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>
67 * Is a socket 'connection oriented' ?
69 static inline int connection_based(struct sock *sk)
71 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
74 static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
78 * Avoid a wakeup if event not interesting for us
80 if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR)))
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, struct sk_buff_head *queue,
88 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(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 * @queue: socket queue from which to receive
214 * @flags: MSG\_ flags
215 * @destructor: invoked under the receive lock on successful dequeue
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,
247 struct sk_buff_head *queue,
249 void (*destructor)(struct sock *sk,
250 struct sk_buff *skb),
252 struct sk_buff **last)
255 unsigned long cpu_flags;
257 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
259 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,
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(queue->prev) != *last);
292 EXPORT_SYMBOL(__skb_try_recv_datagram);
294 struct sk_buff *__skb_recv_datagram(struct sock *sk,
295 struct sk_buff_head *sk_queue,
297 void (*destructor)(struct sock *sk,
298 struct sk_buff *skb),
301 struct sk_buff *skb, *last;
304 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
307 skb = __skb_try_recv_datagram(sk, sk_queue, flags, destructor,
315 !__skb_wait_for_more_packets(sk, sk_queue, err,
320 EXPORT_SYMBOL(__skb_recv_datagram);
322 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
323 int noblock, int *err)
327 return __skb_recv_datagram(sk, &sk->sk_receive_queue,
328 flags | (noblock ? MSG_DONTWAIT : 0),
331 EXPORT_SYMBOL(skb_recv_datagram);
333 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
336 sk_mem_reclaim_partial(sk);
338 EXPORT_SYMBOL(skb_free_datagram);
340 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len)
344 if (!skb_unref(skb)) {
345 sk_peek_offset_bwd(sk, len);
349 slow = lock_sock_fast(sk);
350 sk_peek_offset_bwd(sk, len);
352 sk_mem_reclaim_partial(sk);
353 unlock_sock_fast(sk, slow);
355 /* skb is now orphaned, can be freed outside of locked section */
358 EXPORT_SYMBOL(__skb_free_datagram_locked);
360 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
361 struct sk_buff *skb, unsigned int flags,
362 void (*destructor)(struct sock *sk,
363 struct sk_buff *skb))
367 if (flags & MSG_PEEK) {
369 spin_lock_bh(&sk_queue->lock);
371 __skb_unlink(skb, sk_queue);
372 refcount_dec(&skb->users);
377 spin_unlock_bh(&sk_queue->lock);
380 atomic_inc(&sk->sk_drops);
383 EXPORT_SYMBOL(__sk_queue_drop_skb);
386 * skb_kill_datagram - Free a datagram skbuff forcibly
388 * @skb: datagram skbuff
389 * @flags: MSG\_ flags
391 * This function frees a datagram skbuff that was received by
392 * skb_recv_datagram. The flags argument must match the one
393 * used for skb_recv_datagram.
395 * If the MSG_PEEK flag is set, and the packet is still on the
396 * receive queue of the socket, it will be taken off the queue
397 * before it is freed.
399 * This function currently only disables BH when acquiring the
400 * sk_receive_queue lock. Therefore it must not be used in a
401 * context where that lock is acquired in an IRQ context.
403 * It returns 0 if the packet was removed by us.
406 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
408 int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags,
412 sk_mem_reclaim_partial(sk);
415 EXPORT_SYMBOL(skb_kill_datagram);
417 static int __skb_datagram_iter(const struct sk_buff *skb, int offset,
418 struct iov_iter *to, int len, bool fault_short,
419 size_t (*cb)(const void *, size_t, void *,
420 struct iov_iter *), void *data)
422 int start = skb_headlen(skb);
423 int i, copy = start - offset, start_off = offset, n;
424 struct sk_buff *frag_iter;
430 n = cb(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 = cb(vaddr + skb_frag_off(frag) + offset - start,
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_datagram_iter(frag_iter, offset - start,
474 to, copy, fault_short, cb, data))
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 (fault_short || iov_iter_count(to))
502 * skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator
504 * @skb: buffer to copy
505 * @offset: offset in the buffer to start copying from
506 * @to: iovec iterator to copy to
507 * @len: amount of data to copy from buffer to iovec
508 * @hash: hash request to update
510 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
511 struct iov_iter *to, int len,
512 struct ahash_request *hash)
514 return __skb_datagram_iter(skb, offset, to, len, true,
515 hash_and_copy_to_iter, hash);
517 EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter);
519 static size_t simple_copy_to_iter(const void *addr, size_t bytes,
520 void *data __always_unused, struct iov_iter *i)
522 return copy_to_iter(addr, bytes, i);
526 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
527 * @skb: buffer to copy
528 * @offset: offset in the buffer to start copying from
529 * @to: iovec iterator to copy to
530 * @len: amount of data to copy from buffer to iovec
532 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset,
533 struct iov_iter *to, int len)
535 trace_skb_copy_datagram_iovec(skb, len);
536 return __skb_datagram_iter(skb, offset, to, len, false,
537 simple_copy_to_iter, NULL);
539 EXPORT_SYMBOL(skb_copy_datagram_iter);
542 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
543 * @skb: buffer to copy
544 * @offset: offset in the buffer to start copying to
545 * @from: the copy source
546 * @len: amount of data to copy to buffer from iovec
548 * Returns 0 or -EFAULT.
550 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
551 struct iov_iter *from,
554 int start = skb_headlen(skb);
555 int i, copy = start - offset;
556 struct sk_buff *frag_iter;
562 if (copy_from_iter(skb->data + offset, copy, from) != copy)
564 if ((len -= copy) == 0)
569 /* Copy paged appendix. Hmm... why does this look so complicated? */
570 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
572 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
574 WARN_ON(start > offset + len);
576 end = start + skb_frag_size(frag);
577 if ((copy = end - offset) > 0) {
582 copied = copy_page_from_iter(skb_frag_page(frag),
583 skb_frag_off(frag) + offset - start,
595 skb_walk_frags(skb, frag_iter) {
598 WARN_ON(start > offset + len);
600 end = start + frag_iter->len;
601 if ((copy = end - offset) > 0) {
604 if (skb_copy_datagram_from_iter(frag_iter,
608 if ((len -= copy) == 0)
620 EXPORT_SYMBOL(skb_copy_datagram_from_iter);
622 int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
623 struct iov_iter *from, size_t length)
625 int frag = skb_shinfo(skb)->nr_frags;
627 while (length && iov_iter_count(from)) {
628 struct page *pages[MAX_SKB_FRAGS];
631 unsigned long truesize;
634 if (frag == MAX_SKB_FRAGS)
637 copied = iov_iter_get_pages(from, pages, length,
638 MAX_SKB_FRAGS - frag, &start);
642 iov_iter_advance(from, copied);
645 truesize = PAGE_ALIGN(copied + start);
646 skb->data_len += copied;
648 skb->truesize += truesize;
649 if (sk && sk->sk_type == SOCK_STREAM) {
650 sk_wmem_queued_add(sk, truesize);
651 sk_mem_charge(sk, truesize);
653 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
656 int size = min_t(int, copied, PAGE_SIZE - start);
657 skb_fill_page_desc(skb, frag++, pages[n], start, size);
665 EXPORT_SYMBOL(__zerocopy_sg_from_iter);
668 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
669 * @skb: buffer to copy
670 * @from: the source to copy from
672 * The function will first copy up to headlen, and then pin the userspace
673 * pages and build frags through them.
675 * Returns 0, -EFAULT or -EMSGSIZE.
677 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from)
679 int copy = min_t(int, skb_headlen(skb), iov_iter_count(from));
681 /* copy up to skb headlen */
682 if (skb_copy_datagram_from_iter(skb, 0, from, copy))
685 return __zerocopy_sg_from_iter(NULL, skb, from, ~0U);
687 EXPORT_SYMBOL(zerocopy_sg_from_iter);
690 * skb_copy_and_csum_datagram_iter - Copy datagram to an iovec iterator
691 * and update a checksum.
692 * @skb: buffer to copy
693 * @offset: offset in the buffer to start copying from
694 * @to: iovec iterator to copy to
695 * @len: amount of data to copy from buffer to iovec
696 * @csump: checksum pointer
698 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
699 struct iov_iter *to, int len,
702 return __skb_datagram_iter(skb, offset, to, len, true,
703 csum_and_copy_to_iter, csump);
707 * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec.
709 * @hlen: hardware length
712 * Caller _must_ check that skb will fit to this iovec.
714 * Returns: 0 - success.
715 * -EINVAL - checksum failure.
716 * -EFAULT - fault during copy.
718 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb,
719 int hlen, struct msghdr *msg)
722 int chunk = skb->len - hlen;
727 if (msg_data_left(msg) < chunk) {
728 if (__skb_checksum_complete(skb))
730 if (skb_copy_datagram_msg(skb, hlen, msg, chunk))
733 csum = csum_partial(skb->data, hlen, skb->csum);
734 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter,
738 if (csum_fold(csum)) {
739 iov_iter_revert(&msg->msg_iter, chunk);
743 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
744 !skb->csum_complete_sw)
745 netdev_rx_csum_fault(NULL, skb);
751 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg);
754 * datagram_poll - generic datagram poll
759 * Datagram poll: Again totally generic. This also handles
760 * sequenced packet sockets providing the socket receive queue
761 * is only ever holding data ready to receive.
763 * Note: when you *don't* use this routine for this protocol,
764 * and you use a different write policy from sock_writeable()
765 * then please supply your own write_space callback.
767 __poll_t datagram_poll(struct file *file, struct socket *sock,
770 struct sock *sk = sock->sk;
773 sock_poll_wait(file, sock, wait);
776 /* exceptional events? */
777 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
779 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
781 if (sk->sk_shutdown & RCV_SHUTDOWN)
782 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
783 if (sk->sk_shutdown == SHUTDOWN_MASK)
787 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
788 mask |= EPOLLIN | EPOLLRDNORM;
790 /* Connection-based need to check for termination and startup */
791 if (connection_based(sk)) {
792 if (sk->sk_state == TCP_CLOSE)
794 /* connection hasn't started yet? */
795 if (sk->sk_state == TCP_SYN_SENT)
800 if (sock_writeable(sk))
801 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
803 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
807 EXPORT_SYMBOL(datagram_poll);