Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * SUCS NET3: | |
4 | * | |
5 | * Generic datagram handling routines. These are generic for all | |
6 | * protocols. Possibly a generic IP version on top of these would | |
7 | * make sense. Not tonight however 8-). | |
8 | * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and | |
9 | * NetROM layer all have identical poll code and mostly | |
10 | * identical recvmsg() code. So we share it here. The poll was | |
11 | * shared before but buried in udp.c so I moved it. | |
12 | * | |
113aa838 | 13 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old |
1da177e4 LT |
14 | * udp.c code) |
15 | * | |
16 | * Fixes: | |
17 | * Alan Cox : NULL return from skb_peek_copy() | |
18 | * understood | |
19 | * Alan Cox : Rewrote skb_read_datagram to avoid the | |
20 | * skb_peek_copy stuff. | |
21 | * Alan Cox : Added support for SOCK_SEQPACKET. | |
22 | * IPX can no longer use the SO_TYPE hack | |
23 | * but AX.25 now works right, and SPX is | |
24 | * feasible. | |
25 | * Alan Cox : Fixed write poll of non IP protocol | |
26 | * crash. | |
27 | * Florian La Roche: Changed for my new skbuff handling. | |
28 | * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET. | |
29 | * Linus Torvalds : BSD semantic fixes. | |
30 | * Alan Cox : Datagram iovec handling | |
31 | * Darryl Miles : Fixed non-blocking SOCK_STREAM. | |
32 | * Alan Cox : POSIXisms | |
33 | * Pete Wyckoff : Unconnected accept() fix. | |
34 | * | |
35 | */ | |
36 | ||
37 | #include <linux/module.h> | |
38 | #include <linux/types.h> | |
39 | #include <linux/kernel.h> | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 LT |
41 | #include <linux/mm.h> |
42 | #include <linux/interrupt.h> | |
43 | #include <linux/errno.h> | |
44 | #include <linux/sched.h> | |
45 | #include <linux/inet.h> | |
1da177e4 LT |
46 | #include <linux/netdevice.h> |
47 | #include <linux/rtnetlink.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/highmem.h> | |
3305b80c | 50 | #include <linux/spinlock.h> |
5a0e3ad6 | 51 | #include <linux/slab.h> |
0433547a | 52 | #include <linux/pagemap.h> |
a8f820aa | 53 | #include <linux/uio.h> |
1da177e4 LT |
54 | |
55 | #include <net/protocol.h> | |
56 | #include <linux/skbuff.h> | |
1da177e4 | 57 | |
c752f073 ACM |
58 | #include <net/checksum.h> |
59 | #include <net/sock.h> | |
60 | #include <net/tcp_states.h> | |
e9b3cc1b | 61 | #include <trace/events/skb.h> |
076bb0c8 | 62 | #include <net/busy_poll.h> |
1da177e4 LT |
63 | |
64 | /* | |
65 | * Is a socket 'connection oriented' ? | |
66 | */ | |
67 | static inline int connection_based(struct sock *sk) | |
68 | { | |
69 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
70 | } | |
71 | ||
ac6424b9 | 72 | static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync, |
bf368e4e ED |
73 | void *key) |
74 | { | |
75 | unsigned long bits = (unsigned long)key; | |
76 | ||
77 | /* | |
78 | * Avoid a wakeup if event not interesting for us | |
79 | */ | |
80 | if (bits && !(bits & (POLLIN | POLLERR))) | |
81 | return 0; | |
82 | return autoremove_wake_function(wait, mode, sync, key); | |
83 | } | |
1da177e4 | 84 | /* |
39cc8613 | 85 | * Wait for the last received packet to be different from skb |
1da177e4 | 86 | */ |
ea3793ee RW |
87 | int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, |
88 | const struct sk_buff *skb) | |
1da177e4 LT |
89 | { |
90 | int error; | |
bf368e4e | 91 | DEFINE_WAIT_FUNC(wait, receiver_wake_function); |
1da177e4 | 92 | |
aa395145 | 93 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); |
1da177e4 LT |
94 | |
95 | /* Socket errors? */ | |
96 | error = sock_error(sk); | |
97 | if (error) | |
98 | goto out_err; | |
99 | ||
39cc8613 | 100 | if (sk->sk_receive_queue.prev != skb) |
1da177e4 LT |
101 | goto out; |
102 | ||
103 | /* Socket shut down? */ | |
104 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
105 | goto out_noerr; | |
106 | ||
107 | /* Sequenced packets can come disconnected. | |
108 | * If so we report the problem | |
109 | */ | |
110 | error = -ENOTCONN; | |
111 | if (connection_based(sk) && | |
112 | !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) | |
113 | goto out_err; | |
114 | ||
115 | /* handle signals */ | |
116 | if (signal_pending(current)) | |
117 | goto interrupted; | |
118 | ||
119 | error = 0; | |
120 | *timeo_p = schedule_timeout(*timeo_p); | |
121 | out: | |
aa395145 | 122 | finish_wait(sk_sleep(sk), &wait); |
1da177e4 LT |
123 | return error; |
124 | interrupted: | |
125 | error = sock_intr_errno(*timeo_p); | |
126 | out_err: | |
127 | *err = error; | |
128 | goto out; | |
129 | out_noerr: | |
130 | *err = 0; | |
131 | error = 1; | |
132 | goto out; | |
133 | } | |
ea3793ee | 134 | EXPORT_SYMBOL(__skb_wait_for_more_packets); |
1da177e4 | 135 | |
a0a2a660 | 136 | static struct sk_buff *skb_set_peeked(struct sk_buff *skb) |
738ac1eb HX |
137 | { |
138 | struct sk_buff *nskb; | |
139 | ||
140 | if (skb->peeked) | |
a0a2a660 | 141 | return skb; |
738ac1eb HX |
142 | |
143 | /* We have to unshare an skb before modifying it. */ | |
144 | if (!skb_shared(skb)) | |
145 | goto done; | |
146 | ||
147 | nskb = skb_clone(skb, GFP_ATOMIC); | |
148 | if (!nskb) | |
a0a2a660 | 149 | return ERR_PTR(-ENOMEM); |
738ac1eb HX |
150 | |
151 | skb->prev->next = nskb; | |
152 | skb->next->prev = nskb; | |
153 | nskb->prev = skb->prev; | |
154 | nskb->next = skb->next; | |
155 | ||
156 | consume_skb(skb); | |
157 | skb = nskb; | |
158 | ||
159 | done: | |
160 | skb->peeked = 1; | |
161 | ||
a0a2a660 | 162 | return skb; |
738ac1eb HX |
163 | } |
164 | ||
65101aec PA |
165 | struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, |
166 | struct sk_buff_head *queue, | |
167 | unsigned int flags, | |
168 | void (*destructor)(struct sock *sk, | |
169 | struct sk_buff *skb), | |
170 | int *peeked, int *off, int *err, | |
171 | struct sk_buff **last) | |
172 | { | |
a0917e0b | 173 | bool peek_at_off = false; |
65101aec | 174 | struct sk_buff *skb; |
a0917e0b MD |
175 | int _off = 0; |
176 | ||
177 | if (unlikely(flags & MSG_PEEK && *off >= 0)) { | |
178 | peek_at_off = true; | |
179 | _off = *off; | |
180 | } | |
65101aec PA |
181 | |
182 | *last = queue->prev; | |
183 | skb_queue_walk(queue, skb) { | |
184 | if (flags & MSG_PEEK) { | |
a0917e0b MD |
185 | if (peek_at_off && _off >= skb->len && |
186 | (_off || skb->peeked)) { | |
de321ed3 | 187 | _off -= skb->len; |
65101aec PA |
188 | continue; |
189 | } | |
190 | if (!skb->len) { | |
191 | skb = skb_set_peeked(skb); | |
192 | if (unlikely(IS_ERR(skb))) { | |
193 | *err = PTR_ERR(skb); | |
de321ed3 | 194 | return NULL; |
65101aec PA |
195 | } |
196 | } | |
197 | *peeked = 1; | |
63354797 | 198 | refcount_inc(&skb->users); |
65101aec PA |
199 | } else { |
200 | __skb_unlink(skb, queue); | |
201 | if (destructor) | |
202 | destructor(sk, skb); | |
203 | } | |
de321ed3 | 204 | *off = _off; |
65101aec PA |
205 | return skb; |
206 | } | |
207 | return NULL; | |
208 | } | |
209 | ||
1da177e4 | 210 | /** |
ea3793ee | 211 | * __skb_try_recv_datagram - Receive a datagram skbuff |
4dc3b16b | 212 | * @sk: socket |
d3f6cd9e | 213 | * @flags: MSG\_ flags |
7c13f97f | 214 | * @destructor: invoked under the receive lock on successful dequeue |
39cc8613 | 215 | * @peeked: returns non-zero if this packet has been seen before |
3f518bf7 PE |
216 | * @off: an offset in bytes to peek skb from. Returns an offset |
217 | * within an skb where data actually starts | |
4dc3b16b | 218 | * @err: error code returned |
ea3793ee RW |
219 | * @last: set to last peeked message to inform the wait function |
220 | * what to look for when peeking | |
1da177e4 LT |
221 | * |
222 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
223 | * and possible races. This replaces identical code in packet, raw and | |
224 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
225 | * the long standing peek and read race for datagram sockets. If you | |
226 | * alter this routine remember it must be re-entrant. | |
227 | * | |
ea3793ee RW |
228 | * This function will lock the socket if a skb is returned, so |
229 | * the caller needs to unlock the socket in that case (usually by | |
d651983d | 230 | * calling skb_free_datagram). Returns NULL with @err set to |
ea3793ee RW |
231 | * -EAGAIN if no data was available or to some other value if an |
232 | * error was detected. | |
1da177e4 LT |
233 | * |
234 | * * It does not lock socket since today. This function is | |
235 | * * free of race conditions. This measure should/can improve | |
236 | * * significantly datagram socket latencies at high loads, | |
237 | * * when data copying to user space takes lots of time. | |
238 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
239 | * * 8) Great win.) | |
240 | * * --ANK (980729) | |
241 | * | |
242 | * The order of the tests when we find no data waiting are specified | |
243 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
244 | * the standard around please. | |
245 | */ | |
ea3793ee | 246 | struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags, |
7c13f97f PA |
247 | void (*destructor)(struct sock *sk, |
248 | struct sk_buff *skb), | |
ea3793ee RW |
249 | int *peeked, int *off, int *err, |
250 | struct sk_buff **last) | |
1da177e4 | 251 | { |
738ac1eb | 252 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
ea3793ee | 253 | struct sk_buff *skb; |
738ac1eb | 254 | unsigned long cpu_flags; |
1da177e4 LT |
255 | /* |
256 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
257 | */ | |
258 | int error = sock_error(sk); | |
259 | ||
260 | if (error) | |
261 | goto no_packet; | |
262 | ||
a297569f | 263 | *peeked = 0; |
1da177e4 LT |
264 | do { |
265 | /* Again only user level code calls this function, so nothing | |
266 | * interrupt level will suddenly eat the receive_queue. | |
267 | * | |
268 | * Look at current nfs client by the way... | |
8917a3c0 | 269 | * However, this function was correct in any case. 8) |
1da177e4 | 270 | */ |
4934b032 | 271 | spin_lock_irqsave(&queue->lock, cpu_flags); |
65101aec | 272 | skb = __skb_try_recv_from_queue(sk, queue, flags, destructor, |
de321ed3 | 273 | peeked, off, &error, last); |
3f518bf7 | 274 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
de321ed3 AV |
275 | if (error) |
276 | goto no_packet; | |
65101aec PA |
277 | if (skb) |
278 | return skb; | |
2b5cd0df AD |
279 | |
280 | if (!sk_can_busy_loop(sk)) | |
281 | break; | |
282 | ||
283 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
284 | } while (!skb_queue_empty(&sk->sk_receive_queue)); | |
1da177e4 | 285 | |
ea3793ee | 286 | error = -EAGAIN; |
a5b50476 | 287 | |
ea3793ee RW |
288 | no_packet: |
289 | *err = error; | |
290 | return NULL; | |
291 | } | |
292 | EXPORT_SYMBOL(__skb_try_recv_datagram); | |
1da177e4 | 293 | |
ea3793ee | 294 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
7c13f97f PA |
295 | void (*destructor)(struct sock *sk, |
296 | struct sk_buff *skb), | |
ea3793ee RW |
297 | int *peeked, int *off, int *err) |
298 | { | |
299 | struct sk_buff *skb, *last; | |
300 | long timeo; | |
1da177e4 | 301 | |
ea3793ee RW |
302 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
303 | ||
304 | do { | |
7c13f97f PA |
305 | skb = __skb_try_recv_datagram(sk, flags, destructor, peeked, |
306 | off, err, &last); | |
ea3793ee RW |
307 | if (skb) |
308 | return skb; | |
309 | ||
760a4322 | 310 | if (*err != -EAGAIN) |
ea3793ee RW |
311 | break; |
312 | } while (timeo && | |
313 | !__skb_wait_for_more_packets(sk, err, &timeo, last)); | |
1da177e4 | 314 | |
1da177e4 LT |
315 | return NULL; |
316 | } | |
a59322be HX |
317 | EXPORT_SYMBOL(__skb_recv_datagram); |
318 | ||
95c96174 | 319 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
320 | int noblock, int *err) |
321 | { | |
3f518bf7 | 322 | int peeked, off = 0; |
a59322be HX |
323 | |
324 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
7c13f97f | 325 | NULL, &peeked, &off, err); |
a59322be | 326 | } |
9e34a5b5 | 327 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
328 | |
329 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
330 | { | |
ead2ceb0 | 331 | consume_skb(skb); |
270acefa | 332 | sk_mem_reclaim_partial(sk); |
1da177e4 | 333 | } |
9d410c79 ED |
334 | EXPORT_SYMBOL(skb_free_datagram); |
335 | ||
627d2d6b | 336 | void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len) |
9d410c79 | 337 | { |
8a74ad60 ED |
338 | bool slow; |
339 | ||
3889a803 | 340 | if (!skb_unref(skb)) { |
627d2d6b | 341 | sk_peek_offset_bwd(sk, len); |
93bb64ea | 342 | return; |
627d2d6b | 343 | } |
93bb64ea | 344 | |
8a74ad60 | 345 | slow = lock_sock_fast(sk); |
627d2d6b | 346 | sk_peek_offset_bwd(sk, len); |
4b0b72f7 ED |
347 | skb_orphan(skb); |
348 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 349 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 350 | |
93bb64ea ED |
351 | /* skb is now orphaned, can be freed outside of locked section */ |
352 | __kfree_skb(skb); | |
9d410c79 | 353 | } |
627d2d6b | 354 | EXPORT_SYMBOL(__skb_free_datagram_locked); |
1da177e4 | 355 | |
65101aec PA |
356 | int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, |
357 | struct sk_buff *skb, unsigned int flags, | |
69629464 ED |
358 | void (*destructor)(struct sock *sk, |
359 | struct sk_buff *skb)) | |
f8c3bf00 PA |
360 | { |
361 | int err = 0; | |
362 | ||
363 | if (flags & MSG_PEEK) { | |
364 | err = -ENOENT; | |
65101aec | 365 | spin_lock_bh(&sk_queue->lock); |
fd6055a8 | 366 | if (skb->next) { |
65101aec | 367 | __skb_unlink(skb, sk_queue); |
63354797 | 368 | refcount_dec(&skb->users); |
69629464 ED |
369 | if (destructor) |
370 | destructor(sk, skb); | |
f8c3bf00 PA |
371 | err = 0; |
372 | } | |
65101aec | 373 | spin_unlock_bh(&sk_queue->lock); |
f8c3bf00 PA |
374 | } |
375 | ||
376 | atomic_inc(&sk->sk_drops); | |
377 | return err; | |
378 | } | |
379 | EXPORT_SYMBOL(__sk_queue_drop_skb); | |
380 | ||
3305b80c HX |
381 | /** |
382 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
383 | * @sk: socket | |
384 | * @skb: datagram skbuff | |
d3f6cd9e | 385 | * @flags: MSG\_ flags |
3305b80c HX |
386 | * |
387 | * This function frees a datagram skbuff that was received by | |
388 | * skb_recv_datagram. The flags argument must match the one | |
389 | * used for skb_recv_datagram. | |
390 | * | |
391 | * If the MSG_PEEK flag is set, and the packet is still on the | |
392 | * receive queue of the socket, it will be taken off the queue | |
393 | * before it is freed. | |
394 | * | |
395 | * This function currently only disables BH when acquiring the | |
396 | * sk_receive_queue lock. Therefore it must not be used in a | |
397 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
398 | * |
399 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
400 | */ |
401 | ||
27ab2568 | 402 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 403 | { |
65101aec PA |
404 | int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags, |
405 | NULL); | |
3305b80c | 406 | |
61de71c6 JD |
407 | kfree_skb(skb); |
408 | sk_mem_reclaim_partial(sk); | |
27ab2568 | 409 | return err; |
3305b80c | 410 | } |
3305b80c HX |
411 | EXPORT_SYMBOL(skb_kill_datagram); |
412 | ||
a8f820aa HX |
413 | /** |
414 | * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. | |
415 | * @skb: buffer to copy | |
416 | * @offset: offset in the buffer to start copying from | |
417 | * @to: iovec iterator to copy to | |
418 | * @len: amount of data to copy from buffer to iovec | |
419 | */ | |
420 | int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, | |
421 | struct iov_iter *to, int len) | |
422 | { | |
423 | int start = skb_headlen(skb); | |
32786821 | 424 | int i, copy = start - offset, start_off = offset, n; |
a8f820aa HX |
425 | struct sk_buff *frag_iter; |
426 | ||
427 | trace_skb_copy_datagram_iovec(skb, len); | |
428 | ||
429 | /* Copy header. */ | |
430 | if (copy > 0) { | |
431 | if (copy > len) | |
432 | copy = len; | |
32786821 AV |
433 | n = copy_to_iter(skb->data + offset, copy, to); |
434 | offset += n; | |
435 | if (n != copy) | |
a8f820aa HX |
436 | goto short_copy; |
437 | if ((len -= copy) == 0) | |
438 | return 0; | |
a8f820aa HX |
439 | } |
440 | ||
441 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
442 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
443 | int end; | |
444 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
445 | ||
446 | WARN_ON(start > offset + len); | |
447 | ||
448 | end = start + skb_frag_size(frag); | |
449 | if ((copy = end - offset) > 0) { | |
450 | if (copy > len) | |
451 | copy = len; | |
32786821 | 452 | n = copy_page_to_iter(skb_frag_page(frag), |
a8f820aa | 453 | frag->page_offset + offset - |
32786821 AV |
454 | start, copy, to); |
455 | offset += n; | |
456 | if (n != copy) | |
a8f820aa HX |
457 | goto short_copy; |
458 | if (!(len -= copy)) | |
459 | return 0; | |
a8f820aa HX |
460 | } |
461 | start = end; | |
462 | } | |
463 | ||
464 | skb_walk_frags(skb, frag_iter) { | |
465 | int end; | |
466 | ||
467 | WARN_ON(start > offset + len); | |
468 | ||
469 | end = start + frag_iter->len; | |
470 | if ((copy = end - offset) > 0) { | |
471 | if (copy > len) | |
472 | copy = len; | |
473 | if (skb_copy_datagram_iter(frag_iter, offset - start, | |
474 | to, copy)) | |
475 | goto fault; | |
476 | if ((len -= copy) == 0) | |
477 | return 0; | |
478 | offset += copy; | |
479 | } | |
480 | start = end; | |
481 | } | |
482 | if (!len) | |
483 | return 0; | |
484 | ||
485 | /* This is not really a user copy fault, but rather someone | |
486 | * gave us a bogus length on the skb. We should probably | |
487 | * print a warning here as it may indicate a kernel bug. | |
488 | */ | |
489 | ||
490 | fault: | |
32786821 | 491 | iov_iter_revert(to, offset - start_off); |
a8f820aa HX |
492 | return -EFAULT; |
493 | ||
494 | short_copy: | |
495 | if (iov_iter_count(to)) | |
496 | goto fault; | |
497 | ||
498 | return 0; | |
499 | } | |
500 | EXPORT_SYMBOL(skb_copy_datagram_iter); | |
501 | ||
db543c1f | 502 | /** |
8feb2fb2 | 503 | * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. |
db543c1f RR |
504 | * @skb: buffer to copy |
505 | * @offset: offset in the buffer to start copying to | |
8feb2fb2 | 506 | * @from: the copy source |
db543c1f RR |
507 | * @len: amount of data to copy to buffer from iovec |
508 | * | |
509 | * Returns 0 or -EFAULT. | |
db543c1f | 510 | */ |
3a654f97 AV |
511 | int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, |
512 | struct iov_iter *from, | |
513 | int len) | |
514 | { | |
515 | int start = skb_headlen(skb); | |
516 | int i, copy = start - offset; | |
517 | struct sk_buff *frag_iter; | |
518 | ||
519 | /* Copy header. */ | |
520 | if (copy > 0) { | |
521 | if (copy > len) | |
522 | copy = len; | |
523 | if (copy_from_iter(skb->data + offset, copy, from) != copy) | |
524 | goto fault; | |
525 | if ((len -= copy) == 0) | |
526 | return 0; | |
527 | offset += copy; | |
528 | } | |
529 | ||
530 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
531 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
532 | int end; | |
533 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
534 | ||
535 | WARN_ON(start > offset + len); | |
536 | ||
537 | end = start + skb_frag_size(frag); | |
538 | if ((copy = end - offset) > 0) { | |
539 | size_t copied; | |
540 | ||
541 | if (copy > len) | |
542 | copy = len; | |
543 | copied = copy_page_from_iter(skb_frag_page(frag), | |
544 | frag->page_offset + offset - start, | |
545 | copy, from); | |
546 | if (copied != copy) | |
547 | goto fault; | |
548 | ||
549 | if (!(len -= copy)) | |
550 | return 0; | |
551 | offset += copy; | |
552 | } | |
553 | start = end; | |
554 | } | |
555 | ||
556 | skb_walk_frags(skb, frag_iter) { | |
557 | int end; | |
558 | ||
559 | WARN_ON(start > offset + len); | |
560 | ||
561 | end = start + frag_iter->len; | |
562 | if ((copy = end - offset) > 0) { | |
563 | if (copy > len) | |
564 | copy = len; | |
565 | if (skb_copy_datagram_from_iter(frag_iter, | |
566 | offset - start, | |
567 | from, copy)) | |
568 | goto fault; | |
569 | if ((len -= copy) == 0) | |
570 | return 0; | |
571 | offset += copy; | |
572 | } | |
573 | start = end; | |
574 | } | |
575 | if (!len) | |
576 | return 0; | |
577 | ||
578 | fault: | |
579 | return -EFAULT; | |
580 | } | |
581 | EXPORT_SYMBOL(skb_copy_datagram_from_iter); | |
582 | ||
52267790 WB |
583 | int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, |
584 | struct iov_iter *from, size_t length) | |
3a654f97 | 585 | { |
52267790 | 586 | int frag = skb_shinfo(skb)->nr_frags; |
3a654f97 | 587 | |
52267790 | 588 | while (length && iov_iter_count(from)) { |
3a654f97 AV |
589 | struct page *pages[MAX_SKB_FRAGS]; |
590 | size_t start; | |
591 | ssize_t copied; | |
592 | unsigned long truesize; | |
593 | int n = 0; | |
594 | ||
595 | if (frag == MAX_SKB_FRAGS) | |
596 | return -EMSGSIZE; | |
597 | ||
52267790 | 598 | copied = iov_iter_get_pages(from, pages, length, |
3a654f97 AV |
599 | MAX_SKB_FRAGS - frag, &start); |
600 | if (copied < 0) | |
601 | return -EFAULT; | |
602 | ||
603 | iov_iter_advance(from, copied); | |
52267790 | 604 | length -= copied; |
3a654f97 AV |
605 | |
606 | truesize = PAGE_ALIGN(copied + start); | |
607 | skb->data_len += copied; | |
608 | skb->len += copied; | |
609 | skb->truesize += truesize; | |
52267790 WB |
610 | if (sk && sk->sk_type == SOCK_STREAM) { |
611 | sk->sk_wmem_queued += truesize; | |
612 | sk_mem_charge(sk, truesize); | |
613 | } else { | |
614 | refcount_add(truesize, &skb->sk->sk_wmem_alloc); | |
615 | } | |
3a654f97 AV |
616 | while (copied) { |
617 | int size = min_t(int, copied, PAGE_SIZE - start); | |
618 | skb_fill_page_desc(skb, frag++, pages[n], start, size); | |
619 | start = 0; | |
620 | copied -= size; | |
621 | n++; | |
622 | } | |
623 | } | |
624 | return 0; | |
625 | } | |
52267790 WB |
626 | EXPORT_SYMBOL(__zerocopy_sg_from_iter); |
627 | ||
628 | /** | |
629 | * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter | |
630 | * @skb: buffer to copy | |
631 | * @from: the source to copy from | |
632 | * | |
633 | * The function will first copy up to headlen, and then pin the userspace | |
634 | * pages and build frags through them. | |
635 | * | |
636 | * Returns 0, -EFAULT or -EMSGSIZE. | |
637 | */ | |
638 | int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) | |
639 | { | |
640 | int copy = min_t(int, skb_headlen(skb), iov_iter_count(from)); | |
641 | ||
642 | /* copy up to skb headlen */ | |
643 | if (skb_copy_datagram_from_iter(skb, 0, from, copy)) | |
644 | return -EFAULT; | |
645 | ||
646 | return __zerocopy_sg_from_iter(NULL, skb, from, ~0U); | |
647 | } | |
3a654f97 AV |
648 | EXPORT_SYMBOL(zerocopy_sg_from_iter); |
649 | ||
1da177e4 | 650 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
e5a4b0bb | 651 | struct iov_iter *to, int len, |
5084205f | 652 | __wsum *csump) |
1da177e4 | 653 | { |
1a028e50 | 654 | int start = skb_headlen(skb); |
32786821 | 655 | int i, copy = start - offset, start_off = offset; |
5b1a002a DM |
656 | struct sk_buff *frag_iter; |
657 | int pos = 0; | |
e5a4b0bb | 658 | int n; |
1da177e4 LT |
659 | |
660 | /* Copy header. */ | |
661 | if (copy > 0) { | |
1da177e4 LT |
662 | if (copy > len) |
663 | copy = len; | |
e5a4b0bb | 664 | n = csum_and_copy_to_iter(skb->data + offset, copy, csump, to); |
32786821 | 665 | offset += n; |
e5a4b0bb | 666 | if (n != copy) |
1da177e4 LT |
667 | goto fault; |
668 | if ((len -= copy) == 0) | |
669 | return 0; | |
1da177e4 LT |
670 | pos = copy; |
671 | } | |
672 | ||
673 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1a028e50 | 674 | int end; |
9e903e08 | 675 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1da177e4 | 676 | |
547b792c | 677 | WARN_ON(start > offset + len); |
1a028e50 | 678 | |
9e903e08 | 679 | end = start + skb_frag_size(frag); |
1da177e4 | 680 | if ((copy = end - offset) > 0) { |
e5a4b0bb | 681 | __wsum csum2 = 0; |
ea2ab693 | 682 | struct page *page = skb_frag_page(frag); |
e5a4b0bb | 683 | u8 *vaddr = kmap(page); |
1da177e4 LT |
684 | |
685 | if (copy > len) | |
686 | copy = len; | |
e5a4b0bb AV |
687 | n = csum_and_copy_to_iter(vaddr + frag->page_offset + |
688 | offset - start, copy, | |
689 | &csum2, to); | |
1da177e4 | 690 | kunmap(page); |
32786821 | 691 | offset += n; |
e5a4b0bb | 692 | if (n != copy) |
1da177e4 LT |
693 | goto fault; |
694 | *csump = csum_block_add(*csump, csum2, pos); | |
695 | if (!(len -= copy)) | |
696 | return 0; | |
1da177e4 LT |
697 | pos += copy; |
698 | } | |
1a028e50 | 699 | start = end; |
1da177e4 LT |
700 | } |
701 | ||
5b1a002a DM |
702 | skb_walk_frags(skb, frag_iter) { |
703 | int end; | |
704 | ||
705 | WARN_ON(start > offset + len); | |
706 | ||
707 | end = start + frag_iter->len; | |
708 | if ((copy = end - offset) > 0) { | |
709 | __wsum csum2 = 0; | |
710 | if (copy > len) | |
711 | copy = len; | |
712 | if (skb_copy_and_csum_datagram(frag_iter, | |
713 | offset - start, | |
714 | to, copy, | |
715 | &csum2)) | |
716 | goto fault; | |
717 | *csump = csum_block_add(*csump, csum2, pos); | |
718 | if ((len -= copy) == 0) | |
719 | return 0; | |
720 | offset += copy; | |
5b1a002a | 721 | pos += copy; |
1da177e4 | 722 | } |
5b1a002a | 723 | start = end; |
1da177e4 LT |
724 | } |
725 | if (!len) | |
726 | return 0; | |
727 | ||
728 | fault: | |
32786821 | 729 | iov_iter_revert(to, offset - start_off); |
1da177e4 LT |
730 | return -EFAULT; |
731 | } | |
732 | ||
759e5d00 | 733 | __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) |
fb286bb2 | 734 | { |
d3bc23e7 | 735 | __sum16 sum; |
fb286bb2 | 736 | |
759e5d00 | 737 | sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); |
46fb51eb TH |
738 | if (likely(!sum)) { |
739 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
740 | !skb->csum_complete_sw) | |
741 | netdev_rx_csum_fault(skb->dev); | |
742 | } | |
89c22d8c HX |
743 | if (!skb_shared(skb)) |
744 | skb->csum_valid = !sum; | |
fb286bb2 HX |
745 | return sum; |
746 | } | |
759e5d00 HX |
747 | EXPORT_SYMBOL(__skb_checksum_complete_head); |
748 | ||
749 | __sum16 __skb_checksum_complete(struct sk_buff *skb) | |
750 | { | |
46fb51eb TH |
751 | __wsum csum; |
752 | __sum16 sum; | |
753 | ||
754 | csum = skb_checksum(skb, 0, skb->len, 0); | |
755 | ||
756 | /* skb->csum holds pseudo checksum */ | |
757 | sum = csum_fold(csum_add(skb->csum, csum)); | |
758 | if (likely(!sum)) { | |
759 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
760 | !skb->csum_complete_sw) | |
761 | netdev_rx_csum_fault(skb->dev); | |
762 | } | |
763 | ||
89c22d8c HX |
764 | if (!skb_shared(skb)) { |
765 | /* Save full packet checksum */ | |
766 | skb->csum = csum; | |
767 | skb->ip_summed = CHECKSUM_COMPLETE; | |
768 | skb->csum_complete_sw = 1; | |
769 | skb->csum_valid = !sum; | |
770 | } | |
46fb51eb TH |
771 | |
772 | return sum; | |
759e5d00 | 773 | } |
fb286bb2 HX |
774 | EXPORT_SYMBOL(__skb_checksum_complete); |
775 | ||
1da177e4 | 776 | /** |
e5a4b0bb | 777 | * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. |
4dc3b16b PP |
778 | * @skb: skbuff |
779 | * @hlen: hardware length | |
e5a4b0bb | 780 | * @msg: destination |
4ec93edb | 781 | * |
1da177e4 LT |
782 | * Caller _must_ check that skb will fit to this iovec. |
783 | * | |
784 | * Returns: 0 - success. | |
785 | * -EINVAL - checksum failure. | |
e5a4b0bb | 786 | * -EFAULT - fault during copy. |
1da177e4 | 787 | */ |
e5a4b0bb AV |
788 | int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, |
789 | int hlen, struct msghdr *msg) | |
1da177e4 | 790 | { |
d3bc23e7 | 791 | __wsum csum; |
1da177e4 LT |
792 | int chunk = skb->len - hlen; |
793 | ||
ef8aef55 HX |
794 | if (!chunk) |
795 | return 0; | |
796 | ||
01e97e65 | 797 | if (msg_data_left(msg) < chunk) { |
fb286bb2 | 798 | if (__skb_checksum_complete(skb)) |
a6a59932 | 799 | return -EINVAL; |
e5a4b0bb | 800 | if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) |
1da177e4 LT |
801 | goto fault; |
802 | } else { | |
803 | csum = csum_partial(skb->data, hlen, skb->csum); | |
e5a4b0bb | 804 | if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, |
1da177e4 LT |
805 | chunk, &csum)) |
806 | goto fault; | |
a6a59932 DT |
807 | |
808 | if (csum_fold(csum)) { | |
809 | iov_iter_revert(&msg->msg_iter, chunk); | |
810 | return -EINVAL; | |
811 | } | |
812 | ||
84fa7933 | 813 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) |
fb286bb2 | 814 | netdev_rx_csum_fault(skb->dev); |
1da177e4 LT |
815 | } |
816 | return 0; | |
1da177e4 LT |
817 | fault: |
818 | return -EFAULT; | |
819 | } | |
e5a4b0bb | 820 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); |
1da177e4 LT |
821 | |
822 | /** | |
823 | * datagram_poll - generic datagram poll | |
4dc3b16b PP |
824 | * @file: file struct |
825 | * @sock: socket | |
826 | * @wait: poll table | |
1da177e4 LT |
827 | * |
828 | * Datagram poll: Again totally generic. This also handles | |
829 | * sequenced packet sockets providing the socket receive queue | |
830 | * is only ever holding data ready to receive. | |
831 | * | |
d3f6cd9e | 832 | * Note: when you *don't* use this routine for this protocol, |
1da177e4 LT |
833 | * and you use a different write policy from sock_writeable() |
834 | * then please supply your own write_space callback. | |
835 | */ | |
836 | unsigned int datagram_poll(struct file *file, struct socket *sock, | |
837 | poll_table *wait) | |
838 | { | |
839 | struct sock *sk = sock->sk; | |
840 | unsigned int mask; | |
841 | ||
aa395145 | 842 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 LT |
843 | mask = 0; |
844 | ||
845 | /* exceptional events? */ | |
846 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
7d4c04fc | 847 | mask |= POLLERR | |
8facd5fb | 848 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0); |
7d4c04fc | 849 | |
f348d70a | 850 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
db40980f | 851 | mask |= POLLRDHUP | POLLIN | POLLRDNORM; |
1da177e4 LT |
852 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
853 | mask |= POLLHUP; | |
854 | ||
855 | /* readable? */ | |
db40980f | 856 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
857 | mask |= POLLIN | POLLRDNORM; |
858 | ||
859 | /* Connection-based need to check for termination and startup */ | |
860 | if (connection_based(sk)) { | |
861 | if (sk->sk_state == TCP_CLOSE) | |
862 | mask |= POLLHUP; | |
863 | /* connection hasn't started yet? */ | |
864 | if (sk->sk_state == TCP_SYN_SENT) | |
865 | return mask; | |
866 | } | |
867 | ||
868 | /* writable? */ | |
869 | if (sock_writeable(sk)) | |
870 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
871 | else | |
9cd3e072 | 872 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
873 | |
874 | return mask; | |
875 | } | |
1da177e4 | 876 | EXPORT_SYMBOL(datagram_poll); |