 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1371 static const void * const jumptable[256] __annotate_jump_table = {
1372 [0 ... 255] = &&default_label,
1373 /* Now overwrite non-defaults ... */
1374 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1375 /* Non-UAPI available opcodes. */
1376 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1377 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1378 [BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1379 [BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1380 [BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1381 [BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1383 #undef BPF_INSN_3_LBL
1384 #undef BPF_INSN_2_LBL
1385 u32 tail_call_cnt = 0;
1387 #define CONT ({ insn++; goto select_insn; })
1388 #define CONT_JMP ({ insn++; goto select_insn; })
1391 goto *jumptable[insn->code];
1394 #define ALU(OPCODE, OP) \
1395 ALU64_##OPCODE##_X: \
1399 DST = (u32) DST OP (u32) SRC; \
1401 ALU64_##OPCODE##_K: \
1405 DST = (u32) DST OP (u32) IMM; \
1436 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1440 DST = (u64) (u32) (((s32) DST) >> SRC);
1443 DST = (u64) (u32) (((s32) DST) >> IMM);
1446 (*(s64 *) &DST) >>= SRC;
1449 (*(s64 *) &DST) >>= IMM;
1452 div64_u64_rem(DST, SRC, &AX);
1457 DST = do_div(AX, (u32) SRC);
1460 div64_u64_rem(DST, IMM, &AX);
1465 DST = do_div(AX, (u32) IMM);
1468 DST = div64_u64(DST, SRC);
1472 do_div(AX, (u32) SRC);
1476 DST = div64_u64(DST, IMM);
1480 do_div(AX, (u32) IMM);
1486 DST = (__force u16) cpu_to_be16(DST);
1489 DST = (__force u32) cpu_to_be32(DST);
1492 DST = (__force u64) cpu_to_be64(DST);
1499 DST = (__force u16) cpu_to_le16(DST);
1502 DST = (__force u32) cpu_to_le32(DST);
1505 DST = (__force u64) cpu_to_le64(DST);
1512 /* Function call scratches BPF_R1-BPF_R5 registers,
1513 * preserves BPF_R6-BPF_R9, and stores return value
1516 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1521 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1524 insn + insn->off + 1);
1528 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1529 struct bpf_array *array = container_of(map, struct bpf_array, map);
1530 struct bpf_prog *prog;
1533 if (unlikely(index >= array->map.max_entries))
1535 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1540 prog = READ_ONCE(array->ptrs[index]);
1544 /* ARG1 at this point is guaranteed to point to CTX from
1545 * the verifier side due to the fact that the tail call is
1546 * handled like a helper, that is, bpf_tail_call_proto,
1547 * where arg1_type is ARG_PTR_TO_CTX.
1549 insn = prog->insnsi;
1560 #define COND_JMP(SIGN, OPCODE, CMP_OP) \
1562 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \
1563 insn += insn->off; \
1567 JMP32_##OPCODE##_X: \
1568 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \
1569 insn += insn->off; \
1574 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \
1575 insn += insn->off; \
1579 JMP32_##OPCODE##_K: \
1580 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \
1581 insn += insn->off; \
1585 COND_JMP(u, JEQ, ==)
1586 COND_JMP(u, JNE, !=)
1589 COND_JMP(u, JGE, >=)
1590 COND_JMP(u, JLE, <=)
1591 COND_JMP(u, JSET, &)
1592 COND_JMP(s, JSGT, >)
1593 COND_JMP(s, JSLT, <)
1594 COND_JMP(s, JSGE, >=)
1595 COND_JMP(s, JSLE, <=)
1597 /* STX and ST and LDX*/
1598 #define LDST(SIZEOP, SIZE) \
1600 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1603 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1606 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1614 #define LDX_PROBE(SIZEOP, SIZE) \
1615 LDX_PROBE_MEM_##SIZEOP: \
1616 bpf_probe_read_kernel(&DST, SIZE, (const void *)(long) (SRC + insn->off)); \
1624 #define ATOMIC_ALU_OP(BOP, KOP) \
1626 if (BPF_SIZE(insn->code) == BPF_W) \
1627 atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
1628 (DST + insn->off)); \
1630 atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
1631 (DST + insn->off)); \
1633 case BOP | BPF_FETCH: \
1634 if (BPF_SIZE(insn->code) == BPF_W) \
1635 SRC = (u32) atomic_fetch_##KOP( \
1637 (atomic_t *)(unsigned long) (DST + insn->off)); \
1639 SRC = (u64) atomic64_fetch_##KOP( \
1641 (atomic64_t *)(unsigned long) (DST + insn->off)); \
1647 ATOMIC_ALU_OP(BPF_ADD, add)
1648 ATOMIC_ALU_OP(BPF_AND, and)
1649 ATOMIC_ALU_OP(BPF_OR, or)
1650 ATOMIC_ALU_OP(BPF_XOR, xor)
1651 #undef ATOMIC_ALU_OP
1654 if (BPF_SIZE(insn->code) == BPF_W)
1655 SRC = (u32) atomic_xchg(
1656 (atomic_t *)(unsigned long) (DST + insn->off),
1659 SRC = (u64) atomic64_xchg(
1660 (atomic64_t *)(unsigned long) (DST + insn->off),
1664 if (BPF_SIZE(insn->code) == BPF_W)
1665 BPF_R0 = (u32) atomic_cmpxchg(
1666 (atomic_t *)(unsigned long) (DST + insn->off),
1667 (u32) BPF_R0, (u32) SRC);
1669 BPF_R0 = (u64) atomic64_cmpxchg(
1670 (atomic64_t *)(unsigned long) (DST + insn->off),
1671 (u64) BPF_R0, (u64) SRC);
1680 /* If we ever reach this, we have a bug somewhere. Die hard here
1681 * instead of just returning 0; we could be somewhere in a subprog,
1682 * so execution could continue otherwise which we do /not/ want.
1684 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1686 pr_warn("BPF interpreter: unknown opcode %02x (imm: 0x%x)\n",
1687 insn->code, insn->imm);
1692 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1693 #define DEFINE_BPF_PROG_RUN(stack_size) \
1694 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1696 u64 stack[stack_size / sizeof(u64)]; \
1697 u64 regs[MAX_BPF_EXT_REG]; \
1699 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1700 ARG1 = (u64) (unsigned long) ctx; \
1701 return ___bpf_prog_run(regs, insn, stack); \
1704 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1705 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1706 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1707 const struct bpf_insn *insn) \
1709 u64 stack[stack_size / sizeof(u64)]; \
1710 u64 regs[MAX_BPF_EXT_REG]; \
1712 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1718 return ___bpf_prog_run(regs, insn, stack); \
1721 #define EVAL1(FN, X) FN(X)
1722 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1723 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1724 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1725 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1726 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1728 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1729 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1730 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1732 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1733 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1734 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1736 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1738 static unsigned int (*interpreters[])(const void *ctx,
1739 const struct bpf_insn *insn) = {
1740 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1741 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1742 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1744 #undef PROG_NAME_LIST
1745 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1746 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1747 const struct bpf_insn *insn) = {
1748 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1749 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1750 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1752 #undef PROG_NAME_LIST
1754 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1756 stack_depth = max_t(u32, stack_depth, 1);
1757 insn->off = (s16) insn->imm;
1758 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1759 __bpf_call_base_args;
1760 insn->code = BPF_JMP | BPF_CALL_ARGS;
1764 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1765 const struct bpf_insn *insn)
1767 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1768 * is not working properly, so warn about it!
1775 bool bpf_prog_array_compatible(struct bpf_array *array,
1776 const struct bpf_prog *fp)
1778 if (fp->kprobe_override)
1781 if (!array->aux->type) {
1782 /* There's no owner yet where we could check for
1785 array->aux->type = fp->type;
1786 array->aux->jited = fp->jited;
1790 return array->aux->type == fp->type &&
1791 array->aux->jited == fp->jited;
1794 static int bpf_check_tail_call(const struct bpf_prog *fp)
1796 struct bpf_prog_aux *aux = fp->aux;
1799 mutex_lock(&aux->used_maps_mutex);
1800 for (i = 0; i < aux->used_map_cnt; i++) {
1801 struct bpf_map *map = aux->used_maps[i];
1802 struct bpf_array *array;
1804 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1807 array = container_of(map, struct bpf_array, map);
1808 if (!bpf_prog_array_compatible(array, fp)) {
1815 mutex_unlock(&aux->used_maps_mutex);
1819 static void bpf_prog_select_func(struct bpf_prog *fp)
1821 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1822 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1824 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1826 fp->bpf_func = __bpf_prog_ret0_warn;
1831 * bpf_prog_select_runtime - select exec runtime for BPF program
1832 * @fp: bpf_prog populated with internal BPF program
1833 * @err: pointer to error variable
1835 * Try to JIT eBPF program, if JIT is not available, use interpreter.
1836 * The BPF program will be executed via BPF_PROG_RUN() macro.
1838 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1840 /* In case of BPF to BPF calls, verifier did all the prep
1841 * work with regards to JITing, etc.
1846 bpf_prog_select_func(fp);
1848 /* eBPF JITs can rewrite the program in case constant
1849 * blinding is active. However, in case of error during
1850 * blinding, bpf_int_jit_compile() must always return a
1851 * valid program, which in this case would simply not
1852 * be JITed, but falls back to the interpreter.
1854 if (!bpf_prog_is_dev_bound(fp->aux)) {
1855 *err = bpf_prog_alloc_jited_linfo(fp);
1859 fp = bpf_int_jit_compile(fp);
1860 bpf_prog_jit_attempt_done(fp);
1861 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1868 *err = bpf_prog_offload_compile(fp);
1874 bpf_prog_lock_ro(fp);
1876 /* The tail call compatibility check can only be done at
1877 * this late stage as we need to determine, if we deal
1878 * with JITed or non JITed program concatenations and not
1879 * all eBPF JITs might immediately support all features.
1881 *err = bpf_check_tail_call(fp);
1885 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1887 static unsigned int __bpf_prog_ret1(const void *ctx,
1888 const struct bpf_insn *insn)
1893 static struct bpf_prog_dummy {
1894 struct bpf_prog prog;
1895 } dummy_bpf_prog = {
1897 .bpf_func = __bpf_prog_ret1,
1901 /* to avoid allocating empty bpf_prog_array for cgroups that
1902 * don't have bpf program attached use one global 'empty_prog_array'
1903 * It will not be modified the caller of bpf_prog_array_alloc()
1904 * (since caller requested prog_cnt == 0)
1905 * that pointer should be 'freed' by bpf_prog_array_free()
1908 struct bpf_prog_array hdr;
1909 struct bpf_prog *null_prog;
1910 } empty_prog_array = {
1914 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1917 return kzalloc(sizeof(struct bpf_prog_array) +
1918 sizeof(struct bpf_prog_array_item) *
1922 return &empty_prog_array.hdr;
1925 void bpf_prog_array_free(struct bpf_prog_array *progs)
1927 if (!progs || progs == &empty_prog_array.hdr)
1929 kfree_rcu(progs, rcu);
1932 int bpf_prog_array_length(struct bpf_prog_array *array)
1934 struct bpf_prog_array_item *item;
1937 for (item = array->items; item->prog; item++)
1938 if (item->prog != &dummy_bpf_prog.prog)
1943 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
1945 struct bpf_prog_array_item *item;
1947 for (item = array->items; item->prog; item++)
1948 if (item->prog != &dummy_bpf_prog.prog)
1953 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
1957 struct bpf_prog_array_item *item;
1960 for (item = array->items; item->prog; item++) {
1961 if (item->prog == &dummy_bpf_prog.prog)
1963 prog_ids[i] = item->prog->aux->id;
1964 if (++i == request_cnt) {
1970 return !!(item->prog);
1973 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
1974 __u32 __user *prog_ids, u32 cnt)
1976 unsigned long err = 0;
1980 /* users of this function are doing:
1981 * cnt = bpf_prog_array_length();
1983 * bpf_prog_array_copy_to_user(..., cnt);
1984 * so below kcalloc doesn't need extra cnt > 0 check.
1986 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1989 nospc = bpf_prog_array_copy_core(array, ids, cnt);
1990 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1999 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
2000 struct bpf_prog *old_prog)
2002 struct bpf_prog_array_item *item;
2004 for (item = array->items; item->prog; item++)
2005 if (item->prog == old_prog) {
2006 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
2012 * bpf_prog_array_delete_safe_at() - Replaces the program at the given
2013 * index into the program array with
2014 * a dummy no-op program.
2015 * @array: a bpf_prog_array
2016 * @index: the index of the program to replace
2018 * Skips over dummy programs, by not counting them, when calculating
2019 * the position of the program to replace.
2023 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2024 * * -ENOENT - Index out of range
2026 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
2028 return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
2032 * bpf_prog_array_update_at() - Updates the program at the given index
2033 * into the program array.
2034 * @array: a bpf_prog_array
2035 * @index: the index of the program to update
2036 * @prog: the program to insert into the array
2038 * Skips over dummy programs, by not counting them, when calculating
2039 * the position of the program to update.
2043 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2044 * * -ENOENT - Index out of range
2046 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2047 struct bpf_prog *prog)
2049 struct bpf_prog_array_item *item;
2051 if (unlikely(index < 0))
2054 for (item = array->items; item->prog; item++) {
2055 if (item->prog == &dummy_bpf_prog.prog)
2058 WRITE_ONCE(item->prog, prog);
2066 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2067 struct bpf_prog *exclude_prog,
2068 struct bpf_prog *include_prog,
2069 struct bpf_prog_array **new_array)
2071 int new_prog_cnt, carry_prog_cnt = 0;
2072 struct bpf_prog_array_item *existing;
2073 struct bpf_prog_array *array;
2074 bool found_exclude = false;
2075 int new_prog_idx = 0;
2077 /* Figure out how many existing progs we need to carry over to
2081 existing = old_array->items;
2082 for (; existing->prog; existing++) {
2083 if (existing->prog == exclude_prog) {
2084 found_exclude = true;
2087 if (existing->prog != &dummy_bpf_prog.prog)
2089 if (existing->prog == include_prog)
2094 if (exclude_prog && !found_exclude)
2097 /* How many progs (not NULL) will be in the new array? */
2098 new_prog_cnt = carry_prog_cnt;
2102 /* Do we have any prog (not NULL) in the new array? */
2103 if (!new_prog_cnt) {
2108 /* +1 as the end of prog_array is marked with NULL */
2109 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2113 /* Fill in the new prog array */
2114 if (carry_prog_cnt) {
2115 existing = old_array->items;
2116 for (; existing->prog; existing++)
2117 if (existing->prog != exclude_prog &&
2118 existing->prog != &dummy_bpf_prog.prog) {
2119 array->items[new_prog_idx++].prog =
2124 array->items[new_prog_idx++].prog = include_prog;
2125 array->items[new_prog_idx].prog = NULL;
2130 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2131 u32 *prog_ids, u32 request_cnt,
2137 cnt = bpf_prog_array_length(array);
2141 /* return early if user requested only program count or nothing to copy */
2142 if (!request_cnt || !cnt)
2145 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
2146 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
2150 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2151 struct bpf_map **used_maps, u32 len)
2153 struct bpf_map *map;
2156 for (i = 0; i < len; i++) {
2158 if (map->ops->map_poke_untrack)
2159 map->ops->map_poke_untrack(map, aux);
2164 static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2166 __bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
2167 kfree(aux->used_maps);
2170 void __bpf_free_used_btfs(struct bpf_prog_aux *aux,
2171 struct btf_mod_pair *used_btfs, u32 len)
2173 #ifdef CONFIG_BPF_SYSCALL
2174 struct btf_mod_pair *btf_mod;
2177 for (i = 0; i < len; i++) {
2178 btf_mod = &used_btfs[i];
2179 if (btf_mod->module)
2180 module_put(btf_mod->module);
2181 btf_put(btf_mod->btf);
2186 static void bpf_free_used_btfs(struct bpf_prog_aux *aux)
2188 __bpf_free_used_btfs(aux, aux->used_btfs, aux->used_btf_cnt);
2189 kfree(aux->used_btfs);
2192 static void bpf_prog_free_deferred(struct work_struct *work)
2194 struct bpf_prog_aux *aux;
2197 aux = container_of(work, struct bpf_prog_aux, work);
2198 bpf_free_used_maps(aux);
2199 bpf_free_used_btfs(aux);
2200 if (bpf_prog_is_dev_bound(aux))
2201 bpf_prog_offload_destroy(aux->prog);
2202 #ifdef CONFIG_PERF_EVENTS
2203 if (aux->prog->has_callchain_buf)
2204 put_callchain_buffers();
2206 if (aux->dst_trampoline)
2207 bpf_trampoline_put(aux->dst_trampoline);
2208 for (i = 0; i < aux->func_cnt; i++)
2209 bpf_jit_free(aux->func[i]);
2210 if (aux->func_cnt) {
2212 bpf_prog_unlock_free(aux->prog);
2214 bpf_jit_free(aux->prog);
2218 /* Free internal BPF program */
2219 void bpf_prog_free(struct bpf_prog *fp)
2221 struct bpf_prog_aux *aux = fp->aux;
2224 bpf_prog_put(aux->dst_prog);
2225 INIT_WORK(&aux->work, bpf_prog_free_deferred);
2226 schedule_work(&aux->work);
2228 EXPORT_SYMBOL_GPL(bpf_prog_free);
2230 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2231 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2233 void bpf_user_rnd_init_once(void)
2235 prandom_init_once(&bpf_user_rnd_state);
2238 BPF_CALL_0(bpf_user_rnd_u32)
2240 /* Should someone ever have the rather unwise idea to use some
2241 * of the registers passed into this function, then note that
2242 * this function is called from native eBPF and classic-to-eBPF
2243 * transformations. Register assignments from both sides are
2244 * different, f.e. classic always sets fn(ctx, A, X) here.
2246 struct rnd_state *state;
2249 state = &get_cpu_var(bpf_user_rnd_state);
2250 res = prandom_u32_state(state);
2251 put_cpu_var(bpf_user_rnd_state);
2256 BPF_CALL_0(bpf_get_raw_cpu_id)
2258 return raw_smp_processor_id();
2261 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2262 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2263 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2264 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2265 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2266 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2267 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2268 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2269 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2270 const struct bpf_func_proto bpf_jiffies64_proto __weak;
2272 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2273 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2274 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2275 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2276 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
2277 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto __weak;
2279 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2280 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2281 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2282 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2283 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
2284 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2285 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
2286 const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
2287 const struct bpf_func_proto bpf_seq_printf_btf_proto __weak;
2289 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2295 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2296 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2300 EXPORT_SYMBOL_GPL(bpf_event_output);
2302 /* Always built-in helper functions. */
2303 const struct bpf_func_proto bpf_tail_call_proto = {
2306 .ret_type = RET_VOID,
2307 .arg1_type = ARG_PTR_TO_CTX,
2308 .arg2_type = ARG_CONST_MAP_PTR,
2309 .arg3_type = ARG_ANYTHING,
2312 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2313 * It is encouraged to implement bpf_int_jit_compile() instead, so that
2314 * eBPF and implicitly also cBPF can get JITed!
2316 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2321 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2322 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2324 void __weak bpf_jit_compile(struct bpf_prog *prog)
2328 bool __weak bpf_helper_changes_pkt_data(void *func)
2333 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2334 * analysis code and wants explicit zero extension inserted by verifier.
2335 * Otherwise, return FALSE.
2337 * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
2338 * you don't override this. JITs that don't want these extra insns can detect
2339 * them using insn_is_zext.
2341 bool __weak bpf_jit_needs_zext(void)
2346 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2347 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2349 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2355 int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
2356 void *addr1, void *addr2)
2361 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2362 EXPORT_SYMBOL(bpf_stats_enabled_key);
2364 /* All definitions of tracepoints related to BPF. */
2365 #define CREATE_TRACE_POINTS
2366 #include <linux/bpf_trace.h>
2368 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2369 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);