blob: 56214736fe888090f569739e2e008c72d8785fb0 [file] [log] [blame]
Andy Kingd021c342013-02-06 14:23:56 +00001/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#include <linux/types.h>
Andy Kingd021c342013-02-06 14:23:56 +000017#include <linux/bitops.h>
18#include <linux/cred.h>
19#include <linux/init.h>
20#include <linux/io.h>
21#include <linux/kernel.h>
22#include <linux/kmod.h>
23#include <linux/list.h>
24#include <linux/miscdevice.h>
25#include <linux/module.h>
26#include <linux/mutex.h>
27#include <linux/net.h>
28#include <linux/poll.h>
29#include <linux/skbuff.h>
30#include <linux/smp.h>
31#include <linux/socket.h>
32#include <linux/stddef.h>
33#include <linux/unistd.h>
34#include <linux/wait.h>
35#include <linux/workqueue.h>
36#include <net/sock.h>
Asias He82a54d02013-07-25 17:39:34 +080037#include <net/af_vsock.h>
Andy Kingd021c342013-02-06 14:23:56 +000038
Andy Kingd021c342013-02-06 14:23:56 +000039#include "vmci_transport_notify.h"
40
41static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
42static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
Andy Kingd021c342013-02-06 14:23:56 +000043static void vmci_transport_peer_detach_cb(u32 sub_id,
44 const struct vmci_event_data *ed,
45 void *client_data);
46static void vmci_transport_recv_pkt_work(struct work_struct *work);
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -070047static void vmci_transport_cleanup(struct work_struct *work);
Andy Kingd021c342013-02-06 14:23:56 +000048static int vmci_transport_recv_listen(struct sock *sk,
49 struct vmci_transport_packet *pkt);
50static int vmci_transport_recv_connecting_server(
51 struct sock *sk,
52 struct sock *pending,
53 struct vmci_transport_packet *pkt);
54static int vmci_transport_recv_connecting_client(
55 struct sock *sk,
56 struct vmci_transport_packet *pkt);
57static int vmci_transport_recv_connecting_client_negotiate(
58 struct sock *sk,
59 struct vmci_transport_packet *pkt);
60static int vmci_transport_recv_connecting_client_invalid(
61 struct sock *sk,
62 struct vmci_transport_packet *pkt);
63static int vmci_transport_recv_connected(struct sock *sk,
64 struct vmci_transport_packet *pkt);
65static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
66static u16 vmci_transport_new_proto_supported_versions(void);
67static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
68 bool old_pkt_proto);
69
70struct vmci_transport_recv_pkt_info {
71 struct work_struct work;
72 struct sock *sk;
73 struct vmci_transport_packet pkt;
74};
75
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -070076static LIST_HEAD(vmci_transport_cleanup_list);
77static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
78static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
79
Andy Kingd021c342013-02-06 14:23:56 +000080static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
81 VMCI_INVALID_ID };
82static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
83
84static int PROTOCOL_OVERRIDE = -1;
85
86#define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN 128
87#define VMCI_TRANSPORT_DEFAULT_QP_SIZE 262144
88#define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX 262144
89
90/* The default peer timeout indicates how long we will wait for a peer response
91 * to a control message.
92 */
93#define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
94
Andy Kingd021c342013-02-06 14:23:56 +000095/* Helper function to convert from a VMCI error code to a VSock error code. */
96
97static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
98{
99 int err;
100
101 switch (vmci_error) {
102 case VMCI_ERROR_NO_MEM:
103 err = ENOMEM;
104 break;
105 case VMCI_ERROR_DUPLICATE_ENTRY:
106 case VMCI_ERROR_ALREADY_EXISTS:
107 err = EADDRINUSE;
108 break;
109 case VMCI_ERROR_NO_ACCESS:
110 err = EPERM;
111 break;
112 case VMCI_ERROR_NO_RESOURCES:
113 err = ENOBUFS;
114 break;
115 case VMCI_ERROR_INVALID_RESOURCE:
116 err = EHOSTUNREACH;
117 break;
118 case VMCI_ERROR_INVALID_ARGS:
119 default:
120 err = EINVAL;
121 }
122
123 return err > 0 ? -err : err;
124}
125
Reilly Grant2a89f922013-03-14 11:55:41 +0000126static u32 vmci_transport_peer_rid(u32 peer_cid)
127{
128 if (VMADDR_CID_HYPERVISOR == peer_cid)
129 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
130
131 return VMCI_TRANSPORT_PACKET_RID;
132}
133
Andy Kingd021c342013-02-06 14:23:56 +0000134static inline void
135vmci_transport_packet_init(struct vmci_transport_packet *pkt,
136 struct sockaddr_vm *src,
137 struct sockaddr_vm *dst,
138 u8 type,
139 u64 size,
140 u64 mode,
141 struct vmci_transport_waiting_info *wait,
142 u16 proto,
143 struct vmci_handle handle)
144{
145 /* We register the stream control handler as an any cid handle so we
146 * must always send from a source address of VMADDR_CID_ANY
147 */
148 pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
149 VMCI_TRANSPORT_PACKET_RID);
150 pkt->dg.dst = vmci_make_handle(dst->svm_cid,
Reilly Grant2a89f922013-03-14 11:55:41 +0000151 vmci_transport_peer_rid(dst->svm_cid));
Andy Kingd021c342013-02-06 14:23:56 +0000152 pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
153 pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
154 pkt->type = type;
155 pkt->src_port = src->svm_port;
156 pkt->dst_port = dst->svm_port;
157 memset(&pkt->proto, 0, sizeof(pkt->proto));
158 memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
159
160 switch (pkt->type) {
161 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
162 pkt->u.size = 0;
163 break;
164
165 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
166 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
167 pkt->u.size = size;
168 break;
169
170 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
171 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
172 pkt->u.handle = handle;
173 break;
174
175 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
176 case VMCI_TRANSPORT_PACKET_TYPE_READ:
177 case VMCI_TRANSPORT_PACKET_TYPE_RST:
178 pkt->u.size = 0;
179 break;
180
181 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
182 pkt->u.mode = mode;
183 break;
184
185 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
186 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
187 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
188 break;
189
190 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
191 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
192 pkt->u.size = size;
193 pkt->proto = proto;
194 break;
195 }
196}
197
198static inline void
199vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
200 struct sockaddr_vm *local,
201 struct sockaddr_vm *remote)
202{
203 vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
204 vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
205}
206
207static int
208__vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
209 struct sockaddr_vm *src,
210 struct sockaddr_vm *dst,
211 enum vmci_transport_packet_type type,
212 u64 size,
213 u64 mode,
214 struct vmci_transport_waiting_info *wait,
215 u16 proto,
216 struct vmci_handle handle,
217 bool convert_error)
218{
219 int err;
220
221 vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
222 proto, handle);
223 err = vmci_datagram_send(&pkt->dg);
224 if (convert_error && (err < 0))
225 return vmci_transport_error_to_vsock_error(err);
226
227 return err;
228}
229
230static int
231vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
232 enum vmci_transport_packet_type type,
233 u64 size,
234 u64 mode,
235 struct vmci_transport_waiting_info *wait,
236 struct vmci_handle handle)
237{
238 struct vmci_transport_packet reply;
239 struct sockaddr_vm src, dst;
240
241 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
242 return 0;
243 } else {
244 vmci_transport_packet_get_addresses(pkt, &src, &dst);
245 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
246 type,
247 size, mode, wait,
248 VSOCK_PROTO_INVALID,
249 handle, true);
250 }
251}
252
253static int
254vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
255 struct sockaddr_vm *dst,
256 enum vmci_transport_packet_type type,
257 u64 size,
258 u64 mode,
259 struct vmci_transport_waiting_info *wait,
260 struct vmci_handle handle)
261{
262 /* Note that it is safe to use a single packet across all CPUs since
263 * two tasklets of the same type are guaranteed to not ever run
264 * simultaneously. If that ever changes, or VMCI stops using tasklets,
265 * we can use per-cpu packets.
266 */
267 static struct vmci_transport_packet pkt;
268
269 return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
270 size, mode, wait,
271 VSOCK_PROTO_INVALID, handle,
272 false);
273}
274
275static int
276vmci_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 vmci_transport_packet *pkt;
285 struct vsock_sock *vsk;
286 int err;
287
288 vsk = vsock_sk(sk);
289
290 if (!vsock_addr_bound(&vsk->local_addr))
291 return -EINVAL;
292
293 if (!vsock_addr_bound(&vsk->remote_addr))
294 return -EINVAL;
295
296 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
297 if (!pkt)
298 return -ENOMEM;
299
300 err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
301 &vsk->remote_addr, type, size,
302 mode, wait, proto, handle,
303 true);
304 kfree(pkt);
305
306 return err;
307}
308
309static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
310 struct sockaddr_vm *src,
311 struct vmci_transport_packet *pkt)
312{
313 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
314 return 0;
315 return vmci_transport_send_control_pkt_bh(
316 dst, src,
317 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
318 0, NULL, VMCI_INVALID_HANDLE);
319}
320
321static int vmci_transport_send_reset(struct sock *sk,
322 struct vmci_transport_packet *pkt)
323{
324 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
325 return 0;
326 return vmci_transport_send_control_pkt(sk,
327 VMCI_TRANSPORT_PACKET_TYPE_RST,
328 0, 0, NULL, VSOCK_PROTO_INVALID,
329 VMCI_INVALID_HANDLE);
330}
331
332static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
333{
334 return vmci_transport_send_control_pkt(
335 sk,
336 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
337 size, 0, NULL,
338 VSOCK_PROTO_INVALID,
339 VMCI_INVALID_HANDLE);
340}
341
342static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
343 u16 version)
344{
345 return vmci_transport_send_control_pkt(
346 sk,
347 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
348 size, 0, NULL, version,
349 VMCI_INVALID_HANDLE);
350}
351
352static int vmci_transport_send_qp_offer(struct sock *sk,
353 struct vmci_handle handle)
354{
355 return vmci_transport_send_control_pkt(
356 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
357 0, NULL,
358 VSOCK_PROTO_INVALID, handle);
359}
360
361static int vmci_transport_send_attach(struct sock *sk,
362 struct vmci_handle handle)
363{
364 return vmci_transport_send_control_pkt(
365 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
366 0, 0, NULL, VSOCK_PROTO_INVALID,
367 handle);
368}
369
370static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
371{
372 return vmci_transport_reply_control_pkt_fast(
373 pkt,
374 VMCI_TRANSPORT_PACKET_TYPE_RST,
375 0, 0, NULL,
376 VMCI_INVALID_HANDLE);
377}
378
379static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
380 struct sockaddr_vm *src)
381{
382 return vmci_transport_send_control_pkt_bh(
383 dst, src,
384 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
385 0, 0, NULL, VMCI_INVALID_HANDLE);
386}
387
388int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
389 struct sockaddr_vm *src)
390{
391 return vmci_transport_send_control_pkt_bh(
392 dst, src,
393 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
394 0, NULL, VMCI_INVALID_HANDLE);
395}
396
397int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
398 struct sockaddr_vm *src)
399{
400 return vmci_transport_send_control_pkt_bh(
401 dst, src,
402 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
403 0, NULL, VMCI_INVALID_HANDLE);
404}
405
406int vmci_transport_send_wrote(struct sock *sk)
407{
408 return vmci_transport_send_control_pkt(
409 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
410 0, NULL, VSOCK_PROTO_INVALID,
411 VMCI_INVALID_HANDLE);
412}
413
414int vmci_transport_send_read(struct sock *sk)
415{
416 return vmci_transport_send_control_pkt(
417 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
418 0, NULL, VSOCK_PROTO_INVALID,
419 VMCI_INVALID_HANDLE);
420}
421
422int vmci_transport_send_waiting_write(struct sock *sk,
423 struct vmci_transport_waiting_info *wait)
424{
425 return vmci_transport_send_control_pkt(
426 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
427 0, 0, wait, VSOCK_PROTO_INVALID,
428 VMCI_INVALID_HANDLE);
429}
430
431int vmci_transport_send_waiting_read(struct sock *sk,
432 struct vmci_transport_waiting_info *wait)
433{
434 return vmci_transport_send_control_pkt(
435 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
436 0, 0, wait, VSOCK_PROTO_INVALID,
437 VMCI_INVALID_HANDLE);
438}
439
440static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
441{
442 return vmci_transport_send_control_pkt(
443 &vsk->sk,
444 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
445 0, mode, NULL,
446 VSOCK_PROTO_INVALID,
447 VMCI_INVALID_HANDLE);
448}
449
450static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
451{
452 return vmci_transport_send_control_pkt(sk,
453 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
454 size, 0, NULL,
455 VSOCK_PROTO_INVALID,
456 VMCI_INVALID_HANDLE);
457}
458
459static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
460 u16 version)
461{
462 return vmci_transport_send_control_pkt(
463 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
464 size, 0, NULL, version,
465 VMCI_INVALID_HANDLE);
466}
467
468static struct sock *vmci_transport_get_pending(
469 struct sock *listener,
470 struct vmci_transport_packet *pkt)
471{
472 struct vsock_sock *vlistener;
473 struct vsock_sock *vpending;
474 struct sock *pending;
Reilly Grant990454b2013-04-01 11:41:52 -0700475 struct sockaddr_vm src;
476
477 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
Andy Kingd021c342013-02-06 14:23:56 +0000478
479 vlistener = vsock_sk(listener);
480
481 list_for_each_entry(vpending, &vlistener->pending_links,
482 pending_links) {
Andy Kingd021c342013-02-06 14:23:56 +0000483 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
Reilly Grant990454b2013-04-01 11:41:52 -0700484 pkt->dst_port == vpending->local_addr.svm_port) {
Andy Kingd021c342013-02-06 14:23:56 +0000485 pending = sk_vsock(vpending);
486 sock_hold(pending);
487 goto found;
488 }
489 }
490
491 pending = NULL;
492found:
493 return pending;
494
495}
496
497static void vmci_transport_release_pending(struct sock *pending)
498{
499 sock_put(pending);
500}
501
502/* We allow two kinds of sockets to communicate with a restricted VM: 1)
503 * trusted sockets 2) sockets from applications running as the same user as the
504 * VM (this is only true for the host side and only when using hosted products)
505 */
506
507static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
508{
509 return vsock->trusted ||
510 vmci_is_context_owner(peer_cid, vsock->owner->uid);
511}
512
513/* We allow sending datagrams to and receiving datagrams from a restricted VM
514 * only if it is trusted as described in vmci_transport_is_trusted.
515 */
516
517static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
518{
Reilly Grant2a89f922013-03-14 11:55:41 +0000519 if (VMADDR_CID_HYPERVISOR == peer_cid)
520 return true;
521
Andy Kingd021c342013-02-06 14:23:56 +0000522 if (vsock->cached_peer != peer_cid) {
523 vsock->cached_peer = peer_cid;
524 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
525 (vmci_context_get_priv_flags(peer_cid) &
526 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
527 vsock->cached_peer_allow_dgram = false;
528 } else {
529 vsock->cached_peer_allow_dgram = true;
530 }
531 }
532
533 return vsock->cached_peer_allow_dgram;
534}
535
536static int
537vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
538 struct vmci_handle *handle,
539 u64 produce_size,
540 u64 consume_size,
541 u32 peer, u32 flags, bool trusted)
542{
543 int err = 0;
544
545 if (trusted) {
546 /* Try to allocate our queue pair as trusted. This will only
547 * work if vsock is running in the host.
548 */
549
550 err = vmci_qpair_alloc(qpair, handle, produce_size,
551 consume_size,
552 peer, flags,
553 VMCI_PRIVILEGE_FLAG_TRUSTED);
554 if (err != VMCI_ERROR_NO_ACCESS)
555 goto out;
556
557 }
558
559 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
560 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
561out:
562 if (err < 0) {
563 pr_err("Could not attach to queue pair with %d\n",
564 err);
565 err = vmci_transport_error_to_vsock_error(err);
566 }
567
568 return err;
569}
570
571static int
572vmci_transport_datagram_create_hnd(u32 resource_id,
573 u32 flags,
574 vmci_datagram_recv_cb recv_cb,
575 void *client_data,
576 struct vmci_handle *out_handle)
577{
578 int err = 0;
579
580 /* Try to allocate our datagram handler as trusted. This will only work
581 * if vsock is running in the host.
582 */
583
584 err = vmci_datagram_create_handle_priv(resource_id, flags,
585 VMCI_PRIVILEGE_FLAG_TRUSTED,
586 recv_cb,
587 client_data, out_handle);
588
589 if (err == VMCI_ERROR_NO_ACCESS)
590 err = vmci_datagram_create_handle(resource_id, flags,
591 recv_cb, client_data,
592 out_handle);
593
594 return err;
595}
596
597/* This is invoked as part of a tasklet that's scheduled when the VMCI
598 * interrupt fires. This is run in bottom-half context and if it ever needs to
599 * sleep it should defer that work to a work queue.
600 */
601
602static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
603{
604 struct sock *sk;
605 size_t size;
606 struct sk_buff *skb;
607 struct vsock_sock *vsk;
608
609 sk = (struct sock *)data;
610
611 /* This handler is privileged when this module is running on the host.
612 * We will get datagrams from all endpoints (even VMs that are in a
613 * restricted context). If we get one from a restricted context then
614 * the destination socket must be trusted.
615 *
616 * NOTE: We access the socket struct without holding the lock here.
617 * This is ok because the field we are interested is never modified
618 * outside of the create and destruct socket functions.
619 */
620 vsk = vsock_sk(sk);
621 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
622 return VMCI_ERROR_NO_ACCESS;
623
624 size = VMCI_DG_SIZE(dg);
625
626 /* Attach the packet to the socket's receive queue as an sk_buff. */
627 skb = alloc_skb(size, GFP_ATOMIC);
Asias Hedce1a282013-06-20 17:20:31 +0800628 if (!skb)
629 return VMCI_ERROR_NO_MEM;
630
631 /* sk_receive_skb() will do a sock_put(), so hold here. */
632 sock_hold(sk);
633 skb_put(skb, size);
634 memcpy(skb->data, dg, size);
635 sk_receive_skb(sk, skb, 0);
Andy Kingd021c342013-02-06 14:23:56 +0000636
637 return VMCI_SUCCESS;
638}
639
640static bool vmci_transport_stream_allow(u32 cid, u32 port)
641{
642 static const u32 non_socket_contexts[] = {
Andy Kingd021c342013-02-06 14:23:56 +0000643 VMADDR_CID_RESERVED,
644 };
645 int i;
646
647 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
648
649 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
650 if (cid == non_socket_contexts[i])
651 return false;
652 }
653
654 return true;
655}
656
657/* This is invoked as part of a tasklet that's scheduled when the VMCI
658 * interrupt fires. This is run in bottom-half context but it defers most of
659 * its work to the packet handling work queue.
660 */
661
662static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
663{
664 struct sock *sk;
665 struct sockaddr_vm dst;
666 struct sockaddr_vm src;
667 struct vmci_transport_packet *pkt;
668 struct vsock_sock *vsk;
669 bool bh_process_pkt;
670 int err;
671
672 sk = NULL;
673 err = VMCI_SUCCESS;
674 bh_process_pkt = false;
675
676 /* Ignore incoming packets from contexts without sockets, or resources
677 * that aren't vsock implementations.
678 */
679
680 if (!vmci_transport_stream_allow(dg->src.context, -1)
Reilly Grant2a89f922013-03-14 11:55:41 +0000681 || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
Andy Kingd021c342013-02-06 14:23:56 +0000682 return VMCI_ERROR_NO_ACCESS;
683
684 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
685 /* Drop datagrams that do not contain full VSock packets. */
686 return VMCI_ERROR_INVALID_ARGS;
687
688 pkt = (struct vmci_transport_packet *)dg;
689
690 /* Find the socket that should handle this packet. First we look for a
691 * connected socket and if there is none we look for a socket bound to
692 * the destintation address.
693 */
694 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
695 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
696
697 sk = vsock_find_connected_socket(&src, &dst);
698 if (!sk) {
699 sk = vsock_find_bound_socket(&dst);
700 if (!sk) {
701 /* We could not find a socket for this specified
702 * address. If this packet is a RST, we just drop it.
703 * If it is another packet, we send a RST. Note that
704 * we do not send a RST reply to RSTs so that we do not
705 * continually send RSTs between two endpoints.
706 *
707 * Note that since this is a reply, dst is src and src
708 * is dst.
709 */
710 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
711 pr_err("unable to send reset\n");
712
713 err = VMCI_ERROR_NOT_FOUND;
714 goto out;
715 }
716 }
717
718 /* If the received packet type is beyond all types known to this
719 * implementation, reply with an invalid message. Hopefully this will
720 * help when implementing backwards compatibility in the future.
721 */
722 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
723 vmci_transport_send_invalid_bh(&dst, &src);
724 err = VMCI_ERROR_INVALID_ARGS;
725 goto out;
726 }
727
728 /* This handler is privileged when this module is running on the host.
729 * We will get datagram connect requests from all endpoints (even VMs
730 * that are in a restricted context). If we get one from a restricted
731 * context then the destination socket must be trusted.
732 *
733 * NOTE: We access the socket struct without holding the lock here.
734 * This is ok because the field we are interested is never modified
735 * outside of the create and destruct socket functions.
736 */
737 vsk = vsock_sk(sk);
738 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
739 err = VMCI_ERROR_NO_ACCESS;
740 goto out;
741 }
742
743 /* We do most everything in a work queue, but let's fast path the
744 * notification of reads and writes to help data transfer performance.
745 * We can only do this if there is no process context code executing
746 * for this socket since that may change the state.
747 */
748 bh_lock_sock(sk);
749
Reilly Grant990454b2013-04-01 11:41:52 -0700750 if (!sock_owned_by_user(sk)) {
751 /* The local context ID may be out of date, update it. */
752 vsk->local_addr.svm_cid = dst.svm_cid;
753
754 if (sk->sk_state == SS_CONNECTED)
755 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
756 sk, pkt, true, &dst, &src,
757 &bh_process_pkt);
758 }
Andy Kingd021c342013-02-06 14:23:56 +0000759
760 bh_unlock_sock(sk);
761
762 if (!bh_process_pkt) {
763 struct vmci_transport_recv_pkt_info *recv_pkt_info;
764
765 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
766 if (!recv_pkt_info) {
767 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
768 pr_err("unable to send reset\n");
769
770 err = VMCI_ERROR_NO_MEM;
771 goto out;
772 }
773
774 recv_pkt_info->sk = sk;
775 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
776 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
777
778 schedule_work(&recv_pkt_info->work);
779 /* Clear sk so that the reference count incremented by one of
780 * the Find functions above is not decremented below. We need
781 * that reference count for the packet handler we've scheduled
782 * to run.
783 */
784 sk = NULL;
785 }
786
787out:
788 if (sk)
789 sock_put(sk);
790
791 return err;
792}
793
Andy Kingd021c342013-02-06 14:23:56 +0000794static void vmci_transport_handle_detach(struct sock *sk)
795{
796 struct vsock_sock *vsk;
797
798 vsk = vsock_sk(sk);
799 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
800 sock_set_flag(sk, SOCK_DONE);
801
802 /* On a detach the peer will not be sending or receiving
803 * anymore.
804 */
805 vsk->peer_shutdown = SHUTDOWN_MASK;
806
807 /* We should not be sending anymore since the peer won't be
808 * there to receive, but we can still receive if there is data
809 * left in our consume queue.
810 */
811 if (vsock_stream_has_data(vsk) <= 0) {
812 if (sk->sk_state == SS_CONNECTING) {
813 /* The peer may detach from a queue pair while
814 * we are still in the connecting state, i.e.,
815 * if the peer VM is killed after attaching to
816 * a queue pair, but before we complete the
817 * handshake. In that case, we treat the detach
818 * event like a reset.
819 */
820
821 sk->sk_state = SS_UNCONNECTED;
822 sk->sk_err = ECONNRESET;
823 sk->sk_error_report(sk);
824 return;
825 }
826 sk->sk_state = SS_UNCONNECTED;
827 }
828 sk->sk_state_change(sk);
829 }
830}
831
832static void vmci_transport_peer_detach_cb(u32 sub_id,
833 const struct vmci_event_data *e_data,
834 void *client_data)
835{
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -0700836 struct vmci_transport *trans = client_data;
Andy Kingd021c342013-02-06 14:23:56 +0000837 const struct vmci_event_payload_qp *e_payload;
Andy Kingd021c342013-02-06 14:23:56 +0000838
839 e_payload = vmci_event_data_const_payload(e_data);
Andy Kingd021c342013-02-06 14:23:56 +0000840
841 /* XXX This is lame, we should provide a way to lookup sockets by
842 * qp_handle.
843 */
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -0700844 if (vmci_handle_is_invalid(e_payload->handle) ||
Jorgen Hansen8ab18d72016-04-05 01:59:32 -0700845 !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -0700846 return;
Andy Kingd021c342013-02-06 14:23:56 +0000847
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -0700848 /* We don't ask for delayed CBs when we subscribe to this event (we
849 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
850 * guarantees in that case about what context we might be running in,
851 * so it could be BH or process, blockable or non-blockable. So we
852 * need to account for all possible contexts here.
853 */
854 spin_lock_bh(&trans->lock);
855 if (!trans->sk)
856 goto out;
857
858 /* Apart from here, trans->lock is only grabbed as part of sk destruct,
859 * where trans->sk isn't locked.
860 */
861 bh_lock_sock(trans->sk);
862
863 vmci_transport_handle_detach(trans->sk);
864
865 bh_unlock_sock(trans->sk);
866 out:
867 spin_unlock_bh(&trans->lock);
Andy Kingd021c342013-02-06 14:23:56 +0000868}
869
870static void vmci_transport_qp_resumed_cb(u32 sub_id,
871 const struct vmci_event_data *e_data,
872 void *client_data)
873{
874 vsock_for_each_connected_socket(vmci_transport_handle_detach);
875}
876
877static void vmci_transport_recv_pkt_work(struct work_struct *work)
878{
879 struct vmci_transport_recv_pkt_info *recv_pkt_info;
880 struct vmci_transport_packet *pkt;
881 struct sock *sk;
882
883 recv_pkt_info =
884 container_of(work, struct vmci_transport_recv_pkt_info, work);
885 sk = recv_pkt_info->sk;
886 pkt = &recv_pkt_info->pkt;
887
888 lock_sock(sk);
889
Reilly Grant990454b2013-04-01 11:41:52 -0700890 /* The local context ID may be out of date. */
891 vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
892
Andy Kingd021c342013-02-06 14:23:56 +0000893 switch (sk->sk_state) {
Stefan Hajnocziea3803c2015-10-29 11:57:42 +0000894 case VSOCK_SS_LISTEN:
Andy Kingd021c342013-02-06 14:23:56 +0000895 vmci_transport_recv_listen(sk, pkt);
896 break;
897 case SS_CONNECTING:
898 /* Processing of pending connections for servers goes through
899 * the listening socket, so see vmci_transport_recv_listen()
900 * for that path.
901 */
902 vmci_transport_recv_connecting_client(sk, pkt);
903 break;
904 case SS_CONNECTED:
905 vmci_transport_recv_connected(sk, pkt);
906 break;
907 default:
908 /* Because this function does not run in the same context as
909 * vmci_transport_recv_stream_cb it is possible that the
910 * socket has closed. We need to let the other side know or it
911 * could be sitting in a connect and hang forever. Send a
912 * reset to prevent that.
913 */
914 vmci_transport_send_reset(sk, pkt);
Asias He0fc93242013-06-20 17:20:32 +0800915 break;
Andy Kingd021c342013-02-06 14:23:56 +0000916 }
917
Andy Kingd021c342013-02-06 14:23:56 +0000918 release_sock(sk);
919 kfree(recv_pkt_info);
920 /* Release reference obtained in the stream callback when we fetched
921 * this socket out of the bound or connected list.
922 */
923 sock_put(sk);
924}
925
926static int vmci_transport_recv_listen(struct sock *sk,
927 struct vmci_transport_packet *pkt)
928{
929 struct sock *pending;
930 struct vsock_sock *vpending;
931 int err;
932 u64 qp_size;
933 bool old_request = false;
934 bool old_pkt_proto = false;
935
936 err = 0;
937
938 /* Because we are in the listen state, we could be receiving a packet
939 * for ourself or any previous connection requests that we received.
940 * If it's the latter, we try to find a socket in our list of pending
941 * connections and, if we do, call the appropriate handler for the
942 * state that that socket is in. Otherwise we try to service the
943 * connection request.
944 */
945 pending = vmci_transport_get_pending(sk, pkt);
946 if (pending) {
947 lock_sock(pending);
Reilly Grant990454b2013-04-01 11:41:52 -0700948
949 /* The local context ID may be out of date. */
950 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
951
Andy Kingd021c342013-02-06 14:23:56 +0000952 switch (pending->sk_state) {
953 case SS_CONNECTING:
954 err = vmci_transport_recv_connecting_server(sk,
955 pending,
956 pkt);
957 break;
958 default:
959 vmci_transport_send_reset(pending, pkt);
960 err = -EINVAL;
961 }
962
963 if (err < 0)
964 vsock_remove_pending(sk, pending);
965
966 release_sock(pending);
967 vmci_transport_release_pending(pending);
968
969 return err;
970 }
971
972 /* The listen state only accepts connection requests. Reply with a
973 * reset unless we received a reset.
974 */
975
976 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
977 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
978 vmci_transport_reply_reset(pkt);
979 return -EINVAL;
980 }
981
982 if (pkt->u.size == 0) {
983 vmci_transport_reply_reset(pkt);
984 return -EINVAL;
985 }
986
987 /* If this socket can't accommodate this connection request, we send a
988 * reset. Otherwise we create and initialize a child socket and reply
989 * with a connection negotiation.
990 */
991 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
992 vmci_transport_reply_reset(pkt);
993 return -ECONNREFUSED;
994 }
995
996 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500997 sk->sk_type, 0);
Andy Kingd021c342013-02-06 14:23:56 +0000998 if (!pending) {
999 vmci_transport_send_reset(sk, pkt);
1000 return -ENOMEM;
1001 }
1002
1003 vpending = vsock_sk(pending);
1004
1005 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1006 pkt->dst_port);
1007 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1008 pkt->src_port);
1009
1010 /* If the proposed size fits within our min/max, accept it. Otherwise
1011 * propose our own size.
1012 */
1013 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1014 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1015 qp_size = pkt->u.size;
1016 } else {
1017 qp_size = vmci_trans(vpending)->queue_pair_size;
1018 }
1019
1020 /* Figure out if we are using old or new requests based on the
1021 * overrides pkt types sent by our peer.
1022 */
1023 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1024 old_request = old_pkt_proto;
1025 } else {
1026 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1027 old_request = true;
1028 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1029 old_request = false;
1030
1031 }
1032
1033 if (old_request) {
1034 /* Handle a REQUEST (or override) */
1035 u16 version = VSOCK_PROTO_INVALID;
1036 if (vmci_transport_proto_to_notify_struct(
1037 pending, &version, true))
1038 err = vmci_transport_send_negotiate(pending, qp_size);
1039 else
1040 err = -EINVAL;
1041
1042 } else {
1043 /* Handle a REQUEST2 (or override) */
1044 int proto_int = pkt->proto;
1045 int pos;
1046 u16 active_proto_version = 0;
1047
1048 /* The list of possible protocols is the intersection of all
1049 * protocols the client supports ... plus all the protocols we
1050 * support.
1051 */
1052 proto_int &= vmci_transport_new_proto_supported_versions();
1053
1054 /* We choose the highest possible protocol version and use that
1055 * one.
1056 */
1057 pos = fls(proto_int);
1058 if (pos) {
1059 active_proto_version = (1 << (pos - 1));
1060 if (vmci_transport_proto_to_notify_struct(
1061 pending, &active_proto_version, false))
1062 err = vmci_transport_send_negotiate2(pending,
1063 qp_size,
1064 active_proto_version);
1065 else
1066 err = -EINVAL;
1067
1068 } else {
1069 err = -EINVAL;
1070 }
1071 }
1072
1073 if (err < 0) {
1074 vmci_transport_send_reset(sk, pkt);
1075 sock_put(pending);
1076 err = vmci_transport_error_to_vsock_error(err);
1077 goto out;
1078 }
1079
1080 vsock_add_pending(sk, pending);
1081 sk->sk_ack_backlog++;
1082
1083 pending->sk_state = SS_CONNECTING;
1084 vmci_trans(vpending)->produce_size =
1085 vmci_trans(vpending)->consume_size = qp_size;
1086 vmci_trans(vpending)->queue_pair_size = qp_size;
1087
1088 vmci_trans(vpending)->notify_ops->process_request(pending);
1089
1090 /* We might never receive another message for this socket and it's not
1091 * connected to any process, so we have to ensure it gets cleaned up
1092 * ourself. Our delayed work function will take care of that. Note
1093 * that we do not ever cancel this function since we have few
1094 * guarantees about its state when calling cancel_delayed_work().
1095 * Instead we hold a reference on the socket for that function and make
1096 * it capable of handling cases where it needs to do nothing but
1097 * release that reference.
1098 */
1099 vpending->listener = sk;
1100 sock_hold(sk);
1101 sock_hold(pending);
1102 INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1103 schedule_delayed_work(&vpending->dwork, HZ);
1104
1105out:
1106 return err;
1107}
1108
1109static int
1110vmci_transport_recv_connecting_server(struct sock *listener,
1111 struct sock *pending,
1112 struct vmci_transport_packet *pkt)
1113{
1114 struct vsock_sock *vpending;
1115 struct vmci_handle handle;
1116 struct vmci_qp *qpair;
1117 bool is_local;
1118 u32 flags;
1119 u32 detach_sub_id;
1120 int err;
1121 int skerr;
1122
1123 vpending = vsock_sk(pending);
1124 detach_sub_id = VMCI_INVALID_ID;
1125
1126 switch (pkt->type) {
1127 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1128 if (vmci_handle_is_invalid(pkt->u.handle)) {
1129 vmci_transport_send_reset(pending, pkt);
1130 skerr = EPROTO;
1131 err = -EINVAL;
1132 goto destroy;
1133 }
1134 break;
1135 default:
1136 /* Close and cleanup the connection. */
1137 vmci_transport_send_reset(pending, pkt);
1138 skerr = EPROTO;
1139 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1140 goto destroy;
1141 }
1142
1143 /* In order to complete the connection we need to attach to the offered
1144 * queue pair and send an attach notification. We also subscribe to the
1145 * detach event so we know when our peer goes away, and we do that
1146 * before attaching so we don't miss an event. If all this succeeds,
1147 * we update our state and wakeup anything waiting in accept() for a
1148 * connection.
1149 */
1150
1151 /* We don't care about attach since we ensure the other side has
1152 * attached by specifying the ATTACH_ONLY flag below.
1153 */
1154 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1155 vmci_transport_peer_detach_cb,
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001156 vmci_trans(vpending), &detach_sub_id);
Andy Kingd021c342013-02-06 14:23:56 +00001157 if (err < VMCI_SUCCESS) {
1158 vmci_transport_send_reset(pending, pkt);
1159 err = vmci_transport_error_to_vsock_error(err);
1160 skerr = -err;
1161 goto destroy;
1162 }
1163
1164 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1165
1166 /* Now attach to the queue pair the client created. */
1167 handle = pkt->u.handle;
1168
1169 /* vpending->local_addr always has a context id so we do not need to
1170 * worry about VMADDR_CID_ANY in this case.
1171 */
1172 is_local =
1173 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1174 flags = VMCI_QPFLAG_ATTACH_ONLY;
1175 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1176
1177 err = vmci_transport_queue_pair_alloc(
1178 &qpair,
1179 &handle,
1180 vmci_trans(vpending)->produce_size,
1181 vmci_trans(vpending)->consume_size,
1182 pkt->dg.src.context,
1183 flags,
1184 vmci_transport_is_trusted(
1185 vpending,
1186 vpending->remote_addr.svm_cid));
1187 if (err < 0) {
1188 vmci_transport_send_reset(pending, pkt);
1189 skerr = -err;
1190 goto destroy;
1191 }
1192
1193 vmci_trans(vpending)->qp_handle = handle;
1194 vmci_trans(vpending)->qpair = qpair;
1195
1196 /* When we send the attach message, we must be ready to handle incoming
1197 * control messages on the newly connected socket. So we move the
1198 * pending socket to the connected state before sending the attach
1199 * message. Otherwise, an incoming packet triggered by the attach being
1200 * received by the peer may be processed concurrently with what happens
1201 * below after sending the attach message, and that incoming packet
1202 * will find the listening socket instead of the (currently) pending
1203 * socket. Note that enqueueing the socket increments the reference
1204 * count, so even if a reset comes before the connection is accepted,
1205 * the socket will be valid until it is removed from the queue.
1206 *
1207 * If we fail sending the attach below, we remove the socket from the
1208 * connected list and move the socket to SS_UNCONNECTED before
1209 * releasing the lock, so a pending slow path processing of an incoming
1210 * packet will not see the socket in the connected state in that case.
1211 */
1212 pending->sk_state = SS_CONNECTED;
1213
1214 vsock_insert_connected(vpending);
1215
1216 /* Notify our peer of our attach. */
1217 err = vmci_transport_send_attach(pending, handle);
1218 if (err < 0) {
1219 vsock_remove_connected(vpending);
1220 pr_err("Could not send attach\n");
1221 vmci_transport_send_reset(pending, pkt);
1222 err = vmci_transport_error_to_vsock_error(err);
1223 skerr = -err;
1224 goto destroy;
1225 }
1226
1227 /* We have a connection. Move the now connected socket from the
1228 * listener's pending list to the accept queue so callers of accept()
1229 * can find it.
1230 */
1231 vsock_remove_pending(listener, pending);
1232 vsock_enqueue_accept(listener, pending);
1233
1234 /* Callers of accept() will be be waiting on the listening socket, not
1235 * the pending socket.
1236 */
Stefan Hajnoczi73629452015-11-04 12:58:42 +00001237 listener->sk_data_ready(listener);
Andy Kingd021c342013-02-06 14:23:56 +00001238
1239 return 0;
1240
1241destroy:
1242 pending->sk_err = skerr;
1243 pending->sk_state = SS_UNCONNECTED;
1244 /* As long as we drop our reference, all necessary cleanup will handle
1245 * when the cleanup function drops its reference and our destruct
1246 * implementation is called. Note that since the listen handler will
1247 * remove pending from the pending list upon our failure, the cleanup
1248 * function won't drop the additional reference, which is why we do it
1249 * here.
1250 */
1251 sock_put(pending);
1252
1253 return err;
1254}
1255
1256static int
1257vmci_transport_recv_connecting_client(struct sock *sk,
1258 struct vmci_transport_packet *pkt)
1259{
1260 struct vsock_sock *vsk;
1261 int err;
1262 int skerr;
1263
1264 vsk = vsock_sk(sk);
1265
1266 switch (pkt->type) {
1267 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1268 if (vmci_handle_is_invalid(pkt->u.handle) ||
1269 !vmci_handle_is_equal(pkt->u.handle,
1270 vmci_trans(vsk)->qp_handle)) {
1271 skerr = EPROTO;
1272 err = -EINVAL;
1273 goto destroy;
1274 }
1275
1276 /* Signify the socket is connected and wakeup the waiter in
1277 * connect(). Also place the socket in the connected table for
1278 * accounting (it can already be found since it's in the bound
1279 * table).
1280 */
1281 sk->sk_state = SS_CONNECTED;
1282 sk->sk_socket->state = SS_CONNECTED;
1283 vsock_insert_connected(vsk);
1284 sk->sk_state_change(sk);
1285
1286 break;
1287 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1288 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1289 if (pkt->u.size == 0
1290 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1291 || pkt->src_port != vsk->remote_addr.svm_port
1292 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1293 || vmci_trans(vsk)->qpair
1294 || vmci_trans(vsk)->produce_size != 0
1295 || vmci_trans(vsk)->consume_size != 0
Andy Kingd021c342013-02-06 14:23:56 +00001296 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1297 skerr = EPROTO;
1298 err = -EINVAL;
1299
1300 goto destroy;
1301 }
1302
1303 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1304 if (err) {
1305 skerr = -err;
1306 goto destroy;
1307 }
1308
1309 break;
1310 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1311 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1312 if (err) {
1313 skerr = -err;
1314 goto destroy;
1315 }
1316
1317 break;
1318 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1319 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1320 * continue processing here after they sent an INVALID packet.
1321 * This meant that we got a RST after the INVALID. We ignore a
1322 * RST after an INVALID. The common code doesn't send the RST
1323 * ... so we can hang if an old version of the common code
1324 * fails between getting a REQUEST and sending an OFFER back.
1325 * Not much we can do about it... except hope that it doesn't
1326 * happen.
1327 */
1328 if (vsk->ignore_connecting_rst) {
1329 vsk->ignore_connecting_rst = false;
1330 } else {
1331 skerr = ECONNRESET;
1332 err = 0;
1333 goto destroy;
1334 }
1335
1336 break;
1337 default:
1338 /* Close and cleanup the connection. */
1339 skerr = EPROTO;
1340 err = -EINVAL;
1341 goto destroy;
1342 }
1343
1344 return 0;
1345
1346destroy:
1347 vmci_transport_send_reset(sk, pkt);
1348
1349 sk->sk_state = SS_UNCONNECTED;
1350 sk->sk_err = skerr;
1351 sk->sk_error_report(sk);
1352 return err;
1353}
1354
1355static int vmci_transport_recv_connecting_client_negotiate(
1356 struct sock *sk,
1357 struct vmci_transport_packet *pkt)
1358{
1359 int err;
1360 struct vsock_sock *vsk;
1361 struct vmci_handle handle;
1362 struct vmci_qp *qpair;
Andy Kingd021c342013-02-06 14:23:56 +00001363 u32 detach_sub_id;
1364 bool is_local;
1365 u32 flags;
1366 bool old_proto = true;
1367 bool old_pkt_proto;
1368 u16 version;
1369
1370 vsk = vsock_sk(sk);
1371 handle = VMCI_INVALID_HANDLE;
Andy Kingd021c342013-02-06 14:23:56 +00001372 detach_sub_id = VMCI_INVALID_ID;
1373
1374 /* If we have gotten here then we should be past the point where old
1375 * linux vsock could have sent the bogus rst.
1376 */
1377 vsk->sent_request = false;
1378 vsk->ignore_connecting_rst = false;
1379
1380 /* Verify that we're OK with the proposed queue pair size */
1381 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1382 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1383 err = -EINVAL;
1384 goto destroy;
1385 }
1386
1387 /* At this point we know the CID the peer is using to talk to us. */
1388
1389 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1390 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1391
1392 /* Setup the notify ops to be the highest supported version that both
1393 * the server and the client support.
1394 */
1395
1396 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1397 old_proto = old_pkt_proto;
1398 } else {
1399 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1400 old_proto = true;
1401 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1402 old_proto = false;
1403
1404 }
1405
1406 if (old_proto)
1407 version = VSOCK_PROTO_INVALID;
1408 else
1409 version = pkt->proto;
1410
1411 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1412 err = -EINVAL;
1413 goto destroy;
1414 }
1415
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001416 /* Subscribe to detach events first.
Andy Kingd021c342013-02-06 14:23:56 +00001417 *
1418 * XXX We attach once for each queue pair created for now so it is easy
1419 * to find the socket (it's provided), but later we should only
1420 * subscribe once and add a way to lookup sockets by queue pair handle.
1421 */
Andy Kingd021c342013-02-06 14:23:56 +00001422 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1423 vmci_transport_peer_detach_cb,
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001424 vmci_trans(vsk), &detach_sub_id);
Andy Kingd021c342013-02-06 14:23:56 +00001425 if (err < VMCI_SUCCESS) {
1426 err = vmci_transport_error_to_vsock_error(err);
1427 goto destroy;
1428 }
1429
1430 /* Make VMCI select the handle for us. */
1431 handle = VMCI_INVALID_HANDLE;
1432 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1433 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1434
1435 err = vmci_transport_queue_pair_alloc(&qpair,
1436 &handle,
1437 pkt->u.size,
1438 pkt->u.size,
1439 vsk->remote_addr.svm_cid,
1440 flags,
1441 vmci_transport_is_trusted(
1442 vsk,
1443 vsk->
1444 remote_addr.svm_cid));
1445 if (err < 0)
1446 goto destroy;
1447
1448 err = vmci_transport_send_qp_offer(sk, handle);
1449 if (err < 0) {
1450 err = vmci_transport_error_to_vsock_error(err);
1451 goto destroy;
1452 }
1453
1454 vmci_trans(vsk)->qp_handle = handle;
1455 vmci_trans(vsk)->qpair = qpair;
1456
1457 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1458 pkt->u.size;
1459
Andy Kingd021c342013-02-06 14:23:56 +00001460 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1461
1462 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1463
1464 return 0;
1465
1466destroy:
Andy Kingd021c342013-02-06 14:23:56 +00001467 if (detach_sub_id != VMCI_INVALID_ID)
1468 vmci_event_unsubscribe(detach_sub_id);
1469
1470 if (!vmci_handle_is_invalid(handle))
1471 vmci_qpair_detach(&qpair);
1472
1473 return err;
1474}
1475
1476static int
1477vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1478 struct vmci_transport_packet *pkt)
1479{
1480 int err = 0;
1481 struct vsock_sock *vsk = vsock_sk(sk);
1482
1483 if (vsk->sent_request) {
1484 vsk->sent_request = false;
1485 vsk->ignore_connecting_rst = true;
1486
1487 err = vmci_transport_send_conn_request(
1488 sk, vmci_trans(vsk)->queue_pair_size);
1489 if (err < 0)
1490 err = vmci_transport_error_to_vsock_error(err);
1491 else
1492 err = 0;
1493
1494 }
1495
1496 return err;
1497}
1498
1499static int vmci_transport_recv_connected(struct sock *sk,
1500 struct vmci_transport_packet *pkt)
1501{
1502 struct vsock_sock *vsk;
1503 bool pkt_processed = false;
1504
1505 /* In cases where we are closing the connection, it's sufficient to
1506 * mark the state change (and maybe error) and wake up any waiting
1507 * threads. Since this is a connected socket, it's owned by a user
1508 * process and will be cleaned up when the failure is passed back on
1509 * the current or next system call. Our system call implementations
1510 * must therefore check for error and state changes on entry and when
1511 * being awoken.
1512 */
1513 switch (pkt->type) {
1514 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1515 if (pkt->u.mode) {
1516 vsk = vsock_sk(sk);
1517
1518 vsk->peer_shutdown |= pkt->u.mode;
1519 sk->sk_state_change(sk);
1520 }
1521 break;
1522
1523 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1524 vsk = vsock_sk(sk);
1525 /* It is possible that we sent our peer a message (e.g a
1526 * WAITING_READ) right before we got notified that the peer had
1527 * detached. If that happens then we can get a RST pkt back
1528 * from our peer even though there is data available for us to
1529 * read. In that case, don't shutdown the socket completely but
1530 * instead allow the local client to finish reading data off
1531 * the queuepair. Always treat a RST pkt in connected mode like
1532 * a clean shutdown.
1533 */
1534 sock_set_flag(sk, SOCK_DONE);
1535 vsk->peer_shutdown = SHUTDOWN_MASK;
1536 if (vsock_stream_has_data(vsk) <= 0)
1537 sk->sk_state = SS_DISCONNECTING;
1538
1539 sk->sk_state_change(sk);
1540 break;
1541
1542 default:
1543 vsk = vsock_sk(sk);
1544 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1545 sk, pkt, false, NULL, NULL,
1546 &pkt_processed);
1547 if (!pkt_processed)
1548 return -EINVAL;
1549
1550 break;
1551 }
1552
1553 return 0;
1554}
1555
1556static int vmci_transport_socket_init(struct vsock_sock *vsk,
1557 struct vsock_sock *psk)
1558{
1559 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1560 if (!vsk->trans)
1561 return -ENOMEM;
1562
1563 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1564 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1565 vmci_trans(vsk)->qpair = NULL;
1566 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001567 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
Andy Kingd021c342013-02-06 14:23:56 +00001568 vmci_trans(vsk)->notify_ops = NULL;
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001569 INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1570 vmci_trans(vsk)->sk = &vsk->sk;
Jorgen Hansen8566b862015-10-22 08:25:25 -07001571 spin_lock_init(&vmci_trans(vsk)->lock);
Andy Kingd021c342013-02-06 14:23:56 +00001572 if (psk) {
1573 vmci_trans(vsk)->queue_pair_size =
1574 vmci_trans(psk)->queue_pair_size;
1575 vmci_trans(vsk)->queue_pair_min_size =
1576 vmci_trans(psk)->queue_pair_min_size;
1577 vmci_trans(vsk)->queue_pair_max_size =
1578 vmci_trans(psk)->queue_pair_max_size;
1579 } else {
1580 vmci_trans(vsk)->queue_pair_size =
1581 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1582 vmci_trans(vsk)->queue_pair_min_size =
1583 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1584 vmci_trans(vsk)->queue_pair_max_size =
1585 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1586 }
1587
1588 return 0;
1589}
1590
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001591static void vmci_transport_free_resources(struct list_head *transport_list)
1592{
1593 while (!list_empty(transport_list)) {
1594 struct vmci_transport *transport =
1595 list_first_entry(transport_list, struct vmci_transport,
1596 elem);
1597 list_del(&transport->elem);
1598
1599 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1600 vmci_event_unsubscribe(transport->detach_sub_id);
1601 transport->detach_sub_id = VMCI_INVALID_ID;
1602 }
1603
1604 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1605 vmci_qpair_detach(&transport->qpair);
1606 transport->qp_handle = VMCI_INVALID_HANDLE;
1607 transport->produce_size = 0;
1608 transport->consume_size = 0;
1609 }
1610
1611 kfree(transport);
1612 }
1613}
1614
1615static void vmci_transport_cleanup(struct work_struct *work)
1616{
1617 LIST_HEAD(pending);
1618
1619 spin_lock_bh(&vmci_transport_cleanup_lock);
1620 list_replace_init(&vmci_transport_cleanup_list, &pending);
1621 spin_unlock_bh(&vmci_transport_cleanup_lock);
1622 vmci_transport_free_resources(&pending);
1623}
1624
Andy Kingd021c342013-02-06 14:23:56 +00001625static void vmci_transport_destruct(struct vsock_sock *vsk)
1626{
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001627 /* Ensure that the detach callback doesn't use the sk/vsk
1628 * we are about to destruct.
1629 */
1630 spin_lock_bh(&vmci_trans(vsk)->lock);
1631 vmci_trans(vsk)->sk = NULL;
1632 spin_unlock_bh(&vmci_trans(vsk)->lock);
Andy Kingd021c342013-02-06 14:23:56 +00001633
1634 if (vmci_trans(vsk)->notify_ops)
1635 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1636
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07001637 spin_lock_bh(&vmci_transport_cleanup_lock);
1638 list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1639 spin_unlock_bh(&vmci_transport_cleanup_lock);
1640 schedule_work(&vmci_transport_cleanup_work);
1641
Andy Kingd021c342013-02-06 14:23:56 +00001642 vsk->trans = NULL;
1643}
1644
1645static void vmci_transport_release(struct vsock_sock *vsk)
1646{
1647 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1648 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1649 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1650 }
1651}
1652
1653static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1654 struct sockaddr_vm *addr)
1655{
1656 u32 port;
1657 u32 flags;
1658 int err;
1659
1660 /* VMCI will select a resource ID for us if we provide
1661 * VMCI_INVALID_ID.
1662 */
1663 port = addr->svm_port == VMADDR_PORT_ANY ?
1664 VMCI_INVALID_ID : addr->svm_port;
1665
1666 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1667 return -EACCES;
1668
1669 flags = addr->svm_cid == VMADDR_CID_ANY ?
1670 VMCI_FLAG_ANYCID_DG_HND : 0;
1671
1672 err = vmci_transport_datagram_create_hnd(port, flags,
1673 vmci_transport_recv_dgram_cb,
1674 &vsk->sk,
1675 &vmci_trans(vsk)->dg_handle);
1676 if (err < VMCI_SUCCESS)
1677 return vmci_transport_error_to_vsock_error(err);
1678 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1679 vmci_trans(vsk)->dg_handle.resource);
1680
1681 return 0;
1682}
1683
1684static int vmci_transport_dgram_enqueue(
1685 struct vsock_sock *vsk,
1686 struct sockaddr_vm *remote_addr,
Al Viro0f7db232014-11-20 04:05:34 -05001687 struct msghdr *msg,
Andy Kingd021c342013-02-06 14:23:56 +00001688 size_t len)
1689{
1690 int err;
1691 struct vmci_datagram *dg;
1692
1693 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1694 return -EMSGSIZE;
1695
1696 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1697 return -EPERM;
1698
1699 /* Allocate a buffer for the user's message and our packet header. */
1700 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1701 if (!dg)
1702 return -ENOMEM;
1703
Al Viro0f7db232014-11-20 04:05:34 -05001704 memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
Andy Kingd021c342013-02-06 14:23:56 +00001705
1706 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1707 remote_addr->svm_port);
1708 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1709 vsk->local_addr.svm_port);
1710 dg->payload_size = len;
1711
1712 err = vmci_datagram_send(dg);
1713 kfree(dg);
1714 if (err < 0)
1715 return vmci_transport_error_to_vsock_error(err);
1716
1717 return err - sizeof(*dg);
1718}
1719
Ying Xue1b784142015-03-02 15:37:48 +08001720static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
Andy Kingd021c342013-02-06 14:23:56 +00001721 struct msghdr *msg, size_t len,
1722 int flags)
1723{
1724 int err;
1725 int noblock;
1726 struct vmci_datagram *dg;
1727 size_t payload_len;
1728 struct sk_buff *skb;
1729
1730 noblock = flags & MSG_DONTWAIT;
1731
1732 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1733 return -EOPNOTSUPP;
1734
1735 /* Retrieve the head sk_buff from the socket's receive queue. */
1736 err = 0;
1737 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
Andy Kingd021c342013-02-06 14:23:56 +00001738 if (!skb)
Jorgen Hansen9c995cc2016-04-18 23:58:52 -07001739 return err;
Andy Kingd021c342013-02-06 14:23:56 +00001740
1741 dg = (struct vmci_datagram *)skb->data;
1742 if (!dg)
1743 /* err is 0, meaning we read zero bytes. */
1744 goto out;
1745
1746 payload_len = dg->payload_size;
1747 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1748 if (payload_len != skb->len - sizeof(*dg)) {
1749 err = -EINVAL;
1750 goto out;
1751 }
1752
1753 if (payload_len > len) {
1754 payload_len = len;
1755 msg->msg_flags |= MSG_TRUNC;
1756 }
1757
1758 /* Place the datagram payload in the user's iovec. */
David S. Miller51f3d022014-11-05 16:46:40 -05001759 err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
Andy Kingd021c342013-02-06 14:23:56 +00001760 if (err)
1761 goto out;
1762
Andy Kingd021c342013-02-06 14:23:56 +00001763 if (msg->msg_name) {
Andy Kingd021c342013-02-06 14:23:56 +00001764 /* Provide the address of the sender. */
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001765 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
Andy Kingd021c342013-02-06 14:23:56 +00001766 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1767 msg->msg_namelen = sizeof(*vm_addr);
1768 }
1769 err = payload_len;
1770
1771out:
1772 skb_free_datagram(&vsk->sk, skb);
1773 return err;
1774}
1775
1776static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1777{
1778 if (cid == VMADDR_CID_HYPERVISOR) {
1779 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1780 * state and are allowed.
1781 */
1782 return port == VMCI_UNITY_PBRPC_REGISTER;
1783 }
1784
1785 return true;
1786}
1787
1788static int vmci_transport_connect(struct vsock_sock *vsk)
1789{
1790 int err;
1791 bool old_pkt_proto = false;
1792 struct sock *sk = &vsk->sk;
1793
1794 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1795 old_pkt_proto) {
1796 err = vmci_transport_send_conn_request(
1797 sk, vmci_trans(vsk)->queue_pair_size);
1798 if (err < 0) {
1799 sk->sk_state = SS_UNCONNECTED;
1800 return err;
1801 }
1802 } else {
1803 int supported_proto_versions =
1804 vmci_transport_new_proto_supported_versions();
1805 err = vmci_transport_send_conn_request2(
1806 sk, vmci_trans(vsk)->queue_pair_size,
1807 supported_proto_versions);
1808 if (err < 0) {
1809 sk->sk_state = SS_UNCONNECTED;
1810 return err;
1811 }
1812
1813 vsk->sent_request = true;
1814 }
1815
1816 return err;
1817}
1818
1819static ssize_t vmci_transport_stream_dequeue(
1820 struct vsock_sock *vsk,
Al Viro0f7db232014-11-20 04:05:34 -05001821 struct msghdr *msg,
Andy Kingd021c342013-02-06 14:23:56 +00001822 size_t len,
1823 int flags)
1824{
1825 if (flags & MSG_PEEK)
Al Virod838df22014-11-24 19:32:50 -05001826 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
Andy Kingd021c342013-02-06 14:23:56 +00001827 else
Al Virod838df22014-11-24 19:32:50 -05001828 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
Andy Kingd021c342013-02-06 14:23:56 +00001829}
1830
1831static ssize_t vmci_transport_stream_enqueue(
1832 struct vsock_sock *vsk,
Al Viro0f7db232014-11-20 04:05:34 -05001833 struct msghdr *msg,
Andy Kingd021c342013-02-06 14:23:56 +00001834 size_t len)
1835{
Al Viro4c946d92014-11-27 19:52:04 -05001836 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
Andy Kingd021c342013-02-06 14:23:56 +00001837}
1838
1839static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1840{
1841 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1842}
1843
1844static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1845{
1846 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1847}
1848
1849static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1850{
1851 return vmci_trans(vsk)->consume_size;
1852}
1853
1854static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1855{
1856 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1857}
1858
1859static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1860{
1861 return vmci_trans(vsk)->queue_pair_size;
1862}
1863
1864static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1865{
1866 return vmci_trans(vsk)->queue_pair_min_size;
1867}
1868
1869static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1870{
1871 return vmci_trans(vsk)->queue_pair_max_size;
1872}
1873
1874static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1875{
1876 if (val < vmci_trans(vsk)->queue_pair_min_size)
1877 vmci_trans(vsk)->queue_pair_min_size = val;
1878 if (val > vmci_trans(vsk)->queue_pair_max_size)
1879 vmci_trans(vsk)->queue_pair_max_size = val;
1880 vmci_trans(vsk)->queue_pair_size = val;
1881}
1882
1883static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1884 u64 val)
1885{
1886 if (val > vmci_trans(vsk)->queue_pair_size)
1887 vmci_trans(vsk)->queue_pair_size = val;
1888 vmci_trans(vsk)->queue_pair_min_size = val;
1889}
1890
1891static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1892 u64 val)
1893{
1894 if (val < vmci_trans(vsk)->queue_pair_size)
1895 vmci_trans(vsk)->queue_pair_size = val;
1896 vmci_trans(vsk)->queue_pair_max_size = val;
1897}
1898
1899static int vmci_transport_notify_poll_in(
1900 struct vsock_sock *vsk,
1901 size_t target,
1902 bool *data_ready_now)
1903{
1904 return vmci_trans(vsk)->notify_ops->poll_in(
1905 &vsk->sk, target, data_ready_now);
1906}
1907
1908static int vmci_transport_notify_poll_out(
1909 struct vsock_sock *vsk,
1910 size_t target,
1911 bool *space_available_now)
1912{
1913 return vmci_trans(vsk)->notify_ops->poll_out(
1914 &vsk->sk, target, space_available_now);
1915}
1916
1917static int vmci_transport_notify_recv_init(
1918 struct vsock_sock *vsk,
1919 size_t target,
1920 struct vsock_transport_recv_notify_data *data)
1921{
1922 return vmci_trans(vsk)->notify_ops->recv_init(
1923 &vsk->sk, target,
1924 (struct vmci_transport_recv_notify_data *)data);
1925}
1926
1927static int vmci_transport_notify_recv_pre_block(
1928 struct vsock_sock *vsk,
1929 size_t target,
1930 struct vsock_transport_recv_notify_data *data)
1931{
1932 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1933 &vsk->sk, target,
1934 (struct vmci_transport_recv_notify_data *)data);
1935}
1936
1937static int vmci_transport_notify_recv_pre_dequeue(
1938 struct vsock_sock *vsk,
1939 size_t target,
1940 struct vsock_transport_recv_notify_data *data)
1941{
1942 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1943 &vsk->sk, target,
1944 (struct vmci_transport_recv_notify_data *)data);
1945}
1946
1947static int vmci_transport_notify_recv_post_dequeue(
1948 struct vsock_sock *vsk,
1949 size_t target,
1950 ssize_t copied,
1951 bool data_read,
1952 struct vsock_transport_recv_notify_data *data)
1953{
1954 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1955 &vsk->sk, target, copied, data_read,
1956 (struct vmci_transport_recv_notify_data *)data);
1957}
1958
1959static int vmci_transport_notify_send_init(
1960 struct vsock_sock *vsk,
1961 struct vsock_transport_send_notify_data *data)
1962{
1963 return vmci_trans(vsk)->notify_ops->send_init(
1964 &vsk->sk,
1965 (struct vmci_transport_send_notify_data *)data);
1966}
1967
1968static int vmci_transport_notify_send_pre_block(
1969 struct vsock_sock *vsk,
1970 struct vsock_transport_send_notify_data *data)
1971{
1972 return vmci_trans(vsk)->notify_ops->send_pre_block(
1973 &vsk->sk,
1974 (struct vmci_transport_send_notify_data *)data);
1975}
1976
1977static int vmci_transport_notify_send_pre_enqueue(
1978 struct vsock_sock *vsk,
1979 struct vsock_transport_send_notify_data *data)
1980{
1981 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1982 &vsk->sk,
1983 (struct vmci_transport_send_notify_data *)data);
1984}
1985
1986static int vmci_transport_notify_send_post_enqueue(
1987 struct vsock_sock *vsk,
1988 ssize_t written,
1989 struct vsock_transport_send_notify_data *data)
1990{
1991 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1992 &vsk->sk, written,
1993 (struct vmci_transport_send_notify_data *)data);
1994}
1995
1996static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1997{
1998 if (PROTOCOL_OVERRIDE != -1) {
1999 if (PROTOCOL_OVERRIDE == 0)
2000 *old_pkt_proto = true;
2001 else
2002 *old_pkt_proto = false;
2003
2004 pr_info("Proto override in use\n");
2005 return true;
2006 }
2007
2008 return false;
2009}
2010
2011static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2012 u16 *proto,
2013 bool old_pkt_proto)
2014{
2015 struct vsock_sock *vsk = vsock_sk(sk);
2016
2017 if (old_pkt_proto) {
2018 if (*proto != VSOCK_PROTO_INVALID) {
2019 pr_err("Can't set both an old and new protocol\n");
2020 return false;
2021 }
2022 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2023 goto exit;
2024 }
2025
2026 switch (*proto) {
2027 case VSOCK_PROTO_PKT_ON_NOTIFY:
2028 vmci_trans(vsk)->notify_ops =
2029 &vmci_transport_notify_pkt_q_state_ops;
2030 break;
2031 default:
2032 pr_err("Unknown notify protocol version\n");
2033 return false;
2034 }
2035
2036exit:
2037 vmci_trans(vsk)->notify_ops->socket_init(sk);
2038 return true;
2039}
2040
2041static u16 vmci_transport_new_proto_supported_versions(void)
2042{
2043 if (PROTOCOL_OVERRIDE != -1)
2044 return PROTOCOL_OVERRIDE;
2045
2046 return VSOCK_PROTO_ALL_SUPPORTED;
2047}
2048
2049static u32 vmci_transport_get_local_cid(void)
2050{
2051 return vmci_get_context_id();
2052}
2053
2054static struct vsock_transport vmci_transport = {
2055 .init = vmci_transport_socket_init,
2056 .destruct = vmci_transport_destruct,
2057 .release = vmci_transport_release,
2058 .connect = vmci_transport_connect,
2059 .dgram_bind = vmci_transport_dgram_bind,
2060 .dgram_dequeue = vmci_transport_dgram_dequeue,
2061 .dgram_enqueue = vmci_transport_dgram_enqueue,
2062 .dgram_allow = vmci_transport_dgram_allow,
2063 .stream_dequeue = vmci_transport_stream_dequeue,
2064 .stream_enqueue = vmci_transport_stream_enqueue,
2065 .stream_has_data = vmci_transport_stream_has_data,
2066 .stream_has_space = vmci_transport_stream_has_space,
2067 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2068 .stream_is_active = vmci_transport_stream_is_active,
2069 .stream_allow = vmci_transport_stream_allow,
2070 .notify_poll_in = vmci_transport_notify_poll_in,
2071 .notify_poll_out = vmci_transport_notify_poll_out,
2072 .notify_recv_init = vmci_transport_notify_recv_init,
2073 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2074 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2075 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2076 .notify_send_init = vmci_transport_notify_send_init,
2077 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2078 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2079 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2080 .shutdown = vmci_transport_shutdown,
2081 .set_buffer_size = vmci_transport_set_buffer_size,
2082 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2083 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2084 .get_buffer_size = vmci_transport_get_buffer_size,
2085 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2086 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2087 .get_local_cid = vmci_transport_get_local_cid,
2088};
2089
2090static int __init vmci_transport_init(void)
2091{
2092 int err;
2093
2094 /* Create the datagram handle that we will use to send and receive all
2095 * VSocket control messages for this context.
2096 */
2097 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2098 VMCI_FLAG_ANYCID_DG_HND,
2099 vmci_transport_recv_stream_cb,
2100 NULL,
2101 &vmci_transport_stream_handle);
2102 if (err < VMCI_SUCCESS) {
2103 pr_err("Unable to create datagram handle. (%d)\n", err);
2104 return vmci_transport_error_to_vsock_error(err);
2105 }
2106
2107 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2108 vmci_transport_qp_resumed_cb,
2109 NULL, &vmci_transport_qp_resumed_sub_id);
2110 if (err < VMCI_SUCCESS) {
2111 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2112 err = vmci_transport_error_to_vsock_error(err);
2113 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2114 goto err_destroy_stream_handle;
2115 }
2116
2117 err = vsock_core_init(&vmci_transport);
2118 if (err < 0)
2119 goto err_unsubscribe;
2120
2121 return 0;
2122
2123err_unsubscribe:
2124 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2125err_destroy_stream_handle:
2126 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2127 return err;
2128}
2129module_init(vmci_transport_init);
2130
2131static void __exit vmci_transport_exit(void)
2132{
Jorgen Hansen4ef7ea92015-10-21 04:53:56 -07002133 cancel_work_sync(&vmci_transport_cleanup_work);
2134 vmci_transport_free_resources(&vmci_transport_cleanup_list);
2135
Andy Kingd021c342013-02-06 14:23:56 +00002136 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2137 if (vmci_datagram_destroy_handle(
2138 vmci_transport_stream_handle) != VMCI_SUCCESS)
2139 pr_err("Couldn't destroy datagram handle\n");
2140 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2141 }
2142
2143 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2144 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2145 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2146 }
2147
2148 vsock_core_exit();
2149}
2150module_exit(vmci_transport_exit);
2151
2152MODULE_AUTHOR("VMware, Inc.");
2153MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
Jorgen Hansen9c995cc2016-04-18 23:58:52 -07002154MODULE_VERSION("1.0.4.0-k");
Andy Kingd021c342013-02-06 14:23:56 +00002155MODULE_LICENSE("GPL v2");
2156MODULE_ALIAS("vmware_vsock");
2157MODULE_ALIAS_NETPROTO(PF_VSOCK);