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 /* Returns a new packet on success, otherwise returns NULL.
42 * If NULL is returned, errp is set to a negative errno.
44 static struct sk_buff *
45 virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info,
52 const size_t skb_len = VIRTIO_VSOCK_SKB_HEADROOM + len;
53 struct virtio_vsock_hdr *hdr;
58 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL);
62 hdr = virtio_vsock_hdr(skb);
63 hdr->type = cpu_to_le16(info->type);
64 hdr->op = cpu_to_le16(info->op);
65 hdr->src_cid = cpu_to_le64(src_cid);
66 hdr->dst_cid = cpu_to_le64(dst_cid);
67 hdr->src_port = cpu_to_le32(src_port);
68 hdr->dst_port = cpu_to_le32(dst_port);
69 hdr->flags = cpu_to_le32(info->flags);
70 hdr->len = cpu_to_le32(len);
72 if (info->msg && len > 0) {
73 payload = skb_put(skb, len);
74 err = memcpy_from_msg(payload, info->msg, len);
78 if (msg_data_left(info->msg) == 0 &&
79 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
80 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
82 if (info->msg->msg_flags & MSG_EOR)
83 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
88 virtio_vsock_skb_set_reply(skb);
90 trace_virtio_transport_alloc_pkt(src_cid, src_port,
97 if (info->vsk && !skb_set_owner_sk_safe(skb, sk_vsock(info->vsk))) {
98 WARN_ONCE(1, "failed to allocate skb on vsock socket with sk_refcnt == 0\n");
110 static struct sk_buff *virtio_transport_build_skb(void *opaque)
112 struct virtio_vsock_hdr *pkt_hdr;
113 struct sk_buff *pkt = opaque;
114 struct af_vsockmon_hdr *hdr;
119 /* A packet could be split to fit the RX buffer, so we can retrieve
120 * the payload length from the header and the buffer pointer taking
121 * care of the offset in the original packet.
123 pkt_hdr = virtio_vsock_hdr(pkt);
124 payload_len = pkt->len;
125 payload_buf = pkt->data;
127 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
132 hdr = skb_put(skb, sizeof(*hdr));
134 /* pkt->hdr is little-endian so no need to byteswap here */
135 hdr->src_cid = pkt_hdr->src_cid;
136 hdr->src_port = pkt_hdr->src_port;
137 hdr->dst_cid = pkt_hdr->dst_cid;
138 hdr->dst_port = pkt_hdr->dst_port;
140 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
141 hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
142 memset(hdr->reserved, 0, sizeof(hdr->reserved));
144 switch (le16_to_cpu(pkt_hdr->op)) {
145 case VIRTIO_VSOCK_OP_REQUEST:
146 case VIRTIO_VSOCK_OP_RESPONSE:
147 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
149 case VIRTIO_VSOCK_OP_RST:
150 case VIRTIO_VSOCK_OP_SHUTDOWN:
151 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
153 case VIRTIO_VSOCK_OP_RW:
154 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
156 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
157 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
158 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
161 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
165 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
168 skb_put_data(skb, payload_buf, payload_len);
174 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
176 if (virtio_vsock_skb_tap_delivered(skb))
179 vsock_deliver_tap(virtio_transport_build_skb, skb);
180 virtio_vsock_skb_set_tap_delivered(skb);
182 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
184 static u16 virtio_transport_get_type(struct sock *sk)
186 if (sk->sk_type == SOCK_STREAM)
187 return VIRTIO_VSOCK_TYPE_STREAM;
189 return VIRTIO_VSOCK_TYPE_SEQPACKET;
192 /* This function can only be used on connecting/connected sockets,
193 * since a socket assigned to a transport is required.
195 * Do not use on listener sockets!
197 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
198 struct virtio_vsock_pkt_info *info)
200 u32 src_cid, src_port, dst_cid, dst_port;
201 const struct virtio_transport *t_ops;
202 struct virtio_vsock_sock *vvs;
203 u32 pkt_len = info->pkt_len;
207 info->type = virtio_transport_get_type(sk_vsock(vsk));
209 t_ops = virtio_transport_get_ops(vsk);
210 if (unlikely(!t_ops))
213 src_cid = t_ops->transport.get_local_cid();
214 src_port = vsk->local_addr.svm_port;
215 if (!info->remote_cid) {
216 dst_cid = vsk->remote_addr.svm_cid;
217 dst_port = vsk->remote_addr.svm_port;
219 dst_cid = info->remote_cid;
220 dst_port = info->remote_port;
225 /* virtio_transport_get_credit might return less than pkt_len credit */
226 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
228 /* Do not send zero length OP_RW pkt */
229 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
238 skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, rest_len);
240 skb = virtio_transport_alloc_skb(info, skb_len,
248 virtio_transport_inc_tx_pkt(vvs, skb);
250 ret = t_ops->send_pkt(skb);
254 /* Both virtio and vhost 'send_pkt()' returns 'skb_len',
255 * but for reliability use 'ret' instead of 'skb_len'.
256 * Also if partial send happens (e.g. 'ret' != 'skb_len')
257 * somehow, we break this loop, but account such returned
258 * value in 'virtio_transport_put_credit()'.
262 if (WARN_ONCE(ret != skb_len,
263 "'send_pkt()' returns %i, but %zu expected\n",
268 virtio_transport_put_credit(vvs, rest_len);
270 /* Return number of bytes, if any data has been sent. */
271 if (rest_len != pkt_len)
272 ret = pkt_len - rest_len;
277 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
280 if (vvs->rx_bytes + len > vvs->buf_alloc)
283 vvs->rx_bytes += len;
287 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
290 vvs->rx_bytes -= len;
294 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
296 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
298 spin_lock_bh(&vvs->rx_lock);
299 vvs->last_fwd_cnt = vvs->fwd_cnt;
300 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
301 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
302 spin_unlock_bh(&vvs->rx_lock);
304 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
306 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
313 spin_lock_bh(&vvs->tx_lock);
314 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
318 spin_unlock_bh(&vvs->tx_lock);
322 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
324 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
329 spin_lock_bh(&vvs->tx_lock);
330 vvs->tx_cnt -= credit;
331 spin_unlock_bh(&vvs->tx_lock);
333 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
335 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
337 struct virtio_vsock_pkt_info info = {
338 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
342 return virtio_transport_send_pkt_info(vsk, &info);
346 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
350 struct virtio_vsock_sock *vvs = vsk->trans;
355 spin_lock_bh(&vvs->rx_lock);
357 skb_queue_walk(&vvs->rx_queue, skb) {
361 if (bytes > skb->len)
364 spin_unlock_bh(&vvs->rx_lock);
366 /* sk_lock is held by caller so no one else can dequeue.
367 * Unlock rx_lock since memcpy_to_msg() may sleep.
369 err = memcpy_to_msg(msg, skb->data, bytes);
375 spin_lock_bh(&vvs->rx_lock);
381 spin_unlock_bh(&vvs->rx_lock);
392 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
396 struct virtio_vsock_sock *vvs = vsk->trans;
397 size_t bytes, total = 0;
402 spin_lock_bh(&vvs->rx_lock);
404 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
405 "rx_queue is empty, but rx_bytes is non-zero\n")) {
406 spin_unlock_bh(&vvs->rx_lock);
410 while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
411 skb = skb_peek(&vvs->rx_queue);
414 if (bytes > skb->len)
417 /* sk_lock is held by caller so no one else can dequeue.
418 * Unlock rx_lock since memcpy_to_msg() may sleep.
420 spin_unlock_bh(&vvs->rx_lock);
422 err = memcpy_to_msg(msg, skb->data, bytes);
426 spin_lock_bh(&vvs->rx_lock);
429 skb_pull(skb, bytes);
432 u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
434 virtio_transport_dec_rx_pkt(vvs, pkt_len);
435 __skb_unlink(skb, &vvs->rx_queue);
440 free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
442 spin_unlock_bh(&vvs->rx_lock);
444 /* To reduce the number of credit update messages,
445 * don't update credits as long as lots of space is available.
446 * Note: the limit chosen here is arbitrary. Setting the limit
447 * too high causes extra messages. Too low causes transmitter
448 * stalls. As stalls are in theory more expensive than extra
449 * messages, we set the limit to a high value. TODO: experiment
450 * with different values.
452 if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
453 virtio_transport_send_credit_update(vsk);
464 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
467 struct virtio_vsock_sock *vvs = vsk->trans;
471 spin_lock_bh(&vvs->rx_lock);
473 if (!vvs->msg_count) {
474 spin_unlock_bh(&vvs->rx_lock);
479 len = msg_data_left(msg);
481 skb_queue_walk(&vvs->rx_queue, skb) {
482 struct virtio_vsock_hdr *hdr;
489 if (bytes > skb->len)
492 spin_unlock_bh(&vvs->rx_lock);
494 /* sk_lock is held by caller so no one else can dequeue.
495 * Unlock rx_lock since memcpy_to_msg() may sleep.
497 err = memcpy_to_msg(msg, skb->data, bytes);
501 spin_lock_bh(&vvs->rx_lock);
505 hdr = virtio_vsock_hdr(skb);
507 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
508 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
509 msg->msg_flags |= MSG_EOR;
515 spin_unlock_bh(&vvs->rx_lock);
520 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
524 struct virtio_vsock_sock *vvs = vsk->trans;
525 int dequeued_len = 0;
526 size_t user_buf_len = msg_data_left(msg);
527 bool msg_ready = false;
530 spin_lock_bh(&vvs->rx_lock);
532 if (vvs->msg_count == 0) {
533 spin_unlock_bh(&vvs->rx_lock);
538 struct virtio_vsock_hdr *hdr;
541 skb = __skb_dequeue(&vvs->rx_queue);
544 hdr = virtio_vsock_hdr(skb);
545 pkt_len = (size_t)le32_to_cpu(hdr->len);
547 if (dequeued_len >= 0) {
548 size_t bytes_to_copy;
550 bytes_to_copy = min(user_buf_len, pkt_len);
555 /* sk_lock is held by caller so no one else can dequeue.
556 * Unlock rx_lock since memcpy_to_msg() may sleep.
558 spin_unlock_bh(&vvs->rx_lock);
560 err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
562 /* Copy of message failed. Rest of
563 * fragments will be freed without copy.
567 user_buf_len -= bytes_to_copy;
570 spin_lock_bh(&vvs->rx_lock);
573 if (dequeued_len >= 0)
574 dequeued_len += pkt_len;
577 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
581 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
582 msg->msg_flags |= MSG_EOR;
585 virtio_transport_dec_rx_pkt(vvs, pkt_len);
589 spin_unlock_bh(&vvs->rx_lock);
591 virtio_transport_send_credit_update(vsk);
597 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
599 size_t len, int flags)
601 if (flags & MSG_PEEK)
602 return virtio_transport_stream_do_peek(vsk, msg, len);
604 return virtio_transport_stream_do_dequeue(vsk, msg, len);
606 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
609 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
613 if (flags & MSG_PEEK)
614 return virtio_transport_seqpacket_do_peek(vsk, msg);
616 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
618 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
621 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
625 struct virtio_vsock_sock *vvs = vsk->trans;
627 spin_lock_bh(&vvs->tx_lock);
629 if (len > vvs->peer_buf_alloc) {
630 spin_unlock_bh(&vvs->tx_lock);
634 spin_unlock_bh(&vvs->tx_lock);
636 return virtio_transport_stream_enqueue(vsk, msg, len);
638 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
641 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
643 size_t len, int flags)
647 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
649 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
651 struct virtio_vsock_sock *vvs = vsk->trans;
654 spin_lock_bh(&vvs->rx_lock);
655 bytes = vvs->rx_bytes;
656 spin_unlock_bh(&vvs->rx_lock);
660 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
662 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
664 struct virtio_vsock_sock *vvs = vsk->trans;
667 spin_lock_bh(&vvs->rx_lock);
668 msg_count = vvs->msg_count;
669 spin_unlock_bh(&vvs->rx_lock);
673 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
675 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
677 struct virtio_vsock_sock *vvs = vsk->trans;
680 bytes = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
687 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
689 struct virtio_vsock_sock *vvs = vsk->trans;
692 spin_lock_bh(&vvs->tx_lock);
693 bytes = virtio_transport_has_space(vsk);
694 spin_unlock_bh(&vvs->tx_lock);
698 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
700 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
701 struct vsock_sock *psk)
703 struct virtio_vsock_sock *vvs;
705 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
711 if (psk && psk->trans) {
712 struct virtio_vsock_sock *ptrans = psk->trans;
714 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
717 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
718 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
720 vvs->buf_alloc = vsk->buffer_size;
722 spin_lock_init(&vvs->rx_lock);
723 spin_lock_init(&vvs->tx_lock);
724 skb_queue_head_init(&vvs->rx_queue);
728 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
730 /* sk_lock held by the caller */
731 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
733 struct virtio_vsock_sock *vvs = vsk->trans;
735 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
736 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
738 vvs->buf_alloc = *val;
740 virtio_transport_send_credit_update(vsk);
742 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
745 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
747 bool *data_ready_now)
749 *data_ready_now = vsock_stream_has_data(vsk) >= target;
753 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
756 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
758 bool *space_avail_now)
762 free_space = vsock_stream_has_space(vsk);
764 *space_avail_now = true;
765 else if (free_space == 0)
766 *space_avail_now = false;
770 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
772 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
773 size_t target, struct vsock_transport_recv_notify_data *data)
777 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
779 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
780 size_t target, struct vsock_transport_recv_notify_data *data)
784 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
786 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
787 size_t target, struct vsock_transport_recv_notify_data *data)
791 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
793 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
794 size_t target, ssize_t copied, bool data_read,
795 struct vsock_transport_recv_notify_data *data)
799 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
801 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
802 struct vsock_transport_send_notify_data *data)
806 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
808 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
809 struct vsock_transport_send_notify_data *data)
813 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
815 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
816 struct vsock_transport_send_notify_data *data)
820 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
822 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
823 ssize_t written, struct vsock_transport_send_notify_data *data)
827 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
829 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
831 return vsk->buffer_size;
833 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
835 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
839 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
841 bool virtio_transport_stream_allow(u32 cid, u32 port)
845 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
847 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
848 struct sockaddr_vm *addr)
852 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
854 bool virtio_transport_dgram_allow(u32 cid, u32 port)
858 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
860 int virtio_transport_connect(struct vsock_sock *vsk)
862 struct virtio_vsock_pkt_info info = {
863 .op = VIRTIO_VSOCK_OP_REQUEST,
867 return virtio_transport_send_pkt_info(vsk, &info);
869 EXPORT_SYMBOL_GPL(virtio_transport_connect);
871 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
873 struct virtio_vsock_pkt_info info = {
874 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
875 .flags = (mode & RCV_SHUTDOWN ?
876 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
877 (mode & SEND_SHUTDOWN ?
878 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
882 return virtio_transport_send_pkt_info(vsk, &info);
884 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
887 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
888 struct sockaddr_vm *remote_addr,
894 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
897 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
901 struct virtio_vsock_pkt_info info = {
902 .op = VIRTIO_VSOCK_OP_RW,
908 return virtio_transport_send_pkt_info(vsk, &info);
910 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
912 void virtio_transport_destruct(struct vsock_sock *vsk)
914 struct virtio_vsock_sock *vvs = vsk->trans;
918 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
920 static int virtio_transport_reset(struct vsock_sock *vsk,
923 struct virtio_vsock_pkt_info info = {
924 .op = VIRTIO_VSOCK_OP_RST,
929 /* Send RST only if the original pkt is not a RST pkt */
930 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
933 return virtio_transport_send_pkt_info(vsk, &info);
936 /* Normally packets are associated with a socket. There may be no socket if an
937 * attempt was made to connect to a socket that does not exist.
939 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
942 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
943 struct virtio_vsock_pkt_info info = {
944 .op = VIRTIO_VSOCK_OP_RST,
945 .type = le16_to_cpu(hdr->type),
948 struct sk_buff *reply;
950 /* Send RST only if the original pkt is not a RST pkt */
951 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
957 reply = virtio_transport_alloc_skb(&info, 0,
958 le64_to_cpu(hdr->dst_cid),
959 le32_to_cpu(hdr->dst_port),
960 le64_to_cpu(hdr->src_cid),
961 le32_to_cpu(hdr->src_port));
965 return t->send_pkt(reply);
968 /* This function should be called with sk_lock held and SOCK_DONE set */
969 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
971 struct virtio_vsock_sock *vvs = vsk->trans;
973 /* We don't need to take rx_lock, as the socket is closing and we are
976 __skb_queue_purge(&vvs->rx_queue);
977 vsock_remove_sock(vsk);
980 static void virtio_transport_wait_close(struct sock *sk, long timeout)
983 DEFINE_WAIT_FUNC(wait, woken_wake_function);
985 add_wait_queue(sk_sleep(sk), &wait);
988 if (sk_wait_event(sk, &timeout,
989 sock_flag(sk, SOCK_DONE), &wait))
991 } while (!signal_pending(current) && timeout);
993 remove_wait_queue(sk_sleep(sk), &wait);
997 static void virtio_transport_do_close(struct vsock_sock *vsk,
1000 struct sock *sk = sk_vsock(vsk);
1002 sock_set_flag(sk, SOCK_DONE);
1003 vsk->peer_shutdown = SHUTDOWN_MASK;
1004 if (vsock_stream_has_data(vsk) <= 0)
1005 sk->sk_state = TCP_CLOSING;
1006 sk->sk_state_change(sk);
1008 if (vsk->close_work_scheduled &&
1009 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1010 vsk->close_work_scheduled = false;
1012 virtio_transport_remove_sock(vsk);
1014 /* Release refcnt obtained when we scheduled the timeout */
1019 static void virtio_transport_close_timeout(struct work_struct *work)
1021 struct vsock_sock *vsk =
1022 container_of(work, struct vsock_sock, close_work.work);
1023 struct sock *sk = sk_vsock(vsk);
1028 if (!sock_flag(sk, SOCK_DONE)) {
1029 (void)virtio_transport_reset(vsk, NULL);
1031 virtio_transport_do_close(vsk, false);
1034 vsk->close_work_scheduled = false;
1040 /* User context, vsk->sk is locked */
1041 static bool virtio_transport_close(struct vsock_sock *vsk)
1043 struct sock *sk = &vsk->sk;
1045 if (!(sk->sk_state == TCP_ESTABLISHED ||
1046 sk->sk_state == TCP_CLOSING))
1049 /* Already received SHUTDOWN from peer, reply with RST */
1050 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1051 (void)virtio_transport_reset(vsk, NULL);
1055 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1056 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1058 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1059 virtio_transport_wait_close(sk, sk->sk_lingertime);
1061 if (sock_flag(sk, SOCK_DONE)) {
1066 INIT_DELAYED_WORK(&vsk->close_work,
1067 virtio_transport_close_timeout);
1068 vsk->close_work_scheduled = true;
1069 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1073 void virtio_transport_release(struct vsock_sock *vsk)
1075 struct sock *sk = &vsk->sk;
1076 bool remove_sock = true;
1078 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1079 remove_sock = virtio_transport_close(vsk);
1082 sock_set_flag(sk, SOCK_DONE);
1083 virtio_transport_remove_sock(vsk);
1086 EXPORT_SYMBOL_GPL(virtio_transport_release);
1089 virtio_transport_recv_connecting(struct sock *sk,
1090 struct sk_buff *skb)
1092 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1093 struct vsock_sock *vsk = vsock_sk(sk);
1097 switch (le16_to_cpu(hdr->op)) {
1098 case VIRTIO_VSOCK_OP_RESPONSE:
1099 sk->sk_state = TCP_ESTABLISHED;
1100 sk->sk_socket->state = SS_CONNECTED;
1101 vsock_insert_connected(vsk);
1102 sk->sk_state_change(sk);
1104 case VIRTIO_VSOCK_OP_INVALID:
1106 case VIRTIO_VSOCK_OP_RST:
1118 virtio_transport_reset(vsk, skb);
1119 sk->sk_state = TCP_CLOSE;
1121 sk_error_report(sk);
1126 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1127 struct sk_buff *skb)
1129 struct virtio_vsock_sock *vvs = vsk->trans;
1130 bool can_enqueue, free_pkt = false;
1131 struct virtio_vsock_hdr *hdr;
1134 hdr = virtio_vsock_hdr(skb);
1135 len = le32_to_cpu(hdr->len);
1137 spin_lock_bh(&vvs->rx_lock);
1139 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1145 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1148 /* Try to copy small packets into the buffer of last packet queued,
1149 * to avoid wasting memory queueing the entire buffer with a small
1152 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1153 struct virtio_vsock_hdr *last_hdr;
1154 struct sk_buff *last_skb;
1156 last_skb = skb_peek_tail(&vvs->rx_queue);
1157 last_hdr = virtio_vsock_hdr(last_skb);
1159 /* If there is space in the last packet queued, we copy the
1160 * new packet in its buffer. We avoid this if the last packet
1161 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1162 * delimiter of SEQPACKET message, so 'pkt' is the first packet
1165 if (skb->len < skb_tailroom(last_skb) &&
1166 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1167 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1169 last_hdr->flags |= hdr->flags;
1170 le32_add_cpu(&last_hdr->len, len);
1175 __skb_queue_tail(&vvs->rx_queue, skb);
1178 spin_unlock_bh(&vvs->rx_lock);
1184 virtio_transport_recv_connected(struct sock *sk,
1185 struct sk_buff *skb)
1187 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1188 struct vsock_sock *vsk = vsock_sk(sk);
1191 switch (le16_to_cpu(hdr->op)) {
1192 case VIRTIO_VSOCK_OP_RW:
1193 virtio_transport_recv_enqueue(vsk, skb);
1194 vsock_data_ready(sk);
1196 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1197 virtio_transport_send_credit_update(vsk);
1199 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1200 sk->sk_write_space(sk);
1202 case VIRTIO_VSOCK_OP_SHUTDOWN:
1203 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1204 vsk->peer_shutdown |= RCV_SHUTDOWN;
1205 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1206 vsk->peer_shutdown |= SEND_SHUTDOWN;
1207 if (vsk->peer_shutdown == SHUTDOWN_MASK &&
1208 vsock_stream_has_data(vsk) <= 0 &&
1209 !sock_flag(sk, SOCK_DONE)) {
1210 (void)virtio_transport_reset(vsk, NULL);
1211 virtio_transport_do_close(vsk, true);
1213 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1214 sk->sk_state_change(sk);
1216 case VIRTIO_VSOCK_OP_RST:
1217 virtio_transport_do_close(vsk, true);
1229 virtio_transport_recv_disconnecting(struct sock *sk,
1230 struct sk_buff *skb)
1232 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1233 struct vsock_sock *vsk = vsock_sk(sk);
1235 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1236 virtio_transport_do_close(vsk, true);
1240 virtio_transport_send_response(struct vsock_sock *vsk,
1241 struct sk_buff *skb)
1243 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1244 struct virtio_vsock_pkt_info info = {
1245 .op = VIRTIO_VSOCK_OP_RESPONSE,
1246 .remote_cid = le64_to_cpu(hdr->src_cid),
1247 .remote_port = le32_to_cpu(hdr->src_port),
1252 return virtio_transport_send_pkt_info(vsk, &info);
1255 static bool virtio_transport_space_update(struct sock *sk,
1256 struct sk_buff *skb)
1258 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1259 struct vsock_sock *vsk = vsock_sk(sk);
1260 struct virtio_vsock_sock *vvs = vsk->trans;
1261 bool space_available;
1263 /* Listener sockets are not associated with any transport, so we are
1264 * not able to take the state to see if there is space available in the
1265 * remote peer, but since they are only used to receive requests, we
1266 * can assume that there is always space available in the other peer.
1271 /* buf_alloc and fwd_cnt is always included in the hdr */
1272 spin_lock_bh(&vvs->tx_lock);
1273 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1274 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1275 space_available = virtio_transport_has_space(vsk);
1276 spin_unlock_bh(&vvs->tx_lock);
1277 return space_available;
1280 /* Handle server socket */
1282 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1283 struct virtio_transport *t)
1285 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1286 struct vsock_sock *vsk = vsock_sk(sk);
1287 struct vsock_sock *vchild;
1291 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1292 virtio_transport_reset_no_sock(t, skb);
1296 if (sk_acceptq_is_full(sk)) {
1297 virtio_transport_reset_no_sock(t, skb);
1301 child = vsock_create_connected(sk);
1303 virtio_transport_reset_no_sock(t, skb);
1307 sk_acceptq_added(sk);
1309 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1311 child->sk_state = TCP_ESTABLISHED;
1313 vchild = vsock_sk(child);
1314 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1315 le32_to_cpu(hdr->dst_port));
1316 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1317 le32_to_cpu(hdr->src_port));
1319 ret = vsock_assign_transport(vchild, vsk);
1320 /* Transport assigned (looking at remote_addr) must be the same
1321 * where we received the request.
1323 if (ret || vchild->transport != &t->transport) {
1324 release_sock(child);
1325 virtio_transport_reset_no_sock(t, skb);
1330 if (virtio_transport_space_update(child, skb))
1331 child->sk_write_space(child);
1333 vsock_insert_connected(vchild);
1334 vsock_enqueue_accept(sk, child);
1335 virtio_transport_send_response(vchild, skb);
1337 release_sock(child);
1339 sk->sk_data_ready(sk);
1343 static bool virtio_transport_valid_type(u16 type)
1345 return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1346 (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1349 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1352 void virtio_transport_recv_pkt(struct virtio_transport *t,
1353 struct sk_buff *skb)
1355 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1356 struct sockaddr_vm src, dst;
1357 struct vsock_sock *vsk;
1359 bool space_available;
1361 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1362 le32_to_cpu(hdr->src_port));
1363 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1364 le32_to_cpu(hdr->dst_port));
1366 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1367 dst.svm_cid, dst.svm_port,
1368 le32_to_cpu(hdr->len),
1369 le16_to_cpu(hdr->type),
1370 le16_to_cpu(hdr->op),
1371 le32_to_cpu(hdr->flags),
1372 le32_to_cpu(hdr->buf_alloc),
1373 le32_to_cpu(hdr->fwd_cnt));
1375 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1376 (void)virtio_transport_reset_no_sock(t, skb);
1380 /* The socket must be in connected or bound table
1381 * otherwise send reset back
1383 sk = vsock_find_connected_socket(&src, &dst);
1385 sk = vsock_find_bound_socket(&dst);
1387 (void)virtio_transport_reset_no_sock(t, skb);
1392 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1393 (void)virtio_transport_reset_no_sock(t, skb);
1398 if (!skb_set_owner_sk_safe(skb, sk)) {
1399 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1407 /* Check if sk has been closed before lock_sock */
1408 if (sock_flag(sk, SOCK_DONE)) {
1409 (void)virtio_transport_reset_no_sock(t, skb);
1415 space_available = virtio_transport_space_update(sk, skb);
1417 /* Update CID in case it has changed after a transport reset event */
1418 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1419 vsk->local_addr.svm_cid = dst.svm_cid;
1421 if (space_available)
1422 sk->sk_write_space(sk);
1424 switch (sk->sk_state) {
1426 virtio_transport_recv_listen(sk, skb, t);
1430 virtio_transport_recv_connecting(sk, skb);
1433 case TCP_ESTABLISHED:
1434 virtio_transport_recv_connected(sk, skb);
1437 virtio_transport_recv_disconnecting(sk, skb);
1441 (void)virtio_transport_reset_no_sock(t, skb);
1448 /* Release refcnt obtained when we fetched this socket out of the
1449 * bound or connected list.
1457 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1459 /* Remove skbs found in a queue that have a vsk that matches.
1461 * Each skb is freed.
1463 * Returns the count of skbs that were reply packets.
1465 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1467 struct sk_buff_head freeme;
1468 struct sk_buff *skb, *tmp;
1471 skb_queue_head_init(&freeme);
1473 spin_lock_bh(&queue->lock);
1474 skb_queue_walk_safe(queue, skb, tmp) {
1475 if (vsock_sk(skb->sk) != vsk)
1478 __skb_unlink(skb, queue);
1479 __skb_queue_tail(&freeme, skb);
1481 if (virtio_vsock_skb_reply(skb))
1484 spin_unlock_bh(&queue->lock);
1486 __skb_queue_purge(&freeme);
1490 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1492 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1494 struct virtio_vsock_sock *vvs = vsk->trans;
1495 struct sock *sk = sk_vsock(vsk);
1496 struct sk_buff *skb;
1500 spin_lock_bh(&vvs->rx_lock);
1501 /* Use __skb_recv_datagram() for race-free handling of the receive. It
1502 * works for types other than dgrams.
1504 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1505 spin_unlock_bh(&vvs->rx_lock);
1510 return recv_actor(sk, skb);
1512 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1514 MODULE_LICENSE("GPL v2");
1515 MODULE_AUTHOR("Asias He");
1516 MODULE_DESCRIPTION("common code for virtio vsock");