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) { | |
3f518bf7 PE |
190 | if (*off >= skb->len) { |
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 | 250 | /* skb is now orphaned, can be freed outside of locked section */ |
07dc22e7 | 251 | trace_kfree_skb(skb, skb_free_datagram_locked); |
93bb64ea | 252 | __kfree_skb(skb); |
9d410c79 ED |
253 | } |
254 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 255 | |
3305b80c HX |
256 | /** |
257 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
258 | * @sk: socket | |
259 | * @skb: datagram skbuff | |
260 | * @flags: MSG_ flags | |
261 | * | |
262 | * This function frees a datagram skbuff that was received by | |
263 | * skb_recv_datagram. The flags argument must match the one | |
264 | * used for skb_recv_datagram. | |
265 | * | |
266 | * If the MSG_PEEK flag is set, and the packet is still on the | |
267 | * receive queue of the socket, it will be taken off the queue | |
268 | * before it is freed. | |
269 | * | |
270 | * This function currently only disables BH when acquiring the | |
271 | * sk_receive_queue lock. Therefore it must not be used in a | |
272 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
273 | * |
274 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
275 | */ |
276 | ||
27ab2568 | 277 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 278 | { |
27ab2568 HX |
279 | int err = 0; |
280 | ||
3305b80c | 281 | if (flags & MSG_PEEK) { |
27ab2568 | 282 | err = -ENOENT; |
3305b80c HX |
283 | spin_lock_bh(&sk->sk_receive_queue.lock); |
284 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
285 | __skb_unlink(skb, &sk->sk_receive_queue); | |
286 | atomic_dec(&skb->users); | |
27ab2568 | 287 | err = 0; |
3305b80c HX |
288 | } |
289 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
290 | } | |
291 | ||
61de71c6 | 292 | kfree_skb(skb); |
8edf19c2 | 293 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
294 | sk_mem_reclaim_partial(sk); |
295 | ||
27ab2568 | 296 | return err; |
3305b80c | 297 | } |
3305b80c HX |
298 | EXPORT_SYMBOL(skb_kill_datagram); |
299 | ||
1da177e4 LT |
300 | /** |
301 | * skb_copy_datagram_iovec - Copy a datagram to an iovec. | |
4dc3b16b PP |
302 | * @skb: buffer to copy |
303 | * @offset: offset in the buffer to start copying from | |
67be2dd1 | 304 | * @to: io vector to copy to |
4dc3b16b | 305 | * @len: amount of data to copy from buffer to iovec |
1da177e4 LT |
306 | * |
307 | * Note: the iovec is modified during the copy. | |
308 | */ | |
309 | int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset, | |
310 | struct iovec *to, int len) | |
311 | { | |
1a028e50 DM |
312 | int start = skb_headlen(skb); |
313 | int i, copy = start - offset; | |
5b1a002a | 314 | struct sk_buff *frag_iter; |
c75d721c | 315 | |
e9b3cc1b NH |
316 | trace_skb_copy_datagram_iovec(skb, len); |
317 | ||
b4d9eda0 DM |
318 | /* Copy header. */ |
319 | if (copy > 0) { | |
320 | if (copy > len) | |
321 | copy = len; | |
322 | if (memcpy_toiovec(to, skb->data + offset, copy)) | |
323 | goto fault; | |
324 | if ((len -= copy) == 0) | |
325 | return 0; | |
326 | offset += copy; | |
327 | } | |
c75d721c | 328 | |
b4d9eda0 DM |
329 | /* Copy paged appendix. Hmm... why does this look so complicated? */ |
330 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 331 | int end; |
9e903e08 | 332 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 333 | |
547b792c | 334 | WARN_ON(start > offset + len); |
1a028e50 | 335 | |
9e903e08 | 336 | end = start + skb_frag_size(frag); |
b4d9eda0 DM |
337 | if ((copy = end - offset) > 0) { |
338 | int err; | |
339 | u8 *vaddr; | |
ea2ab693 | 340 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
341 | |
342 | if (copy > len) | |
343 | copy = len; | |
b4d9eda0 | 344 | vaddr = kmap(page); |
1a028e50 DM |
345 | err = memcpy_toiovec(to, vaddr + frag->page_offset + |
346 | offset - start, copy); | |
b4d9eda0 | 347 | kunmap(page); |
1da177e4 LT |
348 | if (err) |
349 | goto fault; | |
350 | if (!(len -= copy)) | |
351 | return 0; | |
352 | offset += copy; | |
353 | } | |
1a028e50 | 354 | start = end; |
1da177e4 | 355 | } |
b4d9eda0 | 356 | |
5b1a002a DM |
357 | skb_walk_frags(skb, frag_iter) { |
358 | int end; | |
359 | ||
360 | WARN_ON(start > offset + len); | |
361 | ||
362 | end = start + frag_iter->len; | |
363 | if ((copy = end - offset) > 0) { | |
364 | if (copy > len) | |
365 | copy = len; | |
366 | if (skb_copy_datagram_iovec(frag_iter, | |
367 | offset - start, | |
368 | to, copy)) | |
369 | goto fault; | |
370 | if ((len -= copy) == 0) | |
371 | return 0; | |
372 | offset += copy; | |
b4d9eda0 | 373 | } |
5b1a002a | 374 | start = end; |
1da177e4 | 375 | } |
b4d9eda0 DM |
376 | if (!len) |
377 | return 0; | |
378 | ||
1da177e4 LT |
379 | fault: |
380 | return -EFAULT; | |
381 | } | |
9e34a5b5 | 382 | EXPORT_SYMBOL(skb_copy_datagram_iovec); |
1da177e4 | 383 | |
0a1ec07a MT |
384 | /** |
385 | * skb_copy_datagram_const_iovec - Copy a datagram to an iovec. | |
386 | * @skb: buffer to copy | |
387 | * @offset: offset in the buffer to start copying from | |
388 | * @to: io vector to copy to | |
389 | * @to_offset: offset in the io vector to start copying to | |
390 | * @len: amount of data to copy from buffer to iovec | |
391 | * | |
392 | * Returns 0 or -EFAULT. | |
393 | * Note: the iovec is not modified during the copy. | |
394 | */ | |
395 | int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset, | |
396 | const struct iovec *to, int to_offset, | |
397 | int len) | |
398 | { | |
399 | int start = skb_headlen(skb); | |
400 | int i, copy = start - offset; | |
5b1a002a | 401 | struct sk_buff *frag_iter; |
0a1ec07a MT |
402 | |
403 | /* Copy header. */ | |
404 | if (copy > 0) { | |
405 | if (copy > len) | |
406 | copy = len; | |
407 | if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy)) | |
408 | goto fault; | |
409 | if ((len -= copy) == 0) | |
410 | return 0; | |
411 | offset += copy; | |
412 | to_offset += copy; | |
413 | } | |
414 | ||
415 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
416 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
417 | int end; | |
9e903e08 | 418 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
0a1ec07a MT |
419 | |
420 | WARN_ON(start > offset + len); | |
421 | ||
9e903e08 | 422 | end = start + skb_frag_size(frag); |
0a1ec07a MT |
423 | if ((copy = end - offset) > 0) { |
424 | int err; | |
425 | u8 *vaddr; | |
ea2ab693 | 426 | struct page *page = skb_frag_page(frag); |
0a1ec07a MT |
427 | |
428 | if (copy > len) | |
429 | copy = len; | |
430 | vaddr = kmap(page); | |
431 | err = memcpy_toiovecend(to, vaddr + frag->page_offset + | |
432 | offset - start, to_offset, copy); | |
433 | kunmap(page); | |
434 | if (err) | |
435 | goto fault; | |
436 | if (!(len -= copy)) | |
437 | return 0; | |
438 | offset += copy; | |
439 | to_offset += copy; | |
440 | } | |
441 | start = end; | |
442 | } | |
443 | ||
5b1a002a DM |
444 | skb_walk_frags(skb, frag_iter) { |
445 | int end; | |
446 | ||
447 | WARN_ON(start > offset + len); | |
448 | ||
449 | end = start + frag_iter->len; | |
450 | if ((copy = end - offset) > 0) { | |
451 | if (copy > len) | |
452 | copy = len; | |
453 | if (skb_copy_datagram_const_iovec(frag_iter, | |
454 | offset - start, | |
455 | to, to_offset, | |
456 | copy)) | |
457 | goto fault; | |
458 | if ((len -= copy) == 0) | |
459 | return 0; | |
460 | offset += copy; | |
461 | to_offset += copy; | |
0a1ec07a | 462 | } |
5b1a002a | 463 | start = end; |
0a1ec07a MT |
464 | } |
465 | if (!len) | |
466 | return 0; | |
467 | ||
468 | fault: | |
469 | return -EFAULT; | |
470 | } | |
471 | EXPORT_SYMBOL(skb_copy_datagram_const_iovec); | |
472 | ||
db543c1f RR |
473 | /** |
474 | * skb_copy_datagram_from_iovec - Copy a datagram from an iovec. | |
475 | * @skb: buffer to copy | |
476 | * @offset: offset in the buffer to start copying to | |
477 | * @from: io vector to copy to | |
6f26c9a7 | 478 | * @from_offset: offset in the io vector to start copying from |
db543c1f RR |
479 | * @len: amount of data to copy to buffer from iovec |
480 | * | |
481 | * Returns 0 or -EFAULT. | |
6f26c9a7 | 482 | * Note: the iovec is not modified during the copy. |
db543c1f RR |
483 | */ |
484 | int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset, | |
6f26c9a7 MT |
485 | const struct iovec *from, int from_offset, |
486 | int len) | |
db543c1f RR |
487 | { |
488 | int start = skb_headlen(skb); | |
489 | int i, copy = start - offset; | |
5b1a002a | 490 | struct sk_buff *frag_iter; |
db543c1f RR |
491 | |
492 | /* Copy header. */ | |
493 | if (copy > 0) { | |
494 | if (copy > len) | |
495 | copy = len; | |
d2d27bfd SS |
496 | if (memcpy_fromiovecend(skb->data + offset, from, from_offset, |
497 | copy)) | |
db543c1f RR |
498 | goto fault; |
499 | if ((len -= copy) == 0) | |
500 | return 0; | |
501 | offset += copy; | |
6f26c9a7 | 502 | from_offset += copy; |
db543c1f RR |
503 | } |
504 | ||
505 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
506 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
507 | int end; | |
9e903e08 | 508 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
db543c1f RR |
509 | |
510 | WARN_ON(start > offset + len); | |
511 | ||
9e903e08 | 512 | end = start + skb_frag_size(frag); |
db543c1f RR |
513 | if ((copy = end - offset) > 0) { |
514 | int err; | |
515 | u8 *vaddr; | |
ea2ab693 | 516 | struct page *page = skb_frag_page(frag); |
db543c1f RR |
517 | |
518 | if (copy > len) | |
519 | copy = len; | |
520 | vaddr = kmap(page); | |
6f26c9a7 MT |
521 | err = memcpy_fromiovecend(vaddr + frag->page_offset + |
522 | offset - start, | |
523 | from, from_offset, copy); | |
db543c1f RR |
524 | kunmap(page); |
525 | if (err) | |
526 | goto fault; | |
527 | ||
528 | if (!(len -= copy)) | |
529 | return 0; | |
530 | offset += copy; | |
6f26c9a7 | 531 | from_offset += copy; |
db543c1f RR |
532 | } |
533 | start = end; | |
534 | } | |
535 | ||
5b1a002a DM |
536 | skb_walk_frags(skb, frag_iter) { |
537 | int end; | |
538 | ||
539 | WARN_ON(start > offset + len); | |
540 | ||
541 | end = start + frag_iter->len; | |
542 | if ((copy = end - offset) > 0) { | |
543 | if (copy > len) | |
544 | copy = len; | |
545 | if (skb_copy_datagram_from_iovec(frag_iter, | |
546 | offset - start, | |
547 | from, | |
548 | from_offset, | |
549 | copy)) | |
550 | goto fault; | |
551 | if ((len -= copy) == 0) | |
552 | return 0; | |
553 | offset += copy; | |
554 | from_offset += copy; | |
db543c1f | 555 | } |
5b1a002a | 556 | start = end; |
db543c1f RR |
557 | } |
558 | if (!len) | |
559 | return 0; | |
560 | ||
561 | fault: | |
562 | return -EFAULT; | |
563 | } | |
564 | EXPORT_SYMBOL(skb_copy_datagram_from_iovec); | |
565 | ||
1da177e4 LT |
566 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
567 | u8 __user *to, int len, | |
5084205f | 568 | __wsum *csump) |
1da177e4 | 569 | { |
1a028e50 | 570 | int start = skb_headlen(skb); |
1a028e50 | 571 | int i, copy = start - offset; |
5b1a002a DM |
572 | struct sk_buff *frag_iter; |
573 | int pos = 0; | |
1da177e4 LT |
574 | |
575 | /* Copy header. */ | |
576 | if (copy > 0) { | |
577 | int err = 0; | |
578 | if (copy > len) | |
579 | copy = len; | |
580 | *csump = csum_and_copy_to_user(skb->data + offset, to, copy, | |
581 | *csump, &err); | |
582 | if (err) | |
583 | goto fault; | |
584 | if ((len -= copy) == 0) | |
585 | return 0; | |
586 | offset += copy; | |
587 | to += copy; | |
588 | pos = copy; | |
589 | } | |
590 | ||
591 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 592 | int end; |
9e903e08 | 593 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 594 | |
547b792c | 595 | WARN_ON(start > offset + len); |
1a028e50 | 596 | |
9e903e08 | 597 | end = start + skb_frag_size(frag); |
1da177e4 | 598 | if ((copy = end - offset) > 0) { |
5084205f | 599 | __wsum csum2; |
1da177e4 LT |
600 | int err = 0; |
601 | u8 *vaddr; | |
ea2ab693 | 602 | struct page *page = skb_frag_page(frag); |
1da177e4 LT |
603 | |
604 | if (copy > len) | |
605 | copy = len; | |
606 | vaddr = kmap(page); | |
607 | csum2 = csum_and_copy_to_user(vaddr + | |
1a028e50 DM |
608 | frag->page_offset + |
609 | offset - start, | |
1da177e4 LT |
610 | to, copy, 0, &err); |
611 | kunmap(page); | |
612 | if (err) | |
613 | goto fault; | |
614 | *csump = csum_block_add(*csump, csum2, pos); | |
615 | if (!(len -= copy)) | |
616 | return 0; | |
617 | offset += copy; | |
618 | to += copy; | |
619 | pos += copy; | |
620 | } | |
1a028e50 | 621 | start = end; |
1da177e4 LT |
622 | } |
623 | ||
5b1a002a DM |
624 | skb_walk_frags(skb, frag_iter) { |
625 | int end; | |
626 | ||
627 | WARN_ON(start > offset + len); | |
628 | ||
629 | end = start + frag_iter->len; | |
630 | if ((copy = end - offset) > 0) { | |
631 | __wsum csum2 = 0; | |
632 | if (copy > len) | |
633 | copy = len; | |
634 | if (skb_copy_and_csum_datagram(frag_iter, | |
635 | offset - start, | |
636 | to, copy, | |
637 | &csum2)) | |
638 | goto fault; | |
639 | *csump = csum_block_add(*csump, csum2, pos); | |
640 | if ((len -= copy) == 0) | |
641 | return 0; | |
642 | offset += copy; | |
643 | to += copy; | |
644 | pos += copy; | |
1da177e4 | 645 | } |
5b1a002a | 646 | start = end; |
1da177e4 LT |
647 | } |
648 | if (!len) | |
649 | return 0; | |
650 | ||
651 | fault: | |
652 | return -EFAULT; | |
653 | } | |
654 | ||
759e5d00 | 655 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 656 | { |
d3bc23e7 | 657 | __sum16 sum; |
fb286bb2 | 658 | |
759e5d00 | 659 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
fb286bb2 | 660 | if (likely(!sum)) { |
84fa7933 | 661 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 HX |
662 | netdev_rx_csum_fault(skb->dev); |
663 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
664 | } | |
665 | return sum; | |
666 | } | |
759e5d00 HX |
667 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
668 | ||
669 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
670 | { | |
671 | return __skb_checksum_complete_head(skb, skb->len); | |
672 | } | |
fb286bb2 HX |
673 | EXPORT_SYMBOL(__skb_checksum_complete); |
674 | ||
1da177e4 LT |
675 | /** |
676 | * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec. | |
4dc3b16b PP |
677 | * @skb: skbuff |
678 | * @hlen: hardware length | |
67be2dd1 | 679 | * @iov: io vector |
4ec93edb | 680 | * |
1da177e4 LT |
681 | * Caller _must_ check that skb will fit to this iovec. |
682 | * | |
683 | * Returns: 0 - success. | |
684 | * -EINVAL - checksum failure. | |
685 | * -EFAULT - fault during copy. Beware, in this case iovec | |
686 | * can be modified! | |
687 | */ | |
fb286bb2 | 688 | int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, |
1da177e4 LT |
689 | int hlen, struct iovec *iov) |
690 | { | |
d3bc23e7 | 691 | __wsum csum; |
1da177e4 LT |
692 | int chunk = skb->len - hlen; |
693 | ||
ef8aef55 HX |
694 | if (!chunk) |
695 | return 0; | |
696 | ||
1da177e4 LT |
697 | /* Skip filled elements. |
698 | * Pretty silly, look at memcpy_toiovec, though 8) | |
699 | */ | |
700 | while (!iov->iov_len) | |
701 | iov++; | |
702 | ||
703 | if (iov->iov_len < chunk) { | |
fb286bb2 | 704 | if (__skb_checksum_complete(skb)) |
1da177e4 LT |
705 | goto csum_error; |
706 | if (skb_copy_datagram_iovec(skb, hlen, iov, chunk)) | |
707 | goto fault; | |
708 | } else { | |
709 | csum = csum_partial(skb->data, hlen, skb->csum); | |
710 | if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base, | |
711 | chunk, &csum)) | |
712 | goto fault; | |
d3bc23e7 | 713 | if (csum_fold(csum)) |
1da177e4 | 714 | goto csum_error; |
84fa7933 | 715 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 716 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
717 | iov->iov_len -= chunk; |
718 | iov->iov_base += chunk; | |
719 | } | |
720 | return 0; | |
721 | csum_error: | |
722 | return -EINVAL; | |
723 | fault: | |
724 | return -EFAULT; | |
725 | } | |
9e34a5b5 | 726 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec); |
1da177e4 LT |
727 | |
728 | /** | |
729 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
730 | * @file: file struct |
731 | * @sock: socket | |
732 | * @wait: poll table | |
1da177e4 LT |
733 | * |
734 | * Datagram poll: Again totally generic. This also handles | |
735 | * sequenced packet sockets providing the socket receive queue | |
736 | * is only ever holding data ready to receive. | |
737 | * | |
738 | * Note: when you _don't_ use this routine for this protocol, | |
739 | * and you use a different write policy from sock_writeable() | |
740 | * then please supply your own write_space callback. | |
741 | */ | |
742 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
743 | poll_table *wait) | |
744 | { | |
745 | struct sock *sk = sock->sk; | |
746 | unsigned int mask; | |
747 | ||
aa395145 | 748 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
749 | mask = 0; |
750 | ||
751 | /* exceptional events? */ | |
752 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
753 | mask |= POLLERR; | |
f348d70a | 754 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 755 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
756 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
757 | mask |= POLLHUP; | |
758 | ||
759 | /* readable? */ | |
db40980f | 760 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
761 | mask |= POLLIN | POLLRDNORM; |
762 | ||
763 | /* Connection-based need to check for termination and startup */ | |
764 | if (connection_based(sk)) { | |
765 | if (sk->sk_state == TCP_CLOSE) | |
766 | mask |= POLLHUP; | |
767 | /* connection hasn't started yet? */ | |
768 | if (sk->sk_state == TCP_SYN_SENT) | |
769 | return mask; | |
770 | } | |
771 | ||
772 | /* writable? */ | |
773 | if (sock_writeable(sk)) | |
774 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
775 | else | |
776 | set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
777 | ||
778 | return mask; | |
779 | } | |
1da177e4 | 780 | EXPORT_SYMBOL(datagram_poll); |