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.
12 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old
16 * Alan Cox : NULL return from skb_peek_copy()
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
24 * Alan Cox : Fixed write poll of non IP protocol
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.
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/kernel.h>
39 #include <asm/uaccess.h>
40 #include <asm/system.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>
53 #include <net/protocol.h>
54 #include <linux/skbuff.h>
56 #include <net/checksum.h>
58 #include <net/tcp_states.h>
59 #include <trace/events/skb.h>
62 * Is a socket 'connection oriented' ?
64 static inline int connection_based(struct sock *sk)
66 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
69 static int receiver_wake_function(wait_queue_t *wait, unsigned mode, int sync,
72 unsigned long bits = (unsigned long)key;
75 * Avoid a wakeup if event not interesting for us
77 if (bits && !(bits & (POLLIN | POLLERR)))
79 return autoremove_wake_function(wait, mode, sync, key);
84 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
87 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
89 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
92 error = sock_error(sk);
96 if (!skb_queue_empty(&sk->sk_receive_queue))
99 /* Socket shut down? */
100 if (sk->sk_shutdown & RCV_SHUTDOWN)
103 /* Sequenced packets can come disconnected.
104 * If so we report the problem
107 if (connection_based(sk) &&
108 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
112 if (signal_pending(current))
116 *timeo_p = schedule_timeout(*timeo_p);
118 finish_wait(sk_sleep(sk), &wait);
121 error = sock_intr_errno(*timeo_p);
132 * __skb_recv_datagram - Receive a datagram skbuff
135 * @peeked: returns non-zero if this packet has been seen before
136 * @err: error code returned
138 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
139 * and possible races. This replaces identical code in packet, raw and
140 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
141 * the long standing peek and read race for datagram sockets. If you
142 * alter this routine remember it must be re-entrant.
144 * This function will lock the socket if a skb is returned, so the caller
145 * needs to unlock the socket in that case (usually by calling
148 * * It does not lock socket since today. This function is
149 * * free of race conditions. This measure should/can improve
150 * * significantly datagram socket latencies at high loads,
151 * * when data copying to user space takes lots of time.
152 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
156 * The order of the tests when we find no data waiting are specified
157 * quite explicitly by POSIX 1003.1g, don't change them without having
158 * the standard around please.
160 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
161 int *peeked, int *err)
166 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
168 int error = sock_error(sk);
173 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
176 /* Again only user level code calls this function, so nothing
177 * interrupt level will suddenly eat the receive_queue.
179 * Look at current nfs client by the way...
180 * However, this function was correct in any case. 8)
182 unsigned long cpu_flags;
183 struct sk_buff_head *queue = &sk->sk_receive_queue;
185 spin_lock_irqsave(&queue->lock, cpu_flags);
186 skb = skb_peek(queue);
188 *peeked = skb->peeked;
189 if (flags & MSG_PEEK) {
191 atomic_inc(&skb->users);
193 __skb_unlink(skb, queue);
195 spin_unlock_irqrestore(&queue->lock, cpu_flags);
200 /* User doesn't want to wait */
205 } while (!wait_for_packet(sk, err, &timeo));
213 EXPORT_SYMBOL(__skb_recv_datagram);
215 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
216 int noblock, int *err)
220 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
223 EXPORT_SYMBOL(skb_recv_datagram);
225 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
228 sk_mem_reclaim_partial(sk);
230 EXPORT_SYMBOL(skb_free_datagram);
232 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
236 if (likely(atomic_read(&skb->users) == 1))
238 else if (likely(!atomic_dec_and_test(&skb->users)))
241 slow = lock_sock_fast(sk);
243 sk_mem_reclaim_partial(sk);
244 unlock_sock_fast(sk, slow);
246 /* skb is now orphaned, can be freed outside of locked section */
247 trace_kfree_skb(skb, skb_free_datagram_locked);
250 EXPORT_SYMBOL(skb_free_datagram_locked);
253 * skb_kill_datagram - Free a datagram skbuff forcibly
255 * @skb: datagram skbuff
258 * This function frees a datagram skbuff that was received by
259 * skb_recv_datagram. The flags argument must match the one
260 * used for skb_recv_datagram.
262 * If the MSG_PEEK flag is set, and the packet is still on the
263 * receive queue of the socket, it will be taken off the queue
264 * before it is freed.
266 * This function currently only disables BH when acquiring the
267 * sk_receive_queue lock. Therefore it must not be used in a
268 * context where that lock is acquired in an IRQ context.
270 * It returns 0 if the packet was removed by us.
273 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
277 if (flags & MSG_PEEK) {
279 spin_lock_bh(&sk->sk_receive_queue.lock);
280 if (skb == skb_peek(&sk->sk_receive_queue)) {
281 __skb_unlink(skb, &sk->sk_receive_queue);
282 atomic_dec(&skb->users);
285 spin_unlock_bh(&sk->sk_receive_queue.lock);
289 atomic_inc(&sk->sk_drops);
290 sk_mem_reclaim_partial(sk);
294 EXPORT_SYMBOL(skb_kill_datagram);
297 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
298 * @skb: buffer to copy
299 * @offset: offset in the buffer to start copying from
300 * @to: io vector to copy to
301 * @len: amount of data to copy from buffer to iovec
303 * Note: the iovec is modified during the copy.
305 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
306 struct iovec *to, int len)
308 int start = skb_headlen(skb);
309 int i, copy = start - offset;
310 struct sk_buff *frag_iter;
312 trace_skb_copy_datagram_iovec(skb, len);
318 if (memcpy_toiovec(to, skb->data + offset, copy))
320 if ((len -= copy) == 0)
325 /* Copy paged appendix. Hmm... why does this look so complicated? */
326 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
328 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
330 WARN_ON(start > offset + len);
332 end = start + skb_frag_size(frag);
333 if ((copy = end - offset) > 0) {
336 struct page *page = skb_frag_page(frag);
341 err = memcpy_toiovec(to, vaddr + frag->page_offset +
342 offset - start, copy);
353 skb_walk_frags(skb, frag_iter) {
356 WARN_ON(start > offset + len);
358 end = start + frag_iter->len;
359 if ((copy = end - offset) > 0) {
362 if (skb_copy_datagram_iovec(frag_iter,
366 if ((len -= copy) == 0)
378 EXPORT_SYMBOL(skb_copy_datagram_iovec);
381 * skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
382 * @skb: buffer to copy
383 * @offset: offset in the buffer to start copying from
384 * @to: io vector to copy to
385 * @to_offset: offset in the io vector to start copying to
386 * @len: amount of data to copy from buffer to iovec
388 * Returns 0 or -EFAULT.
389 * Note: the iovec is not modified during the copy.
391 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
392 const struct iovec *to, int to_offset,
395 int start = skb_headlen(skb);
396 int i, copy = start - offset;
397 struct sk_buff *frag_iter;
403 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
405 if ((len -= copy) == 0)
411 /* Copy paged appendix. Hmm... why does this look so complicated? */
412 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
414 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
416 WARN_ON(start > offset + len);
418 end = start + skb_frag_size(frag);
419 if ((copy = end - offset) > 0) {
422 struct page *page = skb_frag_page(frag);
427 err = memcpy_toiovecend(to, vaddr + frag->page_offset +
428 offset - start, to_offset, copy);
440 skb_walk_frags(skb, frag_iter) {
443 WARN_ON(start > offset + len);
445 end = start + frag_iter->len;
446 if ((copy = end - offset) > 0) {
449 if (skb_copy_datagram_const_iovec(frag_iter,
454 if ((len -= copy) == 0)
467 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
470 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
471 * @skb: buffer to copy
472 * @offset: offset in the buffer to start copying to
473 * @from: io vector to copy to
474 * @from_offset: offset in the io vector to start copying from
475 * @len: amount of data to copy to buffer from iovec
477 * Returns 0 or -EFAULT.
478 * Note: the iovec is not modified during the copy.
480 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
481 const struct iovec *from, int from_offset,
484 int start = skb_headlen(skb);
485 int i, copy = start - offset;
486 struct sk_buff *frag_iter;
492 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
495 if ((len -= copy) == 0)
501 /* Copy paged appendix. Hmm... why does this look so complicated? */
502 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
504 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
506 WARN_ON(start > offset + len);
508 end = start + skb_frag_size(frag);
509 if ((copy = end - offset) > 0) {
512 struct page *page = skb_frag_page(frag);
517 err = memcpy_fromiovecend(vaddr + frag->page_offset +
519 from, from_offset, copy);
532 skb_walk_frags(skb, frag_iter) {
535 WARN_ON(start > offset + len);
537 end = start + frag_iter->len;
538 if ((copy = end - offset) > 0) {
541 if (skb_copy_datagram_from_iovec(frag_iter,
547 if ((len -= copy) == 0)
560 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
562 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
563 u8 __user *to, int len,
566 int start = skb_headlen(skb);
567 int i, copy = start - offset;
568 struct sk_buff *frag_iter;
576 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
580 if ((len -= copy) == 0)
587 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
589 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
591 WARN_ON(start > offset + len);
593 end = start + skb_frag_size(frag);
594 if ((copy = end - offset) > 0) {
598 struct page *page = skb_frag_page(frag);
603 csum2 = csum_and_copy_to_user(vaddr +
610 *csump = csum_block_add(*csump, csum2, pos);
620 skb_walk_frags(skb, frag_iter) {
623 WARN_ON(start > offset + len);
625 end = start + frag_iter->len;
626 if ((copy = end - offset) > 0) {
630 if (skb_copy_and_csum_datagram(frag_iter,
635 *csump = csum_block_add(*csump, csum2, pos);
636 if ((len -= copy) == 0)
651 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
655 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
657 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
658 netdev_rx_csum_fault(skb->dev);
659 skb->ip_summed = CHECKSUM_UNNECESSARY;
663 EXPORT_SYMBOL(__skb_checksum_complete_head);
665 __sum16 __skb_checksum_complete(struct sk_buff *skb)
667 return __skb_checksum_complete_head(skb, skb->len);
669 EXPORT_SYMBOL(__skb_checksum_complete);
672 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
674 * @hlen: hardware length
677 * Caller _must_ check that skb will fit to this iovec.
679 * Returns: 0 - success.
680 * -EINVAL - checksum failure.
681 * -EFAULT - fault during copy. Beware, in this case iovec
684 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
685 int hlen, struct iovec *iov)
688 int chunk = skb->len - hlen;
693 /* Skip filled elements.
694 * Pretty silly, look at memcpy_toiovec, though 8)
696 while (!iov->iov_len)
699 if (iov->iov_len < chunk) {
700 if (__skb_checksum_complete(skb))
702 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
705 csum = csum_partial(skb->data, hlen, skb->csum);
706 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
711 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
712 netdev_rx_csum_fault(skb->dev);
713 iov->iov_len -= chunk;
714 iov->iov_base += chunk;
722 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
725 * datagram_poll - generic datagram poll
730 * Datagram poll: Again totally generic. This also handles
731 * sequenced packet sockets providing the socket receive queue
732 * is only ever holding data ready to receive.
734 * Note: when you _don't_ use this routine for this protocol,
735 * and you use a different write policy from sock_writeable()
736 * then please supply your own write_space callback.
738 unsigned int datagram_poll(struct file *file, struct socket *sock,
741 struct sock *sk = sock->sk;
744 sock_poll_wait(file, sk_sleep(sk), wait);
747 /* exceptional events? */
748 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
750 if (sk->sk_shutdown & RCV_SHUTDOWN)
751 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
752 if (sk->sk_shutdown == SHUTDOWN_MASK)
756 if (!skb_queue_empty(&sk->sk_receive_queue))
757 mask |= POLLIN | POLLRDNORM;
759 /* Connection-based need to check for termination and startup */
760 if (connection_based(sk)) {
761 if (sk->sk_state == TCP_CLOSE)
763 /* connection hasn't started yet? */
764 if (sk->sk_state == TCP_SYN_SENT)
769 if (sock_writeable(sk))
770 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
772 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
776 EXPORT_SYMBOL(datagram_poll);