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)
50 csk->sk_error_report(csk);
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 * synchonization 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)->tx_flags |= SKBTX_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(skb, i, page, offset, size);
842 skb_shinfo(skb)->tx_flags |= SKBTX_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 struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1088 long timeo, int *err)
1090 struct sk_buff *skb;
1092 while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1094 *err = sock_error(sk);
1098 if (sock_flag(sk, SOCK_DONE))
1101 if ((flags & MSG_DONTWAIT) || !timeo) {
1106 sk_wait_data(sk, &timeo, NULL);
1108 /* Handle signals */
1109 if (signal_pending(current)) {
1110 *err = sock_intr_errno(timeo);
1118 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1119 size_t len, int flags)
1121 struct sock *sk = sock->sk;
1122 struct kcm_sock *kcm = kcm_sk(sk);
1125 struct strp_msg *stm;
1127 struct sk_buff *skb;
1129 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1133 skb = kcm_wait_data(sk, flags, timeo, &err);
1137 /* Okay, have a message on the receive queue */
1139 stm = strp_msg(skb);
1141 if (len > stm->full_len)
1142 len = stm->full_len;
1144 err = skb_copy_datagram_msg(skb, stm->offset, msg, len);
1149 if (likely(!(flags & MSG_PEEK))) {
1150 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1151 if (copied < stm->full_len) {
1152 if (sock->type == SOCK_DGRAM) {
1153 /* Truncated message */
1154 msg->msg_flags |= MSG_TRUNC;
1157 stm->offset += copied;
1158 stm->full_len -= copied;
1161 /* Finished with message */
1162 msg->msg_flags |= MSG_EOR;
1163 KCM_STATS_INCR(kcm->stats.rx_msgs);
1164 skb_unlink(skb, &sk->sk_receive_queue);
1172 return copied ? : err;
1175 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1176 struct pipe_inode_info *pipe, size_t len,
1179 struct sock *sk = sock->sk;
1180 struct kcm_sock *kcm = kcm_sk(sk);
1182 struct strp_msg *stm;
1185 struct sk_buff *skb;
1187 /* Only support splice for SOCKSEQPACKET */
1189 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1193 skb = kcm_wait_data(sk, flags, timeo, &err);
1197 /* Okay, have a message on the receive queue */
1199 stm = strp_msg(skb);
1201 if (len > stm->full_len)
1202 len = stm->full_len;
1204 copied = skb_splice_bits(skb, sk, stm->offset, pipe, len, flags);
1210 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1212 stm->offset += copied;
1213 stm->full_len -= copied;
1215 /* We have no way to return MSG_EOR. If all the bytes have been
1216 * read we still leave the message in the receive socket buffer.
1217 * A subsequent recvmsg needs to be done to return MSG_EOR and
1218 * finish reading the message.
1231 /* kcm sock lock held */
1232 static void kcm_recv_disable(struct kcm_sock *kcm)
1234 struct kcm_mux *mux = kcm->mux;
1236 if (kcm->rx_disabled)
1239 spin_lock_bh(&mux->rx_lock);
1241 kcm->rx_disabled = 1;
1243 /* If a psock is reserved we'll do cleanup in unreserve */
1244 if (!kcm->rx_psock) {
1246 list_del(&kcm->wait_rx_list);
1247 /* paired with lockless reads in kcm_rfree() */
1248 WRITE_ONCE(kcm->rx_wait, false);
1251 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1254 spin_unlock_bh(&mux->rx_lock);
1257 /* kcm sock lock held */
1258 static void kcm_recv_enable(struct kcm_sock *kcm)
1260 struct kcm_mux *mux = kcm->mux;
1262 if (!kcm->rx_disabled)
1265 spin_lock_bh(&mux->rx_lock);
1267 kcm->rx_disabled = 0;
1270 spin_unlock_bh(&mux->rx_lock);
1273 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1274 sockptr_t optval, unsigned int optlen)
1276 struct kcm_sock *kcm = kcm_sk(sock->sk);
1280 if (level != SOL_KCM)
1281 return -ENOPROTOOPT;
1283 if (optlen < sizeof(int))
1286 if (copy_from_sockptr(&val, optval, sizeof(int)))
1289 valbool = val ? 1 : 0;
1292 case KCM_RECV_DISABLE:
1293 lock_sock(&kcm->sk);
1295 kcm_recv_disable(kcm);
1297 kcm_recv_enable(kcm);
1298 release_sock(&kcm->sk);
1307 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1308 char __user *optval, int __user *optlen)
1310 struct kcm_sock *kcm = kcm_sk(sock->sk);
1313 if (level != SOL_KCM)
1314 return -ENOPROTOOPT;
1316 if (get_user(len, optlen))
1319 len = min_t(unsigned int, len, sizeof(int));
1324 case KCM_RECV_DISABLE:
1325 val = kcm->rx_disabled;
1328 return -ENOPROTOOPT;
1331 if (put_user(len, optlen))
1333 if (copy_to_user(optval, &val, len))
1338 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1340 struct kcm_sock *tkcm;
1341 struct list_head *head;
1344 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1345 * we set sk_state, otherwise epoll_wait always returns right away with
1348 kcm->sk.sk_state = TCP_ESTABLISHED;
1350 /* Add to mux's kcm sockets list */
1352 spin_lock_bh(&mux->lock);
1354 head = &mux->kcm_socks;
1355 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1356 if (tkcm->index != index)
1358 head = &tkcm->kcm_sock_list;
1362 list_add(&kcm->kcm_sock_list, head);
1365 mux->kcm_socks_cnt++;
1366 spin_unlock_bh(&mux->lock);
1368 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1370 spin_lock_bh(&mux->rx_lock);
1372 spin_unlock_bh(&mux->rx_lock);
1375 static int kcm_attach(struct socket *sock, struct socket *csock,
1376 struct bpf_prog *prog)
1378 struct kcm_sock *kcm = kcm_sk(sock->sk);
1379 struct kcm_mux *mux = kcm->mux;
1381 struct kcm_psock *psock = NULL, *tpsock;
1382 struct list_head *head;
1384 static const struct strp_callbacks cb = {
1385 .rcv_msg = kcm_rcv_strparser,
1386 .parse_msg = kcm_parse_func_strparser,
1387 .read_sock_done = kcm_read_sock_done,
1397 /* Only allow TCP sockets to be attached for now */
1398 if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1399 csk->sk_protocol != IPPROTO_TCP) {
1404 /* Don't allow listeners or closed sockets */
1405 if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1410 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1418 psock->bpf_prog = prog;
1420 write_lock_bh(&csk->sk_callback_lock);
1422 /* Check if sk_user_data is aready by KCM or someone else.
1423 * Must be done under lock to prevent race conditions.
1425 if (csk->sk_user_data) {
1426 write_unlock_bh(&csk->sk_callback_lock);
1427 kmem_cache_free(kcm_psockp, psock);
1432 err = strp_init(&psock->strp, csk, &cb);
1434 write_unlock_bh(&csk->sk_callback_lock);
1435 kmem_cache_free(kcm_psockp, psock);
1439 psock->save_data_ready = csk->sk_data_ready;
1440 psock->save_write_space = csk->sk_write_space;
1441 psock->save_state_change = csk->sk_state_change;
1442 csk->sk_user_data = psock;
1443 csk->sk_data_ready = psock_data_ready;
1444 csk->sk_write_space = psock_write_space;
1445 csk->sk_state_change = psock_state_change;
1447 write_unlock_bh(&csk->sk_callback_lock);
1451 /* Finished initialization, now add the psock to the MUX. */
1452 spin_lock_bh(&mux->lock);
1453 head = &mux->psocks;
1454 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1455 if (tpsock->index != index)
1457 head = &tpsock->psock_list;
1461 list_add(&psock->psock_list, head);
1462 psock->index = index;
1464 KCM_STATS_INCR(mux->stats.psock_attach);
1466 psock_now_avail(psock);
1467 spin_unlock_bh(&mux->lock);
1469 /* Schedule RX work in case there are already bytes queued */
1470 strp_check_rcv(&psock->strp);
1478 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1480 struct socket *csock;
1481 struct bpf_prog *prog;
1484 csock = sockfd_lookup(info->fd, &err);
1488 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1490 err = PTR_ERR(prog);
1494 err = kcm_attach(sock, csock, prog);
1500 /* Keep reference on file also */
1508 static void kcm_unattach(struct kcm_psock *psock)
1510 struct sock *csk = psock->sk;
1511 struct kcm_mux *mux = psock->mux;
1515 /* Stop getting callbacks from TCP socket. After this there should
1516 * be no way to reserve a kcm for this psock.
1518 write_lock_bh(&csk->sk_callback_lock);
1519 csk->sk_user_data = NULL;
1520 csk->sk_data_ready = psock->save_data_ready;
1521 csk->sk_write_space = psock->save_write_space;
1522 csk->sk_state_change = psock->save_state_change;
1523 strp_stop(&psock->strp);
1525 if (WARN_ON(psock->rx_kcm)) {
1526 write_unlock_bh(&csk->sk_callback_lock);
1531 spin_lock_bh(&mux->rx_lock);
1533 /* Stop receiver activities. After this point psock should not be
1534 * able to get onto ready list either through callbacks or work.
1536 if (psock->ready_rx_msg) {
1537 list_del(&psock->psock_ready_list);
1538 kfree_skb(psock->ready_rx_msg);
1539 psock->ready_rx_msg = NULL;
1540 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1543 spin_unlock_bh(&mux->rx_lock);
1545 write_unlock_bh(&csk->sk_callback_lock);
1547 /* Call strp_done without sock lock */
1549 strp_done(&psock->strp);
1552 bpf_prog_put(psock->bpf_prog);
1554 spin_lock_bh(&mux->lock);
1556 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1557 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1559 KCM_STATS_INCR(mux->stats.psock_unattach);
1561 if (psock->tx_kcm) {
1562 /* psock was reserved. Just mark it finished and we will clean
1563 * up in the kcm paths, we need kcm lock which can not be
1566 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1567 spin_unlock_bh(&mux->lock);
1569 /* We are unattaching a socket that is reserved. Abort the
1570 * socket since we may be out of sync in sending on it. We need
1571 * to do this without the mux lock.
1573 kcm_abort_tx_psock(psock, EPIPE, false);
1575 spin_lock_bh(&mux->lock);
1576 if (!psock->tx_kcm) {
1577 /* psock now unreserved in window mux was unlocked */
1582 /* Commit done before queuing work to process it */
1585 /* Queue tx work to make sure psock->done is handled */
1586 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1587 spin_unlock_bh(&mux->lock);
1590 if (!psock->tx_stopped)
1591 list_del(&psock->psock_avail_list);
1592 list_del(&psock->psock_list);
1594 spin_unlock_bh(&mux->lock);
1597 fput(csk->sk_socket->file);
1598 kmem_cache_free(kcm_psockp, psock);
1604 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1606 struct kcm_sock *kcm = kcm_sk(sock->sk);
1607 struct kcm_mux *mux = kcm->mux;
1608 struct kcm_psock *psock;
1609 struct socket *csock;
1613 csock = sockfd_lookup(info->fd, &err);
1625 spin_lock_bh(&mux->lock);
1627 list_for_each_entry(psock, &mux->psocks, psock_list) {
1628 if (psock->sk != csk)
1631 /* Found the matching psock */
1633 if (psock->unattaching || WARN_ON(psock->done)) {
1638 psock->unattaching = 1;
1640 spin_unlock_bh(&mux->lock);
1642 /* Lower socket lock should already be held */
1643 kcm_unattach(psock);
1649 spin_unlock_bh(&mux->lock);
1656 static struct proto kcm_proto = {
1658 .owner = THIS_MODULE,
1659 .obj_size = sizeof(struct kcm_sock),
1662 /* Clone a kcm socket. */
1663 static struct file *kcm_clone(struct socket *osock)
1665 struct socket *newsock;
1668 newsock = sock_alloc();
1670 return ERR_PTR(-ENFILE);
1672 newsock->type = osock->type;
1673 newsock->ops = osock->ops;
1675 __module_get(newsock->ops->owner);
1677 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1680 sock_release(newsock);
1681 return ERR_PTR(-ENOMEM);
1683 sock_init_data(newsock, newsk);
1684 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1686 return sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1689 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1694 case SIOCKCMATTACH: {
1695 struct kcm_attach info;
1697 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1700 err = kcm_attach_ioctl(sock, &info);
1704 case SIOCKCMUNATTACH: {
1705 struct kcm_unattach info;
1707 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1710 err = kcm_unattach_ioctl(sock, &info);
1714 case SIOCKCMCLONE: {
1715 struct kcm_clone info;
1718 info.fd = get_unused_fd_flags(0);
1719 if (unlikely(info.fd < 0))
1722 file = kcm_clone(sock);
1724 put_unused_fd(info.fd);
1725 return PTR_ERR(file);
1727 if (copy_to_user((void __user *)arg, &info,
1729 put_unused_fd(info.fd);
1733 fd_install(info.fd, file);
1745 static void free_mux(struct rcu_head *rcu)
1747 struct kcm_mux *mux = container_of(rcu,
1748 struct kcm_mux, rcu);
1750 kmem_cache_free(kcm_muxp, mux);
1753 static void release_mux(struct kcm_mux *mux)
1755 struct kcm_net *knet = mux->knet;
1756 struct kcm_psock *psock, *tmp_psock;
1758 /* Release psocks */
1759 list_for_each_entry_safe(psock, tmp_psock,
1760 &mux->psocks, psock_list) {
1761 if (!WARN_ON(psock->unattaching))
1762 kcm_unattach(psock);
1765 if (WARN_ON(mux->psocks_cnt))
1768 __skb_queue_purge(&mux->rx_hold_queue);
1770 mutex_lock(&knet->mutex);
1771 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1772 aggregate_psock_stats(&mux->aggregate_psock_stats,
1773 &knet->aggregate_psock_stats);
1774 aggregate_strp_stats(&mux->aggregate_strp_stats,
1775 &knet->aggregate_strp_stats);
1776 list_del_rcu(&mux->kcm_mux_list);
1778 mutex_unlock(&knet->mutex);
1780 call_rcu(&mux->rcu, free_mux);
1783 static void kcm_done(struct kcm_sock *kcm)
1785 struct kcm_mux *mux = kcm->mux;
1786 struct sock *sk = &kcm->sk;
1789 spin_lock_bh(&mux->rx_lock);
1790 if (kcm->rx_psock) {
1791 /* Cleanup in unreserve_rx_kcm */
1793 kcm->rx_disabled = 1;
1795 spin_unlock_bh(&mux->rx_lock);
1800 list_del(&kcm->wait_rx_list);
1801 /* paired with lockless reads in kcm_rfree() */
1802 WRITE_ONCE(kcm->rx_wait, false);
1804 /* Move any pending receive messages to other kcm sockets */
1805 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1807 spin_unlock_bh(&mux->rx_lock);
1809 if (WARN_ON(sk_rmem_alloc_get(sk)))
1812 /* Detach from MUX */
1813 spin_lock_bh(&mux->lock);
1815 list_del(&kcm->kcm_sock_list);
1816 mux->kcm_socks_cnt--;
1817 socks_cnt = mux->kcm_socks_cnt;
1819 spin_unlock_bh(&mux->lock);
1822 /* We are done with the mux now. */
1826 WARN_ON(kcm->rx_wait);
1831 /* Called by kcm_release to close a KCM socket.
1832 * If this is the last KCM socket on the MUX, destroy the MUX.
1834 static int kcm_release(struct socket *sock)
1836 struct sock *sk = sock->sk;
1837 struct kcm_sock *kcm;
1838 struct kcm_mux *mux;
1839 struct kcm_psock *psock;
1848 kfree_skb(kcm->seq_skb);
1851 /* Purge queue under lock to avoid race condition with tx_work trying
1852 * to act when queue is nonempty. If tx_work runs after this point
1853 * it will just return.
1855 __skb_queue_purge(&sk->sk_write_queue);
1857 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1858 * get a writespace callback. This prevents further work being queued
1859 * from the callback (unbinding the psock occurs after canceling work.
1861 kcm->tx_stopped = 1;
1865 spin_lock_bh(&mux->lock);
1867 /* Take of tx_wait list, after this point there should be no way
1868 * that a psock will be assigned to this kcm.
1870 list_del(&kcm->wait_psock_list);
1871 kcm->tx_wait = false;
1873 spin_unlock_bh(&mux->lock);
1875 /* Cancel work. After this point there should be no outside references
1876 * to the kcm socket.
1878 cancel_work_sync(&kcm->tx_work);
1881 psock = kcm->tx_psock;
1883 /* A psock was reserved, so we need to kill it since it
1884 * may already have some bytes queued from a message. We
1885 * need to do this after removing kcm from tx_wait list.
1887 kcm_abort_tx_psock(psock, EPIPE, false);
1888 unreserve_psock(kcm);
1892 WARN_ON(kcm->tx_wait);
1893 WARN_ON(kcm->tx_psock);
1902 static const struct proto_ops kcm_dgram_ops = {
1904 .owner = THIS_MODULE,
1905 .release = kcm_release,
1906 .bind = sock_no_bind,
1907 .connect = sock_no_connect,
1908 .socketpair = sock_no_socketpair,
1909 .accept = sock_no_accept,
1910 .getname = sock_no_getname,
1911 .poll = datagram_poll,
1913 .listen = sock_no_listen,
1914 .shutdown = sock_no_shutdown,
1915 .setsockopt = kcm_setsockopt,
1916 .getsockopt = kcm_getsockopt,
1917 .sendmsg = kcm_sendmsg,
1918 .recvmsg = kcm_recvmsg,
1919 .mmap = sock_no_mmap,
1920 .sendpage = kcm_sendpage,
1923 static const struct proto_ops kcm_seqpacket_ops = {
1925 .owner = THIS_MODULE,
1926 .release = kcm_release,
1927 .bind = sock_no_bind,
1928 .connect = sock_no_connect,
1929 .socketpair = sock_no_socketpair,
1930 .accept = sock_no_accept,
1931 .getname = sock_no_getname,
1932 .poll = datagram_poll,
1934 .listen = sock_no_listen,
1935 .shutdown = sock_no_shutdown,
1936 .setsockopt = kcm_setsockopt,
1937 .getsockopt = kcm_getsockopt,
1938 .sendmsg = kcm_sendmsg,
1939 .recvmsg = kcm_recvmsg,
1940 .mmap = sock_no_mmap,
1941 .sendpage = kcm_sendpage,
1942 .splice_read = kcm_splice_read,
1945 /* Create proto operation for kcm sockets */
1946 static int kcm_create(struct net *net, struct socket *sock,
1947 int protocol, int kern)
1949 struct kcm_net *knet = net_generic(net, kcm_net_id);
1951 struct kcm_mux *mux;
1953 switch (sock->type) {
1955 sock->ops = &kcm_dgram_ops;
1957 case SOCK_SEQPACKET:
1958 sock->ops = &kcm_seqpacket_ops;
1961 return -ESOCKTNOSUPPORT;
1964 if (protocol != KCMPROTO_CONNECTED)
1965 return -EPROTONOSUPPORT;
1967 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1971 /* Allocate a kcm mux, shared between KCM sockets */
1972 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1978 spin_lock_init(&mux->lock);
1979 spin_lock_init(&mux->rx_lock);
1980 INIT_LIST_HEAD(&mux->kcm_socks);
1981 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1982 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1984 INIT_LIST_HEAD(&mux->psocks);
1985 INIT_LIST_HEAD(&mux->psocks_ready);
1986 INIT_LIST_HEAD(&mux->psocks_avail);
1990 /* Add new MUX to list */
1991 mutex_lock(&knet->mutex);
1992 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1994 mutex_unlock(&knet->mutex);
1996 skb_queue_head_init(&mux->rx_hold_queue);
1998 /* Init KCM socket */
1999 sock_init_data(sock, sk);
2000 init_kcm_sock(kcm_sk(sk), mux);
2005 static const struct net_proto_family kcm_family_ops = {
2007 .create = kcm_create,
2008 .owner = THIS_MODULE,
2011 static __net_init int kcm_init_net(struct net *net)
2013 struct kcm_net *knet = net_generic(net, kcm_net_id);
2015 INIT_LIST_HEAD_RCU(&knet->mux_list);
2016 mutex_init(&knet->mutex);
2021 static __net_exit void kcm_exit_net(struct net *net)
2023 struct kcm_net *knet = net_generic(net, kcm_net_id);
2025 /* All KCM sockets should be closed at this point, which should mean
2026 * that all multiplexors and psocks have been destroyed.
2028 WARN_ON(!list_empty(&knet->mux_list));
2031 static struct pernet_operations kcm_net_ops = {
2032 .init = kcm_init_net,
2033 .exit = kcm_exit_net,
2035 .size = sizeof(struct kcm_net),
2038 static int __init kcm_init(void)
2042 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2043 sizeof(struct kcm_mux), 0,
2044 SLAB_HWCACHE_ALIGN, NULL);
2048 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2049 sizeof(struct kcm_psock), 0,
2050 SLAB_HWCACHE_ALIGN, NULL);
2054 kcm_wq = create_singlethread_workqueue("kkcmd");
2058 err = proto_register(&kcm_proto, 1);
2062 err = register_pernet_device(&kcm_net_ops);
2066 err = sock_register(&kcm_family_ops);
2068 goto sock_register_fail;
2070 err = kcm_proc_init();
2072 goto proc_init_fail;
2077 sock_unregister(PF_KCM);
2080 unregister_pernet_device(&kcm_net_ops);
2083 proto_unregister(&kcm_proto);
2086 kmem_cache_destroy(kcm_muxp);
2087 kmem_cache_destroy(kcm_psockp);
2090 destroy_workqueue(kcm_wq);
2095 static void __exit kcm_exit(void)
2098 sock_unregister(PF_KCM);
2099 unregister_pernet_device(&kcm_net_ops);
2100 proto_unregister(&kcm_proto);
2101 destroy_workqueue(kcm_wq);
2103 kmem_cache_destroy(kcm_muxp);
2104 kmem_cache_destroy(kcm_psockp);
2107 module_init(kcm_init);
2108 module_exit(kcm_exit);
2110 MODULE_LICENSE("GPL");
2111 MODULE_ALIAS_NETPROTO(PF_KCM);