1 /*
2  * VMware vSockets Driver
3  *
4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the Free
8  * Software Foundation version 2 and no later version.
9  *
10  * This program is distributed in the hope that it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  */
15 
16 #include <linux/types.h>
17 #include <linux/bitops.h>
18 #include <linux/cred.h>
19 #include <linux/init.h>
20 #include <linux/io.h>
21 #include <linux/kernel.h>
22 #include <linux/kmod.h>
23 #include <linux/list.h>
24 #include <linux/module.h>
25 #include <linux/mutex.h>
26 #include <linux/net.h>
27 #include <linux/poll.h>
28 #include <linux/skbuff.h>
29 #include <linux/smp.h>
30 #include <linux/socket.h>
31 #include <linux/stddef.h>
32 #include <linux/unistd.h>
33 #include <linux/wait.h>
34 #include <linux/workqueue.h>
35 #include <net/sock.h>
36 #include <net/af_vsock.h>
37 
38 #include "vmci_transport_notify.h"
39 
40 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
41 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
42 static void vmci_transport_peer_detach_cb(u32 sub_id,
43 					  const struct vmci_event_data *ed,
44 					  void *client_data);
45 static void vmci_transport_recv_pkt_work(struct work_struct *work);
46 static void vmci_transport_cleanup(struct work_struct *work);
47 static int vmci_transport_recv_listen(struct sock *sk,
48 				      struct vmci_transport_packet *pkt);
49 static int vmci_transport_recv_connecting_server(
50 					struct sock *sk,
51 					struct sock *pending,
52 					struct vmci_transport_packet *pkt);
53 static int vmci_transport_recv_connecting_client(
54 					struct sock *sk,
55 					struct vmci_transport_packet *pkt);
56 static int vmci_transport_recv_connecting_client_negotiate(
57 					struct sock *sk,
58 					struct vmci_transport_packet *pkt);
59 static int vmci_transport_recv_connecting_client_invalid(
60 					struct sock *sk,
61 					struct vmci_transport_packet *pkt);
62 static int vmci_transport_recv_connected(struct sock *sk,
63 					 struct vmci_transport_packet *pkt);
64 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
65 static u16 vmci_transport_new_proto_supported_versions(void);
66 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
67 						  bool old_pkt_proto);
68 
69 struct vmci_transport_recv_pkt_info {
70 	struct work_struct work;
71 	struct sock *sk;
72 	struct vmci_transport_packet pkt;
73 };
74 
75 static LIST_HEAD(vmci_transport_cleanup_list);
76 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
77 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
78 
79 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
80 							   VMCI_INVALID_ID };
81 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
82 
83 static int PROTOCOL_OVERRIDE = -1;
84 
85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
88 
89 /* The default peer timeout indicates how long we will wait for a peer response
90  * to a control message.
91  */
92 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
93 
94 /* Helper function to convert from a VMCI error code to a VSock error code. */
95 
96 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
97 {
98 	switch (vmci_error) {
99 	case VMCI_ERROR_NO_MEM:
100 		return -ENOMEM;
101 	case VMCI_ERROR_DUPLICATE_ENTRY:
102 	case VMCI_ERROR_ALREADY_EXISTS:
103 		return -EADDRINUSE;
104 	case VMCI_ERROR_NO_ACCESS:
105 		return -EPERM;
106 	case VMCI_ERROR_NO_RESOURCES:
107 		return -ENOBUFS;
108 	case VMCI_ERROR_INVALID_RESOURCE:
109 		return -EHOSTUNREACH;
110 	case VMCI_ERROR_INVALID_ARGS:
111 	default:
112 		break;
113 	}
114 	return -EINVAL;
115 }
116 
117 static u32 vmci_transport_peer_rid(u32 peer_cid)
118 {
119 	if (VMADDR_CID_HYPERVISOR == peer_cid)
120 		return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
121 
122 	return VMCI_TRANSPORT_PACKET_RID;
123 }
124 
125 static inline void
126 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
127 			   struct sockaddr_vm *src,
128 			   struct sockaddr_vm *dst,
129 			   u8 type,
130 			   u64 size,
131 			   u64 mode,
132 			   struct vmci_transport_waiting_info *wait,
133 			   u16 proto,
134 			   struct vmci_handle handle)
135 {
136 	/* We register the stream control handler as an any cid handle so we
137 	 * must always send from a source address of VMADDR_CID_ANY
138 	 */
139 	pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
140 				       VMCI_TRANSPORT_PACKET_RID);
141 	pkt->dg.dst = vmci_make_handle(dst->svm_cid,
142 				       vmci_transport_peer_rid(dst->svm_cid));
143 	pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
144 	pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
145 	pkt->type = type;
146 	pkt->src_port = src->svm_port;
147 	pkt->dst_port = dst->svm_port;
148 	memset(&pkt->proto, 0, sizeof(pkt->proto));
149 	memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
150 
151 	switch (pkt->type) {
152 	case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
153 		pkt->u.size = 0;
154 		break;
155 
156 	case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
157 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
158 		pkt->u.size = size;
159 		break;
160 
161 	case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
162 	case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
163 		pkt->u.handle = handle;
164 		break;
165 
166 	case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
167 	case VMCI_TRANSPORT_PACKET_TYPE_READ:
168 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
169 		pkt->u.size = 0;
170 		break;
171 
172 	case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
173 		pkt->u.mode = mode;
174 		break;
175 
176 	case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
177 	case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
178 		memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
179 		break;
180 
181 	case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
182 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
183 		pkt->u.size = size;
184 		pkt->proto = proto;
185 		break;
186 	}
187 }
188 
189 static inline void
190 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
191 				    struct sockaddr_vm *local,
192 				    struct sockaddr_vm *remote)
193 {
194 	vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
195 	vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
196 }
197 
198 static int
199 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
200 				  struct sockaddr_vm *src,
201 				  struct sockaddr_vm *dst,
202 				  enum vmci_transport_packet_type type,
203 				  u64 size,
204 				  u64 mode,
205 				  struct vmci_transport_waiting_info *wait,
206 				  u16 proto,
207 				  struct vmci_handle handle,
208 				  bool convert_error)
209 {
210 	int err;
211 
212 	vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
213 				   proto, handle);
214 	err = vmci_datagram_send(&pkt->dg);
215 	if (convert_error && (err < 0))
216 		return vmci_transport_error_to_vsock_error(err);
217 
218 	return err;
219 }
220 
221 static int
222 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
223 				      enum vmci_transport_packet_type type,
224 				      u64 size,
225 				      u64 mode,
226 				      struct vmci_transport_waiting_info *wait,
227 				      struct vmci_handle handle)
228 {
229 	struct vmci_transport_packet reply;
230 	struct sockaddr_vm src, dst;
231 
232 	if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
233 		return 0;
234 	} else {
235 		vmci_transport_packet_get_addresses(pkt, &src, &dst);
236 		return __vmci_transport_send_control_pkt(&reply, &src, &dst,
237 							 type,
238 							 size, mode, wait,
239 							 VSOCK_PROTO_INVALID,
240 							 handle, true);
241 	}
242 }
243 
244 static int
245 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
246 				   struct sockaddr_vm *dst,
247 				   enum vmci_transport_packet_type type,
248 				   u64 size,
249 				   u64 mode,
250 				   struct vmci_transport_waiting_info *wait,
251 				   struct vmci_handle handle)
252 {
253 	/* Note that it is safe to use a single packet across all CPUs since
254 	 * two tasklets of the same type are guaranteed to not ever run
255 	 * simultaneously. If that ever changes, or VMCI stops using tasklets,
256 	 * we can use per-cpu packets.
257 	 */
258 	static struct vmci_transport_packet pkt;
259 
260 	return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
261 						 size, mode, wait,
262 						 VSOCK_PROTO_INVALID, handle,
263 						 false);
264 }
265 
266 static int
267 vmci_transport_send_control_pkt(struct sock *sk,
268 				enum vmci_transport_packet_type type,
269 				u64 size,
270 				u64 mode,
271 				struct vmci_transport_waiting_info *wait,
272 				u16 proto,
273 				struct vmci_handle handle)
274 {
275 	struct vmci_transport_packet *pkt;
276 	struct vsock_sock *vsk;
277 	int err;
278 
279 	vsk = vsock_sk(sk);
280 
281 	if (!vsock_addr_bound(&vsk->local_addr))
282 		return -EINVAL;
283 
284 	if (!vsock_addr_bound(&vsk->remote_addr))
285 		return -EINVAL;
286 
287 	pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
288 	if (!pkt)
289 		return -ENOMEM;
290 
291 	err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
292 						&vsk->remote_addr, type, size,
293 						mode, wait, proto, handle,
294 						true);
295 	kfree(pkt);
296 
297 	return err;
298 }
299 
300 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
301 					struct sockaddr_vm *src,
302 					struct vmci_transport_packet *pkt)
303 {
304 	if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
305 		return 0;
306 	return vmci_transport_send_control_pkt_bh(
307 					dst, src,
308 					VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
309 					0, NULL, VMCI_INVALID_HANDLE);
310 }
311 
312 static int vmci_transport_send_reset(struct sock *sk,
313 				     struct vmci_transport_packet *pkt)
314 {
315 	if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
316 		return 0;
317 	return vmci_transport_send_control_pkt(sk,
318 					VMCI_TRANSPORT_PACKET_TYPE_RST,
319 					0, 0, NULL, VSOCK_PROTO_INVALID,
320 					VMCI_INVALID_HANDLE);
321 }
322 
323 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
324 {
325 	return vmci_transport_send_control_pkt(
326 					sk,
327 					VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
328 					size, 0, NULL,
329 					VSOCK_PROTO_INVALID,
330 					VMCI_INVALID_HANDLE);
331 }
332 
333 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
334 					  u16 version)
335 {
336 	return vmci_transport_send_control_pkt(
337 					sk,
338 					VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
339 					size, 0, NULL, version,
340 					VMCI_INVALID_HANDLE);
341 }
342 
343 static int vmci_transport_send_qp_offer(struct sock *sk,
344 					struct vmci_handle handle)
345 {
346 	return vmci_transport_send_control_pkt(
347 					sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
348 					0, NULL,
349 					VSOCK_PROTO_INVALID, handle);
350 }
351 
352 static int vmci_transport_send_attach(struct sock *sk,
353 				      struct vmci_handle handle)
354 {
355 	return vmci_transport_send_control_pkt(
356 					sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
357 					0, 0, NULL, VSOCK_PROTO_INVALID,
358 					handle);
359 }
360 
361 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
362 {
363 	return vmci_transport_reply_control_pkt_fast(
364 						pkt,
365 						VMCI_TRANSPORT_PACKET_TYPE_RST,
366 						0, 0, NULL,
367 						VMCI_INVALID_HANDLE);
368 }
369 
370 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
371 					  struct sockaddr_vm *src)
372 {
373 	return vmci_transport_send_control_pkt_bh(
374 					dst, src,
375 					VMCI_TRANSPORT_PACKET_TYPE_INVALID,
376 					0, 0, NULL, VMCI_INVALID_HANDLE);
377 }
378 
379 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
380 				 struct sockaddr_vm *src)
381 {
382 	return vmci_transport_send_control_pkt_bh(
383 					dst, src,
384 					VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
385 					0, NULL, VMCI_INVALID_HANDLE);
386 }
387 
388 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
389 				struct sockaddr_vm *src)
390 {
391 	return vmci_transport_send_control_pkt_bh(
392 					dst, src,
393 					VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
394 					0, NULL, VMCI_INVALID_HANDLE);
395 }
396 
397 int vmci_transport_send_wrote(struct sock *sk)
398 {
399 	return vmci_transport_send_control_pkt(
400 					sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
401 					0, NULL, VSOCK_PROTO_INVALID,
402 					VMCI_INVALID_HANDLE);
403 }
404 
405 int vmci_transport_send_read(struct sock *sk)
406 {
407 	return vmci_transport_send_control_pkt(
408 					sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
409 					0, NULL, VSOCK_PROTO_INVALID,
410 					VMCI_INVALID_HANDLE);
411 }
412 
413 int vmci_transport_send_waiting_write(struct sock *sk,
414 				      struct vmci_transport_waiting_info *wait)
415 {
416 	return vmci_transport_send_control_pkt(
417 				sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
418 				0, 0, wait, VSOCK_PROTO_INVALID,
419 				VMCI_INVALID_HANDLE);
420 }
421 
422 int vmci_transport_send_waiting_read(struct sock *sk,
423 				     struct vmci_transport_waiting_info *wait)
424 {
425 	return vmci_transport_send_control_pkt(
426 				sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
427 				0, 0, wait, VSOCK_PROTO_INVALID,
428 				VMCI_INVALID_HANDLE);
429 }
430 
431 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
432 {
433 	return vmci_transport_send_control_pkt(
434 					&vsk->sk,
435 					VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
436 					0, mode, NULL,
437 					VSOCK_PROTO_INVALID,
438 					VMCI_INVALID_HANDLE);
439 }
440 
441 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
442 {
443 	return vmci_transport_send_control_pkt(sk,
444 					VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
445 					size, 0, NULL,
446 					VSOCK_PROTO_INVALID,
447 					VMCI_INVALID_HANDLE);
448 }
449 
450 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
451 					     u16 version)
452 {
453 	return vmci_transport_send_control_pkt(
454 					sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
455 					size, 0, NULL, version,
456 					VMCI_INVALID_HANDLE);
457 }
458 
459 static struct sock *vmci_transport_get_pending(
460 					struct sock *listener,
461 					struct vmci_transport_packet *pkt)
462 {
463 	struct vsock_sock *vlistener;
464 	struct vsock_sock *vpending;
465 	struct sock *pending;
466 	struct sockaddr_vm src;
467 
468 	vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
469 
470 	vlistener = vsock_sk(listener);
471 
472 	list_for_each_entry(vpending, &vlistener->pending_links,
473 			    pending_links) {
474 		if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
475 		    pkt->dst_port == vpending->local_addr.svm_port) {
476 			pending = sk_vsock(vpending);
477 			sock_hold(pending);
478 			goto found;
479 		}
480 	}
481 
482 	pending = NULL;
483 found:
484 	return pending;
485 
486 }
487 
488 static void vmci_transport_release_pending(struct sock *pending)
489 {
490 	sock_put(pending);
491 }
492 
493 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
494  * trusted sockets 2) sockets from applications running as the same user as the
495  * VM (this is only true for the host side and only when using hosted products)
496  */
497 
498 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
499 {
500 	return vsock->trusted ||
501 	       vmci_is_context_owner(peer_cid, vsock->owner->uid);
502 }
503 
504 /* We allow sending datagrams to and receiving datagrams from a restricted VM
505  * only if it is trusted as described in vmci_transport_is_trusted.
506  */
507 
508 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
509 {
510 	if (VMADDR_CID_HYPERVISOR == peer_cid)
511 		return true;
512 
513 	if (vsock->cached_peer != peer_cid) {
514 		vsock->cached_peer = peer_cid;
515 		if (!vmci_transport_is_trusted(vsock, peer_cid) &&
516 		    (vmci_context_get_priv_flags(peer_cid) &
517 		     VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
518 			vsock->cached_peer_allow_dgram = false;
519 		} else {
520 			vsock->cached_peer_allow_dgram = true;
521 		}
522 	}
523 
524 	return vsock->cached_peer_allow_dgram;
525 }
526 
527 static int
528 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
529 				struct vmci_handle *handle,
530 				u64 produce_size,
531 				u64 consume_size,
532 				u32 peer, u32 flags, bool trusted)
533 {
534 	int err = 0;
535 
536 	if (trusted) {
537 		/* Try to allocate our queue pair as trusted. This will only
538 		 * work if vsock is running in the host.
539 		 */
540 
541 		err = vmci_qpair_alloc(qpair, handle, produce_size,
542 				       consume_size,
543 				       peer, flags,
544 				       VMCI_PRIVILEGE_FLAG_TRUSTED);
545 		if (err != VMCI_ERROR_NO_ACCESS)
546 			goto out;
547 
548 	}
549 
550 	err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
551 			       peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
552 out:
553 	if (err < 0) {
554 		pr_err("Could not attach to queue pair with %d\n",
555 		       err);
556 		err = vmci_transport_error_to_vsock_error(err);
557 	}
558 
559 	return err;
560 }
561 
562 static int
563 vmci_transport_datagram_create_hnd(u32 resource_id,
564 				   u32 flags,
565 				   vmci_datagram_recv_cb recv_cb,
566 				   void *client_data,
567 				   struct vmci_handle *out_handle)
568 {
569 	int err = 0;
570 
571 	/* Try to allocate our datagram handler as trusted. This will only work
572 	 * if vsock is running in the host.
573 	 */
574 
575 	err = vmci_datagram_create_handle_priv(resource_id, flags,
576 					       VMCI_PRIVILEGE_FLAG_TRUSTED,
577 					       recv_cb,
578 					       client_data, out_handle);
579 
580 	if (err == VMCI_ERROR_NO_ACCESS)
581 		err = vmci_datagram_create_handle(resource_id, flags,
582 						  recv_cb, client_data,
583 						  out_handle);
584 
585 	return err;
586 }
587 
588 /* This is invoked as part of a tasklet that's scheduled when the VMCI
589  * interrupt fires.  This is run in bottom-half context and if it ever needs to
590  * sleep it should defer that work to a work queue.
591  */
592 
593 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
594 {
595 	struct sock *sk;
596 	size_t size;
597 	struct sk_buff *skb;
598 	struct vsock_sock *vsk;
599 
600 	sk = (struct sock *)data;
601 
602 	/* This handler is privileged when this module is running on the host.
603 	 * We will get datagrams from all endpoints (even VMs that are in a
604 	 * restricted context). If we get one from a restricted context then
605 	 * the destination socket must be trusted.
606 	 *
607 	 * NOTE: We access the socket struct without holding the lock here.
608 	 * This is ok because the field we are interested is never modified
609 	 * outside of the create and destruct socket functions.
610 	 */
611 	vsk = vsock_sk(sk);
612 	if (!vmci_transport_allow_dgram(vsk, dg->src.context))
613 		return VMCI_ERROR_NO_ACCESS;
614 
615 	size = VMCI_DG_SIZE(dg);
616 
617 	/* Attach the packet to the socket's receive queue as an sk_buff. */
618 	skb = alloc_skb(size, GFP_ATOMIC);
619 	if (!skb)
620 		return VMCI_ERROR_NO_MEM;
621 
622 	/* sk_receive_skb() will do a sock_put(), so hold here. */
623 	sock_hold(sk);
624 	skb_put(skb, size);
625 	memcpy(skb->data, dg, size);
626 	sk_receive_skb(sk, skb, 0);
627 
628 	return VMCI_SUCCESS;
629 }
630 
631 static bool vmci_transport_stream_allow(u32 cid, u32 port)
632 {
633 	static const u32 non_socket_contexts[] = {
634 		VMADDR_CID_RESERVED,
635 	};
636 	int i;
637 
638 	BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
639 
640 	for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
641 		if (cid == non_socket_contexts[i])
642 			return false;
643 	}
644 
645 	return true;
646 }
647 
648 /* This is invoked as part of a tasklet that's scheduled when the VMCI
649  * interrupt fires.  This is run in bottom-half context but it defers most of
650  * its work to the packet handling work queue.
651  */
652 
653 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
654 {
655 	struct sock *sk;
656 	struct sockaddr_vm dst;
657 	struct sockaddr_vm src;
658 	struct vmci_transport_packet *pkt;
659 	struct vsock_sock *vsk;
660 	bool bh_process_pkt;
661 	int err;
662 
663 	sk = NULL;
664 	err = VMCI_SUCCESS;
665 	bh_process_pkt = false;
666 
667 	/* Ignore incoming packets from contexts without sockets, or resources
668 	 * that aren't vsock implementations.
669 	 */
670 
671 	if (!vmci_transport_stream_allow(dg->src.context, -1)
672 	    || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
673 		return VMCI_ERROR_NO_ACCESS;
674 
675 	if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
676 		/* Drop datagrams that do not contain full VSock packets. */
677 		return VMCI_ERROR_INVALID_ARGS;
678 
679 	pkt = (struct vmci_transport_packet *)dg;
680 
681 	/* Find the socket that should handle this packet.  First we look for a
682 	 * connected socket and if there is none we look for a socket bound to
683 	 * the destintation address.
684 	 */
685 	vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
686 	vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
687 
688 	sk = vsock_find_connected_socket(&src, &dst);
689 	if (!sk) {
690 		sk = vsock_find_bound_socket(&dst);
691 		if (!sk) {
692 			/* We could not find a socket for this specified
693 			 * address.  If this packet is a RST, we just drop it.
694 			 * If it is another packet, we send a RST.  Note that
695 			 * we do not send a RST reply to RSTs so that we do not
696 			 * continually send RSTs between two endpoints.
697 			 *
698 			 * Note that since this is a reply, dst is src and src
699 			 * is dst.
700 			 */
701 			if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
702 				pr_err("unable to send reset\n");
703 
704 			err = VMCI_ERROR_NOT_FOUND;
705 			goto out;
706 		}
707 	}
708 
709 	/* If the received packet type is beyond all types known to this
710 	 * implementation, reply with an invalid message.  Hopefully this will
711 	 * help when implementing backwards compatibility in the future.
712 	 */
713 	if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
714 		vmci_transport_send_invalid_bh(&dst, &src);
715 		err = VMCI_ERROR_INVALID_ARGS;
716 		goto out;
717 	}
718 
719 	/* This handler is privileged when this module is running on the host.
720 	 * We will get datagram connect requests from all endpoints (even VMs
721 	 * that are in a restricted context). If we get one from a restricted
722 	 * context then the destination socket must be trusted.
723 	 *
724 	 * NOTE: We access the socket struct without holding the lock here.
725 	 * This is ok because the field we are interested is never modified
726 	 * outside of the create and destruct socket functions.
727 	 */
728 	vsk = vsock_sk(sk);
729 	if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
730 		err = VMCI_ERROR_NO_ACCESS;
731 		goto out;
732 	}
733 
734 	/* We do most everything in a work queue, but let's fast path the
735 	 * notification of reads and writes to help data transfer performance.
736 	 * We can only do this if there is no process context code executing
737 	 * for this socket since that may change the state.
738 	 */
739 	bh_lock_sock(sk);
740 
741 	if (!sock_owned_by_user(sk)) {
742 		/* The local context ID may be out of date, update it. */
743 		vsk->local_addr.svm_cid = dst.svm_cid;
744 
745 		if (sk->sk_state == TCP_ESTABLISHED)
746 			vmci_trans(vsk)->notify_ops->handle_notify_pkt(
747 					sk, pkt, true, &dst, &src,
748 					&bh_process_pkt);
749 	}
750 
751 	bh_unlock_sock(sk);
752 
753 	if (!bh_process_pkt) {
754 		struct vmci_transport_recv_pkt_info *recv_pkt_info;
755 
756 		recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
757 		if (!recv_pkt_info) {
758 			if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
759 				pr_err("unable to send reset\n");
760 
761 			err = VMCI_ERROR_NO_MEM;
762 			goto out;
763 		}
764 
765 		recv_pkt_info->sk = sk;
766 		memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
767 		INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
768 
769 		schedule_work(&recv_pkt_info->work);
770 		/* Clear sk so that the reference count incremented by one of
771 		 * the Find functions above is not decremented below.  We need
772 		 * that reference count for the packet handler we've scheduled
773 		 * to run.
774 		 */
775 		sk = NULL;
776 	}
777 
778 out:
779 	if (sk)
780 		sock_put(sk);
781 
782 	return err;
783 }
784 
785 static void vmci_transport_handle_detach(struct sock *sk)
786 {
787 	struct vsock_sock *vsk;
788 
789 	vsk = vsock_sk(sk);
790 	if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
791 		sock_set_flag(sk, SOCK_DONE);
792 
793 		/* On a detach the peer will not be sending or receiving
794 		 * anymore.
795 		 */
796 		vsk->peer_shutdown = SHUTDOWN_MASK;
797 
798 		/* We should not be sending anymore since the peer won't be
799 		 * there to receive, but we can still receive if there is data
800 		 * left in our consume queue. If the local endpoint is a host,
801 		 * we can't call vsock_stream_has_data, since that may block,
802 		 * but a host endpoint can't read data once the VM has
803 		 * detached, so there is no available data in that case.
804 		 */
805 		if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
806 		    vsock_stream_has_data(vsk) <= 0) {
807 			if (sk->sk_state == TCP_SYN_SENT) {
808 				/* The peer may detach from a queue pair while
809 				 * we are still in the connecting state, i.e.,
810 				 * if the peer VM is killed after attaching to
811 				 * a queue pair, but before we complete the
812 				 * handshake. In that case, we treat the detach
813 				 * event like a reset.
814 				 */
815 
816 				sk->sk_state = TCP_CLOSE;
817 				sk->sk_err = ECONNRESET;
818 				sk->sk_error_report(sk);
819 				return;
820 			}
821 			sk->sk_state = TCP_CLOSE;
822 		}
823 		sk->sk_state_change(sk);
824 	}
825 }
826 
827 static void vmci_transport_peer_detach_cb(u32 sub_id,
828 					  const struct vmci_event_data *e_data,
829 					  void *client_data)
830 {
831 	struct vmci_transport *trans = client_data;
832 	const struct vmci_event_payload_qp *e_payload;
833 
834 	e_payload = vmci_event_data_const_payload(e_data);
835 
836 	/* XXX This is lame, we should provide a way to lookup sockets by
837 	 * qp_handle.
838 	 */
839 	if (vmci_handle_is_invalid(e_payload->handle) ||
840 	    !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
841 		return;
842 
843 	/* We don't ask for delayed CBs when we subscribe to this event (we
844 	 * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
845 	 * guarantees in that case about what context we might be running in,
846 	 * so it could be BH or process, blockable or non-blockable.  So we
847 	 * need to account for all possible contexts here.
848 	 */
849 	spin_lock_bh(&trans->lock);
850 	if (!trans->sk)
851 		goto out;
852 
853 	/* Apart from here, trans->lock is only grabbed as part of sk destruct,
854 	 * where trans->sk isn't locked.
855 	 */
856 	bh_lock_sock(trans->sk);
857 
858 	vmci_transport_handle_detach(trans->sk);
859 
860 	bh_unlock_sock(trans->sk);
861  out:
862 	spin_unlock_bh(&trans->lock);
863 }
864 
865 static void vmci_transport_qp_resumed_cb(u32 sub_id,
866 					 const struct vmci_event_data *e_data,
867 					 void *client_data)
868 {
869 	vsock_for_each_connected_socket(vmci_transport_handle_detach);
870 }
871 
872 static void vmci_transport_recv_pkt_work(struct work_struct *work)
873 {
874 	struct vmci_transport_recv_pkt_info *recv_pkt_info;
875 	struct vmci_transport_packet *pkt;
876 	struct sock *sk;
877 
878 	recv_pkt_info =
879 		container_of(work, struct vmci_transport_recv_pkt_info, work);
880 	sk = recv_pkt_info->sk;
881 	pkt = &recv_pkt_info->pkt;
882 
883 	lock_sock(sk);
884 
885 	/* The local context ID may be out of date. */
886 	vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
887 
888 	switch (sk->sk_state) {
889 	case TCP_LISTEN:
890 		vmci_transport_recv_listen(sk, pkt);
891 		break;
892 	case TCP_SYN_SENT:
893 		/* Processing of pending connections for servers goes through
894 		 * the listening socket, so see vmci_transport_recv_listen()
895 		 * for that path.
896 		 */
897 		vmci_transport_recv_connecting_client(sk, pkt);
898 		break;
899 	case TCP_ESTABLISHED:
900 		vmci_transport_recv_connected(sk, pkt);
901 		break;
902 	default:
903 		/* Because this function does not run in the same context as
904 		 * vmci_transport_recv_stream_cb it is possible that the
905 		 * socket has closed. We need to let the other side know or it
906 		 * could be sitting in a connect and hang forever. Send a
907 		 * reset to prevent that.
908 		 */
909 		vmci_transport_send_reset(sk, pkt);
910 		break;
911 	}
912 
913 	release_sock(sk);
914 	kfree(recv_pkt_info);
915 	/* Release reference obtained in the stream callback when we fetched
916 	 * this socket out of the bound or connected list.
917 	 */
918 	sock_put(sk);
919 }
920 
921 static int vmci_transport_recv_listen(struct sock *sk,
922 				      struct vmci_transport_packet *pkt)
923 {
924 	struct sock *pending;
925 	struct vsock_sock *vpending;
926 	int err;
927 	u64 qp_size;
928 	bool old_request = false;
929 	bool old_pkt_proto = false;
930 
931 	err = 0;
932 
933 	/* Because we are in the listen state, we could be receiving a packet
934 	 * for ourself or any previous connection requests that we received.
935 	 * If it's the latter, we try to find a socket in our list of pending
936 	 * connections and, if we do, call the appropriate handler for the
937 	 * state that that socket is in.  Otherwise we try to service the
938 	 * connection request.
939 	 */
940 	pending = vmci_transport_get_pending(sk, pkt);
941 	if (pending) {
942 		lock_sock(pending);
943 
944 		/* The local context ID may be out of date. */
945 		vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
946 
947 		switch (pending->sk_state) {
948 		case TCP_SYN_SENT:
949 			err = vmci_transport_recv_connecting_server(sk,
950 								    pending,
951 								    pkt);
952 			break;
953 		default:
954 			vmci_transport_send_reset(pending, pkt);
955 			err = -EINVAL;
956 		}
957 
958 		if (err < 0)
959 			vsock_remove_pending(sk, pending);
960 
961 		release_sock(pending);
962 		vmci_transport_release_pending(pending);
963 
964 		return err;
965 	}
966 
967 	/* The listen state only accepts connection requests.  Reply with a
968 	 * reset unless we received a reset.
969 	 */
970 
971 	if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
972 	      pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
973 		vmci_transport_reply_reset(pkt);
974 		return -EINVAL;
975 	}
976 
977 	if (pkt->u.size == 0) {
978 		vmci_transport_reply_reset(pkt);
979 		return -EINVAL;
980 	}
981 
982 	/* If this socket can't accommodate this connection request, we send a
983 	 * reset.  Otherwise we create and initialize a child socket and reply
984 	 * with a connection negotiation.
985 	 */
986 	if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
987 		vmci_transport_reply_reset(pkt);
988 		return -ECONNREFUSED;
989 	}
990 
991 	pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
992 				 sk->sk_type, 0);
993 	if (!pending) {
994 		vmci_transport_send_reset(sk, pkt);
995 		return -ENOMEM;
996 	}
997 
998 	vpending = vsock_sk(pending);
999 
1000 	vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1001 			pkt->dst_port);
1002 	vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1003 			pkt->src_port);
1004 
1005 	/* If the proposed size fits within our min/max, accept it. Otherwise
1006 	 * propose our own size.
1007 	 */
1008 	if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1009 	    pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1010 		qp_size = pkt->u.size;
1011 	} else {
1012 		qp_size = vmci_trans(vpending)->queue_pair_size;
1013 	}
1014 
1015 	/* Figure out if we are using old or new requests based on the
1016 	 * overrides pkt types sent by our peer.
1017 	 */
1018 	if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1019 		old_request = old_pkt_proto;
1020 	} else {
1021 		if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1022 			old_request = true;
1023 		else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1024 			old_request = false;
1025 
1026 	}
1027 
1028 	if (old_request) {
1029 		/* Handle a REQUEST (or override) */
1030 		u16 version = VSOCK_PROTO_INVALID;
1031 		if (vmci_transport_proto_to_notify_struct(
1032 			pending, &version, true))
1033 			err = vmci_transport_send_negotiate(pending, qp_size);
1034 		else
1035 			err = -EINVAL;
1036 
1037 	} else {
1038 		/* Handle a REQUEST2 (or override) */
1039 		int proto_int = pkt->proto;
1040 		int pos;
1041 		u16 active_proto_version = 0;
1042 
1043 		/* The list of possible protocols is the intersection of all
1044 		 * protocols the client supports ... plus all the protocols we
1045 		 * support.
1046 		 */
1047 		proto_int &= vmci_transport_new_proto_supported_versions();
1048 
1049 		/* We choose the highest possible protocol version and use that
1050 		 * one.
1051 		 */
1052 		pos = fls(proto_int);
1053 		if (pos) {
1054 			active_proto_version = (1 << (pos - 1));
1055 			if (vmci_transport_proto_to_notify_struct(
1056 				pending, &active_proto_version, false))
1057 				err = vmci_transport_send_negotiate2(pending,
1058 							qp_size,
1059 							active_proto_version);
1060 			else
1061 				err = -EINVAL;
1062 
1063 		} else {
1064 			err = -EINVAL;
1065 		}
1066 	}
1067 
1068 	if (err < 0) {
1069 		vmci_transport_send_reset(sk, pkt);
1070 		sock_put(pending);
1071 		err = vmci_transport_error_to_vsock_error(err);
1072 		goto out;
1073 	}
1074 
1075 	vsock_add_pending(sk, pending);
1076 	sk->sk_ack_backlog++;
1077 
1078 	pending->sk_state = TCP_SYN_SENT;
1079 	vmci_trans(vpending)->produce_size =
1080 		vmci_trans(vpending)->consume_size = qp_size;
1081 	vmci_trans(vpending)->queue_pair_size = qp_size;
1082 
1083 	vmci_trans(vpending)->notify_ops->process_request(pending);
1084 
1085 	/* We might never receive another message for this socket and it's not
1086 	 * connected to any process, so we have to ensure it gets cleaned up
1087 	 * ourself.  Our delayed work function will take care of that.  Note
1088 	 * that we do not ever cancel this function since we have few
1089 	 * guarantees about its state when calling cancel_delayed_work().
1090 	 * Instead we hold a reference on the socket for that function and make
1091 	 * it capable of handling cases where it needs to do nothing but
1092 	 * release that reference.
1093 	 */
1094 	vpending->listener = sk;
1095 	sock_hold(sk);
1096 	sock_hold(pending);
1097 	schedule_delayed_work(&vpending->pending_work, HZ);
1098 
1099 out:
1100 	return err;
1101 }
1102 
1103 static int
1104 vmci_transport_recv_connecting_server(struct sock *listener,
1105 				      struct sock *pending,
1106 				      struct vmci_transport_packet *pkt)
1107 {
1108 	struct vsock_sock *vpending;
1109 	struct vmci_handle handle;
1110 	struct vmci_qp *qpair;
1111 	bool is_local;
1112 	u32 flags;
1113 	u32 detach_sub_id;
1114 	int err;
1115 	int skerr;
1116 
1117 	vpending = vsock_sk(pending);
1118 	detach_sub_id = VMCI_INVALID_ID;
1119 
1120 	switch (pkt->type) {
1121 	case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1122 		if (vmci_handle_is_invalid(pkt->u.handle)) {
1123 			vmci_transport_send_reset(pending, pkt);
1124 			skerr = EPROTO;
1125 			err = -EINVAL;
1126 			goto destroy;
1127 		}
1128 		break;
1129 	default:
1130 		/* Close and cleanup the connection. */
1131 		vmci_transport_send_reset(pending, pkt);
1132 		skerr = EPROTO;
1133 		err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1134 		goto destroy;
1135 	}
1136 
1137 	/* In order to complete the connection we need to attach to the offered
1138 	 * queue pair and send an attach notification.  We also subscribe to the
1139 	 * detach event so we know when our peer goes away, and we do that
1140 	 * before attaching so we don't miss an event.  If all this succeeds,
1141 	 * we update our state and wakeup anything waiting in accept() for a
1142 	 * connection.
1143 	 */
1144 
1145 	/* We don't care about attach since we ensure the other side has
1146 	 * attached by specifying the ATTACH_ONLY flag below.
1147 	 */
1148 	err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1149 				   vmci_transport_peer_detach_cb,
1150 				   vmci_trans(vpending), &detach_sub_id);
1151 	if (err < VMCI_SUCCESS) {
1152 		vmci_transport_send_reset(pending, pkt);
1153 		err = vmci_transport_error_to_vsock_error(err);
1154 		skerr = -err;
1155 		goto destroy;
1156 	}
1157 
1158 	vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1159 
1160 	/* Now attach to the queue pair the client created. */
1161 	handle = pkt->u.handle;
1162 
1163 	/* vpending->local_addr always has a context id so we do not need to
1164 	 * worry about VMADDR_CID_ANY in this case.
1165 	 */
1166 	is_local =
1167 	    vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1168 	flags = VMCI_QPFLAG_ATTACH_ONLY;
1169 	flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1170 
1171 	err = vmci_transport_queue_pair_alloc(
1172 					&qpair,
1173 					&handle,
1174 					vmci_trans(vpending)->produce_size,
1175 					vmci_trans(vpending)->consume_size,
1176 					pkt->dg.src.context,
1177 					flags,
1178 					vmci_transport_is_trusted(
1179 						vpending,
1180 						vpending->remote_addr.svm_cid));
1181 	if (err < 0) {
1182 		vmci_transport_send_reset(pending, pkt);
1183 		skerr = -err;
1184 		goto destroy;
1185 	}
1186 
1187 	vmci_trans(vpending)->qp_handle = handle;
1188 	vmci_trans(vpending)->qpair = qpair;
1189 
1190 	/* When we send the attach message, we must be ready to handle incoming
1191 	 * control messages on the newly connected socket. So we move the
1192 	 * pending socket to the connected state before sending the attach
1193 	 * message. Otherwise, an incoming packet triggered by the attach being
1194 	 * received by the peer may be processed concurrently with what happens
1195 	 * below after sending the attach message, and that incoming packet
1196 	 * will find the listening socket instead of the (currently) pending
1197 	 * socket. Note that enqueueing the socket increments the reference
1198 	 * count, so even if a reset comes before the connection is accepted,
1199 	 * the socket will be valid until it is removed from the queue.
1200 	 *
1201 	 * If we fail sending the attach below, we remove the socket from the
1202 	 * connected list and move the socket to TCP_CLOSE before
1203 	 * releasing the lock, so a pending slow path processing of an incoming
1204 	 * packet will not see the socket in the connected state in that case.
1205 	 */
1206 	pending->sk_state = TCP_ESTABLISHED;
1207 
1208 	vsock_insert_connected(vpending);
1209 
1210 	/* Notify our peer of our attach. */
1211 	err = vmci_transport_send_attach(pending, handle);
1212 	if (err < 0) {
1213 		vsock_remove_connected(vpending);
1214 		pr_err("Could not send attach\n");
1215 		vmci_transport_send_reset(pending, pkt);
1216 		err = vmci_transport_error_to_vsock_error(err);
1217 		skerr = -err;
1218 		goto destroy;
1219 	}
1220 
1221 	/* We have a connection. Move the now connected socket from the
1222 	 * listener's pending list to the accept queue so callers of accept()
1223 	 * can find it.
1224 	 */
1225 	vsock_remove_pending(listener, pending);
1226 	vsock_enqueue_accept(listener, pending);
1227 
1228 	/* Callers of accept() will be be waiting on the listening socket, not
1229 	 * the pending socket.
1230 	 */
1231 	listener->sk_data_ready(listener);
1232 
1233 	return 0;
1234 
1235 destroy:
1236 	pending->sk_err = skerr;
1237 	pending->sk_state = TCP_CLOSE;
1238 	/* As long as we drop our reference, all necessary cleanup will handle
1239 	 * when the cleanup function drops its reference and our destruct
1240 	 * implementation is called.  Note that since the listen handler will
1241 	 * remove pending from the pending list upon our failure, the cleanup
1242 	 * function won't drop the additional reference, which is why we do it
1243 	 * here.
1244 	 */
1245 	sock_put(pending);
1246 
1247 	return err;
1248 }
1249 
1250 static int
1251 vmci_transport_recv_connecting_client(struct sock *sk,
1252 				      struct vmci_transport_packet *pkt)
1253 {
1254 	struct vsock_sock *vsk;
1255 	int err;
1256 	int skerr;
1257 
1258 	vsk = vsock_sk(sk);
1259 
1260 	switch (pkt->type) {
1261 	case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1262 		if (vmci_handle_is_invalid(pkt->u.handle) ||
1263 		    !vmci_handle_is_equal(pkt->u.handle,
1264 					  vmci_trans(vsk)->qp_handle)) {
1265 			skerr = EPROTO;
1266 			err = -EINVAL;
1267 			goto destroy;
1268 		}
1269 
1270 		/* Signify the socket is connected and wakeup the waiter in
1271 		 * connect(). Also place the socket in the connected table for
1272 		 * accounting (it can already be found since it's in the bound
1273 		 * table).
1274 		 */
1275 		sk->sk_state = TCP_ESTABLISHED;
1276 		sk->sk_socket->state = SS_CONNECTED;
1277 		vsock_insert_connected(vsk);
1278 		sk->sk_state_change(sk);
1279 
1280 		break;
1281 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1282 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1283 		if (pkt->u.size == 0
1284 		    || pkt->dg.src.context != vsk->remote_addr.svm_cid
1285 		    || pkt->src_port != vsk->remote_addr.svm_port
1286 		    || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1287 		    || vmci_trans(vsk)->qpair
1288 		    || vmci_trans(vsk)->produce_size != 0
1289 		    || vmci_trans(vsk)->consume_size != 0
1290 		    || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1291 			skerr = EPROTO;
1292 			err = -EINVAL;
1293 
1294 			goto destroy;
1295 		}
1296 
1297 		err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1298 		if (err) {
1299 			skerr = -err;
1300 			goto destroy;
1301 		}
1302 
1303 		break;
1304 	case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1305 		err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1306 		if (err) {
1307 			skerr = -err;
1308 			goto destroy;
1309 		}
1310 
1311 		break;
1312 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
1313 		/* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1314 		 * continue processing here after they sent an INVALID packet.
1315 		 * This meant that we got a RST after the INVALID. We ignore a
1316 		 * RST after an INVALID. The common code doesn't send the RST
1317 		 * ... so we can hang if an old version of the common code
1318 		 * fails between getting a REQUEST and sending an OFFER back.
1319 		 * Not much we can do about it... except hope that it doesn't
1320 		 * happen.
1321 		 */
1322 		if (vsk->ignore_connecting_rst) {
1323 			vsk->ignore_connecting_rst = false;
1324 		} else {
1325 			skerr = ECONNRESET;
1326 			err = 0;
1327 			goto destroy;
1328 		}
1329 
1330 		break;
1331 	default:
1332 		/* Close and cleanup the connection. */
1333 		skerr = EPROTO;
1334 		err = -EINVAL;
1335 		goto destroy;
1336 	}
1337 
1338 	return 0;
1339 
1340 destroy:
1341 	vmci_transport_send_reset(sk, pkt);
1342 
1343 	sk->sk_state = TCP_CLOSE;
1344 	sk->sk_err = skerr;
1345 	sk->sk_error_report(sk);
1346 	return err;
1347 }
1348 
1349 static int vmci_transport_recv_connecting_client_negotiate(
1350 					struct sock *sk,
1351 					struct vmci_transport_packet *pkt)
1352 {
1353 	int err;
1354 	struct vsock_sock *vsk;
1355 	struct vmci_handle handle;
1356 	struct vmci_qp *qpair;
1357 	u32 detach_sub_id;
1358 	bool is_local;
1359 	u32 flags;
1360 	bool old_proto = true;
1361 	bool old_pkt_proto;
1362 	u16 version;
1363 
1364 	vsk = vsock_sk(sk);
1365 	handle = VMCI_INVALID_HANDLE;
1366 	detach_sub_id = VMCI_INVALID_ID;
1367 
1368 	/* If we have gotten here then we should be past the point where old
1369 	 * linux vsock could have sent the bogus rst.
1370 	 */
1371 	vsk->sent_request = false;
1372 	vsk->ignore_connecting_rst = false;
1373 
1374 	/* Verify that we're OK with the proposed queue pair size */
1375 	if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1376 	    pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1377 		err = -EINVAL;
1378 		goto destroy;
1379 	}
1380 
1381 	/* At this point we know the CID the peer is using to talk to us. */
1382 
1383 	if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1384 		vsk->local_addr.svm_cid = pkt->dg.dst.context;
1385 
1386 	/* Setup the notify ops to be the highest supported version that both
1387 	 * the server and the client support.
1388 	 */
1389 
1390 	if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1391 		old_proto = old_pkt_proto;
1392 	} else {
1393 		if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1394 			old_proto = true;
1395 		else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1396 			old_proto = false;
1397 
1398 	}
1399 
1400 	if (old_proto)
1401 		version = VSOCK_PROTO_INVALID;
1402 	else
1403 		version = pkt->proto;
1404 
1405 	if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1406 		err = -EINVAL;
1407 		goto destroy;
1408 	}
1409 
1410 	/* Subscribe to detach events first.
1411 	 *
1412 	 * XXX We attach once for each queue pair created for now so it is easy
1413 	 * to find the socket (it's provided), but later we should only
1414 	 * subscribe once and add a way to lookup sockets by queue pair handle.
1415 	 */
1416 	err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1417 				   vmci_transport_peer_detach_cb,
1418 				   vmci_trans(vsk), &detach_sub_id);
1419 	if (err < VMCI_SUCCESS) {
1420 		err = vmci_transport_error_to_vsock_error(err);
1421 		goto destroy;
1422 	}
1423 
1424 	/* Make VMCI select the handle for us. */
1425 	handle = VMCI_INVALID_HANDLE;
1426 	is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1427 	flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1428 
1429 	err = vmci_transport_queue_pair_alloc(&qpair,
1430 					      &handle,
1431 					      pkt->u.size,
1432 					      pkt->u.size,
1433 					      vsk->remote_addr.svm_cid,
1434 					      flags,
1435 					      vmci_transport_is_trusted(
1436 						  vsk,
1437 						  vsk->
1438 						  remote_addr.svm_cid));
1439 	if (err < 0)
1440 		goto destroy;
1441 
1442 	err = vmci_transport_send_qp_offer(sk, handle);
1443 	if (err < 0) {
1444 		err = vmci_transport_error_to_vsock_error(err);
1445 		goto destroy;
1446 	}
1447 
1448 	vmci_trans(vsk)->qp_handle = handle;
1449 	vmci_trans(vsk)->qpair = qpair;
1450 
1451 	vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1452 		pkt->u.size;
1453 
1454 	vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1455 
1456 	vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1457 
1458 	return 0;
1459 
1460 destroy:
1461 	if (detach_sub_id != VMCI_INVALID_ID)
1462 		vmci_event_unsubscribe(detach_sub_id);
1463 
1464 	if (!vmci_handle_is_invalid(handle))
1465 		vmci_qpair_detach(&qpair);
1466 
1467 	return err;
1468 }
1469 
1470 static int
1471 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1472 					      struct vmci_transport_packet *pkt)
1473 {
1474 	int err = 0;
1475 	struct vsock_sock *vsk = vsock_sk(sk);
1476 
1477 	if (vsk->sent_request) {
1478 		vsk->sent_request = false;
1479 		vsk->ignore_connecting_rst = true;
1480 
1481 		err = vmci_transport_send_conn_request(
1482 			sk, vmci_trans(vsk)->queue_pair_size);
1483 		if (err < 0)
1484 			err = vmci_transport_error_to_vsock_error(err);
1485 		else
1486 			err = 0;
1487 
1488 	}
1489 
1490 	return err;
1491 }
1492 
1493 static int vmci_transport_recv_connected(struct sock *sk,
1494 					 struct vmci_transport_packet *pkt)
1495 {
1496 	struct vsock_sock *vsk;
1497 	bool pkt_processed = false;
1498 
1499 	/* In cases where we are closing the connection, it's sufficient to
1500 	 * mark the state change (and maybe error) and wake up any waiting
1501 	 * threads. Since this is a connected socket, it's owned by a user
1502 	 * process and will be cleaned up when the failure is passed back on
1503 	 * the current or next system call.  Our system call implementations
1504 	 * must therefore check for error and state changes on entry and when
1505 	 * being awoken.
1506 	 */
1507 	switch (pkt->type) {
1508 	case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1509 		if (pkt->u.mode) {
1510 			vsk = vsock_sk(sk);
1511 
1512 			vsk->peer_shutdown |= pkt->u.mode;
1513 			sk->sk_state_change(sk);
1514 		}
1515 		break;
1516 
1517 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
1518 		vsk = vsock_sk(sk);
1519 		/* It is possible that we sent our peer a message (e.g a
1520 		 * WAITING_READ) right before we got notified that the peer had
1521 		 * detached. If that happens then we can get a RST pkt back
1522 		 * from our peer even though there is data available for us to
1523 		 * read. In that case, don't shutdown the socket completely but
1524 		 * instead allow the local client to finish reading data off
1525 		 * the queuepair. Always treat a RST pkt in connected mode like
1526 		 * a clean shutdown.
1527 		 */
1528 		sock_set_flag(sk, SOCK_DONE);
1529 		vsk->peer_shutdown = SHUTDOWN_MASK;
1530 		if (vsock_stream_has_data(vsk) <= 0)
1531 			sk->sk_state = TCP_CLOSING;
1532 
1533 		sk->sk_state_change(sk);
1534 		break;
1535 
1536 	default:
1537 		vsk = vsock_sk(sk);
1538 		vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1539 				sk, pkt, false, NULL, NULL,
1540 				&pkt_processed);
1541 		if (!pkt_processed)
1542 			return -EINVAL;
1543 
1544 		break;
1545 	}
1546 
1547 	return 0;
1548 }
1549 
1550 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1551 				      struct vsock_sock *psk)
1552 {
1553 	vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1554 	if (!vsk->trans)
1555 		return -ENOMEM;
1556 
1557 	vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1558 	vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1559 	vmci_trans(vsk)->qpair = NULL;
1560 	vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1561 	vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1562 	vmci_trans(vsk)->notify_ops = NULL;
1563 	INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1564 	vmci_trans(vsk)->sk = &vsk->sk;
1565 	spin_lock_init(&vmci_trans(vsk)->lock);
1566 	if (psk) {
1567 		vmci_trans(vsk)->queue_pair_size =
1568 			vmci_trans(psk)->queue_pair_size;
1569 		vmci_trans(vsk)->queue_pair_min_size =
1570 			vmci_trans(psk)->queue_pair_min_size;
1571 		vmci_trans(vsk)->queue_pair_max_size =
1572 			vmci_trans(psk)->queue_pair_max_size;
1573 	} else {
1574 		vmci_trans(vsk)->queue_pair_size =
1575 			VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1576 		vmci_trans(vsk)->queue_pair_min_size =
1577 			 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1578 		vmci_trans(vsk)->queue_pair_max_size =
1579 			VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1580 	}
1581 
1582 	return 0;
1583 }
1584 
1585 static void vmci_transport_free_resources(struct list_head *transport_list)
1586 {
1587 	while (!list_empty(transport_list)) {
1588 		struct vmci_transport *transport =
1589 		    list_first_entry(transport_list, struct vmci_transport,
1590 				     elem);
1591 		list_del(&transport->elem);
1592 
1593 		if (transport->detach_sub_id != VMCI_INVALID_ID) {
1594 			vmci_event_unsubscribe(transport->detach_sub_id);
1595 			transport->detach_sub_id = VMCI_INVALID_ID;
1596 		}
1597 
1598 		if (!vmci_handle_is_invalid(transport->qp_handle)) {
1599 			vmci_qpair_detach(&transport->qpair);
1600 			transport->qp_handle = VMCI_INVALID_HANDLE;
1601 			transport->produce_size = 0;
1602 			transport->consume_size = 0;
1603 		}
1604 
1605 		kfree(transport);
1606 	}
1607 }
1608 
1609 static void vmci_transport_cleanup(struct work_struct *work)
1610 {
1611 	LIST_HEAD(pending);
1612 
1613 	spin_lock_bh(&vmci_transport_cleanup_lock);
1614 	list_replace_init(&vmci_transport_cleanup_list, &pending);
1615 	spin_unlock_bh(&vmci_transport_cleanup_lock);
1616 	vmci_transport_free_resources(&pending);
1617 }
1618 
1619 static void vmci_transport_destruct(struct vsock_sock *vsk)
1620 {
1621 	/* Ensure that the detach callback doesn't use the sk/vsk
1622 	 * we are about to destruct.
1623 	 */
1624 	spin_lock_bh(&vmci_trans(vsk)->lock);
1625 	vmci_trans(vsk)->sk = NULL;
1626 	spin_unlock_bh(&vmci_trans(vsk)->lock);
1627 
1628 	if (vmci_trans(vsk)->notify_ops)
1629 		vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1630 
1631 	spin_lock_bh(&vmci_transport_cleanup_lock);
1632 	list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1633 	spin_unlock_bh(&vmci_transport_cleanup_lock);
1634 	schedule_work(&vmci_transport_cleanup_work);
1635 
1636 	vsk->trans = NULL;
1637 }
1638 
1639 static void vmci_transport_release(struct vsock_sock *vsk)
1640 {
1641 	vsock_remove_sock(vsk);
1642 
1643 	if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1644 		vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1645 		vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1646 	}
1647 }
1648 
1649 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1650 				     struct sockaddr_vm *addr)
1651 {
1652 	u32 port;
1653 	u32 flags;
1654 	int err;
1655 
1656 	/* VMCI will select a resource ID for us if we provide
1657 	 * VMCI_INVALID_ID.
1658 	 */
1659 	port = addr->svm_port == VMADDR_PORT_ANY ?
1660 			VMCI_INVALID_ID : addr->svm_port;
1661 
1662 	if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1663 		return -EACCES;
1664 
1665 	flags = addr->svm_cid == VMADDR_CID_ANY ?
1666 				VMCI_FLAG_ANYCID_DG_HND : 0;
1667 
1668 	err = vmci_transport_datagram_create_hnd(port, flags,
1669 						 vmci_transport_recv_dgram_cb,
1670 						 &vsk->sk,
1671 						 &vmci_trans(vsk)->dg_handle);
1672 	if (err < VMCI_SUCCESS)
1673 		return vmci_transport_error_to_vsock_error(err);
1674 	vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1675 			vmci_trans(vsk)->dg_handle.resource);
1676 
1677 	return 0;
1678 }
1679 
1680 static int vmci_transport_dgram_enqueue(
1681 	struct vsock_sock *vsk,
1682 	struct sockaddr_vm *remote_addr,
1683 	struct msghdr *msg,
1684 	size_t len)
1685 {
1686 	int err;
1687 	struct vmci_datagram *dg;
1688 
1689 	if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1690 		return -EMSGSIZE;
1691 
1692 	if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1693 		return -EPERM;
1694 
1695 	/* Allocate a buffer for the user's message and our packet header. */
1696 	dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1697 	if (!dg)
1698 		return -ENOMEM;
1699 
1700 	memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1701 
1702 	dg->dst = vmci_make_handle(remote_addr->svm_cid,
1703 				   remote_addr->svm_port);
1704 	dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1705 				   vsk->local_addr.svm_port);
1706 	dg->payload_size = len;
1707 
1708 	err = vmci_datagram_send(dg);
1709 	kfree(dg);
1710 	if (err < 0)
1711 		return vmci_transport_error_to_vsock_error(err);
1712 
1713 	return err - sizeof(*dg);
1714 }
1715 
1716 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1717 					struct msghdr *msg, size_t len,
1718 					int flags)
1719 {
1720 	int err;
1721 	int noblock;
1722 	struct vmci_datagram *dg;
1723 	size_t payload_len;
1724 	struct sk_buff *skb;
1725 
1726 	noblock = flags & MSG_DONTWAIT;
1727 
1728 	if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1729 		return -EOPNOTSUPP;
1730 
1731 	/* Retrieve the head sk_buff from the socket's receive queue. */
1732 	err = 0;
1733 	skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1734 	if (!skb)
1735 		return err;
1736 
1737 	dg = (struct vmci_datagram *)skb->data;
1738 	if (!dg)
1739 		/* err is 0, meaning we read zero bytes. */
1740 		goto out;
1741 
1742 	payload_len = dg->payload_size;
1743 	/* Ensure the sk_buff matches the payload size claimed in the packet. */
1744 	if (payload_len != skb->len - sizeof(*dg)) {
1745 		err = -EINVAL;
1746 		goto out;
1747 	}
1748 
1749 	if (payload_len > len) {
1750 		payload_len = len;
1751 		msg->msg_flags |= MSG_TRUNC;
1752 	}
1753 
1754 	/* Place the datagram payload in the user's iovec. */
1755 	err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1756 	if (err)
1757 		goto out;
1758 
1759 	if (msg->msg_name) {
1760 		/* Provide the address of the sender. */
1761 		DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1762 		vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1763 		msg->msg_namelen = sizeof(*vm_addr);
1764 	}
1765 	err = payload_len;
1766 
1767 out:
1768 	skb_free_datagram(&vsk->sk, skb);
1769 	return err;
1770 }
1771 
1772 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1773 {
1774 	if (cid == VMADDR_CID_HYPERVISOR) {
1775 		/* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1776 		 * state and are allowed.
1777 		 */
1778 		return port == VMCI_UNITY_PBRPC_REGISTER;
1779 	}
1780 
1781 	return true;
1782 }
1783 
1784 static int vmci_transport_connect(struct vsock_sock *vsk)
1785 {
1786 	int err;
1787 	bool old_pkt_proto = false;
1788 	struct sock *sk = &vsk->sk;
1789 
1790 	if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1791 		old_pkt_proto) {
1792 		err = vmci_transport_send_conn_request(
1793 			sk, vmci_trans(vsk)->queue_pair_size);
1794 		if (err < 0) {
1795 			sk->sk_state = TCP_CLOSE;
1796 			return err;
1797 		}
1798 	} else {
1799 		int supported_proto_versions =
1800 			vmci_transport_new_proto_supported_versions();
1801 		err = vmci_transport_send_conn_request2(
1802 				sk, vmci_trans(vsk)->queue_pair_size,
1803 				supported_proto_versions);
1804 		if (err < 0) {
1805 			sk->sk_state = TCP_CLOSE;
1806 			return err;
1807 		}
1808 
1809 		vsk->sent_request = true;
1810 	}
1811 
1812 	return err;
1813 }
1814 
1815 static ssize_t vmci_transport_stream_dequeue(
1816 	struct vsock_sock *vsk,
1817 	struct msghdr *msg,
1818 	size_t len,
1819 	int flags)
1820 {
1821 	if (flags & MSG_PEEK)
1822 		return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1823 	else
1824 		return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1825 }
1826 
1827 static ssize_t vmci_transport_stream_enqueue(
1828 	struct vsock_sock *vsk,
1829 	struct msghdr *msg,
1830 	size_t len)
1831 {
1832 	return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1833 }
1834 
1835 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1836 {
1837 	return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1838 }
1839 
1840 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1841 {
1842 	return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1843 }
1844 
1845 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1846 {
1847 	return vmci_trans(vsk)->consume_size;
1848 }
1849 
1850 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1851 {
1852 	return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1853 }
1854 
1855 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1856 {
1857 	return vmci_trans(vsk)->queue_pair_size;
1858 }
1859 
1860 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1861 {
1862 	return vmci_trans(vsk)->queue_pair_min_size;
1863 }
1864 
1865 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1866 {
1867 	return vmci_trans(vsk)->queue_pair_max_size;
1868 }
1869 
1870 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1871 {
1872 	if (val < vmci_trans(vsk)->queue_pair_min_size)
1873 		vmci_trans(vsk)->queue_pair_min_size = val;
1874 	if (val > vmci_trans(vsk)->queue_pair_max_size)
1875 		vmci_trans(vsk)->queue_pair_max_size = val;
1876 	vmci_trans(vsk)->queue_pair_size = val;
1877 }
1878 
1879 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1880 					       u64 val)
1881 {
1882 	if (val > vmci_trans(vsk)->queue_pair_size)
1883 		vmci_trans(vsk)->queue_pair_size = val;
1884 	vmci_trans(vsk)->queue_pair_min_size = val;
1885 }
1886 
1887 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1888 					       u64 val)
1889 {
1890 	if (val < vmci_trans(vsk)->queue_pair_size)
1891 		vmci_trans(vsk)->queue_pair_size = val;
1892 	vmci_trans(vsk)->queue_pair_max_size = val;
1893 }
1894 
1895 static int vmci_transport_notify_poll_in(
1896 	struct vsock_sock *vsk,
1897 	size_t target,
1898 	bool *data_ready_now)
1899 {
1900 	return vmci_trans(vsk)->notify_ops->poll_in(
1901 			&vsk->sk, target, data_ready_now);
1902 }
1903 
1904 static int vmci_transport_notify_poll_out(
1905 	struct vsock_sock *vsk,
1906 	size_t target,
1907 	bool *space_available_now)
1908 {
1909 	return vmci_trans(vsk)->notify_ops->poll_out(
1910 			&vsk->sk, target, space_available_now);
1911 }
1912 
1913 static int vmci_transport_notify_recv_init(
1914 	struct vsock_sock *vsk,
1915 	size_t target,
1916 	struct vsock_transport_recv_notify_data *data)
1917 {
1918 	return vmci_trans(vsk)->notify_ops->recv_init(
1919 			&vsk->sk, target,
1920 			(struct vmci_transport_recv_notify_data *)data);
1921 }
1922 
1923 static int vmci_transport_notify_recv_pre_block(
1924 	struct vsock_sock *vsk,
1925 	size_t target,
1926 	struct vsock_transport_recv_notify_data *data)
1927 {
1928 	return vmci_trans(vsk)->notify_ops->recv_pre_block(
1929 			&vsk->sk, target,
1930 			(struct vmci_transport_recv_notify_data *)data);
1931 }
1932 
1933 static int vmci_transport_notify_recv_pre_dequeue(
1934 	struct vsock_sock *vsk,
1935 	size_t target,
1936 	struct vsock_transport_recv_notify_data *data)
1937 {
1938 	return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1939 			&vsk->sk, target,
1940 			(struct vmci_transport_recv_notify_data *)data);
1941 }
1942 
1943 static int vmci_transport_notify_recv_post_dequeue(
1944 	struct vsock_sock *vsk,
1945 	size_t target,
1946 	ssize_t copied,
1947 	bool data_read,
1948 	struct vsock_transport_recv_notify_data *data)
1949 {
1950 	return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1951 			&vsk->sk, target, copied, data_read,
1952 			(struct vmci_transport_recv_notify_data *)data);
1953 }
1954 
1955 static int vmci_transport_notify_send_init(
1956 	struct vsock_sock *vsk,
1957 	struct vsock_transport_send_notify_data *data)
1958 {
1959 	return vmci_trans(vsk)->notify_ops->send_init(
1960 			&vsk->sk,
1961 			(struct vmci_transport_send_notify_data *)data);
1962 }
1963 
1964 static int vmci_transport_notify_send_pre_block(
1965 	struct vsock_sock *vsk,
1966 	struct vsock_transport_send_notify_data *data)
1967 {
1968 	return vmci_trans(vsk)->notify_ops->send_pre_block(
1969 			&vsk->sk,
1970 			(struct vmci_transport_send_notify_data *)data);
1971 }
1972 
1973 static int vmci_transport_notify_send_pre_enqueue(
1974 	struct vsock_sock *vsk,
1975 	struct vsock_transport_send_notify_data *data)
1976 {
1977 	return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1978 			&vsk->sk,
1979 			(struct vmci_transport_send_notify_data *)data);
1980 }
1981 
1982 static int vmci_transport_notify_send_post_enqueue(
1983 	struct vsock_sock *vsk,
1984 	ssize_t written,
1985 	struct vsock_transport_send_notify_data *data)
1986 {
1987 	return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1988 			&vsk->sk, written,
1989 			(struct vmci_transport_send_notify_data *)data);
1990 }
1991 
1992 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1993 {
1994 	if (PROTOCOL_OVERRIDE != -1) {
1995 		if (PROTOCOL_OVERRIDE == 0)
1996 			*old_pkt_proto = true;
1997 		else
1998 			*old_pkt_proto = false;
1999 
2000 		pr_info("Proto override in use\n");
2001 		return true;
2002 	}
2003 
2004 	return false;
2005 }
2006 
2007 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2008 						  u16 *proto,
2009 						  bool old_pkt_proto)
2010 {
2011 	struct vsock_sock *vsk = vsock_sk(sk);
2012 
2013 	if (old_pkt_proto) {
2014 		if (*proto != VSOCK_PROTO_INVALID) {
2015 			pr_err("Can't set both an old and new protocol\n");
2016 			return false;
2017 		}
2018 		vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2019 		goto exit;
2020 	}
2021 
2022 	switch (*proto) {
2023 	case VSOCK_PROTO_PKT_ON_NOTIFY:
2024 		vmci_trans(vsk)->notify_ops =
2025 			&vmci_transport_notify_pkt_q_state_ops;
2026 		break;
2027 	default:
2028 		pr_err("Unknown notify protocol version\n");
2029 		return false;
2030 	}
2031 
2032 exit:
2033 	vmci_trans(vsk)->notify_ops->socket_init(sk);
2034 	return true;
2035 }
2036 
2037 static u16 vmci_transport_new_proto_supported_versions(void)
2038 {
2039 	if (PROTOCOL_OVERRIDE != -1)
2040 		return PROTOCOL_OVERRIDE;
2041 
2042 	return VSOCK_PROTO_ALL_SUPPORTED;
2043 }
2044 
2045 static u32 vmci_transport_get_local_cid(void)
2046 {
2047 	return vmci_get_context_id();
2048 }
2049 
2050 static const struct vsock_transport vmci_transport = {
2051 	.init = vmci_transport_socket_init,
2052 	.destruct = vmci_transport_destruct,
2053 	.release = vmci_transport_release,
2054 	.connect = vmci_transport_connect,
2055 	.dgram_bind = vmci_transport_dgram_bind,
2056 	.dgram_dequeue = vmci_transport_dgram_dequeue,
2057 	.dgram_enqueue = vmci_transport_dgram_enqueue,
2058 	.dgram_allow = vmci_transport_dgram_allow,
2059 	.stream_dequeue = vmci_transport_stream_dequeue,
2060 	.stream_enqueue = vmci_transport_stream_enqueue,
2061 	.stream_has_data = vmci_transport_stream_has_data,
2062 	.stream_has_space = vmci_transport_stream_has_space,
2063 	.stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2064 	.stream_is_active = vmci_transport_stream_is_active,
2065 	.stream_allow = vmci_transport_stream_allow,
2066 	.notify_poll_in = vmci_transport_notify_poll_in,
2067 	.notify_poll_out = vmci_transport_notify_poll_out,
2068 	.notify_recv_init = vmci_transport_notify_recv_init,
2069 	.notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2070 	.notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2071 	.notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2072 	.notify_send_init = vmci_transport_notify_send_init,
2073 	.notify_send_pre_block = vmci_transport_notify_send_pre_block,
2074 	.notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2075 	.notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2076 	.shutdown = vmci_transport_shutdown,
2077 	.set_buffer_size = vmci_transport_set_buffer_size,
2078 	.set_min_buffer_size = vmci_transport_set_min_buffer_size,
2079 	.set_max_buffer_size = vmci_transport_set_max_buffer_size,
2080 	.get_buffer_size = vmci_transport_get_buffer_size,
2081 	.get_min_buffer_size = vmci_transport_get_min_buffer_size,
2082 	.get_max_buffer_size = vmci_transport_get_max_buffer_size,
2083 	.get_local_cid = vmci_transport_get_local_cid,
2084 };
2085 
2086 static int __init vmci_transport_init(void)
2087 {
2088 	int err;
2089 
2090 	/* Create the datagram handle that we will use to send and receive all
2091 	 * VSocket control messages for this context.
2092 	 */
2093 	err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2094 						 VMCI_FLAG_ANYCID_DG_HND,
2095 						 vmci_transport_recv_stream_cb,
2096 						 NULL,
2097 						 &vmci_transport_stream_handle);
2098 	if (err < VMCI_SUCCESS) {
2099 		pr_err("Unable to create datagram handle. (%d)\n", err);
2100 		return vmci_transport_error_to_vsock_error(err);
2101 	}
2102 
2103 	err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2104 				   vmci_transport_qp_resumed_cb,
2105 				   NULL, &vmci_transport_qp_resumed_sub_id);
2106 	if (err < VMCI_SUCCESS) {
2107 		pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2108 		err = vmci_transport_error_to_vsock_error(err);
2109 		vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2110 		goto err_destroy_stream_handle;
2111 	}
2112 
2113 	err = vsock_core_init(&vmci_transport);
2114 	if (err < 0)
2115 		goto err_unsubscribe;
2116 
2117 	return 0;
2118 
2119 err_unsubscribe:
2120 	vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2121 err_destroy_stream_handle:
2122 	vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2123 	return err;
2124 }
2125 module_init(vmci_transport_init);
2126 
2127 static void __exit vmci_transport_exit(void)
2128 {
2129 	cancel_work_sync(&vmci_transport_cleanup_work);
2130 	vmci_transport_free_resources(&vmci_transport_cleanup_list);
2131 
2132 	if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2133 		if (vmci_datagram_destroy_handle(
2134 			vmci_transport_stream_handle) != VMCI_SUCCESS)
2135 			pr_err("Couldn't destroy datagram handle\n");
2136 		vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2137 	}
2138 
2139 	if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2140 		vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2141 		vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2142 	}
2143 
2144 	vsock_core_exit();
2145 }
2146 module_exit(vmci_transport_exit);
2147 
2148 MODULE_AUTHOR("VMware, Inc.");
2149 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2150 MODULE_VERSION("1.0.5.0-k");
2151 MODULE_LICENSE("GPL v2");
2152 MODULE_ALIAS("vmware_vsock");
2153 MODULE_ALIAS_NETPROTO(PF_VSOCK);
2154