projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'mm-hotfixes-stable-2025-07-11-16-16' of git://git.kernel.org/pub/scm/linux...
[linux-block.git] / kernel / bpf / core.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
f5bffecd
AS		 2/*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
f5bffecd 16 * Andi Kleen - Fix a few bad bugs and races.
4df95ff4 17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
f5bffecd 18 */
738cbe72 19
838e9690 20#include <uapi/linux/btf.h>
f5bffecd
AS		 21#include <linux/filter.h>
22#include <linux/skbuff.h>
60a3b225 23#include <linux/vmalloc.h>
aaedc2ff 24#include <linux/prandom.h>
09756af4 25#include <linux/bpf.h>
838e9690 26#include <linux/btf.h>
00089c04 27#include <linux/objtool.h>
cb01621b 28#include <linux/overflow.h>
74451e66
DB		 29#include <linux/rbtree_latch.h>
30#include <linux/kallsyms.h>
31#include <linux/rcupdate.h>
c195651e 32#include <linux/perf_event.h>
3dec541b 33#include <linux/extable.h>
b7b3fc8d 34#include <linux/log2.h>
2357672c 35#include <linux/bpf_verifier.h>
ef078600 36#include <linux/nodemask.h>
f3dd0c53 37#include <linux/nospec.h>
958cf2e2 38#include <linux/bpf_mem_alloc.h>
bf396508 39#include <linux/memcontrol.h>
12af2b83 40#include <linux/execmem.h>
f5e81d11
DB		 41
42#include <asm/barrier.h>
5f60d5f6 43#include <linux/unaligned.h>
3324b584 44
f5bffecd
AS		 45/* Registers */
46#define BPF_R0 regs[BPF_REG_0]
47#define BPF_R1 regs[BPF_REG_1]
48#define BPF_R2 regs[BPF_REG_2]
49#define BPF_R3 regs[BPF_REG_3]
50#define BPF_R4 regs[BPF_REG_4]
51#define BPF_R5 regs[BPF_REG_5]
52#define BPF_R6 regs[BPF_REG_6]
53#define BPF_R7 regs[BPF_REG_7]
54#define BPF_R8 regs[BPF_REG_8]
55#define BPF_R9 regs[BPF_REG_9]
56#define BPF_R10 regs[BPF_REG_10]
57
58/* Named registers */
59#define DST regs[insn->dst_reg]
60#define SRC regs[insn->src_reg]
61#define FP regs[BPF_REG_FP]
144cd91c 62#define AX regs[BPF_REG_AX]
f5bffecd
AS		 63#define ARG1 regs[BPF_REG_ARG1]
64#define CTX regs[BPF_REG_CTX]
8100928c 65#define OFF insn->off
f5bffecd
AS		 66#define IMM insn->imm
67
1fda5bb6
YS		 68struct bpf_mem_alloc bpf_global_ma;
69bool bpf_global_ma_set;
958cf2e2 70
f5bffecd
AS		 71/* No hurry in this branch
72 *
73 * Exported for the bpf jit load helper.
74 */
75void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
76{
77 u8 *ptr = NULL;
78
0326195f 79 if (k >= SKF_NET_OFF) {
f5bffecd 80 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
0326195f
ED		 81 } else if (k >= SKF_LL_OFF) {
82 if (unlikely(!skb_mac_header_was_set(skb)))
83 return NULL;
f5bffecd 84 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
0326195f 85 }
f5bffecd
AS		 86 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
87 return ptr;
88
89 return NULL;
90}
91
fe506415
AS		 92/* tell bpf programs that include vmlinux.h kernel's PAGE_SIZE */
93enum page_size_enum {
94 __PAGE_SIZE = PAGE_SIZE
95};
96
492ecee8 97struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
60a3b225 98{
bf396508 99 gfp_t gfp_flags = bpf_memcg_flags(GFP_KERNEL | __GFP_ZERO | gfp_extra_flags);
09756af4 100 struct bpf_prog_aux *aux;
60a3b225
DB		 101 struct bpf_prog *fp;
102
fe506415 103 size = round_up(size, __PAGE_SIZE);
88dca4ca 104 fp = __vmalloc(size, gfp_flags);
60a3b225
DB		 105 if (fp == NULL)
106 return NULL;
107
bf396508 108 aux = kzalloc(sizeof(*aux), bpf_memcg_flags(GFP_KERNEL | gfp_extra_flags));
09756af4 109 if (aux == NULL) {
60a3b225
DB		 110 vfree(fp);
111 return NULL;
112 }
bf396508 113 fp->active = alloc_percpu_gfp(int, bpf_memcg_flags(GFP_KERNEL | gfp_extra_flags));
ca06f55b
AS		 114 if (!fp->active) {
115 vfree(fp);
116 kfree(aux);
117 return NULL;
118 }
60a3b225
DB		 119
120 fp->pages = size / PAGE_SIZE;
09756af4 121 fp->aux = aux;
e9d8afa9 122 fp->aux->prog = fp;
60b58afc 123 fp->jit_requested = ebpf_jit_enabled();
d2a3b7c5 124 fp->blinding_requested = bpf_jit_blinding_enabled(fp);
c0e19f2c
SF		 125#ifdef CONFIG_CGROUP_BPF
126 aux->cgroup_atype = CGROUP_BPF_ATTACH_TYPE_INVALID;
127#endif
60a3b225 128
ecb60d1c 129 INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
4f9087f1
PZ		 130#ifdef CONFIG_FINEIBT
131 INIT_LIST_HEAD_RCU(&fp->aux->ksym_prefix.lnode);
132#endif
984fe94f 133 mutex_init(&fp->aux->used_maps_mutex);
d6083f04 134 mutex_init(&fp->aux->ext_mutex);
3aac1ead 135 mutex_init(&fp->aux->dst_mutex);
74451e66 136
60a3b225
DB		 137 return fp;
138}
492ecee8
AS		 139
140struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
141{
bf396508 142 gfp_t gfp_flags = bpf_memcg_flags(GFP_KERNEL | __GFP_ZERO | gfp_extra_flags);
492ecee8 143 struct bpf_prog *prog;
4b911304 144 int cpu;
492ecee8
AS		 145
146 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
147 if (!prog)
148 return NULL;
149
700d4796
AS		 150 prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
151 if (!prog->stats) {
ca06f55b 152 free_percpu(prog->active);
492ecee8
AS		 153 kfree(prog->aux);
154 vfree(prog);
155 return NULL;
156 }
157
4b911304
ED		 158 for_each_possible_cpu(cpu) {
159 struct bpf_prog_stats *pstats;
160
700d4796 161 pstats = per_cpu_ptr(prog->stats, cpu);
4b911304
ED		 162 u64_stats_init(&pstats->syncp);
163 }
492ecee8
AS		 164 return prog;
165}
60a3b225
DB		 166EXPORT_SYMBOL_GPL(bpf_prog_alloc);
167
c454a46b
MKL		 168int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
169{
170 if (!prog->aux->nr_linfo || !prog->jit_requested)
171 return 0;
172
e16301fb
MKL		 173 prog->aux->jited_linfo = kvcalloc(prog->aux->nr_linfo,
174 sizeof(*prog->aux->jited_linfo),
bf396508 175 bpf_memcg_flags(GFP_KERNEL | __GFP_NOWARN));
c454a46b
MKL		 176 if (!prog->aux->jited_linfo)
177 return -ENOMEM;
178
179 return 0;
180}
181
e16301fb 182void bpf_prog_jit_attempt_done(struct bpf_prog *prog)
c454a46b 183{
e16301fb
MKL		 184 if (prog->aux->jited_linfo &&
185 (!prog->jited || !prog->aux->jited_linfo[0])) {
186 kvfree(prog->aux->jited_linfo);
187 prog->aux->jited_linfo = NULL;
188 }
e6ac2450
MKL		 189
190 kfree(prog->aux->kfunc_tab);
191 prog->aux->kfunc_tab = NULL;
c454a46b
MKL		 192}
193
194/* The jit engine is responsible to provide an array
195 * for insn_off to the jited_off mapping (insn_to_jit_off).
196 *
197 * The idx to this array is the insn_off. Hence, the insn_off
198 * here is relative to the prog itself instead of the main prog.
199 * This array has one entry for each xlated bpf insn.
200 *
cc168554 201 * jited_off is the byte off to the end of the jited insn.
c454a46b
MKL		 202 *
203 * Hence, with
204 * insn_start:
205 * The first bpf insn off of the prog. The insn off
206 * here is relative to the main prog.
207 * e.g. if prog is a subprog, insn_start > 0
208 * linfo_idx:
209 * The prog's idx to prog->aux->linfo and jited_linfo
210 *
211 * jited_linfo[linfo_idx] = prog->bpf_func
212 *
213 * For i > linfo_idx,
214 *
215 * jited_linfo[i] = prog->bpf_func +
216 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
217 */
218void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
219 const u32 *insn_to_jit_off)
220{
221 u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
222 const struct bpf_line_info *linfo;
223 void **jited_linfo;
224
335d1c5b 225 if (!prog->aux->jited_linfo || prog->aux->func_idx > prog->aux->func_cnt)
c454a46b
MKL		 226 /* Userspace did not provide linfo */
227 return;
228
229 linfo_idx = prog->aux->linfo_idx;
230 linfo = &prog->aux->linfo[linfo_idx];
231 insn_start = linfo[0].insn_off;
232 insn_end = insn_start + prog->len;
233
234 jited_linfo = &prog->aux->jited_linfo[linfo_idx];
235 jited_linfo[0] = prog->bpf_func;
236
237 nr_linfo = prog->aux->nr_linfo - linfo_idx;
238
239 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
240 /* The verifier ensures that linfo[i].insn_off is
241 * strictly increasing
242 */
243 jited_linfo[i] = prog->bpf_func +
244 insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
245}
246
60a3b225
DB		 247struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
248 gfp_t gfp_extra_flags)
249{
bf396508 250 gfp_t gfp_flags = bpf_memcg_flags(GFP_KERNEL | __GFP_ZERO | gfp_extra_flags);
60a3b225 251 struct bpf_prog *fp;
3ac1f01b 252 u32 pages;
60a3b225 253
60a3b225 254 size = round_up(size, PAGE_SIZE);
5ccb071e
DB		 255 pages = size / PAGE_SIZE;
256 if (pages <= fp_old->pages)
60a3b225
DB		 257 return fp_old;
258
88dca4ca 259 fp = __vmalloc(size, gfp_flags);
3ac1f01b 260 if (fp) {
60a3b225 261 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
5ccb071e 262 fp->pages = pages;
e9d8afa9 263 fp->aux->prog = fp;
60a3b225 264
09756af4 265 /* We keep fp->aux from fp_old around in the new
60a3b225
DB		 266 * reallocated structure.
267 */
09756af4 268 fp_old->aux = NULL;
1336c662
AS		 269 fp_old->stats = NULL;
270 fp_old->active = NULL;
60a3b225
DB		 271 __bpf_prog_free(fp_old);
272 }
273
274 return fp;
275}
60a3b225
DB		 276
277void __bpf_prog_free(struct bpf_prog *fp)
278{
492ecee8 279 if (fp->aux) {
984fe94f 280 mutex_destroy(&fp->aux->used_maps_mutex);
3aac1ead 281 mutex_destroy(&fp->aux->dst_mutex);
a66886fe 282 kfree(fp->aux->poke_tab);
492ecee8
AS		 283 kfree(fp->aux);
284 }
700d4796 285 free_percpu(fp->stats);
ca06f55b 286 free_percpu(fp->active);
60a3b225
DB		 287 vfree(fp);
288}
60a3b225 289
f1f7714e 290int bpf_prog_calc_tag(struct bpf_prog *fp)
7bd509e3 291{
6b0b0fa2 292 const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
f1f7714e 293 u32 raw_size = bpf_prog_tag_scratch_size(fp);
6b0b0fa2
EB		 294 u32 digest[SHA1_DIGEST_WORDS];
295 u32 ws[SHA1_WORKSPACE_WORDS];
7bd509e3 296 u32 i, bsize, psize, blocks;
aafe6ae9 297 struct bpf_insn *dst;
7bd509e3 298 bool was_ld_map;
aafe6ae9 299 u8 *raw, *todo;
7bd509e3
DB		 300 __be32 *result;
301 __be64 *bits;
302
aafe6ae9
DB		 303 raw = vmalloc(raw_size);
304 if (!raw)
305 return -ENOMEM;
306
6b0b0fa2 307 sha1_init(digest);
7bd509e3
DB		 308 memset(ws, 0, sizeof(ws));
309
310 /* We need to take out the map fd for the digest calculation
311 * since they are unstable from user space side.
312 */
aafe6ae9 313 dst = (void *)raw;
7bd509e3
DB		 314 for (i = 0, was_ld_map = false; i < fp->len; i++) {
315 dst[i] = fp->insnsi[i];
316 if (!was_ld_map &&
317 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
d8eca5bb
DB		 318 (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
319 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
7bd509e3
DB		 320 was_ld_map = true;
321 dst[i].imm = 0;
322 } else if (was_ld_map &&
323 dst[i].code == 0 &&
324 dst[i].dst_reg == 0 &&
325 dst[i].src_reg == 0 &&
326 dst[i].off == 0) {
327 was_ld_map = false;
328 dst[i].imm = 0;
329 } else {
330 was_ld_map = false;
331 }
332 }
333
aafe6ae9
DB		 334 psize = bpf_prog_insn_size(fp);
335 memset(&raw[psize], 0, raw_size - psize);
7bd509e3
DB		 336 raw[psize++] = 0x80;
337
6b0b0fa2
EB		 338 bsize = round_up(psize, SHA1_BLOCK_SIZE);
339 blocks = bsize / SHA1_BLOCK_SIZE;
aafe6ae9 340 todo = raw;
7bd509e3
DB		 341 if (bsize - psize >= sizeof(__be64)) {
342 bits = (__be64 *)(todo + bsize - sizeof(__be64));
343 } else {
344 bits = (__be64 *)(todo + bsize + bits_offset);
345 blocks++;
346 }
347 *bits = cpu_to_be64((psize - 1) << 3);
348
349 while (blocks--) {
6b0b0fa2
EB		 350 sha1_transform(digest, todo, ws);
351 todo += SHA1_BLOCK_SIZE;
7bd509e3
DB		 352 }
353
f1f7714e 354 result = (__force __be32 *)digest;
6b0b0fa2 355 for (i = 0; i < SHA1_DIGEST_WORDS; i++)
f1f7714e
DB		 356 result[i] = cpu_to_be32(digest[i]);
357 memcpy(fp->tag, result, sizeof(fp->tag));
aafe6ae9
DB		 358
359 vfree(raw);
360 return 0;
7bd509e3
DB		 361}
362
2cbd95a5 363static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
af959b18 364 s32 end_new, s32 curr, const bool probe_pass)
c237ee5e 365{
050fad7c 366 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
2cbd95a5 367 s32 delta = end_new - end_old;
050fad7c
DB		 368 s64 imm = insn->imm;
369
2cbd95a5 370 if (curr < pos && curr + imm + 1 >= end_old)
050fad7c 371 imm += delta;
2cbd95a5 372 else if (curr >= end_new && curr + imm + 1 < end_new)
050fad7c
DB		 373 imm -= delta;
374 if (imm < imm_min || imm > imm_max)
375 return -ERANGE;
376 if (!probe_pass)
377 insn->imm = imm;
378 return 0;
379}
380
2cbd95a5 381static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
af959b18 382 s32 end_new, s32 curr, const bool probe_pass)
050fad7c 383{
dfce9cb3 384 s64 off_min, off_max, off;
2cbd95a5 385 s32 delta = end_new - end_old;
4cd58e9a 386
dfce9cb3 387 if (insn->code == (BPF_JMP32 | BPF_JA)) {
4cd58e9a 388 off = insn->imm;
dfce9cb3
YS		 389 off_min = S32_MIN;
390 off_max = S32_MAX;
391 } else {
4cd58e9a 392 off = insn->off;
dfce9cb3
YS		 393 off_min = S16_MIN;
394 off_max = S16_MAX;
395 }
050fad7c 396
2cbd95a5 397 if (curr < pos && curr + off + 1 >= end_old)
050fad7c 398 off += delta;
2cbd95a5 399 else if (curr >= end_new && curr + off + 1 < end_new)
050fad7c
DB		 400 off -= delta;
401 if (off < off_min || off > off_max)
402 return -ERANGE;
4cd58e9a
YS		 403 if (!probe_pass) {
404 if (insn->code == (BPF_JMP32 | BPF_JA))
405 insn->imm = off;
406 else
407 insn->off = off;
408 }
050fad7c
DB		 409 return 0;
410}
411
2cbd95a5
JK		 412static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
413 s32 end_new, const bool probe_pass)
050fad7c 414{
2cbd95a5 415 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
c237ee5e 416 struct bpf_insn *insn = prog->insnsi;
050fad7c 417 int ret = 0;
c237ee5e
DB		 418
419 for (i = 0; i < insn_cnt; i++, insn++) {
050fad7c
DB		 420 u8 code;
421
422 /* In the probing pass we still operate on the original,
423 * unpatched image in order to check overflows before we
424 * do any other adjustments. Therefore skip the patchlet.
425 */
426 if (probe_pass && i == pos) {
2cbd95a5
JK		 427 i = end_new;
428 insn = prog->insnsi + end_old;
050fad7c 429 }
3990ed4c
MKL		 430 if (bpf_pseudo_func(insn)) {
431 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
432 end_new, i, probe_pass);
433 if (ret)
434 return ret;
435 continue;
436 }
1ea47e01 437 code = insn->code;
092ed096
JW		 438 if ((BPF_CLASS(code) != BPF_JMP &&
439 BPF_CLASS(code) != BPF_JMP32) ||
050fad7c 440 BPF_OP(code) == BPF_EXIT)
1ea47e01 441 continue;
050fad7c 442 /* Adjust offset of jmps if we cross patch boundaries. */
1ea47e01 443 if (BPF_OP(code) == BPF_CALL) {
050fad7c 444 if (insn->src_reg != BPF_PSEUDO_CALL)
1ea47e01 445 continue;
2cbd95a5
JK		 446 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
447 end_new, i, probe_pass);
1ea47e01 448 } else {
2cbd95a5
JK		 449 ret = bpf_adj_delta_to_off(insn, pos, end_old,
450 end_new, i, probe_pass);
1ea47e01 451 }
050fad7c
DB		 452 if (ret)
453 break;
c237ee5e 454 }
050fad7c
DB		 455
456 return ret;
c237ee5e
DB		 457}
458
c454a46b
MKL		 459static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
460{
461 struct bpf_line_info *linfo;
462 u32 i, nr_linfo;
463
464 nr_linfo = prog->aux->nr_linfo;
465 if (!nr_linfo || !delta)
466 return;
467
468 linfo = prog->aux->linfo;
469
470 for (i = 0; i < nr_linfo; i++)
471 if (off < linfo[i].insn_off)
472 break;
473
474 /* Push all off < linfo[i].insn_off by delta */
475 for (; i < nr_linfo; i++)
476 linfo[i].insn_off += delta;
477}
478
c237ee5e
DB		 479struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
480 const struct bpf_insn *patch, u32 len)
481{
482 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
050fad7c 483 const u32 cnt_max = S16_MAX;
c237ee5e 484 struct bpf_prog *prog_adj;
4f73379e 485 int err;
c237ee5e
DB		 486
487 /* Since our patchlet doesn't expand the image, we're done. */
488 if (insn_delta == 0) {
489 memcpy(prog->insnsi + off, patch, sizeof(*patch));
490 return prog;
491 }
492
493 insn_adj_cnt = prog->len + insn_delta;
494
050fad7c
DB		 495 /* Reject anything that would potentially let the insn->off
496 * target overflow when we have excessive program expansions.
497 * We need to probe here before we do any reallocation where
498 * we afterwards may not fail anymore.
499 */
500 if (insn_adj_cnt > cnt_max &&
4f73379e
AS		 501 (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
502 return ERR_PTR(err);
050fad7c 503
c237ee5e
DB		 504 /* Several new instructions need to be inserted. Make room
505 * for them. Likely, there's no need for a new allocation as
506 * last page could have large enough tailroom.
507 */
508 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
509 GFP_USER);
510 if (!prog_adj)
4f73379e 511 return ERR_PTR(-ENOMEM);
c237ee5e
DB		 512
513 prog_adj->len = insn_adj_cnt;
514
515 /* Patching happens in 3 steps:
516 *
517 * 1) Move over tail of insnsi from next instruction onwards,
518 * so we can patch the single target insn with one or more
519 * new ones (patching is always from 1 to n insns, n > 0).
520 * 2) Inject new instructions at the target location.
521 * 3) Adjust branch offsets if necessary.
522 */
523 insn_rest = insn_adj_cnt - off - len;
524
525 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
526 sizeof(*patch) * insn_rest);
527 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
528
050fad7c
DB		 529 /* We are guaranteed to not fail at this point, otherwise
530 * the ship has sailed to reverse to the original state. An
531 * overflow cannot happen at this point.
532 */
2cbd95a5 533 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
c237ee5e 534
c454a46b
MKL		 535 bpf_adj_linfo(prog_adj, off, insn_delta);
536
c237ee5e
DB		 537 return prog_adj;
538}
539
52875a04
JK		 540int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
541{
c4441ca8
AP		 542 int err;
543
52875a04
JK		 544 /* Branch offsets can't overflow when program is shrinking, no need
545 * to call bpf_adj_branches(..., true) here
546 */
547 memmove(prog->insnsi + off, prog->insnsi + off + cnt,
548 sizeof(struct bpf_insn) * (prog->len - off - cnt));
549 prog->len -= cnt;
550
c4441ca8
AP		 551 err = bpf_adj_branches(prog, off, off + cnt, off, false);
552 WARN_ON_ONCE(err);
553 return err;
52875a04
JK		 554}
555
cd7455f1 556static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
7d1982b4
DB		 557{
558 int i;
559
335d1c5b 560 for (i = 0; i < fp->aux->real_func_cnt; i++)
7d1982b4
DB		 561 bpf_prog_kallsyms_del(fp->aux->func[i]);
562}
563
564void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
565{
566 bpf_prog_kallsyms_del_subprogs(fp);
567 bpf_prog_kallsyms_del(fp);
568}
569
b954d834 570#ifdef CONFIG_BPF_JIT
fa9dd599 571/* All BPF JIT sysctl knobs here. */
81c22041
DB		 572int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
573int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
fa9dd599 574int bpf_jit_harden __read_mostly;
fdadd049 575long bpf_jit_limit __read_mostly;
fadb7ff1 576long bpf_jit_limit_max __read_mostly;
fa9dd599 577
535911c8
JO		 578static void
579bpf_prog_ksym_set_addr(struct bpf_prog *prog)
74451e66 580{
74451e66
DB		 581 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
582
535911c8 583 prog->aux->ksym.start = (unsigned long) prog->bpf_func;
d00c6473 584 prog->aux->ksym.end = prog->aux->ksym.start + prog->jited_len;
74451e66
DB		 585}
586
bfea9a85
JO		 587static void
588bpf_prog_ksym_set_name(struct bpf_prog *prog)
74451e66 589{
bfea9a85 590 char *sym = prog->aux->ksym.name;
368211fb 591 const char *end = sym + KSYM_NAME_LEN;
838e9690
YS		 592 const struct btf_type *type;
593 const char *func_name;
368211fb 594
74451e66 595 BUILD_BUG_ON(sizeof("bpf_prog_") +
368211fb
MKL		 596 sizeof(prog->tag) * 2 +
597 /* name has been null terminated.
598 * We should need +1 for the '_' preceding
599 * the name. However, the null character
600 * is double counted between the name and the
601 * sizeof("bpf_prog_") above, so we omit
602 * the +1 here.
603 */
604 sizeof(prog->aux->name) > KSYM_NAME_LEN);
74451e66
DB		 605
606 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
607 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
838e9690
YS		 608
609 /* prog->aux->name will be ignored if full btf name is available */
335d1c5b 610 if (prog->aux->func_info_cnt && prog->aux->func_idx < prog->aux->func_info_cnt) {
ba64e7d8
YS		 611 type = btf_type_by_id(prog->aux->btf,
612 prog->aux->func_info[prog->aux->func_idx].type_id);
838e9690
YS		 613 func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
614 snprintf(sym, (size_t)(end - sym), "_%s", func_name);
615 return;
616 }
617
368211fb
MKL		 618 if (prog->aux->name[0])
619 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
620 else
621 *sym = 0;
74451e66
DB		 622}
623
ca4424c9 624static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
74451e66 625{
ca4424c9 626 return container_of(n, struct bpf_ksym, tnode)->start;
74451e66
DB		 627}
628
629static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
630 struct latch_tree_node *b)
631{
ca4424c9 632 return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
74451e66
DB		 633}
634
635static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
636{
637 unsigned long val = (unsigned long)key;
ca4424c9 638 const struct bpf_ksym *ksym;
74451e66 639
ca4424c9 640 ksym = container_of(n, struct bpf_ksym, tnode);
74451e66 641
ca4424c9 642 if (val < ksym->start)
74451e66 643 return -1;
66d9111f
KKD		 644 /* Ensure that we detect return addresses as part of the program, when
645 * the final instruction is a call for a program part of the stack
646 * trace. Therefore, do val > ksym->end instead of val >= ksym->end.
647 */
648 if (val > ksym->end)
74451e66
DB		 649 return 1;
650
651 return 0;
652}
653
654static const struct latch_tree_ops bpf_tree_ops = {
655 .less = bpf_tree_less,
656 .comp = bpf_tree_comp,
657};
658
659static DEFINE_SPINLOCK(bpf_lock);
660static LIST_HEAD(bpf_kallsyms);
661static struct latch_tree_root bpf_tree __cacheline_aligned;
662
dba122fb 663void bpf_ksym_add(struct bpf_ksym *ksym)
74451e66 664{
dba122fb
JO		 665 spin_lock_bh(&bpf_lock);
666 WARN_ON_ONCE(!list_empty(&ksym->lnode));
667 list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
668 latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
669 spin_unlock_bh(&bpf_lock);
74451e66
DB		 670}
671
dba122fb 672static void __bpf_ksym_del(struct bpf_ksym *ksym)
74451e66 673{
dba122fb 674 if (list_empty(&ksym->lnode))
74451e66
DB		 675 return;
676
dba122fb
JO		 677 latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
678 list_del_rcu(&ksym->lnode);
679}
680
681void bpf_ksym_del(struct bpf_ksym *ksym)
682{
683 spin_lock_bh(&bpf_lock);
684 __bpf_ksym_del(ksym);
685 spin_unlock_bh(&bpf_lock);
74451e66
DB		 686}
687
688static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
689{
690 return fp->jited && !bpf_prog_was_classic(fp);
691}
692
74451e66
DB		 693void bpf_prog_kallsyms_add(struct bpf_prog *fp)
694{
74451e66 695 if (!bpf_prog_kallsyms_candidate(fp) ||
d79a3549 696 !bpf_token_capable(fp->aux->token, CAP_BPF))
74451e66
DB		 697 return;
698
535911c8 699 bpf_prog_ksym_set_addr(fp);
bfea9a85 700 bpf_prog_ksym_set_name(fp);
cbd76f8d 701 fp->aux->ksym.prog = true;
535911c8 702
dba122fb 703 bpf_ksym_add(&fp->aux->ksym);
4f9087f1
PZ		 704
705#ifdef CONFIG_FINEIBT
706 /*
707 * When FineIBT, code in the __cfi_foo() symbols can get executed
708 * and hence unwinder needs help.
709 */
710 if (cfi_mode != CFI_FINEIBT)
711 return;
712
713 snprintf(fp->aux->ksym_prefix.name, KSYM_NAME_LEN,
714 "__cfi_%s", fp->aux->ksym.name);
715
716 fp->aux->ksym_prefix.start = (unsigned long) fp->bpf_func - 16;
717 fp->aux->ksym_prefix.end = (unsigned long) fp->bpf_func;
718
719 bpf_ksym_add(&fp->aux->ksym_prefix);
720#endif
74451e66
DB		 721}
722
723void bpf_prog_kallsyms_del(struct bpf_prog *fp)
724{
74451e66
DB		 725 if (!bpf_prog_kallsyms_candidate(fp))
726 return;
727
dba122fb 728 bpf_ksym_del(&fp->aux->ksym);
4f9087f1
PZ		 729#ifdef CONFIG_FINEIBT
730 if (cfi_mode != CFI_FINEIBT)
731 return;
732 bpf_ksym_del(&fp->aux->ksym_prefix);
733#endif
74451e66
DB		 734}
735
eda0c929
JO		 736static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
737{
738 struct latch_tree_node *n;
739
740 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
741 return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
742}
743
7e1f4eb9 744int __bpf_address_lookup(unsigned long addr, unsigned long *size,
74451e66
DB		 745 unsigned long *off, char *sym)
746{
eda0c929 747 struct bpf_ksym *ksym;
7e1f4eb9 748 int ret = 0;
74451e66
DB		 749
750 rcu_read_lock();
eda0c929
JO		 751 ksym = bpf_ksym_find(addr);
752 if (ksym) {
753 unsigned long symbol_start = ksym->start;
754 unsigned long symbol_end = ksym->end;
535911c8 755
7e1f4eb9 756 ret = strscpy(sym, ksym->name, KSYM_NAME_LEN);
74451e66 757
74451e66
DB		 758 if (size)
759 *size = symbol_end - symbol_start;
760 if (off)
761 *off = addr - symbol_start;
762 }
763 rcu_read_unlock();
764
765 return ret;
766}
767
768bool is_bpf_text_address(unsigned long addr)
769{
770 bool ret;
771
772 rcu_read_lock();
eda0c929 773 ret = bpf_ksym_find(addr) != NULL;
74451e66
DB		 774 rcu_read_unlock();
775
776 return ret;
777}
778
f18b03fa 779struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
cbd76f8d
JO		 780{
781 struct bpf_ksym *ksym = bpf_ksym_find(addr);
782
783 return ksym && ksym->prog ?
784 container_of(ksym, struct bpf_prog_aux, ksym)->prog :
785 NULL;
786}
787
3dec541b
AS		 788const struct exception_table_entry *search_bpf_extables(unsigned long addr)
789{
790 const struct exception_table_entry *e = NULL;
791 struct bpf_prog *prog;
792
793 rcu_read_lock();
cbd76f8d 794 prog = bpf_prog_ksym_find(addr);
3dec541b
AS		 795 if (!prog)
796 goto out;
797 if (!prog->aux->num_exentries)
798 goto out;
799
800 e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
801out:
802 rcu_read_unlock();
803 return e;
804}
805
74451e66
DB		 806int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
807 char *sym)
808{
ecb60d1c 809 struct bpf_ksym *ksym;
74451e66
DB		 810 unsigned int it = 0;
811 int ret = -ERANGE;
812
813 if (!bpf_jit_kallsyms_enabled())
814 return ret;
815
816 rcu_read_lock();
ecb60d1c 817 list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
74451e66
DB		 818 if (it++ != symnum)
819 continue;
820
2e114248 821 strscpy(sym, ksym->name, KSYM_NAME_LEN);
74451e66 822
ecb60d1c 823 *value = ksym->start;
74451e66
DB		 824 *type = BPF_SYM_ELF_TYPE;
825
826 ret = 0;
827 break;
828 }
829 rcu_read_unlock();
830
831 return ret;
832}
833
a66886fe
DB		 834int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
835 struct bpf_jit_poke_descriptor *poke)
836{
837 struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
838 static const u32 poke_tab_max = 1024;
839 u32 slot = prog->aux->size_poke_tab;
840 u32 size = slot + 1;
841
842 if (size > poke_tab_max)
843 return -ENOSPC;
cf71b174 844 if (poke->tailcall_target || poke->tailcall_target_stable ||
ebf7d1f5 845 poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
a66886fe
DB		 846 return -EINVAL;
847
848 switch (poke->reason) {
849 case BPF_POKE_REASON_TAIL_CALL:
850 if (!poke->tail_call.map)
851 return -EINVAL;
852 break;
853 default:
854 return -EINVAL;
855 }
856
a3034872 857 tab = krealloc_array(tab, size, sizeof(*poke), GFP_KERNEL);
a66886fe
DB		 858 if (!tab)
859 return -ENOMEM;
860
861 memcpy(&tab[slot], poke, sizeof(*poke));
862 prog->aux->size_poke_tab = size;
863 prog->aux->poke_tab = tab;
864
865 return slot;
866}
867
57631054
SL		 868/*
869 * BPF program pack allocator.
870 *
871 * Most BPF programs are pretty small. Allocating a hole page for each
872 * program is sometime a waste. Many small bpf program also adds pressure
873 * to instruction TLB. To solve this issue, we introduce a BPF program pack
874 * allocator. The prog_pack allocator uses HPAGE_PMD_SIZE page (2MB on x86)
875 * to host BPF programs.
876 */
57631054
SL		 877#define BPF_PROG_CHUNK_SHIFT 6
878#define BPF_PROG_CHUNK_SIZE (1 << BPF_PROG_CHUNK_SHIFT)
879#define BPF_PROG_CHUNK_MASK (~(BPF_PROG_CHUNK_SIZE - 1))
57631054
SL		 880
881struct bpf_prog_pack {
882 struct list_head list;
883 void *ptr;
4cc0991a 884 unsigned long bitmap[];
57631054
SL		 885};
886
19c02415
SL		 887void bpf_jit_fill_hole_with_zero(void *area, unsigned int size)
888{
889 memset(area, 0, size);
890}
891
57631054
SL		 892#define BPF_PROG_SIZE_TO_NBITS(size) (round_up(size, BPF_PROG_CHUNK_SIZE) / BPF_PROG_CHUNK_SIZE)
893
894static DEFINE_MUTEX(pack_mutex);
895static LIST_HEAD(pack_list);
896
e5810941
SL		 897/* PMD_SIZE is not available in some special config, e.g. ARCH=arm with
898 * CONFIG_MMU=n. Use PAGE_SIZE in these cases.
899 */
900#ifdef PMD_SIZE
d6170e4a
PM		 901/* PMD_SIZE is really big for some archs. It doesn't make sense to
902 * reserve too much memory in one allocation. Hardcode BPF_PROG_PACK_SIZE to
903 * 2MiB * num_possible_nodes(). On most architectures PMD_SIZE will be
904 * greater than or equal to 2MB.
905 */
906#define BPF_PROG_PACK_SIZE (SZ_2M * num_possible_nodes())
e5810941 907#else
ea2babac 908#define BPF_PROG_PACK_SIZE PAGE_SIZE
e5810941
SL		 909#endif
910
ea2babac 911#define BPF_PROG_CHUNK_COUNT (BPF_PROG_PACK_SIZE / BPF_PROG_CHUNK_SIZE)
ef078600 912
d88bb5ee 913static struct bpf_prog_pack *alloc_new_pack(bpf_jit_fill_hole_t bpf_fill_ill_insns)
57631054
SL		 914{
915 struct bpf_prog_pack *pack;
c733239f 916 int err;
57631054 917
ea2babac 918 pack = kzalloc(struct_size(pack, bitmap, BITS_TO_LONGS(BPF_PROG_CHUNK_COUNT)),
ef078600 919 GFP_KERNEL);
57631054
SL		 920 if (!pack)
921 return NULL;
20e490ad 922 pack->ptr = bpf_jit_alloc_exec(BPF_PROG_PACK_SIZE);
c733239f
CL		 923 if (!pack->ptr)
924 goto out;
ea2babac
SL		 925 bpf_fill_ill_insns(pack->ptr, BPF_PROG_PACK_SIZE);
926 bitmap_zero(pack->bitmap, BPF_PROG_PACK_SIZE / BPF_PROG_CHUNK_SIZE);
57631054
SL		 927
928 set_vm_flush_reset_perms(pack->ptr);
c733239f
CL		 929 err = set_memory_rox((unsigned long)pack->ptr,
930 BPF_PROG_PACK_SIZE / PAGE_SIZE);
931 if (err)
932 goto out;
933 list_add_tail(&pack->list, &pack_list);
57631054 934 return pack;
c733239f
CL		 935
936out:
937 bpf_jit_free_exec(pack->ptr);
938 kfree(pack);
939 return NULL;
57631054
SL		 940}
941
19c02415 942void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns)
57631054
SL		 943{
944 unsigned int nbits = BPF_PROG_SIZE_TO_NBITS(size);
945 struct bpf_prog_pack *pack;
946 unsigned long pos;
947 void *ptr = NULL;
948
ef078600 949 mutex_lock(&pack_mutex);
ea2babac 950 if (size > BPF_PROG_PACK_SIZE) {
57631054 951 size = round_up(size, PAGE_SIZE);
20e490ad 952 ptr = bpf_jit_alloc_exec(size);
57631054 953 if (ptr) {
c733239f
CL		 954 int err;
955
d88bb5ee 956 bpf_fill_ill_insns(ptr, size);
57631054 957 set_vm_flush_reset_perms(ptr);
c733239f
CL		 958 err = set_memory_rox((unsigned long)ptr,
959 size / PAGE_SIZE);
960 if (err) {
961 bpf_jit_free_exec(ptr);
962 ptr = NULL;
963 }
57631054 964 }
ef078600 965 goto out;
57631054 966 }
57631054 967 list_for_each_entry(pack, &pack_list, list) {
ea2babac 968 pos = bitmap_find_next_zero_area(pack->bitmap, BPF_PROG_CHUNK_COUNT, 0,
57631054 969 nbits, 0);
ea2babac 970 if (pos < BPF_PROG_CHUNK_COUNT)
57631054
SL		 971 goto found_free_area;
972 }
973
d88bb5ee 974 pack = alloc_new_pack(bpf_fill_ill_insns);
57631054
SL		 975 if (!pack)
976 goto out;
977
978 pos = 0;
979
980found_free_area:
981 bitmap_set(pack->bitmap, pos, nbits);
982 ptr = (void *)(pack->ptr) + (pos << BPF_PROG_CHUNK_SHIFT);
983
984out:
985 mutex_unlock(&pack_mutex);
986 return ptr;
987}
988
f08a1c65 989void bpf_prog_pack_free(void *ptr, u32 size)
57631054
SL		 990{
991 struct bpf_prog_pack *pack = NULL, *tmp;
992 unsigned int nbits;
993 unsigned long pos;
57631054 994
ef078600 995 mutex_lock(&pack_mutex);
f08a1c65
SL		 996 if (size > BPF_PROG_PACK_SIZE) {
997 bpf_jit_free_exec(ptr);
ef078600 998 goto out;
57631054
SL		 999 }
1000
57631054 1001 list_for_each_entry(tmp, &pack_list, list) {
f08a1c65 1002 if (ptr >= tmp->ptr && (tmp->ptr + BPF_PROG_PACK_SIZE) > ptr) {
57631054
SL		 1003 pack = tmp;
1004 break;
1005 }
1006 }
1007
1008 if (WARN_ONCE(!pack, "bpf_prog_pack bug\n"))
1009 goto out;
1010
f08a1c65
SL		 1011 nbits = BPF_PROG_SIZE_TO_NBITS(size);
1012 pos = ((unsigned long)ptr - (unsigned long)pack->ptr) >> BPF_PROG_CHUNK_SHIFT;
57631054 1013
f08a1c65 1014 WARN_ONCE(bpf_arch_text_invalidate(ptr, size),
fe736565
SL		 1015 "bpf_prog_pack bug: missing bpf_arch_text_invalidate?\n");
1016
57631054 1017 bitmap_clear(pack->bitmap, pos, nbits);
ea2babac
SL		 1018 if (bitmap_find_next_zero_area(pack->bitmap, BPF_PROG_CHUNK_COUNT, 0,
1019 BPF_PROG_CHUNK_COUNT, 0) == 0) {
57631054 1020 list_del(&pack->list);
20e490ad 1021 bpf_jit_free_exec(pack->ptr);
57631054
SL		 1022 kfree(pack);
1023 }
1024out:
1025 mutex_unlock(&pack_mutex);
1026}
1027
ede95a63
DB		 1028static atomic_long_t bpf_jit_current;
1029
fdadd049
DB		 1030/* Can be overridden by an arch's JIT compiler if it has a custom,
1031 * dedicated BPF backend memory area, or if neither of the two
1032 * below apply.
1033 */
1034u64 __weak bpf_jit_alloc_exec_limit(void)
1035{
ede95a63 1036#if defined(MODULES_VADDR)
fdadd049
DB		 1037 return MODULES_END - MODULES_VADDR;
1038#else
1039 return VMALLOC_END - VMALLOC_START;
1040#endif
1041}
1042
ede95a63
DB		 1043static int __init bpf_jit_charge_init(void)
1044{
1045 /* Only used as heuristic here to derive limit. */
fadb7ff1 1046 bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
10ec8ca8 1047 bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 1,
fdadd049 1048 PAGE_SIZE), LONG_MAX);
ede95a63
DB		 1049 return 0;
1050}
1051pure_initcall(bpf_jit_charge_init);
ede95a63 1052
3486bedd 1053int bpf_jit_charge_modmem(u32 size)
ede95a63 1054{
0947ae11 1055 if (atomic_long_add_return(size, &bpf_jit_current) > READ_ONCE(bpf_jit_limit)) {
8a98ae12 1056 if (!bpf_capable()) {
3486bedd 1057 atomic_long_sub(size, &bpf_jit_current);
ede95a63
DB		 1058 return -EPERM;
1059 }
1060 }
1061
1062 return 0;
1063}
1064
3486bedd 1065void bpf_jit_uncharge_modmem(u32 size)
ede95a63 1066{
3486bedd 1067 atomic_long_sub(size, &bpf_jit_current);
ede95a63
DB		 1068}
1069
dc002bb6
AB		 1070void *__weak bpf_jit_alloc_exec(unsigned long size)
1071{
12af2b83 1072 return execmem_alloc(EXECMEM_BPF, size);
dc002bb6
AB		 1073}
1074
1075void __weak bpf_jit_free_exec(void *addr)
1076{
12af2b83 1077 execmem_free(addr);
dc002bb6
AB		 1078}
1079
738cbe72
DB		 1080struct bpf_binary_header *
1081bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1082 unsigned int alignment,
1083 bpf_jit_fill_hole_t bpf_fill_ill_insns)
1084{
1085 struct bpf_binary_header *hdr;
ed2d9e1a 1086 u32 size, hole, start;
738cbe72 1087
b7b3fc8d
IL		 1088 WARN_ON_ONCE(!is_power_of_2(alignment) ||
1089 alignment > BPF_IMAGE_ALIGNMENT);
1090
738cbe72
DB		 1091 /* Most of BPF filters are really small, but if some of them
1092 * fill a page, allow at least 128 extra bytes to insert a
1093 * random section of illegal instructions.
1094 */
1095 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
ede95a63 1096
3486bedd 1097 if (bpf_jit_charge_modmem(size))
ede95a63 1098 return NULL;
dc002bb6 1099 hdr = bpf_jit_alloc_exec(size);
ede95a63 1100 if (!hdr) {
3486bedd 1101 bpf_jit_uncharge_modmem(size);
738cbe72 1102 return NULL;
ede95a63 1103 }
738cbe72
DB		 1104
1105 /* Fill space with illegal/arch-dep instructions. */
1106 bpf_fill_ill_insns(hdr, size);
1107
ed2d9e1a 1108 hdr->size = size;
738cbe72
DB		 1109 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
1110 PAGE_SIZE - sizeof(*hdr));
8032bf12 1111 start = get_random_u32_below(hole) & ~(alignment - 1);
738cbe72
DB		 1112
1113 /* Leave a random number of instructions before BPF code. */
1114 *image_ptr = &hdr->image[start];
1115
1116 return hdr;
1117}
1118
1119void bpf_jit_binary_free(struct bpf_binary_header *hdr)
1120{
ed2d9e1a 1121 u32 size = hdr->size;
ede95a63 1122
dc002bb6 1123 bpf_jit_free_exec(hdr);
ed2d9e1a 1124 bpf_jit_uncharge_modmem(size);
738cbe72 1125}
4f3446bb 1126
33c98058
SL		 1127/* Allocate jit binary from bpf_prog_pack allocator.
1128 * Since the allocated memory is RO+X, the JIT engine cannot write directly
1129 * to the memory. To solve this problem, a RW buffer is also allocated at
1130 * as the same time. The JIT engine should calculate offsets based on the
1131 * RO memory address, but write JITed program to the RW buffer. Once the
1132 * JIT engine finishes, it calls bpf_jit_binary_pack_finalize, which copies
1133 * the JITed program to the RO memory.
1134 */
1135struct bpf_binary_header *
1136bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **image_ptr,
1137 unsigned int alignment,
1138 struct bpf_binary_header **rw_header,
1139 u8 **rw_image,
1140 bpf_jit_fill_hole_t bpf_fill_ill_insns)
1141{
1142 struct bpf_binary_header *ro_header;
1143 u32 size, hole, start;
1144
1145 WARN_ON_ONCE(!is_power_of_2(alignment) ||
1146 alignment > BPF_IMAGE_ALIGNMENT);
1147
1148 /* add 16 bytes for a random section of illegal instructions */
1149 size = round_up(proglen + sizeof(*ro_header) + 16, BPF_PROG_CHUNK_SIZE);
1150
1151 if (bpf_jit_charge_modmem(size))
1152 return NULL;
d88bb5ee 1153 ro_header = bpf_prog_pack_alloc(size, bpf_fill_ill_insns);
33c98058
SL		 1154 if (!ro_header) {
1155 bpf_jit_uncharge_modmem(size);
1156 return NULL;
1157 }
1158
1159 *rw_header = kvmalloc(size, GFP_KERNEL);
1160 if (!*rw_header) {
f08a1c65 1161 bpf_prog_pack_free(ro_header, size);
33c98058
SL		 1162 bpf_jit_uncharge_modmem(size);
1163 return NULL;
1164 }
1165
1166 /* Fill space with illegal/arch-dep instructions. */
1167 bpf_fill_ill_insns(*rw_header, size);
1168 (*rw_header)->size = size;
1169
1170 hole = min_t(unsigned int, size - (proglen + sizeof(*ro_header)),
1171 BPF_PROG_CHUNK_SIZE - sizeof(*ro_header));
8032bf12 1172 start = get_random_u32_below(hole) & ~(alignment - 1);
33c98058
SL		 1173
1174 *image_ptr = &ro_header->image[start];
1175 *rw_image = &(*rw_header)->image[start];
1176
1177 return ro_header;
1178}
1179
1180/* Copy JITed text from rw_header to its final location, the ro_header. */
9919c5c9 1181int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header,
33c98058
SL		 1182 struct bpf_binary_header *rw_header)
1183{
1184 void *ptr;
1185
1186 ptr = bpf_arch_text_copy(ro_header, rw_header, rw_header->size);
1187
1188 kvfree(rw_header);
1189
1190 if (IS_ERR(ptr)) {
f08a1c65 1191 bpf_prog_pack_free(ro_header, ro_header->size);
33c98058
SL		 1192 return PTR_ERR(ptr);
1193 }
33c98058
SL		 1194 return 0;
1195}
1196
1197/* bpf_jit_binary_pack_free is called in two different scenarios:
1198 * 1) when the program is freed after;
1199 * 2) when the JIT engine fails (before bpf_jit_binary_pack_finalize).
1200 * For case 2), we need to free both the RO memory and the RW buffer.
676b2daa
SL		 1201 *
1202 * bpf_jit_binary_pack_free requires proper ro_header->size. However,
1203 * bpf_jit_binary_pack_alloc does not set it. Therefore, ro_header->size
1204 * must be set with either bpf_jit_binary_pack_finalize (normal path) or
1205 * bpf_arch_text_copy (when jit fails).
33c98058
SL		 1206 */
1207void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1208 struct bpf_binary_header *rw_header)
1209{
676b2daa 1210 u32 size = ro_header->size;
33c98058 1211
f08a1c65 1212 bpf_prog_pack_free(ro_header, size);
33c98058
SL		 1213 kvfree(rw_header);
1214 bpf_jit_uncharge_modmem(size);
1215}
1216
1d5f82d9
SL		 1217struct bpf_binary_header *
1218bpf_jit_binary_pack_hdr(const struct bpf_prog *fp)
1219{
1220 unsigned long real_start = (unsigned long)fp->bpf_func;
1221 unsigned long addr;
1222
1223 addr = real_start & BPF_PROG_CHUNK_MASK;
1224 return (void *)addr;
1225}
1226
33c98058
SL		 1227static inline struct bpf_binary_header *
1228bpf_jit_binary_hdr(const struct bpf_prog *fp)
1229{
1230 unsigned long real_start = (unsigned long)fp->bpf_func;
1231 unsigned long addr;
1232
1d5f82d9 1233 addr = real_start & PAGE_MASK;
33c98058
SL		 1234 return (void *)addr;
1235}
1236
74451e66
DB		 1237/* This symbol is only overridden by archs that have different
1238 * requirements than the usual eBPF JITs, f.e. when they only
1239 * implement cBPF JIT, do not set images read-only, etc.
1240 */
1241void __weak bpf_jit_free(struct bpf_prog *fp)
1242{
1243 if (fp->jited) {
1244 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
1245
1d5f82d9 1246 bpf_jit_binary_free(hdr);
74451e66
DB		 1247 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
1248 }
1249
1250 bpf_prog_unlock_free(fp);
1251}
1252
e2c95a61
DB		 1253int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1254 const struct bpf_insn *insn, bool extra_pass,
1255 u64 *func_addr, bool *func_addr_fixed)
1256{
1257 s16 off = insn->off;
1258 s32 imm = insn->imm;
1259 u8 *addr;
1cf3bfc6 1260 int err;
e2c95a61
DB		 1261
1262 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
1263 if (!*func_addr_fixed) {
1264 /* Place-holder address till the last pass has collected
1265 * all addresses for JITed subprograms in which case we
1266 * can pick them up from prog->aux.
1267 */
1268 if (!extra_pass)
1269 addr = NULL;
1270 else if (prog->aux->func &&
335d1c5b 1271 off >= 0 && off < prog->aux->real_func_cnt)
e2c95a61
DB		 1272 addr = (u8 *)prog->aux->func[off]->bpf_func;
1273 else
1274 return -EINVAL;
1cf3bfc6
IL		 1275 } else if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL &&
1276 bpf_jit_supports_far_kfunc_call()) {
1277 err = bpf_get_kfunc_addr(prog, insn->imm, insn->off, &addr);
1278 if (err)
1279 return err;
e2c95a61
DB		 1280 } else {
1281 /* Address of a BPF helper call. Since part of the core
1282 * kernel, it's always at a fixed location. __bpf_call_base
1283 * and the helper with imm relative to it are both in core
1284 * kernel.
1285 */
1286 addr = (u8 *)__bpf_call_base + imm;
1287 }
1288
1289 *func_addr = (unsigned long)addr;
1290 return 0;
1291}
1292
4f3446bb
DB		 1293static int bpf_jit_blind_insn(const struct bpf_insn *from,
1294 const struct bpf_insn *aux,
ede7c460
NR		 1295 struct bpf_insn *to_buff,
1296 bool emit_zext)
4f3446bb
DB		 1297{
1298 struct bpf_insn *to = to_buff;
a251c17a 1299 u32 imm_rnd = get_random_u32();
4f3446bb
DB		 1300 s16 off;
1301
1302 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
1303 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
1304
9b73bfdd
DB		 1305 /* Constraints on AX register:
1306 *
1307 * AX register is inaccessible from user space. It is mapped in
1308 * all JITs, and used here for constant blinding rewrites. It is
1309 * typically "stateless" meaning its contents are only valid within
1310 * the executed instruction, but not across several instructions.
1311 * There are a few exceptions however which are further detailed
1312 * below.
1313 *
1314 * Constant blinding is only used by JITs, not in the interpreter.
1315 * The interpreter uses AX in some occasions as a local temporary
1316 * register e.g. in DIV or MOD instructions.
1317 *
1318 * In restricted circumstances, the verifier can also use the AX
1319 * register for rewrites as long as they do not interfere with
1320 * the above cases!
1321 */
1322 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
1323 goto out;
1324
4f3446bb
DB		 1325 if (from->imm == 0 &&
1326 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
1327 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
1328 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
1329 goto out;
1330 }
1331
1332 switch (from->code) {
1333 case BPF_ALU | BPF_ADD | BPF_K:
1334 case BPF_ALU | BPF_SUB | BPF_K:
1335 case BPF_ALU | BPF_AND | BPF_K:
1336 case BPF_ALU | BPF_OR | BPF_K:
1337 case BPF_ALU | BPF_XOR | BPF_K:
1338 case BPF_ALU | BPF_MUL | BPF_K:
1339 case BPF_ALU | BPF_MOV | BPF_K:
1340 case BPF_ALU | BPF_DIV | BPF_K:
1341 case BPF_ALU | BPF_MOD | BPF_K:
1342 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1343 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
7058e3a3 1344 *to++ = BPF_ALU32_REG_OFF(from->code, from->dst_reg, BPF_REG_AX, from->off);
4f3446bb
DB		 1345 break;
1346
1347 case BPF_ALU64 | BPF_ADD | BPF_K:
1348 case BPF_ALU64 | BPF_SUB | BPF_K:
1349 case BPF_ALU64 | BPF_AND | BPF_K:
1350 case BPF_ALU64 | BPF_OR | BPF_K:
1351 case BPF_ALU64 | BPF_XOR | BPF_K:
1352 case BPF_ALU64 | BPF_MUL | BPF_K:
1353 case BPF_ALU64 | BPF_MOV | BPF_K:
1354 case BPF_ALU64 | BPF_DIV | BPF_K:
1355 case BPF_ALU64 | BPF_MOD | BPF_K:
1356 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1357 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
7058e3a3 1358 *to++ = BPF_ALU64_REG_OFF(from->code, from->dst_reg, BPF_REG_AX, from->off);
4f3446bb
DB		 1359 break;
1360
1361 case BPF_JMP | BPF_JEQ | BPF_K:
1362 case BPF_JMP | BPF_JNE | BPF_K:
1363 case BPF_JMP | BPF_JGT | BPF_K:
92b31a9a 1364 case BPF_JMP | BPF_JLT | BPF_K:
4f3446bb 1365 case BPF_JMP | BPF_JGE | BPF_K:
92b31a9a 1366 case BPF_JMP | BPF_JLE | BPF_K:
4f3446bb 1367 case BPF_JMP | BPF_JSGT | BPF_K:
92b31a9a 1368 case BPF_JMP | BPF_JSLT | BPF_K:
4f3446bb 1369 case BPF_JMP | BPF_JSGE | BPF_K:
92b31a9a 1370 case BPF_JMP | BPF_JSLE | BPF_K:
4f3446bb
DB		 1371 case BPF_JMP | BPF_JSET | BPF_K:
1372 /* Accommodate for extra offset in case of a backjump. */
1373 off = from->off;
1374 if (off < 0)
1375 off -= 2;
1376 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1377 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1378 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1379 break;
1380
a7b76c88
JW		 1381 case BPF_JMP32 | BPF_JEQ | BPF_K:
1382 case BPF_JMP32 | BPF_JNE | BPF_K:
1383 case BPF_JMP32 | BPF_JGT | BPF_K:
1384 case BPF_JMP32 | BPF_JLT | BPF_K:
1385 case BPF_JMP32 | BPF_JGE | BPF_K:
1386 case BPF_JMP32 | BPF_JLE | BPF_K:
1387 case BPF_JMP32 | BPF_JSGT | BPF_K:
1388 case BPF_JMP32 | BPF_JSLT | BPF_K:
1389 case BPF_JMP32 | BPF_JSGE | BPF_K:
1390 case BPF_JMP32 | BPF_JSLE | BPF_K:
1391 case BPF_JMP32 | BPF_JSET | BPF_K:
1392 /* Accommodate for extra offset in case of a backjump. */
1393 off = from->off;
1394 if (off < 0)
1395 off -= 2;
1396 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1397 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1398 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1399 off);
1400 break;
1401
4f3446bb
DB		 1402 case BPF_LD | BPF_IMM | BPF_DW:
1403 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1404 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1405 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1406 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1407 break;
1408 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1409 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1410 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
ede7c460
NR		 1411 if (emit_zext)
1412 *to++ = BPF_ZEXT_REG(BPF_REG_AX);
4f3446bb
DB		 1413 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
1414 break;
1415
1416 case BPF_ST | BPF_MEM | BPF_DW:
1417 case BPF_ST | BPF_MEM | BPF_W:
1418 case BPF_ST | BPF_MEM | BPF_H:
1419 case BPF_ST | BPF_MEM | BPF_B:
1420 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1421 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1422 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1423 break;
1424 }
1425out:
1426 return to - to_buff;
1427}
1428
1429static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1430 gfp_t gfp_extra_flags)
1431{
19809c2d 1432 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
4f3446bb
DB		 1433 struct bpf_prog *fp;
1434
88dca4ca 1435 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
4f3446bb 1436 if (fp != NULL) {
4f3446bb
DB		 1437 /* aux->prog still points to the fp_other one, so
1438 * when promoting the clone to the real program,
1439 * this still needs to be adapted.
1440 */
1441 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1442 }
1443
1444 return fp;
1445}
1446
1447static void bpf_prog_clone_free(struct bpf_prog *fp)
1448{
1449 /* aux was stolen by the other clone, so we cannot free
1450 * it from this path! It will be freed eventually by the
1451 * other program on release.
1452 *
1453 * At this point, we don't need a deferred release since
1454 * clone is guaranteed to not be locked.
1455 */
1456 fp->aux = NULL;
53f523f3
CW		 1457 fp->stats = NULL;
1458 fp->active = NULL;
4f3446bb
DB		 1459 __bpf_prog_free(fp);
1460}
1461
1462void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1463{
1464 /* We have to repoint aux->prog to self, as we don't
1465 * know whether fp here is the clone or the original.
1466 */
1467 fp->aux->prog = fp;
1468 bpf_prog_clone_free(fp_other);
1469}
1470
1471struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1472{
1473 struct bpf_insn insn_buff[16], aux[2];
1474 struct bpf_prog *clone, *tmp;
1475 int insn_delta, insn_cnt;
1476 struct bpf_insn *insn;
1477 int i, rewritten;
1478
d2a3b7c5 1479 if (!prog->blinding_requested || prog->blinded)
4f3446bb
DB		 1480 return prog;
1481
1482 clone = bpf_prog_clone_create(prog, GFP_USER);
1483 if (!clone)
1484 return ERR_PTR(-ENOMEM);
1485
1486 insn_cnt = clone->len;
1487 insn = clone->insnsi;
1488
1489 for (i = 0; i < insn_cnt; i++, insn++) {
4b6313cf
AS		 1490 if (bpf_pseudo_func(insn)) {
1491 /* ld_imm64 with an address of bpf subprog is not
1492 * a user controlled constant. Don't randomize it,
1493 * since it will conflict with jit_subprogs() logic.
1494 */
1495 insn++;
1496 i++;
1497 continue;
1498 }
1499
4f3446bb
DB		 1500 /* We temporarily need to hold the original ld64 insn
1501 * so that we can still access the first part in the
1502 * second blinding run.
1503 */
1504 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1505 insn[1].code == 0)
1506 memcpy(aux, insn, sizeof(aux));
1507
ede7c460
NR		 1508 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1509 clone->aux->verifier_zext);
4f3446bb
DB		 1510 if (!rewritten)
1511 continue;
1512
1513 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
4f73379e 1514 if (IS_ERR(tmp)) {
4f3446bb
DB		 1515 /* Patching may have repointed aux->prog during
1516 * realloc from the original one, so we need to
1517 * fix it up here on error.
1518 */
1519 bpf_jit_prog_release_other(prog, clone);
4f73379e 1520 return tmp;
4f3446bb
DB		 1521 }
1522
1523 clone = tmp;
1524 insn_delta = rewritten - 1;
1525
1526 /* Walk new program and skip insns we just inserted. */
1527 insn = clone->insnsi + i + insn_delta;
1528 insn_cnt += insn_delta;
1529 i += insn_delta;
1530 }
1531
1c2a088a 1532 clone->blinded = 1;
4f3446bb
DB		 1533 return clone;
1534}
b954d834 1535#endif /* CONFIG_BPF_JIT */
738cbe72 1536
f5bffecd
AS		 1537/* Base function for offset calculation. Needs to go into .text section,
1538 * therefore keeping it non-static as well; will also be used by JITs
7105e828
DB		 1539 * anyway later on, so do not let the compiler omit it. This also needs
1540 * to go into kallsyms for correlation from e.g. bpftool, so naming
1541 * must not change.
f5bffecd
AS		 1542 */
1543noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1544{
1545 return 0;
1546}
4d9c5c53 1547EXPORT_SYMBOL_GPL(__bpf_call_base);
f5bffecd 1548
5e581dad
DB		 1549/* All UAPI available opcodes. */
1550#define BPF_INSN_MAP(INSN_2, INSN_3) \
1551 /* 32 bit ALU operations. */ \
1552 /* Register based. */ \
2dc6b100
JW		 1553 INSN_3(ALU, ADD, X), \
1554 INSN_3(ALU, SUB, X), \
1555 INSN_3(ALU, AND, X), \
1556 INSN_3(ALU, OR, X), \
1557 INSN_3(ALU, LSH, X), \
1558 INSN_3(ALU, RSH, X), \
1559 INSN_3(ALU, XOR, X), \
1560 INSN_3(ALU, MUL, X), \
1561 INSN_3(ALU, MOV, X), \
1562 INSN_3(ALU, ARSH, X), \
1563 INSN_3(ALU, DIV, X), \
1564 INSN_3(ALU, MOD, X), \
5e581dad
DB		 1565 INSN_2(ALU, NEG), \
1566 INSN_3(ALU, END, TO_BE), \
1567 INSN_3(ALU, END, TO_LE), \
1568 /* Immediate based. */ \
2dc6b100
JW		 1569 INSN_3(ALU, ADD, K), \
1570 INSN_3(ALU, SUB, K), \
1571 INSN_3(ALU, AND, K), \
1572 INSN_3(ALU, OR, K), \
1573 INSN_3(ALU, LSH, K), \
1574 INSN_3(ALU, RSH, K), \
1575 INSN_3(ALU, XOR, K), \
1576 INSN_3(ALU, MUL, K), \
1577 INSN_3(ALU, MOV, K), \
1578 INSN_3(ALU, ARSH, K), \
1579 INSN_3(ALU, DIV, K), \
1580 INSN_3(ALU, MOD, K), \
5e581dad
DB		 1581 /* 64 bit ALU operations. */ \
1582 /* Register based. */ \
1583 INSN_3(ALU64, ADD, X), \
1584 INSN_3(ALU64, SUB, X), \
1585 INSN_3(ALU64, AND, X), \
1586 INSN_3(ALU64, OR, X), \
1587 INSN_3(ALU64, LSH, X), \
1588 INSN_3(ALU64, RSH, X), \
1589 INSN_3(ALU64, XOR, X), \
1590 INSN_3(ALU64, MUL, X), \
1591 INSN_3(ALU64, MOV, X), \
1592 INSN_3(ALU64, ARSH, X), \
1593 INSN_3(ALU64, DIV, X), \
1594 INSN_3(ALU64, MOD, X), \
1595 INSN_2(ALU64, NEG), \
0845c3db 1596 INSN_3(ALU64, END, TO_LE), \
5e581dad
DB		 1597 /* Immediate based. */ \
1598 INSN_3(ALU64, ADD, K), \
1599 INSN_3(ALU64, SUB, K), \
1600 INSN_3(ALU64, AND, K), \
1601 INSN_3(ALU64, OR, K), \
1602 INSN_3(ALU64, LSH, K), \
1603 INSN_3(ALU64, RSH, K), \
1604 INSN_3(ALU64, XOR, K), \
1605 INSN_3(ALU64, MUL, K), \
1606 INSN_3(ALU64, MOV, K), \
1607 INSN_3(ALU64, ARSH, K), \
1608 INSN_3(ALU64, DIV, K), \
1609 INSN_3(ALU64, MOD, K), \
1610 /* Call instruction. */ \
1611 INSN_2(JMP, CALL), \
1612 /* Exit instruction. */ \
1613 INSN_2(JMP, EXIT), \
503a8865
JW		 1614 /* 32-bit Jump instructions. */ \
1615 /* Register based. */ \
1616 INSN_3(JMP32, JEQ, X), \
1617 INSN_3(JMP32, JNE, X), \
1618 INSN_3(JMP32, JGT, X), \
1619 INSN_3(JMP32, JLT, X), \
1620 INSN_3(JMP32, JGE, X), \
1621 INSN_3(JMP32, JLE, X), \
1622 INSN_3(JMP32, JSGT, X), \
1623 INSN_3(JMP32, JSLT, X), \
1624 INSN_3(JMP32, JSGE, X), \
1625 INSN_3(JMP32, JSLE, X), \
1626 INSN_3(JMP32, JSET, X), \
1627 /* Immediate based. */ \
1628 INSN_3(JMP32, JEQ, K), \
1629 INSN_3(JMP32, JNE, K), \
1630 INSN_3(JMP32, JGT, K), \
1631 INSN_3(JMP32, JLT, K), \
1632 INSN_3(JMP32, JGE, K), \
1633 INSN_3(JMP32, JLE, K), \
1634 INSN_3(JMP32, JSGT, K), \
1635 INSN_3(JMP32, JSLT, K), \
1636 INSN_3(JMP32, JSGE, K), \
1637 INSN_3(JMP32, JSLE, K), \
1638 INSN_3(JMP32, JSET, K), \
5e581dad
DB		 1639 /* Jump instructions. */ \
1640 /* Register based. */ \
1641 INSN_3(JMP, JEQ, X), \
1642 INSN_3(JMP, JNE, X), \
1643 INSN_3(JMP, JGT, X), \
1644 INSN_3(JMP, JLT, X), \
1645 INSN_3(JMP, JGE, X), \
1646 INSN_3(JMP, JLE, X), \
1647 INSN_3(JMP, JSGT, X), \
1648 INSN_3(JMP, JSLT, X), \
1649 INSN_3(JMP, JSGE, X), \
1650 INSN_3(JMP, JSLE, X), \
1651 INSN_3(JMP, JSET, X), \
1652 /* Immediate based. */ \
1653 INSN_3(JMP, JEQ, K), \
1654 INSN_3(JMP, JNE, K), \
1655 INSN_3(JMP, JGT, K), \
1656 INSN_3(JMP, JLT, K), \
1657 INSN_3(JMP, JGE, K), \
1658 INSN_3(JMP, JLE, K), \
1659 INSN_3(JMP, JSGT, K), \
1660 INSN_3(JMP, JSLT, K), \
1661 INSN_3(JMP, JSGE, K), \
1662 INSN_3(JMP, JSLE, K), \
1663 INSN_3(JMP, JSET, K), \
1664 INSN_2(JMP, JA), \
4cd58e9a 1665 INSN_2(JMP32, JA), \
88044230
PY		 1666 /* Atomic operations. */ \
1667 INSN_3(STX, ATOMIC, B), \
1668 INSN_3(STX, ATOMIC, H), \
1669 INSN_3(STX, ATOMIC, W), \
1670 INSN_3(STX, ATOMIC, DW), \
5e581dad
DB		 1671 /* Store instructions. */ \
1672 /* Register based. */ \
1673 INSN_3(STX, MEM, B), \
1674 INSN_3(STX, MEM, H), \
1675 INSN_3(STX, MEM, W), \
1676 INSN_3(STX, MEM, DW), \
5e581dad
DB		 1677 /* Immediate based. */ \
1678 INSN_3(ST, MEM, B), \
1679 INSN_3(ST, MEM, H), \
1680 INSN_3(ST, MEM, W), \
1681 INSN_3(ST, MEM, DW), \
1682 /* Load instructions. */ \
1683 /* Register based. */ \
1684 INSN_3(LDX, MEM, B), \
1685 INSN_3(LDX, MEM, H), \
1686 INSN_3(LDX, MEM, W), \
1687 INSN_3(LDX, MEM, DW), \
1f9a1ea8
YS		 1688 INSN_3(LDX, MEMSX, B), \
1689 INSN_3(LDX, MEMSX, H), \
1690 INSN_3(LDX, MEMSX, W), \
5e581dad 1691 /* Immediate based. */ \
e0cea7ce 1692 INSN_3(LD, IMM, DW)
5e581dad
DB		 1693
1694bool bpf_opcode_in_insntable(u8 code)
1695{
1696#define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1697#define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1698 static const bool public_insntable[256] = {
1699 [0 ... 255] = false,
1700 /* Now overwrite non-defaults ... */
1701 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
e0cea7ce
DB		 1702 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1703 [BPF_LD | BPF_ABS | BPF_B] = true,
1704 [BPF_LD | BPF_ABS | BPF_H] = true,
1705 [BPF_LD | BPF_ABS | BPF_W] = true,
1706 [BPF_LD | BPF_IND | BPF_B] = true,
1707 [BPF_LD | BPF_IND | BPF_H] = true,
1708 [BPF_LD | BPF_IND | BPF_W] = true,
011832b9 1709 [BPF_JMP | BPF_JCOND] = true,
5e581dad
DB		 1710 };
1711#undef BPF_INSN_3_TBL
1712#undef BPF_INSN_2_TBL
1713 return public_insntable[code];
1714}
1715
290af866 1716#ifndef CONFIG_BPF_JIT_ALWAYS_ON
f5bffecd 1717/**
019d0454 1718 * ___bpf_prog_run - run eBPF program on a given context
de1da68d 1719 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
7ae457c1 1720 * @insn: is the array of eBPF instructions
f5bffecd 1721 *
7ae457c1 1722 * Decode and execute eBPF instructions.
019d0454
RD		 1723 *
1724 * Return: whatever value is in %BPF_R0 at program exit
f5bffecd 1725 */
2ec9898e 1726static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
f5bffecd 1727{
5e581dad
DB		 1728#define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1729#define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
e55a7325 1730 static const void * const jumptable[256] __annotate_jump_table = {
f5bffecd
AS		 1731 [0 ... 255] = &&default_label,
1732 /* Now overwrite non-defaults ... */
5e581dad
DB		 1733 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1734 /* Non-UAPI available opcodes. */
1ea47e01 1735 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
71189fa9 1736 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
f5e81d11 1737 [BPF_ST | BPF_NOSPEC] = &&ST_NOSPEC,
2a02759e
AS		 1738 [BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1739 [BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1740 [BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1741 [BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1f9a1ea8
YS		 1742 [BPF_LDX | BPF_PROBE_MEMSX | BPF_B] = &&LDX_PROBE_MEMSX_B,
1743 [BPF_LDX | BPF_PROBE_MEMSX | BPF_H] = &&LDX_PROBE_MEMSX_H,
1744 [BPF_LDX | BPF_PROBE_MEMSX | BPF_W] = &&LDX_PROBE_MEMSX_W,
f5bffecd 1745 };
5e581dad
DB		 1746#undef BPF_INSN_3_LBL
1747#undef BPF_INSN_2_LBL
04fd61ab 1748 u32 tail_call_cnt = 0;
f5bffecd
AS		 1749
1750#define CONT ({ insn++; goto select_insn; })
1751#define CONT_JMP ({ insn++; goto select_insn; })
1752
f5bffecd
AS		 1753select_insn:
1754 goto *jumptable[insn->code];
1755
28131e9d
DB		 1756 /* Explicitly mask the register-based shift amounts with 63 or 31
1757 * to avoid undefined behavior. Normally this won't affect the
1758 * generated code, for example, in case of native 64 bit archs such
1759 * as x86-64 or arm64, the compiler is optimizing the AND away for
1760 * the interpreter. In case of JITs, each of the JIT backends compiles
1761 * the BPF shift operations to machine instructions which produce
1762 * implementation-defined results in such a case; the resulting
1763 * contents of the register may be arbitrary, but program behaviour
1764 * as a whole remains defined. In other words, in case of JIT backends,
1765 * the AND must /not/ be added to the emitted LSH/RSH/ARSH translation.
1766 */
1767 /* ALU (shifts) */
1768#define SHT(OPCODE, OP) \
1769 ALU64_##OPCODE##_X: \
1770 DST = DST OP (SRC & 63); \
1771 CONT; \
1772 ALU_##OPCODE##_X: \
1773 DST = (u32) DST OP ((u32) SRC & 31); \
1774 CONT; \
1775 ALU64_##OPCODE##_K: \
1776 DST = DST OP IMM; \
1777 CONT; \
1778 ALU_##OPCODE##_K: \
1779 DST = (u32) DST OP (u32) IMM; \
1780 CONT;
1781 /* ALU (rest) */
1782#define ALU(OPCODE, OP) \
1783 ALU64_##OPCODE##_X: \
1784 DST = DST OP SRC; \
1785 CONT; \
1786 ALU_##OPCODE##_X: \
1787 DST = (u32) DST OP (u32) SRC; \
1788 CONT; \
1789 ALU64_##OPCODE##_K: \
1790 DST = DST OP IMM; \
1791 CONT; \
1792 ALU_##OPCODE##_K: \
1793 DST = (u32) DST OP (u32) IMM; \
f5bffecd 1794 CONT;
f5bffecd
AS		 1795 ALU(ADD, +)
1796 ALU(SUB, -)
1797 ALU(AND, &)
1798 ALU(OR, |)
f5bffecd
AS		 1799 ALU(XOR, ^)
1800 ALU(MUL, *)
28131e9d
DB		 1801 SHT(LSH, <<)
1802 SHT(RSH, >>)
1803#undef SHT
f5bffecd
AS		 1804#undef ALU
1805 ALU_NEG:
1806 DST = (u32) -DST;
1807 CONT;
1808 ALU64_NEG:
1809 DST = -DST;
1810 CONT;
1811 ALU_MOV_X:
8100928c
YS		 1812 switch (OFF) {
1813 case 0:
1814 DST = (u32) SRC;
1815 break;
1816 case 8:
1817 DST = (u32)(s8) SRC;
1818 break;
1819 case 16:
1820 DST = (u32)(s16) SRC;
1821 break;
1822 }
f5bffecd
AS		 1823 CONT;
1824 ALU_MOV_K:
1825 DST = (u32) IMM;
1826 CONT;
1827 ALU64_MOV_X:
8100928c
YS		 1828 switch (OFF) {
1829 case 0:
1830 DST = SRC;
1831 break;
1832 case 8:
1833 DST = (s8) SRC;
1834 break;
1835 case 16:
1836 DST = (s16) SRC;
1837 break;
1838 case 32:
1839 DST = (s32) SRC;
1840 break;
1841 }
f5bffecd
AS		 1842 CONT;
1843 ALU64_MOV_K:
1844 DST = IMM;
1845 CONT;
02ab695b
AS		 1846 LD_IMM_DW:
1847 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1848 insn++;
1849 CONT;
2dc6b100 1850 ALU_ARSH_X:
28131e9d 1851 DST = (u64) (u32) (((s32) DST) >> (SRC & 31));
2dc6b100
JW		 1852 CONT;
1853 ALU_ARSH_K:
75672dda 1854 DST = (u64) (u32) (((s32) DST) >> IMM);
2dc6b100 1855 CONT;
f5bffecd 1856 ALU64_ARSH_X:
28131e9d 1857 (*(s64 *) &DST) >>= (SRC & 63);
f5bffecd
AS		 1858 CONT;
1859 ALU64_ARSH_K:
1860 (*(s64 *) &DST) >>= IMM;
1861 CONT;
1862 ALU64_MOD_X:
ec0e2da9
YS		 1863 switch (OFF) {
1864 case 0:
1865 div64_u64_rem(DST, SRC, &AX);
1866 DST = AX;
1867 break;
1868 case 1:
1869 AX = div64_s64(DST, SRC);
1870 DST = DST - AX * SRC;
1871 break;
1872 }
f5bffecd
AS		 1873 CONT;
1874 ALU_MOD_X:
ec0e2da9
YS		 1875 switch (OFF) {
1876 case 0:
1877 AX = (u32) DST;
1878 DST = do_div(AX, (u32) SRC);
1879 break;
1880 case 1:
1881 AX = abs((s32)DST);
1882 AX = do_div(AX, abs((s32)SRC));
1883 if ((s32)DST < 0)
1884 DST = (u32)-AX;
1885 else
1886 DST = (u32)AX;
1887 break;
1888 }
f5bffecd
AS		 1889 CONT;
1890 ALU64_MOD_K:
ec0e2da9
YS		 1891 switch (OFF) {
1892 case 0:
1893 div64_u64_rem(DST, IMM, &AX);
1894 DST = AX;
1895 break;
1896 case 1:
1897 AX = div64_s64(DST, IMM);
1898 DST = DST - AX * IMM;
1899 break;
1900 }
f5bffecd
AS		 1901 CONT;
1902 ALU_MOD_K:
ec0e2da9
YS		 1903 switch (OFF) {
1904 case 0:
1905 AX = (u32) DST;
1906 DST = do_div(AX, (u32) IMM);
1907 break;
1908 case 1:
1909 AX = abs((s32)DST);
1910 AX = do_div(AX, abs((s32)IMM));
1911 if ((s32)DST < 0)
1912 DST = (u32)-AX;
1913 else
1914 DST = (u32)AX;
1915 break;
1916 }
f5bffecd
AS		 1917 CONT;
1918 ALU64_DIV_X:
ec0e2da9
YS		 1919 switch (OFF) {
1920 case 0:
1921 DST = div64_u64(DST, SRC);
1922 break;
1923 case 1:
1924 DST = div64_s64(DST, SRC);
1925 break;
1926 }
f5bffecd
AS		 1927 CONT;
1928 ALU_DIV_X:
ec0e2da9
YS		 1929 switch (OFF) {
1930 case 0:
1931 AX = (u32) DST;
1932 do_div(AX, (u32) SRC);
1933 DST = (u32) AX;
1934 break;
1935 case 1:
1936 AX = abs((s32)DST);
1937 do_div(AX, abs((s32)SRC));
09fedc73 1938 if (((s32)DST < 0) == ((s32)SRC < 0))
ec0e2da9
YS		 1939 DST = (u32)AX;
1940 else
1941 DST = (u32)-AX;
1942 break;
1943 }
f5bffecd
AS		 1944 CONT;
1945 ALU64_DIV_K:
ec0e2da9
YS		 1946 switch (OFF) {
1947 case 0:
1948 DST = div64_u64(DST, IMM);
1949 break;
1950 case 1:
1951 DST = div64_s64(DST, IMM);
1952 break;
1953 }
f5bffecd
AS		 1954 CONT;
1955 ALU_DIV_K:
ec0e2da9
YS		 1956 switch (OFF) {
1957 case 0:
1958 AX = (u32) DST;
1959 do_div(AX, (u32) IMM);
1960 DST = (u32) AX;
1961 break;
1962 case 1:
1963 AX = abs((s32)DST);
1964 do_div(AX, abs((s32)IMM));
09fedc73 1965 if (((s32)DST < 0) == ((s32)IMM < 0))
ec0e2da9
YS		 1966 DST = (u32)AX;
1967 else
1968 DST = (u32)-AX;
1969 break;
1970 }
f5bffecd
AS		 1971 CONT;
1972 ALU_END_TO_BE:
1973 switch (IMM) {
1974 case 16:
1975 DST = (__force u16) cpu_to_be16(DST);
1976 break;
1977 case 32:
1978 DST = (__force u32) cpu_to_be32(DST);
1979 break;
1980 case 64:
1981 DST = (__force u64) cpu_to_be64(DST);
1982 break;
1983 }
1984 CONT;
1985 ALU_END_TO_LE:
1986 switch (IMM) {
1987 case 16:
1988 DST = (__force u16) cpu_to_le16(DST);
1989 break;
1990 case 32:
1991 DST = (__force u32) cpu_to_le32(DST);
1992 break;
1993 case 64:
1994 DST = (__force u64) cpu_to_le64(DST);
1995 break;
1996 }
1997 CONT;
0845c3db
YS		 1998 ALU64_END_TO_LE:
1999 switch (IMM) {
2000 case 16:
2001 DST = (__force u16) __swab16(DST);
2002 break;
2003 case 32:
2004 DST = (__force u32) __swab32(DST);
2005 break;
2006 case 64:
2007 DST = (__force u64) __swab64(DST);
2008 break;
2009 }
2010 CONT;
f5bffecd
AS		 2011
2012 /* CALL */
2013 JMP_CALL:
2014 /* Function call scratches BPF_R1-BPF_R5 registers,
2015 * preserves BPF_R6-BPF_R9, and stores return value
2016 * into BPF_R0.
2017 */
2018 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
2019 BPF_R4, BPF_R5);
2020 CONT;
2021
1ea47e01
AS		 2022 JMP_CALL_ARGS:
2023 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
2024 BPF_R3, BPF_R4,
2025 BPF_R5,
2026 insn + insn->off + 1);
2027 CONT;
2028
04fd61ab
AS		 2029 JMP_TAIL_CALL: {
2030 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
2031 struct bpf_array *array = container_of(map, struct bpf_array, map);
2032 struct bpf_prog *prog;
90caccdd 2033 u32 index = BPF_R3;
04fd61ab
AS		 2034
2035 if (unlikely(index >= array->map.max_entries))
2036 goto out;
ebf7f6f0
TY		 2037
2038 if (unlikely(tail_call_cnt >= MAX_TAIL_CALL_CNT))
04fd61ab
AS		 2039 goto out;
2040
2041 tail_call_cnt++;
2042
2a36f0b9 2043 prog = READ_ONCE(array->ptrs[index]);
1ca1cc98 2044 if (!prog)
04fd61ab
AS		 2045 goto out;
2046
c4675f93
DB		 2047 /* ARG1 at this point is guaranteed to point to CTX from
2048 * the verifier side due to the fact that the tail call is
0142dddc 2049 * handled like a helper, that is, bpf_tail_call_proto,
c4675f93
DB		 2050 * where arg1_type is ARG_PTR_TO_CTX.
2051 */
04fd61ab
AS		 2052 insn = prog->insnsi;
2053 goto select_insn;
2054out:
2055 CONT;
2056 }
f5bffecd
AS		 2057 JMP_JA:
2058 insn += insn->off;
2059 CONT;
4cd58e9a
YS		 2060 JMP32_JA:
2061 insn += insn->imm;
2062 CONT;
f5bffecd
AS		 2063 JMP_EXIT:
2064 return BPF_R0;
503a8865
JW		 2065 /* JMP */
2066#define COND_JMP(SIGN, OPCODE, CMP_OP) \
2067 JMP_##OPCODE##_X: \
2068 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \
2069 insn += insn->off; \
2070 CONT_JMP; \
2071 } \
2072 CONT; \
2073 JMP32_##OPCODE##_X: \
2074 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \
2075 insn += insn->off; \
2076 CONT_JMP; \
2077 } \
2078 CONT; \
2079 JMP_##OPCODE##_K: \
2080 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \
2081 insn += insn->off; \
2082 CONT_JMP; \
2083 } \
2084 CONT; \
2085 JMP32_##OPCODE##_K: \
2086 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \
2087 insn += insn->off; \
2088 CONT_JMP; \
2089 } \
2090 CONT;
2091 COND_JMP(u, JEQ, ==)
2092 COND_JMP(u, JNE, !=)
2093 COND_JMP(u, JGT, >)
2094 COND_JMP(u, JLT, <)
2095 COND_JMP(u, JGE, >=)
2096 COND_JMP(u, JLE, <=)
2097 COND_JMP(u, JSET, &)
2098 COND_JMP(s, JSGT, >)
2099 COND_JMP(s, JSLT, <)
2100 COND_JMP(s, JSGE, >=)
2101 COND_JMP(s, JSLE, <=)
2102#undef COND_JMP
f5e81d11
DB		 2103 /* ST, STX and LDX*/
2104 ST_NOSPEC:
2105 /* Speculation barrier for mitigating Speculative Store Bypass.
2106 * In case of arm64, we rely on the firmware mitigation as
2107 * controlled via the ssbd kernel parameter. Whenever the
2108 * mitigation is enabled, it works for all of the kernel code
2109 * with no need to provide any additional instructions here.
2110 * In case of x86, we use 'lfence' insn for mitigation. We
2111 * reuse preexisting logic from Spectre v1 mitigation that
2112 * happens to produce the required code on x86 for v4 as well.
2113 */
f5e81d11 2114 barrier_nospec();
f5e81d11 2115 CONT;
f5bffecd
AS		 2116#define LDST(SIZEOP, SIZE) \
2117 STX_MEM_##SIZEOP: \
2118 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
2119 CONT; \
2120 ST_MEM_##SIZEOP: \
2121 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
2122 CONT; \
2123 LDX_MEM_##SIZEOP: \
2124 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
caff1fa4
MD		 2125 CONT; \
2126 LDX_PROBE_MEM_##SIZEOP: \
6a5a148a
AB		 2127 bpf_probe_read_kernel_common(&DST, sizeof(SIZE), \
2128 (const void *)(long) (SRC + insn->off)); \
caff1fa4 2129 DST = *((SIZE *)&DST); \
f5bffecd
AS		 2130 CONT;
2131
2132 LDST(B, u8)
2133 LDST(H, u16)
2134 LDST(W, u32)
2135 LDST(DW, u64)
2136#undef LDST
2a02759e 2137
1f9a1ea8
YS		 2138#define LDSX(SIZEOP, SIZE) \
2139 LDX_MEMSX_##SIZEOP: \
2140 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
2141 CONT; \
2142 LDX_PROBE_MEMSX_##SIZEOP: \
6a5a148a 2143 bpf_probe_read_kernel_common(&DST, sizeof(SIZE), \
1f9a1ea8
YS		 2144 (const void *)(long) (SRC + insn->off)); \
2145 DST = *((SIZE *)&DST); \
2146 CONT;
2147
2148 LDSX(B, s8)
2149 LDSX(H, s16)
2150 LDSX(W, s32)
2151#undef LDSX
2152
46291067
BJ		 2153#define ATOMIC_ALU_OP(BOP, KOP) \
2154 case BOP: \
2155 if (BPF_SIZE(insn->code) == BPF_W) \
2156 atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
2157 (DST + insn->off)); \
88044230 2158 else if (BPF_SIZE(insn->code) == BPF_DW) \
46291067
BJ		 2159 atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
2160 (DST + insn->off)); \
88044230
PY		 2161 else \
2162 goto default_label; \
46291067
BJ		 2163 break; \
2164 case BOP | BPF_FETCH: \
2165 if (BPF_SIZE(insn->code) == BPF_W) \
2166 SRC = (u32) atomic_fetch_##KOP( \
2167 (u32) SRC, \
2168 (atomic_t *)(unsigned long) (DST + insn->off)); \
88044230 2169 else if (BPF_SIZE(insn->code) == BPF_DW) \
46291067
BJ		 2170 SRC = (u64) atomic64_fetch_##KOP( \
2171 (u64) SRC, \
2172 (atomic64_t *)(unsigned long) (DST + insn->off)); \
88044230
PY		 2173 else \
2174 goto default_label; \
5ffa2550 2175 break;
5ca419f2 2176
91c960b0 2177 STX_ATOMIC_DW:
46291067 2178 STX_ATOMIC_W:
88044230
PY		 2179 STX_ATOMIC_H:
2180 STX_ATOMIC_B:
91c960b0 2181 switch (IMM) {
88044230
PY		 2182 /* Atomic read-modify-write instructions support only W and DW
2183 * size modifiers.
2184 */
46291067 2185 ATOMIC_ALU_OP(BPF_ADD, add)
981f94c3
BJ		 2186 ATOMIC_ALU_OP(BPF_AND, and)
2187 ATOMIC_ALU_OP(BPF_OR, or)
2188 ATOMIC_ALU_OP(BPF_XOR, xor)
46291067
BJ		 2189#undef ATOMIC_ALU_OP
2190
5ffa2550 2191 case BPF_XCHG:
46291067
BJ		 2192 if (BPF_SIZE(insn->code) == BPF_W)
2193 SRC = (u32) atomic_xchg(
2194 (atomic_t *)(unsigned long) (DST + insn->off),
2195 (u32) SRC);
88044230 2196 else if (BPF_SIZE(insn->code) == BPF_DW)
46291067
BJ		 2197 SRC = (u64) atomic64_xchg(
2198 (atomic64_t *)(unsigned long) (DST + insn->off),
2199 (u64) SRC);
88044230
PY		 2200 else
2201 goto default_label;
5ffa2550
BJ		 2202 break;
2203 case BPF_CMPXCHG:
46291067
BJ		 2204 if (BPF_SIZE(insn->code) == BPF_W)
2205 BPF_R0 = (u32) atomic_cmpxchg(
2206 (atomic_t *)(unsigned long) (DST + insn->off),
2207 (u32) BPF_R0, (u32) SRC);
88044230 2208 else if (BPF_SIZE(insn->code) == BPF_DW)
46291067
BJ		 2209 BPF_R0 = (u64) atomic64_cmpxchg(
2210 (atomic64_t *)(unsigned long) (DST + insn->off),
2211 (u64) BPF_R0, (u64) SRC);
88044230
PY		 2212 else
2213 goto default_label;
2214 break;
2215 /* Atomic load and store instructions support all size
2216 * modifiers.
2217 */
2218 case BPF_LOAD_ACQ:
2219 switch (BPF_SIZE(insn->code)) {
2220#define LOAD_ACQUIRE(SIZEOP, SIZE) \
2221 case BPF_##SIZEOP: \
2222 DST = (SIZE)smp_load_acquire( \
2223 (SIZE *)(unsigned long)(SRC + insn->off)); \
2224 break;
2225 LOAD_ACQUIRE(B, u8)
2226 LOAD_ACQUIRE(H, u16)
2227 LOAD_ACQUIRE(W, u32)
2228#ifdef CONFIG_64BIT
2229 LOAD_ACQUIRE(DW, u64)
2230#endif
2231#undef LOAD_ACQUIRE
2232 default:
2233 goto default_label;
2234 }
2235 break;
2236 case BPF_STORE_REL:
2237 switch (BPF_SIZE(insn->code)) {
2238#define STORE_RELEASE(SIZEOP, SIZE) \
2239 case BPF_##SIZEOP: \
2240 smp_store_release( \
2241 (SIZE *)(unsigned long)(DST + insn->off), (SIZE)SRC); \
2242 break;
2243 STORE_RELEASE(B, u8)
2244 STORE_RELEASE(H, u16)
2245 STORE_RELEASE(W, u32)
2246#ifdef CONFIG_64BIT
2247 STORE_RELEASE(DW, u64)
2248#endif
2249#undef STORE_RELEASE
2250 default:
2251 goto default_label;
2252 }
5ffa2550 2253 break;
46291067 2254
91c960b0
BJ		 2255 default:
2256 goto default_label;
2257 }
f5bffecd 2258 CONT;
f5bffecd
AS		 2259
2260 default_label:
5e581dad
DB		 2261 /* If we ever reach this, we have a bug somewhere. Die hard here
2262 * instead of just returning 0; we could be somewhere in a subprog,
2263 * so execution could continue otherwise which we do /not/ want.
2264 *
2265 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
2266 */
91c960b0
BJ		 2267 pr_warn("BPF interpreter: unknown opcode %02x (imm: 0x%x)\n",
2268 insn->code, insn->imm);
5e581dad 2269 BUG_ON(1);
f5bffecd
AS		 2270 return 0;
2271}
f696b8f4 2272
b870aa90
AS		 2273#define PROG_NAME(stack_size) __bpf_prog_run##stack_size
2274#define DEFINE_BPF_PROG_RUN(stack_size) \
2275static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
2276{ \
2277 u64 stack[stack_size / sizeof(u64)]; \
a6a7aaba 2278 u64 regs[MAX_BPF_EXT_REG] = {}; \
b870aa90 2279\
e8742081 2280 kmsan_unpoison_memory(stack, sizeof(stack)); \
b870aa90
AS		 2281 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2282 ARG1 = (u64) (unsigned long) ctx; \
2ec9898e 2283 return ___bpf_prog_run(regs, insn); \
f696b8f4 2284}
f5bffecd 2285
1ea47e01
AS		 2286#define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
2287#define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
2288static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
2289 const struct bpf_insn *insn) \
2290{ \
2291 u64 stack[stack_size / sizeof(u64)]; \
144cd91c 2292 u64 regs[MAX_BPF_EXT_REG]; \
1ea47e01 2293\
e8742081 2294 kmsan_unpoison_memory(stack, sizeof(stack)); \
1ea47e01
AS		 2295 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2296 BPF_R1 = r1; \
2297 BPF_R2 = r2; \
2298 BPF_R3 = r3; \
2299 BPF_R4 = r4; \
2300 BPF_R5 = r5; \
2ec9898e 2301 return ___bpf_prog_run(regs, insn); \
1ea47e01
AS		 2302}
2303
b870aa90
AS		 2304#define EVAL1(FN, X) FN(X)
2305#define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
2306#define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
2307#define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
2308#define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
2309#define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
2310
2311EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
2312EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
2313EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
2314
1ea47e01
AS		 2315EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
2316EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
2317EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
2318
b870aa90
AS		 2319#define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
2320
2321static unsigned int (*interpreters[])(const void *ctx,
2322 const struct bpf_insn *insn) = {
2323EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2324EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2325EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2326};
1ea47e01
AS		 2327#undef PROG_NAME_LIST
2328#define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
ba49f976
AB		 2329static __maybe_unused
2330u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
2331 const struct bpf_insn *insn) = {
1ea47e01
AS		 2332EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2333EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2334EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2335};
2336#undef PROG_NAME_LIST
2337
ba49f976 2338#ifdef CONFIG_BPF_SYSCALL
1ea47e01
AS		 2339void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
2340{
2341 stack_depth = max_t(u32, stack_depth, 1);
2342 insn->off = (s16) insn->imm;
2343 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
2344 __bpf_call_base_args;
2345 insn->code = BPF_JMP | BPF_CALL_ARGS;
2346}
ba49f976 2347#endif
6ebc5030
JC		 2348#endif
2349
fa9dd599
DB		 2350static unsigned int __bpf_prog_ret0_warn(const void *ctx,
2351 const struct bpf_insn *insn)
290af866 2352{
fa9dd599 2353 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
6ebc5030
JC		 2354 * is not working properly, or interpreter is being used when
2355 * prog->jit_requested is not 0, so warn about it!
fa9dd599
DB		 2356 */
2357 WARN_ON_ONCE(1);
290af866
AS		 2358 return 0;
2359}
290af866 2360
714070c4
LB		 2361static bool __bpf_prog_map_compatible(struct bpf_map *map,
2362 const struct bpf_prog *fp)
04fd61ab 2363{
1c123c56 2364 enum bpf_prog_type prog_type = resolve_prog_type(fp);
54713c85 2365 bool ret;
28ead3ea 2366 struct bpf_prog_aux *aux = fp->aux;
54713c85 2367
9802d865
JB		 2368 if (fp->kprobe_override)
2369 return false;
2370
f45d5b6c
THJ		 2371 spin_lock(&map->owner.lock);
2372 if (!map->owner.type) {
3324b584
DB		 2373 /* There's no owner yet where we could check for
2374 * compatibility.
2375 */
1c123c56 2376 map->owner.type = prog_type;
f45d5b6c 2377 map->owner.jited = fp->jited;
28ead3ea
XK		 2378 map->owner.xdp_has_frags = aux->xdp_has_frags;
2379 map->owner.attach_func_proto = aux->attach_func_proto;
54713c85
THJ		 2380 ret = true;
2381 } else {
1c123c56 2382 ret = map->owner.type == prog_type &&
f45d5b6c 2383 map->owner.jited == fp->jited &&
28ead3ea
XK		 2384 map->owner.xdp_has_frags == aux->xdp_has_frags;
2385 if (ret &&
2386 map->owner.attach_func_proto != aux->attach_func_proto) {
2387 switch (prog_type) {
2388 case BPF_PROG_TYPE_TRACING:
2389 case BPF_PROG_TYPE_LSM:
2390 case BPF_PROG_TYPE_EXT:
2391 case BPF_PROG_TYPE_STRUCT_OPS:
2392 ret = false;
2393 break;
2394 default:
2395 break;
2396 }
2397 }
04fd61ab 2398 }
f45d5b6c
THJ		 2399 spin_unlock(&map->owner.lock);
2400
54713c85 2401 return ret;
04fd61ab
AS		 2402}
2403
714070c4
LB		 2404bool bpf_prog_map_compatible(struct bpf_map *map, const struct bpf_prog *fp)
2405{
2406 /* XDP programs inserted into maps are not guaranteed to run on
2407 * a particular netdev (and can run outside driver context entirely
2408 * in the case of devmap and cpumap). Until device checks
2409 * are implemented, prohibit adding dev-bound programs to program maps.
2410 */
2411 if (bpf_prog_is_dev_bound(fp->aux))
2412 return false;
2413
2414 return __bpf_prog_map_compatible(map, fp);
2415}
2416
3324b584 2417static int bpf_check_tail_call(const struct bpf_prog *fp)
04fd61ab
AS		 2418{
2419 struct bpf_prog_aux *aux = fp->aux;
984fe94f 2420 int i, ret = 0;
04fd61ab 2421
984fe94f 2422 mutex_lock(&aux->used_maps_mutex);
04fd61ab 2423 for (i = 0; i < aux->used_map_cnt; i++) {
3324b584 2424 struct bpf_map *map = aux->used_maps[i];
04fd61ab 2425
f45d5b6c 2426 if (!map_type_contains_progs(map))
04fd61ab 2427 continue;
3324b584 2428
714070c4 2429 if (!__bpf_prog_map_compatible(map, fp)) {
984fe94f
YZ		 2430 ret = -EINVAL;
2431 goto out;
2432 }
04fd61ab
AS		 2433 }
2434
984fe94f
YZ		 2435out:
2436 mutex_unlock(&aux->used_maps_mutex);
2437 return ret;
04fd61ab
AS		 2438}
2439
9facc336
DB		 2440static void bpf_prog_select_func(struct bpf_prog *fp)
2441{
2442#ifndef CONFIG_BPF_JIT_ALWAYS_ON
2443 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
6ebc5030 2444 u32 idx = (round_up(stack_depth, 32) / 32) - 1;
9facc336 2445
6ebc5030
JC		 2446 /* may_goto may cause stack size > 512, leading to idx out-of-bounds.
2447 * But for non-JITed programs, we don't need bpf_func, so no bounds
2448 * check needed.
2449 */
2450 if (!fp->jit_requested &&
2451 !WARN_ON_ONCE(idx >= ARRAY_SIZE(interpreters))) {
2452 fp->bpf_func = interpreters[idx];
2453 } else {
2454 fp->bpf_func = __bpf_prog_ret0_warn;
2455 }
9facc336
DB		 2456#else
2457 fp->bpf_func = __bpf_prog_ret0_warn;
2458#endif
2459}
2460
f5bffecd 2461/**
3324b584 2462 * bpf_prog_select_runtime - select exec runtime for BPF program
06edc59c 2463 * @fp: bpf_prog populated with BPF program
d1c55ab5 2464 * @err: pointer to error variable
f5bffecd 2465 *
3324b584 2466 * Try to JIT eBPF program, if JIT is not available, use interpreter.
fb7dd8bc 2467 * The BPF program will be executed via bpf_prog_run() function.
019d0454
RD		 2468 *
2469 * Return: the &fp argument along with &err set to 0 for success or
2470 * a negative errno code on failure
f5bffecd 2471 */
d1c55ab5 2472struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
f5bffecd 2473{
9facc336
DB		 2474 /* In case of BPF to BPF calls, verifier did all the prep
2475 * work with regards to JITing, etc.
2476 */
86bc9c74 2477 bool jit_needed = fp->jit_requested;
e6ac2450 2478
9facc336
DB		 2479 if (fp->bpf_func)
2480 goto finalize;
8007e40a 2481
e6ac2450
MKL		 2482 if (IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) ||
2483 bpf_prog_has_kfunc_call(fp))
2484 jit_needed = true;
2485
9facc336 2486 bpf_prog_select_func(fp);
f5bffecd 2487
d1c55ab5
DB		 2488 /* eBPF JITs can rewrite the program in case constant
2489 * blinding is active. However, in case of error during
2490 * blinding, bpf_int_jit_compile() must always return a
2491 * valid program, which in this case would simply not
2492 * be JITed, but falls back to the interpreter.
2493 */
9d03ebc7 2494 if (!bpf_prog_is_offloaded(fp->aux)) {
c454a46b
MKL		 2495 *err = bpf_prog_alloc_jited_linfo(fp);
2496 if (*err)
2497 return fp;
2498
ab3f0063 2499 fp = bpf_int_jit_compile(fp);
e16301fb 2500 bpf_prog_jit_attempt_done(fp);
e6ac2450 2501 if (!fp->jited && jit_needed) {
290af866
AS		 2502 *err = -ENOTSUPP;
2503 return fp;
c454a46b 2504 }
ab3f0063
JK		 2505 } else {
2506 *err = bpf_prog_offload_compile(fp);
2507 if (*err)
2508 return fp;
2509 }
9facc336
DB		 2510
2511finalize:
7d2cc63e
CL		 2512 *err = bpf_prog_lock_ro(fp);
2513 if (*err)
2514 return fp;
04fd61ab 2515
3324b584
DB		 2516 /* The tail call compatibility check can only be done at
2517 * this late stage as we need to determine, if we deal
2518 * with JITed or non JITed program concatenations and not
2519 * all eBPF JITs might immediately support all features.
2520 */
d1c55ab5 2521 *err = bpf_check_tail_call(fp);
85782e03 2522
d1c55ab5 2523 return fp;
f5bffecd 2524}
7ae457c1 2525EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
f5bffecd 2526
e87c6bc3
YS		 2527static unsigned int __bpf_prog_ret1(const void *ctx,
2528 const struct bpf_insn *insn)
2529{
2530 return 1;
2531}
2532
2533static struct bpf_prog_dummy {
2534 struct bpf_prog prog;
2535} dummy_bpf_prog = {
2536 .prog = {
2537 .bpf_func = __bpf_prog_ret1,
2538 },
2539};
2540
46531a30 2541struct bpf_empty_prog_array bpf_empty_prog_array = {
324bda9e
AS		 2542 .null_prog = NULL,
2543};
46531a30 2544EXPORT_SYMBOL(bpf_empty_prog_array);
324bda9e 2545
d29ab6e1 2546struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
324bda9e 2547{
cb01621b
AS		 2548 struct bpf_prog_array *p;
2549
324bda9e 2550 if (prog_cnt)
cb01621b
AS		 2551 p = kzalloc(struct_size(p, items, prog_cnt + 1), flags);
2552 else
2553 p = &bpf_empty_prog_array.hdr;
324bda9e 2554
cb01621b 2555 return p;
324bda9e
AS		 2556}
2557
54e9c9d4 2558void bpf_prog_array_free(struct bpf_prog_array *progs)
324bda9e 2559{
46531a30 2560 if (!progs || progs == &bpf_empty_prog_array.hdr)
324bda9e
AS		 2561 return;
2562 kfree_rcu(progs, rcu);
2563}
2564
8c7dcb84
DK		 2565static void __bpf_prog_array_free_sleepable_cb(struct rcu_head *rcu)
2566{
2567 struct bpf_prog_array *progs;
2568
4835f9ee
HT		 2569 /* If RCU Tasks Trace grace period implies RCU grace period, there is
2570 * no need to call kfree_rcu(), just call kfree() directly.
2571 */
8c7dcb84 2572 progs = container_of(rcu, struct bpf_prog_array, rcu);
4835f9ee
HT		 2573 if (rcu_trace_implies_rcu_gp())
2574 kfree(progs);
2575 else
2576 kfree_rcu(progs, rcu);
8c7dcb84
DK		 2577}
2578
2579void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs)
2580{
2581 if (!progs || progs == &bpf_empty_prog_array.hdr)
2582 return;
2583 call_rcu_tasks_trace(&progs->rcu, __bpf_prog_array_free_sleepable_cb);
2584}
2585
54e9c9d4 2586int bpf_prog_array_length(struct bpf_prog_array *array)
468e2f64 2587{
394e40a2 2588 struct bpf_prog_array_item *item;
468e2f64
AS		 2589 u32 cnt = 0;
2590
54e9c9d4 2591 for (item = array->items; item->prog; item++)
394e40a2 2592 if (item->prog != &dummy_bpf_prog.prog)
c8c088ba 2593 cnt++;
468e2f64
AS		 2594 return cnt;
2595}
2596
0d01da6a
SF		 2597bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
2598{
2599 struct bpf_prog_array_item *item;
2600
2601 for (item = array->items; item->prog; item++)
2602 if (item->prog != &dummy_bpf_prog.prog)
2603 return false;
2604 return true;
2605}
394e40a2 2606
54e9c9d4 2607static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
3a38bb98
YS		 2608 u32 *prog_ids,
2609 u32 request_cnt)
2610{
394e40a2 2611 struct bpf_prog_array_item *item;
3a38bb98
YS		 2612 int i = 0;
2613
54e9c9d4 2614 for (item = array->items; item->prog; item++) {
394e40a2 2615 if (item->prog == &dummy_bpf_prog.prog)
3a38bb98 2616 continue;
394e40a2 2617 prog_ids[i] = item->prog->aux->id;
3a38bb98 2618 if (++i == request_cnt) {
394e40a2 2619 item++;
3a38bb98
YS		 2620 break;
2621 }
2622 }
2623
394e40a2 2624 return !!(item->prog);
3a38bb98
YS		 2625}
2626
54e9c9d4 2627int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
468e2f64
AS		 2628 __u32 __user *prog_ids, u32 cnt)
2629{
0911287c 2630 unsigned long err = 0;
0911287c 2631 bool nospc;
3a38bb98 2632 u32 *ids;
0911287c
AS		 2633
2634 /* users of this function are doing:
2635 * cnt = bpf_prog_array_length();
2636 * if (cnt > 0)
2637 * bpf_prog_array_copy_to_user(..., cnt);
54e9c9d4 2638 * so below kcalloc doesn't need extra cnt > 0 check.
0911287c 2639 */
9c481b90 2640 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
0911287c
AS		 2641 if (!ids)
2642 return -ENOMEM;
394e40a2 2643 nospc = bpf_prog_array_copy_core(array, ids, cnt);
0911287c
AS		 2644 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
2645 kfree(ids);
2646 if (err)
2647 return -EFAULT;
2648 if (nospc)
468e2f64
AS		 2649 return -ENOSPC;
2650 return 0;
2651}
2652
54e9c9d4 2653void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
e87c6bc3
YS		 2654 struct bpf_prog *old_prog)
2655{
54e9c9d4 2656 struct bpf_prog_array_item *item;
e87c6bc3 2657
54e9c9d4 2658 for (item = array->items; item->prog; item++)
394e40a2
RG		 2659 if (item->prog == old_prog) {
2660 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
e87c6bc3
YS		 2661 break;
2662 }
2663}
2664
ce3aa9cc
JS		 2665/**
2666 * bpf_prog_array_delete_safe_at() - Replaces the program at the given
2667 * index into the program array with
2668 * a dummy no-op program.
2669 * @array: a bpf_prog_array
2670 * @index: the index of the program to replace
2671 *
2672 * Skips over dummy programs, by not counting them, when calculating
b8c1a309 2673 * the position of the program to replace.
ce3aa9cc
JS		 2674 *
2675 * Return:
2676 * * 0 - Success
2677 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2678 * * -ENOENT - Index out of range
2679 */
2680int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
2681{
2682 return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
2683}
2684
2685/**
2686 * bpf_prog_array_update_at() - Updates the program at the given index
2687 * into the program array.
2688 * @array: a bpf_prog_array
2689 * @index: the index of the program to update
2690 * @prog: the program to insert into the array
2691 *
2692 * Skips over dummy programs, by not counting them, when calculating
2693 * the position of the program to update.
2694 *
2695 * Return:
2696 * * 0 - Success
2697 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2698 * * -ENOENT - Index out of range
2699 */
2700int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2701 struct bpf_prog *prog)
2702{
2703 struct bpf_prog_array_item *item;
2704
2705 if (unlikely(index < 0))
2706 return -EINVAL;
2707
2708 for (item = array->items; item->prog; item++) {
2709 if (item->prog == &dummy_bpf_prog.prog)
2710 continue;
2711 if (!index) {
2712 WRITE_ONCE(item->prog, prog);
2713 return 0;
2714 }
2715 index--;
2716 }
2717 return -ENOENT;
2718}
2719
54e9c9d4 2720int bpf_prog_array_copy(struct bpf_prog_array *old_array,
e87c6bc3
YS		 2721 struct bpf_prog *exclude_prog,
2722 struct bpf_prog *include_prog,
82e6b1ee 2723 u64 bpf_cookie,
e87c6bc3
YS		 2724 struct bpf_prog_array **new_array)
2725{
2726 int new_prog_cnt, carry_prog_cnt = 0;
82e6b1ee 2727 struct bpf_prog_array_item *existing, *new;
e87c6bc3 2728 struct bpf_prog_array *array;
170a7e3e 2729 bool found_exclude = false;
e87c6bc3
YS		 2730
2731 /* Figure out how many existing progs we need to carry over to
2732 * the new array.
2733 */
2734 if (old_array) {
394e40a2
RG		 2735 existing = old_array->items;
2736 for (; existing->prog; existing++) {
2737 if (existing->prog == exclude_prog) {
170a7e3e
SY		 2738 found_exclude = true;
2739 continue;
2740 }
394e40a2 2741 if (existing->prog != &dummy_bpf_prog.prog)
e87c6bc3 2742 carry_prog_cnt++;
394e40a2 2743 if (existing->prog == include_prog)
e87c6bc3
YS		 2744 return -EEXIST;
2745 }
2746 }
2747
170a7e3e
SY		 2748 if (exclude_prog && !found_exclude)
2749 return -ENOENT;
2750
e87c6bc3
YS		 2751 /* How many progs (not NULL) will be in the new array? */
2752 new_prog_cnt = carry_prog_cnt;
2753 if (include_prog)
2754 new_prog_cnt += 1;
2755
2756 /* Do we have any prog (not NULL) in the new array? */
2757 if (!new_prog_cnt) {
2758 *new_array = NULL;
2759 return 0;
2760 }
2761
2762 /* +1 as the end of prog_array is marked with NULL */
2763 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2764 if (!array)
2765 return -ENOMEM;
82e6b1ee 2766 new = array->items;
e87c6bc3
YS		 2767
2768 /* Fill in the new prog array */
2769 if (carry_prog_cnt) {
394e40a2 2770 existing = old_array->items;
82e6b1ee
AN		 2771 for (; existing->prog; existing++) {
2772 if (existing->prog == exclude_prog ||
2773 existing->prog == &dummy_bpf_prog.prog)
2774 continue;
2775
2776 new->prog = existing->prog;
2777 new->bpf_cookie = existing->bpf_cookie;
2778 new++;
2779 }
e87c6bc3 2780 }
82e6b1ee
AN		 2781 if (include_prog) {
2782 new->prog = include_prog;
2783 new->bpf_cookie = bpf_cookie;
2784 new++;
2785 }
2786 new->prog = NULL;
e87c6bc3
YS		 2787 *new_array = array;
2788 return 0;
2789}
2790
54e9c9d4 2791int bpf_prog_array_copy_info(struct bpf_prog_array *array,
3a38bb98
YS		 2792 u32 *prog_ids, u32 request_cnt,
2793 u32 *prog_cnt)
f371b304
YS		 2794{
2795 u32 cnt = 0;
2796
2797 if (array)
2798 cnt = bpf_prog_array_length(array);
2799
3a38bb98 2800 *prog_cnt = cnt;
f371b304
YS		 2801
2802 /* return early if user requested only program count or nothing to copy */
2803 if (!request_cnt || !cnt)
2804 return 0;
2805
3a38bb98 2806 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
394e40a2 2807 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
3a38bb98 2808 : 0;
f371b304
YS		 2809}
2810
a2ea0746
DB		 2811void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2812 struct bpf_map **used_maps, u32 len)
6332be04 2813{
da765a2f 2814 struct bpf_map *map;
af66bfd3 2815 bool sleepable;
a2ea0746 2816 u32 i;
6332be04 2817
66c84731 2818 sleepable = aux->prog->sleepable;
a2ea0746
DB		 2819 for (i = 0; i < len; i++) {
2820 map = used_maps[i];
da765a2f
DB		 2821 if (map->ops->map_poke_untrack)
2822 map->ops->map_poke_untrack(map, aux);
af66bfd3
HT		 2823 if (sleepable)
2824 atomic64_dec(&map->sleepable_refcnt);
da765a2f
DB		 2825 bpf_map_put(map);
2826 }
a2ea0746
DB		 2827}
2828
2829static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2830{
2831 __bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
6332be04
DB		 2832 kfree(aux->used_maps);
2833}
2834
ab224b9e 2835void __bpf_free_used_btfs(struct btf_mod_pair *used_btfs, u32 len)
541c3bad
AN		 2836{
2837#ifdef CONFIG_BPF_SYSCALL
2838 struct btf_mod_pair *btf_mod;
2839 u32 i;
2840
2841 for (i = 0; i < len; i++) {
2842 btf_mod = &used_btfs[i];
2843 if (btf_mod->module)
2844 module_put(btf_mod->module);
2845 btf_put(btf_mod->btf);
2846 }
2847#endif
2848}
2849
2850static void bpf_free_used_btfs(struct bpf_prog_aux *aux)
2851{
ab224b9e 2852 __bpf_free_used_btfs(aux->used_btfs, aux->used_btf_cnt);
541c3bad
AN		 2853 kfree(aux->used_btfs);
2854}
2855
60a3b225
DB		 2856static void bpf_prog_free_deferred(struct work_struct *work)
2857{
09756af4 2858 struct bpf_prog_aux *aux;
1c2a088a 2859 int i;
60a3b225 2860
09756af4 2861 aux = container_of(work, struct bpf_prog_aux, work);
2357672c
KKD		 2862#ifdef CONFIG_BPF_SYSCALL
2863 bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
c0e19f2c
SF		 2864#endif
2865#ifdef CONFIG_CGROUP_BPF
2866 if (aux->cgroup_atype != CGROUP_BPF_ATTACH_TYPE_INVALID)
2867 bpf_cgroup_atype_put(aux->cgroup_atype);
2357672c 2868#endif
6332be04 2869 bpf_free_used_maps(aux);
541c3bad 2870 bpf_free_used_btfs(aux);
ab3f0063 2871 if (bpf_prog_is_dev_bound(aux))
2b3486bc 2872 bpf_prog_dev_bound_destroy(aux->prog);
c195651e
YS		 2873#ifdef CONFIG_PERF_EVENTS
2874 if (aux->prog->has_callchain_buf)
2875 put_callchain_buffers();
2876#endif
3aac1ead
THJ		 2877 if (aux->dst_trampoline)
2878 bpf_trampoline_put(aux->dst_trampoline);
335d1c5b 2879 for (i = 0; i < aux->real_func_cnt; i++) {
f263a814
JF		 2880 /* We can just unlink the subprog poke descriptor table as
2881 * it was originally linked to the main program and is also
2882 * released along with it.
2883 */
2884 aux->func[i]->aux->poke_tab = NULL;
1c2a088a 2885 bpf_jit_free(aux->func[i]);
f263a814 2886 }
335d1c5b 2887 if (aux->real_func_cnt) {
1c2a088a
AS		 2888 kfree(aux->func);
2889 bpf_prog_unlock_free(aux->prog);
2890 } else {
2891 bpf_jit_free(aux->prog);
2892 }
60a3b225
DB		 2893}
2894
7ae457c1 2895void bpf_prog_free(struct bpf_prog *fp)
f5bffecd 2896{
09756af4 2897 struct bpf_prog_aux *aux = fp->aux;
60a3b225 2898
3aac1ead
THJ		 2899 if (aux->dst_prog)
2900 bpf_prog_put(aux->dst_prog);
caf8f28e 2901 bpf_token_put(aux->token);
09756af4 2902 INIT_WORK(&aux->work, bpf_prog_free_deferred);
09756af4 2903 schedule_work(&aux->work);
f5bffecd 2904}
7ae457c1 2905EXPORT_SYMBOL_GPL(bpf_prog_free);
f89b7755 2906
a7de265c 2907/* RNG for unprivileged user space with separated state from prandom_u32(). */
3ad00405
DB		 2908static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2909
2910void bpf_user_rnd_init_once(void)
2911{
2912 prandom_init_once(&bpf_user_rnd_state);
2913}
2914
f3694e00 2915BPF_CALL_0(bpf_user_rnd_u32)
3ad00405
DB		 2916{
2917 /* Should someone ever have the rather unwise idea to use some
2918 * of the registers passed into this function, then note that
2919 * this function is called from native eBPF and classic-to-eBPF
2920 * transformations. Register assignments from both sides are
2921 * different, f.e. classic always sets fn(ctx, A, X) here.
2922 */
2923 struct rnd_state *state;
2924 u32 res;
2925
2926 state = &get_cpu_var(bpf_user_rnd_state);
2927 res = prandom_u32_state(state);
b761fe22 2928 put_cpu_var(bpf_user_rnd_state);
3ad00405
DB		 2929
2930 return res;
2931}
2932
6890896b
SF		 2933BPF_CALL_0(bpf_get_raw_cpu_id)
2934{
2935 return raw_smp_processor_id();
2936}
2937
3ba67dab
DB		 2938/* Weak definitions of helper functions in case we don't have bpf syscall. */
2939const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2940const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2941const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
f1a2e44a
MV		 2942const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2943const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2944const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
07343110 2945const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto __weak;
d83525ca
AS		 2946const struct bpf_func_proto bpf_spin_lock_proto __weak;
2947const struct bpf_func_proto bpf_spin_unlock_proto __weak;
5576b991 2948const struct bpf_func_proto bpf_jiffies64_proto __weak;
3ba67dab 2949
03e69b50 2950const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
c04167ce 2951const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2d0e30c3 2952const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
17ca8cbf 2953const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
71d19214 2954const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
d0551261 2955const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto __weak;
c8996c98 2956const struct bpf_func_proto bpf_ktime_get_tai_ns_proto __weak;
bd570ff9 2957
ffeedafb
AS		 2958const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2959const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2960const struct bpf_func_proto bpf_get_current_comm_proto __weak;
bf6fa2c8 2961const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
0f09abd1 2962const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
cd339431 2963const struct bpf_func_proto bpf_get_local_storage_proto __weak;
b4490c5c 2964const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
c4d0bfb4 2965const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
eb411377 2966const struct bpf_func_proto bpf_seq_printf_btf_proto __weak;
69fd337a
SF		 2967const struct bpf_func_proto bpf_set_retval_proto __weak;
2968const struct bpf_func_proto bpf_get_retval_proto __weak;
bd570ff9 2969
0756ea3e
AS		 2970const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2971{
2972 return NULL;
2973}
03e69b50 2974
10aceb62
DM		 2975const struct bpf_func_proto * __weak bpf_get_trace_vprintk_proto(void)
2976{
2977 return NULL;
2978}
2979
ae0a457f
ET		 2980const struct bpf_func_proto * __weak bpf_get_perf_event_read_value_proto(void)
2981{
2982 return NULL;
2983}
2984
555c8a86
DB		 2985u64 __weak
2986bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2987 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
bd570ff9 2988{
555c8a86 2989 return -ENOTSUPP;
bd570ff9 2990}
6cb5fb38 2991EXPORT_SYMBOL_GPL(bpf_event_output);
bd570ff9 2992
3324b584
DB		 2993/* Always built-in helper functions. */
2994const struct bpf_func_proto bpf_tail_call_proto = {
2995 .func = NULL,
2996 .gpl_only = false,
2997 .ret_type = RET_VOID,
2998 .arg1_type = ARG_PTR_TO_CTX,
2999 .arg2_type = ARG_CONST_MAP_PTR,
3000 .arg3_type = ARG_ANYTHING,
3001};
3002
9383191d
DB		 3003/* Stub for JITs that only support cBPF. eBPF programs are interpreted.
3004 * It is encouraged to implement bpf_int_jit_compile() instead, so that
3005 * eBPF and implicitly also cBPF can get JITed!
3006 */
d1c55ab5 3007struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
3324b584 3008{
d1c55ab5 3009 return prog;
3324b584
DB		 3010}
3011
9383191d
DB		 3012/* Stub for JITs that support eBPF. All cBPF code gets transformed into
3013 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
3014 */
3015void __weak bpf_jit_compile(struct bpf_prog *prog)
3016{
3017}
3018
b238e187 3019bool __weak bpf_helper_changes_pkt_data(enum bpf_func_id func_id)
969bf05e
AS		 3020{
3021 return false;
3022}
3023
a4b1d3c1
JW		 3024/* Return TRUE if the JIT backend wants verifier to enable sub-register usage
3025 * analysis code and wants explicit zero extension inserted by verifier.
3026 * Otherwise, return FALSE.
39491867
BJ		 3027 *
3028 * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
3029 * you don't override this. JITs that don't want these extra insns can detect
3030 * them using insn_is_zext.
a4b1d3c1
JW		 3031 */
3032bool __weak bpf_jit_needs_zext(void)
3033{
3034 return false;
3035}
3036
2ddec2c8
PM		 3037/* Return true if the JIT inlines the call to the helper corresponding to
3038 * the imm.
3039 *
3040 * The verifier will not patch the insn->imm for the call to the helper if
3041 * this returns true.
3042 */
3043bool __weak bpf_jit_inlines_helper_call(s32 imm)
3044{
3045 return false;
3046}
3047
95acd881
TA		 3048/* Return TRUE if the JIT backend supports mixing bpf2bpf and tailcalls. */
3049bool __weak bpf_jit_supports_subprog_tailcalls(void)
3050{
3051 return false;
3052}
3053
7bdbf744
AN		 3054bool __weak bpf_jit_supports_percpu_insn(void)
3055{
3056 return false;
3057}
3058
e6ac2450
MKL		 3059bool __weak bpf_jit_supports_kfunc_call(void)
3060{
3061 return false;
3062}
3063
1cf3bfc6
IL		 3064bool __weak bpf_jit_supports_far_kfunc_call(void)
3065{
3066 return false;
3067}
3068
142fd4d2
AS		 3069bool __weak bpf_jit_supports_arena(void)
3070{
3071 return false;
3072}
3073
d503a04f
AS		 3074bool __weak bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
3075{
3076 return false;
3077}
3078
66e13b61
PM		 3079u64 __weak bpf_arch_uaddress_limit(void)
3080{
3081#if defined(CONFIG_64BIT) && defined(CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE)
3082 return TASK_SIZE;
3083#else
3084 return 0;
3085#endif
3086}
3087
7c05e7f3
HT		 3088/* Return TRUE if the JIT backend satisfies the following two conditions:
3089 * 1) JIT backend supports atomic_xchg() on pointer-sized words.
3090 * 2) Under the specific arch, the implementation of xchg() is the same
3091 * as atomic_xchg() on pointer-sized words.
3092 */
3093bool __weak bpf_jit_supports_ptr_xchg(void)
3094{
3095 return false;
3096}
3097
f89b7755
AS		 3098/* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
3099 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
3100 */
3101int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
3102 int len)
3103{
3104 return -EFAULT;
3105}
a67edbf4 3106
5964b200
AS		 3107int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
3108 void *addr1, void *addr2)
3109{
3110 return -ENOTSUPP;
3111}
3112
ebc1415d
SL		 3113void * __weak bpf_arch_text_copy(void *dst, void *src, size_t len)
3114{
3115 return ERR_PTR(-ENOTSUPP);
3116}
3117
fe736565
SL		 3118int __weak bpf_arch_text_invalidate(void *dst, size_t len)
3119{
3120 return -ENOTSUPP;
3121}
3122
fd5d27b7
KKD		 3123bool __weak bpf_jit_supports_exceptions(void)
3124{
3125 return false;
3126}
3127
a76ab573
YS		 3128bool __weak bpf_jit_supports_private_stack(void)
3129{
3130 return false;
3131}
3132
fd5d27b7
KKD		 3133void __weak arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie)
3134{
3135}
3136
e723608b
KKD		 3137bool __weak bpf_jit_supports_timed_may_goto(void)
3138{
3139 return false;
3140}
3141
3142u64 __weak arch_bpf_timed_may_goto(void)
3143{
3144 return 0;
3145}
3146
3147u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *p)
3148{
3149 u64 time = ktime_get_mono_fast_ns();
3150
3151 /* Populate the timestamp for this stack frame, and refresh count. */
3152 if (!p->timestamp) {
3153 p->timestamp = time;
3154 return BPF_MAX_TIMED_LOOPS;
3155 }
3156 /* Check if we've exhausted our time slice, and zero count. */
3157 if (time - p->timestamp >= (NSEC_PER_SEC / 4))
3158 return 0;
3159 /* Refresh the count for the stack frame. */
3160 return BPF_MAX_TIMED_LOOPS;
3161}
3162
31746031
AS		 3163/* for configs without MMU or 32-bit */
3164__weak const struct bpf_map_ops arena_map_ops;
3165__weak u64 bpf_arena_get_user_vm_start(struct bpf_arena *arena)
3166{
3167 return 0;
3168}
3169__weak u64 bpf_arena_get_kern_vm_start(struct bpf_arena *arena)
3170{
3171 return 0;
3172}
3173
958cf2e2
KKD		 3174#ifdef CONFIG_BPF_SYSCALL
3175static int __init bpf_global_ma_init(void)
3176{
3177 int ret;
3178
3179 ret = bpf_mem_alloc_init(&bpf_global_ma, 0, false);
3180 bpf_global_ma_set = !ret;
1fda5bb6 3181 return ret;
958cf2e2
KKD		 3182}
3183late_initcall(bpf_global_ma_init);
3184#endif
3185
492ecee8
AS		 3186DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
3187EXPORT_SYMBOL(bpf_stats_enabled_key);
492ecee8 3188
a67edbf4
DB		 3189/* All definitions of tracepoints related to BPF. */
3190#define CREATE_TRACE_POINTS
3191#include <linux/bpf_trace.h>
3192
3193EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
e7d47989 3194EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
Linux 6.x block layer and io_uring tree(s)