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);
131 static struct sk_buff *skb_set_peeked(struct sk_buff *skb)
133 struct sk_buff *nskb;
138 /* We have to unshare an skb before modifying it. */
139 if (!skb_shared(skb))
142 nskb = skb_clone(skb, GFP_ATOMIC);
144 return ERR_PTR(-ENOMEM);
146 skb->prev->next = nskb;
147 skb->next->prev = nskb;
148 nskb->prev = skb->prev;
149 nskb->next = skb->next;
161 * __skb_recv_datagram - Receive a datagram skbuff
164 * @peeked: returns non-zero if this packet has been seen before
165 * @err: error code returned
167 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
168 * and possible races. This replaces identical code in packet, raw and
169 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
170 * the long standing peek and read race for datagram sockets. If you
171 * alter this routine remember it must be re-entrant.
173 * This function will lock the socket if a skb is returned, so the caller
174 * needs to unlock the socket in that case (usually by calling
177 * * It does not lock socket since today. This function is
178 * * free of race conditions. This measure should/can improve
179 * * significantly datagram socket latencies at high loads,
180 * * when data copying to user space takes lots of time.
181 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
185 * The order of the tests when we find no data waiting are specified
186 * quite explicitly by POSIX 1003.1g, don't change them without having
187 * the standard around please.
189 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
190 int *peeked, int *err)
192 struct sk_buff_head *queue = &sk->sk_receive_queue;
194 unsigned long cpu_flags;
197 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
199 int error = sock_error(sk);
204 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
207 /* Again only user level code calls this function, so nothing
208 * interrupt level will suddenly eat the receive_queue.
210 * Look at current nfs client by the way...
211 * However, this function was correct in any case. 8)
213 spin_lock_irqsave(&queue->lock, cpu_flags);
214 skb = skb_peek(queue);
216 *peeked = skb->peeked;
217 if (flags & MSG_PEEK) {
219 skb = skb_set_peeked(skb);
220 error = PTR_ERR(skb);
224 atomic_inc(&skb->users);
226 __skb_unlink(skb, queue);
228 spin_unlock_irqrestore(&queue->lock, cpu_flags);
233 /* User doesn't want to wait */
238 } while (!wait_for_packet(sk, err, &timeo));
243 spin_unlock_irqrestore(&queue->lock, cpu_flags);
248 EXPORT_SYMBOL(__skb_recv_datagram);
250 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
251 int noblock, int *err)
255 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
258 EXPORT_SYMBOL(skb_recv_datagram);
260 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
263 sk_mem_reclaim_partial(sk);
265 EXPORT_SYMBOL(skb_free_datagram);
267 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
271 if (likely(atomic_read(&skb->users) == 1))
273 else if (likely(!atomic_dec_and_test(&skb->users)))
276 slow = lock_sock_fast(sk);
278 sk_mem_reclaim_partial(sk);
279 unlock_sock_fast(sk, slow);
281 /* skb is now orphaned, can be freed outside of locked section */
282 trace_kfree_skb(skb, skb_free_datagram_locked);
285 EXPORT_SYMBOL(skb_free_datagram_locked);
288 * skb_kill_datagram - Free a datagram skbuff forcibly
290 * @skb: datagram skbuff
293 * This function frees a datagram skbuff that was received by
294 * skb_recv_datagram. The flags argument must match the one
295 * used for skb_recv_datagram.
297 * If the MSG_PEEK flag is set, and the packet is still on the
298 * receive queue of the socket, it will be taken off the queue
299 * before it is freed.
301 * This function currently only disables BH when acquiring the
302 * sk_receive_queue lock. Therefore it must not be used in a
303 * context where that lock is acquired in an IRQ context.
305 * It returns 0 if the packet was removed by us.
308 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
312 if (flags & MSG_PEEK) {
314 spin_lock_bh(&sk->sk_receive_queue.lock);
315 if (skb == skb_peek(&sk->sk_receive_queue)) {
316 __skb_unlink(skb, &sk->sk_receive_queue);
317 atomic_dec(&skb->users);
320 spin_unlock_bh(&sk->sk_receive_queue.lock);
324 atomic_inc(&sk->sk_drops);
325 sk_mem_reclaim_partial(sk);
329 EXPORT_SYMBOL(skb_kill_datagram);
332 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
333 * @skb: buffer to copy
334 * @offset: offset in the buffer to start copying from
335 * @to: io vector to copy to
336 * @len: amount of data to copy from buffer to iovec
338 * Note: the iovec is modified during the copy.
340 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
341 struct iovec *to, int len)
343 int start = skb_headlen(skb);
344 int i, copy = start - offset;
345 struct sk_buff *frag_iter;
347 trace_skb_copy_datagram_iovec(skb, len);
353 if (memcpy_toiovec(to, skb->data + offset, copy))
355 if ((len -= copy) == 0)
360 /* Copy paged appendix. Hmm... why does this look so complicated? */
361 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
363 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
365 WARN_ON(start > offset + len);
367 end = start + skb_frag_size(frag);
368 if ((copy = end - offset) > 0) {
371 struct page *page = skb_frag_page(frag);
376 err = memcpy_toiovec(to, vaddr + frag->page_offset +
377 offset - start, copy);
388 skb_walk_frags(skb, frag_iter) {
391 WARN_ON(start > offset + len);
393 end = start + frag_iter->len;
394 if ((copy = end - offset) > 0) {
397 if (skb_copy_datagram_iovec(frag_iter,
401 if ((len -= copy) == 0)
413 EXPORT_SYMBOL(skb_copy_datagram_iovec);
416 * skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
417 * @skb: buffer to copy
418 * @offset: offset in the buffer to start copying from
419 * @to: io vector to copy to
420 * @to_offset: offset in the io vector to start copying to
421 * @len: amount of data to copy from buffer to iovec
423 * Returns 0 or -EFAULT.
424 * Note: the iovec is not modified during the copy.
426 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
427 const struct iovec *to, int to_offset,
430 int start = skb_headlen(skb);
431 int i, copy = start - offset;
432 struct sk_buff *frag_iter;
438 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
440 if ((len -= copy) == 0)
446 /* Copy paged appendix. Hmm... why does this look so complicated? */
447 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
449 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
451 WARN_ON(start > offset + len);
453 end = start + skb_frag_size(frag);
454 if ((copy = end - offset) > 0) {
457 struct page *page = skb_frag_page(frag);
462 err = memcpy_toiovecend(to, vaddr + frag->page_offset +
463 offset - start, to_offset, copy);
475 skb_walk_frags(skb, frag_iter) {
478 WARN_ON(start > offset + len);
480 end = start + frag_iter->len;
481 if ((copy = end - offset) > 0) {
484 if (skb_copy_datagram_const_iovec(frag_iter,
489 if ((len -= copy) == 0)
502 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
505 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
506 * @skb: buffer to copy
507 * @offset: offset in the buffer to start copying to
508 * @from: io vector to copy to
509 * @from_offset: offset in the io vector to start copying from
510 * @len: amount of data to copy to buffer from iovec
512 * Returns 0 or -EFAULT.
513 * Note: the iovec is not modified during the copy.
515 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
516 const struct iovec *from, int from_offset,
519 int start = skb_headlen(skb);
520 int i, copy = start - offset;
521 struct sk_buff *frag_iter;
527 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
530 if ((len -= copy) == 0)
536 /* Copy paged appendix. Hmm... why does this look so complicated? */
537 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
539 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
541 WARN_ON(start > offset + len);
543 end = start + skb_frag_size(frag);
544 if ((copy = end - offset) > 0) {
547 struct page *page = skb_frag_page(frag);
552 err = memcpy_fromiovecend(vaddr + frag->page_offset +
554 from, from_offset, copy);
567 skb_walk_frags(skb, frag_iter) {
570 WARN_ON(start > offset + len);
572 end = start + frag_iter->len;
573 if ((copy = end - offset) > 0) {
576 if (skb_copy_datagram_from_iovec(frag_iter,
582 if ((len -= copy) == 0)
595 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
597 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
598 u8 __user *to, int len,
601 int start = skb_headlen(skb);
602 int i, copy = start - offset;
603 struct sk_buff *frag_iter;
611 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
615 if ((len -= copy) == 0)
622 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
624 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
626 WARN_ON(start > offset + len);
628 end = start + skb_frag_size(frag);
629 if ((copy = end - offset) > 0) {
633 struct page *page = skb_frag_page(frag);
638 csum2 = csum_and_copy_to_user(vaddr +
645 *csump = csum_block_add(*csump, csum2, pos);
655 skb_walk_frags(skb, frag_iter) {
658 WARN_ON(start > offset + len);
660 end = start + frag_iter->len;
661 if ((copy = end - offset) > 0) {
665 if (skb_copy_and_csum_datagram(frag_iter,
670 *csump = csum_block_add(*csump, csum2, pos);
671 if ((len -= copy) == 0)
686 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
690 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
692 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
693 netdev_rx_csum_fault(skb->dev);
694 if (!skb_shared(skb))
695 skb->ip_summed = CHECKSUM_UNNECESSARY;
699 EXPORT_SYMBOL(__skb_checksum_complete_head);
701 __sum16 __skb_checksum_complete(struct sk_buff *skb)
703 return __skb_checksum_complete_head(skb, skb->len);
705 EXPORT_SYMBOL(__skb_checksum_complete);
708 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
710 * @hlen: hardware length
713 * Caller _must_ check that skb will fit to this iovec.
715 * Returns: 0 - success.
716 * -EINVAL - checksum failure.
717 * -EFAULT - fault during copy. Beware, in this case iovec
720 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
721 int hlen, struct iovec *iov)
724 int chunk = skb->len - hlen;
729 /* Skip filled elements.
730 * Pretty silly, look at memcpy_toiovec, though 8)
732 while (!iov->iov_len)
735 if (iov->iov_len < chunk) {
736 if (__skb_checksum_complete(skb))
738 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
741 csum = csum_partial(skb->data, hlen, skb->csum);
742 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
747 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
748 netdev_rx_csum_fault(skb->dev);
749 iov->iov_len -= chunk;
750 iov->iov_base += chunk;
758 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
761 * datagram_poll - generic datagram poll
766 * Datagram poll: Again totally generic. This also handles
767 * sequenced packet sockets providing the socket receive queue
768 * is only ever holding data ready to receive.
770 * Note: when you _don't_ use this routine for this protocol,
771 * and you use a different write policy from sock_writeable()
772 * then please supply your own write_space callback.
774 unsigned int datagram_poll(struct file *file, struct socket *sock,
777 struct sock *sk = sock->sk;
780 sock_poll_wait(file, sk_sleep(sk), wait);
783 /* exceptional events? */
784 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
786 if (sk->sk_shutdown & RCV_SHUTDOWN)
787 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
788 if (sk->sk_shutdown == SHUTDOWN_MASK)
792 if (!skb_queue_empty(&sk->sk_receive_queue))
793 mask |= POLLIN | POLLRDNORM;
795 /* Connection-based need to check for termination and startup */
796 if (connection_based(sk)) {
797 if (sk->sk_state == TCP_CLOSE)
799 /* connection hasn't started yet? */
800 if (sk->sk_state == TCP_SYN_SENT)
805 if (sock_writeable(sk))
806 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
808 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
812 EXPORT_SYMBOL(datagram_poll);