1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
8 #include <linux/filter.h>
9 #include <linux/scatterlist.h>
10 #include <linux/skbuff.h>
14 #include <net/strparser.h>
16 #define MAX_MSG_FRAGS MAX_SKB_FRAGS
17 #define NR_MSG_FRAG_IDS (MAX_MSG_FRAGS + 1)
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
37 * 2) to chain tailer SG entries after the message.
39 struct scatterlist data[MAX_MSG_FRAGS + 2];
41 static_assert(BITS_PER_LONG >= NR_MSG_FRAG_IDS);
43 /* UAPI in filter.c depends on struct sk_msg_sg being first element. */
52 struct sock *sk_redir;
54 struct list_head list;
57 struct sk_psock_progs {
58 struct bpf_prog *msg_parser;
59 struct bpf_prog *stream_parser;
60 struct bpf_prog *stream_verdict;
61 struct bpf_prog *skb_verdict;
64 enum sk_psock_state_bits {
68 struct sk_psock_link {
69 struct list_head list;
74 struct sk_psock_work_state {
82 struct sock *sk_redir;
87 struct sk_psock_progs progs;
88 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
89 struct strparser strp;
91 struct sk_buff_head ingress_skb;
92 struct list_head ingress_msg;
93 spinlock_t ingress_lock;
95 struct list_head link;
98 void (*saved_unhash)(struct sock *sk);
99 void (*saved_close)(struct sock *sk, long timeout);
100 void (*saved_write_space)(struct sock *sk);
101 void (*saved_data_ready)(struct sock *sk);
102 struct proto *sk_proto;
103 struct mutex work_mutex;
104 struct sk_psock_work_state work_state;
105 struct work_struct work;
106 struct rcu_work rwork;
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,
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,
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);
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);
128 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
130 WARN_ON(i == msg->sg.end && bytes);
133 static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
135 if (psock->apply_bytes) {
136 if (psock->apply_bytes < bytes)
137 psock->apply_bytes = 0;
139 psock->apply_bytes -= bytes;
143 static inline u32 sk_msg_iter_dist(u32 start, u32 end)
145 return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
148 #define sk_msg_iter_var_prev(var) \
151 var = NR_MSG_FRAG_IDS - 1; \
156 #define sk_msg_iter_var_next(var) \
159 if (var == NR_MSG_FRAG_IDS) \
163 #define sk_msg_iter_prev(msg, which) \
164 sk_msg_iter_var_prev(msg->sg.which)
166 #define sk_msg_iter_next(msg, which) \
167 sk_msg_iter_var_next(msg->sg.which)
169 static inline void sk_msg_clear_meta(struct sk_msg *msg)
171 memset(&msg->sg, 0, offsetofend(struct sk_msg_sg, copy));
174 static inline void sk_msg_init(struct sk_msg *msg)
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);
181 static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
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;
192 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
194 memcpy(dst, src, sizeof(*src));
198 static inline bool sk_msg_full(const struct sk_msg *msg)
200 return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
203 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
205 return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
208 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
210 return &msg->sg.data[which];
213 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
215 return msg->sg.data[which];
218 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
220 return sg_page(sk_msg_elem(msg, which));
223 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
225 return msg->flags & BPF_F_INGRESS;
228 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
230 struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
232 if (test_bit(msg->sg.start, &msg->sg.copy)) {
234 msg->data_end = NULL;
236 msg->data = sg_virt(sge);
237 msg->data_end = msg->data + sge->length;
241 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
244 struct scatterlist *sge;
247 sge = sk_msg_elem(msg, msg->sg.end);
248 sg_set_page(sge, page, len, offset);
251 __set_bit(msg->sg.end, &msg->sg.copy);
253 sk_msg_iter_next(msg, end);
256 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
260 __set_bit(i, &msg->sg.copy);
262 __clear_bit(i, &msg->sg.copy);
263 sk_msg_iter_var_next(i);
264 if (i == msg->sg.end)
269 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
271 sk_msg_sg_copy(msg, start, true);
274 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
276 sk_msg_sg_copy(msg, start, false);
279 static inline struct sk_psock *sk_psock(const struct sock *sk)
281 return rcu_dereference_sk_user_data(sk);
284 static inline void sk_psock_queue_msg(struct sk_psock *psock,
287 spin_lock_bh(&psock->ingress_lock);
288 list_add_tail(&msg->list, &psock->ingress_msg);
289 spin_unlock_bh(&psock->ingress_lock);
292 static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
296 spin_lock_bh(&psock->ingress_lock);
297 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
299 list_del(&msg->list);
300 spin_unlock_bh(&psock->ingress_lock);
304 static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
308 spin_lock_bh(&psock->ingress_lock);
309 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
310 spin_unlock_bh(&psock->ingress_lock);
314 static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
319 spin_lock_bh(&psock->ingress_lock);
320 if (list_is_last(&msg->list, &psock->ingress_msg))
323 ret = list_next_entry(msg, list);
324 spin_unlock_bh(&psock->ingress_lock);
328 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
330 return psock ? list_empty(&psock->ingress_msg) : true;
333 static inline void kfree_sk_msg(struct sk_msg *msg)
336 consume_skb(msg->skb);
340 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
342 struct sock *sk = psock->sk;
345 sk->sk_error_report(sk);
348 struct sk_psock *sk_psock_init(struct sock *sk, int node);
349 void sk_psock_stop(struct sk_psock *psock, bool wait);
351 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
352 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
353 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
354 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
356 static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
361 static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
365 static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
370 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
371 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
373 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
376 static inline struct sk_psock_link *sk_psock_init_link(void)
378 return kzalloc(sizeof(struct sk_psock_link),
379 GFP_ATOMIC | __GFP_NOWARN);
382 static inline void sk_psock_free_link(struct sk_psock_link *link)
387 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
389 static inline void sk_psock_cork_free(struct sk_psock *psock)
392 sk_msg_free(psock->sk, psock->cork);
398 static inline void sk_psock_update_proto(struct sock *sk,
399 struct sk_psock *psock,
402 /* Pairs with lockless read in sk_clone_lock() */
403 WRITE_ONCE(sk->sk_prot, ops);
406 static inline void sk_psock_restore_proto(struct sock *sk,
407 struct sk_psock *psock)
409 sk->sk_prot->unhash = psock->saved_unhash;
410 if (inet_csk_has_ulp(sk)) {
411 tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
413 sk->sk_write_space = psock->saved_write_space;
414 /* Pairs with lockless read in sk_clone_lock() */
415 WRITE_ONCE(sk->sk_prot, psock->sk_proto);
419 static inline void sk_psock_set_state(struct sk_psock *psock,
420 enum sk_psock_state_bits bit)
422 set_bit(bit, &psock->state);
425 static inline void sk_psock_clear_state(struct sk_psock *psock,
426 enum sk_psock_state_bits bit)
428 clear_bit(bit, &psock->state);
431 static inline bool sk_psock_test_state(const struct sk_psock *psock,
432 enum sk_psock_state_bits bit)
434 return test_bit(bit, &psock->state);
437 static inline struct sk_psock *sk_psock_get(struct sock *sk)
439 struct sk_psock *psock;
442 psock = sk_psock(sk);
443 if (psock && !refcount_inc_not_zero(&psock->refcnt))
449 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
451 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
453 if (refcount_dec_and_test(&psock->refcnt))
454 sk_psock_drop(sk, psock);
457 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
459 if (psock->saved_data_ready)
460 psock->saved_data_ready(sk);
462 sk->sk_data_ready(sk);
465 static inline void psock_set_prog(struct bpf_prog **pprog,
466 struct bpf_prog *prog)
468 prog = xchg(pprog, prog);
473 static inline int psock_replace_prog(struct bpf_prog **pprog,
474 struct bpf_prog *prog,
475 struct bpf_prog *old)
477 if (cmpxchg(pprog, old, prog) != old)
486 static inline void psock_progs_drop(struct sk_psock_progs *progs)
488 psock_set_prog(&progs->msg_parser, NULL);
489 psock_set_prog(&progs->stream_parser, NULL);
490 psock_set_prog(&progs->stream_verdict, NULL);
491 psock_set_prog(&progs->skb_verdict, NULL);
494 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
496 static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
500 return !!psock->saved_data_ready;
503 #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
505 /* We only have one bit so far. */
506 #define BPF_F_PTR_MASK ~(BPF_F_INGRESS)
508 static inline bool skb_bpf_ingress(const struct sk_buff *skb)
510 unsigned long sk_redir = skb->_sk_redir;
512 return sk_redir & BPF_F_INGRESS;
515 static inline void skb_bpf_set_ingress(struct sk_buff *skb)
517 skb->_sk_redir |= BPF_F_INGRESS;
520 static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
523 skb->_sk_redir = (unsigned long)sk_redir;
525 skb->_sk_redir |= BPF_F_INGRESS;
528 static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
530 unsigned long sk_redir = skb->_sk_redir;
532 return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
535 static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
539 #endif /* CONFIG_NET_SOCK_MSG */
540 #endif /* _LINUX_SKMSG_H */