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         bool                            copy[MAX_MSG_FRAGS];
  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
  42 /* UAPI in filter.c depends on struct sk_msg_sg being first element. */
  43 struct sk_msg {
  44         struct sk_msg_sg                sg;
  45         void                            *data;
  46         void                            *data_end;
  47         u32                             apply_bytes;
  48         u32                             cork_bytes;
  49         u32                             flags;
  50         struct sk_buff                  *skb;
  51         struct sock                     *sk_redir;
  52         struct sock                     *sk;
  53         struct list_head                list;
  54 };
  55
  56 struct sk_psock_progs {
  57         struct bpf_prog                 *msg_parser;
  58         struct bpf_prog                 *skb_parser;
  59         struct bpf_prog                 *skb_verdict;
  60 };
  61
  62 enum sk_psock_state_bits {
  63         SK_PSOCK_TX_ENABLED,
  64 };
  65
  66 struct sk_psock_link {
  67         struct list_head                list;
  68         struct bpf_map                  *map;
  69         void                            *link_raw;
  70 };
  71
  72 struct sk_psock_parser {
  73         struct strparser                strp;
  74         bool                            enabled;
  75         void (*saved_data_ready)(struct sock *sk);
  76 };
  77
  78 struct sk_psock_work_state {
  79         struct sk_buff                  *skb;
  80         u32                             len;
  81         u32                             off;
  82 };
  83
  84 struct sk_psock {
  85         struct sock                     *sk;
  86         struct sock                     *sk_redir;
  87         u32                             apply_bytes;
  88         u32                             cork_bytes;
  89         u32                             eval;
  90         struct sk_msg                   *cork;
  91         struct sk_psock_progs           progs;
  92         struct sk_psock_parser          parser;
  93         struct sk_buff_head             ingress_skb;
  94         struct list_head                ingress_msg;
  95         unsigned long                   state;
  96         struct list_head                link;
  97         spinlock_t                      link_lock;
  98         refcount_t                      refcnt;
  99         void (*saved_unhash)(struct sock *sk);
 100         void (*saved_close)(struct sock *sk, long timeout);
 101         void (*saved_write_space)(struct sock *sk);
 102         struct proto                    *sk_proto;
 103         struct sk_psock_work_state      work_state;
 104         struct work_struct              work;
 105         union {
 106                 struct rcu_head         rcu;
 107                 struct work_struct      gc;
 108         };
 109 };
 110
 111 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
 112                  int elem_first_coalesce);
 113 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
 114                  u32 off, u32 len);
 115 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
 116 int sk_msg_free(struct sock *sk, struct sk_msg *msg);
 117 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
 118 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
 119 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
 120                                   u32 bytes);
 121
 122 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
 123 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
 124
 125 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
 126                               struct sk_msg *msg, u32 bytes);
 127 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
 128                              struct sk_msg *msg, u32 bytes);
 129
 130 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
 131 {
 132         WARN_ON(i == msg->sg.end && bytes);
 133 }
 134
 135 static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
 136 {
 137         if (psock->apply_bytes) {
 138                 if (psock->apply_bytes < bytes)
 139                         psock->apply_bytes = 0;
 140                 else
 141                         psock->apply_bytes -= bytes;
 142         }
 143 }
 144
 145 static inline u32 sk_msg_iter_dist(u32 start, u32 end)
 146 {
 147         return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
 148 }
 149
 150 #define sk_msg_iter_var_prev(var)                       \
 151         do {                                            \
 152                 if (var == 0)                           \
 153                         var = NR_MSG_FRAG_IDS - 1;      \
 154                 else                                    \
 155                         var--;                          \
 156         } while (0)
 157
 158 #define sk_msg_iter_var_next(var)                       \
 159         do {                                            \
 160                 var++;                                  \
 161                 if (var == NR_MSG_FRAG_IDS)             \
 162                         var = 0;                        \
 163         } while (0)
 164
 165 #define sk_msg_iter_prev(msg, which)                    \
 166         sk_msg_iter_var_prev(msg->sg.which)
 167
 168 #define sk_msg_iter_next(msg, which)                    \
 169         sk_msg_iter_var_next(msg->sg.which)
 170
 171 static inline void sk_msg_clear_meta(struct sk_msg *msg)
 172 {
 173         memset(&msg->sg, 0, offsetofend(struct sk_msg_sg, copy));
 174 }
 175
 176 static inline void sk_msg_init(struct sk_msg *msg)
 177 {
 178         BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
 179         memset(msg, 0, sizeof(*msg));
 180         sg_init_marker(msg->sg.data, NR_MSG_FRAG_IDS);
 181 }
 182
 183 static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
 184                                int which, u32 size)
 185 {
 186         dst->sg.data[which] = src->sg.data[which];
 187         dst->sg.data[which].length  = size;
 188         dst->sg.size               += size;
 189         src->sg.data[which].length -= size;
 190         src->sg.data[which].offset += size;
 191 }
 192
 193 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
 194 {
 195         memcpy(dst, src, sizeof(*src));
 196         sk_msg_init(src);
 197 }
 198
 199 static inline bool sk_msg_full(const struct sk_msg *msg)
 200 {
 201         return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
 202 }
 203
 204 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
 205 {
 206         return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
 207 }
 208
 209 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
 210 {
 211         return &msg->sg.data[which];
 212 }
 213
 214 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
 215 {
 216         return msg->sg.data[which];
 217 }
 218
 219 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
 220 {
 221         return sg_page(sk_msg_elem(msg, which));
 222 }
 223
 224 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
 225 {
 226         return msg->flags & BPF_F_INGRESS;
 227 }
 228
 229 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
 230 {
 231         struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
 232
 233         if (msg->sg.copy[msg->sg.start]) {
 234                 msg->data = NULL;
 235                 msg->data_end = NULL;
 236         } else {
 237                 msg->data = sg_virt(sge);
 238                 msg->data_end = msg->data + sge->length;
 239         }
 240 }
 241
 242 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
 243                                    u32 len, u32 offset)
 244 {
 245         struct scatterlist *sge;
 246
 247         get_page(page);
 248         sge = sk_msg_elem(msg, msg->sg.end);
 249         sg_set_page(sge, page, len, offset);
 250         sg_unmark_end(sge);
 251
 252         msg->sg.copy[msg->sg.end] = true;
 253         msg->sg.size += len;
 254         sk_msg_iter_next(msg, end);
 255 }
 256
 257 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
 258 {
 259         do {
 260                 msg->sg.copy[i] = copy_state;
 261                 sk_msg_iter_var_next(i);
 262                 if (i == msg->sg.end)
 263                         break;
 264         } while (1);
 265 }
 266
 267 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
 268 {
 269         sk_msg_sg_copy(msg, start, true);
 270 }
 271
 272 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
 273 {
 274         sk_msg_sg_copy(msg, start, false);
 275 }
 276
 277 static inline struct sk_psock *sk_psock(const struct sock *sk)
 278 {
 279         return rcu_dereference_sk_user_data(sk);
 280 }
 281
 282 static inline void sk_psock_queue_msg(struct sk_psock *psock,
 283                                       struct sk_msg *msg)
 284 {
 285         list_add_tail(&msg->list, &psock->ingress_msg);
 286 }
 287
 288 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
 289 {
 290         return psock ? list_empty(&psock->ingress_msg) : true;
 291 }
 292
 293 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
 294 {
 295         struct sock *sk = psock->sk;
 296
 297         sk->sk_err = err;
 298         sk->sk_error_report(sk);
 299 }
 300
 301 struct sk_psock *sk_psock_init(struct sock *sk, int node);
 302
 303 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
 304 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
 305 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
 306
 307 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
 308                          struct sk_msg *msg);
 309
 310 static inline struct sk_psock_link *sk_psock_init_link(void)
 311 {
 312         return kzalloc(sizeof(struct sk_psock_link),
 313                        GFP_ATOMIC | __GFP_NOWARN);
 314 }
 315
 316 static inline void sk_psock_free_link(struct sk_psock_link *link)
 317 {
 318         kfree(link);
 319 }
 320
 321 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
 322 #if defined(CONFIG_BPF_STREAM_PARSER)
 323 void sk_psock_unlink(struct sock *sk, struct sk_psock_link *link);
 324 #else
 325 static inline void sk_psock_unlink(struct sock *sk,
 326                                    struct sk_psock_link *link)
 327 {
 328 }
 329 #endif
 330
 331 void __sk_psock_purge_ingress_msg(struct sk_psock *psock);
 332
 333 static inline void sk_psock_cork_free(struct sk_psock *psock)
 334 {
 335         if (psock->cork) {
 336                 sk_msg_free(psock->sk, psock->cork);
 337                 kfree(psock->cork);
 338                 psock->cork = NULL;
 339         }
 340 }
 341
 342 static inline void sk_psock_update_proto(struct sock *sk,
 343                                          struct sk_psock *psock,
 344                                          struct proto *ops)
 345 {
 346         psock->saved_unhash = sk->sk_prot->unhash;
 347         psock->saved_close = sk->sk_prot->close;
 348         psock->saved_write_space = sk->sk_write_space;
 349
 350         psock->sk_proto = sk->sk_prot;
 351         sk->sk_prot = ops;
 352 }
 353
 354 static inline void sk_psock_restore_proto(struct sock *sk,
 355                                           struct sk_psock *psock)
 356 {
 357         sk->sk_prot->unhash = psock->saved_unhash;
 358         sk->sk_write_space = psock->saved_write_space;
 359
 360         if (psock->sk_proto) {
 361                 struct inet_connection_sock *icsk = inet_csk(sk);
 362                 bool has_ulp = !!icsk->icsk_ulp_data;
 363
 364                 if (has_ulp)
 365                         tcp_update_ulp(sk, psock->sk_proto);
 366                 else
 367                         sk->sk_prot = psock->sk_proto;
 368                 psock->sk_proto = NULL;
 369         }
 370 }
 371
 372 static inline void sk_psock_set_state(struct sk_psock *psock,
 373                                       enum sk_psock_state_bits bit)
 374 {
 375         set_bit(bit, &psock->state);
 376 }
 377
 378 static inline void sk_psock_clear_state(struct sk_psock *psock,
 379                                         enum sk_psock_state_bits bit)
 380 {
 381         clear_bit(bit, &psock->state);
 382 }
 383
 384 static inline bool sk_psock_test_state(const struct sk_psock *psock,
 385                                        enum sk_psock_state_bits bit)
 386 {
 387         return test_bit(bit, &psock->state);
 388 }
 389
 390 static inline struct sk_psock *sk_psock_get_checked(struct sock *sk)
 391 {
 392         struct sk_psock *psock;
 393
 394         rcu_read_lock();
 395         psock = sk_psock(sk);
 396         if (psock) {
 397                 if (sk->sk_prot->recvmsg != tcp_bpf_recvmsg) {
 398                         psock = ERR_PTR(-EBUSY);
 399                         goto out;
 400                 }
 401
 402                 if (!refcount_inc_not_zero(&psock->refcnt))
 403                         psock = ERR_PTR(-EBUSY);
 404         }
 405 out:
 406         rcu_read_unlock();
 407         return psock;
 408 }
 409
 410 static inline struct sk_psock *sk_psock_get(struct sock *sk)
 411 {
 412         struct sk_psock *psock;
 413
 414         rcu_read_lock();
 415         psock = sk_psock(sk);
 416         if (psock && !refcount_inc_not_zero(&psock->refcnt))
 417                 psock = NULL;
 418         rcu_read_unlock();
 419         return psock;
 420 }
 421
 422 void sk_psock_stop(struct sock *sk, struct sk_psock *psock);
 423 void sk_psock_destroy(struct rcu_head *rcu);
 424 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
 425
 426 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
 427 {
 428         if (refcount_dec_and_test(&psock->refcnt))
 429                 sk_psock_drop(sk, psock);
 430 }
 431
 432 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
 433 {
 434         if (psock->parser.enabled)
 435                 psock->parser.saved_data_ready(sk);
 436         else
 437                 sk->sk_data_ready(sk);
 438 }
 439
 440 static inline void psock_set_prog(struct bpf_prog **pprog,
 441                                   struct bpf_prog *prog)
 442 {
 443         prog = xchg(pprog, prog);
 444         if (prog)
 445                 bpf_prog_put(prog);
 446 }
 447
 448 static inline void psock_progs_drop(struct sk_psock_progs *progs)
 449 {
 450         psock_set_prog(&progs->msg_parser, NULL);
 451         psock_set_prog(&progs->skb_parser, NULL);
 452         psock_set_prog(&progs->skb_verdict, NULL);
 453 }
 454
 455 #endif /* _LINUX_SKMSG_H */