2 * VMware vSockets Driver
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
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.
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
16 #include <linux/types.h>
17 #include <linux/bitops.h>
18 #include <linux/cred.h>
19 #include <linux/init.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>
37 #include <net/af_vsock.h>
39 #include "vmci_transport_notify.h"
41 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
42 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
43 static void vmci_transport_peer_detach_cb(u32 sub_id,
44 const struct vmci_event_data *ed,
46 static void vmci_transport_recv_pkt_work(struct work_struct *work);
47 static void vmci_transport_cleanup(struct work_struct *work);
48 static int vmci_transport_recv_listen(struct sock *sk,
49 struct vmci_transport_packet *pkt);
50 static int vmci_transport_recv_connecting_server(
53 struct vmci_transport_packet *pkt);
54 static int vmci_transport_recv_connecting_client(
56 struct vmci_transport_packet *pkt);
57 static int vmci_transport_recv_connecting_client_negotiate(
59 struct vmci_transport_packet *pkt);
60 static int vmci_transport_recv_connecting_client_invalid(
62 struct vmci_transport_packet *pkt);
63 static int vmci_transport_recv_connected(struct sock *sk,
64 struct vmci_transport_packet *pkt);
65 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
66 static u16 vmci_transport_new_proto_supported_versions(void);
67 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
70 struct vmci_transport_recv_pkt_info {
71 struct work_struct work;
73 struct vmci_transport_packet pkt;
76 static LIST_HEAD(vmci_transport_cleanup_list);
77 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
78 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
80 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
82 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
84 static int PROTOCOL_OVERRIDE = -1;
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
90 /* The default peer timeout indicates how long we will wait for a peer response
91 * to a control message.
93 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
97 /* Helper function to convert from a VMCI error code to a VSock error code. */
99 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
103 switch (vmci_error) {
104 case VMCI_ERROR_NO_MEM:
107 case VMCI_ERROR_DUPLICATE_ENTRY:
108 case VMCI_ERROR_ALREADY_EXISTS:
111 case VMCI_ERROR_NO_ACCESS:
114 case VMCI_ERROR_NO_RESOURCES:
117 case VMCI_ERROR_INVALID_RESOURCE:
120 case VMCI_ERROR_INVALID_ARGS:
125 return err > 0 ? -err : err;
128 static u32 vmci_transport_peer_rid(u32 peer_cid)
130 if (VMADDR_CID_HYPERVISOR == peer_cid)
131 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
133 return VMCI_TRANSPORT_PACKET_RID;
137 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
138 struct sockaddr_vm *src,
139 struct sockaddr_vm *dst,
143 struct vmci_transport_waiting_info *wait,
145 struct vmci_handle handle)
147 /* We register the stream control handler as an any cid handle so we
148 * must always send from a source address of VMADDR_CID_ANY
150 pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
151 VMCI_TRANSPORT_PACKET_RID);
152 pkt->dg.dst = vmci_make_handle(dst->svm_cid,
153 vmci_transport_peer_rid(dst->svm_cid));
154 pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
155 pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
157 pkt->src_port = src->svm_port;
158 pkt->dst_port = dst->svm_port;
159 memset(&pkt->proto, 0, sizeof(pkt->proto));
160 memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
163 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
167 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
168 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
172 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
173 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
174 pkt->u.handle = handle;
177 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
178 case VMCI_TRANSPORT_PACKET_TYPE_READ:
179 case VMCI_TRANSPORT_PACKET_TYPE_RST:
183 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
187 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
188 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
189 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
192 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
193 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
201 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
202 struct sockaddr_vm *local,
203 struct sockaddr_vm *remote)
205 vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
206 vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
210 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
211 struct sockaddr_vm *src,
212 struct sockaddr_vm *dst,
213 enum vmci_transport_packet_type type,
216 struct vmci_transport_waiting_info *wait,
218 struct vmci_handle handle,
223 vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
225 err = vmci_datagram_send(&pkt->dg);
226 if (convert_error && (err < 0))
227 return vmci_transport_error_to_vsock_error(err);
233 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
234 enum vmci_transport_packet_type type,
237 struct vmci_transport_waiting_info *wait,
238 struct vmci_handle handle)
240 struct vmci_transport_packet reply;
241 struct sockaddr_vm src, dst;
243 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
246 vmci_transport_packet_get_addresses(pkt, &src, &dst);
247 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
256 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
257 struct sockaddr_vm *dst,
258 enum vmci_transport_packet_type type,
261 struct vmci_transport_waiting_info *wait,
262 struct vmci_handle handle)
264 /* Note that it is safe to use a single packet across all CPUs since
265 * two tasklets of the same type are guaranteed to not ever run
266 * simultaneously. If that ever changes, or VMCI stops using tasklets,
267 * we can use per-cpu packets.
269 static struct vmci_transport_packet pkt;
271 return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
273 VSOCK_PROTO_INVALID, handle,
278 vmci_transport_send_control_pkt(struct sock *sk,
279 enum vmci_transport_packet_type type,
282 struct vmci_transport_waiting_info *wait,
284 struct vmci_handle handle)
286 struct vmci_transport_packet *pkt;
287 struct vsock_sock *vsk;
292 if (!vsock_addr_bound(&vsk->local_addr))
295 if (!vsock_addr_bound(&vsk->remote_addr))
298 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
302 err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
303 &vsk->remote_addr, type, size,
304 mode, wait, proto, handle,
311 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
312 struct sockaddr_vm *src,
313 struct vmci_transport_packet *pkt)
315 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
317 return vmci_transport_send_control_pkt_bh(
319 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
320 0, NULL, VMCI_INVALID_HANDLE);
323 static int vmci_transport_send_reset(struct sock *sk,
324 struct vmci_transport_packet *pkt)
326 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
328 return vmci_transport_send_control_pkt(sk,
329 VMCI_TRANSPORT_PACKET_TYPE_RST,
330 0, 0, NULL, VSOCK_PROTO_INVALID,
331 VMCI_INVALID_HANDLE);
334 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
336 return vmci_transport_send_control_pkt(
338 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
341 VMCI_INVALID_HANDLE);
344 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
347 return vmci_transport_send_control_pkt(
349 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
350 size, 0, NULL, version,
351 VMCI_INVALID_HANDLE);
354 static int vmci_transport_send_qp_offer(struct sock *sk,
355 struct vmci_handle handle)
357 return vmci_transport_send_control_pkt(
358 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
360 VSOCK_PROTO_INVALID, handle);
363 static int vmci_transport_send_attach(struct sock *sk,
364 struct vmci_handle handle)
366 return vmci_transport_send_control_pkt(
367 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
368 0, 0, NULL, VSOCK_PROTO_INVALID,
372 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
374 return vmci_transport_reply_control_pkt_fast(
376 VMCI_TRANSPORT_PACKET_TYPE_RST,
378 VMCI_INVALID_HANDLE);
381 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
382 struct sockaddr_vm *src)
384 return vmci_transport_send_control_pkt_bh(
386 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
387 0, 0, NULL, VMCI_INVALID_HANDLE);
390 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
391 struct sockaddr_vm *src)
393 return vmci_transport_send_control_pkt_bh(
395 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
396 0, NULL, VMCI_INVALID_HANDLE);
399 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
400 struct sockaddr_vm *src)
402 return vmci_transport_send_control_pkt_bh(
404 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
405 0, NULL, VMCI_INVALID_HANDLE);
408 int vmci_transport_send_wrote(struct sock *sk)
410 return vmci_transport_send_control_pkt(
411 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
412 0, NULL, VSOCK_PROTO_INVALID,
413 VMCI_INVALID_HANDLE);
416 int vmci_transport_send_read(struct sock *sk)
418 return vmci_transport_send_control_pkt(
419 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
420 0, NULL, VSOCK_PROTO_INVALID,
421 VMCI_INVALID_HANDLE);
424 int vmci_transport_send_waiting_write(struct sock *sk,
425 struct vmci_transport_waiting_info *wait)
427 return vmci_transport_send_control_pkt(
428 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
429 0, 0, wait, VSOCK_PROTO_INVALID,
430 VMCI_INVALID_HANDLE);
433 int vmci_transport_send_waiting_read(struct sock *sk,
434 struct vmci_transport_waiting_info *wait)
436 return vmci_transport_send_control_pkt(
437 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
438 0, 0, wait, VSOCK_PROTO_INVALID,
439 VMCI_INVALID_HANDLE);
442 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
444 return vmci_transport_send_control_pkt(
446 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
449 VMCI_INVALID_HANDLE);
452 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
454 return vmci_transport_send_control_pkt(sk,
455 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
458 VMCI_INVALID_HANDLE);
461 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
464 return vmci_transport_send_control_pkt(
465 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
466 size, 0, NULL, version,
467 VMCI_INVALID_HANDLE);
470 static struct sock *vmci_transport_get_pending(
471 struct sock *listener,
472 struct vmci_transport_packet *pkt)
474 struct vsock_sock *vlistener;
475 struct vsock_sock *vpending;
476 struct sock *pending;
477 struct sockaddr_vm src;
479 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
481 vlistener = vsock_sk(listener);
483 list_for_each_entry(vpending, &vlistener->pending_links,
485 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
486 pkt->dst_port == vpending->local_addr.svm_port) {
487 pending = sk_vsock(vpending);
499 static void vmci_transport_release_pending(struct sock *pending)
504 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
505 * trusted sockets 2) sockets from applications running as the same user as the
506 * VM (this is only true for the host side and only when using hosted products)
509 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
511 return vsock->trusted ||
512 vmci_is_context_owner(peer_cid, vsock->owner->uid);
515 /* We allow sending datagrams to and receiving datagrams from a restricted VM
516 * only if it is trusted as described in vmci_transport_is_trusted.
519 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
521 if (VMADDR_CID_HYPERVISOR == peer_cid)
524 if (vsock->cached_peer != peer_cid) {
525 vsock->cached_peer = peer_cid;
526 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
527 (vmci_context_get_priv_flags(peer_cid) &
528 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
529 vsock->cached_peer_allow_dgram = false;
531 vsock->cached_peer_allow_dgram = true;
535 return vsock->cached_peer_allow_dgram;
539 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
540 struct vmci_handle *handle,
543 u32 peer, u32 flags, bool trusted)
548 /* Try to allocate our queue pair as trusted. This will only
549 * work if vsock is running in the host.
552 err = vmci_qpair_alloc(qpair, handle, produce_size,
555 VMCI_PRIVILEGE_FLAG_TRUSTED);
556 if (err != VMCI_ERROR_NO_ACCESS)
561 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
562 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
565 pr_err("Could not attach to queue pair with %d\n",
567 err = vmci_transport_error_to_vsock_error(err);
574 vmci_transport_datagram_create_hnd(u32 resource_id,
576 vmci_datagram_recv_cb recv_cb,
578 struct vmci_handle *out_handle)
582 /* Try to allocate our datagram handler as trusted. This will only work
583 * if vsock is running in the host.
586 err = vmci_datagram_create_handle_priv(resource_id, flags,
587 VMCI_PRIVILEGE_FLAG_TRUSTED,
589 client_data, out_handle);
591 if (err == VMCI_ERROR_NO_ACCESS)
592 err = vmci_datagram_create_handle(resource_id, flags,
593 recv_cb, client_data,
599 /* This is invoked as part of a tasklet that's scheduled when the VMCI
600 * interrupt fires. This is run in bottom-half context and if it ever needs to
601 * sleep it should defer that work to a work queue.
604 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
609 struct vsock_sock *vsk;
611 sk = (struct sock *)data;
613 /* This handler is privileged when this module is running on the host.
614 * We will get datagrams from all endpoints (even VMs that are in a
615 * restricted context). If we get one from a restricted context then
616 * the destination socket must be trusted.
618 * NOTE: We access the socket struct without holding the lock here.
619 * This is ok because the field we are interested is never modified
620 * outside of the create and destruct socket functions.
623 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
624 return VMCI_ERROR_NO_ACCESS;
626 size = VMCI_DG_SIZE(dg);
628 /* Attach the packet to the socket's receive queue as an sk_buff. */
629 skb = alloc_skb(size, GFP_ATOMIC);
631 return VMCI_ERROR_NO_MEM;
633 /* sk_receive_skb() will do a sock_put(), so hold here. */
636 memcpy(skb->data, dg, size);
637 sk_receive_skb(sk, skb, 0);
642 static bool vmci_transport_stream_allow(u32 cid, u32 port)
644 static const u32 non_socket_contexts[] = {
649 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
651 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
652 if (cid == non_socket_contexts[i])
659 /* This is invoked as part of a tasklet that's scheduled when the VMCI
660 * interrupt fires. This is run in bottom-half context but it defers most of
661 * its work to the packet handling work queue.
664 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
667 struct sockaddr_vm dst;
668 struct sockaddr_vm src;
669 struct vmci_transport_packet *pkt;
670 struct vsock_sock *vsk;
676 bh_process_pkt = false;
678 /* Ignore incoming packets from contexts without sockets, or resources
679 * that aren't vsock implementations.
682 if (!vmci_transport_stream_allow(dg->src.context, -1)
683 || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
684 return VMCI_ERROR_NO_ACCESS;
686 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
687 /* Drop datagrams that do not contain full VSock packets. */
688 return VMCI_ERROR_INVALID_ARGS;
690 pkt = (struct vmci_transport_packet *)dg;
692 /* Find the socket that should handle this packet. First we look for a
693 * connected socket and if there is none we look for a socket bound to
694 * the destintation address.
696 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
697 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
699 sk = vsock_find_connected_socket(&src, &dst);
701 sk = vsock_find_bound_socket(&dst);
703 /* We could not find a socket for this specified
704 * address. If this packet is a RST, we just drop it.
705 * If it is another packet, we send a RST. Note that
706 * we do not send a RST reply to RSTs so that we do not
707 * continually send RSTs between two endpoints.
709 * Note that since this is a reply, dst is src and src
712 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
713 pr_err("unable to send reset\n");
715 err = VMCI_ERROR_NOT_FOUND;
720 /* If the received packet type is beyond all types known to this
721 * implementation, reply with an invalid message. Hopefully this will
722 * help when implementing backwards compatibility in the future.
724 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
725 vmci_transport_send_invalid_bh(&dst, &src);
726 err = VMCI_ERROR_INVALID_ARGS;
730 /* This handler is privileged when this module is running on the host.
731 * We will get datagram connect requests from all endpoints (even VMs
732 * that are in a restricted context). If we get one from a restricted
733 * context then the destination socket must be trusted.
735 * NOTE: We access the socket struct without holding the lock here.
736 * This is ok because the field we are interested is never modified
737 * outside of the create and destruct socket functions.
740 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
741 err = VMCI_ERROR_NO_ACCESS;
745 /* We do most everything in a work queue, but let's fast path the
746 * notification of reads and writes to help data transfer performance.
747 * We can only do this if there is no process context code executing
748 * for this socket since that may change the state.
752 if (!sock_owned_by_user(sk)) {
753 /* The local context ID may be out of date, update it. */
754 vsk->local_addr.svm_cid = dst.svm_cid;
756 if (sk->sk_state == SS_CONNECTED)
757 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
758 sk, pkt, true, &dst, &src,
764 if (!bh_process_pkt) {
765 struct vmci_transport_recv_pkt_info *recv_pkt_info;
767 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
768 if (!recv_pkt_info) {
769 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
770 pr_err("unable to send reset\n");
772 err = VMCI_ERROR_NO_MEM;
776 recv_pkt_info->sk = sk;
777 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
778 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
780 schedule_work(&recv_pkt_info->work);
781 /* Clear sk so that the reference count incremented by one of
782 * the Find functions above is not decremented below. We need
783 * that reference count for the packet handler we've scheduled
796 static void vmci_transport_handle_detach(struct sock *sk)
798 struct vsock_sock *vsk;
801 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
802 sock_set_flag(sk, SOCK_DONE);
804 /* On a detach the peer will not be sending or receiving
807 vsk->peer_shutdown = SHUTDOWN_MASK;
809 /* We should not be sending anymore since the peer won't be
810 * there to receive, but we can still receive if there is data
811 * left in our consume queue.
813 if (vsock_stream_has_data(vsk) <= 0) {
814 if (sk->sk_state == SS_CONNECTING) {
815 /* The peer may detach from a queue pair while
816 * we are still in the connecting state, i.e.,
817 * if the peer VM is killed after attaching to
818 * a queue pair, but before we complete the
819 * handshake. In that case, we treat the detach
820 * event like a reset.
823 sk->sk_state = SS_UNCONNECTED;
824 sk->sk_err = ECONNRESET;
825 sk->sk_error_report(sk);
828 sk->sk_state = SS_UNCONNECTED;
830 sk->sk_state_change(sk);
834 static void vmci_transport_peer_detach_cb(u32 sub_id,
835 const struct vmci_event_data *e_data,
838 struct vmci_transport *trans = client_data;
839 const struct vmci_event_payload_qp *e_payload;
841 e_payload = vmci_event_data_const_payload(e_data);
843 /* XXX This is lame, we should provide a way to lookup sockets by
846 if (vmci_handle_is_invalid(e_payload->handle) ||
847 vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
850 /* We don't ask for delayed CBs when we subscribe to this event (we
851 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
852 * guarantees in that case about what context we might be running in,
853 * so it could be BH or process, blockable or non-blockable. So we
854 * need to account for all possible contexts here.
856 spin_lock_bh(&trans->lock);
860 /* Apart from here, trans->lock is only grabbed as part of sk destruct,
861 * where trans->sk isn't locked.
863 bh_lock_sock(trans->sk);
865 vmci_transport_handle_detach(trans->sk);
867 bh_unlock_sock(trans->sk);
869 spin_unlock_bh(&trans->lock);
872 static void vmci_transport_qp_resumed_cb(u32 sub_id,
873 const struct vmci_event_data *e_data,
876 vsock_for_each_connected_socket(vmci_transport_handle_detach);
879 static void vmci_transport_recv_pkt_work(struct work_struct *work)
881 struct vmci_transport_recv_pkt_info *recv_pkt_info;
882 struct vmci_transport_packet *pkt;
886 container_of(work, struct vmci_transport_recv_pkt_info, work);
887 sk = recv_pkt_info->sk;
888 pkt = &recv_pkt_info->pkt;
892 /* The local context ID may be out of date. */
893 vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
895 switch (sk->sk_state) {
897 vmci_transport_recv_listen(sk, pkt);
900 /* Processing of pending connections for servers goes through
901 * the listening socket, so see vmci_transport_recv_listen()
904 vmci_transport_recv_connecting_client(sk, pkt);
907 vmci_transport_recv_connected(sk, pkt);
910 /* Because this function does not run in the same context as
911 * vmci_transport_recv_stream_cb it is possible that the
912 * socket has closed. We need to let the other side know or it
913 * could be sitting in a connect and hang forever. Send a
914 * reset to prevent that.
916 vmci_transport_send_reset(sk, pkt);
921 kfree(recv_pkt_info);
922 /* Release reference obtained in the stream callback when we fetched
923 * this socket out of the bound or connected list.
928 static int vmci_transport_recv_listen(struct sock *sk,
929 struct vmci_transport_packet *pkt)
931 struct sock *pending;
932 struct vsock_sock *vpending;
935 bool old_request = false;
936 bool old_pkt_proto = false;
940 /* Because we are in the listen state, we could be receiving a packet
941 * for ourself or any previous connection requests that we received.
942 * If it's the latter, we try to find a socket in our list of pending
943 * connections and, if we do, call the appropriate handler for the
944 * state that that socket is in. Otherwise we try to service the
945 * connection request.
947 pending = vmci_transport_get_pending(sk, pkt);
951 /* The local context ID may be out of date. */
952 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
954 switch (pending->sk_state) {
956 err = vmci_transport_recv_connecting_server(sk,
961 vmci_transport_send_reset(pending, pkt);
966 vsock_remove_pending(sk, pending);
968 release_sock(pending);
969 vmci_transport_release_pending(pending);
974 /* The listen state only accepts connection requests. Reply with a
975 * reset unless we received a reset.
978 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
979 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
980 vmci_transport_reply_reset(pkt);
984 if (pkt->u.size == 0) {
985 vmci_transport_reply_reset(pkt);
989 /* If this socket can't accommodate this connection request, we send a
990 * reset. Otherwise we create and initialize a child socket and reply
991 * with a connection negotiation.
993 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
994 vmci_transport_reply_reset(pkt);
995 return -ECONNREFUSED;
998 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1001 vmci_transport_send_reset(sk, pkt);
1005 vpending = vsock_sk(pending);
1007 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1009 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1012 /* If the proposed size fits within our min/max, accept it. Otherwise
1013 * propose our own size.
1015 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1016 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1017 qp_size = pkt->u.size;
1019 qp_size = vmci_trans(vpending)->queue_pair_size;
1022 /* Figure out if we are using old or new requests based on the
1023 * overrides pkt types sent by our peer.
1025 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1026 old_request = old_pkt_proto;
1028 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1030 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1031 old_request = false;
1036 /* Handle a REQUEST (or override) */
1037 u16 version = VSOCK_PROTO_INVALID;
1038 if (vmci_transport_proto_to_notify_struct(
1039 pending, &version, true))
1040 err = vmci_transport_send_negotiate(pending, qp_size);
1045 /* Handle a REQUEST2 (or override) */
1046 int proto_int = pkt->proto;
1048 u16 active_proto_version = 0;
1050 /* The list of possible protocols is the intersection of all
1051 * protocols the client supports ... plus all the protocols we
1054 proto_int &= vmci_transport_new_proto_supported_versions();
1056 /* We choose the highest possible protocol version and use that
1059 pos = fls(proto_int);
1061 active_proto_version = (1 << (pos - 1));
1062 if (vmci_transport_proto_to_notify_struct(
1063 pending, &active_proto_version, false))
1064 err = vmci_transport_send_negotiate2(pending,
1066 active_proto_version);
1076 vmci_transport_send_reset(sk, pkt);
1078 err = vmci_transport_error_to_vsock_error(err);
1082 vsock_add_pending(sk, pending);
1083 sk->sk_ack_backlog++;
1085 pending->sk_state = SS_CONNECTING;
1086 vmci_trans(vpending)->produce_size =
1087 vmci_trans(vpending)->consume_size = qp_size;
1088 vmci_trans(vpending)->queue_pair_size = qp_size;
1090 vmci_trans(vpending)->notify_ops->process_request(pending);
1092 /* We might never receive another message for this socket and it's not
1093 * connected to any process, so we have to ensure it gets cleaned up
1094 * ourself. Our delayed work function will take care of that. Note
1095 * that we do not ever cancel this function since we have few
1096 * guarantees about its state when calling cancel_delayed_work().
1097 * Instead we hold a reference on the socket for that function and make
1098 * it capable of handling cases where it needs to do nothing but
1099 * release that reference.
1101 vpending->listener = sk;
1104 INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1105 schedule_delayed_work(&vpending->dwork, HZ);
1112 vmci_transport_recv_connecting_server(struct sock *listener,
1113 struct sock *pending,
1114 struct vmci_transport_packet *pkt)
1116 struct vsock_sock *vpending;
1117 struct vmci_handle handle;
1118 struct vmci_qp *qpair;
1125 vpending = vsock_sk(pending);
1126 detach_sub_id = VMCI_INVALID_ID;
1128 switch (pkt->type) {
1129 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1130 if (vmci_handle_is_invalid(pkt->u.handle)) {
1131 vmci_transport_send_reset(pending, pkt);
1138 /* Close and cleanup the connection. */
1139 vmci_transport_send_reset(pending, pkt);
1141 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1145 /* In order to complete the connection we need to attach to the offered
1146 * queue pair and send an attach notification. We also subscribe to the
1147 * detach event so we know when our peer goes away, and we do that
1148 * before attaching so we don't miss an event. If all this succeeds,
1149 * we update our state and wakeup anything waiting in accept() for a
1153 /* We don't care about attach since we ensure the other side has
1154 * attached by specifying the ATTACH_ONLY flag below.
1156 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1157 vmci_transport_peer_detach_cb,
1158 vmci_trans(vpending), &detach_sub_id);
1159 if (err < VMCI_SUCCESS) {
1160 vmci_transport_send_reset(pending, pkt);
1161 err = vmci_transport_error_to_vsock_error(err);
1166 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1168 /* Now attach to the queue pair the client created. */
1169 handle = pkt->u.handle;
1171 /* vpending->local_addr always has a context id so we do not need to
1172 * worry about VMADDR_CID_ANY in this case.
1175 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1176 flags = VMCI_QPFLAG_ATTACH_ONLY;
1177 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1179 err = vmci_transport_queue_pair_alloc(
1182 vmci_trans(vpending)->produce_size,
1183 vmci_trans(vpending)->consume_size,
1184 pkt->dg.src.context,
1186 vmci_transport_is_trusted(
1188 vpending->remote_addr.svm_cid));
1190 vmci_transport_send_reset(pending, pkt);
1195 vmci_trans(vpending)->qp_handle = handle;
1196 vmci_trans(vpending)->qpair = qpair;
1198 /* When we send the attach message, we must be ready to handle incoming
1199 * control messages on the newly connected socket. So we move the
1200 * pending socket to the connected state before sending the attach
1201 * message. Otherwise, an incoming packet triggered by the attach being
1202 * received by the peer may be processed concurrently with what happens
1203 * below after sending the attach message, and that incoming packet
1204 * will find the listening socket instead of the (currently) pending
1205 * socket. Note that enqueueing the socket increments the reference
1206 * count, so even if a reset comes before the connection is accepted,
1207 * the socket will be valid until it is removed from the queue.
1209 * If we fail sending the attach below, we remove the socket from the
1210 * connected list and move the socket to SS_UNCONNECTED before
1211 * releasing the lock, so a pending slow path processing of an incoming
1212 * packet will not see the socket in the connected state in that case.
1214 pending->sk_state = SS_CONNECTED;
1216 vsock_insert_connected(vpending);
1218 /* Notify our peer of our attach. */
1219 err = vmci_transport_send_attach(pending, handle);
1221 vsock_remove_connected(vpending);
1222 pr_err("Could not send attach\n");
1223 vmci_transport_send_reset(pending, pkt);
1224 err = vmci_transport_error_to_vsock_error(err);
1229 /* We have a connection. Move the now connected socket from the
1230 * listener's pending list to the accept queue so callers of accept()
1233 vsock_remove_pending(listener, pending);
1234 vsock_enqueue_accept(listener, pending);
1236 /* Callers of accept() will be be waiting on the listening socket, not
1237 * the pending socket.
1239 listener->sk_state_change(listener);
1244 pending->sk_err = skerr;
1245 pending->sk_state = SS_UNCONNECTED;
1246 /* As long as we drop our reference, all necessary cleanup will handle
1247 * when the cleanup function drops its reference and our destruct
1248 * implementation is called. Note that since the listen handler will
1249 * remove pending from the pending list upon our failure, the cleanup
1250 * function won't drop the additional reference, which is why we do it
1259 vmci_transport_recv_connecting_client(struct sock *sk,
1260 struct vmci_transport_packet *pkt)
1262 struct vsock_sock *vsk;
1268 switch (pkt->type) {
1269 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1270 if (vmci_handle_is_invalid(pkt->u.handle) ||
1271 !vmci_handle_is_equal(pkt->u.handle,
1272 vmci_trans(vsk)->qp_handle)) {
1278 /* Signify the socket is connected and wakeup the waiter in
1279 * connect(). Also place the socket in the connected table for
1280 * accounting (it can already be found since it's in the bound
1283 sk->sk_state = SS_CONNECTED;
1284 sk->sk_socket->state = SS_CONNECTED;
1285 vsock_insert_connected(vsk);
1286 sk->sk_state_change(sk);
1289 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1290 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1291 if (pkt->u.size == 0
1292 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1293 || pkt->src_port != vsk->remote_addr.svm_port
1294 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1295 || vmci_trans(vsk)->qpair
1296 || vmci_trans(vsk)->produce_size != 0
1297 || vmci_trans(vsk)->consume_size != 0
1298 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1305 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1312 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1313 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1320 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1321 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1322 * continue processing here after they sent an INVALID packet.
1323 * This meant that we got a RST after the INVALID. We ignore a
1324 * RST after an INVALID. The common code doesn't send the RST
1325 * ... so we can hang if an old version of the common code
1326 * fails between getting a REQUEST and sending an OFFER back.
1327 * Not much we can do about it... except hope that it doesn't
1330 if (vsk->ignore_connecting_rst) {
1331 vsk->ignore_connecting_rst = false;
1340 /* Close and cleanup the connection. */
1349 vmci_transport_send_reset(sk, pkt);
1351 sk->sk_state = SS_UNCONNECTED;
1353 sk->sk_error_report(sk);
1357 static int vmci_transport_recv_connecting_client_negotiate(
1359 struct vmci_transport_packet *pkt)
1362 struct vsock_sock *vsk;
1363 struct vmci_handle handle;
1364 struct vmci_qp *qpair;
1368 bool old_proto = true;
1373 handle = VMCI_INVALID_HANDLE;
1374 detach_sub_id = VMCI_INVALID_ID;
1376 /* If we have gotten here then we should be past the point where old
1377 * linux vsock could have sent the bogus rst.
1379 vsk->sent_request = false;
1380 vsk->ignore_connecting_rst = false;
1382 /* Verify that we're OK with the proposed queue pair size */
1383 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1384 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1389 /* At this point we know the CID the peer is using to talk to us. */
1391 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1392 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1394 /* Setup the notify ops to be the highest supported version that both
1395 * the server and the client support.
1398 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1399 old_proto = old_pkt_proto;
1401 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1403 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1409 version = VSOCK_PROTO_INVALID;
1411 version = pkt->proto;
1413 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1418 /* Subscribe to detach events first.
1420 * XXX We attach once for each queue pair created for now so it is easy
1421 * to find the socket (it's provided), but later we should only
1422 * subscribe once and add a way to lookup sockets by queue pair handle.
1424 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1425 vmci_transport_peer_detach_cb,
1426 vmci_trans(vsk), &detach_sub_id);
1427 if (err < VMCI_SUCCESS) {
1428 err = vmci_transport_error_to_vsock_error(err);
1432 /* Make VMCI select the handle for us. */
1433 handle = VMCI_INVALID_HANDLE;
1434 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1435 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1437 err = vmci_transport_queue_pair_alloc(&qpair,
1441 vsk->remote_addr.svm_cid,
1443 vmci_transport_is_trusted(
1446 remote_addr.svm_cid));
1450 err = vmci_transport_send_qp_offer(sk, handle);
1452 err = vmci_transport_error_to_vsock_error(err);
1456 vmci_trans(vsk)->qp_handle = handle;
1457 vmci_trans(vsk)->qpair = qpair;
1459 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1462 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1464 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1469 if (detach_sub_id != VMCI_INVALID_ID)
1470 vmci_event_unsubscribe(detach_sub_id);
1472 if (!vmci_handle_is_invalid(handle))
1473 vmci_qpair_detach(&qpair);
1479 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1480 struct vmci_transport_packet *pkt)
1483 struct vsock_sock *vsk = vsock_sk(sk);
1485 if (vsk->sent_request) {
1486 vsk->sent_request = false;
1487 vsk->ignore_connecting_rst = true;
1489 err = vmci_transport_send_conn_request(
1490 sk, vmci_trans(vsk)->queue_pair_size);
1492 err = vmci_transport_error_to_vsock_error(err);
1501 static int vmci_transport_recv_connected(struct sock *sk,
1502 struct vmci_transport_packet *pkt)
1504 struct vsock_sock *vsk;
1505 bool pkt_processed = false;
1507 /* In cases where we are closing the connection, it's sufficient to
1508 * mark the state change (and maybe error) and wake up any waiting
1509 * threads. Since this is a connected socket, it's owned by a user
1510 * process and will be cleaned up when the failure is passed back on
1511 * the current or next system call. Our system call implementations
1512 * must therefore check for error and state changes on entry and when
1515 switch (pkt->type) {
1516 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1520 vsk->peer_shutdown |= pkt->u.mode;
1521 sk->sk_state_change(sk);
1525 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1527 /* It is possible that we sent our peer a message (e.g a
1528 * WAITING_READ) right before we got notified that the peer had
1529 * detached. If that happens then we can get a RST pkt back
1530 * from our peer even though there is data available for us to
1531 * read. In that case, don't shutdown the socket completely but
1532 * instead allow the local client to finish reading data off
1533 * the queuepair. Always treat a RST pkt in connected mode like
1536 sock_set_flag(sk, SOCK_DONE);
1537 vsk->peer_shutdown = SHUTDOWN_MASK;
1538 if (vsock_stream_has_data(vsk) <= 0)
1539 sk->sk_state = SS_DISCONNECTING;
1541 sk->sk_state_change(sk);
1546 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1547 sk, pkt, false, NULL, NULL,
1558 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1559 struct vsock_sock *psk)
1561 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1565 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1566 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1567 vmci_trans(vsk)->qpair = NULL;
1568 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1569 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1570 vmci_trans(vsk)->notify_ops = NULL;
1571 INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1572 vmci_trans(vsk)->sk = &vsk->sk;
1573 vmci_trans(vsk)->lock = __SPIN_LOCK_UNLOCKED(vmci_trans(vsk)->lock);
1575 vmci_trans(vsk)->queue_pair_size =
1576 vmci_trans(psk)->queue_pair_size;
1577 vmci_trans(vsk)->queue_pair_min_size =
1578 vmci_trans(psk)->queue_pair_min_size;
1579 vmci_trans(vsk)->queue_pair_max_size =
1580 vmci_trans(psk)->queue_pair_max_size;
1582 vmci_trans(vsk)->queue_pair_size =
1583 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1584 vmci_trans(vsk)->queue_pair_min_size =
1585 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1586 vmci_trans(vsk)->queue_pair_max_size =
1587 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1593 static void vmci_transport_free_resources(struct list_head *transport_list)
1595 while (!list_empty(transport_list)) {
1596 struct vmci_transport *transport =
1597 list_first_entry(transport_list, struct vmci_transport,
1599 list_del(&transport->elem);
1601 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1602 vmci_event_unsubscribe(transport->detach_sub_id);
1603 transport->detach_sub_id = VMCI_INVALID_ID;
1606 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1607 vmci_qpair_detach(&transport->qpair);
1608 transport->qp_handle = VMCI_INVALID_HANDLE;
1609 transport->produce_size = 0;
1610 transport->consume_size = 0;
1617 static void vmci_transport_cleanup(struct work_struct *work)
1621 spin_lock_bh(&vmci_transport_cleanup_lock);
1622 list_replace_init(&vmci_transport_cleanup_list, &pending);
1623 spin_unlock_bh(&vmci_transport_cleanup_lock);
1624 vmci_transport_free_resources(&pending);
1627 static void vmci_transport_destruct(struct vsock_sock *vsk)
1629 /* Ensure that the detach callback doesn't use the sk/vsk
1630 * we are about to destruct.
1632 spin_lock_bh(&vmci_trans(vsk)->lock);
1633 vmci_trans(vsk)->sk = NULL;
1634 spin_unlock_bh(&vmci_trans(vsk)->lock);
1636 if (vmci_trans(vsk)->notify_ops)
1637 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1639 spin_lock_bh(&vmci_transport_cleanup_lock);
1640 list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1641 spin_unlock_bh(&vmci_transport_cleanup_lock);
1642 schedule_work(&vmci_transport_cleanup_work);
1647 static void vmci_transport_release(struct vsock_sock *vsk)
1649 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1650 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1651 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1655 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1656 struct sockaddr_vm *addr)
1662 /* VMCI will select a resource ID for us if we provide
1665 port = addr->svm_port == VMADDR_PORT_ANY ?
1666 VMCI_INVALID_ID : addr->svm_port;
1668 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1671 flags = addr->svm_cid == VMADDR_CID_ANY ?
1672 VMCI_FLAG_ANYCID_DG_HND : 0;
1674 err = vmci_transport_datagram_create_hnd(port, flags,
1675 vmci_transport_recv_dgram_cb,
1677 &vmci_trans(vsk)->dg_handle);
1678 if (err < VMCI_SUCCESS)
1679 return vmci_transport_error_to_vsock_error(err);
1680 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1681 vmci_trans(vsk)->dg_handle.resource);
1686 static int vmci_transport_dgram_enqueue(
1687 struct vsock_sock *vsk,
1688 struct sockaddr_vm *remote_addr,
1693 struct vmci_datagram *dg;
1695 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1698 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1701 /* Allocate a buffer for the user's message and our packet header. */
1702 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1706 memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1708 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1709 remote_addr->svm_port);
1710 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1711 vsk->local_addr.svm_port);
1712 dg->payload_size = len;
1714 err = vmci_datagram_send(dg);
1717 return vmci_transport_error_to_vsock_error(err);
1719 return err - sizeof(*dg);
1722 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1723 struct msghdr *msg, size_t len,
1728 struct vmci_datagram *dg;
1730 struct sk_buff *skb;
1732 noblock = flags & MSG_DONTWAIT;
1734 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1737 /* Retrieve the head sk_buff from the socket's receive queue. */
1739 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1746 dg = (struct vmci_datagram *)skb->data;
1748 /* err is 0, meaning we read zero bytes. */
1751 payload_len = dg->payload_size;
1752 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1753 if (payload_len != skb->len - sizeof(*dg)) {
1758 if (payload_len > len) {
1760 msg->msg_flags |= MSG_TRUNC;
1763 /* Place the datagram payload in the user's iovec. */
1764 err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1768 if (msg->msg_name) {
1769 /* Provide the address of the sender. */
1770 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1771 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1772 msg->msg_namelen = sizeof(*vm_addr);
1777 skb_free_datagram(&vsk->sk, skb);
1781 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1783 if (cid == VMADDR_CID_HYPERVISOR) {
1784 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1785 * state and are allowed.
1787 return port == VMCI_UNITY_PBRPC_REGISTER;
1793 static int vmci_transport_connect(struct vsock_sock *vsk)
1796 bool old_pkt_proto = false;
1797 struct sock *sk = &vsk->sk;
1799 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1801 err = vmci_transport_send_conn_request(
1802 sk, vmci_trans(vsk)->queue_pair_size);
1804 sk->sk_state = SS_UNCONNECTED;
1808 int supported_proto_versions =
1809 vmci_transport_new_proto_supported_versions();
1810 err = vmci_transport_send_conn_request2(
1811 sk, vmci_trans(vsk)->queue_pair_size,
1812 supported_proto_versions);
1814 sk->sk_state = SS_UNCONNECTED;
1818 vsk->sent_request = true;
1824 static ssize_t vmci_transport_stream_dequeue(
1825 struct vsock_sock *vsk,
1830 if (flags & MSG_PEEK)
1831 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1833 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1836 static ssize_t vmci_transport_stream_enqueue(
1837 struct vsock_sock *vsk,
1841 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1844 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1846 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1849 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1851 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1854 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1856 return vmci_trans(vsk)->consume_size;
1859 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1861 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1864 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1866 return vmci_trans(vsk)->queue_pair_size;
1869 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1871 return vmci_trans(vsk)->queue_pair_min_size;
1874 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1876 return vmci_trans(vsk)->queue_pair_max_size;
1879 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1881 if (val < vmci_trans(vsk)->queue_pair_min_size)
1882 vmci_trans(vsk)->queue_pair_min_size = val;
1883 if (val > vmci_trans(vsk)->queue_pair_max_size)
1884 vmci_trans(vsk)->queue_pair_max_size = val;
1885 vmci_trans(vsk)->queue_pair_size = val;
1888 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1891 if (val > vmci_trans(vsk)->queue_pair_size)
1892 vmci_trans(vsk)->queue_pair_size = val;
1893 vmci_trans(vsk)->queue_pair_min_size = val;
1896 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1899 if (val < vmci_trans(vsk)->queue_pair_size)
1900 vmci_trans(vsk)->queue_pair_size = val;
1901 vmci_trans(vsk)->queue_pair_max_size = val;
1904 static int vmci_transport_notify_poll_in(
1905 struct vsock_sock *vsk,
1907 bool *data_ready_now)
1909 return vmci_trans(vsk)->notify_ops->poll_in(
1910 &vsk->sk, target, data_ready_now);
1913 static int vmci_transport_notify_poll_out(
1914 struct vsock_sock *vsk,
1916 bool *space_available_now)
1918 return vmci_trans(vsk)->notify_ops->poll_out(
1919 &vsk->sk, target, space_available_now);
1922 static int vmci_transport_notify_recv_init(
1923 struct vsock_sock *vsk,
1925 struct vsock_transport_recv_notify_data *data)
1927 return vmci_trans(vsk)->notify_ops->recv_init(
1929 (struct vmci_transport_recv_notify_data *)data);
1932 static int vmci_transport_notify_recv_pre_block(
1933 struct vsock_sock *vsk,
1935 struct vsock_transport_recv_notify_data *data)
1937 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1939 (struct vmci_transport_recv_notify_data *)data);
1942 static int vmci_transport_notify_recv_pre_dequeue(
1943 struct vsock_sock *vsk,
1945 struct vsock_transport_recv_notify_data *data)
1947 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1949 (struct vmci_transport_recv_notify_data *)data);
1952 static int vmci_transport_notify_recv_post_dequeue(
1953 struct vsock_sock *vsk,
1957 struct vsock_transport_recv_notify_data *data)
1959 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1960 &vsk->sk, target, copied, data_read,
1961 (struct vmci_transport_recv_notify_data *)data);
1964 static int vmci_transport_notify_send_init(
1965 struct vsock_sock *vsk,
1966 struct vsock_transport_send_notify_data *data)
1968 return vmci_trans(vsk)->notify_ops->send_init(
1970 (struct vmci_transport_send_notify_data *)data);
1973 static int vmci_transport_notify_send_pre_block(
1974 struct vsock_sock *vsk,
1975 struct vsock_transport_send_notify_data *data)
1977 return vmci_trans(vsk)->notify_ops->send_pre_block(
1979 (struct vmci_transport_send_notify_data *)data);
1982 static int vmci_transport_notify_send_pre_enqueue(
1983 struct vsock_sock *vsk,
1984 struct vsock_transport_send_notify_data *data)
1986 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1988 (struct vmci_transport_send_notify_data *)data);
1991 static int vmci_transport_notify_send_post_enqueue(
1992 struct vsock_sock *vsk,
1994 struct vsock_transport_send_notify_data *data)
1996 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1998 (struct vmci_transport_send_notify_data *)data);
2001 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2003 if (PROTOCOL_OVERRIDE != -1) {
2004 if (PROTOCOL_OVERRIDE == 0)
2005 *old_pkt_proto = true;
2007 *old_pkt_proto = false;
2009 pr_info("Proto override in use\n");
2016 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2020 struct vsock_sock *vsk = vsock_sk(sk);
2022 if (old_pkt_proto) {
2023 if (*proto != VSOCK_PROTO_INVALID) {
2024 pr_err("Can't set both an old and new protocol\n");
2027 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2032 case VSOCK_PROTO_PKT_ON_NOTIFY:
2033 vmci_trans(vsk)->notify_ops =
2034 &vmci_transport_notify_pkt_q_state_ops;
2037 pr_err("Unknown notify protocol version\n");
2042 vmci_trans(vsk)->notify_ops->socket_init(sk);
2046 static u16 vmci_transport_new_proto_supported_versions(void)
2048 if (PROTOCOL_OVERRIDE != -1)
2049 return PROTOCOL_OVERRIDE;
2051 return VSOCK_PROTO_ALL_SUPPORTED;
2054 static u32 vmci_transport_get_local_cid(void)
2056 return vmci_get_context_id();
2059 static struct vsock_transport vmci_transport = {
2060 .init = vmci_transport_socket_init,
2061 .destruct = vmci_transport_destruct,
2062 .release = vmci_transport_release,
2063 .connect = vmci_transport_connect,
2064 .dgram_bind = vmci_transport_dgram_bind,
2065 .dgram_dequeue = vmci_transport_dgram_dequeue,
2066 .dgram_enqueue = vmci_transport_dgram_enqueue,
2067 .dgram_allow = vmci_transport_dgram_allow,
2068 .stream_dequeue = vmci_transport_stream_dequeue,
2069 .stream_enqueue = vmci_transport_stream_enqueue,
2070 .stream_has_data = vmci_transport_stream_has_data,
2071 .stream_has_space = vmci_transport_stream_has_space,
2072 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2073 .stream_is_active = vmci_transport_stream_is_active,
2074 .stream_allow = vmci_transport_stream_allow,
2075 .notify_poll_in = vmci_transport_notify_poll_in,
2076 .notify_poll_out = vmci_transport_notify_poll_out,
2077 .notify_recv_init = vmci_transport_notify_recv_init,
2078 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2079 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2080 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2081 .notify_send_init = vmci_transport_notify_send_init,
2082 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2083 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2084 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2085 .shutdown = vmci_transport_shutdown,
2086 .set_buffer_size = vmci_transport_set_buffer_size,
2087 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2088 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2089 .get_buffer_size = vmci_transport_get_buffer_size,
2090 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2091 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2092 .get_local_cid = vmci_transport_get_local_cid,
2095 static int __init vmci_transport_init(void)
2099 /* Create the datagram handle that we will use to send and receive all
2100 * VSocket control messages for this context.
2102 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2103 VMCI_FLAG_ANYCID_DG_HND,
2104 vmci_transport_recv_stream_cb,
2106 &vmci_transport_stream_handle);
2107 if (err < VMCI_SUCCESS) {
2108 pr_err("Unable to create datagram handle. (%d)\n", err);
2109 return vmci_transport_error_to_vsock_error(err);
2112 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2113 vmci_transport_qp_resumed_cb,
2114 NULL, &vmci_transport_qp_resumed_sub_id);
2115 if (err < VMCI_SUCCESS) {
2116 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2117 err = vmci_transport_error_to_vsock_error(err);
2118 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2119 goto err_destroy_stream_handle;
2122 err = vsock_core_init(&vmci_transport);
2124 goto err_unsubscribe;
2129 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2130 err_destroy_stream_handle:
2131 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2134 module_init(vmci_transport_init);
2136 static void __exit vmci_transport_exit(void)
2138 cancel_work_sync(&vmci_transport_cleanup_work);
2139 vmci_transport_free_resources(&vmci_transport_cleanup_list);
2141 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2142 if (vmci_datagram_destroy_handle(
2143 vmci_transport_stream_handle) != VMCI_SUCCESS)
2144 pr_err("Couldn't destroy datagram handle\n");
2145 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2148 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2149 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2150 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2155 module_exit(vmci_transport_exit);
2157 MODULE_AUTHOR("VMware, Inc.");
2158 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2159 MODULE_VERSION("1.0.2.0-k");
2160 MODULE_LICENSE("GPL v2");
2161 MODULE_ALIAS("vmware_vsock");
2162 MODULE_ALIAS_NETPROTO(PF_VSOCK);