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>
59 #include <net/ll_poll.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 int 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);
82 * Wait for the last received packet to be different from skb
84 static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
85 const struct sk_buff *skb)
88 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
90 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
93 error = sock_error(sk);
97 if (sk->sk_receive_queue.prev != skb)
100 /* Socket shut down? */
101 if (sk->sk_shutdown & RCV_SHUTDOWN)
104 /* Sequenced packets can come disconnected.
105 * If so we report the problem
108 if (connection_based(sk) &&
109 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
113 if (signal_pending(current))
117 *timeo_p = schedule_timeout(*timeo_p);
119 finish_wait(sk_sleep(sk), &wait);
122 error = sock_intr_errno(*timeo_p);
133 * __skb_recv_datagram - Receive a datagram skbuff
136 * @peeked: returns non-zero if this packet has been seen before
137 * @off: an offset in bytes to peek skb from. Returns an offset
138 * within an skb where data actually starts
139 * @err: error code returned
141 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
142 * and possible races. This replaces identical code in packet, raw and
143 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
144 * the long standing peek and read race for datagram sockets. If you
145 * alter this routine remember it must be re-entrant.
147 * This function will lock the socket if a skb is returned, so the caller
148 * needs to unlock the socket in that case (usually by calling
151 * * It does not lock socket since today. This function is
152 * * free of race conditions. This measure should/can improve
153 * * significantly datagram socket latencies at high loads,
154 * * when data copying to user space takes lots of time.
155 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
159 * The order of the tests when we find no data waiting are specified
160 * quite explicitly by POSIX 1003.1g, don't change them without having
161 * the standard around please.
163 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
164 int *peeked, int *off, int *err)
166 struct sk_buff *skb, *last;
169 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
171 int error = sock_error(sk);
176 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
179 /* Again only user level code calls this function, so nothing
180 * interrupt level will suddenly eat the receive_queue.
182 * Look at current nfs client by the way...
183 * However, this function was correct in any case. 8)
185 unsigned long cpu_flags;
186 struct sk_buff_head *queue = &sk->sk_receive_queue;
189 last = (struct sk_buff *)queue;
190 spin_lock_irqsave(&queue->lock, cpu_flags);
191 skb_queue_walk(queue, skb) {
193 *peeked = skb->peeked;
194 if (flags & MSG_PEEK) {
195 if (_off >= skb->len && (skb->len || _off ||
201 atomic_inc(&skb->users);
203 __skb_unlink(skb, queue);
205 spin_unlock_irqrestore(&queue->lock, cpu_flags);
209 spin_unlock_irqrestore(&queue->lock, cpu_flags);
211 if (sk_valid_ll(sk) && sk_poll_ll(sk, flags & MSG_DONTWAIT))
214 /* User doesn't want to wait */
219 } while (!wait_for_more_packets(sk, err, &timeo, last));
227 EXPORT_SYMBOL(__skb_recv_datagram);
229 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
230 int noblock, int *err)
234 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
237 EXPORT_SYMBOL(skb_recv_datagram);
239 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
242 sk_mem_reclaim_partial(sk);
244 EXPORT_SYMBOL(skb_free_datagram);
246 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
250 if (likely(atomic_read(&skb->users) == 1))
252 else if (likely(!atomic_dec_and_test(&skb->users)))
255 slow = lock_sock_fast(sk);
257 sk_mem_reclaim_partial(sk);
258 unlock_sock_fast(sk, slow);
260 /* skb is now orphaned, can be freed outside of locked section */
263 EXPORT_SYMBOL(skb_free_datagram_locked);
266 * skb_kill_datagram - Free a datagram skbuff forcibly
268 * @skb: datagram skbuff
271 * This function frees a datagram skbuff that was received by
272 * skb_recv_datagram. The flags argument must match the one
273 * used for skb_recv_datagram.
275 * If the MSG_PEEK flag is set, and the packet is still on the
276 * receive queue of the socket, it will be taken off the queue
277 * before it is freed.
279 * This function currently only disables BH when acquiring the
280 * sk_receive_queue lock. Therefore it must not be used in a
281 * context where that lock is acquired in an IRQ context.
283 * It returns 0 if the packet was removed by us.
286 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
290 if (flags & MSG_PEEK) {
292 spin_lock_bh(&sk->sk_receive_queue.lock);
293 if (skb == skb_peek(&sk->sk_receive_queue)) {
294 __skb_unlink(skb, &sk->sk_receive_queue);
295 atomic_dec(&skb->users);
298 spin_unlock_bh(&sk->sk_receive_queue.lock);
302 atomic_inc(&sk->sk_drops);
303 sk_mem_reclaim_partial(sk);
307 EXPORT_SYMBOL(skb_kill_datagram);
310 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
311 * @skb: buffer to copy
312 * @offset: offset in the buffer to start copying from
313 * @to: io vector to copy to
314 * @len: amount of data to copy from buffer to iovec
316 * Note: the iovec is modified during the copy.
318 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
319 struct iovec *to, int len)
321 int start = skb_headlen(skb);
322 int i, copy = start - offset;
323 struct sk_buff *frag_iter;
325 trace_skb_copy_datagram_iovec(skb, len);
331 if (memcpy_toiovec(to, skb->data + offset, copy))
333 if ((len -= copy) == 0)
338 /* Copy paged appendix. Hmm... why does this look so complicated? */
339 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
341 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
343 WARN_ON(start > offset + len);
345 end = start + skb_frag_size(frag);
346 if ((copy = end - offset) > 0) {
349 struct page *page = skb_frag_page(frag);
354 err = memcpy_toiovec(to, vaddr + frag->page_offset +
355 offset - start, copy);
366 skb_walk_frags(skb, frag_iter) {
369 WARN_ON(start > offset + len);
371 end = start + frag_iter->len;
372 if ((copy = end - offset) > 0) {
375 if (skb_copy_datagram_iovec(frag_iter,
379 if ((len -= copy) == 0)
391 EXPORT_SYMBOL(skb_copy_datagram_iovec);
394 * skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
395 * @skb: buffer to copy
396 * @offset: offset in the buffer to start copying from
397 * @to: io vector to copy to
398 * @to_offset: offset in the io vector to start copying to
399 * @len: amount of data to copy from buffer to iovec
401 * Returns 0 or -EFAULT.
402 * Note: the iovec is not modified during the copy.
404 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
405 const struct iovec *to, int to_offset,
408 int start = skb_headlen(skb);
409 int i, copy = start - offset;
410 struct sk_buff *frag_iter;
416 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
418 if ((len -= copy) == 0)
424 /* Copy paged appendix. Hmm... why does this look so complicated? */
425 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
427 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
429 WARN_ON(start > offset + len);
431 end = start + skb_frag_size(frag);
432 if ((copy = end - offset) > 0) {
435 struct page *page = skb_frag_page(frag);
440 err = memcpy_toiovecend(to, vaddr + frag->page_offset +
441 offset - start, to_offset, copy);
453 skb_walk_frags(skb, frag_iter) {
456 WARN_ON(start > offset + len);
458 end = start + frag_iter->len;
459 if ((copy = end - offset) > 0) {
462 if (skb_copy_datagram_const_iovec(frag_iter,
467 if ((len -= copy) == 0)
480 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
483 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
484 * @skb: buffer to copy
485 * @offset: offset in the buffer to start copying to
486 * @from: io vector to copy to
487 * @from_offset: offset in the io vector to start copying from
488 * @len: amount of data to copy to buffer from iovec
490 * Returns 0 or -EFAULT.
491 * Note: the iovec is not modified during the copy.
493 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
494 const struct iovec *from, int from_offset,
497 int start = skb_headlen(skb);
498 int i, copy = start - offset;
499 struct sk_buff *frag_iter;
505 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
508 if ((len -= copy) == 0)
514 /* Copy paged appendix. Hmm... why does this look so complicated? */
515 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
517 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
519 WARN_ON(start > offset + len);
521 end = start + skb_frag_size(frag);
522 if ((copy = end - offset) > 0) {
525 struct page *page = skb_frag_page(frag);
530 err = memcpy_fromiovecend(vaddr + frag->page_offset +
532 from, from_offset, copy);
545 skb_walk_frags(skb, frag_iter) {
548 WARN_ON(start > offset + len);
550 end = start + frag_iter->len;
551 if ((copy = end - offset) > 0) {
554 if (skb_copy_datagram_from_iovec(frag_iter,
560 if ((len -= copy) == 0)
573 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
575 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
576 u8 __user *to, int len,
579 int start = skb_headlen(skb);
580 int i, copy = start - offset;
581 struct sk_buff *frag_iter;
589 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
593 if ((len -= copy) == 0)
600 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
602 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
604 WARN_ON(start > offset + len);
606 end = start + skb_frag_size(frag);
607 if ((copy = end - offset) > 0) {
611 struct page *page = skb_frag_page(frag);
616 csum2 = csum_and_copy_to_user(vaddr +
623 *csump = csum_block_add(*csump, csum2, pos);
633 skb_walk_frags(skb, frag_iter) {
636 WARN_ON(start > offset + len);
638 end = start + frag_iter->len;
639 if ((copy = end - offset) > 0) {
643 if (skb_copy_and_csum_datagram(frag_iter,
648 *csump = csum_block_add(*csump, csum2, pos);
649 if ((len -= copy) == 0)
664 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
668 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
670 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
671 netdev_rx_csum_fault(skb->dev);
672 skb->ip_summed = CHECKSUM_UNNECESSARY;
676 EXPORT_SYMBOL(__skb_checksum_complete_head);
678 __sum16 __skb_checksum_complete(struct sk_buff *skb)
680 return __skb_checksum_complete_head(skb, skb->len);
682 EXPORT_SYMBOL(__skb_checksum_complete);
685 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
687 * @hlen: hardware length
690 * Caller _must_ check that skb will fit to this iovec.
692 * Returns: 0 - success.
693 * -EINVAL - checksum failure.
694 * -EFAULT - fault during copy. Beware, in this case iovec
697 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
698 int hlen, struct iovec *iov)
701 int chunk = skb->len - hlen;
706 /* Skip filled elements.
707 * Pretty silly, look at memcpy_toiovec, though 8)
709 while (!iov->iov_len)
712 if (iov->iov_len < chunk) {
713 if (__skb_checksum_complete(skb))
715 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
718 csum = csum_partial(skb->data, hlen, skb->csum);
719 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
724 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
725 netdev_rx_csum_fault(skb->dev);
726 iov->iov_len -= chunk;
727 iov->iov_base += chunk;
735 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
738 * datagram_poll - generic datagram poll
743 * Datagram poll: Again totally generic. This also handles
744 * sequenced packet sockets providing the socket receive queue
745 * is only ever holding data ready to receive.
747 * Note: when you _don't_ use this routine for this protocol,
748 * and you use a different write policy from sock_writeable()
749 * then please supply your own write_space callback.
751 unsigned int datagram_poll(struct file *file, struct socket *sock,
754 struct sock *sk = sock->sk;
757 sock_poll_wait(file, sk_sleep(sk), wait);
760 /* exceptional events? */
761 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
763 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
765 if (sk->sk_shutdown & RCV_SHUTDOWN)
766 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
767 if (sk->sk_shutdown == SHUTDOWN_MASK)
771 if (!skb_queue_empty(&sk->sk_receive_queue))
772 mask |= POLLIN | POLLRDNORM;
774 /* Connection-based need to check for termination and startup */
775 if (connection_based(sk)) {
776 if (sk->sk_state == TCP_CLOSE)
778 /* connection hasn't started yet? */
779 if (sk->sk_state == TCP_SYN_SENT)
784 if (sock_writeable(sk))
785 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
787 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
791 EXPORT_SYMBOL(datagram_poll);