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> | |
1da177e4 LT |
40 | #include <linux/mm.h> |
41 | #include <linux/interrupt.h> | |
42 | #include <linux/errno.h> | |
43 | #include <linux/sched.h> | |
44 | #include <linux/inet.h> | |
1da177e4 LT |
45 | #include <linux/netdevice.h> |
46 | #include <linux/rtnetlink.h> | |
47 | #include <linux/poll.h> | |
48 | #include <linux/highmem.h> | |
3305b80c | 49 | #include <linux/spinlock.h> |
5a0e3ad6 | 50 | #include <linux/slab.h> |
0433547a | 51 | #include <linux/pagemap.h> |
a8f820aa | 52 | #include <linux/uio.h> |
1da177e4 LT |
53 | |
54 | #include <net/protocol.h> | |
55 | #include <linux/skbuff.h> | |
1da177e4 | 56 | |
c752f073 ACM |
57 | #include <net/checksum.h> |
58 | #include <net/sock.h> | |
59 | #include <net/tcp_states.h> | |
e9b3cc1b | 60 | #include <trace/events/skb.h> |
076bb0c8 | 61 | #include <net/busy_poll.h> |
1da177e4 LT |
62 | |
63 | /* | |
64 | * Is a socket 'connection oriented' ? | |
65 | */ | |
66 | static inline int connection_based(struct sock *sk) | |
67 | { | |
68 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
69 | } | |
70 | ||
95c96174 | 71 | static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync, |
bf368e4e ED |
72 | void *key) |
73 | { | |
74 | unsigned long bits = (unsigned long)key; | |
75 | ||
76 | /* | |
77 | * Avoid a wakeup if event not interesting for us | |
78 | */ | |
79 | if (bits && !(bits & (POLLIN | POLLERR))) | |
80 | return 0; | |
81 | return autoremove_wake_function(wait, mode, sync, key); | |
82 | } | |
1da177e4 | 83 | /* |
39cc8613 | 84 | * Wait for the last received packet to be different from skb |
1da177e4 | 85 | */ |
39cc8613 BP |
86 | static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, |
87 | const struct sk_buff *skb) | |
1da177e4 LT |
88 | { |
89 | int error; | |
bf368e4e | 90 | DEFINE_WAIT_FUNC(wait, receiver_wake_function); |
1da177e4 | 91 | |
aa395145 | 92 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); |
1da177e4 LT |
93 | |
94 | /* Socket errors? */ | |
95 | error = sock_error(sk); | |
96 | if (error) | |
97 | goto out_err; | |
98 | ||
39cc8613 | 99 | if (sk->sk_receive_queue.prev != skb) |
1da177e4 LT |
100 | goto out; |
101 | ||
102 | /* Socket shut down? */ | |
103 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
104 | goto out_noerr; | |
105 | ||
106 | /* Sequenced packets can come disconnected. | |
107 | * If so we report the problem | |
108 | */ | |
109 | error = -ENOTCONN; | |
110 | if (connection_based(sk) && | |
111 | !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) | |
112 | goto out_err; | |
113 | ||
114 | /* handle signals */ | |
115 | if (signal_pending(current)) | |
116 | goto interrupted; | |
117 | ||
118 | error = 0; | |
119 | *timeo_p = schedule_timeout(*timeo_p); | |
120 | out: | |
aa395145 | 121 | finish_wait(sk_sleep(sk), &wait); |
1da177e4 LT |
122 | return error; |
123 | interrupted: | |
124 | error = sock_intr_errno(*timeo_p); | |
125 | out_err: | |
126 | *err = error; | |
127 | goto out; | |
128 | out_noerr: | |
129 | *err = 0; | |
130 | error = 1; | |
131 | goto out; | |
132 | } | |
133 | ||
134 | /** | |
a59322be | 135 | * __skb_recv_datagram - Receive a datagram skbuff |
4dc3b16b PP |
136 | * @sk: socket |
137 | * @flags: MSG_ flags | |
39cc8613 | 138 | * @peeked: returns non-zero if this packet has been seen before |
3f518bf7 PE |
139 | * @off: an offset in bytes to peek skb from. Returns an offset |
140 | * within an skb where data actually starts | |
4dc3b16b | 141 | * @err: error code returned |
1da177e4 LT |
142 | * |
143 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
144 | * and possible races. This replaces identical code in packet, raw and | |
145 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
146 | * the long standing peek and read race for datagram sockets. If you | |
147 | * alter this routine remember it must be re-entrant. | |
148 | * | |
149 | * This function will lock the socket if a skb is returned, so the caller | |
150 | * needs to unlock the socket in that case (usually by calling | |
151 | * skb_free_datagram) | |
152 | * | |
153 | * * It does not lock socket since today. This function is | |
154 | * * free of race conditions. This measure should/can improve | |
155 | * * significantly datagram socket latencies at high loads, | |
156 | * * when data copying to user space takes lots of time. | |
157 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
158 | * * 8) Great win.) | |
159 | * * --ANK (980729) | |
160 | * | |
161 | * The order of the tests when we find no data waiting are specified | |
162 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
163 | * the standard around please. | |
164 | */ | |
95c96174 | 165 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
3f518bf7 | 166 | int *peeked, int *off, int *err) |
1da177e4 | 167 | { |
39cc8613 | 168 | struct sk_buff *skb, *last; |
1da177e4 LT |
169 | long timeo; |
170 | /* | |
171 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
172 | */ | |
173 | int error = sock_error(sk); | |
174 | ||
175 | if (error) | |
176 | goto no_packet; | |
177 | ||
a59322be | 178 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
179 | |
180 | do { | |
181 | /* Again only user level code calls this function, so nothing | |
182 | * interrupt level will suddenly eat the receive_queue. | |
183 | * | |
184 | * Look at current nfs client by the way... | |
8917a3c0 | 185 | * However, this function was correct in any case. 8) |
1da177e4 | 186 | */ |
a59322be | 187 | unsigned long cpu_flags; |
4934b032 | 188 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
39cc8613 | 189 | int _off = *off; |
a59322be | 190 | |
39cc8613 | 191 | last = (struct sk_buff *)queue; |
4934b032 | 192 | spin_lock_irqsave(&queue->lock, cpu_flags); |
3f518bf7 | 193 | skb_queue_walk(queue, skb) { |
39cc8613 | 194 | last = skb; |
a59322be HX |
195 | *peeked = skb->peeked; |
196 | if (flags & MSG_PEEK) { | |
39cc8613 | 197 | if (_off >= skb->len && (skb->len || _off || |
add05ad4 | 198 | skb->peeked)) { |
39cc8613 | 199 | _off -= skb->len; |
3f518bf7 PE |
200 | continue; |
201 | } | |
a59322be | 202 | skb->peeked = 1; |
1da177e4 | 203 | atomic_inc(&skb->users); |
a59322be | 204 | } else |
4934b032 | 205 | __skb_unlink(skb, queue); |
1da177e4 | 206 | |
3f518bf7 | 207 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
39cc8613 | 208 | *off = _off; |
1da177e4 | 209 | return skb; |
3f518bf7 PE |
210 | } |
211 | spin_unlock_irqrestore(&queue->lock, cpu_flags); | |
1da177e4 | 212 | |
cbf55001 ET |
213 | if (sk_can_busy_loop(sk) && |
214 | sk_busy_loop(sk, flags & MSG_DONTWAIT)) | |
a5b50476 ET |
215 | continue; |
216 | ||
1da177e4 LT |
217 | /* User doesn't want to wait */ |
218 | error = -EAGAIN; | |
219 | if (!timeo) | |
220 | goto no_packet; | |
221 | ||
39cc8613 | 222 | } while (!wait_for_more_packets(sk, err, &timeo, last)); |
1da177e4 LT |
223 | |
224 | return NULL; | |
225 | ||
226 | no_packet: | |
227 | *err = error; | |
228 | return NULL; | |
229 | } | |
a59322be HX |
230 | EXPORT_SYMBOL(__skb_recv_datagram); |
231 | ||
95c96174 | 232 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
233 | int noblock, int *err) |
234 | { | |
3f518bf7 | 235 | int peeked, off = 0; |
a59322be HX |
236 | |
237 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
3f518bf7 | 238 | &peeked, &off, err); |
a59322be | 239 | } |
9e34a5b5 | 240 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
241 | |
242 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
243 | { | |
ead2ceb0 | 244 | consume_skb(skb); |
270acefa | 245 | sk_mem_reclaim_partial(sk); |
1da177e4 | 246 | } |
9d410c79 ED |
247 | EXPORT_SYMBOL(skb_free_datagram); |
248 | ||
249 | void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) | |
250 | { | |
8a74ad60 ED |
251 | bool slow; |
252 | ||
93bb64ea ED |
253 | if (likely(atomic_read(&skb->users) == 1)) |
254 | smp_rmb(); | |
255 | else if (likely(!atomic_dec_and_test(&skb->users))) | |
256 | return; | |
257 | ||
8a74ad60 | 258 | slow = lock_sock_fast(sk); |
4b0b72f7 ED |
259 | skb_orphan(skb); |
260 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 261 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 262 | |
93bb64ea ED |
263 | /* skb is now orphaned, can be freed outside of locked section */ |
264 | __kfree_skb(skb); | |
9d410c79 ED |
265 | } |
266 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 267 | |
3305b80c HX |
268 | /** |
269 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
270 | * @sk: socket | |
271 | * @skb: datagram skbuff | |
272 | * @flags: MSG_ flags | |
273 | * | |
274 | * This function frees a datagram skbuff that was received by | |
275 | * skb_recv_datagram. The flags argument must match the one | |
276 | * used for skb_recv_datagram. | |
277 | * | |
278 | * If the MSG_PEEK flag is set, and the packet is still on the | |
279 | * receive queue of the socket, it will be taken off the queue | |
280 | * before it is freed. | |
281 | * | |
282 | * This function currently only disables BH when acquiring the | |
283 | * sk_receive_queue lock. Therefore it must not be used in a | |
284 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
285 | * |
286 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
287 | */ |
288 | ||
27ab2568 | 289 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 290 | { |
27ab2568 HX |
291 | int err = 0; |
292 | ||
3305b80c | 293 | if (flags & MSG_PEEK) { |
27ab2568 | 294 | err = -ENOENT; |
3305b80c HX |
295 | spin_lock_bh(&sk->sk_receive_queue.lock); |
296 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
297 | __skb_unlink(skb, &sk->sk_receive_queue); | |
298 | atomic_dec(&skb->users); | |
27ab2568 | 299 | err = 0; |
3305b80c HX |
300 | } |
301 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
302 | } | |
303 | ||
61de71c6 | 304 | kfree_skb(skb); |
8edf19c2 | 305 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
306 | sk_mem_reclaim_partial(sk); |
307 | ||
27ab2568 | 308 | return err; |
3305b80c | 309 | } |
3305b80c HX |
310 | EXPORT_SYMBOL(skb_kill_datagram); |
311 | ||
a8f820aa HX |
312 | /** |
313 | * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. | |
314 | * @skb: buffer to copy | |
315 | * @offset: offset in the buffer to start copying from | |
316 | * @to: iovec iterator to copy to | |
317 | * @len: amount of data to copy from buffer to iovec | |
318 | */ | |
319 | int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, | |
320 | struct iov_iter *to, int len) | |
321 | { | |
322 | int start = skb_headlen(skb); | |
323 | int i, copy = start - offset; | |
324 | struct sk_buff *frag_iter; | |
325 | ||
326 | trace_skb_copy_datagram_iovec(skb, len); | |
327 | ||
328 | /* Copy header. */ | |
329 | if (copy > 0) { | |
330 | if (copy > len) | |
331 | copy = len; | |
332 | if (copy_to_iter(skb->data + offset, copy, to) != copy) | |
333 | goto short_copy; | |
334 | if ((len -= copy) == 0) | |
335 | return 0; | |
336 | offset += copy; | |
337 | } | |
338 | ||
339 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
340 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
341 | int end; | |
342 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
343 | ||
344 | WARN_ON(start > offset + len); | |
345 | ||
346 | end = start + skb_frag_size(frag); | |
347 | if ((copy = end - offset) > 0) { | |
348 | if (copy > len) | |
349 | copy = len; | |
350 | if (copy_page_to_iter(skb_frag_page(frag), | |
351 | frag->page_offset + offset - | |
352 | start, copy, to) != copy) | |
353 | goto short_copy; | |
354 | if (!(len -= copy)) | |
355 | return 0; | |
356 | offset += copy; | |
357 | } | |
358 | start = end; | |
359 | } | |
360 | ||
361 | skb_walk_frags(skb, frag_iter) { | |
362 | int end; | |
363 | ||
364 | WARN_ON(start > offset + len); | |
365 | ||
366 | end = start + frag_iter->len; | |
367 | if ((copy = end - offset) > 0) { | |
368 | if (copy > len) | |
369 | copy = len; | |
370 | if (skb_copy_datagram_iter(frag_iter, offset - start, | |
371 | to, copy)) | |
372 | goto fault; | |
373 | if ((len -= copy) == 0) | |
374 | return 0; | |
375 | offset += copy; | |
376 | } | |
377 | start = end; | |
378 | } | |
379 | if (!len) | |
380 | return 0; | |
381 | ||
382 | /* This is not really a user copy fault, but rather someone | |
383 | * gave us a bogus length on the skb. We should probably | |
384 | * print a warning here as it may indicate a kernel bug. | |
385 | */ | |
386 | ||
387 | fault: | |
388 | return -EFAULT; | |
389 | ||
390 | short_copy: | |
391 | if (iov_iter_count(to)) | |
392 | goto fault; | |
393 | ||
394 | return 0; | |
395 | } | |
396 | EXPORT_SYMBOL(skb_copy_datagram_iter); | |
397 | ||
db543c1f | 398 | /** |
8feb2fb2 | 399 | * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. |
db543c1f RR |
400 | * @skb: buffer to copy |
401 | * @offset: offset in the buffer to start copying to | |
8feb2fb2 | 402 | * @from: the copy source |
db543c1f RR |
403 | * @len: amount of data to copy to buffer from iovec |
404 | * | |
405 | * Returns 0 or -EFAULT. | |
db543c1f | 406 | */ |
3a654f97 AV |
407 | int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, |
408 | struct iov_iter *from, | |
409 | int len) | |
410 | { | |
411 | int start = skb_headlen(skb); | |
412 | int i, copy = start - offset; | |
413 | struct sk_buff *frag_iter; | |
414 | ||
415 | /* Copy header. */ | |
416 | if (copy > 0) { | |
417 | if (copy > len) | |
418 | copy = len; | |
419 | if (copy_from_iter(skb->data + offset, copy, from) != copy) | |
420 | goto fault; | |
421 | if ((len -= copy) == 0) | |
422 | return 0; | |
423 | offset += copy; | |
424 | } | |
425 | ||
426 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
427 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
428 | int end; | |
429 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
430 | ||
431 | WARN_ON(start > offset + len); | |
432 | ||
433 | end = start + skb_frag_size(frag); | |
434 | if ((copy = end - offset) > 0) { | |
435 | size_t copied; | |
436 | ||
437 | if (copy > len) | |
438 | copy = len; | |
439 | copied = copy_page_from_iter(skb_frag_page(frag), | |
440 | frag->page_offset + offset - start, | |
441 | copy, from); | |
442 | if (copied != copy) | |
443 | goto fault; | |
444 | ||
445 | if (!(len -= copy)) | |
446 | return 0; | |
447 | offset += copy; | |
448 | } | |
449 | start = end; | |
450 | } | |
451 | ||
452 | skb_walk_frags(skb, frag_iter) { | |
453 | int end; | |
454 | ||
455 | WARN_ON(start > offset + len); | |
456 | ||
457 | end = start + frag_iter->len; | |
458 | if ((copy = end - offset) > 0) { | |
459 | if (copy > len) | |
460 | copy = len; | |
461 | if (skb_copy_datagram_from_iter(frag_iter, | |
462 | offset - start, | |
463 | from, copy)) | |
464 | goto fault; | |
465 | if ((len -= copy) == 0) | |
466 | return 0; | |
467 | offset += copy; | |
468 | } | |
469 | start = end; | |
470 | } | |
471 | if (!len) | |
472 | return 0; | |
473 | ||
474 | fault: | |
475 | return -EFAULT; | |
476 | } | |
477 | EXPORT_SYMBOL(skb_copy_datagram_from_iter); | |
478 | ||
c3bdeb5c | 479 | /** |
195e952d | 480 | * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter |
c3bdeb5c | 481 | * @skb: buffer to copy |
195e952d | 482 | * @from: the source to copy from |
c3bdeb5c JW |
483 | * |
484 | * The function will first copy up to headlen, and then pin the userspace | |
485 | * pages and build frags through them. | |
486 | * | |
487 | * Returns 0, -EFAULT or -EMSGSIZE. | |
c3bdeb5c | 488 | */ |
3a654f97 AV |
489 | int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) |
490 | { | |
491 | int len = iov_iter_count(from); | |
492 | int copy = min_t(int, skb_headlen(skb), len); | |
493 | int frag = 0; | |
494 | ||
495 | /* copy up to skb headlen */ | |
496 | if (skb_copy_datagram_from_iter(skb, 0, from, copy)) | |
497 | return -EFAULT; | |
498 | ||
499 | while (iov_iter_count(from)) { | |
500 | struct page *pages[MAX_SKB_FRAGS]; | |
501 | size_t start; | |
502 | ssize_t copied; | |
503 | unsigned long truesize; | |
504 | int n = 0; | |
505 | ||
506 | if (frag == MAX_SKB_FRAGS) | |
507 | return -EMSGSIZE; | |
508 | ||
509 | copied = iov_iter_get_pages(from, pages, ~0U, | |
510 | MAX_SKB_FRAGS - frag, &start); | |
511 | if (copied < 0) | |
512 | return -EFAULT; | |
513 | ||
514 | iov_iter_advance(from, copied); | |
515 | ||
516 | truesize = PAGE_ALIGN(copied + start); | |
517 | skb->data_len += copied; | |
518 | skb->len += copied; | |
519 | skb->truesize += truesize; | |
520 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
521 | while (copied) { | |
522 | int size = min_t(int, copied, PAGE_SIZE - start); | |
523 | skb_fill_page_desc(skb, frag++, pages[n], start, size); | |
524 | start = 0; | |
525 | copied -= size; | |
526 | n++; | |
527 | } | |
528 | } | |
529 | return 0; | |
530 | } | |
531 | EXPORT_SYMBOL(zerocopy_sg_from_iter); | |
532 | ||
1da177e4 | 533 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
e5a4b0bb | 534 | struct iov_iter *to, int len, |
5084205f | 535 | __wsum *csump) |
1da177e4 | 536 | { |
1a028e50 | 537 | int start = skb_headlen(skb); |
1a028e50 | 538 | int i, copy = start - offset; |
5b1a002a DM |
539 | struct sk_buff *frag_iter; |
540 | int pos = 0; | |
e5a4b0bb | 541 | int n; |
1da177e4 LT |
542 | |
543 | /* Copy header. */ | |
544 | if (copy > 0) { | |
1da177e4 LT |
545 | if (copy > len) |
546 | copy = len; | |
e5a4b0bb AV |
547 | n = csum_and_copy_to_iter(skb->data + offset, copy, csump, to); |
548 | if (n != copy) | |
1da177e4 LT |
549 | goto fault; |
550 | if ((len -= copy) == 0) | |
551 | return 0; | |
552 | offset += copy; | |
1da177e4 LT |
553 | pos = copy; |
554 | } | |
555 | ||
556 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 557 | int end; |
9e903e08 | 558 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 559 | |
547b792c | 560 | WARN_ON(start > offset + len); |
1a028e50 | 561 | |
9e903e08 | 562 | end = start + skb_frag_size(frag); |
1da177e4 | 563 | if ((copy = end - offset) > 0) { |
e5a4b0bb | 564 | __wsum csum2 = 0; |
ea2ab693 | 565 | struct page *page = skb_frag_page(frag); |
e5a4b0bb | 566 | u8 *vaddr = kmap(page); |
1da177e4 LT |
567 | |
568 | if (copy > len) | |
569 | copy = len; | |
e5a4b0bb AV |
570 | n = csum_and_copy_to_iter(vaddr + frag->page_offset + |
571 | offset - start, copy, | |
572 | &csum2, to); | |
1da177e4 | 573 | kunmap(page); |
e5a4b0bb | 574 | if (n != copy) |
1da177e4 LT |
575 | goto fault; |
576 | *csump = csum_block_add(*csump, csum2, pos); | |
577 | if (!(len -= copy)) | |
578 | return 0; | |
579 | offset += copy; | |
1da177e4 LT |
580 | pos += copy; |
581 | } | |
1a028e50 | 582 | start = end; |
1da177e4 LT |
583 | } |
584 | ||
5b1a002a DM |
585 | skb_walk_frags(skb, frag_iter) { |
586 | int end; | |
587 | ||
588 | WARN_ON(start > offset + len); | |
589 | ||
590 | end = start + frag_iter->len; | |
591 | if ((copy = end - offset) > 0) { | |
592 | __wsum csum2 = 0; | |
593 | if (copy > len) | |
594 | copy = len; | |
595 | if (skb_copy_and_csum_datagram(frag_iter, | |
596 | offset - start, | |
597 | to, copy, | |
598 | &csum2)) | |
599 | goto fault; | |
600 | *csump = csum_block_add(*csump, csum2, pos); | |
601 | if ((len -= copy) == 0) | |
602 | return 0; | |
603 | offset += copy; | |
5b1a002a | 604 | pos += copy; |
1da177e4 | 605 | } |
5b1a002a | 606 | start = end; |
1da177e4 LT |
607 | } |
608 | if (!len) | |
609 | return 0; | |
610 | ||
611 | fault: | |
612 | return -EFAULT; | |
613 | } | |
614 | ||
759e5d00 | 615 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 616 | { |
d3bc23e7 | 617 | __sum16 sum; |
fb286bb2 | 618 | |
759e5d00 | 619 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
46fb51eb TH |
620 | if (likely(!sum)) { |
621 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
622 | !skb->csum_complete_sw) | |
623 | netdev_rx_csum_fault(skb->dev); | |
624 | } | |
625 | skb->csum_valid = !sum; | |
fb286bb2 HX |
626 | return sum; |
627 | } | |
759e5d00 HX |
628 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
629 | ||
630 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
631 | { | |
46fb51eb TH |
632 | __wsum csum; |
633 | __sum16 sum; | |
634 | ||
635 | csum = skb_checksum(skb, 0, skb->len, 0); | |
636 | ||
637 | /* skb->csum holds pseudo checksum */ | |
638 | sum = csum_fold(csum_add(skb->csum, csum)); | |
639 | if (likely(!sum)) { | |
640 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
641 | !skb->csum_complete_sw) | |
642 | netdev_rx_csum_fault(skb->dev); | |
643 | } | |
644 | ||
645 | /* Save full packet checksum */ | |
646 | skb->csum = csum; | |
647 | skb->ip_summed = CHECKSUM_COMPLETE; | |
648 | skb->csum_complete_sw = 1; | |
649 | skb->csum_valid = !sum; | |
650 | ||
651 | return sum; | |
759e5d00 | 652 | } |
fb286bb2 HX |
653 | EXPORT_SYMBOL(__skb_checksum_complete); |
654 | ||
1da177e4 | 655 | /** |
e5a4b0bb | 656 | * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. |
4dc3b16b PP |
657 | * @skb: skbuff |
658 | * @hlen: hardware length | |
e5a4b0bb | 659 | * @msg: destination |
4ec93edb | 660 | * |
1da177e4 LT |
661 | * Caller _must_ check that skb will fit to this iovec. |
662 | * | |
663 | * Returns: 0 - success. | |
664 | * -EINVAL - checksum failure. | |
e5a4b0bb | 665 | * -EFAULT - fault during copy. |
1da177e4 | 666 | */ |
e5a4b0bb AV |
667 | int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, |
668 | int hlen, struct msghdr *msg) | |
1da177e4 | 669 | { |
d3bc23e7 | 670 | __wsum csum; |
1da177e4 LT |
671 | int chunk = skb->len - hlen; |
672 | ||
ef8aef55 HX |
673 | if (!chunk) |
674 | return 0; | |
675 | ||
e5a4b0bb | 676 | if (iov_iter_count(&msg->msg_iter) < chunk) { |
fb286bb2 | 677 | if (__skb_checksum_complete(skb)) |
1da177e4 | 678 | goto csum_error; |
e5a4b0bb | 679 | if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) |
1da177e4 LT |
680 | goto fault; |
681 | } else { | |
682 | csum = csum_partial(skb->data, hlen, skb->csum); | |
e5a4b0bb | 683 | if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, |
1da177e4 LT |
684 | chunk, &csum)) |
685 | goto fault; | |
d3bc23e7 | 686 | if (csum_fold(csum)) |
1da177e4 | 687 | goto csum_error; |
84fa7933 | 688 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 689 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
690 | } |
691 | return 0; | |
692 | csum_error: | |
693 | return -EINVAL; | |
694 | fault: | |
695 | return -EFAULT; | |
696 | } | |
e5a4b0bb | 697 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); |
1da177e4 LT |
698 | |
699 | /** | |
700 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
701 | * @file: file struct |
702 | * @sock: socket | |
703 | * @wait: poll table | |
1da177e4 LT |
704 | * |
705 | * Datagram poll: Again totally generic. This also handles | |
706 | * sequenced packet sockets providing the socket receive queue | |
707 | * is only ever holding data ready to receive. | |
708 | * | |
709 | * Note: when you _don't_ use this routine for this protocol, | |
710 | * and you use a different write policy from sock_writeable() | |
711 | * then please supply your own write_space callback. | |
712 | */ | |
713 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
714 | poll_table *wait) | |
715 | { | |
716 | struct sock *sk = sock->sk; | |
717 | unsigned int mask; | |
718 | ||
aa395145 | 719 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
720 | mask = 0; |
721 | ||
722 | /* exceptional events? */ | |
723 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
7d4c04fc | 724 | mask |= POLLERR | |
8facd5fb | 725 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0); |
7d4c04fc | 726 | |
f348d70a | 727 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 728 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
729 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
730 | mask |= POLLHUP; | |
731 | ||
732 | /* readable? */ | |
db40980f | 733 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
734 | mask |= POLLIN | POLLRDNORM; |
735 | ||
736 | /* Connection-based need to check for termination and startup */ | |
737 | if (connection_based(sk)) { | |
738 | if (sk->sk_state == TCP_CLOSE) | |
739 | mask |= POLLHUP; | |
740 | /* connection hasn't started yet? */ | |
741 | if (sk->sk_state == TCP_SYN_SENT) | |
742 | return mask; | |
743 | } | |
744 | ||
745 | /* writable? */ | |
746 | if (sock_writeable(sk)) | |
747 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
748 | else | |
749 | set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
750 | ||
751 | return mask; | |
752 | } | |
1da177e4 | 753 | EXPORT_SYMBOL(datagram_poll); |