mm: create the new vm_fault_t type
[linux-2.6-microblaze.git] / net / core / filter.c
1 /*
2  * Linux Socket Filter - Kernel level socket filtering
3  *
4  * Based on the design of the Berkeley Packet Filter. The new
5  * internal format has been designed by PLUMgrid:
6  *
7  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
8  *
9  * Authors:
10  *
11  *      Jay Schulist <jschlst@samba.org>
12  *      Alexei Starovoitov <ast@plumgrid.com>
13  *      Daniel Borkmann <dborkman@redhat.com>
14  *
15  * This program is free software; you can redistribute it and/or
16  * modify it under the terms of the GNU General Public License
17  * as published by the Free Software Foundation; either version
18  * 2 of the License, or (at your option) any later version.
19  *
20  * Andi Kleen - Fix a few bad bugs and races.
21  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
22  */
23
24 #include <linux/module.h>
25 #include <linux/types.h>
26 #include <linux/mm.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
29 #include <linux/sock_diag.h>
30 #include <linux/in.h>
31 #include <linux/inet.h>
32 #include <linux/netdevice.h>
33 #include <linux/if_packet.h>
34 #include <linux/if_arp.h>
35 #include <linux/gfp.h>
36 #include <net/inet_common.h>
37 #include <net/ip.h>
38 #include <net/protocol.h>
39 #include <net/netlink.h>
40 #include <linux/skbuff.h>
41 #include <linux/skmsg.h>
42 #include <net/sock.h>
43 #include <net/flow_dissector.h>
44 #include <linux/errno.h>
45 #include <linux/timer.h>
46 #include <linux/uaccess.h>
47 #include <asm/unaligned.h>
48 #include <asm/cmpxchg.h>
49 #include <linux/filter.h>
50 #include <linux/ratelimit.h>
51 #include <linux/seccomp.h>
52 #include <linux/if_vlan.h>
53 #include <linux/bpf.h>
54 #include <net/sch_generic.h>
55 #include <net/cls_cgroup.h>
56 #include <net/dst_metadata.h>
57 #include <net/dst.h>
58 #include <net/sock_reuseport.h>
59 #include <net/busy_poll.h>
60 #include <net/tcp.h>
61 #include <net/xfrm.h>
62 #include <net/udp.h>
63 #include <linux/bpf_trace.h>
64 #include <net/xdp_sock.h>
65 #include <linux/inetdevice.h>
66 #include <net/inet_hashtables.h>
67 #include <net/inet6_hashtables.h>
68 #include <net/ip_fib.h>
69 #include <net/flow.h>
70 #include <net/arp.h>
71 #include <net/ipv6.h>
72 #include <net/net_namespace.h>
73 #include <linux/seg6_local.h>
74 #include <net/seg6.h>
75 #include <net/seg6_local.h>
76 #include <net/lwtunnel.h>
77
78 /**
79  *      sk_filter_trim_cap - run a packet through a socket filter
80  *      @sk: sock associated with &sk_buff
81  *      @skb: buffer to filter
82  *      @cap: limit on how short the eBPF program may trim the packet
83  *
84  * Run the eBPF program and then cut skb->data to correct size returned by
85  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
86  * than pkt_len we keep whole skb->data. This is the socket level
87  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
88  * be accepted or -EPERM if the packet should be tossed.
89  *
90  */
91 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
92 {
93         int err;
94         struct sk_filter *filter;
95
96         /*
97          * If the skb was allocated from pfmemalloc reserves, only
98          * allow SOCK_MEMALLOC sockets to use it as this socket is
99          * helping free memory
100          */
101         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
102                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
103                 return -ENOMEM;
104         }
105         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
106         if (err)
107                 return err;
108
109         err = security_sock_rcv_skb(sk, skb);
110         if (err)
111                 return err;
112
113         rcu_read_lock();
114         filter = rcu_dereference(sk->sk_filter);
115         if (filter) {
116                 struct sock *save_sk = skb->sk;
117                 unsigned int pkt_len;
118
119                 skb->sk = sk;
120                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
121                 skb->sk = save_sk;
122                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
123         }
124         rcu_read_unlock();
125
126         return err;
127 }
128 EXPORT_SYMBOL(sk_filter_trim_cap);
129
130 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
131 {
132         return skb_get_poff(skb);
133 }
134
135 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
136 {
137         struct nlattr *nla;
138
139         if (skb_is_nonlinear(skb))
140                 return 0;
141
142         if (skb->len < sizeof(struct nlattr))
143                 return 0;
144
145         if (a > skb->len - sizeof(struct nlattr))
146                 return 0;
147
148         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
149         if (nla)
150                 return (void *) nla - (void *) skb->data;
151
152         return 0;
153 }
154
155 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
156 {
157         struct nlattr *nla;
158
159         if (skb_is_nonlinear(skb))
160                 return 0;
161
162         if (skb->len < sizeof(struct nlattr))
163                 return 0;
164
165         if (a > skb->len - sizeof(struct nlattr))
166                 return 0;
167
168         nla = (struct nlattr *) &skb->data[a];
169         if (nla->nla_len > skb->len - a)
170                 return 0;
171
172         nla = nla_find_nested(nla, x);
173         if (nla)
174                 return (void *) nla - (void *) skb->data;
175
176         return 0;
177 }
178
179 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
180            data, int, headlen, int, offset)
181 {
182         u8 tmp, *ptr;
183         const int len = sizeof(tmp);
184
185         if (offset >= 0) {
186                 if (headlen - offset >= len)
187                         return *(u8 *)(data + offset);
188                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
189                         return tmp;
190         } else {
191                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
192                 if (likely(ptr))
193                         return *(u8 *)ptr;
194         }
195
196         return -EFAULT;
197 }
198
199 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
200            int, offset)
201 {
202         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
203                                          offset);
204 }
205
206 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
207            data, int, headlen, int, offset)
208 {
209         u16 tmp, *ptr;
210         const int len = sizeof(tmp);
211
212         if (offset >= 0) {
213                 if (headlen - offset >= len)
214                         return get_unaligned_be16(data + offset);
215                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
216                         return be16_to_cpu(tmp);
217         } else {
218                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
219                 if (likely(ptr))
220                         return get_unaligned_be16(ptr);
221         }
222
223         return -EFAULT;
224 }
225
226 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
227            int, offset)
228 {
229         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
230                                           offset);
231 }
232
233 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
234            data, int, headlen, int, offset)
235 {
236         u32 tmp, *ptr;
237         const int len = sizeof(tmp);
238
239         if (likely(offset >= 0)) {
240                 if (headlen - offset >= len)
241                         return get_unaligned_be32(data + offset);
242                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
243                         return be32_to_cpu(tmp);
244         } else {
245                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
246                 if (likely(ptr))
247                         return get_unaligned_be32(ptr);
248         }
249
250         return -EFAULT;
251 }
252
253 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
254            int, offset)
255 {
256         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
257                                           offset);
258 }
259
260 BPF_CALL_0(bpf_get_raw_cpu_id)
261 {
262         return raw_smp_processor_id();
263 }
264
265 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
266         .func           = bpf_get_raw_cpu_id,
267         .gpl_only       = false,
268         .ret_type       = RET_INTEGER,
269 };
270
271 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
272                               struct bpf_insn *insn_buf)
273 {
274         struct bpf_insn *insn = insn_buf;
275
276         switch (skb_field) {
277         case SKF_AD_MARK:
278                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
279
280                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
281                                       offsetof(struct sk_buff, mark));
282                 break;
283
284         case SKF_AD_PKTTYPE:
285                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
286                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
287 #ifdef __BIG_ENDIAN_BITFIELD
288                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
289 #endif
290                 break;
291
292         case SKF_AD_QUEUE:
293                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
294
295                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
296                                       offsetof(struct sk_buff, queue_mapping));
297                 break;
298
299         case SKF_AD_VLAN_TAG:
300                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
301
302                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
303                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
304                                       offsetof(struct sk_buff, vlan_tci));
305                 break;
306         case SKF_AD_VLAN_TAG_PRESENT:
307                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
308                 if (PKT_VLAN_PRESENT_BIT)
309                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
310                 if (PKT_VLAN_PRESENT_BIT < 7)
311                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
312                 break;
313         }
314
315         return insn - insn_buf;
316 }
317
318 static bool convert_bpf_extensions(struct sock_filter *fp,
319                                    struct bpf_insn **insnp)
320 {
321         struct bpf_insn *insn = *insnp;
322         u32 cnt;
323
324         switch (fp->k) {
325         case SKF_AD_OFF + SKF_AD_PROTOCOL:
326                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
327
328                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
329                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
330                                       offsetof(struct sk_buff, protocol));
331                 /* A = ntohs(A) [emitting a nop or swap16] */
332                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
333                 break;
334
335         case SKF_AD_OFF + SKF_AD_PKTTYPE:
336                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
337                 insn += cnt - 1;
338                 break;
339
340         case SKF_AD_OFF + SKF_AD_IFINDEX:
341         case SKF_AD_OFF + SKF_AD_HATYPE:
342                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
343                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
344
345                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
346                                       BPF_REG_TMP, BPF_REG_CTX,
347                                       offsetof(struct sk_buff, dev));
348                 /* if (tmp != 0) goto pc + 1 */
349                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
350                 *insn++ = BPF_EXIT_INSN();
351                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
352                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
353                                             offsetof(struct net_device, ifindex));
354                 else
355                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
356                                             offsetof(struct net_device, type));
357                 break;
358
359         case SKF_AD_OFF + SKF_AD_MARK:
360                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
361                 insn += cnt - 1;
362                 break;
363
364         case SKF_AD_OFF + SKF_AD_RXHASH:
365                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
366
367                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
368                                     offsetof(struct sk_buff, hash));
369                 break;
370
371         case SKF_AD_OFF + SKF_AD_QUEUE:
372                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
373                 insn += cnt - 1;
374                 break;
375
376         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
377                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
378                                          BPF_REG_A, BPF_REG_CTX, insn);
379                 insn += cnt - 1;
380                 break;
381
382         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
383                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
384                                          BPF_REG_A, BPF_REG_CTX, insn);
385                 insn += cnt - 1;
386                 break;
387
388         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
389                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
390
391                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
392                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
393                                       offsetof(struct sk_buff, vlan_proto));
394                 /* A = ntohs(A) [emitting a nop or swap16] */
395                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
396                 break;
397
398         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
399         case SKF_AD_OFF + SKF_AD_NLATTR:
400         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
401         case SKF_AD_OFF + SKF_AD_CPU:
402         case SKF_AD_OFF + SKF_AD_RANDOM:
403                 /* arg1 = CTX */
404                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
405                 /* arg2 = A */
406                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
407                 /* arg3 = X */
408                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
409                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
410                 switch (fp->k) {
411                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
412                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
413                         break;
414                 case SKF_AD_OFF + SKF_AD_NLATTR:
415                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
416                         break;
417                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
418                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
419                         break;
420                 case SKF_AD_OFF + SKF_AD_CPU:
421                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
422                         break;
423                 case SKF_AD_OFF + SKF_AD_RANDOM:
424                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
425                         bpf_user_rnd_init_once();
426                         break;
427                 }
428                 break;
429
430         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
431                 /* A ^= X */
432                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
433                 break;
434
435         default:
436                 /* This is just a dummy call to avoid letting the compiler
437                  * evict __bpf_call_base() as an optimization. Placed here
438                  * where no-one bothers.
439                  */
440                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
441                 return false;
442         }
443
444         *insnp = insn;
445         return true;
446 }
447
448 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
449 {
450         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
451         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
452         bool endian = BPF_SIZE(fp->code) == BPF_H ||
453                       BPF_SIZE(fp->code) == BPF_W;
454         bool indirect = BPF_MODE(fp->code) == BPF_IND;
455         const int ip_align = NET_IP_ALIGN;
456         struct bpf_insn *insn = *insnp;
457         int offset = fp->k;
458
459         if (!indirect &&
460             ((unaligned_ok && offset >= 0) ||
461              (!unaligned_ok && offset >= 0 &&
462               offset + ip_align >= 0 &&
463               offset + ip_align % size == 0))) {
464                 bool ldx_off_ok = offset <= S16_MAX;
465
466                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
467                 if (offset)
468                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
469                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
470                                       size, 2 + endian + (!ldx_off_ok * 2));
471                 if (ldx_off_ok) {
472                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
473                                               BPF_REG_D, offset);
474                 } else {
475                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
476                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
477                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
478                                               BPF_REG_TMP, 0);
479                 }
480                 if (endian)
481                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
482                 *insn++ = BPF_JMP_A(8);
483         }
484
485         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
486         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
487         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
488         if (!indirect) {
489                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
490         } else {
491                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
492                 if (fp->k)
493                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
494         }
495
496         switch (BPF_SIZE(fp->code)) {
497         case BPF_B:
498                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
499                 break;
500         case BPF_H:
501                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
502                 break;
503         case BPF_W:
504                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
505                 break;
506         default:
507                 return false;
508         }
509
510         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
511         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
512         *insn   = BPF_EXIT_INSN();
513
514         *insnp = insn;
515         return true;
516 }
517
518 /**
519  *      bpf_convert_filter - convert filter program
520  *      @prog: the user passed filter program
521  *      @len: the length of the user passed filter program
522  *      @new_prog: allocated 'struct bpf_prog' or NULL
523  *      @new_len: pointer to store length of converted program
524  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
525  *
526  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
527  * style extended BPF (eBPF).
528  * Conversion workflow:
529  *
530  * 1) First pass for calculating the new program length:
531  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
532  *
533  * 2) 2nd pass to remap in two passes: 1st pass finds new
534  *    jump offsets, 2nd pass remapping:
535  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
536  */
537 static int bpf_convert_filter(struct sock_filter *prog, int len,
538                               struct bpf_prog *new_prog, int *new_len,
539                               bool *seen_ld_abs)
540 {
541         int new_flen = 0, pass = 0, target, i, stack_off;
542         struct bpf_insn *new_insn, *first_insn = NULL;
543         struct sock_filter *fp;
544         int *addrs = NULL;
545         u8 bpf_src;
546
547         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
548         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
549
550         if (len <= 0 || len > BPF_MAXINSNS)
551                 return -EINVAL;
552
553         if (new_prog) {
554                 first_insn = new_prog->insnsi;
555                 addrs = kcalloc(len, sizeof(*addrs),
556                                 GFP_KERNEL | __GFP_NOWARN);
557                 if (!addrs)
558                         return -ENOMEM;
559         }
560
561 do_pass:
562         new_insn = first_insn;
563         fp = prog;
564
565         /* Classic BPF related prologue emission. */
566         if (new_prog) {
567                 /* Classic BPF expects A and X to be reset first. These need
568                  * to be guaranteed to be the first two instructions.
569                  */
570                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
571                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
572
573                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
574                  * In eBPF case it's done by the compiler, here we need to
575                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
576                  */
577                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
578                 if (*seen_ld_abs) {
579                         /* For packet access in classic BPF, cache skb->data
580                          * in callee-saved BPF R8 and skb->len - skb->data_len
581                          * (headlen) in BPF R9. Since classic BPF is read-only
582                          * on CTX, we only need to cache it once.
583                          */
584                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
585                                                   BPF_REG_D, BPF_REG_CTX,
586                                                   offsetof(struct sk_buff, data));
587                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
588                                                   offsetof(struct sk_buff, len));
589                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
590                                                   offsetof(struct sk_buff, data_len));
591                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
592                 }
593         } else {
594                 new_insn += 3;
595         }
596
597         for (i = 0; i < len; fp++, i++) {
598                 struct bpf_insn tmp_insns[32] = { };
599                 struct bpf_insn *insn = tmp_insns;
600
601                 if (addrs)
602                         addrs[i] = new_insn - first_insn;
603
604                 switch (fp->code) {
605                 /* All arithmetic insns and skb loads map as-is. */
606                 case BPF_ALU | BPF_ADD | BPF_X:
607                 case BPF_ALU | BPF_ADD | BPF_K:
608                 case BPF_ALU | BPF_SUB | BPF_X:
609                 case BPF_ALU | BPF_SUB | BPF_K:
610                 case BPF_ALU | BPF_AND | BPF_X:
611                 case BPF_ALU | BPF_AND | BPF_K:
612                 case BPF_ALU | BPF_OR | BPF_X:
613                 case BPF_ALU | BPF_OR | BPF_K:
614                 case BPF_ALU | BPF_LSH | BPF_X:
615                 case BPF_ALU | BPF_LSH | BPF_K:
616                 case BPF_ALU | BPF_RSH | BPF_X:
617                 case BPF_ALU | BPF_RSH | BPF_K:
618                 case BPF_ALU | BPF_XOR | BPF_X:
619                 case BPF_ALU | BPF_XOR | BPF_K:
620                 case BPF_ALU | BPF_MUL | BPF_X:
621                 case BPF_ALU | BPF_MUL | BPF_K:
622                 case BPF_ALU | BPF_DIV | BPF_X:
623                 case BPF_ALU | BPF_DIV | BPF_K:
624                 case BPF_ALU | BPF_MOD | BPF_X:
625                 case BPF_ALU | BPF_MOD | BPF_K:
626                 case BPF_ALU | BPF_NEG:
627                 case BPF_LD | BPF_ABS | BPF_W:
628                 case BPF_LD | BPF_ABS | BPF_H:
629                 case BPF_LD | BPF_ABS | BPF_B:
630                 case BPF_LD | BPF_IND | BPF_W:
631                 case BPF_LD | BPF_IND | BPF_H:
632                 case BPF_LD | BPF_IND | BPF_B:
633                         /* Check for overloaded BPF extension and
634                          * directly convert it if found, otherwise
635                          * just move on with mapping.
636                          */
637                         if (BPF_CLASS(fp->code) == BPF_LD &&
638                             BPF_MODE(fp->code) == BPF_ABS &&
639                             convert_bpf_extensions(fp, &insn))
640                                 break;
641                         if (BPF_CLASS(fp->code) == BPF_LD &&
642                             convert_bpf_ld_abs(fp, &insn)) {
643                                 *seen_ld_abs = true;
644                                 break;
645                         }
646
647                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
648                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
649                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
650                                 /* Error with exception code on div/mod by 0.
651                                  * For cBPF programs, this was always return 0.
652                                  */
653                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
654                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
655                                 *insn++ = BPF_EXIT_INSN();
656                         }
657
658                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
659                         break;
660
661                 /* Jump transformation cannot use BPF block macros
662                  * everywhere as offset calculation and target updates
663                  * require a bit more work than the rest, i.e. jump
664                  * opcodes map as-is, but offsets need adjustment.
665                  */
666
667 #define BPF_EMIT_JMP                                                    \
668         do {                                                            \
669                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
670                 s32 off;                                                \
671                                                                         \
672                 if (target >= len || target < 0)                        \
673                         goto err;                                       \
674                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
675                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
676                 off -= insn - tmp_insns;                                \
677                 /* Reject anything not fitting into insn->off. */       \
678                 if (off < off_min || off > off_max)                     \
679                         goto err;                                       \
680                 insn->off = off;                                        \
681         } while (0)
682
683                 case BPF_JMP | BPF_JA:
684                         target = i + fp->k + 1;
685                         insn->code = fp->code;
686                         BPF_EMIT_JMP;
687                         break;
688
689                 case BPF_JMP | BPF_JEQ | BPF_K:
690                 case BPF_JMP | BPF_JEQ | BPF_X:
691                 case BPF_JMP | BPF_JSET | BPF_K:
692                 case BPF_JMP | BPF_JSET | BPF_X:
693                 case BPF_JMP | BPF_JGT | BPF_K:
694                 case BPF_JMP | BPF_JGT | BPF_X:
695                 case BPF_JMP | BPF_JGE | BPF_K:
696                 case BPF_JMP | BPF_JGE | BPF_X:
697                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
698                                 /* BPF immediates are signed, zero extend
699                                  * immediate into tmp register and use it
700                                  * in compare insn.
701                                  */
702                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
703
704                                 insn->dst_reg = BPF_REG_A;
705                                 insn->src_reg = BPF_REG_TMP;
706                                 bpf_src = BPF_X;
707                         } else {
708                                 insn->dst_reg = BPF_REG_A;
709                                 insn->imm = fp->k;
710                                 bpf_src = BPF_SRC(fp->code);
711                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
712                         }
713
714                         /* Common case where 'jump_false' is next insn. */
715                         if (fp->jf == 0) {
716                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
717                                 target = i + fp->jt + 1;
718                                 BPF_EMIT_JMP;
719                                 break;
720                         }
721
722                         /* Convert some jumps when 'jump_true' is next insn. */
723                         if (fp->jt == 0) {
724                                 switch (BPF_OP(fp->code)) {
725                                 case BPF_JEQ:
726                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
727                                         break;
728                                 case BPF_JGT:
729                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
730                                         break;
731                                 case BPF_JGE:
732                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
733                                         break;
734                                 default:
735                                         goto jmp_rest;
736                                 }
737
738                                 target = i + fp->jf + 1;
739                                 BPF_EMIT_JMP;
740                                 break;
741                         }
742 jmp_rest:
743                         /* Other jumps are mapped into two insns: Jxx and JA. */
744                         target = i + fp->jt + 1;
745                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
746                         BPF_EMIT_JMP;
747                         insn++;
748
749                         insn->code = BPF_JMP | BPF_JA;
750                         target = i + fp->jf + 1;
751                         BPF_EMIT_JMP;
752                         break;
753
754                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
755                 case BPF_LDX | BPF_MSH | BPF_B: {
756                         struct sock_filter tmp = {
757                                 .code   = BPF_LD | BPF_ABS | BPF_B,
758                                 .k      = fp->k,
759                         };
760
761                         *seen_ld_abs = true;
762
763                         /* X = A */
764                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
765                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
766                         convert_bpf_ld_abs(&tmp, &insn);
767                         insn++;
768                         /* A &= 0xf */
769                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
770                         /* A <<= 2 */
771                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
772                         /* tmp = X */
773                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
774                         /* X = A */
775                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
776                         /* A = tmp */
777                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
778                         break;
779                 }
780                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
781                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
782                  */
783                 case BPF_RET | BPF_A:
784                 case BPF_RET | BPF_K:
785                         if (BPF_RVAL(fp->code) == BPF_K)
786                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
787                                                         0, fp->k);
788                         *insn = BPF_EXIT_INSN();
789                         break;
790
791                 /* Store to stack. */
792                 case BPF_ST:
793                 case BPF_STX:
794                         stack_off = fp->k * 4  + 4;
795                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
796                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
797                                             -stack_off);
798                         /* check_load_and_stores() verifies that classic BPF can
799                          * load from stack only after write, so tracking
800                          * stack_depth for ST|STX insns is enough
801                          */
802                         if (new_prog && new_prog->aux->stack_depth < stack_off)
803                                 new_prog->aux->stack_depth = stack_off;
804                         break;
805
806                 /* Load from stack. */
807                 case BPF_LD | BPF_MEM:
808                 case BPF_LDX | BPF_MEM:
809                         stack_off = fp->k * 4  + 4;
810                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
811                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
812                                             -stack_off);
813                         break;
814
815                 /* A = K or X = K */
816                 case BPF_LD | BPF_IMM:
817                 case BPF_LDX | BPF_IMM:
818                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
819                                               BPF_REG_A : BPF_REG_X, fp->k);
820                         break;
821
822                 /* X = A */
823                 case BPF_MISC | BPF_TAX:
824                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
825                         break;
826
827                 /* A = X */
828                 case BPF_MISC | BPF_TXA:
829                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
830                         break;
831
832                 /* A = skb->len or X = skb->len */
833                 case BPF_LD | BPF_W | BPF_LEN:
834                 case BPF_LDX | BPF_W | BPF_LEN:
835                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
836                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
837                                             offsetof(struct sk_buff, len));
838                         break;
839
840                 /* Access seccomp_data fields. */
841                 case BPF_LDX | BPF_ABS | BPF_W:
842                         /* A = *(u32 *) (ctx + K) */
843                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
844                         break;
845
846                 /* Unknown instruction. */
847                 default:
848                         goto err;
849                 }
850
851                 insn++;
852                 if (new_prog)
853                         memcpy(new_insn, tmp_insns,
854                                sizeof(*insn) * (insn - tmp_insns));
855                 new_insn += insn - tmp_insns;
856         }
857
858         if (!new_prog) {
859                 /* Only calculating new length. */
860                 *new_len = new_insn - first_insn;
861                 if (*seen_ld_abs)
862                         *new_len += 4; /* Prologue bits. */
863                 return 0;
864         }
865
866         pass++;
867         if (new_flen != new_insn - first_insn) {
868                 new_flen = new_insn - first_insn;
869                 if (pass > 2)
870                         goto err;
871                 goto do_pass;
872         }
873
874         kfree(addrs);
875         BUG_ON(*new_len != new_flen);
876         return 0;
877 err:
878         kfree(addrs);
879         return -EINVAL;
880 }
881
882 /* Security:
883  *
884  * As we dont want to clear mem[] array for each packet going through
885  * __bpf_prog_run(), we check that filter loaded by user never try to read
886  * a cell if not previously written, and we check all branches to be sure
887  * a malicious user doesn't try to abuse us.
888  */
889 static int check_load_and_stores(const struct sock_filter *filter, int flen)
890 {
891         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
892         int pc, ret = 0;
893
894         BUILD_BUG_ON(BPF_MEMWORDS > 16);
895
896         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
897         if (!masks)
898                 return -ENOMEM;
899
900         memset(masks, 0xff, flen * sizeof(*masks));
901
902         for (pc = 0; pc < flen; pc++) {
903                 memvalid &= masks[pc];
904
905                 switch (filter[pc].code) {
906                 case BPF_ST:
907                 case BPF_STX:
908                         memvalid |= (1 << filter[pc].k);
909                         break;
910                 case BPF_LD | BPF_MEM:
911                 case BPF_LDX | BPF_MEM:
912                         if (!(memvalid & (1 << filter[pc].k))) {
913                                 ret = -EINVAL;
914                                 goto error;
915                         }
916                         break;
917                 case BPF_JMP | BPF_JA:
918                         /* A jump must set masks on target */
919                         masks[pc + 1 + filter[pc].k] &= memvalid;
920                         memvalid = ~0;
921                         break;
922                 case BPF_JMP | BPF_JEQ | BPF_K:
923                 case BPF_JMP | BPF_JEQ | BPF_X:
924                 case BPF_JMP | BPF_JGE | BPF_K:
925                 case BPF_JMP | BPF_JGE | BPF_X:
926                 case BPF_JMP | BPF_JGT | BPF_K:
927                 case BPF_JMP | BPF_JGT | BPF_X:
928                 case BPF_JMP | BPF_JSET | BPF_K:
929                 case BPF_JMP | BPF_JSET | BPF_X:
930                         /* A jump must set masks on targets */
931                         masks[pc + 1 + filter[pc].jt] &= memvalid;
932                         masks[pc + 1 + filter[pc].jf] &= memvalid;
933                         memvalid = ~0;
934                         break;
935                 }
936         }
937 error:
938         kfree(masks);
939         return ret;
940 }
941
942 static bool chk_code_allowed(u16 code_to_probe)
943 {
944         static const bool codes[] = {
945                 /* 32 bit ALU operations */
946                 [BPF_ALU | BPF_ADD | BPF_K] = true,
947                 [BPF_ALU | BPF_ADD | BPF_X] = true,
948                 [BPF_ALU | BPF_SUB | BPF_K] = true,
949                 [BPF_ALU | BPF_SUB | BPF_X] = true,
950                 [BPF_ALU | BPF_MUL | BPF_K] = true,
951                 [BPF_ALU | BPF_MUL | BPF_X] = true,
952                 [BPF_ALU | BPF_DIV | BPF_K] = true,
953                 [BPF_ALU | BPF_DIV | BPF_X] = true,
954                 [BPF_ALU | BPF_MOD | BPF_K] = true,
955                 [BPF_ALU | BPF_MOD | BPF_X] = true,
956                 [BPF_ALU | BPF_AND | BPF_K] = true,
957                 [BPF_ALU | BPF_AND | BPF_X] = true,
958                 [BPF_ALU | BPF_OR | BPF_K] = true,
959                 [BPF_ALU | BPF_OR | BPF_X] = true,
960                 [BPF_ALU | BPF_XOR | BPF_K] = true,
961                 [BPF_ALU | BPF_XOR | BPF_X] = true,
962                 [BPF_ALU | BPF_LSH | BPF_K] = true,
963                 [BPF_ALU | BPF_LSH | BPF_X] = true,
964                 [BPF_ALU | BPF_RSH | BPF_K] = true,
965                 [BPF_ALU | BPF_RSH | BPF_X] = true,
966                 [BPF_ALU | BPF_NEG] = true,
967                 /* Load instructions */
968                 [BPF_LD | BPF_W | BPF_ABS] = true,
969                 [BPF_LD | BPF_H | BPF_ABS] = true,
970                 [BPF_LD | BPF_B | BPF_ABS] = true,
971                 [BPF_LD | BPF_W | BPF_LEN] = true,
972                 [BPF_LD | BPF_W | BPF_IND] = true,
973                 [BPF_LD | BPF_H | BPF_IND] = true,
974                 [BPF_LD | BPF_B | BPF_IND] = true,
975                 [BPF_LD | BPF_IMM] = true,
976                 [BPF_LD | BPF_MEM] = true,
977                 [BPF_LDX | BPF_W | BPF_LEN] = true,
978                 [BPF_LDX | BPF_B | BPF_MSH] = true,
979                 [BPF_LDX | BPF_IMM] = true,
980                 [BPF_LDX | BPF_MEM] = true,
981                 /* Store instructions */
982                 [BPF_ST] = true,
983                 [BPF_STX] = true,
984                 /* Misc instructions */
985                 [BPF_MISC | BPF_TAX] = true,
986                 [BPF_MISC | BPF_TXA] = true,
987                 /* Return instructions */
988                 [BPF_RET | BPF_K] = true,
989                 [BPF_RET | BPF_A] = true,
990                 /* Jump instructions */
991                 [BPF_JMP | BPF_JA] = true,
992                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
993                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
994                 [BPF_JMP | BPF_JGE | BPF_K] = true,
995                 [BPF_JMP | BPF_JGE | BPF_X] = true,
996                 [BPF_JMP | BPF_JGT | BPF_K] = true,
997                 [BPF_JMP | BPF_JGT | BPF_X] = true,
998                 [BPF_JMP | BPF_JSET | BPF_K] = true,
999                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1000         };
1001
1002         if (code_to_probe >= ARRAY_SIZE(codes))
1003                 return false;
1004
1005         return codes[code_to_probe];
1006 }
1007
1008 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1009                                 unsigned int flen)
1010 {
1011         if (filter == NULL)
1012                 return false;
1013         if (flen == 0 || flen > BPF_MAXINSNS)
1014                 return false;
1015
1016         return true;
1017 }
1018
1019 /**
1020  *      bpf_check_classic - verify socket filter code
1021  *      @filter: filter to verify
1022  *      @flen: length of filter
1023  *
1024  * Check the user's filter code. If we let some ugly
1025  * filter code slip through kaboom! The filter must contain
1026  * no references or jumps that are out of range, no illegal
1027  * instructions, and must end with a RET instruction.
1028  *
1029  * All jumps are forward as they are not signed.
1030  *
1031  * Returns 0 if the rule set is legal or -EINVAL if not.
1032  */
1033 static int bpf_check_classic(const struct sock_filter *filter,
1034                              unsigned int flen)
1035 {
1036         bool anc_found;
1037         int pc;
1038
1039         /* Check the filter code now */
1040         for (pc = 0; pc < flen; pc++) {
1041                 const struct sock_filter *ftest = &filter[pc];
1042
1043                 /* May we actually operate on this code? */
1044                 if (!chk_code_allowed(ftest->code))
1045                         return -EINVAL;
1046
1047                 /* Some instructions need special checks */
1048                 switch (ftest->code) {
1049                 case BPF_ALU | BPF_DIV | BPF_K:
1050                 case BPF_ALU | BPF_MOD | BPF_K:
1051                         /* Check for division by zero */
1052                         if (ftest->k == 0)
1053                                 return -EINVAL;
1054                         break;
1055                 case BPF_ALU | BPF_LSH | BPF_K:
1056                 case BPF_ALU | BPF_RSH | BPF_K:
1057                         if (ftest->k >= 32)
1058                                 return -EINVAL;
1059                         break;
1060                 case BPF_LD | BPF_MEM:
1061                 case BPF_LDX | BPF_MEM:
1062                 case BPF_ST:
1063                 case BPF_STX:
1064                         /* Check for invalid memory addresses */
1065                         if (ftest->k >= BPF_MEMWORDS)
1066                                 return -EINVAL;
1067                         break;
1068                 case BPF_JMP | BPF_JA:
1069                         /* Note, the large ftest->k might cause loops.
1070                          * Compare this with conditional jumps below,
1071                          * where offsets are limited. --ANK (981016)
1072                          */
1073                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1074                                 return -EINVAL;
1075                         break;
1076                 case BPF_JMP | BPF_JEQ | BPF_K:
1077                 case BPF_JMP | BPF_JEQ | BPF_X:
1078                 case BPF_JMP | BPF_JGE | BPF_K:
1079                 case BPF_JMP | BPF_JGE | BPF_X:
1080                 case BPF_JMP | BPF_JGT | BPF_K:
1081                 case BPF_JMP | BPF_JGT | BPF_X:
1082                 case BPF_JMP | BPF_JSET | BPF_K:
1083                 case BPF_JMP | BPF_JSET | BPF_X:
1084                         /* Both conditionals must be safe */
1085                         if (pc + ftest->jt + 1 >= flen ||
1086                             pc + ftest->jf + 1 >= flen)
1087                                 return -EINVAL;
1088                         break;
1089                 case BPF_LD | BPF_W | BPF_ABS:
1090                 case BPF_LD | BPF_H | BPF_ABS:
1091                 case BPF_LD | BPF_B | BPF_ABS:
1092                         anc_found = false;
1093                         if (bpf_anc_helper(ftest) & BPF_ANC)
1094                                 anc_found = true;
1095                         /* Ancillary operation unknown or unsupported */
1096                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1097                                 return -EINVAL;
1098                 }
1099         }
1100
1101         /* Last instruction must be a RET code */
1102         switch (filter[flen - 1].code) {
1103         case BPF_RET | BPF_K:
1104         case BPF_RET | BPF_A:
1105                 return check_load_and_stores(filter, flen);
1106         }
1107
1108         return -EINVAL;
1109 }
1110
1111 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1112                                       const struct sock_fprog *fprog)
1113 {
1114         unsigned int fsize = bpf_classic_proglen(fprog);
1115         struct sock_fprog_kern *fkprog;
1116
1117         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1118         if (!fp->orig_prog)
1119                 return -ENOMEM;
1120
1121         fkprog = fp->orig_prog;
1122         fkprog->len = fprog->len;
1123
1124         fkprog->filter = kmemdup(fp->insns, fsize,
1125                                  GFP_KERNEL | __GFP_NOWARN);
1126         if (!fkprog->filter) {
1127                 kfree(fp->orig_prog);
1128                 return -ENOMEM;
1129         }
1130
1131         return 0;
1132 }
1133
1134 static void bpf_release_orig_filter(struct bpf_prog *fp)
1135 {
1136         struct sock_fprog_kern *fprog = fp->orig_prog;
1137
1138         if (fprog) {
1139                 kfree(fprog->filter);
1140                 kfree(fprog);
1141         }
1142 }
1143
1144 static void __bpf_prog_release(struct bpf_prog *prog)
1145 {
1146         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1147                 bpf_prog_put(prog);
1148         } else {
1149                 bpf_release_orig_filter(prog);
1150                 bpf_prog_free(prog);
1151         }
1152 }
1153
1154 static void __sk_filter_release(struct sk_filter *fp)
1155 {
1156         __bpf_prog_release(fp->prog);
1157         kfree(fp);
1158 }
1159
1160 /**
1161  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1162  *      @rcu: rcu_head that contains the sk_filter to free
1163  */
1164 static void sk_filter_release_rcu(struct rcu_head *rcu)
1165 {
1166         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1167
1168         __sk_filter_release(fp);
1169 }
1170
1171 /**
1172  *      sk_filter_release - release a socket filter
1173  *      @fp: filter to remove
1174  *
1175  *      Remove a filter from a socket and release its resources.
1176  */
1177 static void sk_filter_release(struct sk_filter *fp)
1178 {
1179         if (refcount_dec_and_test(&fp->refcnt))
1180                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1181 }
1182
1183 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1184 {
1185         u32 filter_size = bpf_prog_size(fp->prog->len);
1186
1187         atomic_sub(filter_size, &sk->sk_omem_alloc);
1188         sk_filter_release(fp);
1189 }
1190
1191 /* try to charge the socket memory if there is space available
1192  * return true on success
1193  */
1194 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1195 {
1196         u32 filter_size = bpf_prog_size(fp->prog->len);
1197
1198         /* same check as in sock_kmalloc() */
1199         if (filter_size <= sysctl_optmem_max &&
1200             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1201                 atomic_add(filter_size, &sk->sk_omem_alloc);
1202                 return true;
1203         }
1204         return false;
1205 }
1206
1207 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1208 {
1209         if (!refcount_inc_not_zero(&fp->refcnt))
1210                 return false;
1211
1212         if (!__sk_filter_charge(sk, fp)) {
1213                 sk_filter_release(fp);
1214                 return false;
1215         }
1216         return true;
1217 }
1218
1219 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1220 {
1221         struct sock_filter *old_prog;
1222         struct bpf_prog *old_fp;
1223         int err, new_len, old_len = fp->len;
1224         bool seen_ld_abs = false;
1225
1226         /* We are free to overwrite insns et al right here as it
1227          * won't be used at this point in time anymore internally
1228          * after the migration to the internal BPF instruction
1229          * representation.
1230          */
1231         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1232                      sizeof(struct bpf_insn));
1233
1234         /* Conversion cannot happen on overlapping memory areas,
1235          * so we need to keep the user BPF around until the 2nd
1236          * pass. At this time, the user BPF is stored in fp->insns.
1237          */
1238         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1239                            GFP_KERNEL | __GFP_NOWARN);
1240         if (!old_prog) {
1241                 err = -ENOMEM;
1242                 goto out_err;
1243         }
1244
1245         /* 1st pass: calculate the new program length. */
1246         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1247                                  &seen_ld_abs);
1248         if (err)
1249                 goto out_err_free;
1250
1251         /* Expand fp for appending the new filter representation. */
1252         old_fp = fp;
1253         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1254         if (!fp) {
1255                 /* The old_fp is still around in case we couldn't
1256                  * allocate new memory, so uncharge on that one.
1257                  */
1258                 fp = old_fp;
1259                 err = -ENOMEM;
1260                 goto out_err_free;
1261         }
1262
1263         fp->len = new_len;
1264
1265         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1266         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1267                                  &seen_ld_abs);
1268         if (err)
1269                 /* 2nd bpf_convert_filter() can fail only if it fails
1270                  * to allocate memory, remapping must succeed. Note,
1271                  * that at this time old_fp has already been released
1272                  * by krealloc().
1273                  */
1274                 goto out_err_free;
1275
1276         fp = bpf_prog_select_runtime(fp, &err);
1277         if (err)
1278                 goto out_err_free;
1279
1280         kfree(old_prog);
1281         return fp;
1282
1283 out_err_free:
1284         kfree(old_prog);
1285 out_err:
1286         __bpf_prog_release(fp);
1287         return ERR_PTR(err);
1288 }
1289
1290 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1291                                            bpf_aux_classic_check_t trans)
1292 {
1293         int err;
1294
1295         fp->bpf_func = NULL;
1296         fp->jited = 0;
1297
1298         err = bpf_check_classic(fp->insns, fp->len);
1299         if (err) {
1300                 __bpf_prog_release(fp);
1301                 return ERR_PTR(err);
1302         }
1303
1304         /* There might be additional checks and transformations
1305          * needed on classic filters, f.e. in case of seccomp.
1306          */
1307         if (trans) {
1308                 err = trans(fp->insns, fp->len);
1309                 if (err) {
1310                         __bpf_prog_release(fp);
1311                         return ERR_PTR(err);
1312                 }
1313         }
1314
1315         /* Probe if we can JIT compile the filter and if so, do
1316          * the compilation of the filter.
1317          */
1318         bpf_jit_compile(fp);
1319
1320         /* JIT compiler couldn't process this filter, so do the
1321          * internal BPF translation for the optimized interpreter.
1322          */
1323         if (!fp->jited)
1324                 fp = bpf_migrate_filter(fp);
1325
1326         return fp;
1327 }
1328
1329 /**
1330  *      bpf_prog_create - create an unattached filter
1331  *      @pfp: the unattached filter that is created
1332  *      @fprog: the filter program
1333  *
1334  * Create a filter independent of any socket. We first run some
1335  * sanity checks on it to make sure it does not explode on us later.
1336  * If an error occurs or there is insufficient memory for the filter
1337  * a negative errno code is returned. On success the return is zero.
1338  */
1339 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1340 {
1341         unsigned int fsize = bpf_classic_proglen(fprog);
1342         struct bpf_prog *fp;
1343
1344         /* Make sure new filter is there and in the right amounts. */
1345         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1346                 return -EINVAL;
1347
1348         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1349         if (!fp)
1350                 return -ENOMEM;
1351
1352         memcpy(fp->insns, fprog->filter, fsize);
1353
1354         fp->len = fprog->len;
1355         /* Since unattached filters are not copied back to user
1356          * space through sk_get_filter(), we do not need to hold
1357          * a copy here, and can spare us the work.
1358          */
1359         fp->orig_prog = NULL;
1360
1361         /* bpf_prepare_filter() already takes care of freeing
1362          * memory in case something goes wrong.
1363          */
1364         fp = bpf_prepare_filter(fp, NULL);
1365         if (IS_ERR(fp))
1366                 return PTR_ERR(fp);
1367
1368         *pfp = fp;
1369         return 0;
1370 }
1371 EXPORT_SYMBOL_GPL(bpf_prog_create);
1372
1373 /**
1374  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1375  *      @pfp: the unattached filter that is created
1376  *      @fprog: the filter program
1377  *      @trans: post-classic verifier transformation handler
1378  *      @save_orig: save classic BPF program
1379  *
1380  * This function effectively does the same as bpf_prog_create(), only
1381  * that it builds up its insns buffer from user space provided buffer.
1382  * It also allows for passing a bpf_aux_classic_check_t handler.
1383  */
1384 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1385                               bpf_aux_classic_check_t trans, bool save_orig)
1386 {
1387         unsigned int fsize = bpf_classic_proglen(fprog);
1388         struct bpf_prog *fp;
1389         int err;
1390
1391         /* Make sure new filter is there and in the right amounts. */
1392         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1393                 return -EINVAL;
1394
1395         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1396         if (!fp)
1397                 return -ENOMEM;
1398
1399         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1400                 __bpf_prog_free(fp);
1401                 return -EFAULT;
1402         }
1403
1404         fp->len = fprog->len;
1405         fp->orig_prog = NULL;
1406
1407         if (save_orig) {
1408                 err = bpf_prog_store_orig_filter(fp, fprog);
1409                 if (err) {
1410                         __bpf_prog_free(fp);
1411                         return -ENOMEM;
1412                 }
1413         }
1414
1415         /* bpf_prepare_filter() already takes care of freeing
1416          * memory in case something goes wrong.
1417          */
1418         fp = bpf_prepare_filter(fp, trans);
1419         if (IS_ERR(fp))
1420                 return PTR_ERR(fp);
1421
1422         *pfp = fp;
1423         return 0;
1424 }
1425 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1426
1427 void bpf_prog_destroy(struct bpf_prog *fp)
1428 {
1429         __bpf_prog_release(fp);
1430 }
1431 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1432
1433 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1434 {
1435         struct sk_filter *fp, *old_fp;
1436
1437         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1438         if (!fp)
1439                 return -ENOMEM;
1440
1441         fp->prog = prog;
1442
1443         if (!__sk_filter_charge(sk, fp)) {
1444                 kfree(fp);
1445                 return -ENOMEM;
1446         }
1447         refcount_set(&fp->refcnt, 1);
1448
1449         old_fp = rcu_dereference_protected(sk->sk_filter,
1450                                            lockdep_sock_is_held(sk));
1451         rcu_assign_pointer(sk->sk_filter, fp);
1452
1453         if (old_fp)
1454                 sk_filter_uncharge(sk, old_fp);
1455
1456         return 0;
1457 }
1458
1459 static
1460 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1461 {
1462         unsigned int fsize = bpf_classic_proglen(fprog);
1463         struct bpf_prog *prog;
1464         int err;
1465
1466         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1467                 return ERR_PTR(-EPERM);
1468
1469         /* Make sure new filter is there and in the right amounts. */
1470         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1471                 return ERR_PTR(-EINVAL);
1472
1473         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1474         if (!prog)
1475                 return ERR_PTR(-ENOMEM);
1476
1477         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1478                 __bpf_prog_free(prog);
1479                 return ERR_PTR(-EFAULT);
1480         }
1481
1482         prog->len = fprog->len;
1483
1484         err = bpf_prog_store_orig_filter(prog, fprog);
1485         if (err) {
1486                 __bpf_prog_free(prog);
1487                 return ERR_PTR(-ENOMEM);
1488         }
1489
1490         /* bpf_prepare_filter() already takes care of freeing
1491          * memory in case something goes wrong.
1492          */
1493         return bpf_prepare_filter(prog, NULL);
1494 }
1495
1496 /**
1497  *      sk_attach_filter - attach a socket filter
1498  *      @fprog: the filter program
1499  *      @sk: the socket to use
1500  *
1501  * Attach the user's filter code. We first run some sanity checks on
1502  * it to make sure it does not explode on us later. If an error
1503  * occurs or there is insufficient memory for the filter a negative
1504  * errno code is returned. On success the return is zero.
1505  */
1506 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1507 {
1508         struct bpf_prog *prog = __get_filter(fprog, sk);
1509         int err;
1510
1511         if (IS_ERR(prog))
1512                 return PTR_ERR(prog);
1513
1514         err = __sk_attach_prog(prog, sk);
1515         if (err < 0) {
1516                 __bpf_prog_release(prog);
1517                 return err;
1518         }
1519
1520         return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(sk_attach_filter);
1523
1524 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1525 {
1526         struct bpf_prog *prog = __get_filter(fprog, sk);
1527         int err;
1528
1529         if (IS_ERR(prog))
1530                 return PTR_ERR(prog);
1531
1532         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1533                 err = -ENOMEM;
1534         else
1535                 err = reuseport_attach_prog(sk, prog);
1536
1537         if (err)
1538                 __bpf_prog_release(prog);
1539
1540         return err;
1541 }
1542
1543 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1544 {
1545         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1546                 return ERR_PTR(-EPERM);
1547
1548         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1549 }
1550
1551 int sk_attach_bpf(u32 ufd, struct sock *sk)
1552 {
1553         struct bpf_prog *prog = __get_bpf(ufd, sk);
1554         int err;
1555
1556         if (IS_ERR(prog))
1557                 return PTR_ERR(prog);
1558
1559         err = __sk_attach_prog(prog, sk);
1560         if (err < 0) {
1561                 bpf_prog_put(prog);
1562                 return err;
1563         }
1564
1565         return 0;
1566 }
1567
1568 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1569 {
1570         struct bpf_prog *prog;
1571         int err;
1572
1573         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1574                 return -EPERM;
1575
1576         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1577         if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1578                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1579         if (IS_ERR(prog))
1580                 return PTR_ERR(prog);
1581
1582         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1583                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1584                  * bpf prog (e.g. sockmap).  It depends on the
1585                  * limitation imposed by bpf_prog_load().
1586                  * Hence, sysctl_optmem_max is not checked.
1587                  */
1588                 if ((sk->sk_type != SOCK_STREAM &&
1589                      sk->sk_type != SOCK_DGRAM) ||
1590                     (sk->sk_protocol != IPPROTO_UDP &&
1591                      sk->sk_protocol != IPPROTO_TCP) ||
1592                     (sk->sk_family != AF_INET &&
1593                      sk->sk_family != AF_INET6)) {
1594                         err = -ENOTSUPP;
1595                         goto err_prog_put;
1596                 }
1597         } else {
1598                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1599                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1600                         err = -ENOMEM;
1601                         goto err_prog_put;
1602                 }
1603         }
1604
1605         err = reuseport_attach_prog(sk, prog);
1606 err_prog_put:
1607         if (err)
1608                 bpf_prog_put(prog);
1609
1610         return err;
1611 }
1612
1613 void sk_reuseport_prog_free(struct bpf_prog *prog)
1614 {
1615         if (!prog)
1616                 return;
1617
1618         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1619                 bpf_prog_put(prog);
1620         else
1621                 bpf_prog_destroy(prog);
1622 }
1623
1624 struct bpf_scratchpad {
1625         union {
1626                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1627                 u8     buff[MAX_BPF_STACK];
1628         };
1629 };
1630
1631 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1632
1633 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1634                                           unsigned int write_len)
1635 {
1636         return skb_ensure_writable(skb, write_len);
1637 }
1638
1639 static inline int bpf_try_make_writable(struct sk_buff *skb,
1640                                         unsigned int write_len)
1641 {
1642         int err = __bpf_try_make_writable(skb, write_len);
1643
1644         bpf_compute_data_pointers(skb);
1645         return err;
1646 }
1647
1648 static int bpf_try_make_head_writable(struct sk_buff *skb)
1649 {
1650         return bpf_try_make_writable(skb, skb_headlen(skb));
1651 }
1652
1653 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1654 {
1655         if (skb_at_tc_ingress(skb))
1656                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1657 }
1658
1659 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1660 {
1661         if (skb_at_tc_ingress(skb))
1662                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1663 }
1664
1665 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1666            const void *, from, u32, len, u64, flags)
1667 {
1668         void *ptr;
1669
1670         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1671                 return -EINVAL;
1672         if (unlikely(offset > 0xffff))
1673                 return -EFAULT;
1674         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1675                 return -EFAULT;
1676
1677         ptr = skb->data + offset;
1678         if (flags & BPF_F_RECOMPUTE_CSUM)
1679                 __skb_postpull_rcsum(skb, ptr, len, offset);
1680
1681         memcpy(ptr, from, len);
1682
1683         if (flags & BPF_F_RECOMPUTE_CSUM)
1684                 __skb_postpush_rcsum(skb, ptr, len, offset);
1685         if (flags & BPF_F_INVALIDATE_HASH)
1686                 skb_clear_hash(skb);
1687
1688         return 0;
1689 }
1690
1691 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1692         .func           = bpf_skb_store_bytes,
1693         .gpl_only       = false,
1694         .ret_type       = RET_INTEGER,
1695         .arg1_type      = ARG_PTR_TO_CTX,
1696         .arg2_type      = ARG_ANYTHING,
1697         .arg3_type      = ARG_PTR_TO_MEM,
1698         .arg4_type      = ARG_CONST_SIZE,
1699         .arg5_type      = ARG_ANYTHING,
1700 };
1701
1702 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1703            void *, to, u32, len)
1704 {
1705         void *ptr;
1706
1707         if (unlikely(offset > 0xffff))
1708                 goto err_clear;
1709
1710         ptr = skb_header_pointer(skb, offset, len, to);
1711         if (unlikely(!ptr))
1712                 goto err_clear;
1713         if (ptr != to)
1714                 memcpy(to, ptr, len);
1715
1716         return 0;
1717 err_clear:
1718         memset(to, 0, len);
1719         return -EFAULT;
1720 }
1721
1722 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1723         .func           = bpf_skb_load_bytes,
1724         .gpl_only       = false,
1725         .ret_type       = RET_INTEGER,
1726         .arg1_type      = ARG_PTR_TO_CTX,
1727         .arg2_type      = ARG_ANYTHING,
1728         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1729         .arg4_type      = ARG_CONST_SIZE,
1730 };
1731
1732 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1733            u32, offset, void *, to, u32, len, u32, start_header)
1734 {
1735         u8 *end = skb_tail_pointer(skb);
1736         u8 *net = skb_network_header(skb);
1737         u8 *mac = skb_mac_header(skb);
1738         u8 *ptr;
1739
1740         if (unlikely(offset > 0xffff || len > (end - mac)))
1741                 goto err_clear;
1742
1743         switch (start_header) {
1744         case BPF_HDR_START_MAC:
1745                 ptr = mac + offset;
1746                 break;
1747         case BPF_HDR_START_NET:
1748                 ptr = net + offset;
1749                 break;
1750         default:
1751                 goto err_clear;
1752         }
1753
1754         if (likely(ptr >= mac && ptr + len <= end)) {
1755                 memcpy(to, ptr, len);
1756                 return 0;
1757         }
1758
1759 err_clear:
1760         memset(to, 0, len);
1761         return -EFAULT;
1762 }
1763
1764 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1765         .func           = bpf_skb_load_bytes_relative,
1766         .gpl_only       = false,
1767         .ret_type       = RET_INTEGER,
1768         .arg1_type      = ARG_PTR_TO_CTX,
1769         .arg2_type      = ARG_ANYTHING,
1770         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1771         .arg4_type      = ARG_CONST_SIZE,
1772         .arg5_type      = ARG_ANYTHING,
1773 };
1774
1775 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1776 {
1777         /* Idea is the following: should the needed direct read/write
1778          * test fail during runtime, we can pull in more data and redo
1779          * again, since implicitly, we invalidate previous checks here.
1780          *
1781          * Or, since we know how much we need to make read/writeable,
1782          * this can be done once at the program beginning for direct
1783          * access case. By this we overcome limitations of only current
1784          * headroom being accessible.
1785          */
1786         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1787 }
1788
1789 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1790         .func           = bpf_skb_pull_data,
1791         .gpl_only       = false,
1792         .ret_type       = RET_INTEGER,
1793         .arg1_type      = ARG_PTR_TO_CTX,
1794         .arg2_type      = ARG_ANYTHING,
1795 };
1796
1797 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1798 {
1799         sk = sk_to_full_sk(sk);
1800
1801         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1802 }
1803
1804 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1805         .func           = bpf_sk_fullsock,
1806         .gpl_only       = false,
1807         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1808         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1809 };
1810
1811 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1812                                            unsigned int write_len)
1813 {
1814         int err = __bpf_try_make_writable(skb, write_len);
1815
1816         bpf_compute_data_end_sk_skb(skb);
1817         return err;
1818 }
1819
1820 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1821 {
1822         /* Idea is the following: should the needed direct read/write
1823          * test fail during runtime, we can pull in more data and redo
1824          * again, since implicitly, we invalidate previous checks here.
1825          *
1826          * Or, since we know how much we need to make read/writeable,
1827          * this can be done once at the program beginning for direct
1828          * access case. By this we overcome limitations of only current
1829          * headroom being accessible.
1830          */
1831         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1832 }
1833
1834 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1835         .func           = sk_skb_pull_data,
1836         .gpl_only       = false,
1837         .ret_type       = RET_INTEGER,
1838         .arg1_type      = ARG_PTR_TO_CTX,
1839         .arg2_type      = ARG_ANYTHING,
1840 };
1841
1842 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1843            u64, from, u64, to, u64, flags)
1844 {
1845         __sum16 *ptr;
1846
1847         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1848                 return -EINVAL;
1849         if (unlikely(offset > 0xffff || offset & 1))
1850                 return -EFAULT;
1851         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1852                 return -EFAULT;
1853
1854         ptr = (__sum16 *)(skb->data + offset);
1855         switch (flags & BPF_F_HDR_FIELD_MASK) {
1856         case 0:
1857                 if (unlikely(from != 0))
1858                         return -EINVAL;
1859
1860                 csum_replace_by_diff(ptr, to);
1861                 break;
1862         case 2:
1863                 csum_replace2(ptr, from, to);
1864                 break;
1865         case 4:
1866                 csum_replace4(ptr, from, to);
1867                 break;
1868         default:
1869                 return -EINVAL;
1870         }
1871
1872         return 0;
1873 }
1874
1875 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1876         .func           = bpf_l3_csum_replace,
1877         .gpl_only       = false,
1878         .ret_type       = RET_INTEGER,
1879         .arg1_type      = ARG_PTR_TO_CTX,
1880         .arg2_type      = ARG_ANYTHING,
1881         .arg3_type      = ARG_ANYTHING,
1882         .arg4_type      = ARG_ANYTHING,
1883         .arg5_type      = ARG_ANYTHING,
1884 };
1885
1886 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1887            u64, from, u64, to, u64, flags)
1888 {
1889         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1890         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1891         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1892         __sum16 *ptr;
1893
1894         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1895                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1896                 return -EINVAL;
1897         if (unlikely(offset > 0xffff || offset & 1))
1898                 return -EFAULT;
1899         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1900                 return -EFAULT;
1901
1902         ptr = (__sum16 *)(skb->data + offset);
1903         if (is_mmzero && !do_mforce && !*ptr)
1904                 return 0;
1905
1906         switch (flags & BPF_F_HDR_FIELD_MASK) {
1907         case 0:
1908                 if (unlikely(from != 0))
1909                         return -EINVAL;
1910
1911                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1912                 break;
1913         case 2:
1914                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1915                 break;
1916         case 4:
1917                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1918                 break;
1919         default:
1920                 return -EINVAL;
1921         }
1922
1923         if (is_mmzero && !*ptr)
1924                 *ptr = CSUM_MANGLED_0;
1925         return 0;
1926 }
1927
1928 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1929         .func           = bpf_l4_csum_replace,
1930         .gpl_only       = false,
1931         .ret_type       = RET_INTEGER,
1932         .arg1_type      = ARG_PTR_TO_CTX,
1933         .arg2_type      = ARG_ANYTHING,
1934         .arg3_type      = ARG_ANYTHING,
1935         .arg4_type      = ARG_ANYTHING,
1936         .arg5_type      = ARG_ANYTHING,
1937 };
1938
1939 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1940            __be32 *, to, u32, to_size, __wsum, seed)
1941 {
1942         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1943         u32 diff_size = from_size + to_size;
1944         int i, j = 0;
1945
1946         /* This is quite flexible, some examples:
1947          *
1948          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1949          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1950          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
1951          *
1952          * Even for diffing, from_size and to_size don't need to be equal.
1953          */
1954         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1955                      diff_size > sizeof(sp->diff)))
1956                 return -EINVAL;
1957
1958         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1959                 sp->diff[j] = ~from[i];
1960         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
1961                 sp->diff[j] = to[i];
1962
1963         return csum_partial(sp->diff, diff_size, seed);
1964 }
1965
1966 static const struct bpf_func_proto bpf_csum_diff_proto = {
1967         .func           = bpf_csum_diff,
1968         .gpl_only       = false,
1969         .pkt_access     = true,
1970         .ret_type       = RET_INTEGER,
1971         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1972         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1973         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
1974         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
1975         .arg5_type      = ARG_ANYTHING,
1976 };
1977
1978 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1979 {
1980         /* The interface is to be used in combination with bpf_csum_diff()
1981          * for direct packet writes. csum rotation for alignment as well
1982          * as emulating csum_sub() can be done from the eBPF program.
1983          */
1984         if (skb->ip_summed == CHECKSUM_COMPLETE)
1985                 return (skb->csum = csum_add(skb->csum, csum));
1986
1987         return -ENOTSUPP;
1988 }
1989
1990 static const struct bpf_func_proto bpf_csum_update_proto = {
1991         .func           = bpf_csum_update,
1992         .gpl_only       = false,
1993         .ret_type       = RET_INTEGER,
1994         .arg1_type      = ARG_PTR_TO_CTX,
1995         .arg2_type      = ARG_ANYTHING,
1996 };
1997
1998 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1999 {
2000         return dev_forward_skb(dev, skb);
2001 }
2002
2003 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2004                                       struct sk_buff *skb)
2005 {
2006         int ret = ____dev_forward_skb(dev, skb);
2007
2008         if (likely(!ret)) {
2009                 skb->dev = dev;
2010                 ret = netif_rx(skb);
2011         }
2012
2013         return ret;
2014 }
2015
2016 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2017 {
2018         int ret;
2019
2020         if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2021                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2022                 kfree_skb(skb);
2023                 return -ENETDOWN;
2024         }
2025
2026         skb->dev = dev;
2027
2028         __this_cpu_inc(xmit_recursion);
2029         ret = dev_queue_xmit(skb);
2030         __this_cpu_dec(xmit_recursion);
2031
2032         return ret;
2033 }
2034
2035 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2036                                  u32 flags)
2037 {
2038         unsigned int mlen = skb_network_offset(skb);
2039
2040         if (mlen) {
2041                 __skb_pull(skb, mlen);
2042
2043                 /* At ingress, the mac header has already been pulled once.
2044                  * At egress, skb_pospull_rcsum has to be done in case that
2045                  * the skb is originated from ingress (i.e. a forwarded skb)
2046                  * to ensure that rcsum starts at net header.
2047                  */
2048                 if (!skb_at_tc_ingress(skb))
2049                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2050         }
2051         skb_pop_mac_header(skb);
2052         skb_reset_mac_len(skb);
2053         return flags & BPF_F_INGRESS ?
2054                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2055 }
2056
2057 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2058                                  u32 flags)
2059 {
2060         /* Verify that a link layer header is carried */
2061         if (unlikely(skb->mac_header >= skb->network_header)) {
2062                 kfree_skb(skb);
2063                 return -ERANGE;
2064         }
2065
2066         bpf_push_mac_rcsum(skb);
2067         return flags & BPF_F_INGRESS ?
2068                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2069 }
2070
2071 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2072                           u32 flags)
2073 {
2074         if (dev_is_mac_header_xmit(dev))
2075                 return __bpf_redirect_common(skb, dev, flags);
2076         else
2077                 return __bpf_redirect_no_mac(skb, dev, flags);
2078 }
2079
2080 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2081 {
2082         struct net_device *dev;
2083         struct sk_buff *clone;
2084         int ret;
2085
2086         if (unlikely(flags & ~(BPF_F_INGRESS)))
2087                 return -EINVAL;
2088
2089         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2090         if (unlikely(!dev))
2091                 return -EINVAL;
2092
2093         clone = skb_clone(skb, GFP_ATOMIC);
2094         if (unlikely(!clone))
2095                 return -ENOMEM;
2096
2097         /* For direct write, we need to keep the invariant that the skbs
2098          * we're dealing with need to be uncloned. Should uncloning fail
2099          * here, we need to free the just generated clone to unclone once
2100          * again.
2101          */
2102         ret = bpf_try_make_head_writable(skb);
2103         if (unlikely(ret)) {
2104                 kfree_skb(clone);
2105                 return -ENOMEM;
2106         }
2107
2108         return __bpf_redirect(clone, dev, flags);
2109 }
2110
2111 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2112         .func           = bpf_clone_redirect,
2113         .gpl_only       = false,
2114         .ret_type       = RET_INTEGER,
2115         .arg1_type      = ARG_PTR_TO_CTX,
2116         .arg2_type      = ARG_ANYTHING,
2117         .arg3_type      = ARG_ANYTHING,
2118 };
2119
2120 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2121 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2122
2123 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2124 {
2125         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2126
2127         if (unlikely(flags & ~(BPF_F_INGRESS)))
2128                 return TC_ACT_SHOT;
2129
2130         ri->ifindex = ifindex;
2131         ri->flags = flags;
2132
2133         return TC_ACT_REDIRECT;
2134 }
2135
2136 int skb_do_redirect(struct sk_buff *skb)
2137 {
2138         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2139         struct net_device *dev;
2140
2141         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2142         ri->ifindex = 0;
2143         if (unlikely(!dev)) {
2144                 kfree_skb(skb);
2145                 return -EINVAL;
2146         }
2147
2148         return __bpf_redirect(skb, dev, ri->flags);
2149 }
2150
2151 static const struct bpf_func_proto bpf_redirect_proto = {
2152         .func           = bpf_redirect,
2153         .gpl_only       = false,
2154         .ret_type       = RET_INTEGER,
2155         .arg1_type      = ARG_ANYTHING,
2156         .arg2_type      = ARG_ANYTHING,
2157 };
2158
2159 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2160 {
2161         msg->apply_bytes = bytes;
2162         return 0;
2163 }
2164
2165 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2166         .func           = bpf_msg_apply_bytes,
2167         .gpl_only       = false,
2168         .ret_type       = RET_INTEGER,
2169         .arg1_type      = ARG_PTR_TO_CTX,
2170         .arg2_type      = ARG_ANYTHING,
2171 };
2172
2173 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2174 {
2175         msg->cork_bytes = bytes;
2176         return 0;
2177 }
2178
2179 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2180         .func           = bpf_msg_cork_bytes,
2181         .gpl_only       = false,
2182         .ret_type       = RET_INTEGER,
2183         .arg1_type      = ARG_PTR_TO_CTX,
2184         .arg2_type      = ARG_ANYTHING,
2185 };
2186
2187 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2188            u32, end, u64, flags)
2189 {
2190         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2191         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2192         struct scatterlist *sge;
2193         u8 *raw, *to, *from;
2194         struct page *page;
2195
2196         if (unlikely(flags || end <= start))
2197                 return -EINVAL;
2198
2199         /* First find the starting scatterlist element */
2200         i = msg->sg.start;
2201         do {
2202                 len = sk_msg_elem(msg, i)->length;
2203                 if (start < offset + len)
2204                         break;
2205                 offset += len;
2206                 sk_msg_iter_var_next(i);
2207         } while (i != msg->sg.end);
2208
2209         if (unlikely(start >= offset + len))
2210                 return -EINVAL;
2211
2212         first_sge = i;
2213         /* The start may point into the sg element so we need to also
2214          * account for the headroom.
2215          */
2216         bytes_sg_total = start - offset + bytes;
2217         if (!msg->sg.copy[i] && bytes_sg_total <= len)
2218                 goto out;
2219
2220         /* At this point we need to linearize multiple scatterlist
2221          * elements or a single shared page. Either way we need to
2222          * copy into a linear buffer exclusively owned by BPF. Then
2223          * place the buffer in the scatterlist and fixup the original
2224          * entries by removing the entries now in the linear buffer
2225          * and shifting the remaining entries. For now we do not try
2226          * to copy partial entries to avoid complexity of running out
2227          * of sg_entry slots. The downside is reading a single byte
2228          * will copy the entire sg entry.
2229          */
2230         do {
2231                 copy += sk_msg_elem(msg, i)->length;
2232                 sk_msg_iter_var_next(i);
2233                 if (bytes_sg_total <= copy)
2234                         break;
2235         } while (i != msg->sg.end);
2236         last_sge = i;
2237
2238         if (unlikely(bytes_sg_total > copy))
2239                 return -EINVAL;
2240
2241         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2242                            get_order(copy));
2243         if (unlikely(!page))
2244                 return -ENOMEM;
2245
2246         raw = page_address(page);
2247         i = first_sge;
2248         do {
2249                 sge = sk_msg_elem(msg, i);
2250                 from = sg_virt(sge);
2251                 len = sge->length;
2252                 to = raw + poffset;
2253
2254                 memcpy(to, from, len);
2255                 poffset += len;
2256                 sge->length = 0;
2257                 put_page(sg_page(sge));
2258
2259                 sk_msg_iter_var_next(i);
2260         } while (i != last_sge);
2261
2262         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2263
2264         /* To repair sg ring we need to shift entries. If we only
2265          * had a single entry though we can just replace it and
2266          * be done. Otherwise walk the ring and shift the entries.
2267          */
2268         WARN_ON_ONCE(last_sge == first_sge);
2269         shift = last_sge > first_sge ?
2270                 last_sge - first_sge - 1 :
2271                 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2272         if (!shift)
2273                 goto out;
2274
2275         i = first_sge;
2276         sk_msg_iter_var_next(i);
2277         do {
2278                 u32 move_from;
2279
2280                 if (i + shift >= MAX_MSG_FRAGS)
2281                         move_from = i + shift - MAX_MSG_FRAGS;
2282                 else
2283                         move_from = i + shift;
2284                 if (move_from == msg->sg.end)
2285                         break;
2286
2287                 msg->sg.data[i] = msg->sg.data[move_from];
2288                 msg->sg.data[move_from].length = 0;
2289                 msg->sg.data[move_from].page_link = 0;
2290                 msg->sg.data[move_from].offset = 0;
2291                 sk_msg_iter_var_next(i);
2292         } while (1);
2293
2294         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2295                       msg->sg.end - shift + MAX_MSG_FRAGS :
2296                       msg->sg.end - shift;
2297 out:
2298         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2299         msg->data_end = msg->data + bytes;
2300         return 0;
2301 }
2302
2303 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2304         .func           = bpf_msg_pull_data,
2305         .gpl_only       = false,
2306         .ret_type       = RET_INTEGER,
2307         .arg1_type      = ARG_PTR_TO_CTX,
2308         .arg2_type      = ARG_ANYTHING,
2309         .arg3_type      = ARG_ANYTHING,
2310         .arg4_type      = ARG_ANYTHING,
2311 };
2312
2313 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2314            u32, len, u64, flags)
2315 {
2316         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2317         u32 new, i = 0, l, space, copy = 0, offset = 0;
2318         u8 *raw, *to, *from;
2319         struct page *page;
2320
2321         if (unlikely(flags))
2322                 return -EINVAL;
2323
2324         /* First find the starting scatterlist element */
2325         i = msg->sg.start;
2326         do {
2327                 l = sk_msg_elem(msg, i)->length;
2328
2329                 if (start < offset + l)
2330                         break;
2331                 offset += l;
2332                 sk_msg_iter_var_next(i);
2333         } while (i != msg->sg.end);
2334
2335         if (start >= offset + l)
2336                 return -EINVAL;
2337
2338         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2339
2340         /* If no space available will fallback to copy, we need at
2341          * least one scatterlist elem available to push data into
2342          * when start aligns to the beginning of an element or two
2343          * when it falls inside an element. We handle the start equals
2344          * offset case because its the common case for inserting a
2345          * header.
2346          */
2347         if (!space || (space == 1 && start != offset))
2348                 copy = msg->sg.data[i].length;
2349
2350         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2351                            get_order(copy + len));
2352         if (unlikely(!page))
2353                 return -ENOMEM;
2354
2355         if (copy) {
2356                 int front, back;
2357
2358                 raw = page_address(page);
2359
2360                 psge = sk_msg_elem(msg, i);
2361                 front = start - offset;
2362                 back = psge->length - front;
2363                 from = sg_virt(psge);
2364
2365                 if (front)
2366                         memcpy(raw, from, front);
2367
2368                 if (back) {
2369                         from += front;
2370                         to = raw + front + len;
2371
2372                         memcpy(to, from, back);
2373                 }
2374
2375                 put_page(sg_page(psge));
2376         } else if (start - offset) {
2377                 psge = sk_msg_elem(msg, i);
2378                 rsge = sk_msg_elem_cpy(msg, i);
2379
2380                 psge->length = start - offset;
2381                 rsge.length -= psge->length;
2382                 rsge.offset += start;
2383
2384                 sk_msg_iter_var_next(i);
2385                 sg_unmark_end(psge);
2386                 sk_msg_iter_next(msg, end);
2387         }
2388
2389         /* Slot(s) to place newly allocated data */
2390         new = i;
2391
2392         /* Shift one or two slots as needed */
2393         if (!copy) {
2394                 sge = sk_msg_elem_cpy(msg, i);
2395
2396                 sk_msg_iter_var_next(i);
2397                 sg_unmark_end(&sge);
2398                 sk_msg_iter_next(msg, end);
2399
2400                 nsge = sk_msg_elem_cpy(msg, i);
2401                 if (rsge.length) {
2402                         sk_msg_iter_var_next(i);
2403                         nnsge = sk_msg_elem_cpy(msg, i);
2404                 }
2405
2406                 while (i != msg->sg.end) {
2407                         msg->sg.data[i] = sge;
2408                         sge = nsge;
2409                         sk_msg_iter_var_next(i);
2410                         if (rsge.length) {
2411                                 nsge = nnsge;
2412                                 nnsge = sk_msg_elem_cpy(msg, i);
2413                         } else {
2414                                 nsge = sk_msg_elem_cpy(msg, i);
2415                         }
2416                 }
2417         }
2418
2419         /* Place newly allocated data buffer */
2420         sk_mem_charge(msg->sk, len);
2421         msg->sg.size += len;
2422         msg->sg.copy[new] = false;
2423         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2424         if (rsge.length) {
2425                 get_page(sg_page(&rsge));
2426                 sk_msg_iter_var_next(new);
2427                 msg->sg.data[new] = rsge;
2428         }
2429
2430         sk_msg_compute_data_pointers(msg);
2431         return 0;
2432 }
2433
2434 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2435         .func           = bpf_msg_push_data,
2436         .gpl_only       = false,
2437         .ret_type       = RET_INTEGER,
2438         .arg1_type      = ARG_PTR_TO_CTX,
2439         .arg2_type      = ARG_ANYTHING,
2440         .arg3_type      = ARG_ANYTHING,
2441         .arg4_type      = ARG_ANYTHING,
2442 };
2443
2444 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2445 {
2446         int prev;
2447
2448         do {
2449                 prev = i;
2450                 sk_msg_iter_var_next(i);
2451                 msg->sg.data[prev] = msg->sg.data[i];
2452         } while (i != msg->sg.end);
2453
2454         sk_msg_iter_prev(msg, end);
2455 }
2456
2457 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2458 {
2459         struct scatterlist tmp, sge;
2460
2461         sk_msg_iter_next(msg, end);
2462         sge = sk_msg_elem_cpy(msg, i);
2463         sk_msg_iter_var_next(i);
2464         tmp = sk_msg_elem_cpy(msg, i);
2465
2466         while (i != msg->sg.end) {
2467                 msg->sg.data[i] = sge;
2468                 sk_msg_iter_var_next(i);
2469                 sge = tmp;
2470                 tmp = sk_msg_elem_cpy(msg, i);
2471         }
2472 }
2473
2474 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2475            u32, len, u64, flags)
2476 {
2477         u32 i = 0, l, space, offset = 0;
2478         u64 last = start + len;
2479         int pop;
2480
2481         if (unlikely(flags))
2482                 return -EINVAL;
2483
2484         /* First find the starting scatterlist element */
2485         i = msg->sg.start;
2486         do {
2487                 l = sk_msg_elem(msg, i)->length;
2488
2489                 if (start < offset + l)
2490                         break;
2491                 offset += l;
2492                 sk_msg_iter_var_next(i);
2493         } while (i != msg->sg.end);
2494
2495         /* Bounds checks: start and pop must be inside message */
2496         if (start >= offset + l || last >= msg->sg.size)
2497                 return -EINVAL;
2498
2499         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2500
2501         pop = len;
2502         /* --------------| offset
2503          * -| start      |-------- len -------|
2504          *
2505          *  |----- a ----|-------- pop -------|----- b ----|
2506          *  |______________________________________________| length
2507          *
2508          *
2509          * a:   region at front of scatter element to save
2510          * b:   region at back of scatter element to save when length > A + pop
2511          * pop: region to pop from element, same as input 'pop' here will be
2512          *      decremented below per iteration.
2513          *
2514          * Two top-level cases to handle when start != offset, first B is non
2515          * zero and second B is zero corresponding to when a pop includes more
2516          * than one element.
2517          *
2518          * Then if B is non-zero AND there is no space allocate space and
2519          * compact A, B regions into page. If there is space shift ring to
2520          * the rigth free'ing the next element in ring to place B, leaving
2521          * A untouched except to reduce length.
2522          */
2523         if (start != offset) {
2524                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2525                 int a = start;
2526                 int b = sge->length - pop - a;
2527
2528                 sk_msg_iter_var_next(i);
2529
2530                 if (pop < sge->length - a) {
2531                         if (space) {
2532                                 sge->length = a;
2533                                 sk_msg_shift_right(msg, i);
2534                                 nsge = sk_msg_elem(msg, i);
2535                                 get_page(sg_page(sge));
2536                                 sg_set_page(nsge,
2537                                             sg_page(sge),
2538                                             b, sge->offset + pop + a);
2539                         } else {
2540                                 struct page *page, *orig;
2541                                 u8 *to, *from;
2542
2543                                 page = alloc_pages(__GFP_NOWARN |
2544                                                    __GFP_COMP   | GFP_ATOMIC,
2545                                                    get_order(a + b));
2546                                 if (unlikely(!page))
2547                                         return -ENOMEM;
2548
2549                                 sge->length = a;
2550                                 orig = sg_page(sge);
2551                                 from = sg_virt(sge);
2552                                 to = page_address(page);
2553                                 memcpy(to, from, a);
2554                                 memcpy(to + a, from + a + pop, b);
2555                                 sg_set_page(sge, page, a + b, 0);
2556                                 put_page(orig);
2557                         }
2558                         pop = 0;
2559                 } else if (pop >= sge->length - a) {
2560                         sge->length = a;
2561                         pop -= (sge->length - a);
2562                 }
2563         }
2564
2565         /* From above the current layout _must_ be as follows,
2566          *
2567          * -| offset
2568          * -| start
2569          *
2570          *  |---- pop ---|---------------- b ------------|
2571          *  |____________________________________________| length
2572          *
2573          * Offset and start of the current msg elem are equal because in the
2574          * previous case we handled offset != start and either consumed the
2575          * entire element and advanced to the next element OR pop == 0.
2576          *
2577          * Two cases to handle here are first pop is less than the length
2578          * leaving some remainder b above. Simply adjust the element's layout
2579          * in this case. Or pop >= length of the element so that b = 0. In this
2580          * case advance to next element decrementing pop.
2581          */
2582         while (pop) {
2583                 struct scatterlist *sge = sk_msg_elem(msg, i);
2584
2585                 if (pop < sge->length) {
2586                         sge->length -= pop;
2587                         sge->offset += pop;
2588                         pop = 0;
2589                 } else {
2590                         pop -= sge->length;
2591                         sk_msg_shift_left(msg, i);
2592                 }
2593                 sk_msg_iter_var_next(i);
2594         }
2595
2596         sk_mem_uncharge(msg->sk, len - pop);
2597         msg->sg.size -= (len - pop);
2598         sk_msg_compute_data_pointers(msg);
2599         return 0;
2600 }
2601
2602 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2603         .func           = bpf_msg_pop_data,
2604         .gpl_only       = false,
2605         .ret_type       = RET_INTEGER,
2606         .arg1_type      = ARG_PTR_TO_CTX,
2607         .arg2_type      = ARG_ANYTHING,
2608         .arg3_type      = ARG_ANYTHING,
2609         .arg4_type      = ARG_ANYTHING,
2610 };
2611
2612 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2613 {
2614         return task_get_classid(skb);
2615 }
2616
2617 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2618         .func           = bpf_get_cgroup_classid,
2619         .gpl_only       = false,
2620         .ret_type       = RET_INTEGER,
2621         .arg1_type      = ARG_PTR_TO_CTX,
2622 };
2623
2624 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2625 {
2626         return dst_tclassid(skb);
2627 }
2628
2629 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2630         .func           = bpf_get_route_realm,
2631         .gpl_only       = false,
2632         .ret_type       = RET_INTEGER,
2633         .arg1_type      = ARG_PTR_TO_CTX,
2634 };
2635
2636 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2637 {
2638         /* If skb_clear_hash() was called due to mangling, we can
2639          * trigger SW recalculation here. Later access to hash
2640          * can then use the inline skb->hash via context directly
2641          * instead of calling this helper again.
2642          */
2643         return skb_get_hash(skb);
2644 }
2645
2646 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2647         .func           = bpf_get_hash_recalc,
2648         .gpl_only       = false,
2649         .ret_type       = RET_INTEGER,
2650         .arg1_type      = ARG_PTR_TO_CTX,
2651 };
2652
2653 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2654 {
2655         /* After all direct packet write, this can be used once for
2656          * triggering a lazy recalc on next skb_get_hash() invocation.
2657          */
2658         skb_clear_hash(skb);
2659         return 0;
2660 }
2661
2662 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2663         .func           = bpf_set_hash_invalid,
2664         .gpl_only       = false,
2665         .ret_type       = RET_INTEGER,
2666         .arg1_type      = ARG_PTR_TO_CTX,
2667 };
2668
2669 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2670 {
2671         /* Set user specified hash as L4(+), so that it gets returned
2672          * on skb_get_hash() call unless BPF prog later on triggers a
2673          * skb_clear_hash().
2674          */
2675         __skb_set_sw_hash(skb, hash, true);
2676         return 0;
2677 }
2678
2679 static const struct bpf_func_proto bpf_set_hash_proto = {
2680         .func           = bpf_set_hash,
2681         .gpl_only       = false,
2682         .ret_type       = RET_INTEGER,
2683         .arg1_type      = ARG_PTR_TO_CTX,
2684         .arg2_type      = ARG_ANYTHING,
2685 };
2686
2687 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2688            u16, vlan_tci)
2689 {
2690         int ret;
2691
2692         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2693                      vlan_proto != htons(ETH_P_8021AD)))
2694                 vlan_proto = htons(ETH_P_8021Q);
2695
2696         bpf_push_mac_rcsum(skb);
2697         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2698         bpf_pull_mac_rcsum(skb);
2699
2700         bpf_compute_data_pointers(skb);
2701         return ret;
2702 }
2703
2704 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2705         .func           = bpf_skb_vlan_push,
2706         .gpl_only       = false,
2707         .ret_type       = RET_INTEGER,
2708         .arg1_type      = ARG_PTR_TO_CTX,
2709         .arg2_type      = ARG_ANYTHING,
2710         .arg3_type      = ARG_ANYTHING,
2711 };
2712
2713 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2714 {
2715         int ret;
2716
2717         bpf_push_mac_rcsum(skb);
2718         ret = skb_vlan_pop(skb);
2719         bpf_pull_mac_rcsum(skb);
2720
2721         bpf_compute_data_pointers(skb);
2722         return ret;
2723 }
2724
2725 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2726         .func           = bpf_skb_vlan_pop,
2727         .gpl_only       = false,
2728         .ret_type       = RET_INTEGER,
2729         .arg1_type      = ARG_PTR_TO_CTX,
2730 };
2731
2732 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2733 {
2734         /* Caller already did skb_cow() with len as headroom,
2735          * so no need to do it here.
2736          */
2737         skb_push(skb, len);
2738         memmove(skb->data, skb->data + len, off);
2739         memset(skb->data + off, 0, len);
2740
2741         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2742          * needed here as it does not change the skb->csum
2743          * result for checksum complete when summing over
2744          * zeroed blocks.
2745          */
2746         return 0;
2747 }
2748
2749 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2750 {
2751         /* skb_ensure_writable() is not needed here, as we're
2752          * already working on an uncloned skb.
2753          */
2754         if (unlikely(!pskb_may_pull(skb, off + len)))
2755                 return -ENOMEM;
2756
2757         skb_postpull_rcsum(skb, skb->data + off, len);
2758         memmove(skb->data + len, skb->data, off);
2759         __skb_pull(skb, len);
2760
2761         return 0;
2762 }
2763
2764 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2765 {
2766         bool trans_same = skb->transport_header == skb->network_header;
2767         int ret;
2768
2769         /* There's no need for __skb_push()/__skb_pull() pair to
2770          * get to the start of the mac header as we're guaranteed
2771          * to always start from here under eBPF.
2772          */
2773         ret = bpf_skb_generic_push(skb, off, len);
2774         if (likely(!ret)) {
2775                 skb->mac_header -= len;
2776                 skb->network_header -= len;
2777                 if (trans_same)
2778                         skb->transport_header = skb->network_header;
2779         }
2780
2781         return ret;
2782 }
2783
2784 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2785 {
2786         bool trans_same = skb->transport_header == skb->network_header;
2787         int ret;
2788
2789         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2790         ret = bpf_skb_generic_pop(skb, off, len);
2791         if (likely(!ret)) {
2792                 skb->mac_header += len;
2793                 skb->network_header += len;
2794                 if (trans_same)
2795                         skb->transport_header = skb->network_header;
2796         }
2797
2798         return ret;
2799 }
2800
2801 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2802 {
2803         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2804         u32 off = skb_mac_header_len(skb);
2805         int ret;
2806
2807         if (!skb_is_gso_tcp(skb))
2808                 return -ENOTSUPP;
2809
2810         ret = skb_cow(skb, len_diff);
2811         if (unlikely(ret < 0))
2812                 return ret;
2813
2814         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2815         if (unlikely(ret < 0))
2816                 return ret;
2817
2818         if (skb_is_gso(skb)) {
2819                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2820
2821                 /* SKB_GSO_TCPV4 needs to be changed into
2822                  * SKB_GSO_TCPV6.
2823                  */
2824                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2825                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2826                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2827                 }
2828
2829                 /* Due to IPv6 header, MSS needs to be downgraded. */
2830                 skb_decrease_gso_size(shinfo, len_diff);
2831                 /* Header must be checked, and gso_segs recomputed. */
2832                 shinfo->gso_type |= SKB_GSO_DODGY;
2833                 shinfo->gso_segs = 0;
2834         }
2835
2836         skb->protocol = htons(ETH_P_IPV6);
2837         skb_clear_hash(skb);
2838
2839         return 0;
2840 }
2841
2842 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2843 {
2844         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2845         u32 off = skb_mac_header_len(skb);
2846         int ret;
2847
2848         if (!skb_is_gso_tcp(skb))
2849                 return -ENOTSUPP;
2850
2851         ret = skb_unclone(skb, GFP_ATOMIC);
2852         if (unlikely(ret < 0))
2853                 return ret;
2854
2855         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2856         if (unlikely(ret < 0))
2857                 return ret;
2858
2859         if (skb_is_gso(skb)) {
2860                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2861
2862                 /* SKB_GSO_TCPV6 needs to be changed into
2863                  * SKB_GSO_TCPV4.
2864                  */
2865                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2866                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2867                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2868                 }
2869
2870                 /* Due to IPv4 header, MSS can be upgraded. */
2871                 skb_increase_gso_size(shinfo, len_diff);
2872                 /* Header must be checked, and gso_segs recomputed. */
2873                 shinfo->gso_type |= SKB_GSO_DODGY;
2874                 shinfo->gso_segs = 0;
2875         }
2876
2877         skb->protocol = htons(ETH_P_IP);
2878         skb_clear_hash(skb);
2879
2880         return 0;
2881 }
2882
2883 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2884 {
2885         __be16 from_proto = skb->protocol;
2886
2887         if (from_proto == htons(ETH_P_IP) &&
2888               to_proto == htons(ETH_P_IPV6))
2889                 return bpf_skb_proto_4_to_6(skb);
2890
2891         if (from_proto == htons(ETH_P_IPV6) &&
2892               to_proto == htons(ETH_P_IP))
2893                 return bpf_skb_proto_6_to_4(skb);
2894
2895         return -ENOTSUPP;
2896 }
2897
2898 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2899            u64, flags)
2900 {
2901         int ret;
2902
2903         if (unlikely(flags))
2904                 return -EINVAL;
2905
2906         /* General idea is that this helper does the basic groundwork
2907          * needed for changing the protocol, and eBPF program fills the
2908          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2909          * and other helpers, rather than passing a raw buffer here.
2910          *
2911          * The rationale is to keep this minimal and without a need to
2912          * deal with raw packet data. F.e. even if we would pass buffers
2913          * here, the program still needs to call the bpf_lX_csum_replace()
2914          * helpers anyway. Plus, this way we keep also separation of
2915          * concerns, since f.e. bpf_skb_store_bytes() should only take
2916          * care of stores.
2917          *
2918          * Currently, additional options and extension header space are
2919          * not supported, but flags register is reserved so we can adapt
2920          * that. For offloads, we mark packet as dodgy, so that headers
2921          * need to be verified first.
2922          */
2923         ret = bpf_skb_proto_xlat(skb, proto);
2924         bpf_compute_data_pointers(skb);
2925         return ret;
2926 }
2927
2928 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2929         .func           = bpf_skb_change_proto,
2930         .gpl_only       = false,
2931         .ret_type       = RET_INTEGER,
2932         .arg1_type      = ARG_PTR_TO_CTX,
2933         .arg2_type      = ARG_ANYTHING,
2934         .arg3_type      = ARG_ANYTHING,
2935 };
2936
2937 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2938 {
2939         /* We only allow a restricted subset to be changed for now. */
2940         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2941                      !skb_pkt_type_ok(pkt_type)))
2942                 return -EINVAL;
2943
2944         skb->pkt_type = pkt_type;
2945         return 0;
2946 }
2947
2948 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2949         .func           = bpf_skb_change_type,
2950         .gpl_only       = false,
2951         .ret_type       = RET_INTEGER,
2952         .arg1_type      = ARG_PTR_TO_CTX,
2953         .arg2_type      = ARG_ANYTHING,
2954 };
2955
2956 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2957 {
2958         switch (skb->protocol) {
2959         case htons(ETH_P_IP):
2960                 return sizeof(struct iphdr);
2961         case htons(ETH_P_IPV6):
2962                 return sizeof(struct ipv6hdr);
2963         default:
2964                 return ~0U;
2965         }
2966 }
2967
2968 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2969 {
2970         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2971         int ret;
2972
2973         if (!skb_is_gso_tcp(skb))
2974                 return -ENOTSUPP;
2975
2976         ret = skb_cow(skb, len_diff);
2977         if (unlikely(ret < 0))
2978                 return ret;
2979
2980         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2981         if (unlikely(ret < 0))
2982                 return ret;
2983
2984         if (skb_is_gso(skb)) {
2985                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2986
2987                 /* Due to header grow, MSS needs to be downgraded. */
2988                 skb_decrease_gso_size(shinfo, len_diff);
2989                 /* Header must be checked, and gso_segs recomputed. */
2990                 shinfo->gso_type |= SKB_GSO_DODGY;
2991                 shinfo->gso_segs = 0;
2992         }
2993
2994         return 0;
2995 }
2996
2997 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2998 {
2999         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
3000         int ret;
3001
3002         if (!skb_is_gso_tcp(skb))
3003                 return -ENOTSUPP;
3004
3005         ret = skb_unclone(skb, GFP_ATOMIC);
3006         if (unlikely(ret < 0))
3007                 return ret;
3008
3009         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3010         if (unlikely(ret < 0))
3011                 return ret;
3012
3013         if (skb_is_gso(skb)) {
3014                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3015
3016                 /* Due to header shrink, MSS can be upgraded. */
3017                 skb_increase_gso_size(shinfo, len_diff);
3018                 /* Header must be checked, and gso_segs recomputed. */
3019                 shinfo->gso_type |= SKB_GSO_DODGY;
3020                 shinfo->gso_segs = 0;
3021         }
3022
3023         return 0;
3024 }
3025
3026 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3027 {
3028         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3029                           SKB_MAX_ALLOC;
3030 }
3031
3032 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
3033 {
3034         bool trans_same = skb->transport_header == skb->network_header;
3035         u32 len_cur, len_diff_abs = abs(len_diff);
3036         u32 len_min = bpf_skb_net_base_len(skb);
3037         u32 len_max = __bpf_skb_max_len(skb);
3038         __be16 proto = skb->protocol;
3039         bool shrink = len_diff < 0;
3040         int ret;
3041
3042         if (unlikely(len_diff_abs > 0xfffU))
3043                 return -EFAULT;
3044         if (unlikely(proto != htons(ETH_P_IP) &&
3045                      proto != htons(ETH_P_IPV6)))
3046                 return -ENOTSUPP;
3047
3048         len_cur = skb->len - skb_network_offset(skb);
3049         if (skb_transport_header_was_set(skb) && !trans_same)
3050                 len_cur = skb_network_header_len(skb);
3051         if ((shrink && (len_diff_abs >= len_cur ||
3052                         len_cur - len_diff_abs < len_min)) ||
3053             (!shrink && (skb->len + len_diff_abs > len_max &&
3054                          !skb_is_gso(skb))))
3055                 return -ENOTSUPP;
3056
3057         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
3058                        bpf_skb_net_grow(skb, len_diff_abs);
3059
3060         bpf_compute_data_pointers(skb);
3061         return ret;
3062 }
3063
3064 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3065            u32, mode, u64, flags)
3066 {
3067         if (unlikely(flags))
3068                 return -EINVAL;
3069         if (likely(mode == BPF_ADJ_ROOM_NET))
3070                 return bpf_skb_adjust_net(skb, len_diff);
3071
3072         return -ENOTSUPP;
3073 }
3074
3075 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3076         .func           = bpf_skb_adjust_room,
3077         .gpl_only       = false,
3078         .ret_type       = RET_INTEGER,
3079         .arg1_type      = ARG_PTR_TO_CTX,
3080         .arg2_type      = ARG_ANYTHING,
3081         .arg3_type      = ARG_ANYTHING,
3082         .arg4_type      = ARG_ANYTHING,
3083 };
3084
3085 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3086 {
3087         u32 min_len = skb_network_offset(skb);
3088
3089         if (skb_transport_header_was_set(skb))
3090                 min_len = skb_transport_offset(skb);
3091         if (skb->ip_summed == CHECKSUM_PARTIAL)
3092                 min_len = skb_checksum_start_offset(skb) +
3093                           skb->csum_offset + sizeof(__sum16);
3094         return min_len;
3095 }
3096
3097 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3098 {
3099         unsigned int old_len = skb->len;
3100         int ret;
3101
3102         ret = __skb_grow_rcsum(skb, new_len);
3103         if (!ret)
3104                 memset(skb->data + old_len, 0, new_len - old_len);
3105         return ret;
3106 }
3107
3108 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3109 {
3110         return __skb_trim_rcsum(skb, new_len);
3111 }
3112
3113 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3114                                         u64 flags)
3115 {
3116         u32 max_len = __bpf_skb_max_len(skb);
3117         u32 min_len = __bpf_skb_min_len(skb);
3118         int ret;
3119
3120         if (unlikely(flags || new_len > max_len || new_len < min_len))
3121                 return -EINVAL;
3122         if (skb->encapsulation)
3123                 return -ENOTSUPP;
3124
3125         /* The basic idea of this helper is that it's performing the
3126          * needed work to either grow or trim an skb, and eBPF program
3127          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3128          * bpf_lX_csum_replace() and others rather than passing a raw
3129          * buffer here. This one is a slow path helper and intended
3130          * for replies with control messages.
3131          *
3132          * Like in bpf_skb_change_proto(), we want to keep this rather
3133          * minimal and without protocol specifics so that we are able
3134          * to separate concerns as in bpf_skb_store_bytes() should only
3135          * be the one responsible for writing buffers.
3136          *
3137          * It's really expected to be a slow path operation here for
3138          * control message replies, so we're implicitly linearizing,
3139          * uncloning and drop offloads from the skb by this.
3140          */
3141         ret = __bpf_try_make_writable(skb, skb->len);
3142         if (!ret) {
3143                 if (new_len > skb->len)
3144                         ret = bpf_skb_grow_rcsum(skb, new_len);
3145                 else if (new_len < skb->len)
3146                         ret = bpf_skb_trim_rcsum(skb, new_len);
3147                 if (!ret && skb_is_gso(skb))
3148                         skb_gso_reset(skb);
3149         }
3150         return ret;
3151 }
3152
3153 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3154            u64, flags)
3155 {
3156         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3157
3158         bpf_compute_data_pointers(skb);
3159         return ret;
3160 }
3161
3162 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3163         .func           = bpf_skb_change_tail,
3164         .gpl_only       = false,
3165         .ret_type       = RET_INTEGER,
3166         .arg1_type      = ARG_PTR_TO_CTX,
3167         .arg2_type      = ARG_ANYTHING,
3168         .arg3_type      = ARG_ANYTHING,
3169 };
3170
3171 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3172            u64, flags)
3173 {
3174         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3175
3176         bpf_compute_data_end_sk_skb(skb);
3177         return ret;
3178 }
3179
3180 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3181         .func           = sk_skb_change_tail,
3182         .gpl_only       = false,
3183         .ret_type       = RET_INTEGER,
3184         .arg1_type      = ARG_PTR_TO_CTX,
3185         .arg2_type      = ARG_ANYTHING,
3186         .arg3_type      = ARG_ANYTHING,
3187 };
3188
3189 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3190                                         u64 flags)
3191 {
3192         u32 max_len = __bpf_skb_max_len(skb);
3193         u32 new_len = skb->len + head_room;
3194         int ret;
3195
3196         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3197                      new_len < skb->len))
3198                 return -EINVAL;
3199
3200         ret = skb_cow(skb, head_room);
3201         if (likely(!ret)) {
3202                 /* Idea for this helper is that we currently only
3203                  * allow to expand on mac header. This means that
3204                  * skb->protocol network header, etc, stay as is.
3205                  * Compared to bpf_skb_change_tail(), we're more
3206                  * flexible due to not needing to linearize or
3207                  * reset GSO. Intention for this helper is to be
3208                  * used by an L3 skb that needs to push mac header
3209                  * for redirection into L2 device.
3210                  */
3211                 __skb_push(skb, head_room);
3212                 memset(skb->data, 0, head_room);
3213                 skb_reset_mac_header(skb);
3214         }
3215
3216         return ret;
3217 }
3218
3219 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3220            u64, flags)
3221 {
3222         int ret = __bpf_skb_change_head(skb, head_room, flags);
3223
3224         bpf_compute_data_pointers(skb);
3225         return ret;
3226 }
3227
3228 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3229         .func           = bpf_skb_change_head,
3230         .gpl_only       = false,
3231         .ret_type       = RET_INTEGER,
3232         .arg1_type      = ARG_PTR_TO_CTX,
3233         .arg2_type      = ARG_ANYTHING,
3234         .arg3_type      = ARG_ANYTHING,
3235 };
3236
3237 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3238            u64, flags)
3239 {
3240         int ret = __bpf_skb_change_head(skb, head_room, flags);
3241
3242         bpf_compute_data_end_sk_skb(skb);
3243         return ret;
3244 }
3245
3246 static const struct bpf_func_proto sk_skb_change_head_proto = {
3247         .func           = sk_skb_change_head,
3248         .gpl_only       = false,
3249         .ret_type       = RET_INTEGER,
3250         .arg1_type      = ARG_PTR_TO_CTX,
3251         .arg2_type      = ARG_ANYTHING,
3252         .arg3_type      = ARG_ANYTHING,
3253 };
3254 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3255 {
3256         return xdp_data_meta_unsupported(xdp) ? 0 :
3257                xdp->data - xdp->data_meta;
3258 }
3259
3260 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3261 {
3262         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3263         unsigned long metalen = xdp_get_metalen(xdp);
3264         void *data_start = xdp_frame_end + metalen;
3265         void *data = xdp->data + offset;
3266
3267         if (unlikely(data < data_start ||
3268                      data > xdp->data_end - ETH_HLEN))
3269                 return -EINVAL;
3270
3271         if (metalen)
3272                 memmove(xdp->data_meta + offset,
3273                         xdp->data_meta, metalen);
3274         xdp->data_meta += offset;
3275         xdp->data = data;
3276
3277         return 0;
3278 }
3279
3280 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3281         .func           = bpf_xdp_adjust_head,
3282         .gpl_only       = false,
3283         .ret_type       = RET_INTEGER,
3284         .arg1_type      = ARG_PTR_TO_CTX,
3285         .arg2_type      = ARG_ANYTHING,
3286 };
3287
3288 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3289 {
3290         void *data_end = xdp->data_end + offset;
3291
3292         /* only shrinking is allowed for now. */
3293         if (unlikely(offset >= 0))
3294                 return -EINVAL;
3295
3296         if (unlikely(data_end < xdp->data + ETH_HLEN))
3297                 return -EINVAL;
3298
3299         xdp->data_end = data_end;
3300
3301         return 0;
3302 }
3303
3304 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3305         .func           = bpf_xdp_adjust_tail,
3306         .gpl_only       = false,
3307         .ret_type       = RET_INTEGER,
3308         .arg1_type      = ARG_PTR_TO_CTX,
3309         .arg2_type      = ARG_ANYTHING,
3310 };
3311
3312 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3313 {
3314         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3315         void *meta = xdp->data_meta + offset;
3316         unsigned long metalen = xdp->data - meta;
3317
3318         if (xdp_data_meta_unsupported(xdp))
3319                 return -ENOTSUPP;
3320         if (unlikely(meta < xdp_frame_end ||
3321                      meta > xdp->data))
3322                 return -EINVAL;
3323         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3324                      (metalen > 32)))
3325                 return -EACCES;
3326
3327         xdp->data_meta = meta;
3328
3329         return 0;
3330 }
3331
3332 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3333         .func           = bpf_xdp_adjust_meta,
3334         .gpl_only       = false,
3335         .ret_type       = RET_INTEGER,
3336         .arg1_type      = ARG_PTR_TO_CTX,
3337         .arg2_type      = ARG_ANYTHING,
3338 };
3339
3340 static int __bpf_tx_xdp(struct net_device *dev,
3341                         struct bpf_map *map,
3342                         struct xdp_buff *xdp,
3343                         u32 index)
3344 {
3345         struct xdp_frame *xdpf;
3346         int err, sent;
3347
3348         if (!dev->netdev_ops->ndo_xdp_xmit) {
3349                 return -EOPNOTSUPP;
3350         }
3351
3352         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3353         if (unlikely(err))
3354                 return err;
3355
3356         xdpf = convert_to_xdp_frame(xdp);
3357         if (unlikely(!xdpf))
3358                 return -EOVERFLOW;
3359
3360         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3361         if (sent <= 0)
3362                 return sent;
3363         return 0;
3364 }
3365
3366 static noinline int
3367 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3368                      struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3369 {
3370         struct net_device *fwd;
3371         u32 index = ri->ifindex;
3372         int err;
3373
3374         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3375         ri->ifindex = 0;
3376         if (unlikely(!fwd)) {
3377                 err = -EINVAL;
3378                 goto err;
3379         }
3380
3381         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3382         if (unlikely(err))
3383                 goto err;
3384
3385         _trace_xdp_redirect(dev, xdp_prog, index);
3386         return 0;
3387 err:
3388         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3389         return err;
3390 }
3391
3392 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3393                             struct bpf_map *map,
3394                             struct xdp_buff *xdp,
3395                             u32 index)
3396 {
3397         int err;
3398
3399         switch (map->map_type) {
3400         case BPF_MAP_TYPE_DEVMAP: {
3401                 struct bpf_dtab_netdev *dst = fwd;
3402
3403                 err = dev_map_enqueue(dst, xdp, dev_rx);
3404                 if (unlikely(err))
3405                         return err;
3406                 __dev_map_insert_ctx(map, index);
3407                 break;
3408         }
3409         case BPF_MAP_TYPE_CPUMAP: {
3410                 struct bpf_cpu_map_entry *rcpu = fwd;
3411
3412                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3413                 if (unlikely(err))
3414                         return err;
3415                 __cpu_map_insert_ctx(map, index);
3416                 break;
3417         }
3418         case BPF_MAP_TYPE_XSKMAP: {
3419                 struct xdp_sock *xs = fwd;
3420
3421                 err = __xsk_map_redirect(map, xdp, xs);
3422                 return err;
3423         }
3424         default:
3425                 break;
3426         }
3427         return 0;
3428 }
3429
3430 void xdp_do_flush_map(void)
3431 {
3432         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3433         struct bpf_map *map = ri->map_to_flush;
3434
3435         ri->map_to_flush = NULL;
3436         if (map) {
3437                 switch (map->map_type) {
3438                 case BPF_MAP_TYPE_DEVMAP:
3439                         __dev_map_flush(map);
3440                         break;
3441                 case BPF_MAP_TYPE_CPUMAP:
3442                         __cpu_map_flush(map);
3443                         break;
3444                 case BPF_MAP_TYPE_XSKMAP:
3445                         __xsk_map_flush(map);
3446                         break;
3447                 default:
3448                         break;
3449                 }
3450         }
3451 }
3452 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3453
3454 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3455 {
3456         switch (map->map_type) {
3457         case BPF_MAP_TYPE_DEVMAP:
3458                 return __dev_map_lookup_elem(map, index);
3459         case BPF_MAP_TYPE_CPUMAP:
3460                 return __cpu_map_lookup_elem(map, index);
3461         case BPF_MAP_TYPE_XSKMAP:
3462                 return __xsk_map_lookup_elem(map, index);
3463         default:
3464                 return NULL;
3465         }
3466 }
3467
3468 void bpf_clear_redirect_map(struct bpf_map *map)
3469 {
3470         struct bpf_redirect_info *ri;
3471         int cpu;
3472
3473         for_each_possible_cpu(cpu) {
3474                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3475                 /* Avoid polluting remote cacheline due to writes if
3476                  * not needed. Once we pass this test, we need the
3477                  * cmpxchg() to make sure it hasn't been changed in
3478                  * the meantime by remote CPU.
3479                  */
3480                 if (unlikely(READ_ONCE(ri->map) == map))
3481                         cmpxchg(&ri->map, map, NULL);
3482         }
3483 }
3484
3485 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3486                                struct bpf_prog *xdp_prog, struct bpf_map *map,
3487                                struct bpf_redirect_info *ri)
3488 {
3489         u32 index = ri->ifindex;
3490         void *fwd = NULL;
3491         int err;
3492
3493         ri->ifindex = 0;
3494         WRITE_ONCE(ri->map, NULL);
3495
3496         fwd = __xdp_map_lookup_elem(map, index);
3497         if (unlikely(!fwd)) {
3498                 err = -EINVAL;
3499                 goto err;
3500         }
3501         if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3502                 xdp_do_flush_map();
3503
3504         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3505         if (unlikely(err))
3506                 goto err;
3507
3508         ri->map_to_flush = map;
3509         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3510         return 0;
3511 err:
3512         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3513         return err;
3514 }
3515
3516 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3517                     struct bpf_prog *xdp_prog)
3518 {
3519         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3520         struct bpf_map *map = READ_ONCE(ri->map);
3521
3522         if (likely(map))
3523                 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3524
3525         return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3526 }
3527 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3528
3529 static int xdp_do_generic_redirect_map(struct net_device *dev,
3530                                        struct sk_buff *skb,
3531                                        struct xdp_buff *xdp,
3532                                        struct bpf_prog *xdp_prog,
3533                                        struct bpf_map *map)
3534 {
3535         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3536         u32 index = ri->ifindex;
3537         void *fwd = NULL;
3538         int err = 0;
3539
3540         ri->ifindex = 0;
3541         WRITE_ONCE(ri->map, NULL);
3542
3543         fwd = __xdp_map_lookup_elem(map, index);
3544         if (unlikely(!fwd)) {
3545                 err = -EINVAL;
3546                 goto err;
3547         }
3548
3549         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3550                 struct bpf_dtab_netdev *dst = fwd;
3551
3552                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3553                 if (unlikely(err))
3554                         goto err;
3555         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3556                 struct xdp_sock *xs = fwd;
3557
3558                 err = xsk_generic_rcv(xs, xdp);
3559                 if (err)
3560                         goto err;
3561                 consume_skb(skb);
3562         } else {
3563                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3564                 err = -EBADRQC;
3565                 goto err;
3566         }
3567
3568         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3569         return 0;
3570 err:
3571         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3572         return err;
3573 }
3574
3575 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3576                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3577 {
3578         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3579         struct bpf_map *map = READ_ONCE(ri->map);
3580         u32 index = ri->ifindex;
3581         struct net_device *fwd;
3582         int err = 0;
3583
3584         if (map)
3585                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3586                                                    map);
3587         ri->ifindex = 0;
3588         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3589         if (unlikely(!fwd)) {
3590                 err = -EINVAL;
3591                 goto err;
3592         }
3593
3594         err = xdp_ok_fwd_dev(fwd, skb->len);
3595         if (unlikely(err))
3596                 goto err;
3597
3598         skb->dev = fwd;
3599         _trace_xdp_redirect(dev, xdp_prog, index);
3600         generic_xdp_tx(skb, xdp_prog);
3601         return 0;
3602 err:
3603         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3604         return err;
3605 }
3606 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3607
3608 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3609 {
3610         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3611
3612         if (unlikely(flags))
3613                 return XDP_ABORTED;
3614
3615         ri->ifindex = ifindex;
3616         ri->flags = flags;
3617         WRITE_ONCE(ri->map, NULL);
3618
3619         return XDP_REDIRECT;
3620 }
3621
3622 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3623         .func           = bpf_xdp_redirect,
3624         .gpl_only       = false,
3625         .ret_type       = RET_INTEGER,
3626         .arg1_type      = ARG_ANYTHING,
3627         .arg2_type      = ARG_ANYTHING,
3628 };
3629
3630 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3631            u64, flags)
3632 {
3633         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3634
3635         if (unlikely(flags))
3636                 return XDP_ABORTED;
3637
3638         ri->ifindex = ifindex;
3639         ri->flags = flags;
3640         WRITE_ONCE(ri->map, map);
3641
3642         return XDP_REDIRECT;
3643 }
3644
3645 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3646         .func           = bpf_xdp_redirect_map,
3647         .gpl_only       = false,
3648         .ret_type       = RET_INTEGER,
3649         .arg1_type      = ARG_CONST_MAP_PTR,
3650         .arg2_type      = ARG_ANYTHING,
3651         .arg3_type      = ARG_ANYTHING,
3652 };
3653
3654 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3655                                   unsigned long off, unsigned long len)
3656 {
3657         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3658
3659         if (unlikely(!ptr))
3660                 return len;
3661         if (ptr != dst_buff)
3662                 memcpy(dst_buff, ptr, len);
3663
3664         return 0;
3665 }
3666
3667 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3668            u64, flags, void *, meta, u64, meta_size)
3669 {
3670         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3671
3672         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3673                 return -EINVAL;
3674         if (unlikely(skb_size > skb->len))
3675                 return -EFAULT;
3676
3677         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3678                                 bpf_skb_copy);
3679 }
3680
3681 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3682         .func           = bpf_skb_event_output,
3683         .gpl_only       = true,
3684         .ret_type       = RET_INTEGER,
3685         .arg1_type      = ARG_PTR_TO_CTX,
3686         .arg2_type      = ARG_CONST_MAP_PTR,
3687         .arg3_type      = ARG_ANYTHING,
3688         .arg4_type      = ARG_PTR_TO_MEM,
3689         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3690 };
3691
3692 static unsigned short bpf_tunnel_key_af(u64 flags)
3693 {
3694         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3695 }
3696
3697 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3698            u32, size, u64, flags)
3699 {
3700         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3701         u8 compat[sizeof(struct bpf_tunnel_key)];
3702         void *to_orig = to;
3703         int err;
3704
3705         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3706                 err = -EINVAL;
3707                 goto err_clear;
3708         }
3709         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3710                 err = -EPROTO;
3711                 goto err_clear;
3712         }
3713         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3714                 err = -EINVAL;
3715                 switch (size) {
3716                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3717                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3718                         goto set_compat;
3719                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3720                         /* Fixup deprecated structure layouts here, so we have
3721                          * a common path later on.
3722                          */
3723                         if (ip_tunnel_info_af(info) != AF_INET)
3724                                 goto err_clear;
3725 set_compat:
3726                         to = (struct bpf_tunnel_key *)compat;
3727                         break;
3728                 default:
3729                         goto err_clear;
3730                 }
3731         }
3732
3733         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3734         to->tunnel_tos = info->key.tos;
3735         to->tunnel_ttl = info->key.ttl;
3736         to->tunnel_ext = 0;
3737
3738         if (flags & BPF_F_TUNINFO_IPV6) {
3739                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3740                        sizeof(to->remote_ipv6));
3741                 to->tunnel_label = be32_to_cpu(info->key.label);
3742         } else {
3743                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3744                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3745                 to->tunnel_label = 0;
3746         }
3747
3748         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3749                 memcpy(to_orig, to, size);
3750
3751         return 0;
3752 err_clear:
3753         memset(to_orig, 0, size);
3754         return err;
3755 }
3756
3757 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3758         .func           = bpf_skb_get_tunnel_key,
3759         .gpl_only       = false,
3760         .ret_type       = RET_INTEGER,
3761         .arg1_type      = ARG_PTR_TO_CTX,
3762         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3763         .arg3_type      = ARG_CONST_SIZE,
3764         .arg4_type      = ARG_ANYTHING,
3765 };
3766
3767 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3768 {
3769         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3770         int err;
3771
3772         if (unlikely(!info ||
3773                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3774                 err = -ENOENT;
3775                 goto err_clear;
3776         }
3777         if (unlikely(size < info->options_len)) {
3778                 err = -ENOMEM;
3779                 goto err_clear;
3780         }
3781
3782         ip_tunnel_info_opts_get(to, info);
3783         if (size > info->options_len)
3784                 memset(to + info->options_len, 0, size - info->options_len);
3785
3786         return info->options_len;
3787 err_clear:
3788         memset(to, 0, size);
3789         return err;
3790 }
3791
3792 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3793         .func           = bpf_skb_get_tunnel_opt,
3794         .gpl_only       = false,
3795         .ret_type       = RET_INTEGER,
3796         .arg1_type      = ARG_PTR_TO_CTX,
3797         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3798         .arg3_type      = ARG_CONST_SIZE,
3799 };
3800
3801 static struct metadata_dst __percpu *md_dst;
3802
3803 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3804            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3805 {
3806         struct metadata_dst *md = this_cpu_ptr(md_dst);
3807         u8 compat[sizeof(struct bpf_tunnel_key)];
3808         struct ip_tunnel_info *info;
3809
3810         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3811                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3812                 return -EINVAL;
3813         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3814                 switch (size) {
3815                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3816                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3817                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3818                         /* Fixup deprecated structure layouts here, so we have
3819                          * a common path later on.
3820                          */
3821                         memcpy(compat, from, size);
3822                         memset(compat + size, 0, sizeof(compat) - size);
3823                         from = (const struct bpf_tunnel_key *) compat;
3824                         break;
3825                 default:
3826                         return -EINVAL;
3827                 }
3828         }
3829         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3830                      from->tunnel_ext))
3831                 return -EINVAL;
3832
3833         skb_dst_drop(skb);
3834         dst_hold((struct dst_entry *) md);
3835         skb_dst_set(skb, (struct dst_entry *) md);
3836
3837         info = &md->u.tun_info;
3838         memset(info, 0, sizeof(*info));
3839         info->mode = IP_TUNNEL_INFO_TX;
3840
3841         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3842         if (flags & BPF_F_DONT_FRAGMENT)
3843                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3844         if (flags & BPF_F_ZERO_CSUM_TX)
3845                 info->key.tun_flags &= ~TUNNEL_CSUM;
3846         if (flags & BPF_F_SEQ_NUMBER)
3847                 info->key.tun_flags |= TUNNEL_SEQ;
3848
3849         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3850         info->key.tos = from->tunnel_tos;
3851         info->key.ttl = from->tunnel_ttl;
3852
3853         if (flags & BPF_F_TUNINFO_IPV6) {
3854                 info->mode |= IP_TUNNEL_INFO_IPV6;
3855                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3856                        sizeof(from->remote_ipv6));
3857                 info->key.label = cpu_to_be32(from->tunnel_label) &
3858                                   IPV6_FLOWLABEL_MASK;
3859         } else {
3860                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3861         }
3862
3863         return 0;
3864 }
3865
3866 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3867         .func           = bpf_skb_set_tunnel_key,
3868         .gpl_only       = false,
3869         .ret_type       = RET_INTEGER,
3870         .arg1_type      = ARG_PTR_TO_CTX,
3871         .arg2_type      = ARG_PTR_TO_MEM,
3872         .arg3_type      = ARG_CONST_SIZE,
3873         .arg4_type      = ARG_ANYTHING,
3874 };
3875
3876 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3877            const u8 *, from, u32, size)
3878 {
3879         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3880         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3881
3882         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3883                 return -EINVAL;
3884         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3885                 return -ENOMEM;
3886
3887         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3888
3889         return 0;
3890 }
3891
3892 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3893         .func           = bpf_skb_set_tunnel_opt,
3894         .gpl_only       = false,
3895         .ret_type       = RET_INTEGER,
3896         .arg1_type      = ARG_PTR_TO_CTX,
3897         .arg2_type      = ARG_PTR_TO_MEM,
3898         .arg3_type      = ARG_CONST_SIZE,
3899 };
3900
3901 static const struct bpf_func_proto *
3902 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3903 {
3904         if (!md_dst) {
3905                 struct metadata_dst __percpu *tmp;
3906
3907                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3908                                                 METADATA_IP_TUNNEL,
3909                                                 GFP_KERNEL);
3910                 if (!tmp)
3911                         return NULL;
3912                 if (cmpxchg(&md_dst, NULL, tmp))
3913                         metadata_dst_free_percpu(tmp);
3914         }
3915
3916         switch (which) {
3917         case BPF_FUNC_skb_set_tunnel_key:
3918                 return &bpf_skb_set_tunnel_key_proto;
3919         case BPF_FUNC_skb_set_tunnel_opt:
3920                 return &bpf_skb_set_tunnel_opt_proto;
3921         default:
3922                 return NULL;
3923         }
3924 }
3925
3926 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3927            u32, idx)
3928 {
3929         struct bpf_array *array = container_of(map, struct bpf_array, map);
3930         struct cgroup *cgrp;
3931         struct sock *sk;
3932
3933         sk = skb_to_full_sk(skb);
3934         if (!sk || !sk_fullsock(sk))
3935                 return -ENOENT;
3936         if (unlikely(idx >= array->map.max_entries))
3937                 return -E2BIG;
3938
3939         cgrp = READ_ONCE(array->ptrs[idx]);
3940         if (unlikely(!cgrp))
3941                 return -EAGAIN;
3942
3943         return sk_under_cgroup_hierarchy(sk, cgrp);
3944 }
3945
3946 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3947         .func           = bpf_skb_under_cgroup,
3948         .gpl_only       = false,
3949         .ret_type       = RET_INTEGER,
3950         .arg1_type      = ARG_PTR_TO_CTX,
3951         .arg2_type      = ARG_CONST_MAP_PTR,
3952         .arg3_type      = ARG_ANYTHING,
3953 };
3954
3955 #ifdef CONFIG_SOCK_CGROUP_DATA
3956 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3957 {
3958         struct sock *sk = skb_to_full_sk(skb);
3959         struct cgroup *cgrp;
3960
3961         if (!sk || !sk_fullsock(sk))
3962                 return 0;
3963
3964         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3965         return cgrp->kn->id.id;
3966 }
3967
3968 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3969         .func           = bpf_skb_cgroup_id,
3970         .gpl_only       = false,
3971         .ret_type       = RET_INTEGER,
3972         .arg1_type      = ARG_PTR_TO_CTX,
3973 };
3974
3975 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3976            ancestor_level)
3977 {
3978         struct sock *sk = skb_to_full_sk(skb);
3979         struct cgroup *ancestor;
3980         struct cgroup *cgrp;
3981
3982         if (!sk || !sk_fullsock(sk))
3983                 return 0;
3984
3985         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3986         ancestor = cgroup_ancestor(cgrp, ancestor_level);
3987         if (!ancestor)
3988                 return 0;
3989
3990         return ancestor->kn->id.id;
3991 }
3992
3993 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3994         .func           = bpf_skb_ancestor_cgroup_id,
3995         .gpl_only       = false,
3996         .ret_type       = RET_INTEGER,
3997         .arg1_type      = ARG_PTR_TO_CTX,
3998         .arg2_type      = ARG_ANYTHING,
3999 };
4000 #endif
4001
4002 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4003                                   unsigned long off, unsigned long len)
4004 {
4005         memcpy(dst_buff, src_buff + off, len);
4006         return 0;
4007 }
4008
4009 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4010            u64, flags, void *, meta, u64, meta_size)
4011 {
4012         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4013
4014         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4015                 return -EINVAL;
4016         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4017                 return -EFAULT;
4018
4019         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4020                                 xdp_size, bpf_xdp_copy);
4021 }
4022
4023 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4024         .func           = bpf_xdp_event_output,
4025         .gpl_only       = true,
4026         .ret_type       = RET_INTEGER,
4027         .arg1_type      = ARG_PTR_TO_CTX,
4028         .arg2_type      = ARG_CONST_MAP_PTR,
4029         .arg3_type      = ARG_ANYTHING,
4030         .arg4_type      = ARG_PTR_TO_MEM,
4031         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4032 };
4033
4034 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4035 {
4036         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4037 }
4038
4039 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4040         .func           = bpf_get_socket_cookie,
4041         .gpl_only       = false,
4042         .ret_type       = RET_INTEGER,
4043         .arg1_type      = ARG_PTR_TO_CTX,
4044 };
4045
4046 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4047 {
4048         return sock_gen_cookie(ctx->sk);
4049 }
4050
4051 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4052         .func           = bpf_get_socket_cookie_sock_addr,
4053         .gpl_only       = false,
4054         .ret_type       = RET_INTEGER,
4055         .arg1_type      = ARG_PTR_TO_CTX,
4056 };
4057
4058 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4059 {
4060         return sock_gen_cookie(ctx->sk);
4061 }
4062
4063 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4064         .func           = bpf_get_socket_cookie_sock_ops,
4065         .gpl_only       = false,
4066         .ret_type       = RET_INTEGER,
4067         .arg1_type      = ARG_PTR_TO_CTX,
4068 };
4069
4070 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4071 {
4072         struct sock *sk = sk_to_full_sk(skb->sk);
4073         kuid_t kuid;
4074
4075         if (!sk || !sk_fullsock(sk))
4076                 return overflowuid;
4077         kuid = sock_net_uid(sock_net(sk), sk);
4078         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4079 }
4080
4081 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4082         .func           = bpf_get_socket_uid,
4083         .gpl_only       = false,
4084         .ret_type       = RET_INTEGER,
4085         .arg1_type      = ARG_PTR_TO_CTX,
4086 };
4087
4088 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4089            struct bpf_map *, map, u64, flags, void *, data, u64, size)
4090 {
4091         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4092                 return -EINVAL;
4093
4094         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4095 }
4096
4097 static const struct bpf_func_proto bpf_sockopt_event_output_proto =  {
4098         .func           = bpf_sockopt_event_output,
4099         .gpl_only       = true,
4100         .ret_type       = RET_INTEGER,
4101         .arg1_type      = ARG_PTR_TO_CTX,
4102         .arg2_type      = ARG_CONST_MAP_PTR,
4103         .arg3_type      = ARG_ANYTHING,
4104         .arg4_type      = ARG_PTR_TO_MEM,
4105         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4106 };
4107
4108 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4109            int, level, int, optname, char *, optval, int, optlen)
4110 {
4111         struct sock *sk = bpf_sock->sk;
4112         int ret = 0;
4113         int val;
4114
4115         if (!sk_fullsock(sk))
4116                 return -EINVAL;
4117
4118         if (level == SOL_SOCKET) {
4119                 if (optlen != sizeof(int))
4120                         return -EINVAL;
4121                 val = *((int *)optval);
4122
4123                 /* Only some socketops are supported */
4124                 switch (optname) {
4125                 case SO_RCVBUF:
4126                         val = min_t(u32, val, sysctl_rmem_max);
4127                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4128                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4129                         break;
4130                 case SO_SNDBUF:
4131                         val = min_t(u32, val, sysctl_wmem_max);
4132                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4133                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4134                         break;
4135                 case SO_MAX_PACING_RATE: /* 32bit version */
4136                         if (val != ~0U)
4137                                 cmpxchg(&sk->sk_pacing_status,
4138                                         SK_PACING_NONE,
4139                                         SK_PACING_NEEDED);
4140                         sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4141                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4142                                                  sk->sk_max_pacing_rate);
4143                         break;
4144                 case SO_PRIORITY:
4145                         sk->sk_priority = val;
4146                         break;
4147                 case SO_RCVLOWAT:
4148                         if (val < 0)
4149                                 val = INT_MAX;
4150                         sk->sk_rcvlowat = val ? : 1;
4151                         break;
4152                 case SO_MARK:
4153                         if (sk->sk_mark != val) {
4154                                 sk->sk_mark = val;
4155                                 sk_dst_reset(sk);
4156                         }
4157                         break;
4158                 default:
4159                         ret = -EINVAL;
4160                 }
4161 #ifdef CONFIG_INET
4162         } else if (level == SOL_IP) {
4163                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4164                         return -EINVAL;
4165
4166                 val = *((int *)optval);
4167                 /* Only some options are supported */
4168                 switch (optname) {
4169                 case IP_TOS:
4170                         if (val < -1 || val > 0xff) {
4171                                 ret = -EINVAL;
4172                         } else {
4173                                 struct inet_sock *inet = inet_sk(sk);
4174
4175                                 if (val == -1)
4176                                         val = 0;
4177                                 inet->tos = val;
4178                         }
4179                         break;
4180                 default:
4181                         ret = -EINVAL;
4182                 }
4183 #if IS_ENABLED(CONFIG_IPV6)
4184         } else if (level == SOL_IPV6) {
4185                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4186                         return -EINVAL;
4187
4188                 val = *((int *)optval);
4189                 /* Only some options are supported */
4190                 switch (optname) {
4191                 case IPV6_TCLASS:
4192                         if (val < -1 || val > 0xff) {
4193                                 ret = -EINVAL;
4194                         } else {
4195                                 struct ipv6_pinfo *np = inet6_sk(sk);
4196
4197                                 if (val == -1)
4198                                         val = 0;
4199                                 np->tclass = val;
4200                         }
4201                         break;
4202                 default:
4203                         ret = -EINVAL;
4204                 }
4205 #endif
4206         } else if (level == SOL_TCP &&
4207                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4208                 if (optname == TCP_CONGESTION) {
4209                         char name[TCP_CA_NAME_MAX];
4210                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4211
4212                         strncpy(name, optval, min_t(long, optlen,
4213                                                     TCP_CA_NAME_MAX-1));
4214                         name[TCP_CA_NAME_MAX-1] = 0;
4215                         ret = tcp_set_congestion_control(sk, name, false,
4216                                                          reinit);
4217                 } else {
4218                         struct tcp_sock *tp = tcp_sk(sk);
4219
4220                         if (optlen != sizeof(int))
4221                                 return -EINVAL;
4222
4223                         val = *((int *)optval);
4224                         /* Only some options are supported */
4225                         switch (optname) {
4226                         case TCP_BPF_IW:
4227                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4228                                         ret = -EINVAL;
4229                                 else
4230                                         tp->snd_cwnd = val;
4231                                 break;
4232                         case TCP_BPF_SNDCWND_CLAMP:
4233                                 if (val <= 0) {
4234                                         ret = -EINVAL;
4235                                 } else {
4236                                         tp->snd_cwnd_clamp = val;
4237                                         tp->snd_ssthresh = val;
4238                                 }
4239                                 break;
4240                         case TCP_SAVE_SYN:
4241                                 if (val < 0 || val > 1)
4242                                         ret = -EINVAL;
4243                                 else
4244                                         tp->save_syn = val;
4245                                 break;
4246                         default:
4247                                 ret = -EINVAL;
4248                         }
4249                 }
4250 #endif
4251         } else {
4252                 ret = -EINVAL;
4253         }
4254         return ret;
4255 }
4256
4257 static const struct bpf_func_proto bpf_setsockopt_proto = {
4258         .func           = bpf_setsockopt,
4259         .gpl_only       = false,
4260         .ret_type       = RET_INTEGER,
4261         .arg1_type      = ARG_PTR_TO_CTX,
4262         .arg2_type      = ARG_ANYTHING,
4263         .arg3_type      = ARG_ANYTHING,
4264         .arg4_type      = ARG_PTR_TO_MEM,
4265         .arg5_type      = ARG_CONST_SIZE,
4266 };
4267
4268 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4269            int, level, int, optname, char *, optval, int, optlen)
4270 {
4271         struct sock *sk = bpf_sock->sk;
4272
4273         if (!sk_fullsock(sk))
4274                 goto err_clear;
4275 #ifdef CONFIG_INET
4276         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4277                 struct inet_connection_sock *icsk;
4278                 struct tcp_sock *tp;
4279
4280                 switch (optname) {
4281                 case TCP_CONGESTION:
4282                         icsk = inet_csk(sk);
4283
4284                         if (!icsk->icsk_ca_ops || optlen <= 1)
4285                                 goto err_clear;
4286                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4287                         optval[optlen - 1] = 0;
4288                         break;
4289                 case TCP_SAVED_SYN:
4290                         tp = tcp_sk(sk);
4291
4292                         if (optlen <= 0 || !tp->saved_syn ||
4293                             optlen > tp->saved_syn[0])
4294                                 goto err_clear;
4295                         memcpy(optval, tp->saved_syn + 1, optlen);
4296                         break;
4297                 default:
4298                         goto err_clear;
4299                 }
4300         } else if (level == SOL_IP) {
4301                 struct inet_sock *inet = inet_sk(sk);
4302
4303                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4304                         goto err_clear;
4305
4306                 /* Only some options are supported */
4307                 switch (optname) {
4308                 case IP_TOS:
4309                         *((int *)optval) = (int)inet->tos;
4310                         break;
4311                 default:
4312                         goto err_clear;
4313                 }
4314 #if IS_ENABLED(CONFIG_IPV6)
4315         } else if (level == SOL_IPV6) {
4316                 struct ipv6_pinfo *np = inet6_sk(sk);
4317
4318                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4319                         goto err_clear;
4320
4321                 /* Only some options are supported */
4322                 switch (optname) {
4323                 case IPV6_TCLASS:
4324                         *((int *)optval) = (int)np->tclass;
4325                         break;
4326                 default:
4327                         goto err_clear;
4328                 }
4329 #endif
4330         } else {
4331                 goto err_clear;
4332         }
4333         return 0;
4334 #endif
4335 err_clear:
4336         memset(optval, 0, optlen);
4337         return -EINVAL;
4338 }
4339
4340 static const struct bpf_func_proto bpf_getsockopt_proto = {
4341         .func           = bpf_getsockopt,
4342         .gpl_only       = false,
4343         .ret_type       = RET_INTEGER,
4344         .arg1_type      = ARG_PTR_TO_CTX,
4345         .arg2_type      = ARG_ANYTHING,
4346         .arg3_type      = ARG_ANYTHING,
4347         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4348         .arg5_type      = ARG_CONST_SIZE,
4349 };
4350
4351 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4352            int, argval)
4353 {
4354         struct sock *sk = bpf_sock->sk;
4355         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4356
4357         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4358                 return -EINVAL;
4359
4360         if (val)
4361                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4362
4363         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4364 }
4365
4366 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4367         .func           = bpf_sock_ops_cb_flags_set,
4368         .gpl_only       = false,
4369         .ret_type       = RET_INTEGER,
4370         .arg1_type      = ARG_PTR_TO_CTX,
4371         .arg2_type      = ARG_ANYTHING,
4372 };
4373
4374 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4375 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4376
4377 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4378            int, addr_len)
4379 {
4380 #ifdef CONFIG_INET
4381         struct sock *sk = ctx->sk;
4382         int err;
4383
4384         /* Binding to port can be expensive so it's prohibited in the helper.
4385          * Only binding to IP is supported.
4386          */
4387         err = -EINVAL;
4388         if (addr->sa_family == AF_INET) {
4389                 if (addr_len < sizeof(struct sockaddr_in))
4390                         return err;
4391                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4392                         return err;
4393                 return __inet_bind(sk, addr, addr_len, true, false);
4394 #if IS_ENABLED(CONFIG_IPV6)
4395         } else if (addr->sa_family == AF_INET6) {
4396                 if (addr_len < SIN6_LEN_RFC2133)
4397                         return err;
4398                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4399                         return err;
4400                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4401                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4402                  */
4403                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4404 #endif /* CONFIG_IPV6 */
4405         }
4406 #endif /* CONFIG_INET */
4407
4408         return -EAFNOSUPPORT;
4409 }
4410
4411 static const struct bpf_func_proto bpf_bind_proto = {
4412         .func           = bpf_bind,
4413         .gpl_only       = false,
4414         .ret_type       = RET_INTEGER,
4415         .arg1_type      = ARG_PTR_TO_CTX,
4416         .arg2_type      = ARG_PTR_TO_MEM,
4417         .arg3_type      = ARG_CONST_SIZE,
4418 };
4419
4420 #ifdef CONFIG_XFRM
4421 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4422            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4423 {
4424         const struct sec_path *sp = skb_sec_path(skb);
4425         const struct xfrm_state *x;
4426
4427         if (!sp || unlikely(index >= sp->len || flags))
4428                 goto err_clear;
4429
4430         x = sp->xvec[index];
4431
4432         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4433                 goto err_clear;
4434
4435         to->reqid = x->props.reqid;
4436         to->spi = x->id.spi;
4437         to->family = x->props.family;
4438         to->ext = 0;
4439
4440         if (to->family == AF_INET6) {
4441                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4442                        sizeof(to->remote_ipv6));
4443         } else {
4444                 to->remote_ipv4 = x->props.saddr.a4;
4445                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4446         }
4447
4448         return 0;
4449 err_clear:
4450         memset(to, 0, size);
4451         return -EINVAL;
4452 }
4453
4454 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4455         .func           = bpf_skb_get_xfrm_state,
4456         .gpl_only       = false,
4457         .ret_type       = RET_INTEGER,
4458         .arg1_type      = ARG_PTR_TO_CTX,
4459         .arg2_type      = ARG_ANYTHING,
4460         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4461         .arg4_type      = ARG_CONST_SIZE,
4462         .arg5_type      = ARG_ANYTHING,
4463 };
4464 #endif
4465
4466 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4467 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4468                                   const struct neighbour *neigh,
4469                                   const struct net_device *dev)
4470 {
4471         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4472         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4473         params->h_vlan_TCI = 0;
4474         params->h_vlan_proto = 0;
4475         params->ifindex = dev->ifindex;
4476
4477         return 0;
4478 }
4479 #endif
4480
4481 #if IS_ENABLED(CONFIG_INET)
4482 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4483                                u32 flags, bool check_mtu)
4484 {
4485         struct in_device *in_dev;
4486         struct neighbour *neigh;
4487         struct net_device *dev;
4488         struct fib_result res;
4489         struct fib_nh *nh;
4490         struct flowi4 fl4;
4491         int err;
4492         u32 mtu;
4493
4494         dev = dev_get_by_index_rcu(net, params->ifindex);
4495         if (unlikely(!dev))
4496                 return -ENODEV;
4497
4498         /* verify forwarding is enabled on this interface */
4499         in_dev = __in_dev_get_rcu(dev);
4500         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4501                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4502
4503         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4504                 fl4.flowi4_iif = 1;
4505                 fl4.flowi4_oif = params->ifindex;
4506         } else {
4507                 fl4.flowi4_iif = params->ifindex;
4508                 fl4.flowi4_oif = 0;
4509         }
4510         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4511         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4512         fl4.flowi4_flags = 0;
4513
4514         fl4.flowi4_proto = params->l4_protocol;
4515         fl4.daddr = params->ipv4_dst;
4516         fl4.saddr = params->ipv4_src;
4517         fl4.fl4_sport = params->sport;
4518         fl4.fl4_dport = params->dport;
4519
4520         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4521                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4522                 struct fib_table *tb;
4523
4524                 tb = fib_get_table(net, tbid);
4525                 if (unlikely(!tb))
4526                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4527
4528                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4529         } else {
4530                 fl4.flowi4_mark = 0;
4531                 fl4.flowi4_secid = 0;
4532                 fl4.flowi4_tun_key.tun_id = 0;
4533                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4534
4535                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4536         }
4537
4538         if (err) {
4539                 /* map fib lookup errors to RTN_ type */
4540                 if (err == -EINVAL)
4541                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4542                 if (err == -EHOSTUNREACH)
4543                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4544                 if (err == -EACCES)
4545                         return BPF_FIB_LKUP_RET_PROHIBIT;
4546
4547                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4548         }
4549
4550         if (res.type != RTN_UNICAST)
4551                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4552
4553         if (res.fi->fib_nhs > 1)
4554                 fib_select_path(net, &res, &fl4, NULL);
4555
4556         if (check_mtu) {
4557                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4558                 if (params->tot_len > mtu)
4559                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4560         }
4561
4562         nh = &res.fi->fib_nh[res.nh_sel];
4563
4564         /* do not handle lwt encaps right now */
4565         if (nh->nh_lwtstate)
4566                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4567
4568         dev = nh->nh_dev;
4569         if (nh->nh_gw)
4570                 params->ipv4_dst = nh->nh_gw;
4571
4572         params->rt_metric = res.fi->fib_priority;
4573
4574         /* xdp and cls_bpf programs are run in RCU-bh so
4575          * rcu_read_lock_bh is not needed here
4576          */
4577         neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4578         if (!neigh)
4579                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4580
4581         return bpf_fib_set_fwd_params(params, neigh, dev);
4582 }
4583 #endif
4584
4585 #if IS_ENABLED(CONFIG_IPV6)
4586 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4587                                u32 flags, bool check_mtu)
4588 {
4589         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4590         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4591         struct neighbour *neigh;
4592         struct net_device *dev;
4593         struct inet6_dev *idev;
4594         struct fib6_info *f6i;
4595         struct flowi6 fl6;
4596         int strict = 0;
4597         int oif;
4598         u32 mtu;
4599
4600         /* link local addresses are never forwarded */
4601         if (rt6_need_strict(dst) || rt6_need_strict(src))
4602                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4603
4604         dev = dev_get_by_index_rcu(net, params->ifindex);
4605         if (unlikely(!dev))
4606                 return -ENODEV;
4607
4608         idev = __in6_dev_get_safely(dev);
4609         if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4610                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4611
4612         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4613                 fl6.flowi6_iif = 1;
4614                 oif = fl6.flowi6_oif = params->ifindex;
4615         } else {
4616                 oif = fl6.flowi6_iif = params->ifindex;
4617                 fl6.flowi6_oif = 0;
4618                 strict = RT6_LOOKUP_F_HAS_SADDR;
4619         }
4620         fl6.flowlabel = params->flowinfo;
4621         fl6.flowi6_scope = 0;
4622         fl6.flowi6_flags = 0;
4623         fl6.mp_hash = 0;
4624
4625         fl6.flowi6_proto = params->l4_protocol;
4626         fl6.daddr = *dst;
4627         fl6.saddr = *src;
4628         fl6.fl6_sport = params->sport;
4629         fl6.fl6_dport = params->dport;
4630
4631         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4632                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4633                 struct fib6_table *tb;
4634
4635                 tb = ipv6_stub->fib6_get_table(net, tbid);
4636                 if (unlikely(!tb))
4637                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4638
4639                 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4640         } else {
4641                 fl6.flowi6_mark = 0;
4642                 fl6.flowi6_secid = 0;
4643                 fl6.flowi6_tun_key.tun_id = 0;
4644                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4645
4646                 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4647         }
4648
4649         if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4650                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4651
4652         if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4653                 switch (f6i->fib6_type) {
4654                 case RTN_BLACKHOLE:
4655                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4656                 case RTN_UNREACHABLE:
4657                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4658                 case RTN_PROHIBIT:
4659                         return BPF_FIB_LKUP_RET_PROHIBIT;
4660                 default:
4661                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4662                 }
4663         }
4664
4665         if (f6i->fib6_type != RTN_UNICAST)
4666                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4667
4668         if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4669                 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4670                                                        fl6.flowi6_oif, NULL,
4671                                                        strict);
4672
4673         if (check_mtu) {
4674                 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4675                 if (params->tot_len > mtu)
4676                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4677         }
4678
4679         if (f6i->fib6_nh.nh_lwtstate)
4680                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4681
4682         if (f6i->fib6_flags & RTF_GATEWAY)
4683                 *dst = f6i->fib6_nh.nh_gw;
4684
4685         dev = f6i->fib6_nh.nh_dev;
4686         params->rt_metric = f6i->fib6_metric;
4687
4688         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4689          * not needed here. Can not use __ipv6_neigh_lookup_noref here
4690          * because we need to get nd_tbl via the stub
4691          */
4692         neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4693                                       ndisc_hashfn, dst, dev);
4694         if (!neigh)
4695                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4696
4697         return bpf_fib_set_fwd_params(params, neigh, dev);
4698 }
4699 #endif
4700
4701 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4702            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4703 {
4704         if (plen < sizeof(*params))
4705                 return -EINVAL;
4706
4707         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4708                 return -EINVAL;
4709
4710         switch (params->family) {
4711 #if IS_ENABLED(CONFIG_INET)
4712         case AF_INET:
4713                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4714                                            flags, true);
4715 #endif
4716 #if IS_ENABLED(CONFIG_IPV6)
4717         case AF_INET6:
4718                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4719                                            flags, true);
4720 #endif
4721         }
4722         return -EAFNOSUPPORT;
4723 }
4724
4725 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4726         .func           = bpf_xdp_fib_lookup,
4727         .gpl_only       = true,
4728         .ret_type       = RET_INTEGER,
4729         .arg1_type      = ARG_PTR_TO_CTX,
4730         .arg2_type      = ARG_PTR_TO_MEM,
4731         .arg3_type      = ARG_CONST_SIZE,
4732         .arg4_type      = ARG_ANYTHING,
4733 };
4734
4735 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4736            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4737 {
4738         struct net *net = dev_net(skb->dev);
4739         int rc = -EAFNOSUPPORT;
4740
4741         if (plen < sizeof(*params))
4742                 return -EINVAL;
4743
4744         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4745                 return -EINVAL;
4746
4747         switch (params->family) {
4748 #if IS_ENABLED(CONFIG_INET)
4749         case AF_INET:
4750                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4751                 break;
4752 #endif
4753 #if IS_ENABLED(CONFIG_IPV6)
4754         case AF_INET6:
4755                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4756                 break;
4757 #endif
4758         }
4759
4760         if (!rc) {
4761                 struct net_device *dev;
4762
4763                 dev = dev_get_by_index_rcu(net, params->ifindex);
4764                 if (!is_skb_forwardable(dev, skb))
4765                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4766         }
4767
4768         return rc;
4769 }
4770
4771 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4772         .func           = bpf_skb_fib_lookup,
4773         .gpl_only       = true,
4774         .ret_type       = RET_INTEGER,
4775         .arg1_type      = ARG_PTR_TO_CTX,
4776         .arg2_type      = ARG_PTR_TO_MEM,
4777         .arg3_type      = ARG_CONST_SIZE,
4778         .arg4_type      = ARG_ANYTHING,
4779 };
4780
4781 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4782 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4783 {
4784         int err;
4785         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4786
4787         if (!seg6_validate_srh(srh, len))
4788                 return -EINVAL;
4789
4790         switch (type) {
4791         case BPF_LWT_ENCAP_SEG6_INLINE:
4792                 if (skb->protocol != htons(ETH_P_IPV6))
4793                         return -EBADMSG;
4794
4795                 err = seg6_do_srh_inline(skb, srh);
4796                 break;
4797         case BPF_LWT_ENCAP_SEG6:
4798                 skb_reset_inner_headers(skb);
4799                 skb->encapsulation = 1;
4800                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4801                 break;
4802         default:
4803                 return -EINVAL;
4804         }
4805
4806         bpf_compute_data_pointers(skb);
4807         if (err)
4808                 return err;
4809
4810         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4811         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4812
4813         return seg6_lookup_nexthop(skb, NULL, 0);
4814 }
4815 #endif /* CONFIG_IPV6_SEG6_BPF */
4816
4817 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4818 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4819                              bool ingress)
4820 {
4821         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4822 }
4823 #endif
4824
4825 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4826            u32, len)
4827 {
4828         switch (type) {
4829 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4830         case BPF_LWT_ENCAP_SEG6:
4831         case BPF_LWT_ENCAP_SEG6_INLINE:
4832                 return bpf_push_seg6_encap(skb, type, hdr, len);
4833 #endif
4834 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4835         case BPF_LWT_ENCAP_IP:
4836                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4837 #endif
4838         default:
4839                 return -EINVAL;
4840         }
4841 }
4842
4843 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4844            void *, hdr, u32, len)
4845 {
4846         switch (type) {
4847 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4848         case BPF_LWT_ENCAP_IP:
4849                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4850 #endif
4851         default:
4852                 return -EINVAL;
4853         }
4854 }
4855
4856 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4857         .func           = bpf_lwt_in_push_encap,
4858         .gpl_only       = false,
4859         .ret_type       = RET_INTEGER,
4860         .arg1_type      = ARG_PTR_TO_CTX,
4861         .arg2_type      = ARG_ANYTHING,
4862         .arg3_type      = ARG_PTR_TO_MEM,
4863         .arg4_type      = ARG_CONST_SIZE
4864 };
4865
4866 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4867         .func           = bpf_lwt_xmit_push_encap,
4868         .gpl_only       = false,
4869         .ret_type       = RET_INTEGER,
4870         .arg1_type      = ARG_PTR_TO_CTX,
4871         .arg2_type      = ARG_ANYTHING,
4872         .arg3_type      = ARG_PTR_TO_MEM,
4873         .arg4_type      = ARG_CONST_SIZE
4874 };
4875
4876 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4877 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4878            const void *, from, u32, len)
4879 {
4880         struct seg6_bpf_srh_state *srh_state =
4881                 this_cpu_ptr(&seg6_bpf_srh_states);
4882         struct ipv6_sr_hdr *srh = srh_state->srh;
4883         void *srh_tlvs, *srh_end, *ptr;
4884         int srhoff = 0;
4885
4886         if (srh == NULL)
4887                 return -EINVAL;
4888
4889         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4890         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4891
4892         ptr = skb->data + offset;
4893         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4894                 srh_state->valid = false;
4895         else if (ptr < (void *)&srh->flags ||
4896                  ptr + len > (void *)&srh->segments)
4897                 return -EFAULT;
4898
4899         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4900                 return -EFAULT;
4901         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4902                 return -EINVAL;
4903         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4904
4905         memcpy(skb->data + offset, from, len);
4906         return 0;
4907 }
4908
4909 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4910         .func           = bpf_lwt_seg6_store_bytes,
4911         .gpl_only       = false,
4912         .ret_type       = RET_INTEGER,
4913         .arg1_type      = ARG_PTR_TO_CTX,
4914         .arg2_type      = ARG_ANYTHING,
4915         .arg3_type      = ARG_PTR_TO_MEM,
4916         .arg4_type      = ARG_CONST_SIZE
4917 };
4918
4919 static void bpf_update_srh_state(struct sk_buff *skb)
4920 {
4921         struct seg6_bpf_srh_state *srh_state =
4922                 this_cpu_ptr(&seg6_bpf_srh_states);
4923         int srhoff = 0;
4924
4925         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4926                 srh_state->srh = NULL;
4927         } else {
4928                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4929                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4930                 srh_state->valid = true;
4931         }
4932 }
4933
4934 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4935            u32, action, void *, param, u32, param_len)
4936 {
4937         struct seg6_bpf_srh_state *srh_state =
4938                 this_cpu_ptr(&seg6_bpf_srh_states);
4939         int hdroff = 0;
4940         int err;
4941
4942         switch (action) {
4943         case SEG6_LOCAL_ACTION_END_X:
4944                 if (!seg6_bpf_has_valid_srh(skb))
4945                         return -EBADMSG;
4946                 if (param_len != sizeof(struct in6_addr))
4947                         return -EINVAL;
4948                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4949         case SEG6_LOCAL_ACTION_END_T:
4950                 if (!seg6_bpf_has_valid_srh(skb))
4951                         return -EBADMSG;
4952                 if (param_len != sizeof(int))
4953                         return -EINVAL;
4954                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4955         case SEG6_LOCAL_ACTION_END_DT6:
4956                 if (!seg6_bpf_has_valid_srh(skb))
4957                         return -EBADMSG;
4958                 if (param_len != sizeof(int))
4959                         return -EINVAL;
4960
4961                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4962                         return -EBADMSG;
4963                 if (!pskb_pull(skb, hdroff))
4964                         return -EBADMSG;
4965
4966                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4967                 skb_reset_network_header(skb);
4968                 skb_reset_transport_header(skb);
4969                 skb->encapsulation = 0;
4970
4971                 bpf_compute_data_pointers(skb);
4972                 bpf_update_srh_state(skb);
4973                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4974         case SEG6_LOCAL_ACTION_END_B6:
4975                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4976                         return -EBADMSG;
4977                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4978                                           param, param_len);
4979                 if (!err)
4980                         bpf_update_srh_state(skb);
4981
4982                 return err;
4983         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4984                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4985                         return -EBADMSG;
4986                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4987                                           param, param_len);
4988                 if (!err)
4989                         bpf_update_srh_state(skb);
4990
4991                 return err;
4992         default:
4993                 return -EINVAL;
4994         }
4995 }
4996
4997 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4998         .func           = bpf_lwt_seg6_action,
4999         .gpl_only       = false,
5000         .ret_type       = RET_INTEGER,
5001         .arg1_type      = ARG_PTR_TO_CTX,
5002         .arg2_type      = ARG_ANYTHING,
5003         .arg3_type      = ARG_PTR_TO_MEM,
5004         .arg4_type      = ARG_CONST_SIZE
5005 };
5006
5007 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5008            s32, len)
5009 {
5010         struct seg6_bpf_srh_state *srh_state =
5011                 this_cpu_ptr(&seg6_bpf_srh_states);
5012         struct ipv6_sr_hdr *srh = srh_state->srh;
5013         void *srh_end, *srh_tlvs, *ptr;
5014         struct ipv6hdr *hdr;
5015         int srhoff = 0;
5016         int ret;
5017
5018         if (unlikely(srh == NULL))
5019                 return -EINVAL;
5020
5021         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5022                         ((srh->first_segment + 1) << 4));
5023         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5024                         srh_state->hdrlen);
5025         ptr = skb->data + offset;
5026
5027         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5028                 return -EFAULT;
5029         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5030                 return -EFAULT;
5031
5032         if (len > 0) {
5033                 ret = skb_cow_head(skb, len);
5034                 if (unlikely(ret < 0))
5035                         return ret;
5036
5037                 ret = bpf_skb_net_hdr_push(skb, offset, len);
5038         } else {
5039                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5040         }
5041
5042         bpf_compute_data_pointers(skb);
5043         if (unlikely(ret < 0))
5044                 return ret;
5045
5046         hdr = (struct ipv6hdr *)skb->data;
5047         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5048
5049         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5050                 return -EINVAL;
5051         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5052         srh_state->hdrlen += len;
5053         srh_state->valid = false;
5054         return 0;
5055 }
5056
5057 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5058         .func           = bpf_lwt_seg6_adjust_srh,
5059         .gpl_only       = false,
5060         .ret_type       = RET_INTEGER,
5061         .arg1_type      = ARG_PTR_TO_CTX,
5062         .arg2_type      = ARG_ANYTHING,
5063         .arg3_type      = ARG_ANYTHING,
5064 };
5065 #endif /* CONFIG_IPV6_SEG6_BPF */
5066
5067 #define CONVERT_COMMON_TCP_SOCK_FIELDS(md_type, CONVERT)                \
5068 do {                                                                    \
5069         switch (si->off) {                                              \
5070         case offsetof(md_type, snd_cwnd):                               \
5071                 CONVERT(snd_cwnd); break;                               \
5072         case offsetof(md_type, srtt_us):                                \
5073                 CONVERT(srtt_us); break;                                \
5074         case offsetof(md_type, snd_ssthresh):                           \
5075                 CONVERT(snd_ssthresh); break;                           \
5076         case offsetof(md_type, rcv_nxt):                                \
5077                 CONVERT(rcv_nxt); break;                                \
5078         case offsetof(md_type, snd_nxt):                                \
5079                 CONVERT(snd_nxt); break;                                \
5080         case offsetof(md_type, snd_una):                                \
5081                 CONVERT(snd_una); break;                                \
5082         case offsetof(md_type, mss_cache):                              \
5083                 CONVERT(mss_cache); break;                              \
5084         case offsetof(md_type, ecn_flags):                              \
5085                 CONVERT(ecn_flags); break;                              \
5086         case offsetof(md_type, rate_delivered):                         \
5087                 CONVERT(rate_delivered); break;                         \
5088         case offsetof(md_type, rate_interval_us):                       \
5089                 CONVERT(rate_interval_us); break;                       \
5090         case offsetof(md_type, packets_out):                            \
5091                 CONVERT(packets_out); break;                            \
5092         case offsetof(md_type, retrans_out):                            \
5093                 CONVERT(retrans_out); break;                            \
5094         case offsetof(md_type, total_retrans):                          \
5095                 CONVERT(total_retrans); break;                          \
5096         case offsetof(md_type, segs_in):                                \
5097                 CONVERT(segs_in); break;                                \
5098         case offsetof(md_type, data_segs_in):                           \
5099                 CONVERT(data_segs_in); break;                           \
5100         case offsetof(md_type, segs_out):                               \
5101                 CONVERT(segs_out); break;                               \
5102         case offsetof(md_type, data_segs_out):                          \
5103                 CONVERT(data_segs_out); break;                          \
5104         case offsetof(md_type, lost_out):                               \
5105                 CONVERT(lost_out); break;                               \
5106         case offsetof(md_type, sacked_out):                             \
5107                 CONVERT(sacked_out); break;                             \
5108         case offsetof(md_type, bytes_received):                         \
5109                 CONVERT(bytes_received); break;                         \
5110         case offsetof(md_type, bytes_acked):                            \
5111                 CONVERT(bytes_acked); break;                            \
5112         }                                                               \
5113 } while (0)
5114
5115 #ifdef CONFIG_INET
5116 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5117                               int dif, int sdif, u8 family, u8 proto)
5118 {
5119         bool refcounted = false;
5120         struct sock *sk = NULL;
5121
5122         if (family == AF_INET) {
5123                 __be32 src4 = tuple->ipv4.saddr;
5124                 __be32 dst4 = tuple->ipv4.daddr;
5125
5126                 if (proto == IPPROTO_TCP)
5127                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5128                                            src4, tuple->ipv4.sport,
5129                                            dst4, tuple->ipv4.dport,
5130                                            dif, sdif, &refcounted);
5131                 else
5132                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5133                                                dst4, tuple->ipv4.dport,
5134                                                dif, sdif, &udp_table, NULL);
5135 #if IS_ENABLED(CONFIG_IPV6)
5136         } else {
5137                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5138                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5139
5140                 if (proto == IPPROTO_TCP)
5141                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5142                                             src6, tuple->ipv6.sport,
5143                                             dst6, ntohs(tuple->ipv6.dport),
5144                                             dif, sdif, &refcounted);
5145                 else if (likely(ipv6_bpf_stub))
5146                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5147                                                             src6, tuple->ipv6.sport,
5148                                                             dst6, tuple->ipv6.dport,
5149                                                             dif, sdif,
5150                                                             &udp_table, NULL);
5151 #endif
5152         }
5153
5154         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5155                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5156                 sk = NULL;
5157         }
5158         return sk;
5159 }
5160
5161 /* bpf_sk_lookup performs the core lookup for different types of sockets,
5162  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5163  * Returns the socket as an 'unsigned long' to simplify the casting in the
5164  * callers to satisfy BPF_CALL declarations.
5165  */
5166 static unsigned long
5167 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5168                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5169                 u64 flags)
5170 {
5171         struct sock *sk = NULL;
5172         u8 family = AF_UNSPEC;
5173         struct net *net;
5174         int sdif;
5175
5176         family = len == sizeof(tuple->ipv4) ? AF_INET : AF_INET6;
5177         if (unlikely(family == AF_UNSPEC || flags ||
5178                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5179                 goto out;
5180
5181         if (family == AF_INET)
5182                 sdif = inet_sdif(skb);
5183         else
5184                 sdif = inet6_sdif(skb);
5185
5186         if ((s32)netns_id < 0) {
5187                 net = caller_net;
5188                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5189         } else {
5190                 net = get_net_ns_by_id(caller_net, netns_id);
5191                 if (unlikely(!net))
5192                         goto out;
5193                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5194                 put_net(net);
5195         }
5196
5197         if (sk)
5198                 sk = sk_to_full_sk(sk);
5199 out:
5200         return (unsigned long) sk;
5201 }
5202
5203 static unsigned long
5204 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5205               u8 proto, u64 netns_id, u64 flags)
5206 {
5207         struct net *caller_net;
5208         int ifindex;
5209
5210         if (skb->dev) {
5211                 caller_net = dev_net(skb->dev);
5212                 ifindex = skb->dev->ifindex;
5213         } else {
5214                 caller_net = sock_net(skb->sk);
5215                 ifindex = 0;
5216         }
5217
5218         return __bpf_sk_lookup(skb, tuple, len, caller_net, ifindex,
5219                               proto, netns_id, flags);
5220 }
5221
5222 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5223            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5224 {
5225         return bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, netns_id, flags);
5226 }
5227
5228 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5229         .func           = bpf_sk_lookup_tcp,
5230         .gpl_only       = false,
5231         .pkt_access     = true,
5232         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5233         .arg1_type      = ARG_PTR_TO_CTX,
5234         .arg2_type      = ARG_PTR_TO_MEM,
5235         .arg3_type      = ARG_CONST_SIZE,
5236         .arg4_type      = ARG_ANYTHING,
5237         .arg5_type      = ARG_ANYTHING,
5238 };
5239
5240 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5241            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5242 {
5243         return bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, netns_id, flags);
5244 }
5245
5246 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5247         .func           = bpf_sk_lookup_udp,
5248         .gpl_only       = false,
5249         .pkt_access     = true,
5250         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5251         .arg1_type      = ARG_PTR_TO_CTX,
5252         .arg2_type      = ARG_PTR_TO_MEM,
5253         .arg3_type      = ARG_CONST_SIZE,
5254         .arg4_type      = ARG_ANYTHING,
5255         .arg5_type      = ARG_ANYTHING,
5256 };
5257
5258 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5259 {
5260         if (!sock_flag(sk, SOCK_RCU_FREE))
5261                 sock_gen_put(sk);
5262         return 0;
5263 }
5264
5265 static const struct bpf_func_proto bpf_sk_release_proto = {
5266         .func           = bpf_sk_release,
5267         .gpl_only       = false,
5268         .ret_type       = RET_INTEGER,
5269         .arg1_type      = ARG_PTR_TO_SOCKET,
5270 };
5271
5272 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5273            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5274 {
5275         struct net *caller_net = dev_net(ctx->rxq->dev);
5276         int ifindex = ctx->rxq->dev->ifindex;
5277
5278         return __bpf_sk_lookup(NULL, tuple, len, caller_net, ifindex,
5279                               IPPROTO_UDP, netns_id, flags);
5280 }
5281
5282 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5283         .func           = bpf_xdp_sk_lookup_udp,
5284         .gpl_only       = false,
5285         .pkt_access     = true,
5286         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5287         .arg1_type      = ARG_PTR_TO_CTX,
5288         .arg2_type      = ARG_PTR_TO_MEM,
5289         .arg3_type      = ARG_CONST_SIZE,
5290         .arg4_type      = ARG_ANYTHING,
5291         .arg5_type      = ARG_ANYTHING,
5292 };
5293
5294 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5295            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5296 {
5297         struct net *caller_net = dev_net(ctx->rxq->dev);
5298         int ifindex = ctx->rxq->dev->ifindex;
5299
5300         return __bpf_sk_lookup(NULL, tuple, len, caller_net, ifindex,
5301                               IPPROTO_TCP, netns_id, flags);
5302 }
5303
5304 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5305         .func           = bpf_xdp_sk_lookup_tcp,
5306         .gpl_only       = false,
5307         .pkt_access     = true,
5308         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5309         .arg1_type      = ARG_PTR_TO_CTX,
5310         .arg2_type      = ARG_PTR_TO_MEM,
5311         .arg3_type      = ARG_CONST_SIZE,
5312         .arg4_type      = ARG_ANYTHING,
5313         .arg5_type      = ARG_ANYTHING,
5314 };
5315
5316 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5317            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5318 {
5319         return __bpf_sk_lookup(NULL, tuple, len, sock_net(ctx->sk), 0,
5320                                IPPROTO_TCP, netns_id, flags);
5321 }
5322
5323 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5324         .func           = bpf_sock_addr_sk_lookup_tcp,
5325         .gpl_only       = false,
5326         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5327         .arg1_type      = ARG_PTR_TO_CTX,
5328         .arg2_type      = ARG_PTR_TO_MEM,
5329         .arg3_type      = ARG_CONST_SIZE,
5330         .arg4_type      = ARG_ANYTHING,
5331         .arg5_type      = ARG_ANYTHING,
5332 };
5333
5334 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5335            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5336 {
5337         return __bpf_sk_lookup(NULL, tuple, len, sock_net(ctx->sk), 0,
5338                                IPPROTO_UDP, netns_id, flags);
5339 }
5340
5341 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5342         .func           = bpf_sock_addr_sk_lookup_udp,
5343         .gpl_only       = false,
5344         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5345         .arg1_type      = ARG_PTR_TO_CTX,
5346         .arg2_type      = ARG_PTR_TO_MEM,
5347         .arg3_type      = ARG_CONST_SIZE,
5348         .arg4_type      = ARG_ANYTHING,
5349         .arg5_type      = ARG_ANYTHING,
5350 };
5351
5352 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5353                                   struct bpf_insn_access_aux *info)
5354 {
5355         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, bytes_acked))
5356                 return false;
5357
5358         if (off % size != 0)
5359                 return false;
5360
5361         switch (off) {
5362         case offsetof(struct bpf_tcp_sock, bytes_received):
5363         case offsetof(struct bpf_tcp_sock, bytes_acked):
5364                 return size == sizeof(__u64);
5365         default:
5366                 return size == sizeof(__u32);
5367         }
5368 }
5369
5370 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5371                                     const struct bpf_insn *si,
5372                                     struct bpf_insn *insn_buf,
5373                                     struct bpf_prog *prog, u32 *target_size)
5374 {
5375         struct bpf_insn *insn = insn_buf;
5376
5377 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
5378         do {                                                            \
5379                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) >     \
5380                              FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5381                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5382                                       si->dst_reg, si->src_reg,         \
5383                                       offsetof(struct tcp_sock, FIELD)); \
5384         } while (0)
5385
5386         CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_tcp_sock,
5387                                        BPF_TCP_SOCK_GET_COMMON);
5388
5389         if (insn > insn_buf)
5390                 return insn - insn_buf;
5391
5392         switch (si->off) {
5393         case offsetof(struct bpf_tcp_sock, rtt_min):
5394                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
5395                              sizeof(struct minmax));
5396                 BUILD_BUG_ON(sizeof(struct minmax) <
5397                              sizeof(struct minmax_sample));
5398
5399                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5400                                       offsetof(struct tcp_sock, rtt_min) +
5401                                       offsetof(struct minmax_sample, v));
5402                 break;
5403         }
5404
5405         return insn - insn_buf;
5406 }
5407
5408 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5409 {
5410         sk = sk_to_full_sk(sk);
5411
5412         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5413                 return (unsigned long)sk;
5414
5415         return (unsigned long)NULL;
5416 }
5417
5418 static const struct bpf_func_proto bpf_tcp_sock_proto = {
5419         .func           = bpf_tcp_sock,
5420         .gpl_only       = false,
5421         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
5422         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5423 };
5424
5425 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5426 {
5427         unsigned int iphdr_len;
5428
5429         if (skb->protocol == cpu_to_be16(ETH_P_IP))
5430                 iphdr_len = sizeof(struct iphdr);
5431         else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5432                 iphdr_len = sizeof(struct ipv6hdr);
5433         else
5434                 return 0;
5435
5436         if (skb_headlen(skb) < iphdr_len)
5437                 return 0;
5438
5439         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5440                 return 0;
5441
5442         return INET_ECN_set_ce(skb);
5443 }
5444
5445 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5446         .func           = bpf_skb_ecn_set_ce,
5447         .gpl_only       = false,
5448         .ret_type       = RET_INTEGER,
5449         .arg1_type      = ARG_PTR_TO_CTX,
5450 };
5451 #endif /* CONFIG_INET */
5452
5453 bool bpf_helper_changes_pkt_data(void *func)
5454 {
5455         if (func == bpf_skb_vlan_push ||
5456             func == bpf_skb_vlan_pop ||
5457             func == bpf_skb_store_bytes ||
5458             func == bpf_skb_change_proto ||
5459             func == bpf_skb_change_head ||
5460             func == sk_skb_change_head ||
5461             func == bpf_skb_change_tail ||
5462             func == sk_skb_change_tail ||
5463             func == bpf_skb_adjust_room ||
5464             func == bpf_skb_pull_data ||
5465             func == sk_skb_pull_data ||
5466             func == bpf_clone_redirect ||
5467             func == bpf_l3_csum_replace ||
5468             func == bpf_l4_csum_replace ||
5469             func == bpf_xdp_adjust_head ||
5470             func == bpf_xdp_adjust_meta ||
5471             func == bpf_msg_pull_data ||
5472             func == bpf_msg_push_data ||
5473             func == bpf_msg_pop_data ||
5474             func == bpf_xdp_adjust_tail ||
5475 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5476             func == bpf_lwt_seg6_store_bytes ||
5477             func == bpf_lwt_seg6_adjust_srh ||
5478             func == bpf_lwt_seg6_action ||
5479 #endif
5480             func == bpf_lwt_in_push_encap ||
5481             func == bpf_lwt_xmit_push_encap)
5482                 return true;
5483
5484         return false;
5485 }
5486
5487 static const struct bpf_func_proto *
5488 bpf_base_func_proto(enum bpf_func_id func_id)
5489 {
5490         switch (func_id) {
5491         case BPF_FUNC_map_lookup_elem:
5492                 return &bpf_map_lookup_elem_proto;
5493         case BPF_FUNC_map_update_elem:
5494                 return &bpf_map_update_elem_proto;
5495         case BPF_FUNC_map_delete_elem:
5496                 return &bpf_map_delete_elem_proto;
5497         case BPF_FUNC_map_push_elem:
5498                 return &bpf_map_push_elem_proto;
5499         case BPF_FUNC_map_pop_elem:
5500                 return &bpf_map_pop_elem_proto;
5501         case BPF_FUNC_map_peek_elem:
5502                 return &bpf_map_peek_elem_proto;
5503         case BPF_FUNC_get_prandom_u32:
5504                 return &bpf_get_prandom_u32_proto;
5505         case BPF_FUNC_get_smp_processor_id:
5506                 return &bpf_get_raw_smp_processor_id_proto;
5507         case BPF_FUNC_get_numa_node_id:
5508                 return &bpf_get_numa_node_id_proto;
5509         case BPF_FUNC_tail_call:
5510                 return &bpf_tail_call_proto;
5511         case BPF_FUNC_ktime_get_ns:
5512                 return &bpf_ktime_get_ns_proto;
5513         default:
5514                 break;
5515         }
5516
5517         if (!capable(CAP_SYS_ADMIN))
5518                 return NULL;
5519
5520         switch (func_id) {
5521         case BPF_FUNC_spin_lock:
5522                 return &bpf_spin_lock_proto;
5523         case BPF_FUNC_spin_unlock:
5524                 return &bpf_spin_unlock_proto;
5525         case BPF_FUNC_trace_printk:
5526                 return bpf_get_trace_printk_proto();
5527         default:
5528                 return NULL;
5529         }
5530 }
5531
5532 static const struct bpf_func_proto *
5533 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5534 {
5535         switch (func_id) {
5536         /* inet and inet6 sockets are created in a process
5537          * context so there is always a valid uid/gid
5538          */
5539         case BPF_FUNC_get_current_uid_gid:
5540                 return &bpf_get_current_uid_gid_proto;
5541         case BPF_FUNC_get_local_storage:
5542                 return &bpf_get_local_storage_proto;
5543         default:
5544                 return bpf_base_func_proto(func_id);
5545         }
5546 }
5547
5548 static const struct bpf_func_proto *
5549 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5550 {
5551         switch (func_id) {
5552         /* inet and inet6 sockets are created in a process
5553          * context so there is always a valid uid/gid
5554          */
5555         case BPF_FUNC_get_current_uid_gid:
5556                 return &bpf_get_current_uid_gid_proto;
5557         case BPF_FUNC_bind:
5558                 switch (prog->expected_attach_type) {
5559                 case BPF_CGROUP_INET4_CONNECT:
5560                 case BPF_CGROUP_INET6_CONNECT:
5561                         return &bpf_bind_proto;
5562                 default:
5563                         return NULL;
5564                 }
5565         case BPF_FUNC_get_socket_cookie:
5566                 return &bpf_get_socket_cookie_sock_addr_proto;
5567         case BPF_FUNC_get_local_storage:
5568                 return &bpf_get_local_storage_proto;
5569 #ifdef CONFIG_INET
5570         case BPF_FUNC_sk_lookup_tcp:
5571                 return &bpf_sock_addr_sk_lookup_tcp_proto;
5572         case BPF_FUNC_sk_lookup_udp:
5573                 return &bpf_sock_addr_sk_lookup_udp_proto;
5574         case BPF_FUNC_sk_release:
5575                 return &bpf_sk_release_proto;
5576 #endif /* CONFIG_INET */
5577         default:
5578                 return bpf_base_func_proto(func_id);
5579         }
5580 }
5581
5582 static const struct bpf_func_proto *
5583 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5584 {
5585         switch (func_id) {
5586         case BPF_FUNC_skb_load_bytes:
5587                 return &bpf_skb_load_bytes_proto;
5588         case BPF_FUNC_skb_load_bytes_relative:
5589                 return &bpf_skb_load_bytes_relative_proto;
5590         case BPF_FUNC_get_socket_cookie:
5591                 return &bpf_get_socket_cookie_proto;
5592         case BPF_FUNC_get_socket_uid:
5593                 return &bpf_get_socket_uid_proto;
5594         default:
5595                 return bpf_base_func_proto(func_id);
5596         }
5597 }
5598
5599 static const struct bpf_func_proto *
5600 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5601 {
5602         switch (func_id) {
5603         case BPF_FUNC_get_local_storage:
5604                 return &bpf_get_local_storage_proto;
5605         case BPF_FUNC_sk_fullsock:
5606                 return &bpf_sk_fullsock_proto;
5607 #ifdef CONFIG_INET
5608         case BPF_FUNC_tcp_sock:
5609                 return &bpf_tcp_sock_proto;
5610         case BPF_FUNC_skb_ecn_set_ce:
5611                 return &bpf_skb_ecn_set_ce_proto;
5612 #endif
5613         default:
5614                 return sk_filter_func_proto(func_id, prog);
5615         }
5616 }
5617
5618 static const struct bpf_func_proto *
5619 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5620 {
5621         switch (func_id) {
5622         case BPF_FUNC_skb_store_bytes:
5623                 return &bpf_skb_store_bytes_proto;
5624         case BPF_FUNC_skb_load_bytes:
5625                 return &bpf_skb_load_bytes_proto;
5626         case BPF_FUNC_skb_load_bytes_relative:
5627                 return &bpf_skb_load_bytes_relative_proto;
5628         case BPF_FUNC_skb_pull_data:
5629                 return &bpf_skb_pull_data_proto;
5630         case BPF_FUNC_csum_diff:
5631                 return &bpf_csum_diff_proto;
5632         case BPF_FUNC_csum_update:
5633                 return &bpf_csum_update_proto;
5634         case BPF_FUNC_l3_csum_replace:
5635                 return &bpf_l3_csum_replace_proto;
5636         case BPF_FUNC_l4_csum_replace:
5637                 return &bpf_l4_csum_replace_proto;
5638         case BPF_FUNC_clone_redirect:
5639                 return &bpf_clone_redirect_proto;
5640         case BPF_FUNC_get_cgroup_classid:
5641                 return &bpf_get_cgroup_classid_proto;
5642         case BPF_FUNC_skb_vlan_push:
5643                 return &bpf_skb_vlan_push_proto;
5644         case BPF_FUNC_skb_vlan_pop:
5645                 return &bpf_skb_vlan_pop_proto;
5646         case BPF_FUNC_skb_change_proto:
5647                 return &bpf_skb_change_proto_proto;
5648         case BPF_FUNC_skb_change_type:
5649                 return &bpf_skb_change_type_proto;
5650         case BPF_FUNC_skb_adjust_room:
5651                 return &bpf_skb_adjust_room_proto;
5652         case BPF_FUNC_skb_change_tail:
5653                 return &bpf_skb_change_tail_proto;
5654         case BPF_FUNC_skb_get_tunnel_key:
5655                 return &bpf_skb_get_tunnel_key_proto;
5656         case BPF_FUNC_skb_set_tunnel_key:
5657                 return bpf_get_skb_set_tunnel_proto(func_id);
5658         case BPF_FUNC_skb_get_tunnel_opt:
5659                 return &bpf_skb_get_tunnel_opt_proto;
5660         case BPF_FUNC_skb_set_tunnel_opt:
5661                 return bpf_get_skb_set_tunnel_proto(func_id);
5662         case BPF_FUNC_redirect:
5663                 return &bpf_redirect_proto;
5664         case BPF_FUNC_get_route_realm:
5665                 return &bpf_get_route_realm_proto;
5666         case BPF_FUNC_get_hash_recalc:
5667                 return &bpf_get_hash_recalc_proto;
5668         case BPF_FUNC_set_hash_invalid:
5669                 return &bpf_set_hash_invalid_proto;
5670         case BPF_FUNC_set_hash:
5671                 return &bpf_set_hash_proto;
5672         case BPF_FUNC_perf_event_output:
5673                 return &bpf_skb_event_output_proto;
5674         case BPF_FUNC_get_smp_processor_id:
5675                 return &bpf_get_smp_processor_id_proto;
5676         case BPF_FUNC_skb_under_cgroup:
5677                 return &bpf_skb_under_cgroup_proto;
5678         case BPF_FUNC_get_socket_cookie:
5679                 return &bpf_get_socket_cookie_proto;
5680         case BPF_FUNC_get_socket_uid:
5681                 return &bpf_get_socket_uid_proto;
5682         case BPF_FUNC_fib_lookup:
5683                 return &bpf_skb_fib_lookup_proto;
5684         case BPF_FUNC_sk_fullsock:
5685                 return &bpf_sk_fullsock_proto;
5686 #ifdef CONFIG_XFRM
5687         case BPF_FUNC_skb_get_xfrm_state:
5688                 return &bpf_skb_get_xfrm_state_proto;
5689 #endif
5690 #ifdef CONFIG_SOCK_CGROUP_DATA
5691         case BPF_FUNC_skb_cgroup_id:
5692                 return &bpf_skb_cgroup_id_proto;
5693         case BPF_FUNC_skb_ancestor_cgroup_id:
5694                 return &bpf_skb_ancestor_cgroup_id_proto;
5695 #endif
5696 #ifdef CONFIG_INET
5697         case BPF_FUNC_sk_lookup_tcp:
5698                 return &bpf_sk_lookup_tcp_proto;
5699         case BPF_FUNC_sk_lookup_udp:
5700                 return &bpf_sk_lookup_udp_proto;
5701         case BPF_FUNC_sk_release:
5702                 return &bpf_sk_release_proto;
5703         case BPF_FUNC_tcp_sock:
5704                 return &bpf_tcp_sock_proto;
5705 #endif
5706         default:
5707                 return bpf_base_func_proto(func_id);
5708         }
5709 }
5710
5711 static const struct bpf_func_proto *
5712 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5713 {
5714         switch (func_id) {
5715         case BPF_FUNC_perf_event_output:
5716                 return &bpf_xdp_event_output_proto;
5717         case BPF_FUNC_get_smp_processor_id:
5718                 return &bpf_get_smp_processor_id_proto;
5719         case BPF_FUNC_csum_diff:
5720                 return &bpf_csum_diff_proto;
5721         case BPF_FUNC_xdp_adjust_head:
5722                 return &bpf_xdp_adjust_head_proto;
5723         case BPF_FUNC_xdp_adjust_meta:
5724                 return &bpf_xdp_adjust_meta_proto;
5725         case BPF_FUNC_redirect:
5726                 return &bpf_xdp_redirect_proto;
5727         case BPF_FUNC_redirect_map:
5728                 return &bpf_xdp_redirect_map_proto;
5729         case BPF_FUNC_xdp_adjust_tail:
5730                 return &bpf_xdp_adjust_tail_proto;
5731         case BPF_FUNC_fib_lookup:
5732                 return &bpf_xdp_fib_lookup_proto;
5733 #ifdef CONFIG_INET
5734         case BPF_FUNC_sk_lookup_udp:
5735                 return &bpf_xdp_sk_lookup_udp_proto;
5736         case BPF_FUNC_sk_lookup_tcp:
5737                 return &bpf_xdp_sk_lookup_tcp_proto;
5738         case BPF_FUNC_sk_release:
5739                 return &bpf_sk_release_proto;
5740 #endif
5741         default:
5742                 return bpf_base_func_proto(func_id);
5743         }
5744 }
5745
5746 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
5747 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
5748
5749 static const struct bpf_func_proto *
5750 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5751 {
5752         switch (func_id) {
5753         case BPF_FUNC_setsockopt:
5754                 return &bpf_setsockopt_proto;
5755         case BPF_FUNC_getsockopt:
5756                 return &bpf_getsockopt_proto;
5757         case BPF_FUNC_sock_ops_cb_flags_set:
5758                 return &bpf_sock_ops_cb_flags_set_proto;
5759         case BPF_FUNC_sock_map_update:
5760                 return &bpf_sock_map_update_proto;
5761         case BPF_FUNC_sock_hash_update:
5762                 return &bpf_sock_hash_update_proto;
5763         case BPF_FUNC_get_socket_cookie:
5764                 return &bpf_get_socket_cookie_sock_ops_proto;
5765         case BPF_FUNC_get_local_storage:
5766                 return &bpf_get_local_storage_proto;
5767         case BPF_FUNC_perf_event_output:
5768                 return &bpf_sockopt_event_output_proto;
5769         default:
5770                 return bpf_base_func_proto(func_id);
5771         }
5772 }
5773
5774 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
5775 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
5776
5777 static const struct bpf_func_proto *
5778 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5779 {
5780         switch (func_id) {
5781         case BPF_FUNC_msg_redirect_map:
5782                 return &bpf_msg_redirect_map_proto;
5783         case BPF_FUNC_msg_redirect_hash:
5784                 return &bpf_msg_redirect_hash_proto;
5785         case BPF_FUNC_msg_apply_bytes:
5786                 return &bpf_msg_apply_bytes_proto;
5787         case BPF_FUNC_msg_cork_bytes:
5788                 return &bpf_msg_cork_bytes_proto;
5789         case BPF_FUNC_msg_pull_data:
5790                 return &bpf_msg_pull_data_proto;
5791         case BPF_FUNC_msg_push_data:
5792                 return &bpf_msg_push_data_proto;
5793         case BPF_FUNC_msg_pop_data:
5794                 return &bpf_msg_pop_data_proto;
5795         default:
5796                 return bpf_base_func_proto(func_id);
5797         }
5798 }
5799
5800 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
5801 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
5802
5803 static const struct bpf_func_proto *
5804 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5805 {
5806         switch (func_id) {
5807         case BPF_FUNC_skb_store_bytes:
5808                 return &bpf_skb_store_bytes_proto;
5809         case BPF_FUNC_skb_load_bytes:
5810                 return &bpf_skb_load_bytes_proto;
5811         case BPF_FUNC_skb_pull_data:
5812                 return &sk_skb_pull_data_proto;
5813         case BPF_FUNC_skb_change_tail:
5814                 return &sk_skb_change_tail_proto;
5815         case BPF_FUNC_skb_change_head:
5816                 return &sk_skb_change_head_proto;
5817         case BPF_FUNC_get_socket_cookie:
5818                 return &bpf_get_socket_cookie_proto;
5819         case BPF_FUNC_get_socket_uid:
5820                 return &bpf_get_socket_uid_proto;
5821         case BPF_FUNC_sk_redirect_map:
5822                 return &bpf_sk_redirect_map_proto;
5823         case BPF_FUNC_sk_redirect_hash:
5824                 return &bpf_sk_redirect_hash_proto;
5825 #ifdef CONFIG_INET
5826         case BPF_FUNC_sk_lookup_tcp:
5827                 return &bpf_sk_lookup_tcp_proto;
5828         case BPF_FUNC_sk_lookup_udp:
5829                 return &bpf_sk_lookup_udp_proto;
5830         case BPF_FUNC_sk_release:
5831                 return &bpf_sk_release_proto;
5832 #endif
5833         default:
5834                 return bpf_base_func_proto(func_id);
5835         }
5836 }
5837
5838 static const struct bpf_func_proto *
5839 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5840 {
5841         switch (func_id) {
5842         case BPF_FUNC_skb_load_bytes:
5843                 return &bpf_skb_load_bytes_proto;
5844         default:
5845                 return bpf_base_func_proto(func_id);
5846         }
5847 }
5848
5849 static const struct bpf_func_proto *
5850 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5851 {
5852         switch (func_id) {
5853         case BPF_FUNC_skb_load_bytes:
5854                 return &bpf_skb_load_bytes_proto;
5855         case BPF_FUNC_skb_pull_data:
5856                 return &bpf_skb_pull_data_proto;
5857         case BPF_FUNC_csum_diff:
5858                 return &bpf_csum_diff_proto;
5859         case BPF_FUNC_get_cgroup_classid:
5860                 return &bpf_get_cgroup_classid_proto;
5861         case BPF_FUNC_get_route_realm:
5862                 return &bpf_get_route_realm_proto;
5863         case BPF_FUNC_get_hash_recalc:
5864                 return &bpf_get_hash_recalc_proto;
5865         case BPF_FUNC_perf_event_output:
5866                 return &bpf_skb_event_output_proto;
5867         case BPF_FUNC_get_smp_processor_id:
5868                 return &bpf_get_smp_processor_id_proto;
5869         case BPF_FUNC_skb_under_cgroup:
5870                 return &bpf_skb_under_cgroup_proto;
5871         default:
5872                 return bpf_base_func_proto(func_id);
5873         }
5874 }
5875
5876 static const struct bpf_func_proto *
5877 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5878 {
5879         switch (func_id) {
5880         case BPF_FUNC_lwt_push_encap:
5881                 return &bpf_lwt_in_push_encap_proto;
5882         default:
5883                 return lwt_out_func_proto(func_id, prog);
5884         }
5885 }
5886
5887 static const struct bpf_func_proto *
5888 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5889 {
5890         switch (func_id) {
5891         case BPF_FUNC_skb_get_tunnel_key:
5892                 return &bpf_skb_get_tunnel_key_proto;
5893         case BPF_FUNC_skb_set_tunnel_key:
5894                 return bpf_get_skb_set_tunnel_proto(func_id);
5895         case BPF_FUNC_skb_get_tunnel_opt:
5896                 return &bpf_skb_get_tunnel_opt_proto;
5897         case BPF_FUNC_skb_set_tunnel_opt:
5898                 return bpf_get_skb_set_tunnel_proto(func_id);
5899         case BPF_FUNC_redirect:
5900                 return &bpf_redirect_proto;
5901         case BPF_FUNC_clone_redirect:
5902                 return &bpf_clone_redirect_proto;
5903         case BPF_FUNC_skb_change_tail:
5904                 return &bpf_skb_change_tail_proto;
5905         case BPF_FUNC_skb_change_head:
5906                 return &bpf_skb_change_head_proto;
5907         case BPF_FUNC_skb_store_bytes:
5908                 return &bpf_skb_store_bytes_proto;
5909         case BPF_FUNC_csum_update:
5910                 return &bpf_csum_update_proto;
5911         case BPF_FUNC_l3_csum_replace:
5912                 return &bpf_l3_csum_replace_proto;
5913         case BPF_FUNC_l4_csum_replace:
5914                 return &bpf_l4_csum_replace_proto;
5915         case BPF_FUNC_set_hash_invalid:
5916                 return &bpf_set_hash_invalid_proto;
5917         case BPF_FUNC_lwt_push_encap:
5918                 return &bpf_lwt_xmit_push_encap_proto;
5919         default:
5920                 return lwt_out_func_proto(func_id, prog);
5921         }
5922 }
5923
5924 static const struct bpf_func_proto *
5925 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5926 {
5927         switch (func_id) {
5928 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5929         case BPF_FUNC_lwt_seg6_store_bytes:
5930                 return &bpf_lwt_seg6_store_bytes_proto;
5931         case BPF_FUNC_lwt_seg6_action:
5932                 return &bpf_lwt_seg6_action_proto;
5933         case BPF_FUNC_lwt_seg6_adjust_srh:
5934                 return &bpf_lwt_seg6_adjust_srh_proto;
5935 #endif
5936         default:
5937                 return lwt_out_func_proto(func_id, prog);
5938         }
5939 }
5940
5941 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5942                                     const struct bpf_prog *prog,
5943                                     struct bpf_insn_access_aux *info)
5944 {
5945         const int size_default = sizeof(__u32);
5946
5947         if (off < 0 || off >= sizeof(struct __sk_buff))
5948                 return false;
5949
5950         /* The verifier guarantees that size > 0. */
5951         if (off % size != 0)
5952                 return false;
5953
5954         switch (off) {
5955         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5956                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5957                         return false;
5958                 break;
5959         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5960         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5961         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5962         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5963         case bpf_ctx_range(struct __sk_buff, data):
5964         case bpf_ctx_range(struct __sk_buff, data_meta):
5965         case bpf_ctx_range(struct __sk_buff, data_end):
5966                 if (size != size_default)
5967                         return false;
5968                 break;
5969         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5970                 if (size != sizeof(__u64))
5971                         return false;
5972                 break;
5973         case bpf_ctx_range(struct __sk_buff, tstamp):
5974                 if (size != sizeof(__u64))
5975                         return false;
5976                 break;
5977         case offsetof(struct __sk_buff, sk):
5978                 if (type == BPF_WRITE || size != sizeof(__u64))
5979                         return false;
5980                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
5981                 break;
5982         default:
5983                 /* Only narrow read access allowed for now. */
5984                 if (type == BPF_WRITE) {
5985                         if (size != size_default)
5986                                 return false;
5987                 } else {
5988                         bpf_ctx_record_field_size(info, size_default);
5989                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5990                                 return false;
5991                 }
5992         }
5993
5994         return true;
5995 }
5996
5997 static bool sk_filter_is_valid_access(int off, int size,
5998                                       enum bpf_access_type type,
5999                                       const struct bpf_prog *prog,
6000                                       struct bpf_insn_access_aux *info)
6001 {
6002         switch (off) {
6003         case bpf_ctx_range(struct __sk_buff, tc_classid):
6004         case bpf_ctx_range(struct __sk_buff, data):
6005         case bpf_ctx_range(struct __sk_buff, data_meta):
6006         case bpf_ctx_range(struct __sk_buff, data_end):
6007         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6008         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6009         case bpf_ctx_range(struct __sk_buff, tstamp):
6010         case bpf_ctx_range(struct __sk_buff, wire_len):
6011                 return false;
6012         }
6013
6014         if (type == BPF_WRITE) {
6015                 switch (off) {
6016                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6017                         break;
6018                 default:
6019                         return false;
6020                 }
6021         }
6022
6023         return bpf_skb_is_valid_access(off, size, type, prog, info);
6024 }
6025
6026 static bool cg_skb_is_valid_access(int off, int size,
6027                                    enum bpf_access_type type,
6028                                    const struct bpf_prog *prog,
6029                                    struct bpf_insn_access_aux *info)
6030 {
6031         switch (off) {
6032         case bpf_ctx_range(struct __sk_buff, tc_classid):
6033         case bpf_ctx_range(struct __sk_buff, data_meta):
6034         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6035         case bpf_ctx_range(struct __sk_buff, wire_len):
6036                 return false;
6037         case bpf_ctx_range(struct __sk_buff, data):
6038         case bpf_ctx_range(struct __sk_buff, data_end):
6039                 if (!capable(CAP_SYS_ADMIN))
6040                         return false;
6041                 break;
6042         }
6043
6044         if (type == BPF_WRITE) {
6045                 switch (off) {
6046                 case bpf_ctx_range(struct __sk_buff, mark):
6047                 case bpf_ctx_range(struct __sk_buff, priority):
6048                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6049                         break;
6050                 case bpf_ctx_range(struct __sk_buff, tstamp):
6051                         if (!capable(CAP_SYS_ADMIN))
6052                                 return false;
6053                         break;
6054                 default:
6055                         return false;
6056                 }
6057         }
6058
6059         switch (off) {
6060         case bpf_ctx_range(struct __sk_buff, data):
6061                 info->reg_type = PTR_TO_PACKET;
6062                 break;
6063         case bpf_ctx_range(struct __sk_buff, data_end):
6064                 info->reg_type = PTR_TO_PACKET_END;
6065                 break;
6066         }
6067
6068         return bpf_skb_is_valid_access(off, size, type, prog, info);
6069 }
6070
6071 static bool lwt_is_valid_access(int off, int size,
6072                                 enum bpf_access_type type,
6073                                 const struct bpf_prog *prog,
6074                                 struct bpf_insn_access_aux *info)
6075 {
6076         switch (off) {
6077         case bpf_ctx_range(struct __sk_buff, tc_classid):
6078         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6079         case bpf_ctx_range(struct __sk_buff, data_meta):
6080         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6081         case bpf_ctx_range(struct __sk_buff, tstamp):
6082         case bpf_ctx_range(struct __sk_buff, wire_len):
6083                 return false;
6084         }
6085
6086         if (type == BPF_WRITE) {
6087                 switch (off) {
6088                 case bpf_ctx_range(struct __sk_buff, mark):
6089                 case bpf_ctx_range(struct __sk_buff, priority):
6090                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6091                         break;
6092                 default:
6093                         return false;
6094                 }
6095         }
6096
6097         switch (off) {
6098         case bpf_ctx_range(struct __sk_buff, data):
6099                 info->reg_type = PTR_TO_PACKET;
6100                 break;
6101         case bpf_ctx_range(struct __sk_buff, data_end):
6102                 info->reg_type = PTR_TO_PACKET_END;
6103                 break;
6104         }
6105
6106         return bpf_skb_is_valid_access(off, size, type, prog, info);
6107 }
6108
6109 /* Attach type specific accesses */
6110 static bool __sock_filter_check_attach_type(int off,
6111                                             enum bpf_access_type access_type,
6112                                             enum bpf_attach_type attach_type)
6113 {
6114         switch (off) {
6115         case offsetof(struct bpf_sock, bound_dev_if):
6116         case offsetof(struct bpf_sock, mark):
6117         case offsetof(struct bpf_sock, priority):
6118                 switch (attach_type) {
6119                 case BPF_CGROUP_INET_SOCK_CREATE:
6120                         goto full_access;
6121                 default:
6122                         return false;
6123                 }
6124         case bpf_ctx_range(struct bpf_sock, src_ip4):
6125                 switch (attach_type) {
6126                 case BPF_CGROUP_INET4_POST_BIND:
6127                         goto read_only;
6128                 default:
6129                         return false;
6130                 }
6131         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6132                 switch (attach_type) {
6133                 case BPF_CGROUP_INET6_POST_BIND:
6134                         goto read_only;
6135                 default:
6136                         return false;
6137                 }
6138         case bpf_ctx_range(struct bpf_sock, src_port):
6139                 switch (attach_type) {
6140                 case BPF_CGROUP_INET4_POST_BIND:
6141                 case BPF_CGROUP_INET6_POST_BIND:
6142                         goto read_only;
6143                 default:
6144                         return false;
6145                 }
6146         }
6147 read_only:
6148         return access_type == BPF_READ;
6149 full_access:
6150         return true;
6151 }
6152
6153 bool bpf_sock_common_is_valid_access(int off, int size,
6154                                      enum bpf_access_type type,
6155                                      struct bpf_insn_access_aux *info)
6156 {
6157         switch (off) {
6158         case bpf_ctx_range_till(struct bpf_sock, type, priority):
6159                 return false;
6160         default:
6161                 return bpf_sock_is_valid_access(off, size, type, info);
6162         }
6163 }
6164
6165 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6166                               struct bpf_insn_access_aux *info)
6167 {
6168         const int size_default = sizeof(__u32);
6169
6170         if (off < 0 || off >= sizeof(struct bpf_sock))
6171                 return false;
6172         if (off % size != 0)
6173                 return false;
6174
6175         switch (off) {
6176         case offsetof(struct bpf_sock, state):
6177         case offsetof(struct bpf_sock, family):
6178         case offsetof(struct bpf_sock, type):
6179         case offsetof(struct bpf_sock, protocol):
6180         case offsetof(struct bpf_sock, dst_port):
6181         case offsetof(struct bpf_sock, src_port):
6182         case bpf_ctx_range(struct bpf_sock, src_ip4):
6183         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6184         case bpf_ctx_range(struct bpf_sock, dst_ip4):
6185         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6186                 bpf_ctx_record_field_size(info, size_default);
6187                 return bpf_ctx_narrow_access_ok(off, size, size_default);
6188         }
6189
6190         return size == size_default;
6191 }
6192
6193 static bool sock_filter_is_valid_access(int off, int size,
6194                                         enum bpf_access_type type,
6195                                         const struct bpf_prog *prog,
6196                                         struct bpf_insn_access_aux *info)
6197 {
6198         if (!bpf_sock_is_valid_access(off, size, type, info))
6199                 return false;
6200         return __sock_filter_check_attach_type(off, type,
6201                                                prog->expected_attach_type);
6202 }
6203
6204 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6205                              const struct bpf_prog *prog)
6206 {
6207         /* Neither direct read nor direct write requires any preliminary
6208          * action.
6209          */
6210         return 0;
6211 }
6212
6213 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6214                                 const struct bpf_prog *prog, int drop_verdict)
6215 {
6216         struct bpf_insn *insn = insn_buf;
6217
6218         if (!direct_write)
6219                 return 0;
6220
6221         /* if (!skb->cloned)
6222          *       goto start;
6223          *
6224          * (Fast-path, otherwise approximation that we might be
6225          *  a clone, do the rest in helper.)
6226          */
6227         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6228         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6229         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6230
6231         /* ret = bpf_skb_pull_data(skb, 0); */
6232         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6233         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6234         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6235                                BPF_FUNC_skb_pull_data);
6236         /* if (!ret)
6237          *      goto restore;
6238          * return TC_ACT_SHOT;
6239          */
6240         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6241         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6242         *insn++ = BPF_EXIT_INSN();
6243
6244         /* restore: */
6245         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6246         /* start: */
6247         *insn++ = prog->insnsi[0];
6248
6249         return insn - insn_buf;
6250 }
6251
6252 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6253                           struct bpf_insn *insn_buf)
6254 {
6255         bool indirect = BPF_MODE(orig->code) == BPF_IND;
6256         struct bpf_insn *insn = insn_buf;
6257
6258         /* We're guaranteed here that CTX is in R6. */
6259         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6260         if (!indirect) {
6261                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6262         } else {
6263                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6264                 if (orig->imm)
6265                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6266         }
6267
6268         switch (BPF_SIZE(orig->code)) {
6269         case BPF_B:
6270                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6271                 break;
6272         case BPF_H:
6273                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6274                 break;
6275         case BPF_W:
6276                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6277                 break;
6278         }
6279
6280         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6281         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6282         *insn++ = BPF_EXIT_INSN();
6283
6284         return insn - insn_buf;
6285 }
6286
6287 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6288                                const struct bpf_prog *prog)
6289 {
6290         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6291 }
6292
6293 static bool tc_cls_act_is_valid_access(int off, int size,
6294                                        enum bpf_access_type type,
6295                                        const struct bpf_prog *prog,
6296                                        struct bpf_insn_access_aux *info)
6297 {
6298         if (type == BPF_WRITE) {
6299                 switch (off) {
6300                 case bpf_ctx_range(struct __sk_buff, mark):
6301                 case bpf_ctx_range(struct __sk_buff, tc_index):
6302                 case bpf_ctx_range(struct __sk_buff, priority):
6303                 case bpf_ctx_range(struct __sk_buff, tc_classid):
6304                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6305                 case bpf_ctx_range(struct __sk_buff, tstamp):
6306                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6307                         break;
6308                 default:
6309                         return false;
6310                 }
6311         }
6312
6313         switch (off) {
6314         case bpf_ctx_range(struct __sk_buff, data):
6315                 info->reg_type = PTR_TO_PACKET;
6316                 break;
6317         case bpf_ctx_range(struct __sk_buff, data_meta):
6318                 info->reg_type = PTR_TO_PACKET_META;
6319                 break;
6320         case bpf_ctx_range(struct __sk_buff, data_end):
6321                 info->reg_type = PTR_TO_PACKET_END;
6322                 break;
6323         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6324         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6325                 return false;
6326         }
6327
6328         return bpf_skb_is_valid_access(off, size, type, prog, info);
6329 }
6330
6331 static bool __is_valid_xdp_access(int off, int size)
6332 {
6333         if (off < 0 || off >= sizeof(struct xdp_md))
6334                 return false;
6335         if (off % size != 0)
6336                 return false;
6337         if (size != sizeof(__u32))
6338                 return false;
6339
6340         return true;
6341 }
6342
6343 static bool xdp_is_valid_access(int off, int size,
6344                                 enum bpf_access_type type,
6345                                 const struct bpf_prog *prog,
6346                                 struct bpf_insn_access_aux *info)
6347 {
6348         if (type == BPF_WRITE) {
6349                 if (bpf_prog_is_dev_bound(prog->aux)) {
6350                         switch (off) {
6351                         case offsetof(struct xdp_md, rx_queue_index):
6352                                 return __is_valid_xdp_access(off, size);
6353                         }
6354                 }
6355                 return false;
6356         }
6357
6358         switch (off) {
6359         case offsetof(struct xdp_md, data):
6360                 info->reg_type = PTR_TO_PACKET;
6361                 break;
6362         case offsetof(struct xdp_md, data_meta):
6363                 info->reg_type = PTR_TO_PACKET_META;
6364                 break;
6365         case offsetof(struct xdp_md, data_end):
6366                 info->reg_type = PTR_TO_PACKET_END;
6367                 break;
6368         }
6369
6370         return __is_valid_xdp_access(off, size);
6371 }
6372
6373 void bpf_warn_invalid_xdp_action(u32 act)
6374 {
6375         const u32 act_max = XDP_REDIRECT;
6376
6377         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6378                   act > act_max ? "Illegal" : "Driver unsupported",
6379                   act);
6380 }
6381 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6382
6383 static bool sock_addr_is_valid_access(int off, int size,
6384                                       enum bpf_access_type type,
6385                                       const struct bpf_prog *prog,
6386                                       struct bpf_insn_access_aux *info)
6387 {
6388         const int size_default = sizeof(__u32);
6389
6390         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6391                 return false;
6392         if (off % size != 0)
6393                 return false;
6394
6395         /* Disallow access to IPv6 fields from IPv4 contex and vise
6396          * versa.
6397          */
6398         switch (off) {
6399         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6400                 switch (prog->expected_attach_type) {
6401                 case BPF_CGROUP_INET4_BIND:
6402                 case BPF_CGROUP_INET4_CONNECT:
6403                 case BPF_CGROUP_UDP4_SENDMSG:
6404                         break;
6405                 default:
6406                         return false;
6407                 }
6408                 break;
6409         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6410                 switch (prog->expected_attach_type) {
6411                 case BPF_CGROUP_INET6_BIND:
6412                 case BPF_CGROUP_INET6_CONNECT:
6413                 case BPF_CGROUP_UDP6_SENDMSG:
6414                         break;
6415                 default:
6416                         return false;
6417                 }
6418                 break;
6419         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6420                 switch (prog->expected_attach_type) {
6421                 case BPF_CGROUP_UDP4_SENDMSG:
6422                         break;
6423                 default:
6424                         return false;
6425                 }
6426                 break;
6427         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6428                                 msg_src_ip6[3]):
6429                 switch (prog->expected_attach_type) {
6430                 case BPF_CGROUP_UDP6_SENDMSG:
6431                         break;
6432                 default:
6433                         return false;
6434                 }
6435                 break;
6436         }
6437
6438         switch (off) {
6439         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6440         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6441         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6442         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6443                                 msg_src_ip6[3]):
6444                 /* Only narrow read access allowed for now. */
6445                 if (type == BPF_READ) {
6446                         bpf_ctx_record_field_size(info, size_default);
6447                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6448                                 return false;
6449                 } else {
6450                         if (size != size_default)
6451                                 return false;
6452                 }
6453                 break;
6454         case bpf_ctx_range(struct bpf_sock_addr, user_port):
6455                 if (size != size_default)
6456                         return false;
6457                 break;
6458         default:
6459                 if (type == BPF_READ) {
6460                         if (size != size_default)
6461                                 return false;
6462                 } else {
6463                         return false;
6464                 }
6465         }
6466
6467         return true;
6468 }
6469
6470 static bool sock_ops_is_valid_access(int off, int size,
6471                                      enum bpf_access_type type,
6472                                      const struct bpf_prog *prog,
6473                                      struct bpf_insn_access_aux *info)
6474 {
6475         const int size_default = sizeof(__u32);
6476
6477         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6478                 return false;
6479
6480         /* The verifier guarantees that size > 0. */
6481         if (off % size != 0)
6482                 return false;
6483
6484         if (type == BPF_WRITE) {
6485                 switch (off) {
6486                 case offsetof(struct bpf_sock_ops, reply):
6487                 case offsetof(struct bpf_sock_ops, sk_txhash):
6488                         if (size != size_default)
6489                                 return false;
6490                         break;
6491                 default:
6492                         return false;
6493                 }
6494         } else {
6495                 switch (off) {
6496                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6497                                         bytes_acked):
6498                         if (size != sizeof(__u64))
6499                                 return false;
6500                         break;
6501                 default:
6502                         if (size != size_default)
6503                                 return false;
6504                         break;
6505                 }
6506         }
6507
6508         return true;
6509 }
6510
6511 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6512                            const struct bpf_prog *prog)
6513 {
6514         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6515 }
6516
6517 static bool sk_skb_is_valid_access(int off, int size,
6518                                    enum bpf_access_type type,
6519                                    const struct bpf_prog *prog,
6520                                    struct bpf_insn_access_aux *info)
6521 {
6522         switch (off) {
6523         case bpf_ctx_range(struct __sk_buff, tc_classid):
6524         case bpf_ctx_range(struct __sk_buff, data_meta):
6525         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6526         case bpf_ctx_range(struct __sk_buff, tstamp):
6527         case bpf_ctx_range(struct __sk_buff, wire_len):
6528                 return false;
6529         }
6530
6531         if (type == BPF_WRITE) {
6532                 switch (off) {
6533                 case bpf_ctx_range(struct __sk_buff, tc_index):
6534                 case bpf_ctx_range(struct __sk_buff, priority):
6535                         break;
6536                 default:
6537                         return false;
6538                 }
6539         }
6540
6541         switch (off) {
6542         case bpf_ctx_range(struct __sk_buff, mark):
6543                 return false;
6544         case bpf_ctx_range(struct __sk_buff, data):
6545                 info->reg_type = PTR_TO_PACKET;
6546                 break;
6547         case bpf_ctx_range(struct __sk_buff, data_end):
6548                 info->reg_type = PTR_TO_PACKET_END;
6549                 break;
6550         }
6551
6552         return bpf_skb_is_valid_access(off, size, type, prog, info);
6553 }
6554
6555 static bool sk_msg_is_valid_access(int off, int size,
6556                                    enum bpf_access_type type,
6557                                    const struct bpf_prog *prog,
6558                                    struct bpf_insn_access_aux *info)
6559 {
6560         if (type == BPF_WRITE)
6561                 return false;
6562
6563         if (off % size != 0)
6564                 return false;
6565
6566         switch (off) {
6567         case offsetof(struct sk_msg_md, data):
6568                 info->reg_type = PTR_TO_PACKET;
6569                 if (size != sizeof(__u64))
6570                         return false;
6571                 break;
6572         case offsetof(struct sk_msg_md, data_end):
6573                 info->reg_type = PTR_TO_PACKET_END;
6574                 if (size != sizeof(__u64))
6575                         return false;
6576                 break;
6577         case bpf_ctx_range(struct sk_msg_md, family):
6578         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
6579         case bpf_ctx_range(struct sk_msg_md, local_ip4):
6580         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
6581         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
6582         case bpf_ctx_range(struct sk_msg_md, remote_port):
6583         case bpf_ctx_range(struct sk_msg_md, local_port):
6584         case bpf_ctx_range(struct sk_msg_md, size):
6585                 if (size != sizeof(__u32))
6586                         return false;
6587                 break;
6588         default:
6589                 return false;
6590         }
6591         return true;
6592 }
6593
6594 static bool flow_dissector_is_valid_access(int off, int size,
6595                                            enum bpf_access_type type,
6596                                            const struct bpf_prog *prog,
6597                                            struct bpf_insn_access_aux *info)
6598 {
6599         if (type == BPF_WRITE) {
6600                 switch (off) {
6601                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6602                         break;
6603                 default:
6604                         return false;
6605                 }
6606         }
6607
6608         switch (off) {
6609         case bpf_ctx_range(struct __sk_buff, data):
6610                 info->reg_type = PTR_TO_PACKET;
6611                 break;
6612         case bpf_ctx_range(struct __sk_buff, data_end):
6613                 info->reg_type = PTR_TO_PACKET_END;
6614                 break;
6615         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6616                 info->reg_type = PTR_TO_FLOW_KEYS;
6617                 break;
6618         case bpf_ctx_range(struct __sk_buff, tc_classid):
6619         case bpf_ctx_range(struct __sk_buff, data_meta):
6620         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6621         case bpf_ctx_range(struct __sk_buff, tstamp):
6622         case bpf_ctx_range(struct __sk_buff, wire_len):
6623                 return false;
6624         }
6625
6626         return bpf_skb_is_valid_access(off, size, type, prog, info);
6627 }
6628
6629 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
6630                                   const struct bpf_insn *si,
6631                                   struct bpf_insn *insn_buf,
6632                                   struct bpf_prog *prog, u32 *target_size)
6633 {
6634         struct bpf_insn *insn = insn_buf;
6635         int off;
6636
6637         switch (si->off) {
6638         case offsetof(struct __sk_buff, len):
6639                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6640                                       bpf_target_off(struct sk_buff, len, 4,
6641                                                      target_size));
6642                 break;
6643
6644         case offsetof(struct __sk_buff, protocol):
6645                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6646                                       bpf_target_off(struct sk_buff, protocol, 2,
6647                                                      target_size));
6648                 break;
6649
6650         case offsetof(struct __sk_buff, vlan_proto):
6651                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6652                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
6653                                                      target_size));
6654                 break;
6655
6656         case offsetof(struct __sk_buff, priority):
6657                 if (type == BPF_WRITE)
6658                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6659                                               bpf_target_off(struct sk_buff, priority, 4,
6660                                                              target_size));
6661                 else
6662                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6663                                               bpf_target_off(struct sk_buff, priority, 4,
6664                                                              target_size));
6665                 break;
6666
6667         case offsetof(struct __sk_buff, ingress_ifindex):
6668                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6669                                       bpf_target_off(struct sk_buff, skb_iif, 4,
6670                                                      target_size));
6671                 break;
6672
6673         case offsetof(struct __sk_buff, ifindex):
6674                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6675                                       si->dst_reg, si->src_reg,
6676                                       offsetof(struct sk_buff, dev));
6677                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
6678                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6679                                       bpf_target_off(struct net_device, ifindex, 4,
6680                                                      target_size));
6681                 break;
6682
6683         case offsetof(struct __sk_buff, hash):
6684                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6685                                       bpf_target_off(struct sk_buff, hash, 4,
6686                                                      target_size));
6687                 break;
6688
6689         case offsetof(struct __sk_buff, mark):
6690                 if (type == BPF_WRITE)
6691                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6692                                               bpf_target_off(struct sk_buff, mark, 4,
6693                                                              target_size));
6694                 else
6695                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6696                                               bpf_target_off(struct sk_buff, mark, 4,
6697                                                              target_size));
6698                 break;
6699
6700         case offsetof(struct __sk_buff, pkt_type):
6701                 *target_size = 1;
6702                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6703                                       PKT_TYPE_OFFSET());
6704                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
6705 #ifdef __BIG_ENDIAN_BITFIELD
6706                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
6707 #endif
6708                 break;
6709
6710         case offsetof(struct __sk_buff, queue_mapping):
6711                 if (type == BPF_WRITE) {
6712                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
6713                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6714                                               bpf_target_off(struct sk_buff,
6715                                                              queue_mapping,
6716                                                              2, target_size));
6717                 } else {
6718                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6719                                               bpf_target_off(struct sk_buff,
6720                                                              queue_mapping,
6721                                                              2, target_size));
6722                 }
6723                 break;
6724
6725         case offsetof(struct __sk_buff, vlan_present):
6726                 *target_size = 1;
6727                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6728                                       PKT_VLAN_PRESENT_OFFSET());
6729                 if (PKT_VLAN_PRESENT_BIT)
6730                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
6731                 if (PKT_VLAN_PRESENT_BIT < 7)
6732                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
6733                 break;
6734
6735         case offsetof(struct __sk_buff, vlan_tci):
6736                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6737                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
6738                                                      target_size));
6739                 break;
6740
6741         case offsetof(struct __sk_buff, cb[0]) ...
6742              offsetofend(struct __sk_buff, cb[4]) - 1:
6743                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
6744                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
6745                               offsetof(struct qdisc_skb_cb, data)) %
6746                              sizeof(__u64));
6747
6748                 prog->cb_access = 1;
6749                 off  = si->off;
6750                 off -= offsetof(struct __sk_buff, cb[0]);
6751                 off += offsetof(struct sk_buff, cb);
6752                 off += offsetof(struct qdisc_skb_cb, data);
6753                 if (type == BPF_WRITE)
6754                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
6755                                               si->src_reg, off);
6756                 else
6757                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
6758                                               si->src_reg, off);
6759                 break;
6760
6761         case offsetof(struct __sk_buff, tc_classid):
6762                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
6763
6764                 off  = si->off;
6765                 off -= offsetof(struct __sk_buff, tc_classid);
6766                 off += offsetof(struct sk_buff, cb);
6767                 off += offsetof(struct qdisc_skb_cb, tc_classid);
6768                 *target_size = 2;
6769                 if (type == BPF_WRITE)
6770                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
6771                                               si->src_reg, off);
6772                 else
6773                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
6774                                               si->src_reg, off);
6775                 break;
6776
6777         case offsetof(struct __sk_buff, data):
6778                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6779                                       si->dst_reg, si->src_reg,
6780                                       offsetof(struct sk_buff, data));
6781                 break;
6782
6783         case offsetof(struct __sk_buff, data_meta):
6784                 off  = si->off;
6785                 off -= offsetof(struct __sk_buff, data_meta);
6786                 off += offsetof(struct sk_buff, cb);
6787                 off += offsetof(struct bpf_skb_data_end, data_meta);
6788                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6789                                       si->src_reg, off);
6790                 break;
6791
6792         case offsetof(struct __sk_buff, data_end):
6793                 off  = si->off;
6794                 off -= offsetof(struct __sk_buff, data_end);
6795                 off += offsetof(struct sk_buff, cb);
6796                 off += offsetof(struct bpf_skb_data_end, data_end);
6797                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6798                                       si->src_reg, off);
6799                 break;
6800
6801         case offsetof(struct __sk_buff, tc_index):
6802 #ifdef CONFIG_NET_SCHED
6803                 if (type == BPF_WRITE)
6804                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6805                                               bpf_target_off(struct sk_buff, tc_index, 2,
6806                                                              target_size));
6807                 else
6808                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6809                                               bpf_target_off(struct sk_buff, tc_index, 2,
6810                                                              target_size));
6811 #else
6812                 *target_size = 2;
6813                 if (type == BPF_WRITE)
6814                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
6815                 else
6816                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6817 #endif
6818                 break;
6819
6820         case offsetof(struct __sk_buff, napi_id):
6821 #if defined(CONFIG_NET_RX_BUSY_POLL)
6822                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6823                                       bpf_target_off(struct sk_buff, napi_id, 4,
6824                                                      target_size));
6825                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
6826                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6827 #else
6828                 *target_size = 4;
6829                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6830 #endif
6831                 break;
6832         case offsetof(struct __sk_buff, family):
6833                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6834
6835                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6836                                       si->dst_reg, si->src_reg,
6837                                       offsetof(struct sk_buff, sk));
6838                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6839                                       bpf_target_off(struct sock_common,
6840                                                      skc_family,
6841                                                      2, target_size));
6842                 break;
6843         case offsetof(struct __sk_buff, remote_ip4):
6844                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6845
6846                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6847                                       si->dst_reg, si->src_reg,
6848                                       offsetof(struct sk_buff, sk));
6849                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6850                                       bpf_target_off(struct sock_common,
6851                                                      skc_daddr,
6852                                                      4, target_size));
6853                 break;
6854         case offsetof(struct __sk_buff, local_ip4):
6855                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6856                                           skc_rcv_saddr) != 4);
6857
6858                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6859                                       si->dst_reg, si->src_reg,
6860                                       offsetof(struct sk_buff, sk));
6861                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6862                                       bpf_target_off(struct sock_common,
6863                                                      skc_rcv_saddr,
6864                                                      4, target_size));
6865                 break;
6866         case offsetof(struct __sk_buff, remote_ip6[0]) ...
6867              offsetof(struct __sk_buff, remote_ip6[3]):
6868 #if IS_ENABLED(CONFIG_IPV6)
6869                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6870                                           skc_v6_daddr.s6_addr32[0]) != 4);
6871
6872                 off = si->off;
6873                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
6874
6875                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6876                                       si->dst_reg, si->src_reg,
6877                                       offsetof(struct sk_buff, sk));
6878                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6879                                       offsetof(struct sock_common,
6880                                                skc_v6_daddr.s6_addr32[0]) +
6881                                       off);
6882 #else
6883                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6884 #endif
6885                 break;
6886         case offsetof(struct __sk_buff, local_ip6[0]) ...
6887              offsetof(struct __sk_buff, local_ip6[3]):
6888 #if IS_ENABLED(CONFIG_IPV6)
6889                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6890                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6891
6892                 off = si->off;
6893                 off -= offsetof(struct __sk_buff, local_ip6[0]);
6894
6895                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6896                                       si->dst_reg, si->src_reg,
6897                                       offsetof(struct sk_buff, sk));
6898                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6899                                       offsetof(struct sock_common,
6900                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6901                                       off);
6902 #else
6903                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6904 #endif
6905                 break;
6906
6907         case offsetof(struct __sk_buff, remote_port):
6908                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6909
6910                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6911                                       si->dst_reg, si->src_reg,
6912                                       offsetof(struct sk_buff, sk));
6913                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6914                                       bpf_target_off(struct sock_common,
6915                                                      skc_dport,
6916                                                      2, target_size));
6917 #ifndef __BIG_ENDIAN_BITFIELD
6918                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6919 #endif
6920                 break;
6921
6922         case offsetof(struct __sk_buff, local_port):
6923                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6924
6925                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6926                                       si->dst_reg, si->src_reg,
6927                                       offsetof(struct sk_buff, sk));
6928                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6929                                       bpf_target_off(struct sock_common,
6930                                                      skc_num, 2, target_size));
6931                 break;
6932
6933         case offsetof(struct __sk_buff, flow_keys):
6934                 off  = si->off;
6935                 off -= offsetof(struct __sk_buff, flow_keys);
6936                 off += offsetof(struct sk_buff, cb);
6937                 off += offsetof(struct qdisc_skb_cb, flow_keys);
6938                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6939                                       si->src_reg, off);
6940                 break;
6941
6942         case offsetof(struct __sk_buff, tstamp):
6943                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
6944
6945                 if (type == BPF_WRITE)
6946                         *insn++ = BPF_STX_MEM(BPF_DW,
6947                                               si->dst_reg, si->src_reg,
6948                                               bpf_target_off(struct sk_buff,
6949                                                              tstamp, 8,
6950                                                              target_size));
6951                 else
6952                         *insn++ = BPF_LDX_MEM(BPF_DW,
6953                                               si->dst_reg, si->src_reg,
6954                                               bpf_target_off(struct sk_buff,
6955                                                              tstamp, 8,
6956                                                              target_size));
6957                 break;
6958
6959         case offsetof(struct __sk_buff, gso_segs):
6960                 /* si->dst_reg = skb_shinfo(SKB); */
6961 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6962                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
6963                                       si->dst_reg, si->src_reg,
6964                                       offsetof(struct sk_buff, head));
6965                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
6966                                       BPF_REG_AX, si->src_reg,
6967                                       offsetof(struct sk_buff, end));
6968                 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
6969 #else
6970                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
6971                                       si->dst_reg, si->src_reg,
6972                                       offsetof(struct sk_buff, end));
6973 #endif
6974                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
6975                                       si->dst_reg, si->dst_reg,
6976                                       bpf_target_off(struct skb_shared_info,
6977                                                      gso_segs, 2,
6978                                                      target_size));
6979                 break;
6980         case offsetof(struct __sk_buff, wire_len):
6981                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
6982
6983                 off = si->off;
6984                 off -= offsetof(struct __sk_buff, wire_len);
6985                 off += offsetof(struct sk_buff, cb);
6986                 off += offsetof(struct qdisc_skb_cb, pkt_len);
6987                 *target_size = 4;
6988                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
6989                 break;
6990
6991         case offsetof(struct __sk_buff, sk):
6992                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6993                                       si->dst_reg, si->src_reg,
6994                                       offsetof(struct sk_buff, sk));
6995                 break;
6996         }
6997
6998         return insn - insn_buf;
6999 }
7000
7001 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7002                                 const struct bpf_insn *si,
7003                                 struct bpf_insn *insn_buf,
7004                                 struct bpf_prog *prog, u32 *target_size)
7005 {
7006         struct bpf_insn *insn = insn_buf;
7007         int off;
7008
7009         switch (si->off) {
7010         case offsetof(struct bpf_sock, bound_dev_if):
7011                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
7012
7013                 if (type == BPF_WRITE)
7014                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7015                                         offsetof(struct sock, sk_bound_dev_if));
7016                 else
7017                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7018                                       offsetof(struct sock, sk_bound_dev_if));
7019                 break;
7020
7021         case offsetof(struct bpf_sock, mark):
7022                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
7023
7024                 if (type == BPF_WRITE)
7025                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7026                                         offsetof(struct sock, sk_mark));
7027                 else
7028                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7029                                       offsetof(struct sock, sk_mark));
7030                 break;
7031
7032         case offsetof(struct bpf_sock, priority):
7033                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
7034
7035                 if (type == BPF_WRITE)
7036                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7037                                         offsetof(struct sock, sk_priority));
7038                 else
7039                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7040                                       offsetof(struct sock, sk_priority));
7041                 break;
7042
7043         case offsetof(struct bpf_sock, family):
7044                 *insn++ = BPF_LDX_MEM(
7045                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7046                         si->dst_reg, si->src_reg,
7047                         bpf_target_off(struct sock_common,
7048                                        skc_family,
7049                                        FIELD_SIZEOF(struct sock_common,
7050                                                     skc_family),
7051                                        target_size));
7052                 break;
7053
7054         case offsetof(struct bpf_sock, type):
7055                 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7056                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7057                                       offsetof(struct sock, __sk_flags_offset));
7058                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7059                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7060                 *target_size = 2;
7061                 break;
7062
7063         case offsetof(struct bpf_sock, protocol):
7064                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7065                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7066                                       offsetof(struct sock, __sk_flags_offset));
7067                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7068                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7069                 *target_size = 1;
7070                 break;
7071
7072         case offsetof(struct bpf_sock, src_ip4):
7073                 *insn++ = BPF_LDX_MEM(
7074                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7075                         bpf_target_off(struct sock_common, skc_rcv_saddr,
7076                                        FIELD_SIZEOF(struct sock_common,
7077                                                     skc_rcv_saddr),
7078                                        target_size));
7079                 break;
7080
7081         case offsetof(struct bpf_sock, dst_ip4):
7082                 *insn++ = BPF_LDX_MEM(
7083                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7084                         bpf_target_off(struct sock_common, skc_daddr,
7085                                        FIELD_SIZEOF(struct sock_common,
7086                                                     skc_daddr),
7087                                        target_size));
7088                 break;
7089
7090         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7091 #if IS_ENABLED(CONFIG_IPV6)
7092                 off = si->off;
7093                 off -= offsetof(struct bpf_sock, src_ip6[0]);
7094                 *insn++ = BPF_LDX_MEM(
7095                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7096                         bpf_target_off(
7097                                 struct sock_common,
7098                                 skc_v6_rcv_saddr.s6_addr32[0],
7099                                 FIELD_SIZEOF(struct sock_common,
7100                                              skc_v6_rcv_saddr.s6_addr32[0]),
7101                                 target_size) + off);
7102 #else
7103                 (void)off;
7104                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7105 #endif
7106                 break;
7107
7108         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7109 #if IS_ENABLED(CONFIG_IPV6)
7110                 off = si->off;
7111                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7112                 *insn++ = BPF_LDX_MEM(
7113                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7114                         bpf_target_off(struct sock_common,
7115                                        skc_v6_daddr.s6_addr32[0],
7116                                        FIELD_SIZEOF(struct sock_common,
7117                                                     skc_v6_daddr.s6_addr32[0]),
7118                                        target_size) + off);
7119 #else
7120                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7121                 *target_size = 4;
7122 #endif
7123                 break;
7124
7125         case offsetof(struct bpf_sock, src_port):
7126                 *insn++ = BPF_LDX_MEM(
7127                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7128                         si->dst_reg, si->src_reg,
7129                         bpf_target_off(struct sock_common, skc_num,
7130                                        FIELD_SIZEOF(struct sock_common,
7131                                                     skc_num),
7132                                        target_size));
7133                 break;
7134
7135         case offsetof(struct bpf_sock, dst_port):
7136                 *insn++ = BPF_LDX_MEM(
7137                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7138                         si->dst_reg, si->src_reg,
7139                         bpf_target_off(struct sock_common, skc_dport,
7140                                        FIELD_SIZEOF(struct sock_common,
7141                                                     skc_dport),
7142                                        target_size));
7143                 break;
7144
7145         case offsetof(struct bpf_sock, state):
7146                 *insn++ = BPF_LDX_MEM(
7147                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7148                         si->dst_reg, si->src_reg,
7149                         bpf_target_off(struct sock_common, skc_state,
7150                                        FIELD_SIZEOF(struct sock_common,
7151                                                     skc_state),
7152                                        target_size));
7153                 break;
7154         }
7155
7156         return insn - insn_buf;
7157 }
7158
7159 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7160                                          const struct bpf_insn *si,
7161                                          struct bpf_insn *insn_buf,
7162                                          struct bpf_prog *prog, u32 *target_size)
7163 {
7164         struct bpf_insn *insn = insn_buf;
7165
7166         switch (si->off) {
7167         case offsetof(struct __sk_buff, ifindex):
7168                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7169                                       si->dst_reg, si->src_reg,
7170                                       offsetof(struct sk_buff, dev));
7171                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7172                                       bpf_target_off(struct net_device, ifindex, 4,
7173                                                      target_size));
7174                 break;
7175         default:
7176                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7177                                               target_size);
7178         }
7179
7180         return insn - insn_buf;
7181 }
7182
7183 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7184                                   const struct bpf_insn *si,
7185                                   struct bpf_insn *insn_buf,
7186                                   struct bpf_prog *prog, u32 *target_size)
7187 {
7188         struct bpf_insn *insn = insn_buf;
7189
7190         switch (si->off) {
7191         case offsetof(struct xdp_md, data):
7192                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7193                                       si->dst_reg, si->src_reg,
7194                                       offsetof(struct xdp_buff, data));
7195                 break;
7196         case offsetof(struct xdp_md, data_meta):
7197                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7198                                       si->dst_reg, si->src_reg,
7199                                       offsetof(struct xdp_buff, data_meta));
7200                 break;
7201         case offsetof(struct xdp_md, data_end):
7202                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7203                                       si->dst_reg, si->src_reg,
7204                                       offsetof(struct xdp_buff, data_end));
7205                 break;
7206         case offsetof(struct xdp_md, ingress_ifindex):
7207                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7208                                       si->dst_reg, si->src_reg,
7209                                       offsetof(struct xdp_buff, rxq));
7210                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7211                                       si->dst_reg, si->dst_reg,
7212                                       offsetof(struct xdp_rxq_info, dev));
7213                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7214                                       offsetof(struct net_device, ifindex));
7215                 break;
7216         case offsetof(struct xdp_md, rx_queue_index):
7217                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7218                                       si->dst_reg, si->src_reg,
7219                                       offsetof(struct xdp_buff, rxq));
7220                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7221                                       offsetof(struct xdp_rxq_info,
7222                                                queue_index));
7223                 break;
7224         }
7225
7226         return insn - insn_buf;
7227 }
7228
7229 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7230  * context Structure, F is Field in context structure that contains a pointer
7231  * to Nested Structure of type NS that has the field NF.
7232  *
7233  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7234  * sure that SIZE is not greater than actual size of S.F.NF.
7235  *
7236  * If offset OFF is provided, the load happens from that offset relative to
7237  * offset of NF.
7238  */
7239 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
7240         do {                                                                   \
7241                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
7242                                       si->src_reg, offsetof(S, F));            \
7243                 *insn++ = BPF_LDX_MEM(                                         \
7244                         SIZE, si->dst_reg, si->dst_reg,                        \
7245                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
7246                                        target_size)                            \
7247                                 + OFF);                                        \
7248         } while (0)
7249
7250 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
7251         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
7252                                              BPF_FIELD_SIZEOF(NS, NF), 0)
7253
7254 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7255  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7256  *
7257  * It doesn't support SIZE argument though since narrow stores are not
7258  * supported for now.
7259  *
7260  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7261  * "register" since two registers available in convert_ctx_access are not
7262  * enough: we can't override neither SRC, since it contains value to store, nor
7263  * DST since it contains pointer to context that may be used by later
7264  * instructions. But we need a temporary place to save pointer to nested
7265  * structure whose field we want to store to.
7266  */
7267 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF)                \
7268         do {                                                                   \
7269                 int tmp_reg = BPF_REG_9;                                       \
7270                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
7271                         --tmp_reg;                                             \
7272                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
7273                         --tmp_reg;                                             \
7274                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
7275                                       offsetof(S, TF));                        \
7276                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
7277                                       si->dst_reg, offsetof(S, F));            \
7278                 *insn++ = BPF_STX_MEM(                                         \
7279                         BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg,        \
7280                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
7281                                        target_size)                            \
7282                                 + OFF);                                        \
7283                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
7284                                       offsetof(S, TF));                        \
7285         } while (0)
7286
7287 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7288                                                       TF)                      \
7289         do {                                                                   \
7290                 if (type == BPF_WRITE) {                                       \
7291                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF,    \
7292                                                          TF);                  \
7293                 } else {                                                       \
7294                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
7295                                 S, NS, F, NF, SIZE, OFF);  \
7296                 }                                                              \
7297         } while (0)
7298
7299 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
7300         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
7301                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7302
7303 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7304                                         const struct bpf_insn *si,
7305                                         struct bpf_insn *insn_buf,
7306                                         struct bpf_prog *prog, u32 *target_size)
7307 {
7308         struct bpf_insn *insn = insn_buf;
7309         int off;
7310
7311         switch (si->off) {
7312         case offsetof(struct bpf_sock_addr, user_family):
7313                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7314                                             struct sockaddr, uaddr, sa_family);
7315                 break;
7316
7317         case offsetof(struct bpf_sock_addr, user_ip4):
7318                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7319                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7320                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7321                 break;
7322
7323         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7324                 off = si->off;
7325                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7326                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7327                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7328                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7329                         tmp_reg);
7330                 break;
7331
7332         case offsetof(struct bpf_sock_addr, user_port):
7333                 /* To get port we need to know sa_family first and then treat
7334                  * sockaddr as either sockaddr_in or sockaddr_in6.
7335                  * Though we can simplify since port field has same offset and
7336                  * size in both structures.
7337                  * Here we check this invariant and use just one of the
7338                  * structures if it's true.
7339                  */
7340                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7341                              offsetof(struct sockaddr_in6, sin6_port));
7342                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7343                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7344                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7345                                                      struct sockaddr_in6, uaddr,
7346                                                      sin6_port, tmp_reg);
7347                 break;
7348
7349         case offsetof(struct bpf_sock_addr, family):
7350                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7351                                             struct sock, sk, sk_family);
7352                 break;
7353
7354         case offsetof(struct bpf_sock_addr, type):
7355                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7356                         struct bpf_sock_addr_kern, struct sock, sk,
7357                         __sk_flags_offset, BPF_W, 0);
7358                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7359                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7360                 break;
7361
7362         case offsetof(struct bpf_sock_addr, protocol):
7363                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7364                         struct bpf_sock_addr_kern, struct sock, sk,
7365                         __sk_flags_offset, BPF_W, 0);
7366                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7367                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7368                                         SK_FL_PROTO_SHIFT);
7369                 break;
7370
7371         case offsetof(struct bpf_sock_addr, msg_src_ip4):
7372                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7373                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7374                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7375                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7376                 break;
7377
7378         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7379                                 msg_src_ip6[3]):
7380                 off = si->off;
7381                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7382                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7383                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7384                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7385                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7386                 break;
7387         }
7388
7389         return insn - insn_buf;
7390 }
7391
7392 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7393                                        const struct bpf_insn *si,
7394                                        struct bpf_insn *insn_buf,
7395                                        struct bpf_prog *prog,
7396                                        u32 *target_size)
7397 {
7398         struct bpf_insn *insn = insn_buf;
7399         int off;
7400
7401 /* Helper macro for adding read access to tcp_sock or sock fields. */
7402 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7403         do {                                                                  \
7404                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
7405                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
7406                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7407                                                 struct bpf_sock_ops_kern,     \
7408                                                 is_fullsock),                 \
7409                                       si->dst_reg, si->src_reg,               \
7410                                       offsetof(struct bpf_sock_ops_kern,      \
7411                                                is_fullsock));                 \
7412                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
7413                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7414                                                 struct bpf_sock_ops_kern, sk),\
7415                                       si->dst_reg, si->src_reg,               \
7416                                       offsetof(struct bpf_sock_ops_kern, sk));\
7417                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
7418                                                        OBJ_FIELD),            \
7419                                       si->dst_reg, si->dst_reg,               \
7420                                       offsetof(OBJ, OBJ_FIELD));              \
7421         } while (0)
7422
7423 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7424                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7425
7426 /* Helper macro for adding write access to tcp_sock or sock fields.
7427  * The macro is called with two registers, dst_reg which contains a pointer
7428  * to ctx (context) and src_reg which contains the value that should be
7429  * stored. However, we need an additional register since we cannot overwrite
7430  * dst_reg because it may be used later in the program.
7431  * Instead we "borrow" one of the other register. We first save its value
7432  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7433  * it at the end of the macro.
7434  */
7435 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7436         do {                                                                  \
7437                 int reg = BPF_REG_9;                                          \
7438                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
7439                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
7440                 if (si->dst_reg == reg || si->src_reg == reg)                 \
7441                         reg--;                                                \
7442                 if (si->dst_reg == reg || si->src_reg == reg)                 \
7443                         reg--;                                                \
7444                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
7445                                       offsetof(struct bpf_sock_ops_kern,      \
7446                                                temp));                        \
7447                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7448                                                 struct bpf_sock_ops_kern,     \
7449                                                 is_fullsock),                 \
7450                                       reg, si->dst_reg,                       \
7451                                       offsetof(struct bpf_sock_ops_kern,      \
7452                                                is_fullsock));                 \
7453                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
7454                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7455                                                 struct bpf_sock_ops_kern, sk),\
7456                                       reg, si->dst_reg,                       \
7457                                       offsetof(struct bpf_sock_ops_kern, sk));\
7458                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
7459                                       reg, si->src_reg,                       \
7460                                       offsetof(OBJ, OBJ_FIELD));              \
7461                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
7462                                       offsetof(struct bpf_sock_ops_kern,      \
7463                                                temp));                        \
7464         } while (0)
7465
7466 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
7467         do {                                                                  \
7468                 if (TYPE == BPF_WRITE)                                        \
7469                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
7470                 else                                                          \
7471                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
7472         } while (0)
7473
7474         CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_sock_ops,
7475                                        SOCK_OPS_GET_TCP_SOCK_FIELD);
7476
7477         if (insn > insn_buf)
7478                 return insn - insn_buf;
7479
7480         switch (si->off) {
7481         case offsetof(struct bpf_sock_ops, op) ...
7482              offsetof(struct bpf_sock_ops, replylong[3]):
7483                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7484                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7485                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7486                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7487                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7488                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7489                 off = si->off;
7490                 off -= offsetof(struct bpf_sock_ops, op);
7491                 off += offsetof(struct bpf_sock_ops_kern, op);
7492                 if (type == BPF_WRITE)
7493                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7494                                               off);
7495                 else
7496                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7497                                               off);
7498                 break;
7499
7500         case offsetof(struct bpf_sock_ops, family):
7501                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7502
7503                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7504                                               struct bpf_sock_ops_kern, sk),
7505                                       si->dst_reg, si->src_reg,
7506                                       offsetof(struct bpf_sock_ops_kern, sk));
7507                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7508                                       offsetof(struct sock_common, skc_family));
7509                 break;
7510
7511         case offsetof(struct bpf_sock_ops, remote_ip4):
7512                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7513
7514                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7515                                                 struct bpf_sock_ops_kern, sk),
7516                                       si->dst_reg, si->src_reg,
7517                                       offsetof(struct bpf_sock_ops_kern, sk));
7518                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7519                                       offsetof(struct sock_common, skc_daddr));
7520                 break;
7521
7522         case offsetof(struct bpf_sock_ops, local_ip4):
7523                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7524                                           skc_rcv_saddr) != 4);
7525
7526                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7527                                               struct bpf_sock_ops_kern, sk),
7528                                       si->dst_reg, si->src_reg,
7529                                       offsetof(struct bpf_sock_ops_kern, sk));
7530                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7531                                       offsetof(struct sock_common,
7532                                                skc_rcv_saddr));
7533                 break;
7534
7535         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
7536              offsetof(struct bpf_sock_ops, remote_ip6[3]):
7537 #if IS_ENABLED(CONFIG_IPV6)
7538                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7539                                           skc_v6_daddr.s6_addr32[0]) != 4);
7540
7541                 off = si->off;
7542                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
7543                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7544                                                 struct bpf_sock_ops_kern, sk),
7545                                       si->dst_reg, si->src_reg,
7546                                       offsetof(struct bpf_sock_ops_kern, sk));
7547                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7548                                       offsetof(struct sock_common,
7549                                                skc_v6_daddr.s6_addr32[0]) +
7550                                       off);
7551 #else
7552                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7553 #endif
7554                 break;
7555
7556         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
7557              offsetof(struct bpf_sock_ops, local_ip6[3]):
7558 #if IS_ENABLED(CONFIG_IPV6)
7559                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7560                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7561
7562                 off = si->off;
7563                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
7564                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7565                                                 struct bpf_sock_ops_kern, sk),
7566                                       si->dst_reg, si->src_reg,
7567                                       offsetof(struct bpf_sock_ops_kern, sk));
7568                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7569                                       offsetof(struct sock_common,
7570                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7571                                       off);
7572 #else
7573                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7574 #endif
7575                 break;
7576
7577         case offsetof(struct bpf_sock_ops, remote_port):
7578                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7579
7580                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7581                                                 struct bpf_sock_ops_kern, sk),
7582                                       si->dst_reg, si->src_reg,
7583                                       offsetof(struct bpf_sock_ops_kern, sk));
7584                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7585                                       offsetof(struct sock_common, skc_dport));
7586 #ifndef __BIG_ENDIAN_BITFIELD
7587                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7588 #endif
7589                 break;
7590
7591         case offsetof(struct bpf_sock_ops, local_port):
7592                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7593
7594                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7595                                                 struct bpf_sock_ops_kern, sk),
7596                                       si->dst_reg, si->src_reg,
7597                                       offsetof(struct bpf_sock_ops_kern, sk));
7598                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7599                                       offsetof(struct sock_common, skc_num));
7600                 break;
7601
7602         case offsetof(struct bpf_sock_ops, is_fullsock):
7603                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7604                                                 struct bpf_sock_ops_kern,
7605                                                 is_fullsock),
7606                                       si->dst_reg, si->src_reg,
7607                                       offsetof(struct bpf_sock_ops_kern,
7608                                                is_fullsock));
7609                 break;
7610
7611         case offsetof(struct bpf_sock_ops, state):
7612                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
7613
7614                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7615                                                 struct bpf_sock_ops_kern, sk),
7616                                       si->dst_reg, si->src_reg,
7617                                       offsetof(struct bpf_sock_ops_kern, sk));
7618                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
7619                                       offsetof(struct sock_common, skc_state));
7620                 break;
7621
7622         case offsetof(struct bpf_sock_ops, rtt_min):
7623                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
7624                              sizeof(struct minmax));
7625                 BUILD_BUG_ON(sizeof(struct minmax) <
7626                              sizeof(struct minmax_sample));
7627
7628                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7629                                                 struct bpf_sock_ops_kern, sk),
7630                                       si->dst_reg, si->src_reg,
7631                                       offsetof(struct bpf_sock_ops_kern, sk));
7632                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7633                                       offsetof(struct tcp_sock, rtt_min) +
7634                                       FIELD_SIZEOF(struct minmax_sample, t));
7635                 break;
7636
7637         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
7638                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
7639                                    struct tcp_sock);
7640                 break;
7641
7642         case offsetof(struct bpf_sock_ops, sk_txhash):
7643                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
7644                                           struct sock, type);
7645                 break;
7646         }
7647         return insn - insn_buf;
7648 }
7649
7650 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
7651                                      const struct bpf_insn *si,
7652                                      struct bpf_insn *insn_buf,
7653                                      struct bpf_prog *prog, u32 *target_size)
7654 {
7655         struct bpf_insn *insn = insn_buf;
7656         int off;
7657
7658         switch (si->off) {
7659         case offsetof(struct __sk_buff, data_end):
7660                 off  = si->off;
7661                 off -= offsetof(struct __sk_buff, data_end);
7662                 off += offsetof(struct sk_buff, cb);
7663                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
7664                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7665                                       si->src_reg, off);
7666                 break;
7667         default:
7668                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7669                                               target_size);
7670         }
7671
7672         return insn - insn_buf;
7673 }
7674
7675 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
7676                                      const struct bpf_insn *si,
7677                                      struct bpf_insn *insn_buf,
7678                                      struct bpf_prog *prog, u32 *target_size)
7679 {
7680         struct bpf_insn *insn = insn_buf;
7681 #if IS_ENABLED(CONFIG_IPV6)
7682         int off;
7683 #endif
7684
7685         /* convert ctx uses the fact sg element is first in struct */
7686         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
7687
7688         switch (si->off) {
7689         case offsetof(struct sk_msg_md, data):
7690                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
7691                                       si->dst_reg, si->src_reg,
7692                                       offsetof(struct sk_msg, data));
7693                 break;
7694         case offsetof(struct sk_msg_md, data_end):
7695                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
7696                                       si->dst_reg, si->src_reg,
7697                                       offsetof(struct sk_msg, data_end));
7698                 break;
7699         case offsetof(struct sk_msg_md, family):
7700                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7701
7702                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7703                                               struct sk_msg, sk),
7704                                       si->dst_reg, si->src_reg,
7705                                       offsetof(struct sk_msg, sk));
7706                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7707                                       offsetof(struct sock_common, skc_family));
7708                 break;
7709
7710         case offsetof(struct sk_msg_md, remote_ip4):
7711                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7712
7713                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7714                                                 struct sk_msg, sk),
7715                                       si->dst_reg, si->src_reg,
7716                                       offsetof(struct sk_msg, sk));
7717                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7718                                       offsetof(struct sock_common, skc_daddr));
7719                 break;
7720
7721         case offsetof(struct sk_msg_md, local_ip4):
7722                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7723                                           skc_rcv_saddr) != 4);
7724
7725                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7726                                               struct sk_msg, sk),
7727                                       si->dst_reg, si->src_reg,
7728                                       offsetof(struct sk_msg, sk));
7729                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7730                                       offsetof(struct sock_common,
7731                                                skc_rcv_saddr));
7732                 break;
7733
7734         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
7735              offsetof(struct sk_msg_md, remote_ip6[3]):
7736 #if IS_ENABLED(CONFIG_IPV6)
7737                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7738                                           skc_v6_daddr.s6_addr32[0]) != 4);
7739
7740                 off = si->off;
7741                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
7742                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7743                                                 struct sk_msg, sk),
7744                                       si->dst_reg, si->src_reg,
7745                                       offsetof(struct sk_msg, sk));
7746                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7747                                       offsetof(struct sock_common,
7748                                                skc_v6_daddr.s6_addr32[0]) +
7749                                       off);
7750 #else
7751                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7752 #endif
7753                 break;
7754
7755         case offsetof(struct sk_msg_md, local_ip6[0]) ...
7756              offsetof(struct sk_msg_md, local_ip6[3]):
7757 #if IS_ENABLED(CONFIG_IPV6)
7758                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7759                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7760
7761                 off = si->off;
7762                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
7763                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7764                                                 struct sk_msg, sk),
7765                                       si->dst_reg, si->src_reg,
7766                                       offsetof(struct sk_msg, sk));
7767                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7768                                       offsetof(struct sock_common,
7769                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7770                                       off);
7771 #else
7772                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7773 #endif
7774                 break;
7775
7776         case offsetof(struct sk_msg_md, remote_port):
7777                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7778
7779                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7780                                                 struct sk_msg, sk),
7781                                       si->dst_reg, si->src_reg,
7782                                       offsetof(struct sk_msg, sk));
7783                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7784                                       offsetof(struct sock_common, skc_dport));
7785 #ifndef __BIG_ENDIAN_BITFIELD
7786                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7787 #endif
7788                 break;
7789
7790         case offsetof(struct sk_msg_md, local_port):
7791                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7792
7793                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7794                                                 struct sk_msg, sk),
7795                                       si->dst_reg, si->src_reg,
7796                                       offsetof(struct sk_msg, sk));
7797                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7798                                       offsetof(struct sock_common, skc_num));
7799                 break;
7800
7801         case offsetof(struct sk_msg_md, size):
7802                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
7803                                       si->dst_reg, si->src_reg,
7804                                       offsetof(struct sk_msg_sg, size));
7805                 break;
7806         }
7807
7808         return insn - insn_buf;
7809 }
7810
7811 const struct bpf_verifier_ops sk_filter_verifier_ops = {
7812         .get_func_proto         = sk_filter_func_proto,
7813         .is_valid_access        = sk_filter_is_valid_access,
7814         .convert_ctx_access     = bpf_convert_ctx_access,
7815         .gen_ld_abs             = bpf_gen_ld_abs,
7816 };
7817
7818 const struct bpf_prog_ops sk_filter_prog_ops = {
7819         .test_run               = bpf_prog_test_run_skb,
7820 };
7821
7822 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
7823         .get_func_proto         = tc_cls_act_func_proto,
7824         .is_valid_access        = tc_cls_act_is_valid_access,
7825         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
7826         .gen_prologue           = tc_cls_act_prologue,
7827         .gen_ld_abs             = bpf_gen_ld_abs,
7828 };
7829
7830 const struct bpf_prog_ops tc_cls_act_prog_ops = {
7831         .test_run               = bpf_prog_test_run_skb,
7832 };
7833
7834 const struct bpf_verifier_ops xdp_verifier_ops = {
7835         .get_func_proto         = xdp_func_proto,
7836         .is_valid_access        = xdp_is_valid_access,
7837         .convert_ctx_access     = xdp_convert_ctx_access,
7838         .gen_prologue           = bpf_noop_prologue,
7839 };
7840
7841 const struct bpf_prog_ops xdp_prog_ops = {
7842         .test_run               = bpf_prog_test_run_xdp,
7843 };
7844
7845 const struct bpf_verifier_ops cg_skb_verifier_ops = {
7846         .get_func_proto         = cg_skb_func_proto,
7847         .is_valid_access        = cg_skb_is_valid_access,
7848         .convert_ctx_access     = bpf_convert_ctx_access,
7849 };
7850
7851 const struct bpf_prog_ops cg_skb_prog_ops = {
7852         .test_run               = bpf_prog_test_run_skb,
7853 };
7854
7855 const struct bpf_verifier_ops lwt_in_verifier_ops = {
7856         .get_func_proto         = lwt_in_func_proto,
7857         .is_valid_access        = lwt_is_valid_access,
7858         .convert_ctx_access     = bpf_convert_ctx_access,
7859 };
7860
7861 const struct bpf_prog_ops lwt_in_prog_ops = {
7862         .test_run               = bpf_prog_test_run_skb,
7863 };
7864
7865 const struct bpf_verifier_ops lwt_out_verifier_ops = {
7866         .get_func_proto         = lwt_out_func_proto,
7867         .is_valid_access        = lwt_is_valid_access,
7868         .convert_ctx_access     = bpf_convert_ctx_access,
7869 };
7870
7871 const struct bpf_prog_ops lwt_out_prog_ops = {
7872         .test_run               = bpf_prog_test_run_skb,
7873 };
7874
7875 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
7876         .get_func_proto         = lwt_xmit_func_proto,
7877         .is_valid_access        = lwt_is_valid_access,
7878         .convert_ctx_access     = bpf_convert_ctx_access,
7879         .gen_prologue           = tc_cls_act_prologue,
7880 };
7881
7882 const struct bpf_prog_ops lwt_xmit_prog_ops = {
7883         .test_run               = bpf_prog_test_run_skb,
7884 };
7885
7886 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
7887         .get_func_proto         = lwt_seg6local_func_proto,
7888         .is_valid_access        = lwt_is_valid_access,
7889         .convert_ctx_access     = bpf_convert_ctx_access,
7890 };
7891
7892 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
7893         .test_run               = bpf_prog_test_run_skb,
7894 };
7895
7896 const struct bpf_verifier_ops cg_sock_verifier_ops = {
7897         .get_func_proto         = sock_filter_func_proto,
7898         .is_valid_access        = sock_filter_is_valid_access,
7899         .convert_ctx_access     = bpf_sock_convert_ctx_access,
7900 };
7901
7902 const struct bpf_prog_ops cg_sock_prog_ops = {
7903 };
7904
7905 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
7906         .get_func_proto         = sock_addr_func_proto,
7907         .is_valid_access        = sock_addr_is_valid_access,
7908         .convert_ctx_access     = sock_addr_convert_ctx_access,
7909 };
7910
7911 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
7912 };
7913
7914 const struct bpf_verifier_ops sock_ops_verifier_ops = {
7915         .get_func_proto         = sock_ops_func_proto,
7916         .is_valid_access        = sock_ops_is_valid_access,
7917         .convert_ctx_access     = sock_ops_convert_ctx_access,
7918 };
7919
7920 const struct bpf_prog_ops sock_ops_prog_ops = {
7921 };
7922
7923 const struct bpf_verifier_ops sk_skb_verifier_ops = {
7924         .get_func_proto         = sk_skb_func_proto,
7925         .is_valid_access        = sk_skb_is_valid_access,
7926         .convert_ctx_access     = sk_skb_convert_ctx_access,
7927         .gen_prologue           = sk_skb_prologue,
7928 };
7929
7930 const struct bpf_prog_ops sk_skb_prog_ops = {
7931 };
7932
7933 const struct bpf_verifier_ops sk_msg_verifier_ops = {
7934         .get_func_proto         = sk_msg_func_proto,
7935         .is_valid_access        = sk_msg_is_valid_access,
7936         .convert_ctx_access     = sk_msg_convert_ctx_access,
7937         .gen_prologue           = bpf_noop_prologue,
7938 };
7939
7940 const struct bpf_prog_ops sk_msg_prog_ops = {
7941 };
7942
7943 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
7944         .get_func_proto         = flow_dissector_func_proto,
7945         .is_valid_access        = flow_dissector_is_valid_access,
7946         .convert_ctx_access     = bpf_convert_ctx_access,
7947 };
7948
7949 const struct bpf_prog_ops flow_dissector_prog_ops = {
7950         .test_run               = bpf_prog_test_run_flow_dissector,
7951 };
7952
7953 int sk_detach_filter(struct sock *sk)
7954 {
7955         int ret = -ENOENT;
7956         struct sk_filter *filter;
7957
7958         if (sock_flag(sk, SOCK_FILTER_LOCKED))
7959                 return -EPERM;
7960
7961         filter = rcu_dereference_protected(sk->sk_filter,
7962                                            lockdep_sock_is_held(sk));
7963         if (filter) {
7964                 RCU_INIT_POINTER(sk->sk_filter, NULL);
7965                 sk_filter_uncharge(sk, filter);
7966                 ret = 0;
7967         }
7968
7969         return ret;
7970 }
7971 EXPORT_SYMBOL_GPL(sk_detach_filter);
7972
7973 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7974                   unsigned int len)
7975 {
7976         struct sock_fprog_kern *fprog;
7977         struct sk_filter *filter;
7978         int ret = 0;
7979
7980         lock_sock(sk);
7981         filter = rcu_dereference_protected(sk->sk_filter,
7982                                            lockdep_sock_is_held(sk));
7983         if (!filter)
7984                 goto out;
7985
7986         /* We're copying the filter that has been originally attached,
7987          * so no conversion/decode needed anymore. eBPF programs that
7988          * have no original program cannot be dumped through this.
7989          */
7990         ret = -EACCES;
7991         fprog = filter->prog->orig_prog;
7992         if (!fprog)
7993                 goto out;
7994
7995         ret = fprog->len;
7996         if (!len)
7997                 /* User space only enquires number of filter blocks. */
7998                 goto out;
7999
8000         ret = -EINVAL;
8001         if (len < fprog->len)
8002                 goto out;
8003
8004         ret = -EFAULT;
8005         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8006                 goto out;
8007
8008         /* Instead of bytes, the API requests to return the number
8009          * of filter blocks.
8010          */
8011         ret = fprog->len;
8012 out:
8013         release_sock(sk);
8014         return ret;
8015 }
8016
8017 #ifdef CONFIG_INET
8018 struct sk_reuseport_kern {
8019         struct sk_buff *skb;
8020         struct sock *sk;
8021         struct sock *selected_sk;
8022         void *data_end;
8023         u32 hash;
8024         u32 reuseport_id;
8025         bool bind_inany;
8026 };
8027
8028 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8029                                     struct sock_reuseport *reuse,
8030                                     struct sock *sk, struct sk_buff *skb,
8031                                     u32 hash)
8032 {
8033         reuse_kern->skb = skb;
8034         reuse_kern->sk = sk;
8035         reuse_kern->selected_sk = NULL;
8036         reuse_kern->data_end = skb->data + skb_headlen(skb);
8037         reuse_kern->hash = hash;
8038         reuse_kern->reuseport_id = reuse->reuseport_id;
8039         reuse_kern->bind_inany = reuse->bind_inany;
8040 }
8041
8042 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8043                                   struct bpf_prog *prog, struct sk_buff *skb,
8044                                   u32 hash)
8045 {
8046         struct sk_reuseport_kern reuse_kern;
8047         enum sk_action action;
8048
8049         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8050         action = BPF_PROG_RUN(prog, &reuse_kern);
8051
8052         if (action == SK_PASS)
8053                 return reuse_kern.selected_sk;
8054         else
8055                 return ERR_PTR(-ECONNREFUSED);
8056 }
8057
8058 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8059            struct bpf_map *, map, void *, key, u32, flags)
8060 {
8061         struct sock_reuseport *reuse;
8062         struct sock *selected_sk;
8063
8064         selected_sk = map->ops->map_lookup_elem(map, key);
8065         if (!selected_sk)
8066                 return -ENOENT;
8067
8068         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8069         if (!reuse)
8070                 /* selected_sk is unhashed (e.g. by close()) after the
8071                  * above map_lookup_elem().  Treat selected_sk has already
8072                  * been removed from the map.
8073                  */
8074                 return -ENOENT;
8075
8076         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8077                 struct sock *sk;
8078
8079                 if (unlikely(!reuse_kern->reuseport_id))
8080                         /* There is a small race between adding the
8081                          * sk to the map and setting the
8082                          * reuse_kern->reuseport_id.
8083                          * Treat it as the sk has not been added to
8084                          * the bpf map yet.
8085                          */
8086                         return -ENOENT;
8087
8088                 sk = reuse_kern->sk;
8089                 if (sk->sk_protocol != selected_sk->sk_protocol)
8090                         return -EPROTOTYPE;
8091                 else if (sk->sk_family != selected_sk->sk_family)
8092                         return -EAFNOSUPPORT;
8093
8094                 /* Catch all. Likely bound to a different sockaddr. */
8095                 return -EBADFD;
8096         }
8097
8098         reuse_kern->selected_sk = selected_sk;
8099
8100         return 0;
8101 }
8102
8103 static const struct bpf_func_proto sk_select_reuseport_proto = {
8104         .func           = sk_select_reuseport,
8105         .gpl_only       = false,
8106         .ret_type       = RET_INTEGER,
8107         .arg1_type      = ARG_PTR_TO_CTX,
8108         .arg2_type      = ARG_CONST_MAP_PTR,
8109         .arg3_type      = ARG_PTR_TO_MAP_KEY,
8110         .arg4_type      = ARG_ANYTHING,
8111 };
8112
8113 BPF_CALL_4(sk_reuseport_load_bytes,
8114            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8115            void *, to, u32, len)
8116 {
8117         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8118 }
8119
8120 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8121         .func           = sk_reuseport_load_bytes,
8122         .gpl_only       = false,
8123         .ret_type       = RET_INTEGER,
8124         .arg1_type      = ARG_PTR_TO_CTX,
8125         .arg2_type      = ARG_ANYTHING,
8126         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
8127         .arg4_type      = ARG_CONST_SIZE,
8128 };
8129
8130 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8131            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8132            void *, to, u32, len, u32, start_header)
8133 {
8134         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8135                                                len, start_header);
8136 }
8137
8138 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8139         .func           = sk_reuseport_load_bytes_relative,
8140         .gpl_only       = false,
8141         .ret_type       = RET_INTEGER,
8142         .arg1_type      = ARG_PTR_TO_CTX,
8143         .arg2_type      = ARG_ANYTHING,
8144         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
8145         .arg4_type      = ARG_CONST_SIZE,
8146         .arg5_type      = ARG_ANYTHING,
8147 };
8148
8149 static const struct bpf_func_proto *
8150 sk_reuseport_func_proto(enum bpf_func_id func_id,
8151                         const struct bpf_prog *prog)
8152 {
8153         switch (func_id) {
8154         case BPF_FUNC_sk_select_reuseport:
8155                 return &sk_select_reuseport_proto;
8156         case BPF_FUNC_skb_load_bytes:
8157                 return &sk_reuseport_load_bytes_proto;
8158         case BPF_FUNC_skb_load_bytes_relative:
8159                 return &sk_reuseport_load_bytes_relative_proto;
8160         default:
8161                 return bpf_base_func_proto(func_id);
8162         }
8163 }
8164
8165 static bool
8166 sk_reuseport_is_valid_access(int off, int size,
8167                              enum bpf_access_type type,
8168                              const struct bpf_prog *prog,
8169                              struct bpf_insn_access_aux *info)
8170 {
8171         const u32 size_default = sizeof(__u32);
8172
8173         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8174             off % size || type != BPF_READ)
8175                 return false;
8176
8177         switch (off) {
8178         case offsetof(struct sk_reuseport_md, data):
8179                 info->reg_type = PTR_TO_PACKET;
8180                 return size == sizeof(__u64);
8181
8182         case offsetof(struct sk_reuseport_md, data_end):
8183                 info->reg_type = PTR_TO_PACKET_END;
8184                 return size == sizeof(__u64);
8185
8186         case offsetof(struct sk_reuseport_md, hash):
8187                 return size == size_default;
8188
8189         /* Fields that allow narrowing */
8190         case offsetof(struct sk_reuseport_md, eth_protocol):
8191                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8192                         return false;
8193                 /* fall through */
8194         case offsetof(struct sk_reuseport_md, ip_protocol):
8195         case offsetof(struct sk_reuseport_md, bind_inany):
8196         case offsetof(struct sk_reuseport_md, len):
8197                 bpf_ctx_record_field_size(info, size_default);
8198                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8199
8200         default:
8201                 return false;
8202         }
8203 }
8204
8205 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
8206         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8207                               si->dst_reg, si->src_reg,                 \
8208                               bpf_target_off(struct sk_reuseport_kern, F, \
8209                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8210                                              target_size));             \
8211         })
8212
8213 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
8214         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
8215                                     struct sk_buff,                     \
8216                                     skb,                                \
8217                                     SKB_FIELD)
8218
8219 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8220         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
8221                                              struct sock,               \
8222                                              sk,                        \
8223                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
8224
8225 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8226                                            const struct bpf_insn *si,
8227                                            struct bpf_insn *insn_buf,
8228                                            struct bpf_prog *prog,
8229                                            u32 *target_size)
8230 {
8231         struct bpf_insn *insn = insn_buf;
8232
8233         switch (si->off) {
8234         case offsetof(struct sk_reuseport_md, data):
8235                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8236                 break;
8237
8238         case offsetof(struct sk_reuseport_md, len):
8239                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8240                 break;
8241
8242         case offsetof(struct sk_reuseport_md, eth_protocol):
8243                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8244                 break;
8245
8246         case offsetof(struct sk_reuseport_md, ip_protocol):
8247                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8248                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8249                                                     BPF_W, 0);
8250                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8251                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8252                                         SK_FL_PROTO_SHIFT);
8253                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8254                  * aware.  No further narrowing or masking is needed.
8255                  */
8256                 *target_size = 1;
8257                 break;
8258
8259         case offsetof(struct sk_reuseport_md, data_end):
8260                 SK_REUSEPORT_LOAD_FIELD(data_end);
8261                 break;
8262
8263         case offsetof(struct sk_reuseport_md, hash):
8264                 SK_REUSEPORT_LOAD_FIELD(hash);
8265                 break;
8266
8267         case offsetof(struct sk_reuseport_md, bind_inany):
8268                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8269                 break;
8270         }
8271
8272         return insn - insn_buf;
8273 }
8274
8275 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8276         .get_func_proto         = sk_reuseport_func_proto,
8277         .is_valid_access        = sk_reuseport_is_valid_access,
8278         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
8279 };
8280
8281 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8282 };
8283 #endif /* CONFIG_INET */