Commit | Line | Data |
---|---|---|
9b73896a TH |
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 | ||
ab7ac4eb TH |
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> | |
c0338aff | 26 | #include <linux/syscalls.h> |
174cd4b1 IM |
27 | #include <linux/sched/signal.h> |
28 | ||
ab7ac4eb TH |
29 | #include <net/kcm.h> |
30 | #include <net/netns/generic.h> | |
31 | #include <net/sock.h> | |
ab7ac4eb TH |
32 | #include <uapi/linux/kcm.h> |
33 | ||
34 | unsigned int kcm_net_id; | |
35 | ||
36 | static struct kmem_cache *kcm_psockp __read_mostly; | |
37 | static struct kmem_cache *kcm_muxp __read_mostly; | |
38 | static struct workqueue_struct *kcm_wq; | |
39 | ||
40 | static inline struct kcm_sock *kcm_sk(const struct sock *sk) | |
41 | { | |
42 | return (struct kcm_sock *)sk; | |
43 | } | |
44 | ||
45 | static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb) | |
46 | { | |
47 | return (struct kcm_tx_msg *)skb->cb; | |
48 | } | |
49 | ||
ab7ac4eb TH |
50 | static void report_csk_error(struct sock *csk, int err) |
51 | { | |
52 | csk->sk_err = EPIPE; | |
53 | csk->sk_error_report(csk); | |
54 | } | |
55 | ||
ab7ac4eb TH |
56 | static void kcm_abort_tx_psock(struct kcm_psock *psock, int err, |
57 | bool wakeup_kcm) | |
58 | { | |
59 | struct sock *csk = psock->sk; | |
60 | struct kcm_mux *mux = psock->mux; | |
61 | ||
62 | /* Unrecoverable error in transmit */ | |
63 | ||
64 | spin_lock_bh(&mux->lock); | |
65 | ||
66 | if (psock->tx_stopped) { | |
67 | spin_unlock_bh(&mux->lock); | |
68 | return; | |
69 | } | |
70 | ||
71 | psock->tx_stopped = 1; | |
cd6e111b | 72 | KCM_STATS_INCR(psock->stats.tx_aborts); |
ab7ac4eb TH |
73 | |
74 | if (!psock->tx_kcm) { | |
75 | /* Take off psocks_avail list */ | |
76 | list_del(&psock->psock_avail_list); | |
77 | } else if (wakeup_kcm) { | |
78 | /* In this case psock is being aborted while outside of | |
79 | * write_msgs and psock is reserved. Schedule tx_work | |
80 | * to handle the failure there. Need to commit tx_stopped | |
81 | * before queuing work. | |
82 | */ | |
83 | smp_mb(); | |
84 | ||
85 | queue_work(kcm_wq, &psock->tx_kcm->tx_work); | |
86 | } | |
87 | ||
88 | spin_unlock_bh(&mux->lock); | |
89 | ||
90 | /* Report error on lower socket */ | |
91 | report_csk_error(csk, err); | |
92 | } | |
93 | ||
cd6e111b TH |
94 | /* RX mux lock held. */ |
95 | static void kcm_update_rx_mux_stats(struct kcm_mux *mux, | |
96 | struct kcm_psock *psock) | |
97 | { | |
9b73896a TH |
98 | STRP_STATS_ADD(mux->stats.rx_bytes, |
99 | psock->strp.stats.rx_bytes - | |
100 | psock->saved_rx_bytes); | |
cd6e111b | 101 | mux->stats.rx_msgs += |
9b73896a TH |
102 | psock->strp.stats.rx_msgs - psock->saved_rx_msgs; |
103 | psock->saved_rx_msgs = psock->strp.stats.rx_msgs; | |
104 | psock->saved_rx_bytes = psock->strp.stats.rx_bytes; | |
cd6e111b TH |
105 | } |
106 | ||
107 | static void kcm_update_tx_mux_stats(struct kcm_mux *mux, | |
108 | struct kcm_psock *psock) | |
109 | { | |
110 | KCM_STATS_ADD(mux->stats.tx_bytes, | |
111 | psock->stats.tx_bytes - psock->saved_tx_bytes); | |
112 | mux->stats.tx_msgs += | |
113 | psock->stats.tx_msgs - psock->saved_tx_msgs; | |
114 | psock->saved_tx_msgs = psock->stats.tx_msgs; | |
115 | psock->saved_tx_bytes = psock->stats.tx_bytes; | |
116 | } | |
117 | ||
ab7ac4eb TH |
118 | static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); |
119 | ||
120 | /* KCM is ready to receive messages on its queue-- either the KCM is new or | |
121 | * has become unblocked after being blocked on full socket buffer. Queue any | |
122 | * pending ready messages on a psock. RX mux lock held. | |
123 | */ | |
124 | static void kcm_rcv_ready(struct kcm_sock *kcm) | |
125 | { | |
126 | struct kcm_mux *mux = kcm->mux; | |
127 | struct kcm_psock *psock; | |
128 | struct sk_buff *skb; | |
129 | ||
130 | if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled)) | |
131 | return; | |
132 | ||
133 | while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) { | |
134 | if (kcm_queue_rcv_skb(&kcm->sk, skb)) { | |
135 | /* Assuming buffer limit has been reached */ | |
136 | skb_queue_head(&mux->rx_hold_queue, skb); | |
137 | WARN_ON(!sk_rmem_alloc_get(&kcm->sk)); | |
138 | return; | |
139 | } | |
140 | } | |
141 | ||
142 | while (!list_empty(&mux->psocks_ready)) { | |
143 | psock = list_first_entry(&mux->psocks_ready, struct kcm_psock, | |
144 | psock_ready_list); | |
145 | ||
146 | if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) { | |
147 | /* Assuming buffer limit has been reached */ | |
148 | WARN_ON(!sk_rmem_alloc_get(&kcm->sk)); | |
149 | return; | |
150 | } | |
151 | ||
152 | /* Consumed the ready message on the psock. Schedule rx_work to | |
153 | * get more messages. | |
154 | */ | |
155 | list_del(&psock->psock_ready_list); | |
156 | psock->ready_rx_msg = NULL; | |
ab7ac4eb TH |
157 | /* Commit clearing of ready_rx_msg for queuing work */ |
158 | smp_mb(); | |
159 | ||
9b73896a TH |
160 | strp_unpause(&psock->strp); |
161 | strp_check_rcv(&psock->strp); | |
ab7ac4eb TH |
162 | } |
163 | ||
164 | /* Buffer limit is okay now, add to ready list */ | |
165 | list_add_tail(&kcm->wait_rx_list, | |
166 | &kcm->mux->kcm_rx_waiters); | |
167 | kcm->rx_wait = true; | |
168 | } | |
169 | ||
170 | static void kcm_rfree(struct sk_buff *skb) | |
171 | { | |
172 | struct sock *sk = skb->sk; | |
173 | struct kcm_sock *kcm = kcm_sk(sk); | |
174 | struct kcm_mux *mux = kcm->mux; | |
175 | unsigned int len = skb->truesize; | |
176 | ||
177 | sk_mem_uncharge(sk, len); | |
178 | atomic_sub(len, &sk->sk_rmem_alloc); | |
179 | ||
180 | /* For reading rx_wait and rx_psock without holding lock */ | |
181 | smp_mb__after_atomic(); | |
182 | ||
183 | if (!kcm->rx_wait && !kcm->rx_psock && | |
184 | sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) { | |
185 | spin_lock_bh(&mux->rx_lock); | |
186 | kcm_rcv_ready(kcm); | |
187 | spin_unlock_bh(&mux->rx_lock); | |
188 | } | |
189 | } | |
190 | ||
191 | static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) | |
192 | { | |
193 | struct sk_buff_head *list = &sk->sk_receive_queue; | |
194 | ||
195 | if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) | |
196 | return -ENOMEM; | |
197 | ||
198 | if (!sk_rmem_schedule(sk, skb, skb->truesize)) | |
199 | return -ENOBUFS; | |
200 | ||
201 | skb->dev = NULL; | |
202 | ||
203 | skb_orphan(skb); | |
204 | skb->sk = sk; | |
205 | skb->destructor = kcm_rfree; | |
206 | atomic_add(skb->truesize, &sk->sk_rmem_alloc); | |
207 | sk_mem_charge(sk, skb->truesize); | |
208 | ||
209 | skb_queue_tail(list, skb); | |
210 | ||
211 | if (!sock_flag(sk, SOCK_DEAD)) | |
212 | sk->sk_data_ready(sk); | |
213 | ||
214 | return 0; | |
215 | } | |
216 | ||
217 | /* Requeue received messages for a kcm socket to other kcm sockets. This is | |
218 | * called with a kcm socket is receive disabled. | |
219 | * RX mux lock held. | |
220 | */ | |
221 | static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head) | |
222 | { | |
223 | struct sk_buff *skb; | |
224 | struct kcm_sock *kcm; | |
225 | ||
226 | while ((skb = __skb_dequeue(head))) { | |
227 | /* Reset destructor to avoid calling kcm_rcv_ready */ | |
228 | skb->destructor = sock_rfree; | |
229 | skb_orphan(skb); | |
230 | try_again: | |
231 | if (list_empty(&mux->kcm_rx_waiters)) { | |
232 | skb_queue_tail(&mux->rx_hold_queue, skb); | |
233 | continue; | |
234 | } | |
235 | ||
236 | kcm = list_first_entry(&mux->kcm_rx_waiters, | |
237 | struct kcm_sock, wait_rx_list); | |
238 | ||
239 | if (kcm_queue_rcv_skb(&kcm->sk, skb)) { | |
240 | /* Should mean socket buffer full */ | |
241 | list_del(&kcm->wait_rx_list); | |
242 | kcm->rx_wait = false; | |
243 | ||
244 | /* Commit rx_wait to read in kcm_free */ | |
245 | smp_wmb(); | |
246 | ||
247 | goto try_again; | |
248 | } | |
249 | } | |
250 | } | |
251 | ||
252 | /* Lower sock lock held */ | |
253 | static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock, | |
254 | struct sk_buff *head) | |
255 | { | |
256 | struct kcm_mux *mux = psock->mux; | |
257 | struct kcm_sock *kcm; | |
258 | ||
259 | WARN_ON(psock->ready_rx_msg); | |
260 | ||
261 | if (psock->rx_kcm) | |
262 | return psock->rx_kcm; | |
263 | ||
264 | spin_lock_bh(&mux->rx_lock); | |
265 | ||
266 | if (psock->rx_kcm) { | |
267 | spin_unlock_bh(&mux->rx_lock); | |
268 | return psock->rx_kcm; | |
269 | } | |
270 | ||
cd6e111b TH |
271 | kcm_update_rx_mux_stats(mux, psock); |
272 | ||
ab7ac4eb TH |
273 | if (list_empty(&mux->kcm_rx_waiters)) { |
274 | psock->ready_rx_msg = head; | |
9b73896a | 275 | strp_pause(&psock->strp); |
ab7ac4eb TH |
276 | list_add_tail(&psock->psock_ready_list, |
277 | &mux->psocks_ready); | |
278 | spin_unlock_bh(&mux->rx_lock); | |
279 | return NULL; | |
280 | } | |
281 | ||
282 | kcm = list_first_entry(&mux->kcm_rx_waiters, | |
283 | struct kcm_sock, wait_rx_list); | |
284 | list_del(&kcm->wait_rx_list); | |
285 | kcm->rx_wait = false; | |
286 | ||
287 | psock->rx_kcm = kcm; | |
288 | kcm->rx_psock = psock; | |
289 | ||
290 | spin_unlock_bh(&mux->rx_lock); | |
291 | ||
292 | return kcm; | |
293 | } | |
294 | ||
295 | static void kcm_done(struct kcm_sock *kcm); | |
296 | ||
297 | static void kcm_done_work(struct work_struct *w) | |
298 | { | |
299 | kcm_done(container_of(w, struct kcm_sock, done_work)); | |
300 | } | |
301 | ||
302 | /* Lower sock held */ | |
303 | static void unreserve_rx_kcm(struct kcm_psock *psock, | |
304 | bool rcv_ready) | |
305 | { | |
306 | struct kcm_sock *kcm = psock->rx_kcm; | |
307 | struct kcm_mux *mux = psock->mux; | |
308 | ||
309 | if (!kcm) | |
310 | return; | |
311 | ||
312 | spin_lock_bh(&mux->rx_lock); | |
313 | ||
314 | psock->rx_kcm = NULL; | |
315 | kcm->rx_psock = NULL; | |
316 | ||
317 | /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with | |
318 | * kcm_rfree | |
319 | */ | |
320 | smp_mb(); | |
321 | ||
322 | if (unlikely(kcm->done)) { | |
323 | spin_unlock_bh(&mux->rx_lock); | |
324 | ||
325 | /* Need to run kcm_done in a task since we need to qcquire | |
326 | * callback locks which may already be held here. | |
327 | */ | |
328 | INIT_WORK(&kcm->done_work, kcm_done_work); | |
329 | schedule_work(&kcm->done_work); | |
330 | return; | |
331 | } | |
332 | ||
333 | if (unlikely(kcm->rx_disabled)) { | |
334 | requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue); | |
335 | } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) { | |
336 | /* Check for degenerative race with rx_wait that all | |
337 | * data was dequeued (accounted for in kcm_rfree). | |
338 | */ | |
339 | kcm_rcv_ready(kcm); | |
340 | } | |
341 | spin_unlock_bh(&mux->rx_lock); | |
342 | } | |
343 | ||
ab7ac4eb | 344 | /* Lower sock lock held */ |
96a59083 | 345 | static void psock_data_ready(struct sock *sk) |
ab7ac4eb TH |
346 | { |
347 | struct kcm_psock *psock; | |
348 | ||
349 | read_lock_bh(&sk->sk_callback_lock); | |
350 | ||
351 | psock = (struct kcm_psock *)sk->sk_user_data; | |
9b73896a | 352 | if (likely(psock)) |
96a59083 | 353 | strp_data_ready(&psock->strp); |
ab7ac4eb | 354 | |
ab7ac4eb TH |
355 | read_unlock_bh(&sk->sk_callback_lock); |
356 | } | |
357 | ||
9b73896a TH |
358 | /* Called with lower sock held */ |
359 | static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb) | |
ab7ac4eb | 360 | { |
9b73896a TH |
361 | struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp); |
362 | struct kcm_sock *kcm; | |
ab7ac4eb | 363 | |
9b73896a TH |
364 | try_queue: |
365 | kcm = reserve_rx_kcm(psock, skb); | |
366 | if (!kcm) { | |
367 | /* Unable to reserve a KCM, message is held in psock and strp | |
368 | * is paused. | |
369 | */ | |
370 | return; | |
371 | } | |
ab7ac4eb | 372 | |
9b73896a TH |
373 | if (kcm_queue_rcv_skb(&kcm->sk, skb)) { |
374 | /* Should mean socket buffer full */ | |
375 | unreserve_rx_kcm(psock, false); | |
376 | goto try_queue; | |
377 | } | |
ab7ac4eb TH |
378 | } |
379 | ||
9b73896a | 380 | static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb) |
ab7ac4eb | 381 | { |
9b73896a TH |
382 | struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp); |
383 | struct bpf_prog *prog = psock->bpf_prog; | |
384 | ||
385 | return (*prog->bpf_func)(skb, prog->insnsi); | |
ab7ac4eb TH |
386 | } |
387 | ||
9b73896a | 388 | static int kcm_read_sock_done(struct strparser *strp, int err) |
ab7ac4eb | 389 | { |
9b73896a TH |
390 | struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp); |
391 | ||
392 | unreserve_rx_kcm(psock, true); | |
393 | ||
394 | return err; | |
ab7ac4eb TH |
395 | } |
396 | ||
96a59083 | 397 | static void psock_state_change(struct sock *sk) |
ab7ac4eb TH |
398 | { |
399 | /* TCP only does a POLLIN for a half close. Do a POLLHUP here | |
400 | * since application will normally not poll with POLLIN | |
401 | * on the TCP sockets. | |
402 | */ | |
403 | ||
404 | report_csk_error(sk, EPIPE); | |
405 | } | |
406 | ||
96a59083 | 407 | static void psock_write_space(struct sock *sk) |
ab7ac4eb TH |
408 | { |
409 | struct kcm_psock *psock; | |
410 | struct kcm_mux *mux; | |
411 | struct kcm_sock *kcm; | |
412 | ||
413 | read_lock_bh(&sk->sk_callback_lock); | |
414 | ||
415 | psock = (struct kcm_psock *)sk->sk_user_data; | |
416 | if (unlikely(!psock)) | |
417 | goto out; | |
ab7ac4eb TH |
418 | mux = psock->mux; |
419 | ||
420 | spin_lock_bh(&mux->lock); | |
421 | ||
422 | /* Check if the socket is reserved so someone is waiting for sending. */ | |
423 | kcm = psock->tx_kcm; | |
9b73896a | 424 | if (kcm && !unlikely(kcm->tx_stopped)) |
ab7ac4eb TH |
425 | queue_work(kcm_wq, &kcm->tx_work); |
426 | ||
427 | spin_unlock_bh(&mux->lock); | |
428 | out: | |
429 | read_unlock_bh(&sk->sk_callback_lock); | |
430 | } | |
431 | ||
432 | static void unreserve_psock(struct kcm_sock *kcm); | |
433 | ||
434 | /* kcm sock is locked. */ | |
435 | static struct kcm_psock *reserve_psock(struct kcm_sock *kcm) | |
436 | { | |
437 | struct kcm_mux *mux = kcm->mux; | |
438 | struct kcm_psock *psock; | |
439 | ||
440 | psock = kcm->tx_psock; | |
441 | ||
442 | smp_rmb(); /* Must read tx_psock before tx_wait */ | |
443 | ||
444 | if (psock) { | |
445 | WARN_ON(kcm->tx_wait); | |
446 | if (unlikely(psock->tx_stopped)) | |
447 | unreserve_psock(kcm); | |
448 | else | |
449 | return kcm->tx_psock; | |
450 | } | |
451 | ||
452 | spin_lock_bh(&mux->lock); | |
453 | ||
454 | /* Check again under lock to see if psock was reserved for this | |
455 | * psock via psock_unreserve. | |
456 | */ | |
457 | psock = kcm->tx_psock; | |
458 | if (unlikely(psock)) { | |
459 | WARN_ON(kcm->tx_wait); | |
460 | spin_unlock_bh(&mux->lock); | |
461 | return kcm->tx_psock; | |
462 | } | |
463 | ||
464 | if (!list_empty(&mux->psocks_avail)) { | |
465 | psock = list_first_entry(&mux->psocks_avail, | |
466 | struct kcm_psock, | |
467 | psock_avail_list); | |
468 | list_del(&psock->psock_avail_list); | |
469 | if (kcm->tx_wait) { | |
470 | list_del(&kcm->wait_psock_list); | |
471 | kcm->tx_wait = false; | |
472 | } | |
473 | kcm->tx_psock = psock; | |
474 | psock->tx_kcm = kcm; | |
cd6e111b | 475 | KCM_STATS_INCR(psock->stats.reserved); |
ab7ac4eb TH |
476 | } else if (!kcm->tx_wait) { |
477 | list_add_tail(&kcm->wait_psock_list, | |
478 | &mux->kcm_tx_waiters); | |
479 | kcm->tx_wait = true; | |
480 | } | |
481 | ||
482 | spin_unlock_bh(&mux->lock); | |
483 | ||
484 | return psock; | |
485 | } | |
486 | ||
487 | /* mux lock held */ | |
488 | static void psock_now_avail(struct kcm_psock *psock) | |
489 | { | |
490 | struct kcm_mux *mux = psock->mux; | |
491 | struct kcm_sock *kcm; | |
492 | ||
493 | if (list_empty(&mux->kcm_tx_waiters)) { | |
494 | list_add_tail(&psock->psock_avail_list, | |
495 | &mux->psocks_avail); | |
496 | } else { | |
497 | kcm = list_first_entry(&mux->kcm_tx_waiters, | |
498 | struct kcm_sock, | |
499 | wait_psock_list); | |
500 | list_del(&kcm->wait_psock_list); | |
501 | kcm->tx_wait = false; | |
502 | psock->tx_kcm = kcm; | |
503 | ||
504 | /* Commit before changing tx_psock since that is read in | |
505 | * reserve_psock before queuing work. | |
506 | */ | |
507 | smp_mb(); | |
508 | ||
509 | kcm->tx_psock = psock; | |
cd6e111b | 510 | KCM_STATS_INCR(psock->stats.reserved); |
ab7ac4eb TH |
511 | queue_work(kcm_wq, &kcm->tx_work); |
512 | } | |
513 | } | |
514 | ||
515 | /* kcm sock is locked. */ | |
516 | static void unreserve_psock(struct kcm_sock *kcm) | |
517 | { | |
518 | struct kcm_psock *psock; | |
519 | struct kcm_mux *mux = kcm->mux; | |
520 | ||
521 | spin_lock_bh(&mux->lock); | |
522 | ||
523 | psock = kcm->tx_psock; | |
524 | ||
525 | if (WARN_ON(!psock)) { | |
526 | spin_unlock_bh(&mux->lock); | |
527 | return; | |
528 | } | |
529 | ||
530 | smp_rmb(); /* Read tx_psock before tx_wait */ | |
531 | ||
cd6e111b TH |
532 | kcm_update_tx_mux_stats(mux, psock); |
533 | ||
ab7ac4eb TH |
534 | WARN_ON(kcm->tx_wait); |
535 | ||
536 | kcm->tx_psock = NULL; | |
537 | psock->tx_kcm = NULL; | |
cd6e111b | 538 | KCM_STATS_INCR(psock->stats.unreserved); |
ab7ac4eb TH |
539 | |
540 | if (unlikely(psock->tx_stopped)) { | |
541 | if (psock->done) { | |
542 | /* Deferred free */ | |
543 | list_del(&psock->psock_list); | |
544 | mux->psocks_cnt--; | |
545 | sock_put(psock->sk); | |
546 | fput(psock->sk->sk_socket->file); | |
547 | kmem_cache_free(kcm_psockp, psock); | |
548 | } | |
549 | ||
550 | /* Don't put back on available list */ | |
551 | ||
552 | spin_unlock_bh(&mux->lock); | |
553 | ||
554 | return; | |
555 | } | |
556 | ||
557 | psock_now_avail(psock); | |
558 | ||
559 | spin_unlock_bh(&mux->lock); | |
560 | } | |
561 | ||
cd6e111b TH |
562 | static void kcm_report_tx_retry(struct kcm_sock *kcm) |
563 | { | |
564 | struct kcm_mux *mux = kcm->mux; | |
565 | ||
566 | spin_lock_bh(&mux->lock); | |
567 | KCM_STATS_INCR(mux->stats.tx_retries); | |
568 | spin_unlock_bh(&mux->lock); | |
569 | } | |
570 | ||
ab7ac4eb TH |
571 | /* Write any messages ready on the kcm socket. Called with kcm sock lock |
572 | * held. Return bytes actually sent or error. | |
573 | */ | |
574 | static int kcm_write_msgs(struct kcm_sock *kcm) | |
575 | { | |
576 | struct sock *sk = &kcm->sk; | |
577 | struct kcm_psock *psock; | |
578 | struct sk_buff *skb, *head; | |
579 | struct kcm_tx_msg *txm; | |
580 | unsigned short fragidx, frag_offset; | |
581 | unsigned int sent, total_sent = 0; | |
582 | int ret = 0; | |
583 | ||
584 | kcm->tx_wait_more = false; | |
585 | psock = kcm->tx_psock; | |
586 | if (unlikely(psock && psock->tx_stopped)) { | |
587 | /* A reserved psock was aborted asynchronously. Unreserve | |
588 | * it and we'll retry the message. | |
589 | */ | |
590 | unreserve_psock(kcm); | |
cd6e111b | 591 | kcm_report_tx_retry(kcm); |
ab7ac4eb TH |
592 | if (skb_queue_empty(&sk->sk_write_queue)) |
593 | return 0; | |
594 | ||
595 | kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0; | |
596 | ||
597 | } else if (skb_queue_empty(&sk->sk_write_queue)) { | |
598 | return 0; | |
599 | } | |
600 | ||
601 | head = skb_peek(&sk->sk_write_queue); | |
602 | txm = kcm_tx_msg(head); | |
603 | ||
604 | if (txm->sent) { | |
605 | /* Send of first skbuff in queue already in progress */ | |
606 | if (WARN_ON(!psock)) { | |
607 | ret = -EINVAL; | |
608 | goto out; | |
609 | } | |
610 | sent = txm->sent; | |
611 | frag_offset = txm->frag_offset; | |
612 | fragidx = txm->fragidx; | |
613 | skb = txm->frag_skb; | |
614 | ||
615 | goto do_frag; | |
616 | } | |
617 | ||
618 | try_again: | |
619 | psock = reserve_psock(kcm); | |
620 | if (!psock) | |
621 | goto out; | |
622 | ||
623 | do { | |
624 | skb = head; | |
625 | txm = kcm_tx_msg(head); | |
626 | sent = 0; | |
627 | ||
628 | do_frag_list: | |
629 | if (WARN_ON(!skb_shinfo(skb)->nr_frags)) { | |
630 | ret = -EINVAL; | |
631 | goto out; | |
632 | } | |
633 | ||
634 | for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; | |
635 | fragidx++) { | |
636 | skb_frag_t *frag; | |
637 | ||
638 | frag_offset = 0; | |
639 | do_frag: | |
640 | frag = &skb_shinfo(skb)->frags[fragidx]; | |
641 | if (WARN_ON(!frag->size)) { | |
642 | ret = -EINVAL; | |
643 | goto out; | |
644 | } | |
645 | ||
646 | ret = kernel_sendpage(psock->sk->sk_socket, | |
647 | frag->page.p, | |
648 | frag->page_offset + frag_offset, | |
649 | frag->size - frag_offset, | |
650 | MSG_DONTWAIT); | |
651 | if (ret <= 0) { | |
652 | if (ret == -EAGAIN) { | |
653 | /* Save state to try again when there's | |
654 | * write space on the socket | |
655 | */ | |
656 | txm->sent = sent; | |
657 | txm->frag_offset = frag_offset; | |
658 | txm->fragidx = fragidx; | |
659 | txm->frag_skb = skb; | |
660 | ||
661 | ret = 0; | |
662 | goto out; | |
663 | } | |
664 | ||
665 | /* Hard failure in sending message, abort this | |
666 | * psock since it has lost framing | |
667 | * synchonization and retry sending the | |
668 | * message from the beginning. | |
669 | */ | |
670 | kcm_abort_tx_psock(psock, ret ? -ret : EPIPE, | |
671 | true); | |
672 | unreserve_psock(kcm); | |
673 | ||
674 | txm->sent = 0; | |
cd6e111b | 675 | kcm_report_tx_retry(kcm); |
ab7ac4eb TH |
676 | ret = 0; |
677 | ||
678 | goto try_again; | |
679 | } | |
680 | ||
681 | sent += ret; | |
682 | frag_offset += ret; | |
cd6e111b | 683 | KCM_STATS_ADD(psock->stats.tx_bytes, ret); |
ab7ac4eb TH |
684 | if (frag_offset < frag->size) { |
685 | /* Not finished with this frag */ | |
686 | goto do_frag; | |
687 | } | |
688 | } | |
689 | ||
690 | if (skb == head) { | |
691 | if (skb_has_frag_list(skb)) { | |
692 | skb = skb_shinfo(skb)->frag_list; | |
693 | goto do_frag_list; | |
694 | } | |
695 | } else if (skb->next) { | |
696 | skb = skb->next; | |
697 | goto do_frag_list; | |
698 | } | |
699 | ||
700 | /* Successfully sent the whole packet, account for it. */ | |
701 | skb_dequeue(&sk->sk_write_queue); | |
702 | kfree_skb(head); | |
703 | sk->sk_wmem_queued -= sent; | |
704 | total_sent += sent; | |
cd6e111b | 705 | KCM_STATS_INCR(psock->stats.tx_msgs); |
ab7ac4eb TH |
706 | } while ((head = skb_peek(&sk->sk_write_queue))); |
707 | out: | |
708 | if (!head) { | |
709 | /* Done with all queued messages. */ | |
710 | WARN_ON(!skb_queue_empty(&sk->sk_write_queue)); | |
711 | unreserve_psock(kcm); | |
712 | } | |
713 | ||
714 | /* Check if write space is available */ | |
715 | sk->sk_write_space(sk); | |
716 | ||
717 | return total_sent ? : ret; | |
718 | } | |
719 | ||
720 | static void kcm_tx_work(struct work_struct *w) | |
721 | { | |
722 | struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work); | |
723 | struct sock *sk = &kcm->sk; | |
724 | int err; | |
725 | ||
726 | lock_sock(sk); | |
727 | ||
728 | /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx | |
729 | * aborts | |
730 | */ | |
731 | err = kcm_write_msgs(kcm); | |
732 | if (err < 0) { | |
733 | /* Hard failure in write, report error on KCM socket */ | |
734 | pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err); | |
735 | report_csk_error(&kcm->sk, -err); | |
736 | goto out; | |
737 | } | |
738 | ||
739 | /* Primarily for SOCK_SEQPACKET sockets */ | |
740 | if (likely(sk->sk_socket) && | |
741 | test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
742 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
743 | sk->sk_write_space(sk); | |
744 | } | |
745 | ||
746 | out: | |
747 | release_sock(sk); | |
748 | } | |
749 | ||
750 | static void kcm_push(struct kcm_sock *kcm) | |
751 | { | |
752 | if (kcm->tx_wait_more) | |
753 | kcm_write_msgs(kcm); | |
754 | } | |
755 | ||
f29698fc TH |
756 | static ssize_t kcm_sendpage(struct socket *sock, struct page *page, |
757 | int offset, size_t size, int flags) | |
758 | ||
759 | { | |
760 | struct sock *sk = sock->sk; | |
761 | struct kcm_sock *kcm = kcm_sk(sk); | |
762 | struct sk_buff *skb = NULL, *head = NULL; | |
763 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
764 | bool eor; | |
765 | int err = 0; | |
766 | int i; | |
767 | ||
768 | if (flags & MSG_SENDPAGE_NOTLAST) | |
769 | flags |= MSG_MORE; | |
770 | ||
771 | /* No MSG_EOR from splice, only look at MSG_MORE */ | |
772 | eor = !(flags & MSG_MORE); | |
773 | ||
774 | lock_sock(sk); | |
775 | ||
776 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); | |
777 | ||
778 | err = -EPIPE; | |
779 | if (sk->sk_err) | |
780 | goto out_error; | |
781 | ||
782 | if (kcm->seq_skb) { | |
783 | /* Previously opened message */ | |
784 | head = kcm->seq_skb; | |
785 | skb = kcm_tx_msg(head)->last_skb; | |
786 | i = skb_shinfo(skb)->nr_frags; | |
787 | ||
788 | if (skb_can_coalesce(skb, i, page, offset)) { | |
789 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); | |
790 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; | |
791 | goto coalesced; | |
792 | } | |
793 | ||
794 | if (i >= MAX_SKB_FRAGS) { | |
795 | struct sk_buff *tskb; | |
796 | ||
797 | tskb = alloc_skb(0, sk->sk_allocation); | |
798 | while (!tskb) { | |
799 | kcm_push(kcm); | |
800 | err = sk_stream_wait_memory(sk, &timeo); | |
801 | if (err) | |
802 | goto out_error; | |
803 | } | |
804 | ||
805 | if (head == skb) | |
806 | skb_shinfo(head)->frag_list = tskb; | |
807 | else | |
808 | skb->next = tskb; | |
809 | ||
810 | skb = tskb; | |
811 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
812 | i = 0; | |
813 | } | |
814 | } else { | |
815 | /* Call the sk_stream functions to manage the sndbuf mem. */ | |
816 | if (!sk_stream_memory_free(sk)) { | |
817 | kcm_push(kcm); | |
818 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
819 | err = sk_stream_wait_memory(sk, &timeo); | |
820 | if (err) | |
821 | goto out_error; | |
822 | } | |
823 | ||
824 | head = alloc_skb(0, sk->sk_allocation); | |
825 | while (!head) { | |
826 | kcm_push(kcm); | |
827 | err = sk_stream_wait_memory(sk, &timeo); | |
828 | if (err) | |
829 | goto out_error; | |
830 | } | |
831 | ||
832 | skb = head; | |
833 | i = 0; | |
834 | } | |
835 | ||
836 | get_page(page); | |
837 | skb_fill_page_desc(skb, i, page, offset, size); | |
838 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; | |
839 | ||
840 | coalesced: | |
841 | skb->len += size; | |
842 | skb->data_len += size; | |
843 | skb->truesize += size; | |
844 | sk->sk_wmem_queued += size; | |
845 | sk_mem_charge(sk, size); | |
846 | ||
847 | if (head != skb) { | |
848 | head->len += size; | |
849 | head->data_len += size; | |
850 | head->truesize += size; | |
851 | } | |
852 | ||
853 | if (eor) { | |
854 | bool not_busy = skb_queue_empty(&sk->sk_write_queue); | |
855 | ||
856 | /* Message complete, queue it on send buffer */ | |
857 | __skb_queue_tail(&sk->sk_write_queue, head); | |
858 | kcm->seq_skb = NULL; | |
859 | KCM_STATS_INCR(kcm->stats.tx_msgs); | |
860 | ||
861 | if (flags & MSG_BATCH) { | |
862 | kcm->tx_wait_more = true; | |
863 | } else if (kcm->tx_wait_more || not_busy) { | |
864 | err = kcm_write_msgs(kcm); | |
865 | if (err < 0) { | |
866 | /* We got a hard error in write_msgs but have | |
867 | * already queued this message. Report an error | |
868 | * in the socket, but don't affect return value | |
869 | * from sendmsg | |
870 | */ | |
871 | pr_warn("KCM: Hard failure on kcm_write_msgs\n"); | |
872 | report_csk_error(&kcm->sk, -err); | |
873 | } | |
874 | } | |
875 | } else { | |
876 | /* Message not complete, save state */ | |
877 | kcm->seq_skb = head; | |
878 | kcm_tx_msg(head)->last_skb = skb; | |
879 | } | |
880 | ||
881 | KCM_STATS_ADD(kcm->stats.tx_bytes, size); | |
882 | ||
883 | release_sock(sk); | |
884 | return size; | |
885 | ||
886 | out_error: | |
887 | kcm_push(kcm); | |
888 | ||
889 | err = sk_stream_error(sk, flags, err); | |
890 | ||
891 | /* make sure we wake any epoll edge trigger waiter */ | |
892 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN)) | |
893 | sk->sk_write_space(sk); | |
894 | ||
895 | release_sock(sk); | |
896 | return err; | |
897 | } | |
898 | ||
ab7ac4eb TH |
899 | static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) |
900 | { | |
901 | struct sock *sk = sock->sk; | |
902 | struct kcm_sock *kcm = kcm_sk(sk); | |
903 | struct sk_buff *skb = NULL, *head = NULL; | |
904 | size_t copy, copied = 0; | |
905 | long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); | |
906 | int eor = (sock->type == SOCK_DGRAM) ? | |
907 | !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR); | |
908 | int err = -EPIPE; | |
909 | ||
910 | lock_sock(sk); | |
911 | ||
912 | /* Per tcp_sendmsg this should be in poll */ | |
913 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); | |
914 | ||
915 | if (sk->sk_err) | |
916 | goto out_error; | |
917 | ||
918 | if (kcm->seq_skb) { | |
919 | /* Previously opened message */ | |
920 | head = kcm->seq_skb; | |
921 | skb = kcm_tx_msg(head)->last_skb; | |
922 | goto start; | |
923 | } | |
924 | ||
925 | /* Call the sk_stream functions to manage the sndbuf mem. */ | |
926 | if (!sk_stream_memory_free(sk)) { | |
927 | kcm_push(kcm); | |
928 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
929 | err = sk_stream_wait_memory(sk, &timeo); | |
930 | if (err) | |
931 | goto out_error; | |
932 | } | |
933 | ||
98e3862c WC |
934 | if (msg_data_left(msg)) { |
935 | /* New message, alloc head skb */ | |
ab7ac4eb | 936 | head = alloc_skb(0, sk->sk_allocation); |
98e3862c WC |
937 | while (!head) { |
938 | kcm_push(kcm); | |
939 | err = sk_stream_wait_memory(sk, &timeo); | |
940 | if (err) | |
941 | goto out_error; | |
ab7ac4eb | 942 | |
98e3862c WC |
943 | head = alloc_skb(0, sk->sk_allocation); |
944 | } | |
ab7ac4eb | 945 | |
98e3862c WC |
946 | skb = head; |
947 | ||
948 | /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling | |
949 | * csum_and_copy_from_iter from skb_do_copy_data_nocache. | |
950 | */ | |
951 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
952 | } | |
ab7ac4eb TH |
953 | |
954 | start: | |
955 | while (msg_data_left(msg)) { | |
956 | bool merge = true; | |
957 | int i = skb_shinfo(skb)->nr_frags; | |
958 | struct page_frag *pfrag = sk_page_frag(sk); | |
959 | ||
960 | if (!sk_page_frag_refill(sk, pfrag)) | |
961 | goto wait_for_memory; | |
962 | ||
963 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
964 | pfrag->offset)) { | |
965 | if (i == MAX_SKB_FRAGS) { | |
966 | struct sk_buff *tskb; | |
967 | ||
968 | tskb = alloc_skb(0, sk->sk_allocation); | |
969 | if (!tskb) | |
970 | goto wait_for_memory; | |
971 | ||
972 | if (head == skb) | |
973 | skb_shinfo(head)->frag_list = tskb; | |
974 | else | |
975 | skb->next = tskb; | |
976 | ||
977 | skb = tskb; | |
978 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
979 | continue; | |
980 | } | |
981 | merge = false; | |
982 | } | |
983 | ||
984 | copy = min_t(int, msg_data_left(msg), | |
985 | pfrag->size - pfrag->offset); | |
986 | ||
987 | if (!sk_wmem_schedule(sk, copy)) | |
988 | goto wait_for_memory; | |
989 | ||
990 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, | |
991 | pfrag->page, | |
992 | pfrag->offset, | |
993 | copy); | |
994 | if (err) | |
995 | goto out_error; | |
996 | ||
997 | /* Update the skb. */ | |
998 | if (merge) { | |
999 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1000 | } else { | |
1001 | skb_fill_page_desc(skb, i, pfrag->page, | |
1002 | pfrag->offset, copy); | |
1003 | get_page(pfrag->page); | |
1004 | } | |
1005 | ||
1006 | pfrag->offset += copy; | |
1007 | copied += copy; | |
1008 | if (head != skb) { | |
1009 | head->len += copy; | |
1010 | head->data_len += copy; | |
1011 | } | |
1012 | ||
1013 | continue; | |
1014 | ||
1015 | wait_for_memory: | |
1016 | kcm_push(kcm); | |
1017 | err = sk_stream_wait_memory(sk, &timeo); | |
1018 | if (err) | |
1019 | goto out_error; | |
1020 | } | |
1021 | ||
1022 | if (eor) { | |
1023 | bool not_busy = skb_queue_empty(&sk->sk_write_queue); | |
1024 | ||
98e3862c WC |
1025 | if (head) { |
1026 | /* Message complete, queue it on send buffer */ | |
1027 | __skb_queue_tail(&sk->sk_write_queue, head); | |
1028 | kcm->seq_skb = NULL; | |
1029 | KCM_STATS_INCR(kcm->stats.tx_msgs); | |
1030 | } | |
ab7ac4eb TH |
1031 | |
1032 | if (msg->msg_flags & MSG_BATCH) { | |
1033 | kcm->tx_wait_more = true; | |
1034 | } else if (kcm->tx_wait_more || not_busy) { | |
1035 | err = kcm_write_msgs(kcm); | |
1036 | if (err < 0) { | |
1037 | /* We got a hard error in write_msgs but have | |
1038 | * already queued this message. Report an error | |
1039 | * in the socket, but don't affect return value | |
1040 | * from sendmsg | |
1041 | */ | |
1042 | pr_warn("KCM: Hard failure on kcm_write_msgs\n"); | |
1043 | report_csk_error(&kcm->sk, -err); | |
1044 | } | |
1045 | } | |
1046 | } else { | |
1047 | /* Message not complete, save state */ | |
1048 | partial_message: | |
cd27b96b WC |
1049 | if (head) { |
1050 | kcm->seq_skb = head; | |
1051 | kcm_tx_msg(head)->last_skb = skb; | |
1052 | } | |
ab7ac4eb TH |
1053 | } |
1054 | ||
cd6e111b TH |
1055 | KCM_STATS_ADD(kcm->stats.tx_bytes, copied); |
1056 | ||
ab7ac4eb TH |
1057 | release_sock(sk); |
1058 | return copied; | |
1059 | ||
1060 | out_error: | |
1061 | kcm_push(kcm); | |
1062 | ||
1063 | if (copied && sock->type == SOCK_SEQPACKET) { | |
1064 | /* Wrote some bytes before encountering an | |
1065 | * error, return partial success. | |
1066 | */ | |
1067 | goto partial_message; | |
1068 | } | |
1069 | ||
1070 | if (head != kcm->seq_skb) | |
1071 | kfree_skb(head); | |
1072 | ||
1073 | err = sk_stream_error(sk, msg->msg_flags, err); | |
1074 | ||
1075 | /* make sure we wake any epoll edge trigger waiter */ | |
1076 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN)) | |
1077 | sk->sk_write_space(sk); | |
1078 | ||
1079 | release_sock(sk); | |
1080 | return err; | |
1081 | } | |
1082 | ||
1083 | static struct sk_buff *kcm_wait_data(struct sock *sk, int flags, | |
1084 | long timeo, int *err) | |
1085 | { | |
1086 | struct sk_buff *skb; | |
1087 | ||
1088 | while (!(skb = skb_peek(&sk->sk_receive_queue))) { | |
1089 | if (sk->sk_err) { | |
1090 | *err = sock_error(sk); | |
1091 | return NULL; | |
1092 | } | |
1093 | ||
1094 | if (sock_flag(sk, SOCK_DONE)) | |
1095 | return NULL; | |
1096 | ||
1097 | if ((flags & MSG_DONTWAIT) || !timeo) { | |
1098 | *err = -EAGAIN; | |
1099 | return NULL; | |
1100 | } | |
1101 | ||
1102 | sk_wait_data(sk, &timeo, NULL); | |
1103 | ||
1104 | /* Handle signals */ | |
1105 | if (signal_pending(current)) { | |
1106 | *err = sock_intr_errno(timeo); | |
1107 | return NULL; | |
1108 | } | |
1109 | } | |
1110 | ||
1111 | return skb; | |
1112 | } | |
1113 | ||
1114 | static int kcm_recvmsg(struct socket *sock, struct msghdr *msg, | |
1115 | size_t len, int flags) | |
1116 | { | |
1117 | struct sock *sk = sock->sk; | |
cd6e111b | 1118 | struct kcm_sock *kcm = kcm_sk(sk); |
ab7ac4eb TH |
1119 | int err = 0; |
1120 | long timeo; | |
9b73896a | 1121 | struct strp_rx_msg *rxm; |
ab7ac4eb TH |
1122 | int copied = 0; |
1123 | struct sk_buff *skb; | |
1124 | ||
1125 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); | |
1126 | ||
1127 | lock_sock(sk); | |
1128 | ||
1129 | skb = kcm_wait_data(sk, flags, timeo, &err); | |
1130 | if (!skb) | |
1131 | goto out; | |
1132 | ||
1133 | /* Okay, have a message on the receive queue */ | |
1134 | ||
9b73896a | 1135 | rxm = strp_rx_msg(skb); |
ab7ac4eb TH |
1136 | |
1137 | if (len > rxm->full_len) | |
1138 | len = rxm->full_len; | |
1139 | ||
1140 | err = skb_copy_datagram_msg(skb, rxm->offset, msg, len); | |
1141 | if (err < 0) | |
1142 | goto out; | |
1143 | ||
1144 | copied = len; | |
1145 | if (likely(!(flags & MSG_PEEK))) { | |
cd6e111b | 1146 | KCM_STATS_ADD(kcm->stats.rx_bytes, copied); |
ab7ac4eb TH |
1147 | if (copied < rxm->full_len) { |
1148 | if (sock->type == SOCK_DGRAM) { | |
1149 | /* Truncated message */ | |
1150 | msg->msg_flags |= MSG_TRUNC; | |
1151 | goto msg_finished; | |
1152 | } | |
1153 | rxm->offset += copied; | |
1154 | rxm->full_len -= copied; | |
1155 | } else { | |
1156 | msg_finished: | |
1157 | /* Finished with message */ | |
1158 | msg->msg_flags |= MSG_EOR; | |
cd6e111b | 1159 | KCM_STATS_INCR(kcm->stats.rx_msgs); |
ab7ac4eb TH |
1160 | skb_unlink(skb, &sk->sk_receive_queue); |
1161 | kfree_skb(skb); | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | out: | |
1166 | release_sock(sk); | |
1167 | ||
1168 | return copied ? : err; | |
1169 | } | |
1170 | ||
91687355 TH |
1171 | static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos, |
1172 | struct pipe_inode_info *pipe, size_t len, | |
1173 | unsigned int flags) | |
1174 | { | |
1175 | struct sock *sk = sock->sk; | |
1176 | struct kcm_sock *kcm = kcm_sk(sk); | |
1177 | long timeo; | |
9b73896a | 1178 | struct strp_rx_msg *rxm; |
91687355 | 1179 | int err = 0; |
f1971a2e | 1180 | ssize_t copied; |
91687355 TH |
1181 | struct sk_buff *skb; |
1182 | ||
1183 | /* Only support splice for SOCKSEQPACKET */ | |
1184 | ||
1185 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); | |
1186 | ||
1187 | lock_sock(sk); | |
1188 | ||
1189 | skb = kcm_wait_data(sk, flags, timeo, &err); | |
1190 | if (!skb) | |
1191 | goto err_out; | |
1192 | ||
1193 | /* Okay, have a message on the receive queue */ | |
1194 | ||
9b73896a | 1195 | rxm = strp_rx_msg(skb); |
91687355 TH |
1196 | |
1197 | if (len > rxm->full_len) | |
1198 | len = rxm->full_len; | |
1199 | ||
25869262 | 1200 | copied = skb_splice_bits(skb, sk, rxm->offset, pipe, len, flags); |
91687355 TH |
1201 | if (copied < 0) { |
1202 | err = copied; | |
1203 | goto err_out; | |
1204 | } | |
1205 | ||
1206 | KCM_STATS_ADD(kcm->stats.rx_bytes, copied); | |
1207 | ||
1208 | rxm->offset += copied; | |
1209 | rxm->full_len -= copied; | |
1210 | ||
1211 | /* We have no way to return MSG_EOR. If all the bytes have been | |
1212 | * read we still leave the message in the receive socket buffer. | |
1213 | * A subsequent recvmsg needs to be done to return MSG_EOR and | |
1214 | * finish reading the message. | |
1215 | */ | |
1216 | ||
1217 | release_sock(sk); | |
1218 | ||
1219 | return copied; | |
1220 | ||
1221 | err_out: | |
1222 | release_sock(sk); | |
1223 | ||
1224 | return err; | |
1225 | } | |
1226 | ||
ab7ac4eb TH |
1227 | /* kcm sock lock held */ |
1228 | static void kcm_recv_disable(struct kcm_sock *kcm) | |
1229 | { | |
1230 | struct kcm_mux *mux = kcm->mux; | |
1231 | ||
1232 | if (kcm->rx_disabled) | |
1233 | return; | |
1234 | ||
1235 | spin_lock_bh(&mux->rx_lock); | |
1236 | ||
1237 | kcm->rx_disabled = 1; | |
1238 | ||
1239 | /* If a psock is reserved we'll do cleanup in unreserve */ | |
1240 | if (!kcm->rx_psock) { | |
1241 | if (kcm->rx_wait) { | |
1242 | list_del(&kcm->wait_rx_list); | |
1243 | kcm->rx_wait = false; | |
1244 | } | |
1245 | ||
1246 | requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue); | |
1247 | } | |
1248 | ||
1249 | spin_unlock_bh(&mux->rx_lock); | |
1250 | } | |
1251 | ||
1252 | /* kcm sock lock held */ | |
1253 | static void kcm_recv_enable(struct kcm_sock *kcm) | |
1254 | { | |
1255 | struct kcm_mux *mux = kcm->mux; | |
1256 | ||
1257 | if (!kcm->rx_disabled) | |
1258 | return; | |
1259 | ||
1260 | spin_lock_bh(&mux->rx_lock); | |
1261 | ||
1262 | kcm->rx_disabled = 0; | |
1263 | kcm_rcv_ready(kcm); | |
1264 | ||
1265 | spin_unlock_bh(&mux->rx_lock); | |
1266 | } | |
1267 | ||
1268 | static int kcm_setsockopt(struct socket *sock, int level, int optname, | |
1269 | char __user *optval, unsigned int optlen) | |
1270 | { | |
1271 | struct kcm_sock *kcm = kcm_sk(sock->sk); | |
1272 | int val, valbool; | |
1273 | int err = 0; | |
1274 | ||
1275 | if (level != SOL_KCM) | |
1276 | return -ENOPROTOOPT; | |
1277 | ||
1278 | if (optlen < sizeof(int)) | |
1279 | return -EINVAL; | |
1280 | ||
1281 | if (get_user(val, (int __user *)optval)) | |
1282 | return -EINVAL; | |
1283 | ||
1284 | valbool = val ? 1 : 0; | |
1285 | ||
1286 | switch (optname) { | |
1287 | case KCM_RECV_DISABLE: | |
1288 | lock_sock(&kcm->sk); | |
1289 | if (valbool) | |
1290 | kcm_recv_disable(kcm); | |
1291 | else | |
1292 | kcm_recv_enable(kcm); | |
1293 | release_sock(&kcm->sk); | |
1294 | break; | |
1295 | default: | |
1296 | err = -ENOPROTOOPT; | |
1297 | } | |
1298 | ||
1299 | return err; | |
1300 | } | |
1301 | ||
1302 | static int kcm_getsockopt(struct socket *sock, int level, int optname, | |
1303 | char __user *optval, int __user *optlen) | |
1304 | { | |
1305 | struct kcm_sock *kcm = kcm_sk(sock->sk); | |
1306 | int val, len; | |
1307 | ||
1308 | if (level != SOL_KCM) | |
1309 | return -ENOPROTOOPT; | |
1310 | ||
1311 | if (get_user(len, optlen)) | |
1312 | return -EFAULT; | |
1313 | ||
1314 | len = min_t(unsigned int, len, sizeof(int)); | |
1315 | if (len < 0) | |
1316 | return -EINVAL; | |
1317 | ||
1318 | switch (optname) { | |
1319 | case KCM_RECV_DISABLE: | |
1320 | val = kcm->rx_disabled; | |
1321 | break; | |
1322 | default: | |
1323 | return -ENOPROTOOPT; | |
1324 | } | |
1325 | ||
1326 | if (put_user(len, optlen)) | |
1327 | return -EFAULT; | |
1328 | if (copy_to_user(optval, &val, len)) | |
1329 | return -EFAULT; | |
1330 | return 0; | |
1331 | } | |
1332 | ||
1333 | static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux) | |
1334 | { | |
1335 | struct kcm_sock *tkcm; | |
1336 | struct list_head *head; | |
1337 | int index = 0; | |
1338 | ||
1339 | /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so | |
1340 | * we set sk_state, otherwise epoll_wait always returns right away with | |
1341 | * POLLHUP | |
1342 | */ | |
1343 | kcm->sk.sk_state = TCP_ESTABLISHED; | |
1344 | ||
1345 | /* Add to mux's kcm sockets list */ | |
1346 | kcm->mux = mux; | |
1347 | spin_lock_bh(&mux->lock); | |
1348 | ||
1349 | head = &mux->kcm_socks; | |
1350 | list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) { | |
1351 | if (tkcm->index != index) | |
1352 | break; | |
1353 | head = &tkcm->kcm_sock_list; | |
1354 | index++; | |
1355 | } | |
1356 | ||
1357 | list_add(&kcm->kcm_sock_list, head); | |
1358 | kcm->index = index; | |
1359 | ||
1360 | mux->kcm_socks_cnt++; | |
1361 | spin_unlock_bh(&mux->lock); | |
1362 | ||
1363 | INIT_WORK(&kcm->tx_work, kcm_tx_work); | |
1364 | ||
1365 | spin_lock_bh(&mux->rx_lock); | |
1366 | kcm_rcv_ready(kcm); | |
1367 | spin_unlock_bh(&mux->rx_lock); | |
1368 | } | |
1369 | ||
1370 | static int kcm_attach(struct socket *sock, struct socket *csock, | |
1371 | struct bpf_prog *prog) | |
1372 | { | |
1373 | struct kcm_sock *kcm = kcm_sk(sock->sk); | |
1374 | struct kcm_mux *mux = kcm->mux; | |
1375 | struct sock *csk; | |
1376 | struct kcm_psock *psock = NULL, *tpsock; | |
1377 | struct list_head *head; | |
1378 | int index = 0; | |
9b73896a | 1379 | struct strp_callbacks cb; |
96a59083 | 1380 | int err; |
ab7ac4eb TH |
1381 | |
1382 | csk = csock->sk; | |
1383 | if (!csk) | |
1384 | return -EINVAL; | |
1385 | ||
ab7ac4eb TH |
1386 | psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL); |
1387 | if (!psock) | |
1388 | return -ENOMEM; | |
1389 | ||
1390 | psock->mux = mux; | |
1391 | psock->sk = csk; | |
1392 | psock->bpf_prog = prog; | |
29152a34 | 1393 | |
9b73896a TH |
1394 | cb.rcv_msg = kcm_rcv_strparser; |
1395 | cb.abort_parser = NULL; | |
1396 | cb.parse_msg = kcm_parse_func_strparser; | |
1397 | cb.read_sock_done = kcm_read_sock_done; | |
29152a34 | 1398 | |
96a59083 TH |
1399 | err = strp_init(&psock->strp, csk, &cb); |
1400 | if (err) { | |
1401 | kmem_cache_free(kcm_psockp, psock); | |
1402 | return err; | |
1403 | } | |
ab7ac4eb TH |
1404 | |
1405 | sock_hold(csk); | |
1406 | ||
1407 | write_lock_bh(&csk->sk_callback_lock); | |
1408 | psock->save_data_ready = csk->sk_data_ready; | |
1409 | psock->save_write_space = csk->sk_write_space; | |
1410 | psock->save_state_change = csk->sk_state_change; | |
1411 | csk->sk_user_data = psock; | |
96a59083 TH |
1412 | csk->sk_data_ready = psock_data_ready; |
1413 | csk->sk_write_space = psock_write_space; | |
1414 | csk->sk_state_change = psock_state_change; | |
ab7ac4eb TH |
1415 | write_unlock_bh(&csk->sk_callback_lock); |
1416 | ||
1417 | /* Finished initialization, now add the psock to the MUX. */ | |
1418 | spin_lock_bh(&mux->lock); | |
1419 | head = &mux->psocks; | |
1420 | list_for_each_entry(tpsock, &mux->psocks, psock_list) { | |
1421 | if (tpsock->index != index) | |
1422 | break; | |
1423 | head = &tpsock->psock_list; | |
1424 | index++; | |
1425 | } | |
1426 | ||
1427 | list_add(&psock->psock_list, head); | |
1428 | psock->index = index; | |
1429 | ||
cd6e111b | 1430 | KCM_STATS_INCR(mux->stats.psock_attach); |
ab7ac4eb TH |
1431 | mux->psocks_cnt++; |
1432 | psock_now_avail(psock); | |
1433 | spin_unlock_bh(&mux->lock); | |
1434 | ||
1435 | /* Schedule RX work in case there are already bytes queued */ | |
9b73896a | 1436 | strp_check_rcv(&psock->strp); |
ab7ac4eb TH |
1437 | |
1438 | return 0; | |
1439 | } | |
1440 | ||
1441 | static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info) | |
1442 | { | |
1443 | struct socket *csock; | |
1444 | struct bpf_prog *prog; | |
1445 | int err; | |
1446 | ||
1447 | csock = sockfd_lookup(info->fd, &err); | |
1448 | if (!csock) | |
1449 | return -ENOENT; | |
1450 | ||
113214be | 1451 | prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER); |
ab7ac4eb TH |
1452 | if (IS_ERR(prog)) { |
1453 | err = PTR_ERR(prog); | |
1454 | goto out; | |
1455 | } | |
1456 | ||
ab7ac4eb TH |
1457 | err = kcm_attach(sock, csock, prog); |
1458 | if (err) { | |
1459 | bpf_prog_put(prog); | |
1460 | goto out; | |
1461 | } | |
1462 | ||
1463 | /* Keep reference on file also */ | |
1464 | ||
1465 | return 0; | |
1466 | out: | |
1467 | fput(csock->file); | |
1468 | return err; | |
1469 | } | |
1470 | ||
1471 | static void kcm_unattach(struct kcm_psock *psock) | |
1472 | { | |
1473 | struct sock *csk = psock->sk; | |
1474 | struct kcm_mux *mux = psock->mux; | |
1475 | ||
1616b38f TH |
1476 | lock_sock(csk); |
1477 | ||
ab7ac4eb TH |
1478 | /* Stop getting callbacks from TCP socket. After this there should |
1479 | * be no way to reserve a kcm for this psock. | |
1480 | */ | |
1481 | write_lock_bh(&csk->sk_callback_lock); | |
1482 | csk->sk_user_data = NULL; | |
1483 | csk->sk_data_ready = psock->save_data_ready; | |
1484 | csk->sk_write_space = psock->save_write_space; | |
1485 | csk->sk_state_change = psock->save_state_change; | |
9b73896a | 1486 | strp_stop(&psock->strp); |
ab7ac4eb TH |
1487 | |
1488 | if (WARN_ON(psock->rx_kcm)) { | |
1489 | write_unlock_bh(&csk->sk_callback_lock); | |
1490 | return; | |
1491 | } | |
1492 | ||
1493 | spin_lock_bh(&mux->rx_lock); | |
1494 | ||
1495 | /* Stop receiver activities. After this point psock should not be | |
1496 | * able to get onto ready list either through callbacks or work. | |
1497 | */ | |
1498 | if (psock->ready_rx_msg) { | |
1499 | list_del(&psock->psock_ready_list); | |
1500 | kfree_skb(psock->ready_rx_msg); | |
1501 | psock->ready_rx_msg = NULL; | |
cd6e111b | 1502 | KCM_STATS_INCR(mux->stats.rx_ready_drops); |
ab7ac4eb TH |
1503 | } |
1504 | ||
1505 | spin_unlock_bh(&mux->rx_lock); | |
1506 | ||
1507 | write_unlock_bh(&csk->sk_callback_lock); | |
1508 | ||
1616b38f TH |
1509 | /* Call strp_done without sock lock */ |
1510 | release_sock(csk); | |
9b73896a | 1511 | strp_done(&psock->strp); |
1616b38f | 1512 | lock_sock(csk); |
ab7ac4eb TH |
1513 | |
1514 | bpf_prog_put(psock->bpf_prog); | |
1515 | ||
ab7ac4eb TH |
1516 | spin_lock_bh(&mux->lock); |
1517 | ||
cd6e111b | 1518 | aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats); |
9b73896a | 1519 | save_strp_stats(&psock->strp, &mux->aggregate_strp_stats); |
cd6e111b TH |
1520 | |
1521 | KCM_STATS_INCR(mux->stats.psock_unattach); | |
1522 | ||
ab7ac4eb TH |
1523 | if (psock->tx_kcm) { |
1524 | /* psock was reserved. Just mark it finished and we will clean | |
1525 | * up in the kcm paths, we need kcm lock which can not be | |
1526 | * acquired here. | |
1527 | */ | |
cd6e111b | 1528 | KCM_STATS_INCR(mux->stats.psock_unattach_rsvd); |
ab7ac4eb TH |
1529 | spin_unlock_bh(&mux->lock); |
1530 | ||
1531 | /* We are unattaching a socket that is reserved. Abort the | |
1532 | * socket since we may be out of sync in sending on it. We need | |
1533 | * to do this without the mux lock. | |
1534 | */ | |
1535 | kcm_abort_tx_psock(psock, EPIPE, false); | |
1536 | ||
1537 | spin_lock_bh(&mux->lock); | |
1538 | if (!psock->tx_kcm) { | |
1539 | /* psock now unreserved in window mux was unlocked */ | |
1540 | goto no_reserved; | |
1541 | } | |
1542 | psock->done = 1; | |
1543 | ||
1544 | /* Commit done before queuing work to process it */ | |
1545 | smp_mb(); | |
1546 | ||
1547 | /* Queue tx work to make sure psock->done is handled */ | |
1548 | queue_work(kcm_wq, &psock->tx_kcm->tx_work); | |
1549 | spin_unlock_bh(&mux->lock); | |
1550 | } else { | |
1551 | no_reserved: | |
1552 | if (!psock->tx_stopped) | |
1553 | list_del(&psock->psock_avail_list); | |
1554 | list_del(&psock->psock_list); | |
1555 | mux->psocks_cnt--; | |
1556 | spin_unlock_bh(&mux->lock); | |
1557 | ||
1558 | sock_put(csk); | |
1559 | fput(csk->sk_socket->file); | |
1560 | kmem_cache_free(kcm_psockp, psock); | |
1561 | } | |
1616b38f TH |
1562 | |
1563 | release_sock(csk); | |
ab7ac4eb TH |
1564 | } |
1565 | ||
1566 | static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info) | |
1567 | { | |
1568 | struct kcm_sock *kcm = kcm_sk(sock->sk); | |
1569 | struct kcm_mux *mux = kcm->mux; | |
1570 | struct kcm_psock *psock; | |
1571 | struct socket *csock; | |
1572 | struct sock *csk; | |
1573 | int err; | |
1574 | ||
1575 | csock = sockfd_lookup(info->fd, &err); | |
1576 | if (!csock) | |
1577 | return -ENOENT; | |
1578 | ||
1579 | csk = csock->sk; | |
1580 | if (!csk) { | |
1581 | err = -EINVAL; | |
1582 | goto out; | |
1583 | } | |
1584 | ||
1585 | err = -ENOENT; | |
1586 | ||
1587 | spin_lock_bh(&mux->lock); | |
1588 | ||
1589 | list_for_each_entry(psock, &mux->psocks, psock_list) { | |
1590 | if (psock->sk != csk) | |
1591 | continue; | |
1592 | ||
1593 | /* Found the matching psock */ | |
1594 | ||
1595 | if (psock->unattaching || WARN_ON(psock->done)) { | |
1596 | err = -EALREADY; | |
1597 | break; | |
1598 | } | |
1599 | ||
1600 | psock->unattaching = 1; | |
1601 | ||
1602 | spin_unlock_bh(&mux->lock); | |
1603 | ||
9b73896a | 1604 | /* Lower socket lock should already be held */ |
ab7ac4eb TH |
1605 | kcm_unattach(psock); |
1606 | ||
1607 | err = 0; | |
1608 | goto out; | |
1609 | } | |
1610 | ||
1611 | spin_unlock_bh(&mux->lock); | |
1612 | ||
1613 | out: | |
1614 | fput(csock->file); | |
1615 | return err; | |
1616 | } | |
1617 | ||
1618 | static struct proto kcm_proto = { | |
1619 | .name = "KCM", | |
1620 | .owner = THIS_MODULE, | |
1621 | .obj_size = sizeof(struct kcm_sock), | |
1622 | }; | |
1623 | ||
1624 | /* Clone a kcm socket. */ | |
1625 | static int kcm_clone(struct socket *osock, struct kcm_clone *info, | |
1626 | struct socket **newsockp) | |
1627 | { | |
1628 | struct socket *newsock; | |
1629 | struct sock *newsk; | |
1630 | struct file *newfile; | |
1631 | int err, newfd; | |
1632 | ||
1633 | err = -ENFILE; | |
1634 | newsock = sock_alloc(); | |
1635 | if (!newsock) | |
1636 | goto out; | |
1637 | ||
1638 | newsock->type = osock->type; | |
1639 | newsock->ops = osock->ops; | |
1640 | ||
1641 | __module_get(newsock->ops->owner); | |
1642 | ||
1643 | newfd = get_unused_fd_flags(0); | |
1644 | if (unlikely(newfd < 0)) { | |
1645 | err = newfd; | |
1646 | goto out_fd_fail; | |
1647 | } | |
1648 | ||
1649 | newfile = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name); | |
1650 | if (unlikely(IS_ERR(newfile))) { | |
1651 | err = PTR_ERR(newfile); | |
1652 | goto out_sock_alloc_fail; | |
1653 | } | |
1654 | ||
1655 | newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL, | |
1656 | &kcm_proto, true); | |
1657 | if (!newsk) { | |
1658 | err = -ENOMEM; | |
1659 | goto out_sk_alloc_fail; | |
1660 | } | |
1661 | ||
1662 | sock_init_data(newsock, newsk); | |
1663 | init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux); | |
1664 | ||
1665 | fd_install(newfd, newfile); | |
1666 | *newsockp = newsock; | |
1667 | info->fd = newfd; | |
1668 | ||
1669 | return 0; | |
1670 | ||
1671 | out_sk_alloc_fail: | |
1672 | fput(newfile); | |
1673 | out_sock_alloc_fail: | |
1674 | put_unused_fd(newfd); | |
1675 | out_fd_fail: | |
1676 | sock_release(newsock); | |
1677 | out: | |
1678 | return err; | |
1679 | } | |
1680 | ||
1681 | static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) | |
1682 | { | |
1683 | int err; | |
1684 | ||
1685 | switch (cmd) { | |
1686 | case SIOCKCMATTACH: { | |
1687 | struct kcm_attach info; | |
1688 | ||
1689 | if (copy_from_user(&info, (void __user *)arg, sizeof(info))) | |
1690 | err = -EFAULT; | |
1691 | ||
1692 | err = kcm_attach_ioctl(sock, &info); | |
1693 | ||
1694 | break; | |
1695 | } | |
1696 | case SIOCKCMUNATTACH: { | |
1697 | struct kcm_unattach info; | |
1698 | ||
1699 | if (copy_from_user(&info, (void __user *)arg, sizeof(info))) | |
1700 | err = -EFAULT; | |
1701 | ||
1702 | err = kcm_unattach_ioctl(sock, &info); | |
1703 | ||
1704 | break; | |
1705 | } | |
1706 | case SIOCKCMCLONE: { | |
1707 | struct kcm_clone info; | |
1708 | struct socket *newsock = NULL; | |
1709 | ||
1710 | if (copy_from_user(&info, (void __user *)arg, sizeof(info))) | |
1711 | err = -EFAULT; | |
1712 | ||
1713 | err = kcm_clone(sock, &info, &newsock); | |
1714 | ||
1715 | if (!err) { | |
1716 | if (copy_to_user((void __user *)arg, &info, | |
1717 | sizeof(info))) { | |
1718 | err = -EFAULT; | |
c0338aff | 1719 | sys_close(info.fd); |
ab7ac4eb TH |
1720 | } |
1721 | } | |
1722 | ||
1723 | break; | |
1724 | } | |
1725 | default: | |
1726 | err = -ENOIOCTLCMD; | |
1727 | break; | |
1728 | } | |
1729 | ||
1730 | return err; | |
1731 | } | |
1732 | ||
1733 | static void free_mux(struct rcu_head *rcu) | |
1734 | { | |
1735 | struct kcm_mux *mux = container_of(rcu, | |
1736 | struct kcm_mux, rcu); | |
1737 | ||
1738 | kmem_cache_free(kcm_muxp, mux); | |
1739 | } | |
1740 | ||
1741 | static void release_mux(struct kcm_mux *mux) | |
1742 | { | |
1743 | struct kcm_net *knet = mux->knet; | |
1744 | struct kcm_psock *psock, *tmp_psock; | |
1745 | ||
1746 | /* Release psocks */ | |
1747 | list_for_each_entry_safe(psock, tmp_psock, | |
1748 | &mux->psocks, psock_list) { | |
1616b38f | 1749 | if (!WARN_ON(psock->unattaching)) |
ab7ac4eb TH |
1750 | kcm_unattach(psock); |
1751 | } | |
1752 | ||
1753 | if (WARN_ON(mux->psocks_cnt)) | |
1754 | return; | |
1755 | ||
1756 | __skb_queue_purge(&mux->rx_hold_queue); | |
1757 | ||
1758 | mutex_lock(&knet->mutex); | |
cd6e111b TH |
1759 | aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats); |
1760 | aggregate_psock_stats(&mux->aggregate_psock_stats, | |
1761 | &knet->aggregate_psock_stats); | |
9b73896a TH |
1762 | aggregate_strp_stats(&mux->aggregate_strp_stats, |
1763 | &knet->aggregate_strp_stats); | |
ab7ac4eb TH |
1764 | list_del_rcu(&mux->kcm_mux_list); |
1765 | knet->count--; | |
1766 | mutex_unlock(&knet->mutex); | |
1767 | ||
1768 | call_rcu(&mux->rcu, free_mux); | |
1769 | } | |
1770 | ||
1771 | static void kcm_done(struct kcm_sock *kcm) | |
1772 | { | |
1773 | struct kcm_mux *mux = kcm->mux; | |
1774 | struct sock *sk = &kcm->sk; | |
1775 | int socks_cnt; | |
1776 | ||
1777 | spin_lock_bh(&mux->rx_lock); | |
1778 | if (kcm->rx_psock) { | |
1779 | /* Cleanup in unreserve_rx_kcm */ | |
1780 | WARN_ON(kcm->done); | |
1781 | kcm->rx_disabled = 1; | |
1782 | kcm->done = 1; | |
1783 | spin_unlock_bh(&mux->rx_lock); | |
1784 | return; | |
1785 | } | |
1786 | ||
1787 | if (kcm->rx_wait) { | |
1788 | list_del(&kcm->wait_rx_list); | |
1789 | kcm->rx_wait = false; | |
1790 | } | |
1791 | /* Move any pending receive messages to other kcm sockets */ | |
1792 | requeue_rx_msgs(mux, &sk->sk_receive_queue); | |
1793 | ||
1794 | spin_unlock_bh(&mux->rx_lock); | |
1795 | ||
1796 | if (WARN_ON(sk_rmem_alloc_get(sk))) | |
1797 | return; | |
1798 | ||
1799 | /* Detach from MUX */ | |
1800 | spin_lock_bh(&mux->lock); | |
1801 | ||
1802 | list_del(&kcm->kcm_sock_list); | |
1803 | mux->kcm_socks_cnt--; | |
1804 | socks_cnt = mux->kcm_socks_cnt; | |
1805 | ||
1806 | spin_unlock_bh(&mux->lock); | |
1807 | ||
1808 | if (!socks_cnt) { | |
1809 | /* We are done with the mux now. */ | |
1810 | release_mux(mux); | |
1811 | } | |
1812 | ||
1813 | WARN_ON(kcm->rx_wait); | |
1814 | ||
1815 | sock_put(&kcm->sk); | |
1816 | } | |
1817 | ||
1818 | /* Called by kcm_release to close a KCM socket. | |
1819 | * If this is the last KCM socket on the MUX, destroy the MUX. | |
1820 | */ | |
1821 | static int kcm_release(struct socket *sock) | |
1822 | { | |
1823 | struct sock *sk = sock->sk; | |
1824 | struct kcm_sock *kcm; | |
1825 | struct kcm_mux *mux; | |
1826 | struct kcm_psock *psock; | |
1827 | ||
1828 | if (!sk) | |
1829 | return 0; | |
1830 | ||
1831 | kcm = kcm_sk(sk); | |
1832 | mux = kcm->mux; | |
1833 | ||
1834 | sock_orphan(sk); | |
1835 | kfree_skb(kcm->seq_skb); | |
1836 | ||
1837 | lock_sock(sk); | |
1838 | /* Purge queue under lock to avoid race condition with tx_work trying | |
1839 | * to act when queue is nonempty. If tx_work runs after this point | |
1840 | * it will just return. | |
1841 | */ | |
1842 | __skb_queue_purge(&sk->sk_write_queue); | |
9b73896a TH |
1843 | |
1844 | /* Set tx_stopped. This is checked when psock is bound to a kcm and we | |
1845 | * get a writespace callback. This prevents further work being queued | |
1846 | * from the callback (unbinding the psock occurs after canceling work. | |
1847 | */ | |
1848 | kcm->tx_stopped = 1; | |
1849 | ||
ab7ac4eb TH |
1850 | release_sock(sk); |
1851 | ||
1852 | spin_lock_bh(&mux->lock); | |
1853 | if (kcm->tx_wait) { | |
1854 | /* Take of tx_wait list, after this point there should be no way | |
1855 | * that a psock will be assigned to this kcm. | |
1856 | */ | |
1857 | list_del(&kcm->wait_psock_list); | |
1858 | kcm->tx_wait = false; | |
1859 | } | |
1860 | spin_unlock_bh(&mux->lock); | |
1861 | ||
1862 | /* Cancel work. After this point there should be no outside references | |
1863 | * to the kcm socket. | |
1864 | */ | |
1865 | cancel_work_sync(&kcm->tx_work); | |
1866 | ||
1867 | lock_sock(sk); | |
1868 | psock = kcm->tx_psock; | |
1869 | if (psock) { | |
1870 | /* A psock was reserved, so we need to kill it since it | |
1871 | * may already have some bytes queued from a message. We | |
1872 | * need to do this after removing kcm from tx_wait list. | |
1873 | */ | |
1874 | kcm_abort_tx_psock(psock, EPIPE, false); | |
1875 | unreserve_psock(kcm); | |
1876 | } | |
1877 | release_sock(sk); | |
1878 | ||
1879 | WARN_ON(kcm->tx_wait); | |
1880 | WARN_ON(kcm->tx_psock); | |
1881 | ||
1882 | sock->sk = NULL; | |
1883 | ||
1884 | kcm_done(kcm); | |
1885 | ||
1886 | return 0; | |
1887 | } | |
1888 | ||
91687355 | 1889 | static const struct proto_ops kcm_dgram_ops = { |
ab7ac4eb TH |
1890 | .family = PF_KCM, |
1891 | .owner = THIS_MODULE, | |
1892 | .release = kcm_release, | |
1893 | .bind = sock_no_bind, | |
1894 | .connect = sock_no_connect, | |
1895 | .socketpair = sock_no_socketpair, | |
1896 | .accept = sock_no_accept, | |
1897 | .getname = sock_no_getname, | |
1898 | .poll = datagram_poll, | |
1899 | .ioctl = kcm_ioctl, | |
1900 | .listen = sock_no_listen, | |
1901 | .shutdown = sock_no_shutdown, | |
1902 | .setsockopt = kcm_setsockopt, | |
1903 | .getsockopt = kcm_getsockopt, | |
1904 | .sendmsg = kcm_sendmsg, | |
1905 | .recvmsg = kcm_recvmsg, | |
1906 | .mmap = sock_no_mmap, | |
f29698fc | 1907 | .sendpage = kcm_sendpage, |
ab7ac4eb TH |
1908 | }; |
1909 | ||
91687355 TH |
1910 | static const struct proto_ops kcm_seqpacket_ops = { |
1911 | .family = PF_KCM, | |
1912 | .owner = THIS_MODULE, | |
1913 | .release = kcm_release, | |
1914 | .bind = sock_no_bind, | |
1915 | .connect = sock_no_connect, | |
1916 | .socketpair = sock_no_socketpair, | |
1917 | .accept = sock_no_accept, | |
1918 | .getname = sock_no_getname, | |
1919 | .poll = datagram_poll, | |
1920 | .ioctl = kcm_ioctl, | |
1921 | .listen = sock_no_listen, | |
1922 | .shutdown = sock_no_shutdown, | |
1923 | .setsockopt = kcm_setsockopt, | |
1924 | .getsockopt = kcm_getsockopt, | |
1925 | .sendmsg = kcm_sendmsg, | |
1926 | .recvmsg = kcm_recvmsg, | |
1927 | .mmap = sock_no_mmap, | |
f29698fc | 1928 | .sendpage = kcm_sendpage, |
91687355 TH |
1929 | .splice_read = kcm_splice_read, |
1930 | }; | |
1931 | ||
ab7ac4eb TH |
1932 | /* Create proto operation for kcm sockets */ |
1933 | static int kcm_create(struct net *net, struct socket *sock, | |
1934 | int protocol, int kern) | |
1935 | { | |
1936 | struct kcm_net *knet = net_generic(net, kcm_net_id); | |
1937 | struct sock *sk; | |
1938 | struct kcm_mux *mux; | |
1939 | ||
1940 | switch (sock->type) { | |
1941 | case SOCK_DGRAM: | |
91687355 TH |
1942 | sock->ops = &kcm_dgram_ops; |
1943 | break; | |
ab7ac4eb | 1944 | case SOCK_SEQPACKET: |
91687355 | 1945 | sock->ops = &kcm_seqpacket_ops; |
ab7ac4eb TH |
1946 | break; |
1947 | default: | |
1948 | return -ESOCKTNOSUPPORT; | |
1949 | } | |
1950 | ||
1951 | if (protocol != KCMPROTO_CONNECTED) | |
1952 | return -EPROTONOSUPPORT; | |
1953 | ||
1954 | sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern); | |
1955 | if (!sk) | |
1956 | return -ENOMEM; | |
1957 | ||
1958 | /* Allocate a kcm mux, shared between KCM sockets */ | |
1959 | mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL); | |
1960 | if (!mux) { | |
1961 | sk_free(sk); | |
1962 | return -ENOMEM; | |
1963 | } | |
1964 | ||
1965 | spin_lock_init(&mux->lock); | |
1966 | spin_lock_init(&mux->rx_lock); | |
1967 | INIT_LIST_HEAD(&mux->kcm_socks); | |
1968 | INIT_LIST_HEAD(&mux->kcm_rx_waiters); | |
1969 | INIT_LIST_HEAD(&mux->kcm_tx_waiters); | |
1970 | ||
1971 | INIT_LIST_HEAD(&mux->psocks); | |
1972 | INIT_LIST_HEAD(&mux->psocks_ready); | |
1973 | INIT_LIST_HEAD(&mux->psocks_avail); | |
1974 | ||
1975 | mux->knet = knet; | |
1976 | ||
1977 | /* Add new MUX to list */ | |
1978 | mutex_lock(&knet->mutex); | |
1979 | list_add_rcu(&mux->kcm_mux_list, &knet->mux_list); | |
1980 | knet->count++; | |
1981 | mutex_unlock(&knet->mutex); | |
1982 | ||
1983 | skb_queue_head_init(&mux->rx_hold_queue); | |
1984 | ||
1985 | /* Init KCM socket */ | |
1986 | sock_init_data(sock, sk); | |
1987 | init_kcm_sock(kcm_sk(sk), mux); | |
1988 | ||
1989 | return 0; | |
1990 | } | |
1991 | ||
1992 | static struct net_proto_family kcm_family_ops = { | |
1993 | .family = PF_KCM, | |
1994 | .create = kcm_create, | |
1995 | .owner = THIS_MODULE, | |
1996 | }; | |
1997 | ||
1998 | static __net_init int kcm_init_net(struct net *net) | |
1999 | { | |
2000 | struct kcm_net *knet = net_generic(net, kcm_net_id); | |
2001 | ||
2002 | INIT_LIST_HEAD_RCU(&knet->mux_list); | |
2003 | mutex_init(&knet->mutex); | |
2004 | ||
2005 | return 0; | |
2006 | } | |
2007 | ||
2008 | static __net_exit void kcm_exit_net(struct net *net) | |
2009 | { | |
2010 | struct kcm_net *knet = net_generic(net, kcm_net_id); | |
2011 | ||
2012 | /* All KCM sockets should be closed at this point, which should mean | |
2013 | * that all multiplexors and psocks have been destroyed. | |
2014 | */ | |
2015 | WARN_ON(!list_empty(&knet->mux_list)); | |
2016 | } | |
2017 | ||
2018 | static struct pernet_operations kcm_net_ops = { | |
2019 | .init = kcm_init_net, | |
2020 | .exit = kcm_exit_net, | |
2021 | .id = &kcm_net_id, | |
2022 | .size = sizeof(struct kcm_net), | |
2023 | }; | |
2024 | ||
2025 | static int __init kcm_init(void) | |
2026 | { | |
2027 | int err = -ENOMEM; | |
2028 | ||
2029 | kcm_muxp = kmem_cache_create("kcm_mux_cache", | |
2030 | sizeof(struct kcm_mux), 0, | |
2031 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
2032 | if (!kcm_muxp) | |
2033 | goto fail; | |
2034 | ||
2035 | kcm_psockp = kmem_cache_create("kcm_psock_cache", | |
2036 | sizeof(struct kcm_psock), 0, | |
2037 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
2038 | if (!kcm_psockp) | |
2039 | goto fail; | |
2040 | ||
2041 | kcm_wq = create_singlethread_workqueue("kkcmd"); | |
2042 | if (!kcm_wq) | |
2043 | goto fail; | |
2044 | ||
2045 | err = proto_register(&kcm_proto, 1); | |
2046 | if (err) | |
2047 | goto fail; | |
2048 | ||
2049 | err = sock_register(&kcm_family_ops); | |
2050 | if (err) | |
2051 | goto sock_register_fail; | |
2052 | ||
2053 | err = register_pernet_device(&kcm_net_ops); | |
2054 | if (err) | |
2055 | goto net_ops_fail; | |
2056 | ||
cd6e111b TH |
2057 | err = kcm_proc_init(); |
2058 | if (err) | |
2059 | goto proc_init_fail; | |
2060 | ||
ab7ac4eb TH |
2061 | return 0; |
2062 | ||
cd6e111b TH |
2063 | proc_init_fail: |
2064 | unregister_pernet_device(&kcm_net_ops); | |
2065 | ||
ab7ac4eb TH |
2066 | net_ops_fail: |
2067 | sock_unregister(PF_KCM); | |
2068 | ||
2069 | sock_register_fail: | |
2070 | proto_unregister(&kcm_proto); | |
2071 | ||
2072 | fail: | |
2073 | kmem_cache_destroy(kcm_muxp); | |
2074 | kmem_cache_destroy(kcm_psockp); | |
2075 | ||
2076 | if (kcm_wq) | |
2077 | destroy_workqueue(kcm_wq); | |
2078 | ||
2079 | return err; | |
2080 | } | |
2081 | ||
2082 | static void __exit kcm_exit(void) | |
2083 | { | |
cd6e111b | 2084 | kcm_proc_exit(); |
ab7ac4eb TH |
2085 | unregister_pernet_device(&kcm_net_ops); |
2086 | sock_unregister(PF_KCM); | |
2087 | proto_unregister(&kcm_proto); | |
2088 | destroy_workqueue(kcm_wq); | |
2089 | ||
2090 | kmem_cache_destroy(kcm_muxp); | |
2091 | kmem_cache_destroy(kcm_psockp); | |
2092 | } | |
2093 | ||
2094 | module_init(kcm_init); | |
2095 | module_exit(kcm_exit); | |
2096 | ||
2097 | MODULE_LICENSE("GPL"); | |
2098 | MODULE_ALIAS_NETPROTO(PF_KCM); | |
2099 |