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>
26 #include <linux/syscalls.h>
27 #include <linux/sched/signal.h>
30 #include <net/netns/generic.h>
32 #include <uapi/linux/kcm.h>
34 unsigned int kcm_net_id;
36 static struct kmem_cache *kcm_psockp __read_mostly;
37 static struct kmem_cache *kcm_muxp __read_mostly;
38 static struct workqueue_struct *kcm_wq;
40 static inline struct kcm_sock *kcm_sk(const struct sock *sk)
42 return (struct kcm_sock *)sk;
45 static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
47 return (struct kcm_tx_msg *)skb->cb;
50 static void report_csk_error(struct sock *csk, int err)
53 csk->sk_error_report(csk);
56 static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
59 struct sock *csk = psock->sk;
60 struct kcm_mux *mux = psock->mux;
62 /* Unrecoverable error in transmit */
64 spin_lock_bh(&mux->lock);
66 if (psock->tx_stopped) {
67 spin_unlock_bh(&mux->lock);
71 psock->tx_stopped = 1;
72 KCM_STATS_INCR(psock->stats.tx_aborts);
75 /* Take off psocks_avail list */
76 list_del(&psock->psock_avail_list);
77 } else if (wakeup_kcm) {
78 /* In this case psock is being aborted while outside of
79 * write_msgs and psock is reserved. Schedule tx_work
80 * to handle the failure there. Need to commit tx_stopped
81 * before queuing work.
85 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
88 spin_unlock_bh(&mux->lock);
90 /* Report error on lower socket */
91 report_csk_error(csk, err);
94 /* RX mux lock held. */
95 static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
96 struct kcm_psock *psock)
98 STRP_STATS_ADD(mux->stats.rx_bytes,
99 psock->strp.stats.bytes -
100 psock->saved_rx_bytes);
101 mux->stats.rx_msgs +=
102 psock->strp.stats.msgs - psock->saved_rx_msgs;
103 psock->saved_rx_msgs = psock->strp.stats.msgs;
104 psock->saved_rx_bytes = psock->strp.stats.bytes;
107 static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
108 struct kcm_psock *psock)
110 KCM_STATS_ADD(mux->stats.tx_bytes,
111 psock->stats.tx_bytes - psock->saved_tx_bytes);
112 mux->stats.tx_msgs +=
113 psock->stats.tx_msgs - psock->saved_tx_msgs;
114 psock->saved_tx_msgs = psock->stats.tx_msgs;
115 psock->saved_tx_bytes = psock->stats.tx_bytes;
118 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
120 /* KCM is ready to receive messages on its queue-- either the KCM is new or
121 * has become unblocked after being blocked on full socket buffer. Queue any
122 * pending ready messages on a psock. RX mux lock held.
124 static void kcm_rcv_ready(struct kcm_sock *kcm)
126 struct kcm_mux *mux = kcm->mux;
127 struct kcm_psock *psock;
130 if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
133 while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) {
134 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
135 /* Assuming buffer limit has been reached */
136 skb_queue_head(&mux->rx_hold_queue, skb);
137 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
142 while (!list_empty(&mux->psocks_ready)) {
143 psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
146 if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) {
147 /* Assuming buffer limit has been reached */
148 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
152 /* Consumed the ready message on the psock. Schedule rx_work to
155 list_del(&psock->psock_ready_list);
156 psock->ready_rx_msg = NULL;
157 /* Commit clearing of ready_rx_msg for queuing work */
160 strp_unpause(&psock->strp);
161 strp_check_rcv(&psock->strp);
164 /* Buffer limit is okay now, add to ready list */
165 list_add_tail(&kcm->wait_rx_list,
166 &kcm->mux->kcm_rx_waiters);
167 /* paired with lockless reads in kcm_rfree() */
168 WRITE_ONCE(kcm->rx_wait, true);
171 static void kcm_rfree(struct sk_buff *skb)
173 struct sock *sk = skb->sk;
174 struct kcm_sock *kcm = kcm_sk(sk);
175 struct kcm_mux *mux = kcm->mux;
176 unsigned int len = skb->truesize;
178 sk_mem_uncharge(sk, len);
179 atomic_sub(len, &sk->sk_rmem_alloc);
181 /* For reading rx_wait and rx_psock without holding lock */
182 smp_mb__after_atomic();
184 if (!READ_ONCE(kcm->rx_wait) && !READ_ONCE(kcm->rx_psock) &&
185 sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
186 spin_lock_bh(&mux->rx_lock);
188 spin_unlock_bh(&mux->rx_lock);
192 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
194 struct sk_buff_head *list = &sk->sk_receive_queue;
196 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
199 if (!sk_rmem_schedule(sk, skb, skb->truesize))
206 skb->destructor = kcm_rfree;
207 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
208 sk_mem_charge(sk, skb->truesize);
210 skb_queue_tail(list, skb);
212 if (!sock_flag(sk, SOCK_DEAD))
213 sk->sk_data_ready(sk);
218 /* Requeue received messages for a kcm socket to other kcm sockets. This is
219 * called with a kcm socket is receive disabled.
222 static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
225 struct kcm_sock *kcm;
227 while ((skb = skb_dequeue(head))) {
228 /* Reset destructor to avoid calling kcm_rcv_ready */
229 skb->destructor = sock_rfree;
232 if (list_empty(&mux->kcm_rx_waiters)) {
233 skb_queue_tail(&mux->rx_hold_queue, skb);
237 kcm = list_first_entry(&mux->kcm_rx_waiters,
238 struct kcm_sock, wait_rx_list);
240 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
241 /* Should mean socket buffer full */
242 list_del(&kcm->wait_rx_list);
243 /* paired with lockless reads in kcm_rfree() */
244 WRITE_ONCE(kcm->rx_wait, false);
246 /* Commit rx_wait to read in kcm_free */
254 /* Lower sock lock held */
255 static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
256 struct sk_buff *head)
258 struct kcm_mux *mux = psock->mux;
259 struct kcm_sock *kcm;
261 WARN_ON(psock->ready_rx_msg);
264 return psock->rx_kcm;
266 spin_lock_bh(&mux->rx_lock);
269 spin_unlock_bh(&mux->rx_lock);
270 return psock->rx_kcm;
273 kcm_update_rx_mux_stats(mux, psock);
275 if (list_empty(&mux->kcm_rx_waiters)) {
276 psock->ready_rx_msg = head;
277 strp_pause(&psock->strp);
278 list_add_tail(&psock->psock_ready_list,
280 spin_unlock_bh(&mux->rx_lock);
284 kcm = list_first_entry(&mux->kcm_rx_waiters,
285 struct kcm_sock, wait_rx_list);
286 list_del(&kcm->wait_rx_list);
287 /* paired with lockless reads in kcm_rfree() */
288 WRITE_ONCE(kcm->rx_wait, false);
291 /* paired with lockless reads in kcm_rfree() */
292 WRITE_ONCE(kcm->rx_psock, psock);
294 spin_unlock_bh(&mux->rx_lock);
299 static void kcm_done(struct kcm_sock *kcm);
301 static void kcm_done_work(struct work_struct *w)
303 kcm_done(container_of(w, struct kcm_sock, done_work));
306 /* Lower sock held */
307 static void unreserve_rx_kcm(struct kcm_psock *psock,
310 struct kcm_sock *kcm = psock->rx_kcm;
311 struct kcm_mux *mux = psock->mux;
316 spin_lock_bh(&mux->rx_lock);
318 psock->rx_kcm = NULL;
319 /* paired with lockless reads in kcm_rfree() */
320 WRITE_ONCE(kcm->rx_psock, NULL);
322 /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
327 if (unlikely(kcm->done)) {
328 spin_unlock_bh(&mux->rx_lock);
330 /* Need to run kcm_done in a task since we need to qcquire
331 * callback locks which may already be held here.
333 INIT_WORK(&kcm->done_work, kcm_done_work);
334 schedule_work(&kcm->done_work);
338 if (unlikely(kcm->rx_disabled)) {
339 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
340 } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
341 /* Check for degenerative race with rx_wait that all
342 * data was dequeued (accounted for in kcm_rfree).
346 spin_unlock_bh(&mux->rx_lock);
349 /* Lower sock lock held */
350 static void psock_data_ready(struct sock *sk)
352 struct kcm_psock *psock;
354 read_lock_bh(&sk->sk_callback_lock);
356 psock = (struct kcm_psock *)sk->sk_user_data;
358 strp_data_ready(&psock->strp);
360 read_unlock_bh(&sk->sk_callback_lock);
363 /* Called with lower sock held */
364 static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
366 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
367 struct kcm_sock *kcm;
370 kcm = reserve_rx_kcm(psock, skb);
372 /* Unable to reserve a KCM, message is held in psock and strp
378 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
379 /* Should mean socket buffer full */
380 unreserve_rx_kcm(psock, false);
385 static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
387 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
388 struct bpf_prog *prog = psock->bpf_prog;
390 return (*prog->bpf_func)(skb, prog->insnsi);
393 static int kcm_read_sock_done(struct strparser *strp, int err)
395 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
397 unreserve_rx_kcm(psock, true);
402 static void psock_state_change(struct sock *sk)
404 /* TCP only does a POLLIN for a half close. Do a POLLHUP here
405 * since application will normally not poll with POLLIN
406 * on the TCP sockets.
409 report_csk_error(sk, EPIPE);
412 static void psock_write_space(struct sock *sk)
414 struct kcm_psock *psock;
416 struct kcm_sock *kcm;
418 read_lock_bh(&sk->sk_callback_lock);
420 psock = (struct kcm_psock *)sk->sk_user_data;
421 if (unlikely(!psock))
425 spin_lock_bh(&mux->lock);
427 /* Check if the socket is reserved so someone is waiting for sending. */
429 if (kcm && !unlikely(kcm->tx_stopped))
430 queue_work(kcm_wq, &kcm->tx_work);
432 spin_unlock_bh(&mux->lock);
434 read_unlock_bh(&sk->sk_callback_lock);
437 static void unreserve_psock(struct kcm_sock *kcm);
439 /* kcm sock is locked. */
440 static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
442 struct kcm_mux *mux = kcm->mux;
443 struct kcm_psock *psock;
445 psock = kcm->tx_psock;
447 smp_rmb(); /* Must read tx_psock before tx_wait */
450 WARN_ON(kcm->tx_wait);
451 if (unlikely(psock->tx_stopped))
452 unreserve_psock(kcm);
454 return kcm->tx_psock;
457 spin_lock_bh(&mux->lock);
459 /* Check again under lock to see if psock was reserved for this
460 * psock via psock_unreserve.
462 psock = kcm->tx_psock;
463 if (unlikely(psock)) {
464 WARN_ON(kcm->tx_wait);
465 spin_unlock_bh(&mux->lock);
466 return kcm->tx_psock;
469 if (!list_empty(&mux->psocks_avail)) {
470 psock = list_first_entry(&mux->psocks_avail,
473 list_del(&psock->psock_avail_list);
475 list_del(&kcm->wait_psock_list);
476 kcm->tx_wait = false;
478 kcm->tx_psock = psock;
480 KCM_STATS_INCR(psock->stats.reserved);
481 } else if (!kcm->tx_wait) {
482 list_add_tail(&kcm->wait_psock_list,
483 &mux->kcm_tx_waiters);
487 spin_unlock_bh(&mux->lock);
493 static void psock_now_avail(struct kcm_psock *psock)
495 struct kcm_mux *mux = psock->mux;
496 struct kcm_sock *kcm;
498 if (list_empty(&mux->kcm_tx_waiters)) {
499 list_add_tail(&psock->psock_avail_list,
502 kcm = list_first_entry(&mux->kcm_tx_waiters,
505 list_del(&kcm->wait_psock_list);
506 kcm->tx_wait = false;
509 /* Commit before changing tx_psock since that is read in
510 * reserve_psock before queuing work.
514 kcm->tx_psock = psock;
515 KCM_STATS_INCR(psock->stats.reserved);
516 queue_work(kcm_wq, &kcm->tx_work);
520 /* kcm sock is locked. */
521 static void unreserve_psock(struct kcm_sock *kcm)
523 struct kcm_psock *psock;
524 struct kcm_mux *mux = kcm->mux;
526 spin_lock_bh(&mux->lock);
528 psock = kcm->tx_psock;
530 if (WARN_ON(!psock)) {
531 spin_unlock_bh(&mux->lock);
535 smp_rmb(); /* Read tx_psock before tx_wait */
537 kcm_update_tx_mux_stats(mux, psock);
539 WARN_ON(kcm->tx_wait);
541 kcm->tx_psock = NULL;
542 psock->tx_kcm = NULL;
543 KCM_STATS_INCR(psock->stats.unreserved);
545 if (unlikely(psock->tx_stopped)) {
548 list_del(&psock->psock_list);
551 fput(psock->sk->sk_socket->file);
552 kmem_cache_free(kcm_psockp, psock);
555 /* Don't put back on available list */
557 spin_unlock_bh(&mux->lock);
562 psock_now_avail(psock);
564 spin_unlock_bh(&mux->lock);
567 static void kcm_report_tx_retry(struct kcm_sock *kcm)
569 struct kcm_mux *mux = kcm->mux;
571 spin_lock_bh(&mux->lock);
572 KCM_STATS_INCR(mux->stats.tx_retries);
573 spin_unlock_bh(&mux->lock);
576 /* Write any messages ready on the kcm socket. Called with kcm sock lock
577 * held. Return bytes actually sent or error.
579 static int kcm_write_msgs(struct kcm_sock *kcm)
581 struct sock *sk = &kcm->sk;
582 struct kcm_psock *psock;
583 struct sk_buff *skb, *head;
584 struct kcm_tx_msg *txm;
585 unsigned short fragidx, frag_offset;
586 unsigned int sent, total_sent = 0;
589 kcm->tx_wait_more = false;
590 psock = kcm->tx_psock;
591 if (unlikely(psock && psock->tx_stopped)) {
592 /* A reserved psock was aborted asynchronously. Unreserve
593 * it and we'll retry the message.
595 unreserve_psock(kcm);
596 kcm_report_tx_retry(kcm);
597 if (skb_queue_empty(&sk->sk_write_queue))
600 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
602 } else if (skb_queue_empty(&sk->sk_write_queue)) {
606 head = skb_peek(&sk->sk_write_queue);
607 txm = kcm_tx_msg(head);
610 /* Send of first skbuff in queue already in progress */
611 if (WARN_ON(!psock)) {
616 frag_offset = txm->frag_offset;
617 fragidx = txm->fragidx;
624 psock = reserve_psock(kcm);
630 txm = kcm_tx_msg(head);
634 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
639 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
645 frag = &skb_shinfo(skb)->frags[fragidx];
646 if (WARN_ON(!frag->size)) {
651 ret = kernel_sendpage(psock->sk->sk_socket,
653 frag->page_offset + frag_offset,
654 frag->size - frag_offset,
657 if (ret == -EAGAIN) {
658 /* Save state to try again when there's
659 * write space on the socket
662 txm->frag_offset = frag_offset;
663 txm->fragidx = fragidx;
670 /* Hard failure in sending message, abort this
671 * psock since it has lost framing
672 * synchonization and retry sending the
673 * message from the beginning.
675 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
677 unreserve_psock(kcm);
680 kcm_report_tx_retry(kcm);
688 KCM_STATS_ADD(psock->stats.tx_bytes, ret);
689 if (frag_offset < frag->size) {
690 /* Not finished with this frag */
696 if (skb_has_frag_list(skb)) {
697 skb = skb_shinfo(skb)->frag_list;
700 } else if (skb->next) {
705 /* Successfully sent the whole packet, account for it. */
706 skb_dequeue(&sk->sk_write_queue);
708 sk->sk_wmem_queued -= sent;
710 KCM_STATS_INCR(psock->stats.tx_msgs);
711 } while ((head = skb_peek(&sk->sk_write_queue)));
714 /* Done with all queued messages. */
715 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
716 unreserve_psock(kcm);
719 /* Check if write space is available */
720 sk->sk_write_space(sk);
722 return total_sent ? : ret;
725 static void kcm_tx_work(struct work_struct *w)
727 struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
728 struct sock *sk = &kcm->sk;
733 /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
736 err = kcm_write_msgs(kcm);
738 /* Hard failure in write, report error on KCM socket */
739 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
740 report_csk_error(&kcm->sk, -err);
744 /* Primarily for SOCK_SEQPACKET sockets */
745 if (likely(sk->sk_socket) &&
746 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
747 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
748 sk->sk_write_space(sk);
755 static void kcm_push(struct kcm_sock *kcm)
757 if (kcm->tx_wait_more)
761 static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
762 int offset, size_t size, int flags)
765 struct sock *sk = sock->sk;
766 struct kcm_sock *kcm = kcm_sk(sk);
767 struct sk_buff *skb = NULL, *head = NULL;
768 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
773 if (flags & MSG_SENDPAGE_NOTLAST)
776 /* No MSG_EOR from splice, only look at MSG_MORE */
777 eor = !(flags & MSG_MORE);
781 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
788 /* Previously opened message */
790 skb = kcm_tx_msg(head)->last_skb;
791 i = skb_shinfo(skb)->nr_frags;
793 if (skb_can_coalesce(skb, i, page, offset)) {
794 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
795 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
799 if (i >= MAX_SKB_FRAGS) {
800 struct sk_buff *tskb;
802 tskb = alloc_skb(0, sk->sk_allocation);
805 err = sk_stream_wait_memory(sk, &timeo);
811 skb_shinfo(head)->frag_list = tskb;
816 skb->ip_summed = CHECKSUM_UNNECESSARY;
820 /* Call the sk_stream functions to manage the sndbuf mem. */
821 if (!sk_stream_memory_free(sk)) {
823 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
824 err = sk_stream_wait_memory(sk, &timeo);
829 head = alloc_skb(0, sk->sk_allocation);
832 err = sk_stream_wait_memory(sk, &timeo);
842 skb_fill_page_desc(skb, i, page, offset, size);
843 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
847 skb->data_len += size;
848 skb->truesize += size;
849 sk->sk_wmem_queued += size;
850 sk_mem_charge(sk, size);
854 head->data_len += size;
855 head->truesize += size;
859 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
861 /* Message complete, queue it on send buffer */
862 __skb_queue_tail(&sk->sk_write_queue, head);
864 KCM_STATS_INCR(kcm->stats.tx_msgs);
866 if (flags & MSG_BATCH) {
867 kcm->tx_wait_more = true;
868 } else if (kcm->tx_wait_more || not_busy) {
869 err = kcm_write_msgs(kcm);
871 /* We got a hard error in write_msgs but have
872 * already queued this message. Report an error
873 * in the socket, but don't affect return value
876 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
877 report_csk_error(&kcm->sk, -err);
881 /* Message not complete, save state */
883 kcm_tx_msg(head)->last_skb = skb;
886 KCM_STATS_ADD(kcm->stats.tx_bytes, size);
894 err = sk_stream_error(sk, flags, err);
896 /* make sure we wake any epoll edge trigger waiter */
897 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
898 sk->sk_write_space(sk);
904 static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
906 struct sock *sk = sock->sk;
907 struct kcm_sock *kcm = kcm_sk(sk);
908 struct sk_buff *skb = NULL, *head = NULL;
909 size_t copy, copied = 0;
910 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
911 int eor = (sock->type == SOCK_DGRAM) ?
912 !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
917 /* Per tcp_sendmsg this should be in poll */
918 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
924 /* Previously opened message */
926 skb = kcm_tx_msg(head)->last_skb;
930 /* Call the sk_stream functions to manage the sndbuf mem. */
931 if (!sk_stream_memory_free(sk)) {
933 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
934 err = sk_stream_wait_memory(sk, &timeo);
939 if (msg_data_left(msg)) {
940 /* New message, alloc head skb */
941 head = alloc_skb(0, sk->sk_allocation);
944 err = sk_stream_wait_memory(sk, &timeo);
948 head = alloc_skb(0, sk->sk_allocation);
953 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
954 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
956 skb->ip_summed = CHECKSUM_UNNECESSARY;
960 while (msg_data_left(msg)) {
962 int i = skb_shinfo(skb)->nr_frags;
963 struct page_frag *pfrag = sk_page_frag(sk);
965 if (!sk_page_frag_refill(sk, pfrag))
966 goto wait_for_memory;
968 if (!skb_can_coalesce(skb, i, pfrag->page,
970 if (i == MAX_SKB_FRAGS) {
971 struct sk_buff *tskb;
973 tskb = alloc_skb(0, sk->sk_allocation);
975 goto wait_for_memory;
978 skb_shinfo(head)->frag_list = tskb;
983 skb->ip_summed = CHECKSUM_UNNECESSARY;
989 copy = min_t(int, msg_data_left(msg),
990 pfrag->size - pfrag->offset);
992 if (!sk_wmem_schedule(sk, copy))
993 goto wait_for_memory;
995 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1002 /* Update the skb. */
1004 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1006 skb_fill_page_desc(skb, i, pfrag->page,
1007 pfrag->offset, copy);
1008 get_page(pfrag->page);
1011 pfrag->offset += copy;
1015 head->data_len += copy;
1022 err = sk_stream_wait_memory(sk, &timeo);
1028 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1031 /* Message complete, queue it on send buffer */
1032 __skb_queue_tail(&sk->sk_write_queue, head);
1033 kcm->seq_skb = NULL;
1034 KCM_STATS_INCR(kcm->stats.tx_msgs);
1037 if (msg->msg_flags & MSG_BATCH) {
1038 kcm->tx_wait_more = true;
1039 } else if (kcm->tx_wait_more || not_busy) {
1040 err = kcm_write_msgs(kcm);
1042 /* We got a hard error in write_msgs but have
1043 * already queued this message. Report an error
1044 * in the socket, but don't affect return value
1047 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1048 report_csk_error(&kcm->sk, -err);
1052 /* Message not complete, save state */
1055 kcm->seq_skb = head;
1056 kcm_tx_msg(head)->last_skb = skb;
1060 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1068 if (sock->type == SOCK_SEQPACKET) {
1069 /* Wrote some bytes before encountering an
1070 * error, return partial success.
1073 goto partial_message;
1074 if (head != kcm->seq_skb)
1078 kcm->seq_skb = NULL;
1081 err = sk_stream_error(sk, msg->msg_flags, err);
1083 /* make sure we wake any epoll edge trigger waiter */
1084 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1085 sk->sk_write_space(sk);
1091 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1092 size_t len, int flags)
1094 int noblock = flags & MSG_DONTWAIT;
1095 struct sock *sk = sock->sk;
1096 struct kcm_sock *kcm = kcm_sk(sk);
1098 struct strp_msg *stm;
1100 struct sk_buff *skb;
1102 skb = skb_recv_datagram(sk, flags, noblock, &err);
1106 /* Okay, have a message on the receive queue */
1108 stm = strp_msg(skb);
1110 if (len > stm->full_len)
1111 len = stm->full_len;
1113 err = skb_copy_datagram_msg(skb, stm->offset, msg, len);
1118 if (likely(!(flags & MSG_PEEK))) {
1119 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1120 if (copied < stm->full_len) {
1121 if (sock->type == SOCK_DGRAM) {
1122 /* Truncated message */
1123 msg->msg_flags |= MSG_TRUNC;
1126 stm->offset += copied;
1127 stm->full_len -= copied;
1130 /* Finished with message */
1131 msg->msg_flags |= MSG_EOR;
1132 KCM_STATS_INCR(kcm->stats.rx_msgs);
1137 skb_free_datagram(sk, skb);
1138 return copied ? : err;
1141 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1142 struct pipe_inode_info *pipe, size_t len,
1145 int noblock = flags & MSG_DONTWAIT;
1146 struct sock *sk = sock->sk;
1147 struct kcm_sock *kcm = kcm_sk(sk);
1148 struct strp_msg *stm;
1151 struct sk_buff *skb;
1153 /* Only support splice for SOCKSEQPACKET */
1155 skb = skb_recv_datagram(sk, flags, noblock, &err);
1159 /* Okay, have a message on the receive queue */
1161 stm = strp_msg(skb);
1163 if (len > stm->full_len)
1164 len = stm->full_len;
1166 copied = skb_splice_bits(skb, sk, stm->offset, pipe, len, flags);
1172 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1174 stm->offset += copied;
1175 stm->full_len -= copied;
1177 /* We have no way to return MSG_EOR. If all the bytes have been
1178 * read we still leave the message in the receive socket buffer.
1179 * A subsequent recvmsg needs to be done to return MSG_EOR and
1180 * finish reading the message.
1183 skb_free_datagram(sk, skb);
1187 skb_free_datagram(sk, skb);
1191 /* kcm sock lock held */
1192 static void kcm_recv_disable(struct kcm_sock *kcm)
1194 struct kcm_mux *mux = kcm->mux;
1196 if (kcm->rx_disabled)
1199 spin_lock_bh(&mux->rx_lock);
1201 kcm->rx_disabled = 1;
1203 /* If a psock is reserved we'll do cleanup in unreserve */
1204 if (!kcm->rx_psock) {
1206 list_del(&kcm->wait_rx_list);
1207 /* paired with lockless reads in kcm_rfree() */
1208 WRITE_ONCE(kcm->rx_wait, false);
1211 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1214 spin_unlock_bh(&mux->rx_lock);
1217 /* kcm sock lock held */
1218 static void kcm_recv_enable(struct kcm_sock *kcm)
1220 struct kcm_mux *mux = kcm->mux;
1222 if (!kcm->rx_disabled)
1225 spin_lock_bh(&mux->rx_lock);
1227 kcm->rx_disabled = 0;
1230 spin_unlock_bh(&mux->rx_lock);
1233 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1234 char __user *optval, unsigned int optlen)
1236 struct kcm_sock *kcm = kcm_sk(sock->sk);
1240 if (level != SOL_KCM)
1241 return -ENOPROTOOPT;
1243 if (optlen < sizeof(int))
1246 if (get_user(val, (int __user *)optval))
1249 valbool = val ? 1 : 0;
1252 case KCM_RECV_DISABLE:
1253 lock_sock(&kcm->sk);
1255 kcm_recv_disable(kcm);
1257 kcm_recv_enable(kcm);
1258 release_sock(&kcm->sk);
1267 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1268 char __user *optval, int __user *optlen)
1270 struct kcm_sock *kcm = kcm_sk(sock->sk);
1273 if (level != SOL_KCM)
1274 return -ENOPROTOOPT;
1276 if (get_user(len, optlen))
1279 len = min_t(unsigned int, len, sizeof(int));
1284 case KCM_RECV_DISABLE:
1285 val = kcm->rx_disabled;
1288 return -ENOPROTOOPT;
1291 if (put_user(len, optlen))
1293 if (copy_to_user(optval, &val, len))
1298 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1300 struct kcm_sock *tkcm;
1301 struct list_head *head;
1304 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1305 * we set sk_state, otherwise epoll_wait always returns right away with
1308 kcm->sk.sk_state = TCP_ESTABLISHED;
1310 /* Add to mux's kcm sockets list */
1312 spin_lock_bh(&mux->lock);
1314 head = &mux->kcm_socks;
1315 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1316 if (tkcm->index != index)
1318 head = &tkcm->kcm_sock_list;
1322 list_add(&kcm->kcm_sock_list, head);
1325 mux->kcm_socks_cnt++;
1326 spin_unlock_bh(&mux->lock);
1328 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1330 spin_lock_bh(&mux->rx_lock);
1332 spin_unlock_bh(&mux->rx_lock);
1335 static int kcm_attach(struct socket *sock, struct socket *csock,
1336 struct bpf_prog *prog)
1338 struct kcm_sock *kcm = kcm_sk(sock->sk);
1339 struct kcm_mux *mux = kcm->mux;
1341 struct kcm_psock *psock = NULL, *tpsock;
1342 struct list_head *head;
1344 static const struct strp_callbacks cb = {
1345 .rcv_msg = kcm_rcv_strparser,
1346 .parse_msg = kcm_parse_func_strparser,
1347 .read_sock_done = kcm_read_sock_done,
1357 /* Only allow TCP sockets to be attached for now */
1358 if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1359 csk->sk_protocol != IPPROTO_TCP) {
1364 /* Don't allow listeners or closed sockets */
1365 if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1370 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1378 psock->bpf_prog = prog;
1380 write_lock_bh(&csk->sk_callback_lock);
1382 /* Check if sk_user_data is aready by KCM or someone else.
1383 * Must be done under lock to prevent race conditions.
1385 if (csk->sk_user_data) {
1386 write_unlock_bh(&csk->sk_callback_lock);
1387 kmem_cache_free(kcm_psockp, psock);
1392 err = strp_init(&psock->strp, csk, &cb);
1394 write_unlock_bh(&csk->sk_callback_lock);
1395 kmem_cache_free(kcm_psockp, psock);
1399 psock->save_data_ready = csk->sk_data_ready;
1400 psock->save_write_space = csk->sk_write_space;
1401 psock->save_state_change = csk->sk_state_change;
1402 csk->sk_user_data = psock;
1403 csk->sk_data_ready = psock_data_ready;
1404 csk->sk_write_space = psock_write_space;
1405 csk->sk_state_change = psock_state_change;
1407 write_unlock_bh(&csk->sk_callback_lock);
1411 /* Finished initialization, now add the psock to the MUX. */
1412 spin_lock_bh(&mux->lock);
1413 head = &mux->psocks;
1414 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1415 if (tpsock->index != index)
1417 head = &tpsock->psock_list;
1421 list_add(&psock->psock_list, head);
1422 psock->index = index;
1424 KCM_STATS_INCR(mux->stats.psock_attach);
1426 psock_now_avail(psock);
1427 spin_unlock_bh(&mux->lock);
1429 /* Schedule RX work in case there are already bytes queued */
1430 strp_check_rcv(&psock->strp);
1438 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1440 struct socket *csock;
1441 struct bpf_prog *prog;
1444 csock = sockfd_lookup(info->fd, &err);
1448 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1450 err = PTR_ERR(prog);
1454 err = kcm_attach(sock, csock, prog);
1460 /* Keep reference on file also */
1468 static void kcm_unattach(struct kcm_psock *psock)
1470 struct sock *csk = psock->sk;
1471 struct kcm_mux *mux = psock->mux;
1475 /* Stop getting callbacks from TCP socket. After this there should
1476 * be no way to reserve a kcm for this psock.
1478 write_lock_bh(&csk->sk_callback_lock);
1479 csk->sk_user_data = NULL;
1480 csk->sk_data_ready = psock->save_data_ready;
1481 csk->sk_write_space = psock->save_write_space;
1482 csk->sk_state_change = psock->save_state_change;
1483 strp_stop(&psock->strp);
1485 if (WARN_ON(psock->rx_kcm)) {
1486 write_unlock_bh(&csk->sk_callback_lock);
1491 spin_lock_bh(&mux->rx_lock);
1493 /* Stop receiver activities. After this point psock should not be
1494 * able to get onto ready list either through callbacks or work.
1496 if (psock->ready_rx_msg) {
1497 list_del(&psock->psock_ready_list);
1498 kfree_skb(psock->ready_rx_msg);
1499 psock->ready_rx_msg = NULL;
1500 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1503 spin_unlock_bh(&mux->rx_lock);
1505 write_unlock_bh(&csk->sk_callback_lock);
1507 /* Call strp_done without sock lock */
1509 strp_done(&psock->strp);
1512 bpf_prog_put(psock->bpf_prog);
1514 spin_lock_bh(&mux->lock);
1516 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1517 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1519 KCM_STATS_INCR(mux->stats.psock_unattach);
1521 if (psock->tx_kcm) {
1522 /* psock was reserved. Just mark it finished and we will clean
1523 * up in the kcm paths, we need kcm lock which can not be
1526 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1527 spin_unlock_bh(&mux->lock);
1529 /* We are unattaching a socket that is reserved. Abort the
1530 * socket since we may be out of sync in sending on it. We need
1531 * to do this without the mux lock.
1533 kcm_abort_tx_psock(psock, EPIPE, false);
1535 spin_lock_bh(&mux->lock);
1536 if (!psock->tx_kcm) {
1537 /* psock now unreserved in window mux was unlocked */
1542 /* Commit done before queuing work to process it */
1545 /* Queue tx work to make sure psock->done is handled */
1546 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1547 spin_unlock_bh(&mux->lock);
1550 if (!psock->tx_stopped)
1551 list_del(&psock->psock_avail_list);
1552 list_del(&psock->psock_list);
1554 spin_unlock_bh(&mux->lock);
1557 fput(csk->sk_socket->file);
1558 kmem_cache_free(kcm_psockp, psock);
1564 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1566 struct kcm_sock *kcm = kcm_sk(sock->sk);
1567 struct kcm_mux *mux = kcm->mux;
1568 struct kcm_psock *psock;
1569 struct socket *csock;
1573 csock = sockfd_lookup(info->fd, &err);
1585 spin_lock_bh(&mux->lock);
1587 list_for_each_entry(psock, &mux->psocks, psock_list) {
1588 if (psock->sk != csk)
1591 /* Found the matching psock */
1593 if (psock->unattaching || WARN_ON(psock->done)) {
1598 psock->unattaching = 1;
1600 spin_unlock_bh(&mux->lock);
1602 /* Lower socket lock should already be held */
1603 kcm_unattach(psock);
1609 spin_unlock_bh(&mux->lock);
1616 static struct proto kcm_proto = {
1618 .owner = THIS_MODULE,
1619 .obj_size = sizeof(struct kcm_sock),
1622 /* Clone a kcm socket. */
1623 static struct file *kcm_clone(struct socket *osock)
1625 struct socket *newsock;
1629 newsock = sock_alloc();
1631 return ERR_PTR(-ENFILE);
1633 newsock->type = osock->type;
1634 newsock->ops = osock->ops;
1636 __module_get(newsock->ops->owner);
1638 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1641 sock_release(newsock);
1642 return ERR_PTR(-ENOMEM);
1644 sock_init_data(newsock, newsk);
1645 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1647 file = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1649 sock_release(newsock);
1654 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1659 case SIOCKCMATTACH: {
1660 struct kcm_attach info;
1662 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1665 err = kcm_attach_ioctl(sock, &info);
1669 case SIOCKCMUNATTACH: {
1670 struct kcm_unattach info;
1672 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1675 err = kcm_unattach_ioctl(sock, &info);
1679 case SIOCKCMCLONE: {
1680 struct kcm_clone info;
1683 info.fd = get_unused_fd_flags(0);
1684 if (unlikely(info.fd < 0))
1687 file = kcm_clone(sock);
1689 put_unused_fd(info.fd);
1690 return PTR_ERR(file);
1692 if (copy_to_user((void __user *)arg, &info,
1694 put_unused_fd(info.fd);
1698 fd_install(info.fd, file);
1710 static void free_mux(struct rcu_head *rcu)
1712 struct kcm_mux *mux = container_of(rcu,
1713 struct kcm_mux, rcu);
1715 kmem_cache_free(kcm_muxp, mux);
1718 static void release_mux(struct kcm_mux *mux)
1720 struct kcm_net *knet = mux->knet;
1721 struct kcm_psock *psock, *tmp_psock;
1723 /* Release psocks */
1724 list_for_each_entry_safe(psock, tmp_psock,
1725 &mux->psocks, psock_list) {
1726 if (!WARN_ON(psock->unattaching))
1727 kcm_unattach(psock);
1730 if (WARN_ON(mux->psocks_cnt))
1733 __skb_queue_purge(&mux->rx_hold_queue);
1735 mutex_lock(&knet->mutex);
1736 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1737 aggregate_psock_stats(&mux->aggregate_psock_stats,
1738 &knet->aggregate_psock_stats);
1739 aggregate_strp_stats(&mux->aggregate_strp_stats,
1740 &knet->aggregate_strp_stats);
1741 list_del_rcu(&mux->kcm_mux_list);
1743 mutex_unlock(&knet->mutex);
1745 call_rcu(&mux->rcu, free_mux);
1748 static void kcm_done(struct kcm_sock *kcm)
1750 struct kcm_mux *mux = kcm->mux;
1751 struct sock *sk = &kcm->sk;
1754 spin_lock_bh(&mux->rx_lock);
1755 if (kcm->rx_psock) {
1756 /* Cleanup in unreserve_rx_kcm */
1758 kcm->rx_disabled = 1;
1760 spin_unlock_bh(&mux->rx_lock);
1765 list_del(&kcm->wait_rx_list);
1766 /* paired with lockless reads in kcm_rfree() */
1767 WRITE_ONCE(kcm->rx_wait, false);
1769 /* Move any pending receive messages to other kcm sockets */
1770 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1772 spin_unlock_bh(&mux->rx_lock);
1774 if (WARN_ON(sk_rmem_alloc_get(sk)))
1777 /* Detach from MUX */
1778 spin_lock_bh(&mux->lock);
1780 list_del(&kcm->kcm_sock_list);
1781 mux->kcm_socks_cnt--;
1782 socks_cnt = mux->kcm_socks_cnt;
1784 spin_unlock_bh(&mux->lock);
1787 /* We are done with the mux now. */
1791 WARN_ON(kcm->rx_wait);
1796 /* Called by kcm_release to close a KCM socket.
1797 * If this is the last KCM socket on the MUX, destroy the MUX.
1799 static int kcm_release(struct socket *sock)
1801 struct sock *sk = sock->sk;
1802 struct kcm_sock *kcm;
1803 struct kcm_mux *mux;
1804 struct kcm_psock *psock;
1814 kfree_skb(kcm->seq_skb);
1816 /* Purge queue under lock to avoid race condition with tx_work trying
1817 * to act when queue is nonempty. If tx_work runs after this point
1818 * it will just return.
1820 __skb_queue_purge(&sk->sk_write_queue);
1822 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1823 * get a writespace callback. This prevents further work being queued
1824 * from the callback (unbinding the psock occurs after canceling work.
1826 kcm->tx_stopped = 1;
1830 spin_lock_bh(&mux->lock);
1832 /* Take of tx_wait list, after this point there should be no way
1833 * that a psock will be assigned to this kcm.
1835 list_del(&kcm->wait_psock_list);
1836 kcm->tx_wait = false;
1838 spin_unlock_bh(&mux->lock);
1840 /* Cancel work. After this point there should be no outside references
1841 * to the kcm socket.
1843 cancel_work_sync(&kcm->tx_work);
1846 psock = kcm->tx_psock;
1848 /* A psock was reserved, so we need to kill it since it
1849 * may already have some bytes queued from a message. We
1850 * need to do this after removing kcm from tx_wait list.
1852 kcm_abort_tx_psock(psock, EPIPE, false);
1853 unreserve_psock(kcm);
1857 WARN_ON(kcm->tx_wait);
1858 WARN_ON(kcm->tx_psock);
1867 static const struct proto_ops kcm_dgram_ops = {
1869 .owner = THIS_MODULE,
1870 .release = kcm_release,
1871 .bind = sock_no_bind,
1872 .connect = sock_no_connect,
1873 .socketpair = sock_no_socketpair,
1874 .accept = sock_no_accept,
1875 .getname = sock_no_getname,
1876 .poll = datagram_poll,
1878 .listen = sock_no_listen,
1879 .shutdown = sock_no_shutdown,
1880 .setsockopt = kcm_setsockopt,
1881 .getsockopt = kcm_getsockopt,
1882 .sendmsg = kcm_sendmsg,
1883 .recvmsg = kcm_recvmsg,
1884 .mmap = sock_no_mmap,
1885 .sendpage = kcm_sendpage,
1888 static const struct proto_ops kcm_seqpacket_ops = {
1890 .owner = THIS_MODULE,
1891 .release = kcm_release,
1892 .bind = sock_no_bind,
1893 .connect = sock_no_connect,
1894 .socketpair = sock_no_socketpair,
1895 .accept = sock_no_accept,
1896 .getname = sock_no_getname,
1897 .poll = datagram_poll,
1899 .listen = sock_no_listen,
1900 .shutdown = sock_no_shutdown,
1901 .setsockopt = kcm_setsockopt,
1902 .getsockopt = kcm_getsockopt,
1903 .sendmsg = kcm_sendmsg,
1904 .recvmsg = kcm_recvmsg,
1905 .mmap = sock_no_mmap,
1906 .sendpage = kcm_sendpage,
1907 .splice_read = kcm_splice_read,
1910 /* Create proto operation for kcm sockets */
1911 static int kcm_create(struct net *net, struct socket *sock,
1912 int protocol, int kern)
1914 struct kcm_net *knet = net_generic(net, kcm_net_id);
1916 struct kcm_mux *mux;
1918 switch (sock->type) {
1920 sock->ops = &kcm_dgram_ops;
1922 case SOCK_SEQPACKET:
1923 sock->ops = &kcm_seqpacket_ops;
1926 return -ESOCKTNOSUPPORT;
1929 if (protocol != KCMPROTO_CONNECTED)
1930 return -EPROTONOSUPPORT;
1932 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1936 /* Allocate a kcm mux, shared between KCM sockets */
1937 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1943 spin_lock_init(&mux->lock);
1944 spin_lock_init(&mux->rx_lock);
1945 INIT_LIST_HEAD(&mux->kcm_socks);
1946 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1947 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1949 INIT_LIST_HEAD(&mux->psocks);
1950 INIT_LIST_HEAD(&mux->psocks_ready);
1951 INIT_LIST_HEAD(&mux->psocks_avail);
1955 /* Add new MUX to list */
1956 mutex_lock(&knet->mutex);
1957 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1959 mutex_unlock(&knet->mutex);
1961 skb_queue_head_init(&mux->rx_hold_queue);
1963 /* Init KCM socket */
1964 sock_init_data(sock, sk);
1965 init_kcm_sock(kcm_sk(sk), mux);
1970 static const struct net_proto_family kcm_family_ops = {
1972 .create = kcm_create,
1973 .owner = THIS_MODULE,
1976 static __net_init int kcm_init_net(struct net *net)
1978 struct kcm_net *knet = net_generic(net, kcm_net_id);
1980 INIT_LIST_HEAD_RCU(&knet->mux_list);
1981 mutex_init(&knet->mutex);
1986 static __net_exit void kcm_exit_net(struct net *net)
1988 struct kcm_net *knet = net_generic(net, kcm_net_id);
1990 /* All KCM sockets should be closed at this point, which should mean
1991 * that all multiplexors and psocks have been destroyed.
1993 WARN_ON(!list_empty(&knet->mux_list));
1995 mutex_destroy(&knet->mutex);
1998 static struct pernet_operations kcm_net_ops = {
1999 .init = kcm_init_net,
2000 .exit = kcm_exit_net,
2002 .size = sizeof(struct kcm_net),
2005 static int __init kcm_init(void)
2009 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2010 sizeof(struct kcm_mux), 0,
2011 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2015 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2016 sizeof(struct kcm_psock), 0,
2017 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2021 kcm_wq = create_singlethread_workqueue("kkcmd");
2025 err = proto_register(&kcm_proto, 1);
2029 err = register_pernet_device(&kcm_net_ops);
2033 err = sock_register(&kcm_family_ops);
2035 goto sock_register_fail;
2037 err = kcm_proc_init();
2039 goto proc_init_fail;
2044 sock_unregister(PF_KCM);
2047 unregister_pernet_device(&kcm_net_ops);
2050 proto_unregister(&kcm_proto);
2053 kmem_cache_destroy(kcm_muxp);
2054 kmem_cache_destroy(kcm_psockp);
2057 destroy_workqueue(kcm_wq);
2062 static void __exit kcm_exit(void)
2065 sock_unregister(PF_KCM);
2066 unregister_pernet_device(&kcm_net_ops);
2067 proto_unregister(&kcm_proto);
2068 destroy_workqueue(kcm_wq);
2070 kmem_cache_destroy(kcm_muxp);
2071 kmem_cache_destroy(kcm_psockp);
2074 module_init(kcm_init);
2075 module_exit(kcm_exit);
2077 MODULE_LICENSE("GPL");
2078 MODULE_ALIAS_NETPROTO(PF_KCM);