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 DECLARE_BITMAP(copy, MAX_MSG_FRAGS + 2);
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 *stream_parser;
59 struct bpf_prog *stream_verdict;
60 struct bpf_prog *skb_verdict;
61 struct bpf_link *msg_parser_link;
62 struct bpf_link *stream_parser_link;
63 struct bpf_link *stream_verdict_link;
64 struct bpf_link *skb_verdict_link;
65 };
66
67 enum sk_psock_state_bits {
68 SK_PSOCK_TX_ENABLED,
69 SK_PSOCK_RX_STRP_ENABLED,
70 };
71
72 struct sk_psock_link {
73 struct list_head list;
74 struct bpf_map *map;
75 void *link_raw;
76 };
77
78 struct sk_psock_work_state {
79 u32 len;
80 u32 off;
81 };
82
83 struct sk_psock {
84 struct sock *sk;
85 struct sock *sk_redir;
86 u32 apply_bytes;
87 u32 cork_bytes;
88 u32 eval;
89 bool redir_ingress; /* undefined if sk_redir is null */
90 struct sk_msg *cork;
91 struct sk_psock_progs progs;
92 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
93 struct strparser strp;
94 #endif
95 struct sk_buff_head ingress_skb;
96 struct list_head ingress_msg;
97 spinlock_t ingress_lock;
98 unsigned long state;
99 struct list_head link;
100 spinlock_t link_lock;
101 refcount_t refcnt;
102 void (*saved_unhash)(struct sock *sk);
103 void (*saved_destroy)(struct sock *sk);
104 void (*saved_close)(struct sock *sk, long timeout);
105 void (*saved_write_space)(struct sock *sk);
106 void (*saved_data_ready)(struct sock *sk);
107 /* psock_update_sk_prot may be called with restore=false many times
108 * so the handler must be safe for this case. It will be called
109 * exactly once with restore=true when the psock is being destroyed
110 * and psock refcnt is zero, but before an RCU grace period.
111 */
112 int (*psock_update_sk_prot)(struct sock *sk, struct sk_psock *psock,
113 bool restore);
114 struct proto *sk_proto;
115 struct mutex work_mutex;
116 struct sk_psock_work_state work_state;
117 struct delayed_work work;
118 struct sock *sk_pair;
119 struct rcu_work rwork;
120 };
121
122 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
123 int elem_first_coalesce);
124 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
125 u32 off, u32 len);
126 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
127 int sk_msg_free(struct sock *sk, struct sk_msg *msg);
128 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
129 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
130 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
131 u32 bytes);
132
133 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
134 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
135
136 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
137 struct sk_msg *msg, u32 bytes);
138 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
139 struct sk_msg *msg, u32 bytes);
140 int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
141 int len, int flags);
142 bool sk_msg_is_readable(struct sock *sk);
143
sk_msg_check_to_free(struct sk_msg * msg,u32 i,u32 bytes)144 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
145 {
146 WARN_ON(i == msg->sg.end && bytes);
147 }
148
sk_msg_apply_bytes(struct sk_psock * psock,u32 bytes)149 static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
150 {
151 if (psock->apply_bytes) {
152 if (psock->apply_bytes < bytes)
153 psock->apply_bytes = 0;
154 else
155 psock->apply_bytes -= bytes;
156 }
157 }
158
sk_msg_iter_dist(u32 start,u32 end)159 static inline u32 sk_msg_iter_dist(u32 start, u32 end)
160 {
161 return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
162 }
163
164 #define sk_msg_iter_var_prev(var) \
165 do { \
166 if (var == 0) \
167 var = NR_MSG_FRAG_IDS - 1; \
168 else \
169 var--; \
170 } while (0)
171
172 #define sk_msg_iter_var_next(var) \
173 do { \
174 var++; \
175 if (var == NR_MSG_FRAG_IDS) \
176 var = 0; \
177 } while (0)
178
179 #define sk_msg_iter_prev(msg, which) \
180 sk_msg_iter_var_prev(msg->sg.which)
181
182 #define sk_msg_iter_next(msg, which) \
183 sk_msg_iter_var_next(msg->sg.which)
184
sk_msg_init(struct sk_msg * msg)185 static inline void sk_msg_init(struct sk_msg *msg)
186 {
187 BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
188 memset(msg, 0, sizeof(*msg));
189 sg_init_marker(msg->sg.data, NR_MSG_FRAG_IDS);
190 }
191
sk_msg_xfer(struct sk_msg * dst,struct sk_msg * src,int which,u32 size)192 static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
193 int which, u32 size)
194 {
195 dst->sg.data[which] = src->sg.data[which];
196 dst->sg.data[which].length = size;
197 dst->sg.size += size;
198 src->sg.size -= size;
199 src->sg.data[which].length -= size;
200 src->sg.data[which].offset += size;
201 }
202
sk_msg_xfer_full(struct sk_msg * dst,struct sk_msg * src)203 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
204 {
205 memcpy(dst, src, sizeof(*src));
206 sk_msg_init(src);
207 }
208
sk_msg_full(const struct sk_msg * msg)209 static inline bool sk_msg_full(const struct sk_msg *msg)
210 {
211 return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
212 }
213
sk_msg_elem_used(const struct sk_msg * msg)214 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
215 {
216 return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
217 }
218
sk_msg_elem(struct sk_msg * msg,int which)219 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
220 {
221 return &msg->sg.data[which];
222 }
223
sk_msg_elem_cpy(struct sk_msg * msg,int which)224 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
225 {
226 return msg->sg.data[which];
227 }
228
sk_msg_page(struct sk_msg * msg,int which)229 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
230 {
231 return sg_page(sk_msg_elem(msg, which));
232 }
233
sk_msg_to_ingress(const struct sk_msg * msg)234 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
235 {
236 return msg->flags & BPF_F_INGRESS;
237 }
238
sk_msg_compute_data_pointers(struct sk_msg * msg)239 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
240 {
241 struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
242
243 if (test_bit(msg->sg.start, msg->sg.copy)) {
244 msg->data = NULL;
245 msg->data_end = NULL;
246 } else {
247 msg->data = sg_virt(sge);
248 msg->data_end = msg->data + sge->length;
249 }
250 }
251
sk_msg_page_add(struct sk_msg * msg,struct page * page,u32 len,u32 offset)252 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
253 u32 len, u32 offset)
254 {
255 struct scatterlist *sge;
256
257 get_page(page);
258 sge = sk_msg_elem(msg, msg->sg.end);
259 sg_set_page(sge, page, len, offset);
260 sg_unmark_end(sge);
261
262 __set_bit(msg->sg.end, msg->sg.copy);
263 msg->sg.size += len;
264 sk_msg_iter_next(msg, end);
265 }
266
sk_msg_sg_copy(struct sk_msg * msg,u32 i,bool copy_state)267 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
268 {
269 do {
270 if (copy_state)
271 __set_bit(i, msg->sg.copy);
272 else
273 __clear_bit(i, msg->sg.copy);
274 sk_msg_iter_var_next(i);
275 if (i == msg->sg.end)
276 break;
277 } while (1);
278 }
279
sk_msg_sg_copy_set(struct sk_msg * msg,u32 start)280 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
281 {
282 sk_msg_sg_copy(msg, start, true);
283 }
284
sk_msg_sg_copy_clear(struct sk_msg * msg,u32 start)285 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
286 {
287 sk_msg_sg_copy(msg, start, false);
288 }
289
sk_psock(const struct sock * sk)290 static inline struct sk_psock *sk_psock(const struct sock *sk)
291 {
292 return __rcu_dereference_sk_user_data_with_flags(sk,
293 SK_USER_DATA_PSOCK);
294 }
295
sk_psock_set_state(struct sk_psock * psock,enum sk_psock_state_bits bit)296 static inline void sk_psock_set_state(struct sk_psock *psock,
297 enum sk_psock_state_bits bit)
298 {
299 set_bit(bit, &psock->state);
300 }
301
sk_psock_clear_state(struct sk_psock * psock,enum sk_psock_state_bits bit)302 static inline void sk_psock_clear_state(struct sk_psock *psock,
303 enum sk_psock_state_bits bit)
304 {
305 clear_bit(bit, &psock->state);
306 }
307
sk_psock_test_state(const struct sk_psock * psock,enum sk_psock_state_bits bit)308 static inline bool sk_psock_test_state(const struct sk_psock *psock,
309 enum sk_psock_state_bits bit)
310 {
311 return test_bit(bit, &psock->state);
312 }
313
sock_drop(struct sock * sk,struct sk_buff * skb)314 static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
315 {
316 sk_drops_add(sk, skb);
317 kfree_skb(skb);
318 }
319
sk_psock_queue_msg(struct sk_psock * psock,struct sk_msg * msg)320 static inline bool sk_psock_queue_msg(struct sk_psock *psock,
321 struct sk_msg *msg)
322 {
323 bool ret;
324
325 spin_lock_bh(&psock->ingress_lock);
326 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
327 list_add_tail(&msg->list, &psock->ingress_msg);
328 ret = true;
329 } else {
330 sk_msg_free(psock->sk, msg);
331 kfree(msg);
332 ret = false;
333 }
334 spin_unlock_bh(&psock->ingress_lock);
335 return ret;
336 }
337
sk_psock_dequeue_msg(struct sk_psock * psock)338 static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
339 {
340 struct sk_msg *msg;
341
342 spin_lock_bh(&psock->ingress_lock);
343 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
344 if (msg)
345 list_del(&msg->list);
346 spin_unlock_bh(&psock->ingress_lock);
347 return msg;
348 }
349
sk_psock_peek_msg(struct sk_psock * psock)350 static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
351 {
352 struct sk_msg *msg;
353
354 spin_lock_bh(&psock->ingress_lock);
355 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
356 spin_unlock_bh(&psock->ingress_lock);
357 return msg;
358 }
359
sk_psock_next_msg(struct sk_psock * psock,struct sk_msg * msg)360 static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
361 struct sk_msg *msg)
362 {
363 struct sk_msg *ret;
364
365 spin_lock_bh(&psock->ingress_lock);
366 if (list_is_last(&msg->list, &psock->ingress_msg))
367 ret = NULL;
368 else
369 ret = list_next_entry(msg, list);
370 spin_unlock_bh(&psock->ingress_lock);
371 return ret;
372 }
373
sk_psock_queue_empty(const struct sk_psock * psock)374 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
375 {
376 return psock ? list_empty(&psock->ingress_msg) : true;
377 }
378
kfree_sk_msg(struct sk_msg * msg)379 static inline void kfree_sk_msg(struct sk_msg *msg)
380 {
381 if (msg->skb)
382 consume_skb(msg->skb);
383 kfree(msg);
384 }
385
sk_psock_report_error(struct sk_psock * psock,int err)386 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
387 {
388 struct sock *sk = psock->sk;
389
390 sk->sk_err = err;
391 sk_error_report(sk);
392 }
393
394 struct sk_psock *sk_psock_init(struct sock *sk, int node);
395 void sk_psock_stop(struct sk_psock *psock);
396
397 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
398 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
399 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
400 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
401 #else
sk_psock_init_strp(struct sock * sk,struct sk_psock * psock)402 static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
403 {
404 return -EOPNOTSUPP;
405 }
406
sk_psock_start_strp(struct sock * sk,struct sk_psock * psock)407 static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
408 {
409 }
410
sk_psock_stop_strp(struct sock * sk,struct sk_psock * psock)411 static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
412 {
413 }
414 #endif
415
416 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
417 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
418
419 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
420 struct sk_msg *msg);
421
422 /*
423 * This specialized allocator has to be a macro for its allocations to be
424 * accounted separately (to have a separate alloc_tag). The typecast is
425 * intentional to enforce typesafety.
426 */
427 #define sk_psock_init_link() \
428 ((struct sk_psock_link *)kzalloc(sizeof(struct sk_psock_link), \
429 GFP_ATOMIC | __GFP_NOWARN))
430
sk_psock_free_link(struct sk_psock_link * link)431 static inline void sk_psock_free_link(struct sk_psock_link *link)
432 {
433 kfree(link);
434 }
435
436 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
437
sk_psock_cork_free(struct sk_psock * psock)438 static inline void sk_psock_cork_free(struct sk_psock *psock)
439 {
440 if (psock->cork) {
441 sk_msg_free(psock->sk, psock->cork);
442 kfree(psock->cork);
443 psock->cork = NULL;
444 }
445 }
446
sk_psock_restore_proto(struct sock * sk,struct sk_psock * psock)447 static inline void sk_psock_restore_proto(struct sock *sk,
448 struct sk_psock *psock)
449 {
450 if (psock->psock_update_sk_prot)
451 psock->psock_update_sk_prot(sk, psock, true);
452 }
453
sk_psock_get(struct sock * sk)454 static inline struct sk_psock *sk_psock_get(struct sock *sk)
455 {
456 struct sk_psock *psock;
457
458 rcu_read_lock();
459 psock = sk_psock(sk);
460 if (psock && !refcount_inc_not_zero(&psock->refcnt))
461 psock = NULL;
462 rcu_read_unlock();
463 return psock;
464 }
465
466 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
467
sk_psock_put(struct sock * sk,struct sk_psock * psock)468 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
469 {
470 if (refcount_dec_and_test(&psock->refcnt))
471 sk_psock_drop(sk, psock);
472 }
473
sk_psock_data_ready(struct sock * sk,struct sk_psock * psock)474 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
475 {
476 read_lock_bh(&sk->sk_callback_lock);
477 if (psock->saved_data_ready)
478 psock->saved_data_ready(sk);
479 else
480 sk->sk_data_ready(sk);
481 read_unlock_bh(&sk->sk_callback_lock);
482 }
483
psock_set_prog(struct bpf_prog ** pprog,struct bpf_prog * prog)484 static inline void psock_set_prog(struct bpf_prog **pprog,
485 struct bpf_prog *prog)
486 {
487 prog = xchg(pprog, prog);
488 if (prog)
489 bpf_prog_put(prog);
490 }
491
psock_replace_prog(struct bpf_prog ** pprog,struct bpf_prog * prog,struct bpf_prog * old)492 static inline int psock_replace_prog(struct bpf_prog **pprog,
493 struct bpf_prog *prog,
494 struct bpf_prog *old)
495 {
496 if (cmpxchg(pprog, old, prog) != old)
497 return -ENOENT;
498
499 if (old)
500 bpf_prog_put(old);
501
502 return 0;
503 }
504
psock_progs_drop(struct sk_psock_progs * progs)505 static inline void psock_progs_drop(struct sk_psock_progs *progs)
506 {
507 psock_set_prog(&progs->msg_parser, NULL);
508 psock_set_prog(&progs->stream_parser, NULL);
509 psock_set_prog(&progs->stream_verdict, NULL);
510 psock_set_prog(&progs->skb_verdict, NULL);
511 }
512
513 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
514
sk_psock_strp_enabled(struct sk_psock * psock)515 static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
516 {
517 if (!psock)
518 return false;
519 return !!psock->saved_data_ready;
520 }
521
522 #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
523
524 #define BPF_F_STRPARSER (1UL << 1)
525
526 /* We only have two bits so far. */
527 #define BPF_F_PTR_MASK ~(BPF_F_INGRESS | BPF_F_STRPARSER)
528
skb_bpf_strparser(const struct sk_buff * skb)529 static inline bool skb_bpf_strparser(const struct sk_buff *skb)
530 {
531 unsigned long sk_redir = skb->_sk_redir;
532
533 return sk_redir & BPF_F_STRPARSER;
534 }
535
skb_bpf_set_strparser(struct sk_buff * skb)536 static inline void skb_bpf_set_strparser(struct sk_buff *skb)
537 {
538 skb->_sk_redir |= BPF_F_STRPARSER;
539 }
540
skb_bpf_ingress(const struct sk_buff * skb)541 static inline bool skb_bpf_ingress(const struct sk_buff *skb)
542 {
543 unsigned long sk_redir = skb->_sk_redir;
544
545 return sk_redir & BPF_F_INGRESS;
546 }
547
skb_bpf_set_ingress(struct sk_buff * skb)548 static inline void skb_bpf_set_ingress(struct sk_buff *skb)
549 {
550 skb->_sk_redir |= BPF_F_INGRESS;
551 }
552
skb_bpf_set_redir(struct sk_buff * skb,struct sock * sk_redir,bool ingress)553 static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
554 bool ingress)
555 {
556 skb->_sk_redir = (unsigned long)sk_redir;
557 if (ingress)
558 skb->_sk_redir |= BPF_F_INGRESS;
559 }
560
skb_bpf_redirect_fetch(const struct sk_buff * skb)561 static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
562 {
563 unsigned long sk_redir = skb->_sk_redir;
564
565 return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
566 }
567
skb_bpf_redirect_clear(struct sk_buff * skb)568 static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
569 {
570 skb->_sk_redir = 0;
571 }
572 #endif /* CONFIG_NET_SOCK_MSG */
573 #endif /* _LINUX_SKMSG_H */
574