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> |
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 | ||
95c96174 | 68 | static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync, |
bf368e4e ED |
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 | 80 | /* |
39cc8613 | 81 | * Wait for the last received packet to be different from skb |
1da177e4 | 82 | */ |
39cc8613 BP |
83 | static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, |
84 | const struct sk_buff *skb) | |
1da177e4 LT |
85 | { |
86 | int error; | |
bf368e4e | 87 | DEFINE_WAIT_FUNC(wait, receiver_wake_function); |
1da177e4 | 88 | |
aa395145 | 89 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); |
1da177e4 LT |
90 | |
91 | /* Socket errors? */ | |
92 | error = sock_error(sk); | |
93 | if (error) | |
94 | goto out_err; | |
95 | ||
39cc8613 | 96 | if (sk->sk_receive_queue.prev != skb) |
1da177e4 LT |
97 | goto out; |
98 | ||
99 | /* Socket shut down? */ | |
100 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
101 | goto out_noerr; | |
102 | ||
103 | /* Sequenced packets can come disconnected. | |
104 | * If so we report the problem | |
105 | */ | |
106 | error = -ENOTCONN; | |
107 | if (connection_based(sk) && | |
108 | !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) | |
109 | goto out_err; | |
110 | ||
111 | /* handle signals */ | |
112 | if (signal_pending(current)) | |
113 | goto interrupted; | |
114 | ||
115 | error = 0; | |
116 | *timeo_p = schedule_timeout(*timeo_p); | |
117 | out: | |
aa395145 | 118 | finish_wait(sk_sleep(sk), &wait); |
1da177e4 LT |
119 | return error; |
120 | interrupted: | |
121 | error = sock_intr_errno(*timeo_p); | |
122 | out_err: | |
123 | *err = error; | |
124 | goto out; | |
125 | out_noerr: | |
126 | *err = 0; | |
127 | error = 1; | |
128 | goto out; | |
129 | } | |
130 | ||
131 | /** | |
a59322be | 132 | * __skb_recv_datagram - Receive a datagram skbuff |
4dc3b16b PP |
133 | * @sk: socket |
134 | * @flags: MSG_ flags | |
39cc8613 | 135 | * @peeked: returns non-zero if this packet has been seen before |
3f518bf7 PE |
136 | * @off: an offset in bytes to peek skb from. Returns an offset |
137 | * within an skb where data actually starts | |
4dc3b16b | 138 | * @err: error code returned |
1da177e4 LT |
139 | * |
140 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
141 | * and possible races. This replaces identical code in packet, raw and | |
142 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
143 | * the long standing peek and read race for datagram sockets. If you | |
144 | * alter this routine remember it must be re-entrant. | |
145 | * | |
146 | * This function will lock the socket if a skb is returned, so the caller | |
147 | * needs to unlock the socket in that case (usually by calling | |
148 | * skb_free_datagram) | |
149 | * | |
150 | * * It does not lock socket since today. This function is | |
151 | * * free of race conditions. This measure should/can improve | |
152 | * * significantly datagram socket latencies at high loads, | |
153 | * * when data copying to user space takes lots of time. | |
154 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
155 | * * 8) Great win.) | |
156 | * * --ANK (980729) | |
157 | * | |
158 | * The order of the tests when we find no data waiting are specified | |
159 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
160 | * the standard around please. | |
161 | */ | |
95c96174 | 162 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
3f518bf7 | 163 | int *peeked, int *off, int *err) |
1da177e4 | 164 | { |
39cc8613 | 165 | struct sk_buff *skb, *last; |
1da177e4 LT |
166 | long timeo; |
167 | /* | |
168 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
169 | */ | |
170 | int error = sock_error(sk); | |
171 | ||
172 | if (error) | |
173 | goto no_packet; | |
174 | ||
a59322be | 175 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
176 | |
177 | do { | |
178 | /* Again only user level code calls this function, so nothing | |
179 | * interrupt level will suddenly eat the receive_queue. | |
180 | * | |
181 | * Look at current nfs client by the way... | |
8917a3c0 | 182 | * However, this function was correct in any case. 8) |
1da177e4 | 183 | */ |
a59322be | 184 | unsigned long cpu_flags; |
4934b032 | 185 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
39cc8613 | 186 | int _off = *off; |
a59322be | 187 | |
39cc8613 | 188 | last = (struct sk_buff *)queue; |
4934b032 | 189 | spin_lock_irqsave(&queue->lock, cpu_flags); |
3f518bf7 | 190 | skb_queue_walk(queue, skb) { |
39cc8613 | 191 | last = skb; |
a59322be HX |
192 | *peeked = skb->peeked; |
193 | if (flags & MSG_PEEK) { | |
39cc8613 | 194 | if (_off >= skb->len && (skb->len || _off || |
add05ad4 | 195 | skb->peeked)) { |
39cc8613 | 196 | _off -= skb->len; |
3f518bf7 PE |
197 | continue; |
198 | } | |
a59322be | 199 | skb->peeked = 1; |
1da177e4 | 200 | atomic_inc(&skb->users); |
a59322be | 201 | } else |
4934b032 | 202 | __skb_unlink(skb, queue); |
1da177e4 | 203 | |
3f518bf7 | 204 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
39cc8613 | 205 | *off = _off; |
1da177e4 | 206 | return skb; |
3f518bf7 PE |
207 | } |
208 | spin_unlock_irqrestore(&queue->lock, cpu_flags); | |
1da177e4 LT |
209 | |
210 | /* User doesn't want to wait */ | |
211 | error = -EAGAIN; | |
212 | if (!timeo) | |
213 | goto no_packet; | |
214 | ||
39cc8613 | 215 | } while (!wait_for_more_packets(sk, err, &timeo, last)); |
1da177e4 LT |
216 | |
217 | return NULL; | |
218 | ||
219 | no_packet: | |
220 | *err = error; | |
221 | return NULL; | |
222 | } | |
a59322be HX |
223 | EXPORT_SYMBOL(__skb_recv_datagram); |
224 | ||
95c96174 | 225 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
226 | int noblock, int *err) |
227 | { | |
3f518bf7 | 228 | int peeked, off = 0; |
a59322be HX |
229 | |
230 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
3f518bf7 | 231 | &peeked, &off, err); |
a59322be | 232 | } |
9e34a5b5 | 233 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
234 | |
235 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
236 | { | |
ead2ceb0 | 237 | consume_skb(skb); |
270acefa | 238 | sk_mem_reclaim_partial(sk); |
1da177e4 | 239 | } |
9d410c79 ED |
240 | EXPORT_SYMBOL(skb_free_datagram); |
241 | ||
242 | void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) | |
243 | { | |
8a74ad60 ED |
244 | bool slow; |
245 | ||
93bb64ea ED |
246 | if (likely(atomic_read(&skb->users) == 1)) |
247 | smp_rmb(); | |
248 | else if (likely(!atomic_dec_and_test(&skb->users))) | |
249 | return; | |
250 | ||
8a74ad60 | 251 | slow = lock_sock_fast(sk); |
4b0b72f7 ED |
252 | skb_orphan(skb); |
253 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 254 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 255 | |
93bb64ea ED |
256 | /* skb is now orphaned, can be freed outside of locked section */ |
257 | __kfree_skb(skb); | |
9d410c79 ED |
258 | } |
259 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 260 | |
3305b80c HX |
261 | /** |
262 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
263 | * @sk: socket | |
264 | * @skb: datagram skbuff | |
265 | * @flags: MSG_ flags | |
266 | * | |
267 | * This function frees a datagram skbuff that was received by | |
268 | * skb_recv_datagram. The flags argument must match the one | |
269 | * used for skb_recv_datagram. | |
270 | * | |
271 | * If the MSG_PEEK flag is set, and the packet is still on the | |
272 | * receive queue of the socket, it will be taken off the queue | |
273 | * before it is freed. | |
274 | * | |
275 | * This function currently only disables BH when acquiring the | |
276 | * sk_receive_queue lock. Therefore it must not be used in a | |
277 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
278 | * |
279 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
280 | */ |
281 | ||
27ab2568 | 282 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 283 | { |
27ab2568 HX |
284 | int err = 0; |
285 | ||
3305b80c | 286 | if (flags & MSG_PEEK) { |
27ab2568 | 287 | err = -ENOENT; |
3305b80c HX |
288 | spin_lock_bh(&sk->sk_receive_queue.lock); |
289 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
290 | __skb_unlink(skb, &sk->sk_receive_queue); | |
291 | atomic_dec(&skb->users); | |
27ab2568 | 292 | err = 0; |
3305b80c HX |
293 | } |
294 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
295 | } | |
296 | ||
61de71c6 | 297 | kfree_skb(skb); |
8edf19c2 | 298 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
299 | sk_mem_reclaim_partial(sk); |
300 | ||
27ab2568 | 301 | return err; |
3305b80c | 302 | } |
3305b80c HX |
303 | EXPORT_SYMBOL(skb_kill_datagram); |
304 | ||
1da177e4 LT |
305 | /** |
306 | * skb_copy_datagram_iovec - Copy a datagram to an iovec. | |
4dc3b16b PP |
307 | * @skb: buffer to copy |
308 | * @offset: offset in the buffer to start copying from | |
67be2dd1 | 309 | * @to: io vector to copy to |
4dc3b16b | 310 | * @len: amount of data to copy from buffer to iovec |
1da177e4 LT |
311 | * |
312 | * Note: the iovec is modified during the copy. | |
313 | */ | |
314 | int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset, | |
315 | struct iovec *to, int len) | |
316 | { | |
1a028e50 DM |
317 | int start = skb_headlen(skb); |
318 | int i, copy = start - offset; | |
5b1a002a | 319 | struct sk_buff *frag_iter; |
c75d721c | 320 | |
e9b3cc1b NH |
321 | trace_skb_copy_datagram_iovec(skb, len); |
322 | ||
b4d9eda0 DM |
323 | /* Copy header. */ |
324 | if (copy > 0) { | |
325 | if (copy > len) | |
326 | copy = len; | |
327 | if (memcpy_toiovec(to, skb->data + offset, copy)) | |
328 | goto fault; | |
329 | if ((len -= copy) == 0) | |
330 | return 0; | |
331 | offset += copy; | |
332 | } | |
c75d721c | 333 | |
b4d9eda0 DM |
334 | /* Copy paged appendix. Hmm... why does this look so complicated? */ |
335 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 336 | int end; |
9e903e08 | 337 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 338 | |
547b792c | 339 | WARN_ON(start > offset + len); |
1a028e50 | 340 | |
9e903e08 | 341 | end = start + skb_frag_size(frag); |
b4d9eda0 DM |
342 | if ((copy = end - offset) > 0) { |
343 | int err; | |
344 | u8 *vaddr; | |
ea2ab693 | 345 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
346 | |
347 | if (copy > len) | |
348 | copy = len; | |
b4d9eda0 | 349 | vaddr = kmap(page); |
1a028e50 DM |
350 | err = memcpy_toiovec(to, vaddr + frag->page_offset + |
351 | offset - start, copy); | |
b4d9eda0 | 352 | kunmap(page); |
1da177e4 LT |
353 | if (err) |
354 | goto fault; | |
355 | if (!(len -= copy)) | |
356 | return 0; | |
357 | offset += copy; | |
358 | } | |
1a028e50 | 359 | start = end; |
1da177e4 | 360 | } |
b4d9eda0 | 361 | |
5b1a002a DM |
362 | skb_walk_frags(skb, frag_iter) { |
363 | int end; | |
364 | ||
365 | WARN_ON(start > offset + len); | |
366 | ||
367 | end = start + frag_iter->len; | |
368 | if ((copy = end - offset) > 0) { | |
369 | if (copy > len) | |
370 | copy = len; | |
371 | if (skb_copy_datagram_iovec(frag_iter, | |
372 | offset - start, | |
373 | to, copy)) | |
374 | goto fault; | |
375 | if ((len -= copy) == 0) | |
376 | return 0; | |
377 | offset += copy; | |
b4d9eda0 | 378 | } |
5b1a002a | 379 | start = end; |
1da177e4 | 380 | } |
b4d9eda0 DM |
381 | if (!len) |
382 | return 0; | |
383 | ||
1da177e4 LT |
384 | fault: |
385 | return -EFAULT; | |
386 | } | |
9e34a5b5 | 387 | EXPORT_SYMBOL(skb_copy_datagram_iovec); |
1da177e4 | 388 | |
0a1ec07a MT |
389 | /** |
390 | * skb_copy_datagram_const_iovec - Copy a datagram to an iovec. | |
391 | * @skb: buffer to copy | |
392 | * @offset: offset in the buffer to start copying from | |
393 | * @to: io vector to copy to | |
394 | * @to_offset: offset in the io vector to start copying to | |
395 | * @len: amount of data to copy from buffer to iovec | |
396 | * | |
397 | * Returns 0 or -EFAULT. | |
398 | * Note: the iovec is not modified during the copy. | |
399 | */ | |
400 | int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset, | |
401 | const struct iovec *to, int to_offset, | |
402 | int len) | |
403 | { | |
404 | int start = skb_headlen(skb); | |
405 | int i, copy = start - offset; | |
5b1a002a | 406 | struct sk_buff *frag_iter; |
0a1ec07a MT |
407 | |
408 | /* Copy header. */ | |
409 | if (copy > 0) { | |
410 | if (copy > len) | |
411 | copy = len; | |
412 | if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy)) | |
413 | goto fault; | |
414 | if ((len -= copy) == 0) | |
415 | return 0; | |
416 | offset += copy; | |
417 | to_offset += copy; | |
418 | } | |
419 | ||
420 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
421 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
422 | int end; | |
9e903e08 | 423 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
0a1ec07a MT |
424 | |
425 | WARN_ON(start > offset + len); | |
426 | ||
9e903e08 | 427 | end = start + skb_frag_size(frag); |
0a1ec07a MT |
428 | if ((copy = end - offset) > 0) { |
429 | int err; | |
430 | u8 *vaddr; | |
ea2ab693 | 431 | struct page *page = skb_frag_page(frag); |
0a1ec07a MT |
432 | |
433 | if (copy > len) | |
434 | copy = len; | |
435 | vaddr = kmap(page); | |
436 | err = memcpy_toiovecend(to, vaddr + frag->page_offset + | |
437 | offset - start, to_offset, copy); | |
438 | kunmap(page); | |
439 | if (err) | |
440 | goto fault; | |
441 | if (!(len -= copy)) | |
442 | return 0; | |
443 | offset += copy; | |
444 | to_offset += copy; | |
445 | } | |
446 | start = end; | |
447 | } | |
448 | ||
5b1a002a DM |
449 | skb_walk_frags(skb, frag_iter) { |
450 | int end; | |
451 | ||
452 | WARN_ON(start > offset + len); | |
453 | ||
454 | end = start + frag_iter->len; | |
455 | if ((copy = end - offset) > 0) { | |
456 | if (copy > len) | |
457 | copy = len; | |
458 | if (skb_copy_datagram_const_iovec(frag_iter, | |
459 | offset - start, | |
460 | to, to_offset, | |
461 | copy)) | |
462 | goto fault; | |
463 | if ((len -= copy) == 0) | |
464 | return 0; | |
465 | offset += copy; | |
466 | to_offset += copy; | |
0a1ec07a | 467 | } |
5b1a002a | 468 | start = end; |
0a1ec07a MT |
469 | } |
470 | if (!len) | |
471 | return 0; | |
472 | ||
473 | fault: | |
474 | return -EFAULT; | |
475 | } | |
476 | EXPORT_SYMBOL(skb_copy_datagram_const_iovec); | |
477 | ||
db543c1f RR |
478 | /** |
479 | * skb_copy_datagram_from_iovec - Copy a datagram from an iovec. | |
480 | * @skb: buffer to copy | |
481 | * @offset: offset in the buffer to start copying to | |
482 | * @from: io vector to copy to | |
6f26c9a7 | 483 | * @from_offset: offset in the io vector to start copying from |
db543c1f RR |
484 | * @len: amount of data to copy to buffer from iovec |
485 | * | |
486 | * Returns 0 or -EFAULT. | |
6f26c9a7 | 487 | * Note: the iovec is not modified during the copy. |
db543c1f RR |
488 | */ |
489 | int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset, | |
6f26c9a7 MT |
490 | const struct iovec *from, int from_offset, |
491 | int len) | |
db543c1f RR |
492 | { |
493 | int start = skb_headlen(skb); | |
494 | int i, copy = start - offset; | |
5b1a002a | 495 | struct sk_buff *frag_iter; |
db543c1f RR |
496 | |
497 | /* Copy header. */ | |
498 | if (copy > 0) { | |
499 | if (copy > len) | |
500 | copy = len; | |
d2d27bfd SS |
501 | if (memcpy_fromiovecend(skb->data + offset, from, from_offset, |
502 | copy)) | |
db543c1f RR |
503 | goto fault; |
504 | if ((len -= copy) == 0) | |
505 | return 0; | |
506 | offset += copy; | |
6f26c9a7 | 507 | from_offset += copy; |
db543c1f RR |
508 | } |
509 | ||
510 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
511 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
512 | int end; | |
9e903e08 | 513 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
db543c1f RR |
514 | |
515 | WARN_ON(start > offset + len); | |
516 | ||
9e903e08 | 517 | end = start + skb_frag_size(frag); |
db543c1f RR |
518 | if ((copy = end - offset) > 0) { |
519 | int err; | |
520 | u8 *vaddr; | |
ea2ab693 | 521 | struct page *page = skb_frag_page(frag); |
db543c1f RR |
522 | |
523 | if (copy > len) | |
524 | copy = len; | |
525 | vaddr = kmap(page); | |
6f26c9a7 MT |
526 | err = memcpy_fromiovecend(vaddr + frag->page_offset + |
527 | offset - start, | |
528 | from, from_offset, copy); | |
db543c1f RR |
529 | kunmap(page); |
530 | if (err) | |
531 | goto fault; | |
532 | ||
533 | if (!(len -= copy)) | |
534 | return 0; | |
535 | offset += copy; | |
6f26c9a7 | 536 | from_offset += copy; |
db543c1f RR |
537 | } |
538 | start = end; | |
539 | } | |
540 | ||
5b1a002a DM |
541 | skb_walk_frags(skb, frag_iter) { |
542 | int end; | |
543 | ||
544 | WARN_ON(start > offset + len); | |
545 | ||
546 | end = start + frag_iter->len; | |
547 | if ((copy = end - offset) > 0) { | |
548 | if (copy > len) | |
549 | copy = len; | |
550 | if (skb_copy_datagram_from_iovec(frag_iter, | |
551 | offset - start, | |
552 | from, | |
553 | from_offset, | |
554 | copy)) | |
555 | goto fault; | |
556 | if ((len -= copy) == 0) | |
557 | return 0; | |
558 | offset += copy; | |
559 | from_offset += copy; | |
db543c1f | 560 | } |
5b1a002a | 561 | start = end; |
db543c1f RR |
562 | } |
563 | if (!len) | |
564 | return 0; | |
565 | ||
566 | fault: | |
567 | return -EFAULT; | |
568 | } | |
569 | EXPORT_SYMBOL(skb_copy_datagram_from_iovec); | |
570 | ||
1da177e4 LT |
571 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
572 | u8 __user *to, int len, | |
5084205f | 573 | __wsum *csump) |
1da177e4 | 574 | { |
1a028e50 | 575 | int start = skb_headlen(skb); |
1a028e50 | 576 | int i, copy = start - offset; |
5b1a002a DM |
577 | struct sk_buff *frag_iter; |
578 | int pos = 0; | |
1da177e4 LT |
579 | |
580 | /* Copy header. */ | |
581 | if (copy > 0) { | |
582 | int err = 0; | |
583 | if (copy > len) | |
584 | copy = len; | |
585 | *csump = csum_and_copy_to_user(skb->data + offset, to, copy, | |
586 | *csump, &err); | |
587 | if (err) | |
588 | goto fault; | |
589 | if ((len -= copy) == 0) | |
590 | return 0; | |
591 | offset += copy; | |
592 | to += copy; | |
593 | pos = copy; | |
594 | } | |
595 | ||
596 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 597 | int end; |
9e903e08 | 598 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 599 | |
547b792c | 600 | WARN_ON(start > offset + len); |
1a028e50 | 601 | |
9e903e08 | 602 | end = start + skb_frag_size(frag); |
1da177e4 | 603 | if ((copy = end - offset) > 0) { |
5084205f | 604 | __wsum csum2; |
1da177e4 LT |
605 | int err = 0; |
606 | u8 *vaddr; | |
ea2ab693 | 607 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
608 | |
609 | if (copy > len) | |
610 | copy = len; | |
611 | vaddr = kmap(page); | |
612 | csum2 = csum_and_copy_to_user(vaddr + | |
1a028e50 DM |
613 | frag->page_offset + |
614 | offset - start, | |
1da177e4 LT |
615 | to, copy, 0, &err); |
616 | kunmap(page); | |
617 | if (err) | |
618 | goto fault; | |
619 | *csump = csum_block_add(*csump, csum2, pos); | |
620 | if (!(len -= copy)) | |
621 | return 0; | |
622 | offset += copy; | |
623 | to += copy; | |
624 | pos += copy; | |
625 | } | |
1a028e50 | 626 | start = end; |
1da177e4 LT |
627 | } |
628 | ||
5b1a002a DM |
629 | skb_walk_frags(skb, frag_iter) { |
630 | int end; | |
631 | ||
632 | WARN_ON(start > offset + len); | |
633 | ||
634 | end = start + frag_iter->len; | |
635 | if ((copy = end - offset) > 0) { | |
636 | __wsum csum2 = 0; | |
637 | if (copy > len) | |
638 | copy = len; | |
639 | if (skb_copy_and_csum_datagram(frag_iter, | |
640 | offset - start, | |
641 | to, copy, | |
642 | &csum2)) | |
643 | goto fault; | |
644 | *csump = csum_block_add(*csump, csum2, pos); | |
645 | if ((len -= copy) == 0) | |
646 | return 0; | |
647 | offset += copy; | |
648 | to += copy; | |
649 | pos += copy; | |
1da177e4 | 650 | } |
5b1a002a | 651 | start = end; |
1da177e4 LT |
652 | } |
653 | if (!len) | |
654 | return 0; | |
655 | ||
656 | fault: | |
657 | return -EFAULT; | |
658 | } | |
659 | ||
759e5d00 | 660 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 661 | { |
d3bc23e7 | 662 | __sum16 sum; |
fb286bb2 | 663 | |
759e5d00 | 664 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
fb286bb2 | 665 | if (likely(!sum)) { |
84fa7933 | 666 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 HX |
667 | netdev_rx_csum_fault(skb->dev); |
668 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
669 | } | |
670 | return sum; | |
671 | } | |
759e5d00 HX |
672 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
673 | ||
674 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
675 | { | |
676 | return __skb_checksum_complete_head(skb, skb->len); | |
677 | } | |
fb286bb2 HX |
678 | EXPORT_SYMBOL(__skb_checksum_complete); |
679 | ||
1da177e4 LT |
680 | /** |
681 | * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec. | |
4dc3b16b PP |
682 | * @skb: skbuff |
683 | * @hlen: hardware length | |
67be2dd1 | 684 | * @iov: io vector |
4ec93edb | 685 | * |
1da177e4 LT |
686 | * Caller _must_ check that skb will fit to this iovec. |
687 | * | |
688 | * Returns: 0 - success. | |
689 | * -EINVAL - checksum failure. | |
690 | * -EFAULT - fault during copy. Beware, in this case iovec | |
691 | * can be modified! | |
692 | */ | |
fb286bb2 | 693 | int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, |
1da177e4 LT |
694 | int hlen, struct iovec *iov) |
695 | { | |
d3bc23e7 | 696 | __wsum csum; |
1da177e4 LT |
697 | int chunk = skb->len - hlen; |
698 | ||
ef8aef55 HX |
699 | if (!chunk) |
700 | return 0; | |
701 | ||
1da177e4 LT |
702 | /* Skip filled elements. |
703 | * Pretty silly, look at memcpy_toiovec, though 8) | |
704 | */ | |
705 | while (!iov->iov_len) | |
706 | iov++; | |
707 | ||
708 | if (iov->iov_len < chunk) { | |
fb286bb2 | 709 | if (__skb_checksum_complete(skb)) |
1da177e4 LT |
710 | goto csum_error; |
711 | if (skb_copy_datagram_iovec(skb, hlen, iov, chunk)) | |
712 | goto fault; | |
713 | } else { | |
714 | csum = csum_partial(skb->data, hlen, skb->csum); | |
715 | if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base, | |
716 | chunk, &csum)) | |
717 | goto fault; | |
d3bc23e7 | 718 | if (csum_fold(csum)) |
1da177e4 | 719 | goto csum_error; |
84fa7933 | 720 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 721 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
722 | iov->iov_len -= chunk; |
723 | iov->iov_base += chunk; | |
724 | } | |
725 | return 0; | |
726 | csum_error: | |
727 | return -EINVAL; | |
728 | fault: | |
729 | return -EFAULT; | |
730 | } | |
9e34a5b5 | 731 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec); |
1da177e4 LT |
732 | |
733 | /** | |
734 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
735 | * @file: file struct |
736 | * @sock: socket | |
737 | * @wait: poll table | |
1da177e4 LT |
738 | * |
739 | * Datagram poll: Again totally generic. This also handles | |
740 | * sequenced packet sockets providing the socket receive queue | |
741 | * is only ever holding data ready to receive. | |
742 | * | |
743 | * Note: when you _don't_ use this routine for this protocol, | |
744 | * and you use a different write policy from sock_writeable() | |
745 | * then please supply your own write_space callback. | |
746 | */ | |
747 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
748 | poll_table *wait) | |
749 | { | |
750 | struct sock *sk = sock->sk; | |
751 | unsigned int mask; | |
752 | ||
aa395145 | 753 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
754 | mask = 0; |
755 | ||
756 | /* exceptional events? */ | |
757 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
7d4c04fc | 758 | mask |= POLLERR | |
8facd5fb | 759 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0); |
7d4c04fc | 760 | |
f348d70a | 761 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 762 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
763 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
764 | mask |= POLLHUP; | |
765 | ||
766 | /* readable? */ | |
db40980f | 767 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
768 | mask |= POLLIN | POLLRDNORM; |
769 | ||
770 | /* Connection-based need to check for termination and startup */ | |
771 | if (connection_based(sk)) { | |
772 | if (sk->sk_state == TCP_CLOSE) | |
773 | mask |= POLLHUP; | |
774 | /* connection hasn't started yet? */ | |
775 | if (sk->sk_state == TCP_SYN_SENT) | |
776 | return mask; | |
777 | } | |
778 | ||
779 | /* writable? */ | |
780 | if (sock_writeable(sk)) | |
781 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
782 | else | |
783 | set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
784 | ||
785 | return mask; | |
786 | } | |
1da177e4 | 787 | EXPORT_SYMBOL(datagram_poll); |