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>
58 #include <trace/events/skb.h>
61 * Is a socket 'connection oriented' ?
63 static inline int connection_based(struct sock *sk)
65 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
68 static int receiver_wake_function(wait_queue_t *wait, unsigned mode, int sync,
71 unsigned long bits = (unsigned long)key;
74 * Avoid a wakeup if event not interesting for us
76 if (bits && !(bits & (POLLIN | POLLERR)))
78 return autoremove_wake_function(wait, mode, sync, key);
83 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
86 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
88 prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
91 error = sock_error(sk);
95 if (!skb_queue_empty(&sk->sk_receive_queue))
98 /* Socket shut down? */
99 if (sk->sk_shutdown & RCV_SHUTDOWN)
102 /* Sequenced packets can come disconnected.
103 * If so we report the problem
106 if (connection_based(sk) &&
107 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
111 if (signal_pending(current))
115 *timeo_p = schedule_timeout(*timeo_p);
117 finish_wait(sk->sk_sleep, &wait);
120 error = sock_intr_errno(*timeo_p);
131 * __skb_recv_datagram - Receive a datagram skbuff
134 * @peeked: returns non-zero if this packet has been seen before
135 * @err: error code returned
137 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
138 * and possible races. This replaces identical code in packet, raw and
139 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
140 * the long standing peek and read race for datagram sockets. If you
141 * alter this routine remember it must be re-entrant.
143 * This function will lock the socket if a skb is returned, so the caller
144 * needs to unlock the socket in that case (usually by calling
147 * * It does not lock socket since today. This function is
148 * * free of race conditions. This measure should/can improve
149 * * significantly datagram socket latencies at high loads,
150 * * when data copying to user space takes lots of time.
151 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
155 * The order of the tests when we find no data waiting are specified
156 * quite explicitly by POSIX 1003.1g, don't change them without having
157 * the standard around please.
159 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
160 int *peeked, int *err)
165 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
167 int error = sock_error(sk);
172 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
175 /* Again only user level code calls this function, so nothing
176 * interrupt level will suddenly eat the receive_queue.
178 * Look at current nfs client by the way...
179 * However, this function was corrent in any case. 8)
181 unsigned long cpu_flags;
183 spin_lock_irqsave(&sk->sk_receive_queue.lock, cpu_flags);
184 skb = skb_peek(&sk->sk_receive_queue);
186 *peeked = skb->peeked;
187 if (flags & MSG_PEEK) {
189 atomic_inc(&skb->users);
191 __skb_unlink(skb, &sk->sk_receive_queue);
193 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, cpu_flags);
198 /* User doesn't want to wait */
203 } while (!wait_for_packet(sk, err, &timeo));
211 EXPORT_SYMBOL(__skb_recv_datagram);
213 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
214 int noblock, int *err)
218 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
222 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
225 sk_mem_reclaim_partial(sk);
227 EXPORT_SYMBOL(skb_free_datagram);
229 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
232 skb_free_datagram(sk, skb);
235 EXPORT_SYMBOL(skb_free_datagram_locked);
238 * skb_kill_datagram - Free a datagram skbuff forcibly
240 * @skb: datagram skbuff
243 * This function frees a datagram skbuff that was received by
244 * skb_recv_datagram. The flags argument must match the one
245 * used for skb_recv_datagram.
247 * If the MSG_PEEK flag is set, and the packet is still on the
248 * receive queue of the socket, it will be taken off the queue
249 * before it is freed.
251 * This function currently only disables BH when acquiring the
252 * sk_receive_queue lock. Therefore it must not be used in a
253 * context where that lock is acquired in an IRQ context.
255 * It returns 0 if the packet was removed by us.
258 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
262 if (flags & MSG_PEEK) {
264 spin_lock_bh(&sk->sk_receive_queue.lock);
265 if (skb == skb_peek(&sk->sk_receive_queue)) {
266 __skb_unlink(skb, &sk->sk_receive_queue);
267 atomic_dec(&skb->users);
270 spin_unlock_bh(&sk->sk_receive_queue.lock);
274 atomic_inc(&sk->sk_drops);
275 sk_mem_reclaim_partial(sk);
280 EXPORT_SYMBOL(skb_kill_datagram);
283 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
284 * @skb: buffer to copy
285 * @offset: offset in the buffer to start copying from
286 * @to: io vector to copy to
287 * @len: amount of data to copy from buffer to iovec
289 * Note: the iovec is modified during the copy.
291 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
292 struct iovec *to, int len)
294 int start = skb_headlen(skb);
295 int i, copy = start - offset;
296 struct sk_buff *frag_iter;
298 trace_skb_copy_datagram_iovec(skb, len);
304 if (memcpy_toiovec(to, skb->data + offset, copy))
306 if ((len -= copy) == 0)
311 /* Copy paged appendix. Hmm... why does this look so complicated? */
312 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
315 WARN_ON(start > offset + len);
317 end = start + skb_shinfo(skb)->frags[i].size;
318 if ((copy = end - offset) > 0) {
321 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
322 struct page *page = frag->page;
327 err = memcpy_toiovec(to, vaddr + frag->page_offset +
328 offset - start, copy);
339 skb_walk_frags(skb, frag_iter) {
342 WARN_ON(start > offset + len);
344 end = start + frag_iter->len;
345 if ((copy = end - offset) > 0) {
348 if (skb_copy_datagram_iovec(frag_iter,
352 if ((len -= copy) == 0)
366 * skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
367 * @skb: buffer to copy
368 * @offset: offset in the buffer to start copying from
369 * @to: io vector to copy to
370 * @to_offset: offset in the io vector to start copying to
371 * @len: amount of data to copy from buffer to iovec
373 * Returns 0 or -EFAULT.
374 * Note: the iovec is not modified during the copy.
376 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
377 const struct iovec *to, int to_offset,
380 int start = skb_headlen(skb);
381 int i, copy = start - offset;
382 struct sk_buff *frag_iter;
388 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
390 if ((len -= copy) == 0)
396 /* Copy paged appendix. Hmm... why does this look so complicated? */
397 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
400 WARN_ON(start > offset + len);
402 end = start + skb_shinfo(skb)->frags[i].size;
403 if ((copy = end - offset) > 0) {
406 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
407 struct page *page = frag->page;
412 err = memcpy_toiovecend(to, vaddr + frag->page_offset +
413 offset - start, to_offset, copy);
425 skb_walk_frags(skb, frag_iter) {
428 WARN_ON(start > offset + len);
430 end = start + frag_iter->len;
431 if ((copy = end - offset) > 0) {
434 if (skb_copy_datagram_const_iovec(frag_iter,
439 if ((len -= copy) == 0)
452 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
455 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
456 * @skb: buffer to copy
457 * @offset: offset in the buffer to start copying to
458 * @from: io vector to copy to
459 * @from_offset: offset in the io vector to start copying from
460 * @len: amount of data to copy to buffer from iovec
462 * Returns 0 or -EFAULT.
463 * Note: the iovec is not modified during the copy.
465 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
466 const struct iovec *from, int from_offset,
469 int start = skb_headlen(skb);
470 int i, copy = start - offset;
471 struct sk_buff *frag_iter;
477 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
480 if ((len -= copy) == 0)
486 /* Copy paged appendix. Hmm... why does this look so complicated? */
487 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
490 WARN_ON(start > offset + len);
492 end = start + skb_shinfo(skb)->frags[i].size;
493 if ((copy = end - offset) > 0) {
496 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
497 struct page *page = frag->page;
502 err = memcpy_fromiovecend(vaddr + frag->page_offset +
504 from, from_offset, copy);
517 skb_walk_frags(skb, frag_iter) {
520 WARN_ON(start > offset + len);
522 end = start + frag_iter->len;
523 if ((copy = end - offset) > 0) {
526 if (skb_copy_datagram_from_iovec(frag_iter,
532 if ((len -= copy) == 0)
545 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
547 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
548 u8 __user *to, int len,
551 int start = skb_headlen(skb);
552 int i, copy = start - offset;
553 struct sk_buff *frag_iter;
561 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
565 if ((len -= copy) == 0)
572 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
575 WARN_ON(start > offset + len);
577 end = start + skb_shinfo(skb)->frags[i].size;
578 if ((copy = end - offset) > 0) {
582 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
583 struct page *page = frag->page;
588 csum2 = csum_and_copy_to_user(vaddr +
595 *csump = csum_block_add(*csump, csum2, pos);
605 skb_walk_frags(skb, frag_iter) {
608 WARN_ON(start > offset + len);
610 end = start + frag_iter->len;
611 if ((copy = end - offset) > 0) {
615 if (skb_copy_and_csum_datagram(frag_iter,
620 *csump = csum_block_add(*csump, csum2, pos);
621 if ((len -= copy) == 0)
636 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
640 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
642 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
643 netdev_rx_csum_fault(skb->dev);
644 skb->ip_summed = CHECKSUM_UNNECESSARY;
648 EXPORT_SYMBOL(__skb_checksum_complete_head);
650 __sum16 __skb_checksum_complete(struct sk_buff *skb)
652 return __skb_checksum_complete_head(skb, skb->len);
654 EXPORT_SYMBOL(__skb_checksum_complete);
657 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
659 * @hlen: hardware length
662 * Caller _must_ check that skb will fit to this iovec.
664 * Returns: 0 - success.
665 * -EINVAL - checksum failure.
666 * -EFAULT - fault during copy. Beware, in this case iovec
669 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
670 int hlen, struct iovec *iov)
673 int chunk = skb->len - hlen;
678 /* Skip filled elements.
679 * Pretty silly, look at memcpy_toiovec, though 8)
681 while (!iov->iov_len)
684 if (iov->iov_len < chunk) {
685 if (__skb_checksum_complete(skb))
687 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
690 csum = csum_partial(skb->data, hlen, skb->csum);
691 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
696 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
697 netdev_rx_csum_fault(skb->dev);
698 iov->iov_len -= chunk;
699 iov->iov_base += chunk;
709 * datagram_poll - generic datagram poll
714 * Datagram poll: Again totally generic. This also handles
715 * sequenced packet sockets providing the socket receive queue
716 * is only ever holding data ready to receive.
718 * Note: when you _don't_ use this routine for this protocol,
719 * and you use a different write policy from sock_writeable()
720 * then please supply your own write_space callback.
722 unsigned int datagram_poll(struct file *file, struct socket *sock,
725 struct sock *sk = sock->sk;
728 sock_poll_wait(file, sk->sk_sleep, wait);
731 /* exceptional events? */
732 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
734 if (sk->sk_shutdown & RCV_SHUTDOWN)
736 if (sk->sk_shutdown == SHUTDOWN_MASK)
740 if (!skb_queue_empty(&sk->sk_receive_queue) ||
741 (sk->sk_shutdown & RCV_SHUTDOWN))
742 mask |= POLLIN | POLLRDNORM;
744 /* Connection-based need to check for termination and startup */
745 if (connection_based(sk)) {
746 if (sk->sk_state == TCP_CLOSE)
748 /* connection hasn't started yet? */
749 if (sk->sk_state == TCP_SYN_SENT)
754 if (sock_writeable(sk))
755 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
757 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
762 EXPORT_SYMBOL(datagram_poll);
763 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
764 EXPORT_SYMBOL(skb_copy_datagram_iovec);
765 EXPORT_SYMBOL(skb_recv_datagram);