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/ip.h>
  37 #include <net/protocol.h>
  38 #include <net/netlink.h>
  39 #include <linux/skbuff.h>
  40 #include <net/sock.h>
  41 #include <net/flow_dissector.h>
  42 #include <linux/errno.h>
  43 #include <linux/timer.h>
  44 #include <linux/uaccess.h>
  45 #include <asm/unaligned.h>
  46 #include <asm/cmpxchg.h>
  47 #include <linux/filter.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/seccomp.h>
  50 #include <linux/if_vlan.h>
  51 #include <linux/bpf.h>
  52 #include <net/sch_generic.h>
  53 #include <net/cls_cgroup.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/dst.h>
  56 #include <net/sock_reuseport.h>
  57 #include <net/busy_poll.h>
  58 #include <net/tcp.h>
  59 #include <linux/bpf_trace.h>
  60
  61 /**
  62  *      sk_filter_trim_cap - run a packet through a socket filter
  63  *      @sk: sock associated with &sk_buff
  64  *      @skb: buffer to filter
  65  *      @cap: limit on how short the eBPF program may trim the packet
  66  *
  67  * Run the eBPF program and then cut skb->data to correct size returned by
  68  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
  69  * than pkt_len we keep whole skb->data. This is the socket level
  70  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
  71  * be accepted or -EPERM if the packet should be tossed.
  72  *
  73  */
  74 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
  75 {
  76         int err;
  77         struct sk_filter *filter;
  78
  79         /*
  80          * If the skb was allocated from pfmemalloc reserves, only
  81          * allow SOCK_MEMALLOC sockets to use it as this socket is
  82          * helping free memory
  83          */
  84         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
  85                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
  86                 return -ENOMEM;
  87         }
  88         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
  89         if (err)
  90                 return err;
  91
  92         err = security_sock_rcv_skb(sk, skb);
  93         if (err)
  94                 return err;
  95
  96         rcu_read_lock();
  97         filter = rcu_dereference(sk->sk_filter);
  98         if (filter) {
  99                 struct sock *save_sk = skb->sk;
 100                 unsigned int pkt_len;
 101
 102                 skb->sk = sk;
 103                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 104                 skb->sk = save_sk;
 105                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 106         }
 107         rcu_read_unlock();
 108
 109         return err;
 110 }
 111 EXPORT_SYMBOL(sk_filter_trim_cap);
 112
 113 BPF_CALL_1(__skb_get_pay_offset, struct sk_buff *, skb)
 114 {
 115         return skb_get_poff(skb);
 116 }
 117
 118 BPF_CALL_3(__skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 119 {
 120         struct nlattr *nla;
 121
 122         if (skb_is_nonlinear(skb))
 123                 return 0;
 124
 125         if (skb->len < sizeof(struct nlattr))
 126                 return 0;
 127
 128         if (a > skb->len - sizeof(struct nlattr))
 129                 return 0;
 130
 131         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 132         if (nla)
 133                 return (void *) nla - (void *) skb->data;
 134
 135         return 0;
 136 }
 137
 138 BPF_CALL_3(__skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 139 {
 140         struct nlattr *nla;
 141
 142         if (skb_is_nonlinear(skb))
 143                 return 0;
 144
 145         if (skb->len < sizeof(struct nlattr))
 146                 return 0;
 147
 148         if (a > skb->len - sizeof(struct nlattr))
 149                 return 0;
 150
 151         nla = (struct nlattr *) &skb->data[a];
 152         if (nla->nla_len > skb->len - a)
 153                 return 0;
 154
 155         nla = nla_find_nested(nla, x);
 156         if (nla)
 157                 return (void *) nla - (void *) skb->data;
 158
 159         return 0;
 160 }
 161
 162 BPF_CALL_0(__get_raw_cpu_id)
 163 {
 164         return raw_smp_processor_id();
 165 }
 166
 167 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 168         .func           = __get_raw_cpu_id,
 169         .gpl_only       = false,
 170         .ret_type       = RET_INTEGER,
 171 };
 172
 173 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 174                               struct bpf_insn *insn_buf)
 175 {
 176         struct bpf_insn *insn = insn_buf;
 177
 178         switch (skb_field) {
 179         case SKF_AD_MARK:
 180                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
 181
 182                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 183                                       offsetof(struct sk_buff, mark));
 184                 break;
 185
 186         case SKF_AD_PKTTYPE:
 187                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 188                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 189 #ifdef __BIG_ENDIAN_BITFIELD
 190                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 191 #endif
 192                 break;
 193
 194         case SKF_AD_QUEUE:
 195                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
 196
 197                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 198                                       offsetof(struct sk_buff, queue_mapping));
 199                 break;
 200
 201         case SKF_AD_VLAN_TAG:
 202         case SKF_AD_VLAN_TAG_PRESENT:
 203                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
 204                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
 205
 206                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 207                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 208                                       offsetof(struct sk_buff, vlan_tci));
 209                 if (skb_field == SKF_AD_VLAN_TAG) {
 210                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
 211                                                 ~VLAN_TAG_PRESENT);
 212                 } else {
 213                         /* dst_reg >>= 12 */
 214                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
 215                         /* dst_reg &= 1 */
 216                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 217                 }
 218                 break;
 219         }
 220
 221         return insn - insn_buf;
 222 }
 223
 224 static bool convert_bpf_extensions(struct sock_filter *fp,
 225                                    struct bpf_insn **insnp)
 226 {
 227         struct bpf_insn *insn = *insnp;
 228         u32 cnt;
 229
 230         switch (fp->k) {
 231         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 232                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
 233
 234                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 235                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 236                                       offsetof(struct sk_buff, protocol));
 237                 /* A = ntohs(A) [emitting a nop or swap16] */
 238                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 239                 break;
 240
 241         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 242                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 243                 insn += cnt - 1;
 244                 break;
 245
 246         case SKF_AD_OFF + SKF_AD_IFINDEX:
 247         case SKF_AD_OFF + SKF_AD_HATYPE:
 248                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
 249                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
 250
 251                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 252                                       BPF_REG_TMP, BPF_REG_CTX,
 253                                       offsetof(struct sk_buff, dev));
 254                 /* if (tmp != 0) goto pc + 1 */
 255                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 256                 *insn++ = BPF_EXIT_INSN();
 257                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 258                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 259                                             offsetof(struct net_device, ifindex));
 260                 else
 261                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 262                                             offsetof(struct net_device, type));
 263                 break;
 264
 265         case SKF_AD_OFF + SKF_AD_MARK:
 266                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 267                 insn += cnt - 1;
 268                 break;
 269
 270         case SKF_AD_OFF + SKF_AD_RXHASH:
 271                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
 272
 273                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 274                                     offsetof(struct sk_buff, hash));
 275                 break;
 276
 277         case SKF_AD_OFF + SKF_AD_QUEUE:
 278                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 279                 insn += cnt - 1;
 280                 break;
 281
 282         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 283                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 284                                          BPF_REG_A, BPF_REG_CTX, insn);
 285                 insn += cnt - 1;
 286                 break;
 287
 288         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 289                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 290                                          BPF_REG_A, BPF_REG_CTX, insn);
 291                 insn += cnt - 1;
 292                 break;
 293
 294         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 295                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
 296
 297                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 298                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 299                                       offsetof(struct sk_buff, vlan_proto));
 300                 /* A = ntohs(A) [emitting a nop or swap16] */
 301                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 302                 break;
 303
 304         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 305         case SKF_AD_OFF + SKF_AD_NLATTR:
 306         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 307         case SKF_AD_OFF + SKF_AD_CPU:
 308         case SKF_AD_OFF + SKF_AD_RANDOM:
 309                 /* arg1 = CTX */
 310                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 311                 /* arg2 = A */
 312                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 313                 /* arg3 = X */
 314                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 315                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 316                 switch (fp->k) {
 317                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 318                         *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
 319                         break;
 320                 case SKF_AD_OFF + SKF_AD_NLATTR:
 321                         *insn = BPF_EMIT_CALL(__skb_get_nlattr);
 322                         break;
 323                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 324                         *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
 325                         break;
 326                 case SKF_AD_OFF + SKF_AD_CPU:
 327                         *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
 328                         break;
 329                 case SKF_AD_OFF + SKF_AD_RANDOM:
 330                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 331                         bpf_user_rnd_init_once();
 332                         break;
 333                 }
 334                 break;
 335
 336         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 337                 /* A ^= X */
 338                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 339                 break;
 340
 341         default:
 342                 /* This is just a dummy call to avoid letting the compiler
 343                  * evict __bpf_call_base() as an optimization. Placed here
 344                  * where no-one bothers.
 345                  */
 346                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 347                 return false;
 348         }
 349
 350         *insnp = insn;
 351         return true;
 352 }
 353
 354 /**
 355  *      bpf_convert_filter - convert filter program
 356  *      @prog: the user passed filter program
 357  *      @len: the length of the user passed filter program
 358  *      @new_prog: allocated 'struct bpf_prog' or NULL
 359  *      @new_len: pointer to store length of converted program
 360  *
 361  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 362  * style extended BPF (eBPF).
 363  * Conversion workflow:
 364  *
 365  * 1) First pass for calculating the new program length:
 366  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len)
 367  *
 368  * 2) 2nd pass to remap in two passes: 1st pass finds new
 369  *    jump offsets, 2nd pass remapping:
 370  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
 371  */
 372 static int bpf_convert_filter(struct sock_filter *prog, int len,
 373                               struct bpf_prog *new_prog, int *new_len)
 374 {
 375         int new_flen = 0, pass = 0, target, i, stack_off;
 376         struct bpf_insn *new_insn, *first_insn = NULL;
 377         struct sock_filter *fp;
 378         int *addrs = NULL;
 379         u8 bpf_src;
 380
 381         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 382         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 383
 384         if (len <= 0 || len > BPF_MAXINSNS)
 385                 return -EINVAL;
 386
 387         if (new_prog) {
 388                 first_insn = new_prog->insnsi;
 389                 addrs = kcalloc(len, sizeof(*addrs),
 390                                 GFP_KERNEL | __GFP_NOWARN);
 391                 if (!addrs)
 392                         return -ENOMEM;
 393         }
 394
 395 do_pass:
 396         new_insn = first_insn;
 397         fp = prog;
 398
 399         /* Classic BPF related prologue emission. */
 400         if (new_prog) {
 401                 /* Classic BPF expects A and X to be reset first. These need
 402                  * to be guaranteed to be the first two instructions.
 403                  */
 404                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 405                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 406
 407                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 408                  * In eBPF case it's done by the compiler, here we need to
 409                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 410                  */
 411                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 412         } else {
 413                 new_insn += 3;
 414         }
 415
 416         for (i = 0; i < len; fp++, i++) {
 417                 struct bpf_insn tmp_insns[6] = { };
 418                 struct bpf_insn *insn = tmp_insns;
 419
 420                 if (addrs)
 421                         addrs[i] = new_insn - first_insn;
 422
 423                 switch (fp->code) {
 424                 /* All arithmetic insns and skb loads map as-is. */
 425                 case BPF_ALU | BPF_ADD | BPF_X:
 426                 case BPF_ALU | BPF_ADD | BPF_K:
 427                 case BPF_ALU | BPF_SUB | BPF_X:
 428                 case BPF_ALU | BPF_SUB | BPF_K:
 429                 case BPF_ALU | BPF_AND | BPF_X:
 430                 case BPF_ALU | BPF_AND | BPF_K:
 431                 case BPF_ALU | BPF_OR | BPF_X:
 432                 case BPF_ALU | BPF_OR | BPF_K:
 433                 case BPF_ALU | BPF_LSH | BPF_X:
 434                 case BPF_ALU | BPF_LSH | BPF_K:
 435                 case BPF_ALU | BPF_RSH | BPF_X:
 436                 case BPF_ALU | BPF_RSH | BPF_K:
 437                 case BPF_ALU | BPF_XOR | BPF_X:
 438                 case BPF_ALU | BPF_XOR | BPF_K:
 439                 case BPF_ALU | BPF_MUL | BPF_X:
 440                 case BPF_ALU | BPF_MUL | BPF_K:
 441                 case BPF_ALU | BPF_DIV | BPF_X:
 442                 case BPF_ALU | BPF_DIV | BPF_K:
 443                 case BPF_ALU | BPF_MOD | BPF_X:
 444                 case BPF_ALU | BPF_MOD | BPF_K:
 445                 case BPF_ALU | BPF_NEG:
 446                 case BPF_LD | BPF_ABS | BPF_W:
 447                 case BPF_LD | BPF_ABS | BPF_H:
 448                 case BPF_LD | BPF_ABS | BPF_B:
 449                 case BPF_LD | BPF_IND | BPF_W:
 450                 case BPF_LD | BPF_IND | BPF_H:
 451                 case BPF_LD | BPF_IND | BPF_B:
 452                         /* Check for overloaded BPF extension and
 453                          * directly convert it if found, otherwise
 454                          * just move on with mapping.
 455                          */
 456                         if (BPF_CLASS(fp->code) == BPF_LD &&
 457                             BPF_MODE(fp->code) == BPF_ABS &&
 458                             convert_bpf_extensions(fp, &insn))
 459                                 break;
 460
 461                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 462                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 463                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 464                                 /* Error with exception code on div/mod by 0.
 465                                  * For cBPF programs, this was always return 0.
 466                                  */
 467                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 468                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 469                                 *insn++ = BPF_EXIT_INSN();
 470                         }
 471
 472                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 473                         break;
 474
 475                 /* Jump transformation cannot use BPF block macros
 476                  * everywhere as offset calculation and target updates
 477                  * require a bit more work than the rest, i.e. jump
 478                  * opcodes map as-is, but offsets need adjustment.
 479                  */
 480
 481 #define BPF_EMIT_JMP                                                    \
 482         do {                                                            \
 483                 if (target >= len || target < 0)                        \
 484                         goto err;                                       \
 485                 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0;   \
 486                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 487                 insn->off -= insn - tmp_insns;                          \
 488         } while (0)
 489
 490                 case BPF_JMP | BPF_JA:
 491                         target = i + fp->k + 1;
 492                         insn->code = fp->code;
 493                         BPF_EMIT_JMP;
 494                         break;
 495
 496                 case BPF_JMP | BPF_JEQ | BPF_K:
 497                 case BPF_JMP | BPF_JEQ | BPF_X:
 498                 case BPF_JMP | BPF_JSET | BPF_K:
 499                 case BPF_JMP | BPF_JSET | BPF_X:
 500                 case BPF_JMP | BPF_JGT | BPF_K:
 501                 case BPF_JMP | BPF_JGT | BPF_X:
 502                 case BPF_JMP | BPF_JGE | BPF_K:
 503                 case BPF_JMP | BPF_JGE | BPF_X:
 504                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 505                                 /* BPF immediates are signed, zero extend
 506                                  * immediate into tmp register and use it
 507                                  * in compare insn.
 508                                  */
 509                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 510
 511                                 insn->dst_reg = BPF_REG_A;
 512                                 insn->src_reg = BPF_REG_TMP;
 513                                 bpf_src = BPF_X;
 514                         } else {
 515                                 insn->dst_reg = BPF_REG_A;
 516                                 insn->imm = fp->k;
 517                                 bpf_src = BPF_SRC(fp->code);
 518                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 519                         }
 520
 521                         /* Common case where 'jump_false' is next insn. */
 522                         if (fp->jf == 0) {
 523                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 524                                 target = i + fp->jt + 1;
 525                                 BPF_EMIT_JMP;
 526                                 break;
 527                         }
 528
 529                         /* Convert some jumps when 'jump_true' is next insn. */
 530                         if (fp->jt == 0) {
 531                                 switch (BPF_OP(fp->code)) {
 532                                 case BPF_JEQ:
 533                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 534                                         break;
 535                                 case BPF_JGT:
 536                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 537                                         break;
 538                                 case BPF_JGE:
 539                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 540                                         break;
 541                                 default:
 542                                         goto jmp_rest;
 543                                 }
 544
 545                                 target = i + fp->jf + 1;
 546                                 BPF_EMIT_JMP;
 547                                 break;
 548                         }
 549 jmp_rest:
 550                         /* Other jumps are mapped into two insns: Jxx and JA. */
 551                         target = i + fp->jt + 1;
 552                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 553                         BPF_EMIT_JMP;
 554                         insn++;
 555
 556                         insn->code = BPF_JMP | BPF_JA;
 557                         target = i + fp->jf + 1;
 558                         BPF_EMIT_JMP;
 559                         break;
 560
 561                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 562                 case BPF_LDX | BPF_MSH | BPF_B:
 563                         /* tmp = A */
 564                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
 565                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 566                         *insn++ = BPF_LD_ABS(BPF_B, fp->k);
 567                         /* A &= 0xf */
 568                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 569                         /* A <<= 2 */
 570                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 571                         /* X = A */
 572                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 573                         /* A = tmp */
 574                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 575                         break;
 576
 577                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 578                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 579                  */
 580                 case BPF_RET | BPF_A:
 581                 case BPF_RET | BPF_K:
 582                         if (BPF_RVAL(fp->code) == BPF_K)
 583                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 584                                                         0, fp->k);
 585                         *insn = BPF_EXIT_INSN();
 586                         break;
 587
 588                 /* Store to stack. */
 589                 case BPF_ST:
 590                 case BPF_STX:
 591                         stack_off = fp->k * 4  + 4;
 592                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 593                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 594                                             -stack_off);
 595                         /* check_load_and_stores() verifies that classic BPF can
 596                          * load from stack only after write, so tracking
 597                          * stack_depth for ST|STX insns is enough
 598                          */
 599                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 600                                 new_prog->aux->stack_depth = stack_off;
 601                         break;
 602
 603                 /* Load from stack. */
 604                 case BPF_LD | BPF_MEM:
 605                 case BPF_LDX | BPF_MEM:
 606                         stack_off = fp->k * 4  + 4;
 607                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 608                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 609                                             -stack_off);
 610                         break;
 611
 612                 /* A = K or X = K */
 613                 case BPF_LD | BPF_IMM:
 614                 case BPF_LDX | BPF_IMM:
 615                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 616                                               BPF_REG_A : BPF_REG_X, fp->k);
 617                         break;
 618
 619                 /* X = A */
 620                 case BPF_MISC | BPF_TAX:
 621                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 622                         break;
 623
 624                 /* A = X */
 625                 case BPF_MISC | BPF_TXA:
 626                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 627                         break;
 628
 629                 /* A = skb->len or X = skb->len */
 630                 case BPF_LD | BPF_W | BPF_LEN:
 631                 case BPF_LDX | BPF_W | BPF_LEN:
 632                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 633                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 634                                             offsetof(struct sk_buff, len));
 635                         break;
 636
 637                 /* Access seccomp_data fields. */
 638                 case BPF_LDX | BPF_ABS | BPF_W:
 639                         /* A = *(u32 *) (ctx + K) */
 640                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 641                         break;
 642
 643                 /* Unknown instruction. */
 644                 default:
 645                         goto err;
 646                 }
 647
 648                 insn++;
 649                 if (new_prog)
 650                         memcpy(new_insn, tmp_insns,
 651                                sizeof(*insn) * (insn - tmp_insns));
 652                 new_insn += insn - tmp_insns;
 653         }
 654
 655         if (!new_prog) {
 656                 /* Only calculating new length. */
 657                 *new_len = new_insn - first_insn;
 658                 return 0;
 659         }
 660
 661         pass++;
 662         if (new_flen != new_insn - first_insn) {
 663                 new_flen = new_insn - first_insn;
 664                 if (pass > 2)
 665                         goto err;
 666                 goto do_pass;
 667         }
 668
 669         kfree(addrs);
 670         BUG_ON(*new_len != new_flen);
 671         return 0;
 672 err:
 673         kfree(addrs);
 674         return -EINVAL;
 675 }
 676
 677 /* Security:
 678  *
 679  * As we dont want to clear mem[] array for each packet going through
 680  * __bpf_prog_run(), we check that filter loaded by user never try to read
 681  * a cell if not previously written, and we check all branches to be sure
 682  * a malicious user doesn't try to abuse us.
 683  */
 684 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 685 {
 686         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 687         int pc, ret = 0;
 688
 689         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 690
 691         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 692         if (!masks)
 693                 return -ENOMEM;
 694
 695         memset(masks, 0xff, flen * sizeof(*masks));
 696
 697         for (pc = 0; pc < flen; pc++) {
 698                 memvalid &= masks[pc];
 699
 700                 switch (filter[pc].code) {
 701                 case BPF_ST:
 702                 case BPF_STX:
 703                         memvalid |= (1 << filter[pc].k);
 704                         break;
 705                 case BPF_LD | BPF_MEM:
 706                 case BPF_LDX | BPF_MEM:
 707                         if (!(memvalid & (1 << filter[pc].k))) {
 708                                 ret = -EINVAL;
 709                                 goto error;
 710                         }
 711                         break;
 712                 case BPF_JMP | BPF_JA:
 713                         /* A jump must set masks on target */
 714                         masks[pc + 1 + filter[pc].k] &= memvalid;
 715                         memvalid = ~0;
 716                         break;
 717                 case BPF_JMP | BPF_JEQ | BPF_K:
 718                 case BPF_JMP | BPF_JEQ | BPF_X:
 719                 case BPF_JMP | BPF_JGE | BPF_K:
 720                 case BPF_JMP | BPF_JGE | BPF_X:
 721                 case BPF_JMP | BPF_JGT | BPF_K:
 722                 case BPF_JMP | BPF_JGT | BPF_X:
 723                 case BPF_JMP | BPF_JSET | BPF_K:
 724                 case BPF_JMP | BPF_JSET | BPF_X:
 725                         /* A jump must set masks on targets */
 726                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 727                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 728                         memvalid = ~0;
 729                         break;
 730                 }
 731         }
 732 error:
 733         kfree(masks);
 734         return ret;
 735 }
 736
 737 static bool chk_code_allowed(u16 code_to_probe)
 738 {
 739         static const bool codes[] = {
 740                 /* 32 bit ALU operations */
 741                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 742                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 743                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 744                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 745                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 746                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 747                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 748                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 749                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 750                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 751                 [BPF_ALU | BPF_AND | BPF_K] = true,
 752                 [BPF_ALU | BPF_AND | BPF_X] = true,
 753                 [BPF_ALU | BPF_OR | BPF_K] = true,
 754                 [BPF_ALU | BPF_OR | BPF_X] = true,
 755                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 756                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 757                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 758                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 759                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 760                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 761                 [BPF_ALU | BPF_NEG] = true,
 762                 /* Load instructions */
 763                 [BPF_LD | BPF_W | BPF_ABS] = true,
 764                 [BPF_LD | BPF_H | BPF_ABS] = true,
 765                 [BPF_LD | BPF_B | BPF_ABS] = true,
 766                 [BPF_LD | BPF_W | BPF_LEN] = true,
 767                 [BPF_LD | BPF_W | BPF_IND] = true,
 768                 [BPF_LD | BPF_H | BPF_IND] = true,
 769                 [BPF_LD | BPF_B | BPF_IND] = true,
 770                 [BPF_LD | BPF_IMM] = true,
 771                 [BPF_LD | BPF_MEM] = true,
 772                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 773                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 774                 [BPF_LDX | BPF_IMM] = true,
 775                 [BPF_LDX | BPF_MEM] = true,
 776                 /* Store instructions */
 777                 [BPF_ST] = true,
 778                 [BPF_STX] = true,
 779                 /* Misc instructions */
 780                 [BPF_MISC | BPF_TAX] = true,
 781                 [BPF_MISC | BPF_TXA] = true,
 782                 /* Return instructions */
 783                 [BPF_RET | BPF_K] = true,
 784                 [BPF_RET | BPF_A] = true,
 785                 /* Jump instructions */
 786                 [BPF_JMP | BPF_JA] = true,
 787                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
 788                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
 789                 [BPF_JMP | BPF_JGE | BPF_K] = true,
 790                 [BPF_JMP | BPF_JGE | BPF_X] = true,
 791                 [BPF_JMP | BPF_JGT | BPF_K] = true,
 792                 [BPF_JMP | BPF_JGT | BPF_X] = true,
 793                 [BPF_JMP | BPF_JSET | BPF_K] = true,
 794                 [BPF_JMP | BPF_JSET | BPF_X] = true,
 795         };
 796
 797         if (code_to_probe >= ARRAY_SIZE(codes))
 798                 return false;
 799
 800         return codes[code_to_probe];
 801 }
 802
 803 static bool bpf_check_basics_ok(const struct sock_filter *filter,
 804                                 unsigned int flen)
 805 {
 806         if (filter == NULL)
 807                 return false;
 808         if (flen == 0 || flen > BPF_MAXINSNS)
 809                 return false;
 810
 811         return true;
 812 }
 813
 814 /**
 815  *      bpf_check_classic - verify socket filter code
 816  *      @filter: filter to verify
 817  *      @flen: length of filter
 818  *
 819  * Check the user's filter code. If we let some ugly
 820  * filter code slip through kaboom! The filter must contain
 821  * no references or jumps that are out of range, no illegal
 822  * instructions, and must end with a RET instruction.
 823  *
 824  * All jumps are forward as they are not signed.
 825  *
 826  * Returns 0 if the rule set is legal or -EINVAL if not.
 827  */
 828 static int bpf_check_classic(const struct sock_filter *filter,
 829                              unsigned int flen)
 830 {
 831         bool anc_found;
 832         int pc;
 833
 834         /* Check the filter code now */
 835         for (pc = 0; pc < flen; pc++) {
 836                 const struct sock_filter *ftest = &filter[pc];
 837
 838                 /* May we actually operate on this code? */
 839                 if (!chk_code_allowed(ftest->code))
 840                         return -EINVAL;
 841
 842                 /* Some instructions need special checks */
 843                 switch (ftest->code) {
 844                 case BPF_ALU | BPF_DIV | BPF_K:
 845                 case BPF_ALU | BPF_MOD | BPF_K:
 846                         /* Check for division by zero */
 847                         if (ftest->k == 0)
 848                                 return -EINVAL;
 849                         break;
 850                 case BPF_ALU | BPF_LSH | BPF_K:
 851                 case BPF_ALU | BPF_RSH | BPF_K:
 852                         if (ftest->k >= 32)
 853                                 return -EINVAL;
 854                         break;
 855                 case BPF_LD | BPF_MEM:
 856                 case BPF_LDX | BPF_MEM:
 857                 case BPF_ST:
 858                 case BPF_STX:
 859                         /* Check for invalid memory addresses */
 860                         if (ftest->k >= BPF_MEMWORDS)
 861                                 return -EINVAL;
 862                         break;
 863                 case BPF_JMP | BPF_JA:
 864                         /* Note, the large ftest->k might cause loops.
 865                          * Compare this with conditional jumps below,
 866                          * where offsets are limited. --ANK (981016)
 867                          */
 868                         if (ftest->k >= (unsigned int)(flen - pc - 1))
 869                                 return -EINVAL;
 870                         break;
 871                 case BPF_JMP | BPF_JEQ | BPF_K:
 872                 case BPF_JMP | BPF_JEQ | BPF_X:
 873                 case BPF_JMP | BPF_JGE | BPF_K:
 874                 case BPF_JMP | BPF_JGE | BPF_X:
 875                 case BPF_JMP | BPF_JGT | BPF_K:
 876                 case BPF_JMP | BPF_JGT | BPF_X:
 877                 case BPF_JMP | BPF_JSET | BPF_K:
 878                 case BPF_JMP | BPF_JSET | BPF_X:
 879                         /* Both conditionals must be safe */
 880                         if (pc + ftest->jt + 1 >= flen ||
 881                             pc + ftest->jf + 1 >= flen)
 882                                 return -EINVAL;
 883                         break;
 884                 case BPF_LD | BPF_W | BPF_ABS:
 885                 case BPF_LD | BPF_H | BPF_ABS:
 886                 case BPF_LD | BPF_B | BPF_ABS:
 887                         anc_found = false;
 888                         if (bpf_anc_helper(ftest) & BPF_ANC)
 889                                 anc_found = true;
 890                         /* Ancillary operation unknown or unsupported */
 891                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
 892                                 return -EINVAL;
 893                 }
 894         }
 895
 896         /* Last instruction must be a RET code */
 897         switch (filter[flen - 1].code) {
 898         case BPF_RET | BPF_K:
 899         case BPF_RET | BPF_A:
 900                 return check_load_and_stores(filter, flen);
 901         }
 902
 903         return -EINVAL;
 904 }
 905
 906 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
 907                                       const struct sock_fprog *fprog)
 908 {
 909         unsigned int fsize = bpf_classic_proglen(fprog);
 910         struct sock_fprog_kern *fkprog;
 911
 912         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
 913         if (!fp->orig_prog)
 914                 return -ENOMEM;
 915
 916         fkprog = fp->orig_prog;
 917         fkprog->len = fprog->len;
 918
 919         fkprog->filter = kmemdup(fp->insns, fsize,
 920                                  GFP_KERNEL | __GFP_NOWARN);
 921         if (!fkprog->filter) {
 922                 kfree(fp->orig_prog);
 923                 return -ENOMEM;
 924         }
 925
 926         return 0;
 927 }
 928
 929 static void bpf_release_orig_filter(struct bpf_prog *fp)
 930 {
 931         struct sock_fprog_kern *fprog = fp->orig_prog;
 932
 933         if (fprog) {
 934                 kfree(fprog->filter);
 935                 kfree(fprog);
 936         }
 937 }
 938
 939 static void __bpf_prog_release(struct bpf_prog *prog)
 940 {
 941         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
 942                 bpf_prog_put(prog);
 943         } else {
 944                 bpf_release_orig_filter(prog);
 945                 bpf_prog_free(prog);
 946         }
 947 }
 948
 949 static void __sk_filter_release(struct sk_filter *fp)
 950 {
 951         __bpf_prog_release(fp->prog);
 952         kfree(fp);
 953 }
 954
 955 /**
 956  *      sk_filter_release_rcu - Release a socket filter by rcu_head
 957  *      @rcu: rcu_head that contains the sk_filter to free
 958  */
 959 static void sk_filter_release_rcu(struct rcu_head *rcu)
 960 {
 961         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
 962
 963         __sk_filter_release(fp);
 964 }
 965
 966 /**
 967  *      sk_filter_release - release a socket filter
 968  *      @fp: filter to remove
 969  *
 970  *      Remove a filter from a socket and release its resources.
 971  */
 972 static void sk_filter_release(struct sk_filter *fp)
 973 {
 974         if (refcount_dec_and_test(&fp->refcnt))
 975                 call_rcu(&fp->rcu, sk_filter_release_rcu);
 976 }
 977
 978 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
 979 {
 980         u32 filter_size = bpf_prog_size(fp->prog->len);
 981
 982         atomic_sub(filter_size, &sk->sk_omem_alloc);
 983         sk_filter_release(fp);
 984 }
 985
 986 /* try to charge the socket memory if there is space available
 987  * return true on success
 988  */
 989 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
 990 {
 991         u32 filter_size = bpf_prog_size(fp->prog->len);
 992
 993         /* same check as in sock_kmalloc() */
 994         if (filter_size <= sysctl_optmem_max &&
 995             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
 996                 atomic_add(filter_size, &sk->sk_omem_alloc);
 997                 return true;
 998         }
 999         return false;
1000 }
1001
1002 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1003 {
1004         if (!refcount_inc_not_zero(&fp->refcnt))
1005                 return false;
1006
1007         if (!__sk_filter_charge(sk, fp)) {
1008                 sk_filter_release(fp);
1009                 return false;
1010         }
1011         return true;
1012 }
1013
1014 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1015 {
1016         struct sock_filter *old_prog;
1017         struct bpf_prog *old_fp;
1018         int err, new_len, old_len = fp->len;
1019
1020         /* We are free to overwrite insns et al right here as it
1021          * won't be used at this point in time anymore internally
1022          * after the migration to the internal BPF instruction
1023          * representation.
1024          */
1025         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1026                      sizeof(struct bpf_insn));
1027
1028         /* Conversion cannot happen on overlapping memory areas,
1029          * so we need to keep the user BPF around until the 2nd
1030          * pass. At this time, the user BPF is stored in fp->insns.
1031          */
1032         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1033                            GFP_KERNEL | __GFP_NOWARN);
1034         if (!old_prog) {
1035                 err = -ENOMEM;
1036                 goto out_err;
1037         }
1038
1039         /* 1st pass: calculate the new program length. */
1040         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
1041         if (err)
1042                 goto out_err_free;
1043
1044         /* Expand fp for appending the new filter representation. */
1045         old_fp = fp;
1046         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1047         if (!fp) {
1048                 /* The old_fp is still around in case we couldn't
1049                  * allocate new memory, so uncharge on that one.
1050                  */
1051                 fp = old_fp;
1052                 err = -ENOMEM;
1053                 goto out_err_free;
1054         }
1055
1056         fp->len = new_len;
1057
1058         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1059         err = bpf_convert_filter(old_prog, old_len, fp, &new_len);
1060         if (err)
1061                 /* 2nd bpf_convert_filter() can fail only if it fails
1062                  * to allocate memory, remapping must succeed. Note,
1063                  * that at this time old_fp has already been released
1064                  * by krealloc().
1065                  */
1066                 goto out_err_free;
1067
1068         fp = bpf_prog_select_runtime(fp, &err);
1069         if (err)
1070                 goto out_err_free;
1071
1072         kfree(old_prog);
1073         return fp;
1074
1075 out_err_free:
1076         kfree(old_prog);
1077 out_err:
1078         __bpf_prog_release(fp);
1079         return ERR_PTR(err);
1080 }
1081
1082 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1083                                            bpf_aux_classic_check_t trans)
1084 {
1085         int err;
1086
1087         fp->bpf_func = NULL;
1088         fp->jited = 0;
1089
1090         err = bpf_check_classic(fp->insns, fp->len);
1091         if (err) {
1092                 __bpf_prog_release(fp);
1093                 return ERR_PTR(err);
1094         }
1095
1096         /* There might be additional checks and transformations
1097          * needed on classic filters, f.e. in case of seccomp.
1098          */
1099         if (trans) {
1100                 err = trans(fp->insns, fp->len);
1101                 if (err) {
1102                         __bpf_prog_release(fp);
1103                         return ERR_PTR(err);
1104                 }
1105         }
1106
1107         /* Probe if we can JIT compile the filter and if so, do
1108          * the compilation of the filter.
1109          */
1110         bpf_jit_compile(fp);
1111
1112         /* JIT compiler couldn't process this filter, so do the
1113          * internal BPF translation for the optimized interpreter.
1114          */
1115         if (!fp->jited)
1116                 fp = bpf_migrate_filter(fp);
1117
1118         return fp;
1119 }
1120
1121 /**
1122  *      bpf_prog_create - create an unattached filter
1123  *      @pfp: the unattached filter that is created
1124  *      @fprog: the filter program
1125  *
1126  * Create a filter independent of any socket. We first run some
1127  * sanity checks on it to make sure it does not explode on us later.
1128  * If an error occurs or there is insufficient memory for the filter
1129  * a negative errno code is returned. On success the return is zero.
1130  */
1131 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1132 {
1133         unsigned int fsize = bpf_classic_proglen(fprog);
1134         struct bpf_prog *fp;
1135
1136         /* Make sure new filter is there and in the right amounts. */
1137         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1138                 return -EINVAL;
1139
1140         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1141         if (!fp)
1142                 return -ENOMEM;
1143
1144         memcpy(fp->insns, fprog->filter, fsize);
1145
1146         fp->len = fprog->len;
1147         /* Since unattached filters are not copied back to user
1148          * space through sk_get_filter(), we do not need to hold
1149          * a copy here, and can spare us the work.
1150          */
1151         fp->orig_prog = NULL;
1152
1153         /* bpf_prepare_filter() already takes care of freeing
1154          * memory in case something goes wrong.
1155          */
1156         fp = bpf_prepare_filter(fp, NULL);
1157         if (IS_ERR(fp))
1158                 return PTR_ERR(fp);
1159
1160         *pfp = fp;
1161         return 0;
1162 }
1163 EXPORT_SYMBOL_GPL(bpf_prog_create);
1164
1165 /**
1166  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1167  *      @pfp: the unattached filter that is created
1168  *      @fprog: the filter program
1169  *      @trans: post-classic verifier transformation handler
1170  *      @save_orig: save classic BPF program
1171  *
1172  * This function effectively does the same as bpf_prog_create(), only
1173  * that it builds up its insns buffer from user space provided buffer.
1174  * It also allows for passing a bpf_aux_classic_check_t handler.
1175  */
1176 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1177                               bpf_aux_classic_check_t trans, bool save_orig)
1178 {
1179         unsigned int fsize = bpf_classic_proglen(fprog);
1180         struct bpf_prog *fp;
1181         int err;
1182
1183         /* Make sure new filter is there and in the right amounts. */
1184         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1185                 return -EINVAL;
1186
1187         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1188         if (!fp)
1189                 return -ENOMEM;
1190
1191         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1192                 __bpf_prog_free(fp);
1193                 return -EFAULT;
1194         }
1195
1196         fp->len = fprog->len;
1197         fp->orig_prog = NULL;
1198
1199         if (save_orig) {
1200                 err = bpf_prog_store_orig_filter(fp, fprog);
1201                 if (err) {
1202                         __bpf_prog_free(fp);
1203                         return -ENOMEM;
1204                 }
1205         }
1206
1207         /* bpf_prepare_filter() already takes care of freeing
1208          * memory in case something goes wrong.
1209          */
1210         fp = bpf_prepare_filter(fp, trans);
1211         if (IS_ERR(fp))
1212                 return PTR_ERR(fp);
1213
1214         *pfp = fp;
1215         return 0;
1216 }
1217 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1218
1219 void bpf_prog_destroy(struct bpf_prog *fp)
1220 {
1221         __bpf_prog_release(fp);
1222 }
1223 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1224
1225 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1226 {
1227         struct sk_filter *fp, *old_fp;
1228
1229         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1230         if (!fp)
1231                 return -ENOMEM;
1232
1233         fp->prog = prog;
1234
1235         if (!__sk_filter_charge(sk, fp)) {
1236                 kfree(fp);
1237                 return -ENOMEM;
1238         }
1239         refcount_set(&fp->refcnt, 1);
1240
1241         old_fp = rcu_dereference_protected(sk->sk_filter,
1242                                            lockdep_sock_is_held(sk));
1243         rcu_assign_pointer(sk->sk_filter, fp);
1244
1245         if (old_fp)
1246                 sk_filter_uncharge(sk, old_fp);
1247
1248         return 0;
1249 }
1250
1251 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1252 {
1253         struct bpf_prog *old_prog;
1254         int err;
1255
1256         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1257                 return -ENOMEM;
1258
1259         if (sk_unhashed(sk) && sk->sk_reuseport) {
1260                 err = reuseport_alloc(sk);
1261                 if (err)
1262                         return err;
1263         } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1264                 /* The socket wasn't bound with SO_REUSEPORT */
1265                 return -EINVAL;
1266         }
1267
1268         old_prog = reuseport_attach_prog(sk, prog);
1269         if (old_prog)
1270                 bpf_prog_destroy(old_prog);
1271
1272         return 0;
1273 }
1274
1275 static
1276 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1277 {
1278         unsigned int fsize = bpf_classic_proglen(fprog);
1279         struct bpf_prog *prog;
1280         int err;
1281
1282         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1283                 return ERR_PTR(-EPERM);
1284
1285         /* Make sure new filter is there and in the right amounts. */
1286         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1287                 return ERR_PTR(-EINVAL);
1288
1289         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1290         if (!prog)
1291                 return ERR_PTR(-ENOMEM);
1292
1293         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1294                 __bpf_prog_free(prog);
1295                 return ERR_PTR(-EFAULT);
1296         }
1297
1298         prog->len = fprog->len;
1299
1300         err = bpf_prog_store_orig_filter(prog, fprog);
1301         if (err) {
1302                 __bpf_prog_free(prog);
1303                 return ERR_PTR(-ENOMEM);
1304         }
1305
1306         /* bpf_prepare_filter() already takes care of freeing
1307          * memory in case something goes wrong.
1308          */
1309         return bpf_prepare_filter(prog, NULL);
1310 }
1311
1312 /**
1313  *      sk_attach_filter - attach a socket filter
1314  *      @fprog: the filter program
1315  *      @sk: the socket to use
1316  *
1317  * Attach the user's filter code. We first run some sanity checks on
1318  * it to make sure it does not explode on us later. If an error
1319  * occurs or there is insufficient memory for the filter a negative
1320  * errno code is returned. On success the return is zero.
1321  */
1322 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1323 {
1324         struct bpf_prog *prog = __get_filter(fprog, sk);
1325         int err;
1326
1327         if (IS_ERR(prog))
1328                 return PTR_ERR(prog);
1329
1330         err = __sk_attach_prog(prog, sk);
1331         if (err < 0) {
1332                 __bpf_prog_release(prog);
1333                 return err;
1334         }
1335
1336         return 0;
1337 }
1338 EXPORT_SYMBOL_GPL(sk_attach_filter);
1339
1340 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1341 {
1342         struct bpf_prog *prog = __get_filter(fprog, sk);
1343         int err;
1344
1345         if (IS_ERR(prog))
1346                 return PTR_ERR(prog);
1347
1348         err = __reuseport_attach_prog(prog, sk);
1349         if (err < 0) {
1350                 __bpf_prog_release(prog);
1351                 return err;
1352         }
1353
1354         return 0;
1355 }
1356
1357 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1358 {
1359         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1360                 return ERR_PTR(-EPERM);
1361
1362         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1363 }
1364
1365 int sk_attach_bpf(u32 ufd, struct sock *sk)
1366 {
1367         struct bpf_prog *prog = __get_bpf(ufd, sk);
1368         int err;
1369
1370         if (IS_ERR(prog))
1371                 return PTR_ERR(prog);
1372
1373         err = __sk_attach_prog(prog, sk);
1374         if (err < 0) {
1375                 bpf_prog_put(prog);
1376                 return err;
1377         }
1378
1379         return 0;
1380 }
1381
1382 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1383 {
1384         struct bpf_prog *prog = __get_bpf(ufd, sk);
1385         int err;
1386
1387         if (IS_ERR(prog))
1388                 return PTR_ERR(prog);
1389
1390         err = __reuseport_attach_prog(prog, sk);
1391         if (err < 0) {
1392                 bpf_prog_put(prog);
1393                 return err;
1394         }
1395
1396         return 0;
1397 }
1398
1399 struct bpf_scratchpad {
1400         union {
1401                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1402                 u8     buff[MAX_BPF_STACK];
1403         };
1404 };
1405
1406 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1407
1408 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1409                                           unsigned int write_len)
1410 {
1411         return skb_ensure_writable(skb, write_len);
1412 }
1413
1414 static inline int bpf_try_make_writable(struct sk_buff *skb,
1415                                         unsigned int write_len)
1416 {
1417         int err = __bpf_try_make_writable(skb, write_len);
1418
1419         bpf_compute_data_pointers(skb);
1420         return err;
1421 }
1422
1423 static int bpf_try_make_head_writable(struct sk_buff *skb)
1424 {
1425         return bpf_try_make_writable(skb, skb_headlen(skb));
1426 }
1427
1428 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1429 {
1430         if (skb_at_tc_ingress(skb))
1431                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1432 }
1433
1434 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1435 {
1436         if (skb_at_tc_ingress(skb))
1437                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1438 }
1439
1440 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1441            const void *, from, u32, len, u64, flags)
1442 {
1443         void *ptr;
1444
1445         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1446                 return -EINVAL;
1447         if (unlikely(offset > 0xffff))
1448                 return -EFAULT;
1449         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1450                 return -EFAULT;
1451
1452         ptr = skb->data + offset;
1453         if (flags & BPF_F_RECOMPUTE_CSUM)
1454                 __skb_postpull_rcsum(skb, ptr, len, offset);
1455
1456         memcpy(ptr, from, len);
1457
1458         if (flags & BPF_F_RECOMPUTE_CSUM)
1459                 __skb_postpush_rcsum(skb, ptr, len, offset);
1460         if (flags & BPF_F_INVALIDATE_HASH)
1461                 skb_clear_hash(skb);
1462
1463         return 0;
1464 }
1465
1466 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1467         .func           = bpf_skb_store_bytes,
1468         .gpl_only       = false,
1469         .ret_type       = RET_INTEGER,
1470         .arg1_type      = ARG_PTR_TO_CTX,
1471         .arg2_type      = ARG_ANYTHING,
1472         .arg3_type      = ARG_PTR_TO_MEM,
1473         .arg4_type      = ARG_CONST_SIZE,
1474         .arg5_type      = ARG_ANYTHING,
1475 };
1476
1477 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1478            void *, to, u32, len)
1479 {
1480         void *ptr;
1481
1482         if (unlikely(offset > 0xffff))
1483                 goto err_clear;
1484
1485         ptr = skb_header_pointer(skb, offset, len, to);
1486         if (unlikely(!ptr))
1487                 goto err_clear;
1488         if (ptr != to)
1489                 memcpy(to, ptr, len);
1490
1491         return 0;
1492 err_clear:
1493         memset(to, 0, len);
1494         return -EFAULT;
1495 }
1496
1497 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1498         .func           = bpf_skb_load_bytes,
1499         .gpl_only       = false,
1500         .ret_type       = RET_INTEGER,
1501         .arg1_type      = ARG_PTR_TO_CTX,
1502         .arg2_type      = ARG_ANYTHING,
1503         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1504         .arg4_type      = ARG_CONST_SIZE,
1505 };
1506
1507 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1508 {
1509         /* Idea is the following: should the needed direct read/write
1510          * test fail during runtime, we can pull in more data and redo
1511          * again, since implicitly, we invalidate previous checks here.
1512          *
1513          * Or, since we know how much we need to make read/writeable,
1514          * this can be done once at the program beginning for direct
1515          * access case. By this we overcome limitations of only current
1516          * headroom being accessible.
1517          */
1518         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1519 }
1520
1521 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1522         .func           = bpf_skb_pull_data,
1523         .gpl_only       = false,
1524         .ret_type       = RET_INTEGER,
1525         .arg1_type      = ARG_PTR_TO_CTX,
1526         .arg2_type      = ARG_ANYTHING,
1527 };
1528
1529 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1530            u64, from, u64, to, u64, flags)
1531 {
1532         __sum16 *ptr;
1533
1534         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1535                 return -EINVAL;
1536         if (unlikely(offset > 0xffff || offset & 1))
1537                 return -EFAULT;
1538         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1539                 return -EFAULT;
1540
1541         ptr = (__sum16 *)(skb->data + offset);
1542         switch (flags & BPF_F_HDR_FIELD_MASK) {
1543         case 0:
1544                 if (unlikely(from != 0))
1545                         return -EINVAL;
1546
1547                 csum_replace_by_diff(ptr, to);
1548                 break;
1549         case 2:
1550                 csum_replace2(ptr, from, to);
1551                 break;
1552         case 4:
1553                 csum_replace4(ptr, from, to);
1554                 break;
1555         default:
1556                 return -EINVAL;
1557         }
1558
1559         return 0;
1560 }
1561
1562 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1563         .func           = bpf_l3_csum_replace,
1564         .gpl_only       = false,
1565         .ret_type       = RET_INTEGER,
1566         .arg1_type      = ARG_PTR_TO_CTX,
1567         .arg2_type      = ARG_ANYTHING,
1568         .arg3_type      = ARG_ANYTHING,
1569         .arg4_type      = ARG_ANYTHING,
1570         .arg5_type      = ARG_ANYTHING,
1571 };
1572
1573 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1574            u64, from, u64, to, u64, flags)
1575 {
1576         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1577         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1578         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1579         __sum16 *ptr;
1580
1581         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1582                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1583                 return -EINVAL;
1584         if (unlikely(offset > 0xffff || offset & 1))
1585                 return -EFAULT;
1586         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1587                 return -EFAULT;
1588
1589         ptr = (__sum16 *)(skb->data + offset);
1590         if (is_mmzero && !do_mforce && !*ptr)
1591                 return 0;
1592
1593         switch (flags & BPF_F_HDR_FIELD_MASK) {
1594         case 0:
1595                 if (unlikely(from != 0))
1596                         return -EINVAL;
1597
1598                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1599                 break;
1600         case 2:
1601                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1602                 break;
1603         case 4:
1604                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1605                 break;
1606         default:
1607                 return -EINVAL;
1608         }
1609
1610         if (is_mmzero && !*ptr)
1611                 *ptr = CSUM_MANGLED_0;
1612         return 0;
1613 }
1614
1615 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1616         .func           = bpf_l4_csum_replace,
1617         .gpl_only       = false,
1618         .ret_type       = RET_INTEGER,
1619         .arg1_type      = ARG_PTR_TO_CTX,
1620         .arg2_type      = ARG_ANYTHING,
1621         .arg3_type      = ARG_ANYTHING,
1622         .arg4_type      = ARG_ANYTHING,
1623         .arg5_type      = ARG_ANYTHING,
1624 };
1625
1626 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1627            __be32 *, to, u32, to_size, __wsum, seed)
1628 {
1629         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1630         u32 diff_size = from_size + to_size;
1631         int i, j = 0;
1632
1633         /* This is quite flexible, some examples:
1634          *
1635          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1636          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1637          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
1638          *
1639          * Even for diffing, from_size and to_size don't need to be equal.
1640          */
1641         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1642                      diff_size > sizeof(sp->diff)))
1643                 return -EINVAL;
1644
1645         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1646                 sp->diff[j] = ~from[i];
1647         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
1648                 sp->diff[j] = to[i];
1649
1650         return csum_partial(sp->diff, diff_size, seed);
1651 }
1652
1653 static const struct bpf_func_proto bpf_csum_diff_proto = {
1654         .func           = bpf_csum_diff,
1655         .gpl_only       = false,
1656         .pkt_access     = true,
1657         .ret_type       = RET_INTEGER,
1658         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1659         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1660         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
1661         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
1662         .arg5_type      = ARG_ANYTHING,
1663 };
1664
1665 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1666 {
1667         /* The interface is to be used in combination with bpf_csum_diff()
1668          * for direct packet writes. csum rotation for alignment as well
1669          * as emulating csum_sub() can be done from the eBPF program.
1670          */
1671         if (skb->ip_summed == CHECKSUM_COMPLETE)
1672                 return (skb->csum = csum_add(skb->csum, csum));
1673
1674         return -ENOTSUPP;
1675 }
1676
1677 static const struct bpf_func_proto bpf_csum_update_proto = {
1678         .func           = bpf_csum_update,
1679         .gpl_only       = false,
1680         .ret_type       = RET_INTEGER,
1681         .arg1_type      = ARG_PTR_TO_CTX,
1682         .arg2_type      = ARG_ANYTHING,
1683 };
1684
1685 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1686 {
1687         return dev_forward_skb(dev, skb);
1688 }
1689
1690 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1691                                       struct sk_buff *skb)
1692 {
1693         int ret = ____dev_forward_skb(dev, skb);
1694
1695         if (likely(!ret)) {
1696                 skb->dev = dev;
1697                 ret = netif_rx(skb);
1698         }
1699
1700         return ret;
1701 }
1702
1703 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1704 {
1705         int ret;
1706
1707         if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1708                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1709                 kfree_skb(skb);
1710                 return -ENETDOWN;
1711         }
1712
1713         skb->dev = dev;
1714
1715         __this_cpu_inc(xmit_recursion);
1716         ret = dev_queue_xmit(skb);
1717         __this_cpu_dec(xmit_recursion);
1718
1719         return ret;
1720 }
1721
1722 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
1723                                  u32 flags)
1724 {
1725         /* skb->mac_len is not set on normal egress */
1726         unsigned int mlen = skb->network_header - skb->mac_header;
1727
1728         __skb_pull(skb, mlen);
1729
1730         /* At ingress, the mac header has already been pulled once.
1731          * At egress, skb_pospull_rcsum has to be done in case that
1732          * the skb is originated from ingress (i.e. a forwarded skb)
1733          * to ensure that rcsum starts at net header.
1734          */
1735         if (!skb_at_tc_ingress(skb))
1736                 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
1737         skb_pop_mac_header(skb);
1738         skb_reset_mac_len(skb);
1739         return flags & BPF_F_INGRESS ?
1740                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
1741 }
1742
1743 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
1744                                  u32 flags)
1745 {
1746         /* Verify that a link layer header is carried */
1747         if (unlikely(skb->mac_header >= skb->network_header)) {
1748                 kfree_skb(skb);
1749                 return -ERANGE;
1750         }
1751
1752         bpf_push_mac_rcsum(skb);
1753         return flags & BPF_F_INGRESS ?
1754                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1755 }
1756
1757 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
1758                           u32 flags)
1759 {
1760         if (dev_is_mac_header_xmit(dev))
1761                 return __bpf_redirect_common(skb, dev, flags);
1762         else
1763                 return __bpf_redirect_no_mac(skb, dev, flags);
1764 }
1765
1766 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
1767 {
1768         struct net_device *dev;
1769         struct sk_buff *clone;
1770         int ret;
1771
1772         if (unlikely(flags & ~(BPF_F_INGRESS)))
1773                 return -EINVAL;
1774
1775         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1776         if (unlikely(!dev))
1777                 return -EINVAL;
1778
1779         clone = skb_clone(skb, GFP_ATOMIC);
1780         if (unlikely(!clone))
1781                 return -ENOMEM;
1782
1783         /* For direct write, we need to keep the invariant that the skbs
1784          * we're dealing with need to be uncloned. Should uncloning fail
1785          * here, we need to free the just generated clone to unclone once
1786          * again.
1787          */
1788         ret = bpf_try_make_head_writable(skb);
1789         if (unlikely(ret)) {
1790                 kfree_skb(clone);
1791                 return -ENOMEM;
1792         }
1793
1794         return __bpf_redirect(clone, dev, flags);
1795 }
1796
1797 static const struct bpf_func_proto bpf_clone_redirect_proto = {
1798         .func           = bpf_clone_redirect,
1799         .gpl_only       = false,
1800         .ret_type       = RET_INTEGER,
1801         .arg1_type      = ARG_PTR_TO_CTX,
1802         .arg2_type      = ARG_ANYTHING,
1803         .arg3_type      = ARG_ANYTHING,
1804 };
1805
1806 struct redirect_info {
1807         u32 ifindex;
1808         u32 flags;
1809         struct bpf_map *map;
1810         struct bpf_map *map_to_flush;
1811         unsigned long   map_owner;
1812 };
1813
1814 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1815
1816 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
1817 {
1818         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1819
1820         if (unlikely(flags & ~(BPF_F_INGRESS)))
1821                 return TC_ACT_SHOT;
1822
1823         ri->ifindex = ifindex;
1824         ri->flags = flags;
1825
1826         return TC_ACT_REDIRECT;
1827 }
1828
1829 int skb_do_redirect(struct sk_buff *skb)
1830 {
1831         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1832         struct net_device *dev;
1833
1834         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1835         ri->ifindex = 0;
1836         if (unlikely(!dev)) {
1837                 kfree_skb(skb);
1838                 return -EINVAL;
1839         }
1840
1841         return __bpf_redirect(skb, dev, ri->flags);
1842 }
1843
1844 static const struct bpf_func_proto bpf_redirect_proto = {
1845         .func           = bpf_redirect,
1846         .gpl_only       = false,
1847         .ret_type       = RET_INTEGER,
1848         .arg1_type      = ARG_ANYTHING,
1849         .arg2_type      = ARG_ANYTHING,
1850 };
1851
1852 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
1853            struct bpf_map *, map, u32, key, u64, flags)
1854 {
1855         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1856
1857         /* If user passes invalid input drop the packet. */
1858         if (unlikely(flags))
1859                 return SK_DROP;
1860
1861         tcb->bpf.key = key;
1862         tcb->bpf.flags = flags;
1863         tcb->bpf.map = map;
1864
1865         return SK_PASS;
1866 }
1867
1868 struct sock *do_sk_redirect_map(struct sk_buff *skb)
1869 {
1870         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1871         struct sock *sk = NULL;
1872
1873         if (tcb->bpf.map) {
1874                 sk = __sock_map_lookup_elem(tcb->bpf.map, tcb->bpf.key);
1875
1876                 tcb->bpf.key = 0;
1877                 tcb->bpf.map = NULL;
1878         }
1879
1880         return sk;
1881 }
1882
1883 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
1884         .func           = bpf_sk_redirect_map,
1885         .gpl_only       = false,
1886         .ret_type       = RET_INTEGER,
1887         .arg1_type      = ARG_PTR_TO_CTX,
1888         .arg2_type      = ARG_CONST_MAP_PTR,
1889         .arg3_type      = ARG_ANYTHING,
1890         .arg4_type      = ARG_ANYTHING,
1891 };
1892
1893 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
1894 {
1895         return task_get_classid(skb);
1896 }
1897
1898 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1899         .func           = bpf_get_cgroup_classid,
1900         .gpl_only       = false,
1901         .ret_type       = RET_INTEGER,
1902         .arg1_type      = ARG_PTR_TO_CTX,
1903 };
1904
1905 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
1906 {
1907         return dst_tclassid(skb);
1908 }
1909
1910 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1911         .func           = bpf_get_route_realm,
1912         .gpl_only       = false,
1913         .ret_type       = RET_INTEGER,
1914         .arg1_type      = ARG_PTR_TO_CTX,
1915 };
1916
1917 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
1918 {
1919         /* If skb_clear_hash() was called due to mangling, we can
1920          * trigger SW recalculation here. Later access to hash
1921          * can then use the inline skb->hash via context directly
1922          * instead of calling this helper again.
1923          */
1924         return skb_get_hash(skb);
1925 }
1926
1927 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
1928         .func           = bpf_get_hash_recalc,
1929         .gpl_only       = false,
1930         .ret_type       = RET_INTEGER,
1931         .arg1_type      = ARG_PTR_TO_CTX,
1932 };
1933
1934 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
1935 {
1936         /* After all direct packet write, this can be used once for
1937          * triggering a lazy recalc on next skb_get_hash() invocation.
1938          */
1939         skb_clear_hash(skb);
1940         return 0;
1941 }
1942
1943 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
1944         .func           = bpf_set_hash_invalid,
1945         .gpl_only       = false,
1946         .ret_type       = RET_INTEGER,
1947         .arg1_type      = ARG_PTR_TO_CTX,
1948 };
1949
1950 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
1951 {
1952         /* Set user specified hash as L4(+), so that it gets returned
1953          * on skb_get_hash() call unless BPF prog later on triggers a
1954          * skb_clear_hash().
1955          */
1956         __skb_set_sw_hash(skb, hash, true);
1957         return 0;
1958 }
1959
1960 static const struct bpf_func_proto bpf_set_hash_proto = {
1961         .func           = bpf_set_hash,
1962         .gpl_only       = false,
1963         .ret_type       = RET_INTEGER,
1964         .arg1_type      = ARG_PTR_TO_CTX,
1965         .arg2_type      = ARG_ANYTHING,
1966 };
1967
1968 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
1969            u16, vlan_tci)
1970 {
1971         int ret;
1972
1973         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1974                      vlan_proto != htons(ETH_P_8021AD)))
1975                 vlan_proto = htons(ETH_P_8021Q);
1976
1977         bpf_push_mac_rcsum(skb);
1978         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1979         bpf_pull_mac_rcsum(skb);
1980
1981         bpf_compute_data_pointers(skb);
1982         return ret;
1983 }
1984
1985 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1986         .func           = bpf_skb_vlan_push,
1987         .gpl_only       = false,
1988         .ret_type       = RET_INTEGER,
1989         .arg1_type      = ARG_PTR_TO_CTX,
1990         .arg2_type      = ARG_ANYTHING,
1991         .arg3_type      = ARG_ANYTHING,
1992 };
1993 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1994
1995 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
1996 {
1997         int ret;
1998
1999         bpf_push_mac_rcsum(skb);
2000         ret = skb_vlan_pop(skb);
2001         bpf_pull_mac_rcsum(skb);
2002
2003         bpf_compute_data_pointers(skb);
2004         return ret;
2005 }
2006
2007 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2008         .func           = bpf_skb_vlan_pop,
2009         .gpl_only       = false,
2010         .ret_type       = RET_INTEGER,
2011         .arg1_type      = ARG_PTR_TO_CTX,
2012 };
2013 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
2014
2015 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2016 {
2017         /* Caller already did skb_cow() with len as headroom,
2018          * so no need to do it here.
2019          */
2020         skb_push(skb, len);
2021         memmove(skb->data, skb->data + len, off);
2022         memset(skb->data + off, 0, len);
2023
2024         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2025          * needed here as it does not change the skb->csum
2026          * result for checksum complete when summing over
2027          * zeroed blocks.
2028          */
2029         return 0;
2030 }
2031
2032 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2033 {
2034         /* skb_ensure_writable() is not needed here, as we're
2035          * already working on an uncloned skb.
2036          */
2037         if (unlikely(!pskb_may_pull(skb, off + len)))
2038                 return -ENOMEM;
2039
2040         skb_postpull_rcsum(skb, skb->data + off, len);
2041         memmove(skb->data + len, skb->data, off);
2042         __skb_pull(skb, len);
2043
2044         return 0;
2045 }
2046
2047 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2048 {
2049         bool trans_same = skb->transport_header == skb->network_header;
2050         int ret;
2051
2052         /* There's no need for __skb_push()/__skb_pull() pair to
2053          * get to the start of the mac header as we're guaranteed
2054          * to always start from here under eBPF.
2055          */
2056         ret = bpf_skb_generic_push(skb, off, len);
2057         if (likely(!ret)) {
2058                 skb->mac_header -= len;
2059                 skb->network_header -= len;
2060                 if (trans_same)
2061                         skb->transport_header = skb->network_header;
2062         }
2063
2064         return ret;
2065 }
2066
2067 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2068 {
2069         bool trans_same = skb->transport_header == skb->network_header;
2070         int ret;
2071
2072         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2073         ret = bpf_skb_generic_pop(skb, off, len);
2074         if (likely(!ret)) {
2075                 skb->mac_header += len;
2076                 skb->network_header += len;
2077                 if (trans_same)
2078                         skb->transport_header = skb->network_header;
2079         }
2080
2081         return ret;
2082 }
2083
2084 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2085 {
2086         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2087         u32 off = skb_mac_header_len(skb);
2088         int ret;
2089
2090         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2091         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2092                 return -ENOTSUPP;
2093
2094         ret = skb_cow(skb, len_diff);
2095         if (unlikely(ret < 0))
2096                 return ret;
2097
2098         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2099         if (unlikely(ret < 0))
2100                 return ret;
2101
2102         if (skb_is_gso(skb)) {
2103                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2104
2105                 /* SKB_GSO_TCPV4 needs to be changed into
2106                  * SKB_GSO_TCPV6.
2107                  */
2108                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2109                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2110                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2111                 }
2112
2113                 /* Due to IPv6 header, MSS needs to be downgraded. */
2114                 skb_decrease_gso_size(shinfo, len_diff);
2115                 /* Header must be checked, and gso_segs recomputed. */
2116                 shinfo->gso_type |= SKB_GSO_DODGY;
2117                 shinfo->gso_segs = 0;
2118         }
2119
2120         skb->protocol = htons(ETH_P_IPV6);
2121         skb_clear_hash(skb);
2122
2123         return 0;
2124 }
2125
2126 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2127 {
2128         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2129         u32 off = skb_mac_header_len(skb);
2130         int ret;
2131
2132         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2133         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2134                 return -ENOTSUPP;
2135
2136         ret = skb_unclone(skb, GFP_ATOMIC);
2137         if (unlikely(ret < 0))
2138                 return ret;
2139
2140         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2141         if (unlikely(ret < 0))
2142                 return ret;
2143
2144         if (skb_is_gso(skb)) {
2145                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2146
2147                 /* SKB_GSO_TCPV6 needs to be changed into
2148                  * SKB_GSO_TCPV4.
2149                  */
2150                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2151                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2152                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2153                 }
2154
2155                 /* Due to IPv4 header, MSS can be upgraded. */
2156                 skb_increase_gso_size(shinfo, len_diff);
2157                 /* Header must be checked, and gso_segs recomputed. */
2158                 shinfo->gso_type |= SKB_GSO_DODGY;
2159                 shinfo->gso_segs = 0;
2160         }
2161
2162         skb->protocol = htons(ETH_P_IP);
2163         skb_clear_hash(skb);
2164
2165         return 0;
2166 }
2167
2168 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2169 {
2170         __be16 from_proto = skb->protocol;
2171
2172         if (from_proto == htons(ETH_P_IP) &&
2173               to_proto == htons(ETH_P_IPV6))
2174                 return bpf_skb_proto_4_to_6(skb);
2175
2176         if (from_proto == htons(ETH_P_IPV6) &&
2177               to_proto == htons(ETH_P_IP))
2178                 return bpf_skb_proto_6_to_4(skb);
2179
2180         return -ENOTSUPP;
2181 }
2182
2183 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2184            u64, flags)
2185 {
2186         int ret;
2187
2188         if (unlikely(flags))
2189                 return -EINVAL;
2190
2191         /* General idea is that this helper does the basic groundwork
2192          * needed for changing the protocol, and eBPF program fills the
2193          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2194          * and other helpers, rather than passing a raw buffer here.
2195          *
2196          * The rationale is to keep this minimal and without a need to
2197          * deal with raw packet data. F.e. even if we would pass buffers
2198          * here, the program still needs to call the bpf_lX_csum_replace()
2199          * helpers anyway. Plus, this way we keep also separation of
2200          * concerns, since f.e. bpf_skb_store_bytes() should only take
2201          * care of stores.
2202          *
2203          * Currently, additional options and extension header space are
2204          * not supported, but flags register is reserved so we can adapt
2205          * that. For offloads, we mark packet as dodgy, so that headers
2206          * need to be verified first.
2207          */
2208         ret = bpf_skb_proto_xlat(skb, proto);
2209         bpf_compute_data_pointers(skb);
2210         return ret;
2211 }
2212
2213 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2214         .func           = bpf_skb_change_proto,
2215         .gpl_only       = false,
2216         .ret_type       = RET_INTEGER,
2217         .arg1_type      = ARG_PTR_TO_CTX,
2218         .arg2_type      = ARG_ANYTHING,
2219         .arg3_type      = ARG_ANYTHING,
2220 };
2221
2222 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2223 {
2224         /* We only allow a restricted subset to be changed for now. */
2225         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2226                      !skb_pkt_type_ok(pkt_type)))
2227                 return -EINVAL;
2228
2229         skb->pkt_type = pkt_type;
2230         return 0;
2231 }
2232
2233 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2234         .func           = bpf_skb_change_type,
2235         .gpl_only       = false,
2236         .ret_type       = RET_INTEGER,
2237         .arg1_type      = ARG_PTR_TO_CTX,
2238         .arg2_type      = ARG_ANYTHING,
2239 };
2240
2241 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2242 {
2243         switch (skb->protocol) {
2244         case htons(ETH_P_IP):
2245                 return sizeof(struct iphdr);
2246         case htons(ETH_P_IPV6):
2247                 return sizeof(struct ipv6hdr);
2248         default:
2249                 return ~0U;
2250         }
2251 }
2252
2253 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2254 {
2255         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2256         int ret;
2257
2258         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2259         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2260                 return -ENOTSUPP;
2261
2262         ret = skb_cow(skb, len_diff);
2263         if (unlikely(ret < 0))
2264                 return ret;
2265
2266         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2267         if (unlikely(ret < 0))
2268                 return ret;
2269
2270         if (skb_is_gso(skb)) {
2271                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2272
2273                 /* Due to header grow, MSS needs to be downgraded. */
2274                 skb_decrease_gso_size(shinfo, len_diff);
2275                 /* Header must be checked, and gso_segs recomputed. */
2276                 shinfo->gso_type |= SKB_GSO_DODGY;
2277                 shinfo->gso_segs = 0;
2278         }
2279
2280         return 0;
2281 }
2282
2283 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2284 {
2285         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2286         int ret;
2287
2288         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2289         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2290                 return -ENOTSUPP;
2291
2292         ret = skb_unclone(skb, GFP_ATOMIC);
2293         if (unlikely(ret < 0))
2294                 return ret;
2295
2296         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2297         if (unlikely(ret < 0))
2298                 return ret;
2299
2300         if (skb_is_gso(skb)) {
2301                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2302
2303                 /* Due to header shrink, MSS can be upgraded. */
2304                 skb_increase_gso_size(shinfo, len_diff);
2305                 /* Header must be checked, and gso_segs recomputed. */
2306                 shinfo->gso_type |= SKB_GSO_DODGY;
2307                 shinfo->gso_segs = 0;
2308         }
2309
2310         return 0;
2311 }
2312
2313 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2314 {
2315         return skb->dev->mtu + skb->dev->hard_header_len;
2316 }
2317
2318 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2319 {
2320         bool trans_same = skb->transport_header == skb->network_header;
2321         u32 len_cur, len_diff_abs = abs(len_diff);
2322         u32 len_min = bpf_skb_net_base_len(skb);
2323         u32 len_max = __bpf_skb_max_len(skb);
2324         __be16 proto = skb->protocol;
2325         bool shrink = len_diff < 0;
2326         int ret;
2327
2328         if (unlikely(len_diff_abs > 0xfffU))
2329                 return -EFAULT;
2330         if (unlikely(proto != htons(ETH_P_IP) &&
2331                      proto != htons(ETH_P_IPV6)))
2332                 return -ENOTSUPP;
2333
2334         len_cur = skb->len - skb_network_offset(skb);
2335         if (skb_transport_header_was_set(skb) && !trans_same)
2336                 len_cur = skb_network_header_len(skb);
2337         if ((shrink && (len_diff_abs >= len_cur ||
2338                         len_cur - len_diff_abs < len_min)) ||
2339             (!shrink && (skb->len + len_diff_abs > len_max &&
2340                          !skb_is_gso(skb))))
2341                 return -ENOTSUPP;
2342
2343         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2344                        bpf_skb_net_grow(skb, len_diff_abs);
2345
2346         bpf_compute_data_pointers(skb);
2347         return ret;
2348 }
2349
2350 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2351            u32, mode, u64, flags)
2352 {
2353         if (unlikely(flags))
2354                 return -EINVAL;
2355         if (likely(mode == BPF_ADJ_ROOM_NET))
2356                 return bpf_skb_adjust_net(skb, len_diff);
2357
2358         return -ENOTSUPP;
2359 }
2360
2361 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2362         .func           = bpf_skb_adjust_room,
2363         .gpl_only       = false,
2364         .ret_type       = RET_INTEGER,
2365         .arg1_type      = ARG_PTR_TO_CTX,
2366         .arg2_type      = ARG_ANYTHING,
2367         .arg3_type      = ARG_ANYTHING,
2368         .arg4_type      = ARG_ANYTHING,
2369 };
2370
2371 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2372 {
2373         u32 min_len = skb_network_offset(skb);
2374
2375         if (skb_transport_header_was_set(skb))
2376                 min_len = skb_transport_offset(skb);
2377         if (skb->ip_summed == CHECKSUM_PARTIAL)
2378                 min_len = skb_checksum_start_offset(skb) +
2379                           skb->csum_offset + sizeof(__sum16);
2380         return min_len;
2381 }
2382
2383 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2384 {
2385         unsigned int old_len = skb->len;
2386         int ret;
2387
2388         ret = __skb_grow_rcsum(skb, new_len);
2389         if (!ret)
2390                 memset(skb->data + old_len, 0, new_len - old_len);
2391         return ret;
2392 }
2393
2394 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2395 {
2396         return __skb_trim_rcsum(skb, new_len);
2397 }
2398
2399 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2400            u64, flags)
2401 {
2402         u32 max_len = __bpf_skb_max_len(skb);
2403         u32 min_len = __bpf_skb_min_len(skb);
2404         int ret;
2405
2406         if (unlikely(flags || new_len > max_len || new_len < min_len))
2407                 return -EINVAL;
2408         if (skb->encapsulation)
2409                 return -ENOTSUPP;
2410
2411         /* The basic idea of this helper is that it's performing the
2412          * needed work to either grow or trim an skb, and eBPF program
2413          * rewrites the rest via helpers like bpf_skb_store_bytes(),
2414          * bpf_lX_csum_replace() and others rather than passing a raw
2415          * buffer here. This one is a slow path helper and intended
2416          * for replies with control messages.
2417          *
2418          * Like in bpf_skb_change_proto(), we want to keep this rather
2419          * minimal and without protocol specifics so that we are able
2420          * to separate concerns as in bpf_skb_store_bytes() should only
2421          * be the one responsible for writing buffers.
2422          *
2423          * It's really expected to be a slow path operation here for
2424          * control message replies, so we're implicitly linearizing,
2425          * uncloning and drop offloads from the skb by this.
2426          */
2427         ret = __bpf_try_make_writable(skb, skb->len);
2428         if (!ret) {
2429                 if (new_len > skb->len)
2430                         ret = bpf_skb_grow_rcsum(skb, new_len);
2431                 else if (new_len < skb->len)
2432                         ret = bpf_skb_trim_rcsum(skb, new_len);
2433                 if (!ret && skb_is_gso(skb))
2434                         skb_gso_reset(skb);
2435         }
2436
2437         bpf_compute_data_pointers(skb);
2438         return ret;
2439 }
2440
2441 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2442         .func           = bpf_skb_change_tail,
2443         .gpl_only       = false,
2444         .ret_type       = RET_INTEGER,
2445         .arg1_type      = ARG_PTR_TO_CTX,
2446         .arg2_type      = ARG_ANYTHING,
2447         .arg3_type      = ARG_ANYTHING,
2448 };
2449
2450 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
2451            u64, flags)
2452 {
2453         u32 max_len = __bpf_skb_max_len(skb);
2454         u32 new_len = skb->len + head_room;
2455         int ret;
2456
2457         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2458                      new_len < skb->len))
2459                 return -EINVAL;
2460
2461         ret = skb_cow(skb, head_room);
2462         if (likely(!ret)) {
2463                 /* Idea for this helper is that we currently only
2464                  * allow to expand on mac header. This means that
2465                  * skb->protocol network header, etc, stay as is.
2466                  * Compared to bpf_skb_change_tail(), we're more
2467                  * flexible due to not needing to linearize or
2468                  * reset GSO. Intention for this helper is to be
2469                  * used by an L3 skb that needs to push mac header
2470                  * for redirection into L2 device.
2471                  */
2472                 __skb_push(skb, head_room);
2473                 memset(skb->data, 0, head_room);
2474                 skb_reset_mac_header(skb);
2475         }
2476
2477         bpf_compute_data_pointers(skb);
2478         return 0;
2479 }
2480
2481 static const struct bpf_func_proto bpf_skb_change_head_proto = {
2482         .func           = bpf_skb_change_head,
2483         .gpl_only       = false,
2484         .ret_type       = RET_INTEGER,
2485         .arg1_type      = ARG_PTR_TO_CTX,
2486         .arg2_type      = ARG_ANYTHING,
2487         .arg3_type      = ARG_ANYTHING,
2488 };
2489
2490 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
2491 {
2492         return xdp_data_meta_unsupported(xdp) ? 0 :
2493                xdp->data - xdp->data_meta;
2494 }
2495
2496 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
2497 {
2498         unsigned long metalen = xdp_get_metalen(xdp);
2499         void *data_start = xdp->data_hard_start + metalen;
2500         void *data = xdp->data + offset;
2501
2502         if (unlikely(data < data_start ||
2503                      data > xdp->data_end - ETH_HLEN))
2504                 return -EINVAL;
2505
2506         if (metalen)
2507                 memmove(xdp->data_meta + offset,
2508                         xdp->data_meta, metalen);
2509         xdp->data_meta += offset;
2510         xdp->data = data;
2511
2512         return 0;
2513 }
2514
2515 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
2516         .func           = bpf_xdp_adjust_head,
2517         .gpl_only       = false,
2518         .ret_type       = RET_INTEGER,
2519         .arg1_type      = ARG_PTR_TO_CTX,
2520         .arg2_type      = ARG_ANYTHING,
2521 };
2522
2523 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
2524 {
2525         void *meta = xdp->data_meta + offset;
2526         unsigned long metalen = xdp->data - meta;
2527
2528         if (xdp_data_meta_unsupported(xdp))
2529                 return -ENOTSUPP;
2530         if (unlikely(meta < xdp->data_hard_start ||
2531                      meta > xdp->data))
2532                 return -EINVAL;
2533         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
2534                      (metalen > 32)))
2535                 return -EACCES;
2536
2537         xdp->data_meta = meta;
2538
2539         return 0;
2540 }
2541
2542 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
2543         .func           = bpf_xdp_adjust_meta,
2544         .gpl_only       = false,
2545         .ret_type       = RET_INTEGER,
2546         .arg1_type      = ARG_PTR_TO_CTX,
2547         .arg2_type      = ARG_ANYTHING,
2548 };
2549
2550 static int __bpf_tx_xdp(struct net_device *dev,
2551                         struct bpf_map *map,
2552                         struct xdp_buff *xdp,
2553                         u32 index)
2554 {
2555         int err;
2556
2557         if (!dev->netdev_ops->ndo_xdp_xmit) {
2558                 return -EOPNOTSUPP;
2559         }
2560
2561         err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2562         if (err)
2563                 return err;
2564         dev->netdev_ops->ndo_xdp_flush(dev);
2565         return 0;
2566 }
2567
2568 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
2569                             struct bpf_map *map,
2570                             struct xdp_buff *xdp,
2571                             u32 index)
2572 {
2573         int err;
2574
2575         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
2576                 struct net_device *dev = fwd;
2577
2578                 if (!dev->netdev_ops->ndo_xdp_xmit)
2579                         return -EOPNOTSUPP;
2580
2581                 err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2582                 if (err)
2583                         return err;
2584                 __dev_map_insert_ctx(map, index);
2585
2586         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
2587                 struct bpf_cpu_map_entry *rcpu = fwd;
2588
2589                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
2590                 if (err)
2591                         return err;
2592                 __cpu_map_insert_ctx(map, index);
2593         }
2594         return 0;
2595 }
2596
2597 void xdp_do_flush_map(void)
2598 {
2599         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2600         struct bpf_map *map = ri->map_to_flush;
2601
2602         ri->map_to_flush = NULL;
2603         if (map) {
2604                 switch (map->map_type) {
2605                 case BPF_MAP_TYPE_DEVMAP:
2606                         __dev_map_flush(map);
2607                         break;
2608                 case BPF_MAP_TYPE_CPUMAP:
2609                         __cpu_map_flush(map);
2610                         break;
2611                 default:
2612                         break;
2613                 }
2614         }
2615 }
2616 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
2617
2618 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
2619 {
2620         switch (map->map_type) {
2621         case BPF_MAP_TYPE_DEVMAP:
2622                 return __dev_map_lookup_elem(map, index);
2623         case BPF_MAP_TYPE_CPUMAP:
2624                 return __cpu_map_lookup_elem(map, index);
2625         default:
2626                 return NULL;
2627         }
2628 }
2629
2630 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
2631                                    unsigned long aux)
2632 {
2633         return (unsigned long)xdp_prog->aux != aux;
2634 }
2635
2636 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
2637                                struct bpf_prog *xdp_prog)
2638 {
2639         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2640         unsigned long map_owner = ri->map_owner;
2641         struct bpf_map *map = ri->map;
2642         u32 index = ri->ifindex;
2643         void *fwd = NULL;
2644         int err;
2645
2646         ri->ifindex = 0;
2647         ri->map = NULL;
2648         ri->map_owner = 0;
2649
2650         if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2651                 err = -EFAULT;
2652                 map = NULL;
2653                 goto err;
2654         }
2655
2656         fwd = __xdp_map_lookup_elem(map, index);
2657         if (!fwd) {
2658                 err = -EINVAL;
2659                 goto err;
2660         }
2661         if (ri->map_to_flush && ri->map_to_flush != map)
2662                 xdp_do_flush_map();
2663
2664         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
2665         if (unlikely(err))
2666                 goto err;
2667
2668         ri->map_to_flush = map;
2669         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2670         return 0;
2671 err:
2672         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2673         return err;
2674 }
2675
2676 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
2677                     struct bpf_prog *xdp_prog)
2678 {
2679         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2680         struct net_device *fwd;
2681         u32 index = ri->ifindex;
2682         int err;
2683
2684         if (ri->map)
2685                 return xdp_do_redirect_map(dev, xdp, xdp_prog);
2686
2687         fwd = dev_get_by_index_rcu(dev_net(dev), index);
2688         ri->ifindex = 0;
2689         if (unlikely(!fwd)) {
2690                 err = -EINVAL;
2691                 goto err;
2692         }
2693
2694         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
2695         if (unlikely(err))
2696                 goto err;
2697
2698         _trace_xdp_redirect(dev, xdp_prog, index);
2699         return 0;
2700 err:
2701         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2702         return err;
2703 }
2704 EXPORT_SYMBOL_GPL(xdp_do_redirect);
2705
2706 static int __xdp_generic_ok_fwd_dev(struct sk_buff *skb, struct net_device *fwd)
2707 {
2708         unsigned int len;
2709
2710         if (unlikely(!(fwd->flags & IFF_UP)))
2711                 return -ENETDOWN;
2712
2713         len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
2714         if (skb->len > len)
2715                 return -EMSGSIZE;
2716
2717         return 0;
2718 }
2719
2720 static int xdp_do_generic_redirect_map(struct net_device *dev,
2721                                        struct sk_buff *skb,
2722                                        struct bpf_prog *xdp_prog)
2723 {
2724         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2725         unsigned long map_owner = ri->map_owner;
2726         struct bpf_map *map = ri->map;
2727         struct net_device *fwd = NULL;
2728         u32 index = ri->ifindex;
2729         int err = 0;
2730
2731         ri->ifindex = 0;
2732         ri->map = NULL;
2733         ri->map_owner = 0;
2734
2735         if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2736                 err = -EFAULT;
2737                 map = NULL;
2738                 goto err;
2739         }
2740         fwd = __xdp_map_lookup_elem(map, index);
2741         if (unlikely(!fwd)) {
2742                 err = -EINVAL;
2743                 goto err;
2744         }
2745
2746         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
2747                 if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
2748                         goto err;
2749                 skb->dev = fwd;
2750         } else {
2751                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
2752                 err = -EBADRQC;
2753                 goto err;
2754         }
2755
2756         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2757         return 0;
2758 err:
2759         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2760         return err;
2761 }
2762
2763 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
2764                             struct bpf_prog *xdp_prog)
2765 {
2766         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2767         u32 index = ri->ifindex;
2768         struct net_device *fwd;
2769         int err = 0;
2770
2771         if (ri->map)
2772                 return xdp_do_generic_redirect_map(dev, skb, xdp_prog);
2773
2774         ri->ifindex = 0;
2775         fwd = dev_get_by_index_rcu(dev_net(dev), index);
2776         if (unlikely(!fwd)) {
2777                 err = -EINVAL;
2778                 goto err;
2779         }
2780
2781         if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
2782                 goto err;
2783
2784         skb->dev = fwd;
2785         _trace_xdp_redirect(dev, xdp_prog, index);
2786         return 0;
2787 err:
2788         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2789         return err;
2790 }
2791 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
2792
2793 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
2794 {
2795         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2796
2797         if (unlikely(flags))
2798                 return XDP_ABORTED;
2799
2800         ri->ifindex = ifindex;
2801         ri->flags = flags;
2802         ri->map = NULL;
2803         ri->map_owner = 0;
2804
2805         return XDP_REDIRECT;
2806 }
2807
2808 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
2809         .func           = bpf_xdp_redirect,
2810         .gpl_only       = false,
2811         .ret_type       = RET_INTEGER,
2812         .arg1_type      = ARG_ANYTHING,
2813         .arg2_type      = ARG_ANYTHING,
2814 };
2815
2816 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
2817            unsigned long, map_owner)
2818 {
2819         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2820
2821         if (unlikely(flags))
2822                 return XDP_ABORTED;
2823
2824         ri->ifindex = ifindex;
2825         ri->flags = flags;
2826         ri->map = map;
2827         ri->map_owner = map_owner;
2828
2829         return XDP_REDIRECT;
2830 }
2831
2832 /* Note, arg4 is hidden from users and populated by the verifier
2833  * with the right pointer.
2834  */
2835 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
2836         .func           = bpf_xdp_redirect_map,
2837         .gpl_only       = false,
2838         .ret_type       = RET_INTEGER,
2839         .arg1_type      = ARG_CONST_MAP_PTR,
2840         .arg2_type      = ARG_ANYTHING,
2841         .arg3_type      = ARG_ANYTHING,
2842 };
2843
2844 bool bpf_helper_changes_pkt_data(void *func)
2845 {
2846         if (func == bpf_skb_vlan_push ||
2847             func == bpf_skb_vlan_pop ||
2848             func == bpf_skb_store_bytes ||
2849             func == bpf_skb_change_proto ||
2850             func == bpf_skb_change_head ||
2851             func == bpf_skb_change_tail ||
2852             func == bpf_skb_adjust_room ||
2853             func == bpf_skb_pull_data ||
2854             func == bpf_clone_redirect ||
2855             func == bpf_l3_csum_replace ||
2856             func == bpf_l4_csum_replace ||
2857             func == bpf_xdp_adjust_head ||
2858             func == bpf_xdp_adjust_meta)
2859                 return true;
2860
2861         return false;
2862 }
2863
2864 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
2865                                   unsigned long off, unsigned long len)
2866 {
2867         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
2868
2869         if (unlikely(!ptr))
2870                 return len;
2871         if (ptr != dst_buff)
2872                 memcpy(dst_buff, ptr, len);
2873
2874         return 0;
2875 }
2876
2877 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
2878            u64, flags, void *, meta, u64, meta_size)
2879 {
2880         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2881
2882         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2883                 return -EINVAL;
2884         if (unlikely(skb_size > skb->len))
2885                 return -EFAULT;
2886
2887         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2888                                 bpf_skb_copy);
2889 }
2890
2891 static const struct bpf_func_proto bpf_skb_event_output_proto = {
2892         .func           = bpf_skb_event_output,
2893         .gpl_only       = true,
2894         .ret_type       = RET_INTEGER,
2895         .arg1_type      = ARG_PTR_TO_CTX,
2896         .arg2_type      = ARG_CONST_MAP_PTR,
2897         .arg3_type      = ARG_ANYTHING,
2898         .arg4_type      = ARG_PTR_TO_MEM,
2899         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
2900 };
2901
2902 static unsigned short bpf_tunnel_key_af(u64 flags)
2903 {
2904         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2905 }
2906
2907 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
2908            u32, size, u64, flags)
2909 {
2910         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2911         u8 compat[sizeof(struct bpf_tunnel_key)];
2912         void *to_orig = to;
2913         int err;
2914
2915         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2916                 err = -EINVAL;
2917                 goto err_clear;
2918         }
2919         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2920                 err = -EPROTO;
2921                 goto err_clear;
2922         }
2923         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2924                 err = -EINVAL;
2925                 switch (size) {
2926                 case offsetof(struct bpf_tunnel_key, tunnel_label):
2927                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2928                         goto set_compat;
2929                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2930                         /* Fixup deprecated structure layouts here, so we have
2931                          * a common path later on.
2932                          */
2933                         if (ip_tunnel_info_af(info) != AF_INET)
2934                                 goto err_clear;
2935 set_compat:
2936                         to = (struct bpf_tunnel_key *)compat;
2937                         break;
2938                 default:
2939                         goto err_clear;
2940                 }
2941         }
2942
2943         to->tunnel_id = be64_to_cpu(info->key.tun_id);
2944         to->tunnel_tos = info->key.tos;
2945         to->tunnel_ttl = info->key.ttl;
2946
2947         if (flags & BPF_F_TUNINFO_IPV6) {
2948                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2949                        sizeof(to->remote_ipv6));
2950                 to->tunnel_label = be32_to_cpu(info->key.label);
2951         } else {
2952                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2953         }
2954
2955         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2956                 memcpy(to_orig, to, size);
2957
2958         return 0;
2959 err_clear:
2960         memset(to_orig, 0, size);
2961         return err;
2962 }
2963
2964 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2965         .func           = bpf_skb_get_tunnel_key,
2966         .gpl_only       = false,
2967         .ret_type       = RET_INTEGER,
2968         .arg1_type      = ARG_PTR_TO_CTX,
2969         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
2970         .arg3_type      = ARG_CONST_SIZE,
2971         .arg4_type      = ARG_ANYTHING,
2972 };
2973
2974 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
2975 {
2976         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2977         int err;
2978
2979         if (unlikely(!info ||
2980                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2981                 err = -ENOENT;
2982                 goto err_clear;
2983         }
2984         if (unlikely(size < info->options_len)) {
2985                 err = -ENOMEM;
2986                 goto err_clear;
2987         }
2988
2989         ip_tunnel_info_opts_get(to, info);
2990         if (size > info->options_len)
2991                 memset(to + info->options_len, 0, size - info->options_len);
2992
2993         return info->options_len;
2994 err_clear:
2995         memset(to, 0, size);
2996         return err;
2997 }
2998
2999 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3000         .func           = bpf_skb_get_tunnel_opt,
3001         .gpl_only       = false,
3002         .ret_type       = RET_INTEGER,
3003         .arg1_type      = ARG_PTR_TO_CTX,
3004         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3005         .arg3_type      = ARG_CONST_SIZE,
3006 };
3007
3008 static struct metadata_dst __percpu *md_dst;
3009
3010 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3011            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3012 {
3013         struct metadata_dst *md = this_cpu_ptr(md_dst);
3014         u8 compat[sizeof(struct bpf_tunnel_key)];
3015         struct ip_tunnel_info *info;
3016
3017         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3018                                BPF_F_DONT_FRAGMENT)))
3019                 return -EINVAL;
3020         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3021                 switch (size) {
3022                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3023                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3024                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3025                         /* Fixup deprecated structure layouts here, so we have
3026                          * a common path later on.
3027                          */
3028                         memcpy(compat, from, size);
3029                         memset(compat + size, 0, sizeof(compat) - size);
3030                         from = (const struct bpf_tunnel_key *) compat;
3031                         break;
3032                 default:
3033                         return -EINVAL;
3034                 }
3035         }
3036         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3037                      from->tunnel_ext))
3038                 return -EINVAL;
3039
3040         skb_dst_drop(skb);
3041         dst_hold((struct dst_entry *) md);
3042         skb_dst_set(skb, (struct dst_entry *) md);
3043
3044         info = &md->u.tun_info;
3045         info->mode = IP_TUNNEL_INFO_TX;
3046
3047         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3048         if (flags & BPF_F_DONT_FRAGMENT)
3049                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3050         if (flags & BPF_F_ZERO_CSUM_TX)
3051                 info->key.tun_flags &= ~TUNNEL_CSUM;
3052
3053         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3054         info->key.tos = from->tunnel_tos;
3055         info->key.ttl = from->tunnel_ttl;
3056
3057         if (flags & BPF_F_TUNINFO_IPV6) {
3058                 info->mode |= IP_TUNNEL_INFO_IPV6;
3059                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3060                        sizeof(from->remote_ipv6));
3061                 info->key.label = cpu_to_be32(from->tunnel_label) &
3062                                   IPV6_FLOWLABEL_MASK;
3063         } else {
3064                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3065         }
3066
3067         return 0;
3068 }
3069
3070 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3071         .func           = bpf_skb_set_tunnel_key,
3072         .gpl_only       = false,
3073         .ret_type       = RET_INTEGER,
3074         .arg1_type      = ARG_PTR_TO_CTX,
3075         .arg2_type      = ARG_PTR_TO_MEM,
3076         .arg3_type      = ARG_CONST_SIZE,
3077         .arg4_type      = ARG_ANYTHING,
3078 };
3079
3080 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3081            const u8 *, from, u32, size)
3082 {
3083         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3084         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3085
3086         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3087                 return -EINVAL;
3088         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3089                 return -ENOMEM;
3090
3091         ip_tunnel_info_opts_set(info, from, size);
3092
3093         return 0;
3094 }
3095
3096 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3097         .func           = bpf_skb_set_tunnel_opt,
3098         .gpl_only       = false,
3099         .ret_type       = RET_INTEGER,
3100         .arg1_type      = ARG_PTR_TO_CTX,
3101         .arg2_type      = ARG_PTR_TO_MEM,
3102         .arg3_type      = ARG_CONST_SIZE,
3103 };
3104
3105 static const struct bpf_func_proto *
3106 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3107 {
3108         if (!md_dst) {
3109                 struct metadata_dst __percpu *tmp;
3110
3111                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3112                                                 METADATA_IP_TUNNEL,
3113                                                 GFP_KERNEL);
3114                 if (!tmp)
3115                         return NULL;
3116                 if (cmpxchg(&md_dst, NULL, tmp))
3117                         metadata_dst_free_percpu(tmp);
3118         }
3119
3120         switch (which) {
3121         case BPF_FUNC_skb_set_tunnel_key:
3122                 return &bpf_skb_set_tunnel_key_proto;
3123         case BPF_FUNC_skb_set_tunnel_opt:
3124                 return &bpf_skb_set_tunnel_opt_proto;
3125         default:
3126                 return NULL;
3127         }
3128 }
3129
3130 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3131            u32, idx)
3132 {
3133         struct bpf_array *array = container_of(map, struct bpf_array, map);
3134         struct cgroup *cgrp;
3135         struct sock *sk;
3136
3137         sk = skb_to_full_sk(skb);
3138         if (!sk || !sk_fullsock(sk))
3139                 return -ENOENT;
3140         if (unlikely(idx >= array->map.max_entries))
3141                 return -E2BIG;
3142
3143         cgrp = READ_ONCE(array->ptrs[idx]);
3144         if (unlikely(!cgrp))
3145                 return -EAGAIN;
3146
3147         return sk_under_cgroup_hierarchy(sk, cgrp);
3148 }
3149
3150 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3151         .func           = bpf_skb_under_cgroup,
3152         .gpl_only       = false,
3153         .ret_type       = RET_INTEGER,
3154         .arg1_type      = ARG_PTR_TO_CTX,
3155         .arg2_type      = ARG_CONST_MAP_PTR,
3156         .arg3_type      = ARG_ANYTHING,
3157 };
3158
3159 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3160                                   unsigned long off, unsigned long len)
3161 {
3162         memcpy(dst_buff, src_buff + off, len);
3163         return 0;
3164 }
3165
3166 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3167            u64, flags, void *, meta, u64, meta_size)
3168 {
3169         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3170
3171         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3172                 return -EINVAL;
3173         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3174                 return -EFAULT;
3175
3176         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3177                                 xdp_size, bpf_xdp_copy);
3178 }
3179
3180 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3181         .func           = bpf_xdp_event_output,
3182         .gpl_only       = true,
3183         .ret_type       = RET_INTEGER,
3184         .arg1_type      = ARG_PTR_TO_CTX,
3185         .arg2_type      = ARG_CONST_MAP_PTR,
3186         .arg3_type      = ARG_ANYTHING,
3187         .arg4_type      = ARG_PTR_TO_MEM,
3188         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3189 };
3190
3191 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3192 {
3193         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3194 }
3195
3196 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3197         .func           = bpf_get_socket_cookie,
3198         .gpl_only       = false,
3199         .ret_type       = RET_INTEGER,
3200         .arg1_type      = ARG_PTR_TO_CTX,
3201 };
3202
3203 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3204 {
3205         struct sock *sk = sk_to_full_sk(skb->sk);
3206         kuid_t kuid;
3207
3208         if (!sk || !sk_fullsock(sk))
3209                 return overflowuid;
3210         kuid = sock_net_uid(sock_net(sk), sk);
3211         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3212 }
3213
3214 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3215         .func           = bpf_get_socket_uid,
3216         .gpl_only       = false,
3217         .ret_type       = RET_INTEGER,
3218         .arg1_type      = ARG_PTR_TO_CTX,
3219 };
3220
3221 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3222            int, level, int, optname, char *, optval, int, optlen)
3223 {
3224         struct sock *sk = bpf_sock->sk;
3225         int ret = 0;
3226         int val;
3227
3228         if (!sk_fullsock(sk))
3229                 return -EINVAL;
3230
3231         if (level == SOL_SOCKET) {
3232                 if (optlen != sizeof(int))
3233                         return -EINVAL;
3234                 val = *((int *)optval);
3235
3236                 /* Only some socketops are supported */
3237                 switch (optname) {
3238                 case SO_RCVBUF:
3239                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3240                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3241                         break;
3242                 case SO_SNDBUF:
3243                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3244                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3245                         break;
3246                 case SO_MAX_PACING_RATE:
3247                         sk->sk_max_pacing_rate = val;
3248                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3249                                                  sk->sk_max_pacing_rate);
3250                         break;
3251                 case SO_PRIORITY:
3252                         sk->sk_priority = val;
3253                         break;
3254                 case SO_RCVLOWAT:
3255                         if (val < 0)
3256                                 val = INT_MAX;
3257                         sk->sk_rcvlowat = val ? : 1;
3258                         break;
3259                 case SO_MARK:
3260                         sk->sk_mark = val;
3261                         break;
3262                 default:
3263                         ret = -EINVAL;
3264                 }
3265 #ifdef CONFIG_INET
3266 #if IS_ENABLED(CONFIG_IPV6)
3267         } else if (level == SOL_IPV6) {
3268                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3269                         return -EINVAL;
3270
3271                 val = *((int *)optval);
3272                 /* Only some options are supported */
3273                 switch (optname) {
3274                 case IPV6_TCLASS:
3275                         if (val < -1 || val > 0xff) {
3276                                 ret = -EINVAL;
3277                         } else {
3278                                 struct ipv6_pinfo *np = inet6_sk(sk);
3279
3280                                 if (val == -1)
3281                                         val = 0;
3282                                 np->tclass = val;
3283                         }
3284                         break;
3285                 default:
3286                         ret = -EINVAL;
3287                 }
3288 #endif
3289         } else if (level == SOL_TCP &&
3290                    sk->sk_prot->setsockopt == tcp_setsockopt) {
3291                 if (optname == TCP_CONGESTION) {
3292                         char name[TCP_CA_NAME_MAX];
3293                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3294
3295                         strncpy(name, optval, min_t(long, optlen,
3296                                                     TCP_CA_NAME_MAX-1));
3297                         name[TCP_CA_NAME_MAX-1] = 0;
3298                         ret = tcp_set_congestion_control(sk, name, false,
3299                                                          reinit);
3300                 } else {
3301                         struct tcp_sock *tp = tcp_sk(sk);
3302
3303                         if (optlen != sizeof(int))
3304                                 return -EINVAL;
3305
3306                         val = *((int *)optval);
3307                         /* Only some options are supported */
3308                         switch (optname) {
3309                         case TCP_BPF_IW:
3310                                 if (val <= 0 || tp->data_segs_out > 0)
3311                                         ret = -EINVAL;
3312                                 else
3313                                         tp->snd_cwnd = val;
3314                                 break;
3315                         case TCP_BPF_SNDCWND_CLAMP:
3316                                 if (val <= 0) {
3317                                         ret = -EINVAL;
3318                                 } else {
3319                                         tp->snd_cwnd_clamp = val;
3320                                         tp->snd_ssthresh = val;
3321                                 }
3322                                 break;
3323                         default:
3324                                 ret = -EINVAL;
3325                         }
3326                 }
3327 #endif
3328         } else {
3329                 ret = -EINVAL;
3330         }
3331         return ret;
3332 }
3333
3334 static const struct bpf_func_proto bpf_setsockopt_proto = {
3335         .func           = bpf_setsockopt,
3336         .gpl_only       = false,
3337         .ret_type       = RET_INTEGER,
3338         .arg1_type      = ARG_PTR_TO_CTX,
3339         .arg2_type      = ARG_ANYTHING,
3340         .arg3_type      = ARG_ANYTHING,
3341         .arg4_type      = ARG_PTR_TO_MEM,
3342         .arg5_type      = ARG_CONST_SIZE,
3343 };
3344
3345 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3346            int, level, int, optname, char *, optval, int, optlen)
3347 {
3348         struct sock *sk = bpf_sock->sk;
3349
3350         if (!sk_fullsock(sk))
3351                 goto err_clear;
3352
3353 #ifdef CONFIG_INET
3354         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
3355                 if (optname == TCP_CONGESTION) {
3356                         struct inet_connection_sock *icsk = inet_csk(sk);
3357
3358                         if (!icsk->icsk_ca_ops || optlen <= 1)
3359                                 goto err_clear;
3360                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
3361                         optval[optlen - 1] = 0;
3362                 } else {
3363                         goto err_clear;
3364                 }
3365 #if IS_ENABLED(CONFIG_IPV6)
3366         } else if (level == SOL_IPV6) {
3367                 struct ipv6_pinfo *np = inet6_sk(sk);
3368
3369                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3370                         goto err_clear;
3371
3372                 /* Only some options are supported */
3373                 switch (optname) {
3374                 case IPV6_TCLASS:
3375                         *((int *)optval) = (int)np->tclass;
3376                         break;
3377                 default:
3378                         goto err_clear;
3379                 }
3380 #endif
3381         } else {
3382                 goto err_clear;
3383         }
3384         return 0;
3385 #endif
3386 err_clear:
3387         memset(optval, 0, optlen);
3388         return -EINVAL;
3389 }
3390
3391 static const struct bpf_func_proto bpf_getsockopt_proto = {
3392         .func           = bpf_getsockopt,
3393         .gpl_only       = false,
3394         .ret_type       = RET_INTEGER,
3395         .arg1_type      = ARG_PTR_TO_CTX,
3396         .arg2_type      = ARG_ANYTHING,
3397         .arg3_type      = ARG_ANYTHING,
3398         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
3399         .arg5_type      = ARG_CONST_SIZE,
3400 };
3401
3402 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
3403            int, argval)
3404 {
3405         struct sock *sk = bpf_sock->sk;
3406         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
3407
3408         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
3409                 return -EINVAL;
3410
3411         if (val)
3412                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
3413
3414         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
3415 }
3416
3417 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
3418         .func           = bpf_sock_ops_cb_flags_set,
3419         .gpl_only       = false,
3420         .ret_type       = RET_INTEGER,
3421         .arg1_type      = ARG_PTR_TO_CTX,
3422         .arg2_type      = ARG_ANYTHING,
3423 };
3424
3425 static const struct bpf_func_proto *
3426 bpf_base_func_proto(enum bpf_func_id func_id)
3427 {
3428         switch (func_id) {
3429         case BPF_FUNC_map_lookup_elem:
3430                 return &bpf_map_lookup_elem_proto;
3431         case BPF_FUNC_map_update_elem:
3432                 return &bpf_map_update_elem_proto;
3433         case BPF_FUNC_map_delete_elem:
3434                 return &bpf_map_delete_elem_proto;
3435         case BPF_FUNC_get_prandom_u32:
3436                 return &bpf_get_prandom_u32_proto;
3437         case BPF_FUNC_get_smp_processor_id:
3438                 return &bpf_get_raw_smp_processor_id_proto;
3439         case BPF_FUNC_get_numa_node_id:
3440                 return &bpf_get_numa_node_id_proto;
3441         case BPF_FUNC_tail_call:
3442                 return &bpf_tail_call_proto;
3443         case BPF_FUNC_ktime_get_ns:
3444                 return &bpf_ktime_get_ns_proto;
3445         case BPF_FUNC_trace_printk:
3446                 if (capable(CAP_SYS_ADMIN))
3447                         return bpf_get_trace_printk_proto();
3448         default:
3449                 return NULL;
3450         }
3451 }
3452
3453 static const struct bpf_func_proto *
3454 sock_filter_func_proto(enum bpf_func_id func_id)
3455 {
3456         switch (func_id) {
3457         /* inet and inet6 sockets are created in a process
3458          * context so there is always a valid uid/gid
3459          */
3460         case BPF_FUNC_get_current_uid_gid:
3461                 return &bpf_get_current_uid_gid_proto;
3462         default:
3463                 return bpf_base_func_proto(func_id);
3464         }
3465 }
3466
3467 static const struct bpf_func_proto *
3468 sk_filter_func_proto(enum bpf_func_id func_id)
3469 {
3470         switch (func_id) {
3471         case BPF_FUNC_skb_load_bytes:
3472                 return &bpf_skb_load_bytes_proto;
3473         case BPF_FUNC_get_socket_cookie:
3474                 return &bpf_get_socket_cookie_proto;
3475         case BPF_FUNC_get_socket_uid:
3476                 return &bpf_get_socket_uid_proto;
3477         default:
3478                 return bpf_base_func_proto(func_id);
3479         }
3480 }
3481
3482 static const struct bpf_func_proto *
3483 tc_cls_act_func_proto(enum bpf_func_id func_id)
3484 {
3485         switch (func_id) {
3486         case BPF_FUNC_skb_store_bytes:
3487                 return &bpf_skb_store_bytes_proto;
3488         case BPF_FUNC_skb_load_bytes:
3489                 return &bpf_skb_load_bytes_proto;
3490         case BPF_FUNC_skb_pull_data:
3491                 return &bpf_skb_pull_data_proto;
3492         case BPF_FUNC_csum_diff:
3493                 return &bpf_csum_diff_proto;
3494         case BPF_FUNC_csum_update:
3495                 return &bpf_csum_update_proto;
3496         case BPF_FUNC_l3_csum_replace:
3497                 return &bpf_l3_csum_replace_proto;
3498         case BPF_FUNC_l4_csum_replace:
3499                 return &bpf_l4_csum_replace_proto;
3500         case BPF_FUNC_clone_redirect:
3501                 return &bpf_clone_redirect_proto;
3502         case BPF_FUNC_get_cgroup_classid:
3503                 return &bpf_get_cgroup_classid_proto;
3504         case BPF_FUNC_skb_vlan_push:
3505                 return &bpf_skb_vlan_push_proto;
3506         case BPF_FUNC_skb_vlan_pop:
3507                 return &bpf_skb_vlan_pop_proto;
3508         case BPF_FUNC_skb_change_proto:
3509                 return &bpf_skb_change_proto_proto;
3510         case BPF_FUNC_skb_change_type:
3511                 return &bpf_skb_change_type_proto;
3512         case BPF_FUNC_skb_adjust_room:
3513                 return &bpf_skb_adjust_room_proto;
3514         case BPF_FUNC_skb_change_tail:
3515                 return &bpf_skb_change_tail_proto;
3516         case BPF_FUNC_skb_get_tunnel_key:
3517                 return &bpf_skb_get_tunnel_key_proto;
3518         case BPF_FUNC_skb_set_tunnel_key:
3519                 return bpf_get_skb_set_tunnel_proto(func_id);
3520         case BPF_FUNC_skb_get_tunnel_opt:
3521                 return &bpf_skb_get_tunnel_opt_proto;
3522         case BPF_FUNC_skb_set_tunnel_opt:
3523                 return bpf_get_skb_set_tunnel_proto(func_id);
3524         case BPF_FUNC_redirect:
3525                 return &bpf_redirect_proto;
3526         case BPF_FUNC_get_route_realm:
3527                 return &bpf_get_route_realm_proto;
3528         case BPF_FUNC_get_hash_recalc:
3529                 return &bpf_get_hash_recalc_proto;
3530         case BPF_FUNC_set_hash_invalid:
3531                 return &bpf_set_hash_invalid_proto;
3532         case BPF_FUNC_set_hash:
3533                 return &bpf_set_hash_proto;
3534         case BPF_FUNC_perf_event_output:
3535                 return &bpf_skb_event_output_proto;
3536         case BPF_FUNC_get_smp_processor_id:
3537                 return &bpf_get_smp_processor_id_proto;
3538         case BPF_FUNC_skb_under_cgroup:
3539                 return &bpf_skb_under_cgroup_proto;
3540         case BPF_FUNC_get_socket_cookie:
3541                 return &bpf_get_socket_cookie_proto;
3542         case BPF_FUNC_get_socket_uid:
3543                 return &bpf_get_socket_uid_proto;
3544         default:
3545                 return bpf_base_func_proto(func_id);
3546         }
3547 }
3548
3549 static const struct bpf_func_proto *
3550 xdp_func_proto(enum bpf_func_id func_id)
3551 {
3552         switch (func_id) {
3553         case BPF_FUNC_perf_event_output:
3554                 return &bpf_xdp_event_output_proto;
3555         case BPF_FUNC_get_smp_processor_id:
3556                 return &bpf_get_smp_processor_id_proto;
3557         case BPF_FUNC_csum_diff:
3558                 return &bpf_csum_diff_proto;
3559         case BPF_FUNC_xdp_adjust_head:
3560                 return &bpf_xdp_adjust_head_proto;
3561         case BPF_FUNC_xdp_adjust_meta:
3562                 return &bpf_xdp_adjust_meta_proto;
3563         case BPF_FUNC_redirect:
3564                 return &bpf_xdp_redirect_proto;
3565         case BPF_FUNC_redirect_map:
3566                 return &bpf_xdp_redirect_map_proto;
3567         default:
3568                 return bpf_base_func_proto(func_id);
3569         }
3570 }
3571
3572 static const struct bpf_func_proto *
3573 lwt_inout_func_proto(enum bpf_func_id func_id)
3574 {
3575         switch (func_id) {
3576         case BPF_FUNC_skb_load_bytes:
3577                 return &bpf_skb_load_bytes_proto;
3578         case BPF_FUNC_skb_pull_data:
3579                 return &bpf_skb_pull_data_proto;
3580         case BPF_FUNC_csum_diff:
3581                 return &bpf_csum_diff_proto;
3582         case BPF_FUNC_get_cgroup_classid:
3583                 return &bpf_get_cgroup_classid_proto;
3584         case BPF_FUNC_get_route_realm:
3585                 return &bpf_get_route_realm_proto;
3586         case BPF_FUNC_get_hash_recalc:
3587                 return &bpf_get_hash_recalc_proto;
3588         case BPF_FUNC_perf_event_output:
3589                 return &bpf_skb_event_output_proto;
3590         case BPF_FUNC_get_smp_processor_id:
3591                 return &bpf_get_smp_processor_id_proto;
3592         case BPF_FUNC_skb_under_cgroup:
3593                 return &bpf_skb_under_cgroup_proto;
3594         default:
3595                 return bpf_base_func_proto(func_id);
3596         }
3597 }
3598
3599 static const struct bpf_func_proto *
3600         sock_ops_func_proto(enum bpf_func_id func_id)
3601 {
3602         switch (func_id) {
3603         case BPF_FUNC_setsockopt:
3604                 return &bpf_setsockopt_proto;
3605         case BPF_FUNC_getsockopt:
3606                 return &bpf_getsockopt_proto;
3607         case BPF_FUNC_sock_ops_cb_flags_set:
3608                 return &bpf_sock_ops_cb_flags_set_proto;
3609         case BPF_FUNC_sock_map_update:
3610                 return &bpf_sock_map_update_proto;
3611         default:
3612                 return bpf_base_func_proto(func_id);
3613         }
3614 }
3615
3616 static const struct bpf_func_proto *sk_skb_func_proto(enum bpf_func_id func_id)
3617 {
3618         switch (func_id) {
3619         case BPF_FUNC_skb_store_bytes:
3620                 return &bpf_skb_store_bytes_proto;
3621         case BPF_FUNC_skb_load_bytes:
3622                 return &bpf_skb_load_bytes_proto;
3623         case BPF_FUNC_skb_pull_data:
3624                 return &bpf_skb_pull_data_proto;
3625         case BPF_FUNC_skb_change_tail:
3626                 return &bpf_skb_change_tail_proto;
3627         case BPF_FUNC_skb_change_head:
3628                 return &bpf_skb_change_head_proto;
3629         case BPF_FUNC_get_socket_cookie:
3630                 return &bpf_get_socket_cookie_proto;
3631         case BPF_FUNC_get_socket_uid:
3632                 return &bpf_get_socket_uid_proto;
3633         case BPF_FUNC_sk_redirect_map:
3634                 return &bpf_sk_redirect_map_proto;
3635         default:
3636                 return bpf_base_func_proto(func_id);
3637         }
3638 }
3639
3640 static const struct bpf_func_proto *
3641 lwt_xmit_func_proto(enum bpf_func_id func_id)
3642 {
3643         switch (func_id) {
3644         case BPF_FUNC_skb_get_tunnel_key:
3645                 return &bpf_skb_get_tunnel_key_proto;
3646         case BPF_FUNC_skb_set_tunnel_key:
3647                 return bpf_get_skb_set_tunnel_proto(func_id);
3648         case BPF_FUNC_skb_get_tunnel_opt:
3649                 return &bpf_skb_get_tunnel_opt_proto;
3650         case BPF_FUNC_skb_set_tunnel_opt:
3651                 return bpf_get_skb_set_tunnel_proto(func_id);
3652         case BPF_FUNC_redirect:
3653                 return &bpf_redirect_proto;
3654         case BPF_FUNC_clone_redirect:
3655                 return &bpf_clone_redirect_proto;
3656         case BPF_FUNC_skb_change_tail:
3657                 return &bpf_skb_change_tail_proto;
3658         case BPF_FUNC_skb_change_head:
3659                 return &bpf_skb_change_head_proto;
3660         case BPF_FUNC_skb_store_bytes:
3661                 return &bpf_skb_store_bytes_proto;
3662         case BPF_FUNC_csum_update:
3663                 return &bpf_csum_update_proto;
3664         case BPF_FUNC_l3_csum_replace:
3665                 return &bpf_l3_csum_replace_proto;
3666         case BPF_FUNC_l4_csum_replace:
3667                 return &bpf_l4_csum_replace_proto;
3668         case BPF_FUNC_set_hash_invalid:
3669                 return &bpf_set_hash_invalid_proto;
3670         default:
3671                 return lwt_inout_func_proto(func_id);
3672         }
3673 }
3674
3675 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
3676                                     struct bpf_insn_access_aux *info)
3677 {
3678         const int size_default = sizeof(__u32);
3679
3680         if (off < 0 || off >= sizeof(struct __sk_buff))
3681                 return false;
3682
3683         /* The verifier guarantees that size > 0. */
3684         if (off % size != 0)
3685                 return false;
3686
3687         switch (off) {
3688         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3689                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
3690                         return false;
3691                 break;
3692         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
3693         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
3694         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
3695         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
3696         case bpf_ctx_range(struct __sk_buff, data):
3697         case bpf_ctx_range(struct __sk_buff, data_meta):
3698         case bpf_ctx_range(struct __sk_buff, data_end):
3699                 if (size != size_default)
3700                         return false;
3701                 break;
3702         default:
3703                 /* Only narrow read access allowed for now. */
3704                 if (type == BPF_WRITE) {
3705                         if (size != size_default)
3706                                 return false;
3707                 } else {
3708                         bpf_ctx_record_field_size(info, size_default);
3709                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
3710                                 return false;
3711                 }
3712         }
3713
3714         return true;
3715 }
3716
3717 static bool sk_filter_is_valid_access(int off, int size,
3718                                       enum bpf_access_type type,
3719                                       struct bpf_insn_access_aux *info)
3720 {
3721         switch (off) {
3722         case bpf_ctx_range(struct __sk_buff, tc_classid):
3723         case bpf_ctx_range(struct __sk_buff, data):
3724         case bpf_ctx_range(struct __sk_buff, data_meta):
3725         case bpf_ctx_range(struct __sk_buff, data_end):
3726         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3727                 return false;
3728         }
3729
3730         if (type == BPF_WRITE) {
3731                 switch (off) {
3732                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3733                         break;
3734                 default:
3735                         return false;
3736                 }
3737         }
3738
3739         return bpf_skb_is_valid_access(off, size, type, info);
3740 }
3741
3742 static bool lwt_is_valid_access(int off, int size,
3743                                 enum bpf_access_type type,
3744                                 struct bpf_insn_access_aux *info)
3745 {
3746         switch (off) {
3747         case bpf_ctx_range(struct __sk_buff, tc_classid):
3748         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3749         case bpf_ctx_range(struct __sk_buff, data_meta):
3750                 return false;
3751         }
3752
3753         if (type == BPF_WRITE) {
3754                 switch (off) {
3755                 case bpf_ctx_range(struct __sk_buff, mark):
3756                 case bpf_ctx_range(struct __sk_buff, priority):
3757                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3758                         break;
3759                 default:
3760                         return false;
3761                 }
3762         }
3763
3764         switch (off) {
3765         case bpf_ctx_range(struct __sk_buff, data):
3766                 info->reg_type = PTR_TO_PACKET;
3767                 break;
3768         case bpf_ctx_range(struct __sk_buff, data_end):
3769                 info->reg_type = PTR_TO_PACKET_END;
3770                 break;
3771         }
3772
3773         return bpf_skb_is_valid_access(off, size, type, info);
3774 }
3775
3776 static bool sock_filter_is_valid_access(int off, int size,
3777                                         enum bpf_access_type type,
3778                                         struct bpf_insn_access_aux *info)
3779 {
3780         if (type == BPF_WRITE) {
3781                 switch (off) {
3782                 case offsetof(struct bpf_sock, bound_dev_if):
3783                 case offsetof(struct bpf_sock, mark):
3784                 case offsetof(struct bpf_sock, priority):
3785                         break;
3786                 default:
3787                         return false;
3788                 }
3789         }
3790
3791         if (off < 0 || off + size > sizeof(struct bpf_sock))
3792                 return false;
3793         /* The verifier guarantees that size > 0. */
3794         if (off % size != 0)
3795                 return false;
3796         if (size != sizeof(__u32))
3797                 return false;
3798
3799         return true;
3800 }
3801
3802 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
3803                                 const struct bpf_prog *prog, int drop_verdict)
3804 {
3805         struct bpf_insn *insn = insn_buf;
3806
3807         if (!direct_write)
3808                 return 0;
3809
3810         /* if (!skb->cloned)
3811          *       goto start;
3812          *
3813          * (Fast-path, otherwise approximation that we might be
3814          *  a clone, do the rest in helper.)
3815          */
3816         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
3817         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
3818         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
3819
3820         /* ret = bpf_skb_pull_data(skb, 0); */
3821         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
3822         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
3823         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
3824                                BPF_FUNC_skb_pull_data);
3825         /* if (!ret)
3826          *      goto restore;
3827          * return TC_ACT_SHOT;
3828          */
3829         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
3830         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
3831         *insn++ = BPF_EXIT_INSN();
3832
3833         /* restore: */
3834         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
3835         /* start: */
3836         *insn++ = prog->insnsi[0];
3837
3838         return insn - insn_buf;
3839 }
3840
3841 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
3842                                const struct bpf_prog *prog)
3843 {
3844         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
3845 }
3846
3847 static bool tc_cls_act_is_valid_access(int off, int size,
3848                                        enum bpf_access_type type,
3849                                        struct bpf_insn_access_aux *info)
3850 {
3851         if (type == BPF_WRITE) {
3852                 switch (off) {
3853                 case bpf_ctx_range(struct __sk_buff, mark):
3854                 case bpf_ctx_range(struct __sk_buff, tc_index):
3855                 case bpf_ctx_range(struct __sk_buff, priority):
3856                 case bpf_ctx_range(struct __sk_buff, tc_classid):
3857                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3858                         break;
3859                 default:
3860                         return false;
3861                 }
3862         }
3863
3864         switch (off) {
3865         case bpf_ctx_range(struct __sk_buff, data):
3866                 info->reg_type = PTR_TO_PACKET;
3867                 break;
3868         case bpf_ctx_range(struct __sk_buff, data_meta):
3869                 info->reg_type = PTR_TO_PACKET_META;
3870                 break;
3871         case bpf_ctx_range(struct __sk_buff, data_end):
3872                 info->reg_type = PTR_TO_PACKET_END;
3873                 break;
3874         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3875                 return false;
3876         }
3877
3878         return bpf_skb_is_valid_access(off, size, type, info);
3879 }
3880
3881 static bool __is_valid_xdp_access(int off, int size)
3882 {
3883         if (off < 0 || off >= sizeof(struct xdp_md))
3884                 return false;
3885         if (off % size != 0)
3886                 return false;
3887         if (size != sizeof(__u32))
3888                 return false;
3889
3890         return true;
3891 }
3892
3893 static bool xdp_is_valid_access(int off, int size,
3894                                 enum bpf_access_type type,
3895                                 struct bpf_insn_access_aux *info)
3896 {
3897         if (type == BPF_WRITE)
3898                 return false;
3899
3900         switch (off) {
3901         case offsetof(struct xdp_md, data):
3902                 info->reg_type = PTR_TO_PACKET;
3903                 break;
3904         case offsetof(struct xdp_md, data_meta):
3905                 info->reg_type = PTR_TO_PACKET_META;
3906                 break;
3907         case offsetof(struct xdp_md, data_end):
3908                 info->reg_type = PTR_TO_PACKET_END;
3909                 break;
3910         }
3911
3912         return __is_valid_xdp_access(off, size);
3913 }
3914
3915 void bpf_warn_invalid_xdp_action(u32 act)
3916 {
3917         const u32 act_max = XDP_REDIRECT;
3918
3919         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
3920                   act > act_max ? "Illegal" : "Driver unsupported",
3921                   act);
3922 }
3923 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
3924
3925 static bool sock_ops_is_valid_access(int off, int size,
3926                                      enum bpf_access_type type,
3927                                      struct bpf_insn_access_aux *info)
3928 {
3929         const int size_default = sizeof(__u32);
3930
3931         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
3932                 return false;
3933
3934         /* The verifier guarantees that size > 0. */
3935         if (off % size != 0)
3936                 return false;
3937
3938         if (type == BPF_WRITE) {
3939                 switch (off) {
3940                 case offsetof(struct bpf_sock_ops, reply):
3941                 case offsetof(struct bpf_sock_ops, sk_txhash):
3942                         if (size != size_default)
3943                                 return false;
3944                         break;
3945                 default:
3946                         return false;
3947                 }
3948         } else {
3949                 switch (off) {
3950                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
3951                                         bytes_acked):
3952                         if (size != sizeof(__u64))
3953                                 return false;
3954                         break;
3955                 default:
3956                         if (size != size_default)
3957                                 return false;
3958                         break;
3959                 }
3960         }
3961
3962         return true;
3963 }
3964
3965 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
3966                            const struct bpf_prog *prog)
3967 {
3968         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
3969 }
3970
3971 static bool sk_skb_is_valid_access(int off, int size,
3972                                    enum bpf_access_type type,
3973                                    struct bpf_insn_access_aux *info)
3974 {
3975         switch (off) {
3976         case bpf_ctx_range(struct __sk_buff, tc_classid):
3977         case bpf_ctx_range(struct __sk_buff, data_meta):
3978                 return false;
3979         }
3980
3981         if (type == BPF_WRITE) {
3982                 switch (off) {
3983                 case bpf_ctx_range(struct __sk_buff, tc_index):
3984                 case bpf_ctx_range(struct __sk_buff, priority):
3985                         break;
3986                 default:
3987                         return false;
3988                 }
3989         }
3990
3991         switch (off) {
3992         case bpf_ctx_range(struct __sk_buff, mark):
3993                 return false;
3994         case bpf_ctx_range(struct __sk_buff, data):
3995                 info->reg_type = PTR_TO_PACKET;
3996                 break;
3997         case bpf_ctx_range(struct __sk_buff, data_end):
3998                 info->reg_type = PTR_TO_PACKET_END;
3999                 break;
4000         }
4001
4002         return bpf_skb_is_valid_access(off, size, type, info);
4003 }
4004
4005 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
4006                                   const struct bpf_insn *si,
4007                                   struct bpf_insn *insn_buf,
4008                                   struct bpf_prog *prog, u32 *target_size)
4009 {
4010         struct bpf_insn *insn = insn_buf;
4011         int off;
4012
4013         switch (si->off) {
4014         case offsetof(struct __sk_buff, len):
4015                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4016                                       bpf_target_off(struct sk_buff, len, 4,
4017                                                      target_size));
4018                 break;
4019
4020         case offsetof(struct __sk_buff, protocol):
4021                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4022                                       bpf_target_off(struct sk_buff, protocol, 2,
4023                                                      target_size));
4024                 break;
4025
4026         case offsetof(struct __sk_buff, vlan_proto):
4027                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4028                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
4029                                                      target_size));
4030                 break;
4031
4032         case offsetof(struct __sk_buff, priority):
4033                 if (type == BPF_WRITE)
4034                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4035                                               bpf_target_off(struct sk_buff, priority, 4,
4036                                                              target_size));
4037                 else
4038                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4039                                               bpf_target_off(struct sk_buff, priority, 4,
4040                                                              target_size));
4041                 break;
4042
4043         case offsetof(struct __sk_buff, ingress_ifindex):
4044                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4045                                       bpf_target_off(struct sk_buff, skb_iif, 4,
4046                                                      target_size));
4047                 break;
4048
4049         case offsetof(struct __sk_buff, ifindex):
4050                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4051                                       si->dst_reg, si->src_reg,
4052                                       offsetof(struct sk_buff, dev));
4053                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
4054                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4055                                       bpf_target_off(struct net_device, ifindex, 4,
4056                                                      target_size));
4057                 break;
4058
4059         case offsetof(struct __sk_buff, hash):
4060                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4061                                       bpf_target_off(struct sk_buff, hash, 4,
4062                                                      target_size));
4063                 break;
4064
4065         case offsetof(struct __sk_buff, mark):
4066                 if (type == BPF_WRITE)
4067                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4068                                               bpf_target_off(struct sk_buff, mark, 4,
4069                                                              target_size));
4070                 else
4071                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4072                                               bpf_target_off(struct sk_buff, mark, 4,
4073                                                              target_size));
4074                 break;
4075
4076         case offsetof(struct __sk_buff, pkt_type):
4077                 *target_size = 1;
4078                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
4079                                       PKT_TYPE_OFFSET());
4080                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
4081 #ifdef __BIG_ENDIAN_BITFIELD
4082                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
4083 #endif
4084                 break;
4085
4086         case offsetof(struct __sk_buff, queue_mapping):
4087                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4088                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
4089                                                      target_size));
4090                 break;
4091
4092         case offsetof(struct __sk_buff, vlan_present):
4093         case offsetof(struct __sk_buff, vlan_tci):
4094                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
4095
4096                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4097                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
4098                                                      target_size));
4099                 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
4100                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
4101                                                 ~VLAN_TAG_PRESENT);
4102                 } else {
4103                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
4104                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
4105                 }
4106                 break;
4107
4108         case offsetof(struct __sk_buff, cb[0]) ...
4109              offsetofend(struct __sk_buff, cb[4]) - 1:
4110                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
4111                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
4112                               offsetof(struct qdisc_skb_cb, data)) %
4113                              sizeof(__u64));
4114
4115                 prog->cb_access = 1;
4116                 off  = si->off;
4117                 off -= offsetof(struct __sk_buff, cb[0]);
4118                 off += offsetof(struct sk_buff, cb);
4119                 off += offsetof(struct qdisc_skb_cb, data);
4120                 if (type == BPF_WRITE)
4121                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
4122                                               si->src_reg, off);
4123                 else
4124                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
4125                                               si->src_reg, off);
4126                 break;
4127
4128         case offsetof(struct __sk_buff, tc_classid):
4129                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
4130
4131                 off  = si->off;
4132                 off -= offsetof(struct __sk_buff, tc_classid);
4133                 off += offsetof(struct sk_buff, cb);
4134                 off += offsetof(struct qdisc_skb_cb, tc_classid);
4135                 *target_size = 2;
4136                 if (type == BPF_WRITE)
4137                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
4138                                               si->src_reg, off);
4139                 else
4140                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
4141                                               si->src_reg, off);
4142                 break;
4143
4144         case offsetof(struct __sk_buff, data):
4145                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
4146                                       si->dst_reg, si->src_reg,
4147                                       offsetof(struct sk_buff, data));
4148                 break;
4149
4150         case offsetof(struct __sk_buff, data_meta):
4151                 off  = si->off;
4152                 off -= offsetof(struct __sk_buff, data_meta);
4153                 off += offsetof(struct sk_buff, cb);
4154                 off += offsetof(struct bpf_skb_data_end, data_meta);
4155                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4156                                       si->src_reg, off);
4157                 break;
4158
4159         case offsetof(struct __sk_buff, data_end):
4160                 off  = si->off;
4161                 off -= offsetof(struct __sk_buff, data_end);
4162                 off += offsetof(struct sk_buff, cb);
4163                 off += offsetof(struct bpf_skb_data_end, data_end);
4164                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4165                                       si->src_reg, off);
4166                 break;
4167
4168         case offsetof(struct __sk_buff, tc_index):
4169 #ifdef CONFIG_NET_SCHED
4170                 if (type == BPF_WRITE)
4171                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
4172                                               bpf_target_off(struct sk_buff, tc_index, 2,
4173                                                              target_size));
4174                 else
4175                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4176                                               bpf_target_off(struct sk_buff, tc_index, 2,
4177                                                              target_size));
4178 #else
4179                 *target_size = 2;
4180                 if (type == BPF_WRITE)
4181                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
4182                 else
4183                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4184 #endif
4185                 break;
4186
4187         case offsetof(struct __sk_buff, napi_id):
4188 #if defined(CONFIG_NET_RX_BUSY_POLL)
4189                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4190                                       bpf_target_off(struct sk_buff, napi_id, 4,
4191                                                      target_size));
4192                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
4193                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4194 #else
4195                 *target_size = 4;
4196                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4197 #endif
4198                 break;
4199         case offsetof(struct __sk_buff, family):
4200                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4201
4202                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4203                                       si->dst_reg, si->src_reg,
4204                                       offsetof(struct sk_buff, sk));
4205                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4206                                       bpf_target_off(struct sock_common,
4207                                                      skc_family,
4208                                                      2, target_size));
4209                 break;
4210         case offsetof(struct __sk_buff, remote_ip4):
4211                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4212
4213                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4214                                       si->dst_reg, si->src_reg,
4215                                       offsetof(struct sk_buff, sk));
4216                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4217                                       bpf_target_off(struct sock_common,
4218                                                      skc_daddr,
4219                                                      4, target_size));
4220                 break;
4221         case offsetof(struct __sk_buff, local_ip4):
4222                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4223                                           skc_rcv_saddr) != 4);
4224
4225                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4226                                       si->dst_reg, si->src_reg,
4227                                       offsetof(struct sk_buff, sk));
4228                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4229                                       bpf_target_off(struct sock_common,
4230                                                      skc_rcv_saddr,
4231                                                      4, target_size));
4232                 break;
4233         case offsetof(struct __sk_buff, remote_ip6[0]) ...
4234              offsetof(struct __sk_buff, remote_ip6[3]):
4235 #if IS_ENABLED(CONFIG_IPV6)
4236                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4237                                           skc_v6_daddr.s6_addr32[0]) != 4);
4238
4239                 off = si->off;
4240                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
4241
4242                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4243                                       si->dst_reg, si->src_reg,
4244                                       offsetof(struct sk_buff, sk));
4245                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4246                                       offsetof(struct sock_common,
4247                                                skc_v6_daddr.s6_addr32[0]) +
4248                                       off);
4249 #else
4250                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4251 #endif
4252                 break;
4253         case offsetof(struct __sk_buff, local_ip6[0]) ...
4254              offsetof(struct __sk_buff, local_ip6[3]):
4255 #if IS_ENABLED(CONFIG_IPV6)
4256                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4257                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4258
4259                 off = si->off;
4260                 off -= offsetof(struct __sk_buff, local_ip6[0]);
4261
4262                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4263                                       si->dst_reg, si->src_reg,
4264                                       offsetof(struct sk_buff, sk));
4265                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4266                                       offsetof(struct sock_common,
4267                                                skc_v6_rcv_saddr.s6_addr32[0]) +
4268                                       off);
4269 #else
4270                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4271 #endif
4272                 break;
4273
4274         case offsetof(struct __sk_buff, remote_port):
4275                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4276
4277                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4278                                       si->dst_reg, si->src_reg,
4279                                       offsetof(struct sk_buff, sk));
4280                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4281                                       bpf_target_off(struct sock_common,
4282                                                      skc_dport,
4283                                                      2, target_size));
4284 #ifndef __BIG_ENDIAN_BITFIELD
4285                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4286 #endif
4287                 break;
4288
4289         case offsetof(struct __sk_buff, local_port):
4290                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4291
4292                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4293                                       si->dst_reg, si->src_reg,
4294                                       offsetof(struct sk_buff, sk));
4295                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4296                                       bpf_target_off(struct sock_common,
4297                                                      skc_num, 2, target_size));
4298                 break;
4299         }
4300
4301         return insn - insn_buf;
4302 }
4303
4304 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
4305                                           const struct bpf_insn *si,
4306                                           struct bpf_insn *insn_buf,
4307                                           struct bpf_prog *prog, u32 *target_size)
4308 {
4309         struct bpf_insn *insn = insn_buf;
4310
4311         switch (si->off) {
4312         case offsetof(struct bpf_sock, bound_dev_if):
4313                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
4314
4315                 if (type == BPF_WRITE)
4316                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4317                                         offsetof(struct sock, sk_bound_dev_if));
4318                 else
4319                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4320                                       offsetof(struct sock, sk_bound_dev_if));
4321                 break;
4322
4323         case offsetof(struct bpf_sock, mark):
4324                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
4325
4326                 if (type == BPF_WRITE)
4327                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4328                                         offsetof(struct sock, sk_mark));
4329                 else
4330                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4331                                       offsetof(struct sock, sk_mark));
4332                 break;
4333
4334         case offsetof(struct bpf_sock, priority):
4335                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
4336
4337                 if (type == BPF_WRITE)
4338                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4339                                         offsetof(struct sock, sk_priority));
4340                 else
4341                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4342                                       offsetof(struct sock, sk_priority));
4343                 break;
4344
4345         case offsetof(struct bpf_sock, family):
4346                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
4347
4348                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4349                                       offsetof(struct sock, sk_family));
4350                 break;
4351
4352         case offsetof(struct bpf_sock, type):
4353                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4354                                       offsetof(struct sock, __sk_flags_offset));
4355                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
4356                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
4357                 break;
4358
4359         case offsetof(struct bpf_sock, protocol):
4360                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4361                                       offsetof(struct sock, __sk_flags_offset));
4362                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
4363                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
4364                 break;
4365         }
4366
4367         return insn - insn_buf;
4368 }
4369
4370 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
4371                                          const struct bpf_insn *si,
4372                                          struct bpf_insn *insn_buf,
4373                                          struct bpf_prog *prog, u32 *target_size)
4374 {
4375         struct bpf_insn *insn = insn_buf;
4376
4377         switch (si->off) {
4378         case offsetof(struct __sk_buff, ifindex):
4379                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4380                                       si->dst_reg, si->src_reg,
4381                                       offsetof(struct sk_buff, dev));
4382                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4383                                       bpf_target_off(struct net_device, ifindex, 4,
4384                                                      target_size));
4385                 break;
4386         default:
4387                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4388                                               target_size);
4389         }
4390
4391         return insn - insn_buf;
4392 }
4393
4394 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
4395                                   const struct bpf_insn *si,
4396                                   struct bpf_insn *insn_buf,
4397                                   struct bpf_prog *prog, u32 *target_size)
4398 {
4399         struct bpf_insn *insn = insn_buf;
4400
4401         switch (si->off) {
4402         case offsetof(struct xdp_md, data):
4403                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
4404                                       si->dst_reg, si->src_reg,
4405                                       offsetof(struct xdp_buff, data));
4406                 break;
4407         case offsetof(struct xdp_md, data_meta):
4408                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
4409                                       si->dst_reg, si->src_reg,
4410                                       offsetof(struct xdp_buff, data_meta));
4411                 break;
4412         case offsetof(struct xdp_md, data_end):
4413                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
4414                                       si->dst_reg, si->src_reg,
4415                                       offsetof(struct xdp_buff, data_end));
4416                 break;
4417         case offsetof(struct xdp_md, ingress_ifindex):
4418                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
4419                                       si->dst_reg, si->src_reg,
4420                                       offsetof(struct xdp_buff, rxq));
4421                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
4422                                       si->dst_reg, si->dst_reg,
4423                                       offsetof(struct xdp_rxq_info, dev));
4424                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4425                                       offsetof(struct net_device, ifindex));
4426                 break;
4427         case offsetof(struct xdp_md, rx_queue_index):
4428                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
4429                                       si->dst_reg, si->src_reg,
4430                                       offsetof(struct xdp_buff, rxq));
4431                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4432                                       offsetof(struct xdp_rxq_info,
4433                                                queue_index));
4434                 break;
4435         }
4436
4437         return insn - insn_buf;
4438 }
4439
4440 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
4441                                        const struct bpf_insn *si,
4442                                        struct bpf_insn *insn_buf,
4443                                        struct bpf_prog *prog,
4444                                        u32 *target_size)
4445 {
4446         struct bpf_insn *insn = insn_buf;
4447         int off;
4448
4449         switch (si->off) {
4450         case offsetof(struct bpf_sock_ops, op) ...
4451              offsetof(struct bpf_sock_ops, replylong[3]):
4452                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
4453                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
4454                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
4455                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
4456                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
4457                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
4458                 off = si->off;
4459                 off -= offsetof(struct bpf_sock_ops, op);
4460                 off += offsetof(struct bpf_sock_ops_kern, op);
4461                 if (type == BPF_WRITE)
4462                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4463                                               off);
4464                 else
4465                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4466                                               off);
4467                 break;
4468
4469         case offsetof(struct bpf_sock_ops, family):
4470                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4471
4472                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4473                                               struct bpf_sock_ops_kern, sk),
4474                                       si->dst_reg, si->src_reg,
4475                                       offsetof(struct bpf_sock_ops_kern, sk));
4476                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4477                                       offsetof(struct sock_common, skc_family));
4478                 break;
4479
4480         case offsetof(struct bpf_sock_ops, remote_ip4):
4481                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4482
4483                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4484                                                 struct bpf_sock_ops_kern, sk),
4485                                       si->dst_reg, si->src_reg,
4486                                       offsetof(struct bpf_sock_ops_kern, sk));
4487                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4488                                       offsetof(struct sock_common, skc_daddr));
4489                 break;
4490
4491         case offsetof(struct bpf_sock_ops, local_ip4):
4492                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_rcv_saddr) != 4);
4493
4494                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4495                                               struct bpf_sock_ops_kern, sk),
4496                                       si->dst_reg, si->src_reg,
4497                                       offsetof(struct bpf_sock_ops_kern, sk));
4498                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4499                                       offsetof(struct sock_common,
4500                                                skc_rcv_saddr));
4501                 break;
4502
4503         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
4504              offsetof(struct bpf_sock_ops, remote_ip6[3]):
4505 #if IS_ENABLED(CONFIG_IPV6)
4506                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4507                                           skc_v6_daddr.s6_addr32[0]) != 4);
4508
4509                 off = si->off;
4510                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
4511                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4512                                                 struct bpf_sock_ops_kern, sk),
4513                                       si->dst_reg, si->src_reg,
4514                                       offsetof(struct bpf_sock_ops_kern, sk));
4515                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4516                                       offsetof(struct sock_common,
4517                                                skc_v6_daddr.s6_addr32[0]) +
4518                                       off);
4519 #else
4520                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4521 #endif
4522                 break;
4523
4524         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
4525              offsetof(struct bpf_sock_ops, local_ip6[3]):
4526 #if IS_ENABLED(CONFIG_IPV6)
4527                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4528                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4529
4530                 off = si->off;
4531                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
4532                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4533                                                 struct bpf_sock_ops_kern, sk),
4534                                       si->dst_reg, si->src_reg,
4535                                       offsetof(struct bpf_sock_ops_kern, sk));
4536                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4537                                       offsetof(struct sock_common,
4538                                                skc_v6_rcv_saddr.s6_addr32[0]) +
4539                                       off);
4540 #else
4541                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4542 #endif
4543                 break;
4544
4545         case offsetof(struct bpf_sock_ops, remote_port):
4546                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4547
4548                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4549                                                 struct bpf_sock_ops_kern, sk),
4550                                       si->dst_reg, si->src_reg,
4551                                       offsetof(struct bpf_sock_ops_kern, sk));
4552                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4553                                       offsetof(struct sock_common, skc_dport));
4554 #ifndef __BIG_ENDIAN_BITFIELD
4555                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4556 #endif
4557                 break;
4558
4559         case offsetof(struct bpf_sock_ops, local_port):
4560                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4561
4562                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4563                                                 struct bpf_sock_ops_kern, sk),
4564                                       si->dst_reg, si->src_reg,
4565                                       offsetof(struct bpf_sock_ops_kern, sk));
4566                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4567                                       offsetof(struct sock_common, skc_num));
4568                 break;
4569
4570         case offsetof(struct bpf_sock_ops, is_fullsock):
4571                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4572                                                 struct bpf_sock_ops_kern,
4573                                                 is_fullsock),
4574                                       si->dst_reg, si->src_reg,
4575                                       offsetof(struct bpf_sock_ops_kern,
4576                                                is_fullsock));
4577                 break;
4578
4579         case offsetof(struct bpf_sock_ops, state):
4580                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
4581
4582                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4583                                                 struct bpf_sock_ops_kern, sk),
4584                                       si->dst_reg, si->src_reg,
4585                                       offsetof(struct bpf_sock_ops_kern, sk));
4586                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
4587                                       offsetof(struct sock_common, skc_state));
4588                 break;
4589
4590         case offsetof(struct bpf_sock_ops, rtt_min):
4591                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
4592                              sizeof(struct minmax));
4593                 BUILD_BUG_ON(sizeof(struct minmax) <
4594                              sizeof(struct minmax_sample));
4595
4596                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4597                                                 struct bpf_sock_ops_kern, sk),
4598                                       si->dst_reg, si->src_reg,
4599                                       offsetof(struct bpf_sock_ops_kern, sk));
4600                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4601                                       offsetof(struct tcp_sock, rtt_min) +
4602                                       FIELD_SIZEOF(struct minmax_sample, t));
4603                 break;
4604
4605 /* Helper macro for adding read access to tcp_sock or sock fields. */
4606 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
4607         do {                                                                  \
4608                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
4609                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
4610                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4611                                                 struct bpf_sock_ops_kern,     \
4612                                                 is_fullsock),                 \
4613                                       si->dst_reg, si->src_reg,               \
4614                                       offsetof(struct bpf_sock_ops_kern,      \
4615                                                is_fullsock));                 \
4616                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
4617                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4618                                                 struct bpf_sock_ops_kern, sk),\
4619                                       si->dst_reg, si->src_reg,               \
4620                                       offsetof(struct bpf_sock_ops_kern, sk));\
4621                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
4622                                                        OBJ_FIELD),            \
4623                                       si->dst_reg, si->dst_reg,               \
4624                                       offsetof(OBJ, OBJ_FIELD));              \
4625         } while (0)
4626
4627 /* Helper macro for adding write access to tcp_sock or sock fields.
4628  * The macro is called with two registers, dst_reg which contains a pointer
4629  * to ctx (context) and src_reg which contains the value that should be
4630  * stored. However, we need an additional register since we cannot overwrite
4631  * dst_reg because it may be used later in the program.
4632  * Instead we "borrow" one of the other register. We first save its value
4633  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
4634  * it at the end of the macro.
4635  */
4636 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
4637         do {                                                                  \
4638                 int reg = BPF_REG_9;                                          \
4639                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
4640                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
4641                 if (si->dst_reg == reg || si->src_reg == reg)                 \
4642                         reg--;                                                \
4643                 if (si->dst_reg == reg || si->src_reg == reg)                 \
4644                         reg--;                                                \
4645                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
4646                                       offsetof(struct bpf_sock_ops_kern,      \
4647                                                temp));                        \
4648                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4649                                                 struct bpf_sock_ops_kern,     \
4650                                                 is_fullsock),                 \
4651                                       reg, si->dst_reg,                       \
4652                                       offsetof(struct bpf_sock_ops_kern,      \
4653                                                is_fullsock));                 \
4654                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
4655                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4656                                                 struct bpf_sock_ops_kern, sk),\
4657                                       reg, si->dst_reg,                       \
4658                                       offsetof(struct bpf_sock_ops_kern, sk));\
4659                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
4660                                       reg, si->src_reg,                       \
4661                                       offsetof(OBJ, OBJ_FIELD));              \
4662                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
4663                                       offsetof(struct bpf_sock_ops_kern,      \
4664                                                temp));                        \
4665         } while (0)
4666
4667 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
4668         do {                                                                  \
4669                 if (TYPE == BPF_WRITE)                                        \
4670                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
4671                 else                                                          \
4672                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
4673         } while (0)
4674
4675         case offsetof(struct bpf_sock_ops, snd_cwnd):
4676                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
4677                 break;
4678
4679         case offsetof(struct bpf_sock_ops, srtt_us):
4680                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
4681                 break;
4682
4683         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
4684                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
4685                                    struct tcp_sock);
4686                 break;
4687
4688         case offsetof(struct bpf_sock_ops, snd_ssthresh):
4689                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
4690                 break;
4691
4692         case offsetof(struct bpf_sock_ops, rcv_nxt):
4693                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
4694                 break;
4695
4696         case offsetof(struct bpf_sock_ops, snd_nxt):
4697                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
4698                 break;
4699
4700         case offsetof(struct bpf_sock_ops, snd_una):
4701                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
4702                 break;
4703
4704         case offsetof(struct bpf_sock_ops, mss_cache):
4705                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
4706                 break;
4707
4708         case offsetof(struct bpf_sock_ops, ecn_flags):
4709                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
4710                 break;
4711
4712         case offsetof(struct bpf_sock_ops, rate_delivered):
4713                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
4714                                    struct tcp_sock);
4715                 break;
4716
4717         case offsetof(struct bpf_sock_ops, rate_interval_us):
4718                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
4719                                    struct tcp_sock);
4720                 break;
4721
4722         case offsetof(struct bpf_sock_ops, packets_out):
4723                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
4724                 break;
4725
4726         case offsetof(struct bpf_sock_ops, retrans_out):
4727                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
4728                 break;
4729
4730         case offsetof(struct bpf_sock_ops, total_retrans):
4731                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
4732                                    struct tcp_sock);
4733                 break;
4734
4735         case offsetof(struct bpf_sock_ops, segs_in):
4736                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
4737                 break;
4738
4739         case offsetof(struct bpf_sock_ops, data_segs_in):
4740                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
4741                 break;
4742
4743         case offsetof(struct bpf_sock_ops, segs_out):
4744                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
4745                 break;
4746
4747         case offsetof(struct bpf_sock_ops, data_segs_out):
4748                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
4749                                    struct tcp_sock);
4750                 break;
4751
4752         case offsetof(struct bpf_sock_ops, lost_out):
4753                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
4754                 break;
4755
4756         case offsetof(struct bpf_sock_ops, sacked_out):
4757                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
4758                 break;
4759
4760         case offsetof(struct bpf_sock_ops, sk_txhash):
4761                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
4762                                           struct sock, type);
4763                 break;
4764
4765         case offsetof(struct bpf_sock_ops, bytes_received):
4766                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
4767                                    struct tcp_sock);
4768                 break;
4769
4770         case offsetof(struct bpf_sock_ops, bytes_acked):
4771                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
4772                 break;
4773
4774         }
4775         return insn - insn_buf;
4776 }
4777
4778 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
4779                                      const struct bpf_insn *si,
4780                                      struct bpf_insn *insn_buf,
4781                                      struct bpf_prog *prog, u32 *target_size)
4782 {
4783         struct bpf_insn *insn = insn_buf;
4784         int off;
4785
4786         switch (si->off) {
4787         case offsetof(struct __sk_buff, data_end):
4788                 off  = si->off;
4789                 off -= offsetof(struct __sk_buff, data_end);
4790                 off += offsetof(struct sk_buff, cb);
4791                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
4792                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4793                                       si->src_reg, off);
4794                 break;
4795         default:
4796                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4797                                               target_size);
4798         }
4799
4800         return insn - insn_buf;
4801 }
4802
4803 const struct bpf_verifier_ops sk_filter_verifier_ops = {
4804         .get_func_proto         = sk_filter_func_proto,
4805         .is_valid_access        = sk_filter_is_valid_access,
4806         .convert_ctx_access     = bpf_convert_ctx_access,
4807 };
4808
4809 const struct bpf_prog_ops sk_filter_prog_ops = {
4810         .test_run               = bpf_prog_test_run_skb,
4811 };
4812
4813 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
4814         .get_func_proto         = tc_cls_act_func_proto,
4815         .is_valid_access        = tc_cls_act_is_valid_access,
4816         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
4817         .gen_prologue           = tc_cls_act_prologue,
4818 };
4819
4820 const struct bpf_prog_ops tc_cls_act_prog_ops = {
4821         .test_run               = bpf_prog_test_run_skb,
4822 };
4823
4824 const struct bpf_verifier_ops xdp_verifier_ops = {
4825         .get_func_proto         = xdp_func_proto,
4826         .is_valid_access        = xdp_is_valid_access,
4827         .convert_ctx_access     = xdp_convert_ctx_access,
4828 };
4829
4830 const struct bpf_prog_ops xdp_prog_ops = {
4831         .test_run               = bpf_prog_test_run_xdp,
4832 };
4833
4834 const struct bpf_verifier_ops cg_skb_verifier_ops = {
4835         .get_func_proto         = sk_filter_func_proto,
4836         .is_valid_access        = sk_filter_is_valid_access,
4837         .convert_ctx_access     = bpf_convert_ctx_access,
4838 };
4839
4840 const struct bpf_prog_ops cg_skb_prog_ops = {
4841         .test_run               = bpf_prog_test_run_skb,
4842 };
4843
4844 const struct bpf_verifier_ops lwt_inout_verifier_ops = {
4845         .get_func_proto         = lwt_inout_func_proto,
4846         .is_valid_access        = lwt_is_valid_access,
4847         .convert_ctx_access     = bpf_convert_ctx_access,
4848 };
4849
4850 const struct bpf_prog_ops lwt_inout_prog_ops = {
4851         .test_run               = bpf_prog_test_run_skb,
4852 };
4853
4854 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
4855         .get_func_proto         = lwt_xmit_func_proto,
4856         .is_valid_access        = lwt_is_valid_access,
4857         .convert_ctx_access     = bpf_convert_ctx_access,
4858         .gen_prologue           = tc_cls_act_prologue,
4859 };
4860
4861 const struct bpf_prog_ops lwt_xmit_prog_ops = {
4862         .test_run               = bpf_prog_test_run_skb,
4863 };
4864
4865 const struct bpf_verifier_ops cg_sock_verifier_ops = {
4866         .get_func_proto         = sock_filter_func_proto,
4867         .is_valid_access        = sock_filter_is_valid_access,
4868         .convert_ctx_access     = sock_filter_convert_ctx_access,
4869 };
4870
4871 const struct bpf_prog_ops cg_sock_prog_ops = {
4872 };
4873
4874 const struct bpf_verifier_ops sock_ops_verifier_ops = {
4875         .get_func_proto         = sock_ops_func_proto,
4876         .is_valid_access        = sock_ops_is_valid_access,
4877         .convert_ctx_access     = sock_ops_convert_ctx_access,
4878 };
4879
4880 const struct bpf_prog_ops sock_ops_prog_ops = {
4881 };
4882
4883 const struct bpf_verifier_ops sk_skb_verifier_ops = {
4884         .get_func_proto         = sk_skb_func_proto,
4885         .is_valid_access        = sk_skb_is_valid_access,
4886         .convert_ctx_access     = sk_skb_convert_ctx_access,
4887         .gen_prologue           = sk_skb_prologue,
4888 };
4889
4890 const struct bpf_prog_ops sk_skb_prog_ops = {
4891 };
4892
4893 int sk_detach_filter(struct sock *sk)
4894 {
4895         int ret = -ENOENT;
4896         struct sk_filter *filter;
4897
4898         if (sock_flag(sk, SOCK_FILTER_LOCKED))
4899                 return -EPERM;
4900
4901         filter = rcu_dereference_protected(sk->sk_filter,
4902                                            lockdep_sock_is_held(sk));
4903         if (filter) {
4904                 RCU_INIT_POINTER(sk->sk_filter, NULL);
4905                 sk_filter_uncharge(sk, filter);
4906                 ret = 0;
4907         }
4908
4909         return ret;
4910 }
4911 EXPORT_SYMBOL_GPL(sk_detach_filter);
4912
4913 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
4914                   unsigned int len)
4915 {
4916         struct sock_fprog_kern *fprog;
4917         struct sk_filter *filter;
4918         int ret = 0;
4919
4920         lock_sock(sk);
4921         filter = rcu_dereference_protected(sk->sk_filter,
4922                                            lockdep_sock_is_held(sk));
4923         if (!filter)
4924                 goto out;
4925
4926         /* We're copying the filter that has been originally attached,
4927          * so no conversion/decode needed anymore. eBPF programs that
4928          * have no original program cannot be dumped through this.
4929          */
4930         ret = -EACCES;
4931         fprog = filter->prog->orig_prog;
4932         if (!fprog)
4933                 goto out;
4934
4935         ret = fprog->len;
4936         if (!len)
4937                 /* User space only enquires number of filter blocks. */
4938                 goto out;
4939
4940         ret = -EINVAL;
4941         if (len < fprog->len)
4942                 goto out;
4943
4944         ret = -EFAULT;
4945         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
4946                 goto out;
4947
4948         /* Instead of bytes, the API requests to return the number
4949          * of filter blocks.
4950          */
4951         ret = fprog->len;
4952 out:
4953         release_sock(sk);
4954         return ret;
4955 }