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>
52 #include <net/protocol.h>
53 #include <linux/skbuff.h>
55 #include <net/checksum.h>
57 #include <net/tcp_states.h>
60 * Is a socket 'connection oriented' ?
62 static inline int connection_based(struct sock *sk)
64 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
67 static int receiver_wake_function(wait_queue_t *wait, unsigned mode, int sync,
70 unsigned long bits = (unsigned long)key;
73 * Avoid a wakeup if event not interesting for us
75 if (bits && !(bits & (POLLIN | POLLERR)))
77 return autoremove_wake_function(wait, mode, sync, key);
82 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
85 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
87 prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
90 error = sock_error(sk);
94 if (!skb_queue_empty(&sk->sk_receive_queue))
97 /* Socket shut down? */
98 if (sk->sk_shutdown & RCV_SHUTDOWN)
101 /* Sequenced packets can come disconnected.
102 * If so we report the problem
105 if (connection_based(sk) &&
106 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
110 if (signal_pending(current))
114 *timeo_p = schedule_timeout(*timeo_p);
116 finish_wait(sk->sk_sleep, &wait);
119 error = sock_intr_errno(*timeo_p);
130 * __skb_recv_datagram - Receive a datagram skbuff
133 * @peeked: returns non-zero if this packet has been seen before
134 * @err: error code returned
136 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
137 * and possible races. This replaces identical code in packet, raw and
138 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
139 * the long standing peek and read race for datagram sockets. If you
140 * alter this routine remember it must be re-entrant.
142 * This function will lock the socket if a skb is returned, so the caller
143 * needs to unlock the socket in that case (usually by calling
146 * * It does not lock socket since today. This function is
147 * * free of race conditions. This measure should/can improve
148 * * significantly datagram socket latencies at high loads,
149 * * when data copying to user space takes lots of time.
150 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
154 * The order of the tests when we find no data waiting are specified
155 * quite explicitly by POSIX 1003.1g, don't change them without having
156 * the standard around please.
158 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
159 int *peeked, int *err)
164 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
166 int error = sock_error(sk);
171 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
174 /* Again only user level code calls this function, so nothing
175 * interrupt level will suddenly eat the receive_queue.
177 * Look at current nfs client by the way...
178 * However, this function was corrent in any case. 8)
180 unsigned long cpu_flags;
182 spin_lock_irqsave(&sk->sk_receive_queue.lock, cpu_flags);
183 skb = skb_peek(&sk->sk_receive_queue);
185 *peeked = skb->peeked;
186 if (flags & MSG_PEEK) {
188 atomic_inc(&skb->users);
190 __skb_unlink(skb, &sk->sk_receive_queue);
192 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, cpu_flags);
197 /* User doesn't want to wait */
202 } while (!wait_for_packet(sk, err, &timeo));
210 EXPORT_SYMBOL(__skb_recv_datagram);
212 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
213 int noblock, int *err)
217 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
221 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
224 sk_mem_reclaim_partial(sk);
228 * skb_kill_datagram - Free a datagram skbuff forcibly
230 * @skb: datagram skbuff
233 * This function frees a datagram skbuff that was received by
234 * skb_recv_datagram. The flags argument must match the one
235 * used for skb_recv_datagram.
237 * If the MSG_PEEK flag is set, and the packet is still on the
238 * receive queue of the socket, it will be taken off the queue
239 * before it is freed.
241 * This function currently only disables BH when acquiring the
242 * sk_receive_queue lock. Therefore it must not be used in a
243 * context where that lock is acquired in an IRQ context.
245 * It returns 0 if the packet was removed by us.
248 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
252 if (flags & MSG_PEEK) {
254 spin_lock_bh(&sk->sk_receive_queue.lock);
255 if (skb == skb_peek(&sk->sk_receive_queue)) {
256 __skb_unlink(skb, &sk->sk_receive_queue);
257 atomic_dec(&skb->users);
260 spin_unlock_bh(&sk->sk_receive_queue.lock);
264 sk_mem_reclaim_partial(sk);
269 EXPORT_SYMBOL(skb_kill_datagram);
272 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
273 * @skb: buffer to copy
274 * @offset: offset in the buffer to start copying from
275 * @to: io vector to copy to
276 * @len: amount of data to copy from buffer to iovec
278 * Note: the iovec is modified during the copy.
280 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
281 struct iovec *to, int len)
283 int start = skb_headlen(skb);
284 int i, copy = start - offset;
285 struct sk_buff *frag_iter;
291 if (memcpy_toiovec(to, skb->data + offset, copy))
293 if ((len -= copy) == 0)
298 /* Copy paged appendix. Hmm... why does this look so complicated? */
299 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
302 WARN_ON(start > offset + len);
304 end = start + skb_shinfo(skb)->frags[i].size;
305 if ((copy = end - offset) > 0) {
308 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
309 struct page *page = frag->page;
314 err = memcpy_toiovec(to, vaddr + frag->page_offset +
315 offset - start, copy);
326 skb_walk_frags(skb, frag_iter) {
329 WARN_ON(start > offset + len);
331 end = start + frag_iter->len;
332 if ((copy = end - offset) > 0) {
335 if (skb_copy_datagram_iovec(frag_iter,
339 if ((len -= copy) == 0)
353 * skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
354 * @skb: buffer to copy
355 * @offset: offset in the buffer to start copying from
356 * @to: io vector to copy to
357 * @to_offset: offset in the io vector to start copying to
358 * @len: amount of data to copy from buffer to iovec
360 * Returns 0 or -EFAULT.
361 * Note: the iovec is not modified during the copy.
363 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
364 const struct iovec *to, int to_offset,
367 int start = skb_headlen(skb);
368 int i, copy = start - offset;
369 struct sk_buff *frag_iter;
375 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
377 if ((len -= copy) == 0)
383 /* Copy paged appendix. Hmm... why does this look so complicated? */
384 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
387 WARN_ON(start > offset + len);
389 end = start + skb_shinfo(skb)->frags[i].size;
390 if ((copy = end - offset) > 0) {
393 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
394 struct page *page = frag->page;
399 err = memcpy_toiovecend(to, vaddr + frag->page_offset +
400 offset - start, to_offset, copy);
412 skb_walk_frags(skb, frag_iter) {
415 WARN_ON(start > offset + len);
417 end = start + frag_iter->len;
418 if ((copy = end - offset) > 0) {
421 if (skb_copy_datagram_const_iovec(frag_iter,
426 if ((len -= copy) == 0)
439 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
442 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
443 * @skb: buffer to copy
444 * @offset: offset in the buffer to start copying to
445 * @from: io vector to copy to
446 * @from_offset: offset in the io vector to start copying from
447 * @len: amount of data to copy to buffer from iovec
449 * Returns 0 or -EFAULT.
450 * Note: the iovec is not modified during the copy.
452 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
453 const struct iovec *from, int from_offset,
456 int start = skb_headlen(skb);
457 int i, copy = start - offset;
458 struct sk_buff *frag_iter;
464 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
467 if ((len -= copy) == 0)
473 /* Copy paged appendix. Hmm... why does this look so complicated? */
474 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
477 WARN_ON(start > offset + len);
479 end = start + skb_shinfo(skb)->frags[i].size;
480 if ((copy = end - offset) > 0) {
483 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
484 struct page *page = frag->page;
489 err = memcpy_fromiovecend(vaddr + frag->page_offset +
491 from, from_offset, copy);
504 skb_walk_frags(skb, frag_iter) {
507 WARN_ON(start > offset + len);
509 end = start + frag_iter->len;
510 if ((copy = end - offset) > 0) {
513 if (skb_copy_datagram_from_iovec(frag_iter,
519 if ((len -= copy) == 0)
532 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
534 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
535 u8 __user *to, int len,
538 int start = skb_headlen(skb);
539 int i, copy = start - offset;
540 struct sk_buff *frag_iter;
548 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
552 if ((len -= copy) == 0)
559 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
562 WARN_ON(start > offset + len);
564 end = start + skb_shinfo(skb)->frags[i].size;
565 if ((copy = end - offset) > 0) {
569 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
570 struct page *page = frag->page;
575 csum2 = csum_and_copy_to_user(vaddr +
582 *csump = csum_block_add(*csump, csum2, pos);
592 skb_walk_frags(skb, frag_iter) {
595 WARN_ON(start > offset + len);
597 end = start + frag_iter->len;
598 if ((copy = end - offset) > 0) {
602 if (skb_copy_and_csum_datagram(frag_iter,
607 *csump = csum_block_add(*csump, csum2, pos);
608 if ((len -= copy) == 0)
623 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
627 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
629 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
630 netdev_rx_csum_fault(skb->dev);
631 skb->ip_summed = CHECKSUM_UNNECESSARY;
635 EXPORT_SYMBOL(__skb_checksum_complete_head);
637 __sum16 __skb_checksum_complete(struct sk_buff *skb)
639 return __skb_checksum_complete_head(skb, skb->len);
641 EXPORT_SYMBOL(__skb_checksum_complete);
644 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
646 * @hlen: hardware length
649 * Caller _must_ check that skb will fit to this iovec.
651 * Returns: 0 - success.
652 * -EINVAL - checksum failure.
653 * -EFAULT - fault during copy. Beware, in this case iovec
656 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
657 int hlen, struct iovec *iov)
660 int chunk = skb->len - hlen;
665 /* Skip filled elements.
666 * Pretty silly, look at memcpy_toiovec, though 8)
668 while (!iov->iov_len)
671 if (iov->iov_len < chunk) {
672 if (__skb_checksum_complete(skb))
674 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
677 csum = csum_partial(skb->data, hlen, skb->csum);
678 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
683 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
684 netdev_rx_csum_fault(skb->dev);
685 iov->iov_len -= chunk;
686 iov->iov_base += chunk;
696 * datagram_poll - generic datagram poll
701 * Datagram poll: Again totally generic. This also handles
702 * sequenced packet sockets providing the socket receive queue
703 * is only ever holding data ready to receive.
705 * Note: when you _don't_ use this routine for this protocol,
706 * and you use a different write policy from sock_writeable()
707 * then please supply your own write_space callback.
709 unsigned int datagram_poll(struct file *file, struct socket *sock,
712 struct sock *sk = sock->sk;
715 poll_wait(file, sk->sk_sleep, wait);
718 /* exceptional events? */
719 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
721 if (sk->sk_shutdown & RCV_SHUTDOWN)
723 if (sk->sk_shutdown == SHUTDOWN_MASK)
727 if (!skb_queue_empty(&sk->sk_receive_queue) ||
728 (sk->sk_shutdown & RCV_SHUTDOWN))
729 mask |= POLLIN | POLLRDNORM;
731 /* Connection-based need to check for termination and startup */
732 if (connection_based(sk)) {
733 if (sk->sk_state == TCP_CLOSE)
735 /* connection hasn't started yet? */
736 if (sk->sk_state == TCP_SYN_SENT)
741 if (sock_writeable(sk))
742 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
744 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
749 EXPORT_SYMBOL(datagram_poll);
750 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
751 EXPORT_SYMBOL(skb_copy_datagram_iovec);
752 EXPORT_SYMBOL(skb_free_datagram);
753 EXPORT_SYMBOL(skb_recv_datagram);