]>
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 | */ |
ea3793ee RW |
86 | int __skb_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 | } | |
ea3793ee | 133 | EXPORT_SYMBOL(__skb_wait_for_more_packets); |
1da177e4 | 134 | |
a0a2a660 | 135 | static struct sk_buff *skb_set_peeked(struct sk_buff *skb) |
738ac1eb HX |
136 | { |
137 | struct sk_buff *nskb; | |
138 | ||
139 | if (skb->peeked) | |
a0a2a660 | 140 | return skb; |
738ac1eb HX |
141 | |
142 | /* We have to unshare an skb before modifying it. */ | |
143 | if (!skb_shared(skb)) | |
144 | goto done; | |
145 | ||
146 | nskb = skb_clone(skb, GFP_ATOMIC); | |
147 | if (!nskb) | |
a0a2a660 | 148 | return ERR_PTR(-ENOMEM); |
738ac1eb HX |
149 | |
150 | skb->prev->next = nskb; | |
151 | skb->next->prev = nskb; | |
152 | nskb->prev = skb->prev; | |
153 | nskb->next = skb->next; | |
154 | ||
155 | consume_skb(skb); | |
156 | skb = nskb; | |
157 | ||
158 | done: | |
159 | skb->peeked = 1; | |
160 | ||
a0a2a660 | 161 | return skb; |
738ac1eb HX |
162 | } |
163 | ||
1da177e4 | 164 | /** |
ea3793ee | 165 | * __skb_try_recv_datagram - Receive a datagram skbuff |
4dc3b16b PP |
166 | * @sk: socket |
167 | * @flags: MSG_ flags | |
39cc8613 | 168 | * @peeked: returns non-zero if this packet has been seen before |
3f518bf7 PE |
169 | * @off: an offset in bytes to peek skb from. Returns an offset |
170 | * within an skb where data actually starts | |
4dc3b16b | 171 | * @err: error code returned |
ea3793ee RW |
172 | * @last: set to last peeked message to inform the wait function |
173 | * what to look for when peeking | |
1da177e4 LT |
174 | * |
175 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
176 | * and possible races. This replaces identical code in packet, raw and | |
177 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
178 | * the long standing peek and read race for datagram sockets. If you | |
179 | * alter this routine remember it must be re-entrant. | |
180 | * | |
ea3793ee RW |
181 | * This function will lock the socket if a skb is returned, so |
182 | * the caller needs to unlock the socket in that case (usually by | |
183 | * calling skb_free_datagram). Returns NULL with *err set to | |
184 | * -EAGAIN if no data was available or to some other value if an | |
185 | * error was detected. | |
1da177e4 LT |
186 | * |
187 | * * It does not lock socket since today. This function is | |
188 | * * free of race conditions. This measure should/can improve | |
189 | * * significantly datagram socket latencies at high loads, | |
190 | * * when data copying to user space takes lots of time. | |
191 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
192 | * * 8) Great win.) | |
193 | * * --ANK (980729) | |
194 | * | |
195 | * The order of the tests when we find no data waiting are specified | |
196 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
197 | * the standard around please. | |
198 | */ | |
ea3793ee RW |
199 | struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags, |
200 | int *peeked, int *off, int *err, | |
201 | struct sk_buff **last) | |
1da177e4 | 202 | { |
738ac1eb | 203 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
ea3793ee | 204 | struct sk_buff *skb; |
738ac1eb | 205 | unsigned long cpu_flags; |
1da177e4 LT |
206 | /* |
207 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
208 | */ | |
209 | int error = sock_error(sk); | |
210 | ||
211 | if (error) | |
212 | goto no_packet; | |
213 | ||
1da177e4 LT |
214 | do { |
215 | /* Again only user level code calls this function, so nothing | |
216 | * interrupt level will suddenly eat the receive_queue. | |
217 | * | |
218 | * Look at current nfs client by the way... | |
8917a3c0 | 219 | * However, this function was correct in any case. 8) |
1da177e4 | 220 | */ |
39cc8613 | 221 | int _off = *off; |
a59322be | 222 | |
ea3793ee | 223 | *last = (struct sk_buff *)queue; |
4934b032 | 224 | spin_lock_irqsave(&queue->lock, cpu_flags); |
3f518bf7 | 225 | skb_queue_walk(queue, skb) { |
ea3793ee | 226 | *last = skb; |
a59322be HX |
227 | *peeked = skb->peeked; |
228 | if (flags & MSG_PEEK) { | |
39cc8613 | 229 | if (_off >= skb->len && (skb->len || _off || |
add05ad4 | 230 | skb->peeked)) { |
39cc8613 | 231 | _off -= skb->len; |
3f518bf7 PE |
232 | continue; |
233 | } | |
738ac1eb | 234 | |
a0a2a660 HX |
235 | skb = skb_set_peeked(skb); |
236 | error = PTR_ERR(skb); | |
ea3793ee RW |
237 | if (IS_ERR(skb)) { |
238 | spin_unlock_irqrestore(&queue->lock, | |
239 | cpu_flags); | |
240 | goto no_packet; | |
241 | } | |
738ac1eb | 242 | |
1da177e4 | 243 | atomic_inc(&skb->users); |
a59322be | 244 | } else |
4934b032 | 245 | __skb_unlink(skb, queue); |
1da177e4 | 246 | |
3f518bf7 | 247 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
39cc8613 | 248 | *off = _off; |
1da177e4 | 249 | return skb; |
3f518bf7 | 250 | } |
ea3793ee | 251 | |
3f518bf7 | 252 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
ea3793ee RW |
253 | } while (sk_can_busy_loop(sk) && |
254 | sk_busy_loop(sk, flags & MSG_DONTWAIT)); | |
1da177e4 | 255 | |
ea3793ee | 256 | error = -EAGAIN; |
a5b50476 | 257 | |
ea3793ee RW |
258 | no_packet: |
259 | *err = error; | |
260 | return NULL; | |
261 | } | |
262 | EXPORT_SYMBOL(__skb_try_recv_datagram); | |
1da177e4 | 263 | |
ea3793ee RW |
264 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
265 | int *peeked, int *off, int *err) | |
266 | { | |
267 | struct sk_buff *skb, *last; | |
268 | long timeo; | |
1da177e4 | 269 | |
ea3793ee RW |
270 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
271 | ||
272 | do { | |
273 | skb = __skb_try_recv_datagram(sk, flags, peeked, off, err, | |
274 | &last); | |
275 | if (skb) | |
276 | return skb; | |
277 | ||
760a4322 | 278 | if (*err != -EAGAIN) |
ea3793ee RW |
279 | break; |
280 | } while (timeo && | |
281 | !__skb_wait_for_more_packets(sk, err, &timeo, last)); | |
1da177e4 | 282 | |
1da177e4 LT |
283 | return NULL; |
284 | } | |
a59322be HX |
285 | EXPORT_SYMBOL(__skb_recv_datagram); |
286 | ||
95c96174 | 287 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
288 | int noblock, int *err) |
289 | { | |
3f518bf7 | 290 | int peeked, off = 0; |
a59322be HX |
291 | |
292 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
3f518bf7 | 293 | &peeked, &off, err); |
a59322be | 294 | } |
9e34a5b5 | 295 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
296 | |
297 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
298 | { | |
ead2ceb0 | 299 | consume_skb(skb); |
270acefa | 300 | sk_mem_reclaim_partial(sk); |
1da177e4 | 301 | } |
9d410c79 ED |
302 | EXPORT_SYMBOL(skb_free_datagram); |
303 | ||
627d2d6b | 304 | void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len) |
9d410c79 | 305 | { |
8a74ad60 ED |
306 | bool slow; |
307 | ||
93bb64ea ED |
308 | if (likely(atomic_read(&skb->users) == 1)) |
309 | smp_rmb(); | |
627d2d6b | 310 | else if (likely(!atomic_dec_and_test(&skb->users))) { |
311 | sk_peek_offset_bwd(sk, len); | |
93bb64ea | 312 | return; |
627d2d6b | 313 | } |
93bb64ea | 314 | |
8a74ad60 | 315 | slow = lock_sock_fast(sk); |
627d2d6b | 316 | sk_peek_offset_bwd(sk, len); |
4b0b72f7 ED |
317 | skb_orphan(skb); |
318 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 319 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 320 | |
93bb64ea ED |
321 | /* skb is now orphaned, can be freed outside of locked section */ |
322 | __kfree_skb(skb); | |
9d410c79 | 323 | } |
627d2d6b | 324 | EXPORT_SYMBOL(__skb_free_datagram_locked); |
1da177e4 | 325 | |
3305b80c HX |
326 | /** |
327 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
328 | * @sk: socket | |
329 | * @skb: datagram skbuff | |
330 | * @flags: MSG_ flags | |
331 | * | |
332 | * This function frees a datagram skbuff that was received by | |
333 | * skb_recv_datagram. The flags argument must match the one | |
334 | * used for skb_recv_datagram. | |
335 | * | |
336 | * If the MSG_PEEK flag is set, and the packet is still on the | |
337 | * receive queue of the socket, it will be taken off the queue | |
338 | * before it is freed. | |
339 | * | |
340 | * This function currently only disables BH when acquiring the | |
341 | * sk_receive_queue lock. Therefore it must not be used in a | |
342 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
343 | * |
344 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
345 | */ |
346 | ||
27ab2568 | 347 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 348 | { |
27ab2568 HX |
349 | int err = 0; |
350 | ||
3305b80c | 351 | if (flags & MSG_PEEK) { |
27ab2568 | 352 | err = -ENOENT; |
3305b80c HX |
353 | spin_lock_bh(&sk->sk_receive_queue.lock); |
354 | if (skb == skb_peek(&sk->sk_receive_queue)) { | |
355 | __skb_unlink(skb, &sk->sk_receive_queue); | |
356 | atomic_dec(&skb->users); | |
27ab2568 | 357 | err = 0; |
3305b80c HX |
358 | } |
359 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
360 | } | |
361 | ||
61de71c6 | 362 | kfree_skb(skb); |
8edf19c2 | 363 | atomic_inc(&sk->sk_drops); |
61de71c6 JD |
364 | sk_mem_reclaim_partial(sk); |
365 | ||
27ab2568 | 366 | return err; |
3305b80c | 367 | } |
3305b80c HX |
368 | EXPORT_SYMBOL(skb_kill_datagram); |
369 | ||
a8f820aa HX |
370 | /** |
371 | * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. | |
372 | * @skb: buffer to copy | |
373 | * @offset: offset in the buffer to start copying from | |
374 | * @to: iovec iterator to copy to | |
375 | * @len: amount of data to copy from buffer to iovec | |
376 | */ | |
377 | int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, | |
378 | struct iov_iter *to, int len) | |
379 | { | |
380 | int start = skb_headlen(skb); | |
381 | int i, copy = start - offset; | |
382 | struct sk_buff *frag_iter; | |
383 | ||
384 | trace_skb_copy_datagram_iovec(skb, len); | |
385 | ||
386 | /* Copy header. */ | |
387 | if (copy > 0) { | |
388 | if (copy > len) | |
389 | copy = len; | |
390 | if (copy_to_iter(skb->data + offset, copy, to) != copy) | |
391 | goto short_copy; | |
392 | if ((len -= copy) == 0) | |
393 | return 0; | |
394 | offset += copy; | |
395 | } | |
396 | ||
397 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
398 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
399 | int end; | |
400 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
401 | ||
402 | WARN_ON(start > offset + len); | |
403 | ||
404 | end = start + skb_frag_size(frag); | |
405 | if ((copy = end - offset) > 0) { | |
406 | if (copy > len) | |
407 | copy = len; | |
408 | if (copy_page_to_iter(skb_frag_page(frag), | |
409 | frag->page_offset + offset - | |
410 | start, copy, to) != copy) | |
411 | goto short_copy; | |
412 | if (!(len -= copy)) | |
413 | return 0; | |
414 | offset += copy; | |
415 | } | |
416 | start = end; | |
417 | } | |
418 | ||
419 | skb_walk_frags(skb, frag_iter) { | |
420 | int end; | |
421 | ||
422 | WARN_ON(start > offset + len); | |
423 | ||
424 | end = start + frag_iter->len; | |
425 | if ((copy = end - offset) > 0) { | |
426 | if (copy > len) | |
427 | copy = len; | |
428 | if (skb_copy_datagram_iter(frag_iter, offset - start, | |
429 | to, copy)) | |
430 | goto fault; | |
431 | if ((len -= copy) == 0) | |
432 | return 0; | |
433 | offset += copy; | |
434 | } | |
435 | start = end; | |
436 | } | |
437 | if (!len) | |
438 | return 0; | |
439 | ||
440 | /* This is not really a user copy fault, but rather someone | |
441 | * gave us a bogus length on the skb. We should probably | |
442 | * print a warning here as it may indicate a kernel bug. | |
443 | */ | |
444 | ||
445 | fault: | |
446 | return -EFAULT; | |
447 | ||
448 | short_copy: | |
449 | if (iov_iter_count(to)) | |
450 | goto fault; | |
451 | ||
452 | return 0; | |
453 | } | |
454 | EXPORT_SYMBOL(skb_copy_datagram_iter); | |
455 | ||
db543c1f | 456 | /** |
8feb2fb2 | 457 | * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. |
db543c1f RR |
458 | * @skb: buffer to copy |
459 | * @offset: offset in the buffer to start copying to | |
8feb2fb2 | 460 | * @from: the copy source |
db543c1f RR |
461 | * @len: amount of data to copy to buffer from iovec |
462 | * | |
463 | * Returns 0 or -EFAULT. | |
db543c1f | 464 | */ |
3a654f97 AV |
465 | int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, |
466 | struct iov_iter *from, | |
467 | int len) | |
468 | { | |
469 | int start = skb_headlen(skb); | |
470 | int i, copy = start - offset; | |
471 | struct sk_buff *frag_iter; | |
472 | ||
473 | /* Copy header. */ | |
474 | if (copy > 0) { | |
475 | if (copy > len) | |
476 | copy = len; | |
477 | if (copy_from_iter(skb->data + offset, copy, from) != copy) | |
478 | goto fault; | |
479 | if ((len -= copy) == 0) | |
480 | return 0; | |
481 | offset += copy; | |
482 | } | |
483 | ||
484 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
485 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
486 | int end; | |
487 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
488 | ||
489 | WARN_ON(start > offset + len); | |
490 | ||
491 | end = start + skb_frag_size(frag); | |
492 | if ((copy = end - offset) > 0) { | |
493 | size_t copied; | |
494 | ||
495 | if (copy > len) | |
496 | copy = len; | |
497 | copied = copy_page_from_iter(skb_frag_page(frag), | |
498 | frag->page_offset + offset - start, | |
499 | copy, from); | |
500 | if (copied != copy) | |
501 | goto fault; | |
502 | ||
503 | if (!(len -= copy)) | |
504 | return 0; | |
505 | offset += copy; | |
506 | } | |
507 | start = end; | |
508 | } | |
509 | ||
510 | skb_walk_frags(skb, frag_iter) { | |
511 | int end; | |
512 | ||
513 | WARN_ON(start > offset + len); | |
514 | ||
515 | end = start + frag_iter->len; | |
516 | if ((copy = end - offset) > 0) { | |
517 | if (copy > len) | |
518 | copy = len; | |
519 | if (skb_copy_datagram_from_iter(frag_iter, | |
520 | offset - start, | |
521 | from, copy)) | |
522 | goto fault; | |
523 | if ((len -= copy) == 0) | |
524 | return 0; | |
525 | offset += copy; | |
526 | } | |
527 | start = end; | |
528 | } | |
529 | if (!len) | |
530 | return 0; | |
531 | ||
532 | fault: | |
533 | return -EFAULT; | |
534 | } | |
535 | EXPORT_SYMBOL(skb_copy_datagram_from_iter); | |
536 | ||
c3bdeb5c | 537 | /** |
195e952d | 538 | * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter |
c3bdeb5c | 539 | * @skb: buffer to copy |
195e952d | 540 | * @from: the source to copy from |
c3bdeb5c JW |
541 | * |
542 | * The function will first copy up to headlen, and then pin the userspace | |
543 | * pages and build frags through them. | |
544 | * | |
545 | * Returns 0, -EFAULT or -EMSGSIZE. | |
c3bdeb5c | 546 | */ |
3a654f97 AV |
547 | int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) |
548 | { | |
549 | int len = iov_iter_count(from); | |
550 | int copy = min_t(int, skb_headlen(skb), len); | |
551 | int frag = 0; | |
552 | ||
553 | /* copy up to skb headlen */ | |
554 | if (skb_copy_datagram_from_iter(skb, 0, from, copy)) | |
555 | return -EFAULT; | |
556 | ||
557 | while (iov_iter_count(from)) { | |
558 | struct page *pages[MAX_SKB_FRAGS]; | |
559 | size_t start; | |
560 | ssize_t copied; | |
561 | unsigned long truesize; | |
562 | int n = 0; | |
563 | ||
564 | if (frag == MAX_SKB_FRAGS) | |
565 | return -EMSGSIZE; | |
566 | ||
567 | copied = iov_iter_get_pages(from, pages, ~0U, | |
568 | MAX_SKB_FRAGS - frag, &start); | |
569 | if (copied < 0) | |
570 | return -EFAULT; | |
571 | ||
572 | iov_iter_advance(from, copied); | |
573 | ||
574 | truesize = PAGE_ALIGN(copied + start); | |
575 | skb->data_len += copied; | |
576 | skb->len += copied; | |
577 | skb->truesize += truesize; | |
578 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
579 | while (copied) { | |
580 | int size = min_t(int, copied, PAGE_SIZE - start); | |
581 | skb_fill_page_desc(skb, frag++, pages[n], start, size); | |
582 | start = 0; | |
583 | copied -= size; | |
584 | n++; | |
585 | } | |
586 | } | |
587 | return 0; | |
588 | } | |
589 | EXPORT_SYMBOL(zerocopy_sg_from_iter); | |
590 | ||
1da177e4 | 591 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
e5a4b0bb | 592 | struct iov_iter *to, int len, |
5084205f | 593 | __wsum *csump) |
1da177e4 | 594 | { |
1a028e50 | 595 | int start = skb_headlen(skb); |
1a028e50 | 596 | int i, copy = start - offset; |
5b1a002a DM |
597 | struct sk_buff *frag_iter; |
598 | int pos = 0; | |
e5a4b0bb | 599 | int n; |
1da177e4 LT |
600 | |
601 | /* Copy header. */ | |
602 | if (copy > 0) { | |
1da177e4 LT |
603 | if (copy > len) |
604 | copy = len; | |
e5a4b0bb AV |
605 | n = csum_and_copy_to_iter(skb->data + offset, copy, csump, to); |
606 | if (n != copy) | |
1da177e4 LT |
607 | goto fault; |
608 | if ((len -= copy) == 0) | |
609 | return 0; | |
610 | offset += copy; | |
1da177e4 LT |
611 | pos = copy; |
612 | } | |
613 | ||
614 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 615 | int end; |
9e903e08 | 616 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 617 | |
547b792c | 618 | WARN_ON(start > offset + len); |
1a028e50 | 619 | |
9e903e08 | 620 | end = start + skb_frag_size(frag); |
1da177e4 | 621 | if ((copy = end - offset) > 0) { |
e5a4b0bb | 622 | __wsum csum2 = 0; |
ea2ab693 | 623 | struct page *page = skb_frag_page(frag); |
e5a4b0bb | 624 | u8 *vaddr = kmap(page); |
1da177e4 LT |
625 | |
626 | if (copy > len) | |
627 | copy = len; | |
e5a4b0bb AV |
628 | n = csum_and_copy_to_iter(vaddr + frag->page_offset + |
629 | offset - start, copy, | |
630 | &csum2, to); | |
1da177e4 | 631 | kunmap(page); |
e5a4b0bb | 632 | if (n != copy) |
1da177e4 LT |
633 | goto fault; |
634 | *csump = csum_block_add(*csump, csum2, pos); | |
635 | if (!(len -= copy)) | |
636 | return 0; | |
637 | offset += copy; | |
1da177e4 LT |
638 | pos += copy; |
639 | } | |
1a028e50 | 640 | start = end; |
1da177e4 LT |
641 | } |
642 | ||
5b1a002a DM |
643 | skb_walk_frags(skb, frag_iter) { |
644 | int end; | |
645 | ||
646 | WARN_ON(start > offset + len); | |
647 | ||
648 | end = start + frag_iter->len; | |
649 | if ((copy = end - offset) > 0) { | |
650 | __wsum csum2 = 0; | |
651 | if (copy > len) | |
652 | copy = len; | |
653 | if (skb_copy_and_csum_datagram(frag_iter, | |
654 | offset - start, | |
655 | to, copy, | |
656 | &csum2)) | |
657 | goto fault; | |
658 | *csump = csum_block_add(*csump, csum2, pos); | |
659 | if ((len -= copy) == 0) | |
660 | return 0; | |
661 | offset += copy; | |
5b1a002a | 662 | pos += copy; |
1da177e4 | 663 | } |
5b1a002a | 664 | start = end; |
1da177e4 LT |
665 | } |
666 | if (!len) | |
667 | return 0; | |
668 | ||
669 | fault: | |
670 | return -EFAULT; | |
671 | } | |
672 | ||
759e5d00 | 673 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 674 | { |
d3bc23e7 | 675 | __sum16 sum; |
fb286bb2 | 676 | |
759e5d00 | 677 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
46fb51eb TH |
678 | if (likely(!sum)) { |
679 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
680 | !skb->csum_complete_sw) | |
681 | netdev_rx_csum_fault(skb->dev); | |
682 | } | |
89c22d8c HX |
683 | if (!skb_shared(skb)) |
684 | skb->csum_valid = !sum; | |
fb286bb2 HX |
685 | return sum; |
686 | } | |
759e5d00 HX |
687 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
688 | ||
689 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
690 | { | |
46fb51eb TH |
691 | __wsum csum; |
692 | __sum16 sum; | |
693 | ||
694 | csum = skb_checksum(skb, 0, skb->len, 0); | |
695 | ||
696 | /* skb->csum holds pseudo checksum */ | |
697 | sum = csum_fold(csum_add(skb->csum, csum)); | |
698 | if (likely(!sum)) { | |
699 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
700 | !skb->csum_complete_sw) | |
701 | netdev_rx_csum_fault(skb->dev); | |
702 | } | |
703 | ||
89c22d8c HX |
704 | if (!skb_shared(skb)) { |
705 | /* Save full packet checksum */ | |
706 | skb->csum = csum; | |
707 | skb->ip_summed = CHECKSUM_COMPLETE; | |
708 | skb->csum_complete_sw = 1; | |
709 | skb->csum_valid = !sum; | |
710 | } | |
46fb51eb TH |
711 | |
712 | return sum; | |
759e5d00 | 713 | } |
fb286bb2 HX |
714 | EXPORT_SYMBOL(__skb_checksum_complete); |
715 | ||
1da177e4 | 716 | /** |
e5a4b0bb | 717 | * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. |
4dc3b16b PP |
718 | * @skb: skbuff |
719 | * @hlen: hardware length | |
e5a4b0bb | 720 | * @msg: destination |
4ec93edb | 721 | * |
1da177e4 LT |
722 | * Caller _must_ check that skb will fit to this iovec. |
723 | * | |
724 | * Returns: 0 - success. | |
725 | * -EINVAL - checksum failure. | |
e5a4b0bb | 726 | * -EFAULT - fault during copy. |
1da177e4 | 727 | */ |
e5a4b0bb AV |
728 | int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, |
729 | int hlen, struct msghdr *msg) | |
1da177e4 | 730 | { |
d3bc23e7 | 731 | __wsum csum; |
1da177e4 LT |
732 | int chunk = skb->len - hlen; |
733 | ||
ef8aef55 HX |
734 | if (!chunk) |
735 | return 0; | |
736 | ||
01e97e65 | 737 | if (msg_data_left(msg) < chunk) { |
fb286bb2 | 738 | if (__skb_checksum_complete(skb)) |
1da177e4 | 739 | goto csum_error; |
e5a4b0bb | 740 | if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) |
1da177e4 LT |
741 | goto fault; |
742 | } else { | |
743 | csum = csum_partial(skb->data, hlen, skb->csum); | |
e5a4b0bb | 744 | if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, |
1da177e4 LT |
745 | chunk, &csum)) |
746 | goto fault; | |
d3bc23e7 | 747 | if (csum_fold(csum)) |
1da177e4 | 748 | goto csum_error; |
84fa7933 | 749 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 750 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
751 | } |
752 | return 0; | |
753 | csum_error: | |
754 | return -EINVAL; | |
755 | fault: | |
756 | return -EFAULT; | |
757 | } | |
e5a4b0bb | 758 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); |
1da177e4 LT |
759 | |
760 | /** | |
761 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
762 | * @file: file struct |
763 | * @sock: socket | |
764 | * @wait: poll table | |
1da177e4 LT |
765 | * |
766 | * Datagram poll: Again totally generic. This also handles | |
767 | * sequenced packet sockets providing the socket receive queue | |
768 | * is only ever holding data ready to receive. | |
769 | * | |
770 | * Note: when you _don't_ use this routine for this protocol, | |
771 | * and you use a different write policy from sock_writeable() | |
772 | * then please supply your own write_space callback. | |
773 | */ | |
774 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
775 | poll_table *wait) | |
776 | { | |
777 | struct sock *sk = sock->sk; | |
778 | unsigned int mask; | |
779 | ||
aa395145 | 780 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
781 | mask = 0; |
782 | ||
783 | /* exceptional events? */ | |
784 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
7d4c04fc | 785 | mask |= POLLERR | |
8facd5fb | 786 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0); |
7d4c04fc | 787 | |
f348d70a | 788 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 789 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
790 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
791 | mask |= POLLHUP; | |
792 | ||
793 | /* readable? */ | |
db40980f | 794 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
795 | mask |= POLLIN | POLLRDNORM; |
796 | ||
797 | /* Connection-based need to check for termination and startup */ | |
798 | if (connection_based(sk)) { | |
799 | if (sk->sk_state == TCP_CLOSE) | |
800 | mask |= POLLHUP; | |
801 | /* connection hasn't started yet? */ | |
802 | if (sk->sk_state == TCP_SYN_SENT) | |
803 | return mask; | |
804 | } | |
805 | ||
806 | /* writable? */ | |
807 | if (sock_writeable(sk)) | |
808 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
809 | else | |
9cd3e072 | 810 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
811 | |
812 | return mask; | |
813 | } | |
1da177e4 | 814 | EXPORT_SYMBOL(datagram_poll); |