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.h>
15 #include <linux/virtio_ids.h>
16 #include <linux/virtio_config.h>
17 #include <linux/virtio_vsock.h>
18 #include <uapi/linux/vsockmon.h>
21 #include <net/af_vsock.h>
23 #define CREATE_TRACE_POINTS
24 #include <trace/events/vsock_virtio_transport_common.h>
26 /* How long to wait for graceful shutdown of a connection */
27 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
29 /* Threshold for detecting small packets to copy */
30 #define GOOD_COPY_LEN 128
32 static const struct virtio_transport *virtio_transport_get_ops(void)
34 const struct vsock_transport *t = vsock_core_get_transport();
36 return container_of(t, struct virtio_transport, transport);
39 static struct virtio_vsock_pkt *
40 virtio_transport_alloc_pkt(struct virtio_vsock_pkt_info *info,
47 struct virtio_vsock_pkt *pkt;
50 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
54 pkt->hdr.type = cpu_to_le16(info->type);
55 pkt->hdr.op = cpu_to_le16(info->op);
56 pkt->hdr.src_cid = cpu_to_le64(src_cid);
57 pkt->hdr.dst_cid = cpu_to_le64(dst_cid);
58 pkt->hdr.src_port = cpu_to_le32(src_port);
59 pkt->hdr.dst_port = cpu_to_le32(dst_port);
60 pkt->hdr.flags = cpu_to_le32(info->flags);
62 pkt->hdr.len = cpu_to_le32(len);
63 pkt->reply = info->reply;
66 if (info->msg && len > 0) {
67 pkt->buf = kmalloc(len, GFP_KERNEL);
73 err = memcpy_from_msg(pkt->buf, info->msg, len);
78 trace_virtio_transport_alloc_pkt(src_cid, src_port,
95 static struct sk_buff *virtio_transport_build_skb(void *opaque)
97 struct virtio_vsock_pkt *pkt = opaque;
98 struct af_vsockmon_hdr *hdr;
103 /* A packet could be split to fit the RX buffer, so we can retrieve
104 * the payload length from the header and the buffer pointer taking
105 * care of the offset in the original packet.
107 payload_len = le32_to_cpu(pkt->hdr.len);
108 payload_buf = pkt->buf + pkt->off;
110 skb = alloc_skb(sizeof(*hdr) + sizeof(pkt->hdr) + payload_len,
115 hdr = skb_put(skb, sizeof(*hdr));
117 /* pkt->hdr is little-endian so no need to byteswap here */
118 hdr->src_cid = pkt->hdr.src_cid;
119 hdr->src_port = pkt->hdr.src_port;
120 hdr->dst_cid = pkt->hdr.dst_cid;
121 hdr->dst_port = pkt->hdr.dst_port;
123 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
124 hdr->len = cpu_to_le16(sizeof(pkt->hdr));
125 memset(hdr->reserved, 0, sizeof(hdr->reserved));
127 switch (le16_to_cpu(pkt->hdr.op)) {
128 case VIRTIO_VSOCK_OP_REQUEST:
129 case VIRTIO_VSOCK_OP_RESPONSE:
130 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
132 case VIRTIO_VSOCK_OP_RST:
133 case VIRTIO_VSOCK_OP_SHUTDOWN:
134 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
136 case VIRTIO_VSOCK_OP_RW:
137 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
139 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
140 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
141 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
144 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
148 skb_put_data(skb, &pkt->hdr, sizeof(pkt->hdr));
151 skb_put_data(skb, payload_buf, payload_len);
157 void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt)
159 vsock_deliver_tap(virtio_transport_build_skb, pkt);
161 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
163 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
164 struct virtio_vsock_pkt_info *info)
166 u32 src_cid, src_port, dst_cid, dst_port;
167 struct virtio_vsock_sock *vvs;
168 struct virtio_vsock_pkt *pkt;
169 u32 pkt_len = info->pkt_len;
171 src_cid = vm_sockets_get_local_cid();
172 src_port = vsk->local_addr.svm_port;
173 if (!info->remote_cid) {
174 dst_cid = vsk->remote_addr.svm_cid;
175 dst_port = vsk->remote_addr.svm_port;
177 dst_cid = info->remote_cid;
178 dst_port = info->remote_port;
183 /* we can send less than pkt_len bytes */
184 if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
185 pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
187 /* virtio_transport_get_credit might return less than pkt_len credit */
188 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
190 /* Do not send zero length OP_RW pkt */
191 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
194 pkt = virtio_transport_alloc_pkt(info, pkt_len,
198 virtio_transport_put_credit(vvs, pkt_len);
202 virtio_transport_inc_tx_pkt(vvs, pkt);
204 return virtio_transport_get_ops()->send_pkt(pkt);
207 static void virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
208 struct virtio_vsock_pkt *pkt)
210 vvs->rx_bytes += pkt->len;
213 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
214 struct virtio_vsock_pkt *pkt)
216 vvs->rx_bytes -= pkt->len;
217 vvs->fwd_cnt += pkt->len;
220 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct virtio_vsock_pkt *pkt)
222 spin_lock_bh(&vvs->rx_lock);
223 vvs->last_fwd_cnt = vvs->fwd_cnt;
224 pkt->hdr.fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
225 pkt->hdr.buf_alloc = cpu_to_le32(vvs->buf_alloc);
226 spin_unlock_bh(&vvs->rx_lock);
228 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
230 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
234 spin_lock_bh(&vvs->tx_lock);
235 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
239 spin_unlock_bh(&vvs->tx_lock);
243 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
245 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
247 spin_lock_bh(&vvs->tx_lock);
248 vvs->tx_cnt -= credit;
249 spin_unlock_bh(&vvs->tx_lock);
251 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
253 static int virtio_transport_send_credit_update(struct vsock_sock *vsk,
255 struct virtio_vsock_hdr *hdr)
257 struct virtio_vsock_pkt_info info = {
258 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
263 return virtio_transport_send_pkt_info(vsk, &info);
267 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
271 struct virtio_vsock_sock *vvs = vsk->trans;
272 struct virtio_vsock_pkt *pkt;
273 size_t bytes, total = 0;
277 spin_lock_bh(&vvs->rx_lock);
278 while (total < len && !list_empty(&vvs->rx_queue)) {
279 pkt = list_first_entry(&vvs->rx_queue,
280 struct virtio_vsock_pkt, list);
283 if (bytes > pkt->len - pkt->off)
284 bytes = pkt->len - pkt->off;
286 /* sk_lock is held by caller so no one else can dequeue.
287 * Unlock rx_lock since memcpy_to_msg() may sleep.
289 spin_unlock_bh(&vvs->rx_lock);
291 err = memcpy_to_msg(msg, pkt->buf + pkt->off, bytes);
295 spin_lock_bh(&vvs->rx_lock);
299 if (pkt->off == pkt->len) {
300 virtio_transport_dec_rx_pkt(vvs, pkt);
301 list_del(&pkt->list);
302 virtio_transport_free_pkt(pkt);
306 free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
308 spin_unlock_bh(&vvs->rx_lock);
310 /* To reduce the number of credit update messages,
311 * don't update credits as long as lots of space is available.
312 * Note: the limit chosen here is arbitrary. Setting the limit
313 * too high causes extra messages. Too low causes transmitter
314 * stalls. As stalls are in theory more expensive than extra
315 * messages, we set the limit to a high value. TODO: experiment
316 * with different values.
318 if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
319 virtio_transport_send_credit_update(vsk,
320 VIRTIO_VSOCK_TYPE_STREAM,
333 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
335 size_t len, int flags)
337 if (flags & MSG_PEEK)
340 return virtio_transport_stream_do_dequeue(vsk, msg, len);
342 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
345 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
347 size_t len, int flags)
351 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
353 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
355 struct virtio_vsock_sock *vvs = vsk->trans;
358 spin_lock_bh(&vvs->rx_lock);
359 bytes = vvs->rx_bytes;
360 spin_unlock_bh(&vvs->rx_lock);
364 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
366 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
368 struct virtio_vsock_sock *vvs = vsk->trans;
371 bytes = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
378 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
380 struct virtio_vsock_sock *vvs = vsk->trans;
383 spin_lock_bh(&vvs->tx_lock);
384 bytes = virtio_transport_has_space(vsk);
385 spin_unlock_bh(&vvs->tx_lock);
389 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
391 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
392 struct vsock_sock *psk)
394 struct virtio_vsock_sock *vvs;
396 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
403 struct virtio_vsock_sock *ptrans = psk->trans;
405 vvs->buf_size = ptrans->buf_size;
406 vvs->buf_size_min = ptrans->buf_size_min;
407 vvs->buf_size_max = ptrans->buf_size_max;
408 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
410 vvs->buf_size = VIRTIO_VSOCK_DEFAULT_BUF_SIZE;
411 vvs->buf_size_min = VIRTIO_VSOCK_DEFAULT_MIN_BUF_SIZE;
412 vvs->buf_size_max = VIRTIO_VSOCK_DEFAULT_MAX_BUF_SIZE;
415 vvs->buf_alloc = vvs->buf_size;
417 spin_lock_init(&vvs->rx_lock);
418 spin_lock_init(&vvs->tx_lock);
419 INIT_LIST_HEAD(&vvs->rx_queue);
423 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
425 u64 virtio_transport_get_buffer_size(struct vsock_sock *vsk)
427 struct virtio_vsock_sock *vvs = vsk->trans;
429 return vvs->buf_size;
431 EXPORT_SYMBOL_GPL(virtio_transport_get_buffer_size);
433 u64 virtio_transport_get_min_buffer_size(struct vsock_sock *vsk)
435 struct virtio_vsock_sock *vvs = vsk->trans;
437 return vvs->buf_size_min;
439 EXPORT_SYMBOL_GPL(virtio_transport_get_min_buffer_size);
441 u64 virtio_transport_get_max_buffer_size(struct vsock_sock *vsk)
443 struct virtio_vsock_sock *vvs = vsk->trans;
445 return vvs->buf_size_max;
447 EXPORT_SYMBOL_GPL(virtio_transport_get_max_buffer_size);
449 void virtio_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
451 struct virtio_vsock_sock *vvs = vsk->trans;
453 if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
454 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
455 if (val < vvs->buf_size_min)
456 vvs->buf_size_min = val;
457 if (val > vvs->buf_size_max)
458 vvs->buf_size_max = val;
460 vvs->buf_alloc = val;
462 EXPORT_SYMBOL_GPL(virtio_transport_set_buffer_size);
464 void virtio_transport_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
466 struct virtio_vsock_sock *vvs = vsk->trans;
468 if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
469 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
470 if (val > vvs->buf_size)
472 vvs->buf_size_min = val;
474 EXPORT_SYMBOL_GPL(virtio_transport_set_min_buffer_size);
476 void virtio_transport_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
478 struct virtio_vsock_sock *vvs = vsk->trans;
480 if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
481 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
482 if (val < vvs->buf_size)
484 vvs->buf_size_max = val;
486 EXPORT_SYMBOL_GPL(virtio_transport_set_max_buffer_size);
489 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
491 bool *data_ready_now)
493 if (vsock_stream_has_data(vsk))
494 *data_ready_now = true;
496 *data_ready_now = false;
500 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
503 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
505 bool *space_avail_now)
509 free_space = vsock_stream_has_space(vsk);
511 *space_avail_now = true;
512 else if (free_space == 0)
513 *space_avail_now = false;
517 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
519 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
520 size_t target, struct vsock_transport_recv_notify_data *data)
524 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
526 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
527 size_t target, struct vsock_transport_recv_notify_data *data)
531 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
533 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
534 size_t target, struct vsock_transport_recv_notify_data *data)
538 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
540 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
541 size_t target, ssize_t copied, bool data_read,
542 struct vsock_transport_recv_notify_data *data)
546 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
548 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
549 struct vsock_transport_send_notify_data *data)
553 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
555 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
556 struct vsock_transport_send_notify_data *data)
560 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
562 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
563 struct vsock_transport_send_notify_data *data)
567 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
569 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
570 ssize_t written, struct vsock_transport_send_notify_data *data)
574 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
576 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
578 struct virtio_vsock_sock *vvs = vsk->trans;
580 return vvs->buf_size;
582 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
584 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
588 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
590 bool virtio_transport_stream_allow(u32 cid, u32 port)
594 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
596 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
597 struct sockaddr_vm *addr)
601 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
603 bool virtio_transport_dgram_allow(u32 cid, u32 port)
607 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
609 int virtio_transport_connect(struct vsock_sock *vsk)
611 struct virtio_vsock_pkt_info info = {
612 .op = VIRTIO_VSOCK_OP_REQUEST,
613 .type = VIRTIO_VSOCK_TYPE_STREAM,
617 return virtio_transport_send_pkt_info(vsk, &info);
619 EXPORT_SYMBOL_GPL(virtio_transport_connect);
621 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
623 struct virtio_vsock_pkt_info info = {
624 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
625 .type = VIRTIO_VSOCK_TYPE_STREAM,
626 .flags = (mode & RCV_SHUTDOWN ?
627 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
628 (mode & SEND_SHUTDOWN ?
629 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
633 return virtio_transport_send_pkt_info(vsk, &info);
635 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
638 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
639 struct sockaddr_vm *remote_addr,
645 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
648 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
652 struct virtio_vsock_pkt_info info = {
653 .op = VIRTIO_VSOCK_OP_RW,
654 .type = VIRTIO_VSOCK_TYPE_STREAM,
660 return virtio_transport_send_pkt_info(vsk, &info);
662 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
664 void virtio_transport_destruct(struct vsock_sock *vsk)
666 struct virtio_vsock_sock *vvs = vsk->trans;
670 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
672 static int virtio_transport_reset(struct vsock_sock *vsk,
673 struct virtio_vsock_pkt *pkt)
675 struct virtio_vsock_pkt_info info = {
676 .op = VIRTIO_VSOCK_OP_RST,
677 .type = VIRTIO_VSOCK_TYPE_STREAM,
682 /* Send RST only if the original pkt is not a RST pkt */
683 if (pkt && le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
686 return virtio_transport_send_pkt_info(vsk, &info);
689 /* Normally packets are associated with a socket. There may be no socket if an
690 * attempt was made to connect to a socket that does not exist.
692 static int virtio_transport_reset_no_sock(struct virtio_vsock_pkt *pkt)
694 const struct virtio_transport *t;
695 struct virtio_vsock_pkt *reply;
696 struct virtio_vsock_pkt_info info = {
697 .op = VIRTIO_VSOCK_OP_RST,
698 .type = le16_to_cpu(pkt->hdr.type),
702 /* Send RST only if the original pkt is not a RST pkt */
703 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
706 reply = virtio_transport_alloc_pkt(&info, 0,
707 le64_to_cpu(pkt->hdr.dst_cid),
708 le32_to_cpu(pkt->hdr.dst_port),
709 le64_to_cpu(pkt->hdr.src_cid),
710 le32_to_cpu(pkt->hdr.src_port));
714 t = virtio_transport_get_ops();
716 virtio_transport_free_pkt(reply);
720 return t->send_pkt(reply);
723 static void virtio_transport_wait_close(struct sock *sk, long timeout)
726 DEFINE_WAIT_FUNC(wait, woken_wake_function);
728 add_wait_queue(sk_sleep(sk), &wait);
731 if (sk_wait_event(sk, &timeout,
732 sock_flag(sk, SOCK_DONE), &wait))
734 } while (!signal_pending(current) && timeout);
736 remove_wait_queue(sk_sleep(sk), &wait);
740 static void virtio_transport_do_close(struct vsock_sock *vsk,
743 struct sock *sk = sk_vsock(vsk);
745 sock_set_flag(sk, SOCK_DONE);
746 vsk->peer_shutdown = SHUTDOWN_MASK;
747 if (vsock_stream_has_data(vsk) <= 0)
748 sk->sk_state = TCP_CLOSING;
749 sk->sk_state_change(sk);
751 if (vsk->close_work_scheduled &&
752 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
753 vsk->close_work_scheduled = false;
755 vsock_remove_sock(vsk);
757 /* Release refcnt obtained when we scheduled the timeout */
762 static void virtio_transport_close_timeout(struct work_struct *work)
764 struct vsock_sock *vsk =
765 container_of(work, struct vsock_sock, close_work.work);
766 struct sock *sk = sk_vsock(vsk);
771 if (!sock_flag(sk, SOCK_DONE)) {
772 (void)virtio_transport_reset(vsk, NULL);
774 virtio_transport_do_close(vsk, false);
777 vsk->close_work_scheduled = false;
783 /* User context, vsk->sk is locked */
784 static bool virtio_transport_close(struct vsock_sock *vsk)
786 struct sock *sk = &vsk->sk;
788 if (!(sk->sk_state == TCP_ESTABLISHED ||
789 sk->sk_state == TCP_CLOSING))
792 /* Already received SHUTDOWN from peer, reply with RST */
793 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
794 (void)virtio_transport_reset(vsk, NULL);
798 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
799 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
801 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
802 virtio_transport_wait_close(sk, sk->sk_lingertime);
804 if (sock_flag(sk, SOCK_DONE)) {
809 INIT_DELAYED_WORK(&vsk->close_work,
810 virtio_transport_close_timeout);
811 vsk->close_work_scheduled = true;
812 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
816 void virtio_transport_release(struct vsock_sock *vsk)
818 struct virtio_vsock_sock *vvs = vsk->trans;
819 struct virtio_vsock_pkt *pkt, *tmp;
820 struct sock *sk = &vsk->sk;
821 bool remove_sock = true;
824 if (sk->sk_type == SOCK_STREAM)
825 remove_sock = virtio_transport_close(vsk);
827 list_for_each_entry_safe(pkt, tmp, &vvs->rx_queue, list) {
828 list_del(&pkt->list);
829 virtio_transport_free_pkt(pkt);
834 vsock_remove_sock(vsk);
836 EXPORT_SYMBOL_GPL(virtio_transport_release);
839 virtio_transport_recv_connecting(struct sock *sk,
840 struct virtio_vsock_pkt *pkt)
842 struct vsock_sock *vsk = vsock_sk(sk);
846 switch (le16_to_cpu(pkt->hdr.op)) {
847 case VIRTIO_VSOCK_OP_RESPONSE:
848 sk->sk_state = TCP_ESTABLISHED;
849 sk->sk_socket->state = SS_CONNECTED;
850 vsock_insert_connected(vsk);
851 sk->sk_state_change(sk);
853 case VIRTIO_VSOCK_OP_INVALID:
855 case VIRTIO_VSOCK_OP_RST:
867 virtio_transport_reset(vsk, pkt);
868 sk->sk_state = TCP_CLOSE;
870 sk->sk_error_report(sk);
875 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
876 struct virtio_vsock_pkt *pkt)
878 struct virtio_vsock_sock *vvs = vsk->trans;
879 bool free_pkt = false;
881 pkt->len = le32_to_cpu(pkt->hdr.len);
884 spin_lock_bh(&vvs->rx_lock);
886 virtio_transport_inc_rx_pkt(vvs, pkt);
888 /* Try to copy small packets into the buffer of last packet queued,
889 * to avoid wasting memory queueing the entire buffer with a small
892 if (pkt->len <= GOOD_COPY_LEN && !list_empty(&vvs->rx_queue)) {
893 struct virtio_vsock_pkt *last_pkt;
895 last_pkt = list_last_entry(&vvs->rx_queue,
896 struct virtio_vsock_pkt, list);
898 /* If there is space in the last packet queued, we copy the
899 * new packet in its buffer.
901 if (pkt->len <= last_pkt->buf_len - last_pkt->len) {
902 memcpy(last_pkt->buf + last_pkt->len, pkt->buf,
904 last_pkt->len += pkt->len;
910 list_add_tail(&pkt->list, &vvs->rx_queue);
913 spin_unlock_bh(&vvs->rx_lock);
915 virtio_transport_free_pkt(pkt);
919 virtio_transport_recv_connected(struct sock *sk,
920 struct virtio_vsock_pkt *pkt)
922 struct vsock_sock *vsk = vsock_sk(sk);
925 switch (le16_to_cpu(pkt->hdr.op)) {
926 case VIRTIO_VSOCK_OP_RW:
927 virtio_transport_recv_enqueue(vsk, pkt);
928 sk->sk_data_ready(sk);
930 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
931 sk->sk_write_space(sk);
933 case VIRTIO_VSOCK_OP_SHUTDOWN:
934 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
935 vsk->peer_shutdown |= RCV_SHUTDOWN;
936 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
937 vsk->peer_shutdown |= SEND_SHUTDOWN;
938 if (vsk->peer_shutdown == SHUTDOWN_MASK &&
939 vsock_stream_has_data(vsk) <= 0) {
940 sock_set_flag(sk, SOCK_DONE);
941 sk->sk_state = TCP_CLOSING;
943 if (le32_to_cpu(pkt->hdr.flags))
944 sk->sk_state_change(sk);
946 case VIRTIO_VSOCK_OP_RST:
947 virtio_transport_do_close(vsk, true);
954 virtio_transport_free_pkt(pkt);
959 virtio_transport_recv_disconnecting(struct sock *sk,
960 struct virtio_vsock_pkt *pkt)
962 struct vsock_sock *vsk = vsock_sk(sk);
964 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
965 virtio_transport_do_close(vsk, true);
969 virtio_transport_send_response(struct vsock_sock *vsk,
970 struct virtio_vsock_pkt *pkt)
972 struct virtio_vsock_pkt_info info = {
973 .op = VIRTIO_VSOCK_OP_RESPONSE,
974 .type = VIRTIO_VSOCK_TYPE_STREAM,
975 .remote_cid = le64_to_cpu(pkt->hdr.src_cid),
976 .remote_port = le32_to_cpu(pkt->hdr.src_port),
981 return virtio_transport_send_pkt_info(vsk, &info);
984 /* Handle server socket */
986 virtio_transport_recv_listen(struct sock *sk, struct virtio_vsock_pkt *pkt)
988 struct vsock_sock *vsk = vsock_sk(sk);
989 struct vsock_sock *vchild;
992 if (le16_to_cpu(pkt->hdr.op) != VIRTIO_VSOCK_OP_REQUEST) {
993 virtio_transport_reset(vsk, pkt);
997 if (sk_acceptq_is_full(sk)) {
998 virtio_transport_reset(vsk, pkt);
1002 child = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1005 virtio_transport_reset(vsk, pkt);
1009 sk->sk_ack_backlog++;
1011 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1013 child->sk_state = TCP_ESTABLISHED;
1015 vchild = vsock_sk(child);
1016 vsock_addr_init(&vchild->local_addr, le64_to_cpu(pkt->hdr.dst_cid),
1017 le32_to_cpu(pkt->hdr.dst_port));
1018 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(pkt->hdr.src_cid),
1019 le32_to_cpu(pkt->hdr.src_port));
1021 vsock_insert_connected(vchild);
1022 vsock_enqueue_accept(sk, child);
1023 virtio_transport_send_response(vchild, pkt);
1025 release_sock(child);
1027 sk->sk_data_ready(sk);
1031 static bool virtio_transport_space_update(struct sock *sk,
1032 struct virtio_vsock_pkt *pkt)
1034 struct vsock_sock *vsk = vsock_sk(sk);
1035 struct virtio_vsock_sock *vvs = vsk->trans;
1036 bool space_available;
1038 /* buf_alloc and fwd_cnt is always included in the hdr */
1039 spin_lock_bh(&vvs->tx_lock);
1040 vvs->peer_buf_alloc = le32_to_cpu(pkt->hdr.buf_alloc);
1041 vvs->peer_fwd_cnt = le32_to_cpu(pkt->hdr.fwd_cnt);
1042 space_available = virtio_transport_has_space(vsk);
1043 spin_unlock_bh(&vvs->tx_lock);
1044 return space_available;
1047 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1050 void virtio_transport_recv_pkt(struct virtio_vsock_pkt *pkt)
1052 struct sockaddr_vm src, dst;
1053 struct vsock_sock *vsk;
1055 bool space_available;
1057 vsock_addr_init(&src, le64_to_cpu(pkt->hdr.src_cid),
1058 le32_to_cpu(pkt->hdr.src_port));
1059 vsock_addr_init(&dst, le64_to_cpu(pkt->hdr.dst_cid),
1060 le32_to_cpu(pkt->hdr.dst_port));
1062 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1063 dst.svm_cid, dst.svm_port,
1064 le32_to_cpu(pkt->hdr.len),
1065 le16_to_cpu(pkt->hdr.type),
1066 le16_to_cpu(pkt->hdr.op),
1067 le32_to_cpu(pkt->hdr.flags),
1068 le32_to_cpu(pkt->hdr.buf_alloc),
1069 le32_to_cpu(pkt->hdr.fwd_cnt));
1071 if (le16_to_cpu(pkt->hdr.type) != VIRTIO_VSOCK_TYPE_STREAM) {
1072 (void)virtio_transport_reset_no_sock(pkt);
1076 /* The socket must be in connected or bound table
1077 * otherwise send reset back
1079 sk = vsock_find_connected_socket(&src, &dst);
1081 sk = vsock_find_bound_socket(&dst);
1083 (void)virtio_transport_reset_no_sock(pkt);
1090 space_available = virtio_transport_space_update(sk, pkt);
1094 /* Update CID in case it has changed after a transport reset event */
1095 vsk->local_addr.svm_cid = dst.svm_cid;
1097 if (space_available)
1098 sk->sk_write_space(sk);
1100 switch (sk->sk_state) {
1102 virtio_transport_recv_listen(sk, pkt);
1103 virtio_transport_free_pkt(pkt);
1106 virtio_transport_recv_connecting(sk, pkt);
1107 virtio_transport_free_pkt(pkt);
1109 case TCP_ESTABLISHED:
1110 virtio_transport_recv_connected(sk, pkt);
1113 virtio_transport_recv_disconnecting(sk, pkt);
1114 virtio_transport_free_pkt(pkt);
1117 virtio_transport_free_pkt(pkt);
1122 /* Release refcnt obtained when we fetched this socket out of the
1123 * bound or connected list.
1129 virtio_transport_free_pkt(pkt);
1131 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1133 void virtio_transport_free_pkt(struct virtio_vsock_pkt *pkt)
1138 EXPORT_SYMBOL_GPL(virtio_transport_free_pkt);
1140 MODULE_LICENSE("GPL v2");
1141 MODULE_AUTHOR("Asias He");
1142 MODULE_DESCRIPTION("common code for virtio vsock");