datagram.c revision 3305b80c214c642b89cd5c21af83bc91ec13f8bd 
1/*
2 *	SUCS NET3:
3 *
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.
11 *
12 *	Authors:	Alan Cox <alan@redhat.com>. (datagram_poll() from old
13 *						     udp.c code)
14 *
15 *	Fixes:
16 *		Alan Cox	:	NULL return from skb_peek_copy()
17 *					understood
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
23 *					feasible.
24 *		Alan Cox	:	Fixed write poll of non IP protocol
25 *					crash.
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.
33 *
34 */
35
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>
41#include <linux/mm.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>
51
52#include <net/protocol.h>
53#include <linux/skbuff.h>
54
55#include <net/checksum.h>
56#include <net/sock.h>
57#include <net/tcp_states.h>
58
59/*
60 *	Is a socket 'connection oriented' ?
61 */
62static inline int connection_based(struct sock *sk)
63{
64	return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
65}
66
67/*
68 * Wait for a packet..
69 */
70static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
71{
72	int error;
73	DEFINE_WAIT(wait);
74
75	prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
76
77	/* Socket errors? */
78	error = sock_error(sk);
79	if (error)
80		goto out_err;
81
82	if (!skb_queue_empty(&sk->sk_receive_queue))
83		goto out;
84
85	/* Socket shut down? */
86	if (sk->sk_shutdown & RCV_SHUTDOWN)
87		goto out_noerr;
88
89	/* Sequenced packets can come disconnected.
90	 * If so we report the problem
91	 */
92	error = -ENOTCONN;
93	if (connection_based(sk) &&
94	    !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
95		goto out_err;
96
97	/* handle signals */
98	if (signal_pending(current))
99		goto interrupted;
100
101	error = 0;
102	*timeo_p = schedule_timeout(*timeo_p);
103out:
104	finish_wait(sk->sk_sleep, &wait);
105	return error;
106interrupted:
107	error = sock_intr_errno(*timeo_p);
108out_err:
109	*err = error;
110	goto out;
111out_noerr:
112	*err = 0;
113	error = 1;
114	goto out;
115}
116
117/**
118 *	skb_recv_datagram - Receive a datagram skbuff
119 *	@sk: socket
120 *	@flags: MSG_ flags
121 *	@noblock: blocking operation?
122 *	@err: error code returned
123 *
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.
129 *
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
132 *	skb_free_datagram)
133 *
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
139 *	*  8) Great win.)
140 *	*			                    --ANK (980729)
141 *
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.
145 */
146struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
147				  int noblock, int *err)
148{
149	struct sk_buff *skb;
150	long timeo;
151	/*
152	 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
153	 */
154	int error = sock_error(sk);
155
156	if (error)
157		goto no_packet;
158
159	timeo = sock_rcvtimeo(sk, noblock);
160
161	do {
162		/* Again only user level code calls this function, so nothing
163		 * interrupt level will suddenly eat the receive_queue.
164		 *
165		 * Look at current nfs client by the way...
166		 * However, this function was corrent in any case. 8)
167		 */
168		if (flags & MSG_PEEK) {
169			unsigned long cpu_flags;
170
171			spin_lock_irqsave(&sk->sk_receive_queue.lock,
172					  cpu_flags);
173			skb = skb_peek(&sk->sk_receive_queue);
174			if (skb)
175				atomic_inc(&skb->users);
176			spin_unlock_irqrestore(&sk->sk_receive_queue.lock,
177					       cpu_flags);
178		} else
179			skb = skb_dequeue(&sk->sk_receive_queue);
180
181		if (skb)
182			return skb;
183
184		/* User doesn't want to wait */
185		error = -EAGAIN;
186		if (!timeo)
187			goto no_packet;
188
189	} while (!wait_for_packet(sk, err, &timeo));
190
191	return NULL;
192
193no_packet:
194	*err = error;
195	return NULL;
196}
197
198void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
199{
200	kfree_skb(skb);
201}
202
203/**
204 *	skb_kill_datagram - Free a datagram skbuff forcibly
205 *	@sk: socket
206 *	@skb: datagram skbuff
207 *	@flags: MSG_ flags
208 *
209 *	This function frees a datagram skbuff that was received by
210 *	skb_recv_datagram.  The flags argument must match the one
211 *	used for skb_recv_datagram.
212 *
213 *	If the MSG_PEEK flag is set, and the packet is still on the
214 *	receive queue of the socket, it will be taken off the queue
215 *	before it is freed.
216 *
217 *	This function currently only disables BH when acquiring the
218 *	sk_receive_queue lock.  Therefore it must not be used in a
219 *	context where that lock is acquired in an IRQ context.
220 */
221
222void skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
223{
224	if (flags & MSG_PEEK) {
225		spin_lock_bh(&sk->sk_receive_queue.lock);
226		if (skb == skb_peek(&sk->sk_receive_queue)) {
227			__skb_unlink(skb, &sk->sk_receive_queue);
228			atomic_dec(&skb->users);
229		}
230		spin_unlock_bh(&sk->sk_receive_queue.lock);
231	}
232
233	kfree_skb(skb);
234}
235
236EXPORT_SYMBOL(skb_kill_datagram);
237
238/**
239 *	skb_copy_datagram_iovec - Copy a datagram to an iovec.
240 *	@skb: buffer to copy
241 *	@offset: offset in the buffer to start copying from
242 *	@to: io vector to copy to
243 *	@len: amount of data to copy from buffer to iovec
244 *
245 *	Note: the iovec is modified during the copy.
246 */
247int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
248			    struct iovec *to, int len)
249{
250	int i, err, fraglen, end = 0;
251	struct sk_buff *next = skb_shinfo(skb)->frag_list;
252
253	if (!len)
254		return 0;
255
256next_skb:
257	fraglen = skb_headlen(skb);
258	i = -1;
259
260	while (1) {
261		int start = end;
262
263		if ((end += fraglen) > offset) {
264			int copy = end - offset, o = offset - start;
265
266			if (copy > len)
267				copy = len;
268			if (i == -1)
269				err = memcpy_toiovec(to, skb->data + o, copy);
270			else {
271				skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
272				struct page *page = frag->page;
273				void *p = kmap(page) + frag->page_offset + o;
274				err = memcpy_toiovec(to, p, copy);
275				kunmap(page);
276			}
277			if (err)
278				goto fault;
279			if (!(len -= copy))
280				return 0;
281			offset += copy;
282		}
283		if (++i >= skb_shinfo(skb)->nr_frags)
284			break;
285		fraglen = skb_shinfo(skb)->frags[i].size;
286	}
287	if (next) {
288		skb = next;
289		BUG_ON(skb_shinfo(skb)->frag_list);
290		next = skb->next;
291		goto next_skb;
292	}
293fault:
294	return -EFAULT;
295}
296
297static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
298				      u8 __user *to, int len,
299				      unsigned int *csump)
300{
301	int start = skb_headlen(skb);
302	int pos = 0;
303	int i, copy = start - offset;
304
305	/* Copy header. */
306	if (copy > 0) {
307		int err = 0;
308		if (copy > len)
309			copy = len;
310		*csump = csum_and_copy_to_user(skb->data + offset, to, copy,
311					       *csump, &err);
312		if (err)
313			goto fault;
314		if ((len -= copy) == 0)
315			return 0;
316		offset += copy;
317		to += copy;
318		pos = copy;
319	}
320
321	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
322		int end;
323
324		BUG_TRAP(start <= offset + len);
325
326		end = start + skb_shinfo(skb)->frags[i].size;
327		if ((copy = end - offset) > 0) {
328			unsigned int csum2;
329			int err = 0;
330			u8  *vaddr;
331			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
332			struct page *page = frag->page;
333
334			if (copy > len)
335				copy = len;
336			vaddr = kmap(page);
337			csum2 = csum_and_copy_to_user(vaddr +
338							frag->page_offset +
339							offset - start,
340						      to, copy, 0, &err);
341			kunmap(page);
342			if (err)
343				goto fault;
344			*csump = csum_block_add(*csump, csum2, pos);
345			if (!(len -= copy))
346				return 0;
347			offset += copy;
348			to += copy;
349			pos += copy;
350		}
351		start = end;
352	}
353
354	if (skb_shinfo(skb)->frag_list) {
355		struct sk_buff *list = skb_shinfo(skb)->frag_list;
356
357		for (; list; list=list->next) {
358			int end;
359
360			BUG_TRAP(start <= offset + len);
361
362			end = start + list->len;
363			if ((copy = end - offset) > 0) {
364				unsigned int csum2 = 0;
365				if (copy > len)
366					copy = len;
367				if (skb_copy_and_csum_datagram(list,
368							       offset - start,
369							       to, copy,
370							       &csum2))
371					goto fault;
372				*csump = csum_block_add(*csump, csum2, pos);
373				if ((len -= copy) == 0)
374					return 0;
375				offset += copy;
376				to += copy;
377				pos += copy;
378			}
379			start = end;
380		}
381	}
382	if (!len)
383		return 0;
384
385fault:
386	return -EFAULT;
387}
388
389unsigned int __skb_checksum_complete(struct sk_buff *skb)
390{
391	unsigned int sum;
392
393	sum = (u16)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
394	if (likely(!sum)) {
395		if (unlikely(skb->ip_summed == CHECKSUM_HW))
396			netdev_rx_csum_fault(skb->dev);
397		skb->ip_summed = CHECKSUM_UNNECESSARY;
398	}
399	return sum;
400}
401EXPORT_SYMBOL(__skb_checksum_complete);
402
403/**
404 *	skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
405 *	@skb: skbuff
406 *	@hlen: hardware length
407 *	@iov: io vector
408 *
409 *	Caller _must_ check that skb will fit to this iovec.
410 *
411 *	Returns: 0       - success.
412 *		 -EINVAL - checksum failure.
413 *		 -EFAULT - fault during copy. Beware, in this case iovec
414 *			   can be modified!
415 */
416int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
417				     int hlen, struct iovec *iov)
418{
419	unsigned int csum;
420	int chunk = skb->len - hlen;
421
422	/* Skip filled elements.
423	 * Pretty silly, look at memcpy_toiovec, though 8)
424	 */
425	while (!iov->iov_len)
426		iov++;
427
428	if (iov->iov_len < chunk) {
429		if (__skb_checksum_complete(skb))
430			goto csum_error;
431		if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
432			goto fault;
433	} else {
434		csum = csum_partial(skb->data, hlen, skb->csum);
435		if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
436					       chunk, &csum))
437			goto fault;
438		if ((unsigned short)csum_fold(csum))
439			goto csum_error;
440		if (unlikely(skb->ip_summed == CHECKSUM_HW))
441			netdev_rx_csum_fault(skb->dev);
442		iov->iov_len -= chunk;
443		iov->iov_base += chunk;
444	}
445	return 0;
446csum_error:
447	return -EINVAL;
448fault:
449	return -EFAULT;
450}
451
452/**
453 * 	datagram_poll - generic datagram poll
454 *	@file: file struct
455 *	@sock: socket
456 *	@wait: poll table
457 *
458 *	Datagram poll: Again totally generic. This also handles
459 *	sequenced packet sockets providing the socket receive queue
460 *	is only ever holding data ready to receive.
461 *
462 *	Note: when you _don't_ use this routine for this protocol,
463 *	and you use a different write policy from sock_writeable()
464 *	then please supply your own write_space callback.
465 */
466unsigned int datagram_poll(struct file *file, struct socket *sock,
467			   poll_table *wait)
468{
469	struct sock *sk = sock->sk;
470	unsigned int mask;
471
472	poll_wait(file, sk->sk_sleep, wait);
473	mask = 0;
474
475	/* exceptional events? */
476	if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
477		mask |= POLLERR;
478	if (sk->sk_shutdown == SHUTDOWN_MASK)
479		mask |= POLLHUP;
480
481	/* readable? */
482	if (!skb_queue_empty(&sk->sk_receive_queue) ||
483	    (sk->sk_shutdown & RCV_SHUTDOWN))
484		mask |= POLLIN | POLLRDNORM;
485
486	/* Connection-based need to check for termination and startup */
487	if (connection_based(sk)) {
488		if (sk->sk_state == TCP_CLOSE)
489			mask |= POLLHUP;
490		/* connection hasn't started yet? */
491		if (sk->sk_state == TCP_SYN_SENT)
492			return mask;
493	}
494
495	/* writable? */
496	if (sock_writeable(sk))
497		mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
498	else
499		set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
500
501	return mask;
502}
503
504EXPORT_SYMBOL(datagram_poll);
505EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
506EXPORT_SYMBOL(skb_copy_datagram_iovec);
507EXPORT_SYMBOL(skb_free_datagram);
508EXPORT_SYMBOL(skb_recv_datagram);
509