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>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/net.h>
16 #include <linux/netdevice.h>
17 #include <linux/poll.h>
18 #include <linux/rculist.h>
19 #include <linux/skbuff.h>
20 #include <linux/socket.h>
21 #include <linux/uaccess.h>
22 #include <linux/workqueue.h>
23 #include <linux/syscalls.h>
24 #include <linux/sched/signal.h>
27 #include <net/netns/generic.h>
29 #include <uapi/linux/kcm.h>
31 unsigned int kcm_net_id;
33 static struct kmem_cache *kcm_psockp __read_mostly;
34 static struct kmem_cache *kcm_muxp __read_mostly;
35 static struct workqueue_struct *kcm_wq;
37 static inline struct kcm_sock *kcm_sk(const struct sock *sk)
39 return (struct kcm_sock *)sk;
42 static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
44 return (struct kcm_tx_msg *)skb->cb;
47 static void report_csk_error(struct sock *csk, int err)
53 static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
56 struct sock *csk = psock->sk;
57 struct kcm_mux *mux = psock->mux;
59 /* Unrecoverable error in transmit */
61 spin_lock_bh(&mux->lock);
63 if (psock->tx_stopped) {
64 spin_unlock_bh(&mux->lock);
68 psock->tx_stopped = 1;
69 KCM_STATS_INCR(psock->stats.tx_aborts);
72 /* Take off psocks_avail list */
73 list_del(&psock->psock_avail_list);
74 } else if (wakeup_kcm) {
75 /* In this case psock is being aborted while outside of
76 * write_msgs and psock is reserved. Schedule tx_work
77 * to handle the failure there. Need to commit tx_stopped
78 * before queuing work.
82 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
85 spin_unlock_bh(&mux->lock);
87 /* Report error on lower socket */
88 report_csk_error(csk, err);
91 /* RX mux lock held. */
92 static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
93 struct kcm_psock *psock)
95 STRP_STATS_ADD(mux->stats.rx_bytes,
96 psock->strp.stats.bytes -
97 psock->saved_rx_bytes);
99 psock->strp.stats.msgs - psock->saved_rx_msgs;
100 psock->saved_rx_msgs = psock->strp.stats.msgs;
101 psock->saved_rx_bytes = psock->strp.stats.bytes;
104 static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
105 struct kcm_psock *psock)
107 KCM_STATS_ADD(mux->stats.tx_bytes,
108 psock->stats.tx_bytes - psock->saved_tx_bytes);
109 mux->stats.tx_msgs +=
110 psock->stats.tx_msgs - psock->saved_tx_msgs;
111 psock->saved_tx_msgs = psock->stats.tx_msgs;
112 psock->saved_tx_bytes = psock->stats.tx_bytes;
115 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
117 /* KCM is ready to receive messages on its queue-- either the KCM is new or
118 * has become unblocked after being blocked on full socket buffer. Queue any
119 * pending ready messages on a psock. RX mux lock held.
121 static void kcm_rcv_ready(struct kcm_sock *kcm)
123 struct kcm_mux *mux = kcm->mux;
124 struct kcm_psock *psock;
127 if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
130 while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) {
131 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
132 /* Assuming buffer limit has been reached */
133 skb_queue_head(&mux->rx_hold_queue, skb);
134 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
139 while (!list_empty(&mux->psocks_ready)) {
140 psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
143 if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) {
144 /* Assuming buffer limit has been reached */
145 WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
149 /* Consumed the ready message on the psock. Schedule rx_work to
152 list_del(&psock->psock_ready_list);
153 psock->ready_rx_msg = NULL;
154 /* Commit clearing of ready_rx_msg for queuing work */
157 strp_unpause(&psock->strp);
158 strp_check_rcv(&psock->strp);
161 /* Buffer limit is okay now, add to ready list */
162 list_add_tail(&kcm->wait_rx_list,
163 &kcm->mux->kcm_rx_waiters);
164 /* paired with lockless reads in kcm_rfree() */
165 WRITE_ONCE(kcm->rx_wait, true);
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 (!READ_ONCE(kcm->rx_wait) && !READ_ONCE(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 /* paired with lockless reads in kcm_rfree() */
241 WRITE_ONCE(kcm->rx_wait, false);
243 /* Commit rx_wait to read in kcm_free */
251 /* Lower sock lock held */
252 static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
253 struct sk_buff *head)
255 struct kcm_mux *mux = psock->mux;
256 struct kcm_sock *kcm;
258 WARN_ON(psock->ready_rx_msg);
261 return psock->rx_kcm;
263 spin_lock_bh(&mux->rx_lock);
266 spin_unlock_bh(&mux->rx_lock);
267 return psock->rx_kcm;
270 kcm_update_rx_mux_stats(mux, psock);
272 if (list_empty(&mux->kcm_rx_waiters)) {
273 psock->ready_rx_msg = head;
274 strp_pause(&psock->strp);
275 list_add_tail(&psock->psock_ready_list,
277 spin_unlock_bh(&mux->rx_lock);
281 kcm = list_first_entry(&mux->kcm_rx_waiters,
282 struct kcm_sock, wait_rx_list);
283 list_del(&kcm->wait_rx_list);
284 /* paired with lockless reads in kcm_rfree() */
285 WRITE_ONCE(kcm->rx_wait, false);
288 /* paired with lockless reads in kcm_rfree() */
289 WRITE_ONCE(kcm->rx_psock, psock);
291 spin_unlock_bh(&mux->rx_lock);
296 static void kcm_done(struct kcm_sock *kcm);
298 static void kcm_done_work(struct work_struct *w)
300 kcm_done(container_of(w, struct kcm_sock, done_work));
303 /* Lower sock held */
304 static void unreserve_rx_kcm(struct kcm_psock *psock,
307 struct kcm_sock *kcm = psock->rx_kcm;
308 struct kcm_mux *mux = psock->mux;
313 spin_lock_bh(&mux->rx_lock);
315 psock->rx_kcm = NULL;
316 /* paired with lockless reads in kcm_rfree() */
317 WRITE_ONCE(kcm->rx_psock, NULL);
319 /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
324 if (unlikely(kcm->done)) {
325 spin_unlock_bh(&mux->rx_lock);
327 /* Need to run kcm_done in a task since we need to qcquire
328 * callback locks which may already be held here.
330 INIT_WORK(&kcm->done_work, kcm_done_work);
331 schedule_work(&kcm->done_work);
335 if (unlikely(kcm->rx_disabled)) {
336 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
337 } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
338 /* Check for degenerative race with rx_wait that all
339 * data was dequeued (accounted for in kcm_rfree).
343 spin_unlock_bh(&mux->rx_lock);
346 /* Lower sock lock held */
347 static void psock_data_ready(struct sock *sk)
349 struct kcm_psock *psock;
351 read_lock_bh(&sk->sk_callback_lock);
353 psock = (struct kcm_psock *)sk->sk_user_data;
355 strp_data_ready(&psock->strp);
357 read_unlock_bh(&sk->sk_callback_lock);
360 /* Called with lower sock held */
361 static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
363 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
364 struct kcm_sock *kcm;
367 kcm = reserve_rx_kcm(psock, skb);
369 /* Unable to reserve a KCM, message is held in psock and strp
375 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
376 /* Should mean socket buffer full */
377 unreserve_rx_kcm(psock, false);
382 static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
384 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
385 struct bpf_prog *prog = psock->bpf_prog;
388 res = bpf_prog_run_pin_on_cpu(prog, skb);
392 static int kcm_read_sock_done(struct strparser *strp, int err)
394 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
396 unreserve_rx_kcm(psock, true);
401 static void psock_state_change(struct sock *sk)
403 /* TCP only does a EPOLLIN for a half close. Do a EPOLLHUP here
404 * since application will normally not poll with EPOLLIN
405 * on the TCP sockets.
408 report_csk_error(sk, EPIPE);
411 static void psock_write_space(struct sock *sk)
413 struct kcm_psock *psock;
415 struct kcm_sock *kcm;
417 read_lock_bh(&sk->sk_callback_lock);
419 psock = (struct kcm_psock *)sk->sk_user_data;
420 if (unlikely(!psock))
424 spin_lock_bh(&mux->lock);
426 /* Check if the socket is reserved so someone is waiting for sending. */
428 if (kcm && !unlikely(kcm->tx_stopped))
429 queue_work(kcm_wq, &kcm->tx_work);
431 spin_unlock_bh(&mux->lock);
433 read_unlock_bh(&sk->sk_callback_lock);
436 static void unreserve_psock(struct kcm_sock *kcm);
438 /* kcm sock is locked. */
439 static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
441 struct kcm_mux *mux = kcm->mux;
442 struct kcm_psock *psock;
444 psock = kcm->tx_psock;
446 smp_rmb(); /* Must read tx_psock before tx_wait */
449 WARN_ON(kcm->tx_wait);
450 if (unlikely(psock->tx_stopped))
451 unreserve_psock(kcm);
453 return kcm->tx_psock;
456 spin_lock_bh(&mux->lock);
458 /* Check again under lock to see if psock was reserved for this
459 * psock via psock_unreserve.
461 psock = kcm->tx_psock;
462 if (unlikely(psock)) {
463 WARN_ON(kcm->tx_wait);
464 spin_unlock_bh(&mux->lock);
465 return kcm->tx_psock;
468 if (!list_empty(&mux->psocks_avail)) {
469 psock = list_first_entry(&mux->psocks_avail,
472 list_del(&psock->psock_avail_list);
474 list_del(&kcm->wait_psock_list);
475 kcm->tx_wait = false;
477 kcm->tx_psock = psock;
479 KCM_STATS_INCR(psock->stats.reserved);
480 } else if (!kcm->tx_wait) {
481 list_add_tail(&kcm->wait_psock_list,
482 &mux->kcm_tx_waiters);
486 spin_unlock_bh(&mux->lock);
492 static void psock_now_avail(struct kcm_psock *psock)
494 struct kcm_mux *mux = psock->mux;
495 struct kcm_sock *kcm;
497 if (list_empty(&mux->kcm_tx_waiters)) {
498 list_add_tail(&psock->psock_avail_list,
501 kcm = list_first_entry(&mux->kcm_tx_waiters,
504 list_del(&kcm->wait_psock_list);
505 kcm->tx_wait = false;
508 /* Commit before changing tx_psock since that is read in
509 * reserve_psock before queuing work.
513 kcm->tx_psock = psock;
514 KCM_STATS_INCR(psock->stats.reserved);
515 queue_work(kcm_wq, &kcm->tx_work);
519 /* kcm sock is locked. */
520 static void unreserve_psock(struct kcm_sock *kcm)
522 struct kcm_psock *psock;
523 struct kcm_mux *mux = kcm->mux;
525 spin_lock_bh(&mux->lock);
527 psock = kcm->tx_psock;
529 if (WARN_ON(!psock)) {
530 spin_unlock_bh(&mux->lock);
534 smp_rmb(); /* Read tx_psock before tx_wait */
536 kcm_update_tx_mux_stats(mux, psock);
538 WARN_ON(kcm->tx_wait);
540 kcm->tx_psock = NULL;
541 psock->tx_kcm = NULL;
542 KCM_STATS_INCR(psock->stats.unreserved);
544 if (unlikely(psock->tx_stopped)) {
547 list_del(&psock->psock_list);
550 fput(psock->sk->sk_socket->file);
551 kmem_cache_free(kcm_psockp, psock);
554 /* Don't put back on available list */
556 spin_unlock_bh(&mux->lock);
561 psock_now_avail(psock);
563 spin_unlock_bh(&mux->lock);
566 static void kcm_report_tx_retry(struct kcm_sock *kcm)
568 struct kcm_mux *mux = kcm->mux;
570 spin_lock_bh(&mux->lock);
571 KCM_STATS_INCR(mux->stats.tx_retries);
572 spin_unlock_bh(&mux->lock);
575 /* Write any messages ready on the kcm socket. Called with kcm sock lock
576 * held. Return bytes actually sent or error.
578 static int kcm_write_msgs(struct kcm_sock *kcm)
580 struct sock *sk = &kcm->sk;
581 struct kcm_psock *psock;
582 struct sk_buff *skb, *head;
583 struct kcm_tx_msg *txm;
584 unsigned short fragidx, frag_offset;
585 unsigned int sent, total_sent = 0;
588 kcm->tx_wait_more = false;
589 psock = kcm->tx_psock;
590 if (unlikely(psock && psock->tx_stopped)) {
591 /* A reserved psock was aborted asynchronously. Unreserve
592 * it and we'll retry the message.
594 unreserve_psock(kcm);
595 kcm_report_tx_retry(kcm);
596 if (skb_queue_empty(&sk->sk_write_queue))
599 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
601 } else if (skb_queue_empty(&sk->sk_write_queue)) {
605 head = skb_peek(&sk->sk_write_queue);
606 txm = kcm_tx_msg(head);
609 /* Send of first skbuff in queue already in progress */
610 if (WARN_ON(!psock)) {
615 frag_offset = txm->frag_offset;
616 fragidx = txm->fragidx;
623 psock = reserve_psock(kcm);
629 txm = kcm_tx_msg(head);
633 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
638 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
644 frag = &skb_shinfo(skb)->frags[fragidx];
645 if (WARN_ON(!skb_frag_size(frag))) {
650 ret = kernel_sendpage(psock->sk->sk_socket,
652 skb_frag_off(frag) + frag_offset,
653 skb_frag_size(frag) - frag_offset,
656 if (ret == -EAGAIN) {
657 /* Save state to try again when there's
658 * write space on the socket
661 txm->frag_offset = frag_offset;
662 txm->fragidx = fragidx;
669 /* Hard failure in sending message, abort this
670 * psock since it has lost framing
671 * synchronization and retry sending the
672 * message from the beginning.
674 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
676 unreserve_psock(kcm);
679 kcm_report_tx_retry(kcm);
687 KCM_STATS_ADD(psock->stats.tx_bytes, ret);
688 if (frag_offset < skb_frag_size(frag)) {
689 /* Not finished with this frag */
695 if (skb_has_frag_list(skb)) {
696 skb = skb_shinfo(skb)->frag_list;
699 } else if (skb->next) {
704 /* Successfully sent the whole packet, account for it. */
705 skb_dequeue(&sk->sk_write_queue);
707 sk->sk_wmem_queued -= sent;
709 KCM_STATS_INCR(psock->stats.tx_msgs);
710 } while ((head = skb_peek(&sk->sk_write_queue)));
713 /* Done with all queued messages. */
714 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
715 unreserve_psock(kcm);
718 /* Check if write space is available */
719 sk->sk_write_space(sk);
721 return total_sent ? : ret;
724 static void kcm_tx_work(struct work_struct *w)
726 struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
727 struct sock *sk = &kcm->sk;
732 /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
735 err = kcm_write_msgs(kcm);
737 /* Hard failure in write, report error on KCM socket */
738 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
739 report_csk_error(&kcm->sk, -err);
743 /* Primarily for SOCK_SEQPACKET sockets */
744 if (likely(sk->sk_socket) &&
745 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
746 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
747 sk->sk_write_space(sk);
754 static void kcm_push(struct kcm_sock *kcm)
756 if (kcm->tx_wait_more)
760 static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
761 int offset, size_t size, int flags)
764 struct sock *sk = sock->sk;
765 struct kcm_sock *kcm = kcm_sk(sk);
766 struct sk_buff *skb = NULL, *head = NULL;
767 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
772 if (flags & MSG_SENDPAGE_NOTLAST)
775 /* No MSG_EOR from splice, only look at MSG_MORE */
776 eor = !(flags & MSG_MORE);
780 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
787 /* Previously opened message */
789 skb = kcm_tx_msg(head)->last_skb;
790 i = skb_shinfo(skb)->nr_frags;
792 if (skb_can_coalesce(skb, i, page, offset)) {
793 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
794 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG;
798 if (i >= MAX_SKB_FRAGS) {
799 struct sk_buff *tskb;
801 tskb = alloc_skb(0, sk->sk_allocation);
804 err = sk_stream_wait_memory(sk, &timeo);
810 skb_shinfo(head)->frag_list = tskb;
815 skb->ip_summed = CHECKSUM_UNNECESSARY;
819 /* Call the sk_stream functions to manage the sndbuf mem. */
820 if (!sk_stream_memory_free(sk)) {
822 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
823 err = sk_stream_wait_memory(sk, &timeo);
828 head = alloc_skb(0, sk->sk_allocation);
831 err = sk_stream_wait_memory(sk, &timeo);
841 skb_fill_page_desc_noacc(skb, i, page, offset, size);
842 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG;
846 skb->data_len += size;
847 skb->truesize += size;
848 sk->sk_wmem_queued += size;
849 sk_mem_charge(sk, size);
853 head->data_len += size;
854 head->truesize += size;
858 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
860 /* Message complete, queue it on send buffer */
861 __skb_queue_tail(&sk->sk_write_queue, head);
863 KCM_STATS_INCR(kcm->stats.tx_msgs);
865 if (flags & MSG_BATCH) {
866 kcm->tx_wait_more = true;
867 } else if (kcm->tx_wait_more || not_busy) {
868 err = kcm_write_msgs(kcm);
870 /* We got a hard error in write_msgs but have
871 * already queued this message. Report an error
872 * in the socket, but don't affect return value
875 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
876 report_csk_error(&kcm->sk, -err);
880 /* Message not complete, save state */
882 kcm_tx_msg(head)->last_skb = skb;
885 KCM_STATS_ADD(kcm->stats.tx_bytes, size);
893 err = sk_stream_error(sk, flags, err);
895 /* make sure we wake any epoll edge trigger waiter */
896 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
897 sk->sk_write_space(sk);
903 static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
905 struct sock *sk = sock->sk;
906 struct kcm_sock *kcm = kcm_sk(sk);
907 struct sk_buff *skb = NULL, *head = NULL;
908 size_t copy, copied = 0;
909 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
910 int eor = (sock->type == SOCK_DGRAM) ?
911 !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
916 /* Per tcp_sendmsg this should be in poll */
917 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
923 /* Previously opened message */
925 skb = kcm_tx_msg(head)->last_skb;
929 /* Call the sk_stream functions to manage the sndbuf mem. */
930 if (!sk_stream_memory_free(sk)) {
932 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
933 err = sk_stream_wait_memory(sk, &timeo);
938 if (msg_data_left(msg)) {
939 /* New message, alloc head skb */
940 head = alloc_skb(0, sk->sk_allocation);
943 err = sk_stream_wait_memory(sk, &timeo);
947 head = alloc_skb(0, sk->sk_allocation);
952 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
953 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
955 skb->ip_summed = CHECKSUM_UNNECESSARY;
959 while (msg_data_left(msg)) {
961 int i = skb_shinfo(skb)->nr_frags;
962 struct page_frag *pfrag = sk_page_frag(sk);
964 if (!sk_page_frag_refill(sk, pfrag))
965 goto wait_for_memory;
967 if (!skb_can_coalesce(skb, i, pfrag->page,
969 if (i == MAX_SKB_FRAGS) {
970 struct sk_buff *tskb;
972 tskb = alloc_skb(0, sk->sk_allocation);
974 goto wait_for_memory;
977 skb_shinfo(head)->frag_list = tskb;
982 skb->ip_summed = CHECKSUM_UNNECESSARY;
988 copy = min_t(int, msg_data_left(msg),
989 pfrag->size - pfrag->offset);
991 if (!sk_wmem_schedule(sk, copy))
992 goto wait_for_memory;
994 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1001 /* Update the skb. */
1003 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1005 skb_fill_page_desc(skb, i, pfrag->page,
1006 pfrag->offset, copy);
1007 get_page(pfrag->page);
1010 pfrag->offset += copy;
1014 head->data_len += copy;
1021 err = sk_stream_wait_memory(sk, &timeo);
1027 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1030 /* Message complete, queue it on send buffer */
1031 __skb_queue_tail(&sk->sk_write_queue, head);
1032 kcm->seq_skb = NULL;
1033 KCM_STATS_INCR(kcm->stats.tx_msgs);
1036 if (msg->msg_flags & MSG_BATCH) {
1037 kcm->tx_wait_more = true;
1038 } else if (kcm->tx_wait_more || not_busy) {
1039 err = kcm_write_msgs(kcm);
1041 /* We got a hard error in write_msgs but have
1042 * already queued this message. Report an error
1043 * in the socket, but don't affect return value
1046 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1047 report_csk_error(&kcm->sk, -err);
1051 /* Message not complete, save state */
1054 kcm->seq_skb = head;
1055 kcm_tx_msg(head)->last_skb = skb;
1059 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1067 if (copied && sock->type == SOCK_SEQPACKET) {
1068 /* Wrote some bytes before encountering an
1069 * error, return partial success.
1071 goto partial_message;
1074 if (head != kcm->seq_skb)
1077 err = sk_stream_error(sk, msg->msg_flags, err);
1079 /* make sure we wake any epoll edge trigger waiter */
1080 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1081 sk->sk_write_space(sk);
1087 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1088 size_t len, int flags)
1090 int noblock = flags & MSG_DONTWAIT;
1091 struct sock *sk = sock->sk;
1092 struct kcm_sock *kcm = kcm_sk(sk);
1094 struct strp_msg *stm;
1096 struct sk_buff *skb;
1098 skb = skb_recv_datagram(sk, flags, noblock, &err);
1102 /* Okay, have a message on the receive queue */
1104 stm = strp_msg(skb);
1106 if (len > stm->full_len)
1107 len = stm->full_len;
1109 err = skb_copy_datagram_msg(skb, stm->offset, msg, len);
1114 if (likely(!(flags & MSG_PEEK))) {
1115 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1116 if (copied < stm->full_len) {
1117 if (sock->type == SOCK_DGRAM) {
1118 /* Truncated message */
1119 msg->msg_flags |= MSG_TRUNC;
1122 stm->offset += copied;
1123 stm->full_len -= copied;
1126 /* Finished with message */
1127 msg->msg_flags |= MSG_EOR;
1128 KCM_STATS_INCR(kcm->stats.rx_msgs);
1133 skb_free_datagram(sk, skb);
1134 return copied ? : err;
1137 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1138 struct pipe_inode_info *pipe, size_t len,
1141 int noblock = flags & MSG_DONTWAIT;
1142 struct sock *sk = sock->sk;
1143 struct kcm_sock *kcm = kcm_sk(sk);
1144 struct strp_msg *stm;
1147 struct sk_buff *skb;
1149 /* Only support splice for SOCKSEQPACKET */
1151 skb = skb_recv_datagram(sk, flags, noblock, &err);
1155 /* Okay, have a message on the receive queue */
1157 stm = strp_msg(skb);
1159 if (len > stm->full_len)
1160 len = stm->full_len;
1162 copied = skb_splice_bits(skb, sk, stm->offset, pipe, len, flags);
1168 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1170 stm->offset += copied;
1171 stm->full_len -= copied;
1173 /* We have no way to return MSG_EOR. If all the bytes have been
1174 * read we still leave the message in the receive socket buffer.
1175 * A subsequent recvmsg needs to be done to return MSG_EOR and
1176 * finish reading the message.
1179 skb_free_datagram(sk, skb);
1183 skb_free_datagram(sk, skb);
1187 /* kcm sock lock held */
1188 static void kcm_recv_disable(struct kcm_sock *kcm)
1190 struct kcm_mux *mux = kcm->mux;
1192 if (kcm->rx_disabled)
1195 spin_lock_bh(&mux->rx_lock);
1197 kcm->rx_disabled = 1;
1199 /* If a psock is reserved we'll do cleanup in unreserve */
1200 if (!kcm->rx_psock) {
1202 list_del(&kcm->wait_rx_list);
1203 /* paired with lockless reads in kcm_rfree() */
1204 WRITE_ONCE(kcm->rx_wait, false);
1207 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1210 spin_unlock_bh(&mux->rx_lock);
1213 /* kcm sock lock held */
1214 static void kcm_recv_enable(struct kcm_sock *kcm)
1216 struct kcm_mux *mux = kcm->mux;
1218 if (!kcm->rx_disabled)
1221 spin_lock_bh(&mux->rx_lock);
1223 kcm->rx_disabled = 0;
1226 spin_unlock_bh(&mux->rx_lock);
1229 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1230 sockptr_t optval, unsigned int optlen)
1232 struct kcm_sock *kcm = kcm_sk(sock->sk);
1236 if (level != SOL_KCM)
1237 return -ENOPROTOOPT;
1239 if (optlen < sizeof(int))
1242 if (copy_from_sockptr(&val, optval, sizeof(int)))
1245 valbool = val ? 1 : 0;
1248 case KCM_RECV_DISABLE:
1249 lock_sock(&kcm->sk);
1251 kcm_recv_disable(kcm);
1253 kcm_recv_enable(kcm);
1254 release_sock(&kcm->sk);
1263 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1264 char __user *optval, int __user *optlen)
1266 struct kcm_sock *kcm = kcm_sk(sock->sk);
1269 if (level != SOL_KCM)
1270 return -ENOPROTOOPT;
1272 if (get_user(len, optlen))
1275 len = min_t(unsigned int, len, sizeof(int));
1280 case KCM_RECV_DISABLE:
1281 val = kcm->rx_disabled;
1284 return -ENOPROTOOPT;
1287 if (put_user(len, optlen))
1289 if (copy_to_user(optval, &val, len))
1294 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1296 struct kcm_sock *tkcm;
1297 struct list_head *head;
1300 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1301 * we set sk_state, otherwise epoll_wait always returns right away with
1304 kcm->sk.sk_state = TCP_ESTABLISHED;
1306 /* Add to mux's kcm sockets list */
1308 spin_lock_bh(&mux->lock);
1310 head = &mux->kcm_socks;
1311 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1312 if (tkcm->index != index)
1314 head = &tkcm->kcm_sock_list;
1318 list_add(&kcm->kcm_sock_list, head);
1321 mux->kcm_socks_cnt++;
1322 spin_unlock_bh(&mux->lock);
1324 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1326 spin_lock_bh(&mux->rx_lock);
1328 spin_unlock_bh(&mux->rx_lock);
1331 static int kcm_attach(struct socket *sock, struct socket *csock,
1332 struct bpf_prog *prog)
1334 struct kcm_sock *kcm = kcm_sk(sock->sk);
1335 struct kcm_mux *mux = kcm->mux;
1337 struct kcm_psock *psock = NULL, *tpsock;
1338 struct list_head *head;
1340 static const struct strp_callbacks cb = {
1341 .rcv_msg = kcm_rcv_strparser,
1342 .parse_msg = kcm_parse_func_strparser,
1343 .read_sock_done = kcm_read_sock_done,
1353 /* Only allow TCP sockets to be attached for now */
1354 if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1355 csk->sk_protocol != IPPROTO_TCP) {
1360 /* Don't allow listeners or closed sockets */
1361 if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1366 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1374 psock->bpf_prog = prog;
1376 write_lock_bh(&csk->sk_callback_lock);
1378 /* Check if sk_user_data is already by KCM or someone else.
1379 * Must be done under lock to prevent race conditions.
1381 if (csk->sk_user_data) {
1382 write_unlock_bh(&csk->sk_callback_lock);
1383 kmem_cache_free(kcm_psockp, psock);
1388 err = strp_init(&psock->strp, csk, &cb);
1390 write_unlock_bh(&csk->sk_callback_lock);
1391 kmem_cache_free(kcm_psockp, psock);
1395 psock->save_data_ready = csk->sk_data_ready;
1396 psock->save_write_space = csk->sk_write_space;
1397 psock->save_state_change = csk->sk_state_change;
1398 csk->sk_user_data = psock;
1399 csk->sk_data_ready = psock_data_ready;
1400 csk->sk_write_space = psock_write_space;
1401 csk->sk_state_change = psock_state_change;
1403 write_unlock_bh(&csk->sk_callback_lock);
1407 /* Finished initialization, now add the psock to the MUX. */
1408 spin_lock_bh(&mux->lock);
1409 head = &mux->psocks;
1410 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1411 if (tpsock->index != index)
1413 head = &tpsock->psock_list;
1417 list_add(&psock->psock_list, head);
1418 psock->index = index;
1420 KCM_STATS_INCR(mux->stats.psock_attach);
1422 psock_now_avail(psock);
1423 spin_unlock_bh(&mux->lock);
1425 /* Schedule RX work in case there are already bytes queued */
1426 strp_check_rcv(&psock->strp);
1434 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1436 struct socket *csock;
1437 struct bpf_prog *prog;
1440 csock = sockfd_lookup(info->fd, &err);
1444 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1446 err = PTR_ERR(prog);
1450 err = kcm_attach(sock, csock, prog);
1456 /* Keep reference on file also */
1464 static void kcm_unattach(struct kcm_psock *psock)
1466 struct sock *csk = psock->sk;
1467 struct kcm_mux *mux = psock->mux;
1471 /* Stop getting callbacks from TCP socket. After this there should
1472 * be no way to reserve a kcm for this psock.
1474 write_lock_bh(&csk->sk_callback_lock);
1475 csk->sk_user_data = NULL;
1476 csk->sk_data_ready = psock->save_data_ready;
1477 csk->sk_write_space = psock->save_write_space;
1478 csk->sk_state_change = psock->save_state_change;
1479 strp_stop(&psock->strp);
1481 if (WARN_ON(psock->rx_kcm)) {
1482 write_unlock_bh(&csk->sk_callback_lock);
1487 spin_lock_bh(&mux->rx_lock);
1489 /* Stop receiver activities. After this point psock should not be
1490 * able to get onto ready list either through callbacks or work.
1492 if (psock->ready_rx_msg) {
1493 list_del(&psock->psock_ready_list);
1494 kfree_skb(psock->ready_rx_msg);
1495 psock->ready_rx_msg = NULL;
1496 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1499 spin_unlock_bh(&mux->rx_lock);
1501 write_unlock_bh(&csk->sk_callback_lock);
1503 /* Call strp_done without sock lock */
1505 strp_done(&psock->strp);
1508 bpf_prog_put(psock->bpf_prog);
1510 spin_lock_bh(&mux->lock);
1512 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1513 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1515 KCM_STATS_INCR(mux->stats.psock_unattach);
1517 if (psock->tx_kcm) {
1518 /* psock was reserved. Just mark it finished and we will clean
1519 * up in the kcm paths, we need kcm lock which can not be
1522 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1523 spin_unlock_bh(&mux->lock);
1525 /* We are unattaching a socket that is reserved. Abort the
1526 * socket since we may be out of sync in sending on it. We need
1527 * to do this without the mux lock.
1529 kcm_abort_tx_psock(psock, EPIPE, false);
1531 spin_lock_bh(&mux->lock);
1532 if (!psock->tx_kcm) {
1533 /* psock now unreserved in window mux was unlocked */
1538 /* Commit done before queuing work to process it */
1541 /* Queue tx work to make sure psock->done is handled */
1542 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1543 spin_unlock_bh(&mux->lock);
1546 if (!psock->tx_stopped)
1547 list_del(&psock->psock_avail_list);
1548 list_del(&psock->psock_list);
1550 spin_unlock_bh(&mux->lock);
1553 fput(csk->sk_socket->file);
1554 kmem_cache_free(kcm_psockp, psock);
1560 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1562 struct kcm_sock *kcm = kcm_sk(sock->sk);
1563 struct kcm_mux *mux = kcm->mux;
1564 struct kcm_psock *psock;
1565 struct socket *csock;
1569 csock = sockfd_lookup(info->fd, &err);
1581 spin_lock_bh(&mux->lock);
1583 list_for_each_entry(psock, &mux->psocks, psock_list) {
1584 if (psock->sk != csk)
1587 /* Found the matching psock */
1589 if (psock->unattaching || WARN_ON(psock->done)) {
1594 psock->unattaching = 1;
1596 spin_unlock_bh(&mux->lock);
1598 /* Lower socket lock should already be held */
1599 kcm_unattach(psock);
1605 spin_unlock_bh(&mux->lock);
1612 static struct proto kcm_proto = {
1614 .owner = THIS_MODULE,
1615 .obj_size = sizeof(struct kcm_sock),
1618 /* Clone a kcm socket. */
1619 static struct file *kcm_clone(struct socket *osock)
1621 struct socket *newsock;
1624 newsock = sock_alloc();
1626 return ERR_PTR(-ENFILE);
1628 newsock->type = osock->type;
1629 newsock->ops = osock->ops;
1631 __module_get(newsock->ops->owner);
1633 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1636 sock_release(newsock);
1637 return ERR_PTR(-ENOMEM);
1639 sock_init_data(newsock, newsk);
1640 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1642 return sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1645 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1650 case SIOCKCMATTACH: {
1651 struct kcm_attach info;
1653 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1656 err = kcm_attach_ioctl(sock, &info);
1660 case SIOCKCMUNATTACH: {
1661 struct kcm_unattach info;
1663 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1666 err = kcm_unattach_ioctl(sock, &info);
1670 case SIOCKCMCLONE: {
1671 struct kcm_clone info;
1674 info.fd = get_unused_fd_flags(0);
1675 if (unlikely(info.fd < 0))
1678 file = kcm_clone(sock);
1680 put_unused_fd(info.fd);
1681 return PTR_ERR(file);
1683 if (copy_to_user((void __user *)arg, &info,
1685 put_unused_fd(info.fd);
1689 fd_install(info.fd, file);
1701 static void free_mux(struct rcu_head *rcu)
1703 struct kcm_mux *mux = container_of(rcu,
1704 struct kcm_mux, rcu);
1706 kmem_cache_free(kcm_muxp, mux);
1709 static void release_mux(struct kcm_mux *mux)
1711 struct kcm_net *knet = mux->knet;
1712 struct kcm_psock *psock, *tmp_psock;
1714 /* Release psocks */
1715 list_for_each_entry_safe(psock, tmp_psock,
1716 &mux->psocks, psock_list) {
1717 if (!WARN_ON(psock->unattaching))
1718 kcm_unattach(psock);
1721 if (WARN_ON(mux->psocks_cnt))
1724 __skb_queue_purge(&mux->rx_hold_queue);
1726 mutex_lock(&knet->mutex);
1727 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1728 aggregate_psock_stats(&mux->aggregate_psock_stats,
1729 &knet->aggregate_psock_stats);
1730 aggregate_strp_stats(&mux->aggregate_strp_stats,
1731 &knet->aggregate_strp_stats);
1732 list_del_rcu(&mux->kcm_mux_list);
1734 mutex_unlock(&knet->mutex);
1736 call_rcu(&mux->rcu, free_mux);
1739 static void kcm_done(struct kcm_sock *kcm)
1741 struct kcm_mux *mux = kcm->mux;
1742 struct sock *sk = &kcm->sk;
1745 spin_lock_bh(&mux->rx_lock);
1746 if (kcm->rx_psock) {
1747 /* Cleanup in unreserve_rx_kcm */
1749 kcm->rx_disabled = 1;
1751 spin_unlock_bh(&mux->rx_lock);
1756 list_del(&kcm->wait_rx_list);
1757 /* paired with lockless reads in kcm_rfree() */
1758 WRITE_ONCE(kcm->rx_wait, false);
1760 /* Move any pending receive messages to other kcm sockets */
1761 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1763 spin_unlock_bh(&mux->rx_lock);
1765 if (WARN_ON(sk_rmem_alloc_get(sk)))
1768 /* Detach from MUX */
1769 spin_lock_bh(&mux->lock);
1771 list_del(&kcm->kcm_sock_list);
1772 mux->kcm_socks_cnt--;
1773 socks_cnt = mux->kcm_socks_cnt;
1775 spin_unlock_bh(&mux->lock);
1778 /* We are done with the mux now. */
1782 WARN_ON(kcm->rx_wait);
1787 /* Called by kcm_release to close a KCM socket.
1788 * If this is the last KCM socket on the MUX, destroy the MUX.
1790 static int kcm_release(struct socket *sock)
1792 struct sock *sk = sock->sk;
1793 struct kcm_sock *kcm;
1794 struct kcm_mux *mux;
1795 struct kcm_psock *psock;
1805 kfree_skb(kcm->seq_skb);
1807 /* Purge queue under lock to avoid race condition with tx_work trying
1808 * to act when queue is nonempty. If tx_work runs after this point
1809 * it will just return.
1811 __skb_queue_purge(&sk->sk_write_queue);
1813 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1814 * get a writespace callback. This prevents further work being queued
1815 * from the callback (unbinding the psock occurs after canceling work.
1817 kcm->tx_stopped = 1;
1821 spin_lock_bh(&mux->lock);
1823 /* Take of tx_wait list, after this point there should be no way
1824 * that a psock will be assigned to this kcm.
1826 list_del(&kcm->wait_psock_list);
1827 kcm->tx_wait = false;
1829 spin_unlock_bh(&mux->lock);
1831 /* Cancel work. After this point there should be no outside references
1832 * to the kcm socket.
1834 cancel_work_sync(&kcm->tx_work);
1837 psock = kcm->tx_psock;
1839 /* A psock was reserved, so we need to kill it since it
1840 * may already have some bytes queued from a message. We
1841 * need to do this after removing kcm from tx_wait list.
1843 kcm_abort_tx_psock(psock, EPIPE, false);
1844 unreserve_psock(kcm);
1848 WARN_ON(kcm->tx_wait);
1849 WARN_ON(kcm->tx_psock);
1858 static const struct proto_ops kcm_dgram_ops = {
1860 .owner = THIS_MODULE,
1861 .release = kcm_release,
1862 .bind = sock_no_bind,
1863 .connect = sock_no_connect,
1864 .socketpair = sock_no_socketpair,
1865 .accept = sock_no_accept,
1866 .getname = sock_no_getname,
1867 .poll = datagram_poll,
1869 .listen = sock_no_listen,
1870 .shutdown = sock_no_shutdown,
1871 .setsockopt = kcm_setsockopt,
1872 .getsockopt = kcm_getsockopt,
1873 .sendmsg = kcm_sendmsg,
1874 .recvmsg = kcm_recvmsg,
1875 .mmap = sock_no_mmap,
1876 .sendpage = kcm_sendpage,
1879 static const struct proto_ops kcm_seqpacket_ops = {
1881 .owner = THIS_MODULE,
1882 .release = kcm_release,
1883 .bind = sock_no_bind,
1884 .connect = sock_no_connect,
1885 .socketpair = sock_no_socketpair,
1886 .accept = sock_no_accept,
1887 .getname = sock_no_getname,
1888 .poll = datagram_poll,
1890 .listen = sock_no_listen,
1891 .shutdown = sock_no_shutdown,
1892 .setsockopt = kcm_setsockopt,
1893 .getsockopt = kcm_getsockopt,
1894 .sendmsg = kcm_sendmsg,
1895 .recvmsg = kcm_recvmsg,
1896 .mmap = sock_no_mmap,
1897 .sendpage = kcm_sendpage,
1898 .splice_read = kcm_splice_read,
1901 /* Create proto operation for kcm sockets */
1902 static int kcm_create(struct net *net, struct socket *sock,
1903 int protocol, int kern)
1905 struct kcm_net *knet = net_generic(net, kcm_net_id);
1907 struct kcm_mux *mux;
1909 switch (sock->type) {
1911 sock->ops = &kcm_dgram_ops;
1913 case SOCK_SEQPACKET:
1914 sock->ops = &kcm_seqpacket_ops;
1917 return -ESOCKTNOSUPPORT;
1920 if (protocol != KCMPROTO_CONNECTED)
1921 return -EPROTONOSUPPORT;
1923 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1927 /* Allocate a kcm mux, shared between KCM sockets */
1928 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1934 spin_lock_init(&mux->lock);
1935 spin_lock_init(&mux->rx_lock);
1936 INIT_LIST_HEAD(&mux->kcm_socks);
1937 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1938 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1940 INIT_LIST_HEAD(&mux->psocks);
1941 INIT_LIST_HEAD(&mux->psocks_ready);
1942 INIT_LIST_HEAD(&mux->psocks_avail);
1946 /* Add new MUX to list */
1947 mutex_lock(&knet->mutex);
1948 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1950 mutex_unlock(&knet->mutex);
1952 skb_queue_head_init(&mux->rx_hold_queue);
1954 /* Init KCM socket */
1955 sock_init_data(sock, sk);
1956 init_kcm_sock(kcm_sk(sk), mux);
1961 static const struct net_proto_family kcm_family_ops = {
1963 .create = kcm_create,
1964 .owner = THIS_MODULE,
1967 static __net_init int kcm_init_net(struct net *net)
1969 struct kcm_net *knet = net_generic(net, kcm_net_id);
1971 INIT_LIST_HEAD_RCU(&knet->mux_list);
1972 mutex_init(&knet->mutex);
1977 static __net_exit void kcm_exit_net(struct net *net)
1979 struct kcm_net *knet = net_generic(net, kcm_net_id);
1981 /* All KCM sockets should be closed at this point, which should mean
1982 * that all multiplexors and psocks have been destroyed.
1984 WARN_ON(!list_empty(&knet->mux_list));
1987 static struct pernet_operations kcm_net_ops = {
1988 .init = kcm_init_net,
1989 .exit = kcm_exit_net,
1991 .size = sizeof(struct kcm_net),
1994 static int __init kcm_init(void)
1998 kcm_muxp = kmem_cache_create("kcm_mux_cache",
1999 sizeof(struct kcm_mux), 0,
2000 SLAB_HWCACHE_ALIGN, NULL);
2004 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2005 sizeof(struct kcm_psock), 0,
2006 SLAB_HWCACHE_ALIGN, NULL);
2010 kcm_wq = create_singlethread_workqueue("kkcmd");
2014 err = proto_register(&kcm_proto, 1);
2018 err = register_pernet_device(&kcm_net_ops);
2022 err = sock_register(&kcm_family_ops);
2024 goto sock_register_fail;
2026 err = kcm_proc_init();
2028 goto proc_init_fail;
2033 sock_unregister(PF_KCM);
2036 unregister_pernet_device(&kcm_net_ops);
2039 proto_unregister(&kcm_proto);
2042 kmem_cache_destroy(kcm_muxp);
2043 kmem_cache_destroy(kcm_psockp);
2046 destroy_workqueue(kcm_wq);
2051 static void __exit kcm_exit(void)
2054 sock_unregister(PF_KCM);
2055 unregister_pernet_device(&kcm_net_ops);
2056 proto_unregister(&kcm_proto);
2057 destroy_workqueue(kcm_wq);
2059 kmem_cache_destroy(kcm_muxp);
2060 kmem_cache_destroy(kcm_psockp);
2063 module_init(kcm_init);
2064 module_exit(kcm_exit);
2066 MODULE_LICENSE("GPL");
2067 MODULE_ALIAS_NETPROTO(PF_KCM);