1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VMware vSockets Driver
4  *
5  * Copyright (C) 2009-2013 VMware, Inc. All rights reserved.
6  */
7 
8 #include <linux/types.h>
9 #include <linux/socket.h>
10 #include <linux/stddef.h>
11 #include <net/sock.h>
12 
13 #include "vmci_transport_notify.h"
14 
15 #define PKT_FIELD(vsk, field_name) (vmci_trans(vsk)->notify.pkt.field_name)
16 
vmci_transport_notify_waiting_write(struct vsock_sock * vsk)17 static bool vmci_transport_notify_waiting_write(struct vsock_sock *vsk)
18 {
19 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
20 	bool retval;
21 	u64 notify_limit;
22 
23 	if (!PKT_FIELD(vsk, peer_waiting_write))
24 		return false;
25 
26 #ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
27 	/* When the sender blocks, we take that as a sign that the sender is
28 	 * faster than the receiver. To reduce the transmit rate of the sender,
29 	 * we delay the sending of the read notification by decreasing the
30 	 * write_notify_window. The notification is delayed until the number of
31 	 * bytes used in the queue drops below the write_notify_window.
32 	 */
33 
34 	if (!PKT_FIELD(vsk, peer_waiting_write_detected)) {
35 		PKT_FIELD(vsk, peer_waiting_write_detected) = true;
36 		if (PKT_FIELD(vsk, write_notify_window) < PAGE_SIZE) {
37 			PKT_FIELD(vsk, write_notify_window) =
38 			    PKT_FIELD(vsk, write_notify_min_window);
39 		} else {
40 			PKT_FIELD(vsk, write_notify_window) -= PAGE_SIZE;
41 			if (PKT_FIELD(vsk, write_notify_window) <
42 			    PKT_FIELD(vsk, write_notify_min_window))
43 				PKT_FIELD(vsk, write_notify_window) =
44 				    PKT_FIELD(vsk, write_notify_min_window);
45 
46 		}
47 	}
48 	notify_limit = vmci_trans(vsk)->consume_size -
49 		PKT_FIELD(vsk, write_notify_window);
50 #else
51 	notify_limit = 0;
52 #endif
53 
54 	/* For now we ignore the wait information and just see if the free
55 	 * space exceeds the notify limit.  Note that improving this function
56 	 * to be more intelligent will not require a protocol change and will
57 	 * retain compatibility between endpoints with mixed versions of this
58 	 * function.
59 	 *
60 	 * The notify_limit is used to delay notifications in the case where
61 	 * flow control is enabled. Below the test is expressed in terms of
62 	 * free space in the queue: if free_space > ConsumeSize -
63 	 * write_notify_window then notify An alternate way of expressing this
64 	 * is to rewrite the expression to use the data ready in the receive
65 	 * queue: if write_notify_window > bufferReady then notify as
66 	 * free_space == ConsumeSize - bufferReady.
67 	 */
68 	retval = vmci_qpair_consume_free_space(vmci_trans(vsk)->qpair) >
69 		notify_limit;
70 #ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
71 	if (retval) {
72 		/*
73 		 * Once we notify the peer, we reset the detected flag so the
74 		 * next wait will again cause a decrease in the window size.
75 		 */
76 
77 		PKT_FIELD(vsk, peer_waiting_write_detected) = false;
78 	}
79 #endif
80 	return retval;
81 #else
82 	return true;
83 #endif
84 }
85 
vmci_transport_notify_waiting_read(struct vsock_sock * vsk)86 static bool vmci_transport_notify_waiting_read(struct vsock_sock *vsk)
87 {
88 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
89 	if (!PKT_FIELD(vsk, peer_waiting_read))
90 		return false;
91 
92 	/* For now we ignore the wait information and just see if there is any
93 	 * data for our peer to read.  Note that improving this function to be
94 	 * more intelligent will not require a protocol change and will retain
95 	 * compatibility between endpoints with mixed versions of this
96 	 * function.
97 	 */
98 	return vmci_qpair_produce_buf_ready(vmci_trans(vsk)->qpair) > 0;
99 #else
100 	return true;
101 #endif
102 }
103 
104 static void
vmci_transport_handle_waiting_read(struct sock * sk,struct vmci_transport_packet * pkt,bool bottom_half,struct sockaddr_vm * dst,struct sockaddr_vm * src)105 vmci_transport_handle_waiting_read(struct sock *sk,
106 				   struct vmci_transport_packet *pkt,
107 				   bool bottom_half,
108 				   struct sockaddr_vm *dst,
109 				   struct sockaddr_vm *src)
110 {
111 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
112 	struct vsock_sock *vsk;
113 
114 	vsk = vsock_sk(sk);
115 
116 	PKT_FIELD(vsk, peer_waiting_read) = true;
117 	memcpy(&PKT_FIELD(vsk, peer_waiting_read_info), &pkt->u.wait,
118 	       sizeof(PKT_FIELD(vsk, peer_waiting_read_info)));
119 
120 	if (vmci_transport_notify_waiting_read(vsk)) {
121 		bool sent;
122 
123 		if (bottom_half)
124 			sent = vmci_transport_send_wrote_bh(dst, src) > 0;
125 		else
126 			sent = vmci_transport_send_wrote(sk) > 0;
127 
128 		if (sent)
129 			PKT_FIELD(vsk, peer_waiting_read) = false;
130 	}
131 #endif
132 }
133 
134 static void
vmci_transport_handle_waiting_write(struct sock * sk,struct vmci_transport_packet * pkt,bool bottom_half,struct sockaddr_vm * dst,struct sockaddr_vm * src)135 vmci_transport_handle_waiting_write(struct sock *sk,
136 				    struct vmci_transport_packet *pkt,
137 				    bool bottom_half,
138 				    struct sockaddr_vm *dst,
139 				    struct sockaddr_vm *src)
140 {
141 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
142 	struct vsock_sock *vsk;
143 
144 	vsk = vsock_sk(sk);
145 
146 	PKT_FIELD(vsk, peer_waiting_write) = true;
147 	memcpy(&PKT_FIELD(vsk, peer_waiting_write_info), &pkt->u.wait,
148 	       sizeof(PKT_FIELD(vsk, peer_waiting_write_info)));
149 
150 	if (vmci_transport_notify_waiting_write(vsk)) {
151 		bool sent;
152 
153 		if (bottom_half)
154 			sent = vmci_transport_send_read_bh(dst, src) > 0;
155 		else
156 			sent = vmci_transport_send_read(sk) > 0;
157 
158 		if (sent)
159 			PKT_FIELD(vsk, peer_waiting_write) = false;
160 	}
161 #endif
162 }
163 
164 static void
vmci_transport_handle_read(struct sock * sk,struct vmci_transport_packet * pkt,bool bottom_half,struct sockaddr_vm * dst,struct sockaddr_vm * src)165 vmci_transport_handle_read(struct sock *sk,
166 			   struct vmci_transport_packet *pkt,
167 			   bool bottom_half,
168 			   struct sockaddr_vm *dst, struct sockaddr_vm *src)
169 {
170 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
171 	struct vsock_sock *vsk;
172 
173 	vsk = vsock_sk(sk);
174 	PKT_FIELD(vsk, sent_waiting_write) = false;
175 #endif
176 
177 	sk->sk_write_space(sk);
178 }
179 
send_waiting_read(struct sock * sk,u64 room_needed)180 static bool send_waiting_read(struct sock *sk, u64 room_needed)
181 {
182 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
183 	struct vsock_sock *vsk;
184 	struct vmci_transport_waiting_info waiting_info;
185 	u64 tail;
186 	u64 head;
187 	u64 room_left;
188 	bool ret;
189 
190 	vsk = vsock_sk(sk);
191 
192 	if (PKT_FIELD(vsk, sent_waiting_read))
193 		return true;
194 
195 	if (PKT_FIELD(vsk, write_notify_window) <
196 			vmci_trans(vsk)->consume_size)
197 		PKT_FIELD(vsk, write_notify_window) =
198 		    min(PKT_FIELD(vsk, write_notify_window) + PAGE_SIZE,
199 			vmci_trans(vsk)->consume_size);
200 
201 	vmci_qpair_get_consume_indexes(vmci_trans(vsk)->qpair, &tail, &head);
202 	room_left = vmci_trans(vsk)->consume_size - head;
203 	if (room_needed >= room_left) {
204 		waiting_info.offset = room_needed - room_left;
205 		waiting_info.generation =
206 		    PKT_FIELD(vsk, consume_q_generation) + 1;
207 	} else {
208 		waiting_info.offset = head + room_needed;
209 		waiting_info.generation = PKT_FIELD(vsk, consume_q_generation);
210 	}
211 
212 	ret = vmci_transport_send_waiting_read(sk, &waiting_info) > 0;
213 	if (ret)
214 		PKT_FIELD(vsk, sent_waiting_read) = true;
215 
216 	return ret;
217 #else
218 	return true;
219 #endif
220 }
221 
send_waiting_write(struct sock * sk,u64 room_needed)222 static bool send_waiting_write(struct sock *sk, u64 room_needed)
223 {
224 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
225 	struct vsock_sock *vsk;
226 	struct vmci_transport_waiting_info waiting_info;
227 	u64 tail;
228 	u64 head;
229 	u64 room_left;
230 	bool ret;
231 
232 	vsk = vsock_sk(sk);
233 
234 	if (PKT_FIELD(vsk, sent_waiting_write))
235 		return true;
236 
237 	vmci_qpair_get_produce_indexes(vmci_trans(vsk)->qpair, &tail, &head);
238 	room_left = vmci_trans(vsk)->produce_size - tail;
239 	if (room_needed + 1 >= room_left) {
240 		/* Wraps around to current generation. */
241 		waiting_info.offset = room_needed + 1 - room_left;
242 		waiting_info.generation = PKT_FIELD(vsk, produce_q_generation);
243 	} else {
244 		waiting_info.offset = tail + room_needed + 1;
245 		waiting_info.generation =
246 		    PKT_FIELD(vsk, produce_q_generation) - 1;
247 	}
248 
249 	ret = vmci_transport_send_waiting_write(sk, &waiting_info) > 0;
250 	if (ret)
251 		PKT_FIELD(vsk, sent_waiting_write) = true;
252 
253 	return ret;
254 #else
255 	return true;
256 #endif
257 }
258 
vmci_transport_send_read_notification(struct sock * sk)259 static int vmci_transport_send_read_notification(struct sock *sk)
260 {
261 	struct vsock_sock *vsk;
262 	bool sent_read;
263 	unsigned int retries;
264 	int err;
265 
266 	vsk = vsock_sk(sk);
267 	sent_read = false;
268 	retries = 0;
269 	err = 0;
270 
271 	if (vmci_transport_notify_waiting_write(vsk)) {
272 		/* Notify the peer that we have read, retrying the send on
273 		 * failure up to our maximum value.  XXX For now we just log
274 		 * the failure, but later we should schedule a work item to
275 		 * handle the resend until it succeeds.  That would require
276 		 * keeping track of work items in the vsk and cleaning them up
277 		 * upon socket close.
278 		 */
279 		while (!(vsk->peer_shutdown & RCV_SHUTDOWN) &&
280 		       !sent_read &&
281 		       retries < VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
282 			err = vmci_transport_send_read(sk);
283 			if (err >= 0)
284 				sent_read = true;
285 
286 			retries++;
287 		}
288 
289 		if (retries >= VMCI_TRANSPORT_MAX_DGRAM_RESENDS)
290 			pr_err("%p unable to send read notify to peer\n", sk);
291 		else
292 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
293 			PKT_FIELD(vsk, peer_waiting_write) = false;
294 #endif
295 
296 	}
297 	return err;
298 }
299 
300 static void
vmci_transport_handle_wrote(struct sock * sk,struct vmci_transport_packet * pkt,bool bottom_half,struct sockaddr_vm * dst,struct sockaddr_vm * src)301 vmci_transport_handle_wrote(struct sock *sk,
302 			    struct vmci_transport_packet *pkt,
303 			    bool bottom_half,
304 			    struct sockaddr_vm *dst, struct sockaddr_vm *src)
305 {
306 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
307 	struct vsock_sock *vsk = vsock_sk(sk);
308 	PKT_FIELD(vsk, sent_waiting_read) = false;
309 #endif
310 	sk->sk_data_ready(sk);
311 }
312 
vmci_transport_notify_pkt_socket_init(struct sock * sk)313 static void vmci_transport_notify_pkt_socket_init(struct sock *sk)
314 {
315 	struct vsock_sock *vsk = vsock_sk(sk);
316 
317 	PKT_FIELD(vsk, write_notify_window) = PAGE_SIZE;
318 	PKT_FIELD(vsk, write_notify_min_window) = PAGE_SIZE;
319 	PKT_FIELD(vsk, peer_waiting_read) = false;
320 	PKT_FIELD(vsk, peer_waiting_write) = false;
321 	PKT_FIELD(vsk, peer_waiting_write_detected) = false;
322 	PKT_FIELD(vsk, sent_waiting_read) = false;
323 	PKT_FIELD(vsk, sent_waiting_write) = false;
324 	PKT_FIELD(vsk, produce_q_generation) = 0;
325 	PKT_FIELD(vsk, consume_q_generation) = 0;
326 
327 	memset(&PKT_FIELD(vsk, peer_waiting_read_info), 0,
328 	       sizeof(PKT_FIELD(vsk, peer_waiting_read_info)));
329 	memset(&PKT_FIELD(vsk, peer_waiting_write_info), 0,
330 	       sizeof(PKT_FIELD(vsk, peer_waiting_write_info)));
331 }
332 
vmci_transport_notify_pkt_socket_destruct(struct vsock_sock * vsk)333 static void vmci_transport_notify_pkt_socket_destruct(struct vsock_sock *vsk)
334 {
335 }
336 
337 static int
vmci_transport_notify_pkt_poll_in(struct sock * sk,size_t target,bool * data_ready_now)338 vmci_transport_notify_pkt_poll_in(struct sock *sk,
339 				  size_t target, bool *data_ready_now)
340 {
341 	struct vsock_sock *vsk = vsock_sk(sk);
342 
343 	if (vsock_stream_has_data(vsk)) {
344 		*data_ready_now = true;
345 	} else {
346 		/* We can't read right now because there is nothing in the
347 		 * queue. Ask for notifications when there is something to
348 		 * read.
349 		 */
350 		if (sk->sk_state == TCP_ESTABLISHED) {
351 			if (!send_waiting_read(sk, 1))
352 				return -1;
353 
354 		}
355 		*data_ready_now = false;
356 	}
357 
358 	return 0;
359 }
360 
361 static int
vmci_transport_notify_pkt_poll_out(struct sock * sk,size_t target,bool * space_avail_now)362 vmci_transport_notify_pkt_poll_out(struct sock *sk,
363 				   size_t target, bool *space_avail_now)
364 {
365 	s64 produce_q_free_space;
366 	struct vsock_sock *vsk = vsock_sk(sk);
367 
368 	produce_q_free_space = vsock_stream_has_space(vsk);
369 	if (produce_q_free_space > 0) {
370 		*space_avail_now = true;
371 		return 0;
372 	} else if (produce_q_free_space == 0) {
373 		/* This is a connected socket but we can't currently send data.
374 		 * Notify the peer that we are waiting if the queue is full. We
375 		 * only send a waiting write if the queue is full because
376 		 * otherwise we end up in an infinite WAITING_WRITE, READ,
377 		 * WAITING_WRITE, READ, etc. loop. Treat failing to send the
378 		 * notification as a socket error, passing that back through
379 		 * the mask.
380 		 */
381 		if (!send_waiting_write(sk, 1))
382 			return -1;
383 
384 		*space_avail_now = false;
385 	}
386 
387 	return 0;
388 }
389 
390 static int
vmci_transport_notify_pkt_recv_init(struct sock * sk,size_t target,struct vmci_transport_recv_notify_data * data)391 vmci_transport_notify_pkt_recv_init(
392 			struct sock *sk,
393 			size_t target,
394 			struct vmci_transport_recv_notify_data *data)
395 {
396 	struct vsock_sock *vsk = vsock_sk(sk);
397 
398 #ifdef VSOCK_OPTIMIZATION_WAITING_NOTIFY
399 	data->consume_head = 0;
400 	data->produce_tail = 0;
401 #ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
402 	data->notify_on_block = false;
403 
404 	if (PKT_FIELD(vsk, write_notify_min_window) < target + 1) {
405 		PKT_FIELD(vsk, write_notify_min_window) = target + 1;
406 		if (PKT_FIELD(vsk, write_notify_window) <
407 		    PKT_FIELD(vsk, write_notify_min_window)) {
408 			/* If the current window is smaller than the new
409 			 * minimal window size, we need to reevaluate whether
410 			 * we need to notify the sender. If the number of ready
411 			 * bytes are smaller than the new window, we need to
412 			 * send a notification to the sender before we block.
413 			 */
414 
415 			PKT_FIELD(vsk, write_notify_window) =
416 			    PKT_FIELD(vsk, write_notify_min_window);
417 			data->notify_on_block = true;
418 		}
419 	}
420 #endif
421 #endif
422 
423 	return 0;
424 }
425 
426 static int
vmci_transport_notify_pkt_recv_pre_block(struct sock * sk,size_t target,struct vmci_transport_recv_notify_data * data)427 vmci_transport_notify_pkt_recv_pre_block(
428 				struct sock *sk,
429 				size_t target,
430 				struct vmci_transport_recv_notify_data *data)
431 {
432 	int err = 0;
433 
434 	/* Notify our peer that we are waiting for data to read. */
435 	if (!send_waiting_read(sk, target)) {
436 		err = -EHOSTUNREACH;
437 		return err;
438 	}
439 #ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
440 	if (data->notify_on_block) {
441 		err = vmci_transport_send_read_notification(sk);
442 		if (err < 0)
443 			return err;
444 
445 		data->notify_on_block = false;
446 	}
447 #endif
448 
449 	return err;
450 }
451 
452 static int
vmci_transport_notify_pkt_recv_pre_dequeue(struct sock * sk,size_t target,struct vmci_transport_recv_notify_data * data)453 vmci_transport_notify_pkt_recv_pre_dequeue(
454 				struct sock *sk,
455 				size_t target,
456 				struct vmci_transport_recv_notify_data *data)
457 {
458 	struct vsock_sock *vsk = vsock_sk(sk);
459 
460 	/* Now consume up to len bytes from the queue.  Note that since we have
461 	 * the socket locked we should copy at least ready bytes.
462 	 */
463 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
464 	vmci_qpair_get_consume_indexes(vmci_trans(vsk)->qpair,
465 				       &data->produce_tail,
466 				       &data->consume_head);
467 #endif
468 
469 	return 0;
470 }
471 
472 static int
vmci_transport_notify_pkt_recv_post_dequeue(struct sock * sk,size_t target,ssize_t copied,bool data_read,struct vmci_transport_recv_notify_data * data)473 vmci_transport_notify_pkt_recv_post_dequeue(
474 				struct sock *sk,
475 				size_t target,
476 				ssize_t copied,
477 				bool data_read,
478 				struct vmci_transport_recv_notify_data *data)
479 {
480 	struct vsock_sock *vsk;
481 	int err;
482 
483 	vsk = vsock_sk(sk);
484 	err = 0;
485 
486 	if (data_read) {
487 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
488 		/* Detect a wrap-around to maintain queue generation.  Note
489 		 * that this is safe since we hold the socket lock across the
490 		 * two queue pair operations.
491 		 */
492 		if (copied >=
493 			vmci_trans(vsk)->consume_size - data->consume_head)
494 			PKT_FIELD(vsk, consume_q_generation)++;
495 #endif
496 
497 		err = vmci_transport_send_read_notification(sk);
498 		if (err < 0)
499 			return err;
500 
501 	}
502 	return err;
503 }
504 
505 static int
vmci_transport_notify_pkt_send_init(struct sock * sk,struct vmci_transport_send_notify_data * data)506 vmci_transport_notify_pkt_send_init(
507 			struct sock *sk,
508 			struct vmci_transport_send_notify_data *data)
509 {
510 #ifdef VSOCK_OPTIMIZATION_WAITING_NOTIFY
511 	data->consume_head = 0;
512 	data->produce_tail = 0;
513 #endif
514 
515 	return 0;
516 }
517 
518 static int
vmci_transport_notify_pkt_send_pre_block(struct sock * sk,struct vmci_transport_send_notify_data * data)519 vmci_transport_notify_pkt_send_pre_block(
520 				struct sock *sk,
521 				struct vmci_transport_send_notify_data *data)
522 {
523 	/* Notify our peer that we are waiting for room to write. */
524 	if (!send_waiting_write(sk, 1))
525 		return -EHOSTUNREACH;
526 
527 	return 0;
528 }
529 
530 static int
vmci_transport_notify_pkt_send_pre_enqueue(struct sock * sk,struct vmci_transport_send_notify_data * data)531 vmci_transport_notify_pkt_send_pre_enqueue(
532 				struct sock *sk,
533 				struct vmci_transport_send_notify_data *data)
534 {
535 	struct vsock_sock *vsk = vsock_sk(sk);
536 
537 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
538 	vmci_qpair_get_produce_indexes(vmci_trans(vsk)->qpair,
539 				       &data->produce_tail,
540 				       &data->consume_head);
541 #endif
542 
543 	return 0;
544 }
545 
546 static int
vmci_transport_notify_pkt_send_post_enqueue(struct sock * sk,ssize_t written,struct vmci_transport_send_notify_data * data)547 vmci_transport_notify_pkt_send_post_enqueue(
548 				struct sock *sk,
549 				ssize_t written,
550 				struct vmci_transport_send_notify_data *data)
551 {
552 	int err = 0;
553 	struct vsock_sock *vsk;
554 	bool sent_wrote = false;
555 	int retries = 0;
556 
557 	vsk = vsock_sk(sk);
558 
559 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
560 	/* Detect a wrap-around to maintain queue generation.  Note that this
561 	 * is safe since we hold the socket lock across the two queue pair
562 	 * operations.
563 	 */
564 	if (written >= vmci_trans(vsk)->produce_size - data->produce_tail)
565 		PKT_FIELD(vsk, produce_q_generation)++;
566 
567 #endif
568 
569 	if (vmci_transport_notify_waiting_read(vsk)) {
570 		/* Notify the peer that we have written, retrying the send on
571 		 * failure up to our maximum value. See the XXX comment for the
572 		 * corresponding piece of code in StreamRecvmsg() for potential
573 		 * improvements.
574 		 */
575 		while (!(vsk->peer_shutdown & RCV_SHUTDOWN) &&
576 		       !sent_wrote &&
577 		       retries < VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
578 			err = vmci_transport_send_wrote(sk);
579 			if (err >= 0)
580 				sent_wrote = true;
581 
582 			retries++;
583 		}
584 
585 		if (retries >= VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
586 			pr_err("%p unable to send wrote notify to peer\n", sk);
587 			return err;
588 		} else {
589 #if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
590 			PKT_FIELD(vsk, peer_waiting_read) = false;
591 #endif
592 		}
593 	}
594 	return err;
595 }
596 
597 static void
vmci_transport_notify_pkt_handle_pkt(struct sock * sk,struct vmci_transport_packet * pkt,bool bottom_half,struct sockaddr_vm * dst,struct sockaddr_vm * src,bool * pkt_processed)598 vmci_transport_notify_pkt_handle_pkt(
599 			struct sock *sk,
600 			struct vmci_transport_packet *pkt,
601 			bool bottom_half,
602 			struct sockaddr_vm *dst,
603 			struct sockaddr_vm *src, bool *pkt_processed)
604 {
605 	bool processed = false;
606 
607 	switch (pkt->type) {
608 	case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
609 		vmci_transport_handle_wrote(sk, pkt, bottom_half, dst, src);
610 		processed = true;
611 		break;
612 	case VMCI_TRANSPORT_PACKET_TYPE_READ:
613 		vmci_transport_handle_read(sk, pkt, bottom_half, dst, src);
614 		processed = true;
615 		break;
616 	case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
617 		vmci_transport_handle_waiting_write(sk, pkt, bottom_half,
618 						    dst, src);
619 		processed = true;
620 		break;
621 
622 	case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
623 		vmci_transport_handle_waiting_read(sk, pkt, bottom_half,
624 						   dst, src);
625 		processed = true;
626 		break;
627 	}
628 
629 	if (pkt_processed)
630 		*pkt_processed = processed;
631 }
632 
vmci_transport_notify_pkt_process_request(struct sock * sk)633 static void vmci_transport_notify_pkt_process_request(struct sock *sk)
634 {
635 	struct vsock_sock *vsk = vsock_sk(sk);
636 
637 	PKT_FIELD(vsk, write_notify_window) = vmci_trans(vsk)->consume_size;
638 	if (vmci_trans(vsk)->consume_size <
639 		PKT_FIELD(vsk, write_notify_min_window))
640 		PKT_FIELD(vsk, write_notify_min_window) =
641 			vmci_trans(vsk)->consume_size;
642 }
643 
vmci_transport_notify_pkt_process_negotiate(struct sock * sk)644 static void vmci_transport_notify_pkt_process_negotiate(struct sock *sk)
645 {
646 	struct vsock_sock *vsk = vsock_sk(sk);
647 
648 	PKT_FIELD(vsk, write_notify_window) = vmci_trans(vsk)->consume_size;
649 	if (vmci_trans(vsk)->consume_size <
650 		PKT_FIELD(vsk, write_notify_min_window))
651 		PKT_FIELD(vsk, write_notify_min_window) =
652 			vmci_trans(vsk)->consume_size;
653 }
654 
655 /* Socket control packet based operations. */
656 const struct vmci_transport_notify_ops vmci_transport_notify_pkt_ops = {
657 	.socket_init = vmci_transport_notify_pkt_socket_init,
658 	.socket_destruct = vmci_transport_notify_pkt_socket_destruct,
659 	.poll_in = vmci_transport_notify_pkt_poll_in,
660 	.poll_out = vmci_transport_notify_pkt_poll_out,
661 	.handle_notify_pkt = vmci_transport_notify_pkt_handle_pkt,
662 	.recv_init = vmci_transport_notify_pkt_recv_init,
663 	.recv_pre_block = vmci_transport_notify_pkt_recv_pre_block,
664 	.recv_pre_dequeue = vmci_transport_notify_pkt_recv_pre_dequeue,
665 	.recv_post_dequeue = vmci_transport_notify_pkt_recv_post_dequeue,
666 	.send_init = vmci_transport_notify_pkt_send_init,
667 	.send_pre_block = vmci_transport_notify_pkt_send_pre_block,
668 	.send_pre_enqueue = vmci_transport_notify_pkt_send_pre_enqueue,
669 	.send_post_enqueue = vmci_transport_notify_pkt_send_post_enqueue,
670 	.process_request = vmci_transport_notify_pkt_process_request,
671 	.process_negotiate = vmci_transport_notify_pkt_process_negotiate,
672 };
673