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
32 /* The extra element is used for chaining the front and sections when
33 * the list becomes partitioned (e.g. end < start). The crypto APIs
34 * require the chaining.
36 struct scatterlist data[MAX_MSG_FRAGS + 1];
38 static_assert(BITS_PER_LONG >= MAX_MSG_FRAGS);
40 /* UAPI in filter.c depends on struct sk_msg_sg being first element. */
49 struct sock *sk_redir;
51 struct list_head list;
54 struct sk_psock_progs {
55 struct bpf_prog *msg_parser;
56 struct bpf_prog *skb_parser;
57 struct bpf_prog *skb_verdict;
60 enum sk_psock_state_bits {
64 struct sk_psock_link {
65 struct list_head list;
70 struct sk_psock_parser {
71 struct strparser strp;
73 void (*saved_data_ready)(struct sock *sk);
76 struct sk_psock_work_state {
84 struct sock *sk_redir;
89 struct sk_psock_progs progs;
90 struct sk_psock_parser parser;
91 struct sk_buff_head ingress_skb;
92 struct list_head ingress_msg;
94 struct list_head link;
97 void (*saved_unhash)(struct sock *sk);
98 void (*saved_close)(struct sock *sk, long timeout);
99 void (*saved_write_space)(struct sock *sk);
100 struct proto *sk_proto;
101 struct sk_psock_work_state work_state;
102 struct work_struct work;
105 struct work_struct gc;
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 + (MAX_MSG_FRAGS - start);
148 #define sk_msg_iter_var_prev(var) \
151 var = MAX_MSG_FRAGS - 1; \
156 #define sk_msg_iter_var_next(var) \
159 if (var == MAX_MSG_FRAGS) \
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 != MAX_MSG_FRAGS);
177 memset(msg, 0, sizeof(*msg));
178 sg_init_marker(msg->sg.data, MAX_MSG_FRAGS);
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.data[which].length -= size;
188 src->sg.data[which].offset += size;
191 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
193 memcpy(dst, src, sizeof(*src));
197 static inline bool sk_msg_full(const struct sk_msg *msg)
199 return (msg->sg.end == msg->sg.start) && msg->sg.size;
202 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
204 if (sk_msg_full(msg))
205 return MAX_MSG_FRAGS;
207 return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
210 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
212 return &msg->sg.data[which];
215 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
217 return msg->sg.data[which];
220 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
222 return sg_page(sk_msg_elem(msg, which));
225 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
227 return msg->flags & BPF_F_INGRESS;
230 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
232 struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
234 if (test_bit(msg->sg.start, &msg->sg.copy)) {
236 msg->data_end = NULL;
238 msg->data = sg_virt(sge);
239 msg->data_end = msg->data + sge->length;
243 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
246 struct scatterlist *sge;
249 sge = sk_msg_elem(msg, msg->sg.end);
250 sg_set_page(sge, page, len, offset);
253 __set_bit(msg->sg.end, &msg->sg.copy);
255 sk_msg_iter_next(msg, end);
258 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
262 __set_bit(i, &msg->sg.copy);
264 __clear_bit(i, &msg->sg.copy);
265 sk_msg_iter_var_next(i);
266 if (i == msg->sg.end)
271 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
273 sk_msg_sg_copy(msg, start, true);
276 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
278 sk_msg_sg_copy(msg, start, false);
281 static inline struct sk_psock *sk_psock(const struct sock *sk)
283 return rcu_dereference_sk_user_data(sk);
286 static inline void sk_psock_queue_msg(struct sk_psock *psock,
289 list_add_tail(&msg->list, &psock->ingress_msg);
292 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
294 return psock ? list_empty(&psock->ingress_msg) : true;
297 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
299 struct sock *sk = psock->sk;
302 sk->sk_error_report(sk);
305 struct sk_psock *sk_psock_init(struct sock *sk, int node);
307 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
308 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
309 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
311 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
314 static inline struct sk_psock_link *sk_psock_init_link(void)
316 return kzalloc(sizeof(struct sk_psock_link),
317 GFP_ATOMIC | __GFP_NOWARN);
320 static inline void sk_psock_free_link(struct sk_psock_link *link)
325 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
326 #if defined(CONFIG_BPF_STREAM_PARSER)
327 void sk_psock_unlink(struct sock *sk, struct sk_psock_link *link);
329 static inline void sk_psock_unlink(struct sock *sk,
330 struct sk_psock_link *link)
335 void __sk_psock_purge_ingress_msg(struct sk_psock *psock);
337 static inline void sk_psock_cork_free(struct sk_psock *psock)
340 sk_msg_free(psock->sk, psock->cork);
346 static inline void sk_psock_update_proto(struct sock *sk,
347 struct sk_psock *psock,
350 psock->saved_unhash = sk->sk_prot->unhash;
351 psock->saved_close = sk->sk_prot->close;
352 psock->saved_write_space = sk->sk_write_space;
354 psock->sk_proto = sk->sk_prot;
358 static inline void sk_psock_restore_proto(struct sock *sk,
359 struct sk_psock *psock)
361 sk->sk_write_space = psock->saved_write_space;
363 if (psock->sk_proto) {
364 struct inet_connection_sock *icsk = inet_csk(sk);
365 bool has_ulp = !!icsk->icsk_ulp_data;
368 tcp_update_ulp(sk, psock->sk_proto);
370 sk->sk_prot = psock->sk_proto;
371 psock->sk_proto = NULL;
375 static inline void sk_psock_set_state(struct sk_psock *psock,
376 enum sk_psock_state_bits bit)
378 set_bit(bit, &psock->state);
381 static inline void sk_psock_clear_state(struct sk_psock *psock,
382 enum sk_psock_state_bits bit)
384 clear_bit(bit, &psock->state);
387 static inline bool sk_psock_test_state(const struct sk_psock *psock,
388 enum sk_psock_state_bits bit)
390 return test_bit(bit, &psock->state);
393 static inline struct sk_psock *sk_psock_get_checked(struct sock *sk)
395 struct sk_psock *psock;
398 psock = sk_psock(sk);
400 if (sk->sk_prot->recvmsg != tcp_bpf_recvmsg) {
401 psock = ERR_PTR(-EBUSY);
405 if (!refcount_inc_not_zero(&psock->refcnt))
406 psock = ERR_PTR(-EBUSY);
413 static inline struct sk_psock *sk_psock_get(struct sock *sk)
415 struct sk_psock *psock;
418 psock = sk_psock(sk);
419 if (psock && !refcount_inc_not_zero(&psock->refcnt))
425 void sk_psock_stop(struct sock *sk, struct sk_psock *psock);
426 void sk_psock_destroy(struct rcu_head *rcu);
427 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
429 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
431 if (refcount_dec_and_test(&psock->refcnt))
432 sk_psock_drop(sk, psock);
435 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
437 if (psock->parser.enabled)
438 psock->parser.saved_data_ready(sk);
440 sk->sk_data_ready(sk);
443 static inline void psock_set_prog(struct bpf_prog **pprog,
444 struct bpf_prog *prog)
446 prog = xchg(pprog, prog);
451 static inline void psock_progs_drop(struct sk_psock_progs *progs)
453 psock_set_prog(&progs->msg_parser, NULL);
454 psock_set_prog(&progs->skb_parser, NULL);
455 psock_set_prog(&progs->skb_verdict, NULL);
458 #endif /* _LINUX_SKMSG_H */