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>
54 #include <linux/indirect_call_wrapper.h>
56 #include <net/protocol.h>
57 #include <linux/skbuff.h>
59 #include <net/checksum.h>
61 #include <net/tcp_states.h>
62 #include <trace/events/skb.h>
63 #include <net/busy_poll.h>
68 * Is a socket 'connection oriented' ?
70 static inline int connection_based(struct sock *sk)
72 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
75 static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
79 * Avoid a wakeup if event not interesting for us
81 if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR)))
83 return autoremove_wake_function(wait, mode, sync, key);
86 * Wait for the last received packet to be different from skb
88 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
89 const struct sk_buff *skb)
92 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
94 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
97 error = sock_error(sk);
101 if (READ_ONCE(sk->sk_receive_queue.prev) != skb)
104 /* Socket shut down? */
105 if (sk->sk_shutdown & RCV_SHUTDOWN)
108 /* Sequenced packets can come disconnected.
109 * If so we report the problem
112 if (connection_based(sk) &&
113 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
117 if (signal_pending(current))
121 *timeo_p = schedule_timeout(*timeo_p);
123 finish_wait(sk_sleep(sk), &wait);
126 error = sock_intr_errno(*timeo_p);
135 EXPORT_SYMBOL(__skb_wait_for_more_packets);
137 static struct sk_buff *skb_set_peeked(struct sk_buff *skb)
139 struct sk_buff *nskb;
144 /* We have to unshare an skb before modifying it. */
145 if (!skb_shared(skb))
148 nskb = skb_clone(skb, GFP_ATOMIC);
150 return ERR_PTR(-ENOMEM);
152 skb->prev->next = nskb;
153 skb->next->prev = nskb;
154 nskb->prev = skb->prev;
155 nskb->next = skb->next;
166 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
167 struct sk_buff_head *queue,
169 void (*destructor)(struct sock *sk,
170 struct sk_buff *skb),
172 struct sk_buff **last)
174 bool peek_at_off = false;
178 if (unlikely(flags & MSG_PEEK && *off >= 0)) {
184 skb_queue_walk(queue, skb) {
185 if (flags & MSG_PEEK) {
186 if (peek_at_off && _off >= skb->len &&
187 (_off || skb->peeked)) {
192 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 * @off: an offset in bytes to peek skb from. Returns an offset
216 * within an skb where data actually starts
217 * @err: error code returned
218 * @last: set to last peeked message to inform the wait function
219 * what to look for when peeking
221 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
222 * and possible races. This replaces identical code in packet, raw and
223 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
224 * the long standing peek and read race for datagram sockets. If you
225 * alter this routine remember it must be re-entrant.
227 * This function will lock the socket if a skb is returned, so
228 * the caller needs to unlock the socket in that case (usually by
229 * calling skb_free_datagram). Returns NULL with @err set to
230 * -EAGAIN if no data was available or to some other value if an
231 * error was detected.
233 * * It does not lock socket since today. This function is
234 * * free of race conditions. This measure should/can improve
235 * * significantly datagram socket latencies at high loads,
236 * * when data copying to user space takes lots of time.
237 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
241 * The order of the tests when we find no data waiting are specified
242 * quite explicitly by POSIX 1003.1g, don't change them without having
243 * the standard around please.
245 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags,
246 void (*destructor)(struct sock *sk,
247 struct sk_buff *skb),
249 struct sk_buff **last)
251 struct sk_buff_head *queue = &sk->sk_receive_queue;
253 unsigned long cpu_flags;
255 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
257 int error = sock_error(sk);
263 /* Again only user level code calls this function, so nothing
264 * interrupt level will suddenly eat the receive_queue.
266 * Look at current nfs client by the way...
267 * However, this function was correct in any case. 8)
269 spin_lock_irqsave(&queue->lock, cpu_flags);
270 skb = __skb_try_recv_from_queue(sk, queue, flags, destructor,
272 spin_unlock_irqrestore(&queue->lock, cpu_flags);
278 if (!sk_can_busy_loop(sk))
281 sk_busy_loop(sk, flags & MSG_DONTWAIT);
282 } while (READ_ONCE(sk->sk_receive_queue.prev) != *last);
290 EXPORT_SYMBOL(__skb_try_recv_datagram);
292 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
293 void (*destructor)(struct sock *sk,
294 struct sk_buff *skb),
297 struct sk_buff *skb, *last;
300 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
303 skb = __skb_try_recv_datagram(sk, flags, destructor, off, err,
311 !__skb_wait_for_more_packets(sk, err, &timeo, last));
315 EXPORT_SYMBOL(__skb_recv_datagram);
317 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
318 int noblock, int *err)
322 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
325 EXPORT_SYMBOL(skb_recv_datagram);
327 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
330 sk_mem_reclaim_partial(sk);
332 EXPORT_SYMBOL(skb_free_datagram);
334 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len)
338 if (!skb_unref(skb)) {
339 sk_peek_offset_bwd(sk, len);
343 slow = lock_sock_fast(sk);
344 sk_peek_offset_bwd(sk, len);
346 sk_mem_reclaim_partial(sk);
347 unlock_sock_fast(sk, slow);
349 /* skb is now orphaned, can be freed outside of locked section */
352 EXPORT_SYMBOL(__skb_free_datagram_locked);
354 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
355 struct sk_buff *skb, unsigned int flags,
356 void (*destructor)(struct sock *sk,
357 struct sk_buff *skb))
361 if (flags & MSG_PEEK) {
363 spin_lock_bh(&sk_queue->lock);
365 __skb_unlink(skb, sk_queue);
366 refcount_dec(&skb->users);
371 spin_unlock_bh(&sk_queue->lock);
374 atomic_inc(&sk->sk_drops);
377 EXPORT_SYMBOL(__sk_queue_drop_skb);
380 * skb_kill_datagram - Free a datagram skbuff forcibly
382 * @skb: datagram skbuff
383 * @flags: MSG\_ flags
385 * This function frees a datagram skbuff that was received by
386 * skb_recv_datagram. The flags argument must match the one
387 * used for skb_recv_datagram.
389 * If the MSG_PEEK flag is set, and the packet is still on the
390 * receive queue of the socket, it will be taken off the queue
391 * before it is freed.
393 * This function currently only disables BH when acquiring the
394 * sk_receive_queue lock. Therefore it must not be used in a
395 * context where that lock is acquired in an IRQ context.
397 * It returns 0 if the packet was removed by us.
400 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
402 int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags,
406 sk_mem_reclaim_partial(sk);
409 EXPORT_SYMBOL(skb_kill_datagram);
411 INDIRECT_CALLABLE_DECLARE(static size_t simple_copy_to_iter(const void *addr,
413 void *data __always_unused,
414 struct iov_iter *i));
416 static int __skb_datagram_iter(const struct sk_buff *skb, int offset,
417 struct iov_iter *to, int len, bool fault_short,
418 size_t (*cb)(const void *, size_t, void *,
419 struct iov_iter *), void *data)
421 int start = skb_headlen(skb);
422 int i, copy = start - offset, start_off = offset, n;
423 struct sk_buff *frag_iter;
429 n = INDIRECT_CALL_1(cb, simple_copy_to_iter,
430 skb->data + offset, copy, data, to);
434 if ((len -= copy) == 0)
438 /* Copy paged appendix. Hmm... why does this look so complicated? */
439 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
441 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
443 WARN_ON(start > offset + len);
445 end = start + skb_frag_size(frag);
446 if ((copy = end - offset) > 0) {
447 struct page *page = skb_frag_page(frag);
448 u8 *vaddr = kmap(page);
452 n = INDIRECT_CALL_1(cb, simple_copy_to_iter,
453 vaddr + skb_frag_off(frag) + offset - start,
465 skb_walk_frags(skb, frag_iter) {
468 WARN_ON(start > offset + len);
470 end = start + frag_iter->len;
471 if ((copy = end - offset) > 0) {
474 if (__skb_datagram_iter(frag_iter, offset - start,
475 to, copy, fault_short, cb, data))
477 if ((len -= copy) == 0)
486 /* This is not really a user copy fault, but rather someone
487 * gave us a bogus length on the skb. We should probably
488 * print a warning here as it may indicate a kernel bug.
492 iov_iter_revert(to, offset - start_off);
496 if (fault_short || iov_iter_count(to))
503 * skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator
505 * @skb: buffer to copy
506 * @offset: offset in the buffer to start copying from
507 * @to: iovec iterator to copy to
508 * @len: amount of data to copy from buffer to iovec
509 * @hash: hash request to update
511 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
512 struct iov_iter *to, int len,
513 struct ahash_request *hash)
515 return __skb_datagram_iter(skb, offset, to, len, true,
516 hash_and_copy_to_iter, hash);
518 EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter);
520 static size_t simple_copy_to_iter(const void *addr, size_t bytes,
521 void *data __always_unused, struct iov_iter *i)
523 return copy_to_iter(addr, bytes, i);
527 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
528 * @skb: buffer to copy
529 * @offset: offset in the buffer to start copying from
530 * @to: iovec iterator to copy to
531 * @len: amount of data to copy from buffer to iovec
533 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset,
534 struct iov_iter *to, int len)
536 trace_skb_copy_datagram_iovec(skb, len);
537 return __skb_datagram_iter(skb, offset, to, len, false,
538 simple_copy_to_iter, NULL);
540 EXPORT_SYMBOL(skb_copy_datagram_iter);
543 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
544 * @skb: buffer to copy
545 * @offset: offset in the buffer to start copying to
546 * @from: the copy source
547 * @len: amount of data to copy to buffer from iovec
549 * Returns 0 or -EFAULT.
551 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
552 struct iov_iter *from,
555 int start = skb_headlen(skb);
556 int i, copy = start - offset;
557 struct sk_buff *frag_iter;
563 if (copy_from_iter(skb->data + offset, copy, from) != copy)
565 if ((len -= copy) == 0)
570 /* Copy paged appendix. Hmm... why does this look so complicated? */
571 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
573 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
575 WARN_ON(start > offset + len);
577 end = start + skb_frag_size(frag);
578 if ((copy = end - offset) > 0) {
583 copied = copy_page_from_iter(skb_frag_page(frag),
584 skb_frag_off(frag) + offset - start,
596 skb_walk_frags(skb, frag_iter) {
599 WARN_ON(start > offset + len);
601 end = start + frag_iter->len;
602 if ((copy = end - offset) > 0) {
605 if (skb_copy_datagram_from_iter(frag_iter,
609 if ((len -= copy) == 0)
621 EXPORT_SYMBOL(skb_copy_datagram_from_iter);
623 int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
624 struct iov_iter *from, size_t length)
626 int frag = skb_shinfo(skb)->nr_frags;
628 while (length && iov_iter_count(from)) {
629 struct page *pages[MAX_SKB_FRAGS];
632 unsigned long truesize;
635 if (frag == MAX_SKB_FRAGS)
638 copied = iov_iter_get_pages(from, pages, length,
639 MAX_SKB_FRAGS - frag, &start);
643 iov_iter_advance(from, copied);
646 truesize = PAGE_ALIGN(copied + start);
647 skb->data_len += copied;
649 skb->truesize += truesize;
650 if (sk && sk->sk_type == SOCK_STREAM) {
651 sk_wmem_queued_add(sk, truesize);
652 sk_mem_charge(sk, truesize);
654 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
657 int size = min_t(int, copied, PAGE_SIZE - start);
658 skb_fill_page_desc(skb, frag++, pages[n], start, size);
666 EXPORT_SYMBOL(__zerocopy_sg_from_iter);
669 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
670 * @skb: buffer to copy
671 * @from: the source to copy from
673 * The function will first copy up to headlen, and then pin the userspace
674 * pages and build frags through them.
676 * Returns 0, -EFAULT or -EMSGSIZE.
678 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from)
680 int copy = min_t(int, skb_headlen(skb), iov_iter_count(from));
682 /* copy up to skb headlen */
683 if (skb_copy_datagram_from_iter(skb, 0, from, copy))
686 return __zerocopy_sg_from_iter(NULL, skb, from, ~0U);
688 EXPORT_SYMBOL(zerocopy_sg_from_iter);
691 * skb_copy_and_csum_datagram_iter - Copy datagram to an iovec iterator
692 * and update a checksum.
693 * @skb: buffer to copy
694 * @offset: offset in the buffer to start copying from
695 * @to: iovec iterator to copy to
696 * @len: amount of data to copy from buffer to iovec
697 * @csump: checksum pointer
699 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
700 struct iov_iter *to, int len,
703 return __skb_datagram_iter(skb, offset, to, len, true,
704 csum_and_copy_to_iter, csump);
708 * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec.
710 * @hlen: hardware length
713 * Caller _must_ check that skb will fit to this iovec.
715 * Returns: 0 - success.
716 * -EINVAL - checksum failure.
717 * -EFAULT - fault during copy.
719 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb,
720 int hlen, struct msghdr *msg)
723 int chunk = skb->len - hlen;
728 if (msg_data_left(msg) < chunk) {
729 if (__skb_checksum_complete(skb))
731 if (skb_copy_datagram_msg(skb, hlen, msg, chunk))
734 csum = csum_partial(skb->data, hlen, skb->csum);
735 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter,
739 if (csum_fold(csum)) {
740 iov_iter_revert(&msg->msg_iter, chunk);
744 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
745 !skb->csum_complete_sw)
746 netdev_rx_csum_fault(NULL, skb);
752 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg);
755 * datagram_poll - generic datagram poll
760 * Datagram poll: Again totally generic. This also handles
761 * sequenced packet sockets providing the socket receive queue
762 * is only ever holding data ready to receive.
764 * Note: when you *don't* use this routine for this protocol,
765 * and you use a different write policy from sock_writeable()
766 * then please supply your own write_space callback.
768 __poll_t datagram_poll(struct file *file, struct socket *sock,
771 struct sock *sk = sock->sk;
774 sock_poll_wait(file, sock, wait);
777 /* exceptional events? */
778 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
780 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
782 if (sk->sk_shutdown & RCV_SHUTDOWN)
783 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
784 if (sk->sk_shutdown == SHUTDOWN_MASK)
788 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
789 mask |= EPOLLIN | EPOLLRDNORM;
791 /* Connection-based need to check for termination and startup */
792 if (connection_based(sk)) {
793 if (sk->sk_state == TCP_CLOSE)
795 /* connection hasn't started yet? */
796 if (sk->sk_state == TCP_SYN_SENT)
801 if (sock_writeable(sk))
802 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
804 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
808 EXPORT_SYMBOL(datagram_poll);