1 // SPDX-License-Identifier: GPL-2.0-only
3 * common code for virtio vsock
5 * Copyright (C) 2013-2015 Red Hat, Inc.
6 * Author: Asias He <asias@redhat.com>
7 * Stefan Hajnoczi <stefanha@redhat.com>
9 #include <linux/spinlock.h>
10 #include <linux/module.h>
11 #include <linux/sched/signal.h>
12 #include <linux/ctype.h>
13 #include <linux/list.h>
14 #include <linux/virtio_vsock.h>
15 #include <uapi/linux/vsockmon.h>
18 #include <net/af_vsock.h>
20 #define CREATE_TRACE_POINTS
21 #include <trace/events/vsock_virtio_transport_common.h>
23 /* How long to wait for graceful shutdown of a connection */
24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
26 /* Threshold for detecting small packets to copy */
27 #define GOOD_COPY_LEN 128
29 static const struct virtio_transport *
30 virtio_transport_get_ops(struct vsock_sock *vsk)
32 const struct vsock_transport *t = vsock_core_get_transport(vsk);
37 return container_of(t, struct virtio_transport, transport);
40 static struct virtio_vsock_pkt *
41 virtio_transport_alloc_pkt(struct virtio_vsock_pkt_info *info,
48 struct virtio_vsock_pkt *pkt;
51 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
55 pkt->hdr.type = cpu_to_le16(info->type);
56 pkt->hdr.op = cpu_to_le16(info->op);
57 pkt->hdr.src_cid = cpu_to_le64(src_cid);
58 pkt->hdr.dst_cid = cpu_to_le64(dst_cid);
59 pkt->hdr.src_port = cpu_to_le32(src_port);
60 pkt->hdr.dst_port = cpu_to_le32(dst_port);
61 pkt->hdr.flags = cpu_to_le32(info->flags);
63 pkt->hdr.len = cpu_to_le32(len);
64 pkt->reply = info->reply;
67 if (info->msg && len > 0) {
68 pkt->buf = kmalloc(len, GFP_KERNEL);
74 err = memcpy_from_msg(pkt->buf, info->msg, len);
79 trace_virtio_transport_alloc_pkt(src_cid, src_port,
96 static struct sk_buff *virtio_transport_build_skb(void *opaque)
98 struct virtio_vsock_pkt *pkt = opaque;
99 struct af_vsockmon_hdr *hdr;
104 /* A packet could be split to fit the RX buffer, so we can retrieve
105 * the payload length from the header and the buffer pointer taking
106 * care of the offset in the original packet.
108 payload_len = le32_to_cpu(pkt->hdr.len);
109 payload_buf = pkt->buf + pkt->off;
111 skb = alloc_skb(sizeof(*hdr) + sizeof(pkt->hdr) + payload_len,
116 hdr = skb_put(skb, sizeof(*hdr));
118 /* pkt->hdr is little-endian so no need to byteswap here */
119 hdr->src_cid = pkt->hdr.src_cid;
120 hdr->src_port = pkt->hdr.src_port;
121 hdr->dst_cid = pkt->hdr.dst_cid;
122 hdr->dst_port = pkt->hdr.dst_port;
124 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
125 hdr->len = cpu_to_le16(sizeof(pkt->hdr));
126 memset(hdr->reserved, 0, sizeof(hdr->reserved));
128 switch (le16_to_cpu(pkt->hdr.op)) {
129 case VIRTIO_VSOCK_OP_REQUEST:
130 case VIRTIO_VSOCK_OP_RESPONSE:
131 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
133 case VIRTIO_VSOCK_OP_RST:
134 case VIRTIO_VSOCK_OP_SHUTDOWN:
135 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
137 case VIRTIO_VSOCK_OP_RW:
138 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
140 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
141 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
142 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
145 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
149 skb_put_data(skb, &pkt->hdr, sizeof(pkt->hdr));
152 skb_put_data(skb, payload_buf, payload_len);
158 void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt)
160 if (pkt->tap_delivered)
163 vsock_deliver_tap(virtio_transport_build_skb, pkt);
164 pkt->tap_delivered = true;
166 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
168 /* This function can only be used on connecting/connected sockets,
169 * since a socket assigned to a transport is required.
171 * Do not use on listener sockets!
173 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
174 struct virtio_vsock_pkt_info *info)
176 u32 src_cid, src_port, dst_cid, dst_port;
177 const struct virtio_transport *t_ops;
178 struct virtio_vsock_sock *vvs;
179 struct virtio_vsock_pkt *pkt;
180 u32 pkt_len = info->pkt_len;
182 t_ops = virtio_transport_get_ops(vsk);
183 if (unlikely(!t_ops))
186 src_cid = t_ops->transport.get_local_cid();
187 src_port = vsk->local_addr.svm_port;
188 if (!info->remote_cid) {
189 dst_cid = vsk->remote_addr.svm_cid;
190 dst_port = vsk->remote_addr.svm_port;
192 dst_cid = info->remote_cid;
193 dst_port = info->remote_port;
198 /* we can send less than pkt_len bytes */
199 if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
200 pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
202 /* virtio_transport_get_credit might return less than pkt_len credit */
203 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
205 /* Do not send zero length OP_RW pkt */
206 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
209 pkt = virtio_transport_alloc_pkt(info, pkt_len,
213 virtio_transport_put_credit(vvs, pkt_len);
217 virtio_transport_inc_tx_pkt(vvs, pkt);
219 return t_ops->send_pkt(pkt);
222 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
223 struct virtio_vsock_pkt *pkt)
225 if (vvs->rx_bytes + pkt->len > vvs->buf_alloc)
228 vvs->rx_bytes += pkt->len;
232 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
233 struct virtio_vsock_pkt *pkt)
235 vvs->rx_bytes -= pkt->len;
236 vvs->fwd_cnt += pkt->len;
239 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct virtio_vsock_pkt *pkt)
241 spin_lock_bh(&vvs->rx_lock);
242 vvs->last_fwd_cnt = vvs->fwd_cnt;
243 pkt->hdr.fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
244 pkt->hdr.buf_alloc = cpu_to_le32(vvs->buf_alloc);
245 spin_unlock_bh(&vvs->rx_lock);
247 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
249 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
253 spin_lock_bh(&vvs->tx_lock);
254 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
258 spin_unlock_bh(&vvs->tx_lock);
262 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
264 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
266 spin_lock_bh(&vvs->tx_lock);
267 vvs->tx_cnt -= credit;
268 spin_unlock_bh(&vvs->tx_lock);
270 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
272 static int virtio_transport_send_credit_update(struct vsock_sock *vsk,
274 struct virtio_vsock_hdr *hdr)
276 struct virtio_vsock_pkt_info info = {
277 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
282 return virtio_transport_send_pkt_info(vsk, &info);
286 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
290 struct virtio_vsock_sock *vvs = vsk->trans;
291 struct virtio_vsock_pkt *pkt;
292 size_t bytes, total = 0, off;
295 spin_lock_bh(&vvs->rx_lock);
297 list_for_each_entry(pkt, &vvs->rx_queue, list) {
303 while (total < len && off < pkt->len) {
305 if (bytes > pkt->len - off)
306 bytes = pkt->len - off;
308 /* sk_lock is held by caller so no one else can dequeue.
309 * Unlock rx_lock since memcpy_to_msg() may sleep.
311 spin_unlock_bh(&vvs->rx_lock);
313 err = memcpy_to_msg(msg, pkt->buf + off, bytes);
317 spin_lock_bh(&vvs->rx_lock);
324 spin_unlock_bh(&vvs->rx_lock);
335 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
339 struct virtio_vsock_sock *vvs = vsk->trans;
340 struct virtio_vsock_pkt *pkt;
341 size_t bytes, total = 0;
345 spin_lock_bh(&vvs->rx_lock);
346 while (total < len && !list_empty(&vvs->rx_queue)) {
347 pkt = list_first_entry(&vvs->rx_queue,
348 struct virtio_vsock_pkt, list);
351 if (bytes > pkt->len - pkt->off)
352 bytes = pkt->len - pkt->off;
354 /* sk_lock is held by caller so no one else can dequeue.
355 * Unlock rx_lock since memcpy_to_msg() may sleep.
357 spin_unlock_bh(&vvs->rx_lock);
359 err = memcpy_to_msg(msg, pkt->buf + pkt->off, bytes);
363 spin_lock_bh(&vvs->rx_lock);
367 if (pkt->off == pkt->len) {
368 virtio_transport_dec_rx_pkt(vvs, pkt);
369 list_del(&pkt->list);
370 virtio_transport_free_pkt(pkt);
374 free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
376 spin_unlock_bh(&vvs->rx_lock);
378 /* To reduce the number of credit update messages,
379 * don't update credits as long as lots of space is available.
380 * Note: the limit chosen here is arbitrary. Setting the limit
381 * too high causes extra messages. Too low causes transmitter
382 * stalls. As stalls are in theory more expensive than extra
383 * messages, we set the limit to a high value. TODO: experiment
384 * with different values.
386 if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
387 virtio_transport_send_credit_update(vsk,
388 VIRTIO_VSOCK_TYPE_STREAM,
401 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
403 size_t len, int flags)
405 if (flags & MSG_PEEK)
406 return virtio_transport_stream_do_peek(vsk, msg, len);
408 return virtio_transport_stream_do_dequeue(vsk, msg, len);
410 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
413 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
415 size_t len, int flags)
419 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
421 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
423 struct virtio_vsock_sock *vvs = vsk->trans;
426 spin_lock_bh(&vvs->rx_lock);
427 bytes = vvs->rx_bytes;
428 spin_unlock_bh(&vvs->rx_lock);
432 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
434 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
436 struct virtio_vsock_sock *vvs = vsk->trans;
439 bytes = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
446 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
448 struct virtio_vsock_sock *vvs = vsk->trans;
451 spin_lock_bh(&vvs->tx_lock);
452 bytes = virtio_transport_has_space(vsk);
453 spin_unlock_bh(&vvs->tx_lock);
457 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
459 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
460 struct vsock_sock *psk)
462 struct virtio_vsock_sock *vvs;
464 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
470 if (psk && psk->trans) {
471 struct virtio_vsock_sock *ptrans = psk->trans;
473 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
476 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
477 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
479 vvs->buf_alloc = vsk->buffer_size;
481 spin_lock_init(&vvs->rx_lock);
482 spin_lock_init(&vvs->tx_lock);
483 INIT_LIST_HEAD(&vvs->rx_queue);
487 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
489 /* sk_lock held by the caller */
490 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
492 struct virtio_vsock_sock *vvs = vsk->trans;
494 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
495 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
497 vvs->buf_alloc = *val;
499 virtio_transport_send_credit_update(vsk, VIRTIO_VSOCK_TYPE_STREAM,
502 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
505 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
507 bool *data_ready_now)
509 if (vsock_stream_has_data(vsk))
510 *data_ready_now = true;
512 *data_ready_now = false;
516 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
519 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
521 bool *space_avail_now)
525 free_space = vsock_stream_has_space(vsk);
527 *space_avail_now = true;
528 else if (free_space == 0)
529 *space_avail_now = false;
533 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
535 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
536 size_t target, struct vsock_transport_recv_notify_data *data)
540 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
542 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
543 size_t target, struct vsock_transport_recv_notify_data *data)
547 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
549 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
550 size_t target, struct vsock_transport_recv_notify_data *data)
554 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
556 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
557 size_t target, ssize_t copied, bool data_read,
558 struct vsock_transport_recv_notify_data *data)
562 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
564 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
565 struct vsock_transport_send_notify_data *data)
569 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
571 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
572 struct vsock_transport_send_notify_data *data)
576 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
578 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
579 struct vsock_transport_send_notify_data *data)
583 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
585 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
586 ssize_t written, struct vsock_transport_send_notify_data *data)
590 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
592 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
594 return vsk->buffer_size;
596 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
598 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
602 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
604 bool virtio_transport_stream_allow(u32 cid, u32 port)
608 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
610 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
611 struct sockaddr_vm *addr)
615 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
617 bool virtio_transport_dgram_allow(u32 cid, u32 port)
621 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
623 int virtio_transport_connect(struct vsock_sock *vsk)
625 struct virtio_vsock_pkt_info info = {
626 .op = VIRTIO_VSOCK_OP_REQUEST,
627 .type = VIRTIO_VSOCK_TYPE_STREAM,
631 return virtio_transport_send_pkt_info(vsk, &info);
633 EXPORT_SYMBOL_GPL(virtio_transport_connect);
635 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
637 struct virtio_vsock_pkt_info info = {
638 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
639 .type = VIRTIO_VSOCK_TYPE_STREAM,
640 .flags = (mode & RCV_SHUTDOWN ?
641 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
642 (mode & SEND_SHUTDOWN ?
643 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
647 return virtio_transport_send_pkt_info(vsk, &info);
649 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
652 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
653 struct sockaddr_vm *remote_addr,
659 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
662 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
666 struct virtio_vsock_pkt_info info = {
667 .op = VIRTIO_VSOCK_OP_RW,
668 .type = VIRTIO_VSOCK_TYPE_STREAM,
674 return virtio_transport_send_pkt_info(vsk, &info);
676 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
678 void virtio_transport_destruct(struct vsock_sock *vsk)
680 struct virtio_vsock_sock *vvs = vsk->trans;
684 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
686 static int virtio_transport_reset(struct vsock_sock *vsk,
687 struct virtio_vsock_pkt *pkt)
689 struct virtio_vsock_pkt_info info = {
690 .op = VIRTIO_VSOCK_OP_RST,
691 .type = VIRTIO_VSOCK_TYPE_STREAM,
696 /* Send RST only if the original pkt is not a RST pkt */
697 if (pkt && le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
700 return virtio_transport_send_pkt_info(vsk, &info);
703 /* Normally packets are associated with a socket. There may be no socket if an
704 * attempt was made to connect to a socket that does not exist.
706 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
707 struct virtio_vsock_pkt *pkt)
709 struct virtio_vsock_pkt *reply;
710 struct virtio_vsock_pkt_info info = {
711 .op = VIRTIO_VSOCK_OP_RST,
712 .type = le16_to_cpu(pkt->hdr.type),
716 /* Send RST only if the original pkt is not a RST pkt */
717 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
720 reply = virtio_transport_alloc_pkt(&info, 0,
721 le64_to_cpu(pkt->hdr.dst_cid),
722 le32_to_cpu(pkt->hdr.dst_port),
723 le64_to_cpu(pkt->hdr.src_cid),
724 le32_to_cpu(pkt->hdr.src_port));
729 virtio_transport_free_pkt(reply);
733 return t->send_pkt(reply);
736 /* This function should be called with sk_lock held and SOCK_DONE set */
737 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
739 struct virtio_vsock_sock *vvs = vsk->trans;
740 struct virtio_vsock_pkt *pkt, *tmp;
742 /* We don't need to take rx_lock, as the socket is closing and we are
745 list_for_each_entry_safe(pkt, tmp, &vvs->rx_queue, list) {
746 list_del(&pkt->list);
747 virtio_transport_free_pkt(pkt);
750 vsock_remove_sock(vsk);
753 static void virtio_transport_wait_close(struct sock *sk, long timeout)
756 DEFINE_WAIT_FUNC(wait, woken_wake_function);
758 add_wait_queue(sk_sleep(sk), &wait);
761 if (sk_wait_event(sk, &timeout,
762 sock_flag(sk, SOCK_DONE), &wait))
764 } while (!signal_pending(current) && timeout);
766 remove_wait_queue(sk_sleep(sk), &wait);
770 static void virtio_transport_do_close(struct vsock_sock *vsk,
773 struct sock *sk = sk_vsock(vsk);
775 sock_set_flag(sk, SOCK_DONE);
776 vsk->peer_shutdown = SHUTDOWN_MASK;
777 if (vsock_stream_has_data(vsk) <= 0)
778 sk->sk_state = TCP_CLOSING;
779 sk->sk_state_change(sk);
781 if (vsk->close_work_scheduled &&
782 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
783 vsk->close_work_scheduled = false;
785 virtio_transport_remove_sock(vsk);
787 /* Release refcnt obtained when we scheduled the timeout */
792 static void virtio_transport_close_timeout(struct work_struct *work)
794 struct vsock_sock *vsk =
795 container_of(work, struct vsock_sock, close_work.work);
796 struct sock *sk = sk_vsock(vsk);
801 if (!sock_flag(sk, SOCK_DONE)) {
802 (void)virtio_transport_reset(vsk, NULL);
804 virtio_transport_do_close(vsk, false);
807 vsk->close_work_scheduled = false;
813 /* User context, vsk->sk is locked */
814 static bool virtio_transport_close(struct vsock_sock *vsk)
816 struct sock *sk = &vsk->sk;
818 if (!(sk->sk_state == TCP_ESTABLISHED ||
819 sk->sk_state == TCP_CLOSING))
822 /* Already received SHUTDOWN from peer, reply with RST */
823 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
824 (void)virtio_transport_reset(vsk, NULL);
828 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
829 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
831 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
832 virtio_transport_wait_close(sk, sk->sk_lingertime);
834 if (sock_flag(sk, SOCK_DONE)) {
839 INIT_DELAYED_WORK(&vsk->close_work,
840 virtio_transport_close_timeout);
841 vsk->close_work_scheduled = true;
842 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
846 void virtio_transport_release(struct vsock_sock *vsk)
848 struct sock *sk = &vsk->sk;
849 bool remove_sock = true;
851 if (sk->sk_type == SOCK_STREAM)
852 remove_sock = virtio_transport_close(vsk);
855 sock_set_flag(sk, SOCK_DONE);
856 virtio_transport_remove_sock(vsk);
859 EXPORT_SYMBOL_GPL(virtio_transport_release);
862 virtio_transport_recv_connecting(struct sock *sk,
863 struct virtio_vsock_pkt *pkt)
865 struct vsock_sock *vsk = vsock_sk(sk);
869 switch (le16_to_cpu(pkt->hdr.op)) {
870 case VIRTIO_VSOCK_OP_RESPONSE:
871 sk->sk_state = TCP_ESTABLISHED;
872 sk->sk_socket->state = SS_CONNECTED;
873 vsock_insert_connected(vsk);
874 sk->sk_state_change(sk);
876 case VIRTIO_VSOCK_OP_INVALID:
878 case VIRTIO_VSOCK_OP_RST:
890 virtio_transport_reset(vsk, pkt);
891 sk->sk_state = TCP_CLOSE;
893 sk->sk_error_report(sk);
898 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
899 struct virtio_vsock_pkt *pkt)
901 struct virtio_vsock_sock *vvs = vsk->trans;
902 bool can_enqueue, free_pkt = false;
904 pkt->len = le32_to_cpu(pkt->hdr.len);
907 spin_lock_bh(&vvs->rx_lock);
909 can_enqueue = virtio_transport_inc_rx_pkt(vvs, pkt);
915 /* Try to copy small packets into the buffer of last packet queued,
916 * to avoid wasting memory queueing the entire buffer with a small
919 if (pkt->len <= GOOD_COPY_LEN && !list_empty(&vvs->rx_queue)) {
920 struct virtio_vsock_pkt *last_pkt;
922 last_pkt = list_last_entry(&vvs->rx_queue,
923 struct virtio_vsock_pkt, list);
925 /* If there is space in the last packet queued, we copy the
926 * new packet in its buffer.
928 if (pkt->len <= last_pkt->buf_len - last_pkt->len) {
929 memcpy(last_pkt->buf + last_pkt->len, pkt->buf,
931 last_pkt->len += pkt->len;
937 list_add_tail(&pkt->list, &vvs->rx_queue);
940 spin_unlock_bh(&vvs->rx_lock);
942 virtio_transport_free_pkt(pkt);
946 virtio_transport_recv_connected(struct sock *sk,
947 struct virtio_vsock_pkt *pkt)
949 struct vsock_sock *vsk = vsock_sk(sk);
952 switch (le16_to_cpu(pkt->hdr.op)) {
953 case VIRTIO_VSOCK_OP_RW:
954 virtio_transport_recv_enqueue(vsk, pkt);
955 sk->sk_data_ready(sk);
957 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
958 sk->sk_write_space(sk);
960 case VIRTIO_VSOCK_OP_SHUTDOWN:
961 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
962 vsk->peer_shutdown |= RCV_SHUTDOWN;
963 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
964 vsk->peer_shutdown |= SEND_SHUTDOWN;
965 if (vsk->peer_shutdown == SHUTDOWN_MASK &&
966 vsock_stream_has_data(vsk) <= 0 &&
967 !sock_flag(sk, SOCK_DONE)) {
968 (void)virtio_transport_reset(vsk, NULL);
970 virtio_transport_do_close(vsk, true);
972 if (le32_to_cpu(pkt->hdr.flags))
973 sk->sk_state_change(sk);
975 case VIRTIO_VSOCK_OP_RST:
976 virtio_transport_do_close(vsk, true);
983 virtio_transport_free_pkt(pkt);
988 virtio_transport_recv_disconnecting(struct sock *sk,
989 struct virtio_vsock_pkt *pkt)
991 struct vsock_sock *vsk = vsock_sk(sk);
993 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
994 virtio_transport_do_close(vsk, true);
998 virtio_transport_send_response(struct vsock_sock *vsk,
999 struct virtio_vsock_pkt *pkt)
1001 struct virtio_vsock_pkt_info info = {
1002 .op = VIRTIO_VSOCK_OP_RESPONSE,
1003 .type = VIRTIO_VSOCK_TYPE_STREAM,
1004 .remote_cid = le64_to_cpu(pkt->hdr.src_cid),
1005 .remote_port = le32_to_cpu(pkt->hdr.src_port),
1010 return virtio_transport_send_pkt_info(vsk, &info);
1013 static bool virtio_transport_space_update(struct sock *sk,
1014 struct virtio_vsock_pkt *pkt)
1016 struct vsock_sock *vsk = vsock_sk(sk);
1017 struct virtio_vsock_sock *vvs = vsk->trans;
1018 bool space_available;
1020 /* Listener sockets are not associated with any transport, so we are
1021 * not able to take the state to see if there is space available in the
1022 * remote peer, but since they are only used to receive requests, we
1023 * can assume that there is always space available in the other peer.
1028 /* buf_alloc and fwd_cnt is always included in the hdr */
1029 spin_lock_bh(&vvs->tx_lock);
1030 vvs->peer_buf_alloc = le32_to_cpu(pkt->hdr.buf_alloc);
1031 vvs->peer_fwd_cnt = le32_to_cpu(pkt->hdr.fwd_cnt);
1032 space_available = virtio_transport_has_space(vsk);
1033 spin_unlock_bh(&vvs->tx_lock);
1034 return space_available;
1037 /* Handle server socket */
1039 virtio_transport_recv_listen(struct sock *sk, struct virtio_vsock_pkt *pkt,
1040 struct virtio_transport *t)
1042 struct vsock_sock *vsk = vsock_sk(sk);
1043 struct vsock_sock *vchild;
1047 if (le16_to_cpu(pkt->hdr.op) != VIRTIO_VSOCK_OP_REQUEST) {
1048 virtio_transport_reset_no_sock(t, pkt);
1052 if (sk_acceptq_is_full(sk)) {
1053 virtio_transport_reset_no_sock(t, pkt);
1057 child = vsock_create_connected(sk);
1059 virtio_transport_reset_no_sock(t, pkt);
1063 sk_acceptq_added(sk);
1065 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1067 child->sk_state = TCP_ESTABLISHED;
1069 vchild = vsock_sk(child);
1070 vsock_addr_init(&vchild->local_addr, le64_to_cpu(pkt->hdr.dst_cid),
1071 le32_to_cpu(pkt->hdr.dst_port));
1072 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(pkt->hdr.src_cid),
1073 le32_to_cpu(pkt->hdr.src_port));
1075 ret = vsock_assign_transport(vchild, vsk);
1076 /* Transport assigned (looking at remote_addr) must be the same
1077 * where we received the request.
1079 if (ret || vchild->transport != &t->transport) {
1080 release_sock(child);
1081 virtio_transport_reset_no_sock(t, pkt);
1086 if (virtio_transport_space_update(child, pkt))
1087 child->sk_write_space(child);
1089 vsock_insert_connected(vchild);
1090 vsock_enqueue_accept(sk, child);
1091 virtio_transport_send_response(vchild, pkt);
1093 release_sock(child);
1095 sk->sk_data_ready(sk);
1099 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1102 void virtio_transport_recv_pkt(struct virtio_transport *t,
1103 struct virtio_vsock_pkt *pkt)
1105 struct sockaddr_vm src, dst;
1106 struct vsock_sock *vsk;
1108 bool space_available;
1110 vsock_addr_init(&src, le64_to_cpu(pkt->hdr.src_cid),
1111 le32_to_cpu(pkt->hdr.src_port));
1112 vsock_addr_init(&dst, le64_to_cpu(pkt->hdr.dst_cid),
1113 le32_to_cpu(pkt->hdr.dst_port));
1115 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1116 dst.svm_cid, dst.svm_port,
1117 le32_to_cpu(pkt->hdr.len),
1118 le16_to_cpu(pkt->hdr.type),
1119 le16_to_cpu(pkt->hdr.op),
1120 le32_to_cpu(pkt->hdr.flags),
1121 le32_to_cpu(pkt->hdr.buf_alloc),
1122 le32_to_cpu(pkt->hdr.fwd_cnt));
1124 if (le16_to_cpu(pkt->hdr.type) != VIRTIO_VSOCK_TYPE_STREAM) {
1125 (void)virtio_transport_reset_no_sock(t, pkt);
1129 /* The socket must be in connected or bound table
1130 * otherwise send reset back
1132 sk = vsock_find_connected_socket(&src, &dst);
1134 sk = vsock_find_bound_socket(&dst);
1136 (void)virtio_transport_reset_no_sock(t, pkt);
1145 /* Check if sk has been closed before lock_sock */
1146 if (sock_flag(sk, SOCK_DONE)) {
1147 (void)virtio_transport_reset_no_sock(t, pkt);
1153 space_available = virtio_transport_space_update(sk, pkt);
1155 /* Update CID in case it has changed after a transport reset event */
1156 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1157 vsk->local_addr.svm_cid = dst.svm_cid;
1159 if (space_available)
1160 sk->sk_write_space(sk);
1162 switch (sk->sk_state) {
1164 virtio_transport_recv_listen(sk, pkt, t);
1165 virtio_transport_free_pkt(pkt);
1168 virtio_transport_recv_connecting(sk, pkt);
1169 virtio_transport_free_pkt(pkt);
1171 case TCP_ESTABLISHED:
1172 virtio_transport_recv_connected(sk, pkt);
1175 virtio_transport_recv_disconnecting(sk, pkt);
1176 virtio_transport_free_pkt(pkt);
1179 (void)virtio_transport_reset_no_sock(t, pkt);
1180 virtio_transport_free_pkt(pkt);
1186 /* Release refcnt obtained when we fetched this socket out of the
1187 * bound or connected list.
1193 virtio_transport_free_pkt(pkt);
1195 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1197 void virtio_transport_free_pkt(struct virtio_vsock_pkt *pkt)
1202 EXPORT_SYMBOL_GPL(virtio_transport_free_pkt);
1204 MODULE_LICENSE("GPL v2");
1205 MODULE_AUTHOR("Asias He");
1206 MODULE_DESCRIPTION("common code for virtio vsock");