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 *
33 virtio_transport_get_ops(struct vsock_sock *vsk)
35 const struct vsock_transport *t = vsock_core_get_transport(vsk);
40 return container_of(t, struct virtio_transport, transport);
43 static struct virtio_vsock_pkt *
44 virtio_transport_alloc_pkt(struct virtio_vsock_pkt_info *info,
51 struct virtio_vsock_pkt *pkt;
54 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
58 pkt->hdr.type = cpu_to_le16(info->type);
59 pkt->hdr.op = cpu_to_le16(info->op);
60 pkt->hdr.src_cid = cpu_to_le64(src_cid);
61 pkt->hdr.dst_cid = cpu_to_le64(dst_cid);
62 pkt->hdr.src_port = cpu_to_le32(src_port);
63 pkt->hdr.dst_port = cpu_to_le32(dst_port);
64 pkt->hdr.flags = cpu_to_le32(info->flags);
66 pkt->hdr.len = cpu_to_le32(len);
67 pkt->reply = info->reply;
70 if (info->msg && len > 0) {
71 pkt->buf = kmalloc(len, GFP_KERNEL);
77 err = memcpy_from_msg(pkt->buf, info->msg, len);
82 trace_virtio_transport_alloc_pkt(src_cid, src_port,
99 static struct sk_buff *virtio_transport_build_skb(void *opaque)
101 struct virtio_vsock_pkt *pkt = opaque;
102 struct af_vsockmon_hdr *hdr;
107 /* A packet could be split to fit the RX buffer, so we can retrieve
108 * the payload length from the header and the buffer pointer taking
109 * care of the offset in the original packet.
111 payload_len = le32_to_cpu(pkt->hdr.len);
112 payload_buf = pkt->buf + pkt->off;
114 skb = alloc_skb(sizeof(*hdr) + sizeof(pkt->hdr) + payload_len,
119 hdr = skb_put(skb, sizeof(*hdr));
121 /* pkt->hdr is little-endian so no need to byteswap here */
122 hdr->src_cid = pkt->hdr.src_cid;
123 hdr->src_port = pkt->hdr.src_port;
124 hdr->dst_cid = pkt->hdr.dst_cid;
125 hdr->dst_port = pkt->hdr.dst_port;
127 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
128 hdr->len = cpu_to_le16(sizeof(pkt->hdr));
129 memset(hdr->reserved, 0, sizeof(hdr->reserved));
131 switch (le16_to_cpu(pkt->hdr.op)) {
132 case VIRTIO_VSOCK_OP_REQUEST:
133 case VIRTIO_VSOCK_OP_RESPONSE:
134 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
136 case VIRTIO_VSOCK_OP_RST:
137 case VIRTIO_VSOCK_OP_SHUTDOWN:
138 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
140 case VIRTIO_VSOCK_OP_RW:
141 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
143 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
144 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
145 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
148 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
152 skb_put_data(skb, &pkt->hdr, sizeof(pkt->hdr));
155 skb_put_data(skb, payload_buf, payload_len);
161 void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt)
163 vsock_deliver_tap(virtio_transport_build_skb, pkt);
165 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
167 /* This function can only be used on connecting/connected sockets,
168 * since a socket assigned to a transport is required.
170 * Do not use on listener sockets!
172 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
173 struct virtio_vsock_pkt_info *info)
175 u32 src_cid, src_port, dst_cid, dst_port;
176 const struct virtio_transport *t_ops;
177 struct virtio_vsock_sock *vvs;
178 struct virtio_vsock_pkt *pkt;
179 u32 pkt_len = info->pkt_len;
181 t_ops = virtio_transport_get_ops(vsk);
182 if (unlikely(!t_ops))
185 src_cid = t_ops->transport.get_local_cid();
186 src_port = vsk->local_addr.svm_port;
187 if (!info->remote_cid) {
188 dst_cid = vsk->remote_addr.svm_cid;
189 dst_port = vsk->remote_addr.svm_port;
191 dst_cid = info->remote_cid;
192 dst_port = info->remote_port;
197 /* we can send less than pkt_len bytes */
198 if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
199 pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
201 /* virtio_transport_get_credit might return less than pkt_len credit */
202 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
204 /* Do not send zero length OP_RW pkt */
205 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
208 pkt = virtio_transport_alloc_pkt(info, pkt_len,
212 virtio_transport_put_credit(vvs, pkt_len);
216 virtio_transport_inc_tx_pkt(vvs, pkt);
218 return t_ops->send_pkt(pkt);
221 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
222 struct virtio_vsock_pkt *pkt)
224 if (vvs->rx_bytes + pkt->len > vvs->buf_alloc)
227 vvs->rx_bytes += pkt->len;
231 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
232 struct virtio_vsock_pkt *pkt)
234 vvs->rx_bytes -= pkt->len;
235 vvs->fwd_cnt += pkt->len;
238 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct virtio_vsock_pkt *pkt)
240 spin_lock_bh(&vvs->rx_lock);
241 vvs->last_fwd_cnt = vvs->fwd_cnt;
242 pkt->hdr.fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
243 pkt->hdr.buf_alloc = cpu_to_le32(vvs->buf_alloc);
244 spin_unlock_bh(&vvs->rx_lock);
246 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
248 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
252 spin_lock_bh(&vvs->tx_lock);
253 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
257 spin_unlock_bh(&vvs->tx_lock);
261 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
263 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
265 spin_lock_bh(&vvs->tx_lock);
266 vvs->tx_cnt -= credit;
267 spin_unlock_bh(&vvs->tx_lock);
269 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
271 static int virtio_transport_send_credit_update(struct vsock_sock *vsk,
273 struct virtio_vsock_hdr *hdr)
275 struct virtio_vsock_pkt_info info = {
276 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
281 return virtio_transport_send_pkt_info(vsk, &info);
285 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
289 struct virtio_vsock_sock *vvs = vsk->trans;
290 struct virtio_vsock_pkt *pkt;
291 size_t bytes, total = 0, off;
294 spin_lock_bh(&vvs->rx_lock);
296 list_for_each_entry(pkt, &vvs->rx_queue, list) {
302 while (total < len && off < pkt->len) {
304 if (bytes > pkt->len - off)
305 bytes = pkt->len - off;
307 /* sk_lock is held by caller so no one else can dequeue.
308 * Unlock rx_lock since memcpy_to_msg() may sleep.
310 spin_unlock_bh(&vvs->rx_lock);
312 err = memcpy_to_msg(msg, pkt->buf + off, bytes);
316 spin_lock_bh(&vvs->rx_lock);
323 spin_unlock_bh(&vvs->rx_lock);
334 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
338 struct virtio_vsock_sock *vvs = vsk->trans;
339 struct virtio_vsock_pkt *pkt;
340 size_t bytes, total = 0;
344 spin_lock_bh(&vvs->rx_lock);
345 while (total < len && !list_empty(&vvs->rx_queue)) {
346 pkt = list_first_entry(&vvs->rx_queue,
347 struct virtio_vsock_pkt, list);
350 if (bytes > pkt->len - pkt->off)
351 bytes = pkt->len - pkt->off;
353 /* sk_lock is held by caller so no one else can dequeue.
354 * Unlock rx_lock since memcpy_to_msg() may sleep.
356 spin_unlock_bh(&vvs->rx_lock);
358 err = memcpy_to_msg(msg, pkt->buf + pkt->off, bytes);
362 spin_lock_bh(&vvs->rx_lock);
366 if (pkt->off == pkt->len) {
367 virtio_transport_dec_rx_pkt(vvs, pkt);
368 list_del(&pkt->list);
369 virtio_transport_free_pkt(pkt);
373 free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
375 spin_unlock_bh(&vvs->rx_lock);
377 /* To reduce the number of credit update messages,
378 * don't update credits as long as lots of space is available.
379 * Note: the limit chosen here is arbitrary. Setting the limit
380 * too high causes extra messages. Too low causes transmitter
381 * stalls. As stalls are in theory more expensive than extra
382 * messages, we set the limit to a high value. TODO: experiment
383 * with different values.
385 if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
386 virtio_transport_send_credit_update(vsk,
387 VIRTIO_VSOCK_TYPE_STREAM,
400 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
402 size_t len, int flags)
404 if (flags & MSG_PEEK)
405 return virtio_transport_stream_do_peek(vsk, msg, len);
407 return virtio_transport_stream_do_dequeue(vsk, msg, len);
409 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
412 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
414 size_t len, int flags)
418 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
420 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
422 struct virtio_vsock_sock *vvs = vsk->trans;
425 spin_lock_bh(&vvs->rx_lock);
426 bytes = vvs->rx_bytes;
427 spin_unlock_bh(&vvs->rx_lock);
431 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
433 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
435 struct virtio_vsock_sock *vvs = vsk->trans;
438 bytes = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
445 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
447 struct virtio_vsock_sock *vvs = vsk->trans;
450 spin_lock_bh(&vvs->tx_lock);
451 bytes = virtio_transport_has_space(vsk);
452 spin_unlock_bh(&vvs->tx_lock);
456 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
458 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
459 struct vsock_sock *psk)
461 struct virtio_vsock_sock *vvs;
463 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
469 if (psk && psk->trans) {
470 struct virtio_vsock_sock *ptrans = psk->trans;
472 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
475 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
476 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
478 vvs->buf_alloc = vsk->buffer_size;
480 spin_lock_init(&vvs->rx_lock);
481 spin_lock_init(&vvs->tx_lock);
482 INIT_LIST_HEAD(&vvs->rx_queue);
486 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
488 /* sk_lock held by the caller */
489 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
491 struct virtio_vsock_sock *vvs = vsk->trans;
493 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
494 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
496 vvs->buf_alloc = *val;
498 virtio_transport_send_credit_update(vsk, VIRTIO_VSOCK_TYPE_STREAM,
501 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
504 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
506 bool *data_ready_now)
508 if (vsock_stream_has_data(vsk))
509 *data_ready_now = true;
511 *data_ready_now = false;
515 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
518 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
520 bool *space_avail_now)
524 free_space = vsock_stream_has_space(vsk);
526 *space_avail_now = true;
527 else if (free_space == 0)
528 *space_avail_now = false;
532 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
534 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
535 size_t target, struct vsock_transport_recv_notify_data *data)
539 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
541 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
542 size_t target, struct vsock_transport_recv_notify_data *data)
546 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
548 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
549 size_t target, struct vsock_transport_recv_notify_data *data)
553 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
555 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
556 size_t target, ssize_t copied, bool data_read,
557 struct vsock_transport_recv_notify_data *data)
561 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
563 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
564 struct vsock_transport_send_notify_data *data)
568 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
570 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
571 struct vsock_transport_send_notify_data *data)
575 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
577 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
578 struct vsock_transport_send_notify_data *data)
582 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
584 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
585 ssize_t written, struct vsock_transport_send_notify_data *data)
589 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
591 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
593 return vsk->buffer_size;
595 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
597 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
601 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
603 bool virtio_transport_stream_allow(u32 cid, u32 port)
607 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
609 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
610 struct sockaddr_vm *addr)
614 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
616 bool virtio_transport_dgram_allow(u32 cid, u32 port)
620 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
622 int virtio_transport_connect(struct vsock_sock *vsk)
624 struct virtio_vsock_pkt_info info = {
625 .op = VIRTIO_VSOCK_OP_REQUEST,
626 .type = VIRTIO_VSOCK_TYPE_STREAM,
630 return virtio_transport_send_pkt_info(vsk, &info);
632 EXPORT_SYMBOL_GPL(virtio_transport_connect);
634 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
636 struct virtio_vsock_pkt_info info = {
637 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
638 .type = VIRTIO_VSOCK_TYPE_STREAM,
639 .flags = (mode & RCV_SHUTDOWN ?
640 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
641 (mode & SEND_SHUTDOWN ?
642 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
646 return virtio_transport_send_pkt_info(vsk, &info);
648 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
651 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
652 struct sockaddr_vm *remote_addr,
658 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
661 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
665 struct virtio_vsock_pkt_info info = {
666 .op = VIRTIO_VSOCK_OP_RW,
667 .type = VIRTIO_VSOCK_TYPE_STREAM,
673 return virtio_transport_send_pkt_info(vsk, &info);
675 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
677 void virtio_transport_destruct(struct vsock_sock *vsk)
679 struct virtio_vsock_sock *vvs = vsk->trans;
683 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
685 static int virtio_transport_reset(struct vsock_sock *vsk,
686 struct virtio_vsock_pkt *pkt)
688 struct virtio_vsock_pkt_info info = {
689 .op = VIRTIO_VSOCK_OP_RST,
690 .type = VIRTIO_VSOCK_TYPE_STREAM,
695 /* Send RST only if the original pkt is not a RST pkt */
696 if (pkt && le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
699 return virtio_transport_send_pkt_info(vsk, &info);
702 /* Normally packets are associated with a socket. There may be no socket if an
703 * attempt was made to connect to a socket that does not exist.
705 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
706 struct virtio_vsock_pkt *pkt)
708 struct virtio_vsock_pkt *reply;
709 struct virtio_vsock_pkt_info info = {
710 .op = VIRTIO_VSOCK_OP_RST,
711 .type = le16_to_cpu(pkt->hdr.type),
715 /* Send RST only if the original pkt is not a RST pkt */
716 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
719 reply = virtio_transport_alloc_pkt(&info, 0,
720 le64_to_cpu(pkt->hdr.dst_cid),
721 le32_to_cpu(pkt->hdr.dst_port),
722 le64_to_cpu(pkt->hdr.src_cid),
723 le32_to_cpu(pkt->hdr.src_port));
728 virtio_transport_free_pkt(reply);
732 return t->send_pkt(reply);
735 static void virtio_transport_wait_close(struct sock *sk, long timeout)
738 DEFINE_WAIT_FUNC(wait, woken_wake_function);
740 add_wait_queue(sk_sleep(sk), &wait);
743 if (sk_wait_event(sk, &timeout,
744 sock_flag(sk, SOCK_DONE), &wait))
746 } while (!signal_pending(current) && timeout);
748 remove_wait_queue(sk_sleep(sk), &wait);
752 static void virtio_transport_do_close(struct vsock_sock *vsk,
755 struct sock *sk = sk_vsock(vsk);
757 sock_set_flag(sk, SOCK_DONE);
758 vsk->peer_shutdown = SHUTDOWN_MASK;
759 if (vsock_stream_has_data(vsk) <= 0)
760 sk->sk_state = TCP_CLOSING;
761 sk->sk_state_change(sk);
763 if (vsk->close_work_scheduled &&
764 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
765 vsk->close_work_scheduled = false;
767 vsock_remove_sock(vsk);
769 /* Release refcnt obtained when we scheduled the timeout */
774 static void virtio_transport_close_timeout(struct work_struct *work)
776 struct vsock_sock *vsk =
777 container_of(work, struct vsock_sock, close_work.work);
778 struct sock *sk = sk_vsock(vsk);
783 if (!sock_flag(sk, SOCK_DONE)) {
784 (void)virtio_transport_reset(vsk, NULL);
786 virtio_transport_do_close(vsk, false);
789 vsk->close_work_scheduled = false;
795 /* User context, vsk->sk is locked */
796 static bool virtio_transport_close(struct vsock_sock *vsk)
798 struct sock *sk = &vsk->sk;
800 if (!(sk->sk_state == TCP_ESTABLISHED ||
801 sk->sk_state == TCP_CLOSING))
804 /* Already received SHUTDOWN from peer, reply with RST */
805 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
806 (void)virtio_transport_reset(vsk, NULL);
810 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
811 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
813 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
814 virtio_transport_wait_close(sk, sk->sk_lingertime);
816 if (sock_flag(sk, SOCK_DONE)) {
821 INIT_DELAYED_WORK(&vsk->close_work,
822 virtio_transport_close_timeout);
823 vsk->close_work_scheduled = true;
824 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
828 void virtio_transport_release(struct vsock_sock *vsk)
830 struct virtio_vsock_sock *vvs = vsk->trans;
831 struct virtio_vsock_pkt *pkt, *tmp;
832 struct sock *sk = &vsk->sk;
833 bool remove_sock = true;
835 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
836 if (sk->sk_type == SOCK_STREAM)
837 remove_sock = virtio_transport_close(vsk);
839 list_for_each_entry_safe(pkt, tmp, &vvs->rx_queue, list) {
840 list_del(&pkt->list);
841 virtio_transport_free_pkt(pkt);
846 vsock_remove_sock(vsk);
848 EXPORT_SYMBOL_GPL(virtio_transport_release);
851 virtio_transport_recv_connecting(struct sock *sk,
852 struct virtio_vsock_pkt *pkt)
854 struct vsock_sock *vsk = vsock_sk(sk);
858 switch (le16_to_cpu(pkt->hdr.op)) {
859 case VIRTIO_VSOCK_OP_RESPONSE:
860 sk->sk_state = TCP_ESTABLISHED;
861 sk->sk_socket->state = SS_CONNECTED;
862 vsock_insert_connected(vsk);
863 sk->sk_state_change(sk);
865 case VIRTIO_VSOCK_OP_INVALID:
867 case VIRTIO_VSOCK_OP_RST:
879 virtio_transport_reset(vsk, pkt);
880 sk->sk_state = TCP_CLOSE;
882 sk->sk_error_report(sk);
887 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
888 struct virtio_vsock_pkt *pkt)
890 struct virtio_vsock_sock *vvs = vsk->trans;
891 bool can_enqueue, free_pkt = false;
893 pkt->len = le32_to_cpu(pkt->hdr.len);
896 spin_lock_bh(&vvs->rx_lock);
898 can_enqueue = virtio_transport_inc_rx_pkt(vvs, pkt);
904 /* Try to copy small packets into the buffer of last packet queued,
905 * to avoid wasting memory queueing the entire buffer with a small
908 if (pkt->len <= GOOD_COPY_LEN && !list_empty(&vvs->rx_queue)) {
909 struct virtio_vsock_pkt *last_pkt;
911 last_pkt = list_last_entry(&vvs->rx_queue,
912 struct virtio_vsock_pkt, list);
914 /* If there is space in the last packet queued, we copy the
915 * new packet in its buffer.
917 if (pkt->len <= last_pkt->buf_len - last_pkt->len) {
918 memcpy(last_pkt->buf + last_pkt->len, pkt->buf,
920 last_pkt->len += pkt->len;
926 list_add_tail(&pkt->list, &vvs->rx_queue);
929 spin_unlock_bh(&vvs->rx_lock);
931 virtio_transport_free_pkt(pkt);
935 virtio_transport_recv_connected(struct sock *sk,
936 struct virtio_vsock_pkt *pkt)
938 struct vsock_sock *vsk = vsock_sk(sk);
941 switch (le16_to_cpu(pkt->hdr.op)) {
942 case VIRTIO_VSOCK_OP_RW:
943 virtio_transport_recv_enqueue(vsk, pkt);
944 sk->sk_data_ready(sk);
946 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
947 sk->sk_write_space(sk);
949 case VIRTIO_VSOCK_OP_SHUTDOWN:
950 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
951 vsk->peer_shutdown |= RCV_SHUTDOWN;
952 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
953 vsk->peer_shutdown |= SEND_SHUTDOWN;
954 if (vsk->peer_shutdown == SHUTDOWN_MASK &&
955 vsock_stream_has_data(vsk) <= 0 &&
956 !sock_flag(sk, SOCK_DONE)) {
957 (void)virtio_transport_reset(vsk, NULL);
959 virtio_transport_do_close(vsk, true);
961 if (le32_to_cpu(pkt->hdr.flags))
962 sk->sk_state_change(sk);
964 case VIRTIO_VSOCK_OP_RST:
965 virtio_transport_do_close(vsk, true);
972 virtio_transport_free_pkt(pkt);
977 virtio_transport_recv_disconnecting(struct sock *sk,
978 struct virtio_vsock_pkt *pkt)
980 struct vsock_sock *vsk = vsock_sk(sk);
982 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
983 virtio_transport_do_close(vsk, true);
987 virtio_transport_send_response(struct vsock_sock *vsk,
988 struct virtio_vsock_pkt *pkt)
990 struct virtio_vsock_pkt_info info = {
991 .op = VIRTIO_VSOCK_OP_RESPONSE,
992 .type = VIRTIO_VSOCK_TYPE_STREAM,
993 .remote_cid = le64_to_cpu(pkt->hdr.src_cid),
994 .remote_port = le32_to_cpu(pkt->hdr.src_port),
999 return virtio_transport_send_pkt_info(vsk, &info);
1002 static bool virtio_transport_space_update(struct sock *sk,
1003 struct virtio_vsock_pkt *pkt)
1005 struct vsock_sock *vsk = vsock_sk(sk);
1006 struct virtio_vsock_sock *vvs = vsk->trans;
1007 bool space_available;
1009 /* Listener sockets are not associated with any transport, so we are
1010 * not able to take the state to see if there is space available in the
1011 * remote peer, but since they are only used to receive requests, we
1012 * can assume that there is always space available in the other peer.
1017 /* buf_alloc and fwd_cnt is always included in the hdr */
1018 spin_lock_bh(&vvs->tx_lock);
1019 vvs->peer_buf_alloc = le32_to_cpu(pkt->hdr.buf_alloc);
1020 vvs->peer_fwd_cnt = le32_to_cpu(pkt->hdr.fwd_cnt);
1021 space_available = virtio_transport_has_space(vsk);
1022 spin_unlock_bh(&vvs->tx_lock);
1023 return space_available;
1026 /* Handle server socket */
1028 virtio_transport_recv_listen(struct sock *sk, struct virtio_vsock_pkt *pkt,
1029 struct virtio_transport *t)
1031 struct vsock_sock *vsk = vsock_sk(sk);
1032 struct vsock_sock *vchild;
1036 if (le16_to_cpu(pkt->hdr.op) != VIRTIO_VSOCK_OP_REQUEST) {
1037 virtio_transport_reset_no_sock(t, pkt);
1041 if (sk_acceptq_is_full(sk)) {
1042 virtio_transport_reset_no_sock(t, pkt);
1046 child = vsock_create_connected(sk);
1048 virtio_transport_reset_no_sock(t, pkt);
1052 sk_acceptq_added(sk);
1054 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1056 child->sk_state = TCP_ESTABLISHED;
1058 vchild = vsock_sk(child);
1059 vsock_addr_init(&vchild->local_addr, le64_to_cpu(pkt->hdr.dst_cid),
1060 le32_to_cpu(pkt->hdr.dst_port));
1061 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(pkt->hdr.src_cid),
1062 le32_to_cpu(pkt->hdr.src_port));
1064 ret = vsock_assign_transport(vchild, vsk);
1065 /* Transport assigned (looking at remote_addr) must be the same
1066 * where we received the request.
1068 if (ret || vchild->transport != &t->transport) {
1069 release_sock(child);
1070 virtio_transport_reset_no_sock(t, pkt);
1075 if (virtio_transport_space_update(child, pkt))
1076 child->sk_write_space(child);
1078 vsock_insert_connected(vchild);
1079 vsock_enqueue_accept(sk, child);
1080 virtio_transport_send_response(vchild, pkt);
1082 release_sock(child);
1084 sk->sk_data_ready(sk);
1088 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1091 void virtio_transport_recv_pkt(struct virtio_transport *t,
1092 struct virtio_vsock_pkt *pkt)
1094 struct sockaddr_vm src, dst;
1095 struct vsock_sock *vsk;
1097 bool space_available;
1099 vsock_addr_init(&src, le64_to_cpu(pkt->hdr.src_cid),
1100 le32_to_cpu(pkt->hdr.src_port));
1101 vsock_addr_init(&dst, le64_to_cpu(pkt->hdr.dst_cid),
1102 le32_to_cpu(pkt->hdr.dst_port));
1104 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1105 dst.svm_cid, dst.svm_port,
1106 le32_to_cpu(pkt->hdr.len),
1107 le16_to_cpu(pkt->hdr.type),
1108 le16_to_cpu(pkt->hdr.op),
1109 le32_to_cpu(pkt->hdr.flags),
1110 le32_to_cpu(pkt->hdr.buf_alloc),
1111 le32_to_cpu(pkt->hdr.fwd_cnt));
1113 if (le16_to_cpu(pkt->hdr.type) != VIRTIO_VSOCK_TYPE_STREAM) {
1114 (void)virtio_transport_reset_no_sock(t, pkt);
1118 /* The socket must be in connected or bound table
1119 * otherwise send reset back
1121 sk = vsock_find_connected_socket(&src, &dst);
1123 sk = vsock_find_bound_socket(&dst);
1125 (void)virtio_transport_reset_no_sock(t, pkt);
1132 space_available = virtio_transport_space_update(sk, pkt);
1136 /* Update CID in case it has changed after a transport reset event */
1137 vsk->local_addr.svm_cid = dst.svm_cid;
1139 if (space_available)
1140 sk->sk_write_space(sk);
1142 switch (sk->sk_state) {
1144 virtio_transport_recv_listen(sk, pkt, t);
1145 virtio_transport_free_pkt(pkt);
1148 virtio_transport_recv_connecting(sk, pkt);
1149 virtio_transport_free_pkt(pkt);
1151 case TCP_ESTABLISHED:
1152 virtio_transport_recv_connected(sk, pkt);
1155 virtio_transport_recv_disconnecting(sk, pkt);
1156 virtio_transport_free_pkt(pkt);
1159 virtio_transport_free_pkt(pkt);
1165 /* Release refcnt obtained when we fetched this socket out of the
1166 * bound or connected list.
1172 virtio_transport_free_pkt(pkt);
1174 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1176 void virtio_transport_free_pkt(struct virtio_vsock_pkt *pkt)
1181 EXPORT_SYMBOL_GPL(virtio_transport_free_pkt);
1183 MODULE_LICENSE("GPL v2");
1184 MODULE_AUTHOR("Asias He");
1185 MODULE_DESCRIPTION("common code for virtio vsock");