1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel Connection Multiplexor
5 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
9 #include <linux/errno.h>
10 #include <linux/errqueue.h>
11 #include <linux/file.h>
12 #include <linux/filter.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/net.h>
17 #include <linux/netdevice.h>
18 #include <linux/poll.h>
19 #include <linux/rculist.h>
20 #include <linux/skbuff.h>
21 #include <linux/socket.h>
22 #include <linux/uaccess.h>
23 #include <linux/workqueue.h>
24 #include <linux/syscalls.h>
25 #include <linux/sched/signal.h>
28 #include <net/netns/generic.h>
30 #include <uapi/linux/kcm.h>
31 #include <trace/events/sock.h>
33 unsigned int kcm_net_id;
35 static struct kmem_cache *kcm_psockp __read_mostly;
36 static struct kmem_cache *kcm_muxp __read_mostly;
37 static struct workqueue_struct *kcm_wq;
39 static inline struct kcm_sock *kcm_sk(const struct sock *sk)
41 return (struct kcm_sock *)sk;
44 static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
46 return (struct kcm_tx_msg *)skb->cb;
49 static void report_csk_error(struct sock *csk, int err)
55 static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
58 struct sock *csk = psock->sk;
59 struct kcm_mux *mux = psock->mux;
61 /* Unrecoverable error in transmit */
63 spin_lock_bh(&mux->lock);
65 if (psock->tx_stopped) {
66 spin_unlock_bh(&mux->lock);
70 psock->tx_stopped = 1;
71 KCM_STATS_INCR(psock->stats.tx_aborts);
74 /* Take off psocks_avail list */
75 list_del(&psock->psock_avail_list);
76 } else if (wakeup_kcm) {
77 /* In this case psock is being aborted while outside of
78 * write_msgs and psock is reserved. Schedule tx_work
79 * to handle the failure there. Need to commit tx_stopped
80 * before queuing work.
84 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
87 spin_unlock_bh(&mux->lock);
89 /* Report error on lower socket */
90 report_csk_error(csk, err);
93 /* RX mux lock held. */
94 static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
95 struct kcm_psock *psock)
97 STRP_STATS_ADD(mux->stats.rx_bytes,
98 psock->strp.stats.bytes -
99 psock->saved_rx_bytes);
100 mux->stats.rx_msgs +=
101 psock->strp.stats.msgs - psock->saved_rx_msgs;
102 psock->saved_rx_msgs = psock->strp.stats.msgs;
103 psock->saved_rx_bytes = psock->strp.stats.bytes;
106 static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
107 struct kcm_psock *psock)
109 KCM_STATS_ADD(mux->stats.tx_bytes,
110 psock->stats.tx_bytes - psock->saved_tx_bytes);
111 mux->stats.tx_msgs +=
112 psock->stats.tx_msgs - psock->saved_tx_msgs;
113 psock->saved_tx_msgs = psock->stats.tx_msgs;
114 psock->saved_tx_bytes = psock->stats.tx_bytes;
117 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
119 /* KCM is ready to receive messages on its queue-- either the KCM is new or
120 * has become unblocked after being blocked on full socket buffer. Queue any
121 * pending ready messages on a psock. RX mux lock held.
123 static void kcm_rcv_ready(struct kcm_sock *kcm)
125 struct kcm_mux *mux = kcm->mux;
126 struct kcm_psock *psock;
129 if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
132 while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) {
133 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
134 /* Assuming buffer limit has been reached */
135 skb_queue_head(&mux->rx_hold_queue, skb);
136 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
141 while (!list_empty(&mux->psocks_ready)) {
142 psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
145 if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) {
146 /* Assuming buffer limit has been reached */
147 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
151 /* Consumed the ready message on the psock. Schedule rx_work to
154 list_del(&psock->psock_ready_list);
155 psock->ready_rx_msg = NULL;
156 /* Commit clearing of ready_rx_msg for queuing work */
159 strp_unpause(&psock->strp);
160 strp_check_rcv(&psock->strp);
163 /* Buffer limit is okay now, add to ready list */
164 list_add_tail(&kcm->wait_rx_list,
165 &kcm->mux->kcm_rx_waiters);
166 /* paired with lockless reads in kcm_rfree() */
167 WRITE_ONCE(kcm->rx_wait, true);
170 static void kcm_rfree(struct sk_buff *skb)
172 struct sock *sk = skb->sk;
173 struct kcm_sock *kcm = kcm_sk(sk);
174 struct kcm_mux *mux = kcm->mux;
175 unsigned int len = skb->truesize;
177 sk_mem_uncharge(sk, len);
178 atomic_sub(len, &sk->sk_rmem_alloc);
180 /* For reading rx_wait and rx_psock without holding lock */
181 smp_mb__after_atomic();
183 if (!READ_ONCE(kcm->rx_wait) && !READ_ONCE(kcm->rx_psock) &&
184 sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
185 spin_lock_bh(&mux->rx_lock);
187 spin_unlock_bh(&mux->rx_lock);
191 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
193 struct sk_buff_head *list = &sk->sk_receive_queue;
195 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
198 if (!sk_rmem_schedule(sk, skb, skb->truesize))
205 skb->destructor = kcm_rfree;
206 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
207 sk_mem_charge(sk, skb->truesize);
209 skb_queue_tail(list, skb);
211 if (!sock_flag(sk, SOCK_DEAD))
212 sk->sk_data_ready(sk);
217 /* Requeue received messages for a kcm socket to other kcm sockets. This is
218 * called with a kcm socket is receive disabled.
221 static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
224 struct kcm_sock *kcm;
226 while ((skb = skb_dequeue(head))) {
227 /* Reset destructor to avoid calling kcm_rcv_ready */
228 skb->destructor = sock_rfree;
231 if (list_empty(&mux->kcm_rx_waiters)) {
232 skb_queue_tail(&mux->rx_hold_queue, skb);
236 kcm = list_first_entry(&mux->kcm_rx_waiters,
237 struct kcm_sock, wait_rx_list);
239 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
240 /* Should mean socket buffer full */
241 list_del(&kcm->wait_rx_list);
242 /* paired with lockless reads in kcm_rfree() */
243 WRITE_ONCE(kcm->rx_wait, false);
245 /* Commit rx_wait to read in kcm_free */
253 /* Lower sock lock held */
254 static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
255 struct sk_buff *head)
257 struct kcm_mux *mux = psock->mux;
258 struct kcm_sock *kcm;
260 WARN_ON(psock->ready_rx_msg);
263 return psock->rx_kcm;
265 spin_lock_bh(&mux->rx_lock);
268 spin_unlock_bh(&mux->rx_lock);
269 return psock->rx_kcm;
272 kcm_update_rx_mux_stats(mux, psock);
274 if (list_empty(&mux->kcm_rx_waiters)) {
275 psock->ready_rx_msg = head;
276 strp_pause(&psock->strp);
277 list_add_tail(&psock->psock_ready_list,
279 spin_unlock_bh(&mux->rx_lock);
283 kcm = list_first_entry(&mux->kcm_rx_waiters,
284 struct kcm_sock, wait_rx_list);
285 list_del(&kcm->wait_rx_list);
286 /* paired with lockless reads in kcm_rfree() */
287 WRITE_ONCE(kcm->rx_wait, false);
290 /* paired with lockless reads in kcm_rfree() */
291 WRITE_ONCE(kcm->rx_psock, psock);
293 spin_unlock_bh(&mux->rx_lock);
298 static void kcm_done(struct kcm_sock *kcm);
300 static void kcm_done_work(struct work_struct *w)
302 kcm_done(container_of(w, struct kcm_sock, done_work));
305 /* Lower sock held */
306 static void unreserve_rx_kcm(struct kcm_psock *psock,
309 struct kcm_sock *kcm = psock->rx_kcm;
310 struct kcm_mux *mux = psock->mux;
315 spin_lock_bh(&mux->rx_lock);
317 psock->rx_kcm = NULL;
318 /* paired with lockless reads in kcm_rfree() */
319 WRITE_ONCE(kcm->rx_psock, NULL);
321 /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
326 if (unlikely(kcm->done)) {
327 spin_unlock_bh(&mux->rx_lock);
329 /* Need to run kcm_done in a task since we need to qcquire
330 * callback locks which may already be held here.
332 INIT_WORK(&kcm->done_work, kcm_done_work);
333 schedule_work(&kcm->done_work);
337 if (unlikely(kcm->rx_disabled)) {
338 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
339 } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
340 /* Check for degenerative race with rx_wait that all
341 * data was dequeued (accounted for in kcm_rfree).
345 spin_unlock_bh(&mux->rx_lock);
348 /* Lower sock lock held */
349 static void psock_data_ready(struct sock *sk)
351 struct kcm_psock *psock;
353 trace_sk_data_ready(sk);
355 read_lock_bh(&sk->sk_callback_lock);
357 psock = (struct kcm_psock *)sk->sk_user_data;
359 strp_data_ready(&psock->strp);
361 read_unlock_bh(&sk->sk_callback_lock);
364 /* Called with lower sock held */
365 static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
367 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
368 struct kcm_sock *kcm;
371 kcm = reserve_rx_kcm(psock, skb);
373 /* Unable to reserve a KCM, message is held in psock and strp
379 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
380 /* Should mean socket buffer full */
381 unreserve_rx_kcm(psock, false);
386 static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
388 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
389 struct bpf_prog *prog = psock->bpf_prog;
392 res = bpf_prog_run_pin_on_cpu(prog, skb);
396 static int kcm_read_sock_done(struct strparser *strp, int err)
398 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
400 unreserve_rx_kcm(psock, true);
405 static void psock_state_change(struct sock *sk)
407 /* TCP only does a EPOLLIN for a half close. Do a EPOLLHUP here
408 * since application will normally not poll with EPOLLIN
409 * on the TCP sockets.
412 report_csk_error(sk, EPIPE);
415 static void psock_write_space(struct sock *sk)
417 struct kcm_psock *psock;
419 struct kcm_sock *kcm;
421 read_lock_bh(&sk->sk_callback_lock);
423 psock = (struct kcm_psock *)sk->sk_user_data;
424 if (unlikely(!psock))
428 spin_lock_bh(&mux->lock);
430 /* Check if the socket is reserved so someone is waiting for sending. */
432 if (kcm && !unlikely(kcm->tx_stopped))
433 queue_work(kcm_wq, &kcm->tx_work);
435 spin_unlock_bh(&mux->lock);
437 read_unlock_bh(&sk->sk_callback_lock);
440 static void unreserve_psock(struct kcm_sock *kcm);
442 /* kcm sock is locked. */
443 static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
445 struct kcm_mux *mux = kcm->mux;
446 struct kcm_psock *psock;
448 psock = kcm->tx_psock;
450 smp_rmb(); /* Must read tx_psock before tx_wait */
453 WARN_ON(kcm->tx_wait);
454 if (unlikely(psock->tx_stopped))
455 unreserve_psock(kcm);
457 return kcm->tx_psock;
460 spin_lock_bh(&mux->lock);
462 /* Check again under lock to see if psock was reserved for this
463 * psock via psock_unreserve.
465 psock = kcm->tx_psock;
466 if (unlikely(psock)) {
467 WARN_ON(kcm->tx_wait);
468 spin_unlock_bh(&mux->lock);
469 return kcm->tx_psock;
472 if (!list_empty(&mux->psocks_avail)) {
473 psock = list_first_entry(&mux->psocks_avail,
476 list_del(&psock->psock_avail_list);
478 list_del(&kcm->wait_psock_list);
479 kcm->tx_wait = false;
481 kcm->tx_psock = psock;
483 KCM_STATS_INCR(psock->stats.reserved);
484 } else if (!kcm->tx_wait) {
485 list_add_tail(&kcm->wait_psock_list,
486 &mux->kcm_tx_waiters);
490 spin_unlock_bh(&mux->lock);
496 static void psock_now_avail(struct kcm_psock *psock)
498 struct kcm_mux *mux = psock->mux;
499 struct kcm_sock *kcm;
501 if (list_empty(&mux->kcm_tx_waiters)) {
502 list_add_tail(&psock->psock_avail_list,
505 kcm = list_first_entry(&mux->kcm_tx_waiters,
508 list_del(&kcm->wait_psock_list);
509 kcm->tx_wait = false;
512 /* Commit before changing tx_psock since that is read in
513 * reserve_psock before queuing work.
517 kcm->tx_psock = psock;
518 KCM_STATS_INCR(psock->stats.reserved);
519 queue_work(kcm_wq, &kcm->tx_work);
523 /* kcm sock is locked. */
524 static void unreserve_psock(struct kcm_sock *kcm)
526 struct kcm_psock *psock;
527 struct kcm_mux *mux = kcm->mux;
529 spin_lock_bh(&mux->lock);
531 psock = kcm->tx_psock;
533 if (WARN_ON(!psock)) {
534 spin_unlock_bh(&mux->lock);
538 smp_rmb(); /* Read tx_psock before tx_wait */
540 kcm_update_tx_mux_stats(mux, psock);
542 WARN_ON(kcm->tx_wait);
544 kcm->tx_psock = NULL;
545 psock->tx_kcm = NULL;
546 KCM_STATS_INCR(psock->stats.unreserved);
548 if (unlikely(psock->tx_stopped)) {
551 list_del(&psock->psock_list);
554 fput(psock->sk->sk_socket->file);
555 kmem_cache_free(kcm_psockp, psock);
558 /* Don't put back on available list */
560 spin_unlock_bh(&mux->lock);
565 psock_now_avail(psock);
567 spin_unlock_bh(&mux->lock);
570 static void kcm_report_tx_retry(struct kcm_sock *kcm)
572 struct kcm_mux *mux = kcm->mux;
574 spin_lock_bh(&mux->lock);
575 KCM_STATS_INCR(mux->stats.tx_retries);
576 spin_unlock_bh(&mux->lock);
579 /* Write any messages ready on the kcm socket. Called with kcm sock lock
580 * held. Return bytes actually sent or error.
582 static int kcm_write_msgs(struct kcm_sock *kcm)
584 struct sock *sk = &kcm->sk;
585 struct kcm_psock *psock;
586 struct sk_buff *skb, *head;
587 struct kcm_tx_msg *txm;
588 unsigned short fragidx, frag_offset;
589 unsigned int sent, total_sent = 0;
592 kcm->tx_wait_more = false;
593 psock = kcm->tx_psock;
594 if (unlikely(psock && psock->tx_stopped)) {
595 /* A reserved psock was aborted asynchronously. Unreserve
596 * it and we'll retry the message.
598 unreserve_psock(kcm);
599 kcm_report_tx_retry(kcm);
600 if (skb_queue_empty(&sk->sk_write_queue))
603 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
605 } else if (skb_queue_empty(&sk->sk_write_queue)) {
609 head = skb_peek(&sk->sk_write_queue);
610 txm = kcm_tx_msg(head);
613 /* Send of first skbuff in queue already in progress */
614 if (WARN_ON(!psock)) {
619 frag_offset = txm->frag_offset;
620 fragidx = txm->fragidx;
627 psock = reserve_psock(kcm);
633 txm = kcm_tx_msg(head);
637 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
642 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
645 struct msghdr msg = {
646 .msg_flags = MSG_DONTWAIT | MSG_SPLICE_PAGES,
652 frag = &skb_shinfo(skb)->frags[fragidx];
653 if (WARN_ON(!skb_frag_size(frag))) {
660 skb_frag_size(frag) - frag_offset,
661 skb_frag_off(frag) + frag_offset);
662 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1,
664 ret = sock_sendmsg(psock->sk->sk_socket, &msg);
666 if (ret == -EAGAIN) {
667 /* Save state to try again when there's
668 * write space on the socket
671 txm->frag_offset = frag_offset;
672 txm->fragidx = fragidx;
679 /* Hard failure in sending message, abort this
680 * psock since it has lost framing
681 * synchronization and retry sending the
682 * message from the beginning.
684 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
686 unreserve_psock(kcm);
689 kcm_report_tx_retry(kcm);
697 KCM_STATS_ADD(psock->stats.tx_bytes, ret);
698 if (frag_offset < skb_frag_size(frag)) {
699 /* Not finished with this frag */
705 if (skb_has_frag_list(skb)) {
706 skb = skb_shinfo(skb)->frag_list;
709 } else if (skb->next) {
714 /* Successfully sent the whole packet, account for it. */
715 skb_dequeue(&sk->sk_write_queue);
717 sk->sk_wmem_queued -= sent;
719 KCM_STATS_INCR(psock->stats.tx_msgs);
720 } while ((head = skb_peek(&sk->sk_write_queue)));
723 /* Done with all queued messages. */
724 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
725 unreserve_psock(kcm);
728 /* Check if write space is available */
729 sk->sk_write_space(sk);
731 return total_sent ? : ret;
734 static void kcm_tx_work(struct work_struct *w)
736 struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
737 struct sock *sk = &kcm->sk;
742 /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
745 err = kcm_write_msgs(kcm);
747 /* Hard failure in write, report error on KCM socket */
748 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
749 report_csk_error(&kcm->sk, -err);
753 /* Primarily for SOCK_SEQPACKET sockets */
754 if (likely(sk->sk_socket) &&
755 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
756 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
757 sk->sk_write_space(sk);
764 static void kcm_push(struct kcm_sock *kcm)
766 if (kcm->tx_wait_more)
770 static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
772 struct sock *sk = sock->sk;
773 struct kcm_sock *kcm = kcm_sk(sk);
774 struct sk_buff *skb = NULL, *head = NULL;
775 size_t copy, copied = 0;
776 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
777 int eor = (sock->type == SOCK_DGRAM) ?
778 !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
783 /* Per tcp_sendmsg this should be in poll */
784 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
790 /* Previously opened message */
792 skb = kcm_tx_msg(head)->last_skb;
796 /* Call the sk_stream functions to manage the sndbuf mem. */
797 if (!sk_stream_memory_free(sk)) {
799 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
800 err = sk_stream_wait_memory(sk, &timeo);
805 if (msg_data_left(msg)) {
806 /* New message, alloc head skb */
807 head = alloc_skb(0, sk->sk_allocation);
810 err = sk_stream_wait_memory(sk, &timeo);
814 head = alloc_skb(0, sk->sk_allocation);
819 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
820 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
822 skb->ip_summed = CHECKSUM_UNNECESSARY;
826 while (msg_data_left(msg)) {
828 int i = skb_shinfo(skb)->nr_frags;
829 struct page_frag *pfrag = sk_page_frag(sk);
831 if (!sk_page_frag_refill(sk, pfrag))
832 goto wait_for_memory;
834 if (!skb_can_coalesce(skb, i, pfrag->page,
836 if (i == MAX_SKB_FRAGS) {
837 struct sk_buff *tskb;
839 tskb = alloc_skb(0, sk->sk_allocation);
841 goto wait_for_memory;
844 skb_shinfo(head)->frag_list = tskb;
849 skb->ip_summed = CHECKSUM_UNNECESSARY;
855 if (msg->msg_flags & MSG_SPLICE_PAGES) {
856 copy = msg_data_left(msg);
857 if (!sk_wmem_schedule(sk, copy))
858 goto wait_for_memory;
860 err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
863 if (err == -EMSGSIZE)
864 goto wait_for_memory;
869 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG;
870 sk_wmem_queued_add(sk, copy);
871 sk_mem_charge(sk, copy);
874 head->truesize += copy;
876 copy = min_t(int, msg_data_left(msg),
877 pfrag->size - pfrag->offset);
878 if (!sk_wmem_schedule(sk, copy))
879 goto wait_for_memory;
881 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
888 /* Update the skb. */
891 &skb_shinfo(skb)->frags[i - 1], copy);
893 skb_fill_page_desc(skb, i, pfrag->page,
894 pfrag->offset, copy);
895 get_page(pfrag->page);
898 pfrag->offset += copy;
904 head->data_len += copy;
911 err = sk_stream_wait_memory(sk, &timeo);
917 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
920 /* Message complete, queue it on send buffer */
921 __skb_queue_tail(&sk->sk_write_queue, head);
923 KCM_STATS_INCR(kcm->stats.tx_msgs);
926 if (msg->msg_flags & MSG_BATCH) {
927 kcm->tx_wait_more = true;
928 } else if (kcm->tx_wait_more || not_busy) {
929 err = kcm_write_msgs(kcm);
931 /* We got a hard error in write_msgs but have
932 * already queued this message. Report an error
933 * in the socket, but don't affect return value
936 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
937 report_csk_error(&kcm->sk, -err);
941 /* Message not complete, save state */
945 kcm_tx_msg(head)->last_skb = skb;
949 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
957 if (copied && sock->type == SOCK_SEQPACKET) {
958 /* Wrote some bytes before encountering an
959 * error, return partial success.
961 goto partial_message;
964 if (head != kcm->seq_skb)
967 err = sk_stream_error(sk, msg->msg_flags, err);
969 /* make sure we wake any epoll edge trigger waiter */
970 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
971 sk->sk_write_space(sk);
977 static void kcm_splice_eof(struct socket *sock)
979 struct sock *sk = sock->sk;
980 struct kcm_sock *kcm = kcm_sk(sk);
982 if (skb_queue_empty_lockless(&sk->sk_write_queue))
990 static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
991 int offset, size_t size, int flags)
995 struct msghdr msg = { .msg_flags = flags | MSG_SPLICE_PAGES, };
997 if (flags & MSG_SENDPAGE_NOTLAST)
998 msg.msg_flags |= MSG_MORE;
1000 if (flags & MSG_OOB)
1003 bvec_set_page(&bvec, page, size, offset);
1004 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
1005 return kcm_sendmsg(sock, &msg, size);
1008 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1009 size_t len, int flags)
1011 struct sock *sk = sock->sk;
1012 struct kcm_sock *kcm = kcm_sk(sk);
1014 struct strp_msg *stm;
1016 struct sk_buff *skb;
1018 skb = skb_recv_datagram(sk, flags, &err);
1022 /* Okay, have a message on the receive queue */
1024 stm = strp_msg(skb);
1026 if (len > stm->full_len)
1027 len = stm->full_len;
1029 err = skb_copy_datagram_msg(skb, stm->offset, msg, len);
1034 if (likely(!(flags & MSG_PEEK))) {
1035 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1036 if (copied < stm->full_len) {
1037 if (sock->type == SOCK_DGRAM) {
1038 /* Truncated message */
1039 msg->msg_flags |= MSG_TRUNC;
1042 stm->offset += copied;
1043 stm->full_len -= copied;
1046 /* Finished with message */
1047 msg->msg_flags |= MSG_EOR;
1048 KCM_STATS_INCR(kcm->stats.rx_msgs);
1053 skb_free_datagram(sk, skb);
1054 return copied ? : err;
1057 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1058 struct pipe_inode_info *pipe, size_t len,
1061 struct sock *sk = sock->sk;
1062 struct kcm_sock *kcm = kcm_sk(sk);
1063 struct strp_msg *stm;
1066 struct sk_buff *skb;
1068 /* Only support splice for SOCKSEQPACKET */
1070 skb = skb_recv_datagram(sk, flags, &err);
1074 /* Okay, have a message on the receive queue */
1076 stm = strp_msg(skb);
1078 if (len > stm->full_len)
1079 len = stm->full_len;
1081 copied = skb_splice_bits(skb, sk, stm->offset, pipe, len, flags);
1087 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1089 stm->offset += copied;
1090 stm->full_len -= copied;
1092 /* We have no way to return MSG_EOR. If all the bytes have been
1093 * read we still leave the message in the receive socket buffer.
1094 * A subsequent recvmsg needs to be done to return MSG_EOR and
1095 * finish reading the message.
1098 skb_free_datagram(sk, skb);
1102 skb_free_datagram(sk, skb);
1106 /* kcm sock lock held */
1107 static void kcm_recv_disable(struct kcm_sock *kcm)
1109 struct kcm_mux *mux = kcm->mux;
1111 if (kcm->rx_disabled)
1114 spin_lock_bh(&mux->rx_lock);
1116 kcm->rx_disabled = 1;
1118 /* If a psock is reserved we'll do cleanup in unreserve */
1119 if (!kcm->rx_psock) {
1121 list_del(&kcm->wait_rx_list);
1122 /* paired with lockless reads in kcm_rfree() */
1123 WRITE_ONCE(kcm->rx_wait, false);
1126 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1129 spin_unlock_bh(&mux->rx_lock);
1132 /* kcm sock lock held */
1133 static void kcm_recv_enable(struct kcm_sock *kcm)
1135 struct kcm_mux *mux = kcm->mux;
1137 if (!kcm->rx_disabled)
1140 spin_lock_bh(&mux->rx_lock);
1142 kcm->rx_disabled = 0;
1145 spin_unlock_bh(&mux->rx_lock);
1148 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1149 sockptr_t optval, unsigned int optlen)
1151 struct kcm_sock *kcm = kcm_sk(sock->sk);
1155 if (level != SOL_KCM)
1156 return -ENOPROTOOPT;
1158 if (optlen < sizeof(int))
1161 if (copy_from_sockptr(&val, optval, sizeof(int)))
1164 valbool = val ? 1 : 0;
1167 case KCM_RECV_DISABLE:
1168 lock_sock(&kcm->sk);
1170 kcm_recv_disable(kcm);
1172 kcm_recv_enable(kcm);
1173 release_sock(&kcm->sk);
1182 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1183 char __user *optval, int __user *optlen)
1185 struct kcm_sock *kcm = kcm_sk(sock->sk);
1188 if (level != SOL_KCM)
1189 return -ENOPROTOOPT;
1191 if (get_user(len, optlen))
1194 len = min_t(unsigned int, len, sizeof(int));
1199 case KCM_RECV_DISABLE:
1200 val = kcm->rx_disabled;
1203 return -ENOPROTOOPT;
1206 if (put_user(len, optlen))
1208 if (copy_to_user(optval, &val, len))
1213 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1215 struct kcm_sock *tkcm;
1216 struct list_head *head;
1219 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1220 * we set sk_state, otherwise epoll_wait always returns right away with
1223 kcm->sk.sk_state = TCP_ESTABLISHED;
1225 /* Add to mux's kcm sockets list */
1227 spin_lock_bh(&mux->lock);
1229 head = &mux->kcm_socks;
1230 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1231 if (tkcm->index != index)
1233 head = &tkcm->kcm_sock_list;
1237 list_add(&kcm->kcm_sock_list, head);
1240 mux->kcm_socks_cnt++;
1241 spin_unlock_bh(&mux->lock);
1243 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1245 spin_lock_bh(&mux->rx_lock);
1247 spin_unlock_bh(&mux->rx_lock);
1250 static int kcm_attach(struct socket *sock, struct socket *csock,
1251 struct bpf_prog *prog)
1253 struct kcm_sock *kcm = kcm_sk(sock->sk);
1254 struct kcm_mux *mux = kcm->mux;
1256 struct kcm_psock *psock = NULL, *tpsock;
1257 struct list_head *head;
1259 static const struct strp_callbacks cb = {
1260 .rcv_msg = kcm_rcv_strparser,
1261 .parse_msg = kcm_parse_func_strparser,
1262 .read_sock_done = kcm_read_sock_done,
1272 /* Only allow TCP sockets to be attached for now */
1273 if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1274 csk->sk_protocol != IPPROTO_TCP) {
1279 /* Don't allow listeners or closed sockets */
1280 if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1285 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1293 psock->bpf_prog = prog;
1295 write_lock_bh(&csk->sk_callback_lock);
1297 /* Check if sk_user_data is already by KCM or someone else.
1298 * Must be done under lock to prevent race conditions.
1300 if (csk->sk_user_data) {
1301 write_unlock_bh(&csk->sk_callback_lock);
1302 kmem_cache_free(kcm_psockp, psock);
1307 err = strp_init(&psock->strp, csk, &cb);
1309 write_unlock_bh(&csk->sk_callback_lock);
1310 kmem_cache_free(kcm_psockp, psock);
1314 psock->save_data_ready = csk->sk_data_ready;
1315 psock->save_write_space = csk->sk_write_space;
1316 psock->save_state_change = csk->sk_state_change;
1317 csk->sk_user_data = psock;
1318 csk->sk_data_ready = psock_data_ready;
1319 csk->sk_write_space = psock_write_space;
1320 csk->sk_state_change = psock_state_change;
1322 write_unlock_bh(&csk->sk_callback_lock);
1326 /* Finished initialization, now add the psock to the MUX. */
1327 spin_lock_bh(&mux->lock);
1328 head = &mux->psocks;
1329 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1330 if (tpsock->index != index)
1332 head = &tpsock->psock_list;
1336 list_add(&psock->psock_list, head);
1337 psock->index = index;
1339 KCM_STATS_INCR(mux->stats.psock_attach);
1341 psock_now_avail(psock);
1342 spin_unlock_bh(&mux->lock);
1344 /* Schedule RX work in case there are already bytes queued */
1345 strp_check_rcv(&psock->strp);
1353 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1355 struct socket *csock;
1356 struct bpf_prog *prog;
1359 csock = sockfd_lookup(info->fd, &err);
1363 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1365 err = PTR_ERR(prog);
1369 err = kcm_attach(sock, csock, prog);
1375 /* Keep reference on file also */
1383 static void kcm_unattach(struct kcm_psock *psock)
1385 struct sock *csk = psock->sk;
1386 struct kcm_mux *mux = psock->mux;
1390 /* Stop getting callbacks from TCP socket. After this there should
1391 * be no way to reserve a kcm for this psock.
1393 write_lock_bh(&csk->sk_callback_lock);
1394 csk->sk_user_data = NULL;
1395 csk->sk_data_ready = psock->save_data_ready;
1396 csk->sk_write_space = psock->save_write_space;
1397 csk->sk_state_change = psock->save_state_change;
1398 strp_stop(&psock->strp);
1400 if (WARN_ON(psock->rx_kcm)) {
1401 write_unlock_bh(&csk->sk_callback_lock);
1406 spin_lock_bh(&mux->rx_lock);
1408 /* Stop receiver activities. After this point psock should not be
1409 * able to get onto ready list either through callbacks or work.
1411 if (psock->ready_rx_msg) {
1412 list_del(&psock->psock_ready_list);
1413 kfree_skb(psock->ready_rx_msg);
1414 psock->ready_rx_msg = NULL;
1415 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1418 spin_unlock_bh(&mux->rx_lock);
1420 write_unlock_bh(&csk->sk_callback_lock);
1422 /* Call strp_done without sock lock */
1424 strp_done(&psock->strp);
1427 bpf_prog_put(psock->bpf_prog);
1429 spin_lock_bh(&mux->lock);
1431 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1432 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1434 KCM_STATS_INCR(mux->stats.psock_unattach);
1436 if (psock->tx_kcm) {
1437 /* psock was reserved. Just mark it finished and we will clean
1438 * up in the kcm paths, we need kcm lock which can not be
1441 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1442 spin_unlock_bh(&mux->lock);
1444 /* We are unattaching a socket that is reserved. Abort the
1445 * socket since we may be out of sync in sending on it. We need
1446 * to do this without the mux lock.
1448 kcm_abort_tx_psock(psock, EPIPE, false);
1450 spin_lock_bh(&mux->lock);
1451 if (!psock->tx_kcm) {
1452 /* psock now unreserved in window mux was unlocked */
1457 /* Commit done before queuing work to process it */
1460 /* Queue tx work to make sure psock->done is handled */
1461 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1462 spin_unlock_bh(&mux->lock);
1465 if (!psock->tx_stopped)
1466 list_del(&psock->psock_avail_list);
1467 list_del(&psock->psock_list);
1469 spin_unlock_bh(&mux->lock);
1472 fput(csk->sk_socket->file);
1473 kmem_cache_free(kcm_psockp, psock);
1479 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1481 struct kcm_sock *kcm = kcm_sk(sock->sk);
1482 struct kcm_mux *mux = kcm->mux;
1483 struct kcm_psock *psock;
1484 struct socket *csock;
1488 csock = sockfd_lookup(info->fd, &err);
1500 spin_lock_bh(&mux->lock);
1502 list_for_each_entry(psock, &mux->psocks, psock_list) {
1503 if (psock->sk != csk)
1506 /* Found the matching psock */
1508 if (psock->unattaching || WARN_ON(psock->done)) {
1513 psock->unattaching = 1;
1515 spin_unlock_bh(&mux->lock);
1517 /* Lower socket lock should already be held */
1518 kcm_unattach(psock);
1524 spin_unlock_bh(&mux->lock);
1531 static struct proto kcm_proto = {
1533 .owner = THIS_MODULE,
1534 .obj_size = sizeof(struct kcm_sock),
1537 /* Clone a kcm socket. */
1538 static struct file *kcm_clone(struct socket *osock)
1540 struct socket *newsock;
1543 newsock = sock_alloc();
1545 return ERR_PTR(-ENFILE);
1547 newsock->type = osock->type;
1548 newsock->ops = osock->ops;
1550 __module_get(newsock->ops->owner);
1552 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1555 sock_release(newsock);
1556 return ERR_PTR(-ENOMEM);
1558 sock_init_data(newsock, newsk);
1559 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1561 return sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1564 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1569 case SIOCKCMATTACH: {
1570 struct kcm_attach info;
1572 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1575 err = kcm_attach_ioctl(sock, &info);
1579 case SIOCKCMUNATTACH: {
1580 struct kcm_unattach info;
1582 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1585 err = kcm_unattach_ioctl(sock, &info);
1589 case SIOCKCMCLONE: {
1590 struct kcm_clone info;
1593 info.fd = get_unused_fd_flags(0);
1594 if (unlikely(info.fd < 0))
1597 file = kcm_clone(sock);
1599 put_unused_fd(info.fd);
1600 return PTR_ERR(file);
1602 if (copy_to_user((void __user *)arg, &info,
1604 put_unused_fd(info.fd);
1608 fd_install(info.fd, file);
1620 static void free_mux(struct rcu_head *rcu)
1622 struct kcm_mux *mux = container_of(rcu,
1623 struct kcm_mux, rcu);
1625 kmem_cache_free(kcm_muxp, mux);
1628 static void release_mux(struct kcm_mux *mux)
1630 struct kcm_net *knet = mux->knet;
1631 struct kcm_psock *psock, *tmp_psock;
1633 /* Release psocks */
1634 list_for_each_entry_safe(psock, tmp_psock,
1635 &mux->psocks, psock_list) {
1636 if (!WARN_ON(psock->unattaching))
1637 kcm_unattach(psock);
1640 if (WARN_ON(mux->psocks_cnt))
1643 __skb_queue_purge(&mux->rx_hold_queue);
1645 mutex_lock(&knet->mutex);
1646 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1647 aggregate_psock_stats(&mux->aggregate_psock_stats,
1648 &knet->aggregate_psock_stats);
1649 aggregate_strp_stats(&mux->aggregate_strp_stats,
1650 &knet->aggregate_strp_stats);
1651 list_del_rcu(&mux->kcm_mux_list);
1653 mutex_unlock(&knet->mutex);
1655 call_rcu(&mux->rcu, free_mux);
1658 static void kcm_done(struct kcm_sock *kcm)
1660 struct kcm_mux *mux = kcm->mux;
1661 struct sock *sk = &kcm->sk;
1664 spin_lock_bh(&mux->rx_lock);
1665 if (kcm->rx_psock) {
1666 /* Cleanup in unreserve_rx_kcm */
1668 kcm->rx_disabled = 1;
1670 spin_unlock_bh(&mux->rx_lock);
1675 list_del(&kcm->wait_rx_list);
1676 /* paired with lockless reads in kcm_rfree() */
1677 WRITE_ONCE(kcm->rx_wait, false);
1679 /* Move any pending receive messages to other kcm sockets */
1680 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1682 spin_unlock_bh(&mux->rx_lock);
1684 if (WARN_ON(sk_rmem_alloc_get(sk)))
1687 /* Detach from MUX */
1688 spin_lock_bh(&mux->lock);
1690 list_del(&kcm->kcm_sock_list);
1691 mux->kcm_socks_cnt--;
1692 socks_cnt = mux->kcm_socks_cnt;
1694 spin_unlock_bh(&mux->lock);
1697 /* We are done with the mux now. */
1701 WARN_ON(kcm->rx_wait);
1706 /* Called by kcm_release to close a KCM socket.
1707 * If this is the last KCM socket on the MUX, destroy the MUX.
1709 static int kcm_release(struct socket *sock)
1711 struct sock *sk = sock->sk;
1712 struct kcm_sock *kcm;
1713 struct kcm_mux *mux;
1714 struct kcm_psock *psock;
1724 kfree_skb(kcm->seq_skb);
1726 /* Purge queue under lock to avoid race condition with tx_work trying
1727 * to act when queue is nonempty. If tx_work runs after this point
1728 * it will just return.
1730 __skb_queue_purge(&sk->sk_write_queue);
1732 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1733 * get a writespace callback. This prevents further work being queued
1734 * from the callback (unbinding the psock occurs after canceling work.
1736 kcm->tx_stopped = 1;
1740 spin_lock_bh(&mux->lock);
1742 /* Take of tx_wait list, after this point there should be no way
1743 * that a psock will be assigned to this kcm.
1745 list_del(&kcm->wait_psock_list);
1746 kcm->tx_wait = false;
1748 spin_unlock_bh(&mux->lock);
1750 /* Cancel work. After this point there should be no outside references
1751 * to the kcm socket.
1753 cancel_work_sync(&kcm->tx_work);
1756 psock = kcm->tx_psock;
1758 /* A psock was reserved, so we need to kill it since it
1759 * may already have some bytes queued from a message. We
1760 * need to do this after removing kcm from tx_wait list.
1762 kcm_abort_tx_psock(psock, EPIPE, false);
1763 unreserve_psock(kcm);
1767 WARN_ON(kcm->tx_wait);
1768 WARN_ON(kcm->tx_psock);
1777 static const struct proto_ops kcm_dgram_ops = {
1779 .owner = THIS_MODULE,
1780 .release = kcm_release,
1781 .bind = sock_no_bind,
1782 .connect = sock_no_connect,
1783 .socketpair = sock_no_socketpair,
1784 .accept = sock_no_accept,
1785 .getname = sock_no_getname,
1786 .poll = datagram_poll,
1788 .listen = sock_no_listen,
1789 .shutdown = sock_no_shutdown,
1790 .setsockopt = kcm_setsockopt,
1791 .getsockopt = kcm_getsockopt,
1792 .sendmsg = kcm_sendmsg,
1793 .recvmsg = kcm_recvmsg,
1794 .mmap = sock_no_mmap,
1795 .splice_eof = kcm_splice_eof,
1796 .sendpage = kcm_sendpage,
1799 static const struct proto_ops kcm_seqpacket_ops = {
1801 .owner = THIS_MODULE,
1802 .release = kcm_release,
1803 .bind = sock_no_bind,
1804 .connect = sock_no_connect,
1805 .socketpair = sock_no_socketpair,
1806 .accept = sock_no_accept,
1807 .getname = sock_no_getname,
1808 .poll = datagram_poll,
1810 .listen = sock_no_listen,
1811 .shutdown = sock_no_shutdown,
1812 .setsockopt = kcm_setsockopt,
1813 .getsockopt = kcm_getsockopt,
1814 .sendmsg = kcm_sendmsg,
1815 .recvmsg = kcm_recvmsg,
1816 .mmap = sock_no_mmap,
1817 .splice_eof = kcm_splice_eof,
1818 .sendpage = kcm_sendpage,
1819 .splice_read = kcm_splice_read,
1822 /* Create proto operation for kcm sockets */
1823 static int kcm_create(struct net *net, struct socket *sock,
1824 int protocol, int kern)
1826 struct kcm_net *knet = net_generic(net, kcm_net_id);
1828 struct kcm_mux *mux;
1830 switch (sock->type) {
1832 sock->ops = &kcm_dgram_ops;
1834 case SOCK_SEQPACKET:
1835 sock->ops = &kcm_seqpacket_ops;
1838 return -ESOCKTNOSUPPORT;
1841 if (protocol != KCMPROTO_CONNECTED)
1842 return -EPROTONOSUPPORT;
1844 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1848 /* Allocate a kcm mux, shared between KCM sockets */
1849 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1855 spin_lock_init(&mux->lock);
1856 spin_lock_init(&mux->rx_lock);
1857 INIT_LIST_HEAD(&mux->kcm_socks);
1858 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1859 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1861 INIT_LIST_HEAD(&mux->psocks);
1862 INIT_LIST_HEAD(&mux->psocks_ready);
1863 INIT_LIST_HEAD(&mux->psocks_avail);
1867 /* Add new MUX to list */
1868 mutex_lock(&knet->mutex);
1869 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1871 mutex_unlock(&knet->mutex);
1873 skb_queue_head_init(&mux->rx_hold_queue);
1875 /* Init KCM socket */
1876 sock_init_data(sock, sk);
1877 init_kcm_sock(kcm_sk(sk), mux);
1882 static const struct net_proto_family kcm_family_ops = {
1884 .create = kcm_create,
1885 .owner = THIS_MODULE,
1888 static __net_init int kcm_init_net(struct net *net)
1890 struct kcm_net *knet = net_generic(net, kcm_net_id);
1892 INIT_LIST_HEAD_RCU(&knet->mux_list);
1893 mutex_init(&knet->mutex);
1898 static __net_exit void kcm_exit_net(struct net *net)
1900 struct kcm_net *knet = net_generic(net, kcm_net_id);
1902 /* All KCM sockets should be closed at this point, which should mean
1903 * that all multiplexors and psocks have been destroyed.
1905 WARN_ON(!list_empty(&knet->mux_list));
1908 static struct pernet_operations kcm_net_ops = {
1909 .init = kcm_init_net,
1910 .exit = kcm_exit_net,
1912 .size = sizeof(struct kcm_net),
1915 static int __init kcm_init(void)
1919 kcm_muxp = kmem_cache_create("kcm_mux_cache",
1920 sizeof(struct kcm_mux), 0,
1921 SLAB_HWCACHE_ALIGN, NULL);
1925 kcm_psockp = kmem_cache_create("kcm_psock_cache",
1926 sizeof(struct kcm_psock), 0,
1927 SLAB_HWCACHE_ALIGN, NULL);
1931 kcm_wq = create_singlethread_workqueue("kkcmd");
1935 err = proto_register(&kcm_proto, 1);
1939 err = register_pernet_device(&kcm_net_ops);
1943 err = sock_register(&kcm_family_ops);
1945 goto sock_register_fail;
1947 err = kcm_proc_init();
1949 goto proc_init_fail;
1954 sock_unregister(PF_KCM);
1957 unregister_pernet_device(&kcm_net_ops);
1960 proto_unregister(&kcm_proto);
1963 kmem_cache_destroy(kcm_muxp);
1964 kmem_cache_destroy(kcm_psockp);
1967 destroy_workqueue(kcm_wq);
1972 static void __exit kcm_exit(void)
1975 sock_unregister(PF_KCM);
1976 unregister_pernet_device(&kcm_net_ops);
1977 proto_unregister(&kcm_proto);
1978 destroy_workqueue(kcm_wq);
1980 kmem_cache_destroy(kcm_muxp);
1981 kmem_cache_destroy(kcm_psockp);
1984 module_init(kcm_init);
1985 module_exit(kcm_exit);
1987 MODULE_LICENSE("GPL");
1988 MODULE_ALIAS_NETPROTO(PF_KCM);