xref: /linux/net/vmw_vsock/vmci_transport.c (revision 9c39c6ffe0c2945c7cf814814c096bc23b63f53d)
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_once("Could not attach to queue pair with %d\n", err);
572 		err = vmci_transport_error_to_vsock_error(err);
573 	}
574 
575 	return err;
576 }
577 
578 static int
579 vmci_transport_datagram_create_hnd(u32 resource_id,
580 				   u32 flags,
581 				   vmci_datagram_recv_cb recv_cb,
582 				   void *client_data,
583 				   struct vmci_handle *out_handle)
584 {
585 	int err = 0;
586 
587 	/* Try to allocate our datagram handler as trusted. This will only work
588 	 * if vsock is running in the host.
589 	 */
590 
591 	err = vmci_datagram_create_handle_priv(resource_id, flags,
592 					       VMCI_PRIVILEGE_FLAG_TRUSTED,
593 					       recv_cb,
594 					       client_data, out_handle);
595 
596 	if (err == VMCI_ERROR_NO_ACCESS)
597 		err = vmci_datagram_create_handle(resource_id, flags,
598 						  recv_cb, client_data,
599 						  out_handle);
600 
601 	return err;
602 }
603 
604 /* This is invoked as part of a tasklet that's scheduled when the VMCI
605  * interrupt fires.  This is run in bottom-half context and if it ever needs to
606  * sleep it should defer that work to a work queue.
607  */
608 
609 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
610 {
611 	struct sock *sk;
612 	size_t size;
613 	struct sk_buff *skb;
614 	struct vsock_sock *vsk;
615 
616 	sk = (struct sock *)data;
617 
618 	/* This handler is privileged when this module is running on the host.
619 	 * We will get datagrams from all endpoints (even VMs that are in a
620 	 * restricted context). If we get one from a restricted context then
621 	 * the destination socket must be trusted.
622 	 *
623 	 * NOTE: We access the socket struct without holding the lock here.
624 	 * This is ok because the field we are interested is never modified
625 	 * outside of the create and destruct socket functions.
626 	 */
627 	vsk = vsock_sk(sk);
628 	if (!vmci_transport_allow_dgram(vsk, dg->src.context))
629 		return VMCI_ERROR_NO_ACCESS;
630 
631 	size = VMCI_DG_SIZE(dg);
632 
633 	/* Attach the packet to the socket's receive queue as an sk_buff. */
634 	skb = alloc_skb(size, GFP_ATOMIC);
635 	if (!skb)
636 		return VMCI_ERROR_NO_MEM;
637 
638 	/* sk_receive_skb() will do a sock_put(), so hold here. */
639 	sock_hold(sk);
640 	skb_put(skb, size);
641 	memcpy(skb->data, dg, size);
642 	sk_receive_skb(sk, skb, 0);
643 
644 	return VMCI_SUCCESS;
645 }
646 
647 static bool vmci_transport_stream_allow(u32 cid, u32 port)
648 {
649 	static const u32 non_socket_contexts[] = {
650 		VMADDR_CID_LOCAL,
651 	};
652 	int i;
653 
654 	BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
655 
656 	for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
657 		if (cid == non_socket_contexts[i])
658 			return false;
659 	}
660 
661 	return true;
662 }
663 
664 /* This is invoked as part of a tasklet that's scheduled when the VMCI
665  * interrupt fires.  This is run in bottom-half context but it defers most of
666  * its work to the packet handling work queue.
667  */
668 
669 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
670 {
671 	struct sock *sk;
672 	struct sockaddr_vm dst;
673 	struct sockaddr_vm src;
674 	struct vmci_transport_packet *pkt;
675 	struct vsock_sock *vsk;
676 	bool bh_process_pkt;
677 	int err;
678 
679 	sk = NULL;
680 	err = VMCI_SUCCESS;
681 	bh_process_pkt = false;
682 
683 	/* Ignore incoming packets from contexts without sockets, or resources
684 	 * that aren't vsock implementations.
685 	 */
686 
687 	if (!vmci_transport_stream_allow(dg->src.context, -1)
688 	    || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
689 		return VMCI_ERROR_NO_ACCESS;
690 
691 	if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
692 		/* Drop datagrams that do not contain full VSock packets. */
693 		return VMCI_ERROR_INVALID_ARGS;
694 
695 	pkt = (struct vmci_transport_packet *)dg;
696 
697 	/* Find the socket that should handle this packet.  First we look for a
698 	 * connected socket and if there is none we look for a socket bound to
699 	 * the destintation address.
700 	 */
701 	vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
702 	vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
703 
704 	sk = vsock_find_connected_socket(&src, &dst);
705 	if (!sk) {
706 		sk = vsock_find_bound_socket(&dst);
707 		if (!sk) {
708 			/* We could not find a socket for this specified
709 			 * address.  If this packet is a RST, we just drop it.
710 			 * If it is another packet, we send a RST.  Note that
711 			 * we do not send a RST reply to RSTs so that we do not
712 			 * continually send RSTs between two endpoints.
713 			 *
714 			 * Note that since this is a reply, dst is src and src
715 			 * is dst.
716 			 */
717 			if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
718 				pr_err("unable to send reset\n");
719 
720 			err = VMCI_ERROR_NOT_FOUND;
721 			goto out;
722 		}
723 	}
724 
725 	/* If the received packet type is beyond all types known to this
726 	 * implementation, reply with an invalid message.  Hopefully this will
727 	 * help when implementing backwards compatibility in the future.
728 	 */
729 	if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
730 		vmci_transport_send_invalid_bh(&dst, &src);
731 		err = VMCI_ERROR_INVALID_ARGS;
732 		goto out;
733 	}
734 
735 	/* This handler is privileged when this module is running on the host.
736 	 * We will get datagram connect requests from all endpoints (even VMs
737 	 * that are in a restricted context). If we get one from a restricted
738 	 * context then the destination socket must be trusted.
739 	 *
740 	 * NOTE: We access the socket struct without holding the lock here.
741 	 * This is ok because the field we are interested is never modified
742 	 * outside of the create and destruct socket functions.
743 	 */
744 	vsk = vsock_sk(sk);
745 	if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
746 		err = VMCI_ERROR_NO_ACCESS;
747 		goto out;
748 	}
749 
750 	/* We do most everything in a work queue, but let's fast path the
751 	 * notification of reads and writes to help data transfer performance.
752 	 * We can only do this if there is no process context code executing
753 	 * for this socket since that may change the state.
754 	 */
755 	bh_lock_sock(sk);
756 
757 	if (!sock_owned_by_user(sk)) {
758 		/* The local context ID may be out of date, update it. */
759 		vsk->local_addr.svm_cid = dst.svm_cid;
760 
761 		if (sk->sk_state == TCP_ESTABLISHED)
762 			vmci_trans(vsk)->notify_ops->handle_notify_pkt(
763 					sk, pkt, true, &dst, &src,
764 					&bh_process_pkt);
765 	}
766 
767 	bh_unlock_sock(sk);
768 
769 	if (!bh_process_pkt) {
770 		struct vmci_transport_recv_pkt_info *recv_pkt_info;
771 
772 		recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
773 		if (!recv_pkt_info) {
774 			if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
775 				pr_err("unable to send reset\n");
776 
777 			err = VMCI_ERROR_NO_MEM;
778 			goto out;
779 		}
780 
781 		recv_pkt_info->sk = sk;
782 		memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
783 		INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
784 
785 		schedule_work(&recv_pkt_info->work);
786 		/* Clear sk so that the reference count incremented by one of
787 		 * the Find functions above is not decremented below.  We need
788 		 * that reference count for the packet handler we've scheduled
789 		 * to run.
790 		 */
791 		sk = NULL;
792 	}
793 
794 out:
795 	if (sk)
796 		sock_put(sk);
797 
798 	return err;
799 }
800 
801 static void vmci_transport_handle_detach(struct sock *sk)
802 {
803 	struct vsock_sock *vsk;
804 
805 	vsk = vsock_sk(sk);
806 	if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
807 		sock_set_flag(sk, SOCK_DONE);
808 
809 		/* On a detach the peer will not be sending or receiving
810 		 * anymore.
811 		 */
812 		vsk->peer_shutdown = SHUTDOWN_MASK;
813 
814 		/* We should not be sending anymore since the peer won't be
815 		 * there to receive, but we can still receive if there is data
816 		 * left in our consume queue. If the local endpoint is a host,
817 		 * we can't call vsock_stream_has_data, since that may block,
818 		 * but a host endpoint can't read data once the VM has
819 		 * detached, so there is no available data in that case.
820 		 */
821 		if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
822 		    vsock_stream_has_data(vsk) <= 0) {
823 			if (sk->sk_state == TCP_SYN_SENT) {
824 				/* The peer may detach from a queue pair while
825 				 * we are still in the connecting state, i.e.,
826 				 * if the peer VM is killed after attaching to
827 				 * a queue pair, but before we complete the
828 				 * handshake. In that case, we treat the detach
829 				 * event like a reset.
830 				 */
831 
832 				sk->sk_state = TCP_CLOSE;
833 				sk->sk_err = ECONNRESET;
834 				sk->sk_error_report(sk);
835 				return;
836 			}
837 			sk->sk_state = TCP_CLOSE;
838 		}
839 		sk->sk_state_change(sk);
840 	}
841 }
842 
843 static void vmci_transport_peer_detach_cb(u32 sub_id,
844 					  const struct vmci_event_data *e_data,
845 					  void *client_data)
846 {
847 	struct vmci_transport *trans = client_data;
848 	const struct vmci_event_payload_qp *e_payload;
849 
850 	e_payload = vmci_event_data_const_payload(e_data);
851 
852 	/* XXX This is lame, we should provide a way to lookup sockets by
853 	 * qp_handle.
854 	 */
855 	if (vmci_handle_is_invalid(e_payload->handle) ||
856 	    !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
857 		return;
858 
859 	/* We don't ask for delayed CBs when we subscribe to this event (we
860 	 * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
861 	 * guarantees in that case about what context we might be running in,
862 	 * so it could be BH or process, blockable or non-blockable.  So we
863 	 * need to account for all possible contexts here.
864 	 */
865 	spin_lock_bh(&trans->lock);
866 	if (!trans->sk)
867 		goto out;
868 
869 	/* Apart from here, trans->lock is only grabbed as part of sk destruct,
870 	 * where trans->sk isn't locked.
871 	 */
872 	bh_lock_sock(trans->sk);
873 
874 	vmci_transport_handle_detach(trans->sk);
875 
876 	bh_unlock_sock(trans->sk);
877  out:
878 	spin_unlock_bh(&trans->lock);
879 }
880 
881 static void vmci_transport_qp_resumed_cb(u32 sub_id,
882 					 const struct vmci_event_data *e_data,
883 					 void *client_data)
884 {
885 	vsock_for_each_connected_socket(vmci_transport_handle_detach);
886 }
887 
888 static void vmci_transport_recv_pkt_work(struct work_struct *work)
889 {
890 	struct vmci_transport_recv_pkt_info *recv_pkt_info;
891 	struct vmci_transport_packet *pkt;
892 	struct sock *sk;
893 
894 	recv_pkt_info =
895 		container_of(work, struct vmci_transport_recv_pkt_info, work);
896 	sk = recv_pkt_info->sk;
897 	pkt = &recv_pkt_info->pkt;
898 
899 	lock_sock(sk);
900 
901 	/* The local context ID may be out of date. */
902 	vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
903 
904 	switch (sk->sk_state) {
905 	case TCP_LISTEN:
906 		vmci_transport_recv_listen(sk, pkt);
907 		break;
908 	case TCP_SYN_SENT:
909 		/* Processing of pending connections for servers goes through
910 		 * the listening socket, so see vmci_transport_recv_listen()
911 		 * for that path.
912 		 */
913 		vmci_transport_recv_connecting_client(sk, pkt);
914 		break;
915 	case TCP_ESTABLISHED:
916 		vmci_transport_recv_connected(sk, pkt);
917 		break;
918 	default:
919 		/* Because this function does not run in the same context as
920 		 * vmci_transport_recv_stream_cb it is possible that the
921 		 * socket has closed. We need to let the other side know or it
922 		 * could be sitting in a connect and hang forever. Send a
923 		 * reset to prevent that.
924 		 */
925 		vmci_transport_send_reset(sk, pkt);
926 		break;
927 	}
928 
929 	release_sock(sk);
930 	kfree(recv_pkt_info);
931 	/* Release reference obtained in the stream callback when we fetched
932 	 * this socket out of the bound or connected list.
933 	 */
934 	sock_put(sk);
935 }
936 
937 static int vmci_transport_recv_listen(struct sock *sk,
938 				      struct vmci_transport_packet *pkt)
939 {
940 	struct sock *pending;
941 	struct vsock_sock *vpending;
942 	int err;
943 	u64 qp_size;
944 	bool old_request = false;
945 	bool old_pkt_proto = false;
946 
947 	err = 0;
948 
949 	/* Because we are in the listen state, we could be receiving a packet
950 	 * for ourself or any previous connection requests that we received.
951 	 * If it's the latter, we try to find a socket in our list of pending
952 	 * connections and, if we do, call the appropriate handler for the
953 	 * state that that socket is in.  Otherwise we try to service the
954 	 * connection request.
955 	 */
956 	pending = vmci_transport_get_pending(sk, pkt);
957 	if (pending) {
958 		lock_sock(pending);
959 
960 		/* The local context ID may be out of date. */
961 		vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
962 
963 		switch (pending->sk_state) {
964 		case TCP_SYN_SENT:
965 			err = vmci_transport_recv_connecting_server(sk,
966 								    pending,
967 								    pkt);
968 			break;
969 		default:
970 			vmci_transport_send_reset(pending, pkt);
971 			err = -EINVAL;
972 		}
973 
974 		if (err < 0)
975 			vsock_remove_pending(sk, pending);
976 
977 		release_sock(pending);
978 		vmci_transport_release_pending(pending);
979 
980 		return err;
981 	}
982 
983 	/* The listen state only accepts connection requests.  Reply with a
984 	 * reset unless we received a reset.
985 	 */
986 
987 	if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
988 	      pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
989 		vmci_transport_reply_reset(pkt);
990 		return -EINVAL;
991 	}
992 
993 	if (pkt->u.size == 0) {
994 		vmci_transport_reply_reset(pkt);
995 		return -EINVAL;
996 	}
997 
998 	/* If this socket can't accommodate this connection request, we send a
999 	 * reset.  Otherwise we create and initialize a child socket and reply
1000 	 * with a connection negotiation.
1001 	 */
1002 	if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1003 		vmci_transport_reply_reset(pkt);
1004 		return -ECONNREFUSED;
1005 	}
1006 
1007 	pending = vsock_create_connected(sk);
1008 	if (!pending) {
1009 		vmci_transport_send_reset(sk, pkt);
1010 		return -ENOMEM;
1011 	}
1012 
1013 	vpending = vsock_sk(pending);
1014 
1015 	vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1016 			pkt->dst_port);
1017 	vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1018 			pkt->src_port);
1019 
1020 	err = vsock_assign_transport(vpending, vsock_sk(sk));
1021 	/* Transport assigned (looking at remote_addr) must be the same
1022 	 * where we received the request.
1023 	 */
1024 	if (err || !vmci_check_transport(vpending)) {
1025 		vmci_transport_send_reset(sk, pkt);
1026 		sock_put(pending);
1027 		return err;
1028 	}
1029 
1030 	/* If the proposed size fits within our min/max, accept it. Otherwise
1031 	 * propose our own size.
1032 	 */
1033 	if (pkt->u.size >= vpending->buffer_min_size &&
1034 	    pkt->u.size <= vpending->buffer_max_size) {
1035 		qp_size = pkt->u.size;
1036 	} else {
1037 		qp_size = vpending->buffer_size;
1038 	}
1039 
1040 	/* Figure out if we are using old or new requests based on the
1041 	 * overrides pkt types sent by our peer.
1042 	 */
1043 	if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1044 		old_request = old_pkt_proto;
1045 	} else {
1046 		if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1047 			old_request = true;
1048 		else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1049 			old_request = false;
1050 
1051 	}
1052 
1053 	if (old_request) {
1054 		/* Handle a REQUEST (or override) */
1055 		u16 version = VSOCK_PROTO_INVALID;
1056 		if (vmci_transport_proto_to_notify_struct(
1057 			pending, &version, true))
1058 			err = vmci_transport_send_negotiate(pending, qp_size);
1059 		else
1060 			err = -EINVAL;
1061 
1062 	} else {
1063 		/* Handle a REQUEST2 (or override) */
1064 		int proto_int = pkt->proto;
1065 		int pos;
1066 		u16 active_proto_version = 0;
1067 
1068 		/* The list of possible protocols is the intersection of all
1069 		 * protocols the client supports ... plus all the protocols we
1070 		 * support.
1071 		 */
1072 		proto_int &= vmci_transport_new_proto_supported_versions();
1073 
1074 		/* We choose the highest possible protocol version and use that
1075 		 * one.
1076 		 */
1077 		pos = fls(proto_int);
1078 		if (pos) {
1079 			active_proto_version = (1 << (pos - 1));
1080 			if (vmci_transport_proto_to_notify_struct(
1081 				pending, &active_proto_version, false))
1082 				err = vmci_transport_send_negotiate2(pending,
1083 							qp_size,
1084 							active_proto_version);
1085 			else
1086 				err = -EINVAL;
1087 
1088 		} else {
1089 			err = -EINVAL;
1090 		}
1091 	}
1092 
1093 	if (err < 0) {
1094 		vmci_transport_send_reset(sk, pkt);
1095 		sock_put(pending);
1096 		err = vmci_transport_error_to_vsock_error(err);
1097 		goto out;
1098 	}
1099 
1100 	vsock_add_pending(sk, pending);
1101 	sk_acceptq_added(sk);
1102 
1103 	pending->sk_state = TCP_SYN_SENT;
1104 	vmci_trans(vpending)->produce_size =
1105 		vmci_trans(vpending)->consume_size = qp_size;
1106 	vpending->buffer_size = qp_size;
1107 
1108 	vmci_trans(vpending)->notify_ops->process_request(pending);
1109 
1110 	/* We might never receive another message for this socket and it's not
1111 	 * connected to any process, so we have to ensure it gets cleaned up
1112 	 * ourself.  Our delayed work function will take care of that.  Note
1113 	 * that we do not ever cancel this function since we have few
1114 	 * guarantees about its state when calling cancel_delayed_work().
1115 	 * Instead we hold a reference on the socket for that function and make
1116 	 * it capable of handling cases where it needs to do nothing but
1117 	 * release that reference.
1118 	 */
1119 	vpending->listener = sk;
1120 	sock_hold(sk);
1121 	sock_hold(pending);
1122 	schedule_delayed_work(&vpending->pending_work, HZ);
1123 
1124 out:
1125 	return err;
1126 }
1127 
1128 static int
1129 vmci_transport_recv_connecting_server(struct sock *listener,
1130 				      struct sock *pending,
1131 				      struct vmci_transport_packet *pkt)
1132 {
1133 	struct vsock_sock *vpending;
1134 	struct vmci_handle handle;
1135 	struct vmci_qp *qpair;
1136 	bool is_local;
1137 	u32 flags;
1138 	u32 detach_sub_id;
1139 	int err;
1140 	int skerr;
1141 
1142 	vpending = vsock_sk(pending);
1143 	detach_sub_id = VMCI_INVALID_ID;
1144 
1145 	switch (pkt->type) {
1146 	case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1147 		if (vmci_handle_is_invalid(pkt->u.handle)) {
1148 			vmci_transport_send_reset(pending, pkt);
1149 			skerr = EPROTO;
1150 			err = -EINVAL;
1151 			goto destroy;
1152 		}
1153 		break;
1154 	default:
1155 		/* Close and cleanup the connection. */
1156 		vmci_transport_send_reset(pending, pkt);
1157 		skerr = EPROTO;
1158 		err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1159 		goto destroy;
1160 	}
1161 
1162 	/* In order to complete the connection we need to attach to the offered
1163 	 * queue pair and send an attach notification.  We also subscribe to the
1164 	 * detach event so we know when our peer goes away, and we do that
1165 	 * before attaching so we don't miss an event.  If all this succeeds,
1166 	 * we update our state and wakeup anything waiting in accept() for a
1167 	 * connection.
1168 	 */
1169 
1170 	/* We don't care about attach since we ensure the other side has
1171 	 * attached by specifying the ATTACH_ONLY flag below.
1172 	 */
1173 	err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1174 				   vmci_transport_peer_detach_cb,
1175 				   vmci_trans(vpending), &detach_sub_id);
1176 	if (err < VMCI_SUCCESS) {
1177 		vmci_transport_send_reset(pending, pkt);
1178 		err = vmci_transport_error_to_vsock_error(err);
1179 		skerr = -err;
1180 		goto destroy;
1181 	}
1182 
1183 	vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1184 
1185 	/* Now attach to the queue pair the client created. */
1186 	handle = pkt->u.handle;
1187 
1188 	/* vpending->local_addr always has a context id so we do not need to
1189 	 * worry about VMADDR_CID_ANY in this case.
1190 	 */
1191 	is_local =
1192 	    vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1193 	flags = VMCI_QPFLAG_ATTACH_ONLY;
1194 	flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1195 
1196 	err = vmci_transport_queue_pair_alloc(
1197 					&qpair,
1198 					&handle,
1199 					vmci_trans(vpending)->produce_size,
1200 					vmci_trans(vpending)->consume_size,
1201 					pkt->dg.src.context,
1202 					flags,
1203 					vmci_transport_is_trusted(
1204 						vpending,
1205 						vpending->remote_addr.svm_cid));
1206 	if (err < 0) {
1207 		vmci_transport_send_reset(pending, pkt);
1208 		skerr = -err;
1209 		goto destroy;
1210 	}
1211 
1212 	vmci_trans(vpending)->qp_handle = handle;
1213 	vmci_trans(vpending)->qpair = qpair;
1214 
1215 	/* When we send the attach message, we must be ready to handle incoming
1216 	 * control messages on the newly connected socket. So we move the
1217 	 * pending socket to the connected state before sending the attach
1218 	 * message. Otherwise, an incoming packet triggered by the attach being
1219 	 * received by the peer may be processed concurrently with what happens
1220 	 * below after sending the attach message, and that incoming packet
1221 	 * will find the listening socket instead of the (currently) pending
1222 	 * socket. Note that enqueueing the socket increments the reference
1223 	 * count, so even if a reset comes before the connection is accepted,
1224 	 * the socket will be valid until it is removed from the queue.
1225 	 *
1226 	 * If we fail sending the attach below, we remove the socket from the
1227 	 * connected list and move the socket to TCP_CLOSE before
1228 	 * releasing the lock, so a pending slow path processing of an incoming
1229 	 * packet will not see the socket in the connected state in that case.
1230 	 */
1231 	pending->sk_state = TCP_ESTABLISHED;
1232 
1233 	vsock_insert_connected(vpending);
1234 
1235 	/* Notify our peer of our attach. */
1236 	err = vmci_transport_send_attach(pending, handle);
1237 	if (err < 0) {
1238 		vsock_remove_connected(vpending);
1239 		pr_err("Could not send attach\n");
1240 		vmci_transport_send_reset(pending, pkt);
1241 		err = vmci_transport_error_to_vsock_error(err);
1242 		skerr = -err;
1243 		goto destroy;
1244 	}
1245 
1246 	/* We have a connection. Move the now connected socket from the
1247 	 * listener's pending list to the accept queue so callers of accept()
1248 	 * can find it.
1249 	 */
1250 	vsock_remove_pending(listener, pending);
1251 	vsock_enqueue_accept(listener, pending);
1252 
1253 	/* Callers of accept() will be be waiting on the listening socket, not
1254 	 * the pending socket.
1255 	 */
1256 	listener->sk_data_ready(listener);
1257 
1258 	return 0;
1259 
1260 destroy:
1261 	pending->sk_err = skerr;
1262 	pending->sk_state = TCP_CLOSE;
1263 	/* As long as we drop our reference, all necessary cleanup will handle
1264 	 * when the cleanup function drops its reference and our destruct
1265 	 * implementation is called.  Note that since the listen handler will
1266 	 * remove pending from the pending list upon our failure, the cleanup
1267 	 * function won't drop the additional reference, which is why we do it
1268 	 * here.
1269 	 */
1270 	sock_put(pending);
1271 
1272 	return err;
1273 }
1274 
1275 static int
1276 vmci_transport_recv_connecting_client(struct sock *sk,
1277 				      struct vmci_transport_packet *pkt)
1278 {
1279 	struct vsock_sock *vsk;
1280 	int err;
1281 	int skerr;
1282 
1283 	vsk = vsock_sk(sk);
1284 
1285 	switch (pkt->type) {
1286 	case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1287 		if (vmci_handle_is_invalid(pkt->u.handle) ||
1288 		    !vmci_handle_is_equal(pkt->u.handle,
1289 					  vmci_trans(vsk)->qp_handle)) {
1290 			skerr = EPROTO;
1291 			err = -EINVAL;
1292 			goto destroy;
1293 		}
1294 
1295 		/* Signify the socket is connected and wakeup the waiter in
1296 		 * connect(). Also place the socket in the connected table for
1297 		 * accounting (it can already be found since it's in the bound
1298 		 * table).
1299 		 */
1300 		sk->sk_state = TCP_ESTABLISHED;
1301 		sk->sk_socket->state = SS_CONNECTED;
1302 		vsock_insert_connected(vsk);
1303 		sk->sk_state_change(sk);
1304 
1305 		break;
1306 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1307 	case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1308 		if (pkt->u.size == 0
1309 		    || pkt->dg.src.context != vsk->remote_addr.svm_cid
1310 		    || pkt->src_port != vsk->remote_addr.svm_port
1311 		    || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1312 		    || vmci_trans(vsk)->qpair
1313 		    || vmci_trans(vsk)->produce_size != 0
1314 		    || vmci_trans(vsk)->consume_size != 0
1315 		    || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1316 			skerr = EPROTO;
1317 			err = -EINVAL;
1318 
1319 			goto destroy;
1320 		}
1321 
1322 		err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1323 		if (err) {
1324 			skerr = -err;
1325 			goto destroy;
1326 		}
1327 
1328 		break;
1329 	case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1330 		err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1331 		if (err) {
1332 			skerr = -err;
1333 			goto destroy;
1334 		}
1335 
1336 		break;
1337 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
1338 		/* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1339 		 * continue processing here after they sent an INVALID packet.
1340 		 * This meant that we got a RST after the INVALID. We ignore a
1341 		 * RST after an INVALID. The common code doesn't send the RST
1342 		 * ... so we can hang if an old version of the common code
1343 		 * fails between getting a REQUEST and sending an OFFER back.
1344 		 * Not much we can do about it... except hope that it doesn't
1345 		 * happen.
1346 		 */
1347 		if (vsk->ignore_connecting_rst) {
1348 			vsk->ignore_connecting_rst = false;
1349 		} else {
1350 			skerr = ECONNRESET;
1351 			err = 0;
1352 			goto destroy;
1353 		}
1354 
1355 		break;
1356 	default:
1357 		/* Close and cleanup the connection. */
1358 		skerr = EPROTO;
1359 		err = -EINVAL;
1360 		goto destroy;
1361 	}
1362 
1363 	return 0;
1364 
1365 destroy:
1366 	vmci_transport_send_reset(sk, pkt);
1367 
1368 	sk->sk_state = TCP_CLOSE;
1369 	sk->sk_err = skerr;
1370 	sk->sk_error_report(sk);
1371 	return err;
1372 }
1373 
1374 static int vmci_transport_recv_connecting_client_negotiate(
1375 					struct sock *sk,
1376 					struct vmci_transport_packet *pkt)
1377 {
1378 	int err;
1379 	struct vsock_sock *vsk;
1380 	struct vmci_handle handle;
1381 	struct vmci_qp *qpair;
1382 	u32 detach_sub_id;
1383 	bool is_local;
1384 	u32 flags;
1385 	bool old_proto = true;
1386 	bool old_pkt_proto;
1387 	u16 version;
1388 
1389 	vsk = vsock_sk(sk);
1390 	handle = VMCI_INVALID_HANDLE;
1391 	detach_sub_id = VMCI_INVALID_ID;
1392 
1393 	/* If we have gotten here then we should be past the point where old
1394 	 * linux vsock could have sent the bogus rst.
1395 	 */
1396 	vsk->sent_request = false;
1397 	vsk->ignore_connecting_rst = false;
1398 
1399 	/* Verify that we're OK with the proposed queue pair size */
1400 	if (pkt->u.size < vsk->buffer_min_size ||
1401 	    pkt->u.size > vsk->buffer_max_size) {
1402 		err = -EINVAL;
1403 		goto destroy;
1404 	}
1405 
1406 	/* At this point we know the CID the peer is using to talk to us. */
1407 
1408 	if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1409 		vsk->local_addr.svm_cid = pkt->dg.dst.context;
1410 
1411 	/* Setup the notify ops to be the highest supported version that both
1412 	 * the server and the client support.
1413 	 */
1414 
1415 	if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1416 		old_proto = old_pkt_proto;
1417 	} else {
1418 		if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1419 			old_proto = true;
1420 		else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1421 			old_proto = false;
1422 
1423 	}
1424 
1425 	if (old_proto)
1426 		version = VSOCK_PROTO_INVALID;
1427 	else
1428 		version = pkt->proto;
1429 
1430 	if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1431 		err = -EINVAL;
1432 		goto destroy;
1433 	}
1434 
1435 	/* Subscribe to detach events first.
1436 	 *
1437 	 * XXX We attach once for each queue pair created for now so it is easy
1438 	 * to find the socket (it's provided), but later we should only
1439 	 * subscribe once and add a way to lookup sockets by queue pair handle.
1440 	 */
1441 	err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1442 				   vmci_transport_peer_detach_cb,
1443 				   vmci_trans(vsk), &detach_sub_id);
1444 	if (err < VMCI_SUCCESS) {
1445 		err = vmci_transport_error_to_vsock_error(err);
1446 		goto destroy;
1447 	}
1448 
1449 	/* Make VMCI select the handle for us. */
1450 	handle = VMCI_INVALID_HANDLE;
1451 	is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1452 	flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1453 
1454 	err = vmci_transport_queue_pair_alloc(&qpair,
1455 					      &handle,
1456 					      pkt->u.size,
1457 					      pkt->u.size,
1458 					      vsk->remote_addr.svm_cid,
1459 					      flags,
1460 					      vmci_transport_is_trusted(
1461 						  vsk,
1462 						  vsk->
1463 						  remote_addr.svm_cid));
1464 	if (err < 0)
1465 		goto destroy;
1466 
1467 	err = vmci_transport_send_qp_offer(sk, handle);
1468 	if (err < 0) {
1469 		err = vmci_transport_error_to_vsock_error(err);
1470 		goto destroy;
1471 	}
1472 
1473 	vmci_trans(vsk)->qp_handle = handle;
1474 	vmci_trans(vsk)->qpair = qpair;
1475 
1476 	vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1477 		pkt->u.size;
1478 
1479 	vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1480 
1481 	vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1482 
1483 	return 0;
1484 
1485 destroy:
1486 	if (detach_sub_id != VMCI_INVALID_ID)
1487 		vmci_event_unsubscribe(detach_sub_id);
1488 
1489 	if (!vmci_handle_is_invalid(handle))
1490 		vmci_qpair_detach(&qpair);
1491 
1492 	return err;
1493 }
1494 
1495 static int
1496 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1497 					      struct vmci_transport_packet *pkt)
1498 {
1499 	int err = 0;
1500 	struct vsock_sock *vsk = vsock_sk(sk);
1501 
1502 	if (vsk->sent_request) {
1503 		vsk->sent_request = false;
1504 		vsk->ignore_connecting_rst = true;
1505 
1506 		err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1507 		if (err < 0)
1508 			err = vmci_transport_error_to_vsock_error(err);
1509 		else
1510 			err = 0;
1511 
1512 	}
1513 
1514 	return err;
1515 }
1516 
1517 static int vmci_transport_recv_connected(struct sock *sk,
1518 					 struct vmci_transport_packet *pkt)
1519 {
1520 	struct vsock_sock *vsk;
1521 	bool pkt_processed = false;
1522 
1523 	/* In cases where we are closing the connection, it's sufficient to
1524 	 * mark the state change (and maybe error) and wake up any waiting
1525 	 * threads. Since this is a connected socket, it's owned by a user
1526 	 * process and will be cleaned up when the failure is passed back on
1527 	 * the current or next system call.  Our system call implementations
1528 	 * must therefore check for error and state changes on entry and when
1529 	 * being awoken.
1530 	 */
1531 	switch (pkt->type) {
1532 	case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1533 		if (pkt->u.mode) {
1534 			vsk = vsock_sk(sk);
1535 
1536 			vsk->peer_shutdown |= pkt->u.mode;
1537 			sk->sk_state_change(sk);
1538 		}
1539 		break;
1540 
1541 	case VMCI_TRANSPORT_PACKET_TYPE_RST:
1542 		vsk = vsock_sk(sk);
1543 		/* It is possible that we sent our peer a message (e.g a
1544 		 * WAITING_READ) right before we got notified that the peer had
1545 		 * detached. If that happens then we can get a RST pkt back
1546 		 * from our peer even though there is data available for us to
1547 		 * read. In that case, don't shutdown the socket completely but
1548 		 * instead allow the local client to finish reading data off
1549 		 * the queuepair. Always treat a RST pkt in connected mode like
1550 		 * a clean shutdown.
1551 		 */
1552 		sock_set_flag(sk, SOCK_DONE);
1553 		vsk->peer_shutdown = SHUTDOWN_MASK;
1554 		if (vsock_stream_has_data(vsk) <= 0)
1555 			sk->sk_state = TCP_CLOSING;
1556 
1557 		sk->sk_state_change(sk);
1558 		break;
1559 
1560 	default:
1561 		vsk = vsock_sk(sk);
1562 		vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1563 				sk, pkt, false, NULL, NULL,
1564 				&pkt_processed);
1565 		if (!pkt_processed)
1566 			return -EINVAL;
1567 
1568 		break;
1569 	}
1570 
1571 	return 0;
1572 }
1573 
1574 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1575 				      struct vsock_sock *psk)
1576 {
1577 	vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1578 	if (!vsk->trans)
1579 		return -ENOMEM;
1580 
1581 	vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1582 	vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1583 	vmci_trans(vsk)->qpair = NULL;
1584 	vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1585 	vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1586 	vmci_trans(vsk)->notify_ops = NULL;
1587 	INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1588 	vmci_trans(vsk)->sk = &vsk->sk;
1589 	spin_lock_init(&vmci_trans(vsk)->lock);
1590 
1591 	return 0;
1592 }
1593 
1594 static void vmci_transport_free_resources(struct list_head *transport_list)
1595 {
1596 	while (!list_empty(transport_list)) {
1597 		struct vmci_transport *transport =
1598 		    list_first_entry(transport_list, struct vmci_transport,
1599 				     elem);
1600 		list_del(&transport->elem);
1601 
1602 		if (transport->detach_sub_id != VMCI_INVALID_ID) {
1603 			vmci_event_unsubscribe(transport->detach_sub_id);
1604 			transport->detach_sub_id = VMCI_INVALID_ID;
1605 		}
1606 
1607 		if (!vmci_handle_is_invalid(transport->qp_handle)) {
1608 			vmci_qpair_detach(&transport->qpair);
1609 			transport->qp_handle = VMCI_INVALID_HANDLE;
1610 			transport->produce_size = 0;
1611 			transport->consume_size = 0;
1612 		}
1613 
1614 		kfree(transport);
1615 	}
1616 }
1617 
1618 static void vmci_transport_cleanup(struct work_struct *work)
1619 {
1620 	LIST_HEAD(pending);
1621 
1622 	spin_lock_bh(&vmci_transport_cleanup_lock);
1623 	list_replace_init(&vmci_transport_cleanup_list, &pending);
1624 	spin_unlock_bh(&vmci_transport_cleanup_lock);
1625 	vmci_transport_free_resources(&pending);
1626 }
1627 
1628 static void vmci_transport_destruct(struct vsock_sock *vsk)
1629 {
1630 	/* transport can be NULL if we hit a failure at init() time */
1631 	if (!vmci_trans(vsk))
1632 		return;
1633 
1634 	/* Ensure that the detach callback doesn't use the sk/vsk
1635 	 * we are about to destruct.
1636 	 */
1637 	spin_lock_bh(&vmci_trans(vsk)->lock);
1638 	vmci_trans(vsk)->sk = NULL;
1639 	spin_unlock_bh(&vmci_trans(vsk)->lock);
1640 
1641 	if (vmci_trans(vsk)->notify_ops)
1642 		vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1643 
1644 	spin_lock_bh(&vmci_transport_cleanup_lock);
1645 	list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1646 	spin_unlock_bh(&vmci_transport_cleanup_lock);
1647 	schedule_work(&vmci_transport_cleanup_work);
1648 
1649 	vsk->trans = NULL;
1650 }
1651 
1652 static void vmci_transport_release(struct vsock_sock *vsk)
1653 {
1654 	vsock_remove_sock(vsk);
1655 
1656 	if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1657 		vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1658 		vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1659 	}
1660 }
1661 
1662 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1663 				     struct sockaddr_vm *addr)
1664 {
1665 	u32 port;
1666 	u32 flags;
1667 	int err;
1668 
1669 	/* VMCI will select a resource ID for us if we provide
1670 	 * VMCI_INVALID_ID.
1671 	 */
1672 	port = addr->svm_port == VMADDR_PORT_ANY ?
1673 			VMCI_INVALID_ID : addr->svm_port;
1674 
1675 	if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1676 		return -EACCES;
1677 
1678 	flags = addr->svm_cid == VMADDR_CID_ANY ?
1679 				VMCI_FLAG_ANYCID_DG_HND : 0;
1680 
1681 	err = vmci_transport_datagram_create_hnd(port, flags,
1682 						 vmci_transport_recv_dgram_cb,
1683 						 &vsk->sk,
1684 						 &vmci_trans(vsk)->dg_handle);
1685 	if (err < VMCI_SUCCESS)
1686 		return vmci_transport_error_to_vsock_error(err);
1687 	vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1688 			vmci_trans(vsk)->dg_handle.resource);
1689 
1690 	return 0;
1691 }
1692 
1693 static int vmci_transport_dgram_enqueue(
1694 	struct vsock_sock *vsk,
1695 	struct sockaddr_vm *remote_addr,
1696 	struct msghdr *msg,
1697 	size_t len)
1698 {
1699 	int err;
1700 	struct vmci_datagram *dg;
1701 
1702 	if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1703 		return -EMSGSIZE;
1704 
1705 	if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1706 		return -EPERM;
1707 
1708 	/* Allocate a buffer for the user's message and our packet header. */
1709 	dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1710 	if (!dg)
1711 		return -ENOMEM;
1712 
1713 	memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1714 
1715 	dg->dst = vmci_make_handle(remote_addr->svm_cid,
1716 				   remote_addr->svm_port);
1717 	dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1718 				   vsk->local_addr.svm_port);
1719 	dg->payload_size = len;
1720 
1721 	err = vmci_datagram_send(dg);
1722 	kfree(dg);
1723 	if (err < 0)
1724 		return vmci_transport_error_to_vsock_error(err);
1725 
1726 	return err - sizeof(*dg);
1727 }
1728 
1729 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1730 					struct msghdr *msg, size_t len,
1731 					int flags)
1732 {
1733 	int err;
1734 	int noblock;
1735 	struct vmci_datagram *dg;
1736 	size_t payload_len;
1737 	struct sk_buff *skb;
1738 
1739 	noblock = flags & MSG_DONTWAIT;
1740 
1741 	if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1742 		return -EOPNOTSUPP;
1743 
1744 	/* Retrieve the head sk_buff from the socket's receive queue. */
1745 	err = 0;
1746 	skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1747 	if (!skb)
1748 		return err;
1749 
1750 	dg = (struct vmci_datagram *)skb->data;
1751 	if (!dg)
1752 		/* err is 0, meaning we read zero bytes. */
1753 		goto out;
1754 
1755 	payload_len = dg->payload_size;
1756 	/* Ensure the sk_buff matches the payload size claimed in the packet. */
1757 	if (payload_len != skb->len - sizeof(*dg)) {
1758 		err = -EINVAL;
1759 		goto out;
1760 	}
1761 
1762 	if (payload_len > len) {
1763 		payload_len = len;
1764 		msg->msg_flags |= MSG_TRUNC;
1765 	}
1766 
1767 	/* Place the datagram payload in the user's iovec. */
1768 	err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1769 	if (err)
1770 		goto out;
1771 
1772 	if (msg->msg_name) {
1773 		/* Provide the address of the sender. */
1774 		DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1775 		vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1776 		msg->msg_namelen = sizeof(*vm_addr);
1777 	}
1778 	err = payload_len;
1779 
1780 out:
1781 	skb_free_datagram(&vsk->sk, skb);
1782 	return err;
1783 }
1784 
1785 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1786 {
1787 	if (cid == VMADDR_CID_HYPERVISOR) {
1788 		/* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1789 		 * state and are allowed.
1790 		 */
1791 		return port == VMCI_UNITY_PBRPC_REGISTER;
1792 	}
1793 
1794 	return true;
1795 }
1796 
1797 static int vmci_transport_connect(struct vsock_sock *vsk)
1798 {
1799 	int err;
1800 	bool old_pkt_proto = false;
1801 	struct sock *sk = &vsk->sk;
1802 
1803 	if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1804 		old_pkt_proto) {
1805 		err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1806 		if (err < 0) {
1807 			sk->sk_state = TCP_CLOSE;
1808 			return err;
1809 		}
1810 	} else {
1811 		int supported_proto_versions =
1812 			vmci_transport_new_proto_supported_versions();
1813 		err = vmci_transport_send_conn_request2(sk, vsk->buffer_size,
1814 				supported_proto_versions);
1815 		if (err < 0) {
1816 			sk->sk_state = TCP_CLOSE;
1817 			return err;
1818 		}
1819 
1820 		vsk->sent_request = true;
1821 	}
1822 
1823 	return err;
1824 }
1825 
1826 static ssize_t vmci_transport_stream_dequeue(
1827 	struct vsock_sock *vsk,
1828 	struct msghdr *msg,
1829 	size_t len,
1830 	int flags)
1831 {
1832 	if (flags & MSG_PEEK)
1833 		return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1834 	else
1835 		return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1836 }
1837 
1838 static ssize_t vmci_transport_stream_enqueue(
1839 	struct vsock_sock *vsk,
1840 	struct msghdr *msg,
1841 	size_t len)
1842 {
1843 	return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1844 }
1845 
1846 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1847 {
1848 	return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1849 }
1850 
1851 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1852 {
1853 	return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1854 }
1855 
1856 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1857 {
1858 	return vmci_trans(vsk)->consume_size;
1859 }
1860 
1861 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1862 {
1863 	return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1864 }
1865 
1866 static int vmci_transport_notify_poll_in(
1867 	struct vsock_sock *vsk,
1868 	size_t target,
1869 	bool *data_ready_now)
1870 {
1871 	return vmci_trans(vsk)->notify_ops->poll_in(
1872 			&vsk->sk, target, data_ready_now);
1873 }
1874 
1875 static int vmci_transport_notify_poll_out(
1876 	struct vsock_sock *vsk,
1877 	size_t target,
1878 	bool *space_available_now)
1879 {
1880 	return vmci_trans(vsk)->notify_ops->poll_out(
1881 			&vsk->sk, target, space_available_now);
1882 }
1883 
1884 static int vmci_transport_notify_recv_init(
1885 	struct vsock_sock *vsk,
1886 	size_t target,
1887 	struct vsock_transport_recv_notify_data *data)
1888 {
1889 	return vmci_trans(vsk)->notify_ops->recv_init(
1890 			&vsk->sk, target,
1891 			(struct vmci_transport_recv_notify_data *)data);
1892 }
1893 
1894 static int vmci_transport_notify_recv_pre_block(
1895 	struct vsock_sock *vsk,
1896 	size_t target,
1897 	struct vsock_transport_recv_notify_data *data)
1898 {
1899 	return vmci_trans(vsk)->notify_ops->recv_pre_block(
1900 			&vsk->sk, target,
1901 			(struct vmci_transport_recv_notify_data *)data);
1902 }
1903 
1904 static int vmci_transport_notify_recv_pre_dequeue(
1905 	struct vsock_sock *vsk,
1906 	size_t target,
1907 	struct vsock_transport_recv_notify_data *data)
1908 {
1909 	return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1910 			&vsk->sk, target,
1911 			(struct vmci_transport_recv_notify_data *)data);
1912 }
1913 
1914 static int vmci_transport_notify_recv_post_dequeue(
1915 	struct vsock_sock *vsk,
1916 	size_t target,
1917 	ssize_t copied,
1918 	bool data_read,
1919 	struct vsock_transport_recv_notify_data *data)
1920 {
1921 	return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1922 			&vsk->sk, target, copied, data_read,
1923 			(struct vmci_transport_recv_notify_data *)data);
1924 }
1925 
1926 static int vmci_transport_notify_send_init(
1927 	struct vsock_sock *vsk,
1928 	struct vsock_transport_send_notify_data *data)
1929 {
1930 	return vmci_trans(vsk)->notify_ops->send_init(
1931 			&vsk->sk,
1932 			(struct vmci_transport_send_notify_data *)data);
1933 }
1934 
1935 static int vmci_transport_notify_send_pre_block(
1936 	struct vsock_sock *vsk,
1937 	struct vsock_transport_send_notify_data *data)
1938 {
1939 	return vmci_trans(vsk)->notify_ops->send_pre_block(
1940 			&vsk->sk,
1941 			(struct vmci_transport_send_notify_data *)data);
1942 }
1943 
1944 static int vmci_transport_notify_send_pre_enqueue(
1945 	struct vsock_sock *vsk,
1946 	struct vsock_transport_send_notify_data *data)
1947 {
1948 	return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1949 			&vsk->sk,
1950 			(struct vmci_transport_send_notify_data *)data);
1951 }
1952 
1953 static int vmci_transport_notify_send_post_enqueue(
1954 	struct vsock_sock *vsk,
1955 	ssize_t written,
1956 	struct vsock_transport_send_notify_data *data)
1957 {
1958 	return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1959 			&vsk->sk, written,
1960 			(struct vmci_transport_send_notify_data *)data);
1961 }
1962 
1963 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1964 {
1965 	if (PROTOCOL_OVERRIDE != -1) {
1966 		if (PROTOCOL_OVERRIDE == 0)
1967 			*old_pkt_proto = true;
1968 		else
1969 			*old_pkt_proto = false;
1970 
1971 		pr_info("Proto override in use\n");
1972 		return true;
1973 	}
1974 
1975 	return false;
1976 }
1977 
1978 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
1979 						  u16 *proto,
1980 						  bool old_pkt_proto)
1981 {
1982 	struct vsock_sock *vsk = vsock_sk(sk);
1983 
1984 	if (old_pkt_proto) {
1985 		if (*proto != VSOCK_PROTO_INVALID) {
1986 			pr_err("Can't set both an old and new protocol\n");
1987 			return false;
1988 		}
1989 		vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
1990 		goto exit;
1991 	}
1992 
1993 	switch (*proto) {
1994 	case VSOCK_PROTO_PKT_ON_NOTIFY:
1995 		vmci_trans(vsk)->notify_ops =
1996 			&vmci_transport_notify_pkt_q_state_ops;
1997 		break;
1998 	default:
1999 		pr_err("Unknown notify protocol version\n");
2000 		return false;
2001 	}
2002 
2003 exit:
2004 	vmci_trans(vsk)->notify_ops->socket_init(sk);
2005 	return true;
2006 }
2007 
2008 static u16 vmci_transport_new_proto_supported_versions(void)
2009 {
2010 	if (PROTOCOL_OVERRIDE != -1)
2011 		return PROTOCOL_OVERRIDE;
2012 
2013 	return VSOCK_PROTO_ALL_SUPPORTED;
2014 }
2015 
2016 static u32 vmci_transport_get_local_cid(void)
2017 {
2018 	return vmci_get_context_id();
2019 }
2020 
2021 static struct vsock_transport vmci_transport = {
2022 	.module = THIS_MODULE,
2023 	.init = vmci_transport_socket_init,
2024 	.destruct = vmci_transport_destruct,
2025 	.release = vmci_transport_release,
2026 	.connect = vmci_transport_connect,
2027 	.dgram_bind = vmci_transport_dgram_bind,
2028 	.dgram_dequeue = vmci_transport_dgram_dequeue,
2029 	.dgram_enqueue = vmci_transport_dgram_enqueue,
2030 	.dgram_allow = vmci_transport_dgram_allow,
2031 	.stream_dequeue = vmci_transport_stream_dequeue,
2032 	.stream_enqueue = vmci_transport_stream_enqueue,
2033 	.stream_has_data = vmci_transport_stream_has_data,
2034 	.stream_has_space = vmci_transport_stream_has_space,
2035 	.stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2036 	.stream_is_active = vmci_transport_stream_is_active,
2037 	.stream_allow = vmci_transport_stream_allow,
2038 	.notify_poll_in = vmci_transport_notify_poll_in,
2039 	.notify_poll_out = vmci_transport_notify_poll_out,
2040 	.notify_recv_init = vmci_transport_notify_recv_init,
2041 	.notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2042 	.notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2043 	.notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2044 	.notify_send_init = vmci_transport_notify_send_init,
2045 	.notify_send_pre_block = vmci_transport_notify_send_pre_block,
2046 	.notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2047 	.notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2048 	.shutdown = vmci_transport_shutdown,
2049 	.get_local_cid = vmci_transport_get_local_cid,
2050 };
2051 
2052 static bool vmci_check_transport(struct vsock_sock *vsk)
2053 {
2054 	return vsk->transport == &vmci_transport;
2055 }
2056 
2057 static void vmci_vsock_transport_cb(bool is_host)
2058 {
2059 	int features;
2060 
2061 	if (is_host)
2062 		features = VSOCK_TRANSPORT_F_H2G;
2063 	else
2064 		features = VSOCK_TRANSPORT_F_G2H;
2065 
2066 	vsock_core_register(&vmci_transport, features);
2067 }
2068 
2069 static int __init vmci_transport_init(void)
2070 {
2071 	int err;
2072 
2073 	/* Create the datagram handle that we will use to send and receive all
2074 	 * VSocket control messages for this context.
2075 	 */
2076 	err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2077 						 VMCI_FLAG_ANYCID_DG_HND,
2078 						 vmci_transport_recv_stream_cb,
2079 						 NULL,
2080 						 &vmci_transport_stream_handle);
2081 	if (err < VMCI_SUCCESS) {
2082 		pr_err("Unable to create datagram handle. (%d)\n", err);
2083 		return vmci_transport_error_to_vsock_error(err);
2084 	}
2085 	err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2086 				   vmci_transport_qp_resumed_cb,
2087 				   NULL, &vmci_transport_qp_resumed_sub_id);
2088 	if (err < VMCI_SUCCESS) {
2089 		pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2090 		err = vmci_transport_error_to_vsock_error(err);
2091 		vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2092 		goto err_destroy_stream_handle;
2093 	}
2094 
2095 	/* Register only with dgram feature, other features (H2G, G2H) will be
2096 	 * registered when the first host or guest becomes active.
2097 	 */
2098 	err = vsock_core_register(&vmci_transport, VSOCK_TRANSPORT_F_DGRAM);
2099 	if (err < 0)
2100 		goto err_unsubscribe;
2101 
2102 	err = vmci_register_vsock_callback(vmci_vsock_transport_cb);
2103 	if (err < 0)
2104 		goto err_unregister;
2105 
2106 	return 0;
2107 
2108 err_unregister:
2109 	vsock_core_unregister(&vmci_transport);
2110 err_unsubscribe:
2111 	vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2112 err_destroy_stream_handle:
2113 	vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2114 	return err;
2115 }
2116 module_init(vmci_transport_init);
2117 
2118 static void __exit vmci_transport_exit(void)
2119 {
2120 	cancel_work_sync(&vmci_transport_cleanup_work);
2121 	vmci_transport_free_resources(&vmci_transport_cleanup_list);
2122 
2123 	if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2124 		if (vmci_datagram_destroy_handle(
2125 			vmci_transport_stream_handle) != VMCI_SUCCESS)
2126 			pr_err("Couldn't destroy datagram handle\n");
2127 		vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2128 	}
2129 
2130 	if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2131 		vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2132 		vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2133 	}
2134 
2135 	vmci_register_vsock_callback(NULL);
2136 	vsock_core_unregister(&vmci_transport);
2137 }
2138 module_exit(vmci_transport_exit);
2139 
2140 MODULE_AUTHOR("VMware, Inc.");
2141 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2142 MODULE_VERSION("1.0.5.0-k");
2143 MODULE_LICENSE("GPL v2");
2144 MODULE_ALIAS("vmware_vsock");
2145 MODULE_ALIAS_NETPROTO(PF_VSOCK);
2146