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>
28 #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_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_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_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_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 if (msg_data_left(msg)) {
933 /* New message, alloc head skb */
934 head = alloc_skb(0, sk->sk_allocation);
937 err = sk_stream_wait_memory(sk, &timeo);
941 head = alloc_skb(0, sk->sk_allocation);
946 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
947 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
949 skb->ip_summed = CHECKSUM_UNNECESSARY;
953 while (msg_data_left(msg)) {
955 int i = skb_shinfo(skb)->nr_frags;
956 struct page_frag *pfrag = sk_page_frag(sk);
958 if (!sk_page_frag_refill(sk, pfrag))
959 goto wait_for_memory;
961 if (!skb_can_coalesce(skb, i, pfrag->page,
963 if (i == MAX_SKB_FRAGS) {
964 struct sk_buff *tskb;
966 tskb = alloc_skb(0, sk->sk_allocation);
968 goto wait_for_memory;
971 skb_shinfo(head)->frag_list = tskb;
976 skb->ip_summed = CHECKSUM_UNNECESSARY;
982 copy = min_t(int, msg_data_left(msg),
983 pfrag->size - pfrag->offset);
985 if (!sk_wmem_schedule(sk, copy))
986 goto wait_for_memory;
988 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
995 /* Update the skb. */
997 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
999 skb_fill_page_desc(skb, i, pfrag->page,
1000 pfrag->offset, copy);
1001 get_page(pfrag->page);
1004 pfrag->offset += copy;
1008 head->data_len += copy;
1015 err = sk_stream_wait_memory(sk, &timeo);
1021 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1024 /* Message complete, queue it on send buffer */
1025 __skb_queue_tail(&sk->sk_write_queue, head);
1026 kcm->seq_skb = NULL;
1027 KCM_STATS_INCR(kcm->stats.tx_msgs);
1030 if (msg->msg_flags & MSG_BATCH) {
1031 kcm->tx_wait_more = true;
1032 } else if (kcm->tx_wait_more || not_busy) {
1033 err = kcm_write_msgs(kcm);
1035 /* We got a hard error in write_msgs but have
1036 * already queued this message. Report an error
1037 * in the socket, but don't affect return value
1040 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1041 report_csk_error(&kcm->sk, -err);
1045 /* Message not complete, save state */
1048 kcm->seq_skb = head;
1049 kcm_tx_msg(head)->last_skb = skb;
1053 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1061 if (copied && sock->type == SOCK_SEQPACKET) {
1062 /* Wrote some bytes before encountering an
1063 * error, return partial success.
1065 goto partial_message;
1068 if (head != kcm->seq_skb)
1071 err = sk_stream_error(sk, msg->msg_flags, err);
1073 /* make sure we wake any epoll edge trigger waiter */
1074 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1075 sk->sk_write_space(sk);
1081 static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1082 long timeo, int *err)
1084 struct sk_buff *skb;
1086 while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1088 *err = sock_error(sk);
1092 if (sock_flag(sk, SOCK_DONE))
1095 if ((flags & MSG_DONTWAIT) || !timeo) {
1100 sk_wait_data(sk, &timeo, NULL);
1102 /* Handle signals */
1103 if (signal_pending(current)) {
1104 *err = sock_intr_errno(timeo);
1112 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1113 size_t len, int flags)
1115 struct sock *sk = sock->sk;
1116 struct kcm_sock *kcm = kcm_sk(sk);
1119 struct strp_rx_msg *rxm;
1121 struct sk_buff *skb;
1123 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1127 skb = kcm_wait_data(sk, flags, timeo, &err);
1131 /* Okay, have a message on the receive queue */
1133 rxm = strp_rx_msg(skb);
1135 if (len > rxm->full_len)
1136 len = rxm->full_len;
1138 err = skb_copy_datagram_msg(skb, rxm->offset, msg, len);
1143 if (likely(!(flags & MSG_PEEK))) {
1144 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1145 if (copied < rxm->full_len) {
1146 if (sock->type == SOCK_DGRAM) {
1147 /* Truncated message */
1148 msg->msg_flags |= MSG_TRUNC;
1151 rxm->offset += copied;
1152 rxm->full_len -= copied;
1155 /* Finished with message */
1156 msg->msg_flags |= MSG_EOR;
1157 KCM_STATS_INCR(kcm->stats.rx_msgs);
1158 skb_unlink(skb, &sk->sk_receive_queue);
1166 return copied ? : err;
1169 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1170 struct pipe_inode_info *pipe, size_t len,
1173 struct sock *sk = sock->sk;
1174 struct kcm_sock *kcm = kcm_sk(sk);
1176 struct strp_rx_msg *rxm;
1179 struct sk_buff *skb;
1181 /* Only support splice for SOCKSEQPACKET */
1183 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1187 skb = kcm_wait_data(sk, flags, timeo, &err);
1191 /* Okay, have a message on the receive queue */
1193 rxm = strp_rx_msg(skb);
1195 if (len > rxm->full_len)
1196 len = rxm->full_len;
1198 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, len, flags);
1204 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1206 rxm->offset += copied;
1207 rxm->full_len -= copied;
1209 /* We have no way to return MSG_EOR. If all the bytes have been
1210 * read we still leave the message in the receive socket buffer.
1211 * A subsequent recvmsg needs to be done to return MSG_EOR and
1212 * finish reading the message.
1225 /* kcm sock lock held */
1226 static void kcm_recv_disable(struct kcm_sock *kcm)
1228 struct kcm_mux *mux = kcm->mux;
1230 if (kcm->rx_disabled)
1233 spin_lock_bh(&mux->rx_lock);
1235 kcm->rx_disabled = 1;
1237 /* If a psock is reserved we'll do cleanup in unreserve */
1238 if (!kcm->rx_psock) {
1240 list_del(&kcm->wait_rx_list);
1241 kcm->rx_wait = false;
1244 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1247 spin_unlock_bh(&mux->rx_lock);
1250 /* kcm sock lock held */
1251 static void kcm_recv_enable(struct kcm_sock *kcm)
1253 struct kcm_mux *mux = kcm->mux;
1255 if (!kcm->rx_disabled)
1258 spin_lock_bh(&mux->rx_lock);
1260 kcm->rx_disabled = 0;
1263 spin_unlock_bh(&mux->rx_lock);
1266 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1267 char __user *optval, unsigned int optlen)
1269 struct kcm_sock *kcm = kcm_sk(sock->sk);
1273 if (level != SOL_KCM)
1274 return -ENOPROTOOPT;
1276 if (optlen < sizeof(int))
1279 if (get_user(val, (int __user *)optval))
1282 valbool = val ? 1 : 0;
1285 case KCM_RECV_DISABLE:
1286 lock_sock(&kcm->sk);
1288 kcm_recv_disable(kcm);
1290 kcm_recv_enable(kcm);
1291 release_sock(&kcm->sk);
1300 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1301 char __user *optval, int __user *optlen)
1303 struct kcm_sock *kcm = kcm_sk(sock->sk);
1306 if (level != SOL_KCM)
1307 return -ENOPROTOOPT;
1309 if (get_user(len, optlen))
1312 len = min_t(unsigned int, len, sizeof(int));
1317 case KCM_RECV_DISABLE:
1318 val = kcm->rx_disabled;
1321 return -ENOPROTOOPT;
1324 if (put_user(len, optlen))
1326 if (copy_to_user(optval, &val, len))
1331 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1333 struct kcm_sock *tkcm;
1334 struct list_head *head;
1337 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1338 * we set sk_state, otherwise epoll_wait always returns right away with
1341 kcm->sk.sk_state = TCP_ESTABLISHED;
1343 /* Add to mux's kcm sockets list */
1345 spin_lock_bh(&mux->lock);
1347 head = &mux->kcm_socks;
1348 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1349 if (tkcm->index != index)
1351 head = &tkcm->kcm_sock_list;
1355 list_add(&kcm->kcm_sock_list, head);
1358 mux->kcm_socks_cnt++;
1359 spin_unlock_bh(&mux->lock);
1361 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1363 spin_lock_bh(&mux->rx_lock);
1365 spin_unlock_bh(&mux->rx_lock);
1368 static int kcm_attach(struct socket *sock, struct socket *csock,
1369 struct bpf_prog *prog)
1371 struct kcm_sock *kcm = kcm_sk(sock->sk);
1372 struct kcm_mux *mux = kcm->mux;
1374 struct kcm_psock *psock = NULL, *tpsock;
1375 struct list_head *head;
1377 struct strp_callbacks cb;
1384 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1390 psock->bpf_prog = prog;
1392 cb.rcv_msg = kcm_rcv_strparser;
1393 cb.abort_parser = NULL;
1394 cb.parse_msg = kcm_parse_func_strparser;
1395 cb.read_sock_done = kcm_read_sock_done;
1397 err = strp_init(&psock->strp, csk, &cb);
1399 kmem_cache_free(kcm_psockp, psock);
1405 write_lock_bh(&csk->sk_callback_lock);
1406 psock->save_data_ready = csk->sk_data_ready;
1407 psock->save_write_space = csk->sk_write_space;
1408 psock->save_state_change = csk->sk_state_change;
1409 csk->sk_user_data = psock;
1410 csk->sk_data_ready = psock_data_ready;
1411 csk->sk_write_space = psock_write_space;
1412 csk->sk_state_change = psock_state_change;
1413 write_unlock_bh(&csk->sk_callback_lock);
1415 /* Finished initialization, now add the psock to the MUX. */
1416 spin_lock_bh(&mux->lock);
1417 head = &mux->psocks;
1418 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1419 if (tpsock->index != index)
1421 head = &tpsock->psock_list;
1425 list_add(&psock->psock_list, head);
1426 psock->index = index;
1428 KCM_STATS_INCR(mux->stats.psock_attach);
1430 psock_now_avail(psock);
1431 spin_unlock_bh(&mux->lock);
1433 /* Schedule RX work in case there are already bytes queued */
1434 strp_check_rcv(&psock->strp);
1439 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1441 struct socket *csock;
1442 struct bpf_prog *prog;
1445 csock = sockfd_lookup(info->fd, &err);
1449 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1451 err = PTR_ERR(prog);
1455 err = kcm_attach(sock, csock, prog);
1461 /* Keep reference on file also */
1469 static void kcm_unattach(struct kcm_psock *psock)
1471 struct sock *csk = psock->sk;
1472 struct kcm_mux *mux = psock->mux;
1476 /* Stop getting callbacks from TCP socket. After this there should
1477 * be no way to reserve a kcm for this psock.
1479 write_lock_bh(&csk->sk_callback_lock);
1480 csk->sk_user_data = NULL;
1481 csk->sk_data_ready = psock->save_data_ready;
1482 csk->sk_write_space = psock->save_write_space;
1483 csk->sk_state_change = psock->save_state_change;
1484 strp_stop(&psock->strp);
1486 if (WARN_ON(psock->rx_kcm)) {
1487 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 int kcm_clone(struct socket *osock, struct kcm_clone *info,
1624 struct socket **newsockp)
1626 struct socket *newsock;
1628 struct file *newfile;
1632 newsock = sock_alloc();
1636 newsock->type = osock->type;
1637 newsock->ops = osock->ops;
1639 __module_get(newsock->ops->owner);
1641 newfd = get_unused_fd_flags(0);
1642 if (unlikely(newfd < 0)) {
1647 newfile = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1648 if (unlikely(IS_ERR(newfile))) {
1649 err = PTR_ERR(newfile);
1650 goto out_sock_alloc_fail;
1653 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1657 goto out_sk_alloc_fail;
1660 sock_init_data(newsock, newsk);
1661 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1663 fd_install(newfd, newfile);
1664 *newsockp = newsock;
1671 out_sock_alloc_fail:
1672 put_unused_fd(newfd);
1674 sock_release(newsock);
1679 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1684 case SIOCKCMATTACH: {
1685 struct kcm_attach info;
1687 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1690 err = kcm_attach_ioctl(sock, &info);
1694 case SIOCKCMUNATTACH: {
1695 struct kcm_unattach info;
1697 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1700 err = kcm_unattach_ioctl(sock, &info);
1704 case SIOCKCMCLONE: {
1705 struct kcm_clone info;
1706 struct socket *newsock = NULL;
1708 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1711 err = kcm_clone(sock, &info, &newsock);
1714 if (copy_to_user((void __user *)arg, &info,
1731 static void free_mux(struct rcu_head *rcu)
1733 struct kcm_mux *mux = container_of(rcu,
1734 struct kcm_mux, rcu);
1736 kmem_cache_free(kcm_muxp, mux);
1739 static void release_mux(struct kcm_mux *mux)
1741 struct kcm_net *knet = mux->knet;
1742 struct kcm_psock *psock, *tmp_psock;
1744 /* Release psocks */
1745 list_for_each_entry_safe(psock, tmp_psock,
1746 &mux->psocks, psock_list) {
1747 if (!WARN_ON(psock->unattaching))
1748 kcm_unattach(psock);
1751 if (WARN_ON(mux->psocks_cnt))
1754 __skb_queue_purge(&mux->rx_hold_queue);
1756 mutex_lock(&knet->mutex);
1757 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1758 aggregate_psock_stats(&mux->aggregate_psock_stats,
1759 &knet->aggregate_psock_stats);
1760 aggregate_strp_stats(&mux->aggregate_strp_stats,
1761 &knet->aggregate_strp_stats);
1762 list_del_rcu(&mux->kcm_mux_list);
1764 mutex_unlock(&knet->mutex);
1766 call_rcu(&mux->rcu, free_mux);
1769 static void kcm_done(struct kcm_sock *kcm)
1771 struct kcm_mux *mux = kcm->mux;
1772 struct sock *sk = &kcm->sk;
1775 spin_lock_bh(&mux->rx_lock);
1776 if (kcm->rx_psock) {
1777 /* Cleanup in unreserve_rx_kcm */
1779 kcm->rx_disabled = 1;
1781 spin_unlock_bh(&mux->rx_lock);
1786 list_del(&kcm->wait_rx_list);
1787 kcm->rx_wait = false;
1789 /* Move any pending receive messages to other kcm sockets */
1790 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1792 spin_unlock_bh(&mux->rx_lock);
1794 if (WARN_ON(sk_rmem_alloc_get(sk)))
1797 /* Detach from MUX */
1798 spin_lock_bh(&mux->lock);
1800 list_del(&kcm->kcm_sock_list);
1801 mux->kcm_socks_cnt--;
1802 socks_cnt = mux->kcm_socks_cnt;
1804 spin_unlock_bh(&mux->lock);
1807 /* We are done with the mux now. */
1811 WARN_ON(kcm->rx_wait);
1816 /* Called by kcm_release to close a KCM socket.
1817 * If this is the last KCM socket on the MUX, destroy the MUX.
1819 static int kcm_release(struct socket *sock)
1821 struct sock *sk = sock->sk;
1822 struct kcm_sock *kcm;
1823 struct kcm_mux *mux;
1824 struct kcm_psock *psock;
1833 kfree_skb(kcm->seq_skb);
1836 /* Purge queue under lock to avoid race condition with tx_work trying
1837 * to act when queue is nonempty. If tx_work runs after this point
1838 * it will just return.
1840 __skb_queue_purge(&sk->sk_write_queue);
1842 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1843 * get a writespace callback. This prevents further work being queued
1844 * from the callback (unbinding the psock occurs after canceling work.
1846 kcm->tx_stopped = 1;
1850 spin_lock_bh(&mux->lock);
1852 /* Take of tx_wait list, after this point there should be no way
1853 * that a psock will be assigned to this kcm.
1855 list_del(&kcm->wait_psock_list);
1856 kcm->tx_wait = false;
1858 spin_unlock_bh(&mux->lock);
1860 /* Cancel work. After this point there should be no outside references
1861 * to the kcm socket.
1863 cancel_work_sync(&kcm->tx_work);
1866 psock = kcm->tx_psock;
1868 /* A psock was reserved, so we need to kill it since it
1869 * may already have some bytes queued from a message. We
1870 * need to do this after removing kcm from tx_wait list.
1872 kcm_abort_tx_psock(psock, EPIPE, false);
1873 unreserve_psock(kcm);
1877 WARN_ON(kcm->tx_wait);
1878 WARN_ON(kcm->tx_psock);
1887 static const struct proto_ops kcm_dgram_ops = {
1889 .owner = THIS_MODULE,
1890 .release = kcm_release,
1891 .bind = sock_no_bind,
1892 .connect = sock_no_connect,
1893 .socketpair = sock_no_socketpair,
1894 .accept = sock_no_accept,
1895 .getname = sock_no_getname,
1896 .poll = datagram_poll,
1898 .listen = sock_no_listen,
1899 .shutdown = sock_no_shutdown,
1900 .setsockopt = kcm_setsockopt,
1901 .getsockopt = kcm_getsockopt,
1902 .sendmsg = kcm_sendmsg,
1903 .recvmsg = kcm_recvmsg,
1904 .mmap = sock_no_mmap,
1905 .sendpage = kcm_sendpage,
1908 static const struct proto_ops kcm_seqpacket_ops = {
1910 .owner = THIS_MODULE,
1911 .release = kcm_release,
1912 .bind = sock_no_bind,
1913 .connect = sock_no_connect,
1914 .socketpair = sock_no_socketpair,
1915 .accept = sock_no_accept,
1916 .getname = sock_no_getname,
1917 .poll = datagram_poll,
1919 .listen = sock_no_listen,
1920 .shutdown = sock_no_shutdown,
1921 .setsockopt = kcm_setsockopt,
1922 .getsockopt = kcm_getsockopt,
1923 .sendmsg = kcm_sendmsg,
1924 .recvmsg = kcm_recvmsg,
1925 .mmap = sock_no_mmap,
1926 .sendpage = kcm_sendpage,
1927 .splice_read = kcm_splice_read,
1930 /* Create proto operation for kcm sockets */
1931 static int kcm_create(struct net *net, struct socket *sock,
1932 int protocol, int kern)
1934 struct kcm_net *knet = net_generic(net, kcm_net_id);
1936 struct kcm_mux *mux;
1938 switch (sock->type) {
1940 sock->ops = &kcm_dgram_ops;
1942 case SOCK_SEQPACKET:
1943 sock->ops = &kcm_seqpacket_ops;
1946 return -ESOCKTNOSUPPORT;
1949 if (protocol != KCMPROTO_CONNECTED)
1950 return -EPROTONOSUPPORT;
1952 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1956 /* Allocate a kcm mux, shared between KCM sockets */
1957 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1963 spin_lock_init(&mux->lock);
1964 spin_lock_init(&mux->rx_lock);
1965 INIT_LIST_HEAD(&mux->kcm_socks);
1966 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1967 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1969 INIT_LIST_HEAD(&mux->psocks);
1970 INIT_LIST_HEAD(&mux->psocks_ready);
1971 INIT_LIST_HEAD(&mux->psocks_avail);
1975 /* Add new MUX to list */
1976 mutex_lock(&knet->mutex);
1977 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1979 mutex_unlock(&knet->mutex);
1981 skb_queue_head_init(&mux->rx_hold_queue);
1983 /* Init KCM socket */
1984 sock_init_data(sock, sk);
1985 init_kcm_sock(kcm_sk(sk), mux);
1990 static struct net_proto_family kcm_family_ops = {
1992 .create = kcm_create,
1993 .owner = THIS_MODULE,
1996 static __net_init int kcm_init_net(struct net *net)
1998 struct kcm_net *knet = net_generic(net, kcm_net_id);
2000 INIT_LIST_HEAD_RCU(&knet->mux_list);
2001 mutex_init(&knet->mutex);
2006 static __net_exit void kcm_exit_net(struct net *net)
2008 struct kcm_net *knet = net_generic(net, kcm_net_id);
2010 /* All KCM sockets should be closed at this point, which should mean
2011 * that all multiplexors and psocks have been destroyed.
2013 WARN_ON(!list_empty(&knet->mux_list));
2016 static struct pernet_operations kcm_net_ops = {
2017 .init = kcm_init_net,
2018 .exit = kcm_exit_net,
2020 .size = sizeof(struct kcm_net),
2023 static int __init kcm_init(void)
2027 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2028 sizeof(struct kcm_mux), 0,
2029 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2033 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2034 sizeof(struct kcm_psock), 0,
2035 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2039 kcm_wq = create_singlethread_workqueue("kkcmd");
2043 err = proto_register(&kcm_proto, 1);
2047 err = sock_register(&kcm_family_ops);
2049 goto sock_register_fail;
2051 err = register_pernet_device(&kcm_net_ops);
2055 err = kcm_proc_init();
2057 goto proc_init_fail;
2062 unregister_pernet_device(&kcm_net_ops);
2065 sock_unregister(PF_KCM);
2068 proto_unregister(&kcm_proto);
2071 kmem_cache_destroy(kcm_muxp);
2072 kmem_cache_destroy(kcm_psockp);
2075 destroy_workqueue(kcm_wq);
2080 static void __exit kcm_exit(void)
2083 unregister_pernet_device(&kcm_net_ops);
2084 sock_unregister(PF_KCM);
2085 proto_unregister(&kcm_proto);
2086 destroy_workqueue(kcm_wq);
2088 kmem_cache_destroy(kcm_muxp);
2089 kmem_cache_destroy(kcm_psockp);
2092 module_init(kcm_init);
2093 module_exit(kcm_exit);
2095 MODULE_LICENSE("GPL");
2096 MODULE_ALIAS_NETPROTO(PF_KCM);