1 // SPDX-License-Identifier: GPL-2.0-only
3 * VMware vSockets Driver
5 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
8 #include <linux/types.h>
9 #include <linux/bitops.h>
10 #include <linux/cred.h>
11 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/kmod.h>
15 #include <linux/list.h>
16 #include <linux/module.h>
17 #include <linux/mutex.h>
18 #include <linux/net.h>
19 #include <linux/poll.h>
20 #include <linux/skbuff.h>
21 #include <linux/smp.h>
22 #include <linux/socket.h>
23 #include <linux/stddef.h>
24 #include <linux/unistd.h>
25 #include <linux/wait.h>
26 #include <linux/workqueue.h>
28 #include <net/af_vsock.h>
30 #include "vmci_transport_notify.h"
32 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
33 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
34 static void vmci_transport_peer_detach_cb(u32 sub_id,
35 const struct vmci_event_data *ed,
37 static void vmci_transport_recv_pkt_work(struct work_struct *work);
38 static void vmci_transport_cleanup(struct work_struct *work);
39 static int vmci_transport_recv_listen(struct sock *sk,
40 struct vmci_transport_packet *pkt);
41 static int vmci_transport_recv_connecting_server(
44 struct vmci_transport_packet *pkt);
45 static int vmci_transport_recv_connecting_client(
47 struct vmci_transport_packet *pkt);
48 static int vmci_transport_recv_connecting_client_negotiate(
50 struct vmci_transport_packet *pkt);
51 static int vmci_transport_recv_connecting_client_invalid(
53 struct vmci_transport_packet *pkt);
54 static int vmci_transport_recv_connected(struct sock *sk,
55 struct vmci_transport_packet *pkt);
56 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
57 static u16 vmci_transport_new_proto_supported_versions(void);
58 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
61 struct vmci_transport_recv_pkt_info {
62 struct work_struct work;
64 struct vmci_transport_packet pkt;
67 static LIST_HEAD(vmci_transport_cleanup_list);
68 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
69 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
71 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
73 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
75 static int PROTOCOL_OVERRIDE = -1;
77 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN 128
78 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE 262144
79 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX 262144
81 /* The default peer timeout indicates how long we will wait for a peer response
82 * to a control message.
84 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
86 /* Helper function to convert from a VMCI error code to a VSock error code. */
88 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
91 case VMCI_ERROR_NO_MEM:
93 case VMCI_ERROR_DUPLICATE_ENTRY:
94 case VMCI_ERROR_ALREADY_EXISTS:
96 case VMCI_ERROR_NO_ACCESS:
98 case VMCI_ERROR_NO_RESOURCES:
100 case VMCI_ERROR_INVALID_RESOURCE:
101 return -EHOSTUNREACH;
102 case VMCI_ERROR_INVALID_ARGS:
109 static u32 vmci_transport_peer_rid(u32 peer_cid)
111 if (VMADDR_CID_HYPERVISOR == peer_cid)
112 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
114 return VMCI_TRANSPORT_PACKET_RID;
118 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
119 struct sockaddr_vm *src,
120 struct sockaddr_vm *dst,
124 struct vmci_transport_waiting_info *wait,
126 struct vmci_handle handle)
128 /* We register the stream control handler as an any cid handle so we
129 * must always send from a source address of VMADDR_CID_ANY
131 pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
132 VMCI_TRANSPORT_PACKET_RID);
133 pkt->dg.dst = vmci_make_handle(dst->svm_cid,
134 vmci_transport_peer_rid(dst->svm_cid));
135 pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
136 pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
138 pkt->src_port = src->svm_port;
139 pkt->dst_port = dst->svm_port;
140 memset(&pkt->proto, 0, sizeof(pkt->proto));
141 memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
144 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
148 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
149 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
153 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
154 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
155 pkt->u.handle = handle;
158 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
159 case VMCI_TRANSPORT_PACKET_TYPE_READ:
160 case VMCI_TRANSPORT_PACKET_TYPE_RST:
164 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
168 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
169 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
170 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
173 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
174 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
182 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
183 struct sockaddr_vm *local,
184 struct sockaddr_vm *remote)
186 vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
187 vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
191 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
192 struct sockaddr_vm *src,
193 struct sockaddr_vm *dst,
194 enum vmci_transport_packet_type type,
197 struct vmci_transport_waiting_info *wait,
199 struct vmci_handle handle,
204 vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
206 err = vmci_datagram_send(&pkt->dg);
207 if (convert_error && (err < 0))
208 return vmci_transport_error_to_vsock_error(err);
214 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
215 enum vmci_transport_packet_type type,
218 struct vmci_transport_waiting_info *wait,
219 struct vmci_handle handle)
221 struct vmci_transport_packet reply;
222 struct sockaddr_vm src, dst;
224 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
227 vmci_transport_packet_get_addresses(pkt, &src, &dst);
228 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
237 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
238 struct sockaddr_vm *dst,
239 enum vmci_transport_packet_type type,
242 struct vmci_transport_waiting_info *wait,
243 struct vmci_handle handle)
245 /* Note that it is safe to use a single packet across all CPUs since
246 * two tasklets of the same type are guaranteed to not ever run
247 * simultaneously. If that ever changes, or VMCI stops using tasklets,
248 * we can use per-cpu packets.
250 static struct vmci_transport_packet pkt;
252 return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
254 VSOCK_PROTO_INVALID, handle,
259 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
260 struct sockaddr_vm *dst,
261 enum vmci_transport_packet_type type,
264 struct vmci_transport_waiting_info *wait,
266 struct vmci_handle handle)
268 struct vmci_transport_packet *pkt;
271 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
275 err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
276 mode, wait, proto, handle,
284 vmci_transport_send_control_pkt(struct sock *sk,
285 enum vmci_transport_packet_type type,
288 struct vmci_transport_waiting_info *wait,
290 struct vmci_handle handle)
292 struct vsock_sock *vsk;
296 if (!vsock_addr_bound(&vsk->local_addr))
299 if (!vsock_addr_bound(&vsk->remote_addr))
302 return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
305 wait, proto, handle);
308 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
309 struct sockaddr_vm *src,
310 struct vmci_transport_packet *pkt)
312 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
314 return vmci_transport_send_control_pkt_bh(
316 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
317 0, NULL, VMCI_INVALID_HANDLE);
320 static int vmci_transport_send_reset(struct sock *sk,
321 struct vmci_transport_packet *pkt)
323 struct sockaddr_vm *dst_ptr;
324 struct sockaddr_vm dst;
325 struct vsock_sock *vsk;
327 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
332 if (!vsock_addr_bound(&vsk->local_addr))
335 if (vsock_addr_bound(&vsk->remote_addr)) {
336 dst_ptr = &vsk->remote_addr;
338 vsock_addr_init(&dst, pkt->dg.src.context,
342 return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
343 VMCI_TRANSPORT_PACKET_TYPE_RST,
344 0, 0, NULL, VSOCK_PROTO_INVALID,
345 VMCI_INVALID_HANDLE);
348 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
350 return vmci_transport_send_control_pkt(
352 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
355 VMCI_INVALID_HANDLE);
358 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
361 return vmci_transport_send_control_pkt(
363 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
364 size, 0, NULL, version,
365 VMCI_INVALID_HANDLE);
368 static int vmci_transport_send_qp_offer(struct sock *sk,
369 struct vmci_handle handle)
371 return vmci_transport_send_control_pkt(
372 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
374 VSOCK_PROTO_INVALID, handle);
377 static int vmci_transport_send_attach(struct sock *sk,
378 struct vmci_handle handle)
380 return vmci_transport_send_control_pkt(
381 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
382 0, 0, NULL, VSOCK_PROTO_INVALID,
386 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
388 return vmci_transport_reply_control_pkt_fast(
390 VMCI_TRANSPORT_PACKET_TYPE_RST,
392 VMCI_INVALID_HANDLE);
395 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
396 struct sockaddr_vm *src)
398 return vmci_transport_send_control_pkt_bh(
400 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
401 0, 0, NULL, VMCI_INVALID_HANDLE);
404 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
405 struct sockaddr_vm *src)
407 return vmci_transport_send_control_pkt_bh(
409 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
410 0, NULL, VMCI_INVALID_HANDLE);
413 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
414 struct sockaddr_vm *src)
416 return vmci_transport_send_control_pkt_bh(
418 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
419 0, NULL, VMCI_INVALID_HANDLE);
422 int vmci_transport_send_wrote(struct sock *sk)
424 return vmci_transport_send_control_pkt(
425 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
426 0, NULL, VSOCK_PROTO_INVALID,
427 VMCI_INVALID_HANDLE);
430 int vmci_transport_send_read(struct sock *sk)
432 return vmci_transport_send_control_pkt(
433 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
434 0, NULL, VSOCK_PROTO_INVALID,
435 VMCI_INVALID_HANDLE);
438 int vmci_transport_send_waiting_write(struct sock *sk,
439 struct vmci_transport_waiting_info *wait)
441 return vmci_transport_send_control_pkt(
442 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
443 0, 0, wait, VSOCK_PROTO_INVALID,
444 VMCI_INVALID_HANDLE);
447 int vmci_transport_send_waiting_read(struct sock *sk,
448 struct vmci_transport_waiting_info *wait)
450 return vmci_transport_send_control_pkt(
451 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
452 0, 0, wait, VSOCK_PROTO_INVALID,
453 VMCI_INVALID_HANDLE);
456 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
458 return vmci_transport_send_control_pkt(
460 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
463 VMCI_INVALID_HANDLE);
466 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
468 return vmci_transport_send_control_pkt(sk,
469 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
472 VMCI_INVALID_HANDLE);
475 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
478 return vmci_transport_send_control_pkt(
479 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
480 size, 0, NULL, version,
481 VMCI_INVALID_HANDLE);
484 static struct sock *vmci_transport_get_pending(
485 struct sock *listener,
486 struct vmci_transport_packet *pkt)
488 struct vsock_sock *vlistener;
489 struct vsock_sock *vpending;
490 struct sock *pending;
491 struct sockaddr_vm src;
493 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
495 vlistener = vsock_sk(listener);
497 list_for_each_entry(vpending, &vlistener->pending_links,
499 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
500 pkt->dst_port == vpending->local_addr.svm_port) {
501 pending = sk_vsock(vpending);
513 static void vmci_transport_release_pending(struct sock *pending)
518 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
519 * trusted sockets 2) sockets from applications running as the same user as the
520 * VM (this is only true for the host side and only when using hosted products)
523 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
525 return vsock->trusted ||
526 vmci_is_context_owner(peer_cid, vsock->owner->uid);
529 /* We allow sending datagrams to and receiving datagrams from a restricted VM
530 * only if it is trusted as described in vmci_transport_is_trusted.
533 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
535 if (VMADDR_CID_HYPERVISOR == peer_cid)
538 if (vsock->cached_peer != peer_cid) {
539 vsock->cached_peer = peer_cid;
540 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
541 (vmci_context_get_priv_flags(peer_cid) &
542 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
543 vsock->cached_peer_allow_dgram = false;
545 vsock->cached_peer_allow_dgram = true;
549 return vsock->cached_peer_allow_dgram;
553 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
554 struct vmci_handle *handle,
557 u32 peer, u32 flags, bool trusted)
562 /* Try to allocate our queue pair as trusted. This will only
563 * work if vsock is running in the host.
566 err = vmci_qpair_alloc(qpair, handle, produce_size,
569 VMCI_PRIVILEGE_FLAG_TRUSTED);
570 if (err != VMCI_ERROR_NO_ACCESS)
575 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
576 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
579 pr_err("Could not attach to queue pair with %d\n",
581 err = vmci_transport_error_to_vsock_error(err);
588 vmci_transport_datagram_create_hnd(u32 resource_id,
590 vmci_datagram_recv_cb recv_cb,
592 struct vmci_handle *out_handle)
596 /* Try to allocate our datagram handler as trusted. This will only work
597 * if vsock is running in the host.
600 err = vmci_datagram_create_handle_priv(resource_id, flags,
601 VMCI_PRIVILEGE_FLAG_TRUSTED,
603 client_data, out_handle);
605 if (err == VMCI_ERROR_NO_ACCESS)
606 err = vmci_datagram_create_handle(resource_id, flags,
607 recv_cb, client_data,
613 /* This is invoked as part of a tasklet that's scheduled when the VMCI
614 * interrupt fires. This is run in bottom-half context and if it ever needs to
615 * sleep it should defer that work to a work queue.
618 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
623 struct vsock_sock *vsk;
625 sk = (struct sock *)data;
627 /* This handler is privileged when this module is running on the host.
628 * We will get datagrams from all endpoints (even VMs that are in a
629 * restricted context). If we get one from a restricted context then
630 * the destination socket must be trusted.
632 * NOTE: We access the socket struct without holding the lock here.
633 * This is ok because the field we are interested is never modified
634 * outside of the create and destruct socket functions.
637 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
638 return VMCI_ERROR_NO_ACCESS;
640 size = VMCI_DG_SIZE(dg);
642 /* Attach the packet to the socket's receive queue as an sk_buff. */
643 skb = alloc_skb(size, GFP_ATOMIC);
645 return VMCI_ERROR_NO_MEM;
647 /* sk_receive_skb() will do a sock_put(), so hold here. */
650 memcpy(skb->data, dg, size);
651 sk_receive_skb(sk, skb, 0);
656 static bool vmci_transport_stream_allow(u32 cid, u32 port)
658 static const u32 non_socket_contexts[] = {
663 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
665 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
666 if (cid == non_socket_contexts[i])
673 /* This is invoked as part of a tasklet that's scheduled when the VMCI
674 * interrupt fires. This is run in bottom-half context but it defers most of
675 * its work to the packet handling work queue.
678 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
681 struct sockaddr_vm dst;
682 struct sockaddr_vm src;
683 struct vmci_transport_packet *pkt;
684 struct vsock_sock *vsk;
690 bh_process_pkt = false;
692 /* Ignore incoming packets from contexts without sockets, or resources
693 * that aren't vsock implementations.
696 if (!vmci_transport_stream_allow(dg->src.context, -1)
697 || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
698 return VMCI_ERROR_NO_ACCESS;
700 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
701 /* Drop datagrams that do not contain full VSock packets. */
702 return VMCI_ERROR_INVALID_ARGS;
704 pkt = (struct vmci_transport_packet *)dg;
706 /* Find the socket that should handle this packet. First we look for a
707 * connected socket and if there is none we look for a socket bound to
708 * the destintation address.
710 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
711 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
713 sk = vsock_find_connected_socket(&src, &dst);
715 sk = vsock_find_bound_socket(&dst);
717 /* We could not find a socket for this specified
718 * address. If this packet is a RST, we just drop it.
719 * If it is another packet, we send a RST. Note that
720 * we do not send a RST reply to RSTs so that we do not
721 * continually send RSTs between two endpoints.
723 * Note that since this is a reply, dst is src and src
726 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
727 pr_err("unable to send reset\n");
729 err = VMCI_ERROR_NOT_FOUND;
734 /* If the received packet type is beyond all types known to this
735 * implementation, reply with an invalid message. Hopefully this will
736 * help when implementing backwards compatibility in the future.
738 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
739 vmci_transport_send_invalid_bh(&dst, &src);
740 err = VMCI_ERROR_INVALID_ARGS;
744 /* This handler is privileged when this module is running on the host.
745 * We will get datagram connect requests from all endpoints (even VMs
746 * that are in a restricted context). If we get one from a restricted
747 * context then the destination socket must be trusted.
749 * NOTE: We access the socket struct without holding the lock here.
750 * This is ok because the field we are interested is never modified
751 * outside of the create and destruct socket functions.
754 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
755 err = VMCI_ERROR_NO_ACCESS;
759 /* We do most everything in a work queue, but let's fast path the
760 * notification of reads and writes to help data transfer performance.
761 * We can only do this if there is no process context code executing
762 * for this socket since that may change the state.
766 if (!sock_owned_by_user(sk)) {
767 /* The local context ID may be out of date, update it. */
768 vsk->local_addr.svm_cid = dst.svm_cid;
770 if (sk->sk_state == TCP_ESTABLISHED)
771 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
772 sk, pkt, true, &dst, &src,
778 if (!bh_process_pkt) {
779 struct vmci_transport_recv_pkt_info *recv_pkt_info;
781 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
782 if (!recv_pkt_info) {
783 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
784 pr_err("unable to send reset\n");
786 err = VMCI_ERROR_NO_MEM;
790 recv_pkt_info->sk = sk;
791 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
792 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
794 schedule_work(&recv_pkt_info->work);
795 /* Clear sk so that the reference count incremented by one of
796 * the Find functions above is not decremented below. We need
797 * that reference count for the packet handler we've scheduled
810 static void vmci_transport_handle_detach(struct sock *sk)
812 struct vsock_sock *vsk;
815 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
816 sock_set_flag(sk, SOCK_DONE);
818 /* On a detach the peer will not be sending or receiving
821 vsk->peer_shutdown = SHUTDOWN_MASK;
823 /* We should not be sending anymore since the peer won't be
824 * there to receive, but we can still receive if there is data
825 * left in our consume queue. If the local endpoint is a host,
826 * we can't call vsock_stream_has_data, since that may block,
827 * but a host endpoint can't read data once the VM has
828 * detached, so there is no available data in that case.
830 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
831 vsock_stream_has_data(vsk) <= 0) {
832 if (sk->sk_state == TCP_SYN_SENT) {
833 /* The peer may detach from a queue pair while
834 * we are still in the connecting state, i.e.,
835 * if the peer VM is killed after attaching to
836 * a queue pair, but before we complete the
837 * handshake. In that case, we treat the detach
838 * event like a reset.
841 sk->sk_state = TCP_CLOSE;
842 sk->sk_err = ECONNRESET;
843 sk->sk_error_report(sk);
846 sk->sk_state = TCP_CLOSE;
848 sk->sk_state_change(sk);
852 static void vmci_transport_peer_detach_cb(u32 sub_id,
853 const struct vmci_event_data *e_data,
856 struct vmci_transport *trans = client_data;
857 const struct vmci_event_payload_qp *e_payload;
859 e_payload = vmci_event_data_const_payload(e_data);
861 /* XXX This is lame, we should provide a way to lookup sockets by
864 if (vmci_handle_is_invalid(e_payload->handle) ||
865 !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
868 /* We don't ask for delayed CBs when we subscribe to this event (we
869 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
870 * guarantees in that case about what context we might be running in,
871 * so it could be BH or process, blockable or non-blockable. So we
872 * need to account for all possible contexts here.
874 spin_lock_bh(&trans->lock);
878 /* Apart from here, trans->lock is only grabbed as part of sk destruct,
879 * where trans->sk isn't locked.
881 bh_lock_sock(trans->sk);
883 vmci_transport_handle_detach(trans->sk);
885 bh_unlock_sock(trans->sk);
887 spin_unlock_bh(&trans->lock);
890 static void vmci_transport_qp_resumed_cb(u32 sub_id,
891 const struct vmci_event_data *e_data,
894 vsock_for_each_connected_socket(vmci_transport_handle_detach);
897 static void vmci_transport_recv_pkt_work(struct work_struct *work)
899 struct vmci_transport_recv_pkt_info *recv_pkt_info;
900 struct vmci_transport_packet *pkt;
904 container_of(work, struct vmci_transport_recv_pkt_info, work);
905 sk = recv_pkt_info->sk;
906 pkt = &recv_pkt_info->pkt;
910 /* The local context ID may be out of date. */
911 vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
913 switch (sk->sk_state) {
915 vmci_transport_recv_listen(sk, pkt);
918 /* Processing of pending connections for servers goes through
919 * the listening socket, so see vmci_transport_recv_listen()
922 vmci_transport_recv_connecting_client(sk, pkt);
924 case TCP_ESTABLISHED:
925 vmci_transport_recv_connected(sk, pkt);
928 /* Because this function does not run in the same context as
929 * vmci_transport_recv_stream_cb it is possible that the
930 * socket has closed. We need to let the other side know or it
931 * could be sitting in a connect and hang forever. Send a
932 * reset to prevent that.
934 vmci_transport_send_reset(sk, pkt);
939 kfree(recv_pkt_info);
940 /* Release reference obtained in the stream callback when we fetched
941 * this socket out of the bound or connected list.
946 static int vmci_transport_recv_listen(struct sock *sk,
947 struct vmci_transport_packet *pkt)
949 struct sock *pending;
950 struct vsock_sock *vpending;
953 bool old_request = false;
954 bool old_pkt_proto = false;
958 /* Because we are in the listen state, we could be receiving a packet
959 * for ourself or any previous connection requests that we received.
960 * If it's the latter, we try to find a socket in our list of pending
961 * connections and, if we do, call the appropriate handler for the
962 * state that that socket is in. Otherwise we try to service the
963 * connection request.
965 pending = vmci_transport_get_pending(sk, pkt);
969 /* The local context ID may be out of date. */
970 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
972 switch (pending->sk_state) {
974 err = vmci_transport_recv_connecting_server(sk,
979 vmci_transport_send_reset(pending, pkt);
984 vsock_remove_pending(sk, pending);
986 release_sock(pending);
987 vmci_transport_release_pending(pending);
992 /* The listen state only accepts connection requests. Reply with a
993 * reset unless we received a reset.
996 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
997 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
998 vmci_transport_reply_reset(pkt);
1002 if (pkt->u.size == 0) {
1003 vmci_transport_reply_reset(pkt);
1007 /* If this socket can't accommodate this connection request, we send a
1008 * reset. Otherwise we create and initialize a child socket and reply
1009 * with a connection negotiation.
1011 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1012 vmci_transport_reply_reset(pkt);
1013 return -ECONNREFUSED;
1016 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1019 vmci_transport_send_reset(sk, pkt);
1023 vpending = vsock_sk(pending);
1025 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1027 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1030 /* If the proposed size fits within our min/max, accept it. Otherwise
1031 * propose our own size.
1033 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1034 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1035 qp_size = pkt->u.size;
1037 qp_size = vmci_trans(vpending)->queue_pair_size;
1040 /* Figure out if we are using old or new requests based on the
1041 * overrides pkt types sent by our peer.
1043 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1044 old_request = old_pkt_proto;
1046 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1048 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1049 old_request = false;
1054 /* Handle a REQUEST (or override) */
1055 u16 version = VSOCK_PROTO_INVALID;
1056 if (vmci_transport_proto_to_notify_struct(
1057 pending, &version, true))
1058 err = vmci_transport_send_negotiate(pending, qp_size);
1063 /* Handle a REQUEST2 (or override) */
1064 int proto_int = pkt->proto;
1066 u16 active_proto_version = 0;
1068 /* The list of possible protocols is the intersection of all
1069 * protocols the client supports ... plus all the protocols we
1072 proto_int &= vmci_transport_new_proto_supported_versions();
1074 /* We choose the highest possible protocol version and use that
1077 pos = fls(proto_int);
1079 active_proto_version = (1 << (pos - 1));
1080 if (vmci_transport_proto_to_notify_struct(
1081 pending, &active_proto_version, false))
1082 err = vmci_transport_send_negotiate2(pending,
1084 active_proto_version);
1094 vmci_transport_send_reset(sk, pkt);
1096 err = vmci_transport_error_to_vsock_error(err);
1100 vsock_add_pending(sk, pending);
1101 sk->sk_ack_backlog++;
1103 pending->sk_state = TCP_SYN_SENT;
1104 vmci_trans(vpending)->produce_size =
1105 vmci_trans(vpending)->consume_size = qp_size;
1106 vmci_trans(vpending)->queue_pair_size = qp_size;
1108 vmci_trans(vpending)->notify_ops->process_request(pending);
1110 /* We might never receive another message for this socket and it's not
1111 * connected to any process, so we have to ensure it gets cleaned up
1112 * ourself. Our delayed work function will take care of that. Note
1113 * that we do not ever cancel this function since we have few
1114 * guarantees about its state when calling cancel_delayed_work().
1115 * Instead we hold a reference on the socket for that function and make
1116 * it capable of handling cases where it needs to do nothing but
1117 * release that reference.
1119 vpending->listener = sk;
1122 schedule_delayed_work(&vpending->pending_work, HZ);
1129 vmci_transport_recv_connecting_server(struct sock *listener,
1130 struct sock *pending,
1131 struct vmci_transport_packet *pkt)
1133 struct vsock_sock *vpending;
1134 struct vmci_handle handle;
1135 struct vmci_qp *qpair;
1142 vpending = vsock_sk(pending);
1143 detach_sub_id = VMCI_INVALID_ID;
1145 switch (pkt->type) {
1146 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1147 if (vmci_handle_is_invalid(pkt->u.handle)) {
1148 vmci_transport_send_reset(pending, pkt);
1155 /* Close and cleanup the connection. */
1156 vmci_transport_send_reset(pending, pkt);
1158 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1162 /* In order to complete the connection we need to attach to the offered
1163 * queue pair and send an attach notification. We also subscribe to the
1164 * detach event so we know when our peer goes away, and we do that
1165 * before attaching so we don't miss an event. If all this succeeds,
1166 * we update our state and wakeup anything waiting in accept() for a
1170 /* We don't care about attach since we ensure the other side has
1171 * attached by specifying the ATTACH_ONLY flag below.
1173 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1174 vmci_transport_peer_detach_cb,
1175 vmci_trans(vpending), &detach_sub_id);
1176 if (err < VMCI_SUCCESS) {
1177 vmci_transport_send_reset(pending, pkt);
1178 err = vmci_transport_error_to_vsock_error(err);
1183 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1185 /* Now attach to the queue pair the client created. */
1186 handle = pkt->u.handle;
1188 /* vpending->local_addr always has a context id so we do not need to
1189 * worry about VMADDR_CID_ANY in this case.
1192 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1193 flags = VMCI_QPFLAG_ATTACH_ONLY;
1194 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1196 err = vmci_transport_queue_pair_alloc(
1199 vmci_trans(vpending)->produce_size,
1200 vmci_trans(vpending)->consume_size,
1201 pkt->dg.src.context,
1203 vmci_transport_is_trusted(
1205 vpending->remote_addr.svm_cid));
1207 vmci_transport_send_reset(pending, pkt);
1212 vmci_trans(vpending)->qp_handle = handle;
1213 vmci_trans(vpending)->qpair = qpair;
1215 /* When we send the attach message, we must be ready to handle incoming
1216 * control messages on the newly connected socket. So we move the
1217 * pending socket to the connected state before sending the attach
1218 * message. Otherwise, an incoming packet triggered by the attach being
1219 * received by the peer may be processed concurrently with what happens
1220 * below after sending the attach message, and that incoming packet
1221 * will find the listening socket instead of the (currently) pending
1222 * socket. Note that enqueueing the socket increments the reference
1223 * count, so even if a reset comes before the connection is accepted,
1224 * the socket will be valid until it is removed from the queue.
1226 * If we fail sending the attach below, we remove the socket from the
1227 * connected list and move the socket to TCP_CLOSE before
1228 * releasing the lock, so a pending slow path processing of an incoming
1229 * packet will not see the socket in the connected state in that case.
1231 pending->sk_state = TCP_ESTABLISHED;
1233 vsock_insert_connected(vpending);
1235 /* Notify our peer of our attach. */
1236 err = vmci_transport_send_attach(pending, handle);
1238 vsock_remove_connected(vpending);
1239 pr_err("Could not send attach\n");
1240 vmci_transport_send_reset(pending, pkt);
1241 err = vmci_transport_error_to_vsock_error(err);
1246 /* We have a connection. Move the now connected socket from the
1247 * listener's pending list to the accept queue so callers of accept()
1250 vsock_remove_pending(listener, pending);
1251 vsock_enqueue_accept(listener, pending);
1253 /* Callers of accept() will be be waiting on the listening socket, not
1254 * the pending socket.
1256 listener->sk_data_ready(listener);
1261 pending->sk_err = skerr;
1262 pending->sk_state = TCP_CLOSE;
1263 /* As long as we drop our reference, all necessary cleanup will handle
1264 * when the cleanup function drops its reference and our destruct
1265 * implementation is called. Note that since the listen handler will
1266 * remove pending from the pending list upon our failure, the cleanup
1267 * function won't drop the additional reference, which is why we do it
1276 vmci_transport_recv_connecting_client(struct sock *sk,
1277 struct vmci_transport_packet *pkt)
1279 struct vsock_sock *vsk;
1285 switch (pkt->type) {
1286 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1287 if (vmci_handle_is_invalid(pkt->u.handle) ||
1288 !vmci_handle_is_equal(pkt->u.handle,
1289 vmci_trans(vsk)->qp_handle)) {
1295 /* Signify the socket is connected and wakeup the waiter in
1296 * connect(). Also place the socket in the connected table for
1297 * accounting (it can already be found since it's in the bound
1300 sk->sk_state = TCP_ESTABLISHED;
1301 sk->sk_socket->state = SS_CONNECTED;
1302 vsock_insert_connected(vsk);
1303 sk->sk_state_change(sk);
1306 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1307 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1308 if (pkt->u.size == 0
1309 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1310 || pkt->src_port != vsk->remote_addr.svm_port
1311 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1312 || vmci_trans(vsk)->qpair
1313 || vmci_trans(vsk)->produce_size != 0
1314 || vmci_trans(vsk)->consume_size != 0
1315 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1322 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1329 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1330 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1337 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1338 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1339 * continue processing here after they sent an INVALID packet.
1340 * This meant that we got a RST after the INVALID. We ignore a
1341 * RST after an INVALID. The common code doesn't send the RST
1342 * ... so we can hang if an old version of the common code
1343 * fails between getting a REQUEST and sending an OFFER back.
1344 * Not much we can do about it... except hope that it doesn't
1347 if (vsk->ignore_connecting_rst) {
1348 vsk->ignore_connecting_rst = false;
1357 /* Close and cleanup the connection. */
1366 vmci_transport_send_reset(sk, pkt);
1368 sk->sk_state = TCP_CLOSE;
1370 sk->sk_error_report(sk);
1374 static int vmci_transport_recv_connecting_client_negotiate(
1376 struct vmci_transport_packet *pkt)
1379 struct vsock_sock *vsk;
1380 struct vmci_handle handle;
1381 struct vmci_qp *qpair;
1385 bool old_proto = true;
1390 handle = VMCI_INVALID_HANDLE;
1391 detach_sub_id = VMCI_INVALID_ID;
1393 /* If we have gotten here then we should be past the point where old
1394 * linux vsock could have sent the bogus rst.
1396 vsk->sent_request = false;
1397 vsk->ignore_connecting_rst = false;
1399 /* Verify that we're OK with the proposed queue pair size */
1400 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1401 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1406 /* At this point we know the CID the peer is using to talk to us. */
1408 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1409 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1411 /* Setup the notify ops to be the highest supported version that both
1412 * the server and the client support.
1415 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1416 old_proto = old_pkt_proto;
1418 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1420 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1426 version = VSOCK_PROTO_INVALID;
1428 version = pkt->proto;
1430 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1435 /* Subscribe to detach events first.
1437 * XXX We attach once for each queue pair created for now so it is easy
1438 * to find the socket (it's provided), but later we should only
1439 * subscribe once and add a way to lookup sockets by queue pair handle.
1441 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1442 vmci_transport_peer_detach_cb,
1443 vmci_trans(vsk), &detach_sub_id);
1444 if (err < VMCI_SUCCESS) {
1445 err = vmci_transport_error_to_vsock_error(err);
1449 /* Make VMCI select the handle for us. */
1450 handle = VMCI_INVALID_HANDLE;
1451 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1452 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1454 err = vmci_transport_queue_pair_alloc(&qpair,
1458 vsk->remote_addr.svm_cid,
1460 vmci_transport_is_trusted(
1463 remote_addr.svm_cid));
1467 err = vmci_transport_send_qp_offer(sk, handle);
1469 err = vmci_transport_error_to_vsock_error(err);
1473 vmci_trans(vsk)->qp_handle = handle;
1474 vmci_trans(vsk)->qpair = qpair;
1476 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1479 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1481 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1486 if (detach_sub_id != VMCI_INVALID_ID)
1487 vmci_event_unsubscribe(detach_sub_id);
1489 if (!vmci_handle_is_invalid(handle))
1490 vmci_qpair_detach(&qpair);
1496 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1497 struct vmci_transport_packet *pkt)
1500 struct vsock_sock *vsk = vsock_sk(sk);
1502 if (vsk->sent_request) {
1503 vsk->sent_request = false;
1504 vsk->ignore_connecting_rst = true;
1506 err = vmci_transport_send_conn_request(
1507 sk, vmci_trans(vsk)->queue_pair_size);
1509 err = vmci_transport_error_to_vsock_error(err);
1518 static int vmci_transport_recv_connected(struct sock *sk,
1519 struct vmci_transport_packet *pkt)
1521 struct vsock_sock *vsk;
1522 bool pkt_processed = false;
1524 /* In cases where we are closing the connection, it's sufficient to
1525 * mark the state change (and maybe error) and wake up any waiting
1526 * threads. Since this is a connected socket, it's owned by a user
1527 * process and will be cleaned up when the failure is passed back on
1528 * the current or next system call. Our system call implementations
1529 * must therefore check for error and state changes on entry and when
1532 switch (pkt->type) {
1533 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1537 vsk->peer_shutdown |= pkt->u.mode;
1538 sk->sk_state_change(sk);
1542 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1544 /* It is possible that we sent our peer a message (e.g a
1545 * WAITING_READ) right before we got notified that the peer had
1546 * detached. If that happens then we can get a RST pkt back
1547 * from our peer even though there is data available for us to
1548 * read. In that case, don't shutdown the socket completely but
1549 * instead allow the local client to finish reading data off
1550 * the queuepair. Always treat a RST pkt in connected mode like
1553 sock_set_flag(sk, SOCK_DONE);
1554 vsk->peer_shutdown = SHUTDOWN_MASK;
1555 if (vsock_stream_has_data(vsk) <= 0)
1556 sk->sk_state = TCP_CLOSING;
1558 sk->sk_state_change(sk);
1563 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1564 sk, pkt, false, NULL, NULL,
1575 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1576 struct vsock_sock *psk)
1578 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1582 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1583 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1584 vmci_trans(vsk)->qpair = NULL;
1585 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1586 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1587 vmci_trans(vsk)->notify_ops = NULL;
1588 INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1589 vmci_trans(vsk)->sk = &vsk->sk;
1590 spin_lock_init(&vmci_trans(vsk)->lock);
1592 vmci_trans(vsk)->queue_pair_size =
1593 vmci_trans(psk)->queue_pair_size;
1594 vmci_trans(vsk)->queue_pair_min_size =
1595 vmci_trans(psk)->queue_pair_min_size;
1596 vmci_trans(vsk)->queue_pair_max_size =
1597 vmci_trans(psk)->queue_pair_max_size;
1599 vmci_trans(vsk)->queue_pair_size =
1600 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1601 vmci_trans(vsk)->queue_pair_min_size =
1602 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1603 vmci_trans(vsk)->queue_pair_max_size =
1604 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1610 static void vmci_transport_free_resources(struct list_head *transport_list)
1612 while (!list_empty(transport_list)) {
1613 struct vmci_transport *transport =
1614 list_first_entry(transport_list, struct vmci_transport,
1616 list_del(&transport->elem);
1618 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1619 vmci_event_unsubscribe(transport->detach_sub_id);
1620 transport->detach_sub_id = VMCI_INVALID_ID;
1623 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1624 vmci_qpair_detach(&transport->qpair);
1625 transport->qp_handle = VMCI_INVALID_HANDLE;
1626 transport->produce_size = 0;
1627 transport->consume_size = 0;
1634 static void vmci_transport_cleanup(struct work_struct *work)
1638 spin_lock_bh(&vmci_transport_cleanup_lock);
1639 list_replace_init(&vmci_transport_cleanup_list, &pending);
1640 spin_unlock_bh(&vmci_transport_cleanup_lock);
1641 vmci_transport_free_resources(&pending);
1644 static void vmci_transport_destruct(struct vsock_sock *vsk)
1646 /* transport can be NULL if we hit a failure at init() time */
1647 if (!vmci_trans(vsk))
1650 /* Ensure that the detach callback doesn't use the sk/vsk
1651 * we are about to destruct.
1653 spin_lock_bh(&vmci_trans(vsk)->lock);
1654 vmci_trans(vsk)->sk = NULL;
1655 spin_unlock_bh(&vmci_trans(vsk)->lock);
1657 if (vmci_trans(vsk)->notify_ops)
1658 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1660 spin_lock_bh(&vmci_transport_cleanup_lock);
1661 list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1662 spin_unlock_bh(&vmci_transport_cleanup_lock);
1663 schedule_work(&vmci_transport_cleanup_work);
1668 static void vmci_transport_release(struct vsock_sock *vsk)
1670 vsock_remove_sock(vsk);
1672 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1673 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1674 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1678 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1679 struct sockaddr_vm *addr)
1685 /* VMCI will select a resource ID for us if we provide
1688 port = addr->svm_port == VMADDR_PORT_ANY ?
1689 VMCI_INVALID_ID : addr->svm_port;
1691 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1694 flags = addr->svm_cid == VMADDR_CID_ANY ?
1695 VMCI_FLAG_ANYCID_DG_HND : 0;
1697 err = vmci_transport_datagram_create_hnd(port, flags,
1698 vmci_transport_recv_dgram_cb,
1700 &vmci_trans(vsk)->dg_handle);
1701 if (err < VMCI_SUCCESS)
1702 return vmci_transport_error_to_vsock_error(err);
1703 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1704 vmci_trans(vsk)->dg_handle.resource);
1709 static int vmci_transport_dgram_enqueue(
1710 struct vsock_sock *vsk,
1711 struct sockaddr_vm *remote_addr,
1716 struct vmci_datagram *dg;
1718 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1721 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1724 /* Allocate a buffer for the user's message and our packet header. */
1725 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1729 memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1731 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1732 remote_addr->svm_port);
1733 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1734 vsk->local_addr.svm_port);
1735 dg->payload_size = len;
1737 err = vmci_datagram_send(dg);
1740 return vmci_transport_error_to_vsock_error(err);
1742 return err - sizeof(*dg);
1745 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1746 struct msghdr *msg, size_t len,
1751 struct vmci_datagram *dg;
1753 struct sk_buff *skb;
1755 noblock = flags & MSG_DONTWAIT;
1757 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1760 /* Retrieve the head sk_buff from the socket's receive queue. */
1762 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1766 dg = (struct vmci_datagram *)skb->data;
1768 /* err is 0, meaning we read zero bytes. */
1771 payload_len = dg->payload_size;
1772 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1773 if (payload_len != skb->len - sizeof(*dg)) {
1778 if (payload_len > len) {
1780 msg->msg_flags |= MSG_TRUNC;
1783 /* Place the datagram payload in the user's iovec. */
1784 err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1788 if (msg->msg_name) {
1789 /* Provide the address of the sender. */
1790 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1791 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1792 msg->msg_namelen = sizeof(*vm_addr);
1797 skb_free_datagram(&vsk->sk, skb);
1801 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1803 if (cid == VMADDR_CID_HYPERVISOR) {
1804 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1805 * state and are allowed.
1807 return port == VMCI_UNITY_PBRPC_REGISTER;
1813 static int vmci_transport_connect(struct vsock_sock *vsk)
1816 bool old_pkt_proto = false;
1817 struct sock *sk = &vsk->sk;
1819 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1821 err = vmci_transport_send_conn_request(
1822 sk, vmci_trans(vsk)->queue_pair_size);
1824 sk->sk_state = TCP_CLOSE;
1828 int supported_proto_versions =
1829 vmci_transport_new_proto_supported_versions();
1830 err = vmci_transport_send_conn_request2(
1831 sk, vmci_trans(vsk)->queue_pair_size,
1832 supported_proto_versions);
1834 sk->sk_state = TCP_CLOSE;
1838 vsk->sent_request = true;
1844 static ssize_t vmci_transport_stream_dequeue(
1845 struct vsock_sock *vsk,
1850 if (flags & MSG_PEEK)
1851 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1853 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1856 static ssize_t vmci_transport_stream_enqueue(
1857 struct vsock_sock *vsk,
1861 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1864 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1866 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1869 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1871 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1874 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1876 return vmci_trans(vsk)->consume_size;
1879 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1881 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1884 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1886 return vmci_trans(vsk)->queue_pair_size;
1889 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1891 return vmci_trans(vsk)->queue_pair_min_size;
1894 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1896 return vmci_trans(vsk)->queue_pair_max_size;
1899 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1901 if (val < vmci_trans(vsk)->queue_pair_min_size)
1902 vmci_trans(vsk)->queue_pair_min_size = val;
1903 if (val > vmci_trans(vsk)->queue_pair_max_size)
1904 vmci_trans(vsk)->queue_pair_max_size = val;
1905 vmci_trans(vsk)->queue_pair_size = val;
1908 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1911 if (val > vmci_trans(vsk)->queue_pair_size)
1912 vmci_trans(vsk)->queue_pair_size = val;
1913 vmci_trans(vsk)->queue_pair_min_size = val;
1916 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1919 if (val < vmci_trans(vsk)->queue_pair_size)
1920 vmci_trans(vsk)->queue_pair_size = val;
1921 vmci_trans(vsk)->queue_pair_max_size = val;
1924 static int vmci_transport_notify_poll_in(
1925 struct vsock_sock *vsk,
1927 bool *data_ready_now)
1929 return vmci_trans(vsk)->notify_ops->poll_in(
1930 &vsk->sk, target, data_ready_now);
1933 static int vmci_transport_notify_poll_out(
1934 struct vsock_sock *vsk,
1936 bool *space_available_now)
1938 return vmci_trans(vsk)->notify_ops->poll_out(
1939 &vsk->sk, target, space_available_now);
1942 static int vmci_transport_notify_recv_init(
1943 struct vsock_sock *vsk,
1945 struct vsock_transport_recv_notify_data *data)
1947 return vmci_trans(vsk)->notify_ops->recv_init(
1949 (struct vmci_transport_recv_notify_data *)data);
1952 static int vmci_transport_notify_recv_pre_block(
1953 struct vsock_sock *vsk,
1955 struct vsock_transport_recv_notify_data *data)
1957 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1959 (struct vmci_transport_recv_notify_data *)data);
1962 static int vmci_transport_notify_recv_pre_dequeue(
1963 struct vsock_sock *vsk,
1965 struct vsock_transport_recv_notify_data *data)
1967 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1969 (struct vmci_transport_recv_notify_data *)data);
1972 static int vmci_transport_notify_recv_post_dequeue(
1973 struct vsock_sock *vsk,
1977 struct vsock_transport_recv_notify_data *data)
1979 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1980 &vsk->sk, target, copied, data_read,
1981 (struct vmci_transport_recv_notify_data *)data);
1984 static int vmci_transport_notify_send_init(
1985 struct vsock_sock *vsk,
1986 struct vsock_transport_send_notify_data *data)
1988 return vmci_trans(vsk)->notify_ops->send_init(
1990 (struct vmci_transport_send_notify_data *)data);
1993 static int vmci_transport_notify_send_pre_block(
1994 struct vsock_sock *vsk,
1995 struct vsock_transport_send_notify_data *data)
1997 return vmci_trans(vsk)->notify_ops->send_pre_block(
1999 (struct vmci_transport_send_notify_data *)data);
2002 static int vmci_transport_notify_send_pre_enqueue(
2003 struct vsock_sock *vsk,
2004 struct vsock_transport_send_notify_data *data)
2006 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2008 (struct vmci_transport_send_notify_data *)data);
2011 static int vmci_transport_notify_send_post_enqueue(
2012 struct vsock_sock *vsk,
2014 struct vsock_transport_send_notify_data *data)
2016 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2018 (struct vmci_transport_send_notify_data *)data);
2021 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2023 if (PROTOCOL_OVERRIDE != -1) {
2024 if (PROTOCOL_OVERRIDE == 0)
2025 *old_pkt_proto = true;
2027 *old_pkt_proto = false;
2029 pr_info("Proto override in use\n");
2036 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2040 struct vsock_sock *vsk = vsock_sk(sk);
2042 if (old_pkt_proto) {
2043 if (*proto != VSOCK_PROTO_INVALID) {
2044 pr_err("Can't set both an old and new protocol\n");
2047 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2052 case VSOCK_PROTO_PKT_ON_NOTIFY:
2053 vmci_trans(vsk)->notify_ops =
2054 &vmci_transport_notify_pkt_q_state_ops;
2057 pr_err("Unknown notify protocol version\n");
2062 vmci_trans(vsk)->notify_ops->socket_init(sk);
2066 static u16 vmci_transport_new_proto_supported_versions(void)
2068 if (PROTOCOL_OVERRIDE != -1)
2069 return PROTOCOL_OVERRIDE;
2071 return VSOCK_PROTO_ALL_SUPPORTED;
2074 static u32 vmci_transport_get_local_cid(void)
2076 return vmci_get_context_id();
2079 static const struct vsock_transport vmci_transport = {
2080 .init = vmci_transport_socket_init,
2081 .destruct = vmci_transport_destruct,
2082 .release = vmci_transport_release,
2083 .connect = vmci_transport_connect,
2084 .dgram_bind = vmci_transport_dgram_bind,
2085 .dgram_dequeue = vmci_transport_dgram_dequeue,
2086 .dgram_enqueue = vmci_transport_dgram_enqueue,
2087 .dgram_allow = vmci_transport_dgram_allow,
2088 .stream_dequeue = vmci_transport_stream_dequeue,
2089 .stream_enqueue = vmci_transport_stream_enqueue,
2090 .stream_has_data = vmci_transport_stream_has_data,
2091 .stream_has_space = vmci_transport_stream_has_space,
2092 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2093 .stream_is_active = vmci_transport_stream_is_active,
2094 .stream_allow = vmci_transport_stream_allow,
2095 .notify_poll_in = vmci_transport_notify_poll_in,
2096 .notify_poll_out = vmci_transport_notify_poll_out,
2097 .notify_recv_init = vmci_transport_notify_recv_init,
2098 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2099 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2100 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2101 .notify_send_init = vmci_transport_notify_send_init,
2102 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2103 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2104 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2105 .shutdown = vmci_transport_shutdown,
2106 .set_buffer_size = vmci_transport_set_buffer_size,
2107 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2108 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2109 .get_buffer_size = vmci_transport_get_buffer_size,
2110 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2111 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2112 .get_local_cid = vmci_transport_get_local_cid,
2115 static int __init vmci_transport_init(void)
2119 /* Create the datagram handle that we will use to send and receive all
2120 * VSocket control messages for this context.
2122 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2123 VMCI_FLAG_ANYCID_DG_HND,
2124 vmci_transport_recv_stream_cb,
2126 &vmci_transport_stream_handle);
2127 if (err < VMCI_SUCCESS) {
2128 pr_err("Unable to create datagram handle. (%d)\n", err);
2129 return vmci_transport_error_to_vsock_error(err);
2132 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2133 vmci_transport_qp_resumed_cb,
2134 NULL, &vmci_transport_qp_resumed_sub_id);
2135 if (err < VMCI_SUCCESS) {
2136 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2137 err = vmci_transport_error_to_vsock_error(err);
2138 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2139 goto err_destroy_stream_handle;
2142 err = vsock_core_init(&vmci_transport);
2144 goto err_unsubscribe;
2149 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2150 err_destroy_stream_handle:
2151 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2154 module_init(vmci_transport_init);
2156 static void __exit vmci_transport_exit(void)
2158 cancel_work_sync(&vmci_transport_cleanup_work);
2159 vmci_transport_free_resources(&vmci_transport_cleanup_list);
2161 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2162 if (vmci_datagram_destroy_handle(
2163 vmci_transport_stream_handle) != VMCI_SUCCESS)
2164 pr_err("Couldn't destroy datagram handle\n");
2165 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2168 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2169 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2170 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2175 module_exit(vmci_transport_exit);
2177 MODULE_AUTHOR("VMware, Inc.");
2178 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2179 MODULE_VERSION("1.0.5.0-k");
2180 MODULE_LICENSE("GPL v2");
2181 MODULE_ALIAS("vmware_vsock");
2182 MODULE_ALIAS_NETPROTO(PF_VSOCK);