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 <linux/bpf_verifier.h>
37 #include <asm/barrier.h>
38 #include <asm/unaligned.h>
41 #define BPF_R0 regs[BPF_REG_0]
42 #define BPF_R1 regs[BPF_REG_1]
43 #define BPF_R2 regs[BPF_REG_2]
44 #define BPF_R3 regs[BPF_REG_3]
45 #define BPF_R4 regs[BPF_REG_4]
46 #define BPF_R5 regs[BPF_REG_5]
47 #define BPF_R6 regs[BPF_REG_6]
48 #define BPF_R7 regs[BPF_REG_7]
49 #define BPF_R8 regs[BPF_REG_8]
50 #define BPF_R9 regs[BPF_REG_9]
51 #define BPF_R10 regs[BPF_REG_10]
54 #define DST regs[insn->dst_reg]
55 #define SRC regs[insn->src_reg]
56 #define FP regs[BPF_REG_FP]
57 #define AX regs[BPF_REG_AX]
58 #define ARG1 regs[BPF_REG_ARG1]
59 #define CTX regs[BPF_REG_CTX]
62 /* No hurry in this branch
64 * Exported for the bpf jit load helper.
66 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
71 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
72 else if (k >= SKF_LL_OFF)
73 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
75 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
81 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
83 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
84 struct bpf_prog_aux *aux;
87 size = round_up(size, PAGE_SIZE);
88 fp = __vmalloc(size, gfp_flags);
92 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT | gfp_extra_flags);
97 fp->active = alloc_percpu_gfp(int, GFP_KERNEL_ACCOUNT | gfp_extra_flags);
104 fp->pages = size / PAGE_SIZE;
107 fp->jit_requested = ebpf_jit_enabled();
109 INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
110 mutex_init(&fp->aux->used_maps_mutex);
111 mutex_init(&fp->aux->dst_mutex);
116 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
118 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
119 struct bpf_prog *prog;
122 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
126 prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
128 free_percpu(prog->active);
134 for_each_possible_cpu(cpu) {
135 struct bpf_prog_stats *pstats;
137 pstats = per_cpu_ptr(prog->stats, cpu);
138 u64_stats_init(&pstats->syncp);
142 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
144 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
146 if (!prog->aux->nr_linfo || !prog->jit_requested)
149 prog->aux->jited_linfo = kvcalloc(prog->aux->nr_linfo,
150 sizeof(*prog->aux->jited_linfo),
151 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
152 if (!prog->aux->jited_linfo)
158 void bpf_prog_jit_attempt_done(struct bpf_prog *prog)
160 if (prog->aux->jited_linfo &&
161 (!prog->jited || !prog->aux->jited_linfo[0])) {
162 kvfree(prog->aux->jited_linfo);
163 prog->aux->jited_linfo = NULL;
166 kfree(prog->aux->kfunc_tab);
167 prog->aux->kfunc_tab = NULL;
170 /* The jit engine is responsible to provide an array
171 * for insn_off to the jited_off mapping (insn_to_jit_off).
173 * The idx to this array is the insn_off. Hence, the insn_off
174 * here is relative to the prog itself instead of the main prog.
175 * This array has one entry for each xlated bpf insn.
177 * jited_off is the byte off to the last byte of the jited insn.
181 * The first bpf insn off of the prog. The insn off
182 * here is relative to the main prog.
183 * e.g. if prog is a subprog, insn_start > 0
185 * The prog's idx to prog->aux->linfo and jited_linfo
187 * jited_linfo[linfo_idx] = prog->bpf_func
191 * jited_linfo[i] = prog->bpf_func +
192 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
194 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
195 const u32 *insn_to_jit_off)
197 u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
198 const struct bpf_line_info *linfo;
201 if (!prog->aux->jited_linfo)
202 /* Userspace did not provide linfo */
205 linfo_idx = prog->aux->linfo_idx;
206 linfo = &prog->aux->linfo[linfo_idx];
207 insn_start = linfo[0].insn_off;
208 insn_end = insn_start + prog->len;
210 jited_linfo = &prog->aux->jited_linfo[linfo_idx];
211 jited_linfo[0] = prog->bpf_func;
213 nr_linfo = prog->aux->nr_linfo - linfo_idx;
215 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
216 /* The verifier ensures that linfo[i].insn_off is
217 * strictly increasing
219 jited_linfo[i] = prog->bpf_func +
220 insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
223 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
224 gfp_t gfp_extra_flags)
226 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
230 size = round_up(size, PAGE_SIZE);
231 pages = size / PAGE_SIZE;
232 if (pages <= fp_old->pages)
235 fp = __vmalloc(size, gfp_flags);
237 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
241 /* We keep fp->aux from fp_old around in the new
242 * reallocated structure.
245 fp_old->stats = NULL;
246 fp_old->active = NULL;
247 __bpf_prog_free(fp_old);
253 void __bpf_prog_free(struct bpf_prog *fp)
256 mutex_destroy(&fp->aux->used_maps_mutex);
257 mutex_destroy(&fp->aux->dst_mutex);
258 kfree(fp->aux->poke_tab);
261 free_percpu(fp->stats);
262 free_percpu(fp->active);
266 int bpf_prog_calc_tag(struct bpf_prog *fp)
268 const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
269 u32 raw_size = bpf_prog_tag_scratch_size(fp);
270 u32 digest[SHA1_DIGEST_WORDS];
271 u32 ws[SHA1_WORKSPACE_WORDS];
272 u32 i, bsize, psize, blocks;
273 struct bpf_insn *dst;
279 raw = vmalloc(raw_size);
284 memset(ws, 0, sizeof(ws));
286 /* We need to take out the map fd for the digest calculation
287 * since they are unstable from user space side.
290 for (i = 0, was_ld_map = false; i < fp->len; i++) {
291 dst[i] = fp->insnsi[i];
293 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
294 (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
295 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
298 } else if (was_ld_map &&
300 dst[i].dst_reg == 0 &&
301 dst[i].src_reg == 0 &&
310 psize = bpf_prog_insn_size(fp);
311 memset(&raw[psize], 0, raw_size - psize);
314 bsize = round_up(psize, SHA1_BLOCK_SIZE);
315 blocks = bsize / SHA1_BLOCK_SIZE;
317 if (bsize - psize >= sizeof(__be64)) {
318 bits = (__be64 *)(todo + bsize - sizeof(__be64));
320 bits = (__be64 *)(todo + bsize + bits_offset);
323 *bits = cpu_to_be64((psize - 1) << 3);
326 sha1_transform(digest, todo, ws);
327 todo += SHA1_BLOCK_SIZE;
330 result = (__force __be32 *)digest;
331 for (i = 0; i < SHA1_DIGEST_WORDS; i++)
332 result[i] = cpu_to_be32(digest[i]);
333 memcpy(fp->tag, result, sizeof(fp->tag));
339 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
340 s32 end_new, s32 curr, const bool probe_pass)
342 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
343 s32 delta = end_new - end_old;
346 if (curr < pos && curr + imm + 1 >= end_old)
348 else if (curr >= end_new && curr + imm + 1 < end_new)
350 if (imm < imm_min || imm > imm_max)
357 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
358 s32 end_new, s32 curr, const bool probe_pass)
360 const s32 off_min = S16_MIN, off_max = S16_MAX;
361 s32 delta = end_new - end_old;
364 if (curr < pos && curr + off + 1 >= end_old)
366 else if (curr >= end_new && curr + off + 1 < end_new)
368 if (off < off_min || off > off_max)
375 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
376 s32 end_new, const bool probe_pass)
378 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
379 struct bpf_insn *insn = prog->insnsi;
382 for (i = 0; i < insn_cnt; i++, insn++) {
385 /* In the probing pass we still operate on the original,
386 * unpatched image in order to check overflows before we
387 * do any other adjustments. Therefore skip the patchlet.
389 if (probe_pass && i == pos) {
391 insn = prog->insnsi + end_old;
393 if (bpf_pseudo_func(insn)) {
394 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
395 end_new, i, probe_pass);
401 if ((BPF_CLASS(code) != BPF_JMP &&
402 BPF_CLASS(code) != BPF_JMP32) ||
403 BPF_OP(code) == BPF_EXIT)
405 /* Adjust offset of jmps if we cross patch boundaries. */
406 if (BPF_OP(code) == BPF_CALL) {
407 if (insn->src_reg != BPF_PSEUDO_CALL)
409 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
410 end_new, i, probe_pass);
412 ret = bpf_adj_delta_to_off(insn, pos, end_old,
413 end_new, i, probe_pass);
422 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
424 struct bpf_line_info *linfo;
427 nr_linfo = prog->aux->nr_linfo;
428 if (!nr_linfo || !delta)
431 linfo = prog->aux->linfo;
433 for (i = 0; i < nr_linfo; i++)
434 if (off < linfo[i].insn_off)
437 /* Push all off < linfo[i].insn_off by delta */
438 for (; i < nr_linfo; i++)
439 linfo[i].insn_off += delta;
442 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
443 const struct bpf_insn *patch, u32 len)
445 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
446 const u32 cnt_max = S16_MAX;
447 struct bpf_prog *prog_adj;
450 /* Since our patchlet doesn't expand the image, we're done. */
451 if (insn_delta == 0) {
452 memcpy(prog->insnsi + off, patch, sizeof(*patch));
456 insn_adj_cnt = prog->len + insn_delta;
458 /* Reject anything that would potentially let the insn->off
459 * target overflow when we have excessive program expansions.
460 * We need to probe here before we do any reallocation where
461 * we afterwards may not fail anymore.
463 if (insn_adj_cnt > cnt_max &&
464 (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
467 /* Several new instructions need to be inserted. Make room
468 * for them. Likely, there's no need for a new allocation as
469 * last page could have large enough tailroom.
471 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
474 return ERR_PTR(-ENOMEM);
476 prog_adj->len = insn_adj_cnt;
478 /* Patching happens in 3 steps:
480 * 1) Move over tail of insnsi from next instruction onwards,
481 * so we can patch the single target insn with one or more
482 * new ones (patching is always from 1 to n insns, n > 0).
483 * 2) Inject new instructions at the target location.
484 * 3) Adjust branch offsets if necessary.
486 insn_rest = insn_adj_cnt - off - len;
488 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
489 sizeof(*patch) * insn_rest);
490 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
492 /* We are guaranteed to not fail at this point, otherwise
493 * the ship has sailed to reverse to the original state. An
494 * overflow cannot happen at this point.
496 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
498 bpf_adj_linfo(prog_adj, off, insn_delta);
503 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
505 /* Branch offsets can't overflow when program is shrinking, no need
506 * to call bpf_adj_branches(..., true) here
508 memmove(prog->insnsi + off, prog->insnsi + off + cnt,
509 sizeof(struct bpf_insn) * (prog->len - off - cnt));
512 return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
515 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
519 for (i = 0; i < fp->aux->func_cnt; i++)
520 bpf_prog_kallsyms_del(fp->aux->func[i]);
523 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
525 bpf_prog_kallsyms_del_subprogs(fp);
526 bpf_prog_kallsyms_del(fp);
529 #ifdef CONFIG_BPF_JIT
530 /* All BPF JIT sysctl knobs here. */
531 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
532 int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
533 int bpf_jit_harden __read_mostly;
534 long bpf_jit_limit __read_mostly;
535 long bpf_jit_limit_max __read_mostly;
538 bpf_prog_ksym_set_addr(struct bpf_prog *prog)
540 const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
541 unsigned long addr = (unsigned long)hdr;
543 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
545 prog->aux->ksym.start = (unsigned long) prog->bpf_func;
546 prog->aux->ksym.end = addr + hdr->pages * PAGE_SIZE;
550 bpf_prog_ksym_set_name(struct bpf_prog *prog)
552 char *sym = prog->aux->ksym.name;
553 const char *end = sym + KSYM_NAME_LEN;
554 const struct btf_type *type;
555 const char *func_name;
557 BUILD_BUG_ON(sizeof("bpf_prog_") +
558 sizeof(prog->tag) * 2 +
559 /* name has been null terminated.
560 * We should need +1 for the '_' preceding
561 * the name. However, the null character
562 * is double counted between the name and the
563 * sizeof("bpf_prog_") above, so we omit
566 sizeof(prog->aux->name) > KSYM_NAME_LEN);
568 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
569 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
571 /* prog->aux->name will be ignored if full btf name is available */
572 if (prog->aux->func_info_cnt) {
573 type = btf_type_by_id(prog->aux->btf,
574 prog->aux->func_info[prog->aux->func_idx].type_id);
575 func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
576 snprintf(sym, (size_t)(end - sym), "_%s", func_name);
580 if (prog->aux->name[0])
581 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
586 static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
588 return container_of(n, struct bpf_ksym, tnode)->start;
591 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
592 struct latch_tree_node *b)
594 return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
597 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
599 unsigned long val = (unsigned long)key;
600 const struct bpf_ksym *ksym;
602 ksym = container_of(n, struct bpf_ksym, tnode);
604 if (val < ksym->start)
606 if (val >= ksym->end)
612 static const struct latch_tree_ops bpf_tree_ops = {
613 .less = bpf_tree_less,
614 .comp = bpf_tree_comp,
617 static DEFINE_SPINLOCK(bpf_lock);
618 static LIST_HEAD(bpf_kallsyms);
619 static struct latch_tree_root bpf_tree __cacheline_aligned;
621 void bpf_ksym_add(struct bpf_ksym *ksym)
623 spin_lock_bh(&bpf_lock);
624 WARN_ON_ONCE(!list_empty(&ksym->lnode));
625 list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
626 latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
627 spin_unlock_bh(&bpf_lock);
630 static void __bpf_ksym_del(struct bpf_ksym *ksym)
632 if (list_empty(&ksym->lnode))
635 latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
636 list_del_rcu(&ksym->lnode);
639 void bpf_ksym_del(struct bpf_ksym *ksym)
641 spin_lock_bh(&bpf_lock);
642 __bpf_ksym_del(ksym);
643 spin_unlock_bh(&bpf_lock);
646 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
648 return fp->jited && !bpf_prog_was_classic(fp);
651 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
653 return list_empty(&fp->aux->ksym.lnode) ||
654 fp->aux->ksym.lnode.prev == LIST_POISON2;
657 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
659 if (!bpf_prog_kallsyms_candidate(fp) ||
663 bpf_prog_ksym_set_addr(fp);
664 bpf_prog_ksym_set_name(fp);
665 fp->aux->ksym.prog = true;
667 bpf_ksym_add(&fp->aux->ksym);
670 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
672 if (!bpf_prog_kallsyms_candidate(fp))
675 bpf_ksym_del(&fp->aux->ksym);
678 static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
680 struct latch_tree_node *n;
682 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
683 return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
686 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
687 unsigned long *off, char *sym)
689 struct bpf_ksym *ksym;
693 ksym = bpf_ksym_find(addr);
695 unsigned long symbol_start = ksym->start;
696 unsigned long symbol_end = ksym->end;
698 strncpy(sym, ksym->name, KSYM_NAME_LEN);
702 *size = symbol_end - symbol_start;
704 *off = addr - symbol_start;
711 bool is_bpf_text_address(unsigned long addr)
716 ret = bpf_ksym_find(addr) != NULL;
722 static struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
724 struct bpf_ksym *ksym = bpf_ksym_find(addr);
726 return ksym && ksym->prog ?
727 container_of(ksym, struct bpf_prog_aux, ksym)->prog :
731 const struct exception_table_entry *search_bpf_extables(unsigned long addr)
733 const struct exception_table_entry *e = NULL;
734 struct bpf_prog *prog;
737 prog = bpf_prog_ksym_find(addr);
740 if (!prog->aux->num_exentries)
743 e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
749 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
752 struct bpf_ksym *ksym;
756 if (!bpf_jit_kallsyms_enabled())
760 list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
764 strncpy(sym, ksym->name, KSYM_NAME_LEN);
766 *value = ksym->start;
767 *type = BPF_SYM_ELF_TYPE;
777 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
778 struct bpf_jit_poke_descriptor *poke)
780 struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
781 static const u32 poke_tab_max = 1024;
782 u32 slot = prog->aux->size_poke_tab;
785 if (size > poke_tab_max)
787 if (poke->tailcall_target || poke->tailcall_target_stable ||
788 poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
791 switch (poke->reason) {
792 case BPF_POKE_REASON_TAIL_CALL:
793 if (!poke->tail_call.map)
800 tab = krealloc(tab, size * sizeof(*poke), GFP_KERNEL);
804 memcpy(&tab[slot], poke, sizeof(*poke));
805 prog->aux->size_poke_tab = size;
806 prog->aux->poke_tab = tab;
811 static atomic_long_t bpf_jit_current;
813 /* Can be overridden by an arch's JIT compiler if it has a custom,
814 * dedicated BPF backend memory area, or if neither of the two
817 u64 __weak bpf_jit_alloc_exec_limit(void)
819 #if defined(MODULES_VADDR)
820 return MODULES_END - MODULES_VADDR;
822 return VMALLOC_END - VMALLOC_START;
826 static int __init bpf_jit_charge_init(void)
828 /* Only used as heuristic here to derive limit. */
829 bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
830 bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
831 PAGE_SIZE), LONG_MAX);
834 pure_initcall(bpf_jit_charge_init);
836 int bpf_jit_charge_modmem(u32 pages)
838 if (atomic_long_add_return(pages, &bpf_jit_current) >
839 (bpf_jit_limit >> PAGE_SHIFT)) {
840 if (!bpf_capable()) {
841 atomic_long_sub(pages, &bpf_jit_current);
849 void bpf_jit_uncharge_modmem(u32 pages)
851 atomic_long_sub(pages, &bpf_jit_current);
854 void *__weak bpf_jit_alloc_exec(unsigned long size)
856 return module_alloc(size);
859 void __weak bpf_jit_free_exec(void *addr)
861 module_memfree(addr);
864 struct bpf_binary_header *
865 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
866 unsigned int alignment,
867 bpf_jit_fill_hole_t bpf_fill_ill_insns)
869 struct bpf_binary_header *hdr;
870 u32 size, hole, start, pages;
872 WARN_ON_ONCE(!is_power_of_2(alignment) ||
873 alignment > BPF_IMAGE_ALIGNMENT);
875 /* Most of BPF filters are really small, but if some of them
876 * fill a page, allow at least 128 extra bytes to insert a
877 * random section of illegal instructions.
879 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
880 pages = size / PAGE_SIZE;
882 if (bpf_jit_charge_modmem(pages))
884 hdr = bpf_jit_alloc_exec(size);
886 bpf_jit_uncharge_modmem(pages);
890 /* Fill space with illegal/arch-dep instructions. */
891 bpf_fill_ill_insns(hdr, size);
894 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
895 PAGE_SIZE - sizeof(*hdr));
896 start = (get_random_int() % hole) & ~(alignment - 1);
898 /* Leave a random number of instructions before BPF code. */
899 *image_ptr = &hdr->image[start];
904 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
906 u32 pages = hdr->pages;
908 bpf_jit_free_exec(hdr);
909 bpf_jit_uncharge_modmem(pages);
912 /* This symbol is only overridden by archs that have different
913 * requirements than the usual eBPF JITs, f.e. when they only
914 * implement cBPF JIT, do not set images read-only, etc.
916 void __weak bpf_jit_free(struct bpf_prog *fp)
919 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
921 bpf_jit_binary_free(hdr);
923 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
926 bpf_prog_unlock_free(fp);
929 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
930 const struct bpf_insn *insn, bool extra_pass,
931 u64 *func_addr, bool *func_addr_fixed)
937 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
938 if (!*func_addr_fixed) {
939 /* Place-holder address till the last pass has collected
940 * all addresses for JITed subprograms in which case we
941 * can pick them up from prog->aux.
945 else if (prog->aux->func &&
946 off >= 0 && off < prog->aux->func_cnt)
947 addr = (u8 *)prog->aux->func[off]->bpf_func;
951 /* Address of a BPF helper call. Since part of the core
952 * kernel, it's always at a fixed location. __bpf_call_base
953 * and the helper with imm relative to it are both in core
956 addr = (u8 *)__bpf_call_base + imm;
959 *func_addr = (unsigned long)addr;
963 static int bpf_jit_blind_insn(const struct bpf_insn *from,
964 const struct bpf_insn *aux,
965 struct bpf_insn *to_buff,
968 struct bpf_insn *to = to_buff;
969 u32 imm_rnd = get_random_int();
972 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
973 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
975 /* Constraints on AX register:
977 * AX register is inaccessible from user space. It is mapped in
978 * all JITs, and used here for constant blinding rewrites. It is
979 * typically "stateless" meaning its contents are only valid within
980 * the executed instruction, but not across several instructions.
981 * There are a few exceptions however which are further detailed
984 * Constant blinding is only used by JITs, not in the interpreter.
985 * The interpreter uses AX in some occasions as a local temporary
986 * register e.g. in DIV or MOD instructions.
988 * In restricted circumstances, the verifier can also use the AX
989 * register for rewrites as long as they do not interfere with
992 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
995 if (from->imm == 0 &&
996 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
997 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
998 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
1002 switch (from->code) {
1003 case BPF_ALU | BPF_ADD | BPF_K:
1004 case BPF_ALU | BPF_SUB | BPF_K:
1005 case BPF_ALU | BPF_AND | BPF_K:
1006 case BPF_ALU | BPF_OR | BPF_K:
1007 case BPF_ALU | BPF_XOR | BPF_K:
1008 case BPF_ALU | BPF_MUL | BPF_K:
1009 case BPF_ALU | BPF_MOV | BPF_K:
1010 case BPF_ALU | BPF_DIV | BPF_K:
1011 case BPF_ALU | BPF_MOD | BPF_K:
1012 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1013 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1014 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
1017 case BPF_ALU64 | BPF_ADD | BPF_K:
1018 case BPF_ALU64 | BPF_SUB | BPF_K:
1019 case BPF_ALU64 | BPF_AND | BPF_K:
1020 case BPF_ALU64 | BPF_OR | BPF_K:
1021 case BPF_ALU64 | BPF_XOR | BPF_K:
1022 case BPF_ALU64 | BPF_MUL | BPF_K:
1023 case BPF_ALU64 | BPF_MOV | BPF_K:
1024 case BPF_ALU64 | BPF_DIV | BPF_K:
1025 case BPF_ALU64 | BPF_MOD | BPF_K:
1026 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1027 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1028 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
1031 case BPF_JMP | BPF_JEQ | BPF_K:
1032 case BPF_JMP | BPF_JNE | BPF_K:
1033 case BPF_JMP | BPF_JGT | BPF_K:
1034 case BPF_JMP | BPF_JLT | BPF_K:
1035 case BPF_JMP | BPF_JGE | BPF_K:
1036 case BPF_JMP | BPF_JLE | BPF_K:
1037 case BPF_JMP | BPF_JSGT | BPF_K:
1038 case BPF_JMP | BPF_JSLT | BPF_K:
1039 case BPF_JMP | BPF_JSGE | BPF_K:
1040 case BPF_JMP | BPF_JSLE | BPF_K:
1041 case BPF_JMP | BPF_JSET | BPF_K:
1042 /* Accommodate for extra offset in case of a backjump. */
1046 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1047 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1048 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1051 case BPF_JMP32 | BPF_JEQ | BPF_K:
1052 case BPF_JMP32 | BPF_JNE | BPF_K:
1053 case BPF_JMP32 | BPF_JGT | BPF_K:
1054 case BPF_JMP32 | BPF_JLT | BPF_K:
1055 case BPF_JMP32 | BPF_JGE | BPF_K:
1056 case BPF_JMP32 | BPF_JLE | BPF_K:
1057 case BPF_JMP32 | BPF_JSGT | BPF_K:
1058 case BPF_JMP32 | BPF_JSLT | BPF_K:
1059 case BPF_JMP32 | BPF_JSGE | BPF_K:
1060 case BPF_JMP32 | BPF_JSLE | BPF_K:
1061 case BPF_JMP32 | BPF_JSET | BPF_K:
1062 /* Accommodate for extra offset in case of a backjump. */
1066 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1067 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1068 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1072 case BPF_LD | BPF_IMM | BPF_DW:
1073 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1074 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1075 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1076 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1078 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1079 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1080 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1082 *to++ = BPF_ZEXT_REG(BPF_REG_AX);
1083 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
1086 case BPF_ST | BPF_MEM | BPF_DW:
1087 case BPF_ST | BPF_MEM | BPF_W:
1088 case BPF_ST | BPF_MEM | BPF_H:
1089 case BPF_ST | BPF_MEM | BPF_B:
1090 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1091 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1092 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1096 return to - to_buff;
1099 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1100 gfp_t gfp_extra_flags)
1102 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1103 struct bpf_prog *fp;
1105 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
1107 /* aux->prog still points to the fp_other one, so
1108 * when promoting the clone to the real program,
1109 * this still needs to be adapted.
1111 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1117 static void bpf_prog_clone_free(struct bpf_prog *fp)
1119 /* aux was stolen by the other clone, so we cannot free
1120 * it from this path! It will be freed eventually by the
1121 * other program on release.
1123 * At this point, we don't need a deferred release since
1124 * clone is guaranteed to not be locked.
1129 __bpf_prog_free(fp);
1132 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1134 /* We have to repoint aux->prog to self, as we don't
1135 * know whether fp here is the clone or the original.
1138 bpf_prog_clone_free(fp_other);
1141 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1143 struct bpf_insn insn_buff[16], aux[2];
1144 struct bpf_prog *clone, *tmp;
1145 int insn_delta, insn_cnt;
1146 struct bpf_insn *insn;
1149 if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1152 clone = bpf_prog_clone_create(prog, GFP_USER);
1154 return ERR_PTR(-ENOMEM);
1156 insn_cnt = clone->len;
1157 insn = clone->insnsi;
1159 for (i = 0; i < insn_cnt; i++, insn++) {
1160 /* We temporarily need to hold the original ld64 insn
1161 * so that we can still access the first part in the
1162 * second blinding run.
1164 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1166 memcpy(aux, insn, sizeof(aux));
1168 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1169 clone->aux->verifier_zext);
1173 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1175 /* Patching may have repointed aux->prog during
1176 * realloc from the original one, so we need to
1177 * fix it up here on error.
1179 bpf_jit_prog_release_other(prog, clone);
1184 insn_delta = rewritten - 1;
1186 /* Walk new program and skip insns we just inserted. */
1187 insn = clone->insnsi + i + insn_delta;
1188 insn_cnt += insn_delta;
1195 #endif /* CONFIG_BPF_JIT */
1197 /* Base function for offset calculation. Needs to go into .text section,
1198 * therefore keeping it non-static as well; will also be used by JITs
1199 * anyway later on, so do not let the compiler omit it. This also needs
1200 * to go into kallsyms for correlation from e.g. bpftool, so naming
1203 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1207 EXPORT_SYMBOL_GPL(__bpf_call_base);
1209 /* All UAPI available opcodes. */
1210 #define BPF_INSN_MAP(INSN_2, INSN_3) \
1211 /* 32 bit ALU operations. */ \
1212 /* Register based. */ \
1213 INSN_3(ALU, ADD, X), \
1214 INSN_3(ALU, SUB, X), \
1215 INSN_3(ALU, AND, X), \
1216 INSN_3(ALU, OR, X), \
1217 INSN_3(ALU, LSH, X), \
1218 INSN_3(ALU, RSH, X), \
1219 INSN_3(ALU, XOR, X), \
1220 INSN_3(ALU, MUL, X), \
1221 INSN_3(ALU, MOV, X), \
1222 INSN_3(ALU, ARSH, X), \
1223 INSN_3(ALU, DIV, X), \
1224 INSN_3(ALU, MOD, X), \
1226 INSN_3(ALU, END, TO_BE), \
1227 INSN_3(ALU, END, TO_LE), \
1228 /* Immediate based. */ \
1229 INSN_3(ALU, ADD, K), \
1230 INSN_3(ALU, SUB, K), \
1231 INSN_3(ALU, AND, K), \
1232 INSN_3(ALU, OR, K), \
1233 INSN_3(ALU, LSH, K), \
1234 INSN_3(ALU, RSH, K), \
1235 INSN_3(ALU, XOR, K), \
1236 INSN_3(ALU, MUL, K), \
1237 INSN_3(ALU, MOV, K), \
1238 INSN_3(ALU, ARSH, K), \
1239 INSN_3(ALU, DIV, K), \
1240 INSN_3(ALU, MOD, K), \
1241 /* 64 bit ALU operations. */ \
1242 /* Register based. */ \
1243 INSN_3(ALU64, ADD, X), \
1244 INSN_3(ALU64, SUB, X), \
1245 INSN_3(ALU64, AND, X), \
1246 INSN_3(ALU64, OR, X), \
1247 INSN_3(ALU64, LSH, X), \
1248 INSN_3(ALU64, RSH, X), \
1249 INSN_3(ALU64, XOR, X), \
1250 INSN_3(ALU64, MUL, X), \
1251 INSN_3(ALU64, MOV, X), \
1252 INSN_3(ALU64, ARSH, X), \
1253 INSN_3(ALU64, DIV, X), \
1254 INSN_3(ALU64, MOD, X), \
1255 INSN_2(ALU64, NEG), \
1256 /* Immediate based. */ \
1257 INSN_3(ALU64, ADD, K), \
1258 INSN_3(ALU64, SUB, K), \
1259 INSN_3(ALU64, AND, K), \
1260 INSN_3(ALU64, OR, K), \
1261 INSN_3(ALU64, LSH, K), \
1262 INSN_3(ALU64, RSH, K), \
1263 INSN_3(ALU64, XOR, K), \
1264 INSN_3(ALU64, MUL, K), \
1265 INSN_3(ALU64, MOV, K), \
1266 INSN_3(ALU64, ARSH, K), \
1267 INSN_3(ALU64, DIV, K), \
1268 INSN_3(ALU64, MOD, K), \
1269 /* Call instruction. */ \
1270 INSN_2(JMP, CALL), \
1271 /* Exit instruction. */ \
1272 INSN_2(JMP, EXIT), \
1273 /* 32-bit Jump instructions. */ \
1274 /* Register based. */ \
1275 INSN_3(JMP32, JEQ, X), \
1276 INSN_3(JMP32, JNE, X), \
1277 INSN_3(JMP32, JGT, X), \
1278 INSN_3(JMP32, JLT, X), \
1279 INSN_3(JMP32, JGE, X), \
1280 INSN_3(JMP32, JLE, X), \
1281 INSN_3(JMP32, JSGT, X), \
1282 INSN_3(JMP32, JSLT, X), \
1283 INSN_3(JMP32, JSGE, X), \
1284 INSN_3(JMP32, JSLE, X), \
1285 INSN_3(JMP32, JSET, X), \
1286 /* Immediate based. */ \
1287 INSN_3(JMP32, JEQ, K), \
1288 INSN_3(JMP32, JNE, K), \
1289 INSN_3(JMP32, JGT, K), \
1290 INSN_3(JMP32, JLT, K), \
1291 INSN_3(JMP32, JGE, K), \
1292 INSN_3(JMP32, JLE, K), \
1293 INSN_3(JMP32, JSGT, K), \
1294 INSN_3(JMP32, JSLT, K), \
1295 INSN_3(JMP32, JSGE, K), \
1296 INSN_3(JMP32, JSLE, K), \
1297 INSN_3(JMP32, JSET, K), \
1298 /* Jump instructions. */ \
1299 /* Register based. */ \
1300 INSN_3(JMP, JEQ, X), \
1301 INSN_3(JMP, JNE, X), \
1302 INSN_3(JMP, JGT, X), \
1303 INSN_3(JMP, JLT, X), \
1304 INSN_3(JMP, JGE, X), \
1305 INSN_3(JMP, JLE, X), \
1306 INSN_3(JMP, JSGT, X), \
1307 INSN_3(JMP, JSLT, X), \
1308 INSN_3(JMP, JSGE, X), \
1309 INSN_3(JMP, JSLE, X), \
1310 INSN_3(JMP, JSET, X), \
1311 /* Immediate based. */ \
1312 INSN_3(JMP, JEQ, K), \
1313 INSN_3(JMP, JNE, K), \
1314 INSN_3(JMP, JGT, K), \
1315 INSN_3(JMP, JLT, K), \
1316 INSN_3(JMP, JGE, K), \
1317 INSN_3(JMP, JLE, K), \
1318 INSN_3(JMP, JSGT, K), \
1319 INSN_3(JMP, JSLT, K), \
1320 INSN_3(JMP, JSGE, K), \
1321 INSN_3(JMP, JSLE, K), \
1322 INSN_3(JMP, JSET, K), \
1324 /* Store instructions. */ \
1325 /* Register based. */ \
1326 INSN_3(STX, MEM, B), \
1327 INSN_3(STX, MEM, H), \
1328 INSN_3(STX, MEM, W), \
1329 INSN_3(STX, MEM, DW), \
1330 INSN_3(STX, ATOMIC, W), \
1331 INSN_3(STX, ATOMIC, DW), \
1332 /* Immediate based. */ \
1333 INSN_3(ST, MEM, B), \
1334 INSN_3(ST, MEM, H), \
1335 INSN_3(ST, MEM, W), \
1336 INSN_3(ST, MEM, DW), \
1337 /* Load instructions. */ \
1338 /* Register based. */ \
1339 INSN_3(LDX, MEM, B), \
1340 INSN_3(LDX, MEM, H), \
1341 INSN_3(LDX, MEM, W), \
1342 INSN_3(LDX, MEM, DW), \
1343 /* Immediate based. */ \
1346 bool bpf_opcode_in_insntable(u8 code)
1348 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1349 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1350 static const bool public_insntable[256] = {
1351 [0 ... 255] = false,
1352 /* Now overwrite non-defaults ... */
1353 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1354 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1355 [BPF_LD | BPF_ABS | BPF_B] = true,
1356 [BPF_LD | BPF_ABS | BPF_H] = true,
1357 [BPF_LD | BPF_ABS | BPF_W] = true,
1358 [BPF_LD | BPF_IND | BPF_B] = true,
1359 [BPF_LD | BPF_IND | BPF_H] = true,
1360 [BPF_LD | BPF_IND | BPF_W] = true,
1362 #undef BPF_INSN_3_TBL
1363 #undef BPF_INSN_2_TBL
1364 return public_insntable[code];
1367 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1368 u64 __weak bpf_probe_read_kernel(void *dst, u32 size, const void *unsafe_ptr)
1370 memset(dst, 0, size);
1375 * ___bpf_prog_run - run eBPF program on a given context
1376 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1377 * @insn: is the array of eBPF instructions
1379 * Decode and execute eBPF instructions.
1381 * Return: whatever value is in %BPF_R0 at program exit
1383 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
1385 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1386 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1387 static const void * const jumptable[256] __annotate_jump_table = {
1388 [0 ... 255] = &&default_label,
1389 /* Now overwrite non-defaults ... */
1390 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1391 /* Non-UAPI available opcodes. */
1392 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1393 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1394 [BPF_ST | BPF_NOSPEC] = &&ST_NOSPEC,
1395 [BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1396 [BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1397 [BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1398 [BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1400 #undef BPF_INSN_3_LBL
1401 #undef BPF_INSN_2_LBL
1402 u32 tail_call_cnt = 0;
1404 #define CONT ({ insn++; goto select_insn; })
1405 #define CONT_JMP ({ insn++; goto select_insn; })
1408 goto *jumptable[insn->code];
1410 /* Explicitly mask the register-based shift amounts with 63 or 31
1411 * to avoid undefined behavior. Normally this won't affect the
1412 * generated code, for example, in case of native 64 bit archs such
1413 * as x86-64 or arm64, the compiler is optimizing the AND away for
1414 * the interpreter. In case of JITs, each of the JIT backends compiles
1415 * the BPF shift operations to machine instructions which produce
1416 * implementation-defined results in such a case; the resulting
1417 * contents of the register may be arbitrary, but program behaviour
1418 * as a whole remains defined. In other words, in case of JIT backends,
1419 * the AND must /not/ be added to the emitted LSH/RSH/ARSH translation.
1422 #define SHT(OPCODE, OP) \
1423 ALU64_##OPCODE##_X: \
1424 DST = DST OP (SRC & 63); \
1427 DST = (u32) DST OP ((u32) SRC & 31); \
1429 ALU64_##OPCODE##_K: \
1433 DST = (u32) DST OP (u32) IMM; \
1436 #define ALU(OPCODE, OP) \
1437 ALU64_##OPCODE##_X: \
1441 DST = (u32) DST OP (u32) SRC; \
1443 ALU64_##OPCODE##_K: \
1447 DST = (u32) DST OP (u32) IMM; \
1478 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1482 DST = (u64) (u32) (((s32) DST) >> (SRC & 31));
1485 DST = (u64) (u32) (((s32) DST) >> IMM);
1488 (*(s64 *) &DST) >>= (SRC & 63);
1491 (*(s64 *) &DST) >>= IMM;
1494 div64_u64_rem(DST, SRC, &AX);
1499 DST = do_div(AX, (u32) SRC);
1502 div64_u64_rem(DST, IMM, &AX);
1507 DST = do_div(AX, (u32) IMM);
1510 DST = div64_u64(DST, SRC);
1514 do_div(AX, (u32) SRC);
1518 DST = div64_u64(DST, IMM);
1522 do_div(AX, (u32) IMM);
1528 DST = (__force u16) cpu_to_be16(DST);
1531 DST = (__force u32) cpu_to_be32(DST);
1534 DST = (__force u64) cpu_to_be64(DST);
1541 DST = (__force u16) cpu_to_le16(DST);
1544 DST = (__force u32) cpu_to_le32(DST);
1547 DST = (__force u64) cpu_to_le64(DST);
1554 /* Function call scratches BPF_R1-BPF_R5 registers,
1555 * preserves BPF_R6-BPF_R9, and stores return value
1558 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1563 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1566 insn + insn->off + 1);
1570 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1571 struct bpf_array *array = container_of(map, struct bpf_array, map);
1572 struct bpf_prog *prog;
1575 if (unlikely(index >= array->map.max_entries))
1578 if (unlikely(tail_call_cnt >= MAX_TAIL_CALL_CNT))
1583 prog = READ_ONCE(array->ptrs[index]);
1587 /* ARG1 at this point is guaranteed to point to CTX from
1588 * the verifier side due to the fact that the tail call is
1589 * handled like a helper, that is, bpf_tail_call_proto,
1590 * where arg1_type is ARG_PTR_TO_CTX.
1592 insn = prog->insnsi;
1603 #define COND_JMP(SIGN, OPCODE, CMP_OP) \
1605 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \
1606 insn += insn->off; \
1610 JMP32_##OPCODE##_X: \
1611 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \
1612 insn += insn->off; \
1617 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \
1618 insn += insn->off; \
1622 JMP32_##OPCODE##_K: \
1623 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \
1624 insn += insn->off; \
1628 COND_JMP(u, JEQ, ==)
1629 COND_JMP(u, JNE, !=)
1632 COND_JMP(u, JGE, >=)
1633 COND_JMP(u, JLE, <=)
1634 COND_JMP(u, JSET, &)
1635 COND_JMP(s, JSGT, >)
1636 COND_JMP(s, JSLT, <)
1637 COND_JMP(s, JSGE, >=)
1638 COND_JMP(s, JSLE, <=)
1640 /* ST, STX and LDX*/
1642 /* Speculation barrier for mitigating Speculative Store Bypass.
1643 * In case of arm64, we rely on the firmware mitigation as
1644 * controlled via the ssbd kernel parameter. Whenever the
1645 * mitigation is enabled, it works for all of the kernel code
1646 * with no need to provide any additional instructions here.
1647 * In case of x86, we use 'lfence' insn for mitigation. We
1648 * reuse preexisting logic from Spectre v1 mitigation that
1649 * happens to produce the required code on x86 for v4 as well.
1655 #define LDST(SIZEOP, SIZE) \
1657 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1660 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1663 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1671 #define LDX_PROBE(SIZEOP, SIZE) \
1672 LDX_PROBE_MEM_##SIZEOP: \
1673 bpf_probe_read_kernel(&DST, SIZE, (const void *)(long) (SRC + insn->off)); \
1681 #define ATOMIC_ALU_OP(BOP, KOP) \
1683 if (BPF_SIZE(insn->code) == BPF_W) \
1684 atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
1685 (DST + insn->off)); \
1687 atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
1688 (DST + insn->off)); \
1690 case BOP | BPF_FETCH: \
1691 if (BPF_SIZE(insn->code) == BPF_W) \
1692 SRC = (u32) atomic_fetch_##KOP( \
1694 (atomic_t *)(unsigned long) (DST + insn->off)); \
1696 SRC = (u64) atomic64_fetch_##KOP( \
1698 (atomic64_t *)(unsigned long) (DST + insn->off)); \
1704 ATOMIC_ALU_OP(BPF_ADD, add)
1705 ATOMIC_ALU_OP(BPF_AND, and)
1706 ATOMIC_ALU_OP(BPF_OR, or)
1707 ATOMIC_ALU_OP(BPF_XOR, xor)
1708 #undef ATOMIC_ALU_OP
1711 if (BPF_SIZE(insn->code) == BPF_W)
1712 SRC = (u32) atomic_xchg(
1713 (atomic_t *)(unsigned long) (DST + insn->off),
1716 SRC = (u64) atomic64_xchg(
1717 (atomic64_t *)(unsigned long) (DST + insn->off),
1721 if (BPF_SIZE(insn->code) == BPF_W)
1722 BPF_R0 = (u32) atomic_cmpxchg(
1723 (atomic_t *)(unsigned long) (DST + insn->off),
1724 (u32) BPF_R0, (u32) SRC);
1726 BPF_R0 = (u64) atomic64_cmpxchg(
1727 (atomic64_t *)(unsigned long) (DST + insn->off),
1728 (u64) BPF_R0, (u64) SRC);
1737 /* If we ever reach this, we have a bug somewhere. Die hard here
1738 * instead of just returning 0; we could be somewhere in a subprog,
1739 * so execution could continue otherwise which we do /not/ want.
1741 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1743 pr_warn("BPF interpreter: unknown opcode %02x (imm: 0x%x)\n",
1744 insn->code, insn->imm);
1749 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1750 #define DEFINE_BPF_PROG_RUN(stack_size) \
1751 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1753 u64 stack[stack_size / sizeof(u64)]; \
1754 u64 regs[MAX_BPF_EXT_REG]; \
1756 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1757 ARG1 = (u64) (unsigned long) ctx; \
1758 return ___bpf_prog_run(regs, insn); \
1761 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1762 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1763 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1764 const struct bpf_insn *insn) \
1766 u64 stack[stack_size / sizeof(u64)]; \
1767 u64 regs[MAX_BPF_EXT_REG]; \
1769 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1775 return ___bpf_prog_run(regs, insn); \
1778 #define EVAL1(FN, X) FN(X)
1779 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1780 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1781 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1782 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1783 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1785 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1786 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1787 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1789 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1790 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1791 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1793 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1795 static unsigned int (*interpreters[])(const void *ctx,
1796 const struct bpf_insn *insn) = {
1797 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1798 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1799 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1801 #undef PROG_NAME_LIST
1802 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1803 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1804 const struct bpf_insn *insn) = {
1805 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1806 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1807 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1809 #undef PROG_NAME_LIST
1811 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1813 stack_depth = max_t(u32, stack_depth, 1);
1814 insn->off = (s16) insn->imm;
1815 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1816 __bpf_call_base_args;
1817 insn->code = BPF_JMP | BPF_CALL_ARGS;
1821 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1822 const struct bpf_insn *insn)
1824 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1825 * is not working properly, so warn about it!
1832 bool bpf_prog_array_compatible(struct bpf_array *array,
1833 const struct bpf_prog *fp)
1837 if (fp->kprobe_override)
1840 spin_lock(&array->aux->owner.lock);
1842 if (!array->aux->owner.type) {
1843 /* There's no owner yet where we could check for
1846 array->aux->owner.type = fp->type;
1847 array->aux->owner.jited = fp->jited;
1850 ret = array->aux->owner.type == fp->type &&
1851 array->aux->owner.jited == fp->jited;
1853 spin_unlock(&array->aux->owner.lock);
1857 static int bpf_check_tail_call(const struct bpf_prog *fp)
1859 struct bpf_prog_aux *aux = fp->aux;
1862 mutex_lock(&aux->used_maps_mutex);
1863 for (i = 0; i < aux->used_map_cnt; i++) {
1864 struct bpf_map *map = aux->used_maps[i];
1865 struct bpf_array *array;
1867 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1870 array = container_of(map, struct bpf_array, map);
1871 if (!bpf_prog_array_compatible(array, fp)) {
1878 mutex_unlock(&aux->used_maps_mutex);
1882 static void bpf_prog_select_func(struct bpf_prog *fp)
1884 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1885 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1887 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1889 fp->bpf_func = __bpf_prog_ret0_warn;
1894 * bpf_prog_select_runtime - select exec runtime for BPF program
1895 * @fp: bpf_prog populated with BPF program
1896 * @err: pointer to error variable
1898 * Try to JIT eBPF program, if JIT is not available, use interpreter.
1899 * The BPF program will be executed via bpf_prog_run() function.
1901 * Return: the &fp argument along with &err set to 0 for success or
1902 * a negative errno code on failure
1904 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1906 /* In case of BPF to BPF calls, verifier did all the prep
1907 * work with regards to JITing, etc.
1909 bool jit_needed = false;
1914 if (IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) ||
1915 bpf_prog_has_kfunc_call(fp))
1918 bpf_prog_select_func(fp);
1920 /* eBPF JITs can rewrite the program in case constant
1921 * blinding is active. However, in case of error during
1922 * blinding, bpf_int_jit_compile() must always return a
1923 * valid program, which in this case would simply not
1924 * be JITed, but falls back to the interpreter.
1926 if (!bpf_prog_is_dev_bound(fp->aux)) {
1927 *err = bpf_prog_alloc_jited_linfo(fp);
1931 fp = bpf_int_jit_compile(fp);
1932 bpf_prog_jit_attempt_done(fp);
1933 if (!fp->jited && jit_needed) {
1938 *err = bpf_prog_offload_compile(fp);
1944 bpf_prog_lock_ro(fp);
1946 /* The tail call compatibility check can only be done at
1947 * this late stage as we need to determine, if we deal
1948 * with JITed or non JITed program concatenations and not
1949 * all eBPF JITs might immediately support all features.
1951 *err = bpf_check_tail_call(fp);
1955 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1957 static unsigned int __bpf_prog_ret1(const void *ctx,
1958 const struct bpf_insn *insn)
1963 static struct bpf_prog_dummy {
1964 struct bpf_prog prog;
1965 } dummy_bpf_prog = {
1967 .bpf_func = __bpf_prog_ret1,
1971 /* to avoid allocating empty bpf_prog_array for cgroups that
1972 * don't have bpf program attached use one global 'empty_prog_array'
1973 * It will not be modified the caller of bpf_prog_array_alloc()
1974 * (since caller requested prog_cnt == 0)
1975 * that pointer should be 'freed' by bpf_prog_array_free()
1978 struct bpf_prog_array hdr;
1979 struct bpf_prog *null_prog;
1980 } empty_prog_array = {
1984 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1987 return kzalloc(sizeof(struct bpf_prog_array) +
1988 sizeof(struct bpf_prog_array_item) *
1992 return &empty_prog_array.hdr;
1995 void bpf_prog_array_free(struct bpf_prog_array *progs)
1997 if (!progs || progs == &empty_prog_array.hdr)
1999 kfree_rcu(progs, rcu);
2002 int bpf_prog_array_length(struct bpf_prog_array *array)
2004 struct bpf_prog_array_item *item;
2007 for (item = array->items; item->prog; item++)
2008 if (item->prog != &dummy_bpf_prog.prog)
2013 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
2015 struct bpf_prog_array_item *item;
2017 for (item = array->items; item->prog; item++)
2018 if (item->prog != &dummy_bpf_prog.prog)
2023 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
2027 struct bpf_prog_array_item *item;
2030 for (item = array->items; item->prog; item++) {
2031 if (item->prog == &dummy_bpf_prog.prog)
2033 prog_ids[i] = item->prog->aux->id;
2034 if (++i == request_cnt) {
2040 return !!(item->prog);
2043 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
2044 __u32 __user *prog_ids, u32 cnt)
2046 unsigned long err = 0;
2050 /* users of this function are doing:
2051 * cnt = bpf_prog_array_length();
2053 * bpf_prog_array_copy_to_user(..., cnt);
2054 * so below kcalloc doesn't need extra cnt > 0 check.
2056 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
2059 nospc = bpf_prog_array_copy_core(array, ids, cnt);
2060 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
2069 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
2070 struct bpf_prog *old_prog)
2072 struct bpf_prog_array_item *item;
2074 for (item = array->items; item->prog; item++)
2075 if (item->prog == old_prog) {
2076 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
2082 * bpf_prog_array_delete_safe_at() - Replaces the program at the given
2083 * index into the program array with
2084 * a dummy no-op program.
2085 * @array: a bpf_prog_array
2086 * @index: the index of the program to replace
2088 * Skips over dummy programs, by not counting them, when calculating
2089 * the position of the program to replace.
2093 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2094 * * -ENOENT - Index out of range
2096 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
2098 return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
2102 * bpf_prog_array_update_at() - Updates the program at the given index
2103 * into the program array.
2104 * @array: a bpf_prog_array
2105 * @index: the index of the program to update
2106 * @prog: the program to insert into the array
2108 * Skips over dummy programs, by not counting them, when calculating
2109 * the position of the program to update.
2113 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2114 * * -ENOENT - Index out of range
2116 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2117 struct bpf_prog *prog)
2119 struct bpf_prog_array_item *item;
2121 if (unlikely(index < 0))
2124 for (item = array->items; item->prog; item++) {
2125 if (item->prog == &dummy_bpf_prog.prog)
2128 WRITE_ONCE(item->prog, prog);
2136 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2137 struct bpf_prog *exclude_prog,
2138 struct bpf_prog *include_prog,
2140 struct bpf_prog_array **new_array)
2142 int new_prog_cnt, carry_prog_cnt = 0;
2143 struct bpf_prog_array_item *existing, *new;
2144 struct bpf_prog_array *array;
2145 bool found_exclude = false;
2147 /* Figure out how many existing progs we need to carry over to
2151 existing = old_array->items;
2152 for (; existing->prog; existing++) {
2153 if (existing->prog == exclude_prog) {
2154 found_exclude = true;
2157 if (existing->prog != &dummy_bpf_prog.prog)
2159 if (existing->prog == include_prog)
2164 if (exclude_prog && !found_exclude)
2167 /* How many progs (not NULL) will be in the new array? */
2168 new_prog_cnt = carry_prog_cnt;
2172 /* Do we have any prog (not NULL) in the new array? */
2173 if (!new_prog_cnt) {
2178 /* +1 as the end of prog_array is marked with NULL */
2179 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2184 /* Fill in the new prog array */
2185 if (carry_prog_cnt) {
2186 existing = old_array->items;
2187 for (; existing->prog; existing++) {
2188 if (existing->prog == exclude_prog ||
2189 existing->prog == &dummy_bpf_prog.prog)
2192 new->prog = existing->prog;
2193 new->bpf_cookie = existing->bpf_cookie;
2198 new->prog = include_prog;
2199 new->bpf_cookie = bpf_cookie;
2207 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2208 u32 *prog_ids, u32 request_cnt,
2214 cnt = bpf_prog_array_length(array);
2218 /* return early if user requested only program count or nothing to copy */
2219 if (!request_cnt || !cnt)
2222 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
2223 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
2227 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2228 struct bpf_map **used_maps, u32 len)
2230 struct bpf_map *map;
2233 for (i = 0; i < len; i++) {
2235 if (map->ops->map_poke_untrack)
2236 map->ops->map_poke_untrack(map, aux);
2241 static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2243 __bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
2244 kfree(aux->used_maps);
2247 void __bpf_free_used_btfs(struct bpf_prog_aux *aux,
2248 struct btf_mod_pair *used_btfs, u32 len)
2250 #ifdef CONFIG_BPF_SYSCALL
2251 struct btf_mod_pair *btf_mod;
2254 for (i = 0; i < len; i++) {
2255 btf_mod = &used_btfs[i];
2256 if (btf_mod->module)
2257 module_put(btf_mod->module);
2258 btf_put(btf_mod->btf);
2263 static void bpf_free_used_btfs(struct bpf_prog_aux *aux)
2265 __bpf_free_used_btfs(aux, aux->used_btfs, aux->used_btf_cnt);
2266 kfree(aux->used_btfs);
2269 static void bpf_prog_free_deferred(struct work_struct *work)
2271 struct bpf_prog_aux *aux;
2274 aux = container_of(work, struct bpf_prog_aux, work);
2275 #ifdef CONFIG_BPF_SYSCALL
2276 bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
2278 bpf_free_used_maps(aux);
2279 bpf_free_used_btfs(aux);
2280 if (bpf_prog_is_dev_bound(aux))
2281 bpf_prog_offload_destroy(aux->prog);
2282 #ifdef CONFIG_PERF_EVENTS
2283 if (aux->prog->has_callchain_buf)
2284 put_callchain_buffers();
2286 if (aux->dst_trampoline)
2287 bpf_trampoline_put(aux->dst_trampoline);
2288 for (i = 0; i < aux->func_cnt; i++) {
2289 /* We can just unlink the subprog poke descriptor table as
2290 * it was originally linked to the main program and is also
2291 * released along with it.
2293 aux->func[i]->aux->poke_tab = NULL;
2294 bpf_jit_free(aux->func[i]);
2296 if (aux->func_cnt) {
2298 bpf_prog_unlock_free(aux->prog);
2300 bpf_jit_free(aux->prog);
2304 void bpf_prog_free(struct bpf_prog *fp)
2306 struct bpf_prog_aux *aux = fp->aux;
2309 bpf_prog_put(aux->dst_prog);
2310 INIT_WORK(&aux->work, bpf_prog_free_deferred);
2311 schedule_work(&aux->work);
2313 EXPORT_SYMBOL_GPL(bpf_prog_free);
2315 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2316 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2318 void bpf_user_rnd_init_once(void)
2320 prandom_init_once(&bpf_user_rnd_state);
2323 BPF_CALL_0(bpf_user_rnd_u32)
2325 /* Should someone ever have the rather unwise idea to use some
2326 * of the registers passed into this function, then note that
2327 * this function is called from native eBPF and classic-to-eBPF
2328 * transformations. Register assignments from both sides are
2329 * different, f.e. classic always sets fn(ctx, A, X) here.
2331 struct rnd_state *state;
2334 state = &get_cpu_var(bpf_user_rnd_state);
2335 res = prandom_u32_state(state);
2336 put_cpu_var(bpf_user_rnd_state);
2341 BPF_CALL_0(bpf_get_raw_cpu_id)
2343 return raw_smp_processor_id();
2346 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2347 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2348 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2349 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2350 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2351 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2352 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2353 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2354 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2355 const struct bpf_func_proto bpf_jiffies64_proto __weak;
2357 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2358 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2359 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2360 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2361 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
2362 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto __weak;
2364 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2365 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2366 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2367 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2368 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
2369 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2370 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
2371 const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
2372 const struct bpf_func_proto bpf_seq_printf_btf_proto __weak;
2374 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2379 const struct bpf_func_proto * __weak bpf_get_trace_vprintk_proto(void)
2385 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2386 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2390 EXPORT_SYMBOL_GPL(bpf_event_output);
2392 /* Always built-in helper functions. */
2393 const struct bpf_func_proto bpf_tail_call_proto = {
2396 .ret_type = RET_VOID,
2397 .arg1_type = ARG_PTR_TO_CTX,
2398 .arg2_type = ARG_CONST_MAP_PTR,
2399 .arg3_type = ARG_ANYTHING,
2402 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2403 * It is encouraged to implement bpf_int_jit_compile() instead, so that
2404 * eBPF and implicitly also cBPF can get JITed!
2406 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2411 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2412 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2414 void __weak bpf_jit_compile(struct bpf_prog *prog)
2418 bool __weak bpf_helper_changes_pkt_data(void *func)
2423 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2424 * analysis code and wants explicit zero extension inserted by verifier.
2425 * Otherwise, return FALSE.
2427 * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
2428 * you don't override this. JITs that don't want these extra insns can detect
2429 * them using insn_is_zext.
2431 bool __weak bpf_jit_needs_zext(void)
2436 bool __weak bpf_jit_supports_kfunc_call(void)
2441 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2442 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2444 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2450 int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
2451 void *addr1, void *addr2)
2456 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2457 EXPORT_SYMBOL(bpf_stats_enabled_key);
2459 /* All definitions of tracepoints related to BPF. */
2460 #define CREATE_TRACE_POINTS
2461 #include <linux/bpf_trace.h>
2463 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2464 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);