xref: /linux-6.15/include/linux/skmsg.h (revision ef700f2e) 
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 	unsigned long			state;
93 	struct list_head		link;
94 	spinlock_t			link_lock;
95 	refcount_t			refcnt;
96 	void (*saved_unhash)(struct sock *sk);
97 	void (*saved_close)(struct sock *sk, long timeout);
98 	void (*saved_write_space)(struct sock *sk);
99 	void (*saved_data_ready)(struct sock *sk);
100 	struct proto			*sk_proto;
101 	struct sk_psock_work_state	work_state;
102 	struct work_struct		work;
103 	union {
104 		struct rcu_head		rcu;
105 		struct work_struct	gc;
106 	};
107 };
108 
109 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
110 		 int elem_first_coalesce);
111 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
112 		 u32 off, u32 len);
113 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
114 int sk_msg_free(struct sock *sk, struct sk_msg *msg);
115 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
116 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
117 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
118 				  u32 bytes);
119 
120 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
121 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
122 
123 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
124 			      struct sk_msg *msg, u32 bytes);
125 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
126 			     struct sk_msg *msg, u32 bytes);
127 
128 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
129 {
130 	WARN_ON(i == msg->sg.end && bytes);
131 }
132 
133 static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
134 {
135 	if (psock->apply_bytes) {
136 		if (psock->apply_bytes < bytes)
137 			psock->apply_bytes = 0;
138 		else
139 			psock->apply_bytes -= bytes;
140 	}
141 }
142 
143 static inline u32 sk_msg_iter_dist(u32 start, u32 end)
144 {
145 	return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
146 }
147 
148 #define sk_msg_iter_var_prev(var)			\
149 	do {						\
150 		if (var == 0)				\
151 			var = NR_MSG_FRAG_IDS - 1;	\
152 		else					\
153 			var--;				\
154 	} while (0)
155 
156 #define sk_msg_iter_var_next(var)			\
157 	do {						\
158 		var++;					\
159 		if (var == NR_MSG_FRAG_IDS)		\
160 			var = 0;			\
161 	} while (0)
162 
163 #define sk_msg_iter_prev(msg, which)			\
164 	sk_msg_iter_var_prev(msg->sg.which)
165 
166 #define sk_msg_iter_next(msg, which)			\
167 	sk_msg_iter_var_next(msg->sg.which)
168 
169 static inline void sk_msg_clear_meta(struct sk_msg *msg)
170 {
171 	memset(&msg->sg, 0, offsetofend(struct sk_msg_sg, copy));
172 }
173 
174 static inline void sk_msg_init(struct sk_msg *msg)
175 {
176 	BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
177 	memset(msg, 0, sizeof(*msg));
178 	sg_init_marker(msg->sg.data, NR_MSG_FRAG_IDS);
179 }
180 
181 static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
182 			       int which, u32 size)
183 {
184 	dst->sg.data[which] = src->sg.data[which];
185 	dst->sg.data[which].length  = size;
186 	dst->sg.size		   += size;
187 	src->sg.size		   -= size;
188 	src->sg.data[which].length -= size;
189 	src->sg.data[which].offset += size;
190 }
191 
192 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
193 {
194 	memcpy(dst, src, sizeof(*src));
195 	sk_msg_init(src);
196 }
197 
198 static inline bool sk_msg_full(const struct sk_msg *msg)
199 {
200 	return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
201 }
202 
203 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
204 {
205 	return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
206 }
207 
208 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
209 {
210 	return &msg->sg.data[which];
211 }
212 
213 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
214 {
215 	return msg->sg.data[which];
216 }
217 
218 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
219 {
220 	return sg_page(sk_msg_elem(msg, which));
221 }
222 
223 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
224 {
225 	return msg->flags & BPF_F_INGRESS;
226 }
227 
228 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
229 {
230 	struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
231 
232 	if (test_bit(msg->sg.start, &msg->sg.copy)) {
233 		msg->data = NULL;
234 		msg->data_end = NULL;
235 	} else {
236 		msg->data = sg_virt(sge);
237 		msg->data_end = msg->data + sge->length;
238 	}
239 }
240 
241 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
242 				   u32 len, u32 offset)
243 {
244 	struct scatterlist *sge;
245 
246 	get_page(page);
247 	sge = sk_msg_elem(msg, msg->sg.end);
248 	sg_set_page(sge, page, len, offset);
249 	sg_unmark_end(sge);
250 
251 	__set_bit(msg->sg.end, &msg->sg.copy);
252 	msg->sg.size += len;
253 	sk_msg_iter_next(msg, end);
254 }
255 
256 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
257 {
258 	do {
259 		if (copy_state)
260 			__set_bit(i, &msg->sg.copy);
261 		else
262 			__clear_bit(i, &msg->sg.copy);
263 		sk_msg_iter_var_next(i);
264 		if (i == msg->sg.end)
265 			break;
266 	} while (1);
267 }
268 
269 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
270 {
271 	sk_msg_sg_copy(msg, start, true);
272 }
273 
274 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
275 {
276 	sk_msg_sg_copy(msg, start, false);
277 }
278 
279 static inline struct sk_psock *sk_psock(const struct sock *sk)
280 {
281 	return rcu_dereference_sk_user_data(sk);
282 }
283 
284 static inline void sk_psock_queue_msg(struct sk_psock *psock,
285 				      struct sk_msg *msg)
286 {
287 	list_add_tail(&msg->list, &psock->ingress_msg);
288 }
289 
290 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
291 {
292 	return psock ? list_empty(&psock->ingress_msg) : true;
293 }
294 
295 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
296 {
297 	struct sock *sk = psock->sk;
298 
299 	sk->sk_err = err;
300 	sk->sk_error_report(sk);
301 }
302 
303 struct sk_psock *sk_psock_init(struct sock *sk, int node);
304 
305 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
306 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
307 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
308 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
309 #else
310 static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
311 {
312 	return -EOPNOTSUPP;
313 }
314 
315 static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
316 {
317 }
318 
319 static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
320 {
321 }
322 #endif
323 
324 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
325 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
326 
327 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
328 			 struct sk_msg *msg);
329 
330 static inline struct sk_psock_link *sk_psock_init_link(void)
331 {
332 	return kzalloc(sizeof(struct sk_psock_link),
333 		       GFP_ATOMIC | __GFP_NOWARN);
334 }
335 
336 static inline void sk_psock_free_link(struct sk_psock_link *link)
337 {
338 	kfree(link);
339 }
340 
341 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
342 
343 static inline void sk_psock_cork_free(struct sk_psock *psock)
344 {
345 	if (psock->cork) {
346 		sk_msg_free(psock->sk, psock->cork);
347 		kfree(psock->cork);
348 		psock->cork = NULL;
349 	}
350 }
351 
352 static inline void sk_psock_update_proto(struct sock *sk,
353 					 struct sk_psock *psock,
354 					 struct proto *ops)
355 {
356 	/* Pairs with lockless read in sk_clone_lock() */
357 	WRITE_ONCE(sk->sk_prot, ops);
358 }
359 
360 static inline void sk_psock_restore_proto(struct sock *sk,
361 					  struct sk_psock *psock)
362 {
363 	sk->sk_prot->unhash = psock->saved_unhash;
364 	if (inet_csk_has_ulp(sk)) {
365 		tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
366 	} else {
367 		sk->sk_write_space = psock->saved_write_space;
368 		/* Pairs with lockless read in sk_clone_lock() */
369 		WRITE_ONCE(sk->sk_prot, psock->sk_proto);
370 	}
371 }
372 
373 static inline void sk_psock_set_state(struct sk_psock *psock,
374 				      enum sk_psock_state_bits bit)
375 {
376 	set_bit(bit, &psock->state);
377 }
378 
379 static inline void sk_psock_clear_state(struct sk_psock *psock,
380 					enum sk_psock_state_bits bit)
381 {
382 	clear_bit(bit, &psock->state);
383 }
384 
385 static inline bool sk_psock_test_state(const struct sk_psock *psock,
386 				       enum sk_psock_state_bits bit)
387 {
388 	return test_bit(bit, &psock->state);
389 }
390 
391 static inline struct sk_psock *sk_psock_get(struct sock *sk)
392 {
393 	struct sk_psock *psock;
394 
395 	rcu_read_lock();
396 	psock = sk_psock(sk);
397 	if (psock && !refcount_inc_not_zero(&psock->refcnt))
398 		psock = NULL;
399 	rcu_read_unlock();
400 	return psock;
401 }
402 
403 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
404 
405 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
406 {
407 	if (refcount_dec_and_test(&psock->refcnt))
408 		sk_psock_drop(sk, psock);
409 }
410 
411 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
412 {
413 	if (psock->saved_data_ready)
414 		psock->saved_data_ready(sk);
415 	else
416 		sk->sk_data_ready(sk);
417 }
418 
419 static inline void psock_set_prog(struct bpf_prog **pprog,
420 				  struct bpf_prog *prog)
421 {
422 	prog = xchg(pprog, prog);
423 	if (prog)
424 		bpf_prog_put(prog);
425 }
426 
427 static inline int psock_replace_prog(struct bpf_prog **pprog,
428 				     struct bpf_prog *prog,
429 				     struct bpf_prog *old)
430 {
431 	if (cmpxchg(pprog, old, prog) != old)
432 		return -ENOENT;
433 
434 	if (old)
435 		bpf_prog_put(old);
436 
437 	return 0;
438 }
439 
440 static inline void psock_progs_drop(struct sk_psock_progs *progs)
441 {
442 	psock_set_prog(&progs->msg_parser, NULL);
443 	psock_set_prog(&progs->stream_parser, NULL);
444 	psock_set_prog(&progs->stream_verdict, NULL);
445 }
446 
447 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
448 
449 static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
450 {
451 	if (!psock)
452 		return false;
453 	return !!psock->saved_data_ready;
454 }
455 
456 #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
457 
458 /* We only have one bit so far. */
459 #define BPF_F_PTR_MASK ~(BPF_F_INGRESS)
460 
461 static inline bool skb_bpf_ingress(const struct sk_buff *skb)
462 {
463 	unsigned long sk_redir = skb->_sk_redir;
464 
465 	return sk_redir & BPF_F_INGRESS;
466 }
467 
468 static inline void skb_bpf_set_ingress(struct sk_buff *skb)
469 {
470 	skb->_sk_redir |= BPF_F_INGRESS;
471 }
472 
473 static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
474 				     bool ingress)
475 {
476 	skb->_sk_redir = (unsigned long)sk_redir;
477 	if (ingress)
478 		skb->_sk_redir |= BPF_F_INGRESS;
479 }
480 
481 static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
482 {
483 	unsigned long sk_redir = skb->_sk_redir;
484 
485 	return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
486 }
487 
488 static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
489 {
490 	skb->_sk_redir = 0;
491 }
492 #endif /* CONFIG_NET_SOCK_MSG */
493 #endif /* _LINUX_SKMSG_H */
494