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/module.h>
25 #include <linux/mutex.h>
26 #include <linux/net.h>
27 #include <linux/poll.h>
28 #include <linux/skbuff.h>
29 #include <linux/smp.h>
30 #include <linux/socket.h>
31 #include <linux/stddef.h>
32 #include <linux/unistd.h>
33 #include <linux/wait.h>
34 #include <linux/workqueue.h>
36 #include <net/af_vsock.h>
38 #include "vmci_transport_notify.h"
40 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
41 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
42 static void vmci_transport_peer_detach_cb(u32 sub_id,
43 const struct vmci_event_data *ed,
45 static void vmci_transport_recv_pkt_work(struct work_struct *work);
46 static void vmci_transport_cleanup(struct work_struct *work);
47 static int vmci_transport_recv_listen(struct sock *sk,
48 struct vmci_transport_packet *pkt);
49 static int vmci_transport_recv_connecting_server(
52 struct vmci_transport_packet *pkt);
53 static int vmci_transport_recv_connecting_client(
55 struct vmci_transport_packet *pkt);
56 static int vmci_transport_recv_connecting_client_negotiate(
58 struct vmci_transport_packet *pkt);
59 static int vmci_transport_recv_connecting_client_invalid(
61 struct vmci_transport_packet *pkt);
62 static int vmci_transport_recv_connected(struct sock *sk,
63 struct vmci_transport_packet *pkt);
64 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
65 static u16 vmci_transport_new_proto_supported_versions(void);
66 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
69 struct vmci_transport_recv_pkt_info {
70 struct work_struct work;
72 struct vmci_transport_packet pkt;
75 static LIST_HEAD(vmci_transport_cleanup_list);
76 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
77 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
79 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
81 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
83 static int PROTOCOL_OVERRIDE = -1;
85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN 128
86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE 262144
87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX 262144
89 /* The default peer timeout indicates how long we will wait for a peer response
90 * to a control message.
92 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
94 /* Helper function to convert from a VMCI error code to a VSock error code. */
96 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
99 case VMCI_ERROR_NO_MEM:
101 case VMCI_ERROR_DUPLICATE_ENTRY:
102 case VMCI_ERROR_ALREADY_EXISTS:
104 case VMCI_ERROR_NO_ACCESS:
106 case VMCI_ERROR_NO_RESOURCES:
108 case VMCI_ERROR_INVALID_RESOURCE:
109 return -EHOSTUNREACH;
110 case VMCI_ERROR_INVALID_ARGS:
117 static u32 vmci_transport_peer_rid(u32 peer_cid)
119 if (VMADDR_CID_HYPERVISOR == peer_cid)
120 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
122 return VMCI_TRANSPORT_PACKET_RID;
126 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
127 struct sockaddr_vm *src,
128 struct sockaddr_vm *dst,
132 struct vmci_transport_waiting_info *wait,
134 struct vmci_handle handle)
136 /* We register the stream control handler as an any cid handle so we
137 * must always send from a source address of VMADDR_CID_ANY
139 pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
140 VMCI_TRANSPORT_PACKET_RID);
141 pkt->dg.dst = vmci_make_handle(dst->svm_cid,
142 vmci_transport_peer_rid(dst->svm_cid));
143 pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
144 pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
146 pkt->src_port = src->svm_port;
147 pkt->dst_port = dst->svm_port;
148 memset(&pkt->proto, 0, sizeof(pkt->proto));
149 memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
152 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
156 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
157 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
161 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
162 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
163 pkt->u.handle = handle;
166 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
167 case VMCI_TRANSPORT_PACKET_TYPE_READ:
168 case VMCI_TRANSPORT_PACKET_TYPE_RST:
172 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
176 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
177 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
178 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
181 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
182 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
190 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
191 struct sockaddr_vm *local,
192 struct sockaddr_vm *remote)
194 vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
195 vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
199 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
200 struct sockaddr_vm *src,
201 struct sockaddr_vm *dst,
202 enum vmci_transport_packet_type type,
205 struct vmci_transport_waiting_info *wait,
207 struct vmci_handle handle,
212 vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
214 err = vmci_datagram_send(&pkt->dg);
215 if (convert_error && (err < 0))
216 return vmci_transport_error_to_vsock_error(err);
222 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
223 enum vmci_transport_packet_type type,
226 struct vmci_transport_waiting_info *wait,
227 struct vmci_handle handle)
229 struct vmci_transport_packet reply;
230 struct sockaddr_vm src, dst;
232 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
235 vmci_transport_packet_get_addresses(pkt, &src, &dst);
236 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
245 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
246 struct sockaddr_vm *dst,
247 enum vmci_transport_packet_type type,
250 struct vmci_transport_waiting_info *wait,
251 struct vmci_handle handle)
253 /* Note that it is safe to use a single packet across all CPUs since
254 * two tasklets of the same type are guaranteed to not ever run
255 * simultaneously. If that ever changes, or VMCI stops using tasklets,
256 * we can use per-cpu packets.
258 static struct vmci_transport_packet pkt;
260 return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
262 VSOCK_PROTO_INVALID, handle,
267 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
268 struct sockaddr_vm *dst,
269 enum vmci_transport_packet_type type,
272 struct vmci_transport_waiting_info *wait,
274 struct vmci_handle handle)
276 struct vmci_transport_packet *pkt;
279 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
283 err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
284 mode, wait, proto, handle,
292 vmci_transport_send_control_pkt(struct sock *sk,
293 enum vmci_transport_packet_type type,
296 struct vmci_transport_waiting_info *wait,
298 struct vmci_handle handle)
300 struct vsock_sock *vsk;
304 if (!vsock_addr_bound(&vsk->local_addr))
307 if (!vsock_addr_bound(&vsk->remote_addr))
310 return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
313 wait, proto, handle);
316 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
317 struct sockaddr_vm *src,
318 struct vmci_transport_packet *pkt)
320 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
322 return vmci_transport_send_control_pkt_bh(
324 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
325 0, NULL, VMCI_INVALID_HANDLE);
328 static int vmci_transport_send_reset(struct sock *sk,
329 struct vmci_transport_packet *pkt)
331 struct sockaddr_vm *dst_ptr;
332 struct sockaddr_vm dst;
333 struct vsock_sock *vsk;
335 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
340 if (!vsock_addr_bound(&vsk->local_addr))
343 if (vsock_addr_bound(&vsk->remote_addr)) {
344 dst_ptr = &vsk->remote_addr;
346 vsock_addr_init(&dst, pkt->dg.src.context,
350 return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
351 VMCI_TRANSPORT_PACKET_TYPE_RST,
352 0, 0, NULL, VSOCK_PROTO_INVALID,
353 VMCI_INVALID_HANDLE);
356 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
358 return vmci_transport_send_control_pkt(
360 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
363 VMCI_INVALID_HANDLE);
366 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
369 return vmci_transport_send_control_pkt(
371 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
372 size, 0, NULL, version,
373 VMCI_INVALID_HANDLE);
376 static int vmci_transport_send_qp_offer(struct sock *sk,
377 struct vmci_handle handle)
379 return vmci_transport_send_control_pkt(
380 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
382 VSOCK_PROTO_INVALID, handle);
385 static int vmci_transport_send_attach(struct sock *sk,
386 struct vmci_handle handle)
388 return vmci_transport_send_control_pkt(
389 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
390 0, 0, NULL, VSOCK_PROTO_INVALID,
394 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
396 return vmci_transport_reply_control_pkt_fast(
398 VMCI_TRANSPORT_PACKET_TYPE_RST,
400 VMCI_INVALID_HANDLE);
403 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
404 struct sockaddr_vm *src)
406 return vmci_transport_send_control_pkt_bh(
408 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
409 0, 0, NULL, VMCI_INVALID_HANDLE);
412 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
413 struct sockaddr_vm *src)
415 return vmci_transport_send_control_pkt_bh(
417 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
418 0, NULL, VMCI_INVALID_HANDLE);
421 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
422 struct sockaddr_vm *src)
424 return vmci_transport_send_control_pkt_bh(
426 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
427 0, NULL, VMCI_INVALID_HANDLE);
430 int vmci_transport_send_wrote(struct sock *sk)
432 return vmci_transport_send_control_pkt(
433 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
434 0, NULL, VSOCK_PROTO_INVALID,
435 VMCI_INVALID_HANDLE);
438 int vmci_transport_send_read(struct sock *sk)
440 return vmci_transport_send_control_pkt(
441 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
442 0, NULL, VSOCK_PROTO_INVALID,
443 VMCI_INVALID_HANDLE);
446 int vmci_transport_send_waiting_write(struct sock *sk,
447 struct vmci_transport_waiting_info *wait)
449 return vmci_transport_send_control_pkt(
450 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
451 0, 0, wait, VSOCK_PROTO_INVALID,
452 VMCI_INVALID_HANDLE);
455 int vmci_transport_send_waiting_read(struct sock *sk,
456 struct vmci_transport_waiting_info *wait)
458 return vmci_transport_send_control_pkt(
459 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
460 0, 0, wait, VSOCK_PROTO_INVALID,
461 VMCI_INVALID_HANDLE);
464 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
466 return vmci_transport_send_control_pkt(
468 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
471 VMCI_INVALID_HANDLE);
474 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
476 return vmci_transport_send_control_pkt(sk,
477 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
480 VMCI_INVALID_HANDLE);
483 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
486 return vmci_transport_send_control_pkt(
487 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
488 size, 0, NULL, version,
489 VMCI_INVALID_HANDLE);
492 static struct sock *vmci_transport_get_pending(
493 struct sock *listener,
494 struct vmci_transport_packet *pkt)
496 struct vsock_sock *vlistener;
497 struct vsock_sock *vpending;
498 struct sock *pending;
499 struct sockaddr_vm src;
501 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
503 vlistener = vsock_sk(listener);
505 list_for_each_entry(vpending, &vlistener->pending_links,
507 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
508 pkt->dst_port == vpending->local_addr.svm_port) {
509 pending = sk_vsock(vpending);
521 static void vmci_transport_release_pending(struct sock *pending)
526 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
527 * trusted sockets 2) sockets from applications running as the same user as the
528 * VM (this is only true for the host side and only when using hosted products)
531 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
533 return vsock->trusted ||
534 vmci_is_context_owner(peer_cid, vsock->owner->uid);
537 /* We allow sending datagrams to and receiving datagrams from a restricted VM
538 * only if it is trusted as described in vmci_transport_is_trusted.
541 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
543 if (VMADDR_CID_HYPERVISOR == peer_cid)
546 if (vsock->cached_peer != peer_cid) {
547 vsock->cached_peer = peer_cid;
548 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
549 (vmci_context_get_priv_flags(peer_cid) &
550 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
551 vsock->cached_peer_allow_dgram = false;
553 vsock->cached_peer_allow_dgram = true;
557 return vsock->cached_peer_allow_dgram;
561 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
562 struct vmci_handle *handle,
565 u32 peer, u32 flags, bool trusted)
570 /* Try to allocate our queue pair as trusted. This will only
571 * work if vsock is running in the host.
574 err = vmci_qpair_alloc(qpair, handle, produce_size,
577 VMCI_PRIVILEGE_FLAG_TRUSTED);
578 if (err != VMCI_ERROR_NO_ACCESS)
583 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
584 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
587 pr_err_once("Could not attach to queue pair with %d\n", err);
588 err = vmci_transport_error_to_vsock_error(err);
595 vmci_transport_datagram_create_hnd(u32 resource_id,
597 vmci_datagram_recv_cb recv_cb,
599 struct vmci_handle *out_handle)
603 /* Try to allocate our datagram handler as trusted. This will only work
604 * if vsock is running in the host.
607 err = vmci_datagram_create_handle_priv(resource_id, flags,
608 VMCI_PRIVILEGE_FLAG_TRUSTED,
610 client_data, out_handle);
612 if (err == VMCI_ERROR_NO_ACCESS)
613 err = vmci_datagram_create_handle(resource_id, flags,
614 recv_cb, client_data,
620 /* This is invoked as part of a tasklet that's scheduled when the VMCI
621 * interrupt fires. This is run in bottom-half context and if it ever needs to
622 * sleep it should defer that work to a work queue.
625 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
630 struct vsock_sock *vsk;
632 sk = (struct sock *)data;
634 /* This handler is privileged when this module is running on the host.
635 * We will get datagrams from all endpoints (even VMs that are in a
636 * restricted context). If we get one from a restricted context then
637 * the destination socket must be trusted.
639 * NOTE: We access the socket struct without holding the lock here.
640 * This is ok because the field we are interested is never modified
641 * outside of the create and destruct socket functions.
644 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
645 return VMCI_ERROR_NO_ACCESS;
647 size = VMCI_DG_SIZE(dg);
649 /* Attach the packet to the socket's receive queue as an sk_buff. */
650 skb = alloc_skb(size, GFP_ATOMIC);
652 return VMCI_ERROR_NO_MEM;
654 /* sk_receive_skb() will do a sock_put(), so hold here. */
657 memcpy(skb->data, dg, size);
658 sk_receive_skb(sk, skb, 0);
663 static bool vmci_transport_stream_allow(u32 cid, u32 port)
665 static const u32 non_socket_contexts[] = {
670 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
672 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
673 if (cid == non_socket_contexts[i])
680 /* This is invoked as part of a tasklet that's scheduled when the VMCI
681 * interrupt fires. This is run in bottom-half context but it defers most of
682 * its work to the packet handling work queue.
685 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
688 struct sockaddr_vm dst;
689 struct sockaddr_vm src;
690 struct vmci_transport_packet *pkt;
691 struct vsock_sock *vsk;
697 bh_process_pkt = false;
699 /* Ignore incoming packets from contexts without sockets, or resources
700 * that aren't vsock implementations.
703 if (!vmci_transport_stream_allow(dg->src.context, -1)
704 || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
705 return VMCI_ERROR_NO_ACCESS;
707 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
708 /* Drop datagrams that do not contain full VSock packets. */
709 return VMCI_ERROR_INVALID_ARGS;
711 pkt = (struct vmci_transport_packet *)dg;
713 /* Find the socket that should handle this packet. First we look for a
714 * connected socket and if there is none we look for a socket bound to
715 * the destintation address.
717 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
718 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
720 sk = vsock_find_connected_socket(&src, &dst);
722 sk = vsock_find_bound_socket(&dst);
724 /* We could not find a socket for this specified
725 * address. If this packet is a RST, we just drop it.
726 * If it is another packet, we send a RST. Note that
727 * we do not send a RST reply to RSTs so that we do not
728 * continually send RSTs between two endpoints.
730 * Note that since this is a reply, dst is src and src
733 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
734 pr_err("unable to send reset\n");
736 err = VMCI_ERROR_NOT_FOUND;
741 /* If the received packet type is beyond all types known to this
742 * implementation, reply with an invalid message. Hopefully this will
743 * help when implementing backwards compatibility in the future.
745 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
746 vmci_transport_send_invalid_bh(&dst, &src);
747 err = VMCI_ERROR_INVALID_ARGS;
751 /* This handler is privileged when this module is running on the host.
752 * We will get datagram connect requests from all endpoints (even VMs
753 * that are in a restricted context). If we get one from a restricted
754 * context then the destination socket must be trusted.
756 * NOTE: We access the socket struct without holding the lock here.
757 * This is ok because the field we are interested is never modified
758 * outside of the create and destruct socket functions.
761 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
762 err = VMCI_ERROR_NO_ACCESS;
766 /* We do most everything in a work queue, but let's fast path the
767 * notification of reads and writes to help data transfer performance.
768 * We can only do this if there is no process context code executing
769 * for this socket since that may change the state.
773 if (!sock_owned_by_user(sk)) {
774 /* The local context ID may be out of date, update it. */
775 vsk->local_addr.svm_cid = dst.svm_cid;
777 if (sk->sk_state == TCP_ESTABLISHED)
778 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
779 sk, pkt, true, &dst, &src,
785 if (!bh_process_pkt) {
786 struct vmci_transport_recv_pkt_info *recv_pkt_info;
788 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
789 if (!recv_pkt_info) {
790 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
791 pr_err("unable to send reset\n");
793 err = VMCI_ERROR_NO_MEM;
797 recv_pkt_info->sk = sk;
798 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
799 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
801 schedule_work(&recv_pkt_info->work);
802 /* Clear sk so that the reference count incremented by one of
803 * the Find functions above is not decremented below. We need
804 * that reference count for the packet handler we've scheduled
817 static void vmci_transport_handle_detach(struct sock *sk)
819 struct vsock_sock *vsk;
822 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
823 sock_set_flag(sk, SOCK_DONE);
825 /* On a detach the peer will not be sending or receiving
828 vsk->peer_shutdown = SHUTDOWN_MASK;
830 /* We should not be sending anymore since the peer won't be
831 * there to receive, but we can still receive if there is data
832 * left in our consume queue. If the local endpoint is a host,
833 * we can't call vsock_stream_has_data, since that may block,
834 * but a host endpoint can't read data once the VM has
835 * detached, so there is no available data in that case.
837 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
838 vsock_stream_has_data(vsk) <= 0) {
839 if (sk->sk_state == TCP_SYN_SENT) {
840 /* The peer may detach from a queue pair while
841 * we are still in the connecting state, i.e.,
842 * if the peer VM is killed after attaching to
843 * a queue pair, but before we complete the
844 * handshake. In that case, we treat the detach
845 * event like a reset.
848 sk->sk_state = TCP_CLOSE;
849 sk->sk_err = ECONNRESET;
850 sk->sk_error_report(sk);
853 sk->sk_state = TCP_CLOSE;
855 sk->sk_state_change(sk);
859 static void vmci_transport_peer_detach_cb(u32 sub_id,
860 const struct vmci_event_data *e_data,
863 struct vmci_transport *trans = client_data;
864 const struct vmci_event_payload_qp *e_payload;
866 e_payload = vmci_event_data_const_payload(e_data);
868 /* XXX This is lame, we should provide a way to lookup sockets by
871 if (vmci_handle_is_invalid(e_payload->handle) ||
872 !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
875 /* We don't ask for delayed CBs when we subscribe to this event (we
876 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
877 * guarantees in that case about what context we might be running in,
878 * so it could be BH or process, blockable or non-blockable. So we
879 * need to account for all possible contexts here.
881 spin_lock_bh(&trans->lock);
885 /* Apart from here, trans->lock is only grabbed as part of sk destruct,
886 * where trans->sk isn't locked.
888 bh_lock_sock(trans->sk);
890 vmci_transport_handle_detach(trans->sk);
892 bh_unlock_sock(trans->sk);
894 spin_unlock_bh(&trans->lock);
897 static void vmci_transport_qp_resumed_cb(u32 sub_id,
898 const struct vmci_event_data *e_data,
901 vsock_for_each_connected_socket(vmci_transport_handle_detach);
904 static void vmci_transport_recv_pkt_work(struct work_struct *work)
906 struct vmci_transport_recv_pkt_info *recv_pkt_info;
907 struct vmci_transport_packet *pkt;
911 container_of(work, struct vmci_transport_recv_pkt_info, work);
912 sk = recv_pkt_info->sk;
913 pkt = &recv_pkt_info->pkt;
917 /* The local context ID may be out of date. */
918 vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
920 switch (sk->sk_state) {
922 vmci_transport_recv_listen(sk, pkt);
925 /* Processing of pending connections for servers goes through
926 * the listening socket, so see vmci_transport_recv_listen()
929 vmci_transport_recv_connecting_client(sk, pkt);
931 case TCP_ESTABLISHED:
932 vmci_transport_recv_connected(sk, pkt);
935 /* Because this function does not run in the same context as
936 * vmci_transport_recv_stream_cb it is possible that the
937 * socket has closed. We need to let the other side know or it
938 * could be sitting in a connect and hang forever. Send a
939 * reset to prevent that.
941 vmci_transport_send_reset(sk, pkt);
946 kfree(recv_pkt_info);
947 /* Release reference obtained in the stream callback when we fetched
948 * this socket out of the bound or connected list.
953 static int vmci_transport_recv_listen(struct sock *sk,
954 struct vmci_transport_packet *pkt)
956 struct sock *pending;
957 struct vsock_sock *vpending;
960 bool old_request = false;
961 bool old_pkt_proto = false;
965 /* Because we are in the listen state, we could be receiving a packet
966 * for ourself or any previous connection requests that we received.
967 * If it's the latter, we try to find a socket in our list of pending
968 * connections and, if we do, call the appropriate handler for the
969 * state that that socket is in. Otherwise we try to service the
970 * connection request.
972 pending = vmci_transport_get_pending(sk, pkt);
976 /* The local context ID may be out of date. */
977 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
979 switch (pending->sk_state) {
981 err = vmci_transport_recv_connecting_server(sk,
986 vmci_transport_send_reset(pending, pkt);
991 vsock_remove_pending(sk, pending);
993 release_sock(pending);
994 vmci_transport_release_pending(pending);
999 /* The listen state only accepts connection requests. Reply with a
1000 * reset unless we received a reset.
1003 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
1004 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
1005 vmci_transport_reply_reset(pkt);
1009 if (pkt->u.size == 0) {
1010 vmci_transport_reply_reset(pkt);
1014 /* If this socket can't accommodate this connection request, we send a
1015 * reset. Otherwise we create and initialize a child socket and reply
1016 * with a connection negotiation.
1018 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1019 vmci_transport_reply_reset(pkt);
1020 return -ECONNREFUSED;
1023 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1026 vmci_transport_send_reset(sk, pkt);
1030 vpending = vsock_sk(pending);
1032 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1034 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1037 /* If the proposed size fits within our min/max, accept it. Otherwise
1038 * propose our own size.
1040 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1041 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1042 qp_size = pkt->u.size;
1044 qp_size = vmci_trans(vpending)->queue_pair_size;
1047 /* Figure out if we are using old or new requests based on the
1048 * overrides pkt types sent by our peer.
1050 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1051 old_request = old_pkt_proto;
1053 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1055 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1056 old_request = false;
1061 /* Handle a REQUEST (or override) */
1062 u16 version = VSOCK_PROTO_INVALID;
1063 if (vmci_transport_proto_to_notify_struct(
1064 pending, &version, true))
1065 err = vmci_transport_send_negotiate(pending, qp_size);
1070 /* Handle a REQUEST2 (or override) */
1071 int proto_int = pkt->proto;
1073 u16 active_proto_version = 0;
1075 /* The list of possible protocols is the intersection of all
1076 * protocols the client supports ... plus all the protocols we
1079 proto_int &= vmci_transport_new_proto_supported_versions();
1081 /* We choose the highest possible protocol version and use that
1084 pos = fls(proto_int);
1086 active_proto_version = (1 << (pos - 1));
1087 if (vmci_transport_proto_to_notify_struct(
1088 pending, &active_proto_version, false))
1089 err = vmci_transport_send_negotiate2(pending,
1091 active_proto_version);
1101 vmci_transport_send_reset(sk, pkt);
1103 err = vmci_transport_error_to_vsock_error(err);
1107 vsock_add_pending(sk, pending);
1108 sk->sk_ack_backlog++;
1110 pending->sk_state = TCP_SYN_SENT;
1111 vmci_trans(vpending)->produce_size =
1112 vmci_trans(vpending)->consume_size = qp_size;
1113 vmci_trans(vpending)->queue_pair_size = qp_size;
1115 vmci_trans(vpending)->notify_ops->process_request(pending);
1117 /* We might never receive another message for this socket and it's not
1118 * connected to any process, so we have to ensure it gets cleaned up
1119 * ourself. Our delayed work function will take care of that. Note
1120 * that we do not ever cancel this function since we have few
1121 * guarantees about its state when calling cancel_delayed_work().
1122 * Instead we hold a reference on the socket for that function and make
1123 * it capable of handling cases where it needs to do nothing but
1124 * release that reference.
1126 vpending->listener = sk;
1129 schedule_delayed_work(&vpending->pending_work, HZ);
1136 vmci_transport_recv_connecting_server(struct sock *listener,
1137 struct sock *pending,
1138 struct vmci_transport_packet *pkt)
1140 struct vsock_sock *vpending;
1141 struct vmci_handle handle;
1142 struct vmci_qp *qpair;
1149 vpending = vsock_sk(pending);
1150 detach_sub_id = VMCI_INVALID_ID;
1152 switch (pkt->type) {
1153 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1154 if (vmci_handle_is_invalid(pkt->u.handle)) {
1155 vmci_transport_send_reset(pending, pkt);
1162 /* Close and cleanup the connection. */
1163 vmci_transport_send_reset(pending, pkt);
1165 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1169 /* In order to complete the connection we need to attach to the offered
1170 * queue pair and send an attach notification. We also subscribe to the
1171 * detach event so we know when our peer goes away, and we do that
1172 * before attaching so we don't miss an event. If all this succeeds,
1173 * we update our state and wakeup anything waiting in accept() for a
1177 /* We don't care about attach since we ensure the other side has
1178 * attached by specifying the ATTACH_ONLY flag below.
1180 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1181 vmci_transport_peer_detach_cb,
1182 vmci_trans(vpending), &detach_sub_id);
1183 if (err < VMCI_SUCCESS) {
1184 vmci_transport_send_reset(pending, pkt);
1185 err = vmci_transport_error_to_vsock_error(err);
1190 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1192 /* Now attach to the queue pair the client created. */
1193 handle = pkt->u.handle;
1195 /* vpending->local_addr always has a context id so we do not need to
1196 * worry about VMADDR_CID_ANY in this case.
1199 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1200 flags = VMCI_QPFLAG_ATTACH_ONLY;
1201 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1203 err = vmci_transport_queue_pair_alloc(
1206 vmci_trans(vpending)->produce_size,
1207 vmci_trans(vpending)->consume_size,
1208 pkt->dg.src.context,
1210 vmci_transport_is_trusted(
1212 vpending->remote_addr.svm_cid));
1214 vmci_transport_send_reset(pending, pkt);
1219 vmci_trans(vpending)->qp_handle = handle;
1220 vmci_trans(vpending)->qpair = qpair;
1222 /* When we send the attach message, we must be ready to handle incoming
1223 * control messages on the newly connected socket. So we move the
1224 * pending socket to the connected state before sending the attach
1225 * message. Otherwise, an incoming packet triggered by the attach being
1226 * received by the peer may be processed concurrently with what happens
1227 * below after sending the attach message, and that incoming packet
1228 * will find the listening socket instead of the (currently) pending
1229 * socket. Note that enqueueing the socket increments the reference
1230 * count, so even if a reset comes before the connection is accepted,
1231 * the socket will be valid until it is removed from the queue.
1233 * If we fail sending the attach below, we remove the socket from the
1234 * connected list and move the socket to TCP_CLOSE before
1235 * releasing the lock, so a pending slow path processing of an incoming
1236 * packet will not see the socket in the connected state in that case.
1238 pending->sk_state = TCP_ESTABLISHED;
1240 vsock_insert_connected(vpending);
1242 /* Notify our peer of our attach. */
1243 err = vmci_transport_send_attach(pending, handle);
1245 vsock_remove_connected(vpending);
1246 pr_err("Could not send attach\n");
1247 vmci_transport_send_reset(pending, pkt);
1248 err = vmci_transport_error_to_vsock_error(err);
1253 /* We have a connection. Move the now connected socket from the
1254 * listener's pending list to the accept queue so callers of accept()
1257 vsock_remove_pending(listener, pending);
1258 vsock_enqueue_accept(listener, pending);
1260 /* Callers of accept() will be be waiting on the listening socket, not
1261 * the pending socket.
1263 listener->sk_data_ready(listener);
1268 pending->sk_err = skerr;
1269 pending->sk_state = TCP_CLOSE;
1270 /* As long as we drop our reference, all necessary cleanup will handle
1271 * when the cleanup function drops its reference and our destruct
1272 * implementation is called. Note that since the listen handler will
1273 * remove pending from the pending list upon our failure, the cleanup
1274 * function won't drop the additional reference, which is why we do it
1283 vmci_transport_recv_connecting_client(struct sock *sk,
1284 struct vmci_transport_packet *pkt)
1286 struct vsock_sock *vsk;
1292 switch (pkt->type) {
1293 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1294 if (vmci_handle_is_invalid(pkt->u.handle) ||
1295 !vmci_handle_is_equal(pkt->u.handle,
1296 vmci_trans(vsk)->qp_handle)) {
1302 /* Signify the socket is connected and wakeup the waiter in
1303 * connect(). Also place the socket in the connected table for
1304 * accounting (it can already be found since it's in the bound
1307 sk->sk_state = TCP_ESTABLISHED;
1308 sk->sk_socket->state = SS_CONNECTED;
1309 vsock_insert_connected(vsk);
1310 sk->sk_state_change(sk);
1313 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1314 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1315 if (pkt->u.size == 0
1316 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1317 || pkt->src_port != vsk->remote_addr.svm_port
1318 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1319 || vmci_trans(vsk)->qpair
1320 || vmci_trans(vsk)->produce_size != 0
1321 || vmci_trans(vsk)->consume_size != 0
1322 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1329 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1336 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1337 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1344 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1345 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1346 * continue processing here after they sent an INVALID packet.
1347 * This meant that we got a RST after the INVALID. We ignore a
1348 * RST after an INVALID. The common code doesn't send the RST
1349 * ... so we can hang if an old version of the common code
1350 * fails between getting a REQUEST and sending an OFFER back.
1351 * Not much we can do about it... except hope that it doesn't
1354 if (vsk->ignore_connecting_rst) {
1355 vsk->ignore_connecting_rst = false;
1364 /* Close and cleanup the connection. */
1373 vmci_transport_send_reset(sk, pkt);
1375 sk->sk_state = TCP_CLOSE;
1377 sk->sk_error_report(sk);
1381 static int vmci_transport_recv_connecting_client_negotiate(
1383 struct vmci_transport_packet *pkt)
1386 struct vsock_sock *vsk;
1387 struct vmci_handle handle;
1388 struct vmci_qp *qpair;
1392 bool old_proto = true;
1397 handle = VMCI_INVALID_HANDLE;
1398 detach_sub_id = VMCI_INVALID_ID;
1400 /* If we have gotten here then we should be past the point where old
1401 * linux vsock could have sent the bogus rst.
1403 vsk->sent_request = false;
1404 vsk->ignore_connecting_rst = false;
1406 /* Verify that we're OK with the proposed queue pair size */
1407 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1408 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1413 /* At this point we know the CID the peer is using to talk to us. */
1415 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1416 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1418 /* Setup the notify ops to be the highest supported version that both
1419 * the server and the client support.
1422 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1423 old_proto = old_pkt_proto;
1425 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1427 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1433 version = VSOCK_PROTO_INVALID;
1435 version = pkt->proto;
1437 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1442 /* Subscribe to detach events first.
1444 * XXX We attach once for each queue pair created for now so it is easy
1445 * to find the socket (it's provided), but later we should only
1446 * subscribe once and add a way to lookup sockets by queue pair handle.
1448 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1449 vmci_transport_peer_detach_cb,
1450 vmci_trans(vsk), &detach_sub_id);
1451 if (err < VMCI_SUCCESS) {
1452 err = vmci_transport_error_to_vsock_error(err);
1456 /* Make VMCI select the handle for us. */
1457 handle = VMCI_INVALID_HANDLE;
1458 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1459 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1461 err = vmci_transport_queue_pair_alloc(&qpair,
1465 vsk->remote_addr.svm_cid,
1467 vmci_transport_is_trusted(
1470 remote_addr.svm_cid));
1474 err = vmci_transport_send_qp_offer(sk, handle);
1476 err = vmci_transport_error_to_vsock_error(err);
1480 vmci_trans(vsk)->qp_handle = handle;
1481 vmci_trans(vsk)->qpair = qpair;
1483 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1486 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1488 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1493 if (detach_sub_id != VMCI_INVALID_ID)
1494 vmci_event_unsubscribe(detach_sub_id);
1496 if (!vmci_handle_is_invalid(handle))
1497 vmci_qpair_detach(&qpair);
1503 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1504 struct vmci_transport_packet *pkt)
1507 struct vsock_sock *vsk = vsock_sk(sk);
1509 if (vsk->sent_request) {
1510 vsk->sent_request = false;
1511 vsk->ignore_connecting_rst = true;
1513 err = vmci_transport_send_conn_request(
1514 sk, vmci_trans(vsk)->queue_pair_size);
1516 err = vmci_transport_error_to_vsock_error(err);
1525 static int vmci_transport_recv_connected(struct sock *sk,
1526 struct vmci_transport_packet *pkt)
1528 struct vsock_sock *vsk;
1529 bool pkt_processed = false;
1531 /* In cases where we are closing the connection, it's sufficient to
1532 * mark the state change (and maybe error) and wake up any waiting
1533 * threads. Since this is a connected socket, it's owned by a user
1534 * process and will be cleaned up when the failure is passed back on
1535 * the current or next system call. Our system call implementations
1536 * must therefore check for error and state changes on entry and when
1539 switch (pkt->type) {
1540 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1544 vsk->peer_shutdown |= pkt->u.mode;
1545 sk->sk_state_change(sk);
1549 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1551 /* It is possible that we sent our peer a message (e.g a
1552 * WAITING_READ) right before we got notified that the peer had
1553 * detached. If that happens then we can get a RST pkt back
1554 * from our peer even though there is data available for us to
1555 * read. In that case, don't shutdown the socket completely but
1556 * instead allow the local client to finish reading data off
1557 * the queuepair. Always treat a RST pkt in connected mode like
1560 sock_set_flag(sk, SOCK_DONE);
1561 vsk->peer_shutdown = SHUTDOWN_MASK;
1562 if (vsock_stream_has_data(vsk) <= 0)
1563 sk->sk_state = TCP_CLOSING;
1565 sk->sk_state_change(sk);
1570 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1571 sk, pkt, false, NULL, NULL,
1582 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1583 struct vsock_sock *psk)
1585 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1589 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1590 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1591 vmci_trans(vsk)->qpair = NULL;
1592 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1593 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1594 vmci_trans(vsk)->notify_ops = NULL;
1595 INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1596 vmci_trans(vsk)->sk = &vsk->sk;
1597 spin_lock_init(&vmci_trans(vsk)->lock);
1599 vmci_trans(vsk)->queue_pair_size =
1600 vmci_trans(psk)->queue_pair_size;
1601 vmci_trans(vsk)->queue_pair_min_size =
1602 vmci_trans(psk)->queue_pair_min_size;
1603 vmci_trans(vsk)->queue_pair_max_size =
1604 vmci_trans(psk)->queue_pair_max_size;
1606 vmci_trans(vsk)->queue_pair_size =
1607 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1608 vmci_trans(vsk)->queue_pair_min_size =
1609 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1610 vmci_trans(vsk)->queue_pair_max_size =
1611 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1617 static void vmci_transport_free_resources(struct list_head *transport_list)
1619 while (!list_empty(transport_list)) {
1620 struct vmci_transport *transport =
1621 list_first_entry(transport_list, struct vmci_transport,
1623 list_del(&transport->elem);
1625 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1626 vmci_event_unsubscribe(transport->detach_sub_id);
1627 transport->detach_sub_id = VMCI_INVALID_ID;
1630 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1631 vmci_qpair_detach(&transport->qpair);
1632 transport->qp_handle = VMCI_INVALID_HANDLE;
1633 transport->produce_size = 0;
1634 transport->consume_size = 0;
1641 static void vmci_transport_cleanup(struct work_struct *work)
1645 spin_lock_bh(&vmci_transport_cleanup_lock);
1646 list_replace_init(&vmci_transport_cleanup_list, &pending);
1647 spin_unlock_bh(&vmci_transport_cleanup_lock);
1648 vmci_transport_free_resources(&pending);
1651 static void vmci_transport_destruct(struct vsock_sock *vsk)
1653 /* transport can be NULL if we hit a failure at init() time */
1654 if (!vmci_trans(vsk))
1657 /* Ensure that the detach callback doesn't use the sk/vsk
1658 * we are about to destruct.
1660 spin_lock_bh(&vmci_trans(vsk)->lock);
1661 vmci_trans(vsk)->sk = NULL;
1662 spin_unlock_bh(&vmci_trans(vsk)->lock);
1664 if (vmci_trans(vsk)->notify_ops)
1665 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1667 spin_lock_bh(&vmci_transport_cleanup_lock);
1668 list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1669 spin_unlock_bh(&vmci_transport_cleanup_lock);
1670 schedule_work(&vmci_transport_cleanup_work);
1675 static void vmci_transport_release(struct vsock_sock *vsk)
1677 vsock_remove_sock(vsk);
1679 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1680 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1681 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1685 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1686 struct sockaddr_vm *addr)
1692 /* VMCI will select a resource ID for us if we provide
1695 port = addr->svm_port == VMADDR_PORT_ANY ?
1696 VMCI_INVALID_ID : addr->svm_port;
1698 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1701 flags = addr->svm_cid == VMADDR_CID_ANY ?
1702 VMCI_FLAG_ANYCID_DG_HND : 0;
1704 err = vmci_transport_datagram_create_hnd(port, flags,
1705 vmci_transport_recv_dgram_cb,
1707 &vmci_trans(vsk)->dg_handle);
1708 if (err < VMCI_SUCCESS)
1709 return vmci_transport_error_to_vsock_error(err);
1710 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1711 vmci_trans(vsk)->dg_handle.resource);
1716 static int vmci_transport_dgram_enqueue(
1717 struct vsock_sock *vsk,
1718 struct sockaddr_vm *remote_addr,
1723 struct vmci_datagram *dg;
1725 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1728 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1731 /* Allocate a buffer for the user's message and our packet header. */
1732 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1736 memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1738 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1739 remote_addr->svm_port);
1740 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1741 vsk->local_addr.svm_port);
1742 dg->payload_size = len;
1744 err = vmci_datagram_send(dg);
1747 return vmci_transport_error_to_vsock_error(err);
1749 return err - sizeof(*dg);
1752 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1753 struct msghdr *msg, size_t len,
1758 struct vmci_datagram *dg;
1760 struct sk_buff *skb;
1762 noblock = flags & MSG_DONTWAIT;
1764 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1767 /* Retrieve the head sk_buff from the socket's receive queue. */
1769 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1773 dg = (struct vmci_datagram *)skb->data;
1775 /* err is 0, meaning we read zero bytes. */
1778 payload_len = dg->payload_size;
1779 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1780 if (payload_len != skb->len - sizeof(*dg)) {
1785 if (payload_len > len) {
1787 msg->msg_flags |= MSG_TRUNC;
1790 /* Place the datagram payload in the user's iovec. */
1791 err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1795 if (msg->msg_name) {
1796 /* Provide the address of the sender. */
1797 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1798 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1799 msg->msg_namelen = sizeof(*vm_addr);
1804 skb_free_datagram(&vsk->sk, skb);
1808 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1810 if (cid == VMADDR_CID_HYPERVISOR) {
1811 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1812 * state and are allowed.
1814 return port == VMCI_UNITY_PBRPC_REGISTER;
1820 static int vmci_transport_connect(struct vsock_sock *vsk)
1823 bool old_pkt_proto = false;
1824 struct sock *sk = &vsk->sk;
1826 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1828 err = vmci_transport_send_conn_request(
1829 sk, vmci_trans(vsk)->queue_pair_size);
1831 sk->sk_state = TCP_CLOSE;
1835 int supported_proto_versions =
1836 vmci_transport_new_proto_supported_versions();
1837 err = vmci_transport_send_conn_request2(
1838 sk, vmci_trans(vsk)->queue_pair_size,
1839 supported_proto_versions);
1841 sk->sk_state = TCP_CLOSE;
1845 vsk->sent_request = true;
1851 static ssize_t vmci_transport_stream_dequeue(
1852 struct vsock_sock *vsk,
1857 if (flags & MSG_PEEK)
1858 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1860 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1863 static ssize_t vmci_transport_stream_enqueue(
1864 struct vsock_sock *vsk,
1868 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1871 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1873 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1876 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1878 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1881 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1883 return vmci_trans(vsk)->consume_size;
1886 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1888 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1891 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1893 return vmci_trans(vsk)->queue_pair_size;
1896 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1898 return vmci_trans(vsk)->queue_pair_min_size;
1901 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1903 return vmci_trans(vsk)->queue_pair_max_size;
1906 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1908 if (val < vmci_trans(vsk)->queue_pair_min_size)
1909 vmci_trans(vsk)->queue_pair_min_size = val;
1910 if (val > vmci_trans(vsk)->queue_pair_max_size)
1911 vmci_trans(vsk)->queue_pair_max_size = val;
1912 vmci_trans(vsk)->queue_pair_size = val;
1915 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1918 if (val > vmci_trans(vsk)->queue_pair_size)
1919 vmci_trans(vsk)->queue_pair_size = val;
1920 vmci_trans(vsk)->queue_pair_min_size = val;
1923 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1926 if (val < vmci_trans(vsk)->queue_pair_size)
1927 vmci_trans(vsk)->queue_pair_size = val;
1928 vmci_trans(vsk)->queue_pair_max_size = val;
1931 static int vmci_transport_notify_poll_in(
1932 struct vsock_sock *vsk,
1934 bool *data_ready_now)
1936 return vmci_trans(vsk)->notify_ops->poll_in(
1937 &vsk->sk, target, data_ready_now);
1940 static int vmci_transport_notify_poll_out(
1941 struct vsock_sock *vsk,
1943 bool *space_available_now)
1945 return vmci_trans(vsk)->notify_ops->poll_out(
1946 &vsk->sk, target, space_available_now);
1949 static int vmci_transport_notify_recv_init(
1950 struct vsock_sock *vsk,
1952 struct vsock_transport_recv_notify_data *data)
1954 return vmci_trans(vsk)->notify_ops->recv_init(
1956 (struct vmci_transport_recv_notify_data *)data);
1959 static int vmci_transport_notify_recv_pre_block(
1960 struct vsock_sock *vsk,
1962 struct vsock_transport_recv_notify_data *data)
1964 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1966 (struct vmci_transport_recv_notify_data *)data);
1969 static int vmci_transport_notify_recv_pre_dequeue(
1970 struct vsock_sock *vsk,
1972 struct vsock_transport_recv_notify_data *data)
1974 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1976 (struct vmci_transport_recv_notify_data *)data);
1979 static int vmci_transport_notify_recv_post_dequeue(
1980 struct vsock_sock *vsk,
1984 struct vsock_transport_recv_notify_data *data)
1986 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1987 &vsk->sk, target, copied, data_read,
1988 (struct vmci_transport_recv_notify_data *)data);
1991 static int vmci_transport_notify_send_init(
1992 struct vsock_sock *vsk,
1993 struct vsock_transport_send_notify_data *data)
1995 return vmci_trans(vsk)->notify_ops->send_init(
1997 (struct vmci_transport_send_notify_data *)data);
2000 static int vmci_transport_notify_send_pre_block(
2001 struct vsock_sock *vsk,
2002 struct vsock_transport_send_notify_data *data)
2004 return vmci_trans(vsk)->notify_ops->send_pre_block(
2006 (struct vmci_transport_send_notify_data *)data);
2009 static int vmci_transport_notify_send_pre_enqueue(
2010 struct vsock_sock *vsk,
2011 struct vsock_transport_send_notify_data *data)
2013 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2015 (struct vmci_transport_send_notify_data *)data);
2018 static int vmci_transport_notify_send_post_enqueue(
2019 struct vsock_sock *vsk,
2021 struct vsock_transport_send_notify_data *data)
2023 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2025 (struct vmci_transport_send_notify_data *)data);
2028 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2030 if (PROTOCOL_OVERRIDE != -1) {
2031 if (PROTOCOL_OVERRIDE == 0)
2032 *old_pkt_proto = true;
2034 *old_pkt_proto = false;
2036 pr_info("Proto override in use\n");
2043 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2047 struct vsock_sock *vsk = vsock_sk(sk);
2049 if (old_pkt_proto) {
2050 if (*proto != VSOCK_PROTO_INVALID) {
2051 pr_err("Can't set both an old and new protocol\n");
2054 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2059 case VSOCK_PROTO_PKT_ON_NOTIFY:
2060 vmci_trans(vsk)->notify_ops =
2061 &vmci_transport_notify_pkt_q_state_ops;
2064 pr_err("Unknown notify protocol version\n");
2069 vmci_trans(vsk)->notify_ops->socket_init(sk);
2073 static u16 vmci_transport_new_proto_supported_versions(void)
2075 if (PROTOCOL_OVERRIDE != -1)
2076 return PROTOCOL_OVERRIDE;
2078 return VSOCK_PROTO_ALL_SUPPORTED;
2081 static u32 vmci_transport_get_local_cid(void)
2083 return vmci_get_context_id();
2086 static const struct vsock_transport vmci_transport = {
2087 .init = vmci_transport_socket_init,
2088 .destruct = vmci_transport_destruct,
2089 .release = vmci_transport_release,
2090 .connect = vmci_transport_connect,
2091 .dgram_bind = vmci_transport_dgram_bind,
2092 .dgram_dequeue = vmci_transport_dgram_dequeue,
2093 .dgram_enqueue = vmci_transport_dgram_enqueue,
2094 .dgram_allow = vmci_transport_dgram_allow,
2095 .stream_dequeue = vmci_transport_stream_dequeue,
2096 .stream_enqueue = vmci_transport_stream_enqueue,
2097 .stream_has_data = vmci_transport_stream_has_data,
2098 .stream_has_space = vmci_transport_stream_has_space,
2099 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2100 .stream_is_active = vmci_transport_stream_is_active,
2101 .stream_allow = vmci_transport_stream_allow,
2102 .notify_poll_in = vmci_transport_notify_poll_in,
2103 .notify_poll_out = vmci_transport_notify_poll_out,
2104 .notify_recv_init = vmci_transport_notify_recv_init,
2105 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2106 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2107 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2108 .notify_send_init = vmci_transport_notify_send_init,
2109 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2110 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2111 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2112 .shutdown = vmci_transport_shutdown,
2113 .set_buffer_size = vmci_transport_set_buffer_size,
2114 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2115 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2116 .get_buffer_size = vmci_transport_get_buffer_size,
2117 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2118 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2119 .get_local_cid = vmci_transport_get_local_cid,
2122 static int __init vmci_transport_init(void)
2126 /* Create the datagram handle that we will use to send and receive all
2127 * VSocket control messages for this context.
2129 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2130 VMCI_FLAG_ANYCID_DG_HND,
2131 vmci_transport_recv_stream_cb,
2133 &vmci_transport_stream_handle);
2134 if (err < VMCI_SUCCESS) {
2135 pr_err("Unable to create datagram handle. (%d)\n", err);
2136 return vmci_transport_error_to_vsock_error(err);
2139 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2140 vmci_transport_qp_resumed_cb,
2141 NULL, &vmci_transport_qp_resumed_sub_id);
2142 if (err < VMCI_SUCCESS) {
2143 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2144 err = vmci_transport_error_to_vsock_error(err);
2145 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2146 goto err_destroy_stream_handle;
2149 err = vsock_core_init(&vmci_transport);
2151 goto err_unsubscribe;
2156 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2157 err_destroy_stream_handle:
2158 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2161 module_init(vmci_transport_init);
2163 static void __exit vmci_transport_exit(void)
2165 cancel_work_sync(&vmci_transport_cleanup_work);
2166 vmci_transport_free_resources(&vmci_transport_cleanup_list);
2168 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2169 if (vmci_datagram_destroy_handle(
2170 vmci_transport_stream_handle) != VMCI_SUCCESS)
2171 pr_err("Couldn't destroy datagram handle\n");
2172 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2175 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2176 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2177 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2182 module_exit(vmci_transport_exit);
2184 MODULE_AUTHOR("VMware, Inc.");
2185 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2186 MODULE_VERSION("1.0.5.0-k");
2187 MODULE_LICENSE("GPL v2");
2188 MODULE_ALIAS("vmware_vsock");
2189 MODULE_ALIAS_NETPROTO(PF_VSOCK);