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 void sk_psock_queue_msg(struct sk_psock *psock,
321  				      struct sk_msg *msg)
322  {
323  	spin_lock_bh(&psock->ingress_lock);
324  	if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED))
325  		list_add_tail(&msg->list, &psock->ingress_msg);
326  	else {
327  		sk_msg_free(psock->sk, msg);
328  		kfree(msg);
329  	}
330  	spin_unlock_bh(&psock->ingress_lock);
331  }
332  
sk_psock_dequeue_msg(struct sk_psock * psock)333  static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
334  {
335  	struct sk_msg *msg;
336  
337  	spin_lock_bh(&psock->ingress_lock);
338  	msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
339  	if (msg)
340  		list_del(&msg->list);
341  	spin_unlock_bh(&psock->ingress_lock);
342  	return msg;
343  }
344  
sk_psock_peek_msg(struct sk_psock * psock)345  static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
346  {
347  	struct sk_msg *msg;
348  
349  	spin_lock_bh(&psock->ingress_lock);
350  	msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
351  	spin_unlock_bh(&psock->ingress_lock);
352  	return msg;
353  }
354  
sk_psock_next_msg(struct sk_psock * psock,struct sk_msg * msg)355  static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
356  					       struct sk_msg *msg)
357  {
358  	struct sk_msg *ret;
359  
360  	spin_lock_bh(&psock->ingress_lock);
361  	if (list_is_last(&msg->list, &psock->ingress_msg))
362  		ret = NULL;
363  	else
364  		ret = list_next_entry(msg, list);
365  	spin_unlock_bh(&psock->ingress_lock);
366  	return ret;
367  }
368  
sk_psock_queue_empty(const struct sk_psock * psock)369  static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
370  {
371  	return psock ? list_empty(&psock->ingress_msg) : true;
372  }
373  
kfree_sk_msg(struct sk_msg * msg)374  static inline void kfree_sk_msg(struct sk_msg *msg)
375  {
376  	if (msg->skb)
377  		consume_skb(msg->skb);
378  	kfree(msg);
379  }
380  
sk_psock_report_error(struct sk_psock * psock,int err)381  static inline void sk_psock_report_error(struct sk_psock *psock, int err)
382  {
383  	struct sock *sk = psock->sk;
384  
385  	sk->sk_err = err;
386  	sk_error_report(sk);
387  }
388  
389  struct sk_psock *sk_psock_init(struct sock *sk, int node);
390  void sk_psock_stop(struct sk_psock *psock);
391  
392  #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
393  int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
394  void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
395  void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
396  #else
sk_psock_init_strp(struct sock * sk,struct sk_psock * psock)397  static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
398  {
399  	return -EOPNOTSUPP;
400  }
401  
sk_psock_start_strp(struct sock * sk,struct sk_psock * psock)402  static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
403  {
404  }
405  
sk_psock_stop_strp(struct sock * sk,struct sk_psock * psock)406  static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
407  {
408  }
409  #endif
410  
411  void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
412  void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
413  
414  int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
415  			 struct sk_msg *msg);
416  
417  /*
418   * This specialized allocator has to be a macro for its allocations to be
419   * accounted separately (to have a separate alloc_tag). The typecast is
420   * intentional to enforce typesafety.
421   */
422  #define sk_psock_init_link()	\
423  		((struct sk_psock_link *)kzalloc(sizeof(struct sk_psock_link),	\
424  						 GFP_ATOMIC | __GFP_NOWARN))
425  
sk_psock_free_link(struct sk_psock_link * link)426  static inline void sk_psock_free_link(struct sk_psock_link *link)
427  {
428  	kfree(link);
429  }
430  
431  struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
432  
sk_psock_cork_free(struct sk_psock * psock)433  static inline void sk_psock_cork_free(struct sk_psock *psock)
434  {
435  	if (psock->cork) {
436  		sk_msg_free(psock->sk, psock->cork);
437  		kfree(psock->cork);
438  		psock->cork = NULL;
439  	}
440  }
441  
sk_psock_restore_proto(struct sock * sk,struct sk_psock * psock)442  static inline void sk_psock_restore_proto(struct sock *sk,
443  					  struct sk_psock *psock)
444  {
445  	if (psock->psock_update_sk_prot)
446  		psock->psock_update_sk_prot(sk, psock, true);
447  }
448  
sk_psock_get(struct sock * sk)449  static inline struct sk_psock *sk_psock_get(struct sock *sk)
450  {
451  	struct sk_psock *psock;
452  
453  	rcu_read_lock();
454  	psock = sk_psock(sk);
455  	if (psock && !refcount_inc_not_zero(&psock->refcnt))
456  		psock = NULL;
457  	rcu_read_unlock();
458  	return psock;
459  }
460  
461  void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
462  
sk_psock_put(struct sock * sk,struct sk_psock * psock)463  static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
464  {
465  	if (refcount_dec_and_test(&psock->refcnt))
466  		sk_psock_drop(sk, psock);
467  }
468  
sk_psock_data_ready(struct sock * sk,struct sk_psock * psock)469  static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
470  {
471  	read_lock_bh(&sk->sk_callback_lock);
472  	if (psock->saved_data_ready)
473  		psock->saved_data_ready(sk);
474  	else
475  		sk->sk_data_ready(sk);
476  	read_unlock_bh(&sk->sk_callback_lock);
477  }
478  
psock_set_prog(struct bpf_prog ** pprog,struct bpf_prog * prog)479  static inline void psock_set_prog(struct bpf_prog **pprog,
480  				  struct bpf_prog *prog)
481  {
482  	prog = xchg(pprog, prog);
483  	if (prog)
484  		bpf_prog_put(prog);
485  }
486  
psock_replace_prog(struct bpf_prog ** pprog,struct bpf_prog * prog,struct bpf_prog * old)487  static inline int psock_replace_prog(struct bpf_prog **pprog,
488  				     struct bpf_prog *prog,
489  				     struct bpf_prog *old)
490  {
491  	if (cmpxchg(pprog, old, prog) != old)
492  		return -ENOENT;
493  
494  	if (old)
495  		bpf_prog_put(old);
496  
497  	return 0;
498  }
499  
psock_progs_drop(struct sk_psock_progs * progs)500  static inline void psock_progs_drop(struct sk_psock_progs *progs)
501  {
502  	psock_set_prog(&progs->msg_parser, NULL);
503  	psock_set_prog(&progs->stream_parser, NULL);
504  	psock_set_prog(&progs->stream_verdict, NULL);
505  	psock_set_prog(&progs->skb_verdict, NULL);
506  }
507  
508  int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
509  
sk_psock_strp_enabled(struct sk_psock * psock)510  static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
511  {
512  	if (!psock)
513  		return false;
514  	return !!psock->saved_data_ready;
515  }
516  
517  #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
518  
519  #define BPF_F_STRPARSER	(1UL << 1)
520  
521  /* We only have two bits so far. */
522  #define BPF_F_PTR_MASK ~(BPF_F_INGRESS | BPF_F_STRPARSER)
523  
skb_bpf_strparser(const struct sk_buff * skb)524  static inline bool skb_bpf_strparser(const struct sk_buff *skb)
525  {
526  	unsigned long sk_redir = skb->_sk_redir;
527  
528  	return sk_redir & BPF_F_STRPARSER;
529  }
530  
skb_bpf_set_strparser(struct sk_buff * skb)531  static inline void skb_bpf_set_strparser(struct sk_buff *skb)
532  {
533  	skb->_sk_redir |= BPF_F_STRPARSER;
534  }
535  
skb_bpf_ingress(const struct sk_buff * skb)536  static inline bool skb_bpf_ingress(const struct sk_buff *skb)
537  {
538  	unsigned long sk_redir = skb->_sk_redir;
539  
540  	return sk_redir & BPF_F_INGRESS;
541  }
542  
skb_bpf_set_ingress(struct sk_buff * skb)543  static inline void skb_bpf_set_ingress(struct sk_buff *skb)
544  {
545  	skb->_sk_redir |= BPF_F_INGRESS;
546  }
547  
skb_bpf_set_redir(struct sk_buff * skb,struct sock * sk_redir,bool ingress)548  static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
549  				     bool ingress)
550  {
551  	skb->_sk_redir = (unsigned long)sk_redir;
552  	if (ingress)
553  		skb->_sk_redir |= BPF_F_INGRESS;
554  }
555  
skb_bpf_redirect_fetch(const struct sk_buff * skb)556  static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
557  {
558  	unsigned long sk_redir = skb->_sk_redir;
559  
560  	return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
561  }
562  
skb_bpf_redirect_clear(struct sk_buff * skb)563  static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
564  {
565  	skb->_sk_redir = 0;
566  }
567  #endif /* CONFIG_NET_SOCK_MSG */
568  #endif /* _LINUX_SKMSG_H */
569