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 | 63 | |
7b7ed885 BVA |
64 | #include "datagram.h" |
65 | ||
1da177e4 LT |
66 | /* |
67 | * Is a socket 'connection oriented' ? | |
68 | */ | |
69 | static inline int connection_based(struct sock *sk) | |
70 | { | |
71 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
72 | } | |
73 | ||
ac6424b9 | 74 | static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync, |
bf368e4e ED |
75 | void *key) |
76 | { | |
bf368e4e ED |
77 | /* |
78 | * Avoid a wakeup if event not interesting for us | |
79 | */ | |
a9a08845 | 80 | if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR))) |
bf368e4e ED |
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), | |
fd69c399 | 170 | int *off, int *err, |
65101aec PA |
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); | |
98e4fcff | 192 | if (IS_ERR(skb)) { |
65101aec | 193 | *err = PTR_ERR(skb); |
de321ed3 | 194 | return NULL; |
65101aec PA |
195 | } |
196 | } | |
63354797 | 197 | refcount_inc(&skb->users); |
65101aec PA |
198 | } else { |
199 | __skb_unlink(skb, queue); | |
200 | if (destructor) | |
201 | destructor(sk, skb); | |
202 | } | |
de321ed3 | 203 | *off = _off; |
65101aec PA |
204 | return skb; |
205 | } | |
206 | return NULL; | |
207 | } | |
208 | ||
1da177e4 | 209 | /** |
ea3793ee | 210 | * __skb_try_recv_datagram - Receive a datagram skbuff |
4dc3b16b | 211 | * @sk: socket |
d3f6cd9e | 212 | * @flags: MSG\_ flags |
7c13f97f | 213 | * @destructor: invoked under the receive lock on successful dequeue |
3f518bf7 PE |
214 | * @off: an offset in bytes to peek skb from. Returns an offset |
215 | * within an skb where data actually starts | |
4dc3b16b | 216 | * @err: error code returned |
ea3793ee RW |
217 | * @last: set to last peeked message to inform the wait function |
218 | * what to look for when peeking | |
1da177e4 LT |
219 | * |
220 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
221 | * and possible races. This replaces identical code in packet, raw and | |
222 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
223 | * the long standing peek and read race for datagram sockets. If you | |
224 | * alter this routine remember it must be re-entrant. | |
225 | * | |
ea3793ee RW |
226 | * This function will lock the socket if a skb is returned, so |
227 | * the caller needs to unlock the socket in that case (usually by | |
d651983d | 228 | * calling skb_free_datagram). Returns NULL with @err set to |
ea3793ee RW |
229 | * -EAGAIN if no data was available or to some other value if an |
230 | * error was detected. | |
1da177e4 LT |
231 | * |
232 | * * It does not lock socket since today. This function is | |
233 | * * free of race conditions. This measure should/can improve | |
234 | * * significantly datagram socket latencies at high loads, | |
235 | * * when data copying to user space takes lots of time. | |
236 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
237 | * * 8) Great win.) | |
238 | * * --ANK (980729) | |
239 | * | |
240 | * The order of the tests when we find no data waiting are specified | |
241 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
242 | * the standard around please. | |
243 | */ | |
ea3793ee | 244 | struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags, |
7c13f97f PA |
245 | void (*destructor)(struct sock *sk, |
246 | struct sk_buff *skb), | |
fd69c399 | 247 | int *off, int *err, |
ea3793ee | 248 | struct sk_buff **last) |
1da177e4 | 249 | { |
738ac1eb | 250 | struct sk_buff_head *queue = &sk->sk_receive_queue; |
ea3793ee | 251 | struct sk_buff *skb; |
738ac1eb | 252 | unsigned long cpu_flags; |
1da177e4 LT |
253 | /* |
254 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
255 | */ | |
256 | int error = sock_error(sk); | |
257 | ||
258 | if (error) | |
259 | goto no_packet; | |
260 | ||
1da177e4 LT |
261 | do { |
262 | /* Again only user level code calls this function, so nothing | |
263 | * interrupt level will suddenly eat the receive_queue. | |
264 | * | |
265 | * Look at current nfs client by the way... | |
8917a3c0 | 266 | * However, this function was correct in any case. 8) |
1da177e4 | 267 | */ |
4934b032 | 268 | spin_lock_irqsave(&queue->lock, cpu_flags); |
65101aec | 269 | skb = __skb_try_recv_from_queue(sk, queue, flags, destructor, |
fd69c399 | 270 | off, &error, last); |
3f518bf7 | 271 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
de321ed3 AV |
272 | if (error) |
273 | goto no_packet; | |
65101aec PA |
274 | if (skb) |
275 | return skb; | |
2b5cd0df AD |
276 | |
277 | if (!sk_can_busy_loop(sk)) | |
278 | break; | |
279 | ||
280 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
0b91bce1 | 281 | } while (sk->sk_receive_queue.prev != *last); |
1da177e4 | 282 | |
ea3793ee | 283 | error = -EAGAIN; |
a5b50476 | 284 | |
ea3793ee RW |
285 | no_packet: |
286 | *err = error; | |
287 | return NULL; | |
288 | } | |
289 | EXPORT_SYMBOL(__skb_try_recv_datagram); | |
1da177e4 | 290 | |
ea3793ee | 291 | struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, |
7c13f97f PA |
292 | void (*destructor)(struct sock *sk, |
293 | struct sk_buff *skb), | |
fd69c399 | 294 | int *off, int *err) |
ea3793ee RW |
295 | { |
296 | struct sk_buff *skb, *last; | |
297 | long timeo; | |
1da177e4 | 298 | |
ea3793ee RW |
299 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
300 | ||
301 | do { | |
fd69c399 PA |
302 | skb = __skb_try_recv_datagram(sk, flags, destructor, off, err, |
303 | &last); | |
ea3793ee RW |
304 | if (skb) |
305 | return skb; | |
306 | ||
760a4322 | 307 | if (*err != -EAGAIN) |
ea3793ee RW |
308 | break; |
309 | } while (timeo && | |
310 | !__skb_wait_for_more_packets(sk, err, &timeo, last)); | |
1da177e4 | 311 | |
1da177e4 LT |
312 | return NULL; |
313 | } | |
a59322be HX |
314 | EXPORT_SYMBOL(__skb_recv_datagram); |
315 | ||
95c96174 | 316 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
317 | int noblock, int *err) |
318 | { | |
fd69c399 | 319 | int off = 0; |
a59322be HX |
320 | |
321 | return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
fd69c399 | 322 | NULL, &off, err); |
a59322be | 323 | } |
9e34a5b5 | 324 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
325 | |
326 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
327 | { | |
ead2ceb0 | 328 | consume_skb(skb); |
270acefa | 329 | sk_mem_reclaim_partial(sk); |
1da177e4 | 330 | } |
9d410c79 ED |
331 | EXPORT_SYMBOL(skb_free_datagram); |
332 | ||
627d2d6b | 333 | void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len) |
9d410c79 | 334 | { |
8a74ad60 ED |
335 | bool slow; |
336 | ||
3889a803 | 337 | if (!skb_unref(skb)) { |
627d2d6b | 338 | sk_peek_offset_bwd(sk, len); |
93bb64ea | 339 | return; |
627d2d6b | 340 | } |
93bb64ea | 341 | |
8a74ad60 | 342 | slow = lock_sock_fast(sk); |
627d2d6b | 343 | sk_peek_offset_bwd(sk, len); |
4b0b72f7 ED |
344 | skb_orphan(skb); |
345 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 346 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 347 | |
93bb64ea ED |
348 | /* skb is now orphaned, can be freed outside of locked section */ |
349 | __kfree_skb(skb); | |
9d410c79 | 350 | } |
627d2d6b | 351 | EXPORT_SYMBOL(__skb_free_datagram_locked); |
1da177e4 | 352 | |
65101aec PA |
353 | int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, |
354 | struct sk_buff *skb, unsigned int flags, | |
69629464 ED |
355 | void (*destructor)(struct sock *sk, |
356 | struct sk_buff *skb)) | |
f8c3bf00 PA |
357 | { |
358 | int err = 0; | |
359 | ||
360 | if (flags & MSG_PEEK) { | |
361 | err = -ENOENT; | |
65101aec | 362 | spin_lock_bh(&sk_queue->lock); |
fd6055a8 | 363 | if (skb->next) { |
65101aec | 364 | __skb_unlink(skb, sk_queue); |
63354797 | 365 | refcount_dec(&skb->users); |
69629464 ED |
366 | if (destructor) |
367 | destructor(sk, skb); | |
f8c3bf00 PA |
368 | err = 0; |
369 | } | |
65101aec | 370 | spin_unlock_bh(&sk_queue->lock); |
f8c3bf00 PA |
371 | } |
372 | ||
373 | atomic_inc(&sk->sk_drops); | |
374 | return err; | |
375 | } | |
376 | EXPORT_SYMBOL(__sk_queue_drop_skb); | |
377 | ||
3305b80c HX |
378 | /** |
379 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
380 | * @sk: socket | |
381 | * @skb: datagram skbuff | |
d3f6cd9e | 382 | * @flags: MSG\_ flags |
3305b80c HX |
383 | * |
384 | * This function frees a datagram skbuff that was received by | |
385 | * skb_recv_datagram. The flags argument must match the one | |
386 | * used for skb_recv_datagram. | |
387 | * | |
388 | * If the MSG_PEEK flag is set, and the packet is still on the | |
389 | * receive queue of the socket, it will be taken off the queue | |
390 | * before it is freed. | |
391 | * | |
392 | * This function currently only disables BH when acquiring the | |
393 | * sk_receive_queue lock. Therefore it must not be used in a | |
394 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
395 | * |
396 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
397 | */ |
398 | ||
27ab2568 | 399 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 400 | { |
65101aec PA |
401 | int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags, |
402 | NULL); | |
3305b80c | 403 | |
61de71c6 JD |
404 | kfree_skb(skb); |
405 | sk_mem_reclaim_partial(sk); | |
27ab2568 | 406 | return err; |
3305b80c | 407 | } |
3305b80c HX |
408 | EXPORT_SYMBOL(skb_kill_datagram); |
409 | ||
56dc6d63 Y |
410 | static int __skb_datagram_iter(const struct sk_buff *skb, int offset, |
411 | struct iov_iter *to, int len, bool fault_short, | |
412 | size_t (*cb)(const void *, size_t, void *, | |
413 | struct iov_iter *), void *data) | |
a8f820aa HX |
414 | { |
415 | int start = skb_headlen(skb); | |
32786821 | 416 | int i, copy = start - offset, start_off = offset, n; |
a8f820aa HX |
417 | struct sk_buff *frag_iter; |
418 | ||
a8f820aa HX |
419 | /* Copy header. */ |
420 | if (copy > 0) { | |
421 | if (copy > len) | |
422 | copy = len; | |
950fcaec | 423 | n = cb(skb->data + offset, copy, data, to); |
32786821 AV |
424 | offset += n; |
425 | if (n != copy) | |
a8f820aa HX |
426 | goto short_copy; |
427 | if ((len -= copy) == 0) | |
428 | return 0; | |
a8f820aa HX |
429 | } |
430 | ||
431 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
432 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
433 | int end; | |
434 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
435 | ||
436 | WARN_ON(start > offset + len); | |
437 | ||
438 | end = start + skb_frag_size(frag); | |
439 | if ((copy = end - offset) > 0) { | |
0fc07791 SG |
440 | struct page *page = skb_frag_page(frag); |
441 | u8 *vaddr = kmap(page); | |
442 | ||
a8f820aa HX |
443 | if (copy > len) |
444 | copy = len; | |
950fcaec SG |
445 | n = cb(vaddr + frag->page_offset + |
446 | offset - start, copy, data, to); | |
0fc07791 | 447 | kunmap(page); |
32786821 AV |
448 | offset += n; |
449 | if (n != copy) | |
a8f820aa HX |
450 | goto short_copy; |
451 | if (!(len -= copy)) | |
452 | return 0; | |
a8f820aa HX |
453 | } |
454 | start = end; | |
455 | } | |
456 | ||
457 | skb_walk_frags(skb, frag_iter) { | |
458 | int end; | |
459 | ||
460 | WARN_ON(start > offset + len); | |
461 | ||
462 | end = start + frag_iter->len; | |
463 | if ((copy = end - offset) > 0) { | |
464 | if (copy > len) | |
465 | copy = len; | |
950fcaec | 466 | if (__skb_datagram_iter(frag_iter, offset - start, |
65d69e25 | 467 | to, copy, fault_short, cb, data)) |
a8f820aa HX |
468 | goto fault; |
469 | if ((len -= copy) == 0) | |
470 | return 0; | |
471 | offset += copy; | |
472 | } | |
473 | start = end; | |
474 | } | |
475 | if (!len) | |
476 | return 0; | |
477 | ||
478 | /* This is not really a user copy fault, but rather someone | |
479 | * gave us a bogus length on the skb. We should probably | |
480 | * print a warning here as it may indicate a kernel bug. | |
481 | */ | |
482 | ||
483 | fault: | |
32786821 | 484 | iov_iter_revert(to, offset - start_off); |
a8f820aa HX |
485 | return -EFAULT; |
486 | ||
487 | short_copy: | |
950fcaec | 488 | if (fault_short || iov_iter_count(to)) |
a8f820aa HX |
489 | goto fault; |
490 | ||
491 | return 0; | |
492 | } | |
950fcaec | 493 | |
65d69e25 SG |
494 | /** |
495 | * skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator | |
496 | * and update a hash. | |
497 | * @skb: buffer to copy | |
498 | * @offset: offset in the buffer to start copying from | |
499 | * @to: iovec iterator to copy to | |
500 | * @len: amount of data to copy from buffer to iovec | |
501 | * @hash: hash request to update | |
502 | */ | |
503 | int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, | |
504 | struct iov_iter *to, int len, | |
505 | struct ahash_request *hash) | |
506 | { | |
507 | return __skb_datagram_iter(skb, offset, to, len, true, | |
508 | hash_and_copy_to_iter, hash); | |
509 | } | |
510 | EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter); | |
511 | ||
950fcaec SG |
512 | static size_t simple_copy_to_iter(const void *addr, size_t bytes, |
513 | void *data __always_unused, struct iov_iter *i) | |
514 | { | |
515 | return copy_to_iter(addr, bytes, i); | |
516 | } | |
517 | ||
518 | /** | |
519 | * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. | |
520 | * @skb: buffer to copy | |
521 | * @offset: offset in the buffer to start copying from | |
522 | * @to: iovec iterator to copy to | |
523 | * @len: amount of data to copy from buffer to iovec | |
524 | */ | |
525 | int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, | |
526 | struct iov_iter *to, int len) | |
527 | { | |
528 | trace_skb_copy_datagram_iovec(skb, len); | |
529 | return __skb_datagram_iter(skb, offset, to, len, false, | |
530 | simple_copy_to_iter, NULL); | |
531 | } | |
a8f820aa HX |
532 | EXPORT_SYMBOL(skb_copy_datagram_iter); |
533 | ||
db543c1f | 534 | /** |
8feb2fb2 | 535 | * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. |
db543c1f RR |
536 | * @skb: buffer to copy |
537 | * @offset: offset in the buffer to start copying to | |
8feb2fb2 | 538 | * @from: the copy source |
db543c1f RR |
539 | * @len: amount of data to copy to buffer from iovec |
540 | * | |
541 | * Returns 0 or -EFAULT. | |
db543c1f | 542 | */ |
3a654f97 AV |
543 | int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, |
544 | struct iov_iter *from, | |
545 | int len) | |
546 | { | |
547 | int start = skb_headlen(skb); | |
548 | int i, copy = start - offset; | |
549 | struct sk_buff *frag_iter; | |
550 | ||
551 | /* Copy header. */ | |
552 | if (copy > 0) { | |
553 | if (copy > len) | |
554 | copy = len; | |
555 | if (copy_from_iter(skb->data + offset, copy, from) != copy) | |
556 | goto fault; | |
557 | if ((len -= copy) == 0) | |
558 | return 0; | |
559 | offset += copy; | |
560 | } | |
561 | ||
562 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
563 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
564 | int end; | |
565 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
566 | ||
567 | WARN_ON(start > offset + len); | |
568 | ||
569 | end = start + skb_frag_size(frag); | |
570 | if ((copy = end - offset) > 0) { | |
571 | size_t copied; | |
572 | ||
573 | if (copy > len) | |
574 | copy = len; | |
575 | copied = copy_page_from_iter(skb_frag_page(frag), | |
576 | frag->page_offset + offset - start, | |
577 | copy, from); | |
578 | if (copied != copy) | |
579 | goto fault; | |
580 | ||
581 | if (!(len -= copy)) | |
582 | return 0; | |
583 | offset += copy; | |
584 | } | |
585 | start = end; | |
586 | } | |
587 | ||
588 | skb_walk_frags(skb, frag_iter) { | |
589 | int end; | |
590 | ||
591 | WARN_ON(start > offset + len); | |
592 | ||
593 | end = start + frag_iter->len; | |
594 | if ((copy = end - offset) > 0) { | |
595 | if (copy > len) | |
596 | copy = len; | |
597 | if (skb_copy_datagram_from_iter(frag_iter, | |
598 | offset - start, | |
599 | from, copy)) | |
600 | goto fault; | |
601 | if ((len -= copy) == 0) | |
602 | return 0; | |
603 | offset += copy; | |
604 | } | |
605 | start = end; | |
606 | } | |
607 | if (!len) | |
608 | return 0; | |
609 | ||
610 | fault: | |
611 | return -EFAULT; | |
612 | } | |
613 | EXPORT_SYMBOL(skb_copy_datagram_from_iter); | |
614 | ||
52267790 WB |
615 | int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, |
616 | struct iov_iter *from, size_t length) | |
3a654f97 | 617 | { |
52267790 | 618 | int frag = skb_shinfo(skb)->nr_frags; |
3a654f97 | 619 | |
52267790 | 620 | while (length && iov_iter_count(from)) { |
3a654f97 AV |
621 | struct page *pages[MAX_SKB_FRAGS]; |
622 | size_t start; | |
623 | ssize_t copied; | |
624 | unsigned long truesize; | |
625 | int n = 0; | |
626 | ||
627 | if (frag == MAX_SKB_FRAGS) | |
628 | return -EMSGSIZE; | |
629 | ||
52267790 | 630 | copied = iov_iter_get_pages(from, pages, length, |
3a654f97 AV |
631 | MAX_SKB_FRAGS - frag, &start); |
632 | if (copied < 0) | |
633 | return -EFAULT; | |
634 | ||
635 | iov_iter_advance(from, copied); | |
52267790 | 636 | length -= copied; |
3a654f97 AV |
637 | |
638 | truesize = PAGE_ALIGN(copied + start); | |
639 | skb->data_len += copied; | |
640 | skb->len += copied; | |
641 | skb->truesize += truesize; | |
52267790 WB |
642 | if (sk && sk->sk_type == SOCK_STREAM) { |
643 | sk->sk_wmem_queued += truesize; | |
644 | sk_mem_charge(sk, truesize); | |
645 | } else { | |
646 | refcount_add(truesize, &skb->sk->sk_wmem_alloc); | |
647 | } | |
3a654f97 AV |
648 | while (copied) { |
649 | int size = min_t(int, copied, PAGE_SIZE - start); | |
650 | skb_fill_page_desc(skb, frag++, pages[n], start, size); | |
651 | start = 0; | |
652 | copied -= size; | |
653 | n++; | |
654 | } | |
655 | } | |
656 | return 0; | |
657 | } | |
52267790 WB |
658 | EXPORT_SYMBOL(__zerocopy_sg_from_iter); |
659 | ||
660 | /** | |
661 | * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter | |
662 | * @skb: buffer to copy | |
663 | * @from: the source to copy from | |
664 | * | |
665 | * The function will first copy up to headlen, and then pin the userspace | |
666 | * pages and build frags through them. | |
667 | * | |
668 | * Returns 0, -EFAULT or -EMSGSIZE. | |
669 | */ | |
670 | int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) | |
671 | { | |
672 | int copy = min_t(int, skb_headlen(skb), iov_iter_count(from)); | |
673 | ||
674 | /* copy up to skb headlen */ | |
675 | if (skb_copy_datagram_from_iter(skb, 0, from, copy)) | |
676 | return -EFAULT; | |
677 | ||
678 | return __zerocopy_sg_from_iter(NULL, skb, from, ~0U); | |
679 | } | |
3a654f97 AV |
680 | EXPORT_SYMBOL(zerocopy_sg_from_iter); |
681 | ||
950fcaec SG |
682 | /** |
683 | * skb_copy_and_csum_datagram_iter - Copy datagram to an iovec iterator | |
684 | * and update a checksum. | |
685 | * @skb: buffer to copy | |
686 | * @offset: offset in the buffer to start copying from | |
687 | * @to: iovec iterator to copy to | |
688 | * @len: amount of data to copy from buffer to iovec | |
689 | * @csump: checksum pointer | |
690 | */ | |
1da177e4 | 691 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
e5a4b0bb | 692 | struct iov_iter *to, int len, |
5084205f | 693 | __wsum *csump) |
1da177e4 | 694 | { |
950fcaec SG |
695 | return __skb_datagram_iter(skb, offset, to, len, true, |
696 | csum_and_copy_to_iter, csump); | |
1da177e4 LT |
697 | } |
698 | ||
699 | /** | |
e5a4b0bb | 700 | * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. |
4dc3b16b PP |
701 | * @skb: skbuff |
702 | * @hlen: hardware length | |
e5a4b0bb | 703 | * @msg: destination |
4ec93edb | 704 | * |
1da177e4 LT |
705 | * Caller _must_ check that skb will fit to this iovec. |
706 | * | |
707 | * Returns: 0 - success. | |
708 | * -EINVAL - checksum failure. | |
e5a4b0bb | 709 | * -EFAULT - fault during copy. |
1da177e4 | 710 | */ |
e5a4b0bb AV |
711 | int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, |
712 | int hlen, struct msghdr *msg) | |
1da177e4 | 713 | { |
d3bc23e7 | 714 | __wsum csum; |
1da177e4 LT |
715 | int chunk = skb->len - hlen; |
716 | ||
ef8aef55 HX |
717 | if (!chunk) |
718 | return 0; | |
719 | ||
01e97e65 | 720 | if (msg_data_left(msg) < chunk) { |
fb286bb2 | 721 | if (__skb_checksum_complete(skb)) |
a6a59932 | 722 | return -EINVAL; |
e5a4b0bb | 723 | if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) |
1da177e4 LT |
724 | goto fault; |
725 | } else { | |
726 | csum = csum_partial(skb->data, hlen, skb->csum); | |
e5a4b0bb | 727 | if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, |
1da177e4 LT |
728 | chunk, &csum)) |
729 | goto fault; | |
a6a59932 DT |
730 | |
731 | if (csum_fold(csum)) { | |
732 | iov_iter_revert(&msg->msg_iter, chunk); | |
733 | return -EINVAL; | |
734 | } | |
735 | ||
db4f1be3 ST |
736 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && |
737 | !skb->csum_complete_sw) | |
7fe50ac8 | 738 | netdev_rx_csum_fault(NULL, skb); |
1da177e4 LT |
739 | } |
740 | return 0; | |
1da177e4 LT |
741 | fault: |
742 | return -EFAULT; | |
743 | } | |
e5a4b0bb | 744 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); |
1da177e4 LT |
745 | |
746 | /** | |
747 | * datagram_poll - generic datagram poll | |
a11e1d43 | 748 | * @file: file struct |
4dc3b16b | 749 | * @sock: socket |
a11e1d43 | 750 | * @wait: poll table |
1da177e4 LT |
751 | * |
752 | * Datagram poll: Again totally generic. This also handles | |
753 | * sequenced packet sockets providing the socket receive queue | |
754 | * is only ever holding data ready to receive. | |
755 | * | |
d3f6cd9e | 756 | * Note: when you *don't* use this routine for this protocol, |
1da177e4 LT |
757 | * and you use a different write policy from sock_writeable() |
758 | * then please supply your own write_space callback. | |
759 | */ | |
a11e1d43 LT |
760 | __poll_t datagram_poll(struct file *file, struct socket *sock, |
761 | poll_table *wait) | |
1da177e4 LT |
762 | { |
763 | struct sock *sk = sock->sk; | |
a11e1d43 LT |
764 | __poll_t mask; |
765 | ||
89ab066d | 766 | sock_poll_wait(file, sock, wait); |
a11e1d43 | 767 | mask = 0; |
1da177e4 LT |
768 | |
769 | /* exceptional events? */ | |
770 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) | |
a9a08845 LT |
771 | mask |= EPOLLERR | |
772 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0); | |
7d4c04fc | 773 | |
f348d70a | 774 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
a9a08845 | 775 | mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM; |
1da177e4 | 776 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
a9a08845 | 777 | mask |= EPOLLHUP; |
1da177e4 LT |
778 | |
779 | /* readable? */ | |
db40980f | 780 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
a9a08845 | 781 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 LT |
782 | |
783 | /* Connection-based need to check for termination and startup */ | |
784 | if (connection_based(sk)) { | |
785 | if (sk->sk_state == TCP_CLOSE) | |
a9a08845 | 786 | mask |= EPOLLHUP; |
1da177e4 LT |
787 | /* connection hasn't started yet? */ |
788 | if (sk->sk_state == TCP_SYN_SENT) | |
789 | return mask; | |
790 | } | |
791 | ||
792 | /* writable? */ | |
793 | if (sock_writeable(sk)) | |
a9a08845 | 794 | mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND; |
1da177e4 | 795 | else |
9cd3e072 | 796 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
797 | |
798 | return mask; | |
799 | } | |
a11e1d43 | 800 | EXPORT_SYMBOL(datagram_poll); |