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