1 // SPDX-License-Identifier: GPL-2.0
5 * Generic datagram handling routines. These are generic for all
6 * protocols. Possibly a generic IP version on top of these would
7 * make sense. Not tonight however 8-).
8 * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and
9 * NetROM layer all have identical poll code and mostly
10 * identical recvmsg() code. So we share it here. The poll was
11 * shared before but buried in udp.c so I moved it.
13 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old
17 * Alan Cox : NULL return from skb_peek_copy()
19 * Alan Cox : Rewrote skb_read_datagram to avoid the
20 * skb_peek_copy stuff.
21 * Alan Cox : Added support for SOCK_SEQPACKET.
22 * IPX can no longer use the SO_TYPE hack
23 * but AX.25 now works right, and SPX is
25 * Alan Cox : Fixed write poll of non IP protocol
27 * Florian La Roche: Changed for my new skbuff handling.
28 * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET.
29 * Linus Torvalds : BSD semantic fixes.
30 * Alan Cox : Datagram iovec handling
31 * Darryl Miles : Fixed non-blocking SOCK_STREAM.
32 * Alan Cox : POSIXisms
33 * Pete Wyckoff : Unconnected accept() fix.
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
40 #include <linux/uaccess.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 #include <linux/slab.h>
52 #include <linux/pagemap.h>
53 #include <linux/uio.h>
55 #include <net/protocol.h>
56 #include <linux/skbuff.h>
58 #include <net/checksum.h>
60 #include <net/tcp_states.h>
61 #include <trace/events/skb.h>
62 #include <net/busy_poll.h>
65 * Is a socket 'connection oriented' ?
67 static inline int connection_based(struct sock
*sk
)
69 return sk
->sk_type
== SOCK_SEQPACKET
|| sk
->sk_type
== SOCK_STREAM
;
72 static int receiver_wake_function(wait_queue_entry_t
*wait
, unsigned int mode
, int sync
,
75 unsigned long bits
= (unsigned long)key
;
78 * Avoid a wakeup if event not interesting for us
80 if (bits
&& !(bits
& (POLLIN
| POLLERR
)))
82 return autoremove_wake_function(wait
, mode
, sync
, key
);
85 * Wait for the last received packet to be different from skb
87 int __skb_wait_for_more_packets(struct sock
*sk
, int *err
, long *timeo_p
,
88 const struct sk_buff
*skb
)
91 DEFINE_WAIT_FUNC(wait
, receiver_wake_function
);
93 prepare_to_wait_exclusive(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
96 error
= sock_error(sk
);
100 if (READ_ONCE(sk
->sk_receive_queue
.prev
) != skb
)
103 /* Socket shut down? */
104 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
107 /* Sequenced packets can come disconnected.
108 * If so we report the problem
111 if (connection_based(sk
) &&
112 !(sk
->sk_state
== TCP_ESTABLISHED
|| sk
->sk_state
== TCP_LISTEN
))
116 if (signal_pending(current
))
120 *timeo_p
= schedule_timeout(*timeo_p
);
122 finish_wait(sk_sleep(sk
), &wait
);
125 error
= sock_intr_errno(*timeo_p
);
134 EXPORT_SYMBOL(__skb_wait_for_more_packets
);
136 static struct sk_buff
*skb_set_peeked(struct sk_buff
*skb
)
138 struct sk_buff
*nskb
;
143 /* We have to unshare an skb before modifying it. */
144 if (!skb_shared(skb
))
147 nskb
= skb_clone(skb
, GFP_ATOMIC
);
149 return ERR_PTR(-ENOMEM
);
151 skb
->prev
->next
= nskb
;
152 skb
->next
->prev
= nskb
;
153 nskb
->prev
= skb
->prev
;
154 nskb
->next
= skb
->next
;
165 struct sk_buff
*__skb_try_recv_from_queue(struct sock
*sk
,
166 struct sk_buff_head
*queue
,
168 void (*destructor
)(struct sock
*sk
,
169 struct sk_buff
*skb
),
170 int *peeked
, int *off
, int *err
,
171 struct sk_buff
**last
)
173 bool peek_at_off
= false;
177 if (unlikely(flags
& MSG_PEEK
&& *off
>= 0)) {
183 skb_queue_walk(queue
, skb
) {
184 if (flags
& MSG_PEEK
) {
185 if (peek_at_off
&& _off
>= skb
->len
&&
186 (_off
|| skb
->peeked
)) {
191 skb
= skb_set_peeked(skb
);
198 refcount_inc(&skb
->users
);
200 __skb_unlink(skb
, queue
);
211 * __skb_try_recv_datagram - Receive a datagram skbuff
213 * @flags: MSG\_ flags
214 * @destructor: invoked under the receive lock on successful dequeue
215 * @peeked: returns non-zero if this packet has been seen before
216 * @off: an offset in bytes to peek skb from. Returns an offset
217 * within an skb where data actually starts
218 * @err: error code returned
219 * @last: set to last peeked message to inform the wait function
220 * what to look for when peeking
222 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
223 * and possible races. This replaces identical code in packet, raw and
224 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
225 * the long standing peek and read race for datagram sockets. If you
226 * alter this routine remember it must be re-entrant.
228 * This function will lock the socket if a skb is returned, so
229 * the caller needs to unlock the socket in that case (usually by
230 * calling skb_free_datagram). Returns NULL with @err set to
231 * -EAGAIN if no data was available or to some other value if an
232 * error was detected.
234 * * It does not lock socket since today. This function is
235 * * free of race conditions. This measure should/can improve
236 * * significantly datagram socket latencies at high loads,
237 * * when data copying to user space takes lots of time.
238 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
242 * The order of the tests when we find no data waiting are specified
243 * quite explicitly by POSIX 1003.1g, don't change them without having
244 * the standard around please.
246 struct sk_buff
*__skb_try_recv_datagram(struct sock
*sk
, unsigned int flags
,
247 void (*destructor
)(struct sock
*sk
,
248 struct sk_buff
*skb
),
249 int *peeked
, int *off
, int *err
,
250 struct sk_buff
**last
)
252 struct sk_buff_head
*queue
= &sk
->sk_receive_queue
;
254 unsigned long cpu_flags
;
256 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
258 int error
= sock_error(sk
);
265 /* Again only user level code calls this function, so nothing
266 * interrupt level will suddenly eat the receive_queue.
268 * Look at current nfs client by the way...
269 * However, this function was correct in any case. 8)
271 spin_lock_irqsave(&queue
->lock
, cpu_flags
);
272 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
, destructor
,
273 peeked
, off
, &error
, last
);
274 spin_unlock_irqrestore(&queue
->lock
, cpu_flags
);
280 if (!sk_can_busy_loop(sk
))
283 sk_busy_loop(sk
, flags
& MSG_DONTWAIT
);
284 } while (READ_ONCE(sk
->sk_receive_queue
.prev
) != *last
);
292 EXPORT_SYMBOL(__skb_try_recv_datagram
);
294 struct sk_buff
*__skb_recv_datagram(struct sock
*sk
, unsigned int flags
,
295 void (*destructor
)(struct sock
*sk
,
296 struct sk_buff
*skb
),
297 int *peeked
, int *off
, int *err
)
299 struct sk_buff
*skb
, *last
;
302 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
305 skb
= __skb_try_recv_datagram(sk
, flags
, destructor
, peeked
,
313 !__skb_wait_for_more_packets(sk
, err
, &timeo
, last
));
317 EXPORT_SYMBOL(__skb_recv_datagram
);
319 struct sk_buff
*skb_recv_datagram(struct sock
*sk
, unsigned int flags
,
320 int noblock
, int *err
)
324 return __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
325 NULL
, &peeked
, &off
, err
);
327 EXPORT_SYMBOL(skb_recv_datagram
);
329 void skb_free_datagram(struct sock
*sk
, struct sk_buff
*skb
)
332 sk_mem_reclaim_partial(sk
);
334 EXPORT_SYMBOL(skb_free_datagram
);
336 void __skb_free_datagram_locked(struct sock
*sk
, struct sk_buff
*skb
, int len
)
340 if (!skb_unref(skb
)) {
341 sk_peek_offset_bwd(sk
, len
);
345 slow
= lock_sock_fast(sk
);
346 sk_peek_offset_bwd(sk
, len
);
348 sk_mem_reclaim_partial(sk
);
349 unlock_sock_fast(sk
, slow
);
351 /* skb is now orphaned, can be freed outside of locked section */
354 EXPORT_SYMBOL(__skb_free_datagram_locked
);
356 int __sk_queue_drop_skb(struct sock
*sk
, struct sk_buff_head
*sk_queue
,
357 struct sk_buff
*skb
, unsigned int flags
,
358 void (*destructor
)(struct sock
*sk
,
359 struct sk_buff
*skb
))
363 if (flags
& MSG_PEEK
) {
365 spin_lock_bh(&sk_queue
->lock
);
367 __skb_unlink(skb
, sk_queue
);
368 refcount_dec(&skb
->users
);
373 spin_unlock_bh(&sk_queue
->lock
);
376 atomic_inc(&sk
->sk_drops
);
379 EXPORT_SYMBOL(__sk_queue_drop_skb
);
382 * skb_kill_datagram - Free a datagram skbuff forcibly
384 * @skb: datagram skbuff
385 * @flags: MSG\_ flags
387 * This function frees a datagram skbuff that was received by
388 * skb_recv_datagram. The flags argument must match the one
389 * used for skb_recv_datagram.
391 * If the MSG_PEEK flag is set, and the packet is still on the
392 * receive queue of the socket, it will be taken off the queue
393 * before it is freed.
395 * This function currently only disables BH when acquiring the
396 * sk_receive_queue lock. Therefore it must not be used in a
397 * context where that lock is acquired in an IRQ context.
399 * It returns 0 if the packet was removed by us.
402 int skb_kill_datagram(struct sock
*sk
, struct sk_buff
*skb
, unsigned int flags
)
404 int err
= __sk_queue_drop_skb(sk
, &sk
->sk_receive_queue
, skb
, flags
,
408 sk_mem_reclaim_partial(sk
);
411 EXPORT_SYMBOL(skb_kill_datagram
);
414 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
415 * @skb: buffer to copy
416 * @offset: offset in the buffer to start copying from
417 * @to: iovec iterator to copy to
418 * @len: amount of data to copy from buffer to iovec
420 int skb_copy_datagram_iter(const struct sk_buff
*skb
, int offset
,
421 struct iov_iter
*to
, int len
)
423 int start
= skb_headlen(skb
);
424 int i
, copy
= start
- offset
, start_off
= offset
, n
;
425 struct sk_buff
*frag_iter
;
427 trace_skb_copy_datagram_iovec(skb
, len
);
433 n
= copy_to_iter(skb
->data
+ offset
, copy
, to
);
437 if ((len
-= copy
) == 0)
441 /* Copy paged appendix. Hmm... why does this look so complicated? */
442 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
444 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
446 WARN_ON(start
> offset
+ len
);
448 end
= start
+ skb_frag_size(frag
);
449 if ((copy
= end
- offset
) > 0) {
452 n
= copy_page_to_iter(skb_frag_page(frag
),
453 frag
->page_offset
+ offset
-
464 skb_walk_frags(skb
, frag_iter
) {
467 WARN_ON(start
> offset
+ len
);
469 end
= start
+ frag_iter
->len
;
470 if ((copy
= end
- offset
) > 0) {
473 if (skb_copy_datagram_iter(frag_iter
, offset
- start
,
476 if ((len
-= copy
) == 0)
485 /* This is not really a user copy fault, but rather someone
486 * gave us a bogus length on the skb. We should probably
487 * print a warning here as it may indicate a kernel bug.
491 iov_iter_revert(to
, offset
- start_off
);
495 if (iov_iter_count(to
))
500 EXPORT_SYMBOL(skb_copy_datagram_iter
);
503 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
504 * @skb: buffer to copy
505 * @offset: offset in the buffer to start copying to
506 * @from: the copy source
507 * @len: amount of data to copy to buffer from iovec
509 * Returns 0 or -EFAULT.
511 int skb_copy_datagram_from_iter(struct sk_buff
*skb
, int offset
,
512 struct iov_iter
*from
,
515 int start
= skb_headlen(skb
);
516 int i
, copy
= start
- offset
;
517 struct sk_buff
*frag_iter
;
523 if (copy_from_iter(skb
->data
+ offset
, copy
, from
) != copy
)
525 if ((len
-= copy
) == 0)
530 /* Copy paged appendix. Hmm... why does this look so complicated? */
531 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
533 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
535 WARN_ON(start
> offset
+ len
);
537 end
= start
+ skb_frag_size(frag
);
538 if ((copy
= end
- offset
) > 0) {
543 copied
= copy_page_from_iter(skb_frag_page(frag
),
544 frag
->page_offset
+ offset
- start
,
556 skb_walk_frags(skb
, frag_iter
) {
559 WARN_ON(start
> offset
+ len
);
561 end
= start
+ frag_iter
->len
;
562 if ((copy
= end
- offset
) > 0) {
565 if (skb_copy_datagram_from_iter(frag_iter
,
569 if ((len
-= copy
) == 0)
581 EXPORT_SYMBOL(skb_copy_datagram_from_iter
);
583 int __zerocopy_sg_from_iter(struct sock
*sk
, struct sk_buff
*skb
,
584 struct iov_iter
*from
, size_t length
)
586 int frag
= skb_shinfo(skb
)->nr_frags
;
588 while (length
&& iov_iter_count(from
)) {
589 struct page
*pages
[MAX_SKB_FRAGS
];
592 unsigned long truesize
;
595 if (frag
== MAX_SKB_FRAGS
)
598 copied
= iov_iter_get_pages(from
, pages
, length
,
599 MAX_SKB_FRAGS
- frag
, &start
);
603 iov_iter_advance(from
, copied
);
606 truesize
= PAGE_ALIGN(copied
+ start
);
607 skb
->data_len
+= copied
;
609 skb
->truesize
+= truesize
;
610 if (sk
&& sk
->sk_type
== SOCK_STREAM
) {
611 sk
->sk_wmem_queued
+= truesize
;
612 sk_mem_charge(sk
, truesize
);
614 refcount_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
617 int size
= min_t(int, copied
, PAGE_SIZE
- start
);
618 skb_fill_page_desc(skb
, frag
++, pages
[n
], start
, size
);
626 EXPORT_SYMBOL(__zerocopy_sg_from_iter
);
629 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
630 * @skb: buffer to copy
631 * @from: the source to copy from
633 * The function will first copy up to headlen, and then pin the userspace
634 * pages and build frags through them.
636 * Returns 0, -EFAULT or -EMSGSIZE.
638 int zerocopy_sg_from_iter(struct sk_buff
*skb
, struct iov_iter
*from
)
640 int copy
= min_t(int, skb_headlen(skb
), iov_iter_count(from
));
642 /* copy up to skb headlen */
643 if (skb_copy_datagram_from_iter(skb
, 0, from
, copy
))
646 return __zerocopy_sg_from_iter(NULL
, skb
, from
, ~0U);
648 EXPORT_SYMBOL(zerocopy_sg_from_iter
);
650 static int skb_copy_and_csum_datagram(const struct sk_buff
*skb
, int offset
,
651 struct iov_iter
*to
, int len
,
654 int start
= skb_headlen(skb
);
655 int i
, copy
= start
- offset
, start_off
= offset
;
656 struct sk_buff
*frag_iter
;
664 n
= csum_and_copy_to_iter(skb
->data
+ offset
, copy
, csump
, to
);
668 if ((len
-= copy
) == 0)
673 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
675 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
677 WARN_ON(start
> offset
+ len
);
679 end
= start
+ skb_frag_size(frag
);
680 if ((copy
= end
- offset
) > 0) {
682 struct page
*page
= skb_frag_page(frag
);
683 u8
*vaddr
= kmap(page
);
687 n
= csum_and_copy_to_iter(vaddr
+ frag
->page_offset
+
688 offset
- start
, copy
,
694 *csump
= csum_block_add(*csump
, csum2
, pos
);
702 skb_walk_frags(skb
, frag_iter
) {
705 WARN_ON(start
> offset
+ len
);
707 end
= start
+ frag_iter
->len
;
708 if ((copy
= end
- offset
) > 0) {
712 if (skb_copy_and_csum_datagram(frag_iter
,
717 *csump
= csum_block_add(*csump
, csum2
, pos
);
718 if ((len
-= copy
) == 0)
729 iov_iter_revert(to
, offset
- start_off
);
733 __sum16
__skb_checksum_complete_head(struct sk_buff
*skb
, int len
)
737 sum
= csum_fold(skb_checksum(skb
, 0, len
, skb
->csum
));
739 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
740 !skb
->csum_complete_sw
)
741 netdev_rx_csum_fault(skb
->dev
);
743 if (!skb_shared(skb
))
744 skb
->csum_valid
= !sum
;
747 EXPORT_SYMBOL(__skb_checksum_complete_head
);
749 __sum16
__skb_checksum_complete(struct sk_buff
*skb
)
754 csum
= skb_checksum(skb
, 0, skb
->len
, 0);
756 /* skb->csum holds pseudo checksum */
757 sum
= csum_fold(csum_add(skb
->csum
, csum
));
759 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
760 !skb
->csum_complete_sw
)
761 netdev_rx_csum_fault(skb
->dev
);
764 if (!skb_shared(skb
)) {
765 /* Save full packet checksum */
767 skb
->ip_summed
= CHECKSUM_COMPLETE
;
768 skb
->csum_complete_sw
= 1;
769 skb
->csum_valid
= !sum
;
774 EXPORT_SYMBOL(__skb_checksum_complete
);
777 * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec.
779 * @hlen: hardware length
782 * Caller _must_ check that skb will fit to this iovec.
784 * Returns: 0 - success.
785 * -EINVAL - checksum failure.
786 * -EFAULT - fault during copy.
788 int skb_copy_and_csum_datagram_msg(struct sk_buff
*skb
,
789 int hlen
, struct msghdr
*msg
)
792 int chunk
= skb
->len
- hlen
;
797 if (msg_data_left(msg
) < chunk
) {
798 if (__skb_checksum_complete(skb
))
800 if (skb_copy_datagram_msg(skb
, hlen
, msg
, chunk
))
803 csum
= csum_partial(skb
->data
, hlen
, skb
->csum
);
804 if (skb_copy_and_csum_datagram(skb
, hlen
, &msg
->msg_iter
,
808 if (csum_fold(csum
)) {
809 iov_iter_revert(&msg
->msg_iter
, chunk
);
813 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
814 !skb
->csum_complete_sw
)
815 netdev_rx_csum_fault(NULL
);
821 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg
);
824 * datagram_poll - generic datagram poll
829 * Datagram poll: Again totally generic. This also handles
830 * sequenced packet sockets providing the socket receive queue
831 * is only ever holding data ready to receive.
833 * Note: when you *don't* use this routine for this protocol,
834 * and you use a different write policy from sock_writeable()
835 * then please supply your own write_space callback.
837 unsigned int datagram_poll(struct file
*file
, struct socket
*sock
,
840 struct sock
*sk
= sock
->sk
;
843 sock_poll_wait(file
, sk_sleep(sk
), wait
);
846 /* exceptional events? */
847 if (sk
->sk_err
|| !skb_queue_empty_lockless(&sk
->sk_error_queue
))
849 (sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
) ? POLLPRI
: 0);
851 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
852 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
853 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
857 if (!skb_queue_empty_lockless(&sk
->sk_receive_queue
))
858 mask
|= POLLIN
| POLLRDNORM
;
860 /* Connection-based need to check for termination and startup */
861 if (connection_based(sk
)) {
862 if (sk
->sk_state
== TCP_CLOSE
)
864 /* connection hasn't started yet? */
865 if (sk
->sk_state
== TCP_SYN_SENT
)
870 if (sock_writeable(sk
))
871 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
873 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
877 EXPORT_SYMBOL(datagram_poll
);