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 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
67enum sk_psock_state_bits {
68 SK_PSOCK_TX_ENABLED,
69 SK_PSOCK_RX_STRP_ENABLED,
70};
71
72struct sk_psock_link {
73 struct list_head list;
74 struct bpf_map *map;
75 void *link_raw;
76};
77
78struct sk_psock_work_state {
79 u32 len;
80 u32 off;
81};
82
83struct 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 u32 copied_seq;
95 u32 ingress_bytes;
96#endif
97 struct sk_buff_head ingress_skb;
98 struct list_head ingress_msg;
99 spinlock_t ingress_lock;
100 unsigned long state;
101 struct list_head link;
102 spinlock_t link_lock;
103 refcount_t refcnt;
104 void (*saved_unhash)(struct sock *sk);
105 void (*saved_destroy)(struct sock *sk);
106 void (*saved_close)(struct sock *sk, long timeout);
107 void (*saved_write_space)(struct sock *sk);
108 void (*saved_data_ready)(struct sock *sk);
109 /* psock_update_sk_prot may be called with restore=false many times
110 * so the handler must be safe for this case. It will be called
111 * exactly once with restore=true when the psock is being destroyed
112 * and psock refcnt is zero, but before an RCU grace period.
113 */
114 int (*psock_update_sk_prot)(struct sock *sk, struct sk_psock *psock,
115 bool restore);
116 struct proto *sk_proto;
117 struct mutex work_mutex;
118 struct sk_psock_work_state work_state;
119 struct delayed_work work;
120 struct sock *sk_pair;
121 struct rcu_work rwork;
122};
123
124int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
125 int elem_first_coalesce);
126int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
127 u32 off, u32 len);
128void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
129int sk_msg_free(struct sock *sk, struct sk_msg *msg);
130int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
131void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
132void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
133 u32 bytes);
134
135void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
136void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
137
138int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
139 struct sk_msg *msg, u32 bytes);
140int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
141 struct sk_msg *msg, u32 bytes);
142int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
143 int len, int flags);
144bool sk_msg_is_readable(struct sock *sk);
145
146static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
147{
148 WARN_ON(i == msg->sg.end && bytes);
149}
150
151static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
152{
153 if (psock->apply_bytes) {
154 if (psock->apply_bytes < bytes)
155 psock->apply_bytes = 0;
156 else
157 psock->apply_bytes -= bytes;
158 }
159}
160
161static inline u32 sk_msg_iter_dist(u32 start, u32 end)
162{
163 return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
164}
165
166#define sk_msg_iter_var_prev(var) \
167 do { \
168 if (var == 0) \
169 var = NR_MSG_FRAG_IDS - 1; \
170 else \
171 var--; \
172 } while (0)
173
174#define sk_msg_iter_var_next(var) \
175 do { \
176 var++; \
177 if (var == NR_MSG_FRAG_IDS) \
178 var = 0; \
179 } while (0)
180
181#define sk_msg_iter_prev(msg, which) \
182 sk_msg_iter_var_prev(msg->sg.which)
183
184#define sk_msg_iter_next(msg, which) \
185 sk_msg_iter_var_next(msg->sg.which)
186
187static inline void sk_msg_init(struct sk_msg *msg)
188{
189 BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
190 memset(s: msg, c: 0, n: sizeof(*msg));
191 sg_init_marker(sgl: msg->sg.data, NR_MSG_FRAG_IDS);
192}
193
194static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
195 int which, u32 size)
196{
197 dst->sg.data[which] = src->sg.data[which];
198 dst->sg.data[which].length = size;
199 dst->sg.size += size;
200 src->sg.size -= size;
201 src->sg.data[which].length -= size;
202 src->sg.data[which].offset += size;
203}
204
205static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
206{
207 memcpy(to: dst, from: src, len: sizeof(*src));
208 sk_msg_init(msg: src);
209}
210
211static inline bool sk_msg_full(const struct sk_msg *msg)
212{
213 return sk_msg_iter_dist(start: msg->sg.start, end: msg->sg.end) == MAX_MSG_FRAGS;
214}
215
216static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
217{
218 return sk_msg_iter_dist(start: msg->sg.start, end: msg->sg.end);
219}
220
221static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
222{
223 return &msg->sg.data[which];
224}
225
226static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
227{
228 return msg->sg.data[which];
229}
230
231static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
232{
233 return sg_page(sg: sk_msg_elem(msg, which));
234}
235
236static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
237{
238 return msg->flags & BPF_F_INGRESS;
239}
240
241static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
242{
243 struct scatterlist *sge = sk_msg_elem(msg, which: msg->sg.start);
244
245 if (test_bit(msg->sg.start, msg->sg.copy)) {
246 msg->data = NULL;
247 msg->data_end = NULL;
248 } else {
249 msg->data = sg_virt(sg: sge);
250 msg->data_end = msg->data + sge->length;
251 }
252}
253
254static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
255 u32 len, u32 offset)
256{
257 struct scatterlist *sge;
258
259 get_page(page);
260 sge = sk_msg_elem(msg, which: msg->sg.end);
261 sg_set_page(sg: sge, page, len, offset);
262 sg_unmark_end(sg: sge);
263
264 __set_bit(msg->sg.end, msg->sg.copy);
265 msg->sg.size += len;
266 sk_msg_iter_next(msg, end);
267}
268
269static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
270{
271 do {
272 if (copy_state)
273 __set_bit(i, msg->sg.copy);
274 else
275 __clear_bit(i, msg->sg.copy);
276 sk_msg_iter_var_next(i);
277 if (i == msg->sg.end)
278 break;
279 } while (1);
280}
281
282static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
283{
284 sk_msg_sg_copy(msg, i: start, copy_state: true);
285}
286
287static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
288{
289 sk_msg_sg_copy(msg, i: start, copy_state: false);
290}
291
292static inline struct sk_psock *sk_psock(const struct sock *sk)
293{
294 return __rcu_dereference_sk_user_data_with_flags(sk,
295 SK_USER_DATA_PSOCK);
296}
297
298static inline void sk_psock_set_state(struct sk_psock *psock,
299 enum sk_psock_state_bits bit)
300{
301 set_bit(nr: bit, addr: &psock->state);
302}
303
304static inline void sk_psock_clear_state(struct sk_psock *psock,
305 enum sk_psock_state_bits bit)
306{
307 clear_bit(nr: bit, addr: &psock->state);
308}
309
310static inline bool sk_psock_test_state(const struct sk_psock *psock,
311 enum sk_psock_state_bits bit)
312{
313 return test_bit(bit, &psock->state);
314}
315
316static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
317{
318 sk_drops_skbadd(sk, skb);
319 kfree_skb(skb);
320}
321
322static inline bool sk_psock_queue_msg(struct sk_psock *psock,
323 struct sk_msg *msg)
324{
325 bool ret;
326
327 spin_lock_bh(lock: &psock->ingress_lock);
328 if (sk_psock_test_state(psock, bit: SK_PSOCK_TX_ENABLED)) {
329 list_add_tail(new: &msg->list, head: &psock->ingress_msg);
330 ret = true;
331 } else {
332 sk_msg_free(sk: psock->sk, msg);
333 kfree(objp: msg);
334 ret = false;
335 }
336 spin_unlock_bh(lock: &psock->ingress_lock);
337 return ret;
338}
339
340static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
341{
342 struct sk_msg *msg;
343
344 spin_lock_bh(lock: &psock->ingress_lock);
345 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
346 if (msg)
347 list_del(entry: &msg->list);
348 spin_unlock_bh(lock: &psock->ingress_lock);
349 return msg;
350}
351
352static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
353{
354 struct sk_msg *msg;
355
356 spin_lock_bh(lock: &psock->ingress_lock);
357 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
358 spin_unlock_bh(lock: &psock->ingress_lock);
359 return msg;
360}
361
362static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
363 struct sk_msg *msg)
364{
365 struct sk_msg *ret;
366
367 spin_lock_bh(lock: &psock->ingress_lock);
368 if (list_is_last(list: &msg->list, head: &psock->ingress_msg))
369 ret = NULL;
370 else
371 ret = list_next_entry(msg, list);
372 spin_unlock_bh(lock: &psock->ingress_lock);
373 return ret;
374}
375
376static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
377{
378 return psock ? list_empty(head: &psock->ingress_msg) : true;
379}
380
381static inline void kfree_sk_msg(struct sk_msg *msg)
382{
383 if (msg->skb)
384 consume_skb(skb: msg->skb);
385 kfree(objp: msg);
386}
387
388static inline void sk_psock_report_error(struct sk_psock *psock, int err)
389{
390 struct sock *sk = psock->sk;
391
392 sk->sk_err = err;
393 sk_error_report(sk);
394}
395
396struct sk_psock *sk_psock_init(struct sock *sk, int node);
397void sk_psock_stop(struct sk_psock *psock);
398
399#if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
400int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
401void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
402void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
403#else
404static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
405{
406 return -EOPNOTSUPP;
407}
408
409static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
410{
411}
412
413static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
414{
415}
416#endif
417
418void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
419void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
420
421int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
422 struct sk_msg *msg);
423
424/*
425 * This specialized allocator has to be a macro for its allocations to be
426 * accounted separately (to have a separate alloc_tag). The typecast is
427 * intentional to enforce typesafety.
428 */
429#define sk_psock_init_link() \
430 ((struct sk_psock_link *)kzalloc(sizeof(struct sk_psock_link), \
431 GFP_ATOMIC | __GFP_NOWARN))
432
433static inline void sk_psock_free_link(struct sk_psock_link *link)
434{
435 kfree(objp: link);
436}
437
438struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
439
440static inline void sk_psock_cork_free(struct sk_psock *psock)
441{
442 if (psock->cork) {
443 sk_msg_free(sk: psock->sk, msg: psock->cork);
444 kfree(objp: psock->cork);
445 psock->cork = NULL;
446 }
447}
448
449static inline void sk_psock_restore_proto(struct sock *sk,
450 struct sk_psock *psock)
451{
452 if (psock->psock_update_sk_prot)
453 psock->psock_update_sk_prot(sk, psock, true);
454}
455
456static inline struct sk_psock *sk_psock_get(struct sock *sk)
457{
458 struct sk_psock *psock;
459
460 rcu_read_lock();
461 psock = sk_psock(sk);
462 if (psock && !refcount_inc_not_zero(r: &psock->refcnt))
463 psock = NULL;
464 rcu_read_unlock();
465 return psock;
466}
467
468void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
469
470static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
471{
472 if (refcount_dec_and_test(r: &psock->refcnt))
473 sk_psock_drop(sk, psock);
474}
475
476static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
477{
478 read_lock_bh(&sk->sk_callback_lock);
479 if (psock->saved_data_ready)
480 psock->saved_data_ready(sk);
481 else
482 sk->sk_data_ready(sk);
483 read_unlock_bh(&sk->sk_callback_lock);
484}
485
486static inline void psock_set_prog(struct bpf_prog **pprog,
487 struct bpf_prog *prog)
488{
489 prog = xchg(pprog, prog);
490 if (prog)
491 bpf_prog_put(prog);
492}
493
494static inline int psock_replace_prog(struct bpf_prog **pprog,
495 struct bpf_prog *prog,
496 struct bpf_prog *old)
497{
498 if (cmpxchg(pprog, old, prog) != old)
499 return -ENOENT;
500
501 if (old)
502 bpf_prog_put(prog: old);
503
504 return 0;
505}
506
507static inline void psock_progs_drop(struct sk_psock_progs *progs)
508{
509 psock_set_prog(pprog: &progs->msg_parser, NULL);
510 psock_set_prog(pprog: &progs->stream_parser, NULL);
511 psock_set_prog(pprog: &progs->stream_verdict, NULL);
512 psock_set_prog(pprog: &progs->skb_verdict, NULL);
513}
514
515int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
516
517static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
518{
519 if (!psock)
520 return false;
521 return !!psock->saved_data_ready;
522}
523
524#if IS_ENABLED(CONFIG_NET_SOCK_MSG)
525
526#define BPF_F_STRPARSER (1UL << 1)
527
528/* We only have two bits so far. */
529#define BPF_F_PTR_MASK ~(BPF_F_INGRESS | BPF_F_STRPARSER)
530
531static inline bool skb_bpf_strparser(const struct sk_buff *skb)
532{
533 unsigned long sk_redir = skb->_sk_redir;
534
535 return sk_redir & BPF_F_STRPARSER;
536}
537
538static inline void skb_bpf_set_strparser(struct sk_buff *skb)
539{
540 skb->_sk_redir |= BPF_F_STRPARSER;
541}
542
543static inline bool skb_bpf_ingress(const struct sk_buff *skb)
544{
545 unsigned long sk_redir = skb->_sk_redir;
546
547 return sk_redir & BPF_F_INGRESS;
548}
549
550static inline void skb_bpf_set_ingress(struct sk_buff *skb)
551{
552 skb->_sk_redir |= BPF_F_INGRESS;
553}
554
555static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
556 bool ingress)
557{
558 skb->_sk_redir = (unsigned long)sk_redir;
559 if (ingress)
560 skb->_sk_redir |= BPF_F_INGRESS;
561}
562
563static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
564{
565 unsigned long sk_redir = skb->_sk_redir;
566
567 return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
568}
569
570static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
571{
572 skb->_sk_redir = 0;
573}
574#endif /* CONFIG_NET_SOCK_MSG */
575#endif /* _LINUX_SKMSG_H */
576