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)
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
37 * 2) to chain tailer SG entries after the message.
39 struct scatterlist data[MAX_MSG_FRAGS + 2];
42 /* UAPI in filter.c depends on struct sk_msg_sg being first element. */
51 struct sock *sk_redir;
53 struct list_head list;
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;
63 enum sk_psock_state_bits {
65 SK_PSOCK_RX_STRP_ENABLED,
68 struct sk_psock_link {
69 struct list_head list;
74 struct sk_psock_work_state {
81 struct sock *sk_redir;
85 bool redir_ingress; /* undefined if sk_redir is null */
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_destroy)(struct sock *sk);
100 void (*saved_close)(struct sock *sk, long timeout);
101 void (*saved_write_space)(struct sock *sk);
102 void (*saved_data_ready)(struct sock *sk);
103 /* psock_update_sk_prot may be called with restore=false many times
104 * so the handler must be safe for this case. It will be called
105 * exactly once with restore=true when the psock is being destroyed
106 * and psock refcnt is zero, but before an RCU grace period.
108 int (*psock_update_sk_prot)(struct sock *sk, struct sk_psock *psock,
110 struct proto *sk_proto;
111 struct mutex work_mutex;
112 struct sk_psock_work_state work_state;
113 struct delayed_work work;
114 struct sock *sk_pair;
115 struct rcu_work rwork;
118 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
119 int elem_first_coalesce);
120 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
122 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len);
123 int sk_msg_free(struct sock *sk, struct sk_msg *msg);
124 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg);
125 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes);
126 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
129 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes);
130 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes);
132 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
133 struct sk_msg *msg, u32 bytes);
134 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
135 struct sk_msg *msg, u32 bytes);
136 int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
138 bool sk_msg_is_readable(struct sock *sk);
140 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
142 WARN_ON(i == msg->sg.end && bytes);
145 static inline void sk_msg_apply_bytes(struct sk_psock *psock, u32 bytes)
147 if (psock->apply_bytes) {
148 if (psock->apply_bytes < bytes)
149 psock->apply_bytes = 0;
151 psock->apply_bytes -= bytes;
155 static inline u32 sk_msg_iter_dist(u32 start, u32 end)
157 return end >= start ? end - start : end + (NR_MSG_FRAG_IDS - start);
160 #define sk_msg_iter_var_prev(var) \
163 var = NR_MSG_FRAG_IDS - 1; \
168 #define sk_msg_iter_var_next(var) \
171 if (var == NR_MSG_FRAG_IDS) \
175 #define sk_msg_iter_prev(msg, which) \
176 sk_msg_iter_var_prev(msg->sg.which)
178 #define sk_msg_iter_next(msg, which) \
179 sk_msg_iter_var_next(msg->sg.which)
181 static inline void sk_msg_init(struct sk_msg *msg)
183 BUILD_BUG_ON(ARRAY_SIZE(msg->sg.data) - 1 != NR_MSG_FRAG_IDS);
184 memset(msg, 0, sizeof(*msg));
185 sg_init_marker(msg->sg.data, NR_MSG_FRAG_IDS);
188 static inline void sk_msg_xfer(struct sk_msg *dst, struct sk_msg *src,
191 dst->sg.data[which] = src->sg.data[which];
192 dst->sg.data[which].length = size;
193 dst->sg.size += size;
194 src->sg.size -= size;
195 src->sg.data[which].length -= size;
196 src->sg.data[which].offset += size;
199 static inline void sk_msg_xfer_full(struct sk_msg *dst, struct sk_msg *src)
201 memcpy(dst, src, sizeof(*src));
205 static inline bool sk_msg_full(const struct sk_msg *msg)
207 return sk_msg_iter_dist(msg->sg.start, msg->sg.end) == MAX_MSG_FRAGS;
210 static inline u32 sk_msg_elem_used(const struct sk_msg *msg)
212 return sk_msg_iter_dist(msg->sg.start, msg->sg.end);
215 static inline struct scatterlist *sk_msg_elem(struct sk_msg *msg, int which)
217 return &msg->sg.data[which];
220 static inline struct scatterlist sk_msg_elem_cpy(struct sk_msg *msg, int which)
222 return msg->sg.data[which];
225 static inline struct page *sk_msg_page(struct sk_msg *msg, int which)
227 return sg_page(sk_msg_elem(msg, which));
230 static inline bool sk_msg_to_ingress(const struct sk_msg *msg)
232 return msg->flags & BPF_F_INGRESS;
235 static inline void sk_msg_compute_data_pointers(struct sk_msg *msg)
237 struct scatterlist *sge = sk_msg_elem(msg, msg->sg.start);
239 if (test_bit(msg->sg.start, msg->sg.copy)) {
241 msg->data_end = NULL;
243 msg->data = sg_virt(sge);
244 msg->data_end = msg->data + sge->length;
248 static inline void sk_msg_page_add(struct sk_msg *msg, struct page *page,
251 struct scatterlist *sge;
254 sge = sk_msg_elem(msg, msg->sg.end);
255 sg_set_page(sge, page, len, offset);
258 __set_bit(msg->sg.end, msg->sg.copy);
260 sk_msg_iter_next(msg, end);
263 static inline void sk_msg_sg_copy(struct sk_msg *msg, u32 i, bool copy_state)
267 __set_bit(i, msg->sg.copy);
269 __clear_bit(i, msg->sg.copy);
270 sk_msg_iter_var_next(i);
271 if (i == msg->sg.end)
276 static inline void sk_msg_sg_copy_set(struct sk_msg *msg, u32 start)
278 sk_msg_sg_copy(msg, start, true);
281 static inline void sk_msg_sg_copy_clear(struct sk_msg *msg, u32 start)
283 sk_msg_sg_copy(msg, start, false);
286 static inline struct sk_psock *sk_psock(const struct sock *sk)
288 return __rcu_dereference_sk_user_data_with_flags(sk,
292 static inline void sk_psock_set_state(struct sk_psock *psock,
293 enum sk_psock_state_bits bit)
295 set_bit(bit, &psock->state);
298 static inline void sk_psock_clear_state(struct sk_psock *psock,
299 enum sk_psock_state_bits bit)
301 clear_bit(bit, &psock->state);
304 static inline bool sk_psock_test_state(const struct sk_psock *psock,
305 enum sk_psock_state_bits bit)
307 return test_bit(bit, &psock->state);
310 static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
312 sk_drops_add(sk, skb);
316 static inline void sk_psock_queue_msg(struct sk_psock *psock,
319 spin_lock_bh(&psock->ingress_lock);
320 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED))
321 list_add_tail(&msg->list, &psock->ingress_msg);
323 sk_msg_free(psock->sk, msg);
326 spin_unlock_bh(&psock->ingress_lock);
329 static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
333 spin_lock_bh(&psock->ingress_lock);
334 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
336 list_del(&msg->list);
337 spin_unlock_bh(&psock->ingress_lock);
341 static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
345 spin_lock_bh(&psock->ingress_lock);
346 msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
347 spin_unlock_bh(&psock->ingress_lock);
351 static inline struct sk_msg *sk_psock_next_msg(struct sk_psock *psock,
356 spin_lock_bh(&psock->ingress_lock);
357 if (list_is_last(&msg->list, &psock->ingress_msg))
360 ret = list_next_entry(msg, list);
361 spin_unlock_bh(&psock->ingress_lock);
365 static inline bool sk_psock_queue_empty(const struct sk_psock *psock)
367 return psock ? list_empty(&psock->ingress_msg) : true;
370 static inline void kfree_sk_msg(struct sk_msg *msg)
373 consume_skb(msg->skb);
377 static inline void sk_psock_report_error(struct sk_psock *psock, int err)
379 struct sock *sk = psock->sk;
385 struct sk_psock *sk_psock_init(struct sock *sk, int node);
386 void sk_psock_stop(struct sk_psock *psock);
388 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
389 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock);
390 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock);
391 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock);
393 static inline int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
398 static inline void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
402 static inline void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
407 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock);
408 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock);
410 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
413 static inline struct sk_psock_link *sk_psock_init_link(void)
415 return kzalloc(sizeof(struct sk_psock_link),
416 GFP_ATOMIC | __GFP_NOWARN);
419 static inline void sk_psock_free_link(struct sk_psock_link *link)
424 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock);
426 static inline void sk_psock_cork_free(struct sk_psock *psock)
429 sk_msg_free(psock->sk, psock->cork);
435 static inline void sk_psock_restore_proto(struct sock *sk,
436 struct sk_psock *psock)
438 if (psock->psock_update_sk_prot)
439 psock->psock_update_sk_prot(sk, psock, true);
442 static inline struct sk_psock *sk_psock_get(struct sock *sk)
444 struct sk_psock *psock;
447 psock = sk_psock(sk);
448 if (psock && !refcount_inc_not_zero(&psock->refcnt))
454 void sk_psock_drop(struct sock *sk, struct sk_psock *psock);
456 static inline void sk_psock_put(struct sock *sk, struct sk_psock *psock)
458 if (refcount_dec_and_test(&psock->refcnt))
459 sk_psock_drop(sk, psock);
462 static inline void sk_psock_data_ready(struct sock *sk, struct sk_psock *psock)
464 if (psock->saved_data_ready)
465 psock->saved_data_ready(sk);
467 sk->sk_data_ready(sk);
470 static inline void psock_set_prog(struct bpf_prog **pprog,
471 struct bpf_prog *prog)
473 prog = xchg(pprog, prog);
478 static inline int psock_replace_prog(struct bpf_prog **pprog,
479 struct bpf_prog *prog,
480 struct bpf_prog *old)
482 if (cmpxchg(pprog, old, prog) != old)
491 static inline void psock_progs_drop(struct sk_psock_progs *progs)
493 psock_set_prog(&progs->msg_parser, NULL);
494 psock_set_prog(&progs->stream_parser, NULL);
495 psock_set_prog(&progs->stream_verdict, NULL);
496 psock_set_prog(&progs->skb_verdict, NULL);
499 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb);
501 static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
505 return !!psock->saved_data_ready;
508 #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
510 #define BPF_F_STRPARSER (1UL << 1)
512 /* We only have two bits so far. */
513 #define BPF_F_PTR_MASK ~(BPF_F_INGRESS | BPF_F_STRPARSER)
515 static inline bool skb_bpf_strparser(const struct sk_buff *skb)
517 unsigned long sk_redir = skb->_sk_redir;
519 return sk_redir & BPF_F_STRPARSER;
522 static inline void skb_bpf_set_strparser(struct sk_buff *skb)
524 skb->_sk_redir |= BPF_F_STRPARSER;
527 static inline bool skb_bpf_ingress(const struct sk_buff *skb)
529 unsigned long sk_redir = skb->_sk_redir;
531 return sk_redir & BPF_F_INGRESS;
534 static inline void skb_bpf_set_ingress(struct sk_buff *skb)
536 skb->_sk_redir |= BPF_F_INGRESS;
539 static inline void skb_bpf_set_redir(struct sk_buff *skb, struct sock *sk_redir,
542 skb->_sk_redir = (unsigned long)sk_redir;
544 skb->_sk_redir |= BPF_F_INGRESS;
547 static inline struct sock *skb_bpf_redirect_fetch(const struct sk_buff *skb)
549 unsigned long sk_redir = skb->_sk_redir;
551 return (struct sock *)(sk_redir & BPF_F_PTR_MASK);
554 static inline void skb_bpf_redirect_clear(struct sk_buff *skb)
558 #endif /* CONFIG_NET_SOCK_MSG */
559 #endif /* _LINUX_SKMSG_H */