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
 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	/** @msg_tot_len: Total bytes queued in ingress_msg list. */
101	u32				msg_tot_len;
102	unsigned long			state;
103	struct list_head		link;
104	spinlock_t			link_lock;
105	refcount_t			refcnt;
106	void (*saved_unhash)(struct sock *sk);
107	void (*saved_destroy)(struct sock *sk);
108	void (*saved_close)(struct sock *sk, long timeout);
109	void (*saved_write_space)(struct sock *sk);
110	void (*saved_data_ready)(struct sock *sk);
111	/* psock_update_sk_prot may be called with restore=false many times
112	 * so the handler must be safe for this case. It will be called
113	 * exactly once with restore=true when the psock is being destroyed
114	 * and psock refcnt is zero, but before an RCU grace period.
115	 */
116	int  (*psock_update_sk_prot)(struct sock *sk, struct sk_psock *psock,
117				     bool restore);
118	struct proto			*sk_proto;
119	struct mutex			work_mutex;
120	struct sk_psock_work_state	work_state;
121	struct delayed_work		work;
122	struct sock			*sk_pair;
123	struct rcu_work			rwork;
124};
125
126int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
127		 int elem_first_coalesce);
128int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
129		 u32 off, u32 len);
130void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
131int sk_msg_free(struct sock *sk, struct sk_msg *msg);
132int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
133void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
134void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
135				  u32 bytes);
136
137void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
138void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
139
140int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
141			      struct sk_msg *msg, u32 bytes);
142int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
143			     struct sk_msg *msg, u32 bytes);
144int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
145		   int len, int flags);
146int __sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
147		     int len, int flags, int *copied_from_self);
148bool sk_msg_is_readable(struct sock *sk);
149
150static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
151{
152	WARN_ON(i == msg->sg.end && bytes);
153}
154
155static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
156{
157	if (psock->apply_bytes) {
158		if (psock->apply_bytes < bytes)
159			psock->apply_bytes = 0;
160		else
161			psock->apply_bytes -= bytes;
162	}
163}
164
165static inline u32 sk_msg_iter_dist(u32 start, u32 end)
166{
167	return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
168}
169
170#define sk_msg_iter_var_prev(var)			\
171	do {						\
172		if (var == 0)				\
173			var = NR_MSG_FRAG_IDS - 1;	\
174		else					\
175			var--;				\
176	} while (0)
177
178#define sk_msg_iter_var_next(var)			\
179	do {						\
180		var++;					\
181		if (var == NR_MSG_FRAG_IDS)		\
182			var = 0;			\
183	} while (0)
184
185#define sk_msg_iter_prev(msg, which)			\
186	sk_msg_iter_var_prev(msg->sg.which)
187
188#define sk_msg_iter_next(msg, which)			\
189	sk_msg_iter_var_next(msg->sg.which)
190
191static inline void sk_msg_init(struct sk_msg *msg)
192{
193	BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
194	memset(msg, 0, sizeof(*msg));
195	sg_init_marker(msg->sg.data, NR_MSG_FRAG_IDS);
196}
197
198static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
199			       int which, u32 size)
200{
201	dst->sg.data[which] = src->sg.data[which];
202	dst->sg.data[which].length  = size;
203	dst->sg.size		   += size;
204	src->sg.size		   -= size;
205	src->sg.data[which].length -= size;
206	src->sg.data[which].offset += size;
207}
208
209static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
210{
211	memcpy(dst, src, sizeof(*src));
212	sk_msg_init(src);
213}
214
215static inline bool sk_msg_full(const struct sk_msg *msg)
216{
217	return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
218}
219
220static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
221{
222	return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
223}
224
225static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
226{
227	return &msg->sg.data[which];
228}
229
230static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
231{
232	return msg->sg.data[which];
233}
234
235static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
236{
237	return sg_page(sk_msg_elem(msg, which));
238}
239
240static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
241{
242	return msg->flags & BPF_F_INGRESS;
243}
244
245static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
246{
247	struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
248
249	if (test_bit(msg->sg.start, msg->sg.copy)) {
250		msg->data = NULL;
251		msg->data_end = NULL;
252	} else {
253		msg->data = sg_virt(sge);
254		msg->data_end = msg->data + sge->length;
255	}
256}
257
258static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
259				   u32 len, u32 offset)
260{
261	struct scatterlist *sge;
262
263	get_page(page);
264	sge = sk_msg_elem(msg, msg->sg.end);
265	sg_set_page(sge, page, len, offset);
266	sg_unmark_end(sge);
267
268	__set_bit(msg->sg.end, msg->sg.copy);
269	msg->sg.size += len;
270	sk_msg_iter_next(msg, end);
271}
272
273static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
274{
275	do {
276		if (copy_state)
277			__set_bit(i, msg->sg.copy);
278		else
279			__clear_bit(i, msg->sg.copy);
280		sk_msg_iter_var_next(i);
281		if (i == msg->sg.end)
282			break;
283	} while (1);
284}
285
286static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
287{
288	sk_msg_sg_copy(msg, start, true);
289}
290
291static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
292{
293	sk_msg_sg_copy(msg, start, false);
294}
295
296static inline struct sk_psock *sk_psock(const struct sock *sk)
297{
298	return __rcu_dereference_sk_user_data_with_flags(sk,
299							 SK_USER_DATA_PSOCK);
300}
301
302static inline void sk_psock_set_state(struct sk_psock *psock,
303				      enum sk_psock_state_bits bit)
304{
305	set_bit(bit, &psock->state);
306}
307
308static inline void sk_psock_clear_state(struct sk_psock *psock,
309					enum sk_psock_state_bits bit)
310{
311	clear_bit(bit, &psock->state);
312}
313
314static inline bool sk_psock_test_state(const struct sk_psock *psock,
315				       enum sk_psock_state_bits bit)
316{
317	return test_bit(bit, &psock->state);
318}
319
320static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
321{
322	sk_drops_skbadd(sk, skb);
323	kfree_skb(skb);
324}
325
326static inline u32 sk_psock_get_msg_len_nolock(struct sk_psock *psock)
327{
328	/* Used by ioctl to read msg_tot_len only; lock-free for performance */
329	return READ_ONCE(psock->msg_tot_len);
330}
331
332static inline void sk_psock_msg_len_add_locked(struct sk_psock *psock, int diff)
333{
334	/* Use WRITE_ONCE to ensure correct read in sk_psock_get_msg_len_nolock().
335	 * ingress_lock should be held to prevent concurrent updates to msg_tot_len
336	 */
337	WRITE_ONCE(psock->msg_tot_len, psock->msg_tot_len + diff);
338}
339
340static inline void sk_psock_msg_len_add(struct sk_psock *psock, int diff)
341{
342	spin_lock_bh(&psock->ingress_lock);
343	sk_psock_msg_len_add_locked(psock, diff);
344	spin_unlock_bh(&psock->ingress_lock);
345}
346
347static inline bool sk_psock_queue_msg(struct sk_psock *psock,
348				      struct sk_msg *msg)
349{
350	bool ret;
351
352	spin_lock_bh(&psock->ingress_lock);
353	if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
354		list_add_tail(&msg->list, &psock->ingress_msg);
355		sk_psock_msg_len_add_locked(psock, msg->sg.size);
356		ret = true;
357	} else {
358		sk_msg_free(psock->sk, msg);
359		kfree(msg);
360		ret = false;
361	}
362	spin_unlock_bh(&psock->ingress_lock);
363	return ret;
364}
365
366static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
367{
368	struct sk_msg *msg;
369
370	spin_lock_bh(&psock->ingress_lock);
371	msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
372	if (msg) {
373		list_del(&msg->list);
374		sk_psock_msg_len_add_locked(psock, -msg->sg.size);
375	}
376	spin_unlock_bh(&psock->ingress_lock);
377	return msg;
378}
379
380static inline struct sk_msg *sk_psock_peek_msg_locked(struct sk_psock *psock)
381{
382	return list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
383}
384
385static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
386{
387	struct sk_msg *msg;
388
389	spin_lock_bh(&psock->ingress_lock);
390	msg = sk_psock_peek_msg_locked(psock);
391	spin_unlock_bh(&psock->ingress_lock);
392	return msg;
393}
394
395static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
396					       struct sk_msg *msg)
397{
398	struct sk_msg *ret;
399
400	spin_lock_bh(&psock->ingress_lock);
401	if (list_is_last(&msg->list, &psock->ingress_msg))
402		ret = NULL;
403	else
404		ret = list_next_entry(msg, list);
405	spin_unlock_bh(&psock->ingress_lock);
406	return ret;
407}
408
409static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
410{
411	return psock ? list_empty(&psock->ingress_msg) : true;
412}
413
414static inline void kfree_sk_msg(struct sk_msg *msg)
415{
416	if (msg->skb)
417		consume_skb(msg->skb);
418	kfree(msg);
419}
420
421static inline void sk_psock_report_error(struct sk_psock *psock, int err)
422{
423	struct sock *sk = psock->sk;
424
425	sk->sk_err = err;
426	sk_error_report(sk);
427}
428
429struct sk_psock *sk_psock_init(struct sock *sk, int node);
430void sk_psock_stop(struct sk_psock *psock);
431
432#if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
433int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
434void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
435void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
436#else
437static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
438{
439	return -EOPNOTSUPP;
440}
441
442static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
443{
444}
445
446static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
447{
448}
449#endif
450
451void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
452void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
453
454int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
455			 struct sk_msg *msg);
456
457/*
458 * This specialized allocator has to be a macro for its allocations to be
459 * accounted separately (to have a separate alloc_tag). The typecast is
460 * intentional to enforce typesafety.
461 */
462#define sk_psock_init_link()	\
463		((struct sk_psock_link *)kzalloc(sizeof(struct sk_psock_link),	\
464						 GFP_ATOMIC | __GFP_NOWARN))
465
466static inline void sk_psock_free_link(struct sk_psock_link *link)
467{
468	kfree(link);
469}
470
471struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
472
473static inline void sk_psock_cork_free(struct sk_psock *psock)
474{
475	if (psock->cork) {
476		sk_msg_free(psock->sk, psock->cork);
477		kfree(psock->cork);
478		psock->cork = NULL;
479	}
480}
481
482static inline void sk_psock_restore_proto(struct sock *sk,
483					  struct sk_psock *psock)
484{
485	if (psock->psock_update_sk_prot)
486		psock->psock_update_sk_prot(sk, psock, true);
487}
488
489static inline struct sk_psock *sk_psock_get(struct sock *sk)
490{
491	struct sk_psock *psock;
492
493	rcu_read_lock();
494	psock = sk_psock(sk);
495	if (psock && !refcount_inc_not_zero(&psock->refcnt))
496		psock = NULL;
497	rcu_read_unlock();
498	return psock;
499}
500
501void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
502
503static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
504{
505	if (refcount_dec_and_test(&psock->refcnt))
506		sk_psock_drop(sk, psock);
507}
508
509static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
510{
511	read_lock_bh(&sk->sk_callback_lock);
512	if (psock->saved_data_ready)
513		psock->saved_data_ready(sk);
514	else
515		sk->sk_data_ready(sk);
516	read_unlock_bh(&sk->sk_callback_lock);
517}
518
519static inline void psock_set_prog(struct bpf_prog **pprog,
520				  struct bpf_prog *prog)
521{
522	prog = xchg(pprog, prog);
523	if (prog)
524		bpf_prog_put(prog);
525}
526
527static inline int psock_replace_prog(struct bpf_prog **pprog,
528				     struct bpf_prog *prog,
529				     struct bpf_prog *old)
530{
531	if (cmpxchg(pprog, old, prog) != old)
532		return -ENOENT;
533
534	if (old)
535		bpf_prog_put(old);
536
537	return 0;
538}
539
540static inline void psock_progs_drop(struct sk_psock_progs *progs)
541{
542	psock_set_prog(&progs->msg_parser, NULL);
543	psock_set_prog(&progs->stream_parser, NULL);
544	psock_set_prog(&progs->stream_verdict, NULL);
545	psock_set_prog(&progs->skb_verdict, NULL);
546}
547
548int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
549
550static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
551{
552	if (!psock)
553		return false;
554	return !!psock->saved_data_ready;
555}
556
557/* for tcp only, sk is locked */
558static inline ssize_t sk_psock_msg_inq(struct sock *sk)
559{
560	struct sk_psock *psock;
561	ssize_t inq = 0;
562
563	psock = sk_psock_get(sk);
564	if (likely(psock)) {
565		inq = sk_psock_get_msg_len_nolock(psock);
566		sk_psock_put(sk, psock);
567	}
568	return inq;
569}
570
571/* for udp only, sk is not locked */
572static inline ssize_t sk_msg_first_len(struct sock *sk)
573{
574	struct sk_psock *psock;
575	struct sk_msg *msg;
576	ssize_t inq = 0;
577
578	psock = sk_psock_get(sk);
579	if (likely(psock)) {
580		spin_lock_bh(&psock->ingress_lock);
581		msg = sk_psock_peek_msg_locked(psock);
582		if (msg)
583			inq = msg->sg.size;
584		spin_unlock_bh(&psock->ingress_lock);
585		sk_psock_put(sk, psock);
586	}
587	return inq;
588}
589
590#if IS_ENABLED(CONFIG_NET_SOCK_MSG)
591
592#define BPF_F_STRPARSER	(1UL << 1)
593
594/* We only have two bits so far. */
595#define BPF_F_PTR_MASK ~(BPF_F_INGRESS | BPF_F_STRPARSER)
596
597static inline bool skb_bpf_strparser(const struct sk_buff *skb)
598{
599	unsigned long sk_redir = skb->_sk_redir;
600
601	return sk_redir & BPF_F_STRPARSER;
602}
603
604static inline void skb_bpf_set_strparser(struct sk_buff *skb)
605{
606	skb->_sk_redir |= BPF_F_STRPARSER;
607}
608
609static inline bool skb_bpf_ingress(const struct sk_buff *skb)
610{
611	unsigned long sk_redir = skb->_sk_redir;
612
613	return sk_redir & BPF_F_INGRESS;
614}
615
616static inline void skb_bpf_set_ingress(struct sk_buff *skb)
617{
618	skb->_sk_redir |= BPF_F_INGRESS;
619}
620
621static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
622				     bool ingress)
623{
624	skb->_sk_redir = (unsigned long)sk_redir;
625	if (ingress)
626		skb->_sk_redir |= BPF_F_INGRESS;
627}
628
629static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
630{
631	unsigned long sk_redir = skb->_sk_redir;
632
633	return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
634}
635
636static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
637{
638	skb->_sk_redir = 0;
639}
640#endif /* CONFIG_NET_SOCK_MSG */
641#endif /* _LINUX_SKMSG_H */