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>
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_attach_cb(u32 sub_id,
44 const struct vmci_event_data *ed,
46 static void vmci_transport_peer_detach_cb(u32 sub_id,
47 const struct vmci_event_data *ed,
49 static void vmci_transport_recv_pkt_work(struct work_struct *work);
50 static int vmci_transport_recv_listen(struct sock *sk,
51 struct vmci_transport_packet *pkt);
52 static int vmci_transport_recv_connecting_server(
55 struct vmci_transport_packet *pkt);
56 static int vmci_transport_recv_connecting_client(
58 struct vmci_transport_packet *pkt);
59 static int vmci_transport_recv_connecting_client_negotiate(
61 struct vmci_transport_packet *pkt);
62 static int vmci_transport_recv_connecting_client_invalid(
64 struct vmci_transport_packet *pkt);
65 static int vmci_transport_recv_connected(struct sock *sk,
66 struct vmci_transport_packet *pkt);
67 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
68 static u16 vmci_transport_new_proto_supported_versions(void);
69 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
72 struct vmci_transport_recv_pkt_info {
73 struct work_struct work;
75 struct vmci_transport_packet pkt;
78 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
80 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
82 static int PROTOCOL_OVERRIDE = -1;
84 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN 128
85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE 262144
86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX 262144
88 /* The default peer timeout indicates how long we will wait for a peer response
89 * to a control message.
91 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
95 /* Helper function to convert from a VMCI error code to a VSock error code. */
97 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
101 switch (vmci_error) {
102 case VMCI_ERROR_NO_MEM:
105 case VMCI_ERROR_DUPLICATE_ENTRY:
106 case VMCI_ERROR_ALREADY_EXISTS:
109 case VMCI_ERROR_NO_ACCESS:
112 case VMCI_ERROR_NO_RESOURCES:
115 case VMCI_ERROR_INVALID_RESOURCE:
118 case VMCI_ERROR_INVALID_ARGS:
123 return err > 0 ? -err : err;
127 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
128 struct sockaddr_vm *src,
129 struct sockaddr_vm *dst,
133 struct vmci_transport_waiting_info *wait,
135 struct vmci_handle handle)
137 /* We register the stream control handler as an any cid handle so we
138 * must always send from a source address of VMADDR_CID_ANY
140 pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
141 VMCI_TRANSPORT_PACKET_RID);
142 pkt->dg.dst = vmci_make_handle(dst->svm_cid,
143 VMCI_TRANSPORT_PACKET_RID);
144 pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
145 pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
147 pkt->src_port = src->svm_port;
148 pkt->dst_port = dst->svm_port;
149 memset(&pkt->proto, 0, sizeof(pkt->proto));
150 memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
153 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
157 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
158 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
162 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
163 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
164 pkt->u.handle = handle;
167 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
168 case VMCI_TRANSPORT_PACKET_TYPE_READ:
169 case VMCI_TRANSPORT_PACKET_TYPE_RST:
173 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
177 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
178 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
179 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
182 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
183 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
191 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
192 struct sockaddr_vm *local,
193 struct sockaddr_vm *remote)
195 vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
196 vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
200 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
201 struct sockaddr_vm *src,
202 struct sockaddr_vm *dst,
203 enum vmci_transport_packet_type type,
206 struct vmci_transport_waiting_info *wait,
208 struct vmci_handle handle,
213 vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
215 err = vmci_datagram_send(&pkt->dg);
216 if (convert_error && (err < 0))
217 return vmci_transport_error_to_vsock_error(err);
223 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
224 enum vmci_transport_packet_type type,
227 struct vmci_transport_waiting_info *wait,
228 struct vmci_handle handle)
230 struct vmci_transport_packet reply;
231 struct sockaddr_vm src, dst;
233 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
236 vmci_transport_packet_get_addresses(pkt, &src, &dst);
237 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
246 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
247 struct sockaddr_vm *dst,
248 enum vmci_transport_packet_type type,
251 struct vmci_transport_waiting_info *wait,
252 struct vmci_handle handle)
254 /* Note that it is safe to use a single packet across all CPUs since
255 * two tasklets of the same type are guaranteed to not ever run
256 * simultaneously. If that ever changes, or VMCI stops using tasklets,
257 * we can use per-cpu packets.
259 static struct vmci_transport_packet pkt;
261 return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
263 VSOCK_PROTO_INVALID, handle,
268 vmci_transport_send_control_pkt(struct sock *sk,
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;
277 struct vsock_sock *vsk;
282 if (!vsock_addr_bound(&vsk->local_addr))
285 if (!vsock_addr_bound(&vsk->remote_addr))
288 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
292 err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
293 &vsk->remote_addr, type, size,
294 mode, wait, proto, handle,
301 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
302 struct sockaddr_vm *src,
303 struct vmci_transport_packet *pkt)
305 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
307 return vmci_transport_send_control_pkt_bh(
309 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
310 0, NULL, VMCI_INVALID_HANDLE);
313 static int vmci_transport_send_reset(struct sock *sk,
314 struct vmci_transport_packet *pkt)
316 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
318 return vmci_transport_send_control_pkt(sk,
319 VMCI_TRANSPORT_PACKET_TYPE_RST,
320 0, 0, NULL, VSOCK_PROTO_INVALID,
321 VMCI_INVALID_HANDLE);
324 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
326 return vmci_transport_send_control_pkt(
328 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
331 VMCI_INVALID_HANDLE);
334 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
337 return vmci_transport_send_control_pkt(
339 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
340 size, 0, NULL, version,
341 VMCI_INVALID_HANDLE);
344 static int vmci_transport_send_qp_offer(struct sock *sk,
345 struct vmci_handle handle)
347 return vmci_transport_send_control_pkt(
348 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
350 VSOCK_PROTO_INVALID, handle);
353 static int vmci_transport_send_attach(struct sock *sk,
354 struct vmci_handle handle)
356 return vmci_transport_send_control_pkt(
357 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
358 0, 0, NULL, VSOCK_PROTO_INVALID,
362 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
364 return vmci_transport_reply_control_pkt_fast(
366 VMCI_TRANSPORT_PACKET_TYPE_RST,
368 VMCI_INVALID_HANDLE);
371 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
372 struct sockaddr_vm *src)
374 return vmci_transport_send_control_pkt_bh(
376 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
377 0, 0, NULL, VMCI_INVALID_HANDLE);
380 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
381 struct sockaddr_vm *src)
383 return vmci_transport_send_control_pkt_bh(
385 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
386 0, NULL, VMCI_INVALID_HANDLE);
389 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
390 struct sockaddr_vm *src)
392 return vmci_transport_send_control_pkt_bh(
394 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
395 0, NULL, VMCI_INVALID_HANDLE);
398 int vmci_transport_send_wrote(struct sock *sk)
400 return vmci_transport_send_control_pkt(
401 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
402 0, NULL, VSOCK_PROTO_INVALID,
403 VMCI_INVALID_HANDLE);
406 int vmci_transport_send_read(struct sock *sk)
408 return vmci_transport_send_control_pkt(
409 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
410 0, NULL, VSOCK_PROTO_INVALID,
411 VMCI_INVALID_HANDLE);
414 int vmci_transport_send_waiting_write(struct sock *sk,
415 struct vmci_transport_waiting_info *wait)
417 return vmci_transport_send_control_pkt(
418 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
419 0, 0, wait, VSOCK_PROTO_INVALID,
420 VMCI_INVALID_HANDLE);
423 int vmci_transport_send_waiting_read(struct sock *sk,
424 struct vmci_transport_waiting_info *wait)
426 return vmci_transport_send_control_pkt(
427 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
428 0, 0, wait, VSOCK_PROTO_INVALID,
429 VMCI_INVALID_HANDLE);
432 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
434 return vmci_transport_send_control_pkt(
436 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
439 VMCI_INVALID_HANDLE);
442 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
444 return vmci_transport_send_control_pkt(sk,
445 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
448 VMCI_INVALID_HANDLE);
451 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
454 return vmci_transport_send_control_pkt(
455 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
456 size, 0, NULL, version,
457 VMCI_INVALID_HANDLE);
460 static struct sock *vmci_transport_get_pending(
461 struct sock *listener,
462 struct vmci_transport_packet *pkt)
464 struct vsock_sock *vlistener;
465 struct vsock_sock *vpending;
466 struct sock *pending;
468 vlistener = vsock_sk(listener);
470 list_for_each_entry(vpending, &vlistener->pending_links,
472 struct sockaddr_vm src;
473 struct sockaddr_vm dst;
475 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
476 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
478 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
479 vsock_addr_equals_addr(&dst, &vpending->local_addr)) {
480 pending = sk_vsock(vpending);
492 static void vmci_transport_release_pending(struct sock *pending)
497 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
498 * trusted sockets 2) sockets from applications running as the same user as the
499 * VM (this is only true for the host side and only when using hosted products)
502 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
504 return vsock->trusted ||
505 vmci_is_context_owner(peer_cid, vsock->owner->uid);
508 /* We allow sending datagrams to and receiving datagrams from a restricted VM
509 * only if it is trusted as described in vmci_transport_is_trusted.
512 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
514 if (vsock->cached_peer != peer_cid) {
515 vsock->cached_peer = peer_cid;
516 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
517 (vmci_context_get_priv_flags(peer_cid) &
518 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
519 vsock->cached_peer_allow_dgram = false;
521 vsock->cached_peer_allow_dgram = true;
525 return vsock->cached_peer_allow_dgram;
529 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
530 struct vmci_handle *handle,
533 u32 peer, u32 flags, bool trusted)
538 /* Try to allocate our queue pair as trusted. This will only
539 * work if vsock is running in the host.
542 err = vmci_qpair_alloc(qpair, handle, produce_size,
545 VMCI_PRIVILEGE_FLAG_TRUSTED);
546 if (err != VMCI_ERROR_NO_ACCESS)
551 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
552 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
555 pr_err("Could not attach to queue pair with %d\n",
557 err = vmci_transport_error_to_vsock_error(err);
564 vmci_transport_datagram_create_hnd(u32 resource_id,
566 vmci_datagram_recv_cb recv_cb,
568 struct vmci_handle *out_handle)
572 /* Try to allocate our datagram handler as trusted. This will only work
573 * if vsock is running in the host.
576 err = vmci_datagram_create_handle_priv(resource_id, flags,
577 VMCI_PRIVILEGE_FLAG_TRUSTED,
579 client_data, out_handle);
581 if (err == VMCI_ERROR_NO_ACCESS)
582 err = vmci_datagram_create_handle(resource_id, flags,
583 recv_cb, client_data,
589 /* This is invoked as part of a tasklet that's scheduled when the VMCI
590 * interrupt fires. This is run in bottom-half context and if it ever needs to
591 * sleep it should defer that work to a work queue.
594 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
599 struct vsock_sock *vsk;
601 sk = (struct sock *)data;
603 /* This handler is privileged when this module is running on the host.
604 * We will get datagrams from all endpoints (even VMs that are in a
605 * restricted context). If we get one from a restricted context then
606 * the destination socket must be trusted.
608 * NOTE: We access the socket struct without holding the lock here.
609 * This is ok because the field we are interested is never modified
610 * outside of the create and destruct socket functions.
613 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
614 return VMCI_ERROR_NO_ACCESS;
616 size = VMCI_DG_SIZE(dg);
618 /* Attach the packet to the socket's receive queue as an sk_buff. */
619 skb = alloc_skb(size, GFP_ATOMIC);
621 /* sk_receive_skb() will do a sock_put(), so hold here. */
624 memcpy(skb->data, dg, size);
625 sk_receive_skb(sk, skb, 0);
631 static bool vmci_transport_stream_allow(u32 cid, u32 port)
633 static const u32 non_socket_contexts[] = {
634 VMADDR_CID_HYPERVISOR,
639 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
641 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
642 if (cid == non_socket_contexts[i])
649 /* This is invoked as part of a tasklet that's scheduled when the VMCI
650 * interrupt fires. This is run in bottom-half context but it defers most of
651 * its work to the packet handling work queue.
654 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
657 struct sockaddr_vm dst;
658 struct sockaddr_vm src;
659 struct vmci_transport_packet *pkt;
660 struct vsock_sock *vsk;
666 bh_process_pkt = false;
668 /* Ignore incoming packets from contexts without sockets, or resources
669 * that aren't vsock implementations.
672 if (!vmci_transport_stream_allow(dg->src.context, -1)
673 || VMCI_TRANSPORT_PACKET_RID != dg->src.resource)
674 return VMCI_ERROR_NO_ACCESS;
676 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
677 /* Drop datagrams that do not contain full VSock packets. */
678 return VMCI_ERROR_INVALID_ARGS;
680 pkt = (struct vmci_transport_packet *)dg;
682 /* Find the socket that should handle this packet. First we look for a
683 * connected socket and if there is none we look for a socket bound to
684 * the destintation address.
686 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
687 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
689 sk = vsock_find_connected_socket(&src, &dst);
691 sk = vsock_find_bound_socket(&dst);
693 /* We could not find a socket for this specified
694 * address. If this packet is a RST, we just drop it.
695 * If it is another packet, we send a RST. Note that
696 * we do not send a RST reply to RSTs so that we do not
697 * continually send RSTs between two endpoints.
699 * Note that since this is a reply, dst is src and src
702 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
703 pr_err("unable to send reset\n");
705 err = VMCI_ERROR_NOT_FOUND;
710 /* If the received packet type is beyond all types known to this
711 * implementation, reply with an invalid message. Hopefully this will
712 * help when implementing backwards compatibility in the future.
714 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
715 vmci_transport_send_invalid_bh(&dst, &src);
716 err = VMCI_ERROR_INVALID_ARGS;
720 /* This handler is privileged when this module is running on the host.
721 * We will get datagram connect requests from all endpoints (even VMs
722 * that are in a restricted context). If we get one from a restricted
723 * context then the destination socket must be trusted.
725 * NOTE: We access the socket struct without holding the lock here.
726 * This is ok because the field we are interested is never modified
727 * outside of the create and destruct socket functions.
730 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
731 err = VMCI_ERROR_NO_ACCESS;
735 /* We do most everything in a work queue, but let's fast path the
736 * notification of reads and writes to help data transfer performance.
737 * We can only do this if there is no process context code executing
738 * for this socket since that may change the state.
742 if (!sock_owned_by_user(sk) && sk->sk_state == SS_CONNECTED)
743 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
744 sk, pkt, true, &dst, &src,
749 if (!bh_process_pkt) {
750 struct vmci_transport_recv_pkt_info *recv_pkt_info;
752 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
753 if (!recv_pkt_info) {
754 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
755 pr_err("unable to send reset\n");
757 err = VMCI_ERROR_NO_MEM;
761 recv_pkt_info->sk = sk;
762 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
763 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
765 schedule_work(&recv_pkt_info->work);
766 /* Clear sk so that the reference count incremented by one of
767 * the Find functions above is not decremented below. We need
768 * that reference count for the packet handler we've scheduled
781 static void vmci_transport_peer_attach_cb(u32 sub_id,
782 const struct vmci_event_data *e_data,
785 struct sock *sk = client_data;
786 const struct vmci_event_payload_qp *e_payload;
787 struct vsock_sock *vsk;
789 e_payload = vmci_event_data_const_payload(e_data);
793 /* We don't ask for delayed CBs when we subscribe to this event (we
794 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
795 * guarantees in that case about what context we might be running in,
796 * so it could be BH or process, blockable or non-blockable. So we
797 * need to account for all possible contexts here.
802 /* XXX This is lame, we should provide a way to lookup sockets by
805 if (vmci_handle_is_equal(vmci_trans(vsk)->qp_handle,
806 e_payload->handle)) {
807 /* XXX This doesn't do anything, but in the future we may want
808 * to set a flag here to verify the attach really did occur and
809 * we weren't just sent a datagram claiming it was.
819 static void vmci_transport_handle_detach(struct sock *sk)
821 struct vsock_sock *vsk;
824 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
825 sock_set_flag(sk, SOCK_DONE);
827 /* On a detach the peer will not be sending or receiving
830 vsk->peer_shutdown = SHUTDOWN_MASK;
832 /* We should not be sending anymore since the peer won't be
833 * there to receive, but we can still receive if there is data
834 * left in our consume queue.
836 if (vsock_stream_has_data(vsk) <= 0) {
837 if (sk->sk_state == SS_CONNECTING) {
838 /* The peer may detach from a queue pair while
839 * we are still in the connecting state, i.e.,
840 * if the peer VM is killed after attaching to
841 * a queue pair, but before we complete the
842 * handshake. In that case, we treat the detach
843 * event like a reset.
846 sk->sk_state = SS_UNCONNECTED;
847 sk->sk_err = ECONNRESET;
848 sk->sk_error_report(sk);
851 sk->sk_state = SS_UNCONNECTED;
853 sk->sk_state_change(sk);
857 static void vmci_transport_peer_detach_cb(u32 sub_id,
858 const struct vmci_event_data *e_data,
861 struct sock *sk = client_data;
862 const struct vmci_event_payload_qp *e_payload;
863 struct vsock_sock *vsk;
865 e_payload = vmci_event_data_const_payload(e_data);
867 if (vmci_handle_is_invalid(e_payload->handle))
870 /* Same rules for locking as for peer_attach_cb(). */
874 /* XXX This is lame, we should provide a way to lookup sockets by
877 if (vmci_handle_is_equal(vmci_trans(vsk)->qp_handle,
879 vmci_transport_handle_detach(sk);
885 static void vmci_transport_qp_resumed_cb(u32 sub_id,
886 const struct vmci_event_data *e_data,
889 vsock_for_each_connected_socket(vmci_transport_handle_detach);
892 static void vmci_transport_recv_pkt_work(struct work_struct *work)
894 struct vmci_transport_recv_pkt_info *recv_pkt_info;
895 struct vmci_transport_packet *pkt;
899 container_of(work, struct vmci_transport_recv_pkt_info, work);
900 sk = recv_pkt_info->sk;
901 pkt = &recv_pkt_info->pkt;
905 switch (sk->sk_state) {
907 vmci_transport_recv_listen(sk, pkt);
910 /* Processing of pending connections for servers goes through
911 * the listening socket, so see vmci_transport_recv_listen()
914 vmci_transport_recv_connecting_client(sk, pkt);
917 vmci_transport_recv_connected(sk, pkt);
920 /* Because this function does not run in the same context as
921 * vmci_transport_recv_stream_cb it is possible that the
922 * socket has closed. We need to let the other side know or it
923 * could be sitting in a connect and hang forever. Send a
924 * reset to prevent that.
926 vmci_transport_send_reset(sk, pkt);
932 kfree(recv_pkt_info);
933 /* Release reference obtained in the stream callback when we fetched
934 * this socket out of the bound or connected list.
939 static int vmci_transport_recv_listen(struct sock *sk,
940 struct vmci_transport_packet *pkt)
942 struct sock *pending;
943 struct vsock_sock *vpending;
946 bool old_request = false;
947 bool old_pkt_proto = false;
951 /* Because we are in the listen state, we could be receiving a packet
952 * for ourself or any previous connection requests that we received.
953 * If it's the latter, we try to find a socket in our list of pending
954 * connections and, if we do, call the appropriate handler for the
955 * state that that socket is in. Otherwise we try to service the
956 * connection request.
958 pending = vmci_transport_get_pending(sk, pkt);
961 switch (pending->sk_state) {
963 err = vmci_transport_recv_connecting_server(sk,
968 vmci_transport_send_reset(pending, pkt);
973 vsock_remove_pending(sk, pending);
975 release_sock(pending);
976 vmci_transport_release_pending(pending);
981 /* The listen state only accepts connection requests. Reply with a
982 * reset unless we received a reset.
985 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
986 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
987 vmci_transport_reply_reset(pkt);
991 if (pkt->u.size == 0) {
992 vmci_transport_reply_reset(pkt);
996 /* If this socket can't accommodate this connection request, we send a
997 * reset. Otherwise we create and initialize a child socket and reply
998 * with a connection negotiation.
1000 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1001 vmci_transport_reply_reset(pkt);
1002 return -ECONNREFUSED;
1005 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1008 vmci_transport_send_reset(sk, pkt);
1012 vpending = vsock_sk(pending);
1014 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1016 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1019 /* If the proposed size fits within our min/max, accept it. Otherwise
1020 * propose our own size.
1022 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1023 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1024 qp_size = pkt->u.size;
1026 qp_size = vmci_trans(vpending)->queue_pair_size;
1029 /* Figure out if we are using old or new requests based on the
1030 * overrides pkt types sent by our peer.
1032 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1033 old_request = old_pkt_proto;
1035 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1037 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1038 old_request = false;
1043 /* Handle a REQUEST (or override) */
1044 u16 version = VSOCK_PROTO_INVALID;
1045 if (vmci_transport_proto_to_notify_struct(
1046 pending, &version, true))
1047 err = vmci_transport_send_negotiate(pending, qp_size);
1052 /* Handle a REQUEST2 (or override) */
1053 int proto_int = pkt->proto;
1055 u16 active_proto_version = 0;
1057 /* The list of possible protocols is the intersection of all
1058 * protocols the client supports ... plus all the protocols we
1061 proto_int &= vmci_transport_new_proto_supported_versions();
1063 /* We choose the highest possible protocol version and use that
1066 pos = fls(proto_int);
1068 active_proto_version = (1 << (pos - 1));
1069 if (vmci_transport_proto_to_notify_struct(
1070 pending, &active_proto_version, false))
1071 err = vmci_transport_send_negotiate2(pending,
1073 active_proto_version);
1083 vmci_transport_send_reset(sk, pkt);
1085 err = vmci_transport_error_to_vsock_error(err);
1089 vsock_add_pending(sk, pending);
1090 sk->sk_ack_backlog++;
1092 pending->sk_state = SS_CONNECTING;
1093 vmci_trans(vpending)->produce_size =
1094 vmci_trans(vpending)->consume_size = qp_size;
1095 vmci_trans(vpending)->queue_pair_size = qp_size;
1097 vmci_trans(vpending)->notify_ops->process_request(pending);
1099 /* We might never receive another message for this socket and it's not
1100 * connected to any process, so we have to ensure it gets cleaned up
1101 * ourself. Our delayed work function will take care of that. Note
1102 * that we do not ever cancel this function since we have few
1103 * guarantees about its state when calling cancel_delayed_work().
1104 * Instead we hold a reference on the socket for that function and make
1105 * it capable of handling cases where it needs to do nothing but
1106 * release that reference.
1108 vpending->listener = sk;
1111 INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1112 schedule_delayed_work(&vpending->dwork, HZ);
1119 vmci_transport_recv_connecting_server(struct sock *listener,
1120 struct sock *pending,
1121 struct vmci_transport_packet *pkt)
1123 struct vsock_sock *vpending;
1124 struct vmci_handle handle;
1125 struct vmci_qp *qpair;
1132 vpending = vsock_sk(pending);
1133 detach_sub_id = VMCI_INVALID_ID;
1135 switch (pkt->type) {
1136 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1137 if (vmci_handle_is_invalid(pkt->u.handle)) {
1138 vmci_transport_send_reset(pending, pkt);
1145 /* Close and cleanup the connection. */
1146 vmci_transport_send_reset(pending, pkt);
1148 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1152 /* In order to complete the connection we need to attach to the offered
1153 * queue pair and send an attach notification. We also subscribe to the
1154 * detach event so we know when our peer goes away, and we do that
1155 * before attaching so we don't miss an event. If all this succeeds,
1156 * we update our state and wakeup anything waiting in accept() for a
1160 /* We don't care about attach since we ensure the other side has
1161 * attached by specifying the ATTACH_ONLY flag below.
1163 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1164 vmci_transport_peer_detach_cb,
1165 pending, &detach_sub_id);
1166 if (err < VMCI_SUCCESS) {
1167 vmci_transport_send_reset(pending, pkt);
1168 err = vmci_transport_error_to_vsock_error(err);
1173 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1175 /* Now attach to the queue pair the client created. */
1176 handle = pkt->u.handle;
1178 /* vpending->local_addr always has a context id so we do not need to
1179 * worry about VMADDR_CID_ANY in this case.
1182 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1183 flags = VMCI_QPFLAG_ATTACH_ONLY;
1184 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1186 err = vmci_transport_queue_pair_alloc(
1189 vmci_trans(vpending)->produce_size,
1190 vmci_trans(vpending)->consume_size,
1191 pkt->dg.src.context,
1193 vmci_transport_is_trusted(
1195 vpending->remote_addr.svm_cid));
1197 vmci_transport_send_reset(pending, pkt);
1202 vmci_trans(vpending)->qp_handle = handle;
1203 vmci_trans(vpending)->qpair = qpair;
1205 /* When we send the attach message, we must be ready to handle incoming
1206 * control messages on the newly connected socket. So we move the
1207 * pending socket to the connected state before sending the attach
1208 * message. Otherwise, an incoming packet triggered by the attach being
1209 * received by the peer may be processed concurrently with what happens
1210 * below after sending the attach message, and that incoming packet
1211 * will find the listening socket instead of the (currently) pending
1212 * socket. Note that enqueueing the socket increments the reference
1213 * count, so even if a reset comes before the connection is accepted,
1214 * the socket will be valid until it is removed from the queue.
1216 * If we fail sending the attach below, we remove the socket from the
1217 * connected list and move the socket to SS_UNCONNECTED before
1218 * releasing the lock, so a pending slow path processing of an incoming
1219 * packet will not see the socket in the connected state in that case.
1221 pending->sk_state = SS_CONNECTED;
1223 vsock_insert_connected(vpending);
1225 /* Notify our peer of our attach. */
1226 err = vmci_transport_send_attach(pending, handle);
1228 vsock_remove_connected(vpending);
1229 pr_err("Could not send attach\n");
1230 vmci_transport_send_reset(pending, pkt);
1231 err = vmci_transport_error_to_vsock_error(err);
1236 /* We have a connection. Move the now connected socket from the
1237 * listener's pending list to the accept queue so callers of accept()
1240 vsock_remove_pending(listener, pending);
1241 vsock_enqueue_accept(listener, pending);
1243 /* Callers of accept() will be be waiting on the listening socket, not
1244 * the pending socket.
1246 listener->sk_state_change(listener);
1251 pending->sk_err = skerr;
1252 pending->sk_state = SS_UNCONNECTED;
1253 /* As long as we drop our reference, all necessary cleanup will handle
1254 * when the cleanup function drops its reference and our destruct
1255 * implementation is called. Note that since the listen handler will
1256 * remove pending from the pending list upon our failure, the cleanup
1257 * function won't drop the additional reference, which is why we do it
1266 vmci_transport_recv_connecting_client(struct sock *sk,
1267 struct vmci_transport_packet *pkt)
1269 struct vsock_sock *vsk;
1275 switch (pkt->type) {
1276 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1277 if (vmci_handle_is_invalid(pkt->u.handle) ||
1278 !vmci_handle_is_equal(pkt->u.handle,
1279 vmci_trans(vsk)->qp_handle)) {
1285 /* Signify the socket is connected and wakeup the waiter in
1286 * connect(). Also place the socket in the connected table for
1287 * accounting (it can already be found since it's in the bound
1290 sk->sk_state = SS_CONNECTED;
1291 sk->sk_socket->state = SS_CONNECTED;
1292 vsock_insert_connected(vsk);
1293 sk->sk_state_change(sk);
1296 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1297 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1298 if (pkt->u.size == 0
1299 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1300 || pkt->src_port != vsk->remote_addr.svm_port
1301 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1302 || vmci_trans(vsk)->qpair
1303 || vmci_trans(vsk)->produce_size != 0
1304 || vmci_trans(vsk)->consume_size != 0
1305 || vmci_trans(vsk)->attach_sub_id != VMCI_INVALID_ID
1306 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1313 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1320 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1321 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1328 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1329 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1330 * continue processing here after they sent an INVALID packet.
1331 * This meant that we got a RST after the INVALID. We ignore a
1332 * RST after an INVALID. The common code doesn't send the RST
1333 * ... so we can hang if an old version of the common code
1334 * fails between getting a REQUEST and sending an OFFER back.
1335 * Not much we can do about it... except hope that it doesn't
1338 if (vsk->ignore_connecting_rst) {
1339 vsk->ignore_connecting_rst = false;
1348 /* Close and cleanup the connection. */
1357 vmci_transport_send_reset(sk, pkt);
1359 sk->sk_state = SS_UNCONNECTED;
1361 sk->sk_error_report(sk);
1365 static int vmci_transport_recv_connecting_client_negotiate(
1367 struct vmci_transport_packet *pkt)
1370 struct vsock_sock *vsk;
1371 struct vmci_handle handle;
1372 struct vmci_qp *qpair;
1377 bool old_proto = true;
1382 handle = VMCI_INVALID_HANDLE;
1383 attach_sub_id = VMCI_INVALID_ID;
1384 detach_sub_id = VMCI_INVALID_ID;
1386 /* If we have gotten here then we should be past the point where old
1387 * linux vsock could have sent the bogus rst.
1389 vsk->sent_request = false;
1390 vsk->ignore_connecting_rst = false;
1392 /* Verify that we're OK with the proposed queue pair size */
1393 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1394 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1399 /* At this point we know the CID the peer is using to talk to us. */
1401 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1402 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1404 /* Setup the notify ops to be the highest supported version that both
1405 * the server and the client support.
1408 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1409 old_proto = old_pkt_proto;
1411 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1413 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1419 version = VSOCK_PROTO_INVALID;
1421 version = pkt->proto;
1423 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1428 /* Subscribe to attach and detach events first.
1430 * XXX We attach once for each queue pair created for now so it is easy
1431 * to find the socket (it's provided), but later we should only
1432 * subscribe once and add a way to lookup sockets by queue pair handle.
1434 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_ATTACH,
1435 vmci_transport_peer_attach_cb,
1436 sk, &attach_sub_id);
1437 if (err < VMCI_SUCCESS) {
1438 err = vmci_transport_error_to_vsock_error(err);
1442 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1443 vmci_transport_peer_detach_cb,
1444 sk, &detach_sub_id);
1445 if (err < VMCI_SUCCESS) {
1446 err = vmci_transport_error_to_vsock_error(err);
1450 /* Make VMCI select the handle for us. */
1451 handle = VMCI_INVALID_HANDLE;
1452 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1453 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1455 err = vmci_transport_queue_pair_alloc(&qpair,
1459 vsk->remote_addr.svm_cid,
1461 vmci_transport_is_trusted(
1464 remote_addr.svm_cid));
1468 err = vmci_transport_send_qp_offer(sk, handle);
1470 err = vmci_transport_error_to_vsock_error(err);
1474 vmci_trans(vsk)->qp_handle = handle;
1475 vmci_trans(vsk)->qpair = qpair;
1477 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1480 vmci_trans(vsk)->attach_sub_id = attach_sub_id;
1481 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1483 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1488 if (attach_sub_id != VMCI_INVALID_ID)
1489 vmci_event_unsubscribe(attach_sub_id);
1491 if (detach_sub_id != VMCI_INVALID_ID)
1492 vmci_event_unsubscribe(detach_sub_id);
1494 if (!vmci_handle_is_invalid(handle))
1495 vmci_qpair_detach(&qpair);
1501 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1502 struct vmci_transport_packet *pkt)
1505 struct vsock_sock *vsk = vsock_sk(sk);
1507 if (vsk->sent_request) {
1508 vsk->sent_request = false;
1509 vsk->ignore_connecting_rst = true;
1511 err = vmci_transport_send_conn_request(
1512 sk, vmci_trans(vsk)->queue_pair_size);
1514 err = vmci_transport_error_to_vsock_error(err);
1523 static int vmci_transport_recv_connected(struct sock *sk,
1524 struct vmci_transport_packet *pkt)
1526 struct vsock_sock *vsk;
1527 bool pkt_processed = false;
1529 /* In cases where we are closing the connection, it's sufficient to
1530 * mark the state change (and maybe error) and wake up any waiting
1531 * threads. Since this is a connected socket, it's owned by a user
1532 * process and will be cleaned up when the failure is passed back on
1533 * the current or next system call. Our system call implementations
1534 * must therefore check for error and state changes on entry and when
1537 switch (pkt->type) {
1538 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1542 vsk->peer_shutdown |= pkt->u.mode;
1543 sk->sk_state_change(sk);
1547 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1549 /* It is possible that we sent our peer a message (e.g a
1550 * WAITING_READ) right before we got notified that the peer had
1551 * detached. If that happens then we can get a RST pkt back
1552 * from our peer even though there is data available for us to
1553 * read. In that case, don't shutdown the socket completely but
1554 * instead allow the local client to finish reading data off
1555 * the queuepair. Always treat a RST pkt in connected mode like
1558 sock_set_flag(sk, SOCK_DONE);
1559 vsk->peer_shutdown = SHUTDOWN_MASK;
1560 if (vsock_stream_has_data(vsk) <= 0)
1561 sk->sk_state = SS_DISCONNECTING;
1563 sk->sk_state_change(sk);
1568 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1569 sk, pkt, false, NULL, NULL,
1580 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1581 struct vsock_sock *psk)
1583 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1587 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1588 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1589 vmci_trans(vsk)->qpair = NULL;
1590 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1591 vmci_trans(vsk)->attach_sub_id = vmci_trans(vsk)->detach_sub_id =
1593 vmci_trans(vsk)->notify_ops = NULL;
1595 vmci_trans(vsk)->queue_pair_size =
1596 vmci_trans(psk)->queue_pair_size;
1597 vmci_trans(vsk)->queue_pair_min_size =
1598 vmci_trans(psk)->queue_pair_min_size;
1599 vmci_trans(vsk)->queue_pair_max_size =
1600 vmci_trans(psk)->queue_pair_max_size;
1602 vmci_trans(vsk)->queue_pair_size =
1603 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1604 vmci_trans(vsk)->queue_pair_min_size =
1605 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1606 vmci_trans(vsk)->queue_pair_max_size =
1607 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1613 static void vmci_transport_destruct(struct vsock_sock *vsk)
1615 if (vmci_trans(vsk)->attach_sub_id != VMCI_INVALID_ID) {
1616 vmci_event_unsubscribe(vmci_trans(vsk)->attach_sub_id);
1617 vmci_trans(vsk)->attach_sub_id = VMCI_INVALID_ID;
1620 if (vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1621 vmci_event_unsubscribe(vmci_trans(vsk)->detach_sub_id);
1622 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1625 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
1626 vmci_qpair_detach(&vmci_trans(vsk)->qpair);
1627 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1628 vmci_trans(vsk)->produce_size = 0;
1629 vmci_trans(vsk)->consume_size = 0;
1632 if (vmci_trans(vsk)->notify_ops)
1633 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1639 static void vmci_transport_release(struct vsock_sock *vsk)
1641 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1642 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1643 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1647 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1648 struct sockaddr_vm *addr)
1654 /* VMCI will select a resource ID for us if we provide
1657 port = addr->svm_port == VMADDR_PORT_ANY ?
1658 VMCI_INVALID_ID : addr->svm_port;
1660 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1663 flags = addr->svm_cid == VMADDR_CID_ANY ?
1664 VMCI_FLAG_ANYCID_DG_HND : 0;
1666 err = vmci_transport_datagram_create_hnd(port, flags,
1667 vmci_transport_recv_dgram_cb,
1669 &vmci_trans(vsk)->dg_handle);
1670 if (err < VMCI_SUCCESS)
1671 return vmci_transport_error_to_vsock_error(err);
1672 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1673 vmci_trans(vsk)->dg_handle.resource);
1678 static int vmci_transport_dgram_enqueue(
1679 struct vsock_sock *vsk,
1680 struct sockaddr_vm *remote_addr,
1685 struct vmci_datagram *dg;
1687 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1690 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1693 /* Allocate a buffer for the user's message and our packet header. */
1694 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1698 memcpy_fromiovec(VMCI_DG_PAYLOAD(dg), iov, len);
1700 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1701 remote_addr->svm_port);
1702 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1703 vsk->local_addr.svm_port);
1704 dg->payload_size = len;
1706 err = vmci_datagram_send(dg);
1709 return vmci_transport_error_to_vsock_error(err);
1711 return err - sizeof(*dg);
1714 static int vmci_transport_dgram_dequeue(struct kiocb *kiocb,
1715 struct vsock_sock *vsk,
1716 struct msghdr *msg, size_t len,
1721 struct vmci_datagram *dg;
1723 struct sk_buff *skb;
1725 noblock = flags & MSG_DONTWAIT;
1727 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1730 /* Retrieve the head sk_buff from the socket's receive queue. */
1732 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1739 dg = (struct vmci_datagram *)skb->data;
1741 /* err is 0, meaning we read zero bytes. */
1744 payload_len = dg->payload_size;
1745 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1746 if (payload_len != skb->len - sizeof(*dg)) {
1751 if (payload_len > len) {
1753 msg->msg_flags |= MSG_TRUNC;
1756 /* Place the datagram payload in the user's iovec. */
1757 err = skb_copy_datagram_iovec(skb, sizeof(*dg), msg->msg_iov,
1762 msg->msg_namelen = 0;
1763 if (msg->msg_name) {
1764 struct sockaddr_vm *vm_addr;
1766 /* Provide the address of the sender. */
1767 vm_addr = (struct sockaddr_vm *)msg->msg_name;
1768 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1769 msg->msg_namelen = sizeof(*vm_addr);
1774 skb_free_datagram(&vsk->sk, skb);
1778 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1780 if (cid == VMADDR_CID_HYPERVISOR) {
1781 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1782 * state and are allowed.
1784 return port == VMCI_UNITY_PBRPC_REGISTER;
1790 static int vmci_transport_connect(struct vsock_sock *vsk)
1793 bool old_pkt_proto = false;
1794 struct sock *sk = &vsk->sk;
1796 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1798 err = vmci_transport_send_conn_request(
1799 sk, vmci_trans(vsk)->queue_pair_size);
1801 sk->sk_state = SS_UNCONNECTED;
1805 int supported_proto_versions =
1806 vmci_transport_new_proto_supported_versions();
1807 err = vmci_transport_send_conn_request2(
1808 sk, vmci_trans(vsk)->queue_pair_size,
1809 supported_proto_versions);
1811 sk->sk_state = SS_UNCONNECTED;
1815 vsk->sent_request = true;
1821 static ssize_t vmci_transport_stream_dequeue(
1822 struct vsock_sock *vsk,
1827 if (flags & MSG_PEEK)
1828 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, iov, len, 0);
1830 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, iov, len, 0);
1833 static ssize_t vmci_transport_stream_enqueue(
1834 struct vsock_sock *vsk,
1838 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, iov, len, 0);
1841 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1843 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1846 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1848 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1851 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1853 return vmci_trans(vsk)->consume_size;
1856 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1858 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1861 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1863 return vmci_trans(vsk)->queue_pair_size;
1866 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1868 return vmci_trans(vsk)->queue_pair_min_size;
1871 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1873 return vmci_trans(vsk)->queue_pair_max_size;
1876 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1878 if (val < vmci_trans(vsk)->queue_pair_min_size)
1879 vmci_trans(vsk)->queue_pair_min_size = val;
1880 if (val > vmci_trans(vsk)->queue_pair_max_size)
1881 vmci_trans(vsk)->queue_pair_max_size = val;
1882 vmci_trans(vsk)->queue_pair_size = val;
1885 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1888 if (val > vmci_trans(vsk)->queue_pair_size)
1889 vmci_trans(vsk)->queue_pair_size = val;
1890 vmci_trans(vsk)->queue_pair_min_size = val;
1893 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1896 if (val < vmci_trans(vsk)->queue_pair_size)
1897 vmci_trans(vsk)->queue_pair_size = val;
1898 vmci_trans(vsk)->queue_pair_max_size = val;
1901 static int vmci_transport_notify_poll_in(
1902 struct vsock_sock *vsk,
1904 bool *data_ready_now)
1906 return vmci_trans(vsk)->notify_ops->poll_in(
1907 &vsk->sk, target, data_ready_now);
1910 static int vmci_transport_notify_poll_out(
1911 struct vsock_sock *vsk,
1913 bool *space_available_now)
1915 return vmci_trans(vsk)->notify_ops->poll_out(
1916 &vsk->sk, target, space_available_now);
1919 static int vmci_transport_notify_recv_init(
1920 struct vsock_sock *vsk,
1922 struct vsock_transport_recv_notify_data *data)
1924 return vmci_trans(vsk)->notify_ops->recv_init(
1926 (struct vmci_transport_recv_notify_data *)data);
1929 static int vmci_transport_notify_recv_pre_block(
1930 struct vsock_sock *vsk,
1932 struct vsock_transport_recv_notify_data *data)
1934 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1936 (struct vmci_transport_recv_notify_data *)data);
1939 static int vmci_transport_notify_recv_pre_dequeue(
1940 struct vsock_sock *vsk,
1942 struct vsock_transport_recv_notify_data *data)
1944 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1946 (struct vmci_transport_recv_notify_data *)data);
1949 static int vmci_transport_notify_recv_post_dequeue(
1950 struct vsock_sock *vsk,
1954 struct vsock_transport_recv_notify_data *data)
1956 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1957 &vsk->sk, target, copied, data_read,
1958 (struct vmci_transport_recv_notify_data *)data);
1961 static int vmci_transport_notify_send_init(
1962 struct vsock_sock *vsk,
1963 struct vsock_transport_send_notify_data *data)
1965 return vmci_trans(vsk)->notify_ops->send_init(
1967 (struct vmci_transport_send_notify_data *)data);
1970 static int vmci_transport_notify_send_pre_block(
1971 struct vsock_sock *vsk,
1972 struct vsock_transport_send_notify_data *data)
1974 return vmci_trans(vsk)->notify_ops->send_pre_block(
1976 (struct vmci_transport_send_notify_data *)data);
1979 static int vmci_transport_notify_send_pre_enqueue(
1980 struct vsock_sock *vsk,
1981 struct vsock_transport_send_notify_data *data)
1983 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1985 (struct vmci_transport_send_notify_data *)data);
1988 static int vmci_transport_notify_send_post_enqueue(
1989 struct vsock_sock *vsk,
1991 struct vsock_transport_send_notify_data *data)
1993 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1995 (struct vmci_transport_send_notify_data *)data);
1998 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2000 if (PROTOCOL_OVERRIDE != -1) {
2001 if (PROTOCOL_OVERRIDE == 0)
2002 *old_pkt_proto = true;
2004 *old_pkt_proto = false;
2006 pr_info("Proto override in use\n");
2013 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2017 struct vsock_sock *vsk = vsock_sk(sk);
2019 if (old_pkt_proto) {
2020 if (*proto != VSOCK_PROTO_INVALID) {
2021 pr_err("Can't set both an old and new protocol\n");
2024 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2029 case VSOCK_PROTO_PKT_ON_NOTIFY:
2030 vmci_trans(vsk)->notify_ops =
2031 &vmci_transport_notify_pkt_q_state_ops;
2034 pr_err("Unknown notify protocol version\n");
2039 vmci_trans(vsk)->notify_ops->socket_init(sk);
2043 static u16 vmci_transport_new_proto_supported_versions(void)
2045 if (PROTOCOL_OVERRIDE != -1)
2046 return PROTOCOL_OVERRIDE;
2048 return VSOCK_PROTO_ALL_SUPPORTED;
2051 static u32 vmci_transport_get_local_cid(void)
2053 return vmci_get_context_id();
2056 static struct vsock_transport vmci_transport = {
2057 .init = vmci_transport_socket_init,
2058 .destruct = vmci_transport_destruct,
2059 .release = vmci_transport_release,
2060 .connect = vmci_transport_connect,
2061 .dgram_bind = vmci_transport_dgram_bind,
2062 .dgram_dequeue = vmci_transport_dgram_dequeue,
2063 .dgram_enqueue = vmci_transport_dgram_enqueue,
2064 .dgram_allow = vmci_transport_dgram_allow,
2065 .stream_dequeue = vmci_transport_stream_dequeue,
2066 .stream_enqueue = vmci_transport_stream_enqueue,
2067 .stream_has_data = vmci_transport_stream_has_data,
2068 .stream_has_space = vmci_transport_stream_has_space,
2069 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2070 .stream_is_active = vmci_transport_stream_is_active,
2071 .stream_allow = vmci_transport_stream_allow,
2072 .notify_poll_in = vmci_transport_notify_poll_in,
2073 .notify_poll_out = vmci_transport_notify_poll_out,
2074 .notify_recv_init = vmci_transport_notify_recv_init,
2075 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2076 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2077 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2078 .notify_send_init = vmci_transport_notify_send_init,
2079 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2080 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2081 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2082 .shutdown = vmci_transport_shutdown,
2083 .set_buffer_size = vmci_transport_set_buffer_size,
2084 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2085 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2086 .get_buffer_size = vmci_transport_get_buffer_size,
2087 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2088 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2089 .get_local_cid = vmci_transport_get_local_cid,
2092 static int __init vmci_transport_init(void)
2096 /* Create the datagram handle that we will use to send and receive all
2097 * VSocket control messages for this context.
2099 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2100 VMCI_FLAG_ANYCID_DG_HND,
2101 vmci_transport_recv_stream_cb,
2103 &vmci_transport_stream_handle);
2104 if (err < VMCI_SUCCESS) {
2105 pr_err("Unable to create datagram handle. (%d)\n", err);
2106 return vmci_transport_error_to_vsock_error(err);
2109 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2110 vmci_transport_qp_resumed_cb,
2111 NULL, &vmci_transport_qp_resumed_sub_id);
2112 if (err < VMCI_SUCCESS) {
2113 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2114 err = vmci_transport_error_to_vsock_error(err);
2115 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2116 goto err_destroy_stream_handle;
2119 err = vsock_core_init(&vmci_transport);
2121 goto err_unsubscribe;
2126 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2127 err_destroy_stream_handle:
2128 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2131 module_init(vmci_transport_init);
2133 static void __exit vmci_transport_exit(void)
2135 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2136 if (vmci_datagram_destroy_handle(
2137 vmci_transport_stream_handle) != VMCI_SUCCESS)
2138 pr_err("Couldn't destroy datagram handle\n");
2139 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2142 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2143 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2144 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2149 module_exit(vmci_transport_exit);
2151 MODULE_AUTHOR("VMware, Inc.");
2152 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2153 MODULE_LICENSE("GPL v2");
2154 MODULE_ALIAS("vmware_vsock");
2155 MODULE_ALIAS_NETPROTO(PF_VSOCK);