1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel Connection Multiplexor
5 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
9 #include <linux/errno.h>
10 #include <linux/errqueue.h>
11 #include <linux/file.h>
12 #include <linux/filter.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/net.h>
17 #include <linux/netdevice.h>
18 #include <linux/poll.h>
19 #include <linux/rculist.h>
20 #include <linux/skbuff.h>
21 #include <linux/socket.h>
22 #include <linux/uaccess.h>
23 #include <linux/workqueue.h>
24 #include <linux/syscalls.h>
25 #include <linux/sched/signal.h>
28 #include <net/netns/generic.h>
30 #include <uapi/linux/kcm.h>
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)
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.bytes -
98 psock->saved_rx_bytes);
100 psock->strp.stats.msgs - psock->saved_rx_msgs;
101 psock->saved_rx_msgs = psock->strp.stats.msgs;
102 psock->saved_rx_bytes = psock->strp.stats.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);
165 /* paired with lockless reads in kcm_rfree() */
166 WRITE_ONCE(kcm->rx_wait, true);
169 static void kcm_rfree(struct sk_buff *skb)
171 struct sock *sk = skb->sk;
172 struct kcm_sock *kcm = kcm_sk(sk);
173 struct kcm_mux *mux = kcm->mux;
174 unsigned int len = skb->truesize;
176 sk_mem_uncharge(sk, len);
177 atomic_sub(len, &sk->sk_rmem_alloc);
179 /* For reading rx_wait and rx_psock without holding lock */
180 smp_mb__after_atomic();
182 if (!READ_ONCE(kcm->rx_wait) && !READ_ONCE(kcm->rx_psock) &&
183 sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
184 spin_lock_bh(&mux->rx_lock);
186 spin_unlock_bh(&mux->rx_lock);
190 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
192 struct sk_buff_head *list = &sk->sk_receive_queue;
194 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
197 if (!sk_rmem_schedule(sk, skb, skb->truesize))
204 skb->destructor = kcm_rfree;
205 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
206 sk_mem_charge(sk, skb->truesize);
208 skb_queue_tail(list, skb);
210 if (!sock_flag(sk, SOCK_DEAD))
211 sk->sk_data_ready(sk);
216 /* Requeue received messages for a kcm socket to other kcm sockets. This is
217 * called with a kcm socket is receive disabled.
220 static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
223 struct kcm_sock *kcm;
225 while ((skb = __skb_dequeue(head))) {
226 /* Reset destructor to avoid calling kcm_rcv_ready */
227 skb->destructor = sock_rfree;
230 if (list_empty(&mux->kcm_rx_waiters)) {
231 skb_queue_tail(&mux->rx_hold_queue, skb);
235 kcm = list_first_entry(&mux->kcm_rx_waiters,
236 struct kcm_sock, wait_rx_list);
238 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
239 /* Should mean socket buffer full */
240 list_del(&kcm->wait_rx_list);
241 /* paired with lockless reads in kcm_rfree() */
242 WRITE_ONCE(kcm->rx_wait, false);
244 /* Commit rx_wait to read in kcm_free */
252 /* Lower sock lock held */
253 static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
254 struct sk_buff *head)
256 struct kcm_mux *mux = psock->mux;
257 struct kcm_sock *kcm;
259 WARN_ON(psock->ready_rx_msg);
262 return psock->rx_kcm;
264 spin_lock_bh(&mux->rx_lock);
267 spin_unlock_bh(&mux->rx_lock);
268 return psock->rx_kcm;
271 kcm_update_rx_mux_stats(mux, psock);
273 if (list_empty(&mux->kcm_rx_waiters)) {
274 psock->ready_rx_msg = head;
275 strp_pause(&psock->strp);
276 list_add_tail(&psock->psock_ready_list,
278 spin_unlock_bh(&mux->rx_lock);
282 kcm = list_first_entry(&mux->kcm_rx_waiters,
283 struct kcm_sock, wait_rx_list);
284 list_del(&kcm->wait_rx_list);
285 /* paired with lockless reads in kcm_rfree() */
286 WRITE_ONCE(kcm->rx_wait, false);
289 /* paired with lockless reads in kcm_rfree() */
290 WRITE_ONCE(kcm->rx_psock, psock);
292 spin_unlock_bh(&mux->rx_lock);
297 static void kcm_done(struct kcm_sock *kcm);
299 static void kcm_done_work(struct work_struct *w)
301 kcm_done(container_of(w, struct kcm_sock, done_work));
304 /* Lower sock held */
305 static void unreserve_rx_kcm(struct kcm_psock *psock,
308 struct kcm_sock *kcm = psock->rx_kcm;
309 struct kcm_mux *mux = psock->mux;
314 spin_lock_bh(&mux->rx_lock);
316 psock->rx_kcm = NULL;
317 /* paired with lockless reads in kcm_rfree() */
318 WRITE_ONCE(kcm->rx_psock, NULL);
320 /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
325 if (unlikely(kcm->done)) {
326 spin_unlock_bh(&mux->rx_lock);
328 /* Need to run kcm_done in a task since we need to qcquire
329 * callback locks which may already be held here.
331 INIT_WORK(&kcm->done_work, kcm_done_work);
332 schedule_work(&kcm->done_work);
336 if (unlikely(kcm->rx_disabled)) {
337 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
338 } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
339 /* Check for degenerative race with rx_wait that all
340 * data was dequeued (accounted for in kcm_rfree).
344 spin_unlock_bh(&mux->rx_lock);
347 /* Lower sock lock held */
348 static void psock_data_ready(struct sock *sk)
350 struct kcm_psock *psock;
352 read_lock_bh(&sk->sk_callback_lock);
354 psock = (struct kcm_psock *)sk->sk_user_data;
356 strp_data_ready(&psock->strp);
358 read_unlock_bh(&sk->sk_callback_lock);
361 /* Called with lower sock held */
362 static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
364 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
365 struct kcm_sock *kcm;
368 kcm = reserve_rx_kcm(psock, skb);
370 /* Unable to reserve a KCM, message is held in psock and strp
376 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
377 /* Should mean socket buffer full */
378 unreserve_rx_kcm(psock, false);
383 static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
385 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
386 struct bpf_prog *prog = psock->bpf_prog;
389 res = bpf_prog_run_pin_on_cpu(prog, skb);
393 static int kcm_read_sock_done(struct strparser *strp, int err)
395 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
397 unreserve_rx_kcm(psock, true);
402 static void psock_state_change(struct sock *sk)
404 /* TCP only does a EPOLLIN for a half close. Do a EPOLLHUP here
405 * since application will normally not poll with EPOLLIN
406 * on the TCP sockets.
409 report_csk_error(sk, EPIPE);
412 static void psock_write_space(struct sock *sk)
414 struct kcm_psock *psock;
416 struct kcm_sock *kcm;
418 read_lock_bh(&sk->sk_callback_lock);
420 psock = (struct kcm_psock *)sk->sk_user_data;
421 if (unlikely(!psock))
425 spin_lock_bh(&mux->lock);
427 /* Check if the socket is reserved so someone is waiting for sending. */
429 if (kcm && !unlikely(kcm->tx_stopped))
430 queue_work(kcm_wq, &kcm->tx_work);
432 spin_unlock_bh(&mux->lock);
434 read_unlock_bh(&sk->sk_callback_lock);
437 static void unreserve_psock(struct kcm_sock *kcm);
439 /* kcm sock is locked. */
440 static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
442 struct kcm_mux *mux = kcm->mux;
443 struct kcm_psock *psock;
445 psock = kcm->tx_psock;
447 smp_rmb(); /* Must read tx_psock before tx_wait */
450 WARN_ON(kcm->tx_wait);
451 if (unlikely(psock->tx_stopped))
452 unreserve_psock(kcm);
454 return kcm->tx_psock;
457 spin_lock_bh(&mux->lock);
459 /* Check again under lock to see if psock was reserved for this
460 * psock via psock_unreserve.
462 psock = kcm->tx_psock;
463 if (unlikely(psock)) {
464 WARN_ON(kcm->tx_wait);
465 spin_unlock_bh(&mux->lock);
466 return kcm->tx_psock;
469 if (!list_empty(&mux->psocks_avail)) {
470 psock = list_first_entry(&mux->psocks_avail,
473 list_del(&psock->psock_avail_list);
475 list_del(&kcm->wait_psock_list);
476 kcm->tx_wait = false;
478 kcm->tx_psock = psock;
480 KCM_STATS_INCR(psock->stats.reserved);
481 } else if (!kcm->tx_wait) {
482 list_add_tail(&kcm->wait_psock_list,
483 &mux->kcm_tx_waiters);
487 spin_unlock_bh(&mux->lock);
493 static void psock_now_avail(struct kcm_psock *psock)
495 struct kcm_mux *mux = psock->mux;
496 struct kcm_sock *kcm;
498 if (list_empty(&mux->kcm_tx_waiters)) {
499 list_add_tail(&psock->psock_avail_list,
502 kcm = list_first_entry(&mux->kcm_tx_waiters,
505 list_del(&kcm->wait_psock_list);
506 kcm->tx_wait = false;
509 /* Commit before changing tx_psock since that is read in
510 * reserve_psock before queuing work.
514 kcm->tx_psock = psock;
515 KCM_STATS_INCR(psock->stats.reserved);
516 queue_work(kcm_wq, &kcm->tx_work);
520 /* kcm sock is locked. */
521 static void unreserve_psock(struct kcm_sock *kcm)
523 struct kcm_psock *psock;
524 struct kcm_mux *mux = kcm->mux;
526 spin_lock_bh(&mux->lock);
528 psock = kcm->tx_psock;
530 if (WARN_ON(!psock)) {
531 spin_unlock_bh(&mux->lock);
535 smp_rmb(); /* Read tx_psock before tx_wait */
537 kcm_update_tx_mux_stats(mux, psock);
539 WARN_ON(kcm->tx_wait);
541 kcm->tx_psock = NULL;
542 psock->tx_kcm = NULL;
543 KCM_STATS_INCR(psock->stats.unreserved);
545 if (unlikely(psock->tx_stopped)) {
548 list_del(&psock->psock_list);
551 fput(psock->sk->sk_socket->file);
552 kmem_cache_free(kcm_psockp, psock);
555 /* Don't put back on available list */
557 spin_unlock_bh(&mux->lock);
562 psock_now_avail(psock);
564 spin_unlock_bh(&mux->lock);
567 static void kcm_report_tx_retry(struct kcm_sock *kcm)
569 struct kcm_mux *mux = kcm->mux;
571 spin_lock_bh(&mux->lock);
572 KCM_STATS_INCR(mux->stats.tx_retries);
573 spin_unlock_bh(&mux->lock);
576 /* Write any messages ready on the kcm socket. Called with kcm sock lock
577 * held. Return bytes actually sent or error.
579 static int kcm_write_msgs(struct kcm_sock *kcm)
581 struct sock *sk = &kcm->sk;
582 struct kcm_psock *psock;
583 struct sk_buff *skb, *head;
584 struct kcm_tx_msg *txm;
585 unsigned short fragidx, frag_offset;
586 unsigned int sent, total_sent = 0;
589 kcm->tx_wait_more = false;
590 psock = kcm->tx_psock;
591 if (unlikely(psock && psock->tx_stopped)) {
592 /* A reserved psock was aborted asynchronously. Unreserve
593 * it and we'll retry the message.
595 unreserve_psock(kcm);
596 kcm_report_tx_retry(kcm);
597 if (skb_queue_empty(&sk->sk_write_queue))
600 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
602 } else if (skb_queue_empty(&sk->sk_write_queue)) {
606 head = skb_peek(&sk->sk_write_queue);
607 txm = kcm_tx_msg(head);
610 /* Send of first skbuff in queue already in progress */
611 if (WARN_ON(!psock)) {
616 frag_offset = txm->frag_offset;
617 fragidx = txm->fragidx;
624 psock = reserve_psock(kcm);
630 txm = kcm_tx_msg(head);
634 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
639 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
645 frag = &skb_shinfo(skb)->frags[fragidx];
646 if (WARN_ON(!skb_frag_size(frag))) {
651 ret = kernel_sendpage(psock->sk->sk_socket,
653 skb_frag_off(frag) + frag_offset,
654 skb_frag_size(frag) - frag_offset,
657 if (ret == -EAGAIN) {
658 /* Save state to try again when there's
659 * write space on the socket
662 txm->frag_offset = frag_offset;
663 txm->fragidx = fragidx;
670 /* Hard failure in sending message, abort this
671 * psock since it has lost framing
672 * synchronization and retry sending the
673 * message from the beginning.
675 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
677 unreserve_psock(kcm);
680 kcm_report_tx_retry(kcm);
688 KCM_STATS_ADD(psock->stats.tx_bytes, ret);
689 if (frag_offset < skb_frag_size(frag)) {
690 /* Not finished with this frag */
696 if (skb_has_frag_list(skb)) {
697 skb = skb_shinfo(skb)->frag_list;
700 } else if (skb->next) {
705 /* Successfully sent the whole packet, account for it. */
706 skb_dequeue(&sk->sk_write_queue);
708 sk->sk_wmem_queued -= sent;
710 KCM_STATS_INCR(psock->stats.tx_msgs);
711 } while ((head = skb_peek(&sk->sk_write_queue)));
714 /* Done with all queued messages. */
715 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
716 unreserve_psock(kcm);
719 /* Check if write space is available */
720 sk->sk_write_space(sk);
722 return total_sent ? : ret;
725 static void kcm_tx_work(struct work_struct *w)
727 struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
728 struct sock *sk = &kcm->sk;
733 /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
736 err = kcm_write_msgs(kcm);
738 /* Hard failure in write, report error on KCM socket */
739 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
740 report_csk_error(&kcm->sk, -err);
744 /* Primarily for SOCK_SEQPACKET sockets */
745 if (likely(sk->sk_socket) &&
746 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
747 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
748 sk->sk_write_space(sk);
755 static void kcm_push(struct kcm_sock *kcm)
757 if (kcm->tx_wait_more)
761 static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
762 int offset, size_t size, int flags)
765 struct sock *sk = sock->sk;
766 struct kcm_sock *kcm = kcm_sk(sk);
767 struct sk_buff *skb = NULL, *head = NULL;
768 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
773 if (flags & MSG_SENDPAGE_NOTLAST)
776 /* No MSG_EOR from splice, only look at MSG_MORE */
777 eor = !(flags & MSG_MORE);
781 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
788 /* Previously opened message */
790 skb = kcm_tx_msg(head)->last_skb;
791 i = skb_shinfo(skb)->nr_frags;
793 if (skb_can_coalesce(skb, i, page, offset)) {
794 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
795 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG;
799 if (i >= MAX_SKB_FRAGS) {
800 struct sk_buff *tskb;
802 tskb = alloc_skb(0, sk->sk_allocation);
805 err = sk_stream_wait_memory(sk, &timeo);
811 skb_shinfo(head)->frag_list = tskb;
816 skb->ip_summed = CHECKSUM_UNNECESSARY;
820 /* Call the sk_stream functions to manage the sndbuf mem. */
821 if (!sk_stream_memory_free(sk)) {
823 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
824 err = sk_stream_wait_memory(sk, &timeo);
829 head = alloc_skb(0, sk->sk_allocation);
832 err = sk_stream_wait_memory(sk, &timeo);
842 skb_fill_page_desc_noacc(skb, i, page, offset, size);
843 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG;
847 skb->data_len += size;
848 skb->truesize += size;
849 sk->sk_wmem_queued += size;
850 sk_mem_charge(sk, size);
854 head->data_len += size;
855 head->truesize += size;
859 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
861 /* Message complete, queue it on send buffer */
862 __skb_queue_tail(&sk->sk_write_queue, head);
864 KCM_STATS_INCR(kcm->stats.tx_msgs);
866 if (flags & MSG_BATCH) {
867 kcm->tx_wait_more = true;
868 } else if (kcm->tx_wait_more || not_busy) {
869 err = kcm_write_msgs(kcm);
871 /* We got a hard error in write_msgs but have
872 * already queued this message. Report an error
873 * in the socket, but don't affect return value
876 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
877 report_csk_error(&kcm->sk, -err);
881 /* Message not complete, save state */
883 kcm_tx_msg(head)->last_skb = skb;
886 KCM_STATS_ADD(kcm->stats.tx_bytes, size);
894 err = sk_stream_error(sk, flags, err);
896 /* make sure we wake any epoll edge trigger waiter */
897 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
898 sk->sk_write_space(sk);
904 static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
906 struct sock *sk = sock->sk;
907 struct kcm_sock *kcm = kcm_sk(sk);
908 struct sk_buff *skb = NULL, *head = NULL;
909 size_t copy, copied = 0;
910 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
911 int eor = (sock->type == SOCK_DGRAM) ?
912 !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
917 /* Per tcp_sendmsg this should be in poll */
918 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
924 /* Previously opened message */
926 skb = kcm_tx_msg(head)->last_skb;
930 /* Call the sk_stream functions to manage the sndbuf mem. */
931 if (!sk_stream_memory_free(sk)) {
933 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
934 err = sk_stream_wait_memory(sk, &timeo);
939 if (msg_data_left(msg)) {
940 /* New message, alloc head skb */
941 head = alloc_skb(0, sk->sk_allocation);
944 err = sk_stream_wait_memory(sk, &timeo);
948 head = alloc_skb(0, sk->sk_allocation);
953 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
954 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
956 skb->ip_summed = CHECKSUM_UNNECESSARY;
960 while (msg_data_left(msg)) {
962 int i = skb_shinfo(skb)->nr_frags;
963 struct page_frag *pfrag = sk_page_frag(sk);
965 if (!sk_page_frag_refill(sk, pfrag))
966 goto wait_for_memory;
968 if (!skb_can_coalesce(skb, i, pfrag->page,
970 if (i == MAX_SKB_FRAGS) {
971 struct sk_buff *tskb;
973 tskb = alloc_skb(0, sk->sk_allocation);
975 goto wait_for_memory;
978 skb_shinfo(head)->frag_list = tskb;
983 skb->ip_summed = CHECKSUM_UNNECESSARY;
989 copy = min_t(int, msg_data_left(msg),
990 pfrag->size - pfrag->offset);
992 if (!sk_wmem_schedule(sk, copy))
993 goto wait_for_memory;
995 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1002 /* Update the skb. */
1004 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1006 skb_fill_page_desc(skb, i, pfrag->page,
1007 pfrag->offset, copy);
1008 get_page(pfrag->page);
1011 pfrag->offset += copy;
1015 head->data_len += copy;
1022 err = sk_stream_wait_memory(sk, &timeo);
1028 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1031 /* Message complete, queue it on send buffer */
1032 __skb_queue_tail(&sk->sk_write_queue, head);
1033 kcm->seq_skb = NULL;
1034 KCM_STATS_INCR(kcm->stats.tx_msgs);
1037 if (msg->msg_flags & MSG_BATCH) {
1038 kcm->tx_wait_more = true;
1039 } else if (kcm->tx_wait_more || not_busy) {
1040 err = kcm_write_msgs(kcm);
1042 /* We got a hard error in write_msgs but have
1043 * already queued this message. Report an error
1044 * in the socket, but don't affect return value
1047 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1048 report_csk_error(&kcm->sk, -err);
1052 /* Message not complete, save state */
1055 kcm->seq_skb = head;
1056 kcm_tx_msg(head)->last_skb = skb;
1060 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1068 if (copied && sock->type == SOCK_SEQPACKET) {
1069 /* Wrote some bytes before encountering an
1070 * error, return partial success.
1072 goto partial_message;
1075 if (head != kcm->seq_skb)
1078 err = sk_stream_error(sk, msg->msg_flags, err);
1080 /* make sure we wake any epoll edge trigger waiter */
1081 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1082 sk->sk_write_space(sk);
1088 static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1089 long timeo, int *err)
1091 struct sk_buff *skb;
1093 while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1095 *err = sock_error(sk);
1099 if (sock_flag(sk, SOCK_DONE))
1102 if ((flags & MSG_DONTWAIT) || !timeo) {
1107 sk_wait_data(sk, &timeo, NULL);
1109 /* Handle signals */
1110 if (signal_pending(current)) {
1111 *err = sock_intr_errno(timeo);
1119 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1120 size_t len, int flags)
1122 struct sock *sk = sock->sk;
1123 struct kcm_sock *kcm = kcm_sk(sk);
1126 struct strp_msg *stm;
1128 struct sk_buff *skb;
1130 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1134 skb = kcm_wait_data(sk, flags, timeo, &err);
1138 /* Okay, have a message on the receive queue */
1140 stm = strp_msg(skb);
1142 if (len > stm->full_len)
1143 len = stm->full_len;
1145 err = skb_copy_datagram_msg(skb, stm->offset, msg, len);
1150 if (likely(!(flags & MSG_PEEK))) {
1151 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1152 if (copied < stm->full_len) {
1153 if (sock->type == SOCK_DGRAM) {
1154 /* Truncated message */
1155 msg->msg_flags |= MSG_TRUNC;
1158 stm->offset += copied;
1159 stm->full_len -= copied;
1162 /* Finished with message */
1163 msg->msg_flags |= MSG_EOR;
1164 KCM_STATS_INCR(kcm->stats.rx_msgs);
1165 skb_unlink(skb, &sk->sk_receive_queue);
1173 return copied ? : err;
1176 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1177 struct pipe_inode_info *pipe, size_t len,
1180 struct sock *sk = sock->sk;
1181 struct kcm_sock *kcm = kcm_sk(sk);
1183 struct strp_msg *stm;
1186 struct sk_buff *skb;
1188 /* Only support splice for SOCKSEQPACKET */
1190 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1194 skb = kcm_wait_data(sk, flags, timeo, &err);
1198 /* Okay, have a message on the receive queue */
1200 stm = strp_msg(skb);
1202 if (len > stm->full_len)
1203 len = stm->full_len;
1205 copied = skb_splice_bits(skb, sk, stm->offset, pipe, len, flags);
1211 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1213 stm->offset += copied;
1214 stm->full_len -= copied;
1216 /* We have no way to return MSG_EOR. If all the bytes have been
1217 * read we still leave the message in the receive socket buffer.
1218 * A subsequent recvmsg needs to be done to return MSG_EOR and
1219 * finish reading the message.
1232 /* kcm sock lock held */
1233 static void kcm_recv_disable(struct kcm_sock *kcm)
1235 struct kcm_mux *mux = kcm->mux;
1237 if (kcm->rx_disabled)
1240 spin_lock_bh(&mux->rx_lock);
1242 kcm->rx_disabled = 1;
1244 /* If a psock is reserved we'll do cleanup in unreserve */
1245 if (!kcm->rx_psock) {
1247 list_del(&kcm->wait_rx_list);
1248 /* paired with lockless reads in kcm_rfree() */
1249 WRITE_ONCE(kcm->rx_wait, false);
1252 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1255 spin_unlock_bh(&mux->rx_lock);
1258 /* kcm sock lock held */
1259 static void kcm_recv_enable(struct kcm_sock *kcm)
1261 struct kcm_mux *mux = kcm->mux;
1263 if (!kcm->rx_disabled)
1266 spin_lock_bh(&mux->rx_lock);
1268 kcm->rx_disabled = 0;
1271 spin_unlock_bh(&mux->rx_lock);
1274 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1275 sockptr_t optval, unsigned int optlen)
1277 struct kcm_sock *kcm = kcm_sk(sock->sk);
1281 if (level != SOL_KCM)
1282 return -ENOPROTOOPT;
1284 if (optlen < sizeof(int))
1287 if (copy_from_sockptr(&val, optval, sizeof(int)))
1290 valbool = val ? 1 : 0;
1293 case KCM_RECV_DISABLE:
1294 lock_sock(&kcm->sk);
1296 kcm_recv_disable(kcm);
1298 kcm_recv_enable(kcm);
1299 release_sock(&kcm->sk);
1308 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1309 char __user *optval, int __user *optlen)
1311 struct kcm_sock *kcm = kcm_sk(sock->sk);
1314 if (level != SOL_KCM)
1315 return -ENOPROTOOPT;
1317 if (get_user(len, optlen))
1320 len = min_t(unsigned int, len, sizeof(int));
1325 case KCM_RECV_DISABLE:
1326 val = kcm->rx_disabled;
1329 return -ENOPROTOOPT;
1332 if (put_user(len, optlen))
1334 if (copy_to_user(optval, &val, len))
1339 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1341 struct kcm_sock *tkcm;
1342 struct list_head *head;
1345 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1346 * we set sk_state, otherwise epoll_wait always returns right away with
1349 kcm->sk.sk_state = TCP_ESTABLISHED;
1351 /* Add to mux's kcm sockets list */
1353 spin_lock_bh(&mux->lock);
1355 head = &mux->kcm_socks;
1356 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1357 if (tkcm->index != index)
1359 head = &tkcm->kcm_sock_list;
1363 list_add(&kcm->kcm_sock_list, head);
1366 mux->kcm_socks_cnt++;
1367 spin_unlock_bh(&mux->lock);
1369 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1371 spin_lock_bh(&mux->rx_lock);
1373 spin_unlock_bh(&mux->rx_lock);
1376 static int kcm_attach(struct socket *sock, struct socket *csock,
1377 struct bpf_prog *prog)
1379 struct kcm_sock *kcm = kcm_sk(sock->sk);
1380 struct kcm_mux *mux = kcm->mux;
1382 struct kcm_psock *psock = NULL, *tpsock;
1383 struct list_head *head;
1385 static const struct strp_callbacks cb = {
1386 .rcv_msg = kcm_rcv_strparser,
1387 .parse_msg = kcm_parse_func_strparser,
1388 .read_sock_done = kcm_read_sock_done,
1398 /* Only allow TCP sockets to be attached for now */
1399 if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1400 csk->sk_protocol != IPPROTO_TCP) {
1405 /* Don't allow listeners or closed sockets */
1406 if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1411 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1419 psock->bpf_prog = prog;
1421 write_lock_bh(&csk->sk_callback_lock);
1423 /* Check if sk_user_data is already by KCM or someone else.
1424 * Must be done under lock to prevent race conditions.
1426 if (csk->sk_user_data) {
1427 write_unlock_bh(&csk->sk_callback_lock);
1428 kmem_cache_free(kcm_psockp, psock);
1433 err = strp_init(&psock->strp, csk, &cb);
1435 write_unlock_bh(&csk->sk_callback_lock);
1436 kmem_cache_free(kcm_psockp, psock);
1440 psock->save_data_ready = csk->sk_data_ready;
1441 psock->save_write_space = csk->sk_write_space;
1442 psock->save_state_change = csk->sk_state_change;
1443 csk->sk_user_data = psock;
1444 csk->sk_data_ready = psock_data_ready;
1445 csk->sk_write_space = psock_write_space;
1446 csk->sk_state_change = psock_state_change;
1448 write_unlock_bh(&csk->sk_callback_lock);
1452 /* Finished initialization, now add the psock to the MUX. */
1453 spin_lock_bh(&mux->lock);
1454 head = &mux->psocks;
1455 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1456 if (tpsock->index != index)
1458 head = &tpsock->psock_list;
1462 list_add(&psock->psock_list, head);
1463 psock->index = index;
1465 KCM_STATS_INCR(mux->stats.psock_attach);
1467 psock_now_avail(psock);
1468 spin_unlock_bh(&mux->lock);
1470 /* Schedule RX work in case there are already bytes queued */
1471 strp_check_rcv(&psock->strp);
1479 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1481 struct socket *csock;
1482 struct bpf_prog *prog;
1485 csock = sockfd_lookup(info->fd, &err);
1489 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1491 err = PTR_ERR(prog);
1495 err = kcm_attach(sock, csock, prog);
1501 /* Keep reference on file also */
1509 static void kcm_unattach(struct kcm_psock *psock)
1511 struct sock *csk = psock->sk;
1512 struct kcm_mux *mux = psock->mux;
1516 /* Stop getting callbacks from TCP socket. After this there should
1517 * be no way to reserve a kcm for this psock.
1519 write_lock_bh(&csk->sk_callback_lock);
1520 csk->sk_user_data = NULL;
1521 csk->sk_data_ready = psock->save_data_ready;
1522 csk->sk_write_space = psock->save_write_space;
1523 csk->sk_state_change = psock->save_state_change;
1524 strp_stop(&psock->strp);
1526 if (WARN_ON(psock->rx_kcm)) {
1527 write_unlock_bh(&csk->sk_callback_lock);
1532 spin_lock_bh(&mux->rx_lock);
1534 /* Stop receiver activities. After this point psock should not be
1535 * able to get onto ready list either through callbacks or work.
1537 if (psock->ready_rx_msg) {
1538 list_del(&psock->psock_ready_list);
1539 kfree_skb(psock->ready_rx_msg);
1540 psock->ready_rx_msg = NULL;
1541 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1544 spin_unlock_bh(&mux->rx_lock);
1546 write_unlock_bh(&csk->sk_callback_lock);
1548 /* Call strp_done without sock lock */
1550 strp_done(&psock->strp);
1553 bpf_prog_put(psock->bpf_prog);
1555 spin_lock_bh(&mux->lock);
1557 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1558 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1560 KCM_STATS_INCR(mux->stats.psock_unattach);
1562 if (psock->tx_kcm) {
1563 /* psock was reserved. Just mark it finished and we will clean
1564 * up in the kcm paths, we need kcm lock which can not be
1567 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1568 spin_unlock_bh(&mux->lock);
1570 /* We are unattaching a socket that is reserved. Abort the
1571 * socket since we may be out of sync in sending on it. We need
1572 * to do this without the mux lock.
1574 kcm_abort_tx_psock(psock, EPIPE, false);
1576 spin_lock_bh(&mux->lock);
1577 if (!psock->tx_kcm) {
1578 /* psock now unreserved in window mux was unlocked */
1583 /* Commit done before queuing work to process it */
1586 /* Queue tx work to make sure psock->done is handled */
1587 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1588 spin_unlock_bh(&mux->lock);
1591 if (!psock->tx_stopped)
1592 list_del(&psock->psock_avail_list);
1593 list_del(&psock->psock_list);
1595 spin_unlock_bh(&mux->lock);
1598 fput(csk->sk_socket->file);
1599 kmem_cache_free(kcm_psockp, psock);
1605 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1607 struct kcm_sock *kcm = kcm_sk(sock->sk);
1608 struct kcm_mux *mux = kcm->mux;
1609 struct kcm_psock *psock;
1610 struct socket *csock;
1614 csock = sockfd_lookup(info->fd, &err);
1626 spin_lock_bh(&mux->lock);
1628 list_for_each_entry(psock, &mux->psocks, psock_list) {
1629 if (psock->sk != csk)
1632 /* Found the matching psock */
1634 if (psock->unattaching || WARN_ON(psock->done)) {
1639 psock->unattaching = 1;
1641 spin_unlock_bh(&mux->lock);
1643 /* Lower socket lock should already be held */
1644 kcm_unattach(psock);
1650 spin_unlock_bh(&mux->lock);
1657 static struct proto kcm_proto = {
1659 .owner = THIS_MODULE,
1660 .obj_size = sizeof(struct kcm_sock),
1663 /* Clone a kcm socket. */
1664 static struct file *kcm_clone(struct socket *osock)
1666 struct socket *newsock;
1669 newsock = sock_alloc();
1671 return ERR_PTR(-ENFILE);
1673 newsock->type = osock->type;
1674 newsock->ops = osock->ops;
1676 __module_get(newsock->ops->owner);
1678 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1681 sock_release(newsock);
1682 return ERR_PTR(-ENOMEM);
1684 sock_init_data(newsock, newsk);
1685 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1687 return sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1690 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1695 case SIOCKCMATTACH: {
1696 struct kcm_attach info;
1698 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1701 err = kcm_attach_ioctl(sock, &info);
1705 case SIOCKCMUNATTACH: {
1706 struct kcm_unattach info;
1708 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1711 err = kcm_unattach_ioctl(sock, &info);
1715 case SIOCKCMCLONE: {
1716 struct kcm_clone info;
1719 info.fd = get_unused_fd_flags(0);
1720 if (unlikely(info.fd < 0))
1723 file = kcm_clone(sock);
1725 put_unused_fd(info.fd);
1726 return PTR_ERR(file);
1728 if (copy_to_user((void __user *)arg, &info,
1730 put_unused_fd(info.fd);
1734 fd_install(info.fd, file);
1746 static void free_mux(struct rcu_head *rcu)
1748 struct kcm_mux *mux = container_of(rcu,
1749 struct kcm_mux, rcu);
1751 kmem_cache_free(kcm_muxp, mux);
1754 static void release_mux(struct kcm_mux *mux)
1756 struct kcm_net *knet = mux->knet;
1757 struct kcm_psock *psock, *tmp_psock;
1759 /* Release psocks */
1760 list_for_each_entry_safe(psock, tmp_psock,
1761 &mux->psocks, psock_list) {
1762 if (!WARN_ON(psock->unattaching))
1763 kcm_unattach(psock);
1766 if (WARN_ON(mux->psocks_cnt))
1769 __skb_queue_purge(&mux->rx_hold_queue);
1771 mutex_lock(&knet->mutex);
1772 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1773 aggregate_psock_stats(&mux->aggregate_psock_stats,
1774 &knet->aggregate_psock_stats);
1775 aggregate_strp_stats(&mux->aggregate_strp_stats,
1776 &knet->aggregate_strp_stats);
1777 list_del_rcu(&mux->kcm_mux_list);
1779 mutex_unlock(&knet->mutex);
1781 call_rcu(&mux->rcu, free_mux);
1784 static void kcm_done(struct kcm_sock *kcm)
1786 struct kcm_mux *mux = kcm->mux;
1787 struct sock *sk = &kcm->sk;
1790 spin_lock_bh(&mux->rx_lock);
1791 if (kcm->rx_psock) {
1792 /* Cleanup in unreserve_rx_kcm */
1794 kcm->rx_disabled = 1;
1796 spin_unlock_bh(&mux->rx_lock);
1801 list_del(&kcm->wait_rx_list);
1802 /* paired with lockless reads in kcm_rfree() */
1803 WRITE_ONCE(kcm->rx_wait, false);
1805 /* Move any pending receive messages to other kcm sockets */
1806 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1808 spin_unlock_bh(&mux->rx_lock);
1810 if (WARN_ON(sk_rmem_alloc_get(sk)))
1813 /* Detach from MUX */
1814 spin_lock_bh(&mux->lock);
1816 list_del(&kcm->kcm_sock_list);
1817 mux->kcm_socks_cnt--;
1818 socks_cnt = mux->kcm_socks_cnt;
1820 spin_unlock_bh(&mux->lock);
1823 /* We are done with the mux now. */
1827 WARN_ON(kcm->rx_wait);
1832 /* Called by kcm_release to close a KCM socket.
1833 * If this is the last KCM socket on the MUX, destroy the MUX.
1835 static int kcm_release(struct socket *sock)
1837 struct sock *sk = sock->sk;
1838 struct kcm_sock *kcm;
1839 struct kcm_mux *mux;
1840 struct kcm_psock *psock;
1850 kfree_skb(kcm->seq_skb);
1852 /* Purge queue under lock to avoid race condition with tx_work trying
1853 * to act when queue is nonempty. If tx_work runs after this point
1854 * it will just return.
1856 __skb_queue_purge(&sk->sk_write_queue);
1858 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1859 * get a writespace callback. This prevents further work being queued
1860 * from the callback (unbinding the psock occurs after canceling work.
1862 kcm->tx_stopped = 1;
1866 spin_lock_bh(&mux->lock);
1868 /* Take of tx_wait list, after this point there should be no way
1869 * that a psock will be assigned to this kcm.
1871 list_del(&kcm->wait_psock_list);
1872 kcm->tx_wait = false;
1874 spin_unlock_bh(&mux->lock);
1876 /* Cancel work. After this point there should be no outside references
1877 * to the kcm socket.
1879 cancel_work_sync(&kcm->tx_work);
1882 psock = kcm->tx_psock;
1884 /* A psock was reserved, so we need to kill it since it
1885 * may already have some bytes queued from a message. We
1886 * need to do this after removing kcm from tx_wait list.
1888 kcm_abort_tx_psock(psock, EPIPE, false);
1889 unreserve_psock(kcm);
1893 WARN_ON(kcm->tx_wait);
1894 WARN_ON(kcm->tx_psock);
1903 static const struct proto_ops kcm_dgram_ops = {
1905 .owner = THIS_MODULE,
1906 .release = kcm_release,
1907 .bind = sock_no_bind,
1908 .connect = sock_no_connect,
1909 .socketpair = sock_no_socketpair,
1910 .accept = sock_no_accept,
1911 .getname = sock_no_getname,
1912 .poll = datagram_poll,
1914 .listen = sock_no_listen,
1915 .shutdown = sock_no_shutdown,
1916 .setsockopt = kcm_setsockopt,
1917 .getsockopt = kcm_getsockopt,
1918 .sendmsg = kcm_sendmsg,
1919 .recvmsg = kcm_recvmsg,
1920 .mmap = sock_no_mmap,
1921 .sendpage = kcm_sendpage,
1924 static const struct proto_ops kcm_seqpacket_ops = {
1926 .owner = THIS_MODULE,
1927 .release = kcm_release,
1928 .bind = sock_no_bind,
1929 .connect = sock_no_connect,
1930 .socketpair = sock_no_socketpair,
1931 .accept = sock_no_accept,
1932 .getname = sock_no_getname,
1933 .poll = datagram_poll,
1935 .listen = sock_no_listen,
1936 .shutdown = sock_no_shutdown,
1937 .setsockopt = kcm_setsockopt,
1938 .getsockopt = kcm_getsockopt,
1939 .sendmsg = kcm_sendmsg,
1940 .recvmsg = kcm_recvmsg,
1941 .mmap = sock_no_mmap,
1942 .sendpage = kcm_sendpage,
1943 .splice_read = kcm_splice_read,
1946 /* Create proto operation for kcm sockets */
1947 static int kcm_create(struct net *net, struct socket *sock,
1948 int protocol, int kern)
1950 struct kcm_net *knet = net_generic(net, kcm_net_id);
1952 struct kcm_mux *mux;
1954 switch (sock->type) {
1956 sock->ops = &kcm_dgram_ops;
1958 case SOCK_SEQPACKET:
1959 sock->ops = &kcm_seqpacket_ops;
1962 return -ESOCKTNOSUPPORT;
1965 if (protocol != KCMPROTO_CONNECTED)
1966 return -EPROTONOSUPPORT;
1968 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1972 /* Allocate a kcm mux, shared between KCM sockets */
1973 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1979 spin_lock_init(&mux->lock);
1980 spin_lock_init(&mux->rx_lock);
1981 INIT_LIST_HEAD(&mux->kcm_socks);
1982 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1983 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1985 INIT_LIST_HEAD(&mux->psocks);
1986 INIT_LIST_HEAD(&mux->psocks_ready);
1987 INIT_LIST_HEAD(&mux->psocks_avail);
1991 /* Add new MUX to list */
1992 mutex_lock(&knet->mutex);
1993 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1995 mutex_unlock(&knet->mutex);
1997 skb_queue_head_init(&mux->rx_hold_queue);
1999 /* Init KCM socket */
2000 sock_init_data(sock, sk);
2001 init_kcm_sock(kcm_sk(sk), mux);
2006 static const struct net_proto_family kcm_family_ops = {
2008 .create = kcm_create,
2009 .owner = THIS_MODULE,
2012 static __net_init int kcm_init_net(struct net *net)
2014 struct kcm_net *knet = net_generic(net, kcm_net_id);
2016 INIT_LIST_HEAD_RCU(&knet->mux_list);
2017 mutex_init(&knet->mutex);
2022 static __net_exit void kcm_exit_net(struct net *net)
2024 struct kcm_net *knet = net_generic(net, kcm_net_id);
2026 /* All KCM sockets should be closed at this point, which should mean
2027 * that all multiplexors and psocks have been destroyed.
2029 WARN_ON(!list_empty(&knet->mux_list));
2032 static struct pernet_operations kcm_net_ops = {
2033 .init = kcm_init_net,
2034 .exit = kcm_exit_net,
2036 .size = sizeof(struct kcm_net),
2039 static int __init kcm_init(void)
2043 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2044 sizeof(struct kcm_mux), 0,
2045 SLAB_HWCACHE_ALIGN, NULL);
2049 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2050 sizeof(struct kcm_psock), 0,
2051 SLAB_HWCACHE_ALIGN, NULL);
2055 kcm_wq = create_singlethread_workqueue("kkcmd");
2059 err = proto_register(&kcm_proto, 1);
2063 err = register_pernet_device(&kcm_net_ops);
2067 err = sock_register(&kcm_family_ops);
2069 goto sock_register_fail;
2071 err = kcm_proc_init();
2073 goto proc_init_fail;
2078 sock_unregister(PF_KCM);
2081 unregister_pernet_device(&kcm_net_ops);
2084 proto_unregister(&kcm_proto);
2087 kmem_cache_destroy(kcm_muxp);
2088 kmem_cache_destroy(kcm_psockp);
2091 destroy_workqueue(kcm_wq);
2096 static void __exit kcm_exit(void)
2099 sock_unregister(PF_KCM);
2100 unregister_pernet_device(&kcm_net_ops);
2101 proto_unregister(&kcm_proto);
2102 destroy_workqueue(kcm_wq);
2104 kmem_cache_destroy(kcm_muxp);
2105 kmem_cache_destroy(kcm_psockp);
2108 module_init(kcm_init);
2109 module_exit(kcm_exit);
2111 MODULE_LICENSE("GPL");
2112 MODULE_ALIAS_NETPROTO(PF_KCM);