Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * SUCS NET3: | |
3 | * | |
4 | * Generic datagram handling routines. These are generic for all | |
5 | * protocols. Possibly a generic IP version on top of these would | |
6 | * make sense. Not tonight however 8-). | |
7 | * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and | |
8 | * NetROM layer all have identical poll code and mostly | |
9 | * identical recvmsg() code. So we share it here. The poll was | |
10 | * shared before but buried in udp.c so I moved it. | |
11 | * | |
113aa838 | 12 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old |
1da177e4 LT |
13 | * udp.c code) |
14 | * | |
15 | * Fixes: | |
16 | * Alan Cox : NULL return from skb_peek_copy() | |
17 | * understood | |
18 | * Alan Cox : Rewrote skb_read_datagram to avoid the | |
19 | * skb_peek_copy stuff. | |
20 | * Alan Cox : Added support for SOCK_SEQPACKET. | |
21 | * IPX can no longer use the SO_TYPE hack | |
22 | * but AX.25 now works right, and SPX is | |
23 | * feasible. | |
24 | * Alan Cox : Fixed write poll of non IP protocol | |
25 | * crash. | |
26 | * Florian La Roche: Changed for my new skbuff handling. | |
27 | * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET. | |
28 | * Linus Torvalds : BSD semantic fixes. | |
29 | * Alan Cox : Datagram iovec handling | |
30 | * Darryl Miles : Fixed non-blocking SOCK_STREAM. | |
31 | * Alan Cox : POSIXisms | |
32 | * Pete Wyckoff : Unconnected accept() fix. | |
33 | * | |
34 | */ | |
35 | ||
36 | #include <linux/module.h> | |
37 | #include <linux/types.h> | |
38 | #include <linux/kernel.h> | |
39 | #include <asm/uaccess.h> | |
40 | #include <asm/system.h> | |
41 | #include <linux/mm.h> | |
42 | #include <linux/interrupt.h> | |
43 | #include <linux/errno.h> | |
44 | #include <linux/sched.h> | |
45 | #include <linux/inet.h> | |
1da177e4 LT |
46 | #include <linux/netdevice.h> |
47 | #include <linux/rtnetlink.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/highmem.h> | |
3305b80c | 50 | #include <linux/spinlock.h> |
1da177e4 LT |
51 | |
52 | #include <net/protocol.h> | |
53 | #include <linux/skbuff.h> | |
1da177e4 | 54 | |
c752f073 ACM |
55 | #include <net/checksum.h> |
56 | #include <net/sock.h> | |
57 | #include <net/tcp_states.h> | |
e9b3cc1b | 58 | #include <trace/events/skb.h> |
1da177e4 LT |
59 | |
60 | /* | |
61 | * Is a socket 'connection oriented' ? | |
62 | */ | |
63 | static inline int connection_based(struct sock *sk) | |
64 | { | |
65 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
66 | } | |
67 | ||
bf368e4e ED |
68 | static int receiver_wake_function(wait_queue_t *wait, unsigned mode, int sync, |
69 | void *key) | |
70 | { | |
71 | unsigned long bits = (unsigned long)key; | |
72 | ||
73 | /* | |
74 | * Avoid a wakeup if event not interesting for us | |
75 | */ | |
76 | if (bits && !(bits & (POLLIN | POLLERR))) | |
77 | return 0; | |
78 | return autoremove_wake_function(wait, mode, sync, key); | |
79 | } | |
1da177e4 LT |
80 | /* |
81 | * Wait for a packet.. | |
82 | */ | |
83 | static int wait_for_packet(struct sock *sk, int *err, long *timeo_p) | |
84 | { | |
85 | int error; | |
bf368e4e | 86 | DEFINE_WAIT_FUNC(wait, receiver_wake_function); |
1da177e4 LT |
87 | |
88 | prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); | |
89 | ||
90 | /* Socket errors? */ | |
91 | error = sock_error(sk); | |
92 | if (error) | |
93 | goto out_err; | |
94 | ||
95 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
96 | goto out; | |
97 | ||
98 | /* Socket shut down? */ | |
99 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
100 | goto out_noerr; | |
101 | ||
102 | /* Sequenced packets can come disconnected. | |
103 | * If so we report the problem | |
104 | */ | |
105 | error = -ENOTCONN; | |
106 | if (connection_based(sk) && | |
107 | !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) | |
108 | goto out_err; | |
109 | ||
110 | /* handle signals */ | |
111 | if (signal_pending(current)) | |
112 | goto interrupted; | |
113 | ||
114 | error = 0; | |
115 | *timeo_p = schedule_timeout(*timeo_p); | |
116 | out: | |
117 | finish_wait(sk->sk_sleep, &wait); | |
118 | return error; | |
119 | interrupted: | |
120 | error = sock_intr_errno(*timeo_p); | |
121 | out_err: | |
122 | *err = error; | |
123 | goto out; | |
124 | out_noerr: | |
125 | *err = 0; | |
126 | error = 1; | |
127 | goto out; | |
128 | } | |
129 | ||
130 | /** | |
a59322be | 131 | * __skb_recv_datagram - Receive a datagram skbuff |
4dc3b16b PP |
132 | * @sk: socket |
133 | * @flags: MSG_ flags | |
a59322be | 134 | * @peeked: returns non-zero if this packet has been seen before |
4dc3b16b | 135 | * @err: error code returned |
1da177e4 LT |
136 | * |
137 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
138 | * and possible races. This replaces identical code in packet, raw and | |
139 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
140 | * the long standing peek and read race for datagram sockets. If you | |
141 | * alter this routine remember it must be re-entrant. | |
142 | * | |
143 | * This function will lock the socket if a skb is returned, so the caller | |
144 | * needs to unlock the socket in that case (usually by calling | |
145 | * skb_free_datagram) | |
146 | * | |
147 | * * It does not lock socket since today. This function is | |
148 | * * free of race conditions. This measure should/can improve | |
149 | * * significantly datagram socket latencies at high loads, | |
150 | * * when data copying to user space takes lots of time. | |
151 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
152 | * * 8) Great win.) | |
153 | * * --ANK (980729) | |
154 | * | |
155 | * The order of the tests when we find no data waiting are specified | |
156 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
157 | * the standard around please. | |
158 | */ | |
a59322be HX |
159 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, |
160 | int *peeked, int *err) | |
1da177e4 LT |
161 | { |
162 | struct sk_buff *skb; | |
163 | long timeo; | |
164 | /* | |
165 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
166 | */ | |
167 | int error = sock_error(sk); | |
168 | ||
169 | if (error) | |
170 | goto no_packet; | |
171 | ||
a59322be | 172 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
173 | |
174 | do { | |
175 | /* Again only user level code calls this function, so nothing | |
176 | * interrupt level will suddenly eat the receive_queue. | |
177 | * | |
178 | * Look at current nfs client by the way... | |
179 | * However, this function was corrent in any case. 8) | |
180 | */ | |
a59322be HX |
181 | unsigned long cpu_flags; |
182 | ||
183 | spin_lock_irqsave(&sk->sk_receive_queue.lock, cpu_flags); | |
184 | skb = skb_peek(&sk->sk_receive_queue); | |
185 | if (skb) { | |
186 | *peeked = skb->peeked; | |
187 | if (flags & MSG_PEEK) { | |
188 | skb->peeked = 1; | |
1da177e4 | 189 | atomic_inc(&skb->users); |
a59322be HX |
190 | } else |
191 | __skb_unlink(skb, &sk->sk_receive_queue); | |
192 | } | |
193 | spin_unlock_irqrestore(&sk->sk_receive_queue.lock, cpu_flags); | |
1da177e4 LT |
194 | |
195 | if (skb) | |
196 | return skb; | |
197 | ||
198 | /* User doesn't want to wait */ | |
199 | error = -EAGAIN; | |
200 | if (!timeo) | |
201 | goto no_packet; | |
202 | ||
203 | } while (!wait_for_packet(sk, err, &timeo)); | |
204 | ||
205 | return NULL; | |
206 | ||
207 | no_packet: | |
208 | *err = error; | |
209 | return NULL; | |
210 | } | |
a59322be HX |
211 | EXPORT_SYMBOL(__skb_recv_datagram); |
212 | ||
213 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, | |
214 | int noblock, int *err) | |
215 | { | |
216 | int peeked; | |
217 | ||
218 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
219 | &peeked, err); | |
220 | } | |
1da177e4 LT |
221 | |
222 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
223 | { | |
ead2ceb0 | 224 | consume_skb(skb); |
270acefa | 225 | sk_mem_reclaim_partial(sk); |
1da177e4 | 226 | } |
9d410c79 ED |
227 | EXPORT_SYMBOL(skb_free_datagram); |
228 | ||
229 | void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) | |
230 | { | |
231 | lock_sock(sk); | |
232 | skb_free_datagram(sk, skb); | |
233 | release_sock(sk); | |
234 | } | |
235 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 236 | |
3305b80c HX |
237 | /** |
238 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
239 | * @sk: socket | |
240 | * @skb: datagram skbuff | |
241 | * @flags: MSG_ flags | |
242 | * | |
243 | * This function frees a datagram skbuff that was received by | |
244 | * skb_recv_datagram. The flags argument must match the one | |
245 | * used for skb_recv_datagram. | |
246 | * | |
247 | * If the MSG_PEEK flag is set, and the packet is still on the | |
248 | * receive queue of the socket, it will be taken off the queue | |
249 | * before it is freed. | |
250 | * | |
251 | * This function currently only disables BH when acquiring the | |
252 | * sk_receive_queue lock. Therefore it must not be used in a | |
253 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
254 | * |
255 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
256 | */ |
257 | ||
27ab2568 | 258 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 259 | { |
27ab2568 HX |
260 | int err = 0; |
261 | ||
3305b80c | 262 | if (flags & MSG_PEEK) { |
27ab2568 | 263 | err = -ENOENT; |
3305b80c HX |
264 | spin_lock_bh(&sk->sk_receive_queue.lock); |
265 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
266 | __skb_unlink(skb, &sk->sk_receive_queue); | |
267 | atomic_dec(&skb->users); | |
27ab2568 | 268 | err = 0; |
3305b80c HX |
269 | } |
270 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
271 | } | |
272 | ||
61de71c6 | 273 | kfree_skb(skb); |
8edf19c2 | 274 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
275 | sk_mem_reclaim_partial(sk); |
276 | ||
27ab2568 | 277 | return err; |
3305b80c HX |
278 | } |
279 | ||
280 | EXPORT_SYMBOL(skb_kill_datagram); | |
281 | ||
1da177e4 LT |
282 | /** |
283 | * skb_copy_datagram_iovec - Copy a datagram to an iovec. | |
4dc3b16b PP |
284 | * @skb: buffer to copy |
285 | * @offset: offset in the buffer to start copying from | |
67be2dd1 | 286 | * @to: io vector to copy to |
4dc3b16b | 287 | * @len: amount of data to copy from buffer to iovec |
1da177e4 LT |
288 | * |
289 | * Note: the iovec is modified during the copy. | |
290 | */ | |
291 | int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset, | |
292 | struct iovec *to, int len) | |
293 | { | |
1a028e50 DM |
294 | int start = skb_headlen(skb); |
295 | int i, copy = start - offset; | |
5b1a002a | 296 | struct sk_buff *frag_iter; |
c75d721c | 297 | |
e9b3cc1b NH |
298 | trace_skb_copy_datagram_iovec(skb, len); |
299 | ||
b4d9eda0 DM |
300 | /* Copy header. */ |
301 | if (copy > 0) { | |
302 | if (copy > len) | |
303 | copy = len; | |
304 | if (memcpy_toiovec(to, skb->data + offset, copy)) | |
305 | goto fault; | |
306 | if ((len -= copy) == 0) | |
307 | return 0; | |
308 | offset += copy; | |
309 | } | |
c75d721c | 310 | |
b4d9eda0 DM |
311 | /* Copy paged appendix. Hmm... why does this look so complicated? */ |
312 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 313 | int end; |
1da177e4 | 314 | |
547b792c | 315 | WARN_ON(start > offset + len); |
1a028e50 DM |
316 | |
317 | end = start + skb_shinfo(skb)->frags[i].size; | |
b4d9eda0 DM |
318 | if ((copy = end - offset) > 0) { |
319 | int err; | |
320 | u8 *vaddr; | |
321 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
322 | struct page *page = frag->page; | |
1da177e4 LT |
323 | |
324 | if (copy > len) | |
325 | copy = len; | |
b4d9eda0 | 326 | vaddr = kmap(page); |
1a028e50 DM |
327 | err = memcpy_toiovec(to, vaddr + frag->page_offset + |
328 | offset - start, copy); | |
b4d9eda0 | 329 | kunmap(page); |
1da177e4 LT |
330 | if (err) |
331 | goto fault; | |
332 | if (!(len -= copy)) | |
333 | return 0; | |
334 | offset += copy; | |
335 | } | |
1a028e50 | 336 | start = end; |
1da177e4 | 337 | } |
b4d9eda0 | 338 | |
5b1a002a DM |
339 | skb_walk_frags(skb, frag_iter) { |
340 | int end; | |
341 | ||
342 | WARN_ON(start > offset + len); | |
343 | ||
344 | end = start + frag_iter->len; | |
345 | if ((copy = end - offset) > 0) { | |
346 | if (copy > len) | |
347 | copy = len; | |
348 | if (skb_copy_datagram_iovec(frag_iter, | |
349 | offset - start, | |
350 | to, copy)) | |
351 | goto fault; | |
352 | if ((len -= copy) == 0) | |
353 | return 0; | |
354 | offset += copy; | |
b4d9eda0 | 355 | } |
5b1a002a | 356 | start = end; |
1da177e4 | 357 | } |
b4d9eda0 DM |
358 | if (!len) |
359 | return 0; | |
360 | ||
1da177e4 LT |
361 | fault: |
362 | return -EFAULT; | |
363 | } | |
364 | ||
0a1ec07a MT |
365 | /** |
366 | * skb_copy_datagram_const_iovec - Copy a datagram to an iovec. | |
367 | * @skb: buffer to copy | |
368 | * @offset: offset in the buffer to start copying from | |
369 | * @to: io vector to copy to | |
370 | * @to_offset: offset in the io vector to start copying to | |
371 | * @len: amount of data to copy from buffer to iovec | |
372 | * | |
373 | * Returns 0 or -EFAULT. | |
374 | * Note: the iovec is not modified during the copy. | |
375 | */ | |
376 | int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset, | |
377 | const struct iovec *to, int to_offset, | |
378 | int len) | |
379 | { | |
380 | int start = skb_headlen(skb); | |
381 | int i, copy = start - offset; | |
5b1a002a | 382 | struct sk_buff *frag_iter; |
0a1ec07a MT |
383 | |
384 | /* Copy header. */ | |
385 | if (copy > 0) { | |
386 | if (copy > len) | |
387 | copy = len; | |
388 | if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy)) | |
389 | goto fault; | |
390 | if ((len -= copy) == 0) | |
391 | return 0; | |
392 | offset += copy; | |
393 | to_offset += copy; | |
394 | } | |
395 | ||
396 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
397 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
398 | int end; | |
399 | ||
400 | WARN_ON(start > offset + len); | |
401 | ||
402 | end = start + skb_shinfo(skb)->frags[i].size; | |
403 | if ((copy = end - offset) > 0) { | |
404 | int err; | |
405 | u8 *vaddr; | |
406 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
407 | struct page *page = frag->page; | |
408 | ||
409 | if (copy > len) | |
410 | copy = len; | |
411 | vaddr = kmap(page); | |
412 | err = memcpy_toiovecend(to, vaddr + frag->page_offset + | |
413 | offset - start, to_offset, copy); | |
414 | kunmap(page); | |
415 | if (err) | |
416 | goto fault; | |
417 | if (!(len -= copy)) | |
418 | return 0; | |
419 | offset += copy; | |
420 | to_offset += copy; | |
421 | } | |
422 | start = end; | |
423 | } | |
424 | ||
5b1a002a DM |
425 | skb_walk_frags(skb, frag_iter) { |
426 | int end; | |
427 | ||
428 | WARN_ON(start > offset + len); | |
429 | ||
430 | end = start + frag_iter->len; | |
431 | if ((copy = end - offset) > 0) { | |
432 | if (copy > len) | |
433 | copy = len; | |
434 | if (skb_copy_datagram_const_iovec(frag_iter, | |
435 | offset - start, | |
436 | to, to_offset, | |
437 | copy)) | |
438 | goto fault; | |
439 | if ((len -= copy) == 0) | |
440 | return 0; | |
441 | offset += copy; | |
442 | to_offset += copy; | |
0a1ec07a | 443 | } |
5b1a002a | 444 | start = end; |
0a1ec07a MT |
445 | } |
446 | if (!len) | |
447 | return 0; | |
448 | ||
449 | fault: | |
450 | return -EFAULT; | |
451 | } | |
452 | EXPORT_SYMBOL(skb_copy_datagram_const_iovec); | |
453 | ||
db543c1f RR |
454 | /** |
455 | * skb_copy_datagram_from_iovec - Copy a datagram from an iovec. | |
456 | * @skb: buffer to copy | |
457 | * @offset: offset in the buffer to start copying to | |
458 | * @from: io vector to copy to | |
6f26c9a7 | 459 | * @from_offset: offset in the io vector to start copying from |
db543c1f RR |
460 | * @len: amount of data to copy to buffer from iovec |
461 | * | |
462 | * Returns 0 or -EFAULT. | |
6f26c9a7 | 463 | * Note: the iovec is not modified during the copy. |
db543c1f RR |
464 | */ |
465 | int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset, | |
6f26c9a7 MT |
466 | const struct iovec *from, int from_offset, |
467 | int len) | |
db543c1f RR |
468 | { |
469 | int start = skb_headlen(skb); | |
470 | int i, copy = start - offset; | |
5b1a002a | 471 | struct sk_buff *frag_iter; |
db543c1f RR |
472 | |
473 | /* Copy header. */ | |
474 | if (copy > 0) { | |
475 | if (copy > len) | |
476 | copy = len; | |
d2d27bfd SS |
477 | if (memcpy_fromiovecend(skb->data + offset, from, from_offset, |
478 | copy)) | |
db543c1f RR |
479 | goto fault; |
480 | if ((len -= copy) == 0) | |
481 | return 0; | |
482 | offset += copy; | |
6f26c9a7 | 483 | from_offset += copy; |
db543c1f RR |
484 | } |
485 | ||
486 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
487 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
488 | int end; | |
489 | ||
490 | WARN_ON(start > offset + len); | |
491 | ||
492 | end = start + skb_shinfo(skb)->frags[i].size; | |
493 | if ((copy = end - offset) > 0) { | |
494 | int err; | |
495 | u8 *vaddr; | |
496 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
497 | struct page *page = frag->page; | |
498 | ||
499 | if (copy > len) | |
500 | copy = len; | |
501 | vaddr = kmap(page); | |
6f26c9a7 MT |
502 | err = memcpy_fromiovecend(vaddr + frag->page_offset + |
503 | offset - start, | |
504 | from, from_offset, copy); | |
db543c1f RR |
505 | kunmap(page); |
506 | if (err) | |
507 | goto fault; | |
508 | ||
509 | if (!(len -= copy)) | |
510 | return 0; | |
511 | offset += copy; | |
6f26c9a7 | 512 | from_offset += copy; |
db543c1f RR |
513 | } |
514 | start = end; | |
515 | } | |
516 | ||
5b1a002a DM |
517 | skb_walk_frags(skb, frag_iter) { |
518 | int end; | |
519 | ||
520 | WARN_ON(start > offset + len); | |
521 | ||
522 | end = start + frag_iter->len; | |
523 | if ((copy = end - offset) > 0) { | |
524 | if (copy > len) | |
525 | copy = len; | |
526 | if (skb_copy_datagram_from_iovec(frag_iter, | |
527 | offset - start, | |
528 | from, | |
529 | from_offset, | |
530 | copy)) | |
531 | goto fault; | |
532 | if ((len -= copy) == 0) | |
533 | return 0; | |
534 | offset += copy; | |
535 | from_offset += copy; | |
db543c1f | 536 | } |
5b1a002a | 537 | start = end; |
db543c1f RR |
538 | } |
539 | if (!len) | |
540 | return 0; | |
541 | ||
542 | fault: | |
543 | return -EFAULT; | |
544 | } | |
545 | EXPORT_SYMBOL(skb_copy_datagram_from_iovec); | |
546 | ||
1da177e4 LT |
547 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
548 | u8 __user *to, int len, | |
5084205f | 549 | __wsum *csump) |
1da177e4 | 550 | { |
1a028e50 | 551 | int start = skb_headlen(skb); |
1a028e50 | 552 | int i, copy = start - offset; |
5b1a002a DM |
553 | struct sk_buff *frag_iter; |
554 | int pos = 0; | |
1da177e4 LT |
555 | |
556 | /* Copy header. */ | |
557 | if (copy > 0) { | |
558 | int err = 0; | |
559 | if (copy > len) | |
560 | copy = len; | |
561 | *csump = csum_and_copy_to_user(skb->data + offset, to, copy, | |
562 | *csump, &err); | |
563 | if (err) | |
564 | goto fault; | |
565 | if ((len -= copy) == 0) | |
566 | return 0; | |
567 | offset += copy; | |
568 | to += copy; | |
569 | pos = copy; | |
570 | } | |
571 | ||
572 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 573 | int end; |
1da177e4 | 574 | |
547b792c | 575 | WARN_ON(start > offset + len); |
1a028e50 DM |
576 | |
577 | end = start + skb_shinfo(skb)->frags[i].size; | |
1da177e4 | 578 | if ((copy = end - offset) > 0) { |
5084205f | 579 | __wsum csum2; |
1da177e4 LT |
580 | int err = 0; |
581 | u8 *vaddr; | |
582 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
583 | struct page *page = frag->page; | |
584 | ||
585 | if (copy > len) | |
586 | copy = len; | |
587 | vaddr = kmap(page); | |
588 | csum2 = csum_and_copy_to_user(vaddr + | |
1a028e50 DM |
589 | frag->page_offset + |
590 | offset - start, | |
1da177e4 LT |
591 | to, copy, 0, &err); |
592 | kunmap(page); | |
593 | if (err) | |
594 | goto fault; | |
595 | *csump = csum_block_add(*csump, csum2, pos); | |
596 | if (!(len -= copy)) | |
597 | return 0; | |
598 | offset += copy; | |
599 | to += copy; | |
600 | pos += copy; | |
601 | } | |
1a028e50 | 602 | start = end; |
1da177e4 LT |
603 | } |
604 | ||
5b1a002a DM |
605 | skb_walk_frags(skb, frag_iter) { |
606 | int end; | |
607 | ||
608 | WARN_ON(start > offset + len); | |
609 | ||
610 | end = start + frag_iter->len; | |
611 | if ((copy = end - offset) > 0) { | |
612 | __wsum csum2 = 0; | |
613 | if (copy > len) | |
614 | copy = len; | |
615 | if (skb_copy_and_csum_datagram(frag_iter, | |
616 | offset - start, | |
617 | to, copy, | |
618 | &csum2)) | |
619 | goto fault; | |
620 | *csump = csum_block_add(*csump, csum2, pos); | |
621 | if ((len -= copy) == 0) | |
622 | return 0; | |
623 | offset += copy; | |
624 | to += copy; | |
625 | pos += copy; | |
1da177e4 | 626 | } |
5b1a002a | 627 | start = end; |
1da177e4 LT |
628 | } |
629 | if (!len) | |
630 | return 0; | |
631 | ||
632 | fault: | |
633 | return -EFAULT; | |
634 | } | |
635 | ||
759e5d00 | 636 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 637 | { |
d3bc23e7 | 638 | __sum16 sum; |
fb286bb2 | 639 | |
759e5d00 | 640 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
fb286bb2 | 641 | if (likely(!sum)) { |
84fa7933 | 642 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 HX |
643 | netdev_rx_csum_fault(skb->dev); |
644 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
645 | } | |
646 | return sum; | |
647 | } | |
759e5d00 HX |
648 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
649 | ||
650 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
651 | { | |
652 | return __skb_checksum_complete_head(skb, skb->len); | |
653 | } | |
fb286bb2 HX |
654 | EXPORT_SYMBOL(__skb_checksum_complete); |
655 | ||
1da177e4 LT |
656 | /** |
657 | * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec. | |
4dc3b16b PP |
658 | * @skb: skbuff |
659 | * @hlen: hardware length | |
67be2dd1 | 660 | * @iov: io vector |
4ec93edb | 661 | * |
1da177e4 LT |
662 | * Caller _must_ check that skb will fit to this iovec. |
663 | * | |
664 | * Returns: 0 - success. | |
665 | * -EINVAL - checksum failure. | |
666 | * -EFAULT - fault during copy. Beware, in this case iovec | |
667 | * can be modified! | |
668 | */ | |
fb286bb2 | 669 | int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, |
1da177e4 LT |
670 | int hlen, struct iovec *iov) |
671 | { | |
d3bc23e7 | 672 | __wsum csum; |
1da177e4 LT |
673 | int chunk = skb->len - hlen; |
674 | ||
ef8aef55 HX |
675 | if (!chunk) |
676 | return 0; | |
677 | ||
1da177e4 LT |
678 | /* Skip filled elements. |
679 | * Pretty silly, look at memcpy_toiovec, though 8) | |
680 | */ | |
681 | while (!iov->iov_len) | |
682 | iov++; | |
683 | ||
684 | if (iov->iov_len < chunk) { | |
fb286bb2 | 685 | if (__skb_checksum_complete(skb)) |
1da177e4 LT |
686 | goto csum_error; |
687 | if (skb_copy_datagram_iovec(skb, hlen, iov, chunk)) | |
688 | goto fault; | |
689 | } else { | |
690 | csum = csum_partial(skb->data, hlen, skb->csum); | |
691 | if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base, | |
692 | chunk, &csum)) | |
693 | goto fault; | |
d3bc23e7 | 694 | if (csum_fold(csum)) |
1da177e4 | 695 | goto csum_error; |
84fa7933 | 696 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 697 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
698 | iov->iov_len -= chunk; |
699 | iov->iov_base += chunk; | |
700 | } | |
701 | return 0; | |
702 | csum_error: | |
703 | return -EINVAL; | |
704 | fault: | |
705 | return -EFAULT; | |
706 | } | |
707 | ||
708 | /** | |
709 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
710 | * @file: file struct |
711 | * @sock: socket | |
712 | * @wait: poll table | |
1da177e4 LT |
713 | * |
714 | * Datagram poll: Again totally generic. This also handles | |
715 | * sequenced packet sockets providing the socket receive queue | |
716 | * is only ever holding data ready to receive. | |
717 | * | |
718 | * Note: when you _don't_ use this routine for this protocol, | |
719 | * and you use a different write policy from sock_writeable() | |
720 | * then please supply your own write_space callback. | |
721 | */ | |
722 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
723 | poll_table *wait) | |
724 | { | |
725 | struct sock *sk = sock->sk; | |
726 | unsigned int mask; | |
727 | ||
a57de0b4 | 728 | sock_poll_wait(file, sk->sk_sleep, wait); |
1da177e4 LT |
729 | mask = 0; |
730 | ||
731 | /* exceptional events? */ | |
732 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
733 | mask |= POLLERR; | |
f348d70a DL |
734 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
735 | mask |= POLLRDHUP; | |
1da177e4 LT |
736 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
737 | mask |= POLLHUP; | |
738 | ||
739 | /* readable? */ | |
740 | if (!skb_queue_empty(&sk->sk_receive_queue) || | |
741 | (sk->sk_shutdown & RCV_SHUTDOWN)) | |
742 | mask |= POLLIN | POLLRDNORM; | |
743 | ||
744 | /* Connection-based need to check for termination and startup */ | |
745 | if (connection_based(sk)) { | |
746 | if (sk->sk_state == TCP_CLOSE) | |
747 | mask |= POLLHUP; | |
748 | /* connection hasn't started yet? */ | |
749 | if (sk->sk_state == TCP_SYN_SENT) | |
750 | return mask; | |
751 | } | |
752 | ||
753 | /* writable? */ | |
754 | if (sock_writeable(sk)) | |
755 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
756 | else | |
757 | set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
758 | ||
759 | return mask; | |
760 | } | |
761 | ||
762 | EXPORT_SYMBOL(datagram_poll); | |
763 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec); | |
764 | EXPORT_SYMBOL(skb_copy_datagram_iovec); | |
1da177e4 | 765 | EXPORT_SYMBOL(skb_recv_datagram); |