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
);
263 skb_free_datagram(sk
, skb
);
267 EXPORT_SYMBOL(skb_kill_datagram
);
270 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
271 * @skb: buffer to copy
272 * @offset: offset in the buffer to start copying from
273 * @to: io vector to copy to
274 * @len: amount of data to copy from buffer to iovec
276 * Note: the iovec is modified during the copy.
278 int skb_copy_datagram_iovec(const struct sk_buff
*skb
, int offset
,
279 struct iovec
*to
, int len
)
281 int start
= skb_headlen(skb
);
282 int i
, copy
= start
- offset
;
288 if (memcpy_toiovec(to
, skb
->data
+ offset
, copy
))
290 if ((len
-= copy
) == 0)
295 /* Copy paged appendix. Hmm... why does this look so complicated? */
296 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
299 WARN_ON(start
> offset
+ len
);
301 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
302 if ((copy
= end
- offset
) > 0) {
305 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
306 struct page
*page
= frag
->page
;
311 err
= memcpy_toiovec(to
, vaddr
+ frag
->page_offset
+
312 offset
- start
, copy
);
323 if (skb_shinfo(skb
)->frag_list
) {
324 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
326 for (; list
; list
= list
->next
) {
329 WARN_ON(start
> offset
+ len
);
331 end
= start
+ list
->len
;
332 if ((copy
= end
- offset
) > 0) {
335 if (skb_copy_datagram_iovec(list
,
339 if ((len
-= copy
) == 0)
354 * skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
355 * @skb: buffer to copy
356 * @offset: offset in the buffer to start copying to
357 * @from: io vector to copy to
358 * @len: amount of data to copy to buffer from iovec
360 * Returns 0 or -EFAULT.
361 * Note: the iovec is modified during the copy.
363 int skb_copy_datagram_from_iovec(struct sk_buff
*skb
, int offset
,
364 struct iovec
*from
, int len
)
366 int start
= skb_headlen(skb
);
367 int i
, copy
= start
- offset
;
373 if (memcpy_fromiovec(skb
->data
+ offset
, from
, copy
))
375 if ((len
-= copy
) == 0)
380 /* Copy paged appendix. Hmm... why does this look so complicated? */
381 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
384 WARN_ON(start
> offset
+ len
);
386 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
387 if ((copy
= end
- offset
) > 0) {
390 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
391 struct page
*page
= frag
->page
;
396 err
= memcpy_fromiovec(vaddr
+ frag
->page_offset
+
397 offset
- start
, from
, copy
);
409 if (skb_shinfo(skb
)->frag_list
) {
410 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
412 for (; list
; list
= list
->next
) {
415 WARN_ON(start
> offset
+ len
);
417 end
= start
+ list
->len
;
418 if ((copy
= end
- offset
) > 0) {
421 if (skb_copy_datagram_from_iovec(list
,
425 if ((len
-= copy
) == 0)
438 EXPORT_SYMBOL(skb_copy_datagram_from_iovec
);
440 static int skb_copy_and_csum_datagram(const struct sk_buff
*skb
, int offset
,
441 u8 __user
*to
, int len
,
444 int start
= skb_headlen(skb
);
446 int i
, copy
= start
- offset
;
453 *csump
= csum_and_copy_to_user(skb
->data
+ offset
, to
, copy
,
457 if ((len
-= copy
) == 0)
464 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
467 WARN_ON(start
> offset
+ len
);
469 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
470 if ((copy
= end
- offset
) > 0) {
474 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
475 struct page
*page
= frag
->page
;
480 csum2
= csum_and_copy_to_user(vaddr
+
487 *csump
= csum_block_add(*csump
, csum2
, pos
);
497 if (skb_shinfo(skb
)->frag_list
) {
498 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
500 for (; list
; list
=list
->next
) {
503 WARN_ON(start
> offset
+ len
);
505 end
= start
+ list
->len
;
506 if ((copy
= end
- offset
) > 0) {
510 if (skb_copy_and_csum_datagram(list
,
515 *csump
= csum_block_add(*csump
, csum2
, pos
);
516 if ((len
-= copy
) == 0)
532 __sum16
__skb_checksum_complete_head(struct sk_buff
*skb
, int len
)
536 sum
= csum_fold(skb_checksum(skb
, 0, len
, skb
->csum
));
538 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
))
539 netdev_rx_csum_fault(skb
->dev
);
540 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
544 EXPORT_SYMBOL(__skb_checksum_complete_head
);
546 __sum16
__skb_checksum_complete(struct sk_buff
*skb
)
548 return __skb_checksum_complete_head(skb
, skb
->len
);
550 EXPORT_SYMBOL(__skb_checksum_complete
);
553 * skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
555 * @hlen: hardware length
558 * Caller _must_ check that skb will fit to this iovec.
560 * Returns: 0 - success.
561 * -EINVAL - checksum failure.
562 * -EFAULT - fault during copy. Beware, in this case iovec
565 int skb_copy_and_csum_datagram_iovec(struct sk_buff
*skb
,
566 int hlen
, struct iovec
*iov
)
569 int chunk
= skb
->len
- hlen
;
574 /* Skip filled elements.
575 * Pretty silly, look at memcpy_toiovec, though 8)
577 while (!iov
->iov_len
)
580 if (iov
->iov_len
< chunk
) {
581 if (__skb_checksum_complete(skb
))
583 if (skb_copy_datagram_iovec(skb
, hlen
, iov
, chunk
))
586 csum
= csum_partial(skb
->data
, hlen
, skb
->csum
);
587 if (skb_copy_and_csum_datagram(skb
, hlen
, iov
->iov_base
,
592 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
))
593 netdev_rx_csum_fault(skb
->dev
);
594 iov
->iov_len
-= chunk
;
595 iov
->iov_base
+= chunk
;
605 * datagram_poll - generic datagram poll
610 * Datagram poll: Again totally generic. This also handles
611 * sequenced packet sockets providing the socket receive queue
612 * is only ever holding data ready to receive.
614 * Note: when you _don't_ use this routine for this protocol,
615 * and you use a different write policy from sock_writeable()
616 * then please supply your own write_space callback.
618 unsigned int datagram_poll(struct file
*file
, struct socket
*sock
,
621 struct sock
*sk
= sock
->sk
;
624 poll_wait(file
, sk
->sk_sleep
, wait
);
627 /* exceptional events? */
628 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
630 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
632 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
636 if (!skb_queue_empty(&sk
->sk_receive_queue
) ||
637 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
638 mask
|= POLLIN
| POLLRDNORM
;
640 /* Connection-based need to check for termination and startup */
641 if (connection_based(sk
)) {
642 if (sk
->sk_state
== TCP_CLOSE
)
644 /* connection hasn't started yet? */
645 if (sk
->sk_state
== TCP_SYN_SENT
)
650 if (sock_writeable(sk
))
651 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
653 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
658 EXPORT_SYMBOL(datagram_poll
);
659 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec
);
660 EXPORT_SYMBOL(skb_copy_datagram_iovec
);
661 EXPORT_SYMBOL(skb_free_datagram
);
662 EXPORT_SYMBOL(skb_recv_datagram
);