]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/kcm/kcmsock.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'sh-for-4.9' of git://git.libc.org/linux-sh
[mirror_ubuntu-artful-kernel.git] / net / kcm / kcmsock.c
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
32 unsigned int kcm_net_id;
33
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;
37
38 static inline struct kcm_sock *kcm_sk(const struct sock *sk)
39 {
40 return (struct kcm_sock *)sk;
41 }
42
43 static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
44 {
45 return (struct kcm_tx_msg *)skb->cb;
46 }
47
48 static void report_csk_error(struct sock *csk, int err)
49 {
50 csk->sk_err = EPIPE;
51 csk->sk_error_report(csk);
52 }
53
54 static 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. */
93 static 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
105 static 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
116 static 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 */
122 static 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
168 static 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
189 static 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 */
219 static 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);
228 try_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 */
251 static 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
293 static void kcm_done(struct kcm_sock *kcm);
294
295 static 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 */
301 static 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 */
343 static 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 */
357 static 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
362 try_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
378 static 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
386 static 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
395 static 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
405 static 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);
426 out:
427 read_unlock_bh(&sk->sk_callback_lock);
428 }
429
430 static void unreserve_psock(struct kcm_sock *kcm);
431
432 /* kcm sock is locked. */
433 static 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 */
486 static 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. */
514 static 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
560 static 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 */
572 static 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
616 try_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
626 do_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;
637 do_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)));
705 out:
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
718 static 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
744 out:
745 release_sock(sk);
746 }
747
748 static void kcm_push(struct kcm_sock *kcm)
749 {
750 if (kcm->tx_wait_more)
751 kcm_write_msgs(kcm);
752 }
753
754 static 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
838 coalesced:
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
884 out_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
897 static 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 /* New message, alloc head skb */
933 head = alloc_skb(0, sk->sk_allocation);
934 while (!head) {
935 kcm_push(kcm);
936 err = sk_stream_wait_memory(sk, &timeo);
937 if (err)
938 goto out_error;
939
940 head = alloc_skb(0, sk->sk_allocation);
941 }
942
943 skb = head;
944
945 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
946 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
947 */
948 skb->ip_summed = CHECKSUM_UNNECESSARY;
949
950 start:
951 while (msg_data_left(msg)) {
952 bool merge = true;
953 int i = skb_shinfo(skb)->nr_frags;
954 struct page_frag *pfrag = sk_page_frag(sk);
955
956 if (!sk_page_frag_refill(sk, pfrag))
957 goto wait_for_memory;
958
959 if (!skb_can_coalesce(skb, i, pfrag->page,
960 pfrag->offset)) {
961 if (i == MAX_SKB_FRAGS) {
962 struct sk_buff *tskb;
963
964 tskb = alloc_skb(0, sk->sk_allocation);
965 if (!tskb)
966 goto wait_for_memory;
967
968 if (head == skb)
969 skb_shinfo(head)->frag_list = tskb;
970 else
971 skb->next = tskb;
972
973 skb = tskb;
974 skb->ip_summed = CHECKSUM_UNNECESSARY;
975 continue;
976 }
977 merge = false;
978 }
979
980 copy = min_t(int, msg_data_left(msg),
981 pfrag->size - pfrag->offset);
982
983 if (!sk_wmem_schedule(sk, copy))
984 goto wait_for_memory;
985
986 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
987 pfrag->page,
988 pfrag->offset,
989 copy);
990 if (err)
991 goto out_error;
992
993 /* Update the skb. */
994 if (merge) {
995 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
996 } else {
997 skb_fill_page_desc(skb, i, pfrag->page,
998 pfrag->offset, copy);
999 get_page(pfrag->page);
1000 }
1001
1002 pfrag->offset += copy;
1003 copied += copy;
1004 if (head != skb) {
1005 head->len += copy;
1006 head->data_len += copy;
1007 }
1008
1009 continue;
1010
1011 wait_for_memory:
1012 kcm_push(kcm);
1013 err = sk_stream_wait_memory(sk, &timeo);
1014 if (err)
1015 goto out_error;
1016 }
1017
1018 if (eor) {
1019 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1020
1021 /* Message complete, queue it on send buffer */
1022 __skb_queue_tail(&sk->sk_write_queue, head);
1023 kcm->seq_skb = NULL;
1024 KCM_STATS_INCR(kcm->stats.tx_msgs);
1025
1026 if (msg->msg_flags & MSG_BATCH) {
1027 kcm->tx_wait_more = true;
1028 } else if (kcm->tx_wait_more || not_busy) {
1029 err = kcm_write_msgs(kcm);
1030 if (err < 0) {
1031 /* We got a hard error in write_msgs but have
1032 * already queued this message. Report an error
1033 * in the socket, but don't affect return value
1034 * from sendmsg
1035 */
1036 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1037 report_csk_error(&kcm->sk, -err);
1038 }
1039 }
1040 } else {
1041 /* Message not complete, save state */
1042 partial_message:
1043 kcm->seq_skb = head;
1044 kcm_tx_msg(head)->last_skb = skb;
1045 }
1046
1047 KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1048
1049 release_sock(sk);
1050 return copied;
1051
1052 out_error:
1053 kcm_push(kcm);
1054
1055 if (copied && sock->type == SOCK_SEQPACKET) {
1056 /* Wrote some bytes before encountering an
1057 * error, return partial success.
1058 */
1059 goto partial_message;
1060 }
1061
1062 if (head != kcm->seq_skb)
1063 kfree_skb(head);
1064
1065 err = sk_stream_error(sk, msg->msg_flags, err);
1066
1067 /* make sure we wake any epoll edge trigger waiter */
1068 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1069 sk->sk_write_space(sk);
1070
1071 release_sock(sk);
1072 return err;
1073 }
1074
1075 static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1076 long timeo, int *err)
1077 {
1078 struct sk_buff *skb;
1079
1080 while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1081 if (sk->sk_err) {
1082 *err = sock_error(sk);
1083 return NULL;
1084 }
1085
1086 if (sock_flag(sk, SOCK_DONE))
1087 return NULL;
1088
1089 if ((flags & MSG_DONTWAIT) || !timeo) {
1090 *err = -EAGAIN;
1091 return NULL;
1092 }
1093
1094 sk_wait_data(sk, &timeo, NULL);
1095
1096 /* Handle signals */
1097 if (signal_pending(current)) {
1098 *err = sock_intr_errno(timeo);
1099 return NULL;
1100 }
1101 }
1102
1103 return skb;
1104 }
1105
1106 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1107 size_t len, int flags)
1108 {
1109 struct sock *sk = sock->sk;
1110 struct kcm_sock *kcm = kcm_sk(sk);
1111 int err = 0;
1112 long timeo;
1113 struct strp_rx_msg *rxm;
1114 int copied = 0;
1115 struct sk_buff *skb;
1116
1117 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1118
1119 lock_sock(sk);
1120
1121 skb = kcm_wait_data(sk, flags, timeo, &err);
1122 if (!skb)
1123 goto out;
1124
1125 /* Okay, have a message on the receive queue */
1126
1127 rxm = strp_rx_msg(skb);
1128
1129 if (len > rxm->full_len)
1130 len = rxm->full_len;
1131
1132 err = skb_copy_datagram_msg(skb, rxm->offset, msg, len);
1133 if (err < 0)
1134 goto out;
1135
1136 copied = len;
1137 if (likely(!(flags & MSG_PEEK))) {
1138 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1139 if (copied < rxm->full_len) {
1140 if (sock->type == SOCK_DGRAM) {
1141 /* Truncated message */
1142 msg->msg_flags |= MSG_TRUNC;
1143 goto msg_finished;
1144 }
1145 rxm->offset += copied;
1146 rxm->full_len -= copied;
1147 } else {
1148 msg_finished:
1149 /* Finished with message */
1150 msg->msg_flags |= MSG_EOR;
1151 KCM_STATS_INCR(kcm->stats.rx_msgs);
1152 skb_unlink(skb, &sk->sk_receive_queue);
1153 kfree_skb(skb);
1154 }
1155 }
1156
1157 out:
1158 release_sock(sk);
1159
1160 return copied ? : err;
1161 }
1162
1163 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1164 struct pipe_inode_info *pipe, size_t len,
1165 unsigned int flags)
1166 {
1167 struct sock *sk = sock->sk;
1168 struct kcm_sock *kcm = kcm_sk(sk);
1169 long timeo;
1170 struct strp_rx_msg *rxm;
1171 int err = 0;
1172 ssize_t copied;
1173 struct sk_buff *skb;
1174
1175 /* Only support splice for SOCKSEQPACKET */
1176
1177 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1178
1179 lock_sock(sk);
1180
1181 skb = kcm_wait_data(sk, flags, timeo, &err);
1182 if (!skb)
1183 goto err_out;
1184
1185 /* Okay, have a message on the receive queue */
1186
1187 rxm = strp_rx_msg(skb);
1188
1189 if (len > rxm->full_len)
1190 len = rxm->full_len;
1191
1192 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, len, flags);
1193 if (copied < 0) {
1194 err = copied;
1195 goto err_out;
1196 }
1197
1198 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1199
1200 rxm->offset += copied;
1201 rxm->full_len -= copied;
1202
1203 /* We have no way to return MSG_EOR. If all the bytes have been
1204 * read we still leave the message in the receive socket buffer.
1205 * A subsequent recvmsg needs to be done to return MSG_EOR and
1206 * finish reading the message.
1207 */
1208
1209 release_sock(sk);
1210
1211 return copied;
1212
1213 err_out:
1214 release_sock(sk);
1215
1216 return err;
1217 }
1218
1219 /* kcm sock lock held */
1220 static void kcm_recv_disable(struct kcm_sock *kcm)
1221 {
1222 struct kcm_mux *mux = kcm->mux;
1223
1224 if (kcm->rx_disabled)
1225 return;
1226
1227 spin_lock_bh(&mux->rx_lock);
1228
1229 kcm->rx_disabled = 1;
1230
1231 /* If a psock is reserved we'll do cleanup in unreserve */
1232 if (!kcm->rx_psock) {
1233 if (kcm->rx_wait) {
1234 list_del(&kcm->wait_rx_list);
1235 kcm->rx_wait = false;
1236 }
1237
1238 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1239 }
1240
1241 spin_unlock_bh(&mux->rx_lock);
1242 }
1243
1244 /* kcm sock lock held */
1245 static void kcm_recv_enable(struct kcm_sock *kcm)
1246 {
1247 struct kcm_mux *mux = kcm->mux;
1248
1249 if (!kcm->rx_disabled)
1250 return;
1251
1252 spin_lock_bh(&mux->rx_lock);
1253
1254 kcm->rx_disabled = 0;
1255 kcm_rcv_ready(kcm);
1256
1257 spin_unlock_bh(&mux->rx_lock);
1258 }
1259
1260 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1261 char __user *optval, unsigned int optlen)
1262 {
1263 struct kcm_sock *kcm = kcm_sk(sock->sk);
1264 int val, valbool;
1265 int err = 0;
1266
1267 if (level != SOL_KCM)
1268 return -ENOPROTOOPT;
1269
1270 if (optlen < sizeof(int))
1271 return -EINVAL;
1272
1273 if (get_user(val, (int __user *)optval))
1274 return -EINVAL;
1275
1276 valbool = val ? 1 : 0;
1277
1278 switch (optname) {
1279 case KCM_RECV_DISABLE:
1280 lock_sock(&kcm->sk);
1281 if (valbool)
1282 kcm_recv_disable(kcm);
1283 else
1284 kcm_recv_enable(kcm);
1285 release_sock(&kcm->sk);
1286 break;
1287 default:
1288 err = -ENOPROTOOPT;
1289 }
1290
1291 return err;
1292 }
1293
1294 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1295 char __user *optval, int __user *optlen)
1296 {
1297 struct kcm_sock *kcm = kcm_sk(sock->sk);
1298 int val, len;
1299
1300 if (level != SOL_KCM)
1301 return -ENOPROTOOPT;
1302
1303 if (get_user(len, optlen))
1304 return -EFAULT;
1305
1306 len = min_t(unsigned int, len, sizeof(int));
1307 if (len < 0)
1308 return -EINVAL;
1309
1310 switch (optname) {
1311 case KCM_RECV_DISABLE:
1312 val = kcm->rx_disabled;
1313 break;
1314 default:
1315 return -ENOPROTOOPT;
1316 }
1317
1318 if (put_user(len, optlen))
1319 return -EFAULT;
1320 if (copy_to_user(optval, &val, len))
1321 return -EFAULT;
1322 return 0;
1323 }
1324
1325 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1326 {
1327 struct kcm_sock *tkcm;
1328 struct list_head *head;
1329 int index = 0;
1330
1331 /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1332 * we set sk_state, otherwise epoll_wait always returns right away with
1333 * POLLHUP
1334 */
1335 kcm->sk.sk_state = TCP_ESTABLISHED;
1336
1337 /* Add to mux's kcm sockets list */
1338 kcm->mux = mux;
1339 spin_lock_bh(&mux->lock);
1340
1341 head = &mux->kcm_socks;
1342 list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1343 if (tkcm->index != index)
1344 break;
1345 head = &tkcm->kcm_sock_list;
1346 index++;
1347 }
1348
1349 list_add(&kcm->kcm_sock_list, head);
1350 kcm->index = index;
1351
1352 mux->kcm_socks_cnt++;
1353 spin_unlock_bh(&mux->lock);
1354
1355 INIT_WORK(&kcm->tx_work, kcm_tx_work);
1356
1357 spin_lock_bh(&mux->rx_lock);
1358 kcm_rcv_ready(kcm);
1359 spin_unlock_bh(&mux->rx_lock);
1360 }
1361
1362 static int kcm_attach(struct socket *sock, struct socket *csock,
1363 struct bpf_prog *prog)
1364 {
1365 struct kcm_sock *kcm = kcm_sk(sock->sk);
1366 struct kcm_mux *mux = kcm->mux;
1367 struct sock *csk;
1368 struct kcm_psock *psock = NULL, *tpsock;
1369 struct list_head *head;
1370 int index = 0;
1371 struct strp_callbacks cb;
1372 int err;
1373
1374 csk = csock->sk;
1375 if (!csk)
1376 return -EINVAL;
1377
1378 psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1379 if (!psock)
1380 return -ENOMEM;
1381
1382 psock->mux = mux;
1383 psock->sk = csk;
1384 psock->bpf_prog = prog;
1385
1386 cb.rcv_msg = kcm_rcv_strparser;
1387 cb.abort_parser = NULL;
1388 cb.parse_msg = kcm_parse_func_strparser;
1389 cb.read_sock_done = kcm_read_sock_done;
1390
1391 err = strp_init(&psock->strp, csk, &cb);
1392 if (err) {
1393 kmem_cache_free(kcm_psockp, psock);
1394 return err;
1395 }
1396
1397 sock_hold(csk);
1398
1399 write_lock_bh(&csk->sk_callback_lock);
1400 psock->save_data_ready = csk->sk_data_ready;
1401 psock->save_write_space = csk->sk_write_space;
1402 psock->save_state_change = csk->sk_state_change;
1403 csk->sk_user_data = psock;
1404 csk->sk_data_ready = psock_data_ready;
1405 csk->sk_write_space = psock_write_space;
1406 csk->sk_state_change = psock_state_change;
1407 write_unlock_bh(&csk->sk_callback_lock);
1408
1409 /* Finished initialization, now add the psock to the MUX. */
1410 spin_lock_bh(&mux->lock);
1411 head = &mux->psocks;
1412 list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1413 if (tpsock->index != index)
1414 break;
1415 head = &tpsock->psock_list;
1416 index++;
1417 }
1418
1419 list_add(&psock->psock_list, head);
1420 psock->index = index;
1421
1422 KCM_STATS_INCR(mux->stats.psock_attach);
1423 mux->psocks_cnt++;
1424 psock_now_avail(psock);
1425 spin_unlock_bh(&mux->lock);
1426
1427 /* Schedule RX work in case there are already bytes queued */
1428 strp_check_rcv(&psock->strp);
1429
1430 return 0;
1431 }
1432
1433 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1434 {
1435 struct socket *csock;
1436 struct bpf_prog *prog;
1437 int err;
1438
1439 csock = sockfd_lookup(info->fd, &err);
1440 if (!csock)
1441 return -ENOENT;
1442
1443 prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1444 if (IS_ERR(prog)) {
1445 err = PTR_ERR(prog);
1446 goto out;
1447 }
1448
1449 err = kcm_attach(sock, csock, prog);
1450 if (err) {
1451 bpf_prog_put(prog);
1452 goto out;
1453 }
1454
1455 /* Keep reference on file also */
1456
1457 return 0;
1458 out:
1459 fput(csock->file);
1460 return err;
1461 }
1462
1463 static void kcm_unattach(struct kcm_psock *psock)
1464 {
1465 struct sock *csk = psock->sk;
1466 struct kcm_mux *mux = psock->mux;
1467
1468 lock_sock(csk);
1469
1470 /* Stop getting callbacks from TCP socket. After this there should
1471 * be no way to reserve a kcm for this psock.
1472 */
1473 write_lock_bh(&csk->sk_callback_lock);
1474 csk->sk_user_data = NULL;
1475 csk->sk_data_ready = psock->save_data_ready;
1476 csk->sk_write_space = psock->save_write_space;
1477 csk->sk_state_change = psock->save_state_change;
1478 strp_stop(&psock->strp);
1479
1480 if (WARN_ON(psock->rx_kcm)) {
1481 write_unlock_bh(&csk->sk_callback_lock);
1482 return;
1483 }
1484
1485 spin_lock_bh(&mux->rx_lock);
1486
1487 /* Stop receiver activities. After this point psock should not be
1488 * able to get onto ready list either through callbacks or work.
1489 */
1490 if (psock->ready_rx_msg) {
1491 list_del(&psock->psock_ready_list);
1492 kfree_skb(psock->ready_rx_msg);
1493 psock->ready_rx_msg = NULL;
1494 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1495 }
1496
1497 spin_unlock_bh(&mux->rx_lock);
1498
1499 write_unlock_bh(&csk->sk_callback_lock);
1500
1501 /* Call strp_done without sock lock */
1502 release_sock(csk);
1503 strp_done(&psock->strp);
1504 lock_sock(csk);
1505
1506 bpf_prog_put(psock->bpf_prog);
1507
1508 spin_lock_bh(&mux->lock);
1509
1510 aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1511 save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1512
1513 KCM_STATS_INCR(mux->stats.psock_unattach);
1514
1515 if (psock->tx_kcm) {
1516 /* psock was reserved. Just mark it finished and we will clean
1517 * up in the kcm paths, we need kcm lock which can not be
1518 * acquired here.
1519 */
1520 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1521 spin_unlock_bh(&mux->lock);
1522
1523 /* We are unattaching a socket that is reserved. Abort the
1524 * socket since we may be out of sync in sending on it. We need
1525 * to do this without the mux lock.
1526 */
1527 kcm_abort_tx_psock(psock, EPIPE, false);
1528
1529 spin_lock_bh(&mux->lock);
1530 if (!psock->tx_kcm) {
1531 /* psock now unreserved in window mux was unlocked */
1532 goto no_reserved;
1533 }
1534 psock->done = 1;
1535
1536 /* Commit done before queuing work to process it */
1537 smp_mb();
1538
1539 /* Queue tx work to make sure psock->done is handled */
1540 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1541 spin_unlock_bh(&mux->lock);
1542 } else {
1543 no_reserved:
1544 if (!psock->tx_stopped)
1545 list_del(&psock->psock_avail_list);
1546 list_del(&psock->psock_list);
1547 mux->psocks_cnt--;
1548 spin_unlock_bh(&mux->lock);
1549
1550 sock_put(csk);
1551 fput(csk->sk_socket->file);
1552 kmem_cache_free(kcm_psockp, psock);
1553 }
1554
1555 release_sock(csk);
1556 }
1557
1558 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1559 {
1560 struct kcm_sock *kcm = kcm_sk(sock->sk);
1561 struct kcm_mux *mux = kcm->mux;
1562 struct kcm_psock *psock;
1563 struct socket *csock;
1564 struct sock *csk;
1565 int err;
1566
1567 csock = sockfd_lookup(info->fd, &err);
1568 if (!csock)
1569 return -ENOENT;
1570
1571 csk = csock->sk;
1572 if (!csk) {
1573 err = -EINVAL;
1574 goto out;
1575 }
1576
1577 err = -ENOENT;
1578
1579 spin_lock_bh(&mux->lock);
1580
1581 list_for_each_entry(psock, &mux->psocks, psock_list) {
1582 if (psock->sk != csk)
1583 continue;
1584
1585 /* Found the matching psock */
1586
1587 if (psock->unattaching || WARN_ON(psock->done)) {
1588 err = -EALREADY;
1589 break;
1590 }
1591
1592 psock->unattaching = 1;
1593
1594 spin_unlock_bh(&mux->lock);
1595
1596 /* Lower socket lock should already be held */
1597 kcm_unattach(psock);
1598
1599 err = 0;
1600 goto out;
1601 }
1602
1603 spin_unlock_bh(&mux->lock);
1604
1605 out:
1606 fput(csock->file);
1607 return err;
1608 }
1609
1610 static struct proto kcm_proto = {
1611 .name = "KCM",
1612 .owner = THIS_MODULE,
1613 .obj_size = sizeof(struct kcm_sock),
1614 };
1615
1616 /* Clone a kcm socket. */
1617 static int kcm_clone(struct socket *osock, struct kcm_clone *info,
1618 struct socket **newsockp)
1619 {
1620 struct socket *newsock;
1621 struct sock *newsk;
1622 struct file *newfile;
1623 int err, newfd;
1624
1625 err = -ENFILE;
1626 newsock = sock_alloc();
1627 if (!newsock)
1628 goto out;
1629
1630 newsock->type = osock->type;
1631 newsock->ops = osock->ops;
1632
1633 __module_get(newsock->ops->owner);
1634
1635 newfd = get_unused_fd_flags(0);
1636 if (unlikely(newfd < 0)) {
1637 err = newfd;
1638 goto out_fd_fail;
1639 }
1640
1641 newfile = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1642 if (unlikely(IS_ERR(newfile))) {
1643 err = PTR_ERR(newfile);
1644 goto out_sock_alloc_fail;
1645 }
1646
1647 newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1648 &kcm_proto, true);
1649 if (!newsk) {
1650 err = -ENOMEM;
1651 goto out_sk_alloc_fail;
1652 }
1653
1654 sock_init_data(newsock, newsk);
1655 init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1656
1657 fd_install(newfd, newfile);
1658 *newsockp = newsock;
1659 info->fd = newfd;
1660
1661 return 0;
1662
1663 out_sk_alloc_fail:
1664 fput(newfile);
1665 out_sock_alloc_fail:
1666 put_unused_fd(newfd);
1667 out_fd_fail:
1668 sock_release(newsock);
1669 out:
1670 return err;
1671 }
1672
1673 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1674 {
1675 int err;
1676
1677 switch (cmd) {
1678 case SIOCKCMATTACH: {
1679 struct kcm_attach info;
1680
1681 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1682 err = -EFAULT;
1683
1684 err = kcm_attach_ioctl(sock, &info);
1685
1686 break;
1687 }
1688 case SIOCKCMUNATTACH: {
1689 struct kcm_unattach info;
1690
1691 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1692 err = -EFAULT;
1693
1694 err = kcm_unattach_ioctl(sock, &info);
1695
1696 break;
1697 }
1698 case SIOCKCMCLONE: {
1699 struct kcm_clone info;
1700 struct socket *newsock = NULL;
1701
1702 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1703 err = -EFAULT;
1704
1705 err = kcm_clone(sock, &info, &newsock);
1706
1707 if (!err) {
1708 if (copy_to_user((void __user *)arg, &info,
1709 sizeof(info))) {
1710 err = -EFAULT;
1711 sys_close(info.fd);
1712 }
1713 }
1714
1715 break;
1716 }
1717 default:
1718 err = -ENOIOCTLCMD;
1719 break;
1720 }
1721
1722 return err;
1723 }
1724
1725 static void free_mux(struct rcu_head *rcu)
1726 {
1727 struct kcm_mux *mux = container_of(rcu,
1728 struct kcm_mux, rcu);
1729
1730 kmem_cache_free(kcm_muxp, mux);
1731 }
1732
1733 static void release_mux(struct kcm_mux *mux)
1734 {
1735 struct kcm_net *knet = mux->knet;
1736 struct kcm_psock *psock, *tmp_psock;
1737
1738 /* Release psocks */
1739 list_for_each_entry_safe(psock, tmp_psock,
1740 &mux->psocks, psock_list) {
1741 if (!WARN_ON(psock->unattaching))
1742 kcm_unattach(psock);
1743 }
1744
1745 if (WARN_ON(mux->psocks_cnt))
1746 return;
1747
1748 __skb_queue_purge(&mux->rx_hold_queue);
1749
1750 mutex_lock(&knet->mutex);
1751 aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1752 aggregate_psock_stats(&mux->aggregate_psock_stats,
1753 &knet->aggregate_psock_stats);
1754 aggregate_strp_stats(&mux->aggregate_strp_stats,
1755 &knet->aggregate_strp_stats);
1756 list_del_rcu(&mux->kcm_mux_list);
1757 knet->count--;
1758 mutex_unlock(&knet->mutex);
1759
1760 call_rcu(&mux->rcu, free_mux);
1761 }
1762
1763 static void kcm_done(struct kcm_sock *kcm)
1764 {
1765 struct kcm_mux *mux = kcm->mux;
1766 struct sock *sk = &kcm->sk;
1767 int socks_cnt;
1768
1769 spin_lock_bh(&mux->rx_lock);
1770 if (kcm->rx_psock) {
1771 /* Cleanup in unreserve_rx_kcm */
1772 WARN_ON(kcm->done);
1773 kcm->rx_disabled = 1;
1774 kcm->done = 1;
1775 spin_unlock_bh(&mux->rx_lock);
1776 return;
1777 }
1778
1779 if (kcm->rx_wait) {
1780 list_del(&kcm->wait_rx_list);
1781 kcm->rx_wait = false;
1782 }
1783 /* Move any pending receive messages to other kcm sockets */
1784 requeue_rx_msgs(mux, &sk->sk_receive_queue);
1785
1786 spin_unlock_bh(&mux->rx_lock);
1787
1788 if (WARN_ON(sk_rmem_alloc_get(sk)))
1789 return;
1790
1791 /* Detach from MUX */
1792 spin_lock_bh(&mux->lock);
1793
1794 list_del(&kcm->kcm_sock_list);
1795 mux->kcm_socks_cnt--;
1796 socks_cnt = mux->kcm_socks_cnt;
1797
1798 spin_unlock_bh(&mux->lock);
1799
1800 if (!socks_cnt) {
1801 /* We are done with the mux now. */
1802 release_mux(mux);
1803 }
1804
1805 WARN_ON(kcm->rx_wait);
1806
1807 sock_put(&kcm->sk);
1808 }
1809
1810 /* Called by kcm_release to close a KCM socket.
1811 * If this is the last KCM socket on the MUX, destroy the MUX.
1812 */
1813 static int kcm_release(struct socket *sock)
1814 {
1815 struct sock *sk = sock->sk;
1816 struct kcm_sock *kcm;
1817 struct kcm_mux *mux;
1818 struct kcm_psock *psock;
1819
1820 if (!sk)
1821 return 0;
1822
1823 kcm = kcm_sk(sk);
1824 mux = kcm->mux;
1825
1826 sock_orphan(sk);
1827 kfree_skb(kcm->seq_skb);
1828
1829 lock_sock(sk);
1830 /* Purge queue under lock to avoid race condition with tx_work trying
1831 * to act when queue is nonempty. If tx_work runs after this point
1832 * it will just return.
1833 */
1834 __skb_queue_purge(&sk->sk_write_queue);
1835
1836 /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1837 * get a writespace callback. This prevents further work being queued
1838 * from the callback (unbinding the psock occurs after canceling work.
1839 */
1840 kcm->tx_stopped = 1;
1841
1842 release_sock(sk);
1843
1844 spin_lock_bh(&mux->lock);
1845 if (kcm->tx_wait) {
1846 /* Take of tx_wait list, after this point there should be no way
1847 * that a psock will be assigned to this kcm.
1848 */
1849 list_del(&kcm->wait_psock_list);
1850 kcm->tx_wait = false;
1851 }
1852 spin_unlock_bh(&mux->lock);
1853
1854 /* Cancel work. After this point there should be no outside references
1855 * to the kcm socket.
1856 */
1857 cancel_work_sync(&kcm->tx_work);
1858
1859 lock_sock(sk);
1860 psock = kcm->tx_psock;
1861 if (psock) {
1862 /* A psock was reserved, so we need to kill it since it
1863 * may already have some bytes queued from a message. We
1864 * need to do this after removing kcm from tx_wait list.
1865 */
1866 kcm_abort_tx_psock(psock, EPIPE, false);
1867 unreserve_psock(kcm);
1868 }
1869 release_sock(sk);
1870
1871 WARN_ON(kcm->tx_wait);
1872 WARN_ON(kcm->tx_psock);
1873
1874 sock->sk = NULL;
1875
1876 kcm_done(kcm);
1877
1878 return 0;
1879 }
1880
1881 static const struct proto_ops kcm_dgram_ops = {
1882 .family = PF_KCM,
1883 .owner = THIS_MODULE,
1884 .release = kcm_release,
1885 .bind = sock_no_bind,
1886 .connect = sock_no_connect,
1887 .socketpair = sock_no_socketpair,
1888 .accept = sock_no_accept,
1889 .getname = sock_no_getname,
1890 .poll = datagram_poll,
1891 .ioctl = kcm_ioctl,
1892 .listen = sock_no_listen,
1893 .shutdown = sock_no_shutdown,
1894 .setsockopt = kcm_setsockopt,
1895 .getsockopt = kcm_getsockopt,
1896 .sendmsg = kcm_sendmsg,
1897 .recvmsg = kcm_recvmsg,
1898 .mmap = sock_no_mmap,
1899 .sendpage = kcm_sendpage,
1900 };
1901
1902 static const struct proto_ops kcm_seqpacket_ops = {
1903 .family = PF_KCM,
1904 .owner = THIS_MODULE,
1905 .release = kcm_release,
1906 .bind = sock_no_bind,
1907 .connect = sock_no_connect,
1908 .socketpair = sock_no_socketpair,
1909 .accept = sock_no_accept,
1910 .getname = sock_no_getname,
1911 .poll = datagram_poll,
1912 .ioctl = kcm_ioctl,
1913 .listen = sock_no_listen,
1914 .shutdown = sock_no_shutdown,
1915 .setsockopt = kcm_setsockopt,
1916 .getsockopt = kcm_getsockopt,
1917 .sendmsg = kcm_sendmsg,
1918 .recvmsg = kcm_recvmsg,
1919 .mmap = sock_no_mmap,
1920 .sendpage = kcm_sendpage,
1921 .splice_read = kcm_splice_read,
1922 };
1923
1924 /* Create proto operation for kcm sockets */
1925 static int kcm_create(struct net *net, struct socket *sock,
1926 int protocol, int kern)
1927 {
1928 struct kcm_net *knet = net_generic(net, kcm_net_id);
1929 struct sock *sk;
1930 struct kcm_mux *mux;
1931
1932 switch (sock->type) {
1933 case SOCK_DGRAM:
1934 sock->ops = &kcm_dgram_ops;
1935 break;
1936 case SOCK_SEQPACKET:
1937 sock->ops = &kcm_seqpacket_ops;
1938 break;
1939 default:
1940 return -ESOCKTNOSUPPORT;
1941 }
1942
1943 if (protocol != KCMPROTO_CONNECTED)
1944 return -EPROTONOSUPPORT;
1945
1946 sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1947 if (!sk)
1948 return -ENOMEM;
1949
1950 /* Allocate a kcm mux, shared between KCM sockets */
1951 mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1952 if (!mux) {
1953 sk_free(sk);
1954 return -ENOMEM;
1955 }
1956
1957 spin_lock_init(&mux->lock);
1958 spin_lock_init(&mux->rx_lock);
1959 INIT_LIST_HEAD(&mux->kcm_socks);
1960 INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1961 INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1962
1963 INIT_LIST_HEAD(&mux->psocks);
1964 INIT_LIST_HEAD(&mux->psocks_ready);
1965 INIT_LIST_HEAD(&mux->psocks_avail);
1966
1967 mux->knet = knet;
1968
1969 /* Add new MUX to list */
1970 mutex_lock(&knet->mutex);
1971 list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1972 knet->count++;
1973 mutex_unlock(&knet->mutex);
1974
1975 skb_queue_head_init(&mux->rx_hold_queue);
1976
1977 /* Init KCM socket */
1978 sock_init_data(sock, sk);
1979 init_kcm_sock(kcm_sk(sk), mux);
1980
1981 return 0;
1982 }
1983
1984 static struct net_proto_family kcm_family_ops = {
1985 .family = PF_KCM,
1986 .create = kcm_create,
1987 .owner = THIS_MODULE,
1988 };
1989
1990 static __net_init int kcm_init_net(struct net *net)
1991 {
1992 struct kcm_net *knet = net_generic(net, kcm_net_id);
1993
1994 INIT_LIST_HEAD_RCU(&knet->mux_list);
1995 mutex_init(&knet->mutex);
1996
1997 return 0;
1998 }
1999
2000 static __net_exit void kcm_exit_net(struct net *net)
2001 {
2002 struct kcm_net *knet = net_generic(net, kcm_net_id);
2003
2004 /* All KCM sockets should be closed at this point, which should mean
2005 * that all multiplexors and psocks have been destroyed.
2006 */
2007 WARN_ON(!list_empty(&knet->mux_list));
2008 }
2009
2010 static struct pernet_operations kcm_net_ops = {
2011 .init = kcm_init_net,
2012 .exit = kcm_exit_net,
2013 .id = &kcm_net_id,
2014 .size = sizeof(struct kcm_net),
2015 };
2016
2017 static int __init kcm_init(void)
2018 {
2019 int err = -ENOMEM;
2020
2021 kcm_muxp = kmem_cache_create("kcm_mux_cache",
2022 sizeof(struct kcm_mux), 0,
2023 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2024 if (!kcm_muxp)
2025 goto fail;
2026
2027 kcm_psockp = kmem_cache_create("kcm_psock_cache",
2028 sizeof(struct kcm_psock), 0,
2029 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2030 if (!kcm_psockp)
2031 goto fail;
2032
2033 kcm_wq = create_singlethread_workqueue("kkcmd");
2034 if (!kcm_wq)
2035 goto fail;
2036
2037 err = proto_register(&kcm_proto, 1);
2038 if (err)
2039 goto fail;
2040
2041 err = sock_register(&kcm_family_ops);
2042 if (err)
2043 goto sock_register_fail;
2044
2045 err = register_pernet_device(&kcm_net_ops);
2046 if (err)
2047 goto net_ops_fail;
2048
2049 err = kcm_proc_init();
2050 if (err)
2051 goto proc_init_fail;
2052
2053 return 0;
2054
2055 proc_init_fail:
2056 unregister_pernet_device(&kcm_net_ops);
2057
2058 net_ops_fail:
2059 sock_unregister(PF_KCM);
2060
2061 sock_register_fail:
2062 proto_unregister(&kcm_proto);
2063
2064 fail:
2065 kmem_cache_destroy(kcm_muxp);
2066 kmem_cache_destroy(kcm_psockp);
2067
2068 if (kcm_wq)
2069 destroy_workqueue(kcm_wq);
2070
2071 return err;
2072 }
2073
2074 static void __exit kcm_exit(void)
2075 {
2076 kcm_proc_exit();
2077 unregister_pernet_device(&kcm_net_ops);
2078 sock_unregister(PF_KCM);
2079 proto_unregister(&kcm_proto);
2080 destroy_workqueue(kcm_wq);
2081
2082 kmem_cache_destroy(kcm_muxp);
2083 kmem_cache_destroy(kcm_psockp);
2084 }
2085
2086 module_init(kcm_init);
2087 module_exit(kcm_exit);
2088
2089 MODULE_LICENSE("GPL");
2090 MODULE_ALIAS_NETPROTO(PF_KCM);
2091
Ubuntu Artful kernel mirror