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
19enum __sk_action {
20 __SK_DROP = 0,
21 __SK_PASS,
22 __SK_REDIRECT,
23 __SK_NONE,
24};
25
26struct 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. */
43struct 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
56struct 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};
62
63enum sk_psock_state_bits {
64 SK_PSOCK_TX_ENABLED,
65 SK_PSOCK_RX_STRP_ENABLED,
66};
67
68struct sk_psock_link {
69 struct list_head list;
70 struct bpf_map *map;
71 void *link_raw;
72};
73
74struct sk_psock_work_state {
75 u32 len;
76 u32 off;
77};
78
79struct sk_psock {
80 struct sock *sk;
81 struct sock *sk_redir;
82 u32 apply_bytes;
83 u32 cork_bytes;
84 u32 eval;
85 bool redir_ingress; /* undefined if sk_redir is null */
86 struct sk_msg *cork;
87 struct sk_psock_progs progs;
88#if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
89 struct strparser strp;
90#endif
91 struct sk_buff_head ingress_skb;
92 struct list_head ingress_msg;
93 spinlock_t ingress_lock;
94 unsigned long state;
95 struct list_head link;
96 spinlock_t link_lock;
97 refcount_t refcnt;
98 void (*saved_unhash)(struct sock *sk);
99 void (*saved_destroy)(struct sock *sk);
100 void (*saved_close)(struct sock *sk, long timeout);
101 void (*saved_write_space)(struct sock *sk);
102 void (*saved_data_ready)(struct sock *sk);
103 int (*psock_update_sk_prot)(struct sock *sk, struct sk_psock *psock,
104 bool restore);
105 struct proto *sk_proto;
106 struct mutex work_mutex;
107 struct sk_psock_work_state work_state;
108 struct delayed_work work;
109 struct rcu_work rwork;
110};
111
112int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
113 int elem_first_coalesce);
114int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
115 u32 off, u32 len);
116void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
117int sk_msg_free(struct sock *sk, struct sk_msg *msg);
118int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
119void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
120void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
121 u32 bytes);
122
123void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
124void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
125
126int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
127 struct sk_msg *msg, u32 bytes);
128int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
129 struct sk_msg *msg, u32 bytes);
130int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
131 int len, int flags);
132bool sk_msg_is_readable(struct sock *sk);
133
134static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
135{
136 WARN_ON(i == msg->sg.end && bytes);
137}
138
139static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
140{
141 if (psock->apply_bytes) {
142 if (psock->apply_bytes < bytes)
143 psock->apply_bytes = 0;
144 else
145 psock->apply_bytes -= bytes;
146 }
147}
148
149static inline u32 sk_msg_iter_dist(u32 start, u32 end)
150{
151 return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
152}
153
154#define sk_msg_iter_var_prev(var) \
155 do { \
156 if (var == 0) \
157 var = NR_MSG_FRAG_IDS - 1; \
158 else \
159 var--; \
160 } while (0)
161
162#define sk_msg_iter_var_next(var) \
163 do { \
164 var++; \
165 if (var == NR_MSG_FRAG_IDS) \
166 var = 0; \
167 } while (0)
168
169#define sk_msg_iter_prev(msg, which) \
170 sk_msg_iter_var_prev(msg->sg.which)
171
172#define sk_msg_iter_next(msg, which) \
173 sk_msg_iter_var_next(msg->sg.which)
174
175static inline void sk_msg_init(struct sk_msg *msg)
176{
177 BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
178 memset(msg, 0, sizeof(*msg));
179 sg_init_marker(sgl: msg->sg.data, NR_MSG_FRAG_IDS);
180}
181
182static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
183 int which, u32 size)
184{
185 dst->sg.data[which] = src->sg.data[which];
186 dst->sg.data[which].length = size;
187 dst->sg.size += size;
188 src->sg.size -= size;
189 src->sg.data[which].length -= size;
190 src->sg.data[which].offset += size;
191}
192
193static 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(msg: src);
197}
198
199static inline bool sk_msg_full(const struct sk_msg *msg)
200{
201 return sk_msg_iter_dist(start: msg->sg.start, end: msg->sg.end) == MAX_MSG_FRAGS;
202}
203
204static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
205{
206 return sk_msg_iter_dist(start: msg->sg.start, end: msg->sg.end);
207}
208
209static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
210{
211 return &msg->sg.data[which];
212}
213
214static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
215{
216 return msg->sg.data[which];
217}
218
219static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
220{
221 return sg_page(sg: sk_msg_elem(msg, which));
222}
223
224static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
225{
226 return msg->flags & BPF_F_INGRESS;
227}
228
229static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
230{
231 struct scatterlist *sge = sk_msg_elem(msg, which: msg->sg.start);
232
233 if (test_bit(msg->sg.start, msg->sg.copy)) {
234 msg->data = NULL;
235 msg->data_end = NULL;
236 } else {
237 msg->data = sg_virt(sg: sge);
238 msg->data_end = msg->data + sge->length;
239 }
240}
241
242static 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, which: msg->sg.end);
249 sg_set_page(sg: sge, page, len, offset);
250 sg_unmark_end(sg: sge);
251
252 __set_bit(msg->sg.end, msg->sg.copy);
253 msg->sg.size += len;
254 sk_msg_iter_next(msg, end);
255}
256
257static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
258{
259 do {
260 if (copy_state)
261 __set_bit(i, msg->sg.copy);
262 else
263 __clear_bit(i, msg->sg.copy);
264 sk_msg_iter_var_next(i);
265 if (i == msg->sg.end)
266 break;
267 } while (1);
268}
269
270static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
271{
272 sk_msg_sg_copy(msg, i: start, copy_state: true);
273}
274
275static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
276{
277 sk_msg_sg_copy(msg, i: start, copy_state: false);
278}
279
280static inline struct sk_psock *sk_psock(const struct sock *sk)
281{
282 return __rcu_dereference_sk_user_data_with_flags(sk,
283 SK_USER_DATA_PSOCK);
284}
285
286static inline void sk_psock_set_state(struct sk_psock *psock,
287 enum sk_psock_state_bits bit)
288{
289 set_bit(nr: bit, addr: &psock->state);
290}
291
292static inline void sk_psock_clear_state(struct sk_psock *psock,
293 enum sk_psock_state_bits bit)
294{
295 clear_bit(nr: bit, addr: &psock->state);
296}
297
298static inline bool sk_psock_test_state(const struct sk_psock *psock,
299 enum sk_psock_state_bits bit)
300{
301 return test_bit(bit, &psock->state);
302}
303
304static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
305{
306 sk_drops_add(sk, skb);
307 kfree_skb(skb);
308}
309
310static inline void sk_psock_queue_msg(struct sk_psock *psock,
311 struct sk_msg *msg)
312{
313 spin_lock_bh(lock: &psock->ingress_lock);
314 if (sk_psock_test_state(psock, bit: SK_PSOCK_TX_ENABLED))
315 list_add_tail(new: &msg->list, head: &psock->ingress_msg);
316 else {
317 sk_msg_free(sk: psock->sk, msg);
318 kfree(objp: msg);
319 }
320 spin_unlock_bh(lock: &psock->ingress_lock);
321}
322
323static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
324{
325 struct sk_msg *msg;
326
327 spin_lock_bh(lock: &psock->ingress_lock);
328 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
329 if (msg)
330 list_del(entry: &msg->list);
331 spin_unlock_bh(lock: &psock->ingress_lock);
332 return msg;
333}
334
335static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
336{
337 struct sk_msg *msg;
338
339 spin_lock_bh(lock: &psock->ingress_lock);
340 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
341 spin_unlock_bh(lock: &psock->ingress_lock);
342 return msg;
343}
344
345static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
346 struct sk_msg *msg)
347{
348 struct sk_msg *ret;
349
350 spin_lock_bh(lock: &psock->ingress_lock);
351 if (list_is_last(list: &msg->list, head: &psock->ingress_msg))
352 ret = NULL;
353 else
354 ret = list_next_entry(msg, list);
355 spin_unlock_bh(lock: &psock->ingress_lock);
356 return ret;
357}
358
359static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
360{
361 return psock ? list_empty(head: &psock->ingress_msg) : true;
362}
363
364static inline void kfree_sk_msg(struct sk_msg *msg)
365{
366 if (msg->skb)
367 consume_skb(skb: msg->skb);
368 kfree(objp: msg);
369}
370
371static inline void sk_psock_report_error(struct sk_psock *psock, int err)
372{
373 struct sock *sk = psock->sk;
374
375 sk->sk_err = err;
376 sk_error_report(sk);
377}
378
379struct sk_psock *sk_psock_init(struct sock *sk, int node);
380void sk_psock_stop(struct sk_psock *psock);
381
382#if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
383int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
384void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
385void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
386#else
387static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
388{
389 return -EOPNOTSUPP;
390}
391
392static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
393{
394}
395
396static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
397{
398}
399#endif
400
401void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
402void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
403
404int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
405 struct sk_msg *msg);
406
407static inline struct sk_psock_link *sk_psock_init_link(void)
408{
409 return kzalloc(size: sizeof(struct sk_psock_link),
410 GFP_ATOMIC | __GFP_NOWARN);
411}
412
413static inline void sk_psock_free_link(struct sk_psock_link *link)
414{
415 kfree(objp: link);
416}
417
418struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
419
420static inline void sk_psock_cork_free(struct sk_psock *psock)
421{
422 if (psock->cork) {
423 sk_msg_free(sk: psock->sk, msg: psock->cork);
424 kfree(objp: psock->cork);
425 psock->cork = NULL;
426 }
427}
428
429static inline void sk_psock_restore_proto(struct sock *sk,
430 struct sk_psock *psock)
431{
432 if (psock->psock_update_sk_prot)
433 psock->psock_update_sk_prot(sk, psock, true);
434}
435
436static 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(r: &psock->refcnt))
443 psock = NULL;
444 rcu_read_unlock();
445 return psock;
446}
447
448void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
449
450static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
451{
452 if (refcount_dec_and_test(r: &psock->refcnt))
453 sk_psock_drop(sk, psock);
454}
455
456static 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
464static 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
472static 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(prog: old);
481
482 return 0;
483}
484
485static inline void psock_progs_drop(struct sk_psock_progs *progs)
486{
487 psock_set_prog(pprog: &progs->msg_parser, NULL);
488 psock_set_prog(pprog: &progs->stream_parser, NULL);
489 psock_set_prog(pprog: &progs->stream_verdict, NULL);
490 psock_set_prog(pprog: &progs->skb_verdict, NULL);
491}
492
493int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
494
495static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
496{
497 if (!psock)
498 return false;
499 return !!psock->saved_data_ready;
500}
501
502static inline bool sk_is_udp(const struct sock *sk)
503{
504 return sk->sk_type == SOCK_DGRAM &&
505 sk->sk_protocol == IPPROTO_UDP;
506}
507
508#if IS_ENABLED(CONFIG_NET_SOCK_MSG)
509
510#define BPF_F_STRPARSER (1UL << 1)
511
512/* We only have two bits so far. */
513#define BPF_F_PTR_MASK ~(BPF_F_INGRESS | BPF_F_STRPARSER)
514
515static inline bool skb_bpf_strparser(const struct sk_buff *skb)
516{
517 unsigned long sk_redir = skb->_sk_redir;
518
519 return sk_redir & BPF_F_STRPARSER;
520}
521
522static inline void skb_bpf_set_strparser(struct sk_buff *skb)
523{
524 skb->_sk_redir |= BPF_F_STRPARSER;
525}
526
527static inline bool skb_bpf_ingress(const struct sk_buff *skb)
528{
529 unsigned long sk_redir = skb->_sk_redir;
530
531 return sk_redir & BPF_F_INGRESS;
532}
533
534static inline void skb_bpf_set_ingress(struct sk_buff *skb)
535{
536 skb->_sk_redir |= BPF_F_INGRESS;
537}
538
539static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
540 bool ingress)
541{
542 skb->_sk_redir = (unsigned long)sk_redir;
543 if (ingress)
544 skb->_sk_redir |= BPF_F_INGRESS;
545}
546
547static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
548{
549 unsigned long sk_redir = skb->_sk_redir;
550
551 return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
552}
553
554static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
555{
556 skb->_sk_redir = 0;
557}
558#endif /* CONFIG_NET_SOCK_MSG */
559#endif /* _LINUX_SKMSG_H */
560

source code of linux/include/linux/skmsg.h