include/linux/skmsg.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
   3
   4 #ifndef _LINUX_SKMSG_H
   5 #define _LINUX_SKMSG_H
   6
   7 #include <linux/bpf.h>
   8 #include <linux/filter.h>
   9 #include <linux/scatterlist.h>
  10 #include <linux/skbuff.h>
  11
  12 #include <net/sock.h>
  13 #include <net/tcp.h>
  14 #include <net/strparser.h>
  15
  16 #define MAX_MSG_FRAGS                   MAX_SKB_FRAGS
  17 #define NR_MSG_FRAG_IDS                 (MAX_MSG_FRAGS + 1)
  18
  19 enum __sk_action {
  20         __SK_DROP = 0,
  21         __SK_PASS,
  22         __SK_REDIRECT,
  23         __SK_NONE,
  24 };
  25
  26 struct sk_msg_sg {
  27         u32                             start;
  28         u32                             curr;
  29         u32                             end;
  30         u32                             size;
  31         u32                             copybreak;
  32         unsigned long                   copy;
  33         /* The extra two elements:
  34          * 1) used for chaining the front and sections when the list becomes
  35          *    partitioned (e.g. end < start). The crypto APIs require the
  36          *    chaining;
  37          * 2) to chain tailer SG entries after the message.
  38          */
  39         struct scatterlist              data[MAX_MSG_FRAGS + 2];
  40 };
  41 static_assert(BITS_PER_LONG >= NR_MSG_FRAG_IDS);
  42
  43 /* UAPI in filter.c depends on struct sk_msg_sg being first element. */
  44 struct sk_msg {
  45         struct sk_msg_sg                sg;
  46         void                            *data;
  47         void                            *data_end;
  48         u32                             apply_bytes;
  49         u32                             cork_bytes;
  50         u32                             flags;
  51         struct sk_buff                  *skb;
  52         struct sock                     *sk_redir;
  53         struct sock                     *sk;
  54         struct list_head                list;
  55 };
  56
  57 struct sk_psock_progs {
  58         struct bpf_prog                 *msg_parser;
  59         struct bpf_prog                 *stream_parser;
  60         struct bpf_prog                 *stream_verdict;
  61 };
  62
  63 enum sk_psock_state_bits {
  64         SK_PSOCK_TX_ENABLED,
  65 };
  66
  67 struct sk_psock_link {
  68         struct list_head                list;
  69         struct bpf_map                  *map;
  70         void                            *link_raw;
  71 };
  72
  73 struct sk_psock_work_state {
  74         struct sk_buff                  *skb;
  75         u32                             len;
  76         u32                             off;
  77 };
  78
  79 struct sk_psock {
  80         struct sock                     *sk;
  81         struct sock                     *sk_redir;
  82         u32                             apply_bytes;
  83         u32                             cork_bytes;
  84         u32                             eval;
  85         struct sk_msg                   *cork;
  86         struct sk_psock_progs           progs;
  87 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
  88         struct strparser                strp;
  89 #endif
  90         struct sk_buff_head             ingress_skb;
  91         struct list_head                ingress_msg;
  92         spinlock_t                      ingress_lock;
  93         unsigned long                   state;
  94         struct list_head                link;
  95         spinlock_t                      link_lock;
  96         refcount_t                      refcnt;
  97         void (*saved_unhash)(struct sock *sk);
  98         void (*saved_close)(struct sock *sk, long timeout);
  99         void (*saved_write_space)(struct sock *sk);
 100         void (*saved_data_ready)(struct sock *sk);
 101         struct proto                    *sk_proto;
 102         struct mutex                    work_mutex;
 103         struct sk_psock_work_state      work_state;
 104         struct work_struct              work;
 105         struct rcu_work                 rwork;
 106 };
 107
 108 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
 109                  int elem_first_coalesce);
 110 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
 111                  u32 off, u32 len);
 112 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
 113 int sk_msg_free(struct sock *sk, struct sk_msg *msg);
 114 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
 115 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
 116 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
 117                                   u32 bytes);
 118
 119 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
 120 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
 121
 122 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
 123                               struct sk_msg *msg, u32 bytes);
 124 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
 125                              struct sk_msg *msg, u32 bytes);
 126
 127 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
 128 {
 129         WARN_ON(i == msg->sg.end && bytes);
 130 }
 131
 132 static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
 133 {
 134         if (psock->apply_bytes) {
 135                 if (psock->apply_bytes < bytes)
 136                         psock->apply_bytes = 0;
 137                 else
 138                         psock->apply_bytes -= bytes;
 139         }
 140 }
 141
 142 static inline u32 sk_msg_iter_dist(u32 start, u32 end)
 143 {
 144         return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
 145 }
 146
 147 #define sk_msg_iter_var_prev(var)                       \
 148         do {                                            \
 149                 if (var == 0)                           \
 150                         var = NR_MSG_FRAG_IDS - 1;      \
 151                 else                                    \
 152                         var--;                          \
 153         } while (0)
 154
 155 #define sk_msg_iter_var_next(var)                       \
 156         do {                                            \
 157                 var++;                                  \
 158                 if (var == NR_MSG_FRAG_IDS)             \
 159                         var = 0;                        \
 160         } while (0)
 161
 162 #define sk_msg_iter_prev(msg, which)                    \
 163         sk_msg_iter_var_prev(msg->sg.which)
 164
 165 #define sk_msg_iter_next(msg, which)                    \
 166         sk_msg_iter_var_next(msg->sg.which)
 167
 168 static inline void sk_msg_clear_meta(struct sk_msg *msg)
 169 {
 170         memset(&msg->sg, 0, offsetofend(struct sk_msg_sg, copy));
 171 }
 172
 173 static inline void sk_msg_init(struct sk_msg *msg)
 174 {
 175         BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
 176         memset(msg, 0, sizeof(*msg));
 177         sg_init_marker(msg->sg.data, NR_MSG_FRAG_IDS);
 178 }
 179
 180 static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
 181                                int which, u32 size)
 182 {
 183         dst->sg.data[which] = src->sg.data[which];
 184         dst->sg.data[which].length  = size;
 185         dst->sg.size               += size;
 186         src->sg.size               -= size;
 187         src->sg.data[which].length -= size;
 188         src->sg.data[which].offset += size;
 189 }
 190
 191 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
 192 {
 193         memcpy(dst, src, sizeof(*src));
 194         sk_msg_init(src);
 195 }
 196
 197 static inline bool sk_msg_full(const struct sk_msg *msg)
 198 {
 199         return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
 200 }
 201
 202 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
 203 {
 204         return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
 205 }
 206
 207 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
 208 {
 209         return &msg->sg.data[which];
 210 }
 211
 212 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
 213 {
 214         return msg->sg.data[which];
 215 }
 216
 217 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
 218 {
 219         return sg_page(sk_msg_elem(msg, which));
 220 }
 221
 222 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
 223 {
 224         return msg->flags & BPF_F_INGRESS;
 225 }
 226
 227 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
 228 {
 229         struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
 230
 231         if (test_bit(msg->sg.start, &msg->sg.copy)) {
 232                 msg->data = NULL;
 233                 msg->data_end = NULL;
 234         } else {
 235                 msg->data = sg_virt(sge);
 236                 msg->data_end = msg->data + sge->length;
 237         }
 238 }
 239
 240 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
 241                                    u32 len, u32 offset)
 242 {
 243         struct scatterlist *sge;
 244
 245         get_page(page);
 246         sge = sk_msg_elem(msg, msg->sg.end);
 247         sg_set_page(sge, page, len, offset);
 248         sg_unmark_end(sge);
 249
 250         __set_bit(msg->sg.end, &msg->sg.copy);
 251         msg->sg.size += len;
 252         sk_msg_iter_next(msg, end);
 253 }
 254
 255 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
 256 {
 257         do {
 258                 if (copy_state)
 259                         __set_bit(i, &msg->sg.copy);
 260                 else
 261                         __clear_bit(i, &msg->sg.copy);
 262                 sk_msg_iter_var_next(i);
 263                 if (i == msg->sg.end)
 264                         break;
 265         } while (1);
 266 }
 267
 268 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
 269 {
 270         sk_msg_sg_copy(msg, start, true);
 271 }
 272
 273 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
 274 {
 275         sk_msg_sg_copy(msg, start, false);
 276 }
 277
 278 static inline struct sk_psock *sk_psock(const struct sock *sk)
 279 {
 280         return rcu_dereference_sk_user_data(sk);
 281 }
 282
 283 static inline void sk_psock_queue_msg(struct sk_psock *psock,
 284                                       struct sk_msg *msg)
 285 {
 286         spin_lock_bh(&psock->ingress_lock);
 287         list_add_tail(&msg->list, &psock->ingress_msg);
 288         spin_unlock_bh(&psock->ingress_lock);
 289 }
 290
 291 static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
 292 {
 293         struct sk_msg *msg;
 294
 295         spin_lock_bh(&psock->ingress_lock);
 296         msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
 297         if (msg)
 298                 list_del(&msg->list);
 299         spin_unlock_bh(&psock->ingress_lock);
 300         return msg;
 301 }
 302
 303 static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
 304 {
 305         struct sk_msg *msg;
 306
 307         spin_lock_bh(&psock->ingress_lock);
 308         msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
 309         spin_unlock_bh(&psock->ingress_lock);
 310         return msg;
 311 }
 312
 313 static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
 314                                                struct sk_msg *msg)
 315 {
 316         struct sk_msg *ret;
 317
 318         spin_lock_bh(&psock->ingress_lock);
 319         if (list_is_last(&msg->list, &psock->ingress_msg))
 320                 ret = NULL;
 321         else
 322                 ret = list_next_entry(msg, list);
 323         spin_unlock_bh(&psock->ingress_lock);
 324         return ret;
 325 }
 326
 327 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
 328 {
 329         return psock ? list_empty(&psock->ingress_msg) : true;
 330 }
 331
 332 static inline void kfree_sk_msg(struct sk_msg *msg)
 333 {
 334         if (msg->skb)
 335                 consume_skb(msg->skb);
 336         kfree(msg);
 337 }
 338
 339 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
 340 {
 341         struct sock *sk = psock->sk;
 342
 343         sk->sk_err = err;
 344         sk->sk_error_report(sk);
 345 }
 346
 347 struct sk_psock *sk_psock_init(struct sock *sk, int node);
 348 void sk_psock_stop(struct sk_psock *psock, bool wait);
 349
 350 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
 351 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
 352 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
 353 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
 354 #else
 355 static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
 356 {
 357         return -EOPNOTSUPP;
 358 }
 359
 360 static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
 361 {
 362 }
 363
 364 static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
 365 {
 366 }
 367 #endif
 368
 369 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
 370 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
 371
 372 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
 373                          struct sk_msg *msg);
 374
 375 static inline struct sk_psock_link *sk_psock_init_link(void)
 376 {
 377         return kzalloc(sizeof(struct sk_psock_link),
 378                        GFP_ATOMIC | __GFP_NOWARN);
 379 }
 380
 381 static inline void sk_psock_free_link(struct sk_psock_link *link)
 382 {
 383         kfree(link);
 384 }
 385
 386 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
 387
 388 static inline void sk_psock_cork_free(struct sk_psock *psock)
 389 {
 390         if (psock->cork) {
 391                 sk_msg_free(psock->sk, psock->cork);
 392                 kfree(psock->cork);
 393                 psock->cork = NULL;
 394         }
 395 }
 396
 397 static inline void sk_psock_update_proto(struct sock *sk,
 398                                          struct sk_psock *psock,
 399                                          struct proto *ops)
 400 {
 401         /* Pairs with lockless read in sk_clone_lock() */
 402         WRITE_ONCE(sk->sk_prot, ops);
 403 }
 404
 405 static inline void sk_psock_restore_proto(struct sock *sk,
 406                                           struct sk_psock *psock)
 407 {
 408         sk->sk_prot->unhash = psock->saved_unhash;
 409         if (inet_csk_has_ulp(sk)) {
 410                 tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
 411         } else {
 412                 sk->sk_write_space = psock->saved_write_space;
 413                 /* Pairs with lockless read in sk_clone_lock() */
 414                 WRITE_ONCE(sk->sk_prot, psock->sk_proto);
 415         }
 416 }
 417
 418 static inline void sk_psock_set_state(struct sk_psock *psock,
 419                                       enum sk_psock_state_bits bit)
 420 {
 421         set_bit(bit, &psock->state);
 422 }
 423
 424 static inline void sk_psock_clear_state(struct sk_psock *psock,
 425                                         enum sk_psock_state_bits bit)
 426 {
 427         clear_bit(bit, &psock->state);
 428 }
 429
 430 static inline bool sk_psock_test_state(const struct sk_psock *psock,
 431                                        enum sk_psock_state_bits bit)
 432 {
 433         return test_bit(bit, &psock->state);
 434 }
 435
 436 static inline struct sk_psock *sk_psock_get(struct sock *sk)
 437 {
 438         struct sk_psock *psock;
 439
 440         rcu_read_lock();
 441         psock = sk_psock(sk);
 442         if (psock && !refcount_inc_not_zero(&psock->refcnt))
 443                 psock = NULL;
 444         rcu_read_unlock();
 445         return psock;
 446 }
 447
 448 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
 449
 450 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
 451 {
 452         if (refcount_dec_and_test(&psock->refcnt))
 453                 sk_psock_drop(sk, psock);
 454 }
 455
 456 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
 457 {
 458         if (psock->saved_data_ready)
 459                 psock->saved_data_ready(sk);
 460         else
 461                 sk->sk_data_ready(sk);
 462 }
 463
 464 static inline void psock_set_prog(struct bpf_prog **pprog,
 465                                   struct bpf_prog *prog)
 466 {
 467         prog = xchg(pprog, prog);
 468         if (prog)
 469                 bpf_prog_put(prog);
 470 }
 471
 472 static inline int psock_replace_prog(struct bpf_prog **pprog,
 473                                      struct bpf_prog *prog,
 474                                      struct bpf_prog *old)
 475 {
 476         if (cmpxchg(pprog, old, prog) != old)
 477                 return -ENOENT;
 478
 479         if (old)
 480                 bpf_prog_put(old);
 481
 482         return 0;
 483 }
 484
 485 static inline void psock_progs_drop(struct sk_psock_progs *progs)
 486 {
 487         psock_set_prog(&progs->msg_parser, NULL);
 488         psock_set_prog(&progs->stream_parser, NULL);
 489         psock_set_prog(&progs->stream_verdict, NULL);
 490 }
 491
 492 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
 493
 494 static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
 495 {
 496         if (!psock)
 497                 return false;
 498         return !!psock->saved_data_ready;
 499 }
 500
 501 #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
 502
 503 /* We only have one bit so far. */
 504 #define BPF_F_PTR_MASK ~(BPF_F_INGRESS)
 505
 506 static inline bool skb_bpf_ingress(const struct sk_buff *skb)
 507 {
 508         unsigned long sk_redir = skb->_sk_redir;
 509
 510         return sk_redir & BPF_F_INGRESS;
 511 }
 512
 513 static inline void skb_bpf_set_ingress(struct sk_buff *skb)
 514 {
 515         skb->_sk_redir |= BPF_F_INGRESS;
 516 }
 517
 518 static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
 519                                      bool ingress)
 520 {
 521         skb->_sk_redir = (unsigned long)sk_redir;
 522         if (ingress)
 523                 skb->_sk_redir |= BPF_F_INGRESS;
 524 }
 525
 526 static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
 527 {
 528         unsigned long sk_redir = skb->_sk_redir;
 529
 530         return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
 531 }
 532
 533 static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
 534 {
 535         skb->_sk_redir = 0;
 536 }
 537 #endif /* CONFIG_NET_SOCK_MSG */
 538 #endif /* _LINUX_SKMSG_H */