]>
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> |
5a0e3ad6 | 51 | #include <linux/slab.h> |
1da177e4 LT |
52 | |
53 | #include <net/protocol.h> | |
54 | #include <linux/skbuff.h> | |
1da177e4 | 55 | |
c752f073 ACM |
56 | #include <net/checksum.h> |
57 | #include <net/sock.h> | |
58 | #include <net/tcp_states.h> | |
e9b3cc1b | 59 | #include <trace/events/skb.h> |
1da177e4 LT |
60 | |
61 | /* | |
62 | * Is a socket 'connection oriented' ? | |
63 | */ | |
64 | static inline int connection_based(struct sock *sk) | |
65 | { | |
66 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
67 | } | |
68 | ||
bf368e4e ED |
69 | static int receiver_wake_function(wait_queue_t *wait, unsigned mode, int sync, |
70 | void *key) | |
71 | { | |
72 | unsigned long bits = (unsigned long)key; | |
73 | ||
74 | /* | |
75 | * Avoid a wakeup if event not interesting for us | |
76 | */ | |
77 | if (bits && !(bits & (POLLIN | POLLERR))) | |
78 | return 0; | |
79 | return autoremove_wake_function(wait, mode, sync, key); | |
80 | } | |
1da177e4 LT |
81 | /* |
82 | * Wait for a packet.. | |
83 | */ | |
84 | static int wait_for_packet(struct sock *sk, int *err, long *timeo_p) | |
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 | ||
96 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
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 | |
a59322be | 135 | * @peeked: returns non-zero if this packet has been seen before |
4dc3b16b | 136 | * @err: error code returned |
1da177e4 LT |
137 | * |
138 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
139 | * and possible races. This replaces identical code in packet, raw and | |
140 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
141 | * the long standing peek and read race for datagram sockets. If you | |
142 | * alter this routine remember it must be re-entrant. | |
143 | * | |
144 | * This function will lock the socket if a skb is returned, so the caller | |
145 | * needs to unlock the socket in that case (usually by calling | |
146 | * skb_free_datagram) | |
147 | * | |
148 | * * It does not lock socket since today. This function is | |
149 | * * free of race conditions. This measure should/can improve | |
150 | * * significantly datagram socket latencies at high loads, | |
151 | * * when data copying to user space takes lots of time. | |
152 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
153 | * * 8) Great win.) | |
154 | * * --ANK (980729) | |
155 | * | |
156 | * The order of the tests when we find no data waiting are specified | |
157 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
158 | * the standard around please. | |
159 | */ | |
a59322be HX |
160 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, |
161 | int *peeked, int *err) | |
1da177e4 LT |
162 | { |
163 | struct sk_buff *skb; | |
164 | long timeo; | |
165 | /* | |
166 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
167 | */ | |
168 | int error = sock_error(sk); | |
169 | ||
170 | if (error) | |
171 | goto no_packet; | |
172 | ||
a59322be | 173 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
174 | |
175 | do { | |
176 | /* Again only user level code calls this function, so nothing | |
177 | * interrupt level will suddenly eat the receive_queue. | |
178 | * | |
179 | * Look at current nfs client by the way... | |
8917a3c0 | 180 | * However, this function was correct in any case. 8) |
1da177e4 | 181 | */ |
a59322be | 182 | unsigned long cpu_flags; |
4934b032 | 183 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
a59322be | 184 | |
4934b032 PE |
185 | spin_lock_irqsave(&queue->lock, cpu_flags); |
186 | skb = skb_peek(queue); | |
a59322be HX |
187 | if (skb) { |
188 | *peeked = skb->peeked; | |
189 | if (flags & MSG_PEEK) { | |
190 | skb->peeked = 1; | |
1da177e4 | 191 | atomic_inc(&skb->users); |
a59322be | 192 | } else |
4934b032 | 193 | __skb_unlink(skb, queue); |
a59322be | 194 | } |
4934b032 | 195 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
1da177e4 LT |
196 | |
197 | if (skb) | |
198 | return skb; | |
199 | ||
200 | /* User doesn't want to wait */ | |
201 | error = -EAGAIN; | |
202 | if (!timeo) | |
203 | goto no_packet; | |
204 | ||
205 | } while (!wait_for_packet(sk, err, &timeo)); | |
206 | ||
207 | return NULL; | |
208 | ||
209 | no_packet: | |
210 | *err = error; | |
211 | return NULL; | |
212 | } | |
a59322be HX |
213 | EXPORT_SYMBOL(__skb_recv_datagram); |
214 | ||
215 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, | |
216 | int noblock, int *err) | |
217 | { | |
218 | int peeked; | |
219 | ||
220 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
221 | &peeked, err); | |
222 | } | |
9e34a5b5 | 223 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
224 | |
225 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
226 | { | |
ead2ceb0 | 227 | consume_skb(skb); |
270acefa | 228 | sk_mem_reclaim_partial(sk); |
1da177e4 | 229 | } |
9d410c79 ED |
230 | EXPORT_SYMBOL(skb_free_datagram); |
231 | ||
232 | void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) | |
233 | { | |
8a74ad60 ED |
234 | bool slow; |
235 | ||
93bb64ea ED |
236 | if (likely(atomic_read(&skb->users) == 1)) |
237 | smp_rmb(); | |
238 | else if (likely(!atomic_dec_and_test(&skb->users))) | |
239 | return; | |
240 | ||
8a74ad60 | 241 | slow = lock_sock_fast(sk); |
4b0b72f7 ED |
242 | skb_orphan(skb); |
243 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 244 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 245 | |
93bb64ea | 246 | /* skb is now orphaned, can be freed outside of locked section */ |
07dc22e7 | 247 | trace_kfree_skb(skb, skb_free_datagram_locked); |
93bb64ea | 248 | __kfree_skb(skb); |
9d410c79 ED |
249 | } |
250 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 251 | |
3305b80c HX |
252 | /** |
253 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
254 | * @sk: socket | |
255 | * @skb: datagram skbuff | |
256 | * @flags: MSG_ flags | |
257 | * | |
258 | * This function frees a datagram skbuff that was received by | |
259 | * skb_recv_datagram. The flags argument must match the one | |
260 | * used for skb_recv_datagram. | |
261 | * | |
262 | * If the MSG_PEEK flag is set, and the packet is still on the | |
263 | * receive queue of the socket, it will be taken off the queue | |
264 | * before it is freed. | |
265 | * | |
266 | * This function currently only disables BH when acquiring the | |
267 | * sk_receive_queue lock. Therefore it must not be used in a | |
268 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
269 | * |
270 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
271 | */ |
272 | ||
27ab2568 | 273 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 274 | { |
27ab2568 HX |
275 | int err = 0; |
276 | ||
3305b80c | 277 | if (flags & MSG_PEEK) { |
27ab2568 | 278 | err = -ENOENT; |
3305b80c HX |
279 | spin_lock_bh(&sk->sk_receive_queue.lock); |
280 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
281 | __skb_unlink(skb, &sk->sk_receive_queue); | |
282 | atomic_dec(&skb->users); | |
27ab2568 | 283 | err = 0; |
3305b80c HX |
284 | } |
285 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
286 | } | |
287 | ||
61de71c6 | 288 | kfree_skb(skb); |
8edf19c2 | 289 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
290 | sk_mem_reclaim_partial(sk); |
291 | ||
27ab2568 | 292 | return err; |
3305b80c | 293 | } |
3305b80c HX |
294 | EXPORT_SYMBOL(skb_kill_datagram); |
295 | ||
1da177e4 LT |
296 | /** |
297 | * skb_copy_datagram_iovec - Copy a datagram to an iovec. | |
4dc3b16b PP |
298 | * @skb: buffer to copy |
299 | * @offset: offset in the buffer to start copying from | |
67be2dd1 | 300 | * @to: io vector to copy to |
4dc3b16b | 301 | * @len: amount of data to copy from buffer to iovec |
1da177e4 LT |
302 | * |
303 | * Note: the iovec is modified during the copy. | |
304 | */ | |
305 | int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset, | |
306 | struct iovec *to, int len) | |
307 | { | |
1a028e50 DM |
308 | int start = skb_headlen(skb); |
309 | int i, copy = start - offset; | |
5b1a002a | 310 | struct sk_buff *frag_iter; |
c75d721c | 311 | |
e9b3cc1b NH |
312 | trace_skb_copy_datagram_iovec(skb, len); |
313 | ||
b4d9eda0 DM |
314 | /* Copy header. */ |
315 | if (copy > 0) { | |
316 | if (copy > len) | |
317 | copy = len; | |
318 | if (memcpy_toiovec(to, skb->data + offset, copy)) | |
319 | goto fault; | |
320 | if ((len -= copy) == 0) | |
321 | return 0; | |
322 | offset += copy; | |
323 | } | |
c75d721c | 324 | |
b4d9eda0 DM |
325 | /* Copy paged appendix. Hmm... why does this look so complicated? */ |
326 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 327 | int end; |
9e903e08 | 328 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 329 | |
547b792c | 330 | WARN_ON(start > offset + len); |
1a028e50 | 331 | |
9e903e08 | 332 | end = start + skb_frag_size(frag); |
b4d9eda0 DM |
333 | if ((copy = end - offset) > 0) { |
334 | int err; | |
335 | u8 *vaddr; | |
ea2ab693 | 336 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
337 | |
338 | if (copy > len) | |
339 | copy = len; | |
b4d9eda0 | 340 | vaddr = kmap(page); |
1a028e50 DM |
341 | err = memcpy_toiovec(to, vaddr + frag->page_offset + |
342 | offset - start, copy); | |
b4d9eda0 | 343 | kunmap(page); |
1da177e4 LT |
344 | if (err) |
345 | goto fault; | |
346 | if (!(len -= copy)) | |
347 | return 0; | |
348 | offset += copy; | |
349 | } | |
1a028e50 | 350 | start = end; |
1da177e4 | 351 | } |
b4d9eda0 | 352 | |
5b1a002a DM |
353 | skb_walk_frags(skb, frag_iter) { |
354 | int end; | |
355 | ||
356 | WARN_ON(start > offset + len); | |
357 | ||
358 | end = start + frag_iter->len; | |
359 | if ((copy = end - offset) > 0) { | |
360 | if (copy > len) | |
361 | copy = len; | |
362 | if (skb_copy_datagram_iovec(frag_iter, | |
363 | offset - start, | |
364 | to, copy)) | |
365 | goto fault; | |
366 | if ((len -= copy) == 0) | |
367 | return 0; | |
368 | offset += copy; | |
b4d9eda0 | 369 | } |
5b1a002a | 370 | start = end; |
1da177e4 | 371 | } |
b4d9eda0 DM |
372 | if (!len) |
373 | return 0; | |
374 | ||
1da177e4 LT |
375 | fault: |
376 | return -EFAULT; | |
377 | } | |
9e34a5b5 | 378 | EXPORT_SYMBOL(skb_copy_datagram_iovec); |
1da177e4 | 379 | |
0a1ec07a MT |
380 | /** |
381 | * skb_copy_datagram_const_iovec - Copy a datagram to an iovec. | |
382 | * @skb: buffer to copy | |
383 | * @offset: offset in the buffer to start copying from | |
384 | * @to: io vector to copy to | |
385 | * @to_offset: offset in the io vector to start copying to | |
386 | * @len: amount of data to copy from buffer to iovec | |
387 | * | |
388 | * Returns 0 or -EFAULT. | |
389 | * Note: the iovec is not modified during the copy. | |
390 | */ | |
391 | int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset, | |
392 | const struct iovec *to, int to_offset, | |
393 | int len) | |
394 | { | |
395 | int start = skb_headlen(skb); | |
396 | int i, copy = start - offset; | |
5b1a002a | 397 | struct sk_buff *frag_iter; |
0a1ec07a MT |
398 | |
399 | /* Copy header. */ | |
400 | if (copy > 0) { | |
401 | if (copy > len) | |
402 | copy = len; | |
403 | if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy)) | |
404 | goto fault; | |
405 | if ((len -= copy) == 0) | |
406 | return 0; | |
407 | offset += copy; | |
408 | to_offset += copy; | |
409 | } | |
410 | ||
411 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
412 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
413 | int end; | |
9e903e08 | 414 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
0a1ec07a MT |
415 | |
416 | WARN_ON(start > offset + len); | |
417 | ||
9e903e08 | 418 | end = start + skb_frag_size(frag); |
0a1ec07a MT |
419 | if ((copy = end - offset) > 0) { |
420 | int err; | |
421 | u8 *vaddr; | |
ea2ab693 | 422 | struct page *page = skb_frag_page(frag); |
0a1ec07a MT |
423 | |
424 | if (copy > len) | |
425 | copy = len; | |
426 | vaddr = kmap(page); | |
427 | err = memcpy_toiovecend(to, vaddr + frag->page_offset + | |
428 | offset - start, to_offset, copy); | |
429 | kunmap(page); | |
430 | if (err) | |
431 | goto fault; | |
432 | if (!(len -= copy)) | |
433 | return 0; | |
434 | offset += copy; | |
435 | to_offset += copy; | |
436 | } | |
437 | start = end; | |
438 | } | |
439 | ||
5b1a002a DM |
440 | skb_walk_frags(skb, frag_iter) { |
441 | int end; | |
442 | ||
443 | WARN_ON(start > offset + len); | |
444 | ||
445 | end = start + frag_iter->len; | |
446 | if ((copy = end - offset) > 0) { | |
447 | if (copy > len) | |
448 | copy = len; | |
449 | if (skb_copy_datagram_const_iovec(frag_iter, | |
450 | offset - start, | |
451 | to, to_offset, | |
452 | copy)) | |
453 | goto fault; | |
454 | if ((len -= copy) == 0) | |
455 | return 0; | |
456 | offset += copy; | |
457 | to_offset += copy; | |
0a1ec07a | 458 | } |
5b1a002a | 459 | start = end; |
0a1ec07a MT |
460 | } |
461 | if (!len) | |
462 | return 0; | |
463 | ||
464 | fault: | |
465 | return -EFAULT; | |
466 | } | |
467 | EXPORT_SYMBOL(skb_copy_datagram_const_iovec); | |
468 | ||
db543c1f RR |
469 | /** |
470 | * skb_copy_datagram_from_iovec - Copy a datagram from an iovec. | |
471 | * @skb: buffer to copy | |
472 | * @offset: offset in the buffer to start copying to | |
473 | * @from: io vector to copy to | |
6f26c9a7 | 474 | * @from_offset: offset in the io vector to start copying from |
db543c1f RR |
475 | * @len: amount of data to copy to buffer from iovec |
476 | * | |
477 | * Returns 0 or -EFAULT. | |
6f26c9a7 | 478 | * Note: the iovec is not modified during the copy. |
db543c1f RR |
479 | */ |
480 | int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset, | |
6f26c9a7 MT |
481 | const struct iovec *from, int from_offset, |
482 | int len) | |
db543c1f RR |
483 | { |
484 | int start = skb_headlen(skb); | |
485 | int i, copy = start - offset; | |
5b1a002a | 486 | struct sk_buff *frag_iter; |
db543c1f RR |
487 | |
488 | /* Copy header. */ | |
489 | if (copy > 0) { | |
490 | if (copy > len) | |
491 | copy = len; | |
d2d27bfd SS |
492 | if (memcpy_fromiovecend(skb->data + offset, from, from_offset, |
493 | copy)) | |
db543c1f RR |
494 | goto fault; |
495 | if ((len -= copy) == 0) | |
496 | return 0; | |
497 | offset += copy; | |
6f26c9a7 | 498 | from_offset += copy; |
db543c1f RR |
499 | } |
500 | ||
501 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
502 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
503 | int end; | |
9e903e08 | 504 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
db543c1f RR |
505 | |
506 | WARN_ON(start > offset + len); | |
507 | ||
9e903e08 | 508 | end = start + skb_frag_size(frag); |
db543c1f RR |
509 | if ((copy = end - offset) > 0) { |
510 | int err; | |
511 | u8 *vaddr; | |
ea2ab693 | 512 | struct page *page = skb_frag_page(frag); |
db543c1f RR |
513 | |
514 | if (copy > len) | |
515 | copy = len; | |
516 | vaddr = kmap(page); | |
6f26c9a7 MT |
517 | err = memcpy_fromiovecend(vaddr + frag->page_offset + |
518 | offset - start, | |
519 | from, from_offset, copy); | |
db543c1f RR |
520 | kunmap(page); |
521 | if (err) | |
522 | goto fault; | |
523 | ||
524 | if (!(len -= copy)) | |
525 | return 0; | |
526 | offset += copy; | |
6f26c9a7 | 527 | from_offset += copy; |
db543c1f RR |
528 | } |
529 | start = end; | |
530 | } | |
531 | ||
5b1a002a DM |
532 | skb_walk_frags(skb, frag_iter) { |
533 | int end; | |
534 | ||
535 | WARN_ON(start > offset + len); | |
536 | ||
537 | end = start + frag_iter->len; | |
538 | if ((copy = end - offset) > 0) { | |
539 | if (copy > len) | |
540 | copy = len; | |
541 | if (skb_copy_datagram_from_iovec(frag_iter, | |
542 | offset - start, | |
543 | from, | |
544 | from_offset, | |
545 | copy)) | |
546 | goto fault; | |
547 | if ((len -= copy) == 0) | |
548 | return 0; | |
549 | offset += copy; | |
550 | from_offset += copy; | |
db543c1f | 551 | } |
5b1a002a | 552 | start = end; |
db543c1f RR |
553 | } |
554 | if (!len) | |
555 | return 0; | |
556 | ||
557 | fault: | |
558 | return -EFAULT; | |
559 | } | |
560 | EXPORT_SYMBOL(skb_copy_datagram_from_iovec); | |
561 | ||
1da177e4 LT |
562 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
563 | u8 __user *to, int len, | |
5084205f | 564 | __wsum *csump) |
1da177e4 | 565 | { |
1a028e50 | 566 | int start = skb_headlen(skb); |
1a028e50 | 567 | int i, copy = start - offset; |
5b1a002a DM |
568 | struct sk_buff *frag_iter; |
569 | int pos = 0; | |
1da177e4 LT |
570 | |
571 | /* Copy header. */ | |
572 | if (copy > 0) { | |
573 | int err = 0; | |
574 | if (copy > len) | |
575 | copy = len; | |
576 | *csump = csum_and_copy_to_user(skb->data + offset, to, copy, | |
577 | *csump, &err); | |
578 | if (err) | |
579 | goto fault; | |
580 | if ((len -= copy) == 0) | |
581 | return 0; | |
582 | offset += copy; | |
583 | to += copy; | |
584 | pos = copy; | |
585 | } | |
586 | ||
587 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 588 | int end; |
9e903e08 | 589 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 590 | |
547b792c | 591 | WARN_ON(start > offset + len); |
1a028e50 | 592 | |
9e903e08 | 593 | end = start + skb_frag_size(frag); |
1da177e4 | 594 | if ((copy = end - offset) > 0) { |
5084205f | 595 | __wsum csum2; |
1da177e4 LT |
596 | int err = 0; |
597 | u8 *vaddr; | |
ea2ab693 | 598 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
599 | |
600 | if (copy > len) | |
601 | copy = len; | |
602 | vaddr = kmap(page); | |
603 | csum2 = csum_and_copy_to_user(vaddr + | |
1a028e50 DM |
604 | frag->page_offset + |
605 | offset - start, | |
1da177e4 LT |
606 | to, copy, 0, &err); |
607 | kunmap(page); | |
608 | if (err) | |
609 | goto fault; | |
610 | *csump = csum_block_add(*csump, csum2, pos); | |
611 | if (!(len -= copy)) | |
612 | return 0; | |
613 | offset += copy; | |
614 | to += copy; | |
615 | pos += copy; | |
616 | } | |
1a028e50 | 617 | start = end; |
1da177e4 LT |
618 | } |
619 | ||
5b1a002a DM |
620 | skb_walk_frags(skb, frag_iter) { |
621 | int end; | |
622 | ||
623 | WARN_ON(start > offset + len); | |
624 | ||
625 | end = start + frag_iter->len; | |
626 | if ((copy = end - offset) > 0) { | |
627 | __wsum csum2 = 0; | |
628 | if (copy > len) | |
629 | copy = len; | |
630 | if (skb_copy_and_csum_datagram(frag_iter, | |
631 | offset - start, | |
632 | to, copy, | |
633 | &csum2)) | |
634 | goto fault; | |
635 | *csump = csum_block_add(*csump, csum2, pos); | |
636 | if ((len -= copy) == 0) | |
637 | return 0; | |
638 | offset += copy; | |
639 | to += copy; | |
640 | pos += copy; | |
1da177e4 | 641 | } |
5b1a002a | 642 | start = end; |
1da177e4 LT |
643 | } |
644 | if (!len) | |
645 | return 0; | |
646 | ||
647 | fault: | |
648 | return -EFAULT; | |
649 | } | |
650 | ||
759e5d00 | 651 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 652 | { |
d3bc23e7 | 653 | __sum16 sum; |
fb286bb2 | 654 | |
759e5d00 | 655 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
fb286bb2 | 656 | if (likely(!sum)) { |
84fa7933 | 657 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 HX |
658 | netdev_rx_csum_fault(skb->dev); |
659 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
660 | } | |
661 | return sum; | |
662 | } | |
759e5d00 HX |
663 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
664 | ||
665 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
666 | { | |
667 | return __skb_checksum_complete_head(skb, skb->len); | |
668 | } | |
fb286bb2 HX |
669 | EXPORT_SYMBOL(__skb_checksum_complete); |
670 | ||
1da177e4 LT |
671 | /** |
672 | * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec. | |
4dc3b16b PP |
673 | * @skb: skbuff |
674 | * @hlen: hardware length | |
67be2dd1 | 675 | * @iov: io vector |
4ec93edb | 676 | * |
1da177e4 LT |
677 | * Caller _must_ check that skb will fit to this iovec. |
678 | * | |
679 | * Returns: 0 - success. | |
680 | * -EINVAL - checksum failure. | |
681 | * -EFAULT - fault during copy. Beware, in this case iovec | |
682 | * can be modified! | |
683 | */ | |
fb286bb2 | 684 | int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, |
1da177e4 LT |
685 | int hlen, struct iovec *iov) |
686 | { | |
d3bc23e7 | 687 | __wsum csum; |
1da177e4 LT |
688 | int chunk = skb->len - hlen; |
689 | ||
ef8aef55 HX |
690 | if (!chunk) |
691 | return 0; | |
692 | ||
1da177e4 LT |
693 | /* Skip filled elements. |
694 | * Pretty silly, look at memcpy_toiovec, though 8) | |
695 | */ | |
696 | while (!iov->iov_len) | |
697 | iov++; | |
698 | ||
699 | if (iov->iov_len < chunk) { | |
fb286bb2 | 700 | if (__skb_checksum_complete(skb)) |
1da177e4 LT |
701 | goto csum_error; |
702 | if (skb_copy_datagram_iovec(skb, hlen, iov, chunk)) | |
703 | goto fault; | |
704 | } else { | |
705 | csum = csum_partial(skb->data, hlen, skb->csum); | |
706 | if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base, | |
707 | chunk, &csum)) | |
708 | goto fault; | |
d3bc23e7 | 709 | if (csum_fold(csum)) |
1da177e4 | 710 | goto csum_error; |
84fa7933 | 711 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 712 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
713 | iov->iov_len -= chunk; |
714 | iov->iov_base += chunk; | |
715 | } | |
716 | return 0; | |
717 | csum_error: | |
718 | return -EINVAL; | |
719 | fault: | |
720 | return -EFAULT; | |
721 | } | |
9e34a5b5 | 722 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec); |
1da177e4 LT |
723 | |
724 | /** | |
725 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
726 | * @file: file struct |
727 | * @sock: socket | |
728 | * @wait: poll table | |
1da177e4 LT |
729 | * |
730 | * Datagram poll: Again totally generic. This also handles | |
731 | * sequenced packet sockets providing the socket receive queue | |
732 | * is only ever holding data ready to receive. | |
733 | * | |
734 | * Note: when you _don't_ use this routine for this protocol, | |
735 | * and you use a different write policy from sock_writeable() | |
736 | * then please supply your own write_space callback. | |
737 | */ | |
738 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
739 | poll_table *wait) | |
740 | { | |
741 | struct sock *sk = sock->sk; | |
742 | unsigned int mask; | |
743 | ||
aa395145 | 744 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
745 | mask = 0; |
746 | ||
747 | /* exceptional events? */ | |
748 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
749 | mask |= POLLERR; | |
f348d70a | 750 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 751 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
752 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
753 | mask |= POLLHUP; | |
754 | ||
755 | /* readable? */ | |
db40980f | 756 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
757 | mask |= POLLIN | POLLRDNORM; |
758 | ||
759 | /* Connection-based need to check for termination and startup */ | |
760 | if (connection_based(sk)) { | |
761 | if (sk->sk_state == TCP_CLOSE) | |
762 | mask |= POLLHUP; | |
763 | /* connection hasn't started yet? */ | |
764 | if (sk->sk_state == TCP_SYN_SENT) | |
765 | return mask; | |
766 | } | |
767 | ||
768 | /* writable? */ | |
769 | if (sock_writeable(sk)) | |
770 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
771 | else | |
772 | set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
773 | ||
774 | return mask; | |
775 | } | |
1da177e4 | 776 | EXPORT_SYMBOL(datagram_poll); |