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@redhat.com>. (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;
70 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
75 prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
78 error = sock_error(sk);
82 if (!skb_queue_empty(&sk->sk_receive_queue))
85 /* Socket shut down? */
86 if (sk->sk_shutdown & RCV_SHUTDOWN)
89 /* Sequenced packets can come disconnected.
90 * If so we report the problem
93 if (connection_based(sk) &&
94 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
98 if (signal_pending(current))
102 *timeo_p = schedule_timeout(*timeo_p);
104 finish_wait(sk->sk_sleep, &wait);
107 error = sock_intr_errno(*timeo_p);
118 * __skb_recv_datagram - Receive a datagram skbuff
121 * @peeked: returns non-zero if this packet has been seen before
122 * @err: error code returned
124 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
125 * and possible races. This replaces identical code in packet, raw and
126 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
127 * the long standing peek and read race for datagram sockets. If you
128 * alter this routine remember it must be re-entrant.
130 * This function will lock the socket if a skb is returned, so the caller
131 * needs to unlock the socket in that case (usually by calling
134 * * It does not lock socket since today. This function is
135 * * free of race conditions. This measure should/can improve
136 * * significantly datagram socket latencies at high loads,
137 * * when data copying to user space takes lots of time.
138 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
142 * The order of the tests when we find no data waiting are specified
143 * quite explicitly by POSIX 1003.1g, don't change them without having
144 * the standard around please.
146 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
147 int *peeked, int *err)
152 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
154 int error = sock_error(sk);
159 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
162 /* Again only user level code calls this function, so nothing
163 * interrupt level will suddenly eat the receive_queue.
165 * Look at current nfs client by the way...
166 * However, this function was corrent in any case. 8)
168 unsigned long cpu_flags;
170 spin_lock_irqsave(&sk->sk_receive_queue.lock, cpu_flags);
171 skb = skb_peek(&sk->sk_receive_queue);
173 *peeked = skb->peeked;
174 if (flags & MSG_PEEK) {
176 atomic_inc(&skb->users);
178 __skb_unlink(skb, &sk->sk_receive_queue);
180 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, cpu_flags);
185 /* User doesn't want to wait */
190 } while (!wait_for_packet(sk, err, &timeo));
198 EXPORT_SYMBOL(__skb_recv_datagram);
200 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
201 int noblock, int *err)
205 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
209 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
215 * skb_kill_datagram - Free a datagram skbuff forcibly
217 * @skb: datagram skbuff
220 * This function frees a datagram skbuff that was received by
221 * skb_recv_datagram. The flags argument must match the one
222 * used for skb_recv_datagram.
224 * If the MSG_PEEK flag is set, and the packet is still on the
225 * receive queue of the socket, it will be taken off the queue
226 * before it is freed.
228 * This function currently only disables BH when acquiring the
229 * sk_receive_queue lock. Therefore it must not be used in a
230 * context where that lock is acquired in an IRQ context.
232 * It returns 0 if the packet was removed by us.
235 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
239 if (flags & MSG_PEEK) {
241 spin_lock_bh(&sk->sk_receive_queue.lock);
242 if (skb == skb_peek(&sk->sk_receive_queue)) {
243 __skb_unlink(skb, &sk->sk_receive_queue);
244 atomic_dec(&skb->users);
247 spin_unlock_bh(&sk->sk_receive_queue.lock);
254 EXPORT_SYMBOL(skb_kill_datagram);
257 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
258 * @skb: buffer to copy
259 * @offset: offset in the buffer to start copying from
260 * @to: io vector to copy to
261 * @len: amount of data to copy from buffer to iovec
263 * Note: the iovec is modified during the copy.
265 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
266 struct iovec *to, int len)
268 int start = skb_headlen(skb);
269 int i, copy = start - offset;
275 if (memcpy_toiovec(to, skb->data + offset, copy))
277 if ((len -= copy) == 0)
282 /* Copy paged appendix. Hmm... why does this look so complicated? */
283 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
286 BUG_TRAP(start <= offset + len);
288 end = start + skb_shinfo(skb)->frags[i].size;
289 if ((copy = end - offset) > 0) {
292 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
293 struct page *page = frag->page;
298 err = memcpy_toiovec(to, vaddr + frag->page_offset +
299 offset - start, copy);
310 if (skb_shinfo(skb)->frag_list) {
311 struct sk_buff *list = skb_shinfo(skb)->frag_list;
313 for (; list; list = list->next) {
316 BUG_TRAP(start <= offset + len);
318 end = start + list->len;
319 if ((copy = end - offset) > 0) {
322 if (skb_copy_datagram_iovec(list,
326 if ((len -= copy) == 0)
340 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
341 u8 __user *to, int len,
344 int start = skb_headlen(skb);
346 int i, copy = start - offset;
353 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
357 if ((len -= copy) == 0)
364 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
367 BUG_TRAP(start <= offset + len);
369 end = start + skb_shinfo(skb)->frags[i].size;
370 if ((copy = end - offset) > 0) {
374 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
375 struct page *page = frag->page;
380 csum2 = csum_and_copy_to_user(vaddr +
387 *csump = csum_block_add(*csump, csum2, pos);
397 if (skb_shinfo(skb)->frag_list) {
398 struct sk_buff *list = skb_shinfo(skb)->frag_list;
400 for (; list; list=list->next) {
403 BUG_TRAP(start <= offset + len);
405 end = start + list->len;
406 if ((copy = end - offset) > 0) {
410 if (skb_copy_and_csum_datagram(list,
415 *csump = csum_block_add(*csump, csum2, pos);
416 if ((len -= copy) == 0)
432 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
436 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
438 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
439 netdev_rx_csum_fault(skb->dev);
440 skb->ip_summed = CHECKSUM_UNNECESSARY;
444 EXPORT_SYMBOL(__skb_checksum_complete_head);
446 __sum16 __skb_checksum_complete(struct sk_buff *skb)
448 return __skb_checksum_complete_head(skb, skb->len);
450 EXPORT_SYMBOL(__skb_checksum_complete);
453 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
455 * @hlen: hardware length
458 * Caller _must_ check that skb will fit to this iovec.
460 * Returns: 0 - success.
461 * -EINVAL - checksum failure.
462 * -EFAULT - fault during copy. Beware, in this case iovec
465 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
466 int hlen, struct iovec *iov)
469 int chunk = skb->len - hlen;
474 /* Skip filled elements.
475 * Pretty silly, look at memcpy_toiovec, though 8)
477 while (!iov->iov_len)
480 if (iov->iov_len < chunk) {
481 if (__skb_checksum_complete(skb))
483 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
486 csum = csum_partial(skb->data, hlen, skb->csum);
487 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
492 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
493 netdev_rx_csum_fault(skb->dev);
494 iov->iov_len -= chunk;
495 iov->iov_base += chunk;
505 * datagram_poll - generic datagram poll
510 * Datagram poll: Again totally generic. This also handles
511 * sequenced packet sockets providing the socket receive queue
512 * is only ever holding data ready to receive.
514 * Note: when you _don't_ use this routine for this protocol,
515 * and you use a different write policy from sock_writeable()
516 * then please supply your own write_space callback.
518 unsigned int datagram_poll(struct file *file, struct socket *sock,
521 struct sock *sk = sock->sk;
524 poll_wait(file, sk->sk_sleep, wait);
527 /* exceptional events? */
528 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
530 if (sk->sk_shutdown & RCV_SHUTDOWN)
532 if (sk->sk_shutdown == SHUTDOWN_MASK)
536 if (!skb_queue_empty(&sk->sk_receive_queue) ||
537 (sk->sk_shutdown & RCV_SHUTDOWN))
538 mask |= POLLIN | POLLRDNORM;
540 /* Connection-based need to check for termination and startup */
541 if (connection_based(sk)) {
542 if (sk->sk_state == TCP_CLOSE)
544 /* connection hasn't started yet? */
545 if (sk->sk_state == TCP_SYN_SENT)
550 if (sock_writeable(sk))
551 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
553 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
558 EXPORT_SYMBOL(datagram_poll);
559 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
560 EXPORT_SYMBOL(skb_copy_datagram_iovec);
561 EXPORT_SYMBOL(skb_free_datagram);
562 EXPORT_SYMBOL(skb_recv_datagram);