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 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 | 87 | |
aa395145 | 88 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); |
1da177e4 LT |
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: | |
aa395145 | 117 | finish_wait(sk_sleep(sk), &wait); |
1da177e4 LT |
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 | |
3f518bf7 PE |
134 | * @off: an offset in bytes to peek skb from. Returns an offset |
135 | * within an skb where data actually starts | |
a59322be | 136 | * @peeked: returns non-zero if this packet has been seen before |
4dc3b16b | 137 | * @err: error code returned |
1da177e4 LT |
138 | * |
139 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
140 | * and possible races. This replaces identical code in packet, raw and | |
141 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
142 | * the long standing peek and read race for datagram sockets. If you | |
143 | * alter this routine remember it must be re-entrant. | |
144 | * | |
145 | * This function will lock the socket if a skb is returned, so the caller | |
146 | * needs to unlock the socket in that case (usually by calling | |
147 | * skb_free_datagram) | |
148 | * | |
149 | * * It does not lock socket since today. This function is | |
150 | * * free of race conditions. This measure should/can improve | |
151 | * * significantly datagram socket latencies at high loads, | |
152 | * * when data copying to user space takes lots of time. | |
153 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
154 | * * 8) Great win.) | |
155 | * * --ANK (980729) | |
156 | * | |
157 | * The order of the tests when we find no data waiting are specified | |
158 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
159 | * the standard around please. | |
160 | */ | |
95c96174 | 161 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
3f518bf7 | 162 | int *peeked, int *off, int *err) |
1da177e4 LT |
163 | { |
164 | struct sk_buff *skb; | |
165 | long timeo; | |
166 | /* | |
167 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
168 | */ | |
169 | int error = sock_error(sk); | |
170 | ||
171 | if (error) | |
172 | goto no_packet; | |
173 | ||
a59322be | 174 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
175 | |
176 | do { | |
177 | /* Again only user level code calls this function, so nothing | |
178 | * interrupt level will suddenly eat the receive_queue. | |
179 | * | |
180 | * Look at current nfs client by the way... | |
8917a3c0 | 181 | * However, this function was correct in any case. 8) |
1da177e4 | 182 | */ |
a59322be | 183 | unsigned long cpu_flags; |
4934b032 | 184 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
a59322be | 185 | |
4934b032 | 186 | spin_lock_irqsave(&queue->lock, cpu_flags); |
3f518bf7 | 187 | skb_queue_walk(queue, skb) { |
a59322be HX |
188 | *peeked = skb->peeked; |
189 | if (flags & MSG_PEEK) { | |
77c1090f | 190 | if (*off >= skb->len && skb->len) { |
3f518bf7 PE |
191 | *off -= skb->len; |
192 | continue; | |
193 | } | |
a59322be | 194 | skb->peeked = 1; |
1da177e4 | 195 | atomic_inc(&skb->users); |
a59322be | 196 | } else |
4934b032 | 197 | __skb_unlink(skb, queue); |
1da177e4 | 198 | |
3f518bf7 | 199 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
1da177e4 | 200 | return skb; |
3f518bf7 PE |
201 | } |
202 | spin_unlock_irqrestore(&queue->lock, cpu_flags); | |
1da177e4 LT |
203 | |
204 | /* User doesn't want to wait */ | |
205 | error = -EAGAIN; | |
206 | if (!timeo) | |
207 | goto no_packet; | |
208 | ||
209 | } while (!wait_for_packet(sk, err, &timeo)); | |
210 | ||
211 | return NULL; | |
212 | ||
213 | no_packet: | |
214 | *err = error; | |
215 | return NULL; | |
216 | } | |
a59322be HX |
217 | EXPORT_SYMBOL(__skb_recv_datagram); |
218 | ||
95c96174 | 219 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
220 | int noblock, int *err) |
221 | { | |
3f518bf7 | 222 | int peeked, off = 0; |
a59322be HX |
223 | |
224 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
3f518bf7 | 225 | &peeked, &off, err); |
a59322be | 226 | } |
9e34a5b5 | 227 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
228 | |
229 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
230 | { | |
ead2ceb0 | 231 | consume_skb(skb); |
270acefa | 232 | sk_mem_reclaim_partial(sk); |
1da177e4 | 233 | } |
9d410c79 ED |
234 | EXPORT_SYMBOL(skb_free_datagram); |
235 | ||
236 | void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) | |
237 | { | |
8a74ad60 ED |
238 | bool slow; |
239 | ||
93bb64ea ED |
240 | if (likely(atomic_read(&skb->users) == 1)) |
241 | smp_rmb(); | |
242 | else if (likely(!atomic_dec_and_test(&skb->users))) | |
243 | return; | |
244 | ||
8a74ad60 | 245 | slow = lock_sock_fast(sk); |
4b0b72f7 ED |
246 | skb_orphan(skb); |
247 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 248 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 249 | |
93bb64ea ED |
250 | /* skb is now orphaned, can be freed outside of locked section */ |
251 | __kfree_skb(skb); | |
9d410c79 ED |
252 | } |
253 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 254 | |
3305b80c HX |
255 | /** |
256 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
257 | * @sk: socket | |
258 | * @skb: datagram skbuff | |
259 | * @flags: MSG_ flags | |
260 | * | |
261 | * This function frees a datagram skbuff that was received by | |
262 | * skb_recv_datagram. The flags argument must match the one | |
263 | * used for skb_recv_datagram. | |
264 | * | |
265 | * If the MSG_PEEK flag is set, and the packet is still on the | |
266 | * receive queue of the socket, it will be taken off the queue | |
267 | * before it is freed. | |
268 | * | |
269 | * This function currently only disables BH when acquiring the | |
270 | * sk_receive_queue lock. Therefore it must not be used in a | |
271 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
272 | * |
273 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
274 | */ |
275 | ||
27ab2568 | 276 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 277 | { |
27ab2568 HX |
278 | int err = 0; |
279 | ||
3305b80c | 280 | if (flags & MSG_PEEK) { |
27ab2568 | 281 | err = -ENOENT; |
3305b80c HX |
282 | spin_lock_bh(&sk->sk_receive_queue.lock); |
283 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
284 | __skb_unlink(skb, &sk->sk_receive_queue); | |
285 | atomic_dec(&skb->users); | |
27ab2568 | 286 | err = 0; |
3305b80c HX |
287 | } |
288 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
289 | } | |
290 | ||
61de71c6 | 291 | kfree_skb(skb); |
8edf19c2 | 292 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
293 | sk_mem_reclaim_partial(sk); |
294 | ||
27ab2568 | 295 | return err; |
3305b80c | 296 | } |
3305b80c HX |
297 | EXPORT_SYMBOL(skb_kill_datagram); |
298 | ||
1da177e4 LT |
299 | /** |
300 | * skb_copy_datagram_iovec - Copy a datagram to an iovec. | |
4dc3b16b PP |
301 | * @skb: buffer to copy |
302 | * @offset: offset in the buffer to start copying from | |
67be2dd1 | 303 | * @to: io vector to copy to |
4dc3b16b | 304 | * @len: amount of data to copy from buffer to iovec |
1da177e4 LT |
305 | * |
306 | * Note: the iovec is modified during the copy. | |
307 | */ | |
308 | int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset, | |
309 | struct iovec *to, int len) | |
310 | { | |
1a028e50 DM |
311 | int start = skb_headlen(skb); |
312 | int i, copy = start - offset; | |
5b1a002a | 313 | struct sk_buff *frag_iter; |
c75d721c | 314 | |
e9b3cc1b NH |
315 | trace_skb_copy_datagram_iovec(skb, len); |
316 | ||
b4d9eda0 DM |
317 | /* Copy header. */ |
318 | if (copy > 0) { | |
319 | if (copy > len) | |
320 | copy = len; | |
321 | if (memcpy_toiovec(to, skb->data + offset, copy)) | |
322 | goto fault; | |
323 | if ((len -= copy) == 0) | |
324 | return 0; | |
325 | offset += copy; | |
326 | } | |
c75d721c | 327 | |
b4d9eda0 DM |
328 | /* Copy paged appendix. Hmm... why does this look so complicated? */ |
329 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 330 | int end; |
9e903e08 | 331 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 332 | |
547b792c | 333 | WARN_ON(start > offset + len); |
1a028e50 | 334 | |
9e903e08 | 335 | end = start + skb_frag_size(frag); |
b4d9eda0 DM |
336 | if ((copy = end - offset) > 0) { |
337 | int err; | |
338 | u8 *vaddr; | |
ea2ab693 | 339 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
340 | |
341 | if (copy > len) | |
342 | copy = len; | |
b4d9eda0 | 343 | vaddr = kmap(page); |
1a028e50 DM |
344 | err = memcpy_toiovec(to, vaddr + frag->page_offset + |
345 | offset - start, copy); | |
b4d9eda0 | 346 | kunmap(page); |
1da177e4 LT |
347 | if (err) |
348 | goto fault; | |
349 | if (!(len -= copy)) | |
350 | return 0; | |
351 | offset += copy; | |
352 | } | |
1a028e50 | 353 | start = end; |
1da177e4 | 354 | } |
b4d9eda0 | 355 | |
5b1a002a DM |
356 | skb_walk_frags(skb, frag_iter) { |
357 | int end; | |
358 | ||
359 | WARN_ON(start > offset + len); | |
360 | ||
361 | end = start + frag_iter->len; | |
362 | if ((copy = end - offset) > 0) { | |
363 | if (copy > len) | |
364 | copy = len; | |
365 | if (skb_copy_datagram_iovec(frag_iter, | |
366 | offset - start, | |
367 | to, copy)) | |
368 | goto fault; | |
369 | if ((len -= copy) == 0) | |
370 | return 0; | |
371 | offset += copy; | |
b4d9eda0 | 372 | } |
5b1a002a | 373 | start = end; |
1da177e4 | 374 | } |
b4d9eda0 DM |
375 | if (!len) |
376 | return 0; | |
377 | ||
1da177e4 LT |
378 | fault: |
379 | return -EFAULT; | |
380 | } | |
9e34a5b5 | 381 | EXPORT_SYMBOL(skb_copy_datagram_iovec); |
1da177e4 | 382 | |
0a1ec07a MT |
383 | /** |
384 | * skb_copy_datagram_const_iovec - Copy a datagram to an iovec. | |
385 | * @skb: buffer to copy | |
386 | * @offset: offset in the buffer to start copying from | |
387 | * @to: io vector to copy to | |
388 | * @to_offset: offset in the io vector to start copying to | |
389 | * @len: amount of data to copy from buffer to iovec | |
390 | * | |
391 | * Returns 0 or -EFAULT. | |
392 | * Note: the iovec is not modified during the copy. | |
393 | */ | |
394 | int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset, | |
395 | const struct iovec *to, int to_offset, | |
396 | int len) | |
397 | { | |
398 | int start = skb_headlen(skb); | |
399 | int i, copy = start - offset; | |
5b1a002a | 400 | struct sk_buff *frag_iter; |
0a1ec07a MT |
401 | |
402 | /* Copy header. */ | |
403 | if (copy > 0) { | |
404 | if (copy > len) | |
405 | copy = len; | |
406 | if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy)) | |
407 | goto fault; | |
408 | if ((len -= copy) == 0) | |
409 | return 0; | |
410 | offset += copy; | |
411 | to_offset += copy; | |
412 | } | |
413 | ||
414 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
415 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
416 | int end; | |
9e903e08 | 417 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
0a1ec07a MT |
418 | |
419 | WARN_ON(start > offset + len); | |
420 | ||
9e903e08 | 421 | end = start + skb_frag_size(frag); |
0a1ec07a MT |
422 | if ((copy = end - offset) > 0) { |
423 | int err; | |
424 | u8 *vaddr; | |
ea2ab693 | 425 | struct page *page = skb_frag_page(frag); |
0a1ec07a MT |
426 | |
427 | if (copy > len) | |
428 | copy = len; | |
429 | vaddr = kmap(page); | |
430 | err = memcpy_toiovecend(to, vaddr + frag->page_offset + | |
431 | offset - start, to_offset, copy); | |
432 | kunmap(page); | |
433 | if (err) | |
434 | goto fault; | |
435 | if (!(len -= copy)) | |
436 | return 0; | |
437 | offset += copy; | |
438 | to_offset += copy; | |
439 | } | |
440 | start = end; | |
441 | } | |
442 | ||
5b1a002a DM |
443 | skb_walk_frags(skb, frag_iter) { |
444 | int end; | |
445 | ||
446 | WARN_ON(start > offset + len); | |
447 | ||
448 | end = start + frag_iter->len; | |
449 | if ((copy = end - offset) > 0) { | |
450 | if (copy > len) | |
451 | copy = len; | |
452 | if (skb_copy_datagram_const_iovec(frag_iter, | |
453 | offset - start, | |
454 | to, to_offset, | |
455 | copy)) | |
456 | goto fault; | |
457 | if ((len -= copy) == 0) | |
458 | return 0; | |
459 | offset += copy; | |
460 | to_offset += copy; | |
0a1ec07a | 461 | } |
5b1a002a | 462 | start = end; |
0a1ec07a MT |
463 | } |
464 | if (!len) | |
465 | return 0; | |
466 | ||
467 | fault: | |
468 | return -EFAULT; | |
469 | } | |
470 | EXPORT_SYMBOL(skb_copy_datagram_const_iovec); | |
471 | ||
db543c1f RR |
472 | /** |
473 | * skb_copy_datagram_from_iovec - Copy a datagram from an iovec. | |
474 | * @skb: buffer to copy | |
475 | * @offset: offset in the buffer to start copying to | |
476 | * @from: io vector to copy to | |
6f26c9a7 | 477 | * @from_offset: offset in the io vector to start copying from |
db543c1f RR |
478 | * @len: amount of data to copy to buffer from iovec |
479 | * | |
480 | * Returns 0 or -EFAULT. | |
6f26c9a7 | 481 | * Note: the iovec is not modified during the copy. |
db543c1f RR |
482 | */ |
483 | int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset, | |
6f26c9a7 MT |
484 | const struct iovec *from, int from_offset, |
485 | int len) | |
db543c1f RR |
486 | { |
487 | int start = skb_headlen(skb); | |
488 | int i, copy = start - offset; | |
5b1a002a | 489 | struct sk_buff *frag_iter; |
db543c1f RR |
490 | |
491 | /* Copy header. */ | |
492 | if (copy > 0) { | |
493 | if (copy > len) | |
494 | copy = len; | |
d2d27bfd SS |
495 | if (memcpy_fromiovecend(skb->data + offset, from, from_offset, |
496 | copy)) | |
db543c1f RR |
497 | goto fault; |
498 | if ((len -= copy) == 0) | |
499 | return 0; | |
500 | offset += copy; | |
6f26c9a7 | 501 | from_offset += copy; |
db543c1f RR |
502 | } |
503 | ||
504 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
505 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
506 | int end; | |
9e903e08 | 507 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
db543c1f RR |
508 | |
509 | WARN_ON(start > offset + len); | |
510 | ||
9e903e08 | 511 | end = start + skb_frag_size(frag); |
db543c1f RR |
512 | if ((copy = end - offset) > 0) { |
513 | int err; | |
514 | u8 *vaddr; | |
ea2ab693 | 515 | struct page *page = skb_frag_page(frag); |
db543c1f RR |
516 | |
517 | if (copy > len) | |
518 | copy = len; | |
519 | vaddr = kmap(page); | |
6f26c9a7 MT |
520 | err = memcpy_fromiovecend(vaddr + frag->page_offset + |
521 | offset - start, | |
522 | from, from_offset, copy); | |
db543c1f RR |
523 | kunmap(page); |
524 | if (err) | |
525 | goto fault; | |
526 | ||
527 | if (!(len -= copy)) | |
528 | return 0; | |
529 | offset += copy; | |
6f26c9a7 | 530 | from_offset += copy; |
db543c1f RR |
531 | } |
532 | start = end; | |
533 | } | |
534 | ||
5b1a002a DM |
535 | skb_walk_frags(skb, frag_iter) { |
536 | int end; | |
537 | ||
538 | WARN_ON(start > offset + len); | |
539 | ||
540 | end = start + frag_iter->len; | |
541 | if ((copy = end - offset) > 0) { | |
542 | if (copy > len) | |
543 | copy = len; | |
544 | if (skb_copy_datagram_from_iovec(frag_iter, | |
545 | offset - start, | |
546 | from, | |
547 | from_offset, | |
548 | copy)) | |
549 | goto fault; | |
550 | if ((len -= copy) == 0) | |
551 | return 0; | |
552 | offset += copy; | |
553 | from_offset += copy; | |
db543c1f | 554 | } |
5b1a002a | 555 | start = end; |
db543c1f RR |
556 | } |
557 | if (!len) | |
558 | return 0; | |
559 | ||
560 | fault: | |
561 | return -EFAULT; | |
562 | } | |
563 | EXPORT_SYMBOL(skb_copy_datagram_from_iovec); | |
564 | ||
1da177e4 LT |
565 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
566 | u8 __user *to, int len, | |
5084205f | 567 | __wsum *csump) |
1da177e4 | 568 | { |
1a028e50 | 569 | int start = skb_headlen(skb); |
1a028e50 | 570 | int i, copy = start - offset; |
5b1a002a DM |
571 | struct sk_buff *frag_iter; |
572 | int pos = 0; | |
1da177e4 LT |
573 | |
574 | /* Copy header. */ | |
575 | if (copy > 0) { | |
576 | int err = 0; | |
577 | if (copy > len) | |
578 | copy = len; | |
579 | *csump = csum_and_copy_to_user(skb->data + offset, to, copy, | |
580 | *csump, &err); | |
581 | if (err) | |
582 | goto fault; | |
583 | if ((len -= copy) == 0) | |
584 | return 0; | |
585 | offset += copy; | |
586 | to += copy; | |
587 | pos = copy; | |
588 | } | |
589 | ||
590 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 591 | int end; |
9e903e08 | 592 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 593 | |
547b792c | 594 | WARN_ON(start > offset + len); |
1a028e50 | 595 | |
9e903e08 | 596 | end = start + skb_frag_size(frag); |
1da177e4 | 597 | if ((copy = end - offset) > 0) { |
5084205f | 598 | __wsum csum2; |
1da177e4 LT |
599 | int err = 0; |
600 | u8 *vaddr; | |
ea2ab693 | 601 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
602 | |
603 | if (copy > len) | |
604 | copy = len; | |
605 | vaddr = kmap(page); | |
606 | csum2 = csum_and_copy_to_user(vaddr + | |
1a028e50 DM |
607 | frag->page_offset + |
608 | offset - start, | |
1da177e4 LT |
609 | to, copy, 0, &err); |
610 | kunmap(page); | |
611 | if (err) | |
612 | goto fault; | |
613 | *csump = csum_block_add(*csump, csum2, pos); | |
614 | if (!(len -= copy)) | |
615 | return 0; | |
616 | offset += copy; | |
617 | to += copy; | |
618 | pos += copy; | |
619 | } | |
1a028e50 | 620 | start = end; |
1da177e4 LT |
621 | } |
622 | ||
5b1a002a DM |
623 | skb_walk_frags(skb, frag_iter) { |
624 | int end; | |
625 | ||
626 | WARN_ON(start > offset + len); | |
627 | ||
628 | end = start + frag_iter->len; | |
629 | if ((copy = end - offset) > 0) { | |
630 | __wsum csum2 = 0; | |
631 | if (copy > len) | |
632 | copy = len; | |
633 | if (skb_copy_and_csum_datagram(frag_iter, | |
634 | offset - start, | |
635 | to, copy, | |
636 | &csum2)) | |
637 | goto fault; | |
638 | *csump = csum_block_add(*csump, csum2, pos); | |
639 | if ((len -= copy) == 0) | |
640 | return 0; | |
641 | offset += copy; | |
642 | to += copy; | |
643 | pos += copy; | |
1da177e4 | 644 | } |
5b1a002a | 645 | start = end; |
1da177e4 LT |
646 | } |
647 | if (!len) | |
648 | return 0; | |
649 | ||
650 | fault: | |
651 | return -EFAULT; | |
652 | } | |
653 | ||
759e5d00 | 654 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 655 | { |
d3bc23e7 | 656 | __sum16 sum; |
fb286bb2 | 657 | |
759e5d00 | 658 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
fb286bb2 | 659 | if (likely(!sum)) { |
84fa7933 | 660 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 HX |
661 | netdev_rx_csum_fault(skb->dev); |
662 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
663 | } | |
664 | return sum; | |
665 | } | |
759e5d00 HX |
666 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
667 | ||
668 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
669 | { | |
670 | return __skb_checksum_complete_head(skb, skb->len); | |
671 | } | |
fb286bb2 HX |
672 | EXPORT_SYMBOL(__skb_checksum_complete); |
673 | ||
1da177e4 LT |
674 | /** |
675 | * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec. | |
4dc3b16b PP |
676 | * @skb: skbuff |
677 | * @hlen: hardware length | |
67be2dd1 | 678 | * @iov: io vector |
4ec93edb | 679 | * |
1da177e4 LT |
680 | * Caller _must_ check that skb will fit to this iovec. |
681 | * | |
682 | * Returns: 0 - success. | |
683 | * -EINVAL - checksum failure. | |
684 | * -EFAULT - fault during copy. Beware, in this case iovec | |
685 | * can be modified! | |
686 | */ | |
fb286bb2 | 687 | int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, |
1da177e4 LT |
688 | int hlen, struct iovec *iov) |
689 | { | |
d3bc23e7 | 690 | __wsum csum; |
1da177e4 LT |
691 | int chunk = skb->len - hlen; |
692 | ||
ef8aef55 HX |
693 | if (!chunk) |
694 | return 0; | |
695 | ||
1da177e4 LT |
696 | /* Skip filled elements. |
697 | * Pretty silly, look at memcpy_toiovec, though 8) | |
698 | */ | |
699 | while (!iov->iov_len) | |
700 | iov++; | |
701 | ||
702 | if (iov->iov_len < chunk) { | |
fb286bb2 | 703 | if (__skb_checksum_complete(skb)) |
1da177e4 LT |
704 | goto csum_error; |
705 | if (skb_copy_datagram_iovec(skb, hlen, iov, chunk)) | |
706 | goto fault; | |
707 | } else { | |
708 | csum = csum_partial(skb->data, hlen, skb->csum); | |
709 | if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base, | |
710 | chunk, &csum)) | |
711 | goto fault; | |
d3bc23e7 | 712 | if (csum_fold(csum)) |
1da177e4 | 713 | goto csum_error; |
84fa7933 | 714 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 715 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
716 | iov->iov_len -= chunk; |
717 | iov->iov_base += chunk; | |
718 | } | |
719 | return 0; | |
720 | csum_error: | |
721 | return -EINVAL; | |
722 | fault: | |
723 | return -EFAULT; | |
724 | } | |
9e34a5b5 | 725 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec); |
1da177e4 LT |
726 | |
727 | /** | |
728 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
729 | * @file: file struct |
730 | * @sock: socket | |
731 | * @wait: poll table | |
1da177e4 LT |
732 | * |
733 | * Datagram poll: Again totally generic. This also handles | |
734 | * sequenced packet sockets providing the socket receive queue | |
735 | * is only ever holding data ready to receive. | |
736 | * | |
737 | * Note: when you _don't_ use this routine for this protocol, | |
738 | * and you use a different write policy from sock_writeable() | |
739 | * then please supply your own write_space callback. | |
740 | */ | |
741 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
742 | poll_table *wait) | |
743 | { | |
744 | struct sock *sk = sock->sk; | |
745 | unsigned int mask; | |
746 | ||
aa395145 | 747 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
748 | mask = 0; |
749 | ||
750 | /* exceptional events? */ | |
751 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
752 | mask |= POLLERR; | |
f348d70a | 753 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 754 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
755 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
756 | mask |= POLLHUP; | |
757 | ||
758 | /* readable? */ | |
db40980f | 759 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
760 | mask |= POLLIN | POLLRDNORM; |
761 | ||
762 | /* Connection-based need to check for termination and startup */ | |
763 | if (connection_based(sk)) { | |
764 | if (sk->sk_state == TCP_CLOSE) | |
765 | mask |= POLLHUP; | |
766 | /* connection hasn't started yet? */ | |
767 | if (sk->sk_state == TCP_SYN_SENT) | |
768 | return mask; | |
769 | } | |
770 | ||
771 | /* writable? */ | |
772 | if (sock_writeable(sk)) | |
773 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
774 | else | |
775 | set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
776 | ||
777 | return mask; | |
778 | } | |
1da177e4 | 779 | EXPORT_SYMBOL(datagram_poll); |