]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * SUCS NET3: | |
3 | * | |
4 | * Generic datagram handling routines. These are generic for all | |
5 | * protocols. Possibly a generic IP version on top of these would | |
6 | * make sense. Not tonight however 8-). | |
7 | * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and | |
8 | * NetROM layer all have identical poll code and mostly | |
9 | * identical recvmsg() code. So we share it here. The poll was | |
10 | * shared before but buried in udp.c so I moved it. | |
11 | * | |
113aa838 | 12 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old |
1da177e4 LT |
13 | * udp.c code) |
14 | * | |
15 | * Fixes: | |
16 | * Alan Cox : NULL return from skb_peek_copy() | |
17 | * understood | |
18 | * Alan Cox : Rewrote skb_read_datagram to avoid the | |
19 | * skb_peek_copy stuff. | |
20 | * Alan Cox : Added support for SOCK_SEQPACKET. | |
21 | * IPX can no longer use the SO_TYPE hack | |
22 | * but AX.25 now works right, and SPX is | |
23 | * feasible. | |
24 | * Alan Cox : Fixed write poll of non IP protocol | |
25 | * crash. | |
26 | * Florian La Roche: Changed for my new skbuff handling. | |
27 | * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET. | |
28 | * Linus Torvalds : BSD semantic fixes. | |
29 | * Alan Cox : Datagram iovec handling | |
30 | * Darryl Miles : Fixed non-blocking SOCK_STREAM. | |
31 | * Alan Cox : POSIXisms | |
32 | * Pete Wyckoff : Unconnected accept() fix. | |
33 | * | |
34 | */ | |
35 | ||
36 | #include <linux/module.h> | |
37 | #include <linux/types.h> | |
38 | #include <linux/kernel.h> | |
39 | #include <asm/uaccess.h> | |
1da177e4 LT |
40 | #include <linux/mm.h> |
41 | #include <linux/interrupt.h> | |
42 | #include <linux/errno.h> | |
43 | #include <linux/sched.h> | |
44 | #include <linux/inet.h> | |
1da177e4 LT |
45 | #include <linux/netdevice.h> |
46 | #include <linux/rtnetlink.h> | |
47 | #include <linux/poll.h> | |
48 | #include <linux/highmem.h> | |
3305b80c | 49 | #include <linux/spinlock.h> |
5a0e3ad6 | 50 | #include <linux/slab.h> |
0433547a | 51 | #include <linux/pagemap.h> |
a8f820aa | 52 | #include <linux/uio.h> |
1da177e4 LT |
53 | |
54 | #include <net/protocol.h> | |
55 | #include <linux/skbuff.h> | |
1da177e4 | 56 | |
c752f073 ACM |
57 | #include <net/checksum.h> |
58 | #include <net/sock.h> | |
59 | #include <net/tcp_states.h> | |
e9b3cc1b | 60 | #include <trace/events/skb.h> |
076bb0c8 | 61 | #include <net/busy_poll.h> |
1da177e4 LT |
62 | |
63 | /* | |
64 | * Is a socket 'connection oriented' ? | |
65 | */ | |
66 | static inline int connection_based(struct sock *sk) | |
67 | { | |
68 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
69 | } | |
70 | ||
95c96174 | 71 | static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync, |
bf368e4e ED |
72 | void *key) |
73 | { | |
74 | unsigned long bits = (unsigned long)key; | |
75 | ||
76 | /* | |
77 | * Avoid a wakeup if event not interesting for us | |
78 | */ | |
79 | if (bits && !(bits & (POLLIN | POLLERR))) | |
80 | return 0; | |
81 | return autoremove_wake_function(wait, mode, sync, key); | |
82 | } | |
1da177e4 | 83 | /* |
39cc8613 | 84 | * Wait for the last received packet to be different from skb |
1da177e4 | 85 | */ |
39cc8613 BP |
86 | static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, |
87 | const struct sk_buff *skb) | |
1da177e4 LT |
88 | { |
89 | int error; | |
bf368e4e | 90 | DEFINE_WAIT_FUNC(wait, receiver_wake_function); |
1da177e4 | 91 | |
aa395145 | 92 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); |
1da177e4 LT |
93 | |
94 | /* Socket errors? */ | |
95 | error = sock_error(sk); | |
96 | if (error) | |
97 | goto out_err; | |
98 | ||
39cc8613 | 99 | if (sk->sk_receive_queue.prev != skb) |
1da177e4 LT |
100 | goto out; |
101 | ||
102 | /* Socket shut down? */ | |
103 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
104 | goto out_noerr; | |
105 | ||
106 | /* Sequenced packets can come disconnected. | |
107 | * If so we report the problem | |
108 | */ | |
109 | error = -ENOTCONN; | |
110 | if (connection_based(sk) && | |
111 | !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) | |
112 | goto out_err; | |
113 | ||
114 | /* handle signals */ | |
115 | if (signal_pending(current)) | |
116 | goto interrupted; | |
117 | ||
118 | error = 0; | |
119 | *timeo_p = schedule_timeout(*timeo_p); | |
120 | out: | |
aa395145 | 121 | finish_wait(sk_sleep(sk), &wait); |
1da177e4 LT |
122 | return error; |
123 | interrupted: | |
124 | error = sock_intr_errno(*timeo_p); | |
125 | out_err: | |
126 | *err = error; | |
127 | goto out; | |
128 | out_noerr: | |
129 | *err = 0; | |
130 | error = 1; | |
131 | goto out; | |
132 | } | |
133 | ||
a0a2a660 | 134 | static struct sk_buff *skb_set_peeked(struct sk_buff *skb) |
738ac1eb HX |
135 | { |
136 | struct sk_buff *nskb; | |
137 | ||
138 | if (skb->peeked) | |
a0a2a660 | 139 | return skb; |
738ac1eb HX |
140 | |
141 | /* We have to unshare an skb before modifying it. */ | |
142 | if (!skb_shared(skb)) | |
143 | goto done; | |
144 | ||
145 | nskb = skb_clone(skb, GFP_ATOMIC); | |
146 | if (!nskb) | |
a0a2a660 | 147 | return ERR_PTR(-ENOMEM); |
738ac1eb HX |
148 | |
149 | skb->prev->next = nskb; | |
150 | skb->next->prev = nskb; | |
151 | nskb->prev = skb->prev; | |
152 | nskb->next = skb->next; | |
153 | ||
154 | consume_skb(skb); | |
155 | skb = nskb; | |
156 | ||
157 | done: | |
158 | skb->peeked = 1; | |
159 | ||
a0a2a660 | 160 | return skb; |
738ac1eb HX |
161 | } |
162 | ||
1da177e4 | 163 | /** |
a59322be | 164 | * __skb_recv_datagram - Receive a datagram skbuff |
4dc3b16b PP |
165 | * @sk: socket |
166 | * @flags: MSG_ flags | |
39cc8613 | 167 | * @peeked: returns non-zero if this packet has been seen before |
3f518bf7 PE |
168 | * @off: an offset in bytes to peek skb from. Returns an offset |
169 | * within an skb where data actually starts | |
4dc3b16b | 170 | * @err: error code returned |
1da177e4 LT |
171 | * |
172 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
173 | * and possible races. This replaces identical code in packet, raw and | |
174 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
175 | * the long standing peek and read race for datagram sockets. If you | |
176 | * alter this routine remember it must be re-entrant. | |
177 | * | |
178 | * This function will lock the socket if a skb is returned, so the caller | |
179 | * needs to unlock the socket in that case (usually by calling | |
180 | * skb_free_datagram) | |
181 | * | |
182 | * * It does not lock socket since today. This function is | |
183 | * * free of race conditions. This measure should/can improve | |
184 | * * significantly datagram socket latencies at high loads, | |
185 | * * when data copying to user space takes lots of time. | |
186 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
187 | * * 8) Great win.) | |
188 | * * --ANK (980729) | |
189 | * | |
190 | * The order of the tests when we find no data waiting are specified | |
191 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
192 | * the standard around please. | |
193 | */ | |
95c96174 | 194 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
3f518bf7 | 195 | int *peeked, int *off, int *err) |
1da177e4 | 196 | { |
738ac1eb | 197 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
39cc8613 | 198 | struct sk_buff *skb, *last; |
738ac1eb | 199 | unsigned long cpu_flags; |
1da177e4 LT |
200 | long timeo; |
201 | /* | |
202 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
203 | */ | |
204 | int error = sock_error(sk); | |
205 | ||
206 | if (error) | |
207 | goto no_packet; | |
208 | ||
a59322be | 209 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
210 | |
211 | do { | |
212 | /* Again only user level code calls this function, so nothing | |
213 | * interrupt level will suddenly eat the receive_queue. | |
214 | * | |
215 | * Look at current nfs client by the way... | |
8917a3c0 | 216 | * However, this function was correct in any case. 8) |
1da177e4 | 217 | */ |
39cc8613 | 218 | int _off = *off; |
a59322be | 219 | |
39cc8613 | 220 | last = (struct sk_buff *)queue; |
4934b032 | 221 | spin_lock_irqsave(&queue->lock, cpu_flags); |
3f518bf7 | 222 | skb_queue_walk(queue, skb) { |
39cc8613 | 223 | last = skb; |
a59322be HX |
224 | *peeked = skb->peeked; |
225 | if (flags & MSG_PEEK) { | |
39cc8613 | 226 | if (_off >= skb->len && (skb->len || _off || |
add05ad4 | 227 | skb->peeked)) { |
39cc8613 | 228 | _off -= skb->len; |
3f518bf7 PE |
229 | continue; |
230 | } | |
738ac1eb | 231 | |
a0a2a660 HX |
232 | skb = skb_set_peeked(skb); |
233 | error = PTR_ERR(skb); | |
234 | if (IS_ERR(skb)) | |
738ac1eb HX |
235 | goto unlock_err; |
236 | ||
1da177e4 | 237 | atomic_inc(&skb->users); |
a59322be | 238 | } else |
4934b032 | 239 | __skb_unlink(skb, queue); |
1da177e4 | 240 | |
3f518bf7 | 241 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
39cc8613 | 242 | *off = _off; |
1da177e4 | 243 | return skb; |
3f518bf7 PE |
244 | } |
245 | spin_unlock_irqrestore(&queue->lock, cpu_flags); | |
1da177e4 | 246 | |
cbf55001 ET |
247 | if (sk_can_busy_loop(sk) && |
248 | sk_busy_loop(sk, flags & MSG_DONTWAIT)) | |
a5b50476 ET |
249 | continue; |
250 | ||
1da177e4 LT |
251 | /* User doesn't want to wait */ |
252 | error = -EAGAIN; | |
253 | if (!timeo) | |
254 | goto no_packet; | |
255 | ||
39cc8613 | 256 | } while (!wait_for_more_packets(sk, err, &timeo, last)); |
1da177e4 LT |
257 | |
258 | return NULL; | |
259 | ||
738ac1eb HX |
260 | unlock_err: |
261 | spin_unlock_irqrestore(&queue->lock, cpu_flags); | |
1da177e4 LT |
262 | no_packet: |
263 | *err = error; | |
264 | return NULL; | |
265 | } | |
a59322be HX |
266 | EXPORT_SYMBOL(__skb_recv_datagram); |
267 | ||
95c96174 | 268 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
269 | int noblock, int *err) |
270 | { | |
3f518bf7 | 271 | int peeked, off = 0; |
a59322be HX |
272 | |
273 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
3f518bf7 | 274 | &peeked, &off, err); |
a59322be | 275 | } |
9e34a5b5 | 276 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
277 | |
278 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
279 | { | |
ead2ceb0 | 280 | consume_skb(skb); |
270acefa | 281 | sk_mem_reclaim_partial(sk); |
1da177e4 | 282 | } |
9d410c79 ED |
283 | EXPORT_SYMBOL(skb_free_datagram); |
284 | ||
285 | void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) | |
286 | { | |
8a74ad60 ED |
287 | bool slow; |
288 | ||
93bb64ea ED |
289 | if (likely(atomic_read(&skb->users) == 1)) |
290 | smp_rmb(); | |
291 | else if (likely(!atomic_dec_and_test(&skb->users))) | |
292 | return; | |
293 | ||
8a74ad60 | 294 | slow = lock_sock_fast(sk); |
4b0b72f7 ED |
295 | skb_orphan(skb); |
296 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 297 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 298 | |
93bb64ea ED |
299 | /* skb is now orphaned, can be freed outside of locked section */ |
300 | __kfree_skb(skb); | |
9d410c79 ED |
301 | } |
302 | EXPORT_SYMBOL(skb_free_datagram_locked); | |
1da177e4 | 303 | |
3305b80c HX |
304 | /** |
305 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
306 | * @sk: socket | |
307 | * @skb: datagram skbuff | |
308 | * @flags: MSG_ flags | |
309 | * | |
310 | * This function frees a datagram skbuff that was received by | |
311 | * skb_recv_datagram. The flags argument must match the one | |
312 | * used for skb_recv_datagram. | |
313 | * | |
314 | * If the MSG_PEEK flag is set, and the packet is still on the | |
315 | * receive queue of the socket, it will be taken off the queue | |
316 | * before it is freed. | |
317 | * | |
318 | * This function currently only disables BH when acquiring the | |
319 | * sk_receive_queue lock. Therefore it must not be used in a | |
320 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
321 | * |
322 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
323 | */ |
324 | ||
27ab2568 | 325 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 326 | { |
27ab2568 HX |
327 | int err = 0; |
328 | ||
3305b80c | 329 | if (flags & MSG_PEEK) { |
27ab2568 | 330 | err = -ENOENT; |
3305b80c HX |
331 | spin_lock_bh(&sk->sk_receive_queue.lock); |
332 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
333 | __skb_unlink(skb, &sk->sk_receive_queue); | |
334 | atomic_dec(&skb->users); | |
27ab2568 | 335 | err = 0; |
3305b80c HX |
336 | } |
337 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
338 | } | |
339 | ||
61de71c6 | 340 | kfree_skb(skb); |
8edf19c2 | 341 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
342 | sk_mem_reclaim_partial(sk); |
343 | ||
27ab2568 | 344 | return err; |
3305b80c | 345 | } |
3305b80c HX |
346 | EXPORT_SYMBOL(skb_kill_datagram); |
347 | ||
a8f820aa HX |
348 | /** |
349 | * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. | |
350 | * @skb: buffer to copy | |
351 | * @offset: offset in the buffer to start copying from | |
352 | * @to: iovec iterator to copy to | |
353 | * @len: amount of data to copy from buffer to iovec | |
354 | */ | |
355 | int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, | |
356 | struct iov_iter *to, int len) | |
357 | { | |
358 | int start = skb_headlen(skb); | |
359 | int i, copy = start - offset; | |
360 | struct sk_buff *frag_iter; | |
361 | ||
362 | trace_skb_copy_datagram_iovec(skb, len); | |
363 | ||
364 | /* Copy header. */ | |
365 | if (copy > 0) { | |
366 | if (copy > len) | |
367 | copy = len; | |
368 | if (copy_to_iter(skb->data + offset, copy, to) != copy) | |
369 | goto short_copy; | |
370 | if ((len -= copy) == 0) | |
371 | return 0; | |
372 | offset += copy; | |
373 | } | |
374 | ||
375 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
376 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
377 | int end; | |
378 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
379 | ||
380 | WARN_ON(start > offset + len); | |
381 | ||
382 | end = start + skb_frag_size(frag); | |
383 | if ((copy = end - offset) > 0) { | |
384 | if (copy > len) | |
385 | copy = len; | |
386 | if (copy_page_to_iter(skb_frag_page(frag), | |
387 | frag->page_offset + offset - | |
388 | start, copy, to) != copy) | |
389 | goto short_copy; | |
390 | if (!(len -= copy)) | |
391 | return 0; | |
392 | offset += copy; | |
393 | } | |
394 | start = end; | |
395 | } | |
396 | ||
397 | skb_walk_frags(skb, frag_iter) { | |
398 | int end; | |
399 | ||
400 | WARN_ON(start > offset + len); | |
401 | ||
402 | end = start + frag_iter->len; | |
403 | if ((copy = end - offset) > 0) { | |
404 | if (copy > len) | |
405 | copy = len; | |
406 | if (skb_copy_datagram_iter(frag_iter, offset - start, | |
407 | to, copy)) | |
408 | goto fault; | |
409 | if ((len -= copy) == 0) | |
410 | return 0; | |
411 | offset += copy; | |
412 | } | |
413 | start = end; | |
414 | } | |
415 | if (!len) | |
416 | return 0; | |
417 | ||
418 | /* This is not really a user copy fault, but rather someone | |
419 | * gave us a bogus length on the skb. We should probably | |
420 | * print a warning here as it may indicate a kernel bug. | |
421 | */ | |
422 | ||
423 | fault: | |
424 | return -EFAULT; | |
425 | ||
426 | short_copy: | |
427 | if (iov_iter_count(to)) | |
428 | goto fault; | |
429 | ||
430 | return 0; | |
431 | } | |
432 | EXPORT_SYMBOL(skb_copy_datagram_iter); | |
433 | ||
db543c1f | 434 | /** |
8feb2fb2 | 435 | * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. |
db543c1f RR |
436 | * @skb: buffer to copy |
437 | * @offset: offset in the buffer to start copying to | |
8feb2fb2 | 438 | * @from: the copy source |
db543c1f RR |
439 | * @len: amount of data to copy to buffer from iovec |
440 | * | |
441 | * Returns 0 or -EFAULT. | |
db543c1f | 442 | */ |
3a654f97 AV |
443 | int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, |
444 | struct iov_iter *from, | |
445 | int len) | |
446 | { | |
447 | int start = skb_headlen(skb); | |
448 | int i, copy = start - offset; | |
449 | struct sk_buff *frag_iter; | |
450 | ||
451 | /* Copy header. */ | |
452 | if (copy > 0) { | |
453 | if (copy > len) | |
454 | copy = len; | |
455 | if (copy_from_iter(skb->data + offset, copy, from) != copy) | |
456 | goto fault; | |
457 | if ((len -= copy) == 0) | |
458 | return 0; | |
459 | offset += copy; | |
460 | } | |
461 | ||
462 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
463 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
464 | int end; | |
465 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
466 | ||
467 | WARN_ON(start > offset + len); | |
468 | ||
469 | end = start + skb_frag_size(frag); | |
470 | if ((copy = end - offset) > 0) { | |
471 | size_t copied; | |
472 | ||
473 | if (copy > len) | |
474 | copy = len; | |
475 | copied = copy_page_from_iter(skb_frag_page(frag), | |
476 | frag->page_offset + offset - start, | |
477 | copy, from); | |
478 | if (copied != copy) | |
479 | goto fault; | |
480 | ||
481 | if (!(len -= copy)) | |
482 | return 0; | |
483 | offset += copy; | |
484 | } | |
485 | start = end; | |
486 | } | |
487 | ||
488 | skb_walk_frags(skb, frag_iter) { | |
489 | int end; | |
490 | ||
491 | WARN_ON(start > offset + len); | |
492 | ||
493 | end = start + frag_iter->len; | |
494 | if ((copy = end - offset) > 0) { | |
495 | if (copy > len) | |
496 | copy = len; | |
497 | if (skb_copy_datagram_from_iter(frag_iter, | |
498 | offset - start, | |
499 | from, copy)) | |
500 | goto fault; | |
501 | if ((len -= copy) == 0) | |
502 | return 0; | |
503 | offset += copy; | |
504 | } | |
505 | start = end; | |
506 | } | |
507 | if (!len) | |
508 | return 0; | |
509 | ||
510 | fault: | |
511 | return -EFAULT; | |
512 | } | |
513 | EXPORT_SYMBOL(skb_copy_datagram_from_iter); | |
514 | ||
c3bdeb5c | 515 | /** |
195e952d | 516 | * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter |
c3bdeb5c | 517 | * @skb: buffer to copy |
195e952d | 518 | * @from: the source to copy from |
c3bdeb5c JW |
519 | * |
520 | * The function will first copy up to headlen, and then pin the userspace | |
521 | * pages and build frags through them. | |
522 | * | |
523 | * Returns 0, -EFAULT or -EMSGSIZE. | |
c3bdeb5c | 524 | */ |
3a654f97 AV |
525 | int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) |
526 | { | |
527 | int len = iov_iter_count(from); | |
528 | int copy = min_t(int, skb_headlen(skb), len); | |
529 | int frag = 0; | |
530 | ||
531 | /* copy up to skb headlen */ | |
532 | if (skb_copy_datagram_from_iter(skb, 0, from, copy)) | |
533 | return -EFAULT; | |
534 | ||
535 | while (iov_iter_count(from)) { | |
536 | struct page *pages[MAX_SKB_FRAGS]; | |
537 | size_t start; | |
538 | ssize_t copied; | |
539 | unsigned long truesize; | |
540 | int n = 0; | |
541 | ||
542 | if (frag == MAX_SKB_FRAGS) | |
543 | return -EMSGSIZE; | |
544 | ||
545 | copied = iov_iter_get_pages(from, pages, ~0U, | |
546 | MAX_SKB_FRAGS - frag, &start); | |
547 | if (copied < 0) | |
548 | return -EFAULT; | |
549 | ||
550 | iov_iter_advance(from, copied); | |
551 | ||
552 | truesize = PAGE_ALIGN(copied + start); | |
553 | skb->data_len += copied; | |
554 | skb->len += copied; | |
555 | skb->truesize += truesize; | |
556 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
557 | while (copied) { | |
558 | int size = min_t(int, copied, PAGE_SIZE - start); | |
559 | skb_fill_page_desc(skb, frag++, pages[n], start, size); | |
560 | start = 0; | |
561 | copied -= size; | |
562 | n++; | |
563 | } | |
564 | } | |
565 | return 0; | |
566 | } | |
567 | EXPORT_SYMBOL(zerocopy_sg_from_iter); | |
568 | ||
1da177e4 | 569 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
e5a4b0bb | 570 | struct iov_iter *to, int len, |
5084205f | 571 | __wsum *csump) |
1da177e4 | 572 | { |
1a028e50 | 573 | int start = skb_headlen(skb); |
1a028e50 | 574 | int i, copy = start - offset; |
5b1a002a DM |
575 | struct sk_buff *frag_iter; |
576 | int pos = 0; | |
e5a4b0bb | 577 | int n; |
1da177e4 LT |
578 | |
579 | /* Copy header. */ | |
580 | if (copy > 0) { | |
1da177e4 LT |
581 | if (copy > len) |
582 | copy = len; | |
e5a4b0bb AV |
583 | n = csum_and_copy_to_iter(skb->data + offset, copy, csump, to); |
584 | if (n != copy) | |
1da177e4 LT |
585 | goto fault; |
586 | if ((len -= copy) == 0) | |
587 | return 0; | |
588 | offset += copy; | |
1da177e4 LT |
589 | pos = copy; |
590 | } | |
591 | ||
592 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 593 | int end; |
9e903e08 | 594 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 595 | |
547b792c | 596 | WARN_ON(start > offset + len); |
1a028e50 | 597 | |
9e903e08 | 598 | end = start + skb_frag_size(frag); |
1da177e4 | 599 | if ((copy = end - offset) > 0) { |
e5a4b0bb | 600 | __wsum csum2 = 0; |
ea2ab693 | 601 | struct page *page = skb_frag_page(frag); |
e5a4b0bb | 602 | u8 *vaddr = kmap(page); |
1da177e4 LT |
603 | |
604 | if (copy > len) | |
605 | copy = len; | |
e5a4b0bb AV |
606 | n = csum_and_copy_to_iter(vaddr + frag->page_offset + |
607 | offset - start, copy, | |
608 | &csum2, to); | |
1da177e4 | 609 | kunmap(page); |
e5a4b0bb | 610 | if (n != copy) |
1da177e4 LT |
611 | goto fault; |
612 | *csump = csum_block_add(*csump, csum2, pos); | |
613 | if (!(len -= copy)) | |
614 | return 0; | |
615 | offset += copy; | |
1da177e4 LT |
616 | pos += copy; |
617 | } | |
1a028e50 | 618 | start = end; |
1da177e4 LT |
619 | } |
620 | ||
5b1a002a DM |
621 | skb_walk_frags(skb, frag_iter) { |
622 | int end; | |
623 | ||
624 | WARN_ON(start > offset + len); | |
625 | ||
626 | end = start + frag_iter->len; | |
627 | if ((copy = end - offset) > 0) { | |
628 | __wsum csum2 = 0; | |
629 | if (copy > len) | |
630 | copy = len; | |
631 | if (skb_copy_and_csum_datagram(frag_iter, | |
632 | offset - start, | |
633 | to, copy, | |
634 | &csum2)) | |
635 | goto fault; | |
636 | *csump = csum_block_add(*csump, csum2, pos); | |
637 | if ((len -= copy) == 0) | |
638 | return 0; | |
639 | offset += copy; | |
5b1a002a | 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)); |
46fb51eb TH |
656 | if (likely(!sum)) { |
657 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
658 | !skb->csum_complete_sw) | |
659 | netdev_rx_csum_fault(skb->dev); | |
660 | } | |
89c22d8c HX |
661 | if (!skb_shared(skb)) |
662 | skb->csum_valid = !sum; | |
fb286bb2 HX |
663 | return sum; |
664 | } | |
759e5d00 HX |
665 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
666 | ||
667 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
668 | { | |
46fb51eb TH |
669 | __wsum csum; |
670 | __sum16 sum; | |
671 | ||
672 | csum = skb_checksum(skb, 0, skb->len, 0); | |
673 | ||
674 | /* skb->csum holds pseudo checksum */ | |
675 | sum = csum_fold(csum_add(skb->csum, csum)); | |
676 | if (likely(!sum)) { | |
677 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
678 | !skb->csum_complete_sw) | |
679 | netdev_rx_csum_fault(skb->dev); | |
680 | } | |
681 | ||
89c22d8c HX |
682 | if (!skb_shared(skb)) { |
683 | /* Save full packet checksum */ | |
684 | skb->csum = csum; | |
685 | skb->ip_summed = CHECKSUM_COMPLETE; | |
686 | skb->csum_complete_sw = 1; | |
687 | skb->csum_valid = !sum; | |
688 | } | |
46fb51eb TH |
689 | |
690 | return sum; | |
759e5d00 | 691 | } |
fb286bb2 HX |
692 | EXPORT_SYMBOL(__skb_checksum_complete); |
693 | ||
1da177e4 | 694 | /** |
e5a4b0bb | 695 | * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. |
4dc3b16b PP |
696 | * @skb: skbuff |
697 | * @hlen: hardware length | |
e5a4b0bb | 698 | * @msg: destination |
4ec93edb | 699 | * |
1da177e4 LT |
700 | * Caller _must_ check that skb will fit to this iovec. |
701 | * | |
702 | * Returns: 0 - success. | |
703 | * -EINVAL - checksum failure. | |
e5a4b0bb | 704 | * -EFAULT - fault during copy. |
1da177e4 | 705 | */ |
e5a4b0bb AV |
706 | int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, |
707 | int hlen, struct msghdr *msg) | |
1da177e4 | 708 | { |
d3bc23e7 | 709 | __wsum csum; |
1da177e4 LT |
710 | int chunk = skb->len - hlen; |
711 | ||
ef8aef55 HX |
712 | if (!chunk) |
713 | return 0; | |
714 | ||
01e97e65 | 715 | if (msg_data_left(msg) < chunk) { |
fb286bb2 | 716 | if (__skb_checksum_complete(skb)) |
1da177e4 | 717 | goto csum_error; |
e5a4b0bb | 718 | if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) |
1da177e4 LT |
719 | goto fault; |
720 | } else { | |
721 | csum = csum_partial(skb->data, hlen, skb->csum); | |
e5a4b0bb | 722 | if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, |
1da177e4 LT |
723 | chunk, &csum)) |
724 | goto fault; | |
d3bc23e7 | 725 | if (csum_fold(csum)) |
1da177e4 | 726 | goto csum_error; |
84fa7933 | 727 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 728 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
729 | } |
730 | return 0; | |
731 | csum_error: | |
732 | return -EINVAL; | |
733 | fault: | |
734 | return -EFAULT; | |
735 | } | |
e5a4b0bb | 736 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); |
1da177e4 LT |
737 | |
738 | /** | |
739 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
740 | * @file: file struct |
741 | * @sock: socket | |
742 | * @wait: poll table | |
1da177e4 LT |
743 | * |
744 | * Datagram poll: Again totally generic. This also handles | |
745 | * sequenced packet sockets providing the socket receive queue | |
746 | * is only ever holding data ready to receive. | |
747 | * | |
748 | * Note: when you _don't_ use this routine for this protocol, | |
749 | * and you use a different write policy from sock_writeable() | |
750 | * then please supply your own write_space callback. | |
751 | */ | |
752 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
753 | poll_table *wait) | |
754 | { | |
755 | struct sock *sk = sock->sk; | |
756 | unsigned int mask; | |
757 | ||
aa395145 | 758 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
759 | mask = 0; |
760 | ||
761 | /* exceptional events? */ | |
762 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
7d4c04fc | 763 | mask |= POLLERR | |
8facd5fb | 764 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0); |
7d4c04fc | 765 | |
f348d70a | 766 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 767 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
768 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
769 | mask |= POLLHUP; | |
770 | ||
771 | /* readable? */ | |
db40980f | 772 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
773 | mask |= POLLIN | POLLRDNORM; |
774 | ||
775 | /* Connection-based need to check for termination and startup */ | |
776 | if (connection_based(sk)) { | |
777 | if (sk->sk_state == TCP_CLOSE) | |
778 | mask |= POLLHUP; | |
779 | /* connection hasn't started yet? */ | |
780 | if (sk->sk_state == TCP_SYN_SENT) | |
781 | return mask; | |
782 | } | |
783 | ||
784 | /* writable? */ | |
785 | if (sock_writeable(sk)) | |
786 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
787 | else | |
9cd3e072 | 788 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
789 | |
790 | return mask; | |
791 | } | |
1da177e4 | 792 | EXPORT_SYMBOL(datagram_poll); |