2 * Kernel Connection Multiplexor
4 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2
8 * as published by the Free Software Foundation.
11 #include <linux/bpf.h>
12 #include <linux/errno.h>
13 #include <linux/errqueue.h>
14 #include <linux/file.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/net.h>
19 #include <linux/netdevice.h>
20 #include <linux/poll.h>
21 #include <linux/rculist.h>
22 #include <linux/skbuff.h>
23 #include <linux/socket.h>
24 #include <linux/uaccess.h>
25 #include <linux/workqueue.h>
27 #include <net/netns/generic.h>
30 #include <uapi/linux/kcm.h>
32 unsigned int kcm_net_id;
34 static struct kmem_cache *kcm_psockp __read_mostly;
35 static struct kmem_cache *kcm_muxp __read_mostly;
36 static struct workqueue_struct *kcm_wq;
38 static inline struct kcm_sock *kcm_sk(const struct sock *sk)
40 return (struct kcm_sock *)sk;
43 static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
45 return (struct kcm_tx_msg *)skb->cb;
48 static void report_csk_error(struct sock *csk, int err)
51 csk->sk_error_report(csk);
54 static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
57 struct sock *csk = psock->sk;
58 struct kcm_mux *mux = psock->mux;
60 /* Unrecoverable error in transmit */
62 spin_lock_bh(&mux->lock);
64 if (psock->tx_stopped) {
65 spin_unlock_bh(&mux->lock);
69 psock->tx_stopped = 1;
70 KCM_STATS_INCR(psock->stats.tx_aborts);
73 /* Take off psocks_avail list */
74 list_del(&psock->psock_avail_list);
75 } else if (wakeup_kcm) {
76 /* In this case psock is being aborted while outside of
77 * write_msgs and psock is reserved. Schedule tx_work
78 * to handle the failure there. Need to commit tx_stopped
79 * before queuing work.
83 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
86 spin_unlock_bh(&mux->lock);
88 /* Report error on lower socket */
89 report_csk_error(csk, err);
92 /* RX mux lock held. */
93 static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
94 struct kcm_psock *psock)
96 STRP_STATS_ADD(mux->stats.rx_bytes,
97 psock->strp.stats.rx_bytes -
98 psock->saved_rx_bytes);
100 psock->strp.stats.rx_msgs - psock->saved_rx_msgs;
101 psock->saved_rx_msgs = psock->strp.stats.rx_msgs;
102 psock->saved_rx_bytes = psock->strp.stats.rx_bytes;
105 static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
106 struct kcm_psock *psock)
108 KCM_STATS_ADD(mux->stats.tx_bytes,
109 psock->stats.tx_bytes - psock->saved_tx_bytes);
110 mux->stats.tx_msgs +=
111 psock->stats.tx_msgs - psock->saved_tx_msgs;
112 psock->saved_tx_msgs = psock->stats.tx_msgs;
113 psock->saved_tx_bytes = psock->stats.tx_bytes;
116 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
118 /* KCM is ready to receive messages on its queue-- either the KCM is new or
119 * has become unblocked after being blocked on full socket buffer. Queue any
120 * pending ready messages on a psock. RX mux lock held.
122 static void kcm_rcv_ready(struct kcm_sock *kcm)
124 struct kcm_mux *mux = kcm->mux;
125 struct kcm_psock *psock;
128 if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
131 while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) {
132 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
133 /* Assuming buffer limit has been reached */
134 skb_queue_head(&mux->rx_hold_queue, skb);
135 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
140 while (!list_empty(&mux->psocks_ready)) {
141 psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
144 if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) {
145 /* Assuming buffer limit has been reached */
146 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
150 /* Consumed the ready message on the psock. Schedule rx_work to
153 list_del(&psock->psock_ready_list);
154 psock->ready_rx_msg = NULL;
155 /* Commit clearing of ready_rx_msg for queuing work */
158 strp_unpause(&psock->strp);
159 strp_check_rcv(&psock->strp);
162 /* Buffer limit is okay now, add to ready list */
163 list_add_tail(&kcm->wait_rx_list,
164 &kcm->mux->kcm_rx_waiters);
168 static void kcm_rfree(struct sk_buff *skb)
170 struct sock *sk = skb->sk;
171 struct kcm_sock *kcm = kcm_sk(sk);
172 struct kcm_mux *mux = kcm->mux;
173 unsigned int len = skb->truesize;
175 sk_mem_uncharge(sk, len);
176 atomic_sub(len, &sk->sk_rmem_alloc);
178 /* For reading rx_wait and rx_psock without holding lock */
179 smp_mb__after_atomic();
181 if (!kcm->rx_wait && !kcm->rx_psock &&
182 sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
183 spin_lock_bh(&mux->rx_lock);
185 spin_unlock_bh(&mux->rx_lock);
189 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
191 struct sk_buff_head *list = &sk->sk_receive_queue;
193 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
196 if (!sk_rmem_schedule(sk, skb, skb->truesize))
203 skb->destructor = kcm_rfree;
204 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
205 sk_mem_charge(sk, skb->truesize);
207 skb_queue_tail(list, skb);
209 if (!sock_flag(sk, SOCK_DEAD))
210 sk->sk_data_ready(sk);
215 /* Requeue received messages for a kcm socket to other kcm sockets. This is
216 * called with a kcm socket is receive disabled.
219 static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
222 struct kcm_sock *kcm;
224 while ((skb = __skb_dequeue(head))) {
225 /* Reset destructor to avoid calling kcm_rcv_ready */
226 skb->destructor = sock_rfree;
229 if (list_empty(&mux->kcm_rx_waiters)) {
230 skb_queue_tail(&mux->rx_hold_queue, skb);
234 kcm = list_first_entry(&mux->kcm_rx_waiters,
235 struct kcm_sock, wait_rx_list);
237 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
238 /* Should mean socket buffer full */
239 list_del(&kcm->wait_rx_list);
240 kcm->rx_wait = false;
242 /* Commit rx_wait to read in kcm_free */
250 /* Lower sock lock held */
251 static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
252 struct sk_buff *head)
254 struct kcm_mux *mux = psock->mux;
255 struct kcm_sock *kcm;
257 WARN_ON(psock->ready_rx_msg);
260 return psock->rx_kcm;
262 spin_lock_bh(&mux->rx_lock);
265 spin_unlock_bh(&mux->rx_lock);
266 return psock->rx_kcm;
269 kcm_update_rx_mux_stats(mux, psock);
271 if (list_empty(&mux->kcm_rx_waiters)) {
272 psock->ready_rx_msg = head;
273 strp_pause(&psock->strp);
274 list_add_tail(&psock->psock_ready_list,
276 spin_unlock_bh(&mux->rx_lock);
280 kcm = list_first_entry(&mux->kcm_rx_waiters,
281 struct kcm_sock, wait_rx_list);
282 list_del(&kcm->wait_rx_list);
283 kcm->rx_wait = false;
286 kcm->rx_psock = psock;
288 spin_unlock_bh(&mux->rx_lock);
293 static void kcm_done(struct kcm_sock *kcm);
295 static void kcm_done_work(struct work_struct *w)
297 kcm_done(container_of(w, struct kcm_sock, done_work));
300 /* Lower sock held */
301 static void unreserve_rx_kcm(struct kcm_psock *psock,
304 struct kcm_sock *kcm = psock->rx_kcm;
305 struct kcm_mux *mux = psock->mux;
310 spin_lock_bh(&mux->rx_lock);
312 psock->rx_kcm = NULL;
313 kcm->rx_psock = NULL;
315 /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
320 if (unlikely(kcm->done)) {
321 spin_unlock_bh(&mux->rx_lock);
323 /* Need to run kcm_done in a task since we need to qcquire
324 * callback locks which may already be held here.
326 INIT_WORK(&kcm->done_work, kcm_done_work);
327 schedule_work(&kcm->done_work);
331 if (unlikely(kcm->rx_disabled)) {
332 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
333 } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
334 /* Check for degenerative race with rx_wait that all
335 * data was dequeued (accounted for in kcm_rfree).
339 spin_unlock_bh(&mux->rx_lock);
342 /* Lower sock lock held */
343 static void psock_tcp_data_ready(struct sock *sk)
345 struct kcm_psock *psock;
347 read_lock_bh(&sk->sk_callback_lock);
349 psock = (struct kcm_psock *)sk->sk_user_data;
351 strp_tcp_data_ready(&psock->strp);
353 read_unlock_bh(&sk->sk_callback_lock);
356 /* Called with lower sock held */
357 static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
359 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
360 struct kcm_sock *kcm;
363 kcm = reserve_rx_kcm(psock, skb);
365 /* Unable to reserve a KCM, message is held in psock and strp
371 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
372 /* Should mean socket buffer full */
373 unreserve_rx_kcm(psock, false);
378 static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
380 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
381 struct bpf_prog *prog = psock->bpf_prog;
383 return (*prog->bpf_func)(skb, prog->insnsi);
386 static int kcm_read_sock_done(struct strparser *strp, int err)
388 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
390 unreserve_rx_kcm(psock, true);
395 static void psock_tcp_state_change(struct sock *sk)
397 /* TCP only does a POLLIN for a half close. Do a POLLHUP here
398 * since application will normally not poll with POLLIN
399 * on the TCP sockets.
402 report_csk_error(sk, EPIPE);
405 static void psock_tcp_write_space(struct sock *sk)
407 struct kcm_psock *psock;
409 struct kcm_sock *kcm;
411 read_lock_bh(&sk->sk_callback_lock);
413 psock = (struct kcm_psock *)sk->sk_user_data;
414 if (unlikely(!psock))
418 spin_lock_bh(&mux->lock);
420 /* Check if the socket is reserved so someone is waiting for sending. */
422 if (kcm && !unlikely(kcm->tx_stopped))
423 queue_work(kcm_wq, &kcm->tx_work);
425 spin_unlock_bh(&mux->lock);
427 read_unlock_bh(&sk->sk_callback_lock);
430 static void unreserve_psock(struct kcm_sock *kcm);
432 /* kcm sock is locked. */
433 static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
435 struct kcm_mux *mux = kcm->mux;
436 struct kcm_psock *psock;
438 psock = kcm->tx_psock;
440 smp_rmb(); /* Must read tx_psock before tx_wait */
443 WARN_ON(kcm->tx_wait);
444 if (unlikely(psock->tx_stopped))
445 unreserve_psock(kcm);
447 return kcm->tx_psock;
450 spin_lock_bh(&mux->lock);
452 /* Check again under lock to see if psock was reserved for this
453 * psock via psock_unreserve.
455 psock = kcm->tx_psock;
456 if (unlikely(psock)) {
457 WARN_ON(kcm->tx_wait);
458 spin_unlock_bh(&mux->lock);
459 return kcm->tx_psock;
462 if (!list_empty(&mux->psocks_avail)) {
463 psock = list_first_entry(&mux->psocks_avail,
466 list_del(&psock->psock_avail_list);
468 list_del(&kcm->wait_psock_list);
469 kcm->tx_wait = false;
471 kcm->tx_psock = psock;
473 KCM_STATS_INCR(psock->stats.reserved);
474 } else if (!kcm->tx_wait) {
475 list_add_tail(&kcm->wait_psock_list,
476 &mux->kcm_tx_waiters);
480 spin_unlock_bh(&mux->lock);
486 static void psock_now_avail(struct kcm_psock *psock)
488 struct kcm_mux *mux = psock->mux;
489 struct kcm_sock *kcm;
491 if (list_empty(&mux->kcm_tx_waiters)) {
492 list_add_tail(&psock->psock_avail_list,
495 kcm = list_first_entry(&mux->kcm_tx_waiters,
498 list_del(&kcm->wait_psock_list);
499 kcm->tx_wait = false;
502 /* Commit before changing tx_psock since that is read in
503 * reserve_psock before queuing work.
507 kcm->tx_psock = psock;
508 KCM_STATS_INCR(psock->stats.reserved);
509 queue_work(kcm_wq, &kcm->tx_work);
513 /* kcm sock is locked. */
514 static void unreserve_psock(struct kcm_sock *kcm)
516 struct kcm_psock *psock;
517 struct kcm_mux *mux = kcm->mux;
519 spin_lock_bh(&mux->lock);
521 psock = kcm->tx_psock;
523 if (WARN_ON(!psock)) {
524 spin_unlock_bh(&mux->lock);
528 smp_rmb(); /* Read tx_psock before tx_wait */
530 kcm_update_tx_mux_stats(mux, psock);
532 WARN_ON(kcm->tx_wait);
534 kcm->tx_psock = NULL;
535 psock->tx_kcm = NULL;
536 KCM_STATS_INCR(psock->stats.unreserved);
538 if (unlikely(psock->tx_stopped)) {
541 list_del(&psock->psock_list);
544 fput(psock->sk->sk_socket->file);
545 kmem_cache_free(kcm_psockp, psock);
548 /* Don't put back on available list */
550 spin_unlock_bh(&mux->lock);
555 psock_now_avail(psock);
557 spin_unlock_bh(&mux->lock);
560 static void kcm_report_tx_retry(struct kcm_sock *kcm)
562 struct kcm_mux *mux = kcm->mux;
564 spin_lock_bh(&mux->lock);
565 KCM_STATS_INCR(mux->stats.tx_retries);
566 spin_unlock_bh(&mux->lock);
569 /* Write any messages ready on the kcm socket. Called with kcm sock lock
570 * held. Return bytes actually sent or error.
572 static int kcm_write_msgs(struct kcm_sock *kcm)
574 struct sock *sk = &kcm->sk;
575 struct kcm_psock *psock;
576 struct sk_buff *skb, *head;
577 struct kcm_tx_msg *txm;
578 unsigned short fragidx, frag_offset;
579 unsigned int sent, total_sent = 0;
582 kcm->tx_wait_more = false;
583 psock = kcm->tx_psock;
584 if (unlikely(psock && psock->tx_stopped)) {
585 /* A reserved psock was aborted asynchronously. Unreserve
586 * it and we'll retry the message.
588 unreserve_psock(kcm);
589 kcm_report_tx_retry(kcm);
590 if (skb_queue_empty(&sk->sk_write_queue))
593 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
595 } else if (skb_queue_empty(&sk->sk_write_queue)) {
599 head = skb_peek(&sk->sk_write_queue);
600 txm = kcm_tx_msg(head);
603 /* Send of first skbuff in queue already in progress */
604 if (WARN_ON(!psock)) {
609 frag_offset = txm->frag_offset;
610 fragidx = txm->fragidx;
617 psock = reserve_psock(kcm);
623 txm = kcm_tx_msg(head);
627 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
632 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
638 frag = &skb_shinfo(skb)->frags[fragidx];
639 if (WARN_ON(!frag->size)) {
644 ret = kernel_sendpage(psock->sk->sk_socket,
646 frag->page_offset + frag_offset,
647 frag->size - frag_offset,
650 if (ret == -EAGAIN) {
651 /* Save state to try again when there's
652 * write space on the socket
655 txm->frag_offset = frag_offset;
656 txm->fragidx = fragidx;
663 /* Hard failure in sending message, abort this
664 * psock since it has lost framing
665 * synchonization and retry sending the
666 * message from the beginning.
668 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
670 unreserve_psock(kcm);
673 kcm_report_tx_retry(kcm);
681 KCM_STATS_ADD(psock->stats.tx_bytes, ret);
682 if (frag_offset < frag->size) {
683 /* Not finished with this frag */
689 if (skb_has_frag_list(skb)) {
690 skb = skb_shinfo(skb)->frag_list;
693 } else if (skb->next) {
698 /* Successfully sent the whole packet, account for it. */
699 skb_dequeue(&sk->sk_write_queue);
701 sk->sk_wmem_queued -= sent;
703 KCM_STATS_INCR(psock->stats.tx_msgs);
704 } while ((head = skb_peek(&sk->sk_write_queue)));
707 /* Done with all queued messages. */
708 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
709 unreserve_psock(kcm);
712 /* Check if write space is available */
713 sk->sk_write_space(sk);
715 return total_sent ? : ret;
718 static void kcm_tx_work(struct work_struct *w)
720 struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
721 struct sock *sk = &kcm->sk;
726 /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
729 err = kcm_write_msgs(kcm);
731 /* Hard failure in write, report error on KCM socket */
732 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
733 report_csk_error(&kcm->sk, -err);
737 /* Primarily for SOCK_SEQPACKET sockets */
738 if (likely(sk->sk_socket) &&
739 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
740 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
741 sk->sk_write_space(sk);
748 static void kcm_push(struct kcm_sock *kcm)
750 if (kcm->tx_wait_more)
754 static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
755 int offset, size_t size, int flags)
758 struct sock *sk = sock->sk;
759 struct kcm_sock *kcm = kcm_sk(sk);
760 struct sk_buff *skb = NULL, *head = NULL;
761 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
766 if (flags & MSG_SENDPAGE_NOTLAST)
769 /* No MSG_EOR from splice, only look at MSG_MORE */
770 eor = !(flags & MSG_MORE);
774 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
781 /* Previously opened message */
783 skb = kcm_tx_msg(head)->last_skb;
784 i = skb_shinfo(skb)->nr_frags;
786 if (skb_can_coalesce(skb, i, page, offset)) {
787 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
788 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
792 if (i >= MAX_SKB_FRAGS) {
793 struct sk_buff *tskb;
795 tskb = alloc_skb(0, sk->sk_allocation);
798 err = sk_stream_wait_memory(sk, &timeo);
804 skb_shinfo(head)->frag_list = tskb;
809 skb->ip_summed = CHECKSUM_UNNECESSARY;
813 /* Call the sk_stream functions to manage the sndbuf mem. */
814 if (!sk_stream_memory_free(sk)) {
816 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
817 err = sk_stream_wait_memory(sk, &timeo);
822 head = alloc_skb(0, sk->sk_allocation);
825 err = sk_stream_wait_memory(sk, &timeo);
835 skb_fill_page_desc(skb, i, page, offset, size);
836 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
840 skb->data_len += size;
841 skb->truesize += size;
842 sk->sk_wmem_queued += size;
843 sk_mem_charge(sk, size);
847 head->data_len += size;
848 head->truesize += size;
852 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
854 /* Message complete, queue it on send buffer */
855 __skb_queue_tail(&sk->sk_write_queue, head);
857 KCM_STATS_INCR(kcm->stats.tx_msgs);
859 if (flags & MSG_BATCH) {
860 kcm->tx_wait_more = true;
861 } else if (kcm->tx_wait_more || not_busy) {
862 err = kcm_write_msgs(kcm);
864 /* We got a hard error in write_msgs but have
865 * already queued this message. Report an error
866 * in the socket, but don't affect return value
869 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
870 report_csk_error(&kcm->sk, -err);
874 /* Message not complete, save state */
876 kcm_tx_msg(head)->last_skb = skb;
879 KCM_STATS_ADD(kcm->stats.tx_bytes, size);
887 err = sk_stream_error(sk, flags, err);
889 /* make sure we wake any epoll edge trigger waiter */
890 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
891 sk->sk_write_space(sk);
897 static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
899 struct sock *sk = sock->sk;
900 struct kcm_sock *kcm = kcm_sk(sk);
901 struct sk_buff *skb = NULL, *head = NULL;
902 size_t copy, copied = 0;
903 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
904 int eor = (sock->type == SOCK_DGRAM) ?
905 !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
910 /* Per tcp_sendmsg this should be in poll */
911 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
917 /* Previously opened message */
919 skb = kcm_tx_msg(head)->last_skb;
923 /* Call the sk_stream functions to manage the sndbuf mem. */
924 if (!sk_stream_memory_free(sk)) {
926 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
927 err = sk_stream_wait_memory(sk, &timeo);
932 /* New message, alloc head skb */
933 head = alloc_skb(0, sk->sk_allocation);
936 err = sk_stream_wait_memory(sk, &timeo);
940 head = alloc_skb(0, sk->sk_allocation);
945 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
946 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
948 skb->ip_summed = CHECKSUM_UNNECESSARY;
951 while (msg_data_left(msg)) {
953 int i = skb_shinfo(skb)->nr_frags;
954 struct page_frag *pfrag = sk_page_frag(sk);
956 if (!sk_page_frag_refill(sk, pfrag))
957 goto wait_for_memory;
959 if (!skb_can_coalesce(skb, i, pfrag->page,
961 if (i == MAX_SKB_FRAGS) {
962 struct sk_buff *tskb;
964 tskb = alloc_skb(0, sk->sk_allocation);
966 goto wait_for_memory;
969 skb_shinfo(head)->frag_list = tskb;
974 skb->ip_summed = CHECKSUM_UNNECESSARY;
980 copy = min_t(int, msg_data_left(msg),
981 pfrag->size - pfrag->offset);
983 if (!sk_wmem_schedule(sk, copy))
984 goto wait_for_memory;
986 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
993 /* Update the skb. */
995 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
997 skb_fill_page_desc(skb, i, pfrag->page,
998 pfrag->offset, copy);
999 get_page(pfrag->page);
1002 pfrag->offset += copy;
1006 head->data_len += copy;
1013 err = sk_stream_wait_memory(sk, &timeo);
1019 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1021 /* Message complete, queue it on send buffer */
1022 __skb_queue_tail(&sk->sk_write_queue, head);
1023 kcm->seq_skb = NULL;
1024 KCM_STATS_INCR(kcm->stats.tx_msgs);
1026 if (msg->msg_flags & MSG_BATCH) {
1027 kcm->tx_wait_more = true;
1028 } else if (kcm->tx_wait_more || not_busy) {
1029 err = kcm_write_msgs(kcm);
1031 /* We got a hard error in write_msgs but have
1032 * already queued this message. Report an error
1033 * in the socket, but don't affect return value
1036 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1037 report_csk_error(&kcm->sk, -err);
1041 /* Message not complete, save state */
1043 kcm->seq_skb = head;
1044 kcm_tx_msg(head)->last_skb = skb;
1047 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1055 if (copied && sock->type == SOCK_SEQPACKET) {
1056 /* Wrote some bytes before encountering an
1057 * error, return partial success.
1059 goto partial_message;
1062 if (head != kcm->seq_skb)
1065 err = sk_stream_error(sk, msg->msg_flags, err);
1067 /* make sure we wake any epoll edge trigger waiter */
1068 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1069 sk->sk_write_space(sk);
1075 static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1076 long timeo, int *err)
1078 struct sk_buff *skb;
1080 while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1082 *err = sock_error(sk);
1086 if (sock_flag(sk, SOCK_DONE))
1089 if ((flags & MSG_DONTWAIT) || !timeo) {
1094 sk_wait_data(sk, &timeo, NULL);
1096 /* Handle signals */
1097 if (signal_pending(current)) {
1098 *err = sock_intr_errno(timeo);
1106 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1107 size_t len, int flags)
1109 struct sock *sk = sock->sk;
1110 struct kcm_sock *kcm = kcm_sk(sk);
1113 struct strp_rx_msg *rxm;
1115 struct sk_buff *skb;
1117 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1121 skb = kcm_wait_data(sk, flags, timeo, &err);
1125 /* Okay, have a message on the receive queue */
1127 rxm = strp_rx_msg(skb);
1129 if (len > rxm->full_len)
1130 len = rxm->full_len;
1132 err = skb_copy_datagram_msg(skb, rxm->offset, msg, len);
1137 if (likely(!(flags & MSG_PEEK))) {
1138 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1139 if (copied < rxm->full_len) {
1140 if (sock->type == SOCK_DGRAM) {
1141 /* Truncated message */
1142 msg->msg_flags |= MSG_TRUNC;
1145 rxm->offset += copied;
1146 rxm->full_len -= copied;
1149 /* Finished with message */
1150 msg->msg_flags |= MSG_EOR;
1151 KCM_STATS_INCR(kcm->stats.rx_msgs);
1152 skb_unlink(skb, &sk->sk_receive_queue);
1160 return copied ? : err;
1163 static ssize_t kcm_sock_splice(struct sock *sk,
1164 struct pipe_inode_info *pipe,
1165 struct splice_pipe_desc *spd)
1170 ret = splice_to_pipe(pipe, spd);
1176 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1177 struct pipe_inode_info *pipe, size_t len,
1180 struct sock *sk = sock->sk;
1181 struct kcm_sock *kcm = kcm_sk(sk);
1183 struct strp_rx_msg *rxm;
1186 struct sk_buff *skb;
1188 /* Only support splice for SOCKSEQPACKET */
1190 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1194 skb = kcm_wait_data(sk, flags, timeo, &err);
1198 /* Okay, have a message on the receive queue */
1200 rxm = strp_rx_msg(skb);
1202 if (len > rxm->full_len)
1203 len = rxm->full_len;
1205 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, len, flags,
1212 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1214 rxm->offset += copied;
1215 rxm->full_len -= copied;
1217 /* We have no way to return MSG_EOR. If all the bytes have been
1218 * read we still leave the message in the receive socket buffer.
1219 * A subsequent recvmsg needs to be done to return MSG_EOR and
1220 * finish reading the message.
1233 /* kcm sock lock held */
1234 static void kcm_recv_disable(struct kcm_sock *kcm)
1236 struct kcm_mux *mux = kcm->mux;
1238 if (kcm->rx_disabled)
1241 spin_lock_bh(&mux->rx_lock);
1243 kcm->rx_disabled = 1;
1245 /* If a psock is reserved we'll do cleanup in unreserve */
1246 if (!kcm->rx_psock) {
1248 list_del(&kcm->wait_rx_list);
1249 kcm->rx_wait = false;
1252 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1255 spin_unlock_bh(&mux->rx_lock);
1258 /* kcm sock lock held */
1259 static void kcm_recv_enable(struct kcm_sock *kcm)
1261 struct kcm_mux *mux = kcm->mux;
1263 if (!kcm->rx_disabled)
1266 spin_lock_bh(&mux->rx_lock);
1268 kcm->rx_disabled = 0;
1271 spin_unlock_bh(&mux->rx_lock);
1274 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1275 char __user *optval, unsigned int optlen)
1277 struct kcm_sock *kcm = kcm_sk(sock->sk);
1281 if (level != SOL_KCM)
1282 return -ENOPROTOOPT;
1284 if (optlen < sizeof(int))
1287 if (get_user(val, (int __user *)optval))
1290 valbool = val ? 1 : 0;
1293 case KCM_RECV_DISABLE:
1294 lock_sock(&kcm->sk);
1296 kcm_recv_disable(kcm);
1298 kcm_recv_enable(kcm);
1299 release_sock(&kcm->sk);
1308 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1309 char __user *optval, int __user *optlen)
1311 struct kcm_sock *kcm = kcm_sk(sock->sk);
1314 if (level != SOL_KCM)
1315 return -ENOPROTOOPT;
1317 if (get_user(len, optlen))
1320 len = min_t(unsigned int, len, sizeof(int));
1325 case KCM_RECV_DISABLE:
1326 val = kcm->rx_disabled;
1329 return -ENOPROTOOPT;
1332 if (put_user(len, optlen))
1334 if (copy_to_user(optval, &val, len))
1339 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1341 struct kcm_sock *tkcm;
1342 struct list_head *head;
1345 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1346 * we set sk_state, otherwise epoll_wait always returns right away with
1349 kcm->sk.sk_state = TCP_ESTABLISHED;
1351 /* Add to mux's kcm sockets list */
1353 spin_lock_bh(&mux->lock);
1355 head = &mux->kcm_socks;
1356 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1357 if (tkcm->index != index)
1359 head = &tkcm->kcm_sock_list;
1363 list_add(&kcm->kcm_sock_list, head);
1366 mux->kcm_socks_cnt++;
1367 spin_unlock_bh(&mux->lock);
1369 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1371 spin_lock_bh(&mux->rx_lock);
1373 spin_unlock_bh(&mux->rx_lock);
1376 static int kcm_attach(struct socket *sock, struct socket *csock,
1377 struct bpf_prog *prog)
1379 struct kcm_sock *kcm = kcm_sk(sock->sk);
1380 struct kcm_mux *mux = kcm->mux;
1382 struct kcm_psock *psock = NULL, *tpsock;
1383 struct list_head *head;
1385 struct strp_callbacks cb;
1387 if (csock->ops->family != PF_INET &&
1388 csock->ops->family != PF_INET6)
1395 /* Only support TCP for now */
1396 if (csk->sk_protocol != IPPROTO_TCP)
1399 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1405 psock->bpf_prog = prog;
1407 cb.rcv_msg = kcm_rcv_strparser;
1408 cb.abort_parser = NULL;
1409 cb.parse_msg = kcm_parse_func_strparser;
1410 cb.read_sock_done = kcm_read_sock_done;
1412 strp_init(&psock->strp, csk, &cb);
1416 write_lock_bh(&csk->sk_callback_lock);
1417 psock->save_data_ready = csk->sk_data_ready;
1418 psock->save_write_space = csk->sk_write_space;
1419 psock->save_state_change = csk->sk_state_change;
1420 csk->sk_user_data = psock;
1421 csk->sk_data_ready = psock_tcp_data_ready;
1422 csk->sk_write_space = psock_tcp_write_space;
1423 csk->sk_state_change = psock_tcp_state_change;
1424 write_unlock_bh(&csk->sk_callback_lock);
1426 /* Finished initialization, now add the psock to the MUX. */
1427 spin_lock_bh(&mux->lock);
1428 head = &mux->psocks;
1429 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1430 if (tpsock->index != index)
1432 head = &tpsock->psock_list;
1436 list_add(&psock->psock_list, head);
1437 psock->index = index;
1439 KCM_STATS_INCR(mux->stats.psock_attach);
1441 psock_now_avail(psock);
1442 spin_unlock_bh(&mux->lock);
1444 /* Schedule RX work in case there are already bytes queued */
1445 strp_check_rcv(&psock->strp);
1450 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1452 struct socket *csock;
1453 struct bpf_prog *prog;
1456 csock = sockfd_lookup(info->fd, &err);
1460 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1462 err = PTR_ERR(prog);
1466 err = kcm_attach(sock, csock, prog);
1472 /* Keep reference on file also */
1480 /* Lower socket lock held */
1481 static void kcm_unattach(struct kcm_psock *psock)
1483 struct sock *csk = psock->sk;
1484 struct kcm_mux *mux = psock->mux;
1486 /* Stop getting callbacks from TCP socket. After this there should
1487 * be no way to reserve a kcm for this psock.
1489 write_lock_bh(&csk->sk_callback_lock);
1490 csk->sk_user_data = NULL;
1491 csk->sk_data_ready = psock->save_data_ready;
1492 csk->sk_write_space = psock->save_write_space;
1493 csk->sk_state_change = psock->save_state_change;
1494 strp_stop(&psock->strp);
1496 if (WARN_ON(psock->rx_kcm)) {
1497 write_unlock_bh(&csk->sk_callback_lock);
1501 spin_lock_bh(&mux->rx_lock);
1503 /* Stop receiver activities. After this point psock should not be
1504 * able to get onto ready list either through callbacks or work.
1506 if (psock->ready_rx_msg) {
1507 list_del(&psock->psock_ready_list);
1508 kfree_skb(psock->ready_rx_msg);
1509 psock->ready_rx_msg = NULL;
1510 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1513 spin_unlock_bh(&mux->rx_lock);
1515 write_unlock_bh(&csk->sk_callback_lock);
1517 strp_done(&psock->strp);
1519 bpf_prog_put(psock->bpf_prog);
1521 spin_lock_bh(&mux->lock);
1523 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1524 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1526 KCM_STATS_INCR(mux->stats.psock_unattach);
1528 if (psock->tx_kcm) {
1529 /* psock was reserved. Just mark it finished and we will clean
1530 * up in the kcm paths, we need kcm lock which can not be
1533 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1534 spin_unlock_bh(&mux->lock);
1536 /* We are unattaching a socket that is reserved. Abort the
1537 * socket since we may be out of sync in sending on it. We need
1538 * to do this without the mux lock.
1540 kcm_abort_tx_psock(psock, EPIPE, false);
1542 spin_lock_bh(&mux->lock);
1543 if (!psock->tx_kcm) {
1544 /* psock now unreserved in window mux was unlocked */
1549 /* Commit done before queuing work to process it */
1552 /* Queue tx work to make sure psock->done is handled */
1553 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1554 spin_unlock_bh(&mux->lock);
1557 if (!psock->tx_stopped)
1558 list_del(&psock->psock_avail_list);
1559 list_del(&psock->psock_list);
1561 spin_unlock_bh(&mux->lock);
1564 fput(csk->sk_socket->file);
1565 kmem_cache_free(kcm_psockp, psock);
1569 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1571 struct kcm_sock *kcm = kcm_sk(sock->sk);
1572 struct kcm_mux *mux = kcm->mux;
1573 struct kcm_psock *psock;
1574 struct socket *csock;
1578 csock = sockfd_lookup(info->fd, &err);
1590 spin_lock_bh(&mux->lock);
1592 list_for_each_entry(psock, &mux->psocks, psock_list) {
1593 if (psock->sk != csk)
1596 /* Found the matching psock */
1598 if (psock->unattaching || WARN_ON(psock->done)) {
1603 psock->unattaching = 1;
1605 spin_unlock_bh(&mux->lock);
1607 /* Lower socket lock should already be held */
1608 kcm_unattach(psock);
1614 spin_unlock_bh(&mux->lock);
1621 static struct proto kcm_proto = {
1623 .owner = THIS_MODULE,
1624 .obj_size = sizeof(struct kcm_sock),
1627 /* Clone a kcm socket. */
1628 static int kcm_clone(struct socket *osock, struct kcm_clone *info,
1629 struct socket **newsockp)
1631 struct socket *newsock;
1633 struct file *newfile;
1637 newsock = sock_alloc();
1641 newsock->type = osock->type;
1642 newsock->ops = osock->ops;
1644 __module_get(newsock->ops->owner);
1646 newfd = get_unused_fd_flags(0);
1647 if (unlikely(newfd < 0)) {
1652 newfile = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1653 if (unlikely(IS_ERR(newfile))) {
1654 err = PTR_ERR(newfile);
1655 goto out_sock_alloc_fail;
1658 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1662 goto out_sk_alloc_fail;
1665 sock_init_data(newsock, newsk);
1666 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1668 fd_install(newfd, newfile);
1669 *newsockp = newsock;
1676 out_sock_alloc_fail:
1677 put_unused_fd(newfd);
1679 sock_release(newsock);
1684 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1689 case SIOCKCMATTACH: {
1690 struct kcm_attach info;
1692 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1695 err = kcm_attach_ioctl(sock, &info);
1699 case SIOCKCMUNATTACH: {
1700 struct kcm_unattach info;
1702 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1705 err = kcm_unattach_ioctl(sock, &info);
1709 case SIOCKCMCLONE: {
1710 struct kcm_clone info;
1711 struct socket *newsock = NULL;
1713 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1716 err = kcm_clone(sock, &info, &newsock);
1719 if (copy_to_user((void __user *)arg, &info,
1722 sock_release(newsock);
1736 static void free_mux(struct rcu_head *rcu)
1738 struct kcm_mux *mux = container_of(rcu,
1739 struct kcm_mux, rcu);
1741 kmem_cache_free(kcm_muxp, mux);
1744 static void release_mux(struct kcm_mux *mux)
1746 struct kcm_net *knet = mux->knet;
1747 struct kcm_psock *psock, *tmp_psock;
1749 /* Release psocks */
1750 list_for_each_entry_safe(psock, tmp_psock,
1751 &mux->psocks, psock_list) {
1752 if (!WARN_ON(psock->unattaching)) {
1753 lock_sock(psock->strp.sk);
1754 kcm_unattach(psock);
1755 release_sock(psock->strp.sk);
1759 if (WARN_ON(mux->psocks_cnt))
1762 __skb_queue_purge(&mux->rx_hold_queue);
1764 mutex_lock(&knet->mutex);
1765 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1766 aggregate_psock_stats(&mux->aggregate_psock_stats,
1767 &knet->aggregate_psock_stats);
1768 aggregate_strp_stats(&mux->aggregate_strp_stats,
1769 &knet->aggregate_strp_stats);
1770 list_del_rcu(&mux->kcm_mux_list);
1772 mutex_unlock(&knet->mutex);
1774 call_rcu(&mux->rcu, free_mux);
1777 static void kcm_done(struct kcm_sock *kcm)
1779 struct kcm_mux *mux = kcm->mux;
1780 struct sock *sk = &kcm->sk;
1783 spin_lock_bh(&mux->rx_lock);
1784 if (kcm->rx_psock) {
1785 /* Cleanup in unreserve_rx_kcm */
1787 kcm->rx_disabled = 1;
1789 spin_unlock_bh(&mux->rx_lock);
1794 list_del(&kcm->wait_rx_list);
1795 kcm->rx_wait = false;
1797 /* Move any pending receive messages to other kcm sockets */
1798 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1800 spin_unlock_bh(&mux->rx_lock);
1802 if (WARN_ON(sk_rmem_alloc_get(sk)))
1805 /* Detach from MUX */
1806 spin_lock_bh(&mux->lock);
1808 list_del(&kcm->kcm_sock_list);
1809 mux->kcm_socks_cnt--;
1810 socks_cnt = mux->kcm_socks_cnt;
1812 spin_unlock_bh(&mux->lock);
1815 /* We are done with the mux now. */
1819 WARN_ON(kcm->rx_wait);
1824 /* Called by kcm_release to close a KCM socket.
1825 * If this is the last KCM socket on the MUX, destroy the MUX.
1827 static int kcm_release(struct socket *sock)
1829 struct sock *sk = sock->sk;
1830 struct kcm_sock *kcm;
1831 struct kcm_mux *mux;
1832 struct kcm_psock *psock;
1841 kfree_skb(kcm->seq_skb);
1844 /* Purge queue under lock to avoid race condition with tx_work trying
1845 * to act when queue is nonempty. If tx_work runs after this point
1846 * it will just return.
1848 __skb_queue_purge(&sk->sk_write_queue);
1850 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1851 * get a writespace callback. This prevents further work being queued
1852 * from the callback (unbinding the psock occurs after canceling work.
1854 kcm->tx_stopped = 1;
1858 spin_lock_bh(&mux->lock);
1860 /* Take of tx_wait list, after this point there should be no way
1861 * that a psock will be assigned to this kcm.
1863 list_del(&kcm->wait_psock_list);
1864 kcm->tx_wait = false;
1866 spin_unlock_bh(&mux->lock);
1868 /* Cancel work. After this point there should be no outside references
1869 * to the kcm socket.
1871 cancel_work_sync(&kcm->tx_work);
1874 psock = kcm->tx_psock;
1876 /* A psock was reserved, so we need to kill it since it
1877 * may already have some bytes queued from a message. We
1878 * need to do this after removing kcm from tx_wait list.
1880 kcm_abort_tx_psock(psock, EPIPE, false);
1881 unreserve_psock(kcm);
1885 WARN_ON(kcm->tx_wait);
1886 WARN_ON(kcm->tx_psock);
1895 static const struct proto_ops kcm_dgram_ops = {
1897 .owner = THIS_MODULE,
1898 .release = kcm_release,
1899 .bind = sock_no_bind,
1900 .connect = sock_no_connect,
1901 .socketpair = sock_no_socketpair,
1902 .accept = sock_no_accept,
1903 .getname = sock_no_getname,
1904 .poll = datagram_poll,
1906 .listen = sock_no_listen,
1907 .shutdown = sock_no_shutdown,
1908 .setsockopt = kcm_setsockopt,
1909 .getsockopt = kcm_getsockopt,
1910 .sendmsg = kcm_sendmsg,
1911 .recvmsg = kcm_recvmsg,
1912 .mmap = sock_no_mmap,
1913 .sendpage = kcm_sendpage,
1916 static const struct proto_ops kcm_seqpacket_ops = {
1918 .owner = THIS_MODULE,
1919 .release = kcm_release,
1920 .bind = sock_no_bind,
1921 .connect = sock_no_connect,
1922 .socketpair = sock_no_socketpair,
1923 .accept = sock_no_accept,
1924 .getname = sock_no_getname,
1925 .poll = datagram_poll,
1927 .listen = sock_no_listen,
1928 .shutdown = sock_no_shutdown,
1929 .setsockopt = kcm_setsockopt,
1930 .getsockopt = kcm_getsockopt,
1931 .sendmsg = kcm_sendmsg,
1932 .recvmsg = kcm_recvmsg,
1933 .mmap = sock_no_mmap,
1934 .sendpage = kcm_sendpage,
1935 .splice_read = kcm_splice_read,
1938 /* Create proto operation for kcm sockets */
1939 static int kcm_create(struct net *net, struct socket *sock,
1940 int protocol, int kern)
1942 struct kcm_net *knet = net_generic(net, kcm_net_id);
1944 struct kcm_mux *mux;
1946 switch (sock->type) {
1948 sock->ops = &kcm_dgram_ops;
1950 case SOCK_SEQPACKET:
1951 sock->ops = &kcm_seqpacket_ops;
1954 return -ESOCKTNOSUPPORT;
1957 if (protocol != KCMPROTO_CONNECTED)
1958 return -EPROTONOSUPPORT;
1960 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1964 /* Allocate a kcm mux, shared between KCM sockets */
1965 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1971 spin_lock_init(&mux->lock);
1972 spin_lock_init(&mux->rx_lock);
1973 INIT_LIST_HEAD(&mux->kcm_socks);
1974 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1975 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1977 INIT_LIST_HEAD(&mux->psocks);
1978 INIT_LIST_HEAD(&mux->psocks_ready);
1979 INIT_LIST_HEAD(&mux->psocks_avail);
1983 /* Add new MUX to list */
1984 mutex_lock(&knet->mutex);
1985 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1987 mutex_unlock(&knet->mutex);
1989 skb_queue_head_init(&mux->rx_hold_queue);
1991 /* Init KCM socket */
1992 sock_init_data(sock, sk);
1993 init_kcm_sock(kcm_sk(sk), mux);
1998 static struct net_proto_family kcm_family_ops = {
2000 .create = kcm_create,
2001 .owner = THIS_MODULE,
2004 static __net_init int kcm_init_net(struct net *net)
2006 struct kcm_net *knet = net_generic(net, kcm_net_id);
2008 INIT_LIST_HEAD_RCU(&knet->mux_list);
2009 mutex_init(&knet->mutex);
2014 static __net_exit void kcm_exit_net(struct net *net)
2016 struct kcm_net *knet = net_generic(net, kcm_net_id);
2018 /* All KCM sockets should be closed at this point, which should mean
2019 * that all multiplexors and psocks have been destroyed.
2021 WARN_ON(!list_empty(&knet->mux_list));
2024 static struct pernet_operations kcm_net_ops = {
2025 .init = kcm_init_net,
2026 .exit = kcm_exit_net,
2028 .size = sizeof(struct kcm_net),
2031 static int __init kcm_init(void)
2035 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2036 sizeof(struct kcm_mux), 0,
2037 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2041 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2042 sizeof(struct kcm_psock), 0,
2043 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2047 kcm_wq = create_singlethread_workqueue("kkcmd");
2051 err = proto_register(&kcm_proto, 1);
2055 err = sock_register(&kcm_family_ops);
2057 goto sock_register_fail;
2059 err = register_pernet_device(&kcm_net_ops);
2063 err = kcm_proc_init();
2065 goto proc_init_fail;
2070 unregister_pernet_device(&kcm_net_ops);
2073 sock_unregister(PF_KCM);
2076 proto_unregister(&kcm_proto);
2079 kmem_cache_destroy(kcm_muxp);
2080 kmem_cache_destroy(kcm_psockp);
2083 destroy_workqueue(kcm_wq);
2088 static void __exit kcm_exit(void)
2091 unregister_pernet_device(&kcm_net_ops);
2092 sock_unregister(PF_KCM);
2093 proto_unregister(&kcm_proto);
2094 destroy_workqueue(kcm_wq);
2096 kmem_cache_destroy(kcm_muxp);
2097 kmem_cache_destroy(kcm_psockp);
2100 module_init(kcm_init);
2101 module_exit(kcm_exit);
2103 MODULE_LICENSE("GPL");
2104 MODULE_ALIAS_NETPROTO(PF_KCM);