xref: /openbmc/linux/net/strparser/strparser.c (revision e5c86679) 
1 /*
2  * Stream Parser
3  *
4  * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation.
9  */
10 
11 #include <linux/bpf.h>
12 #include <linux/errno.h>
13 #include <linux/errqueue.h>
14 #include <linux/file.h>
15 #include <linux/in.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/net.h>
19 #include <linux/netdevice.h>
20 #include <linux/poll.h>
21 #include <linux/rculist.h>
22 #include <linux/skbuff.h>
23 #include <linux/socket.h>
24 #include <linux/uaccess.h>
25 #include <linux/workqueue.h>
26 #include <net/strparser.h>
27 #include <net/netns/generic.h>
28 #include <net/sock.h>
29 
30 static struct workqueue_struct *strp_wq;
31 
32 struct _strp_rx_msg {
33 	/* Internal cb structure. struct strp_rx_msg must be first for passing
34 	 * to upper layer.
35 	 */
36 	struct strp_rx_msg strp;
37 	int accum_len;
38 	int early_eaten;
39 };
40 
41 static inline struct _strp_rx_msg *_strp_rx_msg(struct sk_buff *skb)
42 {
43 	return (struct _strp_rx_msg *)((void *)skb->cb +
44 		offsetof(struct qdisc_skb_cb, data));
45 }
46 
47 /* Lower lock held */
48 static void strp_abort_rx_strp(struct strparser *strp, int err)
49 {
50 	struct sock *csk = strp->sk;
51 
52 	/* Unrecoverable error in receive */
53 
54 	del_timer(&strp->rx_msg_timer);
55 
56 	if (strp->rx_stopped)
57 		return;
58 
59 	strp->rx_stopped = 1;
60 
61 	/* Report an error on the lower socket */
62 	csk->sk_err = err;
63 	csk->sk_error_report(csk);
64 }
65 
66 static void strp_start_rx_timer(struct strparser *strp)
67 {
68 	if (strp->sk->sk_rcvtimeo)
69 		mod_timer(&strp->rx_msg_timer, strp->sk->sk_rcvtimeo);
70 }
71 
72 /* Lower lock held */
73 static void strp_parser_err(struct strparser *strp, int err,
74 			    read_descriptor_t *desc)
75 {
76 	desc->error = err;
77 	kfree_skb(strp->rx_skb_head);
78 	strp->rx_skb_head = NULL;
79 	strp->cb.abort_parser(strp, err);
80 }
81 
82 static inline int strp_peek_len(struct strparser *strp)
83 {
84 	struct socket *sock = strp->sk->sk_socket;
85 
86 	return sock->ops->peek_len(sock);
87 }
88 
89 /* Lower socket lock held */
90 static int strp_recv(read_descriptor_t *desc, struct sk_buff *orig_skb,
91 		     unsigned int orig_offset, size_t orig_len)
92 {
93 	struct strparser *strp = (struct strparser *)desc->arg.data;
94 	struct _strp_rx_msg *rxm;
95 	struct sk_buff *head, *skb;
96 	size_t eaten = 0, cand_len;
97 	ssize_t extra;
98 	int err;
99 	bool cloned_orig = false;
100 
101 	if (strp->rx_paused)
102 		return 0;
103 
104 	head = strp->rx_skb_head;
105 	if (head) {
106 		/* Message already in progress */
107 
108 		rxm = _strp_rx_msg(head);
109 		if (unlikely(rxm->early_eaten)) {
110 			/* Already some number of bytes on the receive sock
111 			 * data saved in rx_skb_head, just indicate they
112 			 * are consumed.
113 			 */
114 			eaten = orig_len <= rxm->early_eaten ?
115 				orig_len : rxm->early_eaten;
116 			rxm->early_eaten -= eaten;
117 
118 			return eaten;
119 		}
120 
121 		if (unlikely(orig_offset)) {
122 			/* Getting data with a non-zero offset when a message is
123 			 * in progress is not expected. If it does happen, we
124 			 * need to clone and pull since we can't deal with
125 			 * offsets in the skbs for a message expect in the head.
126 			 */
127 			orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
128 			if (!orig_skb) {
129 				STRP_STATS_INCR(strp->stats.rx_mem_fail);
130 				desc->error = -ENOMEM;
131 				return 0;
132 			}
133 			if (!pskb_pull(orig_skb, orig_offset)) {
134 				STRP_STATS_INCR(strp->stats.rx_mem_fail);
135 				kfree_skb(orig_skb);
136 				desc->error = -ENOMEM;
137 				return 0;
138 			}
139 			cloned_orig = true;
140 			orig_offset = 0;
141 		}
142 
143 		if (!strp->rx_skb_nextp) {
144 			/* We are going to append to the frags_list of head.
145 			 * Need to unshare the frag_list.
146 			 */
147 			err = skb_unclone(head, GFP_ATOMIC);
148 			if (err) {
149 				STRP_STATS_INCR(strp->stats.rx_mem_fail);
150 				desc->error = err;
151 				return 0;
152 			}
153 
154 			if (unlikely(skb_shinfo(head)->frag_list)) {
155 				/* We can't append to an sk_buff that already
156 				 * has a frag_list. We create a new head, point
157 				 * the frag_list of that to the old head, and
158 				 * then are able to use the old head->next for
159 				 * appending to the message.
160 				 */
161 				if (WARN_ON(head->next)) {
162 					desc->error = -EINVAL;
163 					return 0;
164 				}
165 
166 				skb = alloc_skb(0, GFP_ATOMIC);
167 				if (!skb) {
168 					STRP_STATS_INCR(strp->stats.rx_mem_fail);
169 					desc->error = -ENOMEM;
170 					return 0;
171 				}
172 				skb->len = head->len;
173 				skb->data_len = head->len;
174 				skb->truesize = head->truesize;
175 				*_strp_rx_msg(skb) = *_strp_rx_msg(head);
176 				strp->rx_skb_nextp = &head->next;
177 				skb_shinfo(skb)->frag_list = head;
178 				strp->rx_skb_head = skb;
179 				head = skb;
180 			} else {
181 				strp->rx_skb_nextp =
182 				    &skb_shinfo(head)->frag_list;
183 			}
184 		}
185 	}
186 
187 	while (eaten < orig_len) {
188 		/* Always clone since we will consume something */
189 		skb = skb_clone(orig_skb, GFP_ATOMIC);
190 		if (!skb) {
191 			STRP_STATS_INCR(strp->stats.rx_mem_fail);
192 			desc->error = -ENOMEM;
193 			break;
194 		}
195 
196 		cand_len = orig_len - eaten;
197 
198 		head = strp->rx_skb_head;
199 		if (!head) {
200 			head = skb;
201 			strp->rx_skb_head = head;
202 			/* Will set rx_skb_nextp on next packet if needed */
203 			strp->rx_skb_nextp = NULL;
204 			rxm = _strp_rx_msg(head);
205 			memset(rxm, 0, sizeof(*rxm));
206 			rxm->strp.offset = orig_offset + eaten;
207 		} else {
208 			/* Unclone since we may be appending to an skb that we
209 			 * already share a frag_list with.
210 			 */
211 			err = skb_unclone(skb, GFP_ATOMIC);
212 			if (err) {
213 				STRP_STATS_INCR(strp->stats.rx_mem_fail);
214 				desc->error = err;
215 				break;
216 			}
217 
218 			rxm = _strp_rx_msg(head);
219 			*strp->rx_skb_nextp = skb;
220 			strp->rx_skb_nextp = &skb->next;
221 			head->data_len += skb->len;
222 			head->len += skb->len;
223 			head->truesize += skb->truesize;
224 		}
225 
226 		if (!rxm->strp.full_len) {
227 			ssize_t len;
228 
229 			len = (*strp->cb.parse_msg)(strp, head);
230 
231 			if (!len) {
232 				/* Need more header to determine length */
233 				if (!rxm->accum_len) {
234 					/* Start RX timer for new message */
235 					strp_start_rx_timer(strp);
236 				}
237 				rxm->accum_len += cand_len;
238 				eaten += cand_len;
239 				STRP_STATS_INCR(strp->stats.rx_need_more_hdr);
240 				WARN_ON(eaten != orig_len);
241 				break;
242 			} else if (len < 0) {
243 				if (len == -ESTRPIPE && rxm->accum_len) {
244 					len = -ENODATA;
245 					strp->rx_unrecov_intr = 1;
246 				} else {
247 					strp->rx_interrupted = 1;
248 				}
249 				strp_parser_err(strp, len, desc);
250 				break;
251 			} else if (len > strp->sk->sk_rcvbuf) {
252 				/* Message length exceeds maximum allowed */
253 				STRP_STATS_INCR(strp->stats.rx_msg_too_big);
254 				strp_parser_err(strp, -EMSGSIZE, desc);
255 				break;
256 			} else if (len <= (ssize_t)head->len -
257 					  skb->len - rxm->strp.offset) {
258 				/* Length must be into new skb (and also
259 				 * greater than zero)
260 				 */
261 				STRP_STATS_INCR(strp->stats.rx_bad_hdr_len);
262 				strp_parser_err(strp, -EPROTO, desc);
263 				break;
264 			}
265 
266 			rxm->strp.full_len = len;
267 		}
268 
269 		extra = (ssize_t)(rxm->accum_len + cand_len) -
270 			rxm->strp.full_len;
271 
272 		if (extra < 0) {
273 			/* Message not complete yet. */
274 			if (rxm->strp.full_len - rxm->accum_len >
275 			    strp_peek_len(strp)) {
276 				/* Don't have the whole messages in the socket
277 				 * buffer. Set strp->rx_need_bytes to wait for
278 				 * the rest of the message. Also, set "early
279 				 * eaten" since we've already buffered the skb
280 				 * but don't consume yet per strp_read_sock.
281 				 */
282 
283 				if (!rxm->accum_len) {
284 					/* Start RX timer for new message */
285 					strp_start_rx_timer(strp);
286 				}
287 
288 				strp->rx_need_bytes = rxm->strp.full_len -
289 						       rxm->accum_len;
290 				rxm->accum_len += cand_len;
291 				rxm->early_eaten = cand_len;
292 				STRP_STATS_ADD(strp->stats.rx_bytes, cand_len);
293 				desc->count = 0; /* Stop reading socket */
294 				break;
295 			}
296 			rxm->accum_len += cand_len;
297 			eaten += cand_len;
298 			WARN_ON(eaten != orig_len);
299 			break;
300 		}
301 
302 		/* Positive extra indicates ore bytes than needed for the
303 		 * message
304 		 */
305 
306 		WARN_ON(extra > cand_len);
307 
308 		eaten += (cand_len - extra);
309 
310 		/* Hurray, we have a new message! */
311 		del_timer(&strp->rx_msg_timer);
312 		strp->rx_skb_head = NULL;
313 		STRP_STATS_INCR(strp->stats.rx_msgs);
314 
315 		/* Give skb to upper layer */
316 		strp->cb.rcv_msg(strp, head);
317 
318 		if (unlikely(strp->rx_paused)) {
319 			/* Upper layer paused strp */
320 			break;
321 		}
322 	}
323 
324 	if (cloned_orig)
325 		kfree_skb(orig_skb);
326 
327 	STRP_STATS_ADD(strp->stats.rx_bytes, eaten);
328 
329 	return eaten;
330 }
331 
332 static int default_read_sock_done(struct strparser *strp, int err)
333 {
334 	return err;
335 }
336 
337 /* Called with lock held on lower socket */
338 static int strp_read_sock(struct strparser *strp)
339 {
340 	struct socket *sock = strp->sk->sk_socket;
341 	read_descriptor_t desc;
342 
343 	desc.arg.data = strp;
344 	desc.error = 0;
345 	desc.count = 1; /* give more than one skb per call */
346 
347 	/* sk should be locked here, so okay to do read_sock */
348 	sock->ops->read_sock(strp->sk, &desc, strp_recv);
349 
350 	desc.error = strp->cb.read_sock_done(strp, desc.error);
351 
352 	return desc.error;
353 }
354 
355 /* Lower sock lock held */
356 void strp_data_ready(struct strparser *strp)
357 {
358 	if (unlikely(strp->rx_stopped))
359 		return;
360 
361 	/* This check is needed to synchronize with do_strp_rx_work.
362 	 * do_strp_rx_work acquires a process lock (lock_sock) whereas
363 	 * the lock held here is bh_lock_sock. The two locks can be
364 	 * held by different threads at the same time, but bh_lock_sock
365 	 * allows a thread in BH context to safely check if the process
366 	 * lock is held. In this case, if the lock is held, queue work.
367 	 */
368 	if (sock_owned_by_user(strp->sk)) {
369 		queue_work(strp_wq, &strp->rx_work);
370 		return;
371 	}
372 
373 	if (strp->rx_paused)
374 		return;
375 
376 	if (strp->rx_need_bytes) {
377 		if (strp_peek_len(strp) >= strp->rx_need_bytes)
378 			strp->rx_need_bytes = 0;
379 		else
380 			return;
381 	}
382 
383 	if (strp_read_sock(strp) == -ENOMEM)
384 		queue_work(strp_wq, &strp->rx_work);
385 }
386 EXPORT_SYMBOL_GPL(strp_data_ready);
387 
388 static void do_strp_rx_work(struct strparser *strp)
389 {
390 	read_descriptor_t rd_desc;
391 	struct sock *csk = strp->sk;
392 
393 	/* We need the read lock to synchronize with strp_data_ready. We
394 	 * need the socket lock for calling strp_read_sock.
395 	 */
396 	lock_sock(csk);
397 
398 	if (unlikely(strp->rx_stopped))
399 		goto out;
400 
401 	if (strp->rx_paused)
402 		goto out;
403 
404 	rd_desc.arg.data = strp;
405 
406 	if (strp_read_sock(strp) == -ENOMEM)
407 		queue_work(strp_wq, &strp->rx_work);
408 
409 out:
410 	release_sock(csk);
411 }
412 
413 static void strp_rx_work(struct work_struct *w)
414 {
415 	do_strp_rx_work(container_of(w, struct strparser, rx_work));
416 }
417 
418 static void strp_rx_msg_timeout(unsigned long arg)
419 {
420 	struct strparser *strp = (struct strparser *)arg;
421 
422 	/* Message assembly timed out */
423 	STRP_STATS_INCR(strp->stats.rx_msg_timeouts);
424 	lock_sock(strp->sk);
425 	strp->cb.abort_parser(strp, ETIMEDOUT);
426 	release_sock(strp->sk);
427 }
428 
429 int strp_init(struct strparser *strp, struct sock *csk,
430 	      struct strp_callbacks *cb)
431 {
432 	struct socket *sock = csk->sk_socket;
433 
434 	if (!cb || !cb->rcv_msg || !cb->parse_msg)
435 		return -EINVAL;
436 
437 	if (!sock->ops->read_sock || !sock->ops->peek_len)
438 		return -EAFNOSUPPORT;
439 
440 	memset(strp, 0, sizeof(*strp));
441 
442 	strp->sk = csk;
443 
444 	setup_timer(&strp->rx_msg_timer, strp_rx_msg_timeout,
445 		    (unsigned long)strp);
446 
447 	INIT_WORK(&strp->rx_work, strp_rx_work);
448 
449 	strp->cb.rcv_msg = cb->rcv_msg;
450 	strp->cb.parse_msg = cb->parse_msg;
451 	strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
452 	strp->cb.abort_parser = cb->abort_parser ? : strp_abort_rx_strp;
453 
454 	return 0;
455 }
456 EXPORT_SYMBOL_GPL(strp_init);
457 
458 void strp_unpause(struct strparser *strp)
459 {
460 	strp->rx_paused = 0;
461 
462 	/* Sync setting rx_paused with RX work */
463 	smp_mb();
464 
465 	queue_work(strp_wq, &strp->rx_work);
466 }
467 EXPORT_SYMBOL_GPL(strp_unpause);
468 
469 /* strp must already be stopped so that strp_recv will no longer be called.
470  * Note that strp_done is not called with the lower socket held.
471  */
472 void strp_done(struct strparser *strp)
473 {
474 	WARN_ON(!strp->rx_stopped);
475 
476 	del_timer_sync(&strp->rx_msg_timer);
477 	cancel_work_sync(&strp->rx_work);
478 
479 	if (strp->rx_skb_head) {
480 		kfree_skb(strp->rx_skb_head);
481 		strp->rx_skb_head = NULL;
482 	}
483 }
484 EXPORT_SYMBOL_GPL(strp_done);
485 
486 void strp_stop(struct strparser *strp)
487 {
488 	strp->rx_stopped = 1;
489 }
490 EXPORT_SYMBOL_GPL(strp_stop);
491 
492 void strp_check_rcv(struct strparser *strp)
493 {
494 	queue_work(strp_wq, &strp->rx_work);
495 }
496 EXPORT_SYMBOL_GPL(strp_check_rcv);
497 
498 static int __init strp_mod_init(void)
499 {
500 	strp_wq = create_singlethread_workqueue("kstrp");
501 
502 	return 0;
503 }
504 
505 static void __exit strp_mod_exit(void)
506 {
507 	destroy_workqueue(strp_wq);
508 }
509 module_init(strp_mod_init);
510 module_exit(strp_mod_exit);
511 MODULE_LICENSE("GPL");
512