1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VMware vSockets Driver
4  *
5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
6  */
7 
8 #include <linux/types.h>
9 #include <linux/bitops.h>
10 #include <linux/cred.h>
11 #include <linux/init.h>
12 #include <linux/io.h>
13 #include <linux/kernel.h>
14 #include <linux/kmod.h>
15 #include <linux/list.h>
16 #include <linux/module.h>
17 #include <linux/mutex.h>
18 #include <linux/net.h>
19 #include <linux/poll.h>
20 #include <linux/skbuff.h>
21 #include <linux/smp.h>
22 #include <linux/socket.h>
23 #include <linux/stddef.h>
24 #include <linux/unistd.h>
25 #include <linux/wait.h>
26 #include <linux/workqueue.h>
27 #include <net/sock.h>
28 #include <net/af_vsock.h>
29 
30 #include "vmci_transport_notify.h"
31 
32 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
33 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
34 static void vmci_transport_peer_detach_cb(u32 sub_id,
35 					  const struct vmci_event_data *ed,
36 					  void *client_data);
37 static void vmci_transport_recv_pkt_work(struct work_struct *work);
38 static void vmci_transport_cleanup(struct work_struct *work);
39 static int vmci_transport_recv_listen(struct sock *sk,
40 				      struct vmci_transport_packet *pkt);
41 static int vmci_transport_recv_connecting_server(
42 					struct sock *sk,
43 					struct sock *pending,
44 					struct vmci_transport_packet *pkt);
45 static int vmci_transport_recv_connecting_client(
46 					struct sock *sk,
47 					struct vmci_transport_packet *pkt);
48 static int vmci_transport_recv_connecting_client_negotiate(
49 					struct sock *sk,
50 					struct vmci_transport_packet *pkt);
51 static int vmci_transport_recv_connecting_client_invalid(
52 					struct sock *sk,
53 					struct vmci_transport_packet *pkt);
54 static int vmci_transport_recv_connected(struct sock *sk,
55 					 struct vmci_transport_packet *pkt);
56 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
57 static u16 vmci_transport_new_proto_supported_versions(void);
58 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
59 						  bool old_pkt_proto);
60 static bool vmci_check_transport(struct vsock_sock *vsk);
61 
62 struct vmci_transport_recv_pkt_info {
63 	struct work_struct work;
64 	struct sock *sk;
65 	struct vmci_transport_packet pkt;
66 };
67 
68 static LIST_HEAD(vmci_transport_cleanup_list);
69 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
70 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
71 
72 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
73 							   VMCI_INVALID_ID };
74 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
75 
76 static int PROTOCOL_OVERRIDE = -1;
77 
78 /* Helper function to convert from a VMCI error code to a VSock error code. */
79 
80 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
81 {
82 	switch (vmci_error) {
83 	case VMCI_ERROR_NO_MEM:
84 		return -ENOMEM;
85 	case VMCI_ERROR_DUPLICATE_ENTRY:
86 	case VMCI_ERROR_ALREADY_EXISTS:
87 		return -EADDRINUSE;
88 	case VMCI_ERROR_NO_ACCESS:
89 		return -EPERM;
90 	case VMCI_ERROR_NO_RESOURCES:
91 		return -ENOBUFS;
92 	case VMCI_ERROR_INVALID_RESOURCE:
93 		return -EHOSTUNREACH;
94 	case VMCI_ERROR_INVALID_ARGS:
95 	default:
96 		break;
97 	}
98 	return -EINVAL;
99 }
100 
101 static u32 vmci_transport_peer_rid(u32 peer_cid)
102 {
103 	if (VMADDR_CID_HYPERVISOR == peer_cid)
104 		return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
105 
106 	return VMCI_TRANSPORT_PACKET_RID;
107 }
108 
109 static inline void
110 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
111 			   struct sockaddr_vm *src,
112 			   struct sockaddr_vm *dst,
113 			   u8 type,
114 			   u64 size,
115 			   u64 mode,
116 			   struct vmci_transport_waiting_info *wait,
117 			   u16 proto,
118 			   struct vmci_handle handle)
119 {
120 	/* We register the stream control handler as an any cid handle so we
121 	 * must always send from a source address of VMADDR_CID_ANY
122 	 */
123 	pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
124 				       VMCI_TRANSPORT_PACKET_RID);
125 	pkt->dg.dst = vmci_make_handle(dst->svm_cid,
126 				       vmci_transport_peer_rid(dst->svm_cid));
127 	pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
128 	pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
129 	pkt->type = type;
130 	pkt->src_port = src->svm_port;
131 	pkt->dst_port = dst->svm_port;
132 	memset(&pkt->proto, 0, sizeof(pkt->proto));
133 	memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
134 
135 	switch (pkt->type) {
136 	case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
137 		pkt->u.size = 0;
138 		break;
139 
140 	case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
141 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
142 		pkt->u.size = size;
143 		break;
144 
145 	case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
146 	case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
147 		pkt->u.handle = handle;
148 		break;
149 
150 	case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
151 	case VMCI_TRANSPORT_PACKET_TYPE_READ:
152 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
153 		pkt->u.size = 0;
154 		break;
155 
156 	case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
157 		pkt->u.mode = mode;
158 		break;
159 
160 	case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
161 	case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
162 		memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
163 		break;
164 
165 	case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
166 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
167 		pkt->u.size = size;
168 		pkt->proto = proto;
169 		break;
170 	}
171 }
172 
173 static inline void
174 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
175 				    struct sockaddr_vm *local,
176 				    struct sockaddr_vm *remote)
177 {
178 	vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
179 	vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
180 }
181 
182 static int
183 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
184 				  struct sockaddr_vm *src,
185 				  struct sockaddr_vm *dst,
186 				  enum vmci_transport_packet_type type,
187 				  u64 size,
188 				  u64 mode,
189 				  struct vmci_transport_waiting_info *wait,
190 				  u16 proto,
191 				  struct vmci_handle handle,
192 				  bool convert_error)
193 {
194 	int err;
195 
196 	vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
197 				   proto, handle);
198 	err = vmci_datagram_send(&pkt->dg);
199 	if (convert_error && (err < 0))
200 		return vmci_transport_error_to_vsock_error(err);
201 
202 	return err;
203 }
204 
205 static int
206 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
207 				      enum vmci_transport_packet_type type,
208 				      u64 size,
209 				      u64 mode,
210 				      struct vmci_transport_waiting_info *wait,
211 				      struct vmci_handle handle)
212 {
213 	struct vmci_transport_packet reply;
214 	struct sockaddr_vm src, dst;
215 
216 	if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
217 		return 0;
218 	} else {
219 		vmci_transport_packet_get_addresses(pkt, &src, &dst);
220 		return __vmci_transport_send_control_pkt(&reply, &src, &dst,
221 							 type,
222 							 size, mode, wait,
223 							 VSOCK_PROTO_INVALID,
224 							 handle, true);
225 	}
226 }
227 
228 static int
229 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
230 				   struct sockaddr_vm *dst,
231 				   enum vmci_transport_packet_type type,
232 				   u64 size,
233 				   u64 mode,
234 				   struct vmci_transport_waiting_info *wait,
235 				   struct vmci_handle handle)
236 {
237 	/* Note that it is safe to use a single packet across all CPUs since
238 	 * two tasklets of the same type are guaranteed to not ever run
239 	 * simultaneously. If that ever changes, or VMCI stops using tasklets,
240 	 * we can use per-cpu packets.
241 	 */
242 	static struct vmci_transport_packet pkt;
243 
244 	return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
245 						 size, mode, wait,
246 						 VSOCK_PROTO_INVALID, handle,
247 						 false);
248 }
249 
250 static int
251 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
252 				      struct sockaddr_vm *dst,
253 				      enum vmci_transport_packet_type type,
254 				      u64 size,
255 				      u64 mode,
256 				      struct vmci_transport_waiting_info *wait,
257 				      u16 proto,
258 				      struct vmci_handle handle)
259 {
260 	struct vmci_transport_packet *pkt;
261 	int err;
262 
263 	pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
264 	if (!pkt)
265 		return -ENOMEM;
266 
267 	err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
268 						mode, wait, proto, handle,
269 						true);
270 	kfree(pkt);
271 
272 	return err;
273 }
274 
275 static int
276 vmci_transport_send_control_pkt(struct sock *sk,
277 				enum vmci_transport_packet_type type,
278 				u64 size,
279 				u64 mode,
280 				struct vmci_transport_waiting_info *wait,
281 				u16 proto,
282 				struct vmci_handle handle)
283 {
284 	struct vsock_sock *vsk;
285 
286 	vsk = vsock_sk(sk);
287 
288 	if (!vsock_addr_bound(&vsk->local_addr))
289 		return -EINVAL;
290 
291 	if (!vsock_addr_bound(&vsk->remote_addr))
292 		return -EINVAL;
293 
294 	return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
295 						     &vsk->remote_addr,
296 						     type, size, mode,
297 						     wait, proto, handle);
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 	struct sockaddr_vm *dst_ptr;
316 	struct sockaddr_vm dst;
317 	struct vsock_sock *vsk;
318 
319 	if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
320 		return 0;
321 
322 	vsk = vsock_sk(sk);
323 
324 	if (!vsock_addr_bound(&vsk->local_addr))
325 		return -EINVAL;
326 
327 	if (vsock_addr_bound(&vsk->remote_addr)) {
328 		dst_ptr = &vsk->remote_addr;
329 	} else {
330 		vsock_addr_init(&dst, pkt->dg.src.context,
331 				pkt->src_port);
332 		dst_ptr = &dst;
333 	}
334 	return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
335 					     VMCI_TRANSPORT_PACKET_TYPE_RST,
336 					     0, 0, NULL, VSOCK_PROTO_INVALID,
337 					     VMCI_INVALID_HANDLE);
338 }
339 
340 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
341 {
342 	return vmci_transport_send_control_pkt(
343 					sk,
344 					VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
345 					size, 0, NULL,
346 					VSOCK_PROTO_INVALID,
347 					VMCI_INVALID_HANDLE);
348 }
349 
350 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
351 					  u16 version)
352 {
353 	return vmci_transport_send_control_pkt(
354 					sk,
355 					VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
356 					size, 0, NULL, version,
357 					VMCI_INVALID_HANDLE);
358 }
359 
360 static int vmci_transport_send_qp_offer(struct sock *sk,
361 					struct vmci_handle handle)
362 {
363 	return vmci_transport_send_control_pkt(
364 					sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
365 					0, NULL,
366 					VSOCK_PROTO_INVALID, handle);
367 }
368 
369 static int vmci_transport_send_attach(struct sock *sk,
370 				      struct vmci_handle handle)
371 {
372 	return vmci_transport_send_control_pkt(
373 					sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
374 					0, 0, NULL, VSOCK_PROTO_INVALID,
375 					handle);
376 }
377 
378 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
379 {
380 	return vmci_transport_reply_control_pkt_fast(
381 						pkt,
382 						VMCI_TRANSPORT_PACKET_TYPE_RST,
383 						0, 0, NULL,
384 						VMCI_INVALID_HANDLE);
385 }
386 
387 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
388 					  struct sockaddr_vm *src)
389 {
390 	return vmci_transport_send_control_pkt_bh(
391 					dst, src,
392 					VMCI_TRANSPORT_PACKET_TYPE_INVALID,
393 					0, 0, NULL, VMCI_INVALID_HANDLE);
394 }
395 
396 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
397 				 struct sockaddr_vm *src)
398 {
399 	return vmci_transport_send_control_pkt_bh(
400 					dst, src,
401 					VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
402 					0, NULL, VMCI_INVALID_HANDLE);
403 }
404 
405 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
406 				struct sockaddr_vm *src)
407 {
408 	return vmci_transport_send_control_pkt_bh(
409 					dst, src,
410 					VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
411 					0, NULL, VMCI_INVALID_HANDLE);
412 }
413 
414 int vmci_transport_send_wrote(struct sock *sk)
415 {
416 	return vmci_transport_send_control_pkt(
417 					sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
418 					0, NULL, VSOCK_PROTO_INVALID,
419 					VMCI_INVALID_HANDLE);
420 }
421 
422 int vmci_transport_send_read(struct sock *sk)
423 {
424 	return vmci_transport_send_control_pkt(
425 					sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
426 					0, NULL, VSOCK_PROTO_INVALID,
427 					VMCI_INVALID_HANDLE);
428 }
429 
430 int vmci_transport_send_waiting_write(struct sock *sk,
431 				      struct vmci_transport_waiting_info *wait)
432 {
433 	return vmci_transport_send_control_pkt(
434 				sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
435 				0, 0, wait, VSOCK_PROTO_INVALID,
436 				VMCI_INVALID_HANDLE);
437 }
438 
439 int vmci_transport_send_waiting_read(struct sock *sk,
440 				     struct vmci_transport_waiting_info *wait)
441 {
442 	return vmci_transport_send_control_pkt(
443 				sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
444 				0, 0, wait, VSOCK_PROTO_INVALID,
445 				VMCI_INVALID_HANDLE);
446 }
447 
448 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
449 {
450 	return vmci_transport_send_control_pkt(
451 					&vsk->sk,
452 					VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
453 					0, mode, NULL,
454 					VSOCK_PROTO_INVALID,
455 					VMCI_INVALID_HANDLE);
456 }
457 
458 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
459 {
460 	return vmci_transport_send_control_pkt(sk,
461 					VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
462 					size, 0, NULL,
463 					VSOCK_PROTO_INVALID,
464 					VMCI_INVALID_HANDLE);
465 }
466 
467 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
468 					     u16 version)
469 {
470 	return vmci_transport_send_control_pkt(
471 					sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
472 					size, 0, NULL, version,
473 					VMCI_INVALID_HANDLE);
474 }
475 
476 static struct sock *vmci_transport_get_pending(
477 					struct sock *listener,
478 					struct vmci_transport_packet *pkt)
479 {
480 	struct vsock_sock *vlistener;
481 	struct vsock_sock *vpending;
482 	struct sock *pending;
483 	struct sockaddr_vm src;
484 
485 	vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
486 
487 	vlistener = vsock_sk(listener);
488 
489 	list_for_each_entry(vpending, &vlistener->pending_links,
490 			    pending_links) {
491 		if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
492 		    pkt->dst_port == vpending->local_addr.svm_port) {
493 			pending = sk_vsock(vpending);
494 			sock_hold(pending);
495 			goto found;
496 		}
497 	}
498 
499 	pending = NULL;
500 found:
501 	return pending;
502 
503 }
504 
505 static void vmci_transport_release_pending(struct sock *pending)
506 {
507 	sock_put(pending);
508 }
509 
510 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
511  * trusted sockets 2) sockets from applications running as the same user as the
512  * VM (this is only true for the host side and only when using hosted products)
513  */
514 
515 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
516 {
517 	return vsock->trusted ||
518 	       vmci_is_context_owner(peer_cid, vsock->owner->uid);
519 }
520 
521 /* We allow sending datagrams to and receiving datagrams from a restricted VM
522  * only if it is trusted as described in vmci_transport_is_trusted.
523  */
524 
525 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
526 {
527 	if (VMADDR_CID_HYPERVISOR == peer_cid)
528 		return true;
529 
530 	if (vsock->cached_peer != peer_cid) {
531 		vsock->cached_peer = peer_cid;
532 		if (!vmci_transport_is_trusted(vsock, peer_cid) &&
533 		    (vmci_context_get_priv_flags(peer_cid) &
534 		     VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
535 			vsock->cached_peer_allow_dgram = false;
536 		} else {
537 			vsock->cached_peer_allow_dgram = true;
538 		}
539 	}
540 
541 	return vsock->cached_peer_allow_dgram;
542 }
543 
544 static int
545 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
546 				struct vmci_handle *handle,
547 				u64 produce_size,
548 				u64 consume_size,
549 				u32 peer, u32 flags, bool trusted)
550 {
551 	int err = 0;
552 
553 	if (trusted) {
554 		/* Try to allocate our queue pair as trusted. This will only
555 		 * work if vsock is running in the host.
556 		 */
557 
558 		err = vmci_qpair_alloc(qpair, handle, produce_size,
559 				       consume_size,
560 				       peer, flags,
561 				       VMCI_PRIVILEGE_FLAG_TRUSTED);
562 		if (err != VMCI_ERROR_NO_ACCESS)
563 			goto out;
564 
565 	}
566 
567 	err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
568 			       peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
569 out:
570 	if (err < 0) {
571 		pr_err("Could not attach to queue pair with %d\n",
572 		       err);
573 		err = vmci_transport_error_to_vsock_error(err);
574 	}
575 
576 	return err;
577 }
578 
579 static int
580 vmci_transport_datagram_create_hnd(u32 resource_id,
581 				   u32 flags,
582 				   vmci_datagram_recv_cb recv_cb,
583 				   void *client_data,
584 				   struct vmci_handle *out_handle)
585 {
586 	int err = 0;
587 
588 	/* Try to allocate our datagram handler as trusted. This will only work
589 	 * if vsock is running in the host.
590 	 */
591 
592 	err = vmci_datagram_create_handle_priv(resource_id, flags,
593 					       VMCI_PRIVILEGE_FLAG_TRUSTED,
594 					       recv_cb,
595 					       client_data, out_handle);
596 
597 	if (err == VMCI_ERROR_NO_ACCESS)
598 		err = vmci_datagram_create_handle(resource_id, flags,
599 						  recv_cb, client_data,
600 						  out_handle);
601 
602 	return err;
603 }
604 
605 /* This is invoked as part of a tasklet that's scheduled when the VMCI
606  * interrupt fires.  This is run in bottom-half context and if it ever needs to
607  * sleep it should defer that work to a work queue.
608  */
609 
610 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
611 {
612 	struct sock *sk;
613 	size_t size;
614 	struct sk_buff *skb;
615 	struct vsock_sock *vsk;
616 
617 	sk = (struct sock *)data;
618 
619 	/* This handler is privileged when this module is running on the host.
620 	 * We will get datagrams from all endpoints (even VMs that are in a
621 	 * restricted context). If we get one from a restricted context then
622 	 * the destination socket must be trusted.
623 	 *
624 	 * NOTE: We access the socket struct without holding the lock here.
625 	 * This is ok because the field we are interested is never modified
626 	 * outside of the create and destruct socket functions.
627 	 */
628 	vsk = vsock_sk(sk);
629 	if (!vmci_transport_allow_dgram(vsk, dg->src.context))
630 		return VMCI_ERROR_NO_ACCESS;
631 
632 	size = VMCI_DG_SIZE(dg);
633 
634 	/* Attach the packet to the socket's receive queue as an sk_buff. */
635 	skb = alloc_skb(size, GFP_ATOMIC);
636 	if (!skb)
637 		return VMCI_ERROR_NO_MEM;
638 
639 	/* sk_receive_skb() will do a sock_put(), so hold here. */
640 	sock_hold(sk);
641 	skb_put(skb, size);
642 	memcpy(skb->data, dg, size);
643 	sk_receive_skb(sk, skb, 0);
644 
645 	return VMCI_SUCCESS;
646 }
647 
648 static bool vmci_transport_stream_allow(u32 cid, u32 port)
649 {
650 	static const u32 non_socket_contexts[] = {
651 		VMADDR_CID_RESERVED,
652 	};
653 	int i;
654 
655 	BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
656 
657 	for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
658 		if (cid == non_socket_contexts[i])
659 			return false;
660 	}
661 
662 	return true;
663 }
664 
665 /* This is invoked as part of a tasklet that's scheduled when the VMCI
666  * interrupt fires.  This is run in bottom-half context but it defers most of
667  * its work to the packet handling work queue.
668  */
669 
670 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
671 {
672 	struct sock *sk;
673 	struct sockaddr_vm dst;
674 	struct sockaddr_vm src;
675 	struct vmci_transport_packet *pkt;
676 	struct vsock_sock *vsk;
677 	bool bh_process_pkt;
678 	int err;
679 
680 	sk = NULL;
681 	err = VMCI_SUCCESS;
682 	bh_process_pkt = false;
683 
684 	/* Ignore incoming packets from contexts without sockets, or resources
685 	 * that aren't vsock implementations.
686 	 */
687 
688 	if (!vmci_transport_stream_allow(dg->src.context, -1)
689 	    || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
690 		return VMCI_ERROR_NO_ACCESS;
691 
692 	if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
693 		/* Drop datagrams that do not contain full VSock packets. */
694 		return VMCI_ERROR_INVALID_ARGS;
695 
696 	pkt = (struct vmci_transport_packet *)dg;
697 
698 	/* Find the socket that should handle this packet.  First we look for a
699 	 * connected socket and if there is none we look for a socket bound to
700 	 * the destintation address.
701 	 */
702 	vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
703 	vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
704 
705 	sk = vsock_find_connected_socket(&src, &dst);
706 	if (!sk) {
707 		sk = vsock_find_bound_socket(&dst);
708 		if (!sk) {
709 			/* We could not find a socket for this specified
710 			 * address.  If this packet is a RST, we just drop it.
711 			 * If it is another packet, we send a RST.  Note that
712 			 * we do not send a RST reply to RSTs so that we do not
713 			 * continually send RSTs between two endpoints.
714 			 *
715 			 * Note that since this is a reply, dst is src and src
716 			 * is dst.
717 			 */
718 			if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
719 				pr_err("unable to send reset\n");
720 
721 			err = VMCI_ERROR_NOT_FOUND;
722 			goto out;
723 		}
724 	}
725 
726 	/* If the received packet type is beyond all types known to this
727 	 * implementation, reply with an invalid message.  Hopefully this will
728 	 * help when implementing backwards compatibility in the future.
729 	 */
730 	if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
731 		vmci_transport_send_invalid_bh(&dst, &src);
732 		err = VMCI_ERROR_INVALID_ARGS;
733 		goto out;
734 	}
735 
736 	/* This handler is privileged when this module is running on the host.
737 	 * We will get datagram connect requests from all endpoints (even VMs
738 	 * that are in a restricted context). If we get one from a restricted
739 	 * context then the destination socket must be trusted.
740 	 *
741 	 * NOTE: We access the socket struct without holding the lock here.
742 	 * This is ok because the field we are interested is never modified
743 	 * outside of the create and destruct socket functions.
744 	 */
745 	vsk = vsock_sk(sk);
746 	if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
747 		err = VMCI_ERROR_NO_ACCESS;
748 		goto out;
749 	}
750 
751 	/* We do most everything in a work queue, but let's fast path the
752 	 * notification of reads and writes to help data transfer performance.
753 	 * We can only do this if there is no process context code executing
754 	 * for this socket since that may change the state.
755 	 */
756 	bh_lock_sock(sk);
757 
758 	if (!sock_owned_by_user(sk)) {
759 		/* The local context ID may be out of date, update it. */
760 		vsk->local_addr.svm_cid = dst.svm_cid;
761 
762 		if (sk->sk_state == TCP_ESTABLISHED)
763 			vmci_trans(vsk)->notify_ops->handle_notify_pkt(
764 					sk, pkt, true, &dst, &src,
765 					&bh_process_pkt);
766 	}
767 
768 	bh_unlock_sock(sk);
769 
770 	if (!bh_process_pkt) {
771 		struct vmci_transport_recv_pkt_info *recv_pkt_info;
772 
773 		recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
774 		if (!recv_pkt_info) {
775 			if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
776 				pr_err("unable to send reset\n");
777 
778 			err = VMCI_ERROR_NO_MEM;
779 			goto out;
780 		}
781 
782 		recv_pkt_info->sk = sk;
783 		memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
784 		INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
785 
786 		schedule_work(&recv_pkt_info->work);
787 		/* Clear sk so that the reference count incremented by one of
788 		 * the Find functions above is not decremented below.  We need
789 		 * that reference count for the packet handler we've scheduled
790 		 * to run.
791 		 */
792 		sk = NULL;
793 	}
794 
795 out:
796 	if (sk)
797 		sock_put(sk);
798 
799 	return err;
800 }
801 
802 static void vmci_transport_handle_detach(struct sock *sk)
803 {
804 	struct vsock_sock *vsk;
805 
806 	vsk = vsock_sk(sk);
807 	if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
808 		sock_set_flag(sk, SOCK_DONE);
809 
810 		/* On a detach the peer will not be sending or receiving
811 		 * anymore.
812 		 */
813 		vsk->peer_shutdown = SHUTDOWN_MASK;
814 
815 		/* We should not be sending anymore since the peer won't be
816 		 * there to receive, but we can still receive if there is data
817 		 * left in our consume queue. If the local endpoint is a host,
818 		 * we can't call vsock_stream_has_data, since that may block,
819 		 * but a host endpoint can't read data once the VM has
820 		 * detached, so there is no available data in that case.
821 		 */
822 		if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
823 		    vsock_stream_has_data(vsk) <= 0) {
824 			if (sk->sk_state == TCP_SYN_SENT) {
825 				/* The peer may detach from a queue pair while
826 				 * we are still in the connecting state, i.e.,
827 				 * if the peer VM is killed after attaching to
828 				 * a queue pair, but before we complete the
829 				 * handshake. In that case, we treat the detach
830 				 * event like a reset.
831 				 */
832 
833 				sk->sk_state = TCP_CLOSE;
834 				sk->sk_err = ECONNRESET;
835 				sk->sk_error_report(sk);
836 				return;
837 			}
838 			sk->sk_state = TCP_CLOSE;
839 		}
840 		sk->sk_state_change(sk);
841 	}
842 }
843 
844 static void vmci_transport_peer_detach_cb(u32 sub_id,
845 					  const struct vmci_event_data *e_data,
846 					  void *client_data)
847 {
848 	struct vmci_transport *trans = client_data;
849 	const struct vmci_event_payload_qp *e_payload;
850 
851 	e_payload = vmci_event_data_const_payload(e_data);
852 
853 	/* XXX This is lame, we should provide a way to lookup sockets by
854 	 * qp_handle.
855 	 */
856 	if (vmci_handle_is_invalid(e_payload->handle) ||
857 	    !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
858 		return;
859 
860 	/* We don't ask for delayed CBs when we subscribe to this event (we
861 	 * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
862 	 * guarantees in that case about what context we might be running in,
863 	 * so it could be BH or process, blockable or non-blockable.  So we
864 	 * need to account for all possible contexts here.
865 	 */
866 	spin_lock_bh(&trans->lock);
867 	if (!trans->sk)
868 		goto out;
869 
870 	/* Apart from here, trans->lock is only grabbed as part of sk destruct,
871 	 * where trans->sk isn't locked.
872 	 */
873 	bh_lock_sock(trans->sk);
874 
875 	vmci_transport_handle_detach(trans->sk);
876 
877 	bh_unlock_sock(trans->sk);
878  out:
879 	spin_unlock_bh(&trans->lock);
880 }
881 
882 static void vmci_transport_qp_resumed_cb(u32 sub_id,
883 					 const struct vmci_event_data *e_data,
884 					 void *client_data)
885 {
886 	vsock_for_each_connected_socket(vmci_transport_handle_detach);
887 }
888 
889 static void vmci_transport_recv_pkt_work(struct work_struct *work)
890 {
891 	struct vmci_transport_recv_pkt_info *recv_pkt_info;
892 	struct vmci_transport_packet *pkt;
893 	struct sock *sk;
894 
895 	recv_pkt_info =
896 		container_of(work, struct vmci_transport_recv_pkt_info, work);
897 	sk = recv_pkt_info->sk;
898 	pkt = &recv_pkt_info->pkt;
899 
900 	lock_sock(sk);
901 
902 	/* The local context ID may be out of date. */
903 	vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
904 
905 	switch (sk->sk_state) {
906 	case TCP_LISTEN:
907 		vmci_transport_recv_listen(sk, pkt);
908 		break;
909 	case TCP_SYN_SENT:
910 		/* Processing of pending connections for servers goes through
911 		 * the listening socket, so see vmci_transport_recv_listen()
912 		 * for that path.
913 		 */
914 		vmci_transport_recv_connecting_client(sk, pkt);
915 		break;
916 	case TCP_ESTABLISHED:
917 		vmci_transport_recv_connected(sk, pkt);
918 		break;
919 	default:
920 		/* Because this function does not run in the same context as
921 		 * vmci_transport_recv_stream_cb it is possible that the
922 		 * socket has closed. We need to let the other side know or it
923 		 * could be sitting in a connect and hang forever. Send a
924 		 * reset to prevent that.
925 		 */
926 		vmci_transport_send_reset(sk, pkt);
927 		break;
928 	}
929 
930 	release_sock(sk);
931 	kfree(recv_pkt_info);
932 	/* Release reference obtained in the stream callback when we fetched
933 	 * this socket out of the bound or connected list.
934 	 */
935 	sock_put(sk);
936 }
937 
938 static int vmci_transport_recv_listen(struct sock *sk,
939 				      struct vmci_transport_packet *pkt)
940 {
941 	struct sock *pending;
942 	struct vsock_sock *vpending;
943 	int err;
944 	u64 qp_size;
945 	bool old_request = false;
946 	bool old_pkt_proto = false;
947 
948 	err = 0;
949 
950 	/* Because we are in the listen state, we could be receiving a packet
951 	 * for ourself or any previous connection requests that we received.
952 	 * If it's the latter, we try to find a socket in our list of pending
953 	 * connections and, if we do, call the appropriate handler for the
954 	 * state that that socket is in.  Otherwise we try to service the
955 	 * connection request.
956 	 */
957 	pending = vmci_transport_get_pending(sk, pkt);
958 	if (pending) {
959 		lock_sock(pending);
960 
961 		/* The local context ID may be out of date. */
962 		vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
963 
964 		switch (pending->sk_state) {
965 		case TCP_SYN_SENT:
966 			err = vmci_transport_recv_connecting_server(sk,
967 								    pending,
968 								    pkt);
969 			break;
970 		default:
971 			vmci_transport_send_reset(pending, pkt);
972 			err = -EINVAL;
973 		}
974 
975 		if (err < 0)
976 			vsock_remove_pending(sk, pending);
977 
978 		release_sock(pending);
979 		vmci_transport_release_pending(pending);
980 
981 		return err;
982 	}
983 
984 	/* The listen state only accepts connection requests.  Reply with a
985 	 * reset unless we received a reset.
986 	 */
987 
988 	if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
989 	      pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
990 		vmci_transport_reply_reset(pkt);
991 		return -EINVAL;
992 	}
993 
994 	if (pkt->u.size == 0) {
995 		vmci_transport_reply_reset(pkt);
996 		return -EINVAL;
997 	}
998 
999 	/* If this socket can't accommodate this connection request, we send a
1000 	 * reset.  Otherwise we create and initialize a child socket and reply
1001 	 * with a connection negotiation.
1002 	 */
1003 	if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1004 		vmci_transport_reply_reset(pkt);
1005 		return -ECONNREFUSED;
1006 	}
1007 
1008 	pending = vsock_create_connected(sk);
1009 	if (!pending) {
1010 		vmci_transport_send_reset(sk, pkt);
1011 		return -ENOMEM;
1012 	}
1013 
1014 	vpending = vsock_sk(pending);
1015 
1016 	vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1017 			pkt->dst_port);
1018 	vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1019 			pkt->src_port);
1020 
1021 	err = vsock_assign_transport(vpending, vsock_sk(sk));
1022 	/* Transport assigned (looking at remote_addr) must be the same
1023 	 * where we received the request.
1024 	 */
1025 	if (err || !vmci_check_transport(vpending)) {
1026 		vmci_transport_send_reset(sk, pkt);
1027 		sock_put(pending);
1028 		return err;
1029 	}
1030 
1031 	/* If the proposed size fits within our min/max, accept it. Otherwise
1032 	 * propose our own size.
1033 	 */
1034 	if (pkt->u.size >= vpending->buffer_min_size &&
1035 	    pkt->u.size <= vpending->buffer_max_size) {
1036 		qp_size = pkt->u.size;
1037 	} else {
1038 		qp_size = vpending->buffer_size;
1039 	}
1040 
1041 	/* Figure out if we are using old or new requests based on the
1042 	 * overrides pkt types sent by our peer.
1043 	 */
1044 	if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1045 		old_request = old_pkt_proto;
1046 	} else {
1047 		if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1048 			old_request = true;
1049 		else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1050 			old_request = false;
1051 
1052 	}
1053 
1054 	if (old_request) {
1055 		/* Handle a REQUEST (or override) */
1056 		u16 version = VSOCK_PROTO_INVALID;
1057 		if (vmci_transport_proto_to_notify_struct(
1058 			pending, &version, true))
1059 			err = vmci_transport_send_negotiate(pending, qp_size);
1060 		else
1061 			err = -EINVAL;
1062 
1063 	} else {
1064 		/* Handle a REQUEST2 (or override) */
1065 		int proto_int = pkt->proto;
1066 		int pos;
1067 		u16 active_proto_version = 0;
1068 
1069 		/* The list of possible protocols is the intersection of all
1070 		 * protocols the client supports ... plus all the protocols we
1071 		 * support.
1072 		 */
1073 		proto_int &= vmci_transport_new_proto_supported_versions();
1074 
1075 		/* We choose the highest possible protocol version and use that
1076 		 * one.
1077 		 */
1078 		pos = fls(proto_int);
1079 		if (pos) {
1080 			active_proto_version = (1 << (pos - 1));
1081 			if (vmci_transport_proto_to_notify_struct(
1082 				pending, &active_proto_version, false))
1083 				err = vmci_transport_send_negotiate2(pending,
1084 							qp_size,
1085 							active_proto_version);
1086 			else
1087 				err = -EINVAL;
1088 
1089 		} else {
1090 			err = -EINVAL;
1091 		}
1092 	}
1093 
1094 	if (err < 0) {
1095 		vmci_transport_send_reset(sk, pkt);
1096 		sock_put(pending);
1097 		err = vmci_transport_error_to_vsock_error(err);
1098 		goto out;
1099 	}
1100 
1101 	vsock_add_pending(sk, pending);
1102 	sk_acceptq_added(sk);
1103 
1104 	pending->sk_state = TCP_SYN_SENT;
1105 	vmci_trans(vpending)->produce_size =
1106 		vmci_trans(vpending)->consume_size = qp_size;
1107 	vpending->buffer_size = qp_size;
1108 
1109 	vmci_trans(vpending)->notify_ops->process_request(pending);
1110 
1111 	/* We might never receive another message for this socket and it's not
1112 	 * connected to any process, so we have to ensure it gets cleaned up
1113 	 * ourself.  Our delayed work function will take care of that.  Note
1114 	 * that we do not ever cancel this function since we have few
1115 	 * guarantees about its state when calling cancel_delayed_work().
1116 	 * Instead we hold a reference on the socket for that function and make
1117 	 * it capable of handling cases where it needs to do nothing but
1118 	 * release that reference.
1119 	 */
1120 	vpending->listener = sk;
1121 	sock_hold(sk);
1122 	sock_hold(pending);
1123 	schedule_delayed_work(&vpending->pending_work, HZ);
1124 
1125 out:
1126 	return err;
1127 }
1128 
1129 static int
1130 vmci_transport_recv_connecting_server(struct sock *listener,
1131 				      struct sock *pending,
1132 				      struct vmci_transport_packet *pkt)
1133 {
1134 	struct vsock_sock *vpending;
1135 	struct vmci_handle handle;
1136 	struct vmci_qp *qpair;
1137 	bool is_local;
1138 	u32 flags;
1139 	u32 detach_sub_id;
1140 	int err;
1141 	int skerr;
1142 
1143 	vpending = vsock_sk(pending);
1144 	detach_sub_id = VMCI_INVALID_ID;
1145 
1146 	switch (pkt->type) {
1147 	case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1148 		if (vmci_handle_is_invalid(pkt->u.handle)) {
1149 			vmci_transport_send_reset(pending, pkt);
1150 			skerr = EPROTO;
1151 			err = -EINVAL;
1152 			goto destroy;
1153 		}
1154 		break;
1155 	default:
1156 		/* Close and cleanup the connection. */
1157 		vmci_transport_send_reset(pending, pkt);
1158 		skerr = EPROTO;
1159 		err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1160 		goto destroy;
1161 	}
1162 
1163 	/* In order to complete the connection we need to attach to the offered
1164 	 * queue pair and send an attach notification.  We also subscribe to the
1165 	 * detach event so we know when our peer goes away, and we do that
1166 	 * before attaching so we don't miss an event.  If all this succeeds,
1167 	 * we update our state and wakeup anything waiting in accept() for a
1168 	 * connection.
1169 	 */
1170 
1171 	/* We don't care about attach since we ensure the other side has
1172 	 * attached by specifying the ATTACH_ONLY flag below.
1173 	 */
1174 	err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1175 				   vmci_transport_peer_detach_cb,
1176 				   vmci_trans(vpending), &detach_sub_id);
1177 	if (err < VMCI_SUCCESS) {
1178 		vmci_transport_send_reset(pending, pkt);
1179 		err = vmci_transport_error_to_vsock_error(err);
1180 		skerr = -err;
1181 		goto destroy;
1182 	}
1183 
1184 	vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1185 
1186 	/* Now attach to the queue pair the client created. */
1187 	handle = pkt->u.handle;
1188 
1189 	/* vpending->local_addr always has a context id so we do not need to
1190 	 * worry about VMADDR_CID_ANY in this case.
1191 	 */
1192 	is_local =
1193 	    vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1194 	flags = VMCI_QPFLAG_ATTACH_ONLY;
1195 	flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1196 
1197 	err = vmci_transport_queue_pair_alloc(
1198 					&qpair,
1199 					&handle,
1200 					vmci_trans(vpending)->produce_size,
1201 					vmci_trans(vpending)->consume_size,
1202 					pkt->dg.src.context,
1203 					flags,
1204 					vmci_transport_is_trusted(
1205 						vpending,
1206 						vpending->remote_addr.svm_cid));
1207 	if (err < 0) {
1208 		vmci_transport_send_reset(pending, pkt);
1209 		skerr = -err;
1210 		goto destroy;
1211 	}
1212 
1213 	vmci_trans(vpending)->qp_handle = handle;
1214 	vmci_trans(vpending)->qpair = qpair;
1215 
1216 	/* When we send the attach message, we must be ready to handle incoming
1217 	 * control messages on the newly connected socket. So we move the
1218 	 * pending socket to the connected state before sending the attach
1219 	 * message. Otherwise, an incoming packet triggered by the attach being
1220 	 * received by the peer may be processed concurrently with what happens
1221 	 * below after sending the attach message, and that incoming packet
1222 	 * will find the listening socket instead of the (currently) pending
1223 	 * socket. Note that enqueueing the socket increments the reference
1224 	 * count, so even if a reset comes before the connection is accepted,
1225 	 * the socket will be valid until it is removed from the queue.
1226 	 *
1227 	 * If we fail sending the attach below, we remove the socket from the
1228 	 * connected list and move the socket to TCP_CLOSE before
1229 	 * releasing the lock, so a pending slow path processing of an incoming
1230 	 * packet will not see the socket in the connected state in that case.
1231 	 */
1232 	pending->sk_state = TCP_ESTABLISHED;
1233 
1234 	vsock_insert_connected(vpending);
1235 
1236 	/* Notify our peer of our attach. */
1237 	err = vmci_transport_send_attach(pending, handle);
1238 	if (err < 0) {
1239 		vsock_remove_connected(vpending);
1240 		pr_err("Could not send attach\n");
1241 		vmci_transport_send_reset(pending, pkt);
1242 		err = vmci_transport_error_to_vsock_error(err);
1243 		skerr = -err;
1244 		goto destroy;
1245 	}
1246 
1247 	/* We have a connection. Move the now connected socket from the
1248 	 * listener's pending list to the accept queue so callers of accept()
1249 	 * can find it.
1250 	 */
1251 	vsock_remove_pending(listener, pending);
1252 	vsock_enqueue_accept(listener, pending);
1253 
1254 	/* Callers of accept() will be be waiting on the listening socket, not
1255 	 * the pending socket.
1256 	 */
1257 	listener->sk_data_ready(listener);
1258 
1259 	return 0;
1260 
1261 destroy:
1262 	pending->sk_err = skerr;
1263 	pending->sk_state = TCP_CLOSE;
1264 	/* As long as we drop our reference, all necessary cleanup will handle
1265 	 * when the cleanup function drops its reference and our destruct
1266 	 * implementation is called.  Note that since the listen handler will
1267 	 * remove pending from the pending list upon our failure, the cleanup
1268 	 * function won't drop the additional reference, which is why we do it
1269 	 * here.
1270 	 */
1271 	sock_put(pending);
1272 
1273 	return err;
1274 }
1275 
1276 static int
1277 vmci_transport_recv_connecting_client(struct sock *sk,
1278 				      struct vmci_transport_packet *pkt)
1279 {
1280 	struct vsock_sock *vsk;
1281 	int err;
1282 	int skerr;
1283 
1284 	vsk = vsock_sk(sk);
1285 
1286 	switch (pkt->type) {
1287 	case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1288 		if (vmci_handle_is_invalid(pkt->u.handle) ||
1289 		    !vmci_handle_is_equal(pkt->u.handle,
1290 					  vmci_trans(vsk)->qp_handle)) {
1291 			skerr = EPROTO;
1292 			err = -EINVAL;
1293 			goto destroy;
1294 		}
1295 
1296 		/* Signify the socket is connected and wakeup the waiter in
1297 		 * connect(). Also place the socket in the connected table for
1298 		 * accounting (it can already be found since it's in the bound
1299 		 * table).
1300 		 */
1301 		sk->sk_state = TCP_ESTABLISHED;
1302 		sk->sk_socket->state = SS_CONNECTED;
1303 		vsock_insert_connected(vsk);
1304 		sk->sk_state_change(sk);
1305 
1306 		break;
1307 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1308 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1309 		if (pkt->u.size == 0
1310 		    || pkt->dg.src.context != vsk->remote_addr.svm_cid
1311 		    || pkt->src_port != vsk->remote_addr.svm_port
1312 		    || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1313 		    || vmci_trans(vsk)->qpair
1314 		    || vmci_trans(vsk)->produce_size != 0
1315 		    || vmci_trans(vsk)->consume_size != 0
1316 		    || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1317 			skerr = EPROTO;
1318 			err = -EINVAL;
1319 
1320 			goto destroy;
1321 		}
1322 
1323 		err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1324 		if (err) {
1325 			skerr = -err;
1326 			goto destroy;
1327 		}
1328 
1329 		break;
1330 	case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1331 		err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1332 		if (err) {
1333 			skerr = -err;
1334 			goto destroy;
1335 		}
1336 
1337 		break;
1338 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
1339 		/* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1340 		 * continue processing here after they sent an INVALID packet.
1341 		 * This meant that we got a RST after the INVALID. We ignore a
1342 		 * RST after an INVALID. The common code doesn't send the RST
1343 		 * ... so we can hang if an old version of the common code
1344 		 * fails between getting a REQUEST and sending an OFFER back.
1345 		 * Not much we can do about it... except hope that it doesn't
1346 		 * happen.
1347 		 */
1348 		if (vsk->ignore_connecting_rst) {
1349 			vsk->ignore_connecting_rst = false;
1350 		} else {
1351 			skerr = ECONNRESET;
1352 			err = 0;
1353 			goto destroy;
1354 		}
1355 
1356 		break;
1357 	default:
1358 		/* Close and cleanup the connection. */
1359 		skerr = EPROTO;
1360 		err = -EINVAL;
1361 		goto destroy;
1362 	}
1363 
1364 	return 0;
1365 
1366 destroy:
1367 	vmci_transport_send_reset(sk, pkt);
1368 
1369 	sk->sk_state = TCP_CLOSE;
1370 	sk->sk_err = skerr;
1371 	sk->sk_error_report(sk);
1372 	return err;
1373 }
1374 
1375 static int vmci_transport_recv_connecting_client_negotiate(
1376 					struct sock *sk,
1377 					struct vmci_transport_packet *pkt)
1378 {
1379 	int err;
1380 	struct vsock_sock *vsk;
1381 	struct vmci_handle handle;
1382 	struct vmci_qp *qpair;
1383 	u32 detach_sub_id;
1384 	bool is_local;
1385 	u32 flags;
1386 	bool old_proto = true;
1387 	bool old_pkt_proto;
1388 	u16 version;
1389 
1390 	vsk = vsock_sk(sk);
1391 	handle = VMCI_INVALID_HANDLE;
1392 	detach_sub_id = VMCI_INVALID_ID;
1393 
1394 	/* If we have gotten here then we should be past the point where old
1395 	 * linux vsock could have sent the bogus rst.
1396 	 */
1397 	vsk->sent_request = false;
1398 	vsk->ignore_connecting_rst = false;
1399 
1400 	/* Verify that we're OK with the proposed queue pair size */
1401 	if (pkt->u.size < vsk->buffer_min_size ||
1402 	    pkt->u.size > vsk->buffer_max_size) {
1403 		err = -EINVAL;
1404 		goto destroy;
1405 	}
1406 
1407 	/* At this point we know the CID the peer is using to talk to us. */
1408 
1409 	if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1410 		vsk->local_addr.svm_cid = pkt->dg.dst.context;
1411 
1412 	/* Setup the notify ops to be the highest supported version that both
1413 	 * the server and the client support.
1414 	 */
1415 
1416 	if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1417 		old_proto = old_pkt_proto;
1418 	} else {
1419 		if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1420 			old_proto = true;
1421 		else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1422 			old_proto = false;
1423 
1424 	}
1425 
1426 	if (old_proto)
1427 		version = VSOCK_PROTO_INVALID;
1428 	else
1429 		version = pkt->proto;
1430 
1431 	if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1432 		err = -EINVAL;
1433 		goto destroy;
1434 	}
1435 
1436 	/* Subscribe to detach events first.
1437 	 *
1438 	 * XXX We attach once for each queue pair created for now so it is easy
1439 	 * to find the socket (it's provided), but later we should only
1440 	 * subscribe once and add a way to lookup sockets by queue pair handle.
1441 	 */
1442 	err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1443 				   vmci_transport_peer_detach_cb,
1444 				   vmci_trans(vsk), &detach_sub_id);
1445 	if (err < VMCI_SUCCESS) {
1446 		err = vmci_transport_error_to_vsock_error(err);
1447 		goto destroy;
1448 	}
1449 
1450 	/* Make VMCI select the handle for us. */
1451 	handle = VMCI_INVALID_HANDLE;
1452 	is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1453 	flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1454 
1455 	err = vmci_transport_queue_pair_alloc(&qpair,
1456 					      &handle,
1457 					      pkt->u.size,
1458 					      pkt->u.size,
1459 					      vsk->remote_addr.svm_cid,
1460 					      flags,
1461 					      vmci_transport_is_trusted(
1462 						  vsk,
1463 						  vsk->
1464 						  remote_addr.svm_cid));
1465 	if (err < 0)
1466 		goto destroy;
1467 
1468 	err = vmci_transport_send_qp_offer(sk, handle);
1469 	if (err < 0) {
1470 		err = vmci_transport_error_to_vsock_error(err);
1471 		goto destroy;
1472 	}
1473 
1474 	vmci_trans(vsk)->qp_handle = handle;
1475 	vmci_trans(vsk)->qpair = qpair;
1476 
1477 	vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1478 		pkt->u.size;
1479 
1480 	vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1481 
1482 	vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1483 
1484 	return 0;
1485 
1486 destroy:
1487 	if (detach_sub_id != VMCI_INVALID_ID)
1488 		vmci_event_unsubscribe(detach_sub_id);
1489 
1490 	if (!vmci_handle_is_invalid(handle))
1491 		vmci_qpair_detach(&qpair);
1492 
1493 	return err;
1494 }
1495 
1496 static int
1497 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1498 					      struct vmci_transport_packet *pkt)
1499 {
1500 	int err = 0;
1501 	struct vsock_sock *vsk = vsock_sk(sk);
1502 
1503 	if (vsk->sent_request) {
1504 		vsk->sent_request = false;
1505 		vsk->ignore_connecting_rst = true;
1506 
1507 		err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1508 		if (err < 0)
1509 			err = vmci_transport_error_to_vsock_error(err);
1510 		else
1511 			err = 0;
1512 
1513 	}
1514 
1515 	return err;
1516 }
1517 
1518 static int vmci_transport_recv_connected(struct sock *sk,
1519 					 struct vmci_transport_packet *pkt)
1520 {
1521 	struct vsock_sock *vsk;
1522 	bool pkt_processed = false;
1523 
1524 	/* In cases where we are closing the connection, it's sufficient to
1525 	 * mark the state change (and maybe error) and wake up any waiting
1526 	 * threads. Since this is a connected socket, it's owned by a user
1527 	 * process and will be cleaned up when the failure is passed back on
1528 	 * the current or next system call.  Our system call implementations
1529 	 * must therefore check for error and state changes on entry and when
1530 	 * being awoken.
1531 	 */
1532 	switch (pkt->type) {
1533 	case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1534 		if (pkt->u.mode) {
1535 			vsk = vsock_sk(sk);
1536 
1537 			vsk->peer_shutdown |= pkt->u.mode;
1538 			sk->sk_state_change(sk);
1539 		}
1540 		break;
1541 
1542 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
1543 		vsk = vsock_sk(sk);
1544 		/* It is possible that we sent our peer a message (e.g a
1545 		 * WAITING_READ) right before we got notified that the peer had
1546 		 * detached. If that happens then we can get a RST pkt back
1547 		 * from our peer even though there is data available for us to
1548 		 * read. In that case, don't shutdown the socket completely but
1549 		 * instead allow the local client to finish reading data off
1550 		 * the queuepair. Always treat a RST pkt in connected mode like
1551 		 * a clean shutdown.
1552 		 */
1553 		sock_set_flag(sk, SOCK_DONE);
1554 		vsk->peer_shutdown = SHUTDOWN_MASK;
1555 		if (vsock_stream_has_data(vsk) <= 0)
1556 			sk->sk_state = TCP_CLOSING;
1557 
1558 		sk->sk_state_change(sk);
1559 		break;
1560 
1561 	default:
1562 		vsk = vsock_sk(sk);
1563 		vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1564 				sk, pkt, false, NULL, NULL,
1565 				&pkt_processed);
1566 		if (!pkt_processed)
1567 			return -EINVAL;
1568 
1569 		break;
1570 	}
1571 
1572 	return 0;
1573 }
1574 
1575 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1576 				      struct vsock_sock *psk)
1577 {
1578 	vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1579 	if (!vsk->trans)
1580 		return -ENOMEM;
1581 
1582 	vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1583 	vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1584 	vmci_trans(vsk)->qpair = NULL;
1585 	vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1586 	vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1587 	vmci_trans(vsk)->notify_ops = NULL;
1588 	INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1589 	vmci_trans(vsk)->sk = &vsk->sk;
1590 	spin_lock_init(&vmci_trans(vsk)->lock);
1591 
1592 	return 0;
1593 }
1594 
1595 static void vmci_transport_free_resources(struct list_head *transport_list)
1596 {
1597 	while (!list_empty(transport_list)) {
1598 		struct vmci_transport *transport =
1599 		    list_first_entry(transport_list, struct vmci_transport,
1600 				     elem);
1601 		list_del(&transport->elem);
1602 
1603 		if (transport->detach_sub_id != VMCI_INVALID_ID) {
1604 			vmci_event_unsubscribe(transport->detach_sub_id);
1605 			transport->detach_sub_id = VMCI_INVALID_ID;
1606 		}
1607 
1608 		if (!vmci_handle_is_invalid(transport->qp_handle)) {
1609 			vmci_qpair_detach(&transport->qpair);
1610 			transport->qp_handle = VMCI_INVALID_HANDLE;
1611 			transport->produce_size = 0;
1612 			transport->consume_size = 0;
1613 		}
1614 
1615 		kfree(transport);
1616 	}
1617 }
1618 
1619 static void vmci_transport_cleanup(struct work_struct *work)
1620 {
1621 	LIST_HEAD(pending);
1622 
1623 	spin_lock_bh(&vmci_transport_cleanup_lock);
1624 	list_replace_init(&vmci_transport_cleanup_list, &pending);
1625 	spin_unlock_bh(&vmci_transport_cleanup_lock);
1626 	vmci_transport_free_resources(&pending);
1627 }
1628 
1629 static void vmci_transport_destruct(struct vsock_sock *vsk)
1630 {
1631 	/* transport can be NULL if we hit a failure at init() time */
1632 	if (!vmci_trans(vsk))
1633 		return;
1634 
1635 	/* Ensure that the detach callback doesn't use the sk/vsk
1636 	 * we are about to destruct.
1637 	 */
1638 	spin_lock_bh(&vmci_trans(vsk)->lock);
1639 	vmci_trans(vsk)->sk = NULL;
1640 	spin_unlock_bh(&vmci_trans(vsk)->lock);
1641 
1642 	if (vmci_trans(vsk)->notify_ops)
1643 		vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1644 
1645 	spin_lock_bh(&vmci_transport_cleanup_lock);
1646 	list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1647 	spin_unlock_bh(&vmci_transport_cleanup_lock);
1648 	schedule_work(&vmci_transport_cleanup_work);
1649 
1650 	vsk->trans = NULL;
1651 }
1652 
1653 static void vmci_transport_release(struct vsock_sock *vsk)
1654 {
1655 	vsock_remove_sock(vsk);
1656 
1657 	if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1658 		vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1659 		vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1660 	}
1661 }
1662 
1663 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1664 				     struct sockaddr_vm *addr)
1665 {
1666 	u32 port;
1667 	u32 flags;
1668 	int err;
1669 
1670 	/* VMCI will select a resource ID for us if we provide
1671 	 * VMCI_INVALID_ID.
1672 	 */
1673 	port = addr->svm_port == VMADDR_PORT_ANY ?
1674 			VMCI_INVALID_ID : addr->svm_port;
1675 
1676 	if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1677 		return -EACCES;
1678 
1679 	flags = addr->svm_cid == VMADDR_CID_ANY ?
1680 				VMCI_FLAG_ANYCID_DG_HND : 0;
1681 
1682 	err = vmci_transport_datagram_create_hnd(port, flags,
1683 						 vmci_transport_recv_dgram_cb,
1684 						 &vsk->sk,
1685 						 &vmci_trans(vsk)->dg_handle);
1686 	if (err < VMCI_SUCCESS)
1687 		return vmci_transport_error_to_vsock_error(err);
1688 	vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1689 			vmci_trans(vsk)->dg_handle.resource);
1690 
1691 	return 0;
1692 }
1693 
1694 static int vmci_transport_dgram_enqueue(
1695 	struct vsock_sock *vsk,
1696 	struct sockaddr_vm *remote_addr,
1697 	struct msghdr *msg,
1698 	size_t len)
1699 {
1700 	int err;
1701 	struct vmci_datagram *dg;
1702 
1703 	if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1704 		return -EMSGSIZE;
1705 
1706 	if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1707 		return -EPERM;
1708 
1709 	/* Allocate a buffer for the user's message and our packet header. */
1710 	dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1711 	if (!dg)
1712 		return -ENOMEM;
1713 
1714 	memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1715 
1716 	dg->dst = vmci_make_handle(remote_addr->svm_cid,
1717 				   remote_addr->svm_port);
1718 	dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1719 				   vsk->local_addr.svm_port);
1720 	dg->payload_size = len;
1721 
1722 	err = vmci_datagram_send(dg);
1723 	kfree(dg);
1724 	if (err < 0)
1725 		return vmci_transport_error_to_vsock_error(err);
1726 
1727 	return err - sizeof(*dg);
1728 }
1729 
1730 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1731 					struct msghdr *msg, size_t len,
1732 					int flags)
1733 {
1734 	int err;
1735 	int noblock;
1736 	struct vmci_datagram *dg;
1737 	size_t payload_len;
1738 	struct sk_buff *skb;
1739 
1740 	noblock = flags & MSG_DONTWAIT;
1741 
1742 	if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1743 		return -EOPNOTSUPP;
1744 
1745 	/* Retrieve the head sk_buff from the socket's receive queue. */
1746 	err = 0;
1747 	skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1748 	if (!skb)
1749 		return err;
1750 
1751 	dg = (struct vmci_datagram *)skb->data;
1752 	if (!dg)
1753 		/* err is 0, meaning we read zero bytes. */
1754 		goto out;
1755 
1756 	payload_len = dg->payload_size;
1757 	/* Ensure the sk_buff matches the payload size claimed in the packet. */
1758 	if (payload_len != skb->len - sizeof(*dg)) {
1759 		err = -EINVAL;
1760 		goto out;
1761 	}
1762 
1763 	if (payload_len > len) {
1764 		payload_len = len;
1765 		msg->msg_flags |= MSG_TRUNC;
1766 	}
1767 
1768 	/* Place the datagram payload in the user's iovec. */
1769 	err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1770 	if (err)
1771 		goto out;
1772 
1773 	if (msg->msg_name) {
1774 		/* Provide the address of the sender. */
1775 		DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1776 		vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1777 		msg->msg_namelen = sizeof(*vm_addr);
1778 	}
1779 	err = payload_len;
1780 
1781 out:
1782 	skb_free_datagram(&vsk->sk, skb);
1783 	return err;
1784 }
1785 
1786 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1787 {
1788 	if (cid == VMADDR_CID_HYPERVISOR) {
1789 		/* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1790 		 * state and are allowed.
1791 		 */
1792 		return port == VMCI_UNITY_PBRPC_REGISTER;
1793 	}
1794 
1795 	return true;
1796 }
1797 
1798 static int vmci_transport_connect(struct vsock_sock *vsk)
1799 {
1800 	int err;
1801 	bool old_pkt_proto = false;
1802 	struct sock *sk = &vsk->sk;
1803 
1804 	if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1805 		old_pkt_proto) {
1806 		err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1807 		if (err < 0) {
1808 			sk->sk_state = TCP_CLOSE;
1809 			return err;
1810 		}
1811 	} else {
1812 		int supported_proto_versions =
1813 			vmci_transport_new_proto_supported_versions();
1814 		err = vmci_transport_send_conn_request2(sk, vsk->buffer_size,
1815 				supported_proto_versions);
1816 		if (err < 0) {
1817 			sk->sk_state = TCP_CLOSE;
1818 			return err;
1819 		}
1820 
1821 		vsk->sent_request = true;
1822 	}
1823 
1824 	return err;
1825 }
1826 
1827 static ssize_t vmci_transport_stream_dequeue(
1828 	struct vsock_sock *vsk,
1829 	struct msghdr *msg,
1830 	size_t len,
1831 	int flags)
1832 {
1833 	if (flags & MSG_PEEK)
1834 		return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1835 	else
1836 		return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1837 }
1838 
1839 static ssize_t vmci_transport_stream_enqueue(
1840 	struct vsock_sock *vsk,
1841 	struct msghdr *msg,
1842 	size_t len)
1843 {
1844 	return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1845 }
1846 
1847 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1848 {
1849 	return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1850 }
1851 
1852 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1853 {
1854 	return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1855 }
1856 
1857 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1858 {
1859 	return vmci_trans(vsk)->consume_size;
1860 }
1861 
1862 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1863 {
1864 	return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1865 }
1866 
1867 static int vmci_transport_notify_poll_in(
1868 	struct vsock_sock *vsk,
1869 	size_t target,
1870 	bool *data_ready_now)
1871 {
1872 	return vmci_trans(vsk)->notify_ops->poll_in(
1873 			&vsk->sk, target, data_ready_now);
1874 }
1875 
1876 static int vmci_transport_notify_poll_out(
1877 	struct vsock_sock *vsk,
1878 	size_t target,
1879 	bool *space_available_now)
1880 {
1881 	return vmci_trans(vsk)->notify_ops->poll_out(
1882 			&vsk->sk, target, space_available_now);
1883 }
1884 
1885 static int vmci_transport_notify_recv_init(
1886 	struct vsock_sock *vsk,
1887 	size_t target,
1888 	struct vsock_transport_recv_notify_data *data)
1889 {
1890 	return vmci_trans(vsk)->notify_ops->recv_init(
1891 			&vsk->sk, target,
1892 			(struct vmci_transport_recv_notify_data *)data);
1893 }
1894 
1895 static int vmci_transport_notify_recv_pre_block(
1896 	struct vsock_sock *vsk,
1897 	size_t target,
1898 	struct vsock_transport_recv_notify_data *data)
1899 {
1900 	return vmci_trans(vsk)->notify_ops->recv_pre_block(
1901 			&vsk->sk, target,
1902 			(struct vmci_transport_recv_notify_data *)data);
1903 }
1904 
1905 static int vmci_transport_notify_recv_pre_dequeue(
1906 	struct vsock_sock *vsk,
1907 	size_t target,
1908 	struct vsock_transport_recv_notify_data *data)
1909 {
1910 	return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1911 			&vsk->sk, target,
1912 			(struct vmci_transport_recv_notify_data *)data);
1913 }
1914 
1915 static int vmci_transport_notify_recv_post_dequeue(
1916 	struct vsock_sock *vsk,
1917 	size_t target,
1918 	ssize_t copied,
1919 	bool data_read,
1920 	struct vsock_transport_recv_notify_data *data)
1921 {
1922 	return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1923 			&vsk->sk, target, copied, data_read,
1924 			(struct vmci_transport_recv_notify_data *)data);
1925 }
1926 
1927 static int vmci_transport_notify_send_init(
1928 	struct vsock_sock *vsk,
1929 	struct vsock_transport_send_notify_data *data)
1930 {
1931 	return vmci_trans(vsk)->notify_ops->send_init(
1932 			&vsk->sk,
1933 			(struct vmci_transport_send_notify_data *)data);
1934 }
1935 
1936 static int vmci_transport_notify_send_pre_block(
1937 	struct vsock_sock *vsk,
1938 	struct vsock_transport_send_notify_data *data)
1939 {
1940 	return vmci_trans(vsk)->notify_ops->send_pre_block(
1941 			&vsk->sk,
1942 			(struct vmci_transport_send_notify_data *)data);
1943 }
1944 
1945 static int vmci_transport_notify_send_pre_enqueue(
1946 	struct vsock_sock *vsk,
1947 	struct vsock_transport_send_notify_data *data)
1948 {
1949 	return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1950 			&vsk->sk,
1951 			(struct vmci_transport_send_notify_data *)data);
1952 }
1953 
1954 static int vmci_transport_notify_send_post_enqueue(
1955 	struct vsock_sock *vsk,
1956 	ssize_t written,
1957 	struct vsock_transport_send_notify_data *data)
1958 {
1959 	return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1960 			&vsk->sk, written,
1961 			(struct vmci_transport_send_notify_data *)data);
1962 }
1963 
1964 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1965 {
1966 	if (PROTOCOL_OVERRIDE != -1) {
1967 		if (PROTOCOL_OVERRIDE == 0)
1968 			*old_pkt_proto = true;
1969 		else
1970 			*old_pkt_proto = false;
1971 
1972 		pr_info("Proto override in use\n");
1973 		return true;
1974 	}
1975 
1976 	return false;
1977 }
1978 
1979 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
1980 						  u16 *proto,
1981 						  bool old_pkt_proto)
1982 {
1983 	struct vsock_sock *vsk = vsock_sk(sk);
1984 
1985 	if (old_pkt_proto) {
1986 		if (*proto != VSOCK_PROTO_INVALID) {
1987 			pr_err("Can't set both an old and new protocol\n");
1988 			return false;
1989 		}
1990 		vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
1991 		goto exit;
1992 	}
1993 
1994 	switch (*proto) {
1995 	case VSOCK_PROTO_PKT_ON_NOTIFY:
1996 		vmci_trans(vsk)->notify_ops =
1997 			&vmci_transport_notify_pkt_q_state_ops;
1998 		break;
1999 	default:
2000 		pr_err("Unknown notify protocol version\n");
2001 		return false;
2002 	}
2003 
2004 exit:
2005 	vmci_trans(vsk)->notify_ops->socket_init(sk);
2006 	return true;
2007 }
2008 
2009 static u16 vmci_transport_new_proto_supported_versions(void)
2010 {
2011 	if (PROTOCOL_OVERRIDE != -1)
2012 		return PROTOCOL_OVERRIDE;
2013 
2014 	return VSOCK_PROTO_ALL_SUPPORTED;
2015 }
2016 
2017 static u32 vmci_transport_get_local_cid(void)
2018 {
2019 	return vmci_get_context_id();
2020 }
2021 
2022 static struct vsock_transport vmci_transport = {
2023 	.module = THIS_MODULE,
2024 	.init = vmci_transport_socket_init,
2025 	.destruct = vmci_transport_destruct,
2026 	.release = vmci_transport_release,
2027 	.connect = vmci_transport_connect,
2028 	.dgram_bind = vmci_transport_dgram_bind,
2029 	.dgram_dequeue = vmci_transport_dgram_dequeue,
2030 	.dgram_enqueue = vmci_transport_dgram_enqueue,
2031 	.dgram_allow = vmci_transport_dgram_allow,
2032 	.stream_dequeue = vmci_transport_stream_dequeue,
2033 	.stream_enqueue = vmci_transport_stream_enqueue,
2034 	.stream_has_data = vmci_transport_stream_has_data,
2035 	.stream_has_space = vmci_transport_stream_has_space,
2036 	.stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2037 	.stream_is_active = vmci_transport_stream_is_active,
2038 	.stream_allow = vmci_transport_stream_allow,
2039 	.notify_poll_in = vmci_transport_notify_poll_in,
2040 	.notify_poll_out = vmci_transport_notify_poll_out,
2041 	.notify_recv_init = vmci_transport_notify_recv_init,
2042 	.notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2043 	.notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2044 	.notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2045 	.notify_send_init = vmci_transport_notify_send_init,
2046 	.notify_send_pre_block = vmci_transport_notify_send_pre_block,
2047 	.notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2048 	.notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2049 	.shutdown = vmci_transport_shutdown,
2050 	.get_local_cid = vmci_transport_get_local_cid,
2051 };
2052 
2053 static bool vmci_check_transport(struct vsock_sock *vsk)
2054 {
2055 	return vsk->transport == &vmci_transport;
2056 }
2057 
2058 void vmci_vsock_transport_cb(bool is_host)
2059 {
2060 	int features;
2061 
2062 	if (is_host)
2063 		features = VSOCK_TRANSPORT_F_H2G;
2064 	else
2065 		features = VSOCK_TRANSPORT_F_G2H;
2066 
2067 	vsock_core_register(&vmci_transport, features);
2068 }
2069 
2070 static int __init vmci_transport_init(void)
2071 {
2072 	int err;
2073 
2074 	/* Create the datagram handle that we will use to send and receive all
2075 	 * VSocket control messages for this context.
2076 	 */
2077 	err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2078 						 VMCI_FLAG_ANYCID_DG_HND,
2079 						 vmci_transport_recv_stream_cb,
2080 						 NULL,
2081 						 &vmci_transport_stream_handle);
2082 	if (err < VMCI_SUCCESS) {
2083 		pr_err("Unable to create datagram handle. (%d)\n", err);
2084 		return vmci_transport_error_to_vsock_error(err);
2085 	}
2086 	err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2087 				   vmci_transport_qp_resumed_cb,
2088 				   NULL, &vmci_transport_qp_resumed_sub_id);
2089 	if (err < VMCI_SUCCESS) {
2090 		pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2091 		err = vmci_transport_error_to_vsock_error(err);
2092 		vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2093 		goto err_destroy_stream_handle;
2094 	}
2095 
2096 	/* Register only with dgram feature, other features (H2G, G2H) will be
2097 	 * registered when the first host or guest becomes active.
2098 	 */
2099 	err = vsock_core_register(&vmci_transport, VSOCK_TRANSPORT_F_DGRAM);
2100 	if (err < 0)
2101 		goto err_unsubscribe;
2102 
2103 	err = vmci_register_vsock_callback(vmci_vsock_transport_cb);
2104 	if (err < 0)
2105 		goto err_unregister;
2106 
2107 	return 0;
2108 
2109 err_unregister:
2110 	vsock_core_unregister(&vmci_transport);
2111 err_unsubscribe:
2112 	vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2113 err_destroy_stream_handle:
2114 	vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2115 	return err;
2116 }
2117 module_init(vmci_transport_init);
2118 
2119 static void __exit vmci_transport_exit(void)
2120 {
2121 	cancel_work_sync(&vmci_transport_cleanup_work);
2122 	vmci_transport_free_resources(&vmci_transport_cleanup_list);
2123 
2124 	if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2125 		if (vmci_datagram_destroy_handle(
2126 			vmci_transport_stream_handle) != VMCI_SUCCESS)
2127 			pr_err("Couldn't destroy datagram handle\n");
2128 		vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2129 	}
2130 
2131 	if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2132 		vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2133 		vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2134 	}
2135 
2136 	vmci_register_vsock_callback(NULL);
2137 	vsock_core_unregister(&vmci_transport);
2138 }
2139 module_exit(vmci_transport_exit);
2140 
2141 MODULE_AUTHOR("VMware, Inc.");
2142 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2143 MODULE_VERSION("1.0.5.0-k");
2144 MODULE_LICENSE("GPL v2");
2145 MODULE_ALIAS("vmware_vsock");
2146 MODULE_ALIAS_NETPROTO(PF_VSOCK);
2147