Loading...
Note: File does not exist in v3.5.6.
1/*
2 * Kernel Connection Multiplexor
3 *
4 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2
8 * as published by the Free Software Foundation.
9 */
10
11#include <linux/bpf.h>
12#include <linux/errno.h>
13#include <linux/errqueue.h>
14#include <linux/file.h>
15#include <linux/in.h>
16#include <linux/kernel.h>
17#include <linux/module.h>
18#include <linux/net.h>
19#include <linux/netdevice.h>
20#include <linux/poll.h>
21#include <linux/rculist.h>
22#include <linux/skbuff.h>
23#include <linux/socket.h>
24#include <linux/uaccess.h>
25#include <linux/workqueue.h>
26#include <linux/syscalls.h>
27#include <net/kcm.h>
28#include <net/netns/generic.h>
29#include <net/sock.h>
30#include <uapi/linux/kcm.h>
31
32unsigned int kcm_net_id;
33
34static struct kmem_cache *kcm_psockp __read_mostly;
35static struct kmem_cache *kcm_muxp __read_mostly;
36static struct workqueue_struct *kcm_wq;
37
38static inline struct kcm_sock *kcm_sk(const struct sock *sk)
39{
40 return (struct kcm_sock *)sk;
41}
42
43static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
44{
45 return (struct kcm_tx_msg *)skb->cb;
46}
47
48static void report_csk_error(struct sock *csk, int err)
49{
50 csk->sk_err = EPIPE;
51 csk->sk_error_report(csk);
52}
53
54static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
55 bool wakeup_kcm)
56{
57 struct sock *csk = psock->sk;
58 struct kcm_mux *mux = psock->mux;
59
60 /* Unrecoverable error in transmit */
61
62 spin_lock_bh(&mux->lock);
63
64 if (psock->tx_stopped) {
65 spin_unlock_bh(&mux->lock);
66 return;
67 }
68
69 psock->tx_stopped = 1;
70 KCM_STATS_INCR(psock->stats.tx_aborts);
71
72 if (!psock->tx_kcm) {
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.
80 */
81 smp_mb();
82
83 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
84 }
85
86 spin_unlock_bh(&mux->lock);
87
88 /* Report error on lower socket */
89 report_csk_error(csk, err);
90}
91
92/* RX mux lock held. */
93static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
94 struct kcm_psock *psock)
95{
96 STRP_STATS_ADD(mux->stats.rx_bytes,
97 psock->strp.stats.rx_bytes -
98 psock->saved_rx_bytes);
99 mux->stats.rx_msgs +=
100 psock->strp.stats.rx_msgs - psock->saved_rx_msgs;
101 psock->saved_rx_msgs = psock->strp.stats.rx_msgs;
102 psock->saved_rx_bytes = psock->strp.stats.rx_bytes;
103}
104
105static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
106 struct kcm_psock *psock)
107{
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;
114}
115
116static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
117
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.
121 */
122static void kcm_rcv_ready(struct kcm_sock *kcm)
123{
124 struct kcm_mux *mux = kcm->mux;
125 struct kcm_psock *psock;
126 struct sk_buff *skb;
127
128 if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
129 return;
130
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));
136 return;
137 }
138 }
139
140 while (!list_empty(&mux->psocks_ready)) {
141 psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
142 psock_ready_list);
143
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));
147 return;
148 }
149
150 /* Consumed the ready message on the psock. Schedule rx_work to
151 * get more messages.
152 */
153 list_del(&psock->psock_ready_list);
154 psock->ready_rx_msg = NULL;
155 /* Commit clearing of ready_rx_msg for queuing work */
156 smp_mb();
157
158 strp_unpause(&psock->strp);
159 strp_check_rcv(&psock->strp);
160 }
161
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 kcm->rx_wait = true;
166}
167
168static void kcm_rfree(struct sk_buff *skb)
169{
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;
174
175 sk_mem_uncharge(sk, len);
176 atomic_sub(len, &sk->sk_rmem_alloc);
177
178 /* For reading rx_wait and rx_psock without holding lock */
179 smp_mb__after_atomic();
180
181 if (!kcm->rx_wait && !kcm->rx_psock &&
182 sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
183 spin_lock_bh(&mux->rx_lock);
184 kcm_rcv_ready(kcm);
185 spin_unlock_bh(&mux->rx_lock);
186 }
187}
188
189static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
190{
191 struct sk_buff_head *list = &sk->sk_receive_queue;
192
193 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
194 return -ENOMEM;
195
196 if (!sk_rmem_schedule(sk, skb, skb->truesize))
197 return -ENOBUFS;
198
199 skb->dev = NULL;
200
201 skb_orphan(skb);
202 skb->sk = sk;
203 skb->destructor = kcm_rfree;
204 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
205 sk_mem_charge(sk, skb->truesize);
206
207 skb_queue_tail(list, skb);
208
209 if (!sock_flag(sk, SOCK_DEAD))
210 sk->sk_data_ready(sk);
211
212 return 0;
213}
214
215/* Requeue received messages for a kcm socket to other kcm sockets. This is
216 * called with a kcm socket is receive disabled.
217 * RX mux lock held.
218 */
219static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
220{
221 struct sk_buff *skb;
222 struct kcm_sock *kcm;
223
224 while ((skb = __skb_dequeue(head))) {
225 /* Reset destructor to avoid calling kcm_rcv_ready */
226 skb->destructor = sock_rfree;
227 skb_orphan(skb);
228try_again:
229 if (list_empty(&mux->kcm_rx_waiters)) {
230 skb_queue_tail(&mux->rx_hold_queue, skb);
231 continue;
232 }
233
234 kcm = list_first_entry(&mux->kcm_rx_waiters,
235 struct kcm_sock, wait_rx_list);
236
237 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
238 /* Should mean socket buffer full */
239 list_del(&kcm->wait_rx_list);
240 kcm->rx_wait = false;
241
242 /* Commit rx_wait to read in kcm_free */
243 smp_wmb();
244
245 goto try_again;
246 }
247 }
248}
249
250/* Lower sock lock held */
251static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
252 struct sk_buff *head)
253{
254 struct kcm_mux *mux = psock->mux;
255 struct kcm_sock *kcm;
256
257 WARN_ON(psock->ready_rx_msg);
258
259 if (psock->rx_kcm)
260 return psock->rx_kcm;
261
262 spin_lock_bh(&mux->rx_lock);
263
264 if (psock->rx_kcm) {
265 spin_unlock_bh(&mux->rx_lock);
266 return psock->rx_kcm;
267 }
268
269 kcm_update_rx_mux_stats(mux, psock);
270
271 if (list_empty(&mux->kcm_rx_waiters)) {
272 psock->ready_rx_msg = head;
273 strp_pause(&psock->strp);
274 list_add_tail(&psock->psock_ready_list,
275 &mux->psocks_ready);
276 spin_unlock_bh(&mux->rx_lock);
277 return NULL;
278 }
279
280 kcm = list_first_entry(&mux->kcm_rx_waiters,
281 struct kcm_sock, wait_rx_list);
282 list_del(&kcm->wait_rx_list);
283 kcm->rx_wait = false;
284
285 psock->rx_kcm = kcm;
286 kcm->rx_psock = psock;
287
288 spin_unlock_bh(&mux->rx_lock);
289
290 return kcm;
291}
292
293static void kcm_done(struct kcm_sock *kcm);
294
295static void kcm_done_work(struct work_struct *w)
296{
297 kcm_done(container_of(w, struct kcm_sock, done_work));
298}
299
300/* Lower sock held */
301static void unreserve_rx_kcm(struct kcm_psock *psock,
302 bool rcv_ready)
303{
304 struct kcm_sock *kcm = psock->rx_kcm;
305 struct kcm_mux *mux = psock->mux;
306
307 if (!kcm)
308 return;
309
310 spin_lock_bh(&mux->rx_lock);
311
312 psock->rx_kcm = NULL;
313 kcm->rx_psock = NULL;
314
315 /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
316 * kcm_rfree
317 */
318 smp_mb();
319
320 if (unlikely(kcm->done)) {
321 spin_unlock_bh(&mux->rx_lock);
322
323 /* Need to run kcm_done in a task since we need to qcquire
324 * callback locks which may already be held here.
325 */
326 INIT_WORK(&kcm->done_work, kcm_done_work);
327 schedule_work(&kcm->done_work);
328 return;
329 }
330
331 if (unlikely(kcm->rx_disabled)) {
332 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
333 } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
334 /* Check for degenerative race with rx_wait that all
335 * data was dequeued (accounted for in kcm_rfree).
336 */
337 kcm_rcv_ready(kcm);
338 }
339 spin_unlock_bh(&mux->rx_lock);
340}
341
342/* Lower sock lock held */
343static void psock_data_ready(struct sock *sk)
344{
345 struct kcm_psock *psock;
346
347 read_lock_bh(&sk->sk_callback_lock);
348
349 psock = (struct kcm_psock *)sk->sk_user_data;
350 if (likely(psock))
351 strp_data_ready(&psock->strp);
352
353 read_unlock_bh(&sk->sk_callback_lock);
354}
355
356/* Called with lower sock held */
357static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
358{
359 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
360 struct kcm_sock *kcm;
361
362try_queue:
363 kcm = reserve_rx_kcm(psock, skb);
364 if (!kcm) {
365 /* Unable to reserve a KCM, message is held in psock and strp
366 * is paused.
367 */
368 return;
369 }
370
371 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
372 /* Should mean socket buffer full */
373 unreserve_rx_kcm(psock, false);
374 goto try_queue;
375 }
376}
377
378static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
379{
380 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
381 struct bpf_prog *prog = psock->bpf_prog;
382
383 return (*prog->bpf_func)(skb, prog->insnsi);
384}
385
386static int kcm_read_sock_done(struct strparser *strp, int err)
387{
388 struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
389
390 unreserve_rx_kcm(psock, true);
391
392 return err;
393}
394
395static void psock_state_change(struct sock *sk)
396{
397 /* TCP only does a POLLIN for a half close. Do a POLLHUP here
398 * since application will normally not poll with POLLIN
399 * on the TCP sockets.
400 */
401
402 report_csk_error(sk, EPIPE);
403}
404
405static void psock_write_space(struct sock *sk)
406{
407 struct kcm_psock *psock;
408 struct kcm_mux *mux;
409 struct kcm_sock *kcm;
410
411 read_lock_bh(&sk->sk_callback_lock);
412
413 psock = (struct kcm_psock *)sk->sk_user_data;
414 if (unlikely(!psock))
415 goto out;
416 mux = psock->mux;
417
418 spin_lock_bh(&mux->lock);
419
420 /* Check if the socket is reserved so someone is waiting for sending. */
421 kcm = psock->tx_kcm;
422 if (kcm && !unlikely(kcm->tx_stopped))
423 queue_work(kcm_wq, &kcm->tx_work);
424
425 spin_unlock_bh(&mux->lock);
426out:
427 read_unlock_bh(&sk->sk_callback_lock);
428}
429
430static void unreserve_psock(struct kcm_sock *kcm);
431
432/* kcm sock is locked. */
433static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
434{
435 struct kcm_mux *mux = kcm->mux;
436 struct kcm_psock *psock;
437
438 psock = kcm->tx_psock;
439
440 smp_rmb(); /* Must read tx_psock before tx_wait */
441
442 if (psock) {
443 WARN_ON(kcm->tx_wait);
444 if (unlikely(psock->tx_stopped))
445 unreserve_psock(kcm);
446 else
447 return kcm->tx_psock;
448 }
449
450 spin_lock_bh(&mux->lock);
451
452 /* Check again under lock to see if psock was reserved for this
453 * psock via psock_unreserve.
454 */
455 psock = kcm->tx_psock;
456 if (unlikely(psock)) {
457 WARN_ON(kcm->tx_wait);
458 spin_unlock_bh(&mux->lock);
459 return kcm->tx_psock;
460 }
461
462 if (!list_empty(&mux->psocks_avail)) {
463 psock = list_first_entry(&mux->psocks_avail,
464 struct kcm_psock,
465 psock_avail_list);
466 list_del(&psock->psock_avail_list);
467 if (kcm->tx_wait) {
468 list_del(&kcm->wait_psock_list);
469 kcm->tx_wait = false;
470 }
471 kcm->tx_psock = psock;
472 psock->tx_kcm = kcm;
473 KCM_STATS_INCR(psock->stats.reserved);
474 } else if (!kcm->tx_wait) {
475 list_add_tail(&kcm->wait_psock_list,
476 &mux->kcm_tx_waiters);
477 kcm->tx_wait = true;
478 }
479
480 spin_unlock_bh(&mux->lock);
481
482 return psock;
483}
484
485/* mux lock held */
486static void psock_now_avail(struct kcm_psock *psock)
487{
488 struct kcm_mux *mux = psock->mux;
489 struct kcm_sock *kcm;
490
491 if (list_empty(&mux->kcm_tx_waiters)) {
492 list_add_tail(&psock->psock_avail_list,
493 &mux->psocks_avail);
494 } else {
495 kcm = list_first_entry(&mux->kcm_tx_waiters,
496 struct kcm_sock,
497 wait_psock_list);
498 list_del(&kcm->wait_psock_list);
499 kcm->tx_wait = false;
500 psock->tx_kcm = kcm;
501
502 /* Commit before changing tx_psock since that is read in
503 * reserve_psock before queuing work.
504 */
505 smp_mb();
506
507 kcm->tx_psock = psock;
508 KCM_STATS_INCR(psock->stats.reserved);
509 queue_work(kcm_wq, &kcm->tx_work);
510 }
511}
512
513/* kcm sock is locked. */
514static void unreserve_psock(struct kcm_sock *kcm)
515{
516 struct kcm_psock *psock;
517 struct kcm_mux *mux = kcm->mux;
518
519 spin_lock_bh(&mux->lock);
520
521 psock = kcm->tx_psock;
522
523 if (WARN_ON(!psock)) {
524 spin_unlock_bh(&mux->lock);
525 return;
526 }
527
528 smp_rmb(); /* Read tx_psock before tx_wait */
529
530 kcm_update_tx_mux_stats(mux, psock);
531
532 WARN_ON(kcm->tx_wait);
533
534 kcm->tx_psock = NULL;
535 psock->tx_kcm = NULL;
536 KCM_STATS_INCR(psock->stats.unreserved);
537
538 if (unlikely(psock->tx_stopped)) {
539 if (psock->done) {
540 /* Deferred free */
541 list_del(&psock->psock_list);
542 mux->psocks_cnt--;
543 sock_put(psock->sk);
544 fput(psock->sk->sk_socket->file);
545 kmem_cache_free(kcm_psockp, psock);
546 }
547
548 /* Don't put back on available list */
549
550 spin_unlock_bh(&mux->lock);
551
552 return;
553 }
554
555 psock_now_avail(psock);
556
557 spin_unlock_bh(&mux->lock);
558}
559
560static void kcm_report_tx_retry(struct kcm_sock *kcm)
561{
562 struct kcm_mux *mux = kcm->mux;
563
564 spin_lock_bh(&mux->lock);
565 KCM_STATS_INCR(mux->stats.tx_retries);
566 spin_unlock_bh(&mux->lock);
567}
568
569/* Write any messages ready on the kcm socket. Called with kcm sock lock
570 * held. Return bytes actually sent or error.
571 */
572static int kcm_write_msgs(struct kcm_sock *kcm)
573{
574 struct sock *sk = &kcm->sk;
575 struct kcm_psock *psock;
576 struct sk_buff *skb, *head;
577 struct kcm_tx_msg *txm;
578 unsigned short fragidx, frag_offset;
579 unsigned int sent, total_sent = 0;
580 int ret = 0;
581
582 kcm->tx_wait_more = false;
583 psock = kcm->tx_psock;
584 if (unlikely(psock && psock->tx_stopped)) {
585 /* A reserved psock was aborted asynchronously. Unreserve
586 * it and we'll retry the message.
587 */
588 unreserve_psock(kcm);
589 kcm_report_tx_retry(kcm);
590 if (skb_queue_empty(&sk->sk_write_queue))
591 return 0;
592
593 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
594
595 } else if (skb_queue_empty(&sk->sk_write_queue)) {
596 return 0;
597 }
598
599 head = skb_peek(&sk->sk_write_queue);
600 txm = kcm_tx_msg(head);
601
602 if (txm->sent) {
603 /* Send of first skbuff in queue already in progress */
604 if (WARN_ON(!psock)) {
605 ret = -EINVAL;
606 goto out;
607 }
608 sent = txm->sent;
609 frag_offset = txm->frag_offset;
610 fragidx = txm->fragidx;
611 skb = txm->frag_skb;
612
613 goto do_frag;
614 }
615
616try_again:
617 psock = reserve_psock(kcm);
618 if (!psock)
619 goto out;
620
621 do {
622 skb = head;
623 txm = kcm_tx_msg(head);
624 sent = 0;
625
626do_frag_list:
627 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
628 ret = -EINVAL;
629 goto out;
630 }
631
632 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
633 fragidx++) {
634 skb_frag_t *frag;
635
636 frag_offset = 0;
637do_frag:
638 frag = &skb_shinfo(skb)->frags[fragidx];
639 if (WARN_ON(!frag->size)) {
640 ret = -EINVAL;
641 goto out;
642 }
643
644 ret = kernel_sendpage(psock->sk->sk_socket,
645 frag->page.p,
646 frag->page_offset + frag_offset,
647 frag->size - frag_offset,
648 MSG_DONTWAIT);
649 if (ret <= 0) {
650 if (ret == -EAGAIN) {
651 /* Save state to try again when there's
652 * write space on the socket
653 */
654 txm->sent = sent;
655 txm->frag_offset = frag_offset;
656 txm->fragidx = fragidx;
657 txm->frag_skb = skb;
658
659 ret = 0;
660 goto out;
661 }
662
663 /* Hard failure in sending message, abort this
664 * psock since it has lost framing
665 * synchonization and retry sending the
666 * message from the beginning.
667 */
668 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
669 true);
670 unreserve_psock(kcm);
671
672 txm->sent = 0;
673 kcm_report_tx_retry(kcm);
674 ret = 0;
675
676 goto try_again;
677 }
678
679 sent += ret;
680 frag_offset += ret;
681 KCM_STATS_ADD(psock->stats.tx_bytes, ret);
682 if (frag_offset < frag->size) {
683 /* Not finished with this frag */
684 goto do_frag;
685 }
686 }
687
688 if (skb == head) {
689 if (skb_has_frag_list(skb)) {
690 skb = skb_shinfo(skb)->frag_list;
691 goto do_frag_list;
692 }
693 } else if (skb->next) {
694 skb = skb->next;
695 goto do_frag_list;
696 }
697
698 /* Successfully sent the whole packet, account for it. */
699 skb_dequeue(&sk->sk_write_queue);
700 kfree_skb(head);
701 sk->sk_wmem_queued -= sent;
702 total_sent += sent;
703 KCM_STATS_INCR(psock->stats.tx_msgs);
704 } while ((head = skb_peek(&sk->sk_write_queue)));
705out:
706 if (!head) {
707 /* Done with all queued messages. */
708 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
709 unreserve_psock(kcm);
710 }
711
712 /* Check if write space is available */
713 sk->sk_write_space(sk);
714
715 return total_sent ? : ret;
716}
717
718static void kcm_tx_work(struct work_struct *w)
719{
720 struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
721 struct sock *sk = &kcm->sk;
722 int err;
723
724 lock_sock(sk);
725
726 /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
727 * aborts
728 */
729 err = kcm_write_msgs(kcm);
730 if (err < 0) {
731 /* Hard failure in write, report error on KCM socket */
732 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
733 report_csk_error(&kcm->sk, -err);
734 goto out;
735 }
736
737 /* Primarily for SOCK_SEQPACKET sockets */
738 if (likely(sk->sk_socket) &&
739 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
740 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
741 sk->sk_write_space(sk);
742 }
743
744out:
745 release_sock(sk);
746}
747
748static void kcm_push(struct kcm_sock *kcm)
749{
750 if (kcm->tx_wait_more)
751 kcm_write_msgs(kcm);
752}
753
754static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
755 int offset, size_t size, int flags)
756
757{
758 struct sock *sk = sock->sk;
759 struct kcm_sock *kcm = kcm_sk(sk);
760 struct sk_buff *skb = NULL, *head = NULL;
761 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
762 bool eor;
763 int err = 0;
764 int i;
765
766 if (flags & MSG_SENDPAGE_NOTLAST)
767 flags |= MSG_MORE;
768
769 /* No MSG_EOR from splice, only look at MSG_MORE */
770 eor = !(flags & MSG_MORE);
771
772 lock_sock(sk);
773
774 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
775
776 err = -EPIPE;
777 if (sk->sk_err)
778 goto out_error;
779
780 if (kcm->seq_skb) {
781 /* Previously opened message */
782 head = kcm->seq_skb;
783 skb = kcm_tx_msg(head)->last_skb;
784 i = skb_shinfo(skb)->nr_frags;
785
786 if (skb_can_coalesce(skb, i, page, offset)) {
787 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
788 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
789 goto coalesced;
790 }
791
792 if (i >= MAX_SKB_FRAGS) {
793 struct sk_buff *tskb;
794
795 tskb = alloc_skb(0, sk->sk_allocation);
796 while (!tskb) {
797 kcm_push(kcm);
798 err = sk_stream_wait_memory(sk, &timeo);
799 if (err)
800 goto out_error;
801 }
802
803 if (head == skb)
804 skb_shinfo(head)->frag_list = tskb;
805 else
806 skb->next = tskb;
807
808 skb = tskb;
809 skb->ip_summed = CHECKSUM_UNNECESSARY;
810 i = 0;
811 }
812 } else {
813 /* Call the sk_stream functions to manage the sndbuf mem. */
814 if (!sk_stream_memory_free(sk)) {
815 kcm_push(kcm);
816 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
817 err = sk_stream_wait_memory(sk, &timeo);
818 if (err)
819 goto out_error;
820 }
821
822 head = alloc_skb(0, sk->sk_allocation);
823 while (!head) {
824 kcm_push(kcm);
825 err = sk_stream_wait_memory(sk, &timeo);
826 if (err)
827 goto out_error;
828 }
829
830 skb = head;
831 i = 0;
832 }
833
834 get_page(page);
835 skb_fill_page_desc(skb, i, page, offset, size);
836 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
837
838coalesced:
839 skb->len += size;
840 skb->data_len += size;
841 skb->truesize += size;
842 sk->sk_wmem_queued += size;
843 sk_mem_charge(sk, size);
844
845 if (head != skb) {
846 head->len += size;
847 head->data_len += size;
848 head->truesize += size;
849 }
850
851 if (eor) {
852 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
853
854 /* Message complete, queue it on send buffer */
855 __skb_queue_tail(&sk->sk_write_queue, head);
856 kcm->seq_skb = NULL;
857 KCM_STATS_INCR(kcm->stats.tx_msgs);
858
859 if (flags & MSG_BATCH) {
860 kcm->tx_wait_more = true;
861 } else if (kcm->tx_wait_more || not_busy) {
862 err = kcm_write_msgs(kcm);
863 if (err < 0) {
864 /* We got a hard error in write_msgs but have
865 * already queued this message. Report an error
866 * in the socket, but don't affect return value
867 * from sendmsg
868 */
869 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
870 report_csk_error(&kcm->sk, -err);
871 }
872 }
873 } else {
874 /* Message not complete, save state */
875 kcm->seq_skb = head;
876 kcm_tx_msg(head)->last_skb = skb;
877 }
878
879 KCM_STATS_ADD(kcm->stats.tx_bytes, size);
880
881 release_sock(sk);
882 return size;
883
884out_error:
885 kcm_push(kcm);
886
887 err = sk_stream_error(sk, flags, err);
888
889 /* make sure we wake any epoll edge trigger waiter */
890 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
891 sk->sk_write_space(sk);
892
893 release_sock(sk);
894 return err;
895}
896
897static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
898{
899 struct sock *sk = sock->sk;
900 struct kcm_sock *kcm = kcm_sk(sk);
901 struct sk_buff *skb = NULL, *head = NULL;
902 size_t copy, copied = 0;
903 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
904 int eor = (sock->type == SOCK_DGRAM) ?
905 !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
906 int err = -EPIPE;
907
908 lock_sock(sk);
909
910 /* Per tcp_sendmsg this should be in poll */
911 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
912
913 if (sk->sk_err)
914 goto out_error;
915
916 if (kcm->seq_skb) {
917 /* Previously opened message */
918 head = kcm->seq_skb;
919 skb = kcm_tx_msg(head)->last_skb;
920 goto start;
921 }
922
923 /* Call the sk_stream functions to manage the sndbuf mem. */
924 if (!sk_stream_memory_free(sk)) {
925 kcm_push(kcm);
926 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
927 err = sk_stream_wait_memory(sk, &timeo);
928 if (err)
929 goto out_error;
930 }
931
932 if (msg_data_left(msg)) {
933 /* New message, alloc head skb */
934 head = alloc_skb(0, sk->sk_allocation);
935 while (!head) {
936 kcm_push(kcm);
937 err = sk_stream_wait_memory(sk, &timeo);
938 if (err)
939 goto out_error;
940
941 head = alloc_skb(0, sk->sk_allocation);
942 }
943
944 skb = head;
945
946 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
947 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
948 */
949 skb->ip_summed = CHECKSUM_UNNECESSARY;
950 }
951
952start:
953 while (msg_data_left(msg)) {
954 bool merge = true;
955 int i = skb_shinfo(skb)->nr_frags;
956 struct page_frag *pfrag = sk_page_frag(sk);
957
958 if (!sk_page_frag_refill(sk, pfrag))
959 goto wait_for_memory;
960
961 if (!skb_can_coalesce(skb, i, pfrag->page,
962 pfrag->offset)) {
963 if (i == MAX_SKB_FRAGS) {
964 struct sk_buff *tskb;
965
966 tskb = alloc_skb(0, sk->sk_allocation);
967 if (!tskb)
968 goto wait_for_memory;
969
970 if (head == skb)
971 skb_shinfo(head)->frag_list = tskb;
972 else
973 skb->next = tskb;
974
975 skb = tskb;
976 skb->ip_summed = CHECKSUM_UNNECESSARY;
977 continue;
978 }
979 merge = false;
980 }
981
982 copy = min_t(int, msg_data_left(msg),
983 pfrag->size - pfrag->offset);
984
985 if (!sk_wmem_schedule(sk, copy))
986 goto wait_for_memory;
987
988 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
989 pfrag->page,
990 pfrag->offset,
991 copy);
992 if (err)
993 goto out_error;
994
995 /* Update the skb. */
996 if (merge) {
997 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
998 } else {
999 skb_fill_page_desc(skb, i, pfrag->page,
1000 pfrag->offset, copy);
1001 get_page(pfrag->page);
1002 }
1003
1004 pfrag->offset += copy;
1005 copied += copy;
1006 if (head != skb) {
1007 head->len += copy;
1008 head->data_len += copy;
1009 }
1010
1011 continue;
1012
1013wait_for_memory:
1014 kcm_push(kcm);
1015 err = sk_stream_wait_memory(sk, &timeo);
1016 if (err)
1017 goto out_error;
1018 }
1019
1020 if (eor) {
1021 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1022
1023 if (head) {
1024 /* Message complete, queue it on send buffer */
1025 __skb_queue_tail(&sk->sk_write_queue, head);
1026 kcm->seq_skb = NULL;
1027 KCM_STATS_INCR(kcm->stats.tx_msgs);
1028 }
1029
1030 if (msg->msg_flags & MSG_BATCH) {
1031 kcm->tx_wait_more = true;
1032 } else if (kcm->tx_wait_more || not_busy) {
1033 err = kcm_write_msgs(kcm);
1034 if (err < 0) {
1035 /* We got a hard error in write_msgs but have
1036 * already queued this message. Report an error
1037 * in the socket, but don't affect return value
1038 * from sendmsg
1039 */
1040 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1041 report_csk_error(&kcm->sk, -err);
1042 }
1043 }
1044 } else {
1045 /* Message not complete, save state */
1046partial_message:
1047 if (head) {
1048 kcm->seq_skb = head;
1049 kcm_tx_msg(head)->last_skb = skb;
1050 }
1051 }
1052
1053 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1054
1055 release_sock(sk);
1056 return copied;
1057
1058out_error:
1059 kcm_push(kcm);
1060
1061 if (copied && sock->type == SOCK_SEQPACKET) {
1062 /* Wrote some bytes before encountering an
1063 * error, return partial success.
1064 */
1065 goto partial_message;
1066 }
1067
1068 if (head != kcm->seq_skb)
1069 kfree_skb(head);
1070
1071 err = sk_stream_error(sk, msg->msg_flags, err);
1072
1073 /* make sure we wake any epoll edge trigger waiter */
1074 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1075 sk->sk_write_space(sk);
1076
1077 release_sock(sk);
1078 return err;
1079}
1080
1081static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1082 long timeo, int *err)
1083{
1084 struct sk_buff *skb;
1085
1086 while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1087 if (sk->sk_err) {
1088 *err = sock_error(sk);
1089 return NULL;
1090 }
1091
1092 if (sock_flag(sk, SOCK_DONE))
1093 return NULL;
1094
1095 if ((flags & MSG_DONTWAIT) || !timeo) {
1096 *err = -EAGAIN;
1097 return NULL;
1098 }
1099
1100 sk_wait_data(sk, &timeo, NULL);
1101
1102 /* Handle signals */
1103 if (signal_pending(current)) {
1104 *err = sock_intr_errno(timeo);
1105 return NULL;
1106 }
1107 }
1108
1109 return skb;
1110}
1111
1112static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1113 size_t len, int flags)
1114{
1115 struct sock *sk = sock->sk;
1116 struct kcm_sock *kcm = kcm_sk(sk);
1117 int err = 0;
1118 long timeo;
1119 struct strp_rx_msg *rxm;
1120 int copied = 0;
1121 struct sk_buff *skb;
1122
1123 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1124
1125 lock_sock(sk);
1126
1127 skb = kcm_wait_data(sk, flags, timeo, &err);
1128 if (!skb)
1129 goto out;
1130
1131 /* Okay, have a message on the receive queue */
1132
1133 rxm = strp_rx_msg(skb);
1134
1135 if (len > rxm->full_len)
1136 len = rxm->full_len;
1137
1138 err = skb_copy_datagram_msg(skb, rxm->offset, msg, len);
1139 if (err < 0)
1140 goto out;
1141
1142 copied = len;
1143 if (likely(!(flags & MSG_PEEK))) {
1144 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1145 if (copied < rxm->full_len) {
1146 if (sock->type == SOCK_DGRAM) {
1147 /* Truncated message */
1148 msg->msg_flags |= MSG_TRUNC;
1149 goto msg_finished;
1150 }
1151 rxm->offset += copied;
1152 rxm->full_len -= copied;
1153 } else {
1154msg_finished:
1155 /* Finished with message */
1156 msg->msg_flags |= MSG_EOR;
1157 KCM_STATS_INCR(kcm->stats.rx_msgs);
1158 skb_unlink(skb, &sk->sk_receive_queue);
1159 kfree_skb(skb);
1160 }
1161 }
1162
1163out:
1164 release_sock(sk);
1165
1166 return copied ? : err;
1167}
1168
1169static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1170 struct pipe_inode_info *pipe, size_t len,
1171 unsigned int flags)
1172{
1173 struct sock *sk = sock->sk;
1174 struct kcm_sock *kcm = kcm_sk(sk);
1175 long timeo;
1176 struct strp_rx_msg *rxm;
1177 int err = 0;
1178 ssize_t copied;
1179 struct sk_buff *skb;
1180
1181 /* Only support splice for SOCKSEQPACKET */
1182
1183 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1184
1185 lock_sock(sk);
1186
1187 skb = kcm_wait_data(sk, flags, timeo, &err);
1188 if (!skb)
1189 goto err_out;
1190
1191 /* Okay, have a message on the receive queue */
1192
1193 rxm = strp_rx_msg(skb);
1194
1195 if (len > rxm->full_len)
1196 len = rxm->full_len;
1197
1198 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, len, flags);
1199 if (copied < 0) {
1200 err = copied;
1201 goto err_out;
1202 }
1203
1204 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1205
1206 rxm->offset += copied;
1207 rxm->full_len -= copied;
1208
1209 /* We have no way to return MSG_EOR. If all the bytes have been
1210 * read we still leave the message in the receive socket buffer.
1211 * A subsequent recvmsg needs to be done to return MSG_EOR and
1212 * finish reading the message.
1213 */
1214
1215 release_sock(sk);
1216
1217 return copied;
1218
1219err_out:
1220 release_sock(sk);
1221
1222 return err;
1223}
1224
1225/* kcm sock lock held */
1226static void kcm_recv_disable(struct kcm_sock *kcm)
1227{
1228 struct kcm_mux *mux = kcm->mux;
1229
1230 if (kcm->rx_disabled)
1231 return;
1232
1233 spin_lock_bh(&mux->rx_lock);
1234
1235 kcm->rx_disabled = 1;
1236
1237 /* If a psock is reserved we'll do cleanup in unreserve */
1238 if (!kcm->rx_psock) {
1239 if (kcm->rx_wait) {
1240 list_del(&kcm->wait_rx_list);
1241 kcm->rx_wait = false;
1242 }
1243
1244 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1245 }
1246
1247 spin_unlock_bh(&mux->rx_lock);
1248}
1249
1250/* kcm sock lock held */
1251static void kcm_recv_enable(struct kcm_sock *kcm)
1252{
1253 struct kcm_mux *mux = kcm->mux;
1254
1255 if (!kcm->rx_disabled)
1256 return;
1257
1258 spin_lock_bh(&mux->rx_lock);
1259
1260 kcm->rx_disabled = 0;
1261 kcm_rcv_ready(kcm);
1262
1263 spin_unlock_bh(&mux->rx_lock);
1264}
1265
1266static int kcm_setsockopt(struct socket *sock, int level, int optname,
1267 char __user *optval, unsigned int optlen)
1268{
1269 struct kcm_sock *kcm = kcm_sk(sock->sk);
1270 int val, valbool;
1271 int err = 0;
1272
1273 if (level != SOL_KCM)
1274 return -ENOPROTOOPT;
1275
1276 if (optlen < sizeof(int))
1277 return -EINVAL;
1278
1279 if (get_user(val, (int __user *)optval))
1280 return -EINVAL;
1281
1282 valbool = val ? 1 : 0;
1283
1284 switch (optname) {
1285 case KCM_RECV_DISABLE:
1286 lock_sock(&kcm->sk);
1287 if (valbool)
1288 kcm_recv_disable(kcm);
1289 else
1290 kcm_recv_enable(kcm);
1291 release_sock(&kcm->sk);
1292 break;
1293 default:
1294 err = -ENOPROTOOPT;
1295 }
1296
1297 return err;
1298}
1299
1300static int kcm_getsockopt(struct socket *sock, int level, int optname,
1301 char __user *optval, int __user *optlen)
1302{
1303 struct kcm_sock *kcm = kcm_sk(sock->sk);
1304 int val, len;
1305
1306 if (level != SOL_KCM)
1307 return -ENOPROTOOPT;
1308
1309 if (get_user(len, optlen))
1310 return -EFAULT;
1311
1312 len = min_t(unsigned int, len, sizeof(int));
1313 if (len < 0)
1314 return -EINVAL;
1315
1316 switch (optname) {
1317 case KCM_RECV_DISABLE:
1318 val = kcm->rx_disabled;
1319 break;
1320 default:
1321 return -ENOPROTOOPT;
1322 }
1323
1324 if (put_user(len, optlen))
1325 return -EFAULT;
1326 if (copy_to_user(optval, &val, len))
1327 return -EFAULT;
1328 return 0;
1329}
1330
1331static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1332{
1333 struct kcm_sock *tkcm;
1334 struct list_head *head;
1335 int index = 0;
1336
1337 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1338 * we set sk_state, otherwise epoll_wait always returns right away with
1339 * POLLHUP
1340 */
1341 kcm->sk.sk_state = TCP_ESTABLISHED;
1342
1343 /* Add to mux's kcm sockets list */
1344 kcm->mux = mux;
1345 spin_lock_bh(&mux->lock);
1346
1347 head = &mux->kcm_socks;
1348 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1349 if (tkcm->index != index)
1350 break;
1351 head = &tkcm->kcm_sock_list;
1352 index++;
1353 }
1354
1355 list_add(&kcm->kcm_sock_list, head);
1356 kcm->index = index;
1357
1358 mux->kcm_socks_cnt++;
1359 spin_unlock_bh(&mux->lock);
1360
1361 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1362
1363 spin_lock_bh(&mux->rx_lock);
1364 kcm_rcv_ready(kcm);
1365 spin_unlock_bh(&mux->rx_lock);
1366}
1367
1368static int kcm_attach(struct socket *sock, struct socket *csock,
1369 struct bpf_prog *prog)
1370{
1371 struct kcm_sock *kcm = kcm_sk(sock->sk);
1372 struct kcm_mux *mux = kcm->mux;
1373 struct sock *csk;
1374 struct kcm_psock *psock = NULL, *tpsock;
1375 struct list_head *head;
1376 int index = 0;
1377 struct strp_callbacks cb;
1378 int err;
1379
1380 csk = csock->sk;
1381 if (!csk)
1382 return -EINVAL;
1383
1384 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1385 if (!psock)
1386 return -ENOMEM;
1387
1388 psock->mux = mux;
1389 psock->sk = csk;
1390 psock->bpf_prog = prog;
1391
1392 cb.rcv_msg = kcm_rcv_strparser;
1393 cb.abort_parser = NULL;
1394 cb.parse_msg = kcm_parse_func_strparser;
1395 cb.read_sock_done = kcm_read_sock_done;
1396
1397 err = strp_init(&psock->strp, csk, &cb);
1398 if (err) {
1399 kmem_cache_free(kcm_psockp, psock);
1400 return err;
1401 }
1402
1403 sock_hold(csk);
1404
1405 write_lock_bh(&csk->sk_callback_lock);
1406 psock->save_data_ready = csk->sk_data_ready;
1407 psock->save_write_space = csk->sk_write_space;
1408 psock->save_state_change = csk->sk_state_change;
1409 csk->sk_user_data = psock;
1410 csk->sk_data_ready = psock_data_ready;
1411 csk->sk_write_space = psock_write_space;
1412 csk->sk_state_change = psock_state_change;
1413 write_unlock_bh(&csk->sk_callback_lock);
1414
1415 /* Finished initialization, now add the psock to the MUX. */
1416 spin_lock_bh(&mux->lock);
1417 head = &mux->psocks;
1418 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1419 if (tpsock->index != index)
1420 break;
1421 head = &tpsock->psock_list;
1422 index++;
1423 }
1424
1425 list_add(&psock->psock_list, head);
1426 psock->index = index;
1427
1428 KCM_STATS_INCR(mux->stats.psock_attach);
1429 mux->psocks_cnt++;
1430 psock_now_avail(psock);
1431 spin_unlock_bh(&mux->lock);
1432
1433 /* Schedule RX work in case there are already bytes queued */
1434 strp_check_rcv(&psock->strp);
1435
1436 return 0;
1437}
1438
1439static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1440{
1441 struct socket *csock;
1442 struct bpf_prog *prog;
1443 int err;
1444
1445 csock = sockfd_lookup(info->fd, &err);
1446 if (!csock)
1447 return -ENOENT;
1448
1449 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1450 if (IS_ERR(prog)) {
1451 err = PTR_ERR(prog);
1452 goto out;
1453 }
1454
1455 err = kcm_attach(sock, csock, prog);
1456 if (err) {
1457 bpf_prog_put(prog);
1458 goto out;
1459 }
1460
1461 /* Keep reference on file also */
1462
1463 return 0;
1464out:
1465 fput(csock->file);
1466 return err;
1467}
1468
1469static void kcm_unattach(struct kcm_psock *psock)
1470{
1471 struct sock *csk = psock->sk;
1472 struct kcm_mux *mux = psock->mux;
1473
1474 lock_sock(csk);
1475
1476 /* Stop getting callbacks from TCP socket. After this there should
1477 * be no way to reserve a kcm for this psock.
1478 */
1479 write_lock_bh(&csk->sk_callback_lock);
1480 csk->sk_user_data = NULL;
1481 csk->sk_data_ready = psock->save_data_ready;
1482 csk->sk_write_space = psock->save_write_space;
1483 csk->sk_state_change = psock->save_state_change;
1484 strp_stop(&psock->strp);
1485
1486 if (WARN_ON(psock->rx_kcm)) {
1487 write_unlock_bh(&csk->sk_callback_lock);
1488 return;
1489 }
1490
1491 spin_lock_bh(&mux->rx_lock);
1492
1493 /* Stop receiver activities. After this point psock should not be
1494 * able to get onto ready list either through callbacks or work.
1495 */
1496 if (psock->ready_rx_msg) {
1497 list_del(&psock->psock_ready_list);
1498 kfree_skb(psock->ready_rx_msg);
1499 psock->ready_rx_msg = NULL;
1500 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1501 }
1502
1503 spin_unlock_bh(&mux->rx_lock);
1504
1505 write_unlock_bh(&csk->sk_callback_lock);
1506
1507 /* Call strp_done without sock lock */
1508 release_sock(csk);
1509 strp_done(&psock->strp);
1510 lock_sock(csk);
1511
1512 bpf_prog_put(psock->bpf_prog);
1513
1514 spin_lock_bh(&mux->lock);
1515
1516 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1517 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1518
1519 KCM_STATS_INCR(mux->stats.psock_unattach);
1520
1521 if (psock->tx_kcm) {
1522 /* psock was reserved. Just mark it finished and we will clean
1523 * up in the kcm paths, we need kcm lock which can not be
1524 * acquired here.
1525 */
1526 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1527 spin_unlock_bh(&mux->lock);
1528
1529 /* We are unattaching a socket that is reserved. Abort the
1530 * socket since we may be out of sync in sending on it. We need
1531 * to do this without the mux lock.
1532 */
1533 kcm_abort_tx_psock(psock, EPIPE, false);
1534
1535 spin_lock_bh(&mux->lock);
1536 if (!psock->tx_kcm) {
1537 /* psock now unreserved in window mux was unlocked */
1538 goto no_reserved;
1539 }
1540 psock->done = 1;
1541
1542 /* Commit done before queuing work to process it */
1543 smp_mb();
1544
1545 /* Queue tx work to make sure psock->done is handled */
1546 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1547 spin_unlock_bh(&mux->lock);
1548 } else {
1549no_reserved:
1550 if (!psock->tx_stopped)
1551 list_del(&psock->psock_avail_list);
1552 list_del(&psock->psock_list);
1553 mux->psocks_cnt--;
1554 spin_unlock_bh(&mux->lock);
1555
1556 sock_put(csk);
1557 fput(csk->sk_socket->file);
1558 kmem_cache_free(kcm_psockp, psock);
1559 }
1560
1561 release_sock(csk);
1562}
1563
1564static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1565{
1566 struct kcm_sock *kcm = kcm_sk(sock->sk);
1567 struct kcm_mux *mux = kcm->mux;
1568 struct kcm_psock *psock;
1569 struct socket *csock;
1570 struct sock *csk;
1571 int err;
1572
1573 csock = sockfd_lookup(info->fd, &err);
1574 if (!csock)
1575 return -ENOENT;
1576
1577 csk = csock->sk;
1578 if (!csk) {
1579 err = -EINVAL;
1580 goto out;
1581 }
1582
1583 err = -ENOENT;
1584
1585 spin_lock_bh(&mux->lock);
1586
1587 list_for_each_entry(psock, &mux->psocks, psock_list) {
1588 if (psock->sk != csk)
1589 continue;
1590
1591 /* Found the matching psock */
1592
1593 if (psock->unattaching || WARN_ON(psock->done)) {
1594 err = -EALREADY;
1595 break;
1596 }
1597
1598 psock->unattaching = 1;
1599
1600 spin_unlock_bh(&mux->lock);
1601
1602 /* Lower socket lock should already be held */
1603 kcm_unattach(psock);
1604
1605 err = 0;
1606 goto out;
1607 }
1608
1609 spin_unlock_bh(&mux->lock);
1610
1611out:
1612 fput(csock->file);
1613 return err;
1614}
1615
1616static struct proto kcm_proto = {
1617 .name = "KCM",
1618 .owner = THIS_MODULE,
1619 .obj_size = sizeof(struct kcm_sock),
1620};
1621
1622/* Clone a kcm socket. */
1623static int kcm_clone(struct socket *osock, struct kcm_clone *info,
1624 struct socket **newsockp)
1625{
1626 struct socket *newsock;
1627 struct sock *newsk;
1628 struct file *newfile;
1629 int err, newfd;
1630
1631 err = -ENFILE;
1632 newsock = sock_alloc();
1633 if (!newsock)
1634 goto out;
1635
1636 newsock->type = osock->type;
1637 newsock->ops = osock->ops;
1638
1639 __module_get(newsock->ops->owner);
1640
1641 newfd = get_unused_fd_flags(0);
1642 if (unlikely(newfd < 0)) {
1643 err = newfd;
1644 goto out_fd_fail;
1645 }
1646
1647 newfile = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1648 if (unlikely(IS_ERR(newfile))) {
1649 err = PTR_ERR(newfile);
1650 goto out_sock_alloc_fail;
1651 }
1652
1653 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1654 &kcm_proto, true);
1655 if (!newsk) {
1656 err = -ENOMEM;
1657 goto out_sk_alloc_fail;
1658 }
1659
1660 sock_init_data(newsock, newsk);
1661 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1662
1663 fd_install(newfd, newfile);
1664 *newsockp = newsock;
1665 info->fd = newfd;
1666
1667 return 0;
1668
1669out_sk_alloc_fail:
1670 fput(newfile);
1671out_sock_alloc_fail:
1672 put_unused_fd(newfd);
1673out_fd_fail:
1674 sock_release(newsock);
1675out:
1676 return err;
1677}
1678
1679static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1680{
1681 int err;
1682
1683 switch (cmd) {
1684 case SIOCKCMATTACH: {
1685 struct kcm_attach info;
1686
1687 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1688 err = -EFAULT;
1689
1690 err = kcm_attach_ioctl(sock, &info);
1691
1692 break;
1693 }
1694 case SIOCKCMUNATTACH: {
1695 struct kcm_unattach info;
1696
1697 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1698 err = -EFAULT;
1699
1700 err = kcm_unattach_ioctl(sock, &info);
1701
1702 break;
1703 }
1704 case SIOCKCMCLONE: {
1705 struct kcm_clone info;
1706 struct socket *newsock = NULL;
1707
1708 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1709 err = -EFAULT;
1710
1711 err = kcm_clone(sock, &info, &newsock);
1712
1713 if (!err) {
1714 if (copy_to_user((void __user *)arg, &info,
1715 sizeof(info))) {
1716 err = -EFAULT;
1717 sys_close(info.fd);
1718 }
1719 }
1720
1721 break;
1722 }
1723 default:
1724 err = -ENOIOCTLCMD;
1725 break;
1726 }
1727
1728 return err;
1729}
1730
1731static void free_mux(struct rcu_head *rcu)
1732{
1733 struct kcm_mux *mux = container_of(rcu,
1734 struct kcm_mux, rcu);
1735
1736 kmem_cache_free(kcm_muxp, mux);
1737}
1738
1739static void release_mux(struct kcm_mux *mux)
1740{
1741 struct kcm_net *knet = mux->knet;
1742 struct kcm_psock *psock, *tmp_psock;
1743
1744 /* Release psocks */
1745 list_for_each_entry_safe(psock, tmp_psock,
1746 &mux->psocks, psock_list) {
1747 if (!WARN_ON(psock->unattaching))
1748 kcm_unattach(psock);
1749 }
1750
1751 if (WARN_ON(mux->psocks_cnt))
1752 return;
1753
1754 __skb_queue_purge(&mux->rx_hold_queue);
1755
1756 mutex_lock(&knet->mutex);
1757 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1758 aggregate_psock_stats(&mux->aggregate_psock_stats,
1759 &knet->aggregate_psock_stats);
1760 aggregate_strp_stats(&mux->aggregate_strp_stats,
1761 &knet->aggregate_strp_stats);
1762 list_del_rcu(&mux->kcm_mux_list);
1763 knet->count--;
1764 mutex_unlock(&knet->mutex);
1765
1766 call_rcu(&mux->rcu, free_mux);
1767}
1768
1769static void kcm_done(struct kcm_sock *kcm)
1770{
1771 struct kcm_mux *mux = kcm->mux;
1772 struct sock *sk = &kcm->sk;
1773 int socks_cnt;
1774
1775 spin_lock_bh(&mux->rx_lock);
1776 if (kcm->rx_psock) {
1777 /* Cleanup in unreserve_rx_kcm */
1778 WARN_ON(kcm->done);
1779 kcm->rx_disabled = 1;
1780 kcm->done = 1;
1781 spin_unlock_bh(&mux->rx_lock);
1782 return;
1783 }
1784
1785 if (kcm->rx_wait) {
1786 list_del(&kcm->wait_rx_list);
1787 kcm->rx_wait = false;
1788 }
1789 /* Move any pending receive messages to other kcm sockets */
1790 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1791
1792 spin_unlock_bh(&mux->rx_lock);
1793
1794 if (WARN_ON(sk_rmem_alloc_get(sk)))
1795 return;
1796
1797 /* Detach from MUX */
1798 spin_lock_bh(&mux->lock);
1799
1800 list_del(&kcm->kcm_sock_list);
1801 mux->kcm_socks_cnt--;
1802 socks_cnt = mux->kcm_socks_cnt;
1803
1804 spin_unlock_bh(&mux->lock);
1805
1806 if (!socks_cnt) {
1807 /* We are done with the mux now. */
1808 release_mux(mux);
1809 }
1810
1811 WARN_ON(kcm->rx_wait);
1812
1813 sock_put(&kcm->sk);
1814}
1815
1816/* Called by kcm_release to close a KCM socket.
1817 * If this is the last KCM socket on the MUX, destroy the MUX.
1818 */
1819static int kcm_release(struct socket *sock)
1820{
1821 struct sock *sk = sock->sk;
1822 struct kcm_sock *kcm;
1823 struct kcm_mux *mux;
1824 struct kcm_psock *psock;
1825
1826 if (!sk)
1827 return 0;
1828
1829 kcm = kcm_sk(sk);
1830 mux = kcm->mux;
1831
1832 sock_orphan(sk);
1833 kfree_skb(kcm->seq_skb);
1834
1835 lock_sock(sk);
1836 /* Purge queue under lock to avoid race condition with tx_work trying
1837 * to act when queue is nonempty. If tx_work runs after this point
1838 * it will just return.
1839 */
1840 __skb_queue_purge(&sk->sk_write_queue);
1841
1842 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1843 * get a writespace callback. This prevents further work being queued
1844 * from the callback (unbinding the psock occurs after canceling work.
1845 */
1846 kcm->tx_stopped = 1;
1847
1848 release_sock(sk);
1849
1850 spin_lock_bh(&mux->lock);
1851 if (kcm->tx_wait) {
1852 /* Take of tx_wait list, after this point there should be no way
1853 * that a psock will be assigned to this kcm.
1854 */
1855 list_del(&kcm->wait_psock_list);
1856 kcm->tx_wait = false;
1857 }
1858 spin_unlock_bh(&mux->lock);
1859
1860 /* Cancel work. After this point there should be no outside references
1861 * to the kcm socket.
1862 */
1863 cancel_work_sync(&kcm->tx_work);
1864
1865 lock_sock(sk);
1866 psock = kcm->tx_psock;
1867 if (psock) {
1868 /* A psock was reserved, so we need to kill it since it
1869 * may already have some bytes queued from a message. We
1870 * need to do this after removing kcm from tx_wait list.
1871 */
1872 kcm_abort_tx_psock(psock, EPIPE, false);
1873 unreserve_psock(kcm);
1874 }
1875 release_sock(sk);
1876
1877 WARN_ON(kcm->tx_wait);
1878 WARN_ON(kcm->tx_psock);
1879
1880 sock->sk = NULL;
1881
1882 kcm_done(kcm);
1883
1884 return 0;
1885}
1886
1887static const struct proto_ops kcm_dgram_ops = {
1888 .family = PF_KCM,
1889 .owner = THIS_MODULE,
1890 .release = kcm_release,
1891 .bind = sock_no_bind,
1892 .connect = sock_no_connect,
1893 .socketpair = sock_no_socketpair,
1894 .accept = sock_no_accept,
1895 .getname = sock_no_getname,
1896 .poll = datagram_poll,
1897 .ioctl = kcm_ioctl,
1898 .listen = sock_no_listen,
1899 .shutdown = sock_no_shutdown,
1900 .setsockopt = kcm_setsockopt,
1901 .getsockopt = kcm_getsockopt,
1902 .sendmsg = kcm_sendmsg,
1903 .recvmsg = kcm_recvmsg,
1904 .mmap = sock_no_mmap,
1905 .sendpage = kcm_sendpage,
1906};
1907
1908static const struct proto_ops kcm_seqpacket_ops = {
1909 .family = PF_KCM,
1910 .owner = THIS_MODULE,
1911 .release = kcm_release,
1912 .bind = sock_no_bind,
1913 .connect = sock_no_connect,
1914 .socketpair = sock_no_socketpair,
1915 .accept = sock_no_accept,
1916 .getname = sock_no_getname,
1917 .poll = datagram_poll,
1918 .ioctl = kcm_ioctl,
1919 .listen = sock_no_listen,
1920 .shutdown = sock_no_shutdown,
1921 .setsockopt = kcm_setsockopt,
1922 .getsockopt = kcm_getsockopt,
1923 .sendmsg = kcm_sendmsg,
1924 .recvmsg = kcm_recvmsg,
1925 .mmap = sock_no_mmap,
1926 .sendpage = kcm_sendpage,
1927 .splice_read = kcm_splice_read,
1928};
1929
1930/* Create proto operation for kcm sockets */
1931static int kcm_create(struct net *net, struct socket *sock,
1932 int protocol, int kern)
1933{
1934 struct kcm_net *knet = net_generic(net, kcm_net_id);
1935 struct sock *sk;
1936 struct kcm_mux *mux;
1937
1938 switch (sock->type) {
1939 case SOCK_DGRAM:
1940 sock->ops = &kcm_dgram_ops;
1941 break;
1942 case SOCK_SEQPACKET:
1943 sock->ops = &kcm_seqpacket_ops;
1944 break;
1945 default:
1946 return -ESOCKTNOSUPPORT;
1947 }
1948
1949 if (protocol != KCMPROTO_CONNECTED)
1950 return -EPROTONOSUPPORT;
1951
1952 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1953 if (!sk)
1954 return -ENOMEM;
1955
1956 /* Allocate a kcm mux, shared between KCM sockets */
1957 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1958 if (!mux) {
1959 sk_free(sk);
1960 return -ENOMEM;
1961 }
1962
1963 spin_lock_init(&mux->lock);
1964 spin_lock_init(&mux->rx_lock);
1965 INIT_LIST_HEAD(&mux->kcm_socks);
1966 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1967 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1968
1969 INIT_LIST_HEAD(&mux->psocks);
1970 INIT_LIST_HEAD(&mux->psocks_ready);
1971 INIT_LIST_HEAD(&mux->psocks_avail);
1972
1973 mux->knet = knet;
1974
1975 /* Add new MUX to list */
1976 mutex_lock(&knet->mutex);
1977 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1978 knet->count++;
1979 mutex_unlock(&knet->mutex);
1980
1981 skb_queue_head_init(&mux->rx_hold_queue);
1982
1983 /* Init KCM socket */
1984 sock_init_data(sock, sk);
1985 init_kcm_sock(kcm_sk(sk), mux);
1986
1987 return 0;
1988}
1989
1990static struct net_proto_family kcm_family_ops = {
1991 .family = PF_KCM,
1992 .create = kcm_create,
1993 .owner = THIS_MODULE,
1994};
1995
1996static __net_init int kcm_init_net(struct net *net)
1997{
1998 struct kcm_net *knet = net_generic(net, kcm_net_id);
1999
2000 INIT_LIST_HEAD_RCU(&knet->mux_list);
2001 mutex_init(&knet->mutex);
2002
2003 return 0;
2004}
2005
2006static __net_exit void kcm_exit_net(struct net *net)
2007{
2008 struct kcm_net *knet = net_generic(net, kcm_net_id);
2009
2010 /* All KCM sockets should be closed at this point, which should mean
2011 * that all multiplexors and psocks have been destroyed.
2012 */
2013 WARN_ON(!list_empty(&knet->mux_list));
2014}
2015
2016static struct pernet_operations kcm_net_ops = {
2017 .init = kcm_init_net,
2018 .exit = kcm_exit_net,
2019 .id = &kcm_net_id,
2020 .size = sizeof(struct kcm_net),
2021};
2022
2023static int __init kcm_init(void)
2024{
2025 int err = -ENOMEM;
2026
2027 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2028 sizeof(struct kcm_mux), 0,
2029 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2030 if (!kcm_muxp)
2031 goto fail;
2032
2033 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2034 sizeof(struct kcm_psock), 0,
2035 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2036 if (!kcm_psockp)
2037 goto fail;
2038
2039 kcm_wq = create_singlethread_workqueue("kkcmd");
2040 if (!kcm_wq)
2041 goto fail;
2042
2043 err = proto_register(&kcm_proto, 1);
2044 if (err)
2045 goto fail;
2046
2047 err = sock_register(&kcm_family_ops);
2048 if (err)
2049 goto sock_register_fail;
2050
2051 err = register_pernet_device(&kcm_net_ops);
2052 if (err)
2053 goto net_ops_fail;
2054
2055 err = kcm_proc_init();
2056 if (err)
2057 goto proc_init_fail;
2058
2059 return 0;
2060
2061proc_init_fail:
2062 unregister_pernet_device(&kcm_net_ops);
2063
2064net_ops_fail:
2065 sock_unregister(PF_KCM);
2066
2067sock_register_fail:
2068 proto_unregister(&kcm_proto);
2069
2070fail:
2071 kmem_cache_destroy(kcm_muxp);
2072 kmem_cache_destroy(kcm_psockp);
2073
2074 if (kcm_wq)
2075 destroy_workqueue(kcm_wq);
2076
2077 return err;
2078}
2079
2080static void __exit kcm_exit(void)
2081{
2082 kcm_proc_exit();
2083 unregister_pernet_device(&kcm_net_ops);
2084 sock_unregister(PF_KCM);
2085 proto_unregister(&kcm_proto);
2086 destroy_workqueue(kcm_wq);
2087
2088 kmem_cache_destroy(kcm_muxp);
2089 kmem_cache_destroy(kcm_psockp);
2090}
2091
2092module_init(kcm_init);
2093module_exit(kcm_exit);
2094
2095MODULE_LICENSE("GPL");
2096MODULE_ALIAS_NETPROTO(PF_KCM);
2097