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);
71 hdr->buf_alloc = cpu_to_le32(0);
72 hdr->fwd_cnt = cpu_to_le32(0);
74 if (info->msg && len > 0) {
75 payload = skb_put(skb, len);
76 err = memcpy_from_msg(payload, info->msg, len);
80 if (msg_data_left(info->msg) == 0 &&
81 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
82 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
84 if (info->msg->msg_flags & MSG_EOR)
85 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
90 virtio_vsock_skb_set_reply(skb);
92 trace_virtio_transport_alloc_pkt(src_cid, src_port,
99 if (info->vsk && !skb_set_owner_sk_safe(skb, sk_vsock(info->vsk))) {
100 WARN_ONCE(1, "failed to allocate skb on vsock socket with sk_refcnt == 0\n");
112 static struct sk_buff *virtio_transport_build_skb(void *opaque)
114 struct virtio_vsock_hdr *pkt_hdr;
115 struct sk_buff *pkt = opaque;
116 struct af_vsockmon_hdr *hdr;
121 /* A packet could be split to fit the RX buffer, so we can retrieve
122 * the payload length from the header and the buffer pointer taking
123 * care of the offset in the original packet.
125 pkt_hdr = virtio_vsock_hdr(pkt);
126 payload_len = pkt->len;
127 payload_buf = pkt->data;
129 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
134 hdr = skb_put(skb, sizeof(*hdr));
136 /* pkt->hdr is little-endian so no need to byteswap here */
137 hdr->src_cid = pkt_hdr->src_cid;
138 hdr->src_port = pkt_hdr->src_port;
139 hdr->dst_cid = pkt_hdr->dst_cid;
140 hdr->dst_port = pkt_hdr->dst_port;
142 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
143 hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
144 memset(hdr->reserved, 0, sizeof(hdr->reserved));
146 switch (le16_to_cpu(pkt_hdr->op)) {
147 case VIRTIO_VSOCK_OP_REQUEST:
148 case VIRTIO_VSOCK_OP_RESPONSE:
149 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
151 case VIRTIO_VSOCK_OP_RST:
152 case VIRTIO_VSOCK_OP_SHUTDOWN:
153 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
155 case VIRTIO_VSOCK_OP_RW:
156 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
158 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
159 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
160 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
163 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
167 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
170 skb_put_data(skb, payload_buf, payload_len);
176 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
178 if (virtio_vsock_skb_tap_delivered(skb))
181 vsock_deliver_tap(virtio_transport_build_skb, skb);
182 virtio_vsock_skb_set_tap_delivered(skb);
184 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
186 static u16 virtio_transport_get_type(struct sock *sk)
188 if (sk->sk_type == SOCK_STREAM)
189 return VIRTIO_VSOCK_TYPE_STREAM;
191 return VIRTIO_VSOCK_TYPE_SEQPACKET;
194 /* This function can only be used on connecting/connected sockets,
195 * since a socket assigned to a transport is required.
197 * Do not use on listener sockets!
199 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
200 struct virtio_vsock_pkt_info *info)
202 u32 src_cid, src_port, dst_cid, dst_port;
203 const struct virtio_transport *t_ops;
204 struct virtio_vsock_sock *vvs;
205 u32 pkt_len = info->pkt_len;
209 info->type = virtio_transport_get_type(sk_vsock(vsk));
211 t_ops = virtio_transport_get_ops(vsk);
212 if (unlikely(!t_ops))
215 src_cid = t_ops->transport.get_local_cid();
216 src_port = vsk->local_addr.svm_port;
217 if (!info->remote_cid) {
218 dst_cid = vsk->remote_addr.svm_cid;
219 dst_port = vsk->remote_addr.svm_port;
221 dst_cid = info->remote_cid;
222 dst_port = info->remote_port;
227 /* virtio_transport_get_credit might return less than pkt_len credit */
228 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
230 /* Do not send zero length OP_RW pkt */
231 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
240 skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, rest_len);
242 skb = virtio_transport_alloc_skb(info, skb_len,
250 virtio_transport_inc_tx_pkt(vvs, skb);
252 ret = t_ops->send_pkt(skb);
256 /* Both virtio and vhost 'send_pkt()' returns 'skb_len',
257 * but for reliability use 'ret' instead of 'skb_len'.
258 * Also if partial send happens (e.g. 'ret' != 'skb_len')
259 * somehow, we break this loop, but account such returned
260 * value in 'virtio_transport_put_credit()'.
264 if (WARN_ONCE(ret != skb_len,
265 "'send_pkt()' returns %i, but %zu expected\n",
270 virtio_transport_put_credit(vvs, rest_len);
272 /* Return number of bytes, if any data has been sent. */
273 if (rest_len != pkt_len)
274 ret = pkt_len - rest_len;
279 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
282 if (vvs->rx_bytes + len > vvs->buf_alloc)
285 vvs->rx_bytes += len;
289 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
292 vvs->rx_bytes -= len;
296 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
298 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
300 spin_lock_bh(&vvs->rx_lock);
301 vvs->last_fwd_cnt = vvs->fwd_cnt;
302 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
303 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
304 spin_unlock_bh(&vvs->rx_lock);
306 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
308 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
315 spin_lock_bh(&vvs->tx_lock);
316 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
320 spin_unlock_bh(&vvs->tx_lock);
324 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
326 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
331 spin_lock_bh(&vvs->tx_lock);
332 vvs->tx_cnt -= credit;
333 spin_unlock_bh(&vvs->tx_lock);
335 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
337 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
339 struct virtio_vsock_pkt_info info = {
340 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
344 return virtio_transport_send_pkt_info(vsk, &info);
348 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
352 struct virtio_vsock_sock *vvs = vsk->trans;
357 spin_lock_bh(&vvs->rx_lock);
359 skb_queue_walk(&vvs->rx_queue, skb) {
363 if (bytes > skb->len)
366 spin_unlock_bh(&vvs->rx_lock);
368 /* sk_lock is held by caller so no one else can dequeue.
369 * Unlock rx_lock since memcpy_to_msg() may sleep.
371 err = memcpy_to_msg(msg, skb->data, bytes);
377 spin_lock_bh(&vvs->rx_lock);
383 spin_unlock_bh(&vvs->rx_lock);
394 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
398 struct virtio_vsock_sock *vvs = vsk->trans;
399 size_t bytes, total = 0;
406 spin_lock_bh(&vvs->rx_lock);
408 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
409 "rx_queue is empty, but rx_bytes is non-zero\n")) {
410 spin_unlock_bh(&vvs->rx_lock);
414 while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
415 skb = skb_peek(&vvs->rx_queue);
418 if (bytes > skb->len)
421 /* sk_lock is held by caller so no one else can dequeue.
422 * Unlock rx_lock since memcpy_to_msg() may sleep.
424 spin_unlock_bh(&vvs->rx_lock);
426 err = memcpy_to_msg(msg, skb->data, bytes);
430 spin_lock_bh(&vvs->rx_lock);
433 skb_pull(skb, bytes);
436 u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
438 virtio_transport_dec_rx_pkt(vvs, pkt_len);
439 __skb_unlink(skb, &vvs->rx_queue);
444 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt;
445 free_space = vvs->buf_alloc - fwd_cnt_delta;
446 low_rx_bytes = (vvs->rx_bytes <
447 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX));
449 spin_unlock_bh(&vvs->rx_lock);
451 /* To reduce the number of credit update messages,
452 * don't update credits as long as lots of space is available.
453 * Note: the limit chosen here is arbitrary. Setting the limit
454 * too high causes extra messages. Too low causes transmitter
455 * stalls. As stalls are in theory more expensive than extra
456 * messages, we set the limit to a high value. TODO: experiment
457 * with different values. Also send credit update message when
458 * number of bytes in rx queue is not enough to wake up reader.
461 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes))
462 virtio_transport_send_credit_update(vsk);
473 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
476 struct virtio_vsock_sock *vvs = vsk->trans;
480 spin_lock_bh(&vvs->rx_lock);
482 if (!vvs->msg_count) {
483 spin_unlock_bh(&vvs->rx_lock);
488 len = msg_data_left(msg);
490 skb_queue_walk(&vvs->rx_queue, skb) {
491 struct virtio_vsock_hdr *hdr;
498 if (bytes > skb->len)
501 spin_unlock_bh(&vvs->rx_lock);
503 /* sk_lock is held by caller so no one else can dequeue.
504 * Unlock rx_lock since memcpy_to_msg() may sleep.
506 err = memcpy_to_msg(msg, skb->data, bytes);
510 spin_lock_bh(&vvs->rx_lock);
514 hdr = virtio_vsock_hdr(skb);
516 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
517 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
518 msg->msg_flags |= MSG_EOR;
524 spin_unlock_bh(&vvs->rx_lock);
529 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
533 struct virtio_vsock_sock *vvs = vsk->trans;
534 int dequeued_len = 0;
535 size_t user_buf_len = msg_data_left(msg);
536 bool msg_ready = false;
539 spin_lock_bh(&vvs->rx_lock);
541 if (vvs->msg_count == 0) {
542 spin_unlock_bh(&vvs->rx_lock);
547 struct virtio_vsock_hdr *hdr;
550 skb = __skb_dequeue(&vvs->rx_queue);
553 hdr = virtio_vsock_hdr(skb);
554 pkt_len = (size_t)le32_to_cpu(hdr->len);
556 if (dequeued_len >= 0) {
557 size_t bytes_to_copy;
559 bytes_to_copy = min(user_buf_len, pkt_len);
564 /* sk_lock is held by caller so no one else can dequeue.
565 * Unlock rx_lock since memcpy_to_msg() may sleep.
567 spin_unlock_bh(&vvs->rx_lock);
569 err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
571 /* Copy of message failed. Rest of
572 * fragments will be freed without copy.
576 user_buf_len -= bytes_to_copy;
579 spin_lock_bh(&vvs->rx_lock);
582 if (dequeued_len >= 0)
583 dequeued_len += pkt_len;
586 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
590 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
591 msg->msg_flags |= MSG_EOR;
594 virtio_transport_dec_rx_pkt(vvs, pkt_len);
598 spin_unlock_bh(&vvs->rx_lock);
600 virtio_transport_send_credit_update(vsk);
606 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
608 size_t len, int flags)
610 if (flags & MSG_PEEK)
611 return virtio_transport_stream_do_peek(vsk, msg, len);
613 return virtio_transport_stream_do_dequeue(vsk, msg, len);
615 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
618 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
622 if (flags & MSG_PEEK)
623 return virtio_transport_seqpacket_do_peek(vsk, msg);
625 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
627 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
630 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
634 struct virtio_vsock_sock *vvs = vsk->trans;
636 spin_lock_bh(&vvs->tx_lock);
638 if (len > vvs->peer_buf_alloc) {
639 spin_unlock_bh(&vvs->tx_lock);
643 spin_unlock_bh(&vvs->tx_lock);
645 return virtio_transport_stream_enqueue(vsk, msg, len);
647 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
650 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
652 size_t len, int flags)
656 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
658 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
660 struct virtio_vsock_sock *vvs = vsk->trans;
663 spin_lock_bh(&vvs->rx_lock);
664 bytes = vvs->rx_bytes;
665 spin_unlock_bh(&vvs->rx_lock);
669 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
671 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
673 struct virtio_vsock_sock *vvs = vsk->trans;
676 spin_lock_bh(&vvs->rx_lock);
677 msg_count = vvs->msg_count;
678 spin_unlock_bh(&vvs->rx_lock);
682 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
684 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
686 struct virtio_vsock_sock *vvs = vsk->trans;
689 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
696 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
698 struct virtio_vsock_sock *vvs = vsk->trans;
701 spin_lock_bh(&vvs->tx_lock);
702 bytes = virtio_transport_has_space(vsk);
703 spin_unlock_bh(&vvs->tx_lock);
707 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
709 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
710 struct vsock_sock *psk)
712 struct virtio_vsock_sock *vvs;
714 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
720 if (psk && psk->trans) {
721 struct virtio_vsock_sock *ptrans = psk->trans;
723 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
726 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
727 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
729 vvs->buf_alloc = vsk->buffer_size;
731 spin_lock_init(&vvs->rx_lock);
732 spin_lock_init(&vvs->tx_lock);
733 skb_queue_head_init(&vvs->rx_queue);
737 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
739 /* sk_lock held by the caller */
740 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
742 struct virtio_vsock_sock *vvs = vsk->trans;
744 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
745 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
747 vvs->buf_alloc = *val;
749 virtio_transport_send_credit_update(vsk);
751 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
754 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
756 bool *data_ready_now)
758 *data_ready_now = vsock_stream_has_data(vsk) >= target;
762 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
765 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
767 bool *space_avail_now)
771 free_space = vsock_stream_has_space(vsk);
773 *space_avail_now = true;
774 else if (free_space == 0)
775 *space_avail_now = false;
779 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
781 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
782 size_t target, struct vsock_transport_recv_notify_data *data)
786 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
788 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
789 size_t target, struct vsock_transport_recv_notify_data *data)
793 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
795 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
796 size_t target, struct vsock_transport_recv_notify_data *data)
800 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
802 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
803 size_t target, ssize_t copied, bool data_read,
804 struct vsock_transport_recv_notify_data *data)
808 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
810 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
811 struct vsock_transport_send_notify_data *data)
815 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
817 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
818 struct vsock_transport_send_notify_data *data)
822 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
824 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
825 struct vsock_transport_send_notify_data *data)
829 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
831 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
832 ssize_t written, struct vsock_transport_send_notify_data *data)
836 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
838 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
840 return vsk->buffer_size;
842 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
844 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
848 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
850 bool virtio_transport_stream_allow(u32 cid, u32 port)
854 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
856 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
857 struct sockaddr_vm *addr)
861 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
863 bool virtio_transport_dgram_allow(u32 cid, u32 port)
867 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
869 int virtio_transport_connect(struct vsock_sock *vsk)
871 struct virtio_vsock_pkt_info info = {
872 .op = VIRTIO_VSOCK_OP_REQUEST,
876 return virtio_transport_send_pkt_info(vsk, &info);
878 EXPORT_SYMBOL_GPL(virtio_transport_connect);
880 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
882 struct virtio_vsock_pkt_info info = {
883 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
884 .flags = (mode & RCV_SHUTDOWN ?
885 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
886 (mode & SEND_SHUTDOWN ?
887 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
891 return virtio_transport_send_pkt_info(vsk, &info);
893 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
896 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
897 struct sockaddr_vm *remote_addr,
903 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
906 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
910 struct virtio_vsock_pkt_info info = {
911 .op = VIRTIO_VSOCK_OP_RW,
917 return virtio_transport_send_pkt_info(vsk, &info);
919 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
921 void virtio_transport_destruct(struct vsock_sock *vsk)
923 struct virtio_vsock_sock *vvs = vsk->trans;
927 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
929 static int virtio_transport_reset(struct vsock_sock *vsk,
932 struct virtio_vsock_pkt_info info = {
933 .op = VIRTIO_VSOCK_OP_RST,
938 /* Send RST only if the original pkt is not a RST pkt */
939 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
942 return virtio_transport_send_pkt_info(vsk, &info);
945 /* Normally packets are associated with a socket. There may be no socket if an
946 * attempt was made to connect to a socket that does not exist.
948 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
951 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
952 struct virtio_vsock_pkt_info info = {
953 .op = VIRTIO_VSOCK_OP_RST,
954 .type = le16_to_cpu(hdr->type),
957 struct sk_buff *reply;
959 /* Send RST only if the original pkt is not a RST pkt */
960 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
966 reply = virtio_transport_alloc_skb(&info, 0,
967 le64_to_cpu(hdr->dst_cid),
968 le32_to_cpu(hdr->dst_port),
969 le64_to_cpu(hdr->src_cid),
970 le32_to_cpu(hdr->src_port));
974 return t->send_pkt(reply);
977 /* This function should be called with sk_lock held and SOCK_DONE set */
978 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
980 struct virtio_vsock_sock *vvs = vsk->trans;
982 /* We don't need to take rx_lock, as the socket is closing and we are
985 __skb_queue_purge(&vvs->rx_queue);
986 vsock_remove_sock(vsk);
989 static void virtio_transport_wait_close(struct sock *sk, long timeout)
992 DEFINE_WAIT_FUNC(wait, woken_wake_function);
994 add_wait_queue(sk_sleep(sk), &wait);
997 if (sk_wait_event(sk, &timeout,
998 sock_flag(sk, SOCK_DONE), &wait))
1000 } while (!signal_pending(current) && timeout);
1002 remove_wait_queue(sk_sleep(sk), &wait);
1006 static void virtio_transport_do_close(struct vsock_sock *vsk,
1007 bool cancel_timeout)
1009 struct sock *sk = sk_vsock(vsk);
1011 sock_set_flag(sk, SOCK_DONE);
1012 vsk->peer_shutdown = SHUTDOWN_MASK;
1013 if (vsock_stream_has_data(vsk) <= 0)
1014 sk->sk_state = TCP_CLOSING;
1015 sk->sk_state_change(sk);
1017 if (vsk->close_work_scheduled &&
1018 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1019 vsk->close_work_scheduled = false;
1021 virtio_transport_remove_sock(vsk);
1023 /* Release refcnt obtained when we scheduled the timeout */
1028 static void virtio_transport_close_timeout(struct work_struct *work)
1030 struct vsock_sock *vsk =
1031 container_of(work, struct vsock_sock, close_work.work);
1032 struct sock *sk = sk_vsock(vsk);
1037 if (!sock_flag(sk, SOCK_DONE)) {
1038 (void)virtio_transport_reset(vsk, NULL);
1040 virtio_transport_do_close(vsk, false);
1043 vsk->close_work_scheduled = false;
1049 /* User context, vsk->sk is locked */
1050 static bool virtio_transport_close(struct vsock_sock *vsk)
1052 struct sock *sk = &vsk->sk;
1054 if (!(sk->sk_state == TCP_ESTABLISHED ||
1055 sk->sk_state == TCP_CLOSING))
1058 /* Already received SHUTDOWN from peer, reply with RST */
1059 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1060 (void)virtio_transport_reset(vsk, NULL);
1064 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1065 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1067 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1068 virtio_transport_wait_close(sk, sk->sk_lingertime);
1070 if (sock_flag(sk, SOCK_DONE)) {
1075 INIT_DELAYED_WORK(&vsk->close_work,
1076 virtio_transport_close_timeout);
1077 vsk->close_work_scheduled = true;
1078 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1082 void virtio_transport_release(struct vsock_sock *vsk)
1084 struct sock *sk = &vsk->sk;
1085 bool remove_sock = true;
1087 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1088 remove_sock = virtio_transport_close(vsk);
1091 sock_set_flag(sk, SOCK_DONE);
1092 virtio_transport_remove_sock(vsk);
1095 EXPORT_SYMBOL_GPL(virtio_transport_release);
1098 virtio_transport_recv_connecting(struct sock *sk,
1099 struct sk_buff *skb)
1101 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1102 struct vsock_sock *vsk = vsock_sk(sk);
1106 switch (le16_to_cpu(hdr->op)) {
1107 case VIRTIO_VSOCK_OP_RESPONSE:
1108 sk->sk_state = TCP_ESTABLISHED;
1109 sk->sk_socket->state = SS_CONNECTED;
1110 vsock_insert_connected(vsk);
1111 sk->sk_state_change(sk);
1113 case VIRTIO_VSOCK_OP_INVALID:
1115 case VIRTIO_VSOCK_OP_RST:
1127 virtio_transport_reset(vsk, skb);
1128 sk->sk_state = TCP_CLOSE;
1130 sk_error_report(sk);
1135 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1136 struct sk_buff *skb)
1138 struct virtio_vsock_sock *vvs = vsk->trans;
1139 bool can_enqueue, free_pkt = false;
1140 struct virtio_vsock_hdr *hdr;
1143 hdr = virtio_vsock_hdr(skb);
1144 len = le32_to_cpu(hdr->len);
1146 spin_lock_bh(&vvs->rx_lock);
1148 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1154 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1157 /* Try to copy small packets into the buffer of last packet queued,
1158 * to avoid wasting memory queueing the entire buffer with a small
1161 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1162 struct virtio_vsock_hdr *last_hdr;
1163 struct sk_buff *last_skb;
1165 last_skb = skb_peek_tail(&vvs->rx_queue);
1166 last_hdr = virtio_vsock_hdr(last_skb);
1168 /* If there is space in the last packet queued, we copy the
1169 * new packet in its buffer. We avoid this if the last packet
1170 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1171 * delimiter of SEQPACKET message, so 'pkt' is the first packet
1174 if (skb->len < skb_tailroom(last_skb) &&
1175 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1176 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1178 last_hdr->flags |= hdr->flags;
1179 le32_add_cpu(&last_hdr->len, len);
1184 __skb_queue_tail(&vvs->rx_queue, skb);
1187 spin_unlock_bh(&vvs->rx_lock);
1193 virtio_transport_recv_connected(struct sock *sk,
1194 struct sk_buff *skb)
1196 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1197 struct vsock_sock *vsk = vsock_sk(sk);
1200 switch (le16_to_cpu(hdr->op)) {
1201 case VIRTIO_VSOCK_OP_RW:
1202 virtio_transport_recv_enqueue(vsk, skb);
1203 vsock_data_ready(sk);
1205 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1206 virtio_transport_send_credit_update(vsk);
1208 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1209 sk->sk_write_space(sk);
1211 case VIRTIO_VSOCK_OP_SHUTDOWN:
1212 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1213 vsk->peer_shutdown |= RCV_SHUTDOWN;
1214 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1215 vsk->peer_shutdown |= SEND_SHUTDOWN;
1216 if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1217 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1218 (void)virtio_transport_reset(vsk, NULL);
1219 virtio_transport_do_close(vsk, true);
1221 /* Remove this socket anyway because the remote peer sent
1222 * the shutdown. This way a new connection will succeed
1223 * if the remote peer uses the same source port,
1224 * even if the old socket is still unreleased, but now disconnected.
1226 vsock_remove_sock(vsk);
1228 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1229 sk->sk_state_change(sk);
1231 case VIRTIO_VSOCK_OP_RST:
1232 virtio_transport_do_close(vsk, true);
1244 virtio_transport_recv_disconnecting(struct sock *sk,
1245 struct sk_buff *skb)
1247 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1248 struct vsock_sock *vsk = vsock_sk(sk);
1250 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1251 virtio_transport_do_close(vsk, true);
1255 virtio_transport_send_response(struct vsock_sock *vsk,
1256 struct sk_buff *skb)
1258 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1259 struct virtio_vsock_pkt_info info = {
1260 .op = VIRTIO_VSOCK_OP_RESPONSE,
1261 .remote_cid = le64_to_cpu(hdr->src_cid),
1262 .remote_port = le32_to_cpu(hdr->src_port),
1267 return virtio_transport_send_pkt_info(vsk, &info);
1270 static bool virtio_transport_space_update(struct sock *sk,
1271 struct sk_buff *skb)
1273 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1274 struct vsock_sock *vsk = vsock_sk(sk);
1275 struct virtio_vsock_sock *vvs = vsk->trans;
1276 bool space_available;
1278 /* Listener sockets are not associated with any transport, so we are
1279 * not able to take the state to see if there is space available in the
1280 * remote peer, but since they are only used to receive requests, we
1281 * can assume that there is always space available in the other peer.
1286 /* buf_alloc and fwd_cnt is always included in the hdr */
1287 spin_lock_bh(&vvs->tx_lock);
1288 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1289 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1290 space_available = virtio_transport_has_space(vsk);
1291 spin_unlock_bh(&vvs->tx_lock);
1292 return space_available;
1295 /* Handle server socket */
1297 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1298 struct virtio_transport *t)
1300 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1301 struct vsock_sock *vsk = vsock_sk(sk);
1302 struct vsock_sock *vchild;
1306 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1307 virtio_transport_reset_no_sock(t, skb);
1311 if (sk_acceptq_is_full(sk)) {
1312 virtio_transport_reset_no_sock(t, skb);
1316 child = vsock_create_connected(sk);
1318 virtio_transport_reset_no_sock(t, skb);
1322 sk_acceptq_added(sk);
1324 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1326 child->sk_state = TCP_ESTABLISHED;
1328 vchild = vsock_sk(child);
1329 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1330 le32_to_cpu(hdr->dst_port));
1331 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1332 le32_to_cpu(hdr->src_port));
1334 ret = vsock_assign_transport(vchild, vsk);
1335 /* Transport assigned (looking at remote_addr) must be the same
1336 * where we received the request.
1338 if (ret || vchild->transport != &t->transport) {
1339 release_sock(child);
1340 virtio_transport_reset_no_sock(t, skb);
1345 if (virtio_transport_space_update(child, skb))
1346 child->sk_write_space(child);
1348 vsock_insert_connected(vchild);
1349 vsock_enqueue_accept(sk, child);
1350 virtio_transport_send_response(vchild, skb);
1352 release_sock(child);
1354 sk->sk_data_ready(sk);
1358 static bool virtio_transport_valid_type(u16 type)
1360 return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1361 (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1364 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1367 void virtio_transport_recv_pkt(struct virtio_transport *t,
1368 struct sk_buff *skb)
1370 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1371 struct sockaddr_vm src, dst;
1372 struct vsock_sock *vsk;
1374 bool space_available;
1376 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1377 le32_to_cpu(hdr->src_port));
1378 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1379 le32_to_cpu(hdr->dst_port));
1381 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1382 dst.svm_cid, dst.svm_port,
1383 le32_to_cpu(hdr->len),
1384 le16_to_cpu(hdr->type),
1385 le16_to_cpu(hdr->op),
1386 le32_to_cpu(hdr->flags),
1387 le32_to_cpu(hdr->buf_alloc),
1388 le32_to_cpu(hdr->fwd_cnt));
1390 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1391 (void)virtio_transport_reset_no_sock(t, skb);
1395 /* The socket must be in connected or bound table
1396 * otherwise send reset back
1398 sk = vsock_find_connected_socket(&src, &dst);
1400 sk = vsock_find_bound_socket(&dst);
1402 (void)virtio_transport_reset_no_sock(t, skb);
1407 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1408 (void)virtio_transport_reset_no_sock(t, skb);
1413 if (!skb_set_owner_sk_safe(skb, sk)) {
1414 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1422 /* Check if sk has been closed before lock_sock */
1423 if (sock_flag(sk, SOCK_DONE)) {
1424 (void)virtio_transport_reset_no_sock(t, skb);
1430 space_available = virtio_transport_space_update(sk, skb);
1432 /* Update CID in case it has changed after a transport reset event */
1433 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1434 vsk->local_addr.svm_cid = dst.svm_cid;
1436 if (space_available)
1437 sk->sk_write_space(sk);
1439 switch (sk->sk_state) {
1441 virtio_transport_recv_listen(sk, skb, t);
1445 virtio_transport_recv_connecting(sk, skb);
1448 case TCP_ESTABLISHED:
1449 virtio_transport_recv_connected(sk, skb);
1452 virtio_transport_recv_disconnecting(sk, skb);
1456 (void)virtio_transport_reset_no_sock(t, skb);
1463 /* Release refcnt obtained when we fetched this socket out of the
1464 * bound or connected list.
1472 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1474 /* Remove skbs found in a queue that have a vsk that matches.
1476 * Each skb is freed.
1478 * Returns the count of skbs that were reply packets.
1480 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1482 struct sk_buff_head freeme;
1483 struct sk_buff *skb, *tmp;
1486 skb_queue_head_init(&freeme);
1488 spin_lock_bh(&queue->lock);
1489 skb_queue_walk_safe(queue, skb, tmp) {
1490 if (vsock_sk(skb->sk) != vsk)
1493 __skb_unlink(skb, queue);
1494 __skb_queue_tail(&freeme, skb);
1496 if (virtio_vsock_skb_reply(skb))
1499 spin_unlock_bh(&queue->lock);
1501 __skb_queue_purge(&freeme);
1505 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1507 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1509 struct virtio_vsock_sock *vvs = vsk->trans;
1510 struct sock *sk = sk_vsock(vsk);
1511 struct sk_buff *skb;
1515 spin_lock_bh(&vvs->rx_lock);
1516 /* Use __skb_recv_datagram() for race-free handling of the receive. It
1517 * works for types other than dgrams.
1519 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1520 spin_unlock_bh(&vvs->rx_lock);
1525 return recv_actor(sk, skb);
1527 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1529 MODULE_LICENSE("GPL v2");
1530 MODULE_AUTHOR("Asias He");
1531 MODULE_DESCRIPTION("common code for virtio vsock");