2 * VMware vSockets Driver
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 #include <linux/types.h>
17 #include <linux/bitops.h>
18 #include <linux/cred.h>
19 #include <linux/init.h>
21 #include <linux/kernel.h>
22 #include <linux/kmod.h>
23 #include <linux/list.h>
24 #include <linux/miscdevice.h>
25 #include <linux/module.h>
26 #include <linux/mutex.h>
27 #include <linux/net.h>
28 #include <linux/poll.h>
29 #include <linux/skbuff.h>
30 #include <linux/smp.h>
31 #include <linux/socket.h>
32 #include <linux/stddef.h>
33 #include <linux/unistd.h>
34 #include <linux/wait.h>
35 #include <linux/workqueue.h>
37 #include <net/af_vsock.h>
39 #include "vmci_transport_notify.h"
41 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
42 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
43 static void vmci_transport_peer_detach_cb(u32 sub_id,
44 const struct vmci_event_data *ed,
46 static void vmci_transport_recv_pkt_work(struct work_struct *work);
47 static void vmci_transport_cleanup(struct work_struct *work);
48 static int vmci_transport_recv_listen(struct sock *sk,
49 struct vmci_transport_packet *pkt);
50 static int vmci_transport_recv_connecting_server(
53 struct vmci_transport_packet *pkt);
54 static int vmci_transport_recv_connecting_client(
56 struct vmci_transport_packet *pkt);
57 static int vmci_transport_recv_connecting_client_negotiate(
59 struct vmci_transport_packet *pkt);
60 static int vmci_transport_recv_connecting_client_invalid(
62 struct vmci_transport_packet *pkt);
63 static int vmci_transport_recv_connected(struct sock *sk,
64 struct vmci_transport_packet *pkt);
65 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
66 static u16 vmci_transport_new_proto_supported_versions(void);
67 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
70 struct vmci_transport_recv_pkt_info {
71 struct work_struct work;
73 struct vmci_transport_packet pkt;
76 static LIST_HEAD(vmci_transport_cleanup_list);
77 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
78 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
80 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
82 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
84 static int PROTOCOL_OVERRIDE = -1;
86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN 128
87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE 262144
88 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX 262144
90 /* The default peer timeout indicates how long we will wait for a peer response
91 * to a control message.
93 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
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)
100 case VMCI_ERROR_NO_MEM:
102 case VMCI_ERROR_DUPLICATE_ENTRY:
103 case VMCI_ERROR_ALREADY_EXISTS:
105 case VMCI_ERROR_NO_ACCESS:
107 case VMCI_ERROR_NO_RESOURCES:
109 case VMCI_ERROR_INVALID_RESOURCE:
110 return -EHOSTUNREACH;
111 case VMCI_ERROR_INVALID_ARGS:
118 static u32 vmci_transport_peer_rid(u32 peer_cid)
120 if (VMADDR_CID_HYPERVISOR == peer_cid)
121 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
123 return VMCI_TRANSPORT_PACKET_RID;
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_peer_rid(dst->svm_cid));
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_alloc_send_control_pkt(struct sockaddr_vm *src,
269 struct sockaddr_vm *dst,
270 enum vmci_transport_packet_type type,
273 struct vmci_transport_waiting_info *wait,
275 struct vmci_handle handle)
277 struct vmci_transport_packet *pkt;
280 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
284 err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
285 mode, wait, proto, handle,
293 vmci_transport_send_control_pkt(struct sock *sk,
294 enum vmci_transport_packet_type type,
297 struct vmci_transport_waiting_info *wait,
299 struct vmci_handle handle)
301 struct vsock_sock *vsk;
305 if (!vsock_addr_bound(&vsk->local_addr))
308 if (!vsock_addr_bound(&vsk->remote_addr))
311 return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
314 wait, proto, handle);
317 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
318 struct sockaddr_vm *src,
319 struct vmci_transport_packet *pkt)
321 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
323 return vmci_transport_send_control_pkt_bh(
325 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
326 0, NULL, VMCI_INVALID_HANDLE);
329 static int vmci_transport_send_reset(struct sock *sk,
330 struct vmci_transport_packet *pkt)
332 struct sockaddr_vm *dst_ptr;
333 struct sockaddr_vm dst;
334 struct vsock_sock *vsk;
336 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
341 if (!vsock_addr_bound(&vsk->local_addr))
344 if (vsock_addr_bound(&vsk->remote_addr)) {
345 dst_ptr = &vsk->remote_addr;
347 vsock_addr_init(&dst, pkt->dg.src.context,
351 return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
352 VMCI_TRANSPORT_PACKET_TYPE_RST,
353 0, 0, NULL, VSOCK_PROTO_INVALID,
354 VMCI_INVALID_HANDLE);
357 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
359 return vmci_transport_send_control_pkt(
361 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
364 VMCI_INVALID_HANDLE);
367 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
370 return vmci_transport_send_control_pkt(
372 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
373 size, 0, NULL, version,
374 VMCI_INVALID_HANDLE);
377 static int vmci_transport_send_qp_offer(struct sock *sk,
378 struct vmci_handle handle)
380 return vmci_transport_send_control_pkt(
381 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
383 VSOCK_PROTO_INVALID, handle);
386 static int vmci_transport_send_attach(struct sock *sk,
387 struct vmci_handle handle)
389 return vmci_transport_send_control_pkt(
390 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
391 0, 0, NULL, VSOCK_PROTO_INVALID,
395 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
397 return vmci_transport_reply_control_pkt_fast(
399 VMCI_TRANSPORT_PACKET_TYPE_RST,
401 VMCI_INVALID_HANDLE);
404 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
405 struct sockaddr_vm *src)
407 return vmci_transport_send_control_pkt_bh(
409 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
410 0, 0, NULL, VMCI_INVALID_HANDLE);
413 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
414 struct sockaddr_vm *src)
416 return vmci_transport_send_control_pkt_bh(
418 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
419 0, NULL, VMCI_INVALID_HANDLE);
422 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
423 struct sockaddr_vm *src)
425 return vmci_transport_send_control_pkt_bh(
427 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
428 0, NULL, VMCI_INVALID_HANDLE);
431 int vmci_transport_send_wrote(struct sock *sk)
433 return vmci_transport_send_control_pkt(
434 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
435 0, NULL, VSOCK_PROTO_INVALID,
436 VMCI_INVALID_HANDLE);
439 int vmci_transport_send_read(struct sock *sk)
441 return vmci_transport_send_control_pkt(
442 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
443 0, NULL, VSOCK_PROTO_INVALID,
444 VMCI_INVALID_HANDLE);
447 int vmci_transport_send_waiting_write(struct sock *sk,
448 struct vmci_transport_waiting_info *wait)
450 return vmci_transport_send_control_pkt(
451 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
452 0, 0, wait, VSOCK_PROTO_INVALID,
453 VMCI_INVALID_HANDLE);
456 int vmci_transport_send_waiting_read(struct sock *sk,
457 struct vmci_transport_waiting_info *wait)
459 return vmci_transport_send_control_pkt(
460 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
461 0, 0, wait, VSOCK_PROTO_INVALID,
462 VMCI_INVALID_HANDLE);
465 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
467 return vmci_transport_send_control_pkt(
469 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
472 VMCI_INVALID_HANDLE);
475 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
477 return vmci_transport_send_control_pkt(sk,
478 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
481 VMCI_INVALID_HANDLE);
484 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
487 return vmci_transport_send_control_pkt(
488 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
489 size, 0, NULL, version,
490 VMCI_INVALID_HANDLE);
493 static struct sock *vmci_transport_get_pending(
494 struct sock *listener,
495 struct vmci_transport_packet *pkt)
497 struct vsock_sock *vlistener;
498 struct vsock_sock *vpending;
499 struct sock *pending;
500 struct sockaddr_vm src;
502 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
504 vlistener = vsock_sk(listener);
506 list_for_each_entry(vpending, &vlistener->pending_links,
508 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
509 pkt->dst_port == vpending->local_addr.svm_port) {
510 pending = sk_vsock(vpending);
522 static void vmci_transport_release_pending(struct sock *pending)
527 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
528 * trusted sockets 2) sockets from applications running as the same user as the
529 * VM (this is only true for the host side and only when using hosted products)
532 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
534 return vsock->trusted ||
535 vmci_is_context_owner(peer_cid, vsock->owner->uid);
538 /* We allow sending datagrams to and receiving datagrams from a restricted VM
539 * only if it is trusted as described in vmci_transport_is_trusted.
542 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
544 if (VMADDR_CID_HYPERVISOR == peer_cid)
547 if (vsock->cached_peer != peer_cid) {
548 vsock->cached_peer = peer_cid;
549 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
550 (vmci_context_get_priv_flags(peer_cid) &
551 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
552 vsock->cached_peer_allow_dgram = false;
554 vsock->cached_peer_allow_dgram = true;
558 return vsock->cached_peer_allow_dgram;
562 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
563 struct vmci_handle *handle,
566 u32 peer, u32 flags, bool trusted)
571 /* Try to allocate our queue pair as trusted. This will only
572 * work if vsock is running in the host.
575 err = vmci_qpair_alloc(qpair, handle, produce_size,
578 VMCI_PRIVILEGE_FLAG_TRUSTED);
579 if (err != VMCI_ERROR_NO_ACCESS)
584 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
585 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
588 pr_err_once("Could not attach to queue pair with %d\n", err);
589 err = vmci_transport_error_to_vsock_error(err);
596 vmci_transport_datagram_create_hnd(u32 resource_id,
598 vmci_datagram_recv_cb recv_cb,
600 struct vmci_handle *out_handle)
604 /* Try to allocate our datagram handler as trusted. This will only work
605 * if vsock is running in the host.
608 err = vmci_datagram_create_handle_priv(resource_id, flags,
609 VMCI_PRIVILEGE_FLAG_TRUSTED,
611 client_data, out_handle);
613 if (err == VMCI_ERROR_NO_ACCESS)
614 err = vmci_datagram_create_handle(resource_id, flags,
615 recv_cb, client_data,
621 /* This is invoked as part of a tasklet that's scheduled when the VMCI
622 * interrupt fires. This is run in bottom-half context and if it ever needs to
623 * sleep it should defer that work to a work queue.
626 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
631 struct vsock_sock *vsk;
633 sk = (struct sock *)data;
635 /* This handler is privileged when this module is running on the host.
636 * We will get datagrams from all endpoints (even VMs that are in a
637 * restricted context). If we get one from a restricted context then
638 * the destination socket must be trusted.
640 * NOTE: We access the socket struct without holding the lock here.
641 * This is ok because the field we are interested is never modified
642 * outside of the create and destruct socket functions.
645 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
646 return VMCI_ERROR_NO_ACCESS;
648 size = VMCI_DG_SIZE(dg);
650 /* Attach the packet to the socket's receive queue as an sk_buff. */
651 skb = alloc_skb(size, GFP_ATOMIC);
653 return VMCI_ERROR_NO_MEM;
655 /* sk_receive_skb() will do a sock_put(), so hold here. */
658 memcpy(skb->data, dg, size);
659 sk_receive_skb(sk, skb, 0);
664 static bool vmci_transport_stream_allow(u32 cid, u32 port)
666 static const u32 non_socket_contexts[] = {
671 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
673 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
674 if (cid == non_socket_contexts[i])
681 /* This is invoked as part of a tasklet that's scheduled when the VMCI
682 * interrupt fires. This is run in bottom-half context but it defers most of
683 * its work to the packet handling work queue.
686 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
689 struct sockaddr_vm dst;
690 struct sockaddr_vm src;
691 struct vmci_transport_packet *pkt;
692 struct vsock_sock *vsk;
698 bh_process_pkt = false;
700 /* Ignore incoming packets from contexts without sockets, or resources
701 * that aren't vsock implementations.
704 if (!vmci_transport_stream_allow(dg->src.context, -1)
705 || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
706 return VMCI_ERROR_NO_ACCESS;
708 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
709 /* Drop datagrams that do not contain full VSock packets. */
710 return VMCI_ERROR_INVALID_ARGS;
712 pkt = (struct vmci_transport_packet *)dg;
714 /* Find the socket that should handle this packet. First we look for a
715 * connected socket and if there is none we look for a socket bound to
716 * the destintation address.
718 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
719 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
721 sk = vsock_find_connected_socket(&src, &dst);
723 sk = vsock_find_bound_socket(&dst);
725 /* We could not find a socket for this specified
726 * address. If this packet is a RST, we just drop it.
727 * If it is another packet, we send a RST. Note that
728 * we do not send a RST reply to RSTs so that we do not
729 * continually send RSTs between two endpoints.
731 * Note that since this is a reply, dst is src and src
734 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
735 pr_err("unable to send reset\n");
737 err = VMCI_ERROR_NOT_FOUND;
742 /* If the received packet type is beyond all types known to this
743 * implementation, reply with an invalid message. Hopefully this will
744 * help when implementing backwards compatibility in the future.
746 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
747 vmci_transport_send_invalid_bh(&dst, &src);
748 err = VMCI_ERROR_INVALID_ARGS;
752 /* This handler is privileged when this module is running on the host.
753 * We will get datagram connect requests from all endpoints (even VMs
754 * that are in a restricted context). If we get one from a restricted
755 * context then the destination socket must be trusted.
757 * NOTE: We access the socket struct without holding the lock here.
758 * This is ok because the field we are interested is never modified
759 * outside of the create and destruct socket functions.
762 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
763 err = VMCI_ERROR_NO_ACCESS;
767 /* We do most everything in a work queue, but let's fast path the
768 * notification of reads and writes to help data transfer performance.
769 * We can only do this if there is no process context code executing
770 * for this socket since that may change the state.
774 if (!sock_owned_by_user(sk)) {
775 /* The local context ID may be out of date, update it. */
776 vsk->local_addr.svm_cid = dst.svm_cid;
778 if (sk->sk_state == TCP_ESTABLISHED)
779 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
780 sk, pkt, true, &dst, &src,
786 if (!bh_process_pkt) {
787 struct vmci_transport_recv_pkt_info *recv_pkt_info;
789 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
790 if (!recv_pkt_info) {
791 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
792 pr_err("unable to send reset\n");
794 err = VMCI_ERROR_NO_MEM;
798 recv_pkt_info->sk = sk;
799 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
800 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
802 schedule_work(&recv_pkt_info->work);
803 /* Clear sk so that the reference count incremented by one of
804 * the Find functions above is not decremented below. We need
805 * that reference count for the packet handler we've scheduled
818 static void vmci_transport_handle_detach(struct sock *sk)
820 struct vsock_sock *vsk;
823 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
824 sock_set_flag(sk, SOCK_DONE);
826 /* On a detach the peer will not be sending or receiving
829 vsk->peer_shutdown = SHUTDOWN_MASK;
831 /* We should not be sending anymore since the peer won't be
832 * there to receive, but we can still receive if there is data
833 * left in our consume queue.
835 if (vsock_stream_has_data(vsk) <= 0) {
836 sk->sk_state = TCP_CLOSE;
838 if (sk->sk_state == TCP_SYN_SENT) {
839 /* The peer may detach from a queue pair while
840 * we are still in the connecting state, i.e.,
841 * if the peer VM is killed after attaching to
842 * a queue pair, but before we complete the
843 * handshake. In that case, we treat the detach
844 * event like a reset.
847 sk->sk_err = ECONNRESET;
848 sk->sk_error_report(sk);
852 sk->sk_state_change(sk);
856 static void vmci_transport_peer_detach_cb(u32 sub_id,
857 const struct vmci_event_data *e_data,
860 struct vmci_transport *trans = client_data;
861 const struct vmci_event_payload_qp *e_payload;
863 e_payload = vmci_event_data_const_payload(e_data);
865 /* XXX This is lame, we should provide a way to lookup sockets by
868 if (vmci_handle_is_invalid(e_payload->handle) ||
869 !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
872 /* We don't ask for delayed CBs when we subscribe to this event (we
873 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
874 * guarantees in that case about what context we might be running in,
875 * so it could be BH or process, blockable or non-blockable. So we
876 * need to account for all possible contexts here.
878 spin_lock_bh(&trans->lock);
882 /* Apart from here, trans->lock is only grabbed as part of sk destruct,
883 * where trans->sk isn't locked.
885 bh_lock_sock(trans->sk);
887 vmci_transport_handle_detach(trans->sk);
889 bh_unlock_sock(trans->sk);
891 spin_unlock_bh(&trans->lock);
894 static void vmci_transport_qp_resumed_cb(u32 sub_id,
895 const struct vmci_event_data *e_data,
898 vsock_for_each_connected_socket(vmci_transport_handle_detach);
901 static void vmci_transport_recv_pkt_work(struct work_struct *work)
903 struct vmci_transport_recv_pkt_info *recv_pkt_info;
904 struct vmci_transport_packet *pkt;
908 container_of(work, struct vmci_transport_recv_pkt_info, work);
909 sk = recv_pkt_info->sk;
910 pkt = &recv_pkt_info->pkt;
914 /* The local context ID may be out of date. */
915 vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
917 switch (sk->sk_state) {
919 vmci_transport_recv_listen(sk, pkt);
922 /* Processing of pending connections for servers goes through
923 * the listening socket, so see vmci_transport_recv_listen()
926 vmci_transport_recv_connecting_client(sk, pkt);
928 case TCP_ESTABLISHED:
929 vmci_transport_recv_connected(sk, pkt);
932 /* Because this function does not run in the same context as
933 * vmci_transport_recv_stream_cb it is possible that the
934 * socket has closed. We need to let the other side know or it
935 * could be sitting in a connect and hang forever. Send a
936 * reset to prevent that.
938 vmci_transport_send_reset(sk, pkt);
943 kfree(recv_pkt_info);
944 /* Release reference obtained in the stream callback when we fetched
945 * this socket out of the bound or connected list.
950 static int vmci_transport_recv_listen(struct sock *sk,
951 struct vmci_transport_packet *pkt)
953 struct sock *pending;
954 struct vsock_sock *vpending;
957 bool old_request = false;
958 bool old_pkt_proto = false;
962 /* Because we are in the listen state, we could be receiving a packet
963 * for ourself or any previous connection requests that we received.
964 * If it's the latter, we try to find a socket in our list of pending
965 * connections and, if we do, call the appropriate handler for the
966 * state that that socket is in. Otherwise we try to service the
967 * connection request.
969 pending = vmci_transport_get_pending(sk, pkt);
973 /* The local context ID may be out of date. */
974 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
976 switch (pending->sk_state) {
978 err = vmci_transport_recv_connecting_server(sk,
983 vmci_transport_send_reset(pending, pkt);
988 vsock_remove_pending(sk, pending);
990 release_sock(pending);
991 vmci_transport_release_pending(pending);
996 /* The listen state only accepts connection requests. Reply with a
997 * reset unless we received a reset.
1000 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
1001 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
1002 vmci_transport_reply_reset(pkt);
1006 if (pkt->u.size == 0) {
1007 vmci_transport_reply_reset(pkt);
1011 /* If this socket can't accommodate this connection request, we send a
1012 * reset. Otherwise we create and initialize a child socket and reply
1013 * with a connection negotiation.
1015 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1016 vmci_transport_reply_reset(pkt);
1017 return -ECONNREFUSED;
1020 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1023 vmci_transport_send_reset(sk, pkt);
1027 vpending = vsock_sk(pending);
1029 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1031 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1034 /* If the proposed size fits within our min/max, accept it. Otherwise
1035 * propose our own size.
1037 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1038 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1039 qp_size = pkt->u.size;
1041 qp_size = vmci_trans(vpending)->queue_pair_size;
1044 /* Figure out if we are using old or new requests based on the
1045 * overrides pkt types sent by our peer.
1047 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1048 old_request = old_pkt_proto;
1050 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1052 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1053 old_request = false;
1058 /* Handle a REQUEST (or override) */
1059 u16 version = VSOCK_PROTO_INVALID;
1060 if (vmci_transport_proto_to_notify_struct(
1061 pending, &version, true))
1062 err = vmci_transport_send_negotiate(pending, qp_size);
1067 /* Handle a REQUEST2 (or override) */
1068 int proto_int = pkt->proto;
1070 u16 active_proto_version = 0;
1072 /* The list of possible protocols is the intersection of all
1073 * protocols the client supports ... plus all the protocols we
1076 proto_int &= vmci_transport_new_proto_supported_versions();
1078 /* We choose the highest possible protocol version and use that
1081 pos = fls(proto_int);
1083 active_proto_version = (1 << (pos - 1));
1084 if (vmci_transport_proto_to_notify_struct(
1085 pending, &active_proto_version, false))
1086 err = vmci_transport_send_negotiate2(pending,
1088 active_proto_version);
1098 vmci_transport_send_reset(sk, pkt);
1100 err = vmci_transport_error_to_vsock_error(err);
1104 vsock_add_pending(sk, pending);
1105 sk->sk_ack_backlog++;
1107 pending->sk_state = TCP_SYN_SENT;
1108 vmci_trans(vpending)->produce_size =
1109 vmci_trans(vpending)->consume_size = qp_size;
1110 vmci_trans(vpending)->queue_pair_size = qp_size;
1112 vmci_trans(vpending)->notify_ops->process_request(pending);
1114 /* We might never receive another message for this socket and it's not
1115 * connected to any process, so we have to ensure it gets cleaned up
1116 * ourself. Our delayed work function will take care of that. Note
1117 * that we do not ever cancel this function since we have few
1118 * guarantees about its state when calling cancel_delayed_work().
1119 * Instead we hold a reference on the socket for that function and make
1120 * it capable of handling cases where it needs to do nothing but
1121 * release that reference.
1123 vpending->listener = sk;
1126 schedule_delayed_work(&vpending->pending_work, HZ);
1133 vmci_transport_recv_connecting_server(struct sock *listener,
1134 struct sock *pending,
1135 struct vmci_transport_packet *pkt)
1137 struct vsock_sock *vpending;
1138 struct vmci_handle handle;
1139 struct vmci_qp *qpair;
1146 vpending = vsock_sk(pending);
1147 detach_sub_id = VMCI_INVALID_ID;
1149 switch (pkt->type) {
1150 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1151 if (vmci_handle_is_invalid(pkt->u.handle)) {
1152 vmci_transport_send_reset(pending, pkt);
1159 /* Close and cleanup the connection. */
1160 vmci_transport_send_reset(pending, pkt);
1162 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1166 /* In order to complete the connection we need to attach to the offered
1167 * queue pair and send an attach notification. We also subscribe to the
1168 * detach event so we know when our peer goes away, and we do that
1169 * before attaching so we don't miss an event. If all this succeeds,
1170 * we update our state and wakeup anything waiting in accept() for a
1174 /* We don't care about attach since we ensure the other side has
1175 * attached by specifying the ATTACH_ONLY flag below.
1177 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1178 vmci_transport_peer_detach_cb,
1179 vmci_trans(vpending), &detach_sub_id);
1180 if (err < VMCI_SUCCESS) {
1181 vmci_transport_send_reset(pending, pkt);
1182 err = vmci_transport_error_to_vsock_error(err);
1187 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1189 /* Now attach to the queue pair the client created. */
1190 handle = pkt->u.handle;
1192 /* vpending->local_addr always has a context id so we do not need to
1193 * worry about VMADDR_CID_ANY in this case.
1196 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1197 flags = VMCI_QPFLAG_ATTACH_ONLY;
1198 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1200 err = vmci_transport_queue_pair_alloc(
1203 vmci_trans(vpending)->produce_size,
1204 vmci_trans(vpending)->consume_size,
1205 pkt->dg.src.context,
1207 vmci_transport_is_trusted(
1209 vpending->remote_addr.svm_cid));
1211 vmci_transport_send_reset(pending, pkt);
1216 vmci_trans(vpending)->qp_handle = handle;
1217 vmci_trans(vpending)->qpair = qpair;
1219 /* When we send the attach message, we must be ready to handle incoming
1220 * control messages on the newly connected socket. So we move the
1221 * pending socket to the connected state before sending the attach
1222 * message. Otherwise, an incoming packet triggered by the attach being
1223 * received by the peer may be processed concurrently with what happens
1224 * below after sending the attach message, and that incoming packet
1225 * will find the listening socket instead of the (currently) pending
1226 * socket. Note that enqueueing the socket increments the reference
1227 * count, so even if a reset comes before the connection is accepted,
1228 * the socket will be valid until it is removed from the queue.
1230 * If we fail sending the attach below, we remove the socket from the
1231 * connected list and move the socket to TCP_CLOSE before
1232 * releasing the lock, so a pending slow path processing of an incoming
1233 * packet will not see the socket in the connected state in that case.
1235 pending->sk_state = TCP_ESTABLISHED;
1237 vsock_insert_connected(vpending);
1239 /* Notify our peer of our attach. */
1240 err = vmci_transport_send_attach(pending, handle);
1242 vsock_remove_connected(vpending);
1243 pr_err("Could not send attach\n");
1244 vmci_transport_send_reset(pending, pkt);
1245 err = vmci_transport_error_to_vsock_error(err);
1250 /* We have a connection. Move the now connected socket from the
1251 * listener's pending list to the accept queue so callers of accept()
1254 vsock_remove_pending(listener, pending);
1255 vsock_enqueue_accept(listener, pending);
1257 /* Callers of accept() will be be waiting on the listening socket, not
1258 * the pending socket.
1260 listener->sk_data_ready(listener);
1265 pending->sk_err = skerr;
1266 pending->sk_state = TCP_CLOSE;
1267 /* As long as we drop our reference, all necessary cleanup will handle
1268 * when the cleanup function drops its reference and our destruct
1269 * implementation is called. Note that since the listen handler will
1270 * remove pending from the pending list upon our failure, the cleanup
1271 * function won't drop the additional reference, which is why we do it
1280 vmci_transport_recv_connecting_client(struct sock *sk,
1281 struct vmci_transport_packet *pkt)
1283 struct vsock_sock *vsk;
1289 switch (pkt->type) {
1290 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1291 if (vmci_handle_is_invalid(pkt->u.handle) ||
1292 !vmci_handle_is_equal(pkt->u.handle,
1293 vmci_trans(vsk)->qp_handle)) {
1299 /* Signify the socket is connected and wakeup the waiter in
1300 * connect(). Also place the socket in the connected table for
1301 * accounting (it can already be found since it's in the bound
1304 sk->sk_state = TCP_ESTABLISHED;
1305 sk->sk_socket->state = SS_CONNECTED;
1306 vsock_insert_connected(vsk);
1307 sk->sk_state_change(sk);
1310 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1311 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1312 if (pkt->u.size == 0
1313 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1314 || pkt->src_port != vsk->remote_addr.svm_port
1315 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1316 || vmci_trans(vsk)->qpair
1317 || vmci_trans(vsk)->produce_size != 0
1318 || vmci_trans(vsk)->consume_size != 0
1319 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1326 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1333 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1334 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1341 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1342 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1343 * continue processing here after they sent an INVALID packet.
1344 * This meant that we got a RST after the INVALID. We ignore a
1345 * RST after an INVALID. The common code doesn't send the RST
1346 * ... so we can hang if an old version of the common code
1347 * fails between getting a REQUEST and sending an OFFER back.
1348 * Not much we can do about it... except hope that it doesn't
1351 if (vsk->ignore_connecting_rst) {
1352 vsk->ignore_connecting_rst = false;
1361 /* Close and cleanup the connection. */
1370 vmci_transport_send_reset(sk, pkt);
1372 sk->sk_state = TCP_CLOSE;
1374 sk->sk_error_report(sk);
1378 static int vmci_transport_recv_connecting_client_negotiate(
1380 struct vmci_transport_packet *pkt)
1383 struct vsock_sock *vsk;
1384 struct vmci_handle handle;
1385 struct vmci_qp *qpair;
1389 bool old_proto = true;
1394 handle = VMCI_INVALID_HANDLE;
1395 detach_sub_id = VMCI_INVALID_ID;
1397 /* If we have gotten here then we should be past the point where old
1398 * linux vsock could have sent the bogus rst.
1400 vsk->sent_request = false;
1401 vsk->ignore_connecting_rst = false;
1403 /* Verify that we're OK with the proposed queue pair size */
1404 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1405 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1410 /* At this point we know the CID the peer is using to talk to us. */
1412 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1413 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1415 /* Setup the notify ops to be the highest supported version that both
1416 * the server and the client support.
1419 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1420 old_proto = old_pkt_proto;
1422 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1424 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1430 version = VSOCK_PROTO_INVALID;
1432 version = pkt->proto;
1434 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1439 /* Subscribe to detach events first.
1441 * XXX We attach once for each queue pair created for now so it is easy
1442 * to find the socket (it's provided), but later we should only
1443 * subscribe once and add a way to lookup sockets by queue pair handle.
1445 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1446 vmci_transport_peer_detach_cb,
1447 vmci_trans(vsk), &detach_sub_id);
1448 if (err < VMCI_SUCCESS) {
1449 err = vmci_transport_error_to_vsock_error(err);
1453 /* Make VMCI select the handle for us. */
1454 handle = VMCI_INVALID_HANDLE;
1455 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1456 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1458 err = vmci_transport_queue_pair_alloc(&qpair,
1462 vsk->remote_addr.svm_cid,
1464 vmci_transport_is_trusted(
1467 remote_addr.svm_cid));
1471 err = vmci_transport_send_qp_offer(sk, handle);
1473 err = vmci_transport_error_to_vsock_error(err);
1477 vmci_trans(vsk)->qp_handle = handle;
1478 vmci_trans(vsk)->qpair = qpair;
1480 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1483 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1485 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1490 if (detach_sub_id != VMCI_INVALID_ID)
1491 vmci_event_unsubscribe(detach_sub_id);
1493 if (!vmci_handle_is_invalid(handle))
1494 vmci_qpair_detach(&qpair);
1500 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1501 struct vmci_transport_packet *pkt)
1504 struct vsock_sock *vsk = vsock_sk(sk);
1506 if (vsk->sent_request) {
1507 vsk->sent_request = false;
1508 vsk->ignore_connecting_rst = true;
1510 err = vmci_transport_send_conn_request(
1511 sk, vmci_trans(vsk)->queue_pair_size);
1513 err = vmci_transport_error_to_vsock_error(err);
1522 static int vmci_transport_recv_connected(struct sock *sk,
1523 struct vmci_transport_packet *pkt)
1525 struct vsock_sock *vsk;
1526 bool pkt_processed = false;
1528 /* In cases where we are closing the connection, it's sufficient to
1529 * mark the state change (and maybe error) and wake up any waiting
1530 * threads. Since this is a connected socket, it's owned by a user
1531 * process and will be cleaned up when the failure is passed back on
1532 * the current or next system call. Our system call implementations
1533 * must therefore check for error and state changes on entry and when
1536 switch (pkt->type) {
1537 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1541 vsk->peer_shutdown |= pkt->u.mode;
1542 sk->sk_state_change(sk);
1546 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1548 /* It is possible that we sent our peer a message (e.g a
1549 * WAITING_READ) right before we got notified that the peer had
1550 * detached. If that happens then we can get a RST pkt back
1551 * from our peer even though there is data available for us to
1552 * read. In that case, don't shutdown the socket completely but
1553 * instead allow the local client to finish reading data off
1554 * the queuepair. Always treat a RST pkt in connected mode like
1557 sock_set_flag(sk, SOCK_DONE);
1558 vsk->peer_shutdown = SHUTDOWN_MASK;
1559 if (vsock_stream_has_data(vsk) <= 0)
1560 sk->sk_state = TCP_CLOSING;
1562 sk->sk_state_change(sk);
1567 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1568 sk, pkt, false, NULL, NULL,
1579 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1580 struct vsock_sock *psk)
1582 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1586 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1587 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1588 vmci_trans(vsk)->qpair = NULL;
1589 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1590 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1591 vmci_trans(vsk)->notify_ops = NULL;
1592 INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1593 vmci_trans(vsk)->sk = &vsk->sk;
1594 spin_lock_init(&vmci_trans(vsk)->lock);
1596 vmci_trans(vsk)->queue_pair_size =
1597 vmci_trans(psk)->queue_pair_size;
1598 vmci_trans(vsk)->queue_pair_min_size =
1599 vmci_trans(psk)->queue_pair_min_size;
1600 vmci_trans(vsk)->queue_pair_max_size =
1601 vmci_trans(psk)->queue_pair_max_size;
1603 vmci_trans(vsk)->queue_pair_size =
1604 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1605 vmci_trans(vsk)->queue_pair_min_size =
1606 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1607 vmci_trans(vsk)->queue_pair_max_size =
1608 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1614 static void vmci_transport_free_resources(struct list_head *transport_list)
1616 while (!list_empty(transport_list)) {
1617 struct vmci_transport *transport =
1618 list_first_entry(transport_list, struct vmci_transport,
1620 list_del(&transport->elem);
1622 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1623 vmci_event_unsubscribe(transport->detach_sub_id);
1624 transport->detach_sub_id = VMCI_INVALID_ID;
1627 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1628 vmci_qpair_detach(&transport->qpair);
1629 transport->qp_handle = VMCI_INVALID_HANDLE;
1630 transport->produce_size = 0;
1631 transport->consume_size = 0;
1638 static void vmci_transport_cleanup(struct work_struct *work)
1642 spin_lock_bh(&vmci_transport_cleanup_lock);
1643 list_replace_init(&vmci_transport_cleanup_list, &pending);
1644 spin_unlock_bh(&vmci_transport_cleanup_lock);
1645 vmci_transport_free_resources(&pending);
1648 static void vmci_transport_destruct(struct vsock_sock *vsk)
1650 /* transport can be NULL if we hit a failure at init() time */
1651 if (!vmci_trans(vsk))
1654 /* Ensure that the detach callback doesn't use the sk/vsk
1655 * we are about to destruct.
1657 spin_lock_bh(&vmci_trans(vsk)->lock);
1658 vmci_trans(vsk)->sk = NULL;
1659 spin_unlock_bh(&vmci_trans(vsk)->lock);
1661 if (vmci_trans(vsk)->notify_ops)
1662 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1664 spin_lock_bh(&vmci_transport_cleanup_lock);
1665 list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1666 spin_unlock_bh(&vmci_transport_cleanup_lock);
1667 schedule_work(&vmci_transport_cleanup_work);
1672 static void vmci_transport_release(struct vsock_sock *vsk)
1674 vsock_remove_sock(vsk);
1676 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1677 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1678 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1682 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1683 struct sockaddr_vm *addr)
1689 /* VMCI will select a resource ID for us if we provide
1692 port = addr->svm_port == VMADDR_PORT_ANY ?
1693 VMCI_INVALID_ID : addr->svm_port;
1695 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1698 flags = addr->svm_cid == VMADDR_CID_ANY ?
1699 VMCI_FLAG_ANYCID_DG_HND : 0;
1701 err = vmci_transport_datagram_create_hnd(port, flags,
1702 vmci_transport_recv_dgram_cb,
1704 &vmci_trans(vsk)->dg_handle);
1705 if (err < VMCI_SUCCESS)
1706 return vmci_transport_error_to_vsock_error(err);
1707 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1708 vmci_trans(vsk)->dg_handle.resource);
1713 static int vmci_transport_dgram_enqueue(
1714 struct vsock_sock *vsk,
1715 struct sockaddr_vm *remote_addr,
1720 struct vmci_datagram *dg;
1722 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1725 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1728 /* Allocate a buffer for the user's message and our packet header. */
1729 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1733 err = memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1739 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1740 remote_addr->svm_port);
1741 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1742 vsk->local_addr.svm_port);
1743 dg->payload_size = len;
1745 err = vmci_datagram_send(dg);
1748 return vmci_transport_error_to_vsock_error(err);
1750 return err - sizeof(*dg);
1753 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1754 struct msghdr *msg, size_t len,
1759 struct vmci_datagram *dg;
1761 struct sk_buff *skb;
1763 noblock = flags & MSG_DONTWAIT;
1765 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1768 /* Retrieve the head sk_buff from the socket's receive queue. */
1770 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1774 dg = (struct vmci_datagram *)skb->data;
1776 /* err is 0, meaning we read zero bytes. */
1779 payload_len = dg->payload_size;
1780 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1781 if (payload_len != skb->len - sizeof(*dg)) {
1786 if (payload_len > len) {
1788 msg->msg_flags |= MSG_TRUNC;
1791 /* Place the datagram payload in the user's iovec. */
1792 err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1796 if (msg->msg_name) {
1797 /* Provide the address of the sender. */
1798 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1799 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1800 msg->msg_namelen = sizeof(*vm_addr);
1805 skb_free_datagram(&vsk->sk, skb);
1809 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1811 if (cid == VMADDR_CID_HYPERVISOR) {
1812 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1813 * state and are allowed.
1815 return port == VMCI_UNITY_PBRPC_REGISTER;
1821 static int vmci_transport_connect(struct vsock_sock *vsk)
1824 bool old_pkt_proto = false;
1825 struct sock *sk = &vsk->sk;
1827 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1829 err = vmci_transport_send_conn_request(
1830 sk, vmci_trans(vsk)->queue_pair_size);
1832 sk->sk_state = TCP_CLOSE;
1836 int supported_proto_versions =
1837 vmci_transport_new_proto_supported_versions();
1838 err = vmci_transport_send_conn_request2(
1839 sk, vmci_trans(vsk)->queue_pair_size,
1840 supported_proto_versions);
1842 sk->sk_state = TCP_CLOSE;
1846 vsk->sent_request = true;
1852 static ssize_t vmci_transport_stream_dequeue(
1853 struct vsock_sock *vsk,
1858 if (flags & MSG_PEEK)
1859 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1861 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1864 static ssize_t vmci_transport_stream_enqueue(
1865 struct vsock_sock *vsk,
1869 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1872 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1874 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1877 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1879 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1882 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1884 return vmci_trans(vsk)->consume_size;
1887 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1889 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1892 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1894 return vmci_trans(vsk)->queue_pair_size;
1897 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1899 return vmci_trans(vsk)->queue_pair_min_size;
1902 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1904 return vmci_trans(vsk)->queue_pair_max_size;
1907 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1909 if (val < vmci_trans(vsk)->queue_pair_min_size)
1910 vmci_trans(vsk)->queue_pair_min_size = val;
1911 if (val > vmci_trans(vsk)->queue_pair_max_size)
1912 vmci_trans(vsk)->queue_pair_max_size = val;
1913 vmci_trans(vsk)->queue_pair_size = val;
1916 static void vmci_transport_set_min_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_min_size = val;
1924 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1927 if (val < vmci_trans(vsk)->queue_pair_size)
1928 vmci_trans(vsk)->queue_pair_size = val;
1929 vmci_trans(vsk)->queue_pair_max_size = val;
1932 static int vmci_transport_notify_poll_in(
1933 struct vsock_sock *vsk,
1935 bool *data_ready_now)
1937 return vmci_trans(vsk)->notify_ops->poll_in(
1938 &vsk->sk, target, data_ready_now);
1941 static int vmci_transport_notify_poll_out(
1942 struct vsock_sock *vsk,
1944 bool *space_available_now)
1946 return vmci_trans(vsk)->notify_ops->poll_out(
1947 &vsk->sk, target, space_available_now);
1950 static int vmci_transport_notify_recv_init(
1951 struct vsock_sock *vsk,
1953 struct vsock_transport_recv_notify_data *data)
1955 return vmci_trans(vsk)->notify_ops->recv_init(
1957 (struct vmci_transport_recv_notify_data *)data);
1960 static int vmci_transport_notify_recv_pre_block(
1961 struct vsock_sock *vsk,
1963 struct vsock_transport_recv_notify_data *data)
1965 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1967 (struct vmci_transport_recv_notify_data *)data);
1970 static int vmci_transport_notify_recv_pre_dequeue(
1971 struct vsock_sock *vsk,
1973 struct vsock_transport_recv_notify_data *data)
1975 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1977 (struct vmci_transport_recv_notify_data *)data);
1980 static int vmci_transport_notify_recv_post_dequeue(
1981 struct vsock_sock *vsk,
1985 struct vsock_transport_recv_notify_data *data)
1987 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1988 &vsk->sk, target, copied, data_read,
1989 (struct vmci_transport_recv_notify_data *)data);
1992 static int vmci_transport_notify_send_init(
1993 struct vsock_sock *vsk,
1994 struct vsock_transport_send_notify_data *data)
1996 return vmci_trans(vsk)->notify_ops->send_init(
1998 (struct vmci_transport_send_notify_data *)data);
2001 static int vmci_transport_notify_send_pre_block(
2002 struct vsock_sock *vsk,
2003 struct vsock_transport_send_notify_data *data)
2005 return vmci_trans(vsk)->notify_ops->send_pre_block(
2007 (struct vmci_transport_send_notify_data *)data);
2010 static int vmci_transport_notify_send_pre_enqueue(
2011 struct vsock_sock *vsk,
2012 struct vsock_transport_send_notify_data *data)
2014 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2016 (struct vmci_transport_send_notify_data *)data);
2019 static int vmci_transport_notify_send_post_enqueue(
2020 struct vsock_sock *vsk,
2022 struct vsock_transport_send_notify_data *data)
2024 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2026 (struct vmci_transport_send_notify_data *)data);
2029 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2031 if (PROTOCOL_OVERRIDE != -1) {
2032 if (PROTOCOL_OVERRIDE == 0)
2033 *old_pkt_proto = true;
2035 *old_pkt_proto = false;
2037 pr_info("Proto override in use\n");
2044 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2048 struct vsock_sock *vsk = vsock_sk(sk);
2050 if (old_pkt_proto) {
2051 if (*proto != VSOCK_PROTO_INVALID) {
2052 pr_err("Can't set both an old and new protocol\n");
2055 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2060 case VSOCK_PROTO_PKT_ON_NOTIFY:
2061 vmci_trans(vsk)->notify_ops =
2062 &vmci_transport_notify_pkt_q_state_ops;
2065 pr_err("Unknown notify protocol version\n");
2070 vmci_trans(vsk)->notify_ops->socket_init(sk);
2074 static u16 vmci_transport_new_proto_supported_versions(void)
2076 if (PROTOCOL_OVERRIDE != -1)
2077 return PROTOCOL_OVERRIDE;
2079 return VSOCK_PROTO_ALL_SUPPORTED;
2082 static u32 vmci_transport_get_local_cid(void)
2084 return vmci_get_context_id();
2087 static const struct vsock_transport vmci_transport = {
2088 .init = vmci_transport_socket_init,
2089 .destruct = vmci_transport_destruct,
2090 .release = vmci_transport_release,
2091 .connect = vmci_transport_connect,
2092 .dgram_bind = vmci_transport_dgram_bind,
2093 .dgram_dequeue = vmci_transport_dgram_dequeue,
2094 .dgram_enqueue = vmci_transport_dgram_enqueue,
2095 .dgram_allow = vmci_transport_dgram_allow,
2096 .stream_dequeue = vmci_transport_stream_dequeue,
2097 .stream_enqueue = vmci_transport_stream_enqueue,
2098 .stream_has_data = vmci_transport_stream_has_data,
2099 .stream_has_space = vmci_transport_stream_has_space,
2100 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2101 .stream_is_active = vmci_transport_stream_is_active,
2102 .stream_allow = vmci_transport_stream_allow,
2103 .notify_poll_in = vmci_transport_notify_poll_in,
2104 .notify_poll_out = vmci_transport_notify_poll_out,
2105 .notify_recv_init = vmci_transport_notify_recv_init,
2106 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2107 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2108 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2109 .notify_send_init = vmci_transport_notify_send_init,
2110 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2111 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2112 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2113 .shutdown = vmci_transport_shutdown,
2114 .set_buffer_size = vmci_transport_set_buffer_size,
2115 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2116 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2117 .get_buffer_size = vmci_transport_get_buffer_size,
2118 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2119 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2120 .get_local_cid = vmci_transport_get_local_cid,
2123 static int __init vmci_transport_init(void)
2127 /* Create the datagram handle that we will use to send and receive all
2128 * VSocket control messages for this context.
2130 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2131 VMCI_FLAG_ANYCID_DG_HND,
2132 vmci_transport_recv_stream_cb,
2134 &vmci_transport_stream_handle);
2135 if (err < VMCI_SUCCESS) {
2136 pr_err("Unable to create datagram handle. (%d)\n", err);
2137 return vmci_transport_error_to_vsock_error(err);
2140 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2141 vmci_transport_qp_resumed_cb,
2142 NULL, &vmci_transport_qp_resumed_sub_id);
2143 if (err < VMCI_SUCCESS) {
2144 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2145 err = vmci_transport_error_to_vsock_error(err);
2146 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2147 goto err_destroy_stream_handle;
2150 err = vsock_core_init(&vmci_transport);
2152 goto err_unsubscribe;
2157 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2158 err_destroy_stream_handle:
2159 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2162 module_init(vmci_transport_init);
2164 static void __exit vmci_transport_exit(void)
2166 cancel_work_sync(&vmci_transport_cleanup_work);
2167 vmci_transport_free_resources(&vmci_transport_cleanup_list);
2169 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2170 if (vmci_datagram_destroy_handle(
2171 vmci_transport_stream_handle) != VMCI_SUCCESS)
2172 pr_err("Couldn't destroy datagram handle\n");
2173 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2176 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2177 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2178 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2183 module_exit(vmci_transport_exit);
2185 MODULE_AUTHOR("VMware, Inc.");
2186 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2187 MODULE_VERSION("1.0.4.0-k");
2188 MODULE_LICENSE("GPL v2");
2189 MODULE_ALIAS("vmware_vsock");
2190 MODULE_ALIAS_NETPROTO(PF_VSOCK);