Loading...
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
30static struct workqueue_struct *strp_wq;
31
32struct _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
41static 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 */
48static 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
66static 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 */
73static 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
82static 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 */
90static 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
332static int default_read_sock_done(struct strparser *strp, int err)
333{
334 return err;
335}
336
337/* Called with lock held on lower socket */
338static 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 */
356void 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}
386EXPORT_SYMBOL_GPL(strp_data_ready);
387
388static 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
409out:
410 release_sock(csk);
411}
412
413static void strp_rx_work(struct work_struct *w)
414{
415 do_strp_rx_work(container_of(w, struct strparser, rx_work));
416}
417
418static 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
429int 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}
456EXPORT_SYMBOL_GPL(strp_init);
457
458void 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}
467EXPORT_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 */
472void 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}
484EXPORT_SYMBOL_GPL(strp_done);
485
486void strp_stop(struct strparser *strp)
487{
488 strp->rx_stopped = 1;
489}
490EXPORT_SYMBOL_GPL(strp_stop);
491
492void strp_check_rcv(struct strparser *strp)
493{
494 queue_work(strp_wq, &strp->rx_work);
495}
496EXPORT_SYMBOL_GPL(strp_check_rcv);
497
498static int __init strp_mod_init(void)
499{
500 strp_wq = create_singlethread_workqueue("kstrp");
501
502 return 0;
503}
504
505static void __exit strp_mod_exit(void)
506{
507 destroy_workqueue(strp_wq);
508}
509module_init(strp_mod_init);
510module_exit(strp_mod_exit);
511MODULE_LICENSE("GPL");
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
30static struct workqueue_struct *strp_wq;
31
32struct _strp_msg {
33 /* Internal cb structure. struct strp_msg must be first for passing
34 * to upper layer.
35 */
36 struct strp_msg strp;
37 int accum_len;
38 int early_eaten;
39};
40
41static inline struct _strp_msg *_strp_msg(struct sk_buff *skb)
42{
43 return (struct _strp_msg *)((void *)skb->cb +
44 offsetof(struct qdisc_skb_cb, data));
45}
46
47/* Lower lock held */
48static void strp_abort_strp(struct strparser *strp, int err)
49{
50 /* Unrecoverable error in receive */
51
52 cancel_delayed_work(&strp->msg_timer_work);
53
54 if (strp->stopped)
55 return;
56
57 strp->stopped = 1;
58
59 if (strp->sk) {
60 struct sock *sk = strp->sk;
61
62 /* Report an error on the lower socket */
63 sk->sk_err = -err;
64 sk->sk_error_report(sk);
65 }
66}
67
68static void strp_start_timer(struct strparser *strp, long timeo)
69{
70 if (timeo && timeo != LONG_MAX)
71 mod_delayed_work(strp_wq, &strp->msg_timer_work, timeo);
72}
73
74/* Lower lock held */
75static void strp_parser_err(struct strparser *strp, int err,
76 read_descriptor_t *desc)
77{
78 desc->error = err;
79 kfree_skb(strp->skb_head);
80 strp->skb_head = NULL;
81 strp->cb.abort_parser(strp, err);
82}
83
84static inline int strp_peek_len(struct strparser *strp)
85{
86 if (strp->sk) {
87 struct socket *sock = strp->sk->sk_socket;
88
89 return sock->ops->peek_len(sock);
90 }
91
92 /* If we don't have an associated socket there's nothing to peek.
93 * Return int max to avoid stopping the strparser.
94 */
95
96 return INT_MAX;
97}
98
99/* Lower socket lock held */
100static int __strp_recv(read_descriptor_t *desc, struct sk_buff *orig_skb,
101 unsigned int orig_offset, size_t orig_len,
102 size_t max_msg_size, long timeo)
103{
104 struct strparser *strp = (struct strparser *)desc->arg.data;
105 struct _strp_msg *stm;
106 struct sk_buff *head, *skb;
107 size_t eaten = 0, cand_len;
108 ssize_t extra;
109 int err;
110 bool cloned_orig = false;
111
112 if (strp->paused)
113 return 0;
114
115 head = strp->skb_head;
116 if (head) {
117 /* Message already in progress */
118
119 stm = _strp_msg(head);
120 if (unlikely(stm->early_eaten)) {
121 /* Already some number of bytes on the receive sock
122 * data saved in skb_head, just indicate they
123 * are consumed.
124 */
125 eaten = orig_len <= stm->early_eaten ?
126 orig_len : stm->early_eaten;
127 stm->early_eaten -= eaten;
128
129 return eaten;
130 }
131
132 if (unlikely(orig_offset)) {
133 /* Getting data with a non-zero offset when a message is
134 * in progress is not expected. If it does happen, we
135 * need to clone and pull since we can't deal with
136 * offsets in the skbs for a message expect in the head.
137 */
138 orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
139 if (!orig_skb) {
140 STRP_STATS_INCR(strp->stats.mem_fail);
141 desc->error = -ENOMEM;
142 return 0;
143 }
144 if (!pskb_pull(orig_skb, orig_offset)) {
145 STRP_STATS_INCR(strp->stats.mem_fail);
146 kfree_skb(orig_skb);
147 desc->error = -ENOMEM;
148 return 0;
149 }
150 cloned_orig = true;
151 orig_offset = 0;
152 }
153
154 if (!strp->skb_nextp) {
155 /* We are going to append to the frags_list of head.
156 * Need to unshare the frag_list.
157 */
158 err = skb_unclone(head, GFP_ATOMIC);
159 if (err) {
160 STRP_STATS_INCR(strp->stats.mem_fail);
161 desc->error = err;
162 return 0;
163 }
164
165 if (unlikely(skb_shinfo(head)->frag_list)) {
166 /* We can't append to an sk_buff that already
167 * has a frag_list. We create a new head, point
168 * the frag_list of that to the old head, and
169 * then are able to use the old head->next for
170 * appending to the message.
171 */
172 if (WARN_ON(head->next)) {
173 desc->error = -EINVAL;
174 return 0;
175 }
176
177 skb = alloc_skb(0, GFP_ATOMIC);
178 if (!skb) {
179 STRP_STATS_INCR(strp->stats.mem_fail);
180 desc->error = -ENOMEM;
181 return 0;
182 }
183 skb->len = head->len;
184 skb->data_len = head->len;
185 skb->truesize = head->truesize;
186 *_strp_msg(skb) = *_strp_msg(head);
187 strp->skb_nextp = &head->next;
188 skb_shinfo(skb)->frag_list = head;
189 strp->skb_head = skb;
190 head = skb;
191 } else {
192 strp->skb_nextp =
193 &skb_shinfo(head)->frag_list;
194 }
195 }
196 }
197
198 while (eaten < orig_len) {
199 /* Always clone since we will consume something */
200 skb = skb_clone(orig_skb, GFP_ATOMIC);
201 if (!skb) {
202 STRP_STATS_INCR(strp->stats.mem_fail);
203 desc->error = -ENOMEM;
204 break;
205 }
206
207 cand_len = orig_len - eaten;
208
209 head = strp->skb_head;
210 if (!head) {
211 head = skb;
212 strp->skb_head = head;
213 /* Will set skb_nextp on next packet if needed */
214 strp->skb_nextp = NULL;
215 stm = _strp_msg(head);
216 memset(stm, 0, sizeof(*stm));
217 stm->strp.offset = orig_offset + eaten;
218 } else {
219 /* Unclone since we may be appending to an skb that we
220 * already share a frag_list with.
221 */
222 err = skb_unclone(skb, GFP_ATOMIC);
223 if (err) {
224 STRP_STATS_INCR(strp->stats.mem_fail);
225 desc->error = err;
226 break;
227 }
228
229 stm = _strp_msg(head);
230 *strp->skb_nextp = skb;
231 strp->skb_nextp = &skb->next;
232 head->data_len += skb->len;
233 head->len += skb->len;
234 head->truesize += skb->truesize;
235 }
236
237 if (!stm->strp.full_len) {
238 ssize_t len;
239
240 len = (*strp->cb.parse_msg)(strp, head);
241
242 if (!len) {
243 /* Need more header to determine length */
244 if (!stm->accum_len) {
245 /* Start RX timer for new message */
246 strp_start_timer(strp, timeo);
247 }
248 stm->accum_len += cand_len;
249 eaten += cand_len;
250 STRP_STATS_INCR(strp->stats.need_more_hdr);
251 WARN_ON(eaten != orig_len);
252 break;
253 } else if (len < 0) {
254 if (len == -ESTRPIPE && stm->accum_len) {
255 len = -ENODATA;
256 strp->unrecov_intr = 1;
257 } else {
258 strp->interrupted = 1;
259 }
260 strp_parser_err(strp, len, desc);
261 break;
262 } else if (len > max_msg_size) {
263 /* Message length exceeds maximum allowed */
264 STRP_STATS_INCR(strp->stats.msg_too_big);
265 strp_parser_err(strp, -EMSGSIZE, desc);
266 break;
267 } else if (len <= (ssize_t)head->len -
268 skb->len - stm->strp.offset) {
269 /* Length must be into new skb (and also
270 * greater than zero)
271 */
272 STRP_STATS_INCR(strp->stats.bad_hdr_len);
273 strp_parser_err(strp, -EPROTO, desc);
274 break;
275 }
276
277 stm->strp.full_len = len;
278 }
279
280 extra = (ssize_t)(stm->accum_len + cand_len) -
281 stm->strp.full_len;
282
283 if (extra < 0) {
284 /* Message not complete yet. */
285 if (stm->strp.full_len - stm->accum_len >
286 strp_peek_len(strp)) {
287 /* Don't have the whole message in the socket
288 * buffer. Set strp->need_bytes to wait for
289 * the rest of the message. Also, set "early
290 * eaten" since we've already buffered the skb
291 * but don't consume yet per strp_read_sock.
292 */
293
294 if (!stm->accum_len) {
295 /* Start RX timer for new message */
296 strp_start_timer(strp, timeo);
297 }
298
299 stm->accum_len += cand_len;
300 strp->need_bytes = stm->strp.full_len -
301 stm->accum_len;
302 stm->early_eaten = cand_len;
303 STRP_STATS_ADD(strp->stats.bytes, cand_len);
304 desc->count = 0; /* Stop reading socket */
305 break;
306 }
307 stm->accum_len += cand_len;
308 eaten += cand_len;
309 WARN_ON(eaten != orig_len);
310 break;
311 }
312
313 /* Positive extra indicates ore bytes than needed for the
314 * message
315 */
316
317 WARN_ON(extra > cand_len);
318
319 eaten += (cand_len - extra);
320
321 /* Hurray, we have a new message! */
322 cancel_delayed_work(&strp->msg_timer_work);
323 strp->skb_head = NULL;
324 strp->need_bytes = 0;
325 STRP_STATS_INCR(strp->stats.msgs);
326
327 /* Give skb to upper layer */
328 strp->cb.rcv_msg(strp, head);
329
330 if (unlikely(strp->paused)) {
331 /* Upper layer paused strp */
332 break;
333 }
334 }
335
336 if (cloned_orig)
337 kfree_skb(orig_skb);
338
339 STRP_STATS_ADD(strp->stats.bytes, eaten);
340
341 return eaten;
342}
343
344int strp_process(struct strparser *strp, struct sk_buff *orig_skb,
345 unsigned int orig_offset, size_t orig_len,
346 size_t max_msg_size, long timeo)
347{
348 read_descriptor_t desc; /* Dummy arg to strp_recv */
349
350 desc.arg.data = strp;
351
352 return __strp_recv(&desc, orig_skb, orig_offset, orig_len,
353 max_msg_size, timeo);
354}
355EXPORT_SYMBOL_GPL(strp_process);
356
357static int strp_recv(read_descriptor_t *desc, struct sk_buff *orig_skb,
358 unsigned int orig_offset, size_t orig_len)
359{
360 struct strparser *strp = (struct strparser *)desc->arg.data;
361
362 return __strp_recv(desc, orig_skb, orig_offset, orig_len,
363 strp->sk->sk_rcvbuf, strp->sk->sk_rcvtimeo);
364}
365
366static int default_read_sock_done(struct strparser *strp, int err)
367{
368 return err;
369}
370
371/* Called with lock held on lower socket */
372static int strp_read_sock(struct strparser *strp)
373{
374 struct socket *sock = strp->sk->sk_socket;
375 read_descriptor_t desc;
376
377 if (unlikely(!sock || !sock->ops || !sock->ops->read_sock))
378 return -EBUSY;
379
380 desc.arg.data = strp;
381 desc.error = 0;
382 desc.count = 1; /* give more than one skb per call */
383
384 /* sk should be locked here, so okay to do read_sock */
385 sock->ops->read_sock(strp->sk, &desc, strp_recv);
386
387 desc.error = strp->cb.read_sock_done(strp, desc.error);
388
389 return desc.error;
390}
391
392/* Lower sock lock held */
393void strp_data_ready(struct strparser *strp)
394{
395 if (unlikely(strp->stopped))
396 return;
397
398 /* This check is needed to synchronize with do_strp_work.
399 * do_strp_work acquires a process lock (lock_sock) whereas
400 * the lock held here is bh_lock_sock. The two locks can be
401 * held by different threads at the same time, but bh_lock_sock
402 * allows a thread in BH context to safely check if the process
403 * lock is held. In this case, if the lock is held, queue work.
404 */
405 if (sock_owned_by_user_nocheck(strp->sk)) {
406 queue_work(strp_wq, &strp->work);
407 return;
408 }
409
410 if (strp->paused)
411 return;
412
413 if (strp->need_bytes) {
414 if (strp_peek_len(strp) < strp->need_bytes)
415 return;
416 }
417
418 if (strp_read_sock(strp) == -ENOMEM)
419 queue_work(strp_wq, &strp->work);
420}
421EXPORT_SYMBOL_GPL(strp_data_ready);
422
423static void do_strp_work(struct strparser *strp)
424{
425 read_descriptor_t rd_desc;
426
427 /* We need the read lock to synchronize with strp_data_ready. We
428 * need the socket lock for calling strp_read_sock.
429 */
430 strp->cb.lock(strp);
431
432 if (unlikely(strp->stopped))
433 goto out;
434
435 if (strp->paused)
436 goto out;
437
438 rd_desc.arg.data = strp;
439
440 if (strp_read_sock(strp) == -ENOMEM)
441 queue_work(strp_wq, &strp->work);
442
443out:
444 strp->cb.unlock(strp);
445}
446
447static void strp_work(struct work_struct *w)
448{
449 do_strp_work(container_of(w, struct strparser, work));
450}
451
452static void strp_msg_timeout(struct work_struct *w)
453{
454 struct strparser *strp = container_of(w, struct strparser,
455 msg_timer_work.work);
456
457 /* Message assembly timed out */
458 STRP_STATS_INCR(strp->stats.msg_timeouts);
459 strp->cb.lock(strp);
460 strp->cb.abort_parser(strp, -ETIMEDOUT);
461 strp->cb.unlock(strp);
462}
463
464static void strp_sock_lock(struct strparser *strp)
465{
466 lock_sock(strp->sk);
467}
468
469static void strp_sock_unlock(struct strparser *strp)
470{
471 release_sock(strp->sk);
472}
473
474int strp_init(struct strparser *strp, struct sock *sk,
475 const struct strp_callbacks *cb)
476{
477
478 if (!cb || !cb->rcv_msg || !cb->parse_msg)
479 return -EINVAL;
480
481 /* The sk (sock) arg determines the mode of the stream parser.
482 *
483 * If the sock is set then the strparser is in receive callback mode.
484 * The upper layer calls strp_data_ready to kick receive processing
485 * and strparser calls the read_sock function on the socket to
486 * get packets.
487 *
488 * If the sock is not set then the strparser is in general mode.
489 * The upper layer calls strp_process for each skb to be parsed.
490 */
491
492 if (!sk) {
493 if (!cb->lock || !cb->unlock)
494 return -EINVAL;
495 }
496
497 memset(strp, 0, sizeof(*strp));
498
499 strp->sk = sk;
500
501 strp->cb.lock = cb->lock ? : strp_sock_lock;
502 strp->cb.unlock = cb->unlock ? : strp_sock_unlock;
503 strp->cb.rcv_msg = cb->rcv_msg;
504 strp->cb.parse_msg = cb->parse_msg;
505 strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
506 strp->cb.abort_parser = cb->abort_parser ? : strp_abort_strp;
507
508 INIT_DELAYED_WORK(&strp->msg_timer_work, strp_msg_timeout);
509 INIT_WORK(&strp->work, strp_work);
510
511 return 0;
512}
513EXPORT_SYMBOL_GPL(strp_init);
514
515void strp_unpause(struct strparser *strp)
516{
517 strp->paused = 0;
518
519 /* Sync setting paused with RX work */
520 smp_mb();
521
522 queue_work(strp_wq, &strp->work);
523}
524EXPORT_SYMBOL_GPL(strp_unpause);
525
526/* strp must already be stopped so that strp_recv will no longer be called.
527 * Note that strp_done is not called with the lower socket held.
528 */
529void strp_done(struct strparser *strp)
530{
531 WARN_ON(!strp->stopped);
532
533 cancel_delayed_work_sync(&strp->msg_timer_work);
534 cancel_work_sync(&strp->work);
535
536 if (strp->skb_head) {
537 kfree_skb(strp->skb_head);
538 strp->skb_head = NULL;
539 }
540}
541EXPORT_SYMBOL_GPL(strp_done);
542
543void strp_stop(struct strparser *strp)
544{
545 strp->stopped = 1;
546}
547EXPORT_SYMBOL_GPL(strp_stop);
548
549void strp_check_rcv(struct strparser *strp)
550{
551 queue_work(strp_wq, &strp->work);
552}
553EXPORT_SYMBOL_GPL(strp_check_rcv);
554
555static int __init strp_mod_init(void)
556{
557 strp_wq = create_singlethread_workqueue("kstrp");
558
559 return 0;
560}
561
562static void __exit strp_mod_exit(void)
563{
564 destroy_workqueue(strp_wq);
565}
566module_init(strp_mod_init);
567module_exit(strp_mod_exit);
568MODULE_LICENSE("GPL");