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         ret = skb_cow(skb, len_diff);
2091         if (unlikely(ret < 0))
2092                 return ret;
2093
2094         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2095         if (unlikely(ret < 0))
2096                 return ret;
2097
2098         if (skb_is_gso(skb)) {
2099                 /* SKB_GSO_TCPV4 needs to be changed into
2100                  * SKB_GSO_TCPV6.
2101                  */
2102                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2103                         skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
2104                         skb_shinfo(skb)->gso_type |=  SKB_GSO_TCPV6;
2105                 }
2106
2107                 /* Due to IPv6 header, MSS needs to be downgraded. */
2108                 skb_shinfo(skb)->gso_size -= len_diff;
2109                 /* Header must be checked, and gso_segs recomputed. */
2110                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2111                 skb_shinfo(skb)->gso_segs = 0;
2112         }
2113
2114         skb->protocol = htons(ETH_P_IPV6);
2115         skb_clear_hash(skb);
2116
2117         return 0;
2118 }
2119
2120 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2121 {
2122         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2123         u32 off = skb_mac_header_len(skb);
2124         int ret;
2125
2126         ret = skb_unclone(skb, GFP_ATOMIC);
2127         if (unlikely(ret < 0))
2128                 return ret;
2129
2130         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2131         if (unlikely(ret < 0))
2132                 return ret;
2133
2134         if (skb_is_gso(skb)) {
2135                 /* SKB_GSO_TCPV6 needs to be changed into
2136                  * SKB_GSO_TCPV4.
2137                  */
2138                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
2139                         skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
2140                         skb_shinfo(skb)->gso_type |=  SKB_GSO_TCPV4;
2141                 }
2142
2143                 /* Due to IPv4 header, MSS can be upgraded. */
2144                 skb_shinfo(skb)->gso_size += len_diff;
2145                 /* Header must be checked, and gso_segs recomputed. */
2146                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2147                 skb_shinfo(skb)->gso_segs = 0;
2148         }
2149
2150         skb->protocol = htons(ETH_P_IP);
2151         skb_clear_hash(skb);
2152
2153         return 0;
2154 }
2155
2156 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2157 {
2158         __be16 from_proto = skb->protocol;
2159
2160         if (from_proto == htons(ETH_P_IP) &&
2161               to_proto == htons(ETH_P_IPV6))
2162                 return bpf_skb_proto_4_to_6(skb);
2163
2164         if (from_proto == htons(ETH_P_IPV6) &&
2165               to_proto == htons(ETH_P_IP))
2166                 return bpf_skb_proto_6_to_4(skb);
2167
2168         return -ENOTSUPP;
2169 }
2170
2171 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2172            u64, flags)
2173 {
2174         int ret;
2175
2176         if (unlikely(flags))
2177                 return -EINVAL;
2178
2179         /* General idea is that this helper does the basic groundwork
2180          * needed for changing the protocol, and eBPF program fills the
2181          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2182          * and other helpers, rather than passing a raw buffer here.
2183          *
2184          * The rationale is to keep this minimal and without a need to
2185          * deal with raw packet data. F.e. even if we would pass buffers
2186          * here, the program still needs to call the bpf_lX_csum_replace()
2187          * helpers anyway. Plus, this way we keep also separation of
2188          * concerns, since f.e. bpf_skb_store_bytes() should only take
2189          * care of stores.
2190          *
2191          * Currently, additional options and extension header space are
2192          * not supported, but flags register is reserved so we can adapt
2193          * that. For offloads, we mark packet as dodgy, so that headers
2194          * need to be verified first.
2195          */
2196         ret = bpf_skb_proto_xlat(skb, proto);
2197         bpf_compute_data_pointers(skb);
2198         return ret;
2199 }
2200
2201 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2202         .func           = bpf_skb_change_proto,
2203         .gpl_only       = false,
2204         .ret_type       = RET_INTEGER,
2205         .arg1_type      = ARG_PTR_TO_CTX,
2206         .arg2_type      = ARG_ANYTHING,
2207         .arg3_type      = ARG_ANYTHING,
2208 };
2209
2210 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2211 {
2212         /* We only allow a restricted subset to be changed for now. */
2213         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2214                      !skb_pkt_type_ok(pkt_type)))
2215                 return -EINVAL;
2216
2217         skb->pkt_type = pkt_type;
2218         return 0;
2219 }
2220
2221 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2222         .func           = bpf_skb_change_type,
2223         .gpl_only       = false,
2224         .ret_type       = RET_INTEGER,
2225         .arg1_type      = ARG_PTR_TO_CTX,
2226         .arg2_type      = ARG_ANYTHING,
2227 };
2228
2229 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2230 {
2231         switch (skb->protocol) {
2232         case htons(ETH_P_IP):
2233                 return sizeof(struct iphdr);
2234         case htons(ETH_P_IPV6):
2235                 return sizeof(struct ipv6hdr);
2236         default:
2237                 return ~0U;
2238         }
2239 }
2240
2241 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2242 {
2243         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2244         int ret;
2245
2246         ret = skb_cow(skb, len_diff);
2247         if (unlikely(ret < 0))
2248                 return ret;
2249
2250         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2251         if (unlikely(ret < 0))
2252                 return ret;
2253
2254         if (skb_is_gso(skb)) {
2255                 /* Due to header grow, MSS needs to be downgraded. */
2256                 skb_shinfo(skb)->gso_size -= len_diff;
2257                 /* Header must be checked, and gso_segs recomputed. */
2258                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2259                 skb_shinfo(skb)->gso_segs = 0;
2260         }
2261
2262         return 0;
2263 }
2264
2265 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2266 {
2267         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2268         int ret;
2269
2270         ret = skb_unclone(skb, GFP_ATOMIC);
2271         if (unlikely(ret < 0))
2272                 return ret;
2273
2274         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2275         if (unlikely(ret < 0))
2276                 return ret;
2277
2278         if (skb_is_gso(skb)) {
2279                 /* Due to header shrink, MSS can be upgraded. */
2280                 skb_shinfo(skb)->gso_size += len_diff;
2281                 /* Header must be checked, and gso_segs recomputed. */
2282                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2283                 skb_shinfo(skb)->gso_segs = 0;
2284         }
2285
2286         return 0;
2287 }
2288
2289 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2290 {
2291         return skb->dev->mtu + skb->dev->hard_header_len;
2292 }
2293
2294 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2295 {
2296         bool trans_same = skb->transport_header == skb->network_header;
2297         u32 len_cur, len_diff_abs = abs(len_diff);
2298         u32 len_min = bpf_skb_net_base_len(skb);
2299         u32 len_max = __bpf_skb_max_len(skb);
2300         __be16 proto = skb->protocol;
2301         bool shrink = len_diff < 0;
2302         int ret;
2303
2304         if (unlikely(len_diff_abs > 0xfffU))
2305                 return -EFAULT;
2306         if (unlikely(proto != htons(ETH_P_IP) &&
2307                      proto != htons(ETH_P_IPV6)))
2308                 return -ENOTSUPP;
2309
2310         len_cur = skb->len - skb_network_offset(skb);
2311         if (skb_transport_header_was_set(skb) && !trans_same)
2312                 len_cur = skb_network_header_len(skb);
2313         if ((shrink && (len_diff_abs >= len_cur ||
2314                         len_cur - len_diff_abs < len_min)) ||
2315             (!shrink && (skb->len + len_diff_abs > len_max &&
2316                          !skb_is_gso(skb))))
2317                 return -ENOTSUPP;
2318
2319         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2320                        bpf_skb_net_grow(skb, len_diff_abs);
2321
2322         bpf_compute_data_pointers(skb);
2323         return ret;
2324 }
2325
2326 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2327            u32, mode, u64, flags)
2328 {
2329         if (unlikely(flags))
2330                 return -EINVAL;
2331         if (likely(mode == BPF_ADJ_ROOM_NET))
2332                 return bpf_skb_adjust_net(skb, len_diff);
2333
2334         return -ENOTSUPP;
2335 }
2336
2337 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2338         .func           = bpf_skb_adjust_room,
2339         .gpl_only       = false,
2340         .ret_type       = RET_INTEGER,
2341         .arg1_type      = ARG_PTR_TO_CTX,
2342         .arg2_type      = ARG_ANYTHING,
2343         .arg3_type      = ARG_ANYTHING,
2344         .arg4_type      = ARG_ANYTHING,
2345 };
2346
2347 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2348 {
2349         u32 min_len = skb_network_offset(skb);
2350
2351         if (skb_transport_header_was_set(skb))
2352                 min_len = skb_transport_offset(skb);
2353         if (skb->ip_summed == CHECKSUM_PARTIAL)
2354                 min_len = skb_checksum_start_offset(skb) +
2355                           skb->csum_offset + sizeof(__sum16);
2356         return min_len;
2357 }
2358
2359 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2360 {
2361         unsigned int old_len = skb->len;
2362         int ret;
2363
2364         ret = __skb_grow_rcsum(skb, new_len);
2365         if (!ret)
2366                 memset(skb->data + old_len, 0, new_len - old_len);
2367         return ret;
2368 }
2369
2370 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2371 {
2372         return __skb_trim_rcsum(skb, new_len);
2373 }
2374
2375 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2376            u64, flags)
2377 {
2378         u32 max_len = __bpf_skb_max_len(skb);
2379         u32 min_len = __bpf_skb_min_len(skb);
2380         int ret;
2381
2382         if (unlikely(flags || new_len > max_len || new_len < min_len))
2383                 return -EINVAL;
2384         if (skb->encapsulation)
2385                 return -ENOTSUPP;
2386
2387         /* The basic idea of this helper is that it's performing the
2388          * needed work to either grow or trim an skb, and eBPF program
2389          * rewrites the rest via helpers like bpf_skb_store_bytes(),
2390          * bpf_lX_csum_replace() and others rather than passing a raw
2391          * buffer here. This one is a slow path helper and intended
2392          * for replies with control messages.
2393          *
2394          * Like in bpf_skb_change_proto(), we want to keep this rather
2395          * minimal and without protocol specifics so that we are able
2396          * to separate concerns as in bpf_skb_store_bytes() should only
2397          * be the one responsible for writing buffers.
2398          *
2399          * It's really expected to be a slow path operation here for
2400          * control message replies, so we're implicitly linearizing,
2401          * uncloning and drop offloads from the skb by this.
2402          */
2403         ret = __bpf_try_make_writable(skb, skb->len);
2404         if (!ret) {
2405                 if (new_len > skb->len)
2406                         ret = bpf_skb_grow_rcsum(skb, new_len);
2407                 else if (new_len < skb->len)
2408                         ret = bpf_skb_trim_rcsum(skb, new_len);
2409                 if (!ret && skb_is_gso(skb))
2410                         skb_gso_reset(skb);
2411         }
2412
2413         bpf_compute_data_pointers(skb);
2414         return ret;
2415 }
2416
2417 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2418         .func           = bpf_skb_change_tail,
2419         .gpl_only       = false,
2420         .ret_type       = RET_INTEGER,
2421         .arg1_type      = ARG_PTR_TO_CTX,
2422         .arg2_type      = ARG_ANYTHING,
2423         .arg3_type      = ARG_ANYTHING,
2424 };
2425
2426 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
2427            u64, flags)
2428 {
2429         u32 max_len = __bpf_skb_max_len(skb);
2430         u32 new_len = skb->len + head_room;
2431         int ret;
2432
2433         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2434                      new_len < skb->len))
2435                 return -EINVAL;
2436
2437         ret = skb_cow(skb, head_room);
2438         if (likely(!ret)) {
2439                 /* Idea for this helper is that we currently only
2440                  * allow to expand on mac header. This means that
2441                  * skb->protocol network header, etc, stay as is.
2442                  * Compared to bpf_skb_change_tail(), we're more
2443                  * flexible due to not needing to linearize or
2444                  * reset GSO. Intention for this helper is to be
2445                  * used by an L3 skb that needs to push mac header
2446                  * for redirection into L2 device.
2447                  */
2448                 __skb_push(skb, head_room);
2449                 memset(skb->data, 0, head_room);
2450                 skb_reset_mac_header(skb);
2451         }
2452
2453         bpf_compute_data_pointers(skb);
2454         return 0;
2455 }
2456
2457 static const struct bpf_func_proto bpf_skb_change_head_proto = {
2458         .func           = bpf_skb_change_head,
2459         .gpl_only       = false,
2460         .ret_type       = RET_INTEGER,
2461         .arg1_type      = ARG_PTR_TO_CTX,
2462         .arg2_type      = ARG_ANYTHING,
2463         .arg3_type      = ARG_ANYTHING,
2464 };
2465
2466 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
2467 {
2468         return xdp_data_meta_unsupported(xdp) ? 0 :
2469                xdp->data - xdp->data_meta;
2470 }
2471
2472 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
2473 {
2474         unsigned long metalen = xdp_get_metalen(xdp);
2475         void *data_start = xdp->data_hard_start + metalen;
2476         void *data = xdp->data + offset;
2477
2478         if (unlikely(data < data_start ||
2479                      data > xdp->data_end - ETH_HLEN))
2480                 return -EINVAL;
2481
2482         if (metalen)
2483                 memmove(xdp->data_meta + offset,
2484                         xdp->data_meta, metalen);
2485         xdp->data_meta += offset;
2486         xdp->data = data;
2487
2488         return 0;
2489 }
2490
2491 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
2492         .func           = bpf_xdp_adjust_head,
2493         .gpl_only       = false,
2494         .ret_type       = RET_INTEGER,
2495         .arg1_type      = ARG_PTR_TO_CTX,
2496         .arg2_type      = ARG_ANYTHING,
2497 };
2498
2499 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
2500 {
2501         void *meta = xdp->data_meta + offset;
2502         unsigned long metalen = xdp->data - meta;
2503
2504         if (xdp_data_meta_unsupported(xdp))
2505                 return -ENOTSUPP;
2506         if (unlikely(meta < xdp->data_hard_start ||
2507                      meta > xdp->data))
2508                 return -EINVAL;
2509         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
2510                      (metalen > 32)))
2511                 return -EACCES;
2512
2513         xdp->data_meta = meta;
2514
2515         return 0;
2516 }
2517
2518 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
2519         .func           = bpf_xdp_adjust_meta,
2520         .gpl_only       = false,
2521         .ret_type       = RET_INTEGER,
2522         .arg1_type      = ARG_PTR_TO_CTX,
2523         .arg2_type      = ARG_ANYTHING,
2524 };
2525
2526 static int __bpf_tx_xdp(struct net_device *dev,
2527                         struct bpf_map *map,
2528                         struct xdp_buff *xdp,
2529                         u32 index)
2530 {
2531         int err;
2532
2533         if (!dev->netdev_ops->ndo_xdp_xmit) {
2534                 return -EOPNOTSUPP;
2535         }
2536
2537         err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2538         if (err)
2539                 return err;
2540         dev->netdev_ops->ndo_xdp_flush(dev);
2541         return 0;
2542 }
2543
2544 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
2545                             struct bpf_map *map,
2546                             struct xdp_buff *xdp,
2547                             u32 index)
2548 {
2549         int err;
2550
2551         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
2552                 struct net_device *dev = fwd;
2553
2554                 if (!dev->netdev_ops->ndo_xdp_xmit)
2555                         return -EOPNOTSUPP;
2556
2557                 err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2558                 if (err)
2559                         return err;
2560                 __dev_map_insert_ctx(map, index);
2561
2562         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
2563                 struct bpf_cpu_map_entry *rcpu = fwd;
2564
2565                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
2566                 if (err)
2567                         return err;
2568                 __cpu_map_insert_ctx(map, index);
2569         }
2570         return 0;
2571 }
2572
2573 void xdp_do_flush_map(void)
2574 {
2575         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2576         struct bpf_map *map = ri->map_to_flush;
2577
2578         ri->map_to_flush = NULL;
2579         if (map) {
2580                 switch (map->map_type) {
2581                 case BPF_MAP_TYPE_DEVMAP:
2582                         __dev_map_flush(map);
2583                         break;
2584                 case BPF_MAP_TYPE_CPUMAP:
2585                         __cpu_map_flush(map);
2586                         break;
2587                 default:
2588                         break;
2589                 }
2590         }
2591 }
2592 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
2593
2594 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
2595 {
2596         switch (map->map_type) {
2597         case BPF_MAP_TYPE_DEVMAP:
2598                 return __dev_map_lookup_elem(map, index);
2599         case BPF_MAP_TYPE_CPUMAP:
2600                 return __cpu_map_lookup_elem(map, index);
2601         default:
2602                 return NULL;
2603         }
2604 }
2605
2606 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
2607                                    unsigned long aux)
2608 {
2609         return (unsigned long)xdp_prog->aux != aux;
2610 }
2611
2612 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
2613                                struct bpf_prog *xdp_prog)
2614 {
2615         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2616         unsigned long map_owner = ri->map_owner;
2617         struct bpf_map *map = ri->map;
2618         u32 index = ri->ifindex;
2619         void *fwd = NULL;
2620         int err;
2621
2622         ri->ifindex = 0;
2623         ri->map = NULL;
2624         ri->map_owner = 0;
2625
2626         if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2627                 err = -EFAULT;
2628                 map = NULL;
2629                 goto err;
2630         }
2631
2632         fwd = __xdp_map_lookup_elem(map, index);
2633         if (!fwd) {
2634                 err = -EINVAL;
2635                 goto err;
2636         }
2637         if (ri->map_to_flush && ri->map_to_flush != map)
2638                 xdp_do_flush_map();
2639
2640         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
2641         if (unlikely(err))
2642                 goto err;
2643
2644         ri->map_to_flush = map;
2645         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2646         return 0;
2647 err:
2648         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2649         return err;
2650 }
2651
2652 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
2653                     struct bpf_prog *xdp_prog)
2654 {
2655         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2656         struct net_device *fwd;
2657         u32 index = ri->ifindex;
2658         int err;
2659
2660         if (ri->map)
2661                 return xdp_do_redirect_map(dev, xdp, xdp_prog);
2662
2663         fwd = dev_get_by_index_rcu(dev_net(dev), index);
2664         ri->ifindex = 0;
2665         if (unlikely(!fwd)) {
2666                 err = -EINVAL;
2667                 goto err;
2668         }
2669
2670         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
2671         if (unlikely(err))
2672                 goto err;
2673
2674         _trace_xdp_redirect(dev, xdp_prog, index);
2675         return 0;
2676 err:
2677         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2678         return err;
2679 }
2680 EXPORT_SYMBOL_GPL(xdp_do_redirect);
2681
2682 static int __xdp_generic_ok_fwd_dev(struct sk_buff *skb, struct net_device *fwd)
2683 {
2684         unsigned int len;
2685
2686         if (unlikely(!(fwd->flags & IFF_UP)))
2687                 return -ENETDOWN;
2688
2689         len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
2690         if (skb->len > len)
2691                 return -EMSGSIZE;
2692
2693         return 0;
2694 }
2695
2696 static int xdp_do_generic_redirect_map(struct net_device *dev,
2697                                        struct sk_buff *skb,
2698                                        struct bpf_prog *xdp_prog)
2699 {
2700         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2701         unsigned long map_owner = ri->map_owner;
2702         struct bpf_map *map = ri->map;
2703         struct net_device *fwd = NULL;
2704         u32 index = ri->ifindex;
2705         int err = 0;
2706
2707         ri->ifindex = 0;
2708         ri->map = NULL;
2709         ri->map_owner = 0;
2710
2711         if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2712                 err = -EFAULT;
2713                 map = NULL;
2714                 goto err;
2715         }
2716         fwd = __xdp_map_lookup_elem(map, index);
2717         if (unlikely(!fwd)) {
2718                 err = -EINVAL;
2719                 goto err;
2720         }
2721
2722         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
2723                 if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
2724                         goto err;
2725                 skb->dev = fwd;
2726         } else {
2727                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
2728                 err = -EBADRQC;
2729                 goto err;
2730         }
2731
2732         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2733         return 0;
2734 err:
2735         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2736         return err;
2737 }
2738
2739 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
2740                             struct bpf_prog *xdp_prog)
2741 {
2742         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2743         u32 index = ri->ifindex;
2744         struct net_device *fwd;
2745         int err = 0;
2746
2747         if (ri->map)
2748                 return xdp_do_generic_redirect_map(dev, skb, xdp_prog);
2749
2750         ri->ifindex = 0;
2751         fwd = dev_get_by_index_rcu(dev_net(dev), index);
2752         if (unlikely(!fwd)) {
2753                 err = -EINVAL;
2754                 goto err;
2755         }
2756
2757         if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
2758                 goto err;
2759
2760         skb->dev = fwd;
2761         _trace_xdp_redirect(dev, xdp_prog, index);
2762         return 0;
2763 err:
2764         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2765         return err;
2766 }
2767 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
2768
2769 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
2770 {
2771         struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2772
2773         if (unlikely(flags))
2774                 return XDP_ABORTED;
2775
2776         ri->ifindex = ifindex;
2777         ri->flags = flags;
2778         ri->map = NULL;
2779         ri->map_owner = 0;
2780
2781         return XDP_REDIRECT;
2782 }
2783
2784 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
2785         .func           = bpf_xdp_redirect,
2786         .gpl_only       = false,
2787         .ret_type       = RET_INTEGER,
2788         .arg1_type      = ARG_ANYTHING,
2789         .arg2_type      = ARG_ANYTHING,
2790 };
2791
2792 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
2793            unsigned long, map_owner)
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 = map;
2803         ri->map_owner = map_owner;
2804
2805         return XDP_REDIRECT;
2806 }
2807
2808 /* Note, arg4 is hidden from users and populated by the verifier
2809  * with the right pointer.
2810  */
2811 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
2812         .func           = bpf_xdp_redirect_map,
2813         .gpl_only       = false,
2814         .ret_type       = RET_INTEGER,
2815         .arg1_type      = ARG_CONST_MAP_PTR,
2816         .arg2_type      = ARG_ANYTHING,
2817         .arg3_type      = ARG_ANYTHING,
2818 };
2819
2820 bool bpf_helper_changes_pkt_data(void *func)
2821 {
2822         if (func == bpf_skb_vlan_push ||
2823             func == bpf_skb_vlan_pop ||
2824             func == bpf_skb_store_bytes ||
2825             func == bpf_skb_change_proto ||
2826             func == bpf_skb_change_head ||
2827             func == bpf_skb_change_tail ||
2828             func == bpf_skb_adjust_room ||
2829             func == bpf_skb_pull_data ||
2830             func == bpf_clone_redirect ||
2831             func == bpf_l3_csum_replace ||
2832             func == bpf_l4_csum_replace ||
2833             func == bpf_xdp_adjust_head ||
2834             func == bpf_xdp_adjust_meta)
2835                 return true;
2836
2837         return false;
2838 }
2839
2840 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
2841                                   unsigned long off, unsigned long len)
2842 {
2843         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
2844
2845         if (unlikely(!ptr))
2846                 return len;
2847         if (ptr != dst_buff)
2848                 memcpy(dst_buff, ptr, len);
2849
2850         return 0;
2851 }
2852
2853 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
2854            u64, flags, void *, meta, u64, meta_size)
2855 {
2856         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2857
2858         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2859                 return -EINVAL;
2860         if (unlikely(skb_size > skb->len))
2861                 return -EFAULT;
2862
2863         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2864                                 bpf_skb_copy);
2865 }
2866
2867 static const struct bpf_func_proto bpf_skb_event_output_proto = {
2868         .func           = bpf_skb_event_output,
2869         .gpl_only       = true,
2870         .ret_type       = RET_INTEGER,
2871         .arg1_type      = ARG_PTR_TO_CTX,
2872         .arg2_type      = ARG_CONST_MAP_PTR,
2873         .arg3_type      = ARG_ANYTHING,
2874         .arg4_type      = ARG_PTR_TO_MEM,
2875         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
2876 };
2877
2878 static unsigned short bpf_tunnel_key_af(u64 flags)
2879 {
2880         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2881 }
2882
2883 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
2884            u32, size, u64, flags)
2885 {
2886         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2887         u8 compat[sizeof(struct bpf_tunnel_key)];
2888         void *to_orig = to;
2889         int err;
2890
2891         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2892                 err = -EINVAL;
2893                 goto err_clear;
2894         }
2895         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2896                 err = -EPROTO;
2897                 goto err_clear;
2898         }
2899         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2900                 err = -EINVAL;
2901                 switch (size) {
2902                 case offsetof(struct bpf_tunnel_key, tunnel_label):
2903                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2904                         goto set_compat;
2905                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2906                         /* Fixup deprecated structure layouts here, so we have
2907                          * a common path later on.
2908                          */
2909                         if (ip_tunnel_info_af(info) != AF_INET)
2910                                 goto err_clear;
2911 set_compat:
2912                         to = (struct bpf_tunnel_key *)compat;
2913                         break;
2914                 default:
2915                         goto err_clear;
2916                 }
2917         }
2918
2919         to->tunnel_id = be64_to_cpu(info->key.tun_id);
2920         to->tunnel_tos = info->key.tos;
2921         to->tunnel_ttl = info->key.ttl;
2922
2923         if (flags & BPF_F_TUNINFO_IPV6) {
2924                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2925                        sizeof(to->remote_ipv6));
2926                 to->tunnel_label = be32_to_cpu(info->key.label);
2927         } else {
2928                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2929         }
2930
2931         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2932                 memcpy(to_orig, to, size);
2933
2934         return 0;
2935 err_clear:
2936         memset(to_orig, 0, size);
2937         return err;
2938 }
2939
2940 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2941         .func           = bpf_skb_get_tunnel_key,
2942         .gpl_only       = false,
2943         .ret_type       = RET_INTEGER,
2944         .arg1_type      = ARG_PTR_TO_CTX,
2945         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
2946         .arg3_type      = ARG_CONST_SIZE,
2947         .arg4_type      = ARG_ANYTHING,
2948 };
2949
2950 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
2951 {
2952         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2953         int err;
2954
2955         if (unlikely(!info ||
2956                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2957                 err = -ENOENT;
2958                 goto err_clear;
2959         }
2960         if (unlikely(size < info->options_len)) {
2961                 err = -ENOMEM;
2962                 goto err_clear;
2963         }
2964
2965         ip_tunnel_info_opts_get(to, info);
2966         if (size > info->options_len)
2967                 memset(to + info->options_len, 0, size - info->options_len);
2968
2969         return info->options_len;
2970 err_clear:
2971         memset(to, 0, size);
2972         return err;
2973 }
2974
2975 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2976         .func           = bpf_skb_get_tunnel_opt,
2977         .gpl_only       = false,
2978         .ret_type       = RET_INTEGER,
2979         .arg1_type      = ARG_PTR_TO_CTX,
2980         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
2981         .arg3_type      = ARG_CONST_SIZE,
2982 };
2983
2984 static struct metadata_dst __percpu *md_dst;
2985
2986 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
2987            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
2988 {
2989         struct metadata_dst *md = this_cpu_ptr(md_dst);
2990         u8 compat[sizeof(struct bpf_tunnel_key)];
2991         struct ip_tunnel_info *info;
2992
2993         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2994                                BPF_F_DONT_FRAGMENT)))
2995                 return -EINVAL;
2996         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2997                 switch (size) {
2998                 case offsetof(struct bpf_tunnel_key, tunnel_label):
2999                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3000                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3001                         /* Fixup deprecated structure layouts here, so we have
3002                          * a common path later on.
3003                          */
3004                         memcpy(compat, from, size);
3005                         memset(compat + size, 0, sizeof(compat) - size);
3006                         from = (const struct bpf_tunnel_key *) compat;
3007                         break;
3008                 default:
3009                         return -EINVAL;
3010                 }
3011         }
3012         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3013                      from->tunnel_ext))
3014                 return -EINVAL;
3015
3016         skb_dst_drop(skb);
3017         dst_hold((struct dst_entry *) md);
3018         skb_dst_set(skb, (struct dst_entry *) md);
3019
3020         info = &md->u.tun_info;
3021         info->mode = IP_TUNNEL_INFO_TX;
3022
3023         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3024         if (flags & BPF_F_DONT_FRAGMENT)
3025                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3026         if (flags & BPF_F_ZERO_CSUM_TX)
3027                 info->key.tun_flags &= ~TUNNEL_CSUM;
3028
3029         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3030         info->key.tos = from->tunnel_tos;
3031         info->key.ttl = from->tunnel_ttl;
3032
3033         if (flags & BPF_F_TUNINFO_IPV6) {
3034                 info->mode |= IP_TUNNEL_INFO_IPV6;
3035                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3036                        sizeof(from->remote_ipv6));
3037                 info->key.label = cpu_to_be32(from->tunnel_label) &
3038                                   IPV6_FLOWLABEL_MASK;
3039         } else {
3040                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3041         }
3042
3043         return 0;
3044 }
3045
3046 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3047         .func           = bpf_skb_set_tunnel_key,
3048         .gpl_only       = false,
3049         .ret_type       = RET_INTEGER,
3050         .arg1_type      = ARG_PTR_TO_CTX,
3051         .arg2_type      = ARG_PTR_TO_MEM,
3052         .arg3_type      = ARG_CONST_SIZE,
3053         .arg4_type      = ARG_ANYTHING,
3054 };
3055
3056 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3057            const u8 *, from, u32, size)
3058 {
3059         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3060         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3061
3062         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3063                 return -EINVAL;
3064         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3065                 return -ENOMEM;
3066
3067         ip_tunnel_info_opts_set(info, from, size);
3068
3069         return 0;
3070 }
3071
3072 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3073         .func           = bpf_skb_set_tunnel_opt,
3074         .gpl_only       = false,
3075         .ret_type       = RET_INTEGER,
3076         .arg1_type      = ARG_PTR_TO_CTX,
3077         .arg2_type      = ARG_PTR_TO_MEM,
3078         .arg3_type      = ARG_CONST_SIZE,
3079 };
3080
3081 static const struct bpf_func_proto *
3082 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3083 {
3084         if (!md_dst) {
3085                 struct metadata_dst __percpu *tmp;
3086
3087                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3088                                                 METADATA_IP_TUNNEL,
3089                                                 GFP_KERNEL);
3090                 if (!tmp)
3091                         return NULL;
3092                 if (cmpxchg(&md_dst, NULL, tmp))
3093                         metadata_dst_free_percpu(tmp);
3094         }
3095
3096         switch (which) {
3097         case BPF_FUNC_skb_set_tunnel_key:
3098                 return &bpf_skb_set_tunnel_key_proto;
3099         case BPF_FUNC_skb_set_tunnel_opt:
3100                 return &bpf_skb_set_tunnel_opt_proto;
3101         default:
3102                 return NULL;
3103         }
3104 }
3105
3106 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3107            u32, idx)
3108 {
3109         struct bpf_array *array = container_of(map, struct bpf_array, map);
3110         struct cgroup *cgrp;
3111         struct sock *sk;
3112
3113         sk = skb_to_full_sk(skb);
3114         if (!sk || !sk_fullsock(sk))
3115                 return -ENOENT;
3116         if (unlikely(idx >= array->map.max_entries))
3117                 return -E2BIG;
3118
3119         cgrp = READ_ONCE(array->ptrs[idx]);
3120         if (unlikely(!cgrp))
3121                 return -EAGAIN;
3122
3123         return sk_under_cgroup_hierarchy(sk, cgrp);
3124 }
3125
3126 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3127         .func           = bpf_skb_under_cgroup,
3128         .gpl_only       = false,
3129         .ret_type       = RET_INTEGER,
3130         .arg1_type      = ARG_PTR_TO_CTX,
3131         .arg2_type      = ARG_CONST_MAP_PTR,
3132         .arg3_type      = ARG_ANYTHING,
3133 };
3134
3135 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3136                                   unsigned long off, unsigned long len)
3137 {
3138         memcpy(dst_buff, src_buff + off, len);
3139         return 0;
3140 }
3141
3142 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3143            u64, flags, void *, meta, u64, meta_size)
3144 {
3145         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3146
3147         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3148                 return -EINVAL;
3149         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3150                 return -EFAULT;
3151
3152         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3153                                 xdp_size, bpf_xdp_copy);
3154 }
3155
3156 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3157         .func           = bpf_xdp_event_output,
3158         .gpl_only       = true,
3159         .ret_type       = RET_INTEGER,
3160         .arg1_type      = ARG_PTR_TO_CTX,
3161         .arg2_type      = ARG_CONST_MAP_PTR,
3162         .arg3_type      = ARG_ANYTHING,
3163         .arg4_type      = ARG_PTR_TO_MEM,
3164         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3165 };
3166
3167 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3168 {
3169         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3170 }
3171
3172 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3173         .func           = bpf_get_socket_cookie,
3174         .gpl_only       = false,
3175         .ret_type       = RET_INTEGER,
3176         .arg1_type      = ARG_PTR_TO_CTX,
3177 };
3178
3179 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3180 {
3181         struct sock *sk = sk_to_full_sk(skb->sk);
3182         kuid_t kuid;
3183
3184         if (!sk || !sk_fullsock(sk))
3185                 return overflowuid;
3186         kuid = sock_net_uid(sock_net(sk), sk);
3187         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3188 }
3189
3190 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3191         .func           = bpf_get_socket_uid,
3192         .gpl_only       = false,
3193         .ret_type       = RET_INTEGER,
3194         .arg1_type      = ARG_PTR_TO_CTX,
3195 };
3196
3197 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3198            int, level, int, optname, char *, optval, int, optlen)
3199 {
3200         struct sock *sk = bpf_sock->sk;
3201         int ret = 0;
3202         int val;
3203
3204         if (!sk_fullsock(sk))
3205                 return -EINVAL;
3206
3207         if (level == SOL_SOCKET) {
3208                 if (optlen != sizeof(int))
3209                         return -EINVAL;
3210                 val = *((int *)optval);
3211
3212                 /* Only some socketops are supported */
3213                 switch (optname) {
3214                 case SO_RCVBUF:
3215                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3216                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3217                         break;
3218                 case SO_SNDBUF:
3219                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3220                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3221                         break;
3222                 case SO_MAX_PACING_RATE:
3223                         sk->sk_max_pacing_rate = val;
3224                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3225                                                  sk->sk_max_pacing_rate);
3226                         break;
3227                 case SO_PRIORITY:
3228                         sk->sk_priority = val;
3229                         break;
3230                 case SO_RCVLOWAT:
3231                         if (val < 0)
3232                                 val = INT_MAX;
3233                         sk->sk_rcvlowat = val ? : 1;
3234                         break;
3235                 case SO_MARK:
3236                         sk->sk_mark = val;
3237                         break;
3238                 default:
3239                         ret = -EINVAL;
3240                 }
3241 #ifdef CONFIG_INET
3242 #if IS_ENABLED(CONFIG_IPV6)
3243         } else if (level == SOL_IPV6) {
3244                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3245                         return -EINVAL;
3246
3247                 val = *((int *)optval);
3248                 /* Only some options are supported */
3249                 switch (optname) {
3250                 case IPV6_TCLASS:
3251                         if (val < -1 || val > 0xff) {
3252                                 ret = -EINVAL;
3253                         } else {
3254                                 struct ipv6_pinfo *np = inet6_sk(sk);
3255
3256                                 if (val == -1)
3257                                         val = 0;
3258                                 np->tclass = val;
3259                         }
3260                         break;
3261                 default:
3262                         ret = -EINVAL;
3263                 }
3264 #endif
3265         } else if (level == SOL_TCP &&
3266                    sk->sk_prot->setsockopt == tcp_setsockopt) {
3267                 if (optname == TCP_CONGESTION) {
3268                         char name[TCP_CA_NAME_MAX];
3269                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3270
3271                         strncpy(name, optval, min_t(long, optlen,
3272                                                     TCP_CA_NAME_MAX-1));
3273                         name[TCP_CA_NAME_MAX-1] = 0;
3274                         ret = tcp_set_congestion_control(sk, name, false,
3275                                                          reinit);
3276                 } else {
3277                         struct tcp_sock *tp = tcp_sk(sk);
3278
3279                         if (optlen != sizeof(int))
3280                                 return -EINVAL;
3281
3282                         val = *((int *)optval);
3283                         /* Only some options are supported */
3284                         switch (optname) {
3285                         case TCP_BPF_IW:
3286                                 if (val <= 0 || tp->data_segs_out > 0)
3287                                         ret = -EINVAL;
3288                                 else
3289                                         tp->snd_cwnd = val;
3290                                 break;
3291                         case TCP_BPF_SNDCWND_CLAMP:
3292                                 if (val <= 0) {
3293                                         ret = -EINVAL;
3294                                 } else {
3295                                         tp->snd_cwnd_clamp = val;
3296                                         tp->snd_ssthresh = val;
3297                                 }
3298                                 break;
3299                         default:
3300                                 ret = -EINVAL;
3301                         }
3302                 }
3303 #endif
3304         } else {
3305                 ret = -EINVAL;
3306         }
3307         return ret;
3308 }
3309
3310 static const struct bpf_func_proto bpf_setsockopt_proto = {
3311         .func           = bpf_setsockopt,
3312         .gpl_only       = false,
3313         .ret_type       = RET_INTEGER,
3314         .arg1_type      = ARG_PTR_TO_CTX,
3315         .arg2_type      = ARG_ANYTHING,
3316         .arg3_type      = ARG_ANYTHING,
3317         .arg4_type      = ARG_PTR_TO_MEM,
3318         .arg5_type      = ARG_CONST_SIZE,
3319 };
3320
3321 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3322            int, level, int, optname, char *, optval, int, optlen)
3323 {
3324         struct sock *sk = bpf_sock->sk;
3325
3326         if (!sk_fullsock(sk))
3327                 goto err_clear;
3328
3329 #ifdef CONFIG_INET
3330         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
3331                 if (optname == TCP_CONGESTION) {
3332                         struct inet_connection_sock *icsk = inet_csk(sk);
3333
3334                         if (!icsk->icsk_ca_ops || optlen <= 1)
3335                                 goto err_clear;
3336                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
3337                         optval[optlen - 1] = 0;
3338                 } else {
3339                         goto err_clear;
3340                 }
3341 #if IS_ENABLED(CONFIG_IPV6)
3342         } else if (level == SOL_IPV6) {
3343                 struct ipv6_pinfo *np = inet6_sk(sk);
3344
3345                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3346                         goto err_clear;
3347
3348                 /* Only some options are supported */
3349                 switch (optname) {
3350                 case IPV6_TCLASS:
3351                         *((int *)optval) = (int)np->tclass;
3352                         break;
3353                 default:
3354                         goto err_clear;
3355                 }
3356 #endif
3357         } else {
3358                 goto err_clear;
3359         }
3360         return 0;
3361 #endif
3362 err_clear:
3363         memset(optval, 0, optlen);
3364         return -EINVAL;
3365 }
3366
3367 static const struct bpf_func_proto bpf_getsockopt_proto = {
3368         .func           = bpf_getsockopt,
3369         .gpl_only       = false,
3370         .ret_type       = RET_INTEGER,
3371         .arg1_type      = ARG_PTR_TO_CTX,
3372         .arg2_type      = ARG_ANYTHING,
3373         .arg3_type      = ARG_ANYTHING,
3374         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
3375         .arg5_type      = ARG_CONST_SIZE,
3376 };
3377
3378 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
3379            int, argval)
3380 {
3381         struct sock *sk = bpf_sock->sk;
3382         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
3383
3384         if (!sk_fullsock(sk))
3385                 return -EINVAL;
3386
3387 #ifdef CONFIG_INET
3388         if (val)
3389                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
3390
3391         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
3392 #else
3393         return -EINVAL;
3394 #endif
3395 }
3396
3397 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
3398         .func           = bpf_sock_ops_cb_flags_set,
3399         .gpl_only       = false,
3400         .ret_type       = RET_INTEGER,
3401         .arg1_type      = ARG_PTR_TO_CTX,
3402         .arg2_type      = ARG_ANYTHING,
3403 };
3404
3405 static const struct bpf_func_proto *
3406 bpf_base_func_proto(enum bpf_func_id func_id)
3407 {
3408         switch (func_id) {
3409         case BPF_FUNC_map_lookup_elem:
3410                 return &bpf_map_lookup_elem_proto;
3411         case BPF_FUNC_map_update_elem:
3412                 return &bpf_map_update_elem_proto;
3413         case BPF_FUNC_map_delete_elem:
3414                 return &bpf_map_delete_elem_proto;
3415         case BPF_FUNC_get_prandom_u32:
3416                 return &bpf_get_prandom_u32_proto;
3417         case BPF_FUNC_get_smp_processor_id:
3418                 return &bpf_get_raw_smp_processor_id_proto;
3419         case BPF_FUNC_get_numa_node_id:
3420                 return &bpf_get_numa_node_id_proto;
3421         case BPF_FUNC_tail_call:
3422                 return &bpf_tail_call_proto;
3423         case BPF_FUNC_ktime_get_ns:
3424                 return &bpf_ktime_get_ns_proto;
3425         case BPF_FUNC_trace_printk:
3426                 if (capable(CAP_SYS_ADMIN))
3427                         return bpf_get_trace_printk_proto();
3428         default:
3429                 return NULL;
3430         }
3431 }
3432
3433 static const struct bpf_func_proto *
3434 sock_filter_func_proto(enum bpf_func_id func_id)
3435 {
3436         switch (func_id) {
3437         /* inet and inet6 sockets are created in a process
3438          * context so there is always a valid uid/gid
3439          */
3440         case BPF_FUNC_get_current_uid_gid:
3441                 return &bpf_get_current_uid_gid_proto;
3442         default:
3443                 return bpf_base_func_proto(func_id);
3444         }
3445 }
3446
3447 static const struct bpf_func_proto *
3448 sk_filter_func_proto(enum bpf_func_id func_id)
3449 {
3450         switch (func_id) {
3451         case BPF_FUNC_skb_load_bytes:
3452                 return &bpf_skb_load_bytes_proto;
3453         case BPF_FUNC_get_socket_cookie:
3454                 return &bpf_get_socket_cookie_proto;
3455         case BPF_FUNC_get_socket_uid:
3456                 return &bpf_get_socket_uid_proto;
3457         default:
3458                 return bpf_base_func_proto(func_id);
3459         }
3460 }
3461
3462 static const struct bpf_func_proto *
3463 tc_cls_act_func_proto(enum bpf_func_id func_id)
3464 {
3465         switch (func_id) {
3466         case BPF_FUNC_skb_store_bytes:
3467                 return &bpf_skb_store_bytes_proto;
3468         case BPF_FUNC_skb_load_bytes:
3469                 return &bpf_skb_load_bytes_proto;
3470         case BPF_FUNC_skb_pull_data:
3471                 return &bpf_skb_pull_data_proto;
3472         case BPF_FUNC_csum_diff:
3473                 return &bpf_csum_diff_proto;
3474         case BPF_FUNC_csum_update:
3475                 return &bpf_csum_update_proto;
3476         case BPF_FUNC_l3_csum_replace:
3477                 return &bpf_l3_csum_replace_proto;
3478         case BPF_FUNC_l4_csum_replace:
3479                 return &bpf_l4_csum_replace_proto;
3480         case BPF_FUNC_clone_redirect:
3481                 return &bpf_clone_redirect_proto;
3482         case BPF_FUNC_get_cgroup_classid:
3483                 return &bpf_get_cgroup_classid_proto;
3484         case BPF_FUNC_skb_vlan_push:
3485                 return &bpf_skb_vlan_push_proto;
3486         case BPF_FUNC_skb_vlan_pop:
3487                 return &bpf_skb_vlan_pop_proto;
3488         case BPF_FUNC_skb_change_proto:
3489                 return &bpf_skb_change_proto_proto;
3490         case BPF_FUNC_skb_change_type:
3491                 return &bpf_skb_change_type_proto;
3492         case BPF_FUNC_skb_adjust_room:
3493                 return &bpf_skb_adjust_room_proto;
3494         case BPF_FUNC_skb_change_tail:
3495                 return &bpf_skb_change_tail_proto;
3496         case BPF_FUNC_skb_get_tunnel_key:
3497                 return &bpf_skb_get_tunnel_key_proto;
3498         case BPF_FUNC_skb_set_tunnel_key:
3499                 return bpf_get_skb_set_tunnel_proto(func_id);
3500         case BPF_FUNC_skb_get_tunnel_opt:
3501                 return &bpf_skb_get_tunnel_opt_proto;
3502         case BPF_FUNC_skb_set_tunnel_opt:
3503                 return bpf_get_skb_set_tunnel_proto(func_id);
3504         case BPF_FUNC_redirect:
3505                 return &bpf_redirect_proto;
3506         case BPF_FUNC_get_route_realm:
3507                 return &bpf_get_route_realm_proto;
3508         case BPF_FUNC_get_hash_recalc:
3509                 return &bpf_get_hash_recalc_proto;
3510         case BPF_FUNC_set_hash_invalid:
3511                 return &bpf_set_hash_invalid_proto;
3512         case BPF_FUNC_set_hash:
3513                 return &bpf_set_hash_proto;
3514         case BPF_FUNC_perf_event_output:
3515                 return &bpf_skb_event_output_proto;
3516         case BPF_FUNC_get_smp_processor_id:
3517                 return &bpf_get_smp_processor_id_proto;
3518         case BPF_FUNC_skb_under_cgroup:
3519                 return &bpf_skb_under_cgroup_proto;
3520         case BPF_FUNC_get_socket_cookie:
3521                 return &bpf_get_socket_cookie_proto;
3522         case BPF_FUNC_get_socket_uid:
3523                 return &bpf_get_socket_uid_proto;
3524         default:
3525                 return bpf_base_func_proto(func_id);
3526         }
3527 }
3528
3529 static const struct bpf_func_proto *
3530 xdp_func_proto(enum bpf_func_id func_id)
3531 {
3532         switch (func_id) {
3533         case BPF_FUNC_perf_event_output:
3534                 return &bpf_xdp_event_output_proto;
3535         case BPF_FUNC_get_smp_processor_id:
3536                 return &bpf_get_smp_processor_id_proto;
3537         case BPF_FUNC_csum_diff:
3538                 return &bpf_csum_diff_proto;
3539         case BPF_FUNC_xdp_adjust_head:
3540                 return &bpf_xdp_adjust_head_proto;
3541         case BPF_FUNC_xdp_adjust_meta:
3542                 return &bpf_xdp_adjust_meta_proto;
3543         case BPF_FUNC_redirect:
3544                 return &bpf_xdp_redirect_proto;
3545         case BPF_FUNC_redirect_map:
3546                 return &bpf_xdp_redirect_map_proto;
3547         default:
3548                 return bpf_base_func_proto(func_id);
3549         }
3550 }
3551
3552 static const struct bpf_func_proto *
3553 lwt_inout_func_proto(enum bpf_func_id func_id)
3554 {
3555         switch (func_id) {
3556         case BPF_FUNC_skb_load_bytes:
3557                 return &bpf_skb_load_bytes_proto;
3558         case BPF_FUNC_skb_pull_data:
3559                 return &bpf_skb_pull_data_proto;
3560         case BPF_FUNC_csum_diff:
3561                 return &bpf_csum_diff_proto;
3562         case BPF_FUNC_get_cgroup_classid:
3563                 return &bpf_get_cgroup_classid_proto;
3564         case BPF_FUNC_get_route_realm:
3565                 return &bpf_get_route_realm_proto;
3566         case BPF_FUNC_get_hash_recalc:
3567                 return &bpf_get_hash_recalc_proto;
3568         case BPF_FUNC_perf_event_output:
3569                 return &bpf_skb_event_output_proto;
3570         case BPF_FUNC_get_smp_processor_id:
3571                 return &bpf_get_smp_processor_id_proto;
3572         case BPF_FUNC_skb_under_cgroup:
3573                 return &bpf_skb_under_cgroup_proto;
3574         default:
3575                 return bpf_base_func_proto(func_id);
3576         }
3577 }
3578
3579 static const struct bpf_func_proto *
3580         sock_ops_func_proto(enum bpf_func_id func_id)
3581 {
3582         switch (func_id) {
3583         case BPF_FUNC_setsockopt:
3584                 return &bpf_setsockopt_proto;
3585         case BPF_FUNC_getsockopt:
3586                 return &bpf_getsockopt_proto;
3587         case BPF_FUNC_sock_ops_cb_flags_set:
3588                 return &bpf_sock_ops_cb_flags_set_proto;
3589         case BPF_FUNC_sock_map_update:
3590                 return &bpf_sock_map_update_proto;
3591         default:
3592                 return bpf_base_func_proto(func_id);
3593         }
3594 }
3595
3596 static const struct bpf_func_proto *sk_skb_func_proto(enum bpf_func_id func_id)
3597 {
3598         switch (func_id) {
3599         case BPF_FUNC_skb_store_bytes:
3600                 return &bpf_skb_store_bytes_proto;
3601         case BPF_FUNC_skb_load_bytes:
3602                 return &bpf_skb_load_bytes_proto;
3603         case BPF_FUNC_skb_pull_data:
3604                 return &bpf_skb_pull_data_proto;
3605         case BPF_FUNC_skb_change_tail:
3606                 return &bpf_skb_change_tail_proto;
3607         case BPF_FUNC_skb_change_head:
3608                 return &bpf_skb_change_head_proto;
3609         case BPF_FUNC_get_socket_cookie:
3610                 return &bpf_get_socket_cookie_proto;
3611         case BPF_FUNC_get_socket_uid:
3612                 return &bpf_get_socket_uid_proto;
3613         case BPF_FUNC_sk_redirect_map:
3614                 return &bpf_sk_redirect_map_proto;
3615         default:
3616                 return bpf_base_func_proto(func_id);
3617         }
3618 }
3619
3620 static const struct bpf_func_proto *
3621 lwt_xmit_func_proto(enum bpf_func_id func_id)
3622 {
3623         switch (func_id) {
3624         case BPF_FUNC_skb_get_tunnel_key:
3625                 return &bpf_skb_get_tunnel_key_proto;
3626         case BPF_FUNC_skb_set_tunnel_key:
3627                 return bpf_get_skb_set_tunnel_proto(func_id);
3628         case BPF_FUNC_skb_get_tunnel_opt:
3629                 return &bpf_skb_get_tunnel_opt_proto;
3630         case BPF_FUNC_skb_set_tunnel_opt:
3631                 return bpf_get_skb_set_tunnel_proto(func_id);
3632         case BPF_FUNC_redirect:
3633                 return &bpf_redirect_proto;
3634         case BPF_FUNC_clone_redirect:
3635                 return &bpf_clone_redirect_proto;
3636         case BPF_FUNC_skb_change_tail:
3637                 return &bpf_skb_change_tail_proto;
3638         case BPF_FUNC_skb_change_head:
3639                 return &bpf_skb_change_head_proto;
3640         case BPF_FUNC_skb_store_bytes:
3641                 return &bpf_skb_store_bytes_proto;
3642         case BPF_FUNC_csum_update:
3643                 return &bpf_csum_update_proto;
3644         case BPF_FUNC_l3_csum_replace:
3645                 return &bpf_l3_csum_replace_proto;
3646         case BPF_FUNC_l4_csum_replace:
3647                 return &bpf_l4_csum_replace_proto;
3648         case BPF_FUNC_set_hash_invalid:
3649                 return &bpf_set_hash_invalid_proto;
3650         default:
3651                 return lwt_inout_func_proto(func_id);
3652         }
3653 }
3654
3655 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
3656                                     struct bpf_insn_access_aux *info)
3657 {
3658         const int size_default = sizeof(__u32);
3659
3660         if (off < 0 || off >= sizeof(struct __sk_buff))
3661                 return false;
3662
3663         /* The verifier guarantees that size > 0. */
3664         if (off % size != 0)
3665                 return false;
3666
3667         switch (off) {
3668         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3669                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
3670                         return false;
3671                 break;
3672         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
3673         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
3674         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
3675         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
3676         case bpf_ctx_range(struct __sk_buff, data):
3677         case bpf_ctx_range(struct __sk_buff, data_meta):
3678         case bpf_ctx_range(struct __sk_buff, data_end):
3679                 if (size != size_default)
3680                         return false;
3681                 break;
3682         default:
3683                 /* Only narrow read access allowed for now. */
3684                 if (type == BPF_WRITE) {
3685                         if (size != size_default)
3686                                 return false;
3687                 } else {
3688                         bpf_ctx_record_field_size(info, size_default);
3689                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
3690                                 return false;
3691                 }
3692         }
3693
3694         return true;
3695 }
3696
3697 static bool sk_filter_is_valid_access(int off, int size,
3698                                       enum bpf_access_type type,
3699                                       struct bpf_insn_access_aux *info)
3700 {
3701         switch (off) {
3702         case bpf_ctx_range(struct __sk_buff, tc_classid):
3703         case bpf_ctx_range(struct __sk_buff, data):
3704         case bpf_ctx_range(struct __sk_buff, data_meta):
3705         case bpf_ctx_range(struct __sk_buff, data_end):
3706         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3707                 return false;
3708         }
3709
3710         if (type == BPF_WRITE) {
3711                 switch (off) {
3712                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3713                         break;
3714                 default:
3715                         return false;
3716                 }
3717         }
3718
3719         return bpf_skb_is_valid_access(off, size, type, info);
3720 }
3721
3722 static bool lwt_is_valid_access(int off, int size,
3723                                 enum bpf_access_type type,
3724                                 struct bpf_insn_access_aux *info)
3725 {
3726         switch (off) {
3727         case bpf_ctx_range(struct __sk_buff, tc_classid):
3728         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3729         case bpf_ctx_range(struct __sk_buff, data_meta):
3730                 return false;
3731         }
3732
3733         if (type == BPF_WRITE) {
3734                 switch (off) {
3735                 case bpf_ctx_range(struct __sk_buff, mark):
3736                 case bpf_ctx_range(struct __sk_buff, priority):
3737                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3738                         break;
3739                 default:
3740                         return false;
3741                 }
3742         }
3743
3744         switch (off) {
3745         case bpf_ctx_range(struct __sk_buff, data):
3746                 info->reg_type = PTR_TO_PACKET;
3747                 break;
3748         case bpf_ctx_range(struct __sk_buff, data_end):
3749                 info->reg_type = PTR_TO_PACKET_END;
3750                 break;
3751         }
3752
3753         return bpf_skb_is_valid_access(off, size, type, info);
3754 }
3755
3756 static bool sock_filter_is_valid_access(int off, int size,
3757                                         enum bpf_access_type type,
3758                                         struct bpf_insn_access_aux *info)
3759 {
3760         if (type == BPF_WRITE) {
3761                 switch (off) {
3762                 case offsetof(struct bpf_sock, bound_dev_if):
3763                 case offsetof(struct bpf_sock, mark):
3764                 case offsetof(struct bpf_sock, priority):
3765                         break;
3766                 default:
3767                         return false;
3768                 }
3769         }
3770
3771         if (off < 0 || off + size > sizeof(struct bpf_sock))
3772                 return false;
3773         /* The verifier guarantees that size > 0. */
3774         if (off % size != 0)
3775                 return false;
3776         if (size != sizeof(__u32))
3777                 return false;
3778
3779         return true;
3780 }
3781
3782 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
3783                                 const struct bpf_prog *prog, int drop_verdict)
3784 {
3785         struct bpf_insn *insn = insn_buf;
3786
3787         if (!direct_write)
3788                 return 0;
3789
3790         /* if (!skb->cloned)
3791          *       goto start;
3792          *
3793          * (Fast-path, otherwise approximation that we might be
3794          *  a clone, do the rest in helper.)
3795          */
3796         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
3797         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
3798         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
3799
3800         /* ret = bpf_skb_pull_data(skb, 0); */
3801         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
3802         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
3803         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
3804                                BPF_FUNC_skb_pull_data);
3805         /* if (!ret)
3806          *      goto restore;
3807          * return TC_ACT_SHOT;
3808          */
3809         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
3810         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
3811         *insn++ = BPF_EXIT_INSN();
3812
3813         /* restore: */
3814         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
3815         /* start: */
3816         *insn++ = prog->insnsi[0];
3817
3818         return insn - insn_buf;
3819 }
3820
3821 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
3822                                const struct bpf_prog *prog)
3823 {
3824         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
3825 }
3826
3827 static bool tc_cls_act_is_valid_access(int off, int size,
3828                                        enum bpf_access_type type,
3829                                        struct bpf_insn_access_aux *info)
3830 {
3831         if (type == BPF_WRITE) {
3832                 switch (off) {
3833                 case bpf_ctx_range(struct __sk_buff, mark):
3834                 case bpf_ctx_range(struct __sk_buff, tc_index):
3835                 case bpf_ctx_range(struct __sk_buff, priority):
3836                 case bpf_ctx_range(struct __sk_buff, tc_classid):
3837                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3838                         break;
3839                 default:
3840                         return false;
3841                 }
3842         }
3843
3844         switch (off) {
3845         case bpf_ctx_range(struct __sk_buff, data):
3846                 info->reg_type = PTR_TO_PACKET;
3847                 break;
3848         case bpf_ctx_range(struct __sk_buff, data_meta):
3849                 info->reg_type = PTR_TO_PACKET_META;
3850                 break;
3851         case bpf_ctx_range(struct __sk_buff, data_end):
3852                 info->reg_type = PTR_TO_PACKET_END;
3853                 break;
3854         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3855                 return false;
3856         }
3857
3858         return bpf_skb_is_valid_access(off, size, type, info);
3859 }
3860
3861 static bool __is_valid_xdp_access(int off, int size)
3862 {
3863         if (off < 0 || off >= sizeof(struct xdp_md))
3864                 return false;
3865         if (off % size != 0)
3866                 return false;
3867         if (size != sizeof(__u32))
3868                 return false;
3869
3870         return true;
3871 }
3872
3873 static bool xdp_is_valid_access(int off, int size,
3874                                 enum bpf_access_type type,
3875                                 struct bpf_insn_access_aux *info)
3876 {
3877         if (type == BPF_WRITE)
3878                 return false;
3879
3880         switch (off) {
3881         case offsetof(struct xdp_md, data):
3882                 info->reg_type = PTR_TO_PACKET;
3883                 break;
3884         case offsetof(struct xdp_md, data_meta):
3885                 info->reg_type = PTR_TO_PACKET_META;
3886                 break;
3887         case offsetof(struct xdp_md, data_end):
3888                 info->reg_type = PTR_TO_PACKET_END;
3889                 break;
3890         }
3891
3892         return __is_valid_xdp_access(off, size);
3893 }
3894
3895 void bpf_warn_invalid_xdp_action(u32 act)
3896 {
3897         const u32 act_max = XDP_REDIRECT;
3898
3899         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
3900                   act > act_max ? "Illegal" : "Driver unsupported",
3901                   act);
3902 }
3903 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
3904
3905 static bool sock_ops_is_valid_access(int off, int size,
3906                                      enum bpf_access_type type,
3907                                      struct bpf_insn_access_aux *info)
3908 {
3909         const int size_default = sizeof(__u32);
3910
3911         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
3912                 return false;
3913
3914         /* The verifier guarantees that size > 0. */
3915         if (off % size != 0)
3916                 return false;
3917
3918         if (type == BPF_WRITE) {
3919                 switch (off) {
3920                 case offsetof(struct bpf_sock_ops, reply):
3921                 case offsetof(struct bpf_sock_ops, sk_txhash):
3922                         if (size != size_default)
3923                                 return false;
3924                         break;
3925                 default:
3926                         return false;
3927                 }
3928         } else {
3929                 switch (off) {
3930                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
3931                                         bytes_acked):
3932                         if (size != sizeof(__u64))
3933                                 return false;
3934                         break;
3935                 default:
3936                         if (size != size_default)
3937                                 return false;
3938                         break;
3939                 }
3940         }
3941
3942         return true;
3943 }
3944
3945 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
3946                            const struct bpf_prog *prog)
3947 {
3948         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
3949 }
3950
3951 static bool sk_skb_is_valid_access(int off, int size,
3952                                    enum bpf_access_type type,
3953                                    struct bpf_insn_access_aux *info)
3954 {
3955         switch (off) {
3956         case bpf_ctx_range(struct __sk_buff, tc_classid):
3957         case bpf_ctx_range(struct __sk_buff, data_meta):
3958                 return false;
3959         }
3960
3961         if (type == BPF_WRITE) {
3962                 switch (off) {
3963                 case bpf_ctx_range(struct __sk_buff, tc_index):
3964                 case bpf_ctx_range(struct __sk_buff, priority):
3965                         break;
3966                 default:
3967                         return false;
3968                 }
3969         }
3970
3971         switch (off) {
3972         case bpf_ctx_range(struct __sk_buff, mark):
3973                 return false;
3974         case bpf_ctx_range(struct __sk_buff, data):
3975                 info->reg_type = PTR_TO_PACKET;
3976                 break;
3977         case bpf_ctx_range(struct __sk_buff, data_end):
3978                 info->reg_type = PTR_TO_PACKET_END;
3979                 break;
3980         }
3981
3982         return bpf_skb_is_valid_access(off, size, type, info);
3983 }
3984
3985 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
3986                                   const struct bpf_insn *si,
3987                                   struct bpf_insn *insn_buf,
3988                                   struct bpf_prog *prog, u32 *target_size)
3989 {
3990         struct bpf_insn *insn = insn_buf;
3991         int off;
3992
3993         switch (si->off) {
3994         case offsetof(struct __sk_buff, len):
3995                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3996                                       bpf_target_off(struct sk_buff, len, 4,
3997                                                      target_size));
3998                 break;
3999
4000         case offsetof(struct __sk_buff, protocol):
4001                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4002                                       bpf_target_off(struct sk_buff, protocol, 2,
4003                                                      target_size));
4004                 break;
4005
4006         case offsetof(struct __sk_buff, vlan_proto):
4007                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4008                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
4009                                                      target_size));
4010                 break;
4011
4012         case offsetof(struct __sk_buff, priority):
4013                 if (type == BPF_WRITE)
4014                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4015                                               bpf_target_off(struct sk_buff, priority, 4,
4016                                                              target_size));
4017                 else
4018                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4019                                               bpf_target_off(struct sk_buff, priority, 4,
4020                                                              target_size));
4021                 break;
4022
4023         case offsetof(struct __sk_buff, ingress_ifindex):
4024                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4025                                       bpf_target_off(struct sk_buff, skb_iif, 4,
4026                                                      target_size));
4027                 break;
4028
4029         case offsetof(struct __sk_buff, ifindex):
4030                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4031                                       si->dst_reg, si->src_reg,
4032                                       offsetof(struct sk_buff, dev));
4033                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
4034                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4035                                       bpf_target_off(struct net_device, ifindex, 4,
4036                                                      target_size));
4037                 break;
4038
4039         case offsetof(struct __sk_buff, hash):
4040                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4041                                       bpf_target_off(struct sk_buff, hash, 4,
4042                                                      target_size));
4043                 break;
4044
4045         case offsetof(struct __sk_buff, mark):
4046                 if (type == BPF_WRITE)
4047                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4048                                               bpf_target_off(struct sk_buff, mark, 4,
4049                                                              target_size));
4050                 else
4051                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4052                                               bpf_target_off(struct sk_buff, mark, 4,
4053                                                              target_size));
4054                 break;
4055
4056         case offsetof(struct __sk_buff, pkt_type):
4057                 *target_size = 1;
4058                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
4059                                       PKT_TYPE_OFFSET());
4060                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
4061 #ifdef __BIG_ENDIAN_BITFIELD
4062                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
4063 #endif
4064                 break;
4065
4066         case offsetof(struct __sk_buff, queue_mapping):
4067                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4068                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
4069                                                      target_size));
4070                 break;
4071
4072         case offsetof(struct __sk_buff, vlan_present):
4073         case offsetof(struct __sk_buff, vlan_tci):
4074                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
4075
4076                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4077                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
4078                                                      target_size));
4079                 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
4080                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
4081                                                 ~VLAN_TAG_PRESENT);
4082                 } else {
4083                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
4084                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
4085                 }
4086                 break;
4087
4088         case offsetof(struct __sk_buff, cb[0]) ...
4089              offsetofend(struct __sk_buff, cb[4]) - 1:
4090                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
4091                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
4092                               offsetof(struct qdisc_skb_cb, data)) %
4093                              sizeof(__u64));
4094
4095                 prog->cb_access = 1;
4096                 off  = si->off;
4097                 off -= offsetof(struct __sk_buff, cb[0]);
4098                 off += offsetof(struct sk_buff, cb);
4099                 off += offsetof(struct qdisc_skb_cb, data);
4100                 if (type == BPF_WRITE)
4101                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
4102                                               si->src_reg, off);
4103                 else
4104                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
4105                                               si->src_reg, off);
4106                 break;
4107
4108         case offsetof(struct __sk_buff, tc_classid):
4109                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
4110
4111                 off  = si->off;
4112                 off -= offsetof(struct __sk_buff, tc_classid);
4113                 off += offsetof(struct sk_buff, cb);
4114                 off += offsetof(struct qdisc_skb_cb, tc_classid);
4115                 *target_size = 2;
4116                 if (type == BPF_WRITE)
4117                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
4118                                               si->src_reg, off);
4119                 else
4120                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
4121                                               si->src_reg, off);
4122                 break;
4123
4124         case offsetof(struct __sk_buff, data):
4125                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
4126                                       si->dst_reg, si->src_reg,
4127                                       offsetof(struct sk_buff, data));
4128                 break;
4129
4130         case offsetof(struct __sk_buff, data_meta):
4131                 off  = si->off;
4132                 off -= offsetof(struct __sk_buff, data_meta);
4133                 off += offsetof(struct sk_buff, cb);
4134                 off += offsetof(struct bpf_skb_data_end, data_meta);
4135                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4136                                       si->src_reg, off);
4137                 break;
4138
4139         case offsetof(struct __sk_buff, data_end):
4140                 off  = si->off;
4141                 off -= offsetof(struct __sk_buff, data_end);
4142                 off += offsetof(struct sk_buff, cb);
4143                 off += offsetof(struct bpf_skb_data_end, data_end);
4144                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4145                                       si->src_reg, off);
4146                 break;
4147
4148         case offsetof(struct __sk_buff, tc_index):
4149 #ifdef CONFIG_NET_SCHED
4150                 if (type == BPF_WRITE)
4151                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
4152                                               bpf_target_off(struct sk_buff, tc_index, 2,
4153                                                              target_size));
4154                 else
4155                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4156                                               bpf_target_off(struct sk_buff, tc_index, 2,
4157                                                              target_size));
4158 #else
4159                 *target_size = 2;
4160                 if (type == BPF_WRITE)
4161                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
4162                 else
4163                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4164 #endif
4165                 break;
4166
4167         case offsetof(struct __sk_buff, napi_id):
4168 #if defined(CONFIG_NET_RX_BUSY_POLL)
4169                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4170                                       bpf_target_off(struct sk_buff, napi_id, 4,
4171                                                      target_size));
4172                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
4173                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4174 #else
4175                 *target_size = 4;
4176                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4177 #endif
4178                 break;
4179         case offsetof(struct __sk_buff, family):
4180                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4181
4182                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4183                                       si->dst_reg, si->src_reg,
4184                                       offsetof(struct sk_buff, sk));
4185                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4186                                       bpf_target_off(struct sock_common,
4187                                                      skc_family,
4188                                                      2, target_size));
4189                 break;
4190         case offsetof(struct __sk_buff, remote_ip4):
4191                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4192
4193                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4194                                       si->dst_reg, si->src_reg,
4195                                       offsetof(struct sk_buff, sk));
4196                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4197                                       bpf_target_off(struct sock_common,
4198                                                      skc_daddr,
4199                                                      4, target_size));
4200                 break;
4201         case offsetof(struct __sk_buff, local_ip4):
4202                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4203                                           skc_rcv_saddr) != 4);
4204
4205                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4206                                       si->dst_reg, si->src_reg,
4207                                       offsetof(struct sk_buff, sk));
4208                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4209                                       bpf_target_off(struct sock_common,
4210                                                      skc_rcv_saddr,
4211                                                      4, target_size));
4212                 break;
4213         case offsetof(struct __sk_buff, remote_ip6[0]) ...
4214              offsetof(struct __sk_buff, remote_ip6[3]):
4215 #if IS_ENABLED(CONFIG_IPV6)
4216                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4217                                           skc_v6_daddr.s6_addr32[0]) != 4);
4218
4219                 off = si->off;
4220                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
4221
4222                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4223                                       si->dst_reg, si->src_reg,
4224                                       offsetof(struct sk_buff, sk));
4225                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4226                                       offsetof(struct sock_common,
4227                                                skc_v6_daddr.s6_addr32[0]) +
4228                                       off);
4229 #else
4230                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4231 #endif
4232                 break;
4233         case offsetof(struct __sk_buff, local_ip6[0]) ...
4234              offsetof(struct __sk_buff, local_ip6[3]):
4235 #if IS_ENABLED(CONFIG_IPV6)
4236                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4237                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4238
4239                 off = si->off;
4240                 off -= offsetof(struct __sk_buff, local_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_rcv_saddr.s6_addr32[0]) +
4248                                       off);
4249 #else
4250                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4251 #endif
4252                 break;
4253
4254         case offsetof(struct __sk_buff, remote_port):
4255                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4256
4257                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4258                                       si->dst_reg, si->src_reg,
4259                                       offsetof(struct sk_buff, sk));
4260                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4261                                       bpf_target_off(struct sock_common,
4262                                                      skc_dport,
4263                                                      2, target_size));
4264 #ifndef __BIG_ENDIAN_BITFIELD
4265                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4266 #endif
4267                 break;
4268
4269         case offsetof(struct __sk_buff, local_port):
4270                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4271
4272                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4273                                       si->dst_reg, si->src_reg,
4274                                       offsetof(struct sk_buff, sk));
4275                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4276                                       bpf_target_off(struct sock_common,
4277                                                      skc_num, 2, target_size));
4278                 break;
4279         }
4280
4281         return insn - insn_buf;
4282 }
4283
4284 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
4285                                           const struct bpf_insn *si,
4286                                           struct bpf_insn *insn_buf,
4287                                           struct bpf_prog *prog, u32 *target_size)
4288 {
4289         struct bpf_insn *insn = insn_buf;
4290
4291         switch (si->off) {
4292         case offsetof(struct bpf_sock, bound_dev_if):
4293                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
4294
4295                 if (type == BPF_WRITE)
4296                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4297                                         offsetof(struct sock, sk_bound_dev_if));
4298                 else
4299                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4300                                       offsetof(struct sock, sk_bound_dev_if));
4301                 break;
4302
4303         case offsetof(struct bpf_sock, mark):
4304                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
4305
4306                 if (type == BPF_WRITE)
4307                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4308                                         offsetof(struct sock, sk_mark));
4309                 else
4310                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4311                                       offsetof(struct sock, sk_mark));
4312                 break;
4313
4314         case offsetof(struct bpf_sock, priority):
4315                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
4316
4317                 if (type == BPF_WRITE)
4318                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4319                                         offsetof(struct sock, sk_priority));
4320                 else
4321                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4322                                       offsetof(struct sock, sk_priority));
4323                 break;
4324
4325         case offsetof(struct bpf_sock, family):
4326                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
4327
4328                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4329                                       offsetof(struct sock, sk_family));
4330                 break;
4331
4332         case offsetof(struct bpf_sock, type):
4333                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4334                                       offsetof(struct sock, __sk_flags_offset));
4335                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
4336                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
4337                 break;
4338
4339         case offsetof(struct bpf_sock, protocol):
4340                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4341                                       offsetof(struct sock, __sk_flags_offset));
4342                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
4343                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
4344                 break;
4345         }
4346
4347         return insn - insn_buf;
4348 }
4349
4350 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
4351                                          const struct bpf_insn *si,
4352                                          struct bpf_insn *insn_buf,
4353                                          struct bpf_prog *prog, u32 *target_size)
4354 {
4355         struct bpf_insn *insn = insn_buf;
4356
4357         switch (si->off) {
4358         case offsetof(struct __sk_buff, ifindex):
4359                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4360                                       si->dst_reg, si->src_reg,
4361                                       offsetof(struct sk_buff, dev));
4362                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4363                                       bpf_target_off(struct net_device, ifindex, 4,
4364                                                      target_size));
4365                 break;
4366         default:
4367                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4368                                               target_size);
4369         }
4370
4371         return insn - insn_buf;
4372 }
4373
4374 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
4375                                   const struct bpf_insn *si,
4376                                   struct bpf_insn *insn_buf,
4377                                   struct bpf_prog *prog, u32 *target_size)
4378 {
4379         struct bpf_insn *insn = insn_buf;
4380
4381         switch (si->off) {
4382         case offsetof(struct xdp_md, data):
4383                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
4384                                       si->dst_reg, si->src_reg,
4385                                       offsetof(struct xdp_buff, data));
4386                 break;
4387         case offsetof(struct xdp_md, data_meta):
4388                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
4389                                       si->dst_reg, si->src_reg,
4390                                       offsetof(struct xdp_buff, data_meta));
4391                 break;
4392         case offsetof(struct xdp_md, data_end):
4393                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
4394                                       si->dst_reg, si->src_reg,
4395                                       offsetof(struct xdp_buff, data_end));
4396                 break;
4397         case offsetof(struct xdp_md, ingress_ifindex):
4398                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
4399                                       si->dst_reg, si->src_reg,
4400                                       offsetof(struct xdp_buff, rxq));
4401                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
4402                                       si->dst_reg, si->dst_reg,
4403                                       offsetof(struct xdp_rxq_info, dev));
4404                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4405                                       offsetof(struct net_device, ifindex));
4406                 break;
4407         case offsetof(struct xdp_md, rx_queue_index):
4408                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
4409                                       si->dst_reg, si->src_reg,
4410                                       offsetof(struct xdp_buff, rxq));
4411                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4412                                       offsetof(struct xdp_rxq_info,
4413                                                queue_index));
4414                 break;
4415         }
4416
4417         return insn - insn_buf;
4418 }
4419
4420 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
4421                                        const struct bpf_insn *si,
4422                                        struct bpf_insn *insn_buf,
4423                                        struct bpf_prog *prog,
4424                                        u32 *target_size)
4425 {
4426         struct bpf_insn *insn = insn_buf;
4427         int off;
4428
4429         switch (si->off) {
4430         case offsetof(struct bpf_sock_ops, op) ...
4431              offsetof(struct bpf_sock_ops, replylong[3]):
4432                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
4433                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
4434                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
4435                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
4436                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
4437                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
4438                 off = si->off;
4439                 off -= offsetof(struct bpf_sock_ops, op);
4440                 off += offsetof(struct bpf_sock_ops_kern, op);
4441                 if (type == BPF_WRITE)
4442                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4443                                               off);
4444                 else
4445                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4446                                               off);
4447                 break;
4448
4449         case offsetof(struct bpf_sock_ops, family):
4450                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4451
4452                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4453                                               struct bpf_sock_ops_kern, sk),
4454                                       si->dst_reg, si->src_reg,
4455                                       offsetof(struct bpf_sock_ops_kern, sk));
4456                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4457                                       offsetof(struct sock_common, skc_family));
4458                 break;
4459
4460         case offsetof(struct bpf_sock_ops, remote_ip4):
4461                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4462
4463                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4464                                                 struct bpf_sock_ops_kern, sk),
4465                                       si->dst_reg, si->src_reg,
4466                                       offsetof(struct bpf_sock_ops_kern, sk));
4467                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4468                                       offsetof(struct sock_common, skc_daddr));
4469                 break;
4470
4471         case offsetof(struct bpf_sock_ops, local_ip4):
4472                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_rcv_saddr) != 4);
4473
4474                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4475                                               struct bpf_sock_ops_kern, sk),
4476                                       si->dst_reg, si->src_reg,
4477                                       offsetof(struct bpf_sock_ops_kern, sk));
4478                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4479                                       offsetof(struct sock_common,
4480                                                skc_rcv_saddr));
4481                 break;
4482
4483         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
4484              offsetof(struct bpf_sock_ops, remote_ip6[3]):
4485 #if IS_ENABLED(CONFIG_IPV6)
4486                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4487                                           skc_v6_daddr.s6_addr32[0]) != 4);
4488
4489                 off = si->off;
4490                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
4491                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4492                                                 struct bpf_sock_ops_kern, sk),
4493                                       si->dst_reg, si->src_reg,
4494                                       offsetof(struct bpf_sock_ops_kern, sk));
4495                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4496                                       offsetof(struct sock_common,
4497                                                skc_v6_daddr.s6_addr32[0]) +
4498                                       off);
4499 #else
4500                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4501 #endif
4502                 break;
4503
4504         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
4505              offsetof(struct bpf_sock_ops, local_ip6[3]):
4506 #if IS_ENABLED(CONFIG_IPV6)
4507                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4508                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4509
4510                 off = si->off;
4511                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
4512                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4513                                                 struct bpf_sock_ops_kern, sk),
4514                                       si->dst_reg, si->src_reg,
4515                                       offsetof(struct bpf_sock_ops_kern, sk));
4516                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4517                                       offsetof(struct sock_common,
4518                                                skc_v6_rcv_saddr.s6_addr32[0]) +
4519                                       off);
4520 #else
4521                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4522 #endif
4523                 break;
4524
4525         case offsetof(struct bpf_sock_ops, remote_port):
4526                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4527
4528                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4529                                                 struct bpf_sock_ops_kern, sk),
4530                                       si->dst_reg, si->src_reg,
4531                                       offsetof(struct bpf_sock_ops_kern, sk));
4532                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4533                                       offsetof(struct sock_common, skc_dport));
4534 #ifndef __BIG_ENDIAN_BITFIELD
4535                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4536 #endif
4537                 break;
4538
4539         case offsetof(struct bpf_sock_ops, local_port):
4540                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4541
4542                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4543                                                 struct bpf_sock_ops_kern, sk),
4544                                       si->dst_reg, si->src_reg,
4545                                       offsetof(struct bpf_sock_ops_kern, sk));
4546                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4547                                       offsetof(struct sock_common, skc_num));
4548                 break;
4549
4550         case offsetof(struct bpf_sock_ops, is_fullsock):
4551                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4552                                                 struct bpf_sock_ops_kern,
4553                                                 is_fullsock),
4554                                       si->dst_reg, si->src_reg,
4555                                       offsetof(struct bpf_sock_ops_kern,
4556                                                is_fullsock));
4557                 break;
4558
4559         case offsetof(struct bpf_sock_ops, state):
4560                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
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_B, si->dst_reg, si->dst_reg,
4567                                       offsetof(struct sock_common, skc_state));
4568                 break;
4569
4570         case offsetof(struct bpf_sock_ops, rtt_min):
4571                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
4572                              sizeof(struct minmax));
4573                 BUILD_BUG_ON(sizeof(struct minmax) <
4574                              sizeof(struct minmax_sample));
4575
4576                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4577                                                 struct bpf_sock_ops_kern, sk),
4578                                       si->dst_reg, si->src_reg,
4579                                       offsetof(struct bpf_sock_ops_kern, sk));
4580                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4581                                       offsetof(struct tcp_sock, rtt_min) +
4582                                       FIELD_SIZEOF(struct minmax_sample, t));
4583                 break;
4584
4585 /* Helper macro for adding read access to tcp_sock or sock fields. */
4586 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
4587         do {                                                                  \
4588                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
4589                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
4590                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4591                                                 struct bpf_sock_ops_kern,     \
4592                                                 is_fullsock),                 \
4593                                       si->dst_reg, si->src_reg,               \
4594                                       offsetof(struct bpf_sock_ops_kern,      \
4595                                                is_fullsock));                 \
4596                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
4597                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4598                                                 struct bpf_sock_ops_kern, sk),\
4599                                       si->dst_reg, si->src_reg,               \
4600                                       offsetof(struct bpf_sock_ops_kern, sk));\
4601                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
4602                                                        OBJ_FIELD),            \
4603                                       si->dst_reg, si->dst_reg,               \
4604                                       offsetof(OBJ, OBJ_FIELD));              \
4605         } while (0)
4606
4607 /* Helper macro for adding write access to tcp_sock or sock fields.
4608  * The macro is called with two registers, dst_reg which contains a pointer
4609  * to ctx (context) and src_reg which contains the value that should be
4610  * stored. However, we need an additional register since we cannot overwrite
4611  * dst_reg because it may be used later in the program.
4612  * Instead we "borrow" one of the other register. We first save its value
4613  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
4614  * it at the end of the macro.
4615  */
4616 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
4617         do {                                                                  \
4618                 int reg = BPF_REG_9;                                          \
4619                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
4620                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
4621                 if (si->dst_reg == reg || si->src_reg == reg)                 \
4622                         reg--;                                                \
4623                 if (si->dst_reg == reg || si->src_reg == reg)                 \
4624                         reg--;                                                \
4625                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
4626                                       offsetof(struct bpf_sock_ops_kern,      \
4627                                                temp));                        \
4628                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4629                                                 struct bpf_sock_ops_kern,     \
4630                                                 is_fullsock),                 \
4631                                       reg, si->dst_reg,                       \
4632                                       offsetof(struct bpf_sock_ops_kern,      \
4633                                                is_fullsock));                 \
4634                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
4635                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
4636                                                 struct bpf_sock_ops_kern, sk),\
4637                                       reg, si->dst_reg,                       \
4638                                       offsetof(struct bpf_sock_ops_kern, sk));\
4639                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
4640                                       reg, si->src_reg,                       \
4641                                       offsetof(OBJ, OBJ_FIELD));              \
4642                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
4643                                       offsetof(struct bpf_sock_ops_kern,      \
4644                                                temp));                        \
4645         } while (0)
4646
4647 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
4648         do {                                                                  \
4649                 if (TYPE == BPF_WRITE)                                        \
4650                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
4651                 else                                                          \
4652                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
4653         } while (0)
4654
4655         case offsetof(struct bpf_sock_ops, snd_cwnd):
4656                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
4657                 break;
4658
4659         case offsetof(struct bpf_sock_ops, srtt_us):
4660                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
4661                 break;
4662
4663         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
4664                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
4665                                    struct tcp_sock);
4666                 break;
4667
4668         case offsetof(struct bpf_sock_ops, snd_ssthresh):
4669                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
4670                 break;
4671
4672         case offsetof(struct bpf_sock_ops, rcv_nxt):
4673                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
4674                 break;
4675
4676         case offsetof(struct bpf_sock_ops, snd_nxt):
4677                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
4678                 break;
4679
4680         case offsetof(struct bpf_sock_ops, snd_una):
4681                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
4682                 break;
4683
4684         case offsetof(struct bpf_sock_ops, mss_cache):
4685                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
4686                 break;
4687
4688         case offsetof(struct bpf_sock_ops, ecn_flags):
4689                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
4690                 break;
4691
4692         case offsetof(struct bpf_sock_ops, rate_delivered):
4693                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
4694                                    struct tcp_sock);
4695                 break;
4696
4697         case offsetof(struct bpf_sock_ops, rate_interval_us):
4698                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
4699                                    struct tcp_sock);
4700                 break;
4701
4702         case offsetof(struct bpf_sock_ops, packets_out):
4703                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
4704                 break;
4705
4706         case offsetof(struct bpf_sock_ops, retrans_out):
4707                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
4708                 break;
4709
4710         case offsetof(struct bpf_sock_ops, total_retrans):
4711                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
4712                                    struct tcp_sock);
4713                 break;
4714
4715         case offsetof(struct bpf_sock_ops, segs_in):
4716                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
4717                 break;
4718
4719         case offsetof(struct bpf_sock_ops, data_segs_in):
4720                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
4721                 break;
4722
4723         case offsetof(struct bpf_sock_ops, segs_out):
4724                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
4725                 break;
4726
4727         case offsetof(struct bpf_sock_ops, data_segs_out):
4728                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
4729                                    struct tcp_sock);
4730                 break;
4731
4732         case offsetof(struct bpf_sock_ops, lost_out):
4733                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
4734                 break;
4735
4736         case offsetof(struct bpf_sock_ops, sacked_out):
4737                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
4738                 break;
4739
4740         case offsetof(struct bpf_sock_ops, sk_txhash):
4741                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
4742                                           struct sock, type);
4743                 break;
4744
4745         case offsetof(struct bpf_sock_ops, bytes_received):
4746                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
4747                                    struct tcp_sock);
4748                 break;
4749
4750         case offsetof(struct bpf_sock_ops, bytes_acked):
4751                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
4752                 break;
4753
4754         }
4755         return insn - insn_buf;
4756 }
4757
4758 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
4759                                      const struct bpf_insn *si,
4760                                      struct bpf_insn *insn_buf,
4761                                      struct bpf_prog *prog, u32 *target_size)
4762 {
4763         struct bpf_insn *insn = insn_buf;
4764         int off;
4765
4766         switch (si->off) {
4767         case offsetof(struct __sk_buff, data_end):
4768                 off  = si->off;
4769                 off -= offsetof(struct __sk_buff, data_end);
4770                 off += offsetof(struct sk_buff, cb);
4771                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
4772                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4773                                       si->src_reg, off);
4774                 break;
4775         default:
4776                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4777                                               target_size);
4778         }
4779
4780         return insn - insn_buf;
4781 }
4782
4783 const struct bpf_verifier_ops sk_filter_verifier_ops = {
4784         .get_func_proto         = sk_filter_func_proto,
4785         .is_valid_access        = sk_filter_is_valid_access,
4786         .convert_ctx_access     = bpf_convert_ctx_access,
4787 };
4788
4789 const struct bpf_prog_ops sk_filter_prog_ops = {
4790         .test_run               = bpf_prog_test_run_skb,
4791 };
4792
4793 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
4794         .get_func_proto         = tc_cls_act_func_proto,
4795         .is_valid_access        = tc_cls_act_is_valid_access,
4796         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
4797         .gen_prologue           = tc_cls_act_prologue,
4798 };
4799
4800 const struct bpf_prog_ops tc_cls_act_prog_ops = {
4801         .test_run               = bpf_prog_test_run_skb,
4802 };
4803
4804 const struct bpf_verifier_ops xdp_verifier_ops = {
4805         .get_func_proto         = xdp_func_proto,
4806         .is_valid_access        = xdp_is_valid_access,
4807         .convert_ctx_access     = xdp_convert_ctx_access,
4808 };
4809
4810 const struct bpf_prog_ops xdp_prog_ops = {
4811         .test_run               = bpf_prog_test_run_xdp,
4812 };
4813
4814 const struct bpf_verifier_ops cg_skb_verifier_ops = {
4815         .get_func_proto         = sk_filter_func_proto,
4816         .is_valid_access        = sk_filter_is_valid_access,
4817         .convert_ctx_access     = bpf_convert_ctx_access,
4818 };
4819
4820 const struct bpf_prog_ops cg_skb_prog_ops = {
4821         .test_run               = bpf_prog_test_run_skb,
4822 };
4823
4824 const struct bpf_verifier_ops lwt_inout_verifier_ops = {
4825         .get_func_proto         = lwt_inout_func_proto,
4826         .is_valid_access        = lwt_is_valid_access,
4827         .convert_ctx_access     = bpf_convert_ctx_access,
4828 };
4829
4830 const struct bpf_prog_ops lwt_inout_prog_ops = {
4831         .test_run               = bpf_prog_test_run_skb,
4832 };
4833
4834 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
4835         .get_func_proto         = lwt_xmit_func_proto,
4836         .is_valid_access        = lwt_is_valid_access,
4837         .convert_ctx_access     = bpf_convert_ctx_access,
4838         .gen_prologue           = tc_cls_act_prologue,
4839 };
4840
4841 const struct bpf_prog_ops lwt_xmit_prog_ops = {
4842         .test_run               = bpf_prog_test_run_skb,
4843 };
4844
4845 const struct bpf_verifier_ops cg_sock_verifier_ops = {
4846         .get_func_proto         = sock_filter_func_proto,
4847         .is_valid_access        = sock_filter_is_valid_access,
4848         .convert_ctx_access     = sock_filter_convert_ctx_access,
4849 };
4850
4851 const struct bpf_prog_ops cg_sock_prog_ops = {
4852 };
4853
4854 const struct bpf_verifier_ops sock_ops_verifier_ops = {
4855         .get_func_proto         = sock_ops_func_proto,
4856         .is_valid_access        = sock_ops_is_valid_access,
4857         .convert_ctx_access     = sock_ops_convert_ctx_access,
4858 };
4859
4860 const struct bpf_prog_ops sock_ops_prog_ops = {
4861 };
4862
4863 const struct bpf_verifier_ops sk_skb_verifier_ops = {
4864         .get_func_proto         = sk_skb_func_proto,
4865         .is_valid_access        = sk_skb_is_valid_access,
4866         .convert_ctx_access     = sk_skb_convert_ctx_access,
4867         .gen_prologue           = sk_skb_prologue,
4868 };
4869
4870 const struct bpf_prog_ops sk_skb_prog_ops = {
4871 };
4872
4873 int sk_detach_filter(struct sock *sk)
4874 {
4875         int ret = -ENOENT;
4876         struct sk_filter *filter;
4877
4878         if (sock_flag(sk, SOCK_FILTER_LOCKED))
4879                 return -EPERM;
4880
4881         filter = rcu_dereference_protected(sk->sk_filter,
4882                                            lockdep_sock_is_held(sk));
4883         if (filter) {
4884                 RCU_INIT_POINTER(sk->sk_filter, NULL);
4885                 sk_filter_uncharge(sk, filter);
4886                 ret = 0;
4887         }
4888
4889         return ret;
4890 }
4891 EXPORT_SYMBOL_GPL(sk_detach_filter);
4892
4893 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
4894                   unsigned int len)
4895 {
4896         struct sock_fprog_kern *fprog;
4897         struct sk_filter *filter;
4898         int ret = 0;
4899
4900         lock_sock(sk);
4901         filter = rcu_dereference_protected(sk->sk_filter,
4902                                            lockdep_sock_is_held(sk));
4903         if (!filter)
4904                 goto out;
4905
4906         /* We're copying the filter that has been originally attached,
4907          * so no conversion/decode needed anymore. eBPF programs that
4908          * have no original program cannot be dumped through this.
4909          */
4910         ret = -EACCES;
4911         fprog = filter->prog->orig_prog;
4912         if (!fprog)
4913                 goto out;
4914
4915         ret = fprog->len;
4916         if (!len)
4917                 /* User space only enquires number of filter blocks. */
4918                 goto out;
4919
4920         ret = -EINVAL;
4921         if (len < fprog->len)
4922                 goto out;
4923
4924         ret = -EFAULT;
4925         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
4926                 goto out;
4927
4928         /* Instead of bytes, the API requests to return the number
4929          * of filter blocks.
4930          */
4931         ret = fprog->len;
4932 out:
4933         release_sock(sk);
4934         return ret;
4935 }