1 // SPDX-License-Identifier: GPL-2.0
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 *
13 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old
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>
40 #include <linux/uaccess.h>
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>
46 #include <linux/netdevice.h>
47 #include <linux/rtnetlink.h>
48 #include <linux/poll.h>
49 #include <linux/highmem.h>
50 #include <linux/spinlock.h>
51 #include <linux/slab.h>
52 #include <linux/pagemap.h>
53 #include <linux/iov_iter.h>
54 #include <linux/indirect_call_wrapper.h>
55
56 #include <net/protocol.h>
57 #include <linux/skbuff.h>
58
59 #include <net/checksum.h>
60 #include <net/sock.h>
61 #include <net/tcp_states.h>
62 #include <trace/events/skb.h>
63 #include <net/busy_poll.h>
64 #include <crypto/hash.h>
65
66 /*
67 * Is a socket 'connection oriented' ?
68 */
connection_based(struct sock * sk)69 static inline int connection_based(struct sock *sk)
70 {
71 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
72 }
73
receiver_wake_function(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)74 static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
75 void *key)
76 {
77 /*
78 * Avoid a wakeup if event not interesting for us
79 */
80 if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR)))
81 return 0;
82 return autoremove_wake_function(wait, mode, sync, key);
83 }
84 /*
85 * Wait for the last received packet to be different from skb
86 */
__skb_wait_for_more_packets(struct sock * sk,struct sk_buff_head * queue,int * err,long * timeo_p,const struct sk_buff * skb)87 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
88 int *err, long *timeo_p,
89 const struct sk_buff *skb)
90 {
91 int error;
92 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
93
94 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
95
96 /* Socket errors? */
97 error = sock_error(sk);
98 if (error)
99 goto out_err;
100
101 if (READ_ONCE(queue->prev) != skb)
102 goto out;
103
104 /* Socket shut down? */
105 if (sk->sk_shutdown & RCV_SHUTDOWN)
106 goto out_noerr;
107
108 /* Sequenced packets can come disconnected.
109 * If so we report the problem
110 */
111 error = -ENOTCONN;
112 if (connection_based(sk) &&
113 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
114 goto out_err;
115
116 /* handle signals */
117 if (signal_pending(current))
118 goto interrupted;
119
120 error = 0;
121 *timeo_p = schedule_timeout(*timeo_p);
122 out:
123 finish_wait(sk_sleep(sk), &wait);
124 return error;
125 interrupted:
126 error = sock_intr_errno(*timeo_p);
127 out_err:
128 *err = error;
129 goto out;
130 out_noerr:
131 *err = 0;
132 error = 1;
133 goto out;
134 }
135 EXPORT_SYMBOL(__skb_wait_for_more_packets);
136
skb_set_peeked(struct sk_buff * skb)137 static struct sk_buff *skb_set_peeked(struct sk_buff *skb)
138 {
139 struct sk_buff *nskb;
140
141 if (skb->peeked)
142 return skb;
143
144 /* We have to unshare an skb before modifying it. */
145 if (!skb_shared(skb))
146 goto done;
147
148 nskb = skb_clone(skb, GFP_ATOMIC);
149 if (!nskb)
150 return ERR_PTR(-ENOMEM);
151
152 skb->prev->next = nskb;
153 skb->next->prev = nskb;
154 nskb->prev = skb->prev;
155 nskb->next = skb->next;
156
157 consume_skb(skb);
158 skb = nskb;
159
160 done:
161 skb->peeked = 1;
162
163 return skb;
164 }
165
__skb_try_recv_from_queue(struct sock * sk,struct sk_buff_head * queue,unsigned int flags,int * off,int * err,struct sk_buff ** last)166 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
167 struct sk_buff_head *queue,
168 unsigned int flags,
169 int *off, int *err,
170 struct sk_buff **last)
171 {
172 bool peek_at_off = false;
173 struct sk_buff *skb;
174 int _off = 0;
175
176 if (unlikely(flags & MSG_PEEK && *off >= 0)) {
177 peek_at_off = true;
178 _off = *off;
179 }
180
181 *last = queue->prev;
182 skb_queue_walk(queue, skb) {
183 if (flags & MSG_PEEK) {
184 if (peek_at_off && _off >= skb->len &&
185 (_off || skb->peeked)) {
186 _off -= skb->len;
187 continue;
188 }
189 if (!skb->len) {
190 skb = skb_set_peeked(skb);
191 if (IS_ERR(skb)) {
192 *err = PTR_ERR(skb);
193 return NULL;
194 }
195 }
196 refcount_inc(&skb->users);
197 } else {
198 __skb_unlink(skb, queue);
199 }
200 *off = _off;
201 return skb;
202 }
203 return NULL;
204 }
205
206 /**
207 * __skb_try_recv_datagram - Receive a datagram skbuff
208 * @sk: socket
209 * @queue: socket queue from which to receive
210 * @flags: MSG\_ flags
211 * @off: an offset in bytes to peek skb from. Returns an offset
212 * within an skb where data actually starts
213 * @err: error code returned
214 * @last: set to last peeked message to inform the wait function
215 * what to look for when peeking
216 *
217 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
218 * and possible races. This replaces identical code in packet, raw and
219 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
220 * the long standing peek and read race for datagram sockets. If you
221 * alter this routine remember it must be re-entrant.
222 *
223 * This function will lock the socket if a skb is returned, so
224 * the caller needs to unlock the socket in that case (usually by
225 * calling skb_free_datagram). Returns NULL with @err set to
226 * -EAGAIN if no data was available or to some other value if an
227 * error was detected.
228 *
229 * * It does not lock socket since today. This function is
230 * * free of race conditions. This measure should/can improve
231 * * significantly datagram socket latencies at high loads,
232 * * when data copying to user space takes lots of time.
233 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
234 * * 8) Great win.)
235 * * --ANK (980729)
236 *
237 * The order of the tests when we find no data waiting are specified
238 * quite explicitly by POSIX 1003.1g, don't change them without having
239 * the standard around please.
240 */
__skb_try_recv_datagram(struct sock * sk,struct sk_buff_head * queue,unsigned int flags,int * off,int * err,struct sk_buff ** last)241 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
242 struct sk_buff_head *queue,
243 unsigned int flags, int *off, int *err,
244 struct sk_buff **last)
245 {
246 struct sk_buff *skb;
247 unsigned long cpu_flags;
248 /*
249 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
250 */
251 int error = sock_error(sk);
252
253 if (error)
254 goto no_packet;
255
256 do {
257 /* Again only user level code calls this function, so nothing
258 * interrupt level will suddenly eat the receive_queue.
259 *
260 * Look at current nfs client by the way...
261 * However, this function was correct in any case. 8)
262 */
263 spin_lock_irqsave(&queue->lock, cpu_flags);
264 skb = __skb_try_recv_from_queue(sk, queue, flags, off, &error,
265 last);
266 spin_unlock_irqrestore(&queue->lock, cpu_flags);
267 if (error)
268 goto no_packet;
269 if (skb)
270 return skb;
271
272 if (!sk_can_busy_loop(sk))
273 break;
274
275 sk_busy_loop(sk, flags & MSG_DONTWAIT);
276 } while (READ_ONCE(queue->prev) != *last);
277
278 error = -EAGAIN;
279
280 no_packet:
281 *err = error;
282 return NULL;
283 }
284 EXPORT_SYMBOL(__skb_try_recv_datagram);
285
__skb_recv_datagram(struct sock * sk,struct sk_buff_head * sk_queue,unsigned int flags,int * off,int * err)286 struct sk_buff *__skb_recv_datagram(struct sock *sk,
287 struct sk_buff_head *sk_queue,
288 unsigned int flags, int *off, int *err)
289 {
290 struct sk_buff *skb, *last;
291 long timeo;
292
293 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
294
295 do {
296 skb = __skb_try_recv_datagram(sk, sk_queue, flags, off, err,
297 &last);
298 if (skb)
299 return skb;
300
301 if (*err != -EAGAIN)
302 break;
303 } while (timeo &&
304 !__skb_wait_for_more_packets(sk, sk_queue, err,
305 &timeo, last));
306
307 return NULL;
308 }
309 EXPORT_SYMBOL(__skb_recv_datagram);
310
skb_recv_datagram(struct sock * sk,unsigned int flags,int * err)311 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
312 int *err)
313 {
314 int off = 0;
315
316 return __skb_recv_datagram(sk, &sk->sk_receive_queue, flags,
317 &off, err);
318 }
319 EXPORT_SYMBOL(skb_recv_datagram);
320
skb_free_datagram(struct sock * sk,struct sk_buff * skb)321 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
322 {
323 consume_skb(skb);
324 }
325 EXPORT_SYMBOL(skb_free_datagram);
326
__sk_queue_drop_skb(struct sock * sk,struct sk_buff_head * sk_queue,struct sk_buff * skb,unsigned int flags,void (* destructor)(struct sock * sk,struct sk_buff * skb))327 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
328 struct sk_buff *skb, unsigned int flags,
329 void (*destructor)(struct sock *sk,
330 struct sk_buff *skb))
331 {
332 int err = 0;
333
334 if (flags & MSG_PEEK) {
335 err = -ENOENT;
336 spin_lock_bh(&sk_queue->lock);
337 if (skb->next) {
338 __skb_unlink(skb, sk_queue);
339 refcount_dec(&skb->users);
340 if (destructor)
341 destructor(sk, skb);
342 err = 0;
343 }
344 spin_unlock_bh(&sk_queue->lock);
345 }
346
347 atomic_inc(&sk->sk_drops);
348 return err;
349 }
350 EXPORT_SYMBOL(__sk_queue_drop_skb);
351
352 /**
353 * skb_kill_datagram - Free a datagram skbuff forcibly
354 * @sk: socket
355 * @skb: datagram skbuff
356 * @flags: MSG\_ flags
357 *
358 * This function frees a datagram skbuff that was received by
359 * skb_recv_datagram. The flags argument must match the one
360 * used for skb_recv_datagram.
361 *
362 * If the MSG_PEEK flag is set, and the packet is still on the
363 * receive queue of the socket, it will be taken off the queue
364 * before it is freed.
365 *
366 * This function currently only disables BH when acquiring the
367 * sk_receive_queue lock. Therefore it must not be used in a
368 * context where that lock is acquired in an IRQ context.
369 *
370 * It returns 0 if the packet was removed by us.
371 */
372
skb_kill_datagram(struct sock * sk,struct sk_buff * skb,unsigned int flags)373 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
374 {
375 int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags,
376 NULL);
377
378 kfree_skb(skb);
379 return err;
380 }
381 EXPORT_SYMBOL(skb_kill_datagram);
382
383 INDIRECT_CALLABLE_DECLARE(static size_t simple_copy_to_iter(const void *addr,
384 size_t bytes,
385 void *data __always_unused,
386 struct iov_iter *i));
387
__skb_datagram_iter(const struct sk_buff * skb,int offset,struct iov_iter * to,int len,bool fault_short,size_t (* cb)(const void *,size_t,void *,struct iov_iter *),void * data)388 static int __skb_datagram_iter(const struct sk_buff *skb, int offset,
389 struct iov_iter *to, int len, bool fault_short,
390 size_t (*cb)(const void *, size_t, void *,
391 struct iov_iter *), void *data)
392 {
393 int start = skb_headlen(skb);
394 int i, copy = start - offset, start_off = offset, n;
395 struct sk_buff *frag_iter;
396
397 /* Copy header. */
398 if (copy > 0) {
399 if (copy > len)
400 copy = len;
401 n = INDIRECT_CALL_1(cb, simple_copy_to_iter,
402 skb->data + offset, copy, data, to);
403 offset += n;
404 if (n != copy)
405 goto short_copy;
406 if ((len -= copy) == 0)
407 return 0;
408 }
409
410 /* Copy paged appendix. Hmm... why does this look so complicated? */
411 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
412 int end;
413 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
414
415 WARN_ON(start > offset + len);
416
417 end = start + skb_frag_size(frag);
418 if ((copy = end - offset) > 0) {
419 struct page *page = skb_frag_page(frag);
420 u8 *vaddr = kmap(page);
421
422 if (copy > len)
423 copy = len;
424 n = INDIRECT_CALL_1(cb, simple_copy_to_iter,
425 vaddr + skb_frag_off(frag) + offset - start,
426 copy, data, to);
427 kunmap(page);
428 offset += n;
429 if (n != copy)
430 goto short_copy;
431 if (!(len -= copy))
432 return 0;
433 }
434 start = end;
435 }
436
437 skb_walk_frags(skb, frag_iter) {
438 int end;
439
440 WARN_ON(start > offset + len);
441
442 end = start + frag_iter->len;
443 if ((copy = end - offset) > 0) {
444 if (copy > len)
445 copy = len;
446 if (__skb_datagram_iter(frag_iter, offset - start,
447 to, copy, fault_short, cb, data))
448 goto fault;
449 if ((len -= copy) == 0)
450 return 0;
451 offset += copy;
452 }
453 start = end;
454 }
455 if (!len)
456 return 0;
457
458 /* This is not really a user copy fault, but rather someone
459 * gave us a bogus length on the skb. We should probably
460 * print a warning here as it may indicate a kernel bug.
461 */
462
463 fault:
464 iov_iter_revert(to, offset - start_off);
465 return -EFAULT;
466
467 short_copy:
468 if (fault_short || iov_iter_count(to))
469 goto fault;
470
471 return 0;
472 }
473
hash_and_copy_to_iter(const void * addr,size_t bytes,void * hashp,struct iov_iter * i)474 static size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
475 struct iov_iter *i)
476 {
477 #ifdef CONFIG_CRYPTO_HASH
478 struct ahash_request *hash = hashp;
479 struct scatterlist sg;
480 size_t copied;
481
482 copied = copy_to_iter(addr, bytes, i);
483 sg_init_one(&sg, addr, copied);
484 ahash_request_set_crypt(hash, &sg, NULL, copied);
485 crypto_ahash_update(hash);
486 return copied;
487 #else
488 return 0;
489 #endif
490 }
491
492 /**
493 * skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator
494 * and update a hash.
495 * @skb: buffer to copy
496 * @offset: offset in the buffer to start copying from
497 * @to: iovec iterator to copy to
498 * @len: amount of data to copy from buffer to iovec
499 * @hash: hash request to update
500 */
skb_copy_and_hash_datagram_iter(const struct sk_buff * skb,int offset,struct iov_iter * to,int len,struct ahash_request * hash)501 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
502 struct iov_iter *to, int len,
503 struct ahash_request *hash)
504 {
505 return __skb_datagram_iter(skb, offset, to, len, true,
506 hash_and_copy_to_iter, hash);
507 }
508 EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter);
509
simple_copy_to_iter(const void * addr,size_t bytes,void * data __always_unused,struct iov_iter * i)510 static size_t simple_copy_to_iter(const void *addr, size_t bytes,
511 void *data __always_unused, struct iov_iter *i)
512 {
513 return copy_to_iter(addr, bytes, i);
514 }
515
516 /**
517 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
518 * @skb: buffer to copy
519 * @offset: offset in the buffer to start copying from
520 * @to: iovec iterator to copy to
521 * @len: amount of data to copy from buffer to iovec
522 */
skb_copy_datagram_iter(const struct sk_buff * skb,int offset,struct iov_iter * to,int len)523 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset,
524 struct iov_iter *to, int len)
525 {
526 trace_skb_copy_datagram_iovec(skb, len);
527 return __skb_datagram_iter(skb, offset, to, len, false,
528 simple_copy_to_iter, NULL);
529 }
530 EXPORT_SYMBOL(skb_copy_datagram_iter);
531
532 /**
533 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
534 * @skb: buffer to copy
535 * @offset: offset in the buffer to start copying to
536 * @from: the copy source
537 * @len: amount of data to copy to buffer from iovec
538 *
539 * Returns 0 or -EFAULT.
540 */
skb_copy_datagram_from_iter(struct sk_buff * skb,int offset,struct iov_iter * from,int len)541 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
542 struct iov_iter *from,
543 int len)
544 {
545 int start = skb_headlen(skb);
546 int i, copy = start - offset;
547 struct sk_buff *frag_iter;
548
549 /* Copy header. */
550 if (copy > 0) {
551 if (copy > len)
552 copy = len;
553 if (copy_from_iter(skb->data + offset, copy, from) != copy)
554 goto fault;
555 if ((len -= copy) == 0)
556 return 0;
557 offset += copy;
558 }
559
560 /* Copy paged appendix. Hmm... why does this look so complicated? */
561 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
562 int end;
563 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
564
565 WARN_ON(start > offset + len);
566
567 end = start + skb_frag_size(frag);
568 if ((copy = end - offset) > 0) {
569 size_t copied;
570
571 if (copy > len)
572 copy = len;
573 copied = copy_page_from_iter(skb_frag_page(frag),
574 skb_frag_off(frag) + offset - start,
575 copy, from);
576 if (copied != copy)
577 goto fault;
578
579 if (!(len -= copy))
580 return 0;
581 offset += copy;
582 }
583 start = end;
584 }
585
586 skb_walk_frags(skb, frag_iter) {
587 int end;
588
589 WARN_ON(start > offset + len);
590
591 end = start + frag_iter->len;
592 if ((copy = end - offset) > 0) {
593 if (copy > len)
594 copy = len;
595 if (skb_copy_datagram_from_iter(frag_iter,
596 offset - start,
597 from, copy))
598 goto fault;
599 if ((len -= copy) == 0)
600 return 0;
601 offset += copy;
602 }
603 start = end;
604 }
605 if (!len)
606 return 0;
607
608 fault:
609 return -EFAULT;
610 }
611 EXPORT_SYMBOL(skb_copy_datagram_from_iter);
612
__zerocopy_sg_from_iter(struct msghdr * msg,struct sock * sk,struct sk_buff * skb,struct iov_iter * from,size_t length)613 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
614 struct sk_buff *skb, struct iov_iter *from,
615 size_t length)
616 {
617 int frag;
618
619 if (msg && msg->msg_ubuf && msg->sg_from_iter)
620 return msg->sg_from_iter(sk, skb, from, length);
621
622 frag = skb_shinfo(skb)->nr_frags;
623
624 while (length && iov_iter_count(from)) {
625 struct page *head, *last_head = NULL;
626 struct page *pages[MAX_SKB_FRAGS];
627 int refs, order, n = 0;
628 size_t start;
629 ssize_t copied;
630 unsigned long truesize;
631
632 if (frag == MAX_SKB_FRAGS)
633 return -EMSGSIZE;
634
635 copied = iov_iter_get_pages2(from, pages, length,
636 MAX_SKB_FRAGS - frag, &start);
637 if (copied < 0)
638 return -EFAULT;
639
640 length -= copied;
641
642 truesize = PAGE_ALIGN(copied + start);
643 skb->data_len += copied;
644 skb->len += copied;
645 skb->truesize += truesize;
646 if (sk && sk->sk_type == SOCK_STREAM) {
647 sk_wmem_queued_add(sk, truesize);
648 if (!skb_zcopy_pure(skb))
649 sk_mem_charge(sk, truesize);
650 } else {
651 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
652 }
653
654 head = compound_head(pages[n]);
655 order = compound_order(head);
656
657 for (refs = 0; copied != 0; start = 0) {
658 int size = min_t(int, copied, PAGE_SIZE - start);
659
660 if (pages[n] - head > (1UL << order) - 1) {
661 head = compound_head(pages[n]);
662 order = compound_order(head);
663 }
664
665 start += (pages[n] - head) << PAGE_SHIFT;
666 copied -= size;
667 n++;
668 if (frag) {
669 skb_frag_t *last = &skb_shinfo(skb)->frags[frag - 1];
670
671 if (head == skb_frag_page(last) &&
672 start == skb_frag_off(last) + skb_frag_size(last)) {
673 skb_frag_size_add(last, size);
674 /* We combined this page, we need to release
675 * a reference. Since compound pages refcount
676 * is shared among many pages, batch the refcount
677 * adjustments to limit false sharing.
678 */
679 last_head = head;
680 refs++;
681 continue;
682 }
683 }
684 if (refs) {
685 page_ref_sub(last_head, refs);
686 refs = 0;
687 }
688 skb_fill_page_desc_noacc(skb, frag++, head, start, size);
689 }
690 if (refs)
691 page_ref_sub(last_head, refs);
692 }
693 return 0;
694 }
695 EXPORT_SYMBOL(__zerocopy_sg_from_iter);
696
697 /**
698 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
699 * @skb: buffer to copy
700 * @from: the source to copy from
701 *
702 * The function will first copy up to headlen, and then pin the userspace
703 * pages and build frags through them.
704 *
705 * Returns 0, -EFAULT or -EMSGSIZE.
706 */
zerocopy_sg_from_iter(struct sk_buff * skb,struct iov_iter * from)707 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from)
708 {
709 int copy = min_t(int, skb_headlen(skb), iov_iter_count(from));
710
711 /* copy up to skb headlen */
712 if (skb_copy_datagram_from_iter(skb, 0, from, copy))
713 return -EFAULT;
714
715 return __zerocopy_sg_from_iter(NULL, NULL, skb, from, ~0U);
716 }
717 EXPORT_SYMBOL(zerocopy_sg_from_iter);
718
719 static __always_inline
copy_to_user_iter_csum(void __user * iter_to,size_t progress,size_t len,void * from,void * priv2)720 size_t copy_to_user_iter_csum(void __user *iter_to, size_t progress,
721 size_t len, void *from, void *priv2)
722 {
723 __wsum next, *csum = priv2;
724
725 next = csum_and_copy_to_user(from + progress, iter_to, len);
726 *csum = csum_block_add(*csum, next, progress);
727 return next ? 0 : len;
728 }
729
730 static __always_inline
memcpy_to_iter_csum(void * iter_to,size_t progress,size_t len,void * from,void * priv2)731 size_t memcpy_to_iter_csum(void *iter_to, size_t progress,
732 size_t len, void *from, void *priv2)
733 {
734 __wsum *csum = priv2;
735 __wsum next = csum_partial_copy_nocheck(from + progress, iter_to, len);
736
737 *csum = csum_block_add(*csum, next, progress);
738 return 0;
739 }
740
741 struct csum_state {
742 __wsum csum;
743 size_t off;
744 };
745
csum_and_copy_to_iter(const void * addr,size_t bytes,void * _csstate,struct iov_iter * i)746 static size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
747 struct iov_iter *i)
748 {
749 struct csum_state *csstate = _csstate;
750 __wsum sum;
751
752 if (WARN_ON_ONCE(i->data_source))
753 return 0;
754 if (unlikely(iov_iter_is_discard(i))) {
755 // can't use csum_memcpy() for that one - data is not copied
756 csstate->csum = csum_block_add(csstate->csum,
757 csum_partial(addr, bytes, 0),
758 csstate->off);
759 csstate->off += bytes;
760 return bytes;
761 }
762
763 sum = csum_shift(csstate->csum, csstate->off);
764
765 bytes = iterate_and_advance2(i, bytes, (void *)addr, &sum,
766 copy_to_user_iter_csum,
767 memcpy_to_iter_csum);
768 csstate->csum = csum_shift(sum, csstate->off);
769 csstate->off += bytes;
770 return bytes;
771 }
772
773 /**
774 * skb_copy_and_csum_datagram - Copy datagram to an iovec iterator
775 * and update a checksum.
776 * @skb: buffer to copy
777 * @offset: offset in the buffer to start copying from
778 * @to: iovec iterator to copy to
779 * @len: amount of data to copy from buffer to iovec
780 * @csump: checksum pointer
781 */
skb_copy_and_csum_datagram(const struct sk_buff * skb,int offset,struct iov_iter * to,int len,__wsum * csump)782 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
783 struct iov_iter *to, int len,
784 __wsum *csump)
785 {
786 struct csum_state csdata = { .csum = *csump };
787 int ret;
788
789 ret = __skb_datagram_iter(skb, offset, to, len, true,
790 csum_and_copy_to_iter, &csdata);
791 if (ret)
792 return ret;
793
794 *csump = csdata.csum;
795 return 0;
796 }
797
798 /**
799 * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec.
800 * @skb: skbuff
801 * @hlen: hardware length
802 * @msg: destination
803 *
804 * Caller _must_ check that skb will fit to this iovec.
805 *
806 * Returns: 0 - success.
807 * -EINVAL - checksum failure.
808 * -EFAULT - fault during copy.
809 */
skb_copy_and_csum_datagram_msg(struct sk_buff * skb,int hlen,struct msghdr * msg)810 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb,
811 int hlen, struct msghdr *msg)
812 {
813 __wsum csum;
814 int chunk = skb->len - hlen;
815
816 if (!chunk)
817 return 0;
818
819 if (msg_data_left(msg) < chunk) {
820 if (__skb_checksum_complete(skb))
821 return -EINVAL;
822 if (skb_copy_datagram_msg(skb, hlen, msg, chunk))
823 goto fault;
824 } else {
825 csum = csum_partial(skb->data, hlen, skb->csum);
826 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter,
827 chunk, &csum))
828 goto fault;
829
830 if (csum_fold(csum)) {
831 iov_iter_revert(&msg->msg_iter, chunk);
832 return -EINVAL;
833 }
834
835 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
836 !skb->csum_complete_sw)
837 netdev_rx_csum_fault(NULL, skb);
838 }
839 return 0;
840 fault:
841 return -EFAULT;
842 }
843 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg);
844
845 /**
846 * datagram_poll - generic datagram poll
847 * @file: file struct
848 * @sock: socket
849 * @wait: poll table
850 *
851 * Datagram poll: Again totally generic. This also handles
852 * sequenced packet sockets providing the socket receive queue
853 * is only ever holding data ready to receive.
854 *
855 * Note: when you *don't* use this routine for this protocol,
856 * and you use a different write policy from sock_writeable()
857 * then please supply your own write_space callback.
858 */
datagram_poll(struct file * file,struct socket * sock,poll_table * wait)859 __poll_t datagram_poll(struct file *file, struct socket *sock,
860 poll_table *wait)
861 {
862 struct sock *sk = sock->sk;
863 __poll_t mask;
864 u8 shutdown;
865
866 sock_poll_wait(file, sock, wait);
867 mask = 0;
868
869 /* exceptional events? */
870 if (READ_ONCE(sk->sk_err) ||
871 !skb_queue_empty_lockless(&sk->sk_error_queue))
872 mask |= EPOLLERR |
873 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
874
875 shutdown = READ_ONCE(sk->sk_shutdown);
876 if (shutdown & RCV_SHUTDOWN)
877 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
878 if (shutdown == SHUTDOWN_MASK)
879 mask |= EPOLLHUP;
880
881 /* readable? */
882 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
883 mask |= EPOLLIN | EPOLLRDNORM;
884
885 /* Connection-based need to check for termination and startup */
886 if (connection_based(sk)) {
887 int state = READ_ONCE(sk->sk_state);
888
889 if (state == TCP_CLOSE)
890 mask |= EPOLLHUP;
891 /* connection hasn't started yet? */
892 if (state == TCP_SYN_SENT)
893 return mask;
894 }
895
896 /* writable? */
897 if (sock_writeable(sk))
898 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
899 else
900 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
901
902 return mask;
903 }
904 EXPORT_SYMBOL(datagram_poll);
905