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>
41 #include <linux/interrupt.h>
42 #include <linux/errno.h>
43 #include <linux/sched.h>
44 #include <linux/inet.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/poll.h>
48 #include <linux/highmem.h>
49 #include <linux/spinlock.h>
50 #include <linux/slab.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 int 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);
81 * Wait for the last received packet to be different from skb
83 static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
84 const struct sk_buff *skb)
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 (sk->sk_receive_queue.prev != skb)
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 * @off: an offset in bytes to peek skb from. Returns an offset
137 * within an skb where data actually starts
138 * @err: error code returned
140 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
141 * and possible races. This replaces identical code in packet, raw and
142 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
143 * the long standing peek and read race for datagram sockets. If you
144 * alter this routine remember it must be re-entrant.
146 * This function will lock the socket if a skb is returned, so the caller
147 * needs to unlock the socket in that case (usually by calling
150 * * It does not lock socket since today. This function is
151 * * free of race conditions. This measure should/can improve
152 * * significantly datagram socket latencies at high loads,
153 * * when data copying to user space takes lots of time.
154 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
158 * The order of the tests when we find no data waiting are specified
159 * quite explicitly by POSIX 1003.1g, don't change them without having
160 * the standard around please.
162 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
163 int *peeked, int *off, int *err)
165 struct sk_buff *skb, *last;
168 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
170 int error = sock_error(sk);
175 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
178 /* Again only user level code calls this function, so nothing
179 * interrupt level will suddenly eat the receive_queue.
181 * Look at current nfs client by the way...
182 * However, this function was correct in any case. 8)
184 unsigned long cpu_flags;
185 struct sk_buff_head *queue = &sk->sk_receive_queue;
188 last = (struct sk_buff *)queue;
189 spin_lock_irqsave(&queue->lock, cpu_flags);
190 skb_queue_walk(queue, skb) {
192 *peeked = skb->peeked;
193 if (flags & MSG_PEEK) {
194 if (_off >= skb->len && (skb->len || _off ||
200 atomic_inc(&skb->users);
202 __skb_unlink(skb, queue);
204 spin_unlock_irqrestore(&queue->lock, cpu_flags);
208 spin_unlock_irqrestore(&queue->lock, cpu_flags);
210 /* User doesn't want to wait */
215 } while (!wait_for_more_packets(sk, err, &timeo, last));
223 EXPORT_SYMBOL(__skb_recv_datagram);
225 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
226 int noblock, int *err)
230 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
233 EXPORT_SYMBOL(skb_recv_datagram);
235 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
238 sk_mem_reclaim_partial(sk);
240 EXPORT_SYMBOL(skb_free_datagram);
242 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
246 if (likely(atomic_read(&skb->users) == 1))
248 else if (likely(!atomic_dec_and_test(&skb->users)))
251 slow = lock_sock_fast(sk);
253 sk_mem_reclaim_partial(sk);
254 unlock_sock_fast(sk, slow);
256 /* skb is now orphaned, can be freed outside of locked section */
259 EXPORT_SYMBOL(skb_free_datagram_locked);
262 * skb_kill_datagram - Free a datagram skbuff forcibly
264 * @skb: datagram skbuff
267 * This function frees a datagram skbuff that was received by
268 * skb_recv_datagram. The flags argument must match the one
269 * used for skb_recv_datagram.
271 * If the MSG_PEEK flag is set, and the packet is still on the
272 * receive queue of the socket, it will be taken off the queue
273 * before it is freed.
275 * This function currently only disables BH when acquiring the
276 * sk_receive_queue lock. Therefore it must not be used in a
277 * context where that lock is acquired in an IRQ context.
279 * It returns 0 if the packet was removed by us.
282 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
286 if (flags & MSG_PEEK) {
288 spin_lock_bh(&sk->sk_receive_queue.lock);
289 if (skb == skb_peek(&sk->sk_receive_queue)) {
290 __skb_unlink(skb, &sk->sk_receive_queue);
291 atomic_dec(&skb->users);
294 spin_unlock_bh(&sk->sk_receive_queue.lock);
298 atomic_inc(&sk->sk_drops);
299 sk_mem_reclaim_partial(sk);
303 EXPORT_SYMBOL(skb_kill_datagram);
306 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
307 * @skb: buffer to copy
308 * @offset: offset in the buffer to start copying from
309 * @to: io vector to copy to
310 * @len: amount of data to copy from buffer to iovec
312 * Note: the iovec is modified during the copy.
314 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
315 struct iovec *to, int len)
317 int start = skb_headlen(skb);
318 int i, copy = start - offset;
319 struct sk_buff *frag_iter;
321 trace_skb_copy_datagram_iovec(skb, len);
327 if (memcpy_toiovec(to, skb->data + offset, copy))
329 if ((len -= copy) == 0)
334 /* Copy paged appendix. Hmm... why does this look so complicated? */
335 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
337 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
339 WARN_ON(start > offset + len);
341 end = start + skb_frag_size(frag);
342 if ((copy = end - offset) > 0) {
345 struct page *page = skb_frag_page(frag);
350 err = memcpy_toiovec(to, vaddr + frag->page_offset +
351 offset - start, copy);
362 skb_walk_frags(skb, frag_iter) {
365 WARN_ON(start > offset + len);
367 end = start + frag_iter->len;
368 if ((copy = end - offset) > 0) {
371 if (skb_copy_datagram_iovec(frag_iter,
375 if ((len -= copy) == 0)
387 EXPORT_SYMBOL(skb_copy_datagram_iovec);
390 * skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
391 * @skb: buffer to copy
392 * @offset: offset in the buffer to start copying from
393 * @to: io vector to copy to
394 * @to_offset: offset in the io vector to start copying to
395 * @len: amount of data to copy from buffer to iovec
397 * Returns 0 or -EFAULT.
398 * Note: the iovec is not modified during the copy.
400 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
401 const struct iovec *to, int to_offset,
404 int start = skb_headlen(skb);
405 int i, copy = start - offset;
406 struct sk_buff *frag_iter;
412 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
414 if ((len -= copy) == 0)
420 /* Copy paged appendix. Hmm... why does this look so complicated? */
421 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
423 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
425 WARN_ON(start > offset + len);
427 end = start + skb_frag_size(frag);
428 if ((copy = end - offset) > 0) {
431 struct page *page = skb_frag_page(frag);
436 err = memcpy_toiovecend(to, vaddr + frag->page_offset +
437 offset - start, to_offset, copy);
449 skb_walk_frags(skb, frag_iter) {
452 WARN_ON(start > offset + len);
454 end = start + frag_iter->len;
455 if ((copy = end - offset) > 0) {
458 if (skb_copy_datagram_const_iovec(frag_iter,
463 if ((len -= copy) == 0)
476 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
479 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
480 * @skb: buffer to copy
481 * @offset: offset in the buffer to start copying to
482 * @from: io vector to copy to
483 * @from_offset: offset in the io vector to start copying from
484 * @len: amount of data to copy to buffer from iovec
486 * Returns 0 or -EFAULT.
487 * Note: the iovec is not modified during the copy.
489 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
490 const struct iovec *from, int from_offset,
493 int start = skb_headlen(skb);
494 int i, copy = start - offset;
495 struct sk_buff *frag_iter;
501 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
504 if ((len -= copy) == 0)
510 /* Copy paged appendix. Hmm... why does this look so complicated? */
511 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
513 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
515 WARN_ON(start > offset + len);
517 end = start + skb_frag_size(frag);
518 if ((copy = end - offset) > 0) {
521 struct page *page = skb_frag_page(frag);
526 err = memcpy_fromiovecend(vaddr + frag->page_offset +
528 from, from_offset, copy);
541 skb_walk_frags(skb, frag_iter) {
544 WARN_ON(start > offset + len);
546 end = start + frag_iter->len;
547 if ((copy = end - offset) > 0) {
550 if (skb_copy_datagram_from_iovec(frag_iter,
556 if ((len -= copy) == 0)
569 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
571 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
572 u8 __user *to, int len,
575 int start = skb_headlen(skb);
576 int i, copy = start - offset;
577 struct sk_buff *frag_iter;
585 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
589 if ((len -= copy) == 0)
596 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
598 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
600 WARN_ON(start > offset + len);
602 end = start + skb_frag_size(frag);
603 if ((copy = end - offset) > 0) {
607 struct page *page = skb_frag_page(frag);
612 csum2 = csum_and_copy_to_user(vaddr +
619 *csump = csum_block_add(*csump, csum2, pos);
629 skb_walk_frags(skb, frag_iter) {
632 WARN_ON(start > offset + len);
634 end = start + frag_iter->len;
635 if ((copy = end - offset) > 0) {
639 if (skb_copy_and_csum_datagram(frag_iter,
644 *csump = csum_block_add(*csump, csum2, pos);
645 if ((len -= copy) == 0)
660 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
664 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
666 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
667 netdev_rx_csum_fault(skb->dev);
668 skb->ip_summed = CHECKSUM_UNNECESSARY;
672 EXPORT_SYMBOL(__skb_checksum_complete_head);
674 __sum16 __skb_checksum_complete(struct sk_buff *skb)
676 return __skb_checksum_complete_head(skb, skb->len);
678 EXPORT_SYMBOL(__skb_checksum_complete);
681 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
683 * @hlen: hardware length
686 * Caller _must_ check that skb will fit to this iovec.
688 * Returns: 0 - success.
689 * -EINVAL - checksum failure.
690 * -EFAULT - fault during copy. Beware, in this case iovec
693 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
694 int hlen, struct iovec *iov)
697 int chunk = skb->len - hlen;
702 /* Skip filled elements.
703 * Pretty silly, look at memcpy_toiovec, though 8)
705 while (!iov->iov_len)
708 if (iov->iov_len < chunk) {
709 if (__skb_checksum_complete(skb))
711 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
714 csum = csum_partial(skb->data, hlen, skb->csum);
715 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
720 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
721 netdev_rx_csum_fault(skb->dev);
722 iov->iov_len -= chunk;
723 iov->iov_base += chunk;
731 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
734 * datagram_poll - generic datagram poll
739 * Datagram poll: Again totally generic. This also handles
740 * sequenced packet sockets providing the socket receive queue
741 * is only ever holding data ready to receive.
743 * Note: when you _don't_ use this routine for this protocol,
744 * and you use a different write policy from sock_writeable()
745 * then please supply your own write_space callback.
747 unsigned int datagram_poll(struct file *file, struct socket *sock,
750 struct sock *sk = sock->sk;
753 sock_poll_wait(file, sk_sleep(sk), wait);
756 /* exceptional events? */
757 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
759 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
761 if (sk->sk_shutdown & RCV_SHUTDOWN)
762 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
763 if (sk->sk_shutdown == SHUTDOWN_MASK)
767 if (!skb_queue_empty(&sk->sk_receive_queue))
768 mask |= POLLIN | POLLRDNORM;
770 /* Connection-based need to check for termination and startup */
771 if (connection_based(sk)) {
772 if (sk->sk_state == TCP_CLOSE)
774 /* connection hasn't started yet? */
775 if (sk->sk_state == TCP_SYN_SENT)
780 if (sock_writeable(sk))
781 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
783 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
787 EXPORT_SYMBOL(datagram_poll);